THESE de DOCTORAT de l’ECOLE POLYTECHNIQUE

presentee par

Johannes QUAAS

L’effet indirect des aerosols:Parametrisation dans des modeles de grande echelle

et evaluation avec des donnees satellitales.

The aerosol indirect effect:Parameterization in large-scale models

and evaluation with satellite data.

Soutenue le 17 novembre 2003, devant le jury compose de :

M. Olivier BOUCHER Co-Directeur de theseM. Jean-Louis BRENGUIER Membre inviteM. Yves FOUQUART Rapporteur, President du juryM. Herve LE TREUT Directeur de theseMme Ulrike LOHMANN RapporteurM. Serge PLANTON ExaminateurM. Michael SCHULZ Examinateur

LMDLaboratoire de Meteorologie Dynamique Institut Pierre Simon Laplace Ecole Polytechnique

2

Contents

I. Introduction 7

1. Introduction 9

1.1. The climate system and climate change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.2. Anthropogenic impacts on climate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.3. The role of clouds in the climate system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101.4. Representation of cloud processes in large-scale models . . . . . . . . . . . . . . . . . . . . . . 111.5. The role of cloud microphysics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.5.1. Processes in liquid water clouds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121.5.2. Processes in ice clouds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131.5.3. Processes in mixed clouds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131.5.4. Impacts of microphysical processes on global climate . . . . . . . . . . . . . . . . . . . 131.5.5. Representation of cloud microphysics in models . . . . . . . . . . . . . . . . . . . . . . 13

1.6. Anthropogenic aerosols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141.7. Aerosol indirect effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.7.1. Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151.7.2. Observational evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.8. Concept of radiative forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191.9. Microphysics and aerosol effects in different GCMs: A review . . . . . . . . . . . . . . . . . . 201.10. Outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221.11. Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2. The LMDZ GCM 29

2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.1.1. Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.1.2. Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.2. Clouds and Precipitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.2.1. Standard model treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.2.2. The microphysical scheme of Boucher et al. . . . . . . . . . . . . . . . . . . . . . . . . 32

2.3. Comparison standard vs. Boucher scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.4. Conclusion: The aerosol indirect effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352.5. Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3. POLDER satellite observations 39

3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.2. The POLDER instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.3. Derived physical data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3

Contents Contents

3.4. Comparison of LMDZ and POLDER data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403.5. Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423.6. Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

II. Model evaluation and process studies 47

4. Evaluating parameterizations of cloud-aerosol processes with single column models and ACE-2

CLOUDYCOLUMN observations 49

4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494.2. Observational data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504.3. The SCMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504.4. Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.4.1. One-day SCM simulations: Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524.4.2. One-day SCM simulations: Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524.4.3. Single-step simulations: Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554.4.4. Single-step simulations: Results and Discussion . . . . . . . . . . . . . . . . . . . . . . 57

4.5. Summary and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 604.6. Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

5. Aerosol indirect effects in POLDER satellite data and in the LMDZ GCM 65

5.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 655.2. Observational data from POLDER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665.3. Model description and “satellite-like” output . . . . . . . . . . . . . . . . . . . . . . . . . . . 665.4. Cloud droplet radius - aerosol index relationship . . . . . . . . . . . . . . . . . . . . . . . . . 685.5. Cloud liquid water path - aerosol index relationship . . . . . . . . . . . . . . . . . . . . . . . 705.6. Indirect effects of anthropogenic aerosols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 715.7. Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725.8. Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

III. Impacts of aerosol forcings on climate change 77

6. Impacts of greenhouse gases and aerosol direct and indirect effects on clouds and radiation in

atmospheric GCM simulations of the 1930 - 1989 period 79

6.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 796.2. Model description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

6.2.1. The atmospheric model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 806.2.2. Parameterization of the aerosol indirect effects . . . . . . . . . . . . . . . . . . . . . . 806.2.3. Experiment setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 816.2.4. Radiation diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

6.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 836.3.1. Radiative forcings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 836.3.2. Radiative impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

6.4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 866.4.1. Trends in cloudiness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 866.4.2. Influence of clouds on the SW radiative impact . . . . . . . . . . . . . . . . . . . . . . 866.4.3. The impact of the second aerosol indirect effect . . . . . . . . . . . . . . . . . . . . . . 87

4

Contents Contents

6.5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 886.6. Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

7. Relative importance of greenhouse gas and aerosol forcings for climate trends of the 1950 - 1989

period in atmospheric GCM simulations. 93

7.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 937.2. Model simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 937.3. Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 947.4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

7.4.1. Land surface temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 947.4.2. Diurnal temperature range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

7.5. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 987.6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1007.7. Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

IV. Parameterization of ice phase processes 105

8. Evaluation of the cloud thermodynamic phase parameterization in the LMDZ GCM by using

POLDER satellite data 107

8.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1078.2. Tools and methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

8.2.1. POLDER-1 satellite observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1078.2.2. The LMD GCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1088.2.3. Simulation of satellite observations from the model results . . . . . . . . . . . . . . . . 108

8.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1098.3.1. Cloud phase to cloud temperature relationship in the satellite observations . . . . . . 1098.3.2. Model relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1108.3.3. Parameter fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

8.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1118.5. Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

9. A new liquid- and ice phase microphysical scheme for the LMDZ GCM: Description and prelimi-

nary results 115

9.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1159.2. Description of the scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

9.2.1. Condensation/Evaporation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1179.2.2. Activation of droplets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1189.2.3. Nucleation of ice crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1189.2.4. Collisions/Coalescence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1199.2.5. Freezing/Melting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1199.2.6. Falling of precipitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1199.2.7. Precipitation fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1209.2.8. Evaporation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

9.3. Results of 1D case studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1219.3.1. Test cases for stratocumulus clouds: ACE-2 . . . . . . . . . . . . . . . . . . . . . . . . 1219.3.2. Test cases for cirrus clouds: GCSS WG2 . . . . . . . . . . . . . . . . . . . . . . . . . . 123

9.4. Preliminary 3D results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

5

Contents Contents

9.5. Summary and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1259.6. Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

10.Conclusions and perspectives 129

10.1. Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12910.2. Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13210.3. Resume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

A. Abbreviation and notation list 139

A.1. Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139A.2. Notation list . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

Bibliography 144

6

Part I.

Introduction

7

1. Introduction

1.1. The climate system and

climate change

Global climate change is considered to be one of themost serious concerns of humankind (United Nations,1992; United Nations, 2002). Anthropogenic green-house gases and aerosols impact considerably the en-ergy balance of the Earth system, possibly provokingadverse effects on social, ecological, and economicalequilibria. This is one of the main reasons why theunderstanding of the Earth’s climate system is of ma-jor importance. If better predictions of the responseof the climate system to anthropogenic perturbationswere available, political decisions against negativeimpacts could be taken, and social adaptations tochanged climate conditions would be possible.

The main scientific tools to understand the climatesystem and to simulate climate change on a globalscale are Earth System Models (ESMs). An ESM iscomposed of several parts, each of which describesa component of the climate system, as atmosphere,ocean, continental surfaces, biosphere, sea ice, orcryosphere, to mention the most important ones. Inthe present work, we focus on the atmospheric part ofthe climate system, which is represented by a generalcirculation model (GCM) of the atmosphere. Nu-merically solving the basic equations that describethe circulation, and incorporating diabatic physicalprocesses by parameterizations, GCMs of the atmo-sphere are able to simulate a multitude of atmo-spheric processes. However, due to limited computerpower and limited understanding of processes, sim-ulations remain unrealistic in many aspects. Tech-niques to improve the computational efficiency inclimate modeling (e.g., Quaas, 2000),

Figure 1.1.: Evolution of the global mean surfacetemperature, shown as the deviationfrom the 1960 to 1990 average from (fromFolland et al., 2001).

and improved computer hardware may help to in-crease resolution and complexity of the models. Thepresent work is meant to contribute to the improve-ment of the understanding of the physical processesfocusing on the aerosol indirect effects.

1.2. Anthropogenic impacts on

climate

Two main perturbations of the atmosphere composi-tion are provoked by humankind. The first and mostimportant one is the increase in tropospheric “green-house gases” (GHG). Gases that absorb terrestriallongwave (LW) radiation in the atmosphere re-emitthe energy at lower temperatures. This causes theso-called “greenhouse effect” (GHE), which results ina warming of the Earth’s surface to keep the energybudget in balance. The main anthropogenic contri-bution to the concentration of greenhouse gases inthe atmosphere stems from carbon dioxide (CO2),which originates from the combustion of fossil fuel

9

1.3. The role of clouds in the climate system 1. Introduction

and burning of biomass. Other anthropogenic green-house gases are methane (CH4), nitrous oxide (N2O),chlorofluorocarbons (CFC), and tropospheric ozone(O3). All of these have increased since the begin-ning of the industrialization in about 1750 (Fig. 6.1;Myhre et al., 2001). In agreement with the expectedwarming due to the anthropogenic greenhouse effect,an increase in global mean surface temperature hasbeen observed (Fig. 1.1; Folland et al., 2001).However, the observed temperature increase of 0.6

K for the period 1900 to 2000 is somewhat smallerthan the warming expected due to the greenhouse ef-fect (Mitchell et al., 2001). This may be explainedby important internal variability or the existence ofanother forcing of climate change. A negative forcingdue to anthropogenic aerosols could be responsible foran offset of part of the warming GHE (Wigley , 1989;Charlson et al., 1989; Boucher and Anderson, 1995).Since about two decades, the impact of anthropogenicaerosols has been studied. However, in the latest In-tergovernmental Panel on Climate Change (IPCC)report, the level of scientific understanding of theimpact of aerosols on the radiation balance has stillbeen assessed as “low” or even “very low”. Never-theless, it is commonly accepted that anthropogenicaerosols may have an impact strong enough to offsetthe warming by anthropogenic greenhouse gases atleast to a significant fraction. The impact of aerosolson radiation is twofold. Firstly, a “direct” radiativeeffect exists, as aerosols scatter and absorb sunlight,thus generally enhancing the planetary albedo. Sec-ondly, aerosols may act as cloud condensation nu-clei thus changing cloud properties. These so-called“aerosol indirect effects” will be defined in Section1.7. Besides these direct and indirect effects, absorb-ing aerosols may also cause a “semi-direct effect”.When radiation is absorbed by aerosols in the at-mosphere, the surrounding air is heated and thenevaporates clouds or hinders their formation. As aresult it occurs a decrease in cloud water content orin cloudiness decreases the albedo, which generallycauses a warming. This process has been modeled forconditions observed during the INDOEX experiment(Ackerman et al., 2000a). Recent studies reveal thataerosols in large quantities, like dust, also have a non-negligible effect in the longwave spectrum (Dufresne

et al., 2002).

1.3. The role of clouds in the

climate system

From a meteorological point of view, water is themost important tracer element of the atmosphere, inwhich water exists in all three thermodynamic states.The atmospheric source of water is evaporation overland and ocean surfaces. Water vapor is distributedthroughout the troposphere and interacts with theenergy balance. In the process of evaporation andcondensation, latent heat is transported from thesurface to the atmosphere. As a gas absorbing inthe LW spectrum, water vapor causes an importantgreenhouse effect.

Water vapor eventually condensates to the liquid orsolid phase thus forming clouds. In the Earth’s atmo-sphere, clouds exist at all layers of the troposphere,up to about 11 km. At temperatures above 0◦C, theyconsist of liquid water, below about –35◦C to –40◦C,purely of ice, and in between, mixed-phase cloudsare observed (Houze, 1993). On the global scale, thearea cloud cover is about 60%. When water vaporcondenses to form clouds, latent heat is released,and the respective layer of the atmosphere is thusheated. When cloud water re-evaporates, latent heatis uptaken from the environment. Clouds also in-teract with radiation. In the solar spectrum, cloudsreflect sunlight. As the cloud albedo generally islarger than the albedo of the underlying surface, theoccurrence of clouds enhances the planetary albedo.In the terrestrial spectrum, clouds absorb radiationand cause a greenhouse effect. These so-called “cloudradiative forcings” (CRF) in the SW and LW spectraare of opposite sign (Kandel et al., 1998). The neteffect is positive for high-level clouds and negativefor low-level clouds. On a global scale, clouds have anet negative radiative forcing (Kiehl and Trenberth,1997).

Precipitation is formed through cloud microphysi-cal processes. Thus, clouds are at the origin of oneof the socially, ecologically, and economically mostimportant meteorological quantities. While conden-

10

1. Introduction 1.4. Representation of cloud processes in large-scale models

sation from water vapor is the source for cloud water,precipitation formation constitutes its final sink. Therain-forming processes thus influence the lifetime andwater content of clouds.

1.4. Representation of cloud

processes in large-scale

models

In contrast to atmospheric motion, the condensationand cloud forming processes cannot be described bya fundamental equation in a GCM. They need to berepresented as a function of the variables explicitlycalculated in the model, a method known as pa-rameterization. Among the first schemes applied inGCMs which parameterized clouds rather than sim-ply imposing to the model an observed distributionwas the one of Sundqvist (1978). He calculated cloudwater as a prognostic variable in his model, and in-troduced a fractional cloudiness. While some GCMcloud schemes use a binary cloud fraction of zero orunity (e.g., Fowler et al. (1996)), most others use asimilar approach to define in a grid box an entirelycloudy and a cloud-free fraction (e.g., Slingo, 1987;Del Genio et al., 1996). However, in the vertical,at the exception of the schemes of Del Genio et al.(1996) and Jakob and Klein (1999), a cloud is con-sidered to fill the model layer completely.

The fractional cloudiness is often calculated fromthe grid box mean relative humidity (RH) using a“critical value”. For example, Lohmann and Roeck-ner (1996) use a vertically varying critical relativehumidity which decreases from RHcrit=99% near thesurface to RHcrit=60% in the free troposphere. Otherschemes use a probability density function (PDF) ofthe total water mixing ratio in each grid box to de-fine the saturated fraction of the grid cell (Sommeriaand Deardorff , 1977). The form of the PDF can beconsidered for example of uniform (Le Treut and Li ,1991) or triangular (Smith, 1990) shape. However,schemes using such fixed PDFs do not differ substan-tially from parameterizations using a critical RH.As a first attempt to introduce a subgrid scale dis-tribution of cloudiness based on physical processes,

Tompkins (2002) defines a PDF of the total watermixing ratio using a Beta function. The schemeuses skewness and variance of the PDF as prognosticmodel variables, which are parameterized in terms ofconvection, turbulence, wind shear, and cloud micro-physics.

The clouds defined in the cloudy fraction of a gridcell are generally considered homogeneous. This iscalled the assumption of plane parallel homogeneous(PPH) clouds. In several studies, it has already beenshown that this assumption introduces large biaseson the radiation transfer (e.g., Pincus and Klein,2000; Calahan, 1994; Carlin et al., 2002; Stubenrauchet al., 1997), as well as on the precipitation formation(Jakob and Klein, 1999; Wood et al., 2002)). Recentstudies try also to deal with a possible introductionof statistical distributions in the vertical direction(Di Giuseppe, 2003). However, the treatment of thesubgrid-scale variability is to a large extent beyondthe scope of the present study and will be addressedin future work.

In the vertical, clouds are generally assumed to over-lap either independently between each two layers(random overlap assumption), to overlap to the maxi-mum possible extent (maximum overlap assumption),or to overlap in a maximum way in adjacent layersand randomly otherwise (maximum-random overlapassumption). Recently, Hogan and Illingworth (2000)derived from radar observations a parameterizationbased on the random-maximum assumption, but in-cluding an additional parameter taking into accountthe vertical separation of two cloudy layers. Theknowledge of the overlap of clouds in different layersis necessary at least for the radiation transfer andmay be used as well in the microphysics (see Chapter9).

Most GCMs distinguish between convective andstratiform cloudiness. In a scheme parameterizingthe deep convection, precipitation stemming from theconvective events is diagnosed. A convective cloudcover may be derived using this quantity (Trenberth,1995), while the so-called “stratiform” cloud coveris calculated using the different schemes described

11

1.5. The role of cloud microphysics 1. Introduction

above. It is in this stratiform cloud parameterizationthat the cloud and precipitation microphysics is gen-erally treated as well. Recent studies try to overcomethe separation of stratiform and convective schemes(Bony and Emanuel , 2001).

1.5. The role of cloud

microphysics

All those processes which act on the scale of in-dividual cloud particles are known as “cloud mi-crophysics”. They include the formation of clouddroplets and ice crystals by the activation of hygro-scopic aerosol particles, their growth by condensationdeposition, and the processes which lead to precipita-tion formation and growth of precipitation particles.

Microphysical processes for warm (liquid water) andcold (ice) clouds may be distinguished.

1.5.1. Processes in liquid water clouds

Liquid cloud droplets can be considered spherical to avery good approximation. They are formed by depo-sition of water vapor on hydrophilic aerosol particles.Such aerosol particles are called activated, when theirsize exceeds a certain critical size. The number ofdroplets activated and subsequently the size of eachdroplet is determined by (Twomey , 1959)

1. The number concentration of hydrophilicaerosol

2. The supply of water vapor to condense, whichin turn is given by the supersaturation of theambient air.

The Kohler theory describes the supersaturation ofthe ambient air, S, in dependence of the size of awetted aerosol and its chemical composition:

S =(

1− b

r3d

)exp

(a

rd

)(1.1)

where exp(a/rd) is the “curvature term” which de-scribes the decrease in necessary supersaturation fora droplet to be activated with increasing droplet ra-dius, rd, and (1− b/r3

d) is the “solution term” whichdescribes the dependence on the relative quantity of

dissolved aerosol material and is dominant for smalldroplets. The variable a depends on temperature anddensity of the moist air, and b on the temperatureand the chemical properties of the solution (Prup-pacher and Klett , 1997). Recently, Charlson et al.(2001) proposed an important additional impact ofsoluble gases on the formation of clouds, which couldrender the process of activation more efficient.

Supersaturation in an air parcel is generated by cool-ing of wet air. This is generally the case in ascendingair motion. The supersaturation in a parcel is de-pendent on the updraft speed. With this, the clouddroplet number concentration (CDNC) can be calcu-lated as a function of aerosol number concentrationand updraft speed, w:

Nd ∝ f (Na;w) (1.2)

where Nd is the cloud droplet number concentration[m-3], Na the number concentration of hydrophilicaerosol [m-3], and w, the updraft velocity [m s−1].

Cloud droplets have a typical size of about 3 - 20µm. Eventually, raindrops may form from clouddroplets by collision and coalescence (e.g., Bartlett ,1966)). For two spherical particles, the collision andcoalescence is given by:

ρair∂q

∂t=

∫π E(r, r′) (r + r′)2 (1.3)

|V (r)− V (r′)| q(r′) n(r′) dr′

where q is the mixing ratio [kg kg−1], ρair the air den-sity [mg m−3], r the particle radius [m], Er,r′ is thecollision efficiency, which also determines the collisioncross section (e.g., Hocking , 1959), and V (r) fall ve-locity of a particle of radius r. A distinction betweencloud droplets and raindrops may be defined by theirfall velocity. While cloud droplets have a very smallterminal fall velocity, raindrops fall fast enough toleave the cloud. Typical raindrops are in the rangeof 100 µm to 1 mm. Comparing the sizes of clouddroplets and raindrops, 102 to 107 cloud droplets areneeded to form a raindrop. If cloud droplets andraindrops are considered as two different species ofwater, two processes can be distinguished, by whichrainwater is created from cloud water. The collision

12

1. Introduction 1.5. The role of cloud microphysics

and coalescence of cloud droplets is known as auto-conversion. This process is the initiation of rain. Forthis to be efficient, droplets need to have differentfall velocities, which is determined by the size of thedroplets. For an onset of autoconversion, thus, thedroplet size spectrum has to be relatively broad. Inaddition, to form larger raindrops, colliding dropletshave to be large enough. From observational studiesit has been deduced that both conditions are given,when the particle radius exceeds a certain “critical”size (Kessler , 1969), which has been determined tobe about 10 µm (Pawlowska and Brenguier , 2003).The second process is the collision and coalescence ofraindrops and cloud droplets. While falling througha cloud, raindrops collect by this process plenty ofcloud droplets, growing considerably.

1.5.2. Processes in ice clouds

Below the freezing point of 0◦C, clouds may includeice crystals, and for temperatures below –40◦C, allcloud water consists of ice. Ice crystals are formedby different nucleation processes.

Homogeneous nucleation : A wetted aerosolfreezes, and grows by deposition of water vapor.

Heterogeneous nucleation : A supercooled wettedaerosol freezes by interaction with a “freezingnucleus”. Contact nucleation, deposition nu-cleation, and immersion nucleation are distin-guished (Vali , 1985), which refers to the colli-sion of a supercooled droplet with an aerosol,the deposition of water vapor on a wettedaerosol, and the freezing by the activity of aninsoluble aerosol inside a droplet, respectively.

The ability of different aerosol species to act as freez-ing nuclei is still very unclear and the subject ofrecent research (DeMott et al., 2003).

In modeling studies, ice crystals are often consid-ered to have spherical shape. This approximation is,however, generally not valid and may introduce largeerrors in some cases (Doutriaux-Boucher et al., 2000).

Ice crystals may form snow flakes by collision and coa-lescence processes. The formation of snow from cloud

ice has been studied much less intensively comparedto the corresponding liquid water process. Whenassuming spherical particles, this process can be de-scribed by an equation similar to the one given earlierfor liquid particles.

1.5.3. Processes in mixed clouds

When frozen and liquid particles collide, rimed par-ticles are produced. When rime is formed, ice crys-tal splinters are produced. This so-called “secondaryice production” creates a large number of new icecrystals and has first been observed by Hallett andMossop (1974). It is therefore also known as theHallett-Mossop process. The saturation water vapormixing ratio is lower when considering ice surfacescompared to liquid water surfaces. This implies thatair may be supersaturated with respect to ice, evenwhen its subsaturated w.r.t. liquid water. This has asa consequence the so-called Bergeron-Findeisen pro-cess by which in mixed-phase clouds, ice crystals growat the expense of liquid cloud droplets (Bergeron,1933). Further, while the atmosphere almost neveris supersaturated w.r.t. liquid water at spatial scaleslarger than a few cubic meters, supersaturation w.r.t.ice is common and may reach as much as 70%. How-ever, so far, almost no GCM is able to treat the icesupersaturation.

1.5.4. Impacts of microphysical processes

on global climate

Microphysical processes are expected to react toor impact on global climate change. For example,aerosol indirect effects are commonly considered tobe an anthropogenic perturbation of the climate.As cloud microphysics are complex processes andtheir impacts on global scale climate still are poorlyunderstood, microphysical processes also serve to ex-plain theories which try to link the observed climatechange to natural phenomena or to show mechanismsinhibiting a profound climate change. For example,the so-called “Iris hypothesis” claims that with ris-ing sea surface temperature (SST) due to greenhousewarming, cirrus cloudiness in the tropics would bereduced (Lindzen et al., 2001). Enhancement ofprecipitation-forming microphysical processes due to

13

1.6. Anthropogenic aerosols 1. Introduction

increased cloud water content would be responsi-ble for such a mechanism. Another study claimsfeedback of cloud microphysics to cosmic radiationvariability to explain the observed climate changevia the altering in aerosol particle charge (Svensmarkand Friis-Christensen, 1997). These two theoriesremain largely controversial (e.g., Lin et al., 2002;Hartmann and Michelsen, 2003; Sun and Bradley ,2002).

1.5.5. Representation of cloud

microphysics in models

Several properties of the different water species andaerosols have to be considered to model cloud micro-physical processes. These include the size distribu-tion and chemical composition of aerosols, and thesize distribution of cloud droplets, ice crystals, andprecipitating particles. Some approaches exist to ex-plicitly model cloud microphysical processes, with aparticular focus on the rather complicated mecha-nism of activation of aerosol particles (e.g., Wobrocket al., 2001).

Generally, two types of microphysical schemes may bedistinguished, those resolving the size distributions ofaerosol and water particles in size bins (e.g., Young ,1974a), and those which assume a fixed size distribu-tion and partition the water into separate categoriesfor cloud water and precipitation. The latter ones arecalled “bulk” schemes (e.g., Kessler , 1969). In thescope of the present study the latter type of param-eterizations is rather of interest. Often, such moreor less sophisticated parameterizations have beendeveloped in the framework of cloud-resolving mod-els (CRMs) or large-eddy simulation models (LES),which are able to resolve structures with a typicalsize of a few to several hundreds of meters. However,with such resolutions, microphysical processes stillcannot explicitly be modeled. Even on such scales,they need to be parameterized. Schemes suitableto be used in CRMs have been adapted in the pastseveral years for application in large-scale models forthe sake of more physical realism in GCMs, and to beable to treat the influence of aerosols on clouds. Byadapting CRM microphysical schemes, parameters

often have to be adjusted for large-scale models to beable to simulate realistic cloud properties and pre-cipitation distributions. An example is the “criticalradius” in the autoconversion formulation, which isset to a lower value in GCMs compared to the valuein CRMs or the measured values from observations(Rotstayn, 2000). The reason is, that at small scales,some droplets may exceed the critical value and onsetprecipitation, which then by accretion forms largerrain rates. At the coarse resolution of a GCM, how-ever, the rather large critical radii very rarely areexceeded, or very large cloud water contents wouldhave to be present on large scales.

1.6. Anthropogenic aerosols

In the context of general circulation modeling, fivedifferent species of aerosols are distinguished, whichdiffer in chemical composition and origin. Of par-ticular interest for their impact on climate are theirhydrophilic or hydrophobic properties, which deter-mine their ability to serve as cloud condensation nu-clei (CCN), and the degree to which they absorb ra-diation. In the framework of anthropogenic climatechange, and to quantify the contribution of anthro-pogenic aerosols to climate perturbations, it is impor-tant to distinguish natural and anthropogenic aerosolsources.

• Sulfate aerosols (SO=4 ) are typically of anthro-

pogenic origin. Sources include the combustionof sulfur-containing fossil fuel and the burn-ing of biomass. Sulfate particles are relativelysmall, and generally hydrophilic. When incor-porated in cloud droplets and subsequently inrain drops, sulfate aerosols also cause the so-called “acid rain”, as they dissolve in water toform sulfuric acid.

• Organic carbon, which originates for examplefrom the burning of biomass. Some organicaerosols have hydrophilic properties. Organiccarbon aerosols are slightly absorbing in theSW spectrum.

• Black carbon (soot) is at least near its origingenerally hydrophobic. Soot particles consti-tute the most absorbing aerosols.

14

1. Introduction 1.7. Aerosol indirect effects

• Sea salt, which is entirely of natural origin.However, possibly increasing wind speeds inan anthropogenically altered climate would in-crease the source strength for sea salt aerosols.Sea salt is hydrophilic. In remote oceanicregions, sea salt almost constitutes the onlyspecies of aerosol.

• Dust often contributes most to the total massof aerosols. A main source for dust aerosols aredeserts thus largely a natural origin. Land-usechange or increase windspeed due to anthro-pogenic climate change, however, may consti-tute an anthropogenic source for dust aerosols.Generally, dust aerosols are hydrophobic. Theymay constitute ice nuclei (DeMott et al., 2003).Dust is slightly absorbing.

Individual aerosol particles may be constituted of dif-ferent chemical compositions. Different aerosol typesmay be mixed either “internally” or “externally”,referring to aerosol particles consisting of several dif-ferent chemical compounds, or an aerosol populationwhere individual particles consist of pure chemicalspecies.

It shall be noted that aerosols have a relativelyshort lifetime in the atmosphere. As troposphericaerosol particles have a considerable mass, they aredeposited after about a week at the surface. Hy-drophilic aerosols which are included in cloud andsubsequently precipitation particles can be depositedat the surface by washout. In consequence, aerosolconcentration in the atmosphere is large particularlynear the main source regions, and very low over re-mote maritime regions. Another consequence is thatif the anthropogenic production of aerosol stops, anydirect impact on climate will vanish as well after acouple of days or weeks.

1.7. Aerosol indirect effects

The impact of aerosols on clouds is known as aerosol“indirect” effect (AIE). The impact of aerosols onlow-level liquid water clouds has been the first one tobe studied. We will limit this section to those cloudsas well.

Topopause

Surface

natural + anthropogenic aerosolsnatural aerosols

SW radiation

(a) ADE

(b) First AIE

natural aerosols natural + anthopogenic aerosols

Surface

TropopauseSW radiation

(c) Second AIE

natural aerosols natural + anthopogenic aerosols

Surface

TropopauseSW radiation

(d) Semi−direct eff.

natural aerosols natural + anthopogenic aerosols

Surface

TropopauseSW radiation

Figure 1.2.: Schematic description of (a) the aerosoldirect effect, (b) the first indirect effect,(c) the second indirect effect, and (d)the semi-direct effect. Dots symbolizeaerosols (absorbing aerosols are red andhydrophilic aerosols blue), light blue cir-cles cloud droplets, wavy lines of differ-ent thickness indicate radiative transferof different intensity, arrows (−→) indi-cate a temporal evolution, and the moreor less densely hatched area below the“cloud” indicates precipitation of differ-ent intensity.

15

1.7. Aerosol indirect effects 1. Introduction

However, aerosols may also have an impact on iceclouds (e.g., Georgii , 1959). It has been arguedthat aerosol from air traffic exhaust may contributeto the increase in cirrus cloudiness due to air traffic(Boucher , 1999). As in the IPCC definition and un-less specified otherwise explicitly, the terms of aerosoldirect and indirect effects refer here to the impacts ofanthropogenic only aerosol.

1.7.1. Definition

Hydrophilic aerosols constitute cloud condensationnuclei (CCN). Generally, such particles are necessaryfor any cloud particle to form, as the spontaneous for-mation of liquid or solid particles from water vaporwould need supersaturations of the order of severalhundred percents, which never occurs in the atmo-sphere (Houze, 1993). An increase in aerosol numberconcentration makes more CCN available and morecloud particles may be formed. When the cloud liq-uid water content is constant the individual clouddroplets become smaller. This effect causes an in-crease in cloud albedo and is called alternativelythe first AIE, cloud albedo effect, or Twomey effect(Twomey , 1974).

As mentioned earlier, autoconversion of clouddroplets to form rain drops depends on the size ofthe cloud droplets. For smaller droplets, the precipi-tation formation is less efficient or may even be sup-pressed. So, the lifetime of clouds is enhanced. Forreflecting clouds the planetary albedo is increased.This is called second AIE or cloud lifetime effect(Albrecht , 1989).

1.7.2. Observational evidence

Field studies

A first estimate of the anthropogenic contributionto CCN is given by Squires (1966). From aircraftmeasurements, they deduce a source strength of an-thropogenic CCN of the order of 14% of the naturalCCN sources, when they extrapolate their measure-ments to the USA region, and of 5% for the entirenorthern hemisphere.

Brenguier et al. (2000b) measure several microphys-

ical parameters in maritime stratocumulus cloudsduring the ACE-2 campaign. They find much largerCDNC (up to more than 400 cm−3) in polluted com-pared to clean conditions (less than 100 cm−3). Op-tical thickness of clouds, τc, is shown to dependstrongly on cloud geometrical thickness, H. Theyfind a relationship close to the the theoretical one de-rived for an adiabatically rising air parcel of τc ∝ H

53

for constant CDNC (Brenguier , 1991). Pawlowskaet al. (1999) show the dependence of τc on CDNC,Nd, following the theoretical value of τc ∝ N

13d when

scaling the optical thickness by H53 . Brenguier et al.

(2003) analyze data of Pawlowska and Brenguier(2003) and Schuller et al. (2003), also from ACE-2 measurements. They relate cloud optical thick-ness, cloud droplet radius, and cloud droplet numberconcentration on a large scale (60x60 km2) for sev-eral cases, from clean to heavily polluted conditions.Looking at clean to moderately polluted conditions,they find positive correlations between τc and Nd andre and Nd, respectively, with the behavior of τc ∝ N

13d

and re ∝ N− 1

3d expected from theory. However, when

adding the strongly polluted cases, a different rela-tionship is found. They attribute this finding to thedifferences in cloud geometrical thicknesses in thepolluted compared to clean cases. Again using ACE-2 data, Snider and Brenguier (2000) show from aclosure study that the Twomey equation (Twomey ,1959) of droplet activation was valid in the casesmeasured. The data derived from ACE-2 on thelarge scale are used in Chapters 4 and 9.

Hudson and Yum (2001) carry out an in situ studywith aircrafts measuring CCN, CDNC, and drizzlewater content above a coastal site in Florida. Distin-guishing airmasses advected from the continent andthe ocean, they show that clouds in continental air-masses generally contain more but smaller dropletsand less drizzle. These two findings support thehypotheses for the first and second aerosol indirecteffects, respectively.

Liepert and Lohmann (2001) compare observations ofradiation fluxes at the surface measured from ground-based instruments to radiation fluxes calculated withtwo GCM simulations, one calculating aerosol indi-

16

1. Introduction 1.7. Aerosol indirect effects

rect effects from natural and anthropogenic aerosols,and a second one using natural aerosols only. Theyfind some features indicating that the anthropogenicaerosol indirect effect may be necessary to simulaterealistically the observed radiation fluxes. However,their results remain ambiguous due to deficiencies inthe simulated cloudiness.

Peng et al. (2002) examine data from liquid watercloud cases collected in the RACE and FIRE.ACEcampaigns. They find cloud droplet effective radii(CDR) by 2 - 6 µm smaller in polluted compared toclean cases. In most cases, cloud optical depth andcloud albedo are found to be larger in polluted com-pared to clean cases. They establish a relationshipbetween CDR and cloud albedo, which is found tobe negative for clean cases (in agreement with thefirst aerosol indirect effect). For polluted cases, how-ever, a positive relationship between cloud albedoand CDR is observed, which they attribute to thesecond indirect effect. Earlier investigation of similarrelationships with their model confirms this finding(Lohmann et al., 2000b).

Ramanathan et al. (2001a) report from the IndianOcean Experiment (INDOEX) important indirecteffects of aerosols originating from the Indo-Asianhaze. In polluted compared to clean clouds, whileliquid water content was approximately constant,much larger CDNC (315 cm−3 compared to 90 cm−3)and smaller CDR (≤ 6 µm compared to ≥ 7.5 µm)are observed, giving a radiative forcing due to thefirst aerosol indirect effect of –5 Wm−2 at the topof the atmosphere. Ramanathan et al. (2001a) showa logarithmic dependence of cloud optical thicknesson aerosol optical thickness, with τc ≈14 for largeaerosol optical thicknesses (τa ≥ 0.5). From a mod-eling study of the INDOEX campaign, an increase inlow-level cloud lifetime results in an increase in low-level cloudiness of 2 % and a corresponding radiativeimpact due to the second aerosol indirect effect of –2Wm −2. The same study introduces the “semi-direct”effect of aerosols, which would evaporate cloud wateror suppress the formation of clouds through heatingof the atmosphere by absorbing aerosols. This effectis estimated to correspond to a radiative forcing of 4

Wm−2. Ramanathan et al. (2001b) further interpretthe INDOEX results in cloud-radiation models andanalyze the impact of different aerosol species. Forthe global mean, they deduce a first AIE of –0.3 to–1 Wm−2 of sulfate aerosols and of –0.5 to –2 Wm−2

when adding carbonaceous aerosols.

Satellite data

Some of the first evidence for impacts on clouds byanthropogenic aerosols has been provided by analysisof ship tracks. Looking at TIROS satellite images.Conover (1966) find “anomalous cloud lines”, lin-early shaped lines of brighter clouds of up to 500 kmlength and a width of 25 km. They argue that theproduction of aerosols by ship exhaust was respon-sible for an altering of atmospheric concentration ofCCN which could increase the cloud albedo by up to20%.

Using two channels of the Advanced Very High Res-olution Radiometer (AVHRR), Coakley Jr. et al.(1987) show the coincidence of smaller cloud parti-cles and larger cloud albedo in ship tracks comparedto the environment. More recently, Taylor et al.(2000) investigate ship tracks in the Monterey AreaShip Track (MAST) experiment. For relatively cleanclouds, they find CDNC changes of up to a factor of2 by comparing clouds affected by ship exhaust withthose in the cleaner environment. In some cases,they find a suppression of drizzle and subsequently aconsiderable increase in LWP. A combination of bothaerosol effects yields an increase in cloud radiativeforcing at the top of the atmosphere of up to a factorof 4.

In a pioneering study Han et al. (1994) derive cloudtop droplet radii for liquid water clouds from AVHRRsatellite measurements over oceans on a global scalein the framework of the International Satellite CloudClimatology Project (ISCCP). They show that CDRover continents are smaller by 2 - 3 µm than overoceans. Comparing maritime clouds in the southernand northern hemispheres, a difference of 1 µm, withthe smaller droplets in the NH, is also found. Han

17

1.7. Aerosol indirect effects 1. Introduction

et al. (1998) examine the relationship between CDRand cloud albedo. They show, that cloud albedoincreases with decreasing CDR for most continen-tal clouds and for optically thick clouds. However,for optically thin clouds over oceans, a decrease incloud albedo is observed with increasing CDR. As apossible explanation, Lohmann et al. (2000b) deducefrom a GCM modeling study that precipitating ver-sus non-precipitating clouds would be responsible forthis behavior. Han et al. (1998) also show a generalpositive correlation between cloud liquid water path(LWP) and CDR. Han et al. (2002) further investi-gate the dependence of LWP on CDNC. They findin only one third of the cases examined an increasein LWP with increasing CDNC as expected due tothe second aerosol indirect effect. Another third ofthe cases shows approximately constant LWP, con-sistent with the Twomey hypothesis. Finally, a thirdof the cases shows a decrease in LWP with increasingCDNC, a process possibly related to boundary layerdynamics.

Kaufman and Nakajima (1993) evaluate AVHRRmeasurements of CDR over the Brazilian AmazonBasin and show that smoky (smoke optical depthof τs=2.0) compared to non-smoky (τs ≤ 0.1) re-duce the CDR from 15 to 8 µm, where the smokecomes from forest fires and supplies efficient CCN.On the other hand, absorption by black carbon inthe smoke reduces the reflectance. The net effect ofsmoke in this case is a reduction of the planetaryalbedo. Kaufman and Fraser (1997), however, againevaluating AVHRR data, show a decrease in CDRand an increase in cloud reflectance with increasingsmoke optical depth at constant latitude. The re-lationship between smoke optical depth and cloudalbedo and CDR, respectively, shows steeper slopesfor larger integrated water vapor contents in the at-mosphere (precipitable water, PW). They furthershow that a strong decrease in CDR with increasingsmoke optical depth corresponds to a strong increasein cloud albedo. The radiative forcing caused by thisimpact of smoke on cloud albedo is evaluated to be–2 Wm−2, which is a rather low value given that a re-gion with high pollution is examined. Feingold et al.(2001), analyzing the same data, suggest a saturation

effect for these very large aerosol concentrations. Forthe same experiment, Reid et al. (1999) measured insitu CDR using an aircraft. They find CDR values of3 to 8 µm, thus much smaller than the ones derivedfrom the satellite data.

Nakajima et al. (2001) relate several cloud micro-physical quantities to aerosol number concentra-tion from AVHRR data over oceans on a globalscale. They show an increase in column-integratedCDNC and a decrease in CDR with increasing aerosolnumber concentration, in accordance with the firstaerosol indirect effect. For large aerosol concentra-tions (Na > 107.8m−3), an increase in cloud opticalthickness is also derived with increasing aerosol con-centration. However, no correlation is found betweenNa and cloud liquid water path. Kawamoto andNakajima (2003) identify possible artifacts in thederivation of CDR from AVHRR data looking at along time series of measurements (1985-1994). Theyalso identify a slight decrease in CDR from 1985 to1994, which they attribute to an increase in globalmean aerosol concentration.

Wetzel and Stowe (1999) show a negative correlationbetween CDR and aerosol optical thickness, τa, de-rived from AVHRR data for low-level marine stratusclouds. Additionally, they show a seasonal and re-gional variability that compares well with variationsin aerosol sources. They find a saturation effect inthe CDR to aerosol burden relationship for τa > 0.12.The interpretation of their results is somehow diffi-cult as they look at largely averaged data. Thus,correlations may be due to other influences such asgeographical ones. Other studies such as the one ofNakajima et al. (2001) may be criticized in a similarmanner.

Schwartz et al. (2002) correlate AVHRR data of CDRand cloud albedo with sulfate aerosol concentrationssimulated with a hemispheric chemistry transportmodel (CTM) for the North Atlantic ocean for twoperiods of several days in April 1987. They finda negative correlation between sulfate aerosol con-centration and CDR, but no correlation of aerosolswith cloud optical thickness or cloud albedo. They

18

1. Introduction 1.8. Concept of radiative forcing

attribute this to the large variability in LWP whichmasks possible aerosol influences on cloud albedo.

Chameides et al. (2002) correlate ISCCP cloud opti-cal thickness and column burdens of anthropogenicaerosol modeled with a coupled regional climate -chemistry transport model. For annually averageddata from 1990 - 1993 over the East Asian region,they find a correlation coefficient of r2 > 0.6 betweenthe two quantities, which corresponds to a confidencelevel of >99.99%, so that 60% of the variability incloud optical thickness are explained by variationin aerosol burdens. However, they do not observea consistent correlation between aerosol burden andcloud cover, indicating that either the second aerosolindirect effect does not play an important role orwas offset by other processes such as the semi-directeffect.

Breon et al. (2002) relate CDR and an aerosol in-dex inferred from data of the POLarization and Di-rectionality of the Earth’s Reflectances (POLDER)instrument. They find a decrease in CDR with in-creasing aerosol index, which is steeper over oceansthan over land. The data from Breon et al. (2002) isused in Chapter 5.

Coakley, Jr. and Walsh (2002) analyze several hun-dred cases of ship tracks using AVHRR data. Theyfind a statistically significant increase in cloud re-flectance corresponding to a decrease in cloud dropletradius of 2 µm or more. Clouds with relativelysmall albedo and large droplet sizes are found tobe the most susceptible. The change in cloud opti-cal thickness found is not as large as expected fora given change in CDR. They argue that either thecloud LWP decreased in polluted compared to cleanclouds, or that absorption by aerosols takes place,which would mask part of the cloud albedo increase.Coakley, Jr. and Walsh (2002) do not find a coher-ent variation in cloud top height in polluted versusclean environments, in contrast to other studies (e.g.,Brenguier et al., 2000b).

Rosenfeld and Lensky (1998) use multi-spectralAVHRR measurements to relate CDR to the tem-

perature in convective clouds, the temperature beinga measure for the height in the cloud. They identifylarge differences between continental and maritimeclouds. In maritime clouds, CDR increase rapidlywith height in a cloud due to coalescence up to amaximum size where precipitation onsets. This isnot observed in continental clouds. They also findthat glaciation onsets at much lower temperatures incontinental compared to maritime clouds. They ob-serve an important influence of pollution in biomassburning and urban regions. Using Tropical RainfallMeasuring Mission (TRMM) observations, Rosenfeld(1999) shows that clouds affected by smoke fromforest fires do not form precipitation where similarclouds in the neighborhood do. With a similar ap-proach, Rosenfeld (2000) shows for several differentgeographical locations that similarly, a suppressionof precipitation in clouds can be inferred by urbanand industrial pollution.

1.8. Concept of radiative forcing

At least in the framework of climate modeling, theconcept of “radiative forcing” of an external pertur-bation of the climate system is very useful. A radia-tive forcing is defined as the difference in radiationflux at the tropopause between perturbed and un-perturbed conditions. It is calculated for fixed tro-pospheric temperature and humidity profiles, lettingthe stratosphere adjust. With this definition, it hasbeen shown (e.g., Hansen et al., 1997), that the radia-tive forcing, ∆F , due to any small perturbation of theclimate system is linearly related to a correspondingchange in global mean surface temperature, ∆Ts

∆Ts = λ ∆F (1.4)

where λ is the so-called sensitivity parameter[K W−1 m2], which depends on the model used,but is approximately constant for any perturbationfor which the radiative forcing is calculated. Theradiative forcing is especially useful to compare themagnitude of different perturbations of the climatesystem.

Concerning the radiative forcing of the aerosol in-direct effects, the two different effects have to be

19

1.9. Microphysics and aerosol effects in different GCMs: A review 1. Introduction

distinguished. The first aerosol indirect effect is con-sidered to change the cloud droplet size of low-levelliquid water clouds. This change impacts directlyon the cloud optical thickness, which in turn altersthe cloud albedo. The impact on the LW radiativespectrum is small and is neglected here. Thereforethe radiative forcing of the first aerosol indirect effectis generally calculated on-line in a model simulation,doing the radiation transfer calculation twice, a firsttime with a profile of the cloud optical thickness asit were in natural conditions, and a second time us-ing the profile which is influenced by the additionalanthropogenic aerosols. The radiative forcing of thefirst aerosol indirect effect is then given as the dif-ference in the radiative fluxes calculated for the topof the atmosphere (TOA), as an adjustment of thestratosphere profiles can be neglected when lookingat the SW spectrum (Hansen et al., 1997). The ra-diative forcing due to the first aerosol indirect effectis estimated to be of the order of 0 to –2 Wm−2

(Ramaswamy et al., 2001), with a considerable un-certainty (Boucher and Haywood , 2001; Haywood andBoucher , 2000).

For the second aerosol indirect effect, such a ra-diative forcing cannot be defined, as there are time-dependent processes involved. A cloud cover increaseand subsequently an enhancement of cloud albedoas a reaction to an increase in aerosol concentrationtakes place only after some minutes to hours. Todeal with this, Rotstayn and Penner (2001) define a“quasi-forcing” due to the second aerosol indirect ef-fect. They carry out two model simulations which areidentical except for aerosol concentration consideredin the precipitation scheme of the model. In one ofthe simulations natural only aerosol concentrationsare applied and natural plus anthropogenic aerosolsin the second one. The difference in SW radiationflux at the top of the atmosphere (TOA) is takenas the quasi-forcing due to the second aerosol indi-rect effect. They find that with this definition, bothaerosol indirect effects are of comparable magnitude.However, as shown in Chapter 6, this definition of aquasi-forcing may be too ambiguous to satisfactorilydescribe a radiative forcing due to the second aerosolindirect effect, as other cloud feedback processes

may be active as well, e.g., concerning cirrus clouds.A slightly different approach has been proposed byKristjansson (2002), who calculate two cloud fieldsin a single model simulation, one of which includesthe second aerosol indirect effect of anthropogenicaerosols, and the other one does not, the former onebeing used for the model integration, and the latterone stored for diagnostics.

1.9. The representation of cloud

and precipitation

microphysics and aerosol

indirect effects in different

GCMs: A review

Based on the cloud parameterizations in GCMs de-scribed earlier (Section 1.4), and with the aim of in-corporating the aerosol indirect effects, several GCMsintroduced comprehensive microphysical schemes.For an estimation of the first AIE, the parameteri-zation of the link between aerosol concentration andCDNC is of main interest, as the choice of the aerosolspecies taken into account. The formulae which con-nect aerosol concentrations and CDNC can be di-vided into two categories, namely empirically basedand physically based approaches. The former onesdiagnose CDNC from aerosol mass or aerosol numberconcentration fitting a relationship to observations;the latter ones take into account model variables likethe turbulent kinetic energy to incorporate some in-formation about the updraft velocity in the clouds.Complexity in the aerosol concentrations consideredranges from prescribed fields of sulfate aerosols onlyto comprehensive cycles of different aerosol speciescalculated interactively in the GCM.

For the second aerosol indirect effect, the precipi-tation microphysics is of interest. GCMs that treatmicrophysics generally use bulk parameterizations.The most comprehensive schemes of precipitationmicrophysics include several different species of wa-ter like liquid and frozen cloud water, and rain, snow,hail or graupel as precipitation species. Multiple pro-cesses of conversion between the water species may

20

1. Introduction 1.9. Microphysics and aerosol effects in different GCMs: A review

be taken into account.

Lohmann and Roeckner (1996) introduce a micro-physical scheme in the GCM of the Max PlanckInstitute for Meteorology, the European Centre forMedium Range Weather Forecasting - HAMburgGCM (ECHAM). In their parameterization, mixingratios of water vapor, cloud liquid and ice water, andCDNC are prognostic variables. CDNC is linked tosulfate aerosol mass concentration following Boucherand Lohmann (1995). Autoconversion is parame-terized as in Beheng (1994). The scheme has beenextended to include a physically based activationscheme of droplets by Lohmann et al. (1999) andLohmann et al. (2000a), which depends on aerosolnumber concentration and a subgrid-scale updraftvelocity calculated from the turbulent kinetic energy.Sulfate and carbonaceous aerosols are considered.Using this model Feichter et al. (1997) evaluate theradiative forcing due to the first aerosol indirect ef-fect by sulfate aerosols of –0.76 Wm−2. Lohmann(2002b), Lohmann and Karcher (2002) and Karcherand Lohmann (2003) further introduce several pro-cesses of ice crystal nucleation in the ECHAM model.

Jones et al. (1994) introduce an empirical formulato diagnose CDNC from sulfate aerosol mass in theHadley Centre GCM. This formula is fitted to obser-vational data over various continental and maritimeregions from Martin et al. (1994). The CDR is cal-culated from the liquid water content in stratiformclouds while holding the CDR in convective cloudsfixed. Jones et al. (1994) calculate a radiative forc-ing due to the first AIE of –1.3 Wm−2. Jones andSlingo (1996) examine several other empirical formu-lae of the aerosol to CDNC link, and find radiativeforcings due to the first AIE in the range of –0.5 to–1.5 Wm−2. Wilson and Ballard (1999) introduce acloud microphysical scheme into the Hadley Centremodel, which is based on the Rutledge and Hobbs(1983) parameterization. They use four different wa-ter species, vapor, liquid water, rain water, and “ice”which includes cloud ice and snow. An originality oftheir scheme is that condensation and evaporationis calculated for liquid water clouds only using theSmith (1990) scheme which applies a triangular PDF.

Ice is formed by freezing of liquid cloud water, thus,no temperature-dependent parameterization of thepartitioning between liquid and ice water is neces-sary. In their model, ice crystals are allowed to havenon-spherical shape in the radiation parameteriza-tion following Kristjansson et al. (1999). Autocon-version of liquid droplets to rain is calculated usingthe Tripoli and Cotton (1980) formula. A schemeparameterizing subgrid-scale saturation water vapormixing ratio is introduced by Cusack et al. (1999),which they derive from simulations with their modelin high resolution weather forecasting mode. Joneset al. (2001) use this microphysical scheme to eval-uate the radiative impact of both aerosol indirecteffects by comparing two simulations and found animpact of –1.9 Wm−2. Williams et al. (2001) extendthis study and find the first indirect effect in theirmodel to be much larger than the second one. Theygive a global mean radiative impact of both AIE of–1.6 Wm−2.

Rotstayn (1997) introduce a microphysical schemein the Commonwealth Scientific and Industrial Re-search Organization (CSIRO) GCM. He use watervapor and cloud liquid and ice water as prognosticvariables, diagnosing rain and snow as precipita-tion species. Convective clouds are diagnosed ina separate convection scheme. Clouds are consid-ered to overlap randomly in their model. For iceclouds, rather than assuming spherical particles, theeffective radius is calculated using an empirical pa-rameterization of re = 0.051 q0.667

i . Condensationand evaporation is calculated using the scheme ofSmith (1990). The autoconversion scheme is similarto the Tripoli and Cotton (1980) scheme and usesa critical liquid water mixing ratio, which dependson CDNC. Using this model and the empirical for-mula of Boucher and Lohmann (1995) to link sulfateaerosol mass to CDNC, Rotstayn and Penner (2001)evaluate the sum of both aerosol indirect effects tobe –1.56 Wm−2.

Del Genio et al. (1996) introduce a microphysicalscheme in the Goddard Institute for Space Studies(GISS) GCM. They use water vapor, cloud water,and cloud ice as variables. Fractional cloudiness is

21

1.10. Outline of the thesis 1. Introduction

allowed to occur in the vertical as well as in thehorizontal dimensions by scaling vertically the cloudfraction, f , as f ′ = f

23 , and the cloud optical thick-

ness as τ ′c = f13 τc. In the radiation parameteriza-

tion, a grid box is considered either cloudy or clearcomparing the cloud fraction to a random numberbetween zero and one, which in the temporal meanapproaches the calculated cloud fraction. Similarly,clouds are considered either entirely liquid or frozen,comparing a random number to a calculated ice frac-tion which varies quadratically between zero at −4◦Cover ocean and −10◦C over land to unity at −40◦C.The Bergeron-Findeisen process is accounted for bycalculating the possibility of glaciation in each cloud.Menon et al. (2002) introduce an empirical formulawhich links mass concentrations of sulfate and or-ganic matter to CDNC into this model. Ice crystalnumber concentration is assumed to be constantly0.06 cm−3 in their model. The microphysical schemeis extended to include the Tripoli and Cotton (1980)autoconversion scheme. Menon et al. (2002) evaluatethe sum of both AIE to be –1.55 to –4.36 Wm−2,depending on the choice of parameters in their mi-crophysical scheme.

Ghan et al. (1997b) apply the cloud microphysicsparameterization of the Colorado State Univer-sity (CSU) Regional Atmospheric Modeling System(RAMS) (Flatau et al., 1989) in the National Cen-ter for Atmospheric Research (NCAR) CommunityClimate Model (CCM). In this scheme, water vapor,cloud liquid and ice water, rain and snow as wellas ice crystal number concentration are prognosticvariables. Autoconversion is parameterized followingZiegler (1985). Ghan et al. (1997a) further apply aphysically based link between aerosol number concen-tration, which depends on updraft velocity, aerosolnumber concentration, size distribution and chem-ical composition (Abdul-Razzak et al., 1998). Thisscheme is further extended for multiple aerosol typesby Abdul-Razzak and Ghan (1999). With this GCM,Ghan et al. (2001) quantify the aerosol indirect effectsto be of –1.5 to –3.0 Wm−2, depending on micro-physical parameters and model resolution. Anothermicrophysical scheme has been added to the NCARCCM by Rasch and Kristjansson (1998), based on

the scheme of Boucher et al. (1995a). Kristjansson(2002) use this scheme and introduce further a for-mula to predict CDNC from sulfate and black carbon.They find a radiative forcing of –1.8 Wm−2 for bothaerosol indirect effects, where the first AIE is threetimes larger than the second. They confirm that theaerosol indirect effects caused by black carbon arenegligible.

Takemura et al. (2000) describe a microphysicalscheme in the Center for Climate System Research(CCSR) GCM. In their model, CDNC is linked toaerosol number concentration using an empirical for-mula. Precipitation formation is calculated fromliquid water content using a characteristic time scaledependent on liquid water content and CDNC. Withthis model, the aerosol indirect effects are quantifiedto be –2.0 Wm−2.

Smith and Randall (1992) and Fowler et al. (1996) in-troduce a comprehensive microphysical scheme in theColorado State University (CSU) GCM. This schemeis adapted from the Rutledge and Hobbs (1983) pa-rameterization and includes mixing ratios of watervapor, liquid and ice cloud water, rain and snow asprognostic variables. Autoconversion is parameter-ized following Kessler (1969). This model has notyet been used to investigate aerosol indirect effects.

1.10. Outline of the thesis

The thesis is subdivided into three partitions:

1. A comparison and evaluation of the GCM in-cluding the microphysical scheme of Boucheret al. (1995a) and the empirical link betweensulfate aerosol mass and CDNC (Boucher andLohmann, 1995) using satellite data and in situmeasurements in order to study processes of theaerosol indirect effects,

2. An investigation of the climatic impacts ofaerosol and greenhouse gas effects in GCM sce-nario simulations of the 20th century, and

3. A study of ice phase parameterizations, particu-larly including the development and evaluation

22

1. Introduction 1.10. Outline of the thesis

Aerosols Link CDNC-aerosols Microphysics NudgingChapter 2 Online SO4, OM, BC, SS BL95, sulfate Boucher et al. noneChapter 3 Online SO4, OM, BC, SS BL95, sulfate Boucher et al. T , ~v

Chapter 4 As observed BL95 and modifications Boucher et al. T , ~v

Chapter 5 Online SO4, OM, BC, SS BL95, maximum hydrophilic aer. Boucher et al. T , ~v

Chapter 6 Monthly mean SO4 BL95, sulfate Boucher et al. noneChapter 7 Monthly mean SO4 BL95, sulfate Boucher et al. noneChapter 8 none none standard none

Table 1.1.: Model versions used throughout the thesis. BL95 refers to Boucher and Lohmann (1995). Intheir formula, generally, sulfate aerosol mass concentrations are used except for Chapter 5, wherewe use the maximum of all hydrophilic aerosol mass concentrations instead.

of a comprehensive liquid and ice phase micro-physical scheme.

The LMDZ GCM which is used in this study ispresented in Chapter 2. Of particular interest is thewarm cloud microphysical scheme of Boucher et al.(1995a), which has been implemented in the model.Table 1.1 summarizes differences in the model ver-sions used throughout the thesis. These include theaerosol distributions applied (e.g., whether sulfateonly or also other aerosols are used, calculated onlinein the model or precalculated monthly mean fields),the link between cloud droplet number concentrationand aerosol distributions, the application or not ofthe microphysical scheme, and whether model vari-ables are nudged to analysis data or not. Chapter3 describes the satellite instrument POLDER, thedata of which is widely used in the thesis. A di-rect comparison is done between model-calculatedand POLDER-derived cloud and aerosol quantities.In Chapter 4, a single-column version of the LMDZmodel is evaluated using data from the ACE-2 fieldcampaign, and comparing its results to those of othermodels. This chapter cites the work of Menon et al.(2003), and adds some sensitivity studies with theLMDZ GCM. Using POLDER satellite observationsand the LMDZ model, aerosol indirect effects arestudied by establishing statistical relationships be-tween aerosol index and cloud droplet radius andcloud liquid water path, respectively. This part,Chapter 5, is taken from Quaas et al. (2003a).

In the two following chapters, the relative impacts

of aerosols and greenhouse gases on the climate ofthe 20th century are studied analyzing three fixed-SST atmospheric GCM ensemble simulations for the1930 to 1989 period. In Chapter 6, we focus onthe radiative impacts of the two anthropogenic per-turbations, with a special interest in the response ofclouds. Particularly, the second aerosol indirect effectis examined. Chapter 7 looks at trends in observedvariables like surface temperature and diurnal tem-perature range and identifies different behavior of themodel when forced with different perturbations. Theresults in these two chapters are taken from Quaaset al. (2003b) and Quaas et al. (2003c).

In the third part, the parameterization in the LMDZGCM is evaluated with POLDER satellite data whichtreats the repartition of condensed water in liquid wa-ter and cloud ice (Chapter 8). The partitioning isparameterized as a function of the local tempera-ture. This study is submitted as Boucher and Quaas(2003). A new cloud and precipitation microphysicalscheme is developed for both, liquid and ice phases.The scheme is described and evaluated in Chapter9. Four one-dimensional cases are examined, twocases of marine stratocumulus clouds observed in theACE-2 campaign (see also Chapter 4), and two casesof thin cirrus clouds used in a model intercomparisonproject. A preliminary evaluation of the scheme inthe global model using satellite observations is pre-sented in Chapter 9 as well.

Chapter 10 summarizes the work and outlinespromising approaches for future work.

23

1.11. Resume 1. Introduction

1.11. Introduction : Resume

Le systeme climatique et le changement

climatique

Le changement climatique est considere comme l’undes problemes les plus urgents de l’humanite (NationsUnies, 1992 ; 2000). Gaz a effet de serre et aerosolsmodifient fortement le bilan radiatif du systeme cli-matique et provoquent ainsi des effets dangereux pourles equilibres sociaux, ecologiques et economiques.Cela est l’une des raisons pourquoi la comprehensiondu systeme climatique est d’une importance partic-uliere. Si de meilleures previsions des reponses dusysteme climatique aux perturbations anthropiquesetaient disponibles, des decisions politiques contre lesimpacts negatifs pourraient etre prises ou des adap-tations de la societe seraient possibles.

Les outils les plus importants pour la comprehensiondu systeme climatique global et pour la simula-tion du changement climatique sont les modeles dusysteme terrestre (ESM en Anglais). Ce sont desmodeles qui couplent plusieurs composantes, parmilesquels des modeles de l’atmosphere, de l’ocean,des surfaces continentales avec la vegetation, desglaces de mer, de la biosphere et de la cryosphere.Cette etude sera concentree sur la composante atmo-spherique representee par un modele de circulationgenerale de l’atmosphere. Resolvant numeriquementles equations qui gouvernent la dynamique del’atmosphere et incorporant des processus adia-batiques, ces modeles sont capables de simulerune multitude de processus physiques. Pourtant,du a la capacite limitee des ordinateurs et a lacomprehension limitee des processus, les simula-tions actuelles du systeme atmospherique restentirrealistes sous plusieurs aspects. La resolution et lacomplexite des modeles peuvent etre augmentees pardes meilleures methodes numeriques de modelisation(par example, Quaas , 2000) et des ordinateurs pluspuissants. L’etude presente est quant a elle dedieea l’amelioration de la comprehension des processusphysiques et en particulier celle des effets indirectsdes aerosols.

Les impacts anthropiques sur le climat

Deux modifications majeures de la composition del’atmosphere sont provoquees par l’humanite. Lapremiere et la plus importante est l’augmentationde la concentration tropospherique en gaz a effetde serre, qui absorbent le rayonnement terrestre etle re-emettent a une temperature plus basse. Celaprovoque l’effet de serre qui resulte en une augmen-tation de la temperature de la surface terrestre pourmaintenir l’equilibre radiatif de la planete. Parmi cesgaz, le dioxyde de carbone, produit par la combustiondes combustibles fossiles, est la composante la plusimportante. Les autres principaux gaz a effet de serreanthropiques sont le methane, le protoxyde d’azote,les chlorofluorocarbones et l’ozone tropospherique.La concentration de tous ces gaz a augmente depuisle debut de l’ere industrielle (Fig. 6.1 ; Myhre et al.,2001) et une augmentation de la temperature de lasurface a ete observee (Fig. 1.1 ; Folland et al., 2001).

Pourtant, cette augmentation n’a pas ete aussi fortequ’attendu a partir de simulations avec des modelesde circulation generale (MCG) incluant seulementl’effet de serre (par exemple , Mitchell et al., 2001).Ce fait pourrait etre du a une importante vari-abilite interne du systeme ou a l’existence d’autresforcages. Un forcage negatif du aux aerosols an-thropiques pourrait avoir contrebalance une partiedu rechauffement par effet de serre (par exemple,Wigley, 1989 ; Charlson et al., 1989 ; Boucher et An-derson, 1995). Depuis deux decennies, les differentseffets radiatifs des aerosols ont ete etudies. Pourtant,dans le rapport recent du Groupe Intergouvernemen-tal d’Experts sur le Changement Climatique (GIECou IPCC en anglais), le niveau de la comprehensionscientifique des effets des aerosols a toujours ete qual-ifie de � bas� voire de � tres bas� . Pourtant,il est largement accepte que les aerosols jouent unrole majeur sur le changement climatique. En cequi concerne les impacts des aerosols sur le rayon-nement, les effets � directs� , par modification del’albedo planetaire par diffusion et absorption, et� indirects� par modification des proprietes desnuages, peuvent etre distingues.

24

1. Introduction 1.11. Resume

Les nuages dans le systeme climatique

D’un point de vue meteorologique, l’eau est le traceurle plus important dans l’atmosphere, dans laquelleelle existe dans les trois phases thermodynamiques.La source principale de l’eau dans l’atmosphereest l’evaporation aux surfaces oceaniques et con-tinentales. La vapeur d’eau est ensuite distribueepartout dans l’atmosphere. Elle interagit avec le bi-lan energetique. Par evaporation et condensation,la chaleur latente est distribuee dans l’atmosphere.Comme un gaz absorbant dans le spectre infrarougedu rayonnement, la vapeur d’eau provoque un effetde serre important.

La vapeur d’eau se condense en eau liquide et solidepour former des nuages. Des nuages existent danstoutes les couches de la troposphere jusqu’a environ11 km de hauteur. A des temperatures superieures a0◦C, les nuages sont composes d’eau liquide, pour destemperatures inferieures a –35 - –40◦C, de glace, etentre ces deux temperatures, des nuages mixtes peu-vent exister (par exemple, Houze, 1993). Les nuagescouvrent le globe a hauteur de 60% environ a chaqueinstant. Par condensation et evaporation et donc enredistribuant de la chaleur latente, ils interagissentavec le bilan energetique. Dans le spectre solaire, ilsreflechissent le rayonnement et augmentent l’albedoplanetaire, car ils sont en general plus reflechissantsque la surface terrestre. Cela est un forcage negatif.Dans le spectre infrarouge, des nuages exercent uneffet de serre et donc un forcage positif. L’effet netdes nuages sur le rayonnement est negatif pour lesnuages bas, et il est positif pour les nuages hauts. Enmoyenne globale, leur effet net est negatif (Kiehl etTrenberth, 1997).

Les precipitations se forment dans les nuages pardes processus microphysiques. Les nuages sont donca l’origine de cette quantite meteorologique impor-tante. La precipitation est le puits final de la vapeurd’eau atmospherique et influence ainsi la duree de viedes nuages et leur contenu en eau.

Representation des nuages dans des

modeles de grande echelle

Contrairement aux mouvements atmospheriquesd’echelle synoptique, les processus de condensationet de la formation des nuages ne peuvent pas etredecrits explicitement dans les modeles de grandeechelle. Ils doivent donc etre representes a partir desvariables explicitement calculees dans le modele, unemethode qu’on appelle � parametrisation� . Dansles premiers MCGs, les nuages etaient simplementprescrits en utilisant des observations. Depuis, unemultitude de parametrisations a ete creee. Quelquesmodeles considerent chaque maille comme etant rem-plie entierement par un nuage ou pas du tout (parexemple, Fowler et al., 1996), alors que d’autres in-troduisent une fraction nuageuse entre 0 et 100%(par exemple, Slingo, 1987 ; Le Treut et Li, 1991).A l’exception de peu de modeles (Del Genio et al.,1996 ; Jakob et Klein, 1999), un nuage est considereremplir une couche entierement sur la verticale.

La fraction nuageuse peut etre calculee en utilisantune � valeur critique� de l’humidite relative (parexemple, Sundqvist, 1978 ; Lohmann et Roeckner,1996). D’autres suivent l’approche de Sommeria etDeardorff (1977) qui ont propose une distribution del’eau totale dans chaque maille qui suit une fonctionde probabilite de densite (PDF en anglais). Une tellePDF peut etre consideree comme etant par exem-ple uniforme (Le Treut et Li, 1991) ou triangulaire(Smith, 1990). Des approches recentes introduisentune PDF pronostique qui predit des proprietes dela fonction a partir des processus physiques simulescomme la turbulence, la microphysique des nuages,et la convection (Tompkins, 2002).

Des nuages sont consideres homogenes partout dansla fraction nuageuse de la maille, ce qu’on appellel’hypothese plan-parallele homogene (PPH). Cettehypothese peut introduire des erreurs systematiquesdans l’estimation de l’impact radiatif des nuages, enparticulier dans le spectre des ondes courtes (parexemple, Pincus et Klein, 2000 ; Cahalan, 1994 ;Stubenrauch et al., 1997). Des approches recentes ex-istent qui traitent cette problematique qui pourtantrestera en dehors de cette etude.

25

1.11. Resume 1. Introduction

Les nuages representes par des couches discretes serecouvrent sur la direction verticale. Pour calculer cerecouvrement, qui est d’importance pour l’estimationdes flux radiatifs et de la precipitation, on peut faireles hypotheses differentes tel qu’un recouvrementaleatoire, maximal ou un melange entre les deux.Recemment, Hogan et Illingworth (2000) ont proposeune nouvelle parametrisation qu’ils ont creee a partird’observations radar.

Des GCM distinguent souvent des nuages� convectifs� des nuages � stratiformes� . Lespremiers sont diagnostiques par le schema de con-vection, le dernier par des schemas de nuages et deprecipitation du type de ceux decrits ci-dessus. Onpeut estimer la couverture nuageuse et le contenuen eau des nuages convectifs en utilisant le taux deprecipitation predit par le schema de convection.

La microphysique des nuages

L’ensemble de tous les processus lies aux nuagesou a la precipitation et qui jouent a l’echelle desparticules individuelles est connu sous le terme de� microphysique des nuages� . Ces processus en-globent la formation des gouttelettes des nuages etdes cristaux de glace par activation d’aerosols hy-drophiles ou nucleation, leur croissance par conden-sation, les processus qui conduisent a la formationde la precipitation et la croissance des particulesde precipitation, et les transformations entre lesdifferentes especes d’eau. On peut distinguer lesprocessus de la phase liquide (parfois appelee phasechaude) et de phase glace (froide).

L’activation de nouvelles gouttelettes depend de laconcentration en aerosols hydrophiles, et de la sur-saturation qui, elle, est determinee par la vitesseverticale a l’echelle du nuage (par exemple, Twomey,1959). Les gouttelettes formees grandissent ensuitepar condensation de la vapeur d’eau sur les partic-ules. Si les gouttelettes ont des tailles differentes etdonc des vitesses de chute differentes, elles peuvententrer en collision et former des gouttes plus grandes.On parle d’autoconversion des gouttelettes de nuagespour former des gouttes de pluie. On peut considerer

qu’il y a precipitation lorsque la taille des gouttelettesexcede un � rayon critique� (Kessler, 1969). Apartir des observations, ce rayon a ete determine a10 µm (Pawlowska et Brenguier, 2003). Une fois queles gouttes de pluie existent, elles tombent a traversdes populations de gouttelettes qu’elles collectentpar le processus d’accretion. La taille typique d’unegouttelette de nuages est de l’ordre de 3 a 20 µm alorsque les gouttes de pluie ont une taille typique de 100µm a 1 mm. Il faut donc de 102 a 107 gouttelettesd’eau nuageuse pour former une goutte de pluie, sil’on considere ces particules spheriques, ce qui esteffectivement une tres bonne approximation.

Des cristaux de glace peuvent etre formes par lanucleation homogene, ou un aerosol humide croıtpar la deposition de la vapeur d’eau. La nucleationheterogene est la congelation d’une petite goutteletteinitialement d’eau liquide par interaction avec unaerosol qu’on appelle � noyau de congelation� ou� noyau glacogene� . Par un processus similaire al’autoconversion, les cristaux forment des flocons deneige quand ils atteignent une certaine taille. Les par-ticules de glace sont souvent consideres comme etantspheriques dans des modeles de grande echelle, bienque ce ne soit probablement pas toujours une bonneapproximation (par exemple, Doutriaux-Boucher etal., 2000).

Dans les nuages de phase mixte, des particules givreespeuvent etre formees quand des particules liquides etsolides entrent en collision. Dans ce cas, des cristaux� secondaires� sont formees par le processus deHallett-Mossop (Hallett et Mossop, 1974). Comme lerapport de melange de la vapeur d’eau de saturationest superieur par rapport a l’eau liquide que par rap-port a la glace, des cristaux de glace vont grandir audepens des gouttelettes, un processus que l’on appelleprocessus de Bergeron-Findeisen (Bergeron, 1933).

Dans des modeles, on peut traiter la microphysiqueen resolvant la taille des particules en � size bins�(par exemple, Young, 1974a), ou dans un schema� bulk� en distinguant differentes especes d’eautelles que l’eau liquide des nuages, la pluie, etc. (parexemple, Kessler, 1969). Les parametrisations con-

26

1. Introduction 1.11. Resume

siderees dans cette etude appartiennent a ce secondtype. De tels schemas ont souvent ete concus pourdes modeles capables de resoudre des processus dansles nuages (cloud resolving models en anglais), puisadaptes pour les modeles a grande echelle. Parmi lesmodeles pour lesquels ont ete developpes des schemasde microphysique se trouvent ECHAM (Lohmann etRoeckner, 1996), le MCG du Hadley Centre (Jones etal., 1994 ; Wilson et Ballard, 1999), celui du CSIRO(Rotstayn, 1997), du GISS (Del Genio et al., 1996),le CCM du NCAR (Ghan et al., 1997b ; Rasch etKristjansson, 1998), le MCG du CCSR (Takemuraet al., 2000), ou encore celui de CSU (Fowler et al.,1996).

Aerosols anthropiques et effets indirects

des aerosols

Dans le contexte de la modelisation a l’echelle globale,on distingue habituellement six especes d’aerosols,les sulfates, les aerosols carbones organiques, le car-bone suie, les sels de mer, et les poussieres. Leursproprietes les plus importantes sont leur origine (na-turelle ou anthropique), leur caractere hydrophile ouhydrophobe, leur distribution en taille et leur ca-pacite a absorber le rayonnement solaire.

Une propriete importante des aerosols est leur taillequi fait qu’ils sont deposes a la surface plus ou moinsfacilement. Des aerosols hydrophiles peuvent aussietre incorpores dans les gouttes de pluie et deposespar la precipitation. Les forcages radiatifs dus auxaerosols sont donc concentres spatialement a prox-imite des sources.

Les aerosols hydrophiles servent comme noyaux decondensation dans les nuages. Une augmentation

de la concentration en nombre de ces aerosols peutainsi provoquer une augmentation de la concentra-tion en nombre des gouttelettes. Pour un contenuen eau liquide fixe, le nuage contient ainsi plus degouttelettes qui sont plus petites, et l’albedo dunuage augmente (premier effet indirect ; Twomey,1974). Le processus d’autoconversion est moins ef-ficace pour des gouttelettes plus petites, comme ill’a ete explique ci-dessus. Donc, l’eau d’un nuagecontenant des gouttelettes plus petites est moins vitetransformee en pluie, ce qui augmente sa duree devie, et son contenu en eau liquide moyen (deuxiemeeffet indirect ; Albrecht, 1989). Historiquement, leseffets indirects des aerosols ont ete etudies pour lesnuages d’eau liquide, et c’est egalement le cas danscette etude. Pourtant, des etudes recentes posentla question d’un impact possible des aerosols sur lesnuages de glace par exemple du aux avions (Boucher,1999) ou aux poussieres desertiques (De Mott et al.,2003).

Les deux effets indirects des aerosols ont ete misen evidence dans de nombreuses etudes. Parmi lesprincipales campagnes de mesures, on peut citer leprojet ACE-2 avec les donnees duquel le chapitre 2travaille (Brenguier et al., 2000b). D’autres etudesimportantes sont basees sur la campagne INDOEX(Ramanathan et al., 2001a). Plusieurs etudes satel-litaires ont etudie le contraste terre-mer, entre leshemispheres du Nord et Sud (par exemple, Hanet al., 1994), entre les sillages de bateaux (� shiptracks� en anglais) et leur environnement (par ex-emple, Conover, 1966), ou entre conditions pollueeset non-polluees (Rosenfeld, 2000). D’autres etudesetablirent des relations statistiques entre aerosols etproprietes des nuages (Kaufman et Fraser, 1997 ;Nakajima et al., 2001 ; Breon et al., 2002).

27

1.11. Resume 1. Introduction

28

2. The LMDZ GCM

2.1. Introduction

The atmospheric general circulation model of theLaboratoire de Meteorologie Dynamique has been de-veloped since the late 1970s. The recent version,named LMDZ.3.3 (Li , 1999), is the result of a newconception primarily of the dynamics of the GCMsince the late 1980s. It constitutes the atmosphericcomponent of the Earth System Model of the Insti-tut Pierre Simon Laplace (IPSL; see Fig. 2.1). Othercomponents of this model include

• the oceanic GCM Ocean Parallelise (OPA;Guilyardi and Madec, 1997), which includes amodel for sea ice;

• a model simulating continental surfaces andbiospheric processes, ORCHIDEE (OrganizingCarbon and Hydrology In Dynamic Ecosys-tEms; Maynard and Polcher , 2002; de Rosnayet al., 2002; Verant et al., 2003);

• a comprehensive atmospheric chemistry modelINCA (Interaction with Chemistry andAerosols; Hauglustaine et al., 2003), which in-cludes 99 tracers and about 300 photochemicalreactions as well as aerosol cycles for sulfate,mineral dust, black and organic carbon, andsea-salt aerosols;

• an aerosol module treating the cycles of sulfate(Boucher et al., 2002), carbonaceous (Reddyand Boucher , 2003), dust and sea salt aerosols;

• a model for lakes (Krinner , 2003).

In this thesis, for practical reasons, the aerosol cy-cles of Boucher et al. (2002) and Reddy and Boucher(2003) for sulfate, black and organic carbon, sea salt,and dust, will be used instead of INCA.

Figure 2.1.: Illustration of the Earth System Modelof the IPSL.

The atmospheric model may be divided into a “dy-namics” part which simulates the large-scale dynami-cal processes in the atmosphere, and a “physics” part,which describes the adiabatic processes. To a goodapproximation, the processes treated in the physicspart at each point of the model grid can be consideredindependent from the surrounding cells, thus reduc-ing the problem to one (the vertical) dimension.

2.1.1. Dynamics

The so-called “primitive equations”, a set of non-linear partial differential equations, describe the dy-namics of the atmosphere. This set consists of theequations of motion which predict the horizontalwinds, the thermodynamic equation which treatsthe time evolution of the temperature, the hydro-static equation which links pressure to temperature,and the continuity equation which derives the verti-cal wind from the horizontal winds (e.g., Trenberth,1995). The dynamical model variables are the hor-izontal winds, surface pressure, temperature, and

29

2.2. Clouds and Precipitation 2. The LMDZ GCM

total water content.

The primitive equations are solved numerically ona regular latitude-longitude grid. This grid, however,can be zoomed to increase the resolution to up tofour times the standard resolution over any region ofparticular interest (Le Van, 1989). This techniqueis applied e.g. in Chapter 4. The variables of themodel are evaluated using an Arakawa C grid. In thevertical, the equations are discretized using a hybridsigma-pressure coordinate system. The timestep de-pends on the resolution and the choice of zooming ornot the grid. In the studies presented, it is chosen as180 s.

2.1.2. Physics

As mentioned earlier, the “physical” part of themodel is assumed to be treatable independently ineach grid column, using just the vertical dimension.The physics processes are simulated using a longertimestep compared to the dynamics part of 1800 s.

To a large extent, the physics of the Integrated Fore-casting System (IFS) of the European Centre forMedium-Range Weather Forecasting (ECMWF) hasbeen adapted in the LMDZ GCM (White, 2002).The planetary boundary layer turbulence is param-eterized using an eddy-diffusivity approach, wherethe mixing coefficient is parameterized in terms ofthe Richardson number. The surface processes areevaluated in a bucket model, using the energy bal-ance equation to diagnose the surface temperature.A review of this scheme is given in Cheinet (2002).A new approach to simulate convective boundarylayers has been developed using a single column ver-sion of the LMDZ GCM (Cheinet , 2003; Cheinet andTeixara, 2003)), but is not yet implemented in thethree-dimensional version.

For the deep convection parameterization, two ap-proaches exist in the recent version of the LMDZGCM, a scheme following Tiedtke (1989), and the pa-rameterization of Emanuel (1991). Both of them aremass-flux schemes. As so far, the Emanuel schemedoes not allow for the transport of tracers like aerosolsin the model, the Tiedtke convection scheme is used

in the present studies.

Radiative transfer is calculated in the model usinga two-stream approximation, dividing the radiationin an upwelling and a downwelling flux. For thesolar (SW) spectrum, a parameterization followingFouquart and Bonnel (1980) is used. This schemetakes into account two spectral bands of 0.25 to 0.68µm and 0.68 to 4.0 µm. Cloud optical properties ateach layer accounted for assuming a vertically ran-dom overlap of the fractional clouds. Cloud opticalproperties considered in the SW radiation scheme arecloud optical thickness, τc, single scattering albedo,ω0, and cloud particle asymmetry parameter, g. Thecode also allows for scattering and absorption ef-fects of aerosols in the SW radiation (aerosol directeffect). For the terrestrial (LW) spectrum, a codebased on Morcrette (1991) is used in the GCM. Thisscheme considers six bands and calculates absorp-tion and emission by molecules and clouds. Gaseswhose absorption is taken into account are watervapor, carbon dioxide, methane, nitrous oxide, andchlorofluorocarbons. The cloud radiative propertyof interest in the LW spectrum is its emissivity, ε.As radiation is computationally very expensive, it isapplied only each 12th timestep of the physics, i.e.,with a timestep of 6 hours.

2.2. Clouds and Precipitation

2.2.1. Standard model treatment

The GCM uses the total water content, qt, as prog-nostic variable. In the model, all condensed wateris re-evaporated before applying the physics parame-terization. Following Le Treut and Li (1991), a “top-hat” probability density function (PDF) of the totalwater is assumed in each grid box (see Fig. 9.2), giv-ing a uniform distribution between (qt (1−∆qs

)) and(qt (1 + ∆qs

)), where qt is the grid box mean totalwater mixing ratio [kg kg−1], and ∆qs

is a param-eter, which is considered to take vertically varyingbut otherwise constant values. Using this PDF, thehorizontal fractional cloudiness in the grid box, f , is

30

2. The LMDZ GCM 2.2. Clouds and Precipitation

calculated as

f =

0 qs > qt (1 + ∆qs

qt (1+∆qs )−qs

2 ∆qs qtqs ∈ [qt (1−∆qs

), qt (1 + ∆qs)]

1 qs < qt (1−∆qs

(2.1)where qs is the saturation water vapor mixing ratiocalculated in dependence of the temperature usingthe Tetens formula (Tetens, 1930). The total watercontent in the cloud covered fraction of the grid cellis determined as

qcct =

qs + qt + ∆q

2(2.2)

The total water content in the cloud-free part of thegrid cell, qcf

t , is calculated as the residual. The con-densed water in the grid box is given by

qc = f (qcct − qs)

(1 +

L

cp

∂qs

∂T

)−1

(2.3)

where the last term on the right-hand side takesinto account the heating by latent heat release inthe condensation process. The derivative ∂qs/∂T isobtained from the Clausius-Clapeyron equation, andL is the latent heat taken as Lv, the latent heat ofevaporation, for T > 0◦C, and as Ls, the latent heatof sublimation, for T < 0◦C.

When deep convection in a grid cell is simulatedby the convection scheme, additional cloud watermay be contributed by convective clouds. The cloudwater from convective clouds is diagnosed using baseand top of the convective cloud and convective rainrate calculated by the convection scheme. If thecloud water and cloud fraction calculated by this di-agnostics exceed those of the stratiform scheme thenthey are used instead.

The condensed water is divided into liquid waterand ice, depending on the local temperature, wherethe liquid fraction is given by

xliq =

0 T < Tice(

T−TiceT0−Tice

)nxT ∈ [Tice, T0]

1 T > T0

(2.4)

The parameters are taken as T0 = 0◦C, Tice =−15◦C, and nx = 6. In Chapter 8, this parame-terization is evaluated using POLDER satellite data,

and more realistic values of Tice and nx are derived.

Temperature is adjusted taking into account thelatent heat of condensation or sublimation, respec-tively. The combination of the re-evaporation andthe condensation scheme given above thus accountsfor both, condensation of cloud water from watervapor and evaporation of cloud water.

Stratiform precipitation rate for liquid water cloudsis calculated as

Pr =ql

τp

[1− exp

(−(

ql

Cl

)2)]

(2.5)

where ql is the grid box mean liquid water mix-ing ratio given by ql = xl qc, with xl and qc

from the condensation scheme, τp a characteris-tic timescale for precipitation taken as 1800 s, andCl = 2× 10−4 kg kg−1 a precipitation threshold.

Snow formation in ice clouds is applied as a sedi-mentation term

Ps =qi

∆zVi (2.6)

where qi is the cloud ice mixing ratio (qi = (1−xl) qc),∆z the layer thickness, and Vi = 3.29 (ρair qi)0.16

the terminal fall-velocity of ice crystals (Heymsfieldand Donner , 1990).

The evaporation of rain water is derived from a pa-rameterization of Kessler (1969):

Er = α

(1− qt

qs

)√Pr ∆z (2.7)

with a coefficient α = 2. × 10−5ms−1, and the pre-cipitation flux, Pr and the layer thickness, ∆z, takenwithout dimension. Both rain and snow are leavingthe grid column in a single timestep. Whether itis rain or snow that reaches the surface is decidedby the temperature of the lowest model layer beinglower or larger than 0◦C.

Concerning the cloud optical properties, the asym-metry parameter, g, is taken to be constant in themodel, using g = 0.865 in the first and g = 0.910 inthe second of the two spectral bands in the schemeof Fouquart and Bonnel (1980). Cloud optical thick-ness (COT) is defined as the vertical integral of the

31

2.2. Clouds and Precipitation 2. The LMDZ GCM

extinction, σ, of a cloud, which for spherical particlescan be described as:

τc =∫ ztop

zbase

σdz = 2π

∫ ztop

zbase

∫ ∞

0

r2 n(r)drdz (2.8)

= 2π

∫ ztop

zbase

(∫ ∞

0

r3 n(r)dr

)(∫r2 n(r)dr∫r3n(r)dr

)dz

where n(r) is the droplet size distribution, and zbase

and ztop are cloud base and cloud top height, respec-tively. One can define an “effective radius”, re, of thedroplets as the ratio

re =∫

r3 n(r)dr∫r2n(r)dr

(2.9)

With the assumption of a constant re throughout thecloud and the definition of the liquid water path,W , the optical thickness may be parameterized as(Stephens, 1978)

τc =32

W

ρwater re(2.10)

In the model, ice crystals are assumed to be of spher-ical shape with a constant effective radius of 35 µm,while the effective radius of liquid droplets is assumedto be 9 µm in the three lowest layers of the modeland 13 µm otherwise.

The single scattering albedo of a cloud, ω, is pa-rameterized in terms of COT:

ω0 = a1 − a2 × 10a3 exp(−a4 τc) (2.11)

where the coefficients a1-a4 are slightly different inthe two SW spectral bands. Finally, the emissivity ofthe clouds, ε, necessary in the LW radiative transferis parameterized in terms of the liquid water path

ε = 1− exp (−α W ) (2.12)

where the coefficient α is 0.13 for liquid and 0.09 forice clouds.

2.2.2. The microphysical scheme of

Boucher et al.

A refinement for studying aerosol effects compared tothe standard scheme has been introduced by Boucherand Lohmann (1995), who diagnose the cloud dropletnumber concentration (CDNC, Nd) of liquid droplets

from sulfate aerosol mass concentration, ma, given in[µg sulfate m−3] (formula “D”):

Nd = 10a0+a1 log (ma) (2.13)

where the empirical constants are a0 = 2.21 anda1 = 0.41, which were obtained from a fit to simulta-neously measured sulfate aerosol masses and CDNCat various locations on the globe. Sulfate is used as asurrogate for all hydrophilic aerosols, bearing in mindthat at least organic and sea salt aerosols constituteefficient CCN as well (Novakov and Penner , 1993).CDNC is given in m−3. Assuming spherical particles,the volume-mean droplet radius, rdv is calculated as

rdv = 3

√ql ρair

4π3 ρwater Nd

(2.14)

The effective radius of a droplet is linked to thevolume-mean droplet as re = 1.1 rdv (Pontikis andHicks, 1993).

The microphysics scheme of Boucher et al. (1995a)keeps the condensation/evaporation scheme and thusthe determination of fractional cloudiness as wellas the microphysical processes related to ice clouds,i.e., the snow formation. It introduces, however, mi-crophysical processes of precipitation formation forliquid water clouds using autoconversion of clouddroplets to form rain and accretion of raindrops bycollection of cloud droplets. The autoconversion rate,rll, is parameterized as

rll = c1 H(re − r0) ql r4e Nd (2.15)

where H(x−x0) is the Heaviside function. The “crit-ical radius” r0 is set to 8 µm in the model, a valueslightly smaller than that observed from Pawlowskaand Brenguier (2003). The parameter c1 is the prod-uct of several constants and set to c1 = 0.1 in themodel. From Eqs. 2.14 and 2.15 it can be de-duced that the autoconversion rate depends linearlyon CDNC for re > r0. The process of collision andcoalescence of raindrops and cloud droplets is knownas accretion, rrl. This process is parameterized as

rrl = c2 qccl qr

√ρair

Dr(2.16)

where qr is the rain water mixing ratio, and Dr

the characteristic raindrop diameter of a Marshall-Palmer size distribution (Marshall and Palmer ,

32

2. The LMDZ GCM 2.3. Comparison standard vs. Boucher scheme

1948). The coefficient c2 is the product of severalconstants, and is set to c2 = 0.7 in the model. Evap-oration of rain water is calculated following Liu andOrville (1969) as

Er = −α

2

(qcft

qs− 1)

qr

ρwater D2r

(2.17)

where ρwater the density of liquid water. The coeffi-cient α = 0.42× 10−5km m−1s−1 is given by Liu andOrville (1969).

2.3. Comparison standard vs.

Boucher scheme

The new scheme is compared with the standardmodel scheme and evaluated against observations inFigs. 2.2 to 2.5. For the total cloud cover, data fromthe International Satellite Cloud Climatology Project(ISCCP; Rossow and Schiffer , 1991) is averaged over1984 to 1992. Cloud radiative forcings are taken fromthe Scanner of the Earth Radiation Budget (ScaRaB;Kandel et al., 1998), for the March 1994 to February1995 period, with the data for October 1994 missing.Precipitation over continents is used from the GlobalPrecipitation Climatology Project (GPCP; Huffmanet al., 1997), for 1997. The model output is from aone year simulation forced with SST for 1997 fromReynolds and Smith (1995). We show zonal meanvalues for the annual global mean (in case of pre-cipitation, over continents only), for northern hemi-sphere summer (June-July-August, JJA), for north-ern hemisphere winter (December-January-February,DJF), the annual mean over continents, and overoceans. Generally, the scheme of Boucher et al. sim-ulates cloud and precipitation distributions which arerather similar to those simulated by the standardmodel scheme, and compares well with the observa-tions. The scheme of Boucher et al. underestimatescloud cover in the annual mean by up to 5% in someregions of the northern hemisphere and up to 10%in some regions in the southern hemisphere, whichis due to a too small cloud amount over land in thenorthern hemisphere and over oceans in the south-ern hemisphere. However, except maybe for a smallregion over land (around 20◦N), it does generally abetter job than the model’s standard scheme.

EQ30S60SSP 60N NP30NLatitude

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

Clo

udin

ess

[%]

ISCCPStandardBoucher

(a) annual mean(a) annual mean(a) annual mean

EQ30S60SSP 60N NP30NLatitude

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

Clo

udin

ess

[%]

ISCCPStandardBoucher

(b) JJA(b) JJA(b) JJA

EQ30S60SSP 60N NP30NLatitude

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

Clo

udin

ess

[%]

ISCCPStandardBoucher

(c) DJF(c) DJF(c) DJF

EQ30S60SSP 60N NP30NLatitude

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

Clo

udin

ess

[%]

ISCCPStandardBoucher

(d) Land(d) Land(d) Land

EQ30S60SSP 60N NP30NLatitude

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

Clo

udin

ess

[%]

ISCCPStandardBoucher

(e) Sea(e) Sea(e) Sea

Figure 2.2.: Zonal mean cloudiness, from ISCCP ob-servations (solid, black), for the stan-dard scheme of the model (dot-dashed,blue), and for the scheme of Boucheret al. (dashed, red). For (a) annualglobal mean, (b) JJA global mean, (c)DJF global mean, (d) annual mean overcontinents, (e) over oceans.

33

2.3. Comparison standard vs. Boucher scheme 2. The LMDZ GCM

EQ30S60SSP 60N NP30NLatitude

-140

-120

-100

-80

-60

-40

-20

0

SW

CR

F [W

m-2

]

ScaRaBStandardBoucher

(a) annual mean(a) annual mean(a) annual mean

EQ30S60SSP 60N NP30NLatitude

-140

-120

-100

-80

-60

-40

-20

0

SW

CR

F [W

m-2

]

ScaRaBStandardBoucher

(b) JJA(b) JJA(b) JJA

EQ30S60SSP 60N NP30NLatitude

-140

-120

-100

-80

-60

-40

-20

0

SW

CR

F [W

m-2

]

ScaRaBStandardBoucher

(c) DJF(c) DJF(c) DJF

EQ30S60SSP 60N NP30NLatitude

-140

-120

-100

-80

-60

-40

-20

0

SW

CR

F [W

m-2

]

ScaRaBStandardBoucher

(d) Land(d) Land(d) Land

EQ30S60SSP 60N NP30NLatitude

-140

-120

-100

-80

-60

-40

-20

0

SW

CR

F [W

m-2

]

ScaRaBStandardBoucher

(e) Sea(e) Sea(e) Sea

Figure 2.3.: Zonal mean SW cloud radiative forc-ing at the top of the atmosphere, fromScaRaB observations (solid, black), forthe standard scheme of the model (dot-dashed, blue), and for the scheme ofBoucher et al. (dashed, red). For (a) an-nual global mean, (b) JJA global mean,(c) DJF global mean, (d) annual meanover continents, (e) over oceans.

EQ30S60SSP 60N NP30NLatitude

-60

-50

-40

-30

-20

-10

0

LW C

RF

[W m

-2]

ScaRaBStandardBoucher

(a) annual mean(a) annual mean(a) annual mean

EQ30S60SSP 60N NP30NLatitude

-60

-50

-40

-30

-20

-10

0

LW C

RF

[W m

-2]

ScaRaBStandardBoucher

(b) JJA(b) JJA(b) JJA

EQ30S60SSP 60N NP30NLatitude

-60

-50

-40

-30

-20

-10

0

LW C

RF

[W m

-2]

ScaRaBStandardBoucher

(c) DJF(c) DJF(c) DJF

EQ30S60SSP 60N NP30NLatitude

-60

-50

-40

-30

-20

-10

0

LW C

RF

[W m

-2]

ScaRaBStandardBoucher

(d) Land(d) Land(d) Land

EQ30S60SSP 60N NP30NLatitude

-60

-50

-40

-30

-20

-10

0

LW C

RF

[W m

-2]

ScaRaBStandardBoucher

(e) Sea(e) Sea(e) Sea

Figure 2.4.: As Fig. 2.3, but for LW cloud radiativeforcing.

34

2. The LMDZ GCM 2.4. Conclusion: The aerosol indirect effects

EQ30S60SSP 60N NP30NLatitude

0

20

40

60

80

100

120

140

160

180

200

Pre

cipi

tatio

n [m

m m

onth

-1]

GPCPStandardBoucher

(a) annual mean(a) annual mean(a) annual mean

EQ30S60SSP 60N NP30NLatitude

0

20

40

60

80

100

120

140

160

180

200

Pre

cipi

tatio

n [m

m m

onth

-1]

GPCPStandardBoucher

(b) JJA(b) JJA(b) JJA

EQ30S60SSP 60N NP30NLatitude

0

20

40

60

80

100

120

140

160

180

200

Pre

cipi

tatio

n [m

m m

onth

-1]

GPCPStandardBoucher

(c) DJF(c) DJF(c) DJF

Figure 2.5.: Zonal mean precipitation [mm month−1],from GPCP observations (solid, black),for the standard scheme of the model(dot-dashed, blue), and for the scheme ofBoucher et al. (dashed, red). For (a) an-nual global mean, (b) JJA global mean,(c) DJF global mean.

Generally, the SW cloud radiative forcing simulatedby the scheme of Boucher et al. is too strong, by sev-eral percent. This overestimation is better than forthe standard scheme for JJA, but slightly worse forDJF and over also oceans for the annual mean. How-ever, the LW CRF is much better using the schemeof Boucher et al. compared to the standard schemeeven though both model versions show too strong LWcloud radiative forcing compared to the ScaRaB ob-servations. Precipitation generally is simulated wellby both schemes, with a slight underestimation of therainfall at low latitudes by the scheme of Boucheret al..

Figure 2.6.: Aerosol indirect effects by sulfateaerosols in the annual mean in theLMDZ using the scheme of Boucheret al. [Wm−2]. (a) the first aerosolindirect radiative forcing and (b) thesecond aerosol indirect effect diagnosedas the “quasi-forcing”.

For high and midlatitudes, however, the new micro-physics scheme is closer to the observations, which isremarkable in particular for the southern hemisphere.It should be noted that some of the differences be-tween the schemes and observations may be due tointernal variability as a particular year has been sim-ulated.

2.4. Conclusion: The aerosol

indirect effects

Using the microphysical scheme of Boucher et al. andthe empirical link between sulfate aerosol mass, theaerosol indirect effects have been evaluated in themodel. Figure 2.6 shows the distributions of the

35

2.4. Conclusion: The aerosol indirect effects 2. The LMDZ GCM

EQ30S60SSP 60N NP30NLatitude

-7

-6

-5

-4

-3

-2

-1

0

1

Firs

t AIE

[Wm

-2]

MeanContinentsOceans

First aerosol indirect effectFirst aerosol indirect effectFirst aerosol indirect effect

EQ30S60SSP 60N NP30NLatitude

-7

-6

-5

-4

-3

-2

-1

0

1

Sec

ond

AIE

[Wm

-2]

MeanContinentsOceans

Second aerosol indirect effectSecond aerosol indirect effectSecond aerosol indirect effect

Figure 2.7.: Zonal mean aerosol indirect radiative(quasi-)forcings [Wm−2], over land (red,dashed), over sea (blue, dash-dotted),and for the whole area (black, solid), forthe annual mean (a) first aerosol indirecteffect and (b) second aerosol indirect ef-fect.

annual mean of both aerosol indirect effects, wherethe second indirect effect is estimated using the“quasi-forcing” definition of Rotstayn and Penner(2001). The first aerosol indirect effect is strongestin the northern hemisphere midlatitudes, where themain source regions for anthropogenic aerosols arelocated. In particular, marine clouds near the indus-trialized zones of North America, Western Europe,and East Asia are affected, showing an aerosol indi-rect effect of up to –12 Wm−2. Over land, there is abelt of strong aerosol indirect forcing over the mid-latitudes of the Eurasian continent. For the secondindirect effect, much more scatter is observed. Dueto the “quasi-forcing” definition, also cloud feedbackprocesses other than the second aerosol indirect ef-fect are included (see Chapter 6 for a discussion ofthis difficulty). Nevertheless, it can be identified thatthere are strong effects of aerosols particularly in thenorthern hemisphere, with quasi-forcing values of upto –25 Wm−2 over certain regions even in the annualmean.

EQ30S60SSP 60N NP30NLatitude

-7

-6

-5

-4

-3

-2

-1

0

1

Firs

t AIE

[Wm

-2]

MeanDJFJJA

First aerosol indirect effectFirst aerosol indirect effectFirst aerosol indirect effect

EQ30S60SSP 60N NP30NLatitude

-7

-6

-5

-4

-3

-2

-1

0

1

Sec

ond

AIE

[Wm

-2]

MeanDJFJJA

Second aerosol indirect effectSecond aerosol indirect effectSecond aerosol indirect effect

Figure 2.8.: Zonal mean aerosol indirect effects[Wm−2], for June-July-August (red,dashed), for December-January-February (blue, dash-dotted), andfor the annual mean (black, solid), forthe annual mean (a) first aerosol indirecteffect and (b) second aerosol indirecteffect.

Figure 2.7 shows the zonal annual mean of the aerosolindirect effects over land, oceans, and for the wholearea. The first AIE clearly is stronger over oceans, asexpected, as low-level marine stratocumulus cloudsare the most affected. For the second AIE, a similarresult can be found, although it is less pronounced.The second AIE is found to be of equal magnitudeas the first AIE. Figure 2.8 shows the zonal mean ofthe aerosol indirect effects for northern hemispheresummertime (June-July-August, JJA) and winter-time (December-January-February, DJF), and for theannual mean. It is clearly found that both AIEs arestronger in summertime, as they are concentratedin the northern hemisphere and are active in thesolar spectrum. For the global annual mean, thefirst aerosol indirect radiative forcing in our modelis for sulfate aerosol concentrations at the end of the20th century (≈ 1990) of –1.1 Wm−2, and the quasi-forcing due to the second aerosol indirect effect of–1.2 Wm−2.

36

2. The LMDZ GCM 2.5. Resume

2.5. Le modele LMDZ : Resume

Introduction

Le modele de circulation generale atmospheriquedu Laboratoire de Meteorologie Dynamique estdeveloppe depuis les annees 1970. Il represente lapartie atmospherique du modele du systeme terrestrede l’Institut Pierre Simon Laplace (IPSL, voir Fig.2.1). D’autres parties de ce modele couple sont lemodele de l’ocean OPA (Guilyardi et Madec, 1997),un modele de glace, le modele de la surface conti-nentale et de la vegetation ORCHIDEE (Maynardet Polcher, 2002 ; de Rosnay et al., 2002 ; Verantet al., 2003), le modele de la chimie atmospheriqueINCA (Hauglustaine et al., 2003) et un modele delacs (Krinner, 2003). Les cycles biogeochimiquesdes aerosols soufres (Boucher et al., 2002), carbones(Reddy and Boucher, 2003), des sels de mer et despoussieres ont egalement ete developpes pour LMDZ.Pour des raisons pratiques, c’est ce dernier moduleet non INCA qui sera utilise dans cette these.

On peut diviser le modele atmospherique enune partie � dynamique� qui traite les proces-sus dynamiques a grande echelle, et une partie� physique� qui decrit les processus adiabatiques.La derniere partie peut etre traitee de maniere uni-dimensionnelle en faisant l’hypothese que chaquecolonne de la maille horizontale est independante desautres.

Dans la partie dynamique, le jeu d’equations� primitives� est resolu de maniere numerique. Ilcontient les equations de mouvement pour les ventshorizontaux, l’equation thermodynamique pour latemperature, l’equation hydrostatique qui lie la pres-sion et la temperature, et l’equation de continuitepour calculer le vent vertical a partir des vents hor-izontaux (voir par exemple, Trenberth, 1995). Lagrille horizontale du modele est une grille latitude- longitude qui peut etre zoomee sur une regiond’interet particulier ou la resolution peut etre aug-mentee jusqu’a un facteur 4. Dans la direction verti-cale, une coordonnee hybride de pression et niveauxsigma est utilisee.

La partie physique a ete en partie adaptee du modeleIFS de l’ECMWF. La turbulence de la couche limiteplanetaire est parametree en utilisant l’approche ditede � eddy-diffiusivity� et le coefficient de melangeest calcule en utilisant le nombre de Richardson. Ala surface continentale, la temperature de la sur-face est diagnostiquee a partir de l’equation du bilanenergetique. Pour le traitement de la convection pro-fonde dans le modele, deux schemas de flux de massepeuvent etre appliques, celui de Tiedtke (1989) etcelui d’Emmanuel (1991). Le premier est utilise danstoutes les etudes de cette these.

Le transfert du rayonnement est parametre avecl’approximation de deux courants ascendants et de-scendants, et deux bandes spectrales dans la partiesolaire (ondes courtes, shortwave ou SW en anglais)et quatre dans la partie terrestre (ondes longues,longwave ou LW). Dans le spectre des ondes courtes,LMDZ utilise le schema de Fouquart et Bonnel (1980)qui introduit deux bandes spectrales. L’epaisseuroptique, τc, l’albedo de diffusion simple, ω0, et leparametre d’asymetrie, g, sont utilises pour calculerle flux du rayonnement dans le ciel nuageux. Surla verticale, la couverture nuageuse fractionnairedes differentes couches est supposee se recouvrir demaniere aleatoire. Dans les ondes longues, le flux durayonnement est calcule suivant la parametrisationde Morcrette (1991), avec six bandes. Pour les nu-ages, l’emissivite, ε, est utilisee pour tenir comptede leur absorption et emission de rayonnement ondeslongues dans l’atmosphere.

Nuages et precipitation

Le modele utilise le rapport de melange de l’eau totalecomme variable pronostique. Toute l’eau condenseeest re-evaporee avant que le schema de nuages et deprecipitation ne soit applique. Suivant Le Treut et Li(1991), une fonction de densite de probabilite (PDFen Anglais) uniforme est consideree pour l’eau danschaque maille. La partie de cette PDF qui depassele rapport de melange de saturation donne la frac-tion nuageuse de la maille, et la partie d’eau qui setrouve dans cette partie et qui excede la saturationest condensee. Au cas ou le modele simule de laconvection profonde, une fraction nuageuse et une

37

2.5. Resume 2. The LMDZ GCM

quantite d’eau condensee sont calculees a partir de laprecipitation predite par le schema de convection. Sices valeurs depassent celles qui sont calculees par leschema � stratiforme� decrit auparavant, celles-cisont utilisees au lieu des dernieres. L’eau condenseeest partagee en eau liquide et en glace a partir dela temperature locale. Si celle-ci depasse 0◦C, toutel’eau est liquide, si elle est inferieure a –15◦C, toutel’eau est glace, et entre les deux, une transition estconsideree.

Dans la version standard du modele, la precipitationest calculee de maniere simple et depend du contenud’eau condensee dans la maille. Quand elle tombea travers des couches claires, la precipitation peuts’evaporer. Concernant les proprietes optiques desnuages, l’epaisseur optique est liee au rapport en-tre la colonne d’eau condensee et le rayon effectifdes particules dans la maille en considerant des par-ticules spheriques de taille constante. L’albedo dediffusion simple est une fonction de l’epaisseur op-tique et le parametre d’asymetrie est pris constant.Enfin, l’emissivite est calculee a partir de la colonned’eau condensee.

Le schema microphysique de Boucher et

al.

Le changement principal entre le schema micro-physique des nuages presente ici et le schema stan-dard est le fait que la concentration en nombre desgouttelettes (CDNC en anglais) est calculee et nonplus implicitement diagnostiquee. Elle est liee ala concentration en masse des aerosols de sulfatessuivant l’approche � empirique� de Boucher etLohmann (1995) qui ont examine des mesures simul-tanees de CDNC et de masse de sulfates. La tailledes particules, considerees comme etant spheriques,peut etre calculee a partir du nombre de goutteletteset de l’eau condensee. Cela ouvre la possibilitede calculer le premier effet indirect des aerosols.

Boucher et al. (1995) ont egalement introduit desformules pour parametriser la precipitation liquideen decrivant l’autoconversion des gouttelettes pourformer des gouttes de pluie, la collision et la coa-lescence entre gouttelettes des nuages et gouttes depluie qui aussi transforment gouttelettes en pluie. Laparametrisation des nuages de glace et de la neigen’a pourtant pas ete changee.

Dans le chapitre 2 deux simulations avec LMDZ ontete comparees aux observations, dont une utilisait laversion standard et l’autre incluait le schema micro-physique. Il a ete montre que la couverture nuageusene change pas beaucoup entre les deux simulationset qu’elle est proche des observations (fig. 2.2). Leforcage radiatif des nuages dans les ondes courteset dans les ondes longues est bien simule dans lesdeux versions du modele, et la version qui inclut lamicrophysique est meme un peu meilleure dans lesondes longues (figs. 2.3 et 2.4). La precipitation surles continents est bien simulee dans les deux versionsdu modele (fig. 2.5).

Les deux effets indirects ont ete estimes en utilisantle schema de microphysique (fig. 2.6). Le premiereffet indirect est concentre sur les moyennes latitudesde l’hemisphere nord ou se trouvent les principalessources des aerosols. Il atteint –12 Wm−2 dans lesregions des stratocumulus maritimes au large descotes Est des continents de l’hemisphere nord. Leforcage est particulierement fort dans les moyenneslatitudes du continent eurasiatique. Le quasi-forcagepar le deuxieme effet indirect, ainsi qu’il a ete definipar Rotstayn et Penner (2001), est beaucoup plusbruite. Il atteint jusqu’a –25 Wm−2 dans certainsregions. Les deux effets indirects sont plus marquessur ocean que sur continent (fig. 2.7), et ils sont plusforts en ete de l’hemisphere nord (juin-juillet-aot,JJA) qu’en hiver (decembre-janvier-fevrier, DJF ;fig. 2.8). En moyenne globale et annuelle, on trouveun forcage de –1.1 Wm−2 pour le premier effet indi-rect et de –1.2 Wm−2 pour le second effet indirectpour des concentrations en aerosols de l’annee 1990.

38

3. POLDER satellite observations

3.1. Introduction

Satellite instruments generally measure radiances atthe top of the atmosphere. As the radiances dependon the state of the Earth surface and the atmo-sphere, information can be deduced about a varietyof atmospheric and surface quantities. The radiancesmeasured by satellite instruments in the shortwavespectrum consists of the part of the solar radiationreflected by the Earth’s surface or by different com-ponents of the atmosphere. The longwave (LW) orterrestrial part of the spectrum refers to wavelengthsof 4 to 25 µm, which corresponds to the radiationemitted by the Earth’s surface or by molecules inthe atmosphere. Common instruments measure theintensity of the radiation in different bands of wave-lengths, also known as “channels” of the instrument.From the knowledge about scattering and absorp-tion properties of different constituents of the atmo-sphere, some information about the distribution ofthese components can be deduced from the measuredsignal.

Satellites may be divided into two categories. Geo-stationary satellites are positioned above the Equatorat the specific height at which their rotation speed isidentical to that of the Earth. Thus, the same regionof the Earth is always observed. The advantage isthat in principle a continuous measurement is possi-ble. A disadvantage is that only a part of the globe isobserved and that in particular for high latitudes themeasurement is not very reliable. Further on, theirrelatively high altitude makes the measured signalweaker. The second class of satellites are polar-orbiting platforms. Such satellites rotate at a fixedpath, letting the Earth turn below them. Typically,each point of the Earths surface is passed about once

a day for a swath of about 2500 km, where the highlatitudes are measured several times a day and thelow latitudes more sparsely. Instruments may beable to measure also along or across the track of thesatellite.

3.2. The POLDER instrument

The POLarization and Directionality of the Earth’sReflectance (POLDER) instrument was developedby the Laboratoire d’Optique Atmospherique (LOA)and the French Centre National d’Etudes Spatiales(CNES). It is the first satellite instrument to beable to measure spectral, directional and polarizedcharacteristics of the reflected solar radiation simul-taneously (Deschamps et al., 1994). In particularthe polarization of the radiation measured by thesatellite instrument provides a new possibility to de-duce properties of the constituents of the atmosphere.

The instrument uses 15 spectral bands between 0.44and 0.91 µm. It measures in a range of ± 43◦ alongthe track of the satellite and ± 51◦ across the track,yielding a swath of 2400 km in width. The resolutionat nadir is of 6 x 7 km2. A “superpixel” in the dataobtained consists of 9 image pixels giving a resolutionof about 0.5◦x0.5◦ at the Equator.

The POLDER-1 instrument was mounted on theJapanese ADEOS-1 platform. Data from November1996 until June 1997 are available. Since December2002, the POLDER-2 instrument is operating on theADEOS-2 satellite.

39

3.3. Derived physical data 3. POLDER satellite observations

3.3. Derived physical data

A multitude of physical information can be derivedfrom the POLDER measurements, including landsurface properties, ocean color, radiation budgetmeasurements, and informations about clouds andaerosols. The latter two are of particular interestin the present study. Concerning the aerosols, theaerosol optical thickness, τa, is derived in the 865 nmchannel over oceans (Deuze et al., 1999). Furtheron, the Angstrom coefficient, α, which is a measureof the particle size, is derived over both oceans andcontinents using the polarized measurements in the865 and 670 nm channels. A POLDER aerosol index(AI) is defined as the product of Angstrom coeffi-cient and aerosol optical thickness, ατa. This indexis calculated over both land and ocean (Deuze et al.,2000; Goloub and Arino, 2000; Deuze et al. 2001Tanre et al., 2001). All aerosol products are derivedfor cloud-free pixels according to the cloud screeningalgorithm of Breon and Colzy (2000).

In this study the cloud optical thickness, clouddroplet effective radius, cloud top temperature, andcloud top thermodynamic phase will be used. In or-der to deduce the cloud top temperature, first, thecloud top Rayleigh pressure is first retrieved from thepolarized signal at 443 nm (Vanbauce et al., 1998).The temperature is then obtained by using the six-hourly ECMWF analysis data. The cloud thermo-dynamic phase retrieval is based on the angular andpolarized signatures of cloud reflected radiances atscattering angles near 140◦ (Goloub et al., 2000; Riediet al., 2000). Cloud optical thickness is derived in amanner similar to that used in the ISCCP (Rossowand Schiffer , 1991), assuming a homogeneous plane-parallel liquid water cloud with an effective radius of10 µm (Buriez et al., 1997; Parol et al., 2000), usingup to 14 viewing directions in the 670 nm channel.For liquid water clouds homogeneous at a large scale(150 x 150 km2), Breon and Goloub (1998) derivefrom the multi-directional polarized radiance mea-surements a cloud top droplet effective radius.

3.4. Comparison of LMDZ and

POLDER data

In order to compare POLDER-derived physical quan-tities to those simulated by the LMDZ GCM, anintegration of the model was carried out for theperiod of the observations from November 1996 toDecember 1997. To get realistic conditions of thedynamics, the sea surface temperature and sea iceextent were imposed from the Reynolds and Smith(1995) and the HADISST1.1 (Rayner et al., 2003)datasets. Horizontal winds and temperature in themodel were nudged to ECMWF reanalysis data us-ing relaxation time constants of 0.1 day for the windsand 1 day for the temperature, respectively. Theswath of the satellite is sampled in the model togive a daily coverage of the globe in the same waythe POLDER instrument does it. Finally, for clouddroplet radius, cloud top temperature, and cloud topthermodynamic phase, the cloud top quantities arecalculated for the top of the atmosphere using therandom overlap assumption for the fractional cloudi-ness in the vertical. Figure 3.1 shows the comparisonof the distributions obtained for model simulationsand POLDER observations, averaged over the whole8-month period. POLDER data is averaged to thecoarser model grid.

For the aerosol index, the model simulates well theland-sea contrast in the aerosol burden. However, theaerosol burden diminishes too strongly away from themain source regions. Where the POLDER observa-tions show relatively large aerosol burdens over al-most all continental surfaces, the model shows ratherlow values for some regions of the South and NorthAmerican continents and over Australia. Negativeaerosol indices represent large aerosol particles. It ishowever rather difficult to interpret the differencesin model and observations as in particular for smallaerosol burdens the observations may not be veryreliable either.

Cloud optical thickness generally is overestimatedby the model, a problem common to GCMs, which ispartly due to the coarse vertical resolution and theassumption of plane-parallel homogeneous clouds.

40

3. POLDER satellite observations 3.4. Comparison of LMDZ and POLDER data

Figure 3.1.: Comparison of aerosol index for (a) the LMDZ model and (b) POLDER data, (c-d) cloud opticalthickness, (e-f) cloud top droplet radius [µm], (g-h) cloud top thermodynamical phase, and (i-k)cloud top temperature [◦C], for the period November 1996 to June 1997.

41

3.5. Summary and conclusions 3. POLDER satellite observations

However, several features are well simulated. Opticalthicknesses are larger in the northern hemisphere,and generally somewhat larger over land than overoceans. The larger optical thickness in the ITCZregion is well simulated.

The cloud droplet effective radius is estimated forlarge scale liquid water clouds in both model andobservations and taken at the cloud top. The modelshows generally too low CDR values, underestimat-ing the size of the droplets by 2-3 µm on average. Thecontrast of larger droplets over oceans than over landis observed in both LMDZ model and POLDER data.Similarily, a contrast between both hemispheres iswell simulated in the model. Particularly small radiiover the industrialized regions of the northern hemi-sphere, the eastern part of North America, Europe,and East Asia, and off the east coasts of the NorthAmerican and Asian continents is found in both,model and observations.

Concerning the cloud top thermodynamical phase,the model often simulates too many ice clouds (lowvalues in Fig. 3.1). However, it is able to simulatemore ice clouds over the continents than over oceans,as observed, and more ice clouds in the northernthan in the southern hemisphere. The high level iceclouds in the ITCZ region are somewhat overesti-mated. The maritime stratocumulus regions off thewest coasts of the continents with large liquid wa-ter fractions particularly in the southern hemisphereare very well captured by the model. The cloud toptemperature is related to the thermodynamical phase(see also Chapter 8). Thus it is similarly found that

clouds are higher over land than over ocean, andgenerally higher in the northern than in the southernhemisphere. However, the model generally places theclouds a little bit too high in the atmosphere.

3.5. Summary and conclusions

As the POLDER instrument is able to observe avariety of aerosol and cloud properties, the datasetsare particularly useful to investigate cloud propertiesand aerosol-cloud interactions on a global scale. Theapproach to do this applied here is the evaluation ofa global-scale model using the satellite observations,and to use the combination of model and observa-tions to improve the understanding of the processes(see Chapters 8 and 5 for further results).

A comparison of a nudged simulation with the GCMand the satellite observations reveals some skills andsome deficiencies of the model. Aerosol indices, cloudthermodynamical phase and cloud height are ratherwell captured by the model, showing very similarfeatures compared to the POLDER observations, inparticular when looking at hemispheric and land-sea contrasts. The latter ones are well captured forcloud optical thickness and cloud droplet size as well.However, systematic biases are found in these twoquantities, with too small droplets, and too largecloud optical thicknesses.

Unfortunately, so far, only eight months of POLDERmeasurements are available. However, since Decem-ber 2002, the POLDER-2 instrument is in space, andnew data will be available soon.

42

3. POLDER satellite observations 3.6. Resume

3.6. Observations spatiales

POLDER : Resume

Introduction

Les instruments satellitaires de type radiometremesurent des luminances au sommet de l’atmosphere.Comme ces luminances dependent de l’etat de la sur-face terrestre et de l’atmosphere, de l’informationpeut etre deduite sur une multitude de proprietesde la surface et de l’atmosphere. Les luminancesmesurees par des instruments satellitaires dans lagamme des ondes courtes sont composees de la par-tie du rayonnement solaire qui est reflechie par lasurface terrestre ou par les differentes composantesde l’atmosphere. La partie � terrestre� du spec-tre est celui des longueurs d’onde d’environ 4 a 25µm, et est composee par le rayonnement emis par lasurface, par certaines molecules dans l’atmosphere,les aerosols et les nuages. Des instruments typiquesmesurent l’intensite du rayonnement dans differentesbandes de longueurs d’ondes, qu’on appelle aussi descanaux de l’instrument. si les proprietes de diffusionet d’absorption des composantes de l’atmospheresont connues, des informations sur la distribution deces composantes peuvent etre deduites des signauxmesures.

On distingue deux types de satellites. Les satel-lites geostationnaires sont positionnes au-dessus del’equateur a une hauteur specifique ou leur vitesse derotation est identique a celle de la terre. La memeregion de la surface de la terre est ainsi toujoursobservee. L’avantage de cette methode est que desmesures continues sont en principe possibles. Parcontre, seulement une partie du globe peut etre ob-servee, et en particulier les observations des hauteslatitudes ne sont pas tres fiables. De plus, l’altituderelativement elevee de ces satellites affaiblit le signalmesure. La deuxieme classe de satellites est celle dessatellites defilants polaires qui tournent a une alti-tude donnee, la terre tournant au-dessous du satel-lite. Chaque point de la surface terrestre est ainsisurvole a peu pres une fois par jour avec une faucheeau sol d’environ 2500 km, les hautes latitudes etantobservees plusieurs fois par jour et les basses lati-tudes moins frequemment. Des instruments peuvent

egalement etre capables de mesurer perpendiculaire-ment et le long de la trace.

L’instrument � POLarization and Directional-ity of the Earth’s Reflectances� (POLDER) aete developpe par le Laboratoire d’Optique Atmo-spherique (LOA) et le Centre National de EtudesSpatiales (CNES). C’est le premier instrument ca-pable de mesurer simultanement les caracteristiquesmuli-spectrales, multi-directionnelles et polarisees durayonnement solaire reflechi (Deschamps et al., 1994).En particulier, la polarisation du rayonnement ob-servee donne des nouvelles possibilites de deduire desproprietes des composantes de l’atmosphere.

L’instrument POLDER utilise 15 bandes spectralesentre 0.44 et 0.91 µm. Il mesure dans une zone de± 43◦ le long de et ± 51◦ perpendiculairement a latrace du satellite, ce qui donne une fauchee au sol de2400 km. La resolution au nadir est de 6 × 7 km2.Un � superpixel� des donnees derivees est com-pose de neuf pixels de l’image donnant une resolutiond’environ 0.5 × 0.5◦ a l’equateur.

L’instrument POLDER-1 a ete monte a bord dusatellite Japonais ADEOS-1 et a fourni des donneesde novembre 1996 a juin 1997. La deuxieme version(POLDER-2) de l’instrument a vole sur le satelliteADEOS-2 de decembre 2002 a octobre 2003.

Donnees physiques

Une multitude d’informations physiques peut etrededuite des mesures de POLDER comme des pro-prietes de la surface continentale, la couleur del’ocean, le bilan radiatif, et des informations sur desnuages et des aerosols. Ce sont les deux dernieresqui sont d’interet particulier pour cette etude. Con-cernant les aerosols, l’epaisseur optique des aerosols,τa, est deduit des canaux a 670 et a 865 nm au-dessus des oceans (Deuze et al., 1999). De plus lecoefficient d’Angstrom, α, qui est une mesure de lataille des particules, est egalement derive au-dessusdes oceans. Un indice des aerosols POLDER, AI, estdefini comme le produit du coefficient Angstrom etde l’epaisseur optique des aerosols, τa. Cet indice estcalcule au-dessus des continents et des oceans (Deuze

43

3.6. Resume 3. POLDER satellite observations

et al., 2000 ; Goloub et Arino, 2000 ; Deuze et al.,2001 ; Tanre et al., 2001). Dans cette etude, concer-nant les proprietes des nuages, l’epaisseur optique,le rayon effectif des gouttelettes, la temperature,et la phase thermodynamique au sommet des nu-ages seront utilises. Afin de deduire la temperatureau sommet des nuages, la pression � Rayleigh�est d’abord deduite du signal polarise a 443 nm(Vanbauce et al., 1998). La temperature est en-suite derivee en utilisant des analyses du CentreEuropeen des Previsions Meteorologiques a MoyenTerme (CEPMMT, ECMWF en anglais). La phasethermodynamique au sommet des nuages est estimesur la base des luminances polarisees a des anglesde diffusion autour de 140◦ (Goloub et al., 2000; Riedi et al., 2000). L’epaisseur optique des nu-ages est obtenu pareillement a la procedure d’ISCCP(Rossow et Schiffer, 1991), en supposant des nuageshomogenes, d’eau liquide et composes de particulesspheriques d’un rayon effectif de 10 µm, et en util-isant jusqu’a 14 directions de visee dans le canal a670 nm (Buriez et al., 1997 ; Parol et al., 2000). Pourdes nuages en eau liquide homogenes a une echelle de150 × 150 km2, Breon et Goloub (1998) deduisent lerayon effectif des gouttelettes au sommet des nuagesa partir de l’analyse du signal polarise.

Comparaison donnees POLDER - modele

LMDZ

Afin de comparer les quantites physiques obtenuespar l’instrument POLDER a celles simulees parle modele de circulation generale LMDZ, uneintegration du modele a ete effectuee pour la periodedes observations de novembre 1996 a juin 1997.Pour assurer des conditions realistes de la partie dy-namique du modele, la temperature de la surface desoceans et l’extension de la glace de mer ont ete im-posees au modele a partir des observations (Reynoldset Smith, 1995 ; Rayner et al., 2003). Les vents hori-zontaux et la temperature ont ete guides en utilisantdes constantes de relaxation de 0.1 jour pour les ventset de 1 jour pour la temperature. La fauchee du satel-lite a ete echantillonnee dans le modele afin d’obtenirune vue du globe journaliere de la meme manierequ’elle se fait avec l’instrument spatial. Les quan-tites au sommet des nuages sont calculees en utilisant

une hypothese de recouvrement aleatoire des frac-tions nuageuses. Des distributions moyennees sur laperiode simulee de l’indice des aerosols, de l’epaisseuroptique des nuages, du rayon effectif des gouttelettes,de la phase thermodynamique et de la temperatureau sommet des nuages sont comparees aux quantitesobservees par POLDER (fig. 3.1).

Le modele simule bien le contraste terre-mer dansle contenu en aerosols, meme si les valeurs eleveessont trop concentrees autour des regions sources.L’epaisseur optique des nuages est surestimee engeneral par le modele, ce qui est un probleme com-mun a beaucoup de modeles et qui est en particulierd a la resolution verticale grossiere des modeles ainsiqu’a l’hypothese de l’homogeneite plan-parallele desnuages. Pourtant, plusieurs caracteristiques sont biensimulees par le modele. Les epaisseurs optiques sontplus grandes dans l’hemisphere nord, et en generalplus grand au-dessus des continents qu’au dessus desoceans. L’epaisseur optique elevee dans l’ITCZ estbien simulee.

Le modele en general simule des rayons effectifs troppetits d’environ 2 a 3 µm en moyenne. Le contrasteentre des gouttelettes plus grandes au-dessus desoceans et plus petites au-dessus des continents estbien observe a la fois dans les donnees POLDER etdans LMDZ. Pareillement, le contraste entre les deuxhemispheres est bien simule dans le modele. Des par-ticules particulierement petites au-dessus des regionsindustrialisees de l’hemisphere nord, la partie est del’Amerique du Nord, Europe et l’Asie de l’Est, sontidentifiees a la fois dans les observations et dans lemodele.

Concernant la phase thermodynamique au sommetdes nuages, le modele simule souvent trop de nu-ages de glace. Pourtant, il est capable de donnerplus de nuages de glace au-dessus des continentsqu’au-dessus des oceans, comme c’est observe, etplus de nuages de glace dans l’hemisphere nord quedans l’hemisphere sud. Les nuages de glace dansl’ITCZ sont legerement surestimes. Les regions destratocumulus marins au large des cotes ouest descontinents avec des fractions importantes des nuages

44

3. POLDER satellite observations 3.6. Resume

d’eau liquide en particulier dans l’hemisphere sudsont bien simulees dans le modele. Comme la phasethermodynamique est correlee a la temperature, des

resultats similaires sont trouves pour la distributionde la temperature au sommet des nuages.

45

3.6. Resume 3. POLDER satellite observations

46

Part II.

Model evaluation and process studies

47

4. Evaluating parameterizations of cloud-aerosol

processes with single column models and

ACE-2 CLOUDYCOLUMN observations

4.1. Introduction

Contributions to the first aerosol indirect effect areconsidered to be largest for low-level stratiformclouds of moderate optical thickness. Low-levelclouds with liquid water paths large enough to pro-duce drizzle or rain in clean conditions would be mostsusceptible to the second AIE. Both conditions arefulfilled for marine stratocumulus clouds as observedparticularly off the west coasts of the continents.Some of the first evidence for the aerosol indirecteffects was provided by so-called ship tracks in ma-rine stratocumulus clouds. The exhaust from shipsresults in brighter clouds along the track of a ship,observable in satellite observations (e.g., Coakley Jr.et al., 2000; Durkee et al., 2000; Ackerman et al.,2000b). An explanation for this observation given inthe references cited is an increase in cloud dropletnumber concentration due to the first AIE, and anincrease in cloud liquid water content due to the sec-ond AIE.

An observational method well suited to provide ev-idence for and insight into cloud processes are fieldcampaigns, in which in situ measurements of a mul-titude of different cloud and aerosol properties canbe measured using aircraft.

The Second Aerosol Characterization Experiment(ACE-2) was dedicated to the measurement of aerosoleffects in a marine stratocumulus cloud environmentat midlatitudes. The ACE-2 campaign took place inthe in June/July 1997, between Portugal and the Ca-

nary Islands. Five airplanes were measuring in situaerosol and cloud properties along with radiationfluxes. In addition, aerosol measurements from thePunta del Hidalgo station on the Tenerife Island areavailable. In the CLOUDYCOLUMN project, severalcloud microphysical properties have been measuredin totally eight case studies (Brenguier et al., 2000a).In the present study, data from two particular daysare selected. On June 26th, a cyclone over westernEurope advected pristine maritime airmasses fromthe North Atlantic Ocean to the ACE-2 region. Theday of July 9th was the most polluted case, where ahigh pressure ridge carried polluted airmasses fromthe European continent to the region of measure-ments.

The study presented in this chapter was carried outin the framework of the European Project namedParameterization of the Aerosol indirect ClimaticEffects (PACE). Six single-column model (SCM) ver-sions of GCMs were used to study the behavior ofthe different parameterizations in the two ACE-2CLOUDYCOLUMN cases, June 26th and July 9th,1997. The results of this study are published inMenon et al. (2003). While the evaluation of thefirst AIE is rather simple, as just the influence ofthe aerosol concentration on cloud droplet numberconcentration (CDNC) has to be considered, both ofwhich are observable quantities, the evaluation of thesecond AIE is much more complicated. The secondAIE acts via the microphysical processes of precipi-tation formation, and it results in changes of cloudliquid water content, cloud lifetime, and precipita-

49

4.2. Observational data 4. Evaluation of cloud-aerosol parameterizations with ACE-2

Date ma [µg m−3] Na [cm−3] SO4 [%] ss [%] OC [%]June 26th 1.2 700 21 15 25July 9th 5.8 4400 47 10 20

Table 4.1.: The aerosol mass concentration, ma, number concentration, Na, and the mass fractions of SO4,seasalt (ss) and organic carbon (OC), for the clean (June 26th) and the polluted (July 9th) casesof the CLOUDYCOLUMN experiment (from Guibert et al., 2003)

Date Nd [cm−3] rv [µm] LWP [g m−2] H [m] HB [m] τc αc

June 26th 52±16 7.77±3.64 18.5±17.8 202 1283 3.99±2.29 0.19±0.01July 9th 256±38 4.73±1.67 11.0±10.8 167 817 4.23±2.54 0.27±0.02

Table 4.2.: Cloud droplet number concentrations, Nd, cloud droplet volume mean radius, rv, cloud liq-uid water path, LWP, cloud geometrical thickness, H, the cloud base height, HB , cloud opticalthickness, τc, and cloud albedo, αc, for the clean (June 26th) and polluted (July 9th) cases ofthe CLOUDYCOLUMN experiment. Nd, rv, LWC, H, and HB are deduced from in situ mea-surements (Brenguier et al., 2003; Pawlowska and Brenguier , 2000). LWP is calculated fromLWC using the maximum overlap assumption. τc and αc are obtained from remote sensing data(Schuller et al., 2003)

tion rate. Thus, a perfectly suited evaluation wouldrequire the knowledge of the distributions of thesequantities, which in contrast to CDNC and aerosolconcentration are highly variable both spatially andtemporally. While we are aware of the problems dueto the assumptions of homogeneity, we rather focuson the evaluation and comparison of different param-eterizations of the autoconversion parameterization,which is the most important process for the onset andproduction of precipitation (The distinction betweenautoconversion between droplets to form raindropsand accretion of falling raindrops by the collection ofdroplets is different in large and small scale models.In small scale models, the accretion process is oftenof much more importance in the rain formation, afact due to the higher resolution of these models).

4.2. Observational data

Guibert et al. (2003) report the aerosol concentrationsand types for the different ACE-2 cases, measured atthe Punto del Hidalgo ground station and using air-craft measurements. Table 4.1 summarizes the val-ues given in their study. Pawlowska and Brenguier(2000) established a dataset of cloud parameters on

a region of 60x60 km2 covering the full track of theflights. They select the (almost) adiabatic cases fromall available measurements and exclude cases wheredrizzle has been observed. The mean droplet num-ber concentration and volume mean droplet radiusare deduced from the mean droplet spectra. Simi-larly, Brenguier et al. (2003) give the mean values forcloud droplet radius, cloud droplet number concen-tration, and cloud liquid water content, for five sub-layers of the stratocumulus cloud. Table 4.2 summa-rizes the characteristic values of the cloud quantitiesfor the two clean and polluted cases. Together withmeasurements of the rainfall rate a complete cloudand precipitation microphysics and aerosol datasetis given at the scale of a GCM grid box for the twoACE-2 CLOUDYCOLUMN situations, a clear (June,26th) and a polluted (July, 9th) one.

4.3. The SCMs

Six SCM versions of GCMs were used in thePACE project, named CSIRO, GISS, MetO, PNNL,ECHAM, and the LMDZ. Descriptions of the re-spective models are summarized in Table 4.3. Asthe situations of interest are non-convective, three

50

4. Evaluation of cloud-aerosol parameterizations with ACE-2 4.3. The SCMs

Set

-up

CSIR

OG

ISS

Met

OP

NN

LE

CH

AM

LM

DZ

Inves

tigato

rJ.Pen

ner

,X

.Liu

;

Univ

.M

ichig

an

S.M

enon,A

.D

el

Gen

io;G

ISS

P.

Dav

ison,

D.

Rober

ts;

Hadle

y

Cen

tre

S.G

han;P

NN

LU

.Lohm

ann;

Dalh

ousi

eU

niv

.,

J.

Fei

chte

r;M

ax

Pla

nck

Inst

itute

J.

Quaas;

LM

D/IP

SL,

O.B

ouch

er;LO

A

#le

vel

s(#

bel

ow

720hPa)

18

(6)

31

(8)

30

(13)

24

(11)

19

(6)

19

(6)

Init

ializa

tion

tim

e

0Z

25/06

and

0Z

08/07

0Z

25/06

and

0Z

08/07

0Z

26/06

and

0Z

09/07

0Z

26/06

and

0Z

09/07

Apri

l1

0Z

26/06

and

0Z

09/07

Tre

atm

ent

ofsu

r-

face

fluxes

Dia

gnose

dby

SC

Mfr

om

win

d

and

tem

per

atu

re

Dia

gnose

dby

SC

Mfr

om

sur-

face

win

dand

tem

per

atu

re

Dia

gnose

dby

SC

M

Dia

gnose

dby

SC

Mfr

om

win

d

and

tem

per

atu

re

Sta

ndard

GC

M

trea

tmen

t

Dia

gnose

dby

SC

Mfr

om

win

d

and

tem

per

atu

re

Nudgin

gY

es.

Yes

.N

o.

No.

Yes

.Y

es.

Dro

ple

tnucl

e-

ati

on

Physi

cally

base

d

usi

ng

updra

ft

vel

oci

ty,

aer

oso

l

num

ber

and

anth

ropogen

ic

sulfate

mass

(Chuang

etal.,

1997).

Em

pir

ically

base

din

term

sof

mass

of

sulfate

,

org

anic

and

sea-

salt

aer

oso

lsand

cloud

turb

ule

nce

(Men

on

etal.,

2002)

Em

pir

ically

base

d(J

ones

etal.,2001)

Physi

cally

base

d

usi

ng

updra

ft

vel

oci

ty,

aer

oso

l

num

ber

con-

centr

ati

on

and

dis

trib

u-

tion

pro

per

ties

(Abd

ul-Razz

ak

and

Ghan,1999)

Physi

cally

base

d,

modifi

edver

sion

ofth

eG

han

etal.

(1997a)-

Sch

eme

(Lohm

ann,

2002a)

Em

pir

ically

base

dusi

ng

sul-

fate

aer

oso

lm

ass

(Bouch

erand

Lohm

ann,1995)

Subgri

dq l

var.

in

auto

conv.

Tri

angula

rP

DF.

None.

None.

Tri

angula

rP

DF.

None.

None.

Ref

eren

ces

Rots

tayn

(1997);

Rots

tayn

(2000)

Hanse

net

al.

(1997);

Del

Ge-

nio

etal.

(1996);

Men

on

etal.

(2002)

Wilso

nand

Ballard

(1999);

Jones

etal.

(2001);

Loc

k

etal.

(2000)

Ghan

etal.

(1997a);

Abd

ul-

Razz

ak

and

Ghan

(1999)

Lohm

ann

etal.

(1999);

Lohm

ann

(2002a)

Li

(1999);

Le

Tre

ut

and

Li

(1991);

Bouch

er

etal.

(1995a)

Tab

le4.

3.:Su

mm

ariz

edde

scri

ptio

nof

the

SCM

sus

edin

the

PAC

Est

udy.

51

4.4. Simulations 4. Evaluation of cloud-aerosol parameterizations with ACE-2

parameterizations are more particularly of interestin the SCMs, namely the boundary layer, radiation,and microphysics schemes. The focus of the study ison the cloud droplet activation scheme, which con-tributes to the first AIE, and on the autoconversionscheme, which has the main influence on the sec-ond AIE in the model. Thus, we will address somedifferences and similarities of the SCMs in the micro-physics parameterization in more detail.

Some of the set-up parameters of the one-day sim-ulations are summarized in Table 4.3. The verticalresolution of all GCMs is rather coarse, varying from18 to 31 layers, with roughly a third of them in theboundary layer (BL). The BL scheme in all SCMsdiagnoses the latent heat fluxes at the ocean surfaceusing the wind and temperature profiles. Radiationis parameterized in all SCMs using the two streamapproximation with spectral bands in the solar andterrestrial spectra. The schemes are described insome more detail in Menon et al. (2003) and thereferences given therein.

4.4. Simulations

4.4.1. One-day SCM simulations:

Method

One-day simulations with the six SCMs were carriedout in the framework of the PACE project for each ofthe two ACE-2 CLOUDYCOLUMN cases. For theCSIRO, GISS, MetO and PNNL SCMs, the initialconditions were prescribed from ECMWF reanalysisdata; the CSIRO and GISS models were also forcedwith ECMWF data. In contrast to this, the LMDZSCM and the ECHAM SCM were initialized andforced with profiles from a nudged 3D-run of theGCM.

For the LMDZ model, daily mean SST values wereimposed from the Reynolds SST dataset. Tempera-ture and horizontal winds were nudged to ECMWFreanalysis data using relaxation time constants of 1day and 0.1 day, respectively. The model grid waszoomed over the ACE-2 region with the center at16.5◦W and 29.5◦N, obtaining a resolution of 140x90

km2 in the zoom region. Values at the grid point(17.4◦W/29.5◦N) were taken to initialize and forcethe SCM. While winds, surface pressure and SSTare imposed on the SCM, temperature and humidityare nudged using a relaxation constant deduced fromwind speed and grid size (Ghan et al., 1999).

4.4.2. One-day SCM simulations: Results

Figures 4.1 and 4.2 show the cloud amount and thecloud liquid water content simulated by the six SCMs.Along with this, Fig. 4.3 shows the cloud amountfrom ISCCP satellite measurements on a scale similarto the model resolution.

The results from the six models differ in variousaspects. Three of the models (CSIRO, PNNL, LMD)tend to simulate continuously a cloud, which is inagreement with the observations. For the clean case(June 26th), MetO and ECHAM models simulatea continuous cloud as well. The daily mean cloudamount varies among the models between < 50%and >80% for June 26th, and shows even more scat-ter, between 10 and >80% on July 9th, the LMDmodel being at the upper end of this scale in bothcases. ISCCP satellite observations give a cloudamounts of 80% and 82%, respectively. Compar-ing the two cases, CSIRO, PNNL, and LMD modelssimulate rather similar diurnal cycles of the cloudin both cases, PNNL placing the cloud lower in thepolluted case. The diurnal cycle given by the LMDmodel is consistent with the ISCCP observations forJune 26th, while on July 9th, the minimum at night-time and early morning is not captured. For theclean case, it can be seen that the two models forcingthe SCM with profiles from a nudged 3D-simulation(ECHAM and LMD models) simulate a similarlypersistent cloud, so maybe the consistency in the 3Dfield enables the model to simulate a continuous cloudcover as observed. The two models that are able toresolve the vertical structure of the boundary layerclouds better than the other ones (MetO through abetter resolution, and GISS through a vertical sub-grid cloud fraction parameterization) simulate thesmallest cloud amounts in both cases.

Similarly, it is in these two models, that the high-

52

4. Evaluation of cloud-aerosol parameterizations with ACE-2 4.4. Simulations

Figure 4.1.: Time-pressure plot of the predicted cloud fraction for (a) the clean (June 26) and (b) polluted(July 9) cases. Observed cloud base and top levels are reported in the upper left graph (CSIRO).Cloud fraction from observations with the casi radiometer are also indicated on the color bar.

est cloud bases are simulated (Figure 4.4), in betteragreement with the observations. This feature is,however, more pronounced on June 26th than onJuly 9th. Again probably due to the coarse verticalresolution of the models, cloud geometrical thicknessis overestimated by all models, the MetO and GISSmodels again being somewhat closer to the obser-vations. It is remarkable that the simple verticalsubgrid parameterization in the GISS model givesresults very similar to the MetO model results whichuses a considerably better resolution (13 against 8layers in the boundary layer).

As shown in Fig. 4.5, all models strongly under-estimate the measured rainfall rate at the surface. It

can be noted, however, that the two models forcedfrom a consistent 3D field (ECHAM and LMD) areable to produce a larger rain rate than the othermodels, in better agreement with the observations.

Concerning CDNC and CDR, all models success-fully simulate more and smaller cloud droplets in thepolluted compared to the clean case. SCM parame-terizations differ primarily in the way of predictingNd from aerosol concentrations, PNNL and ECHAMusing physically based, the other models empiricallybased relationships. The PNNL model simulates toolarge CDNC in both cases, and ECHAM too lowvalues in the polluted case. The empirical schemesare rather close to each other, showing, however,

53

4.4. Simulations 4. Evaluation of cloud-aerosol parameterizations with ACE-2

Figure 4.2.: As Fig. 4.1, but for cloud water mixing ratio [g kg-1]. Observations from the casi radiometerare also indicated on the color bar.

0 6 12 18 24Time (hrs)

700

750

800

850

ISC

CP

Clo

ud T

op (h

Pa)

June 26July 09

0 6 12 18 24Time (hrs)

0

20

40

60

80

100

ISC

CP

Clo

ud A

mou

nt (%

)

June 26July 09

Figure 4.3.: ISCCP derived cloud amount and cloud top for the clean and polluted cases.

54

4. Evaluation of cloud-aerosol parameterizations with ACE-2 4.4. Simulations

Figure 4.4.: Cloud base [m] and cloud geometrical thickness [m] simulated by the six SCMs. The observationsare shown as a full circle.

Figure 4.5.: Rain rate at the surface [mm day−1].

different behaviors in response to the increase inaerosol burden from June 26th to July 9th. TheLMD model has the smallest sensitivity to increasedaerosol concentration, the GISS model the strongestone. Concerning the CDR, it is remarkable that allmodels rather overestimate the droplet size althoughseveral of them are within or close to the uncertainty

range given for the observations. It should be notedthat due to adaptation of the autoconversion schemesusing a “critical radius” for the onset of precipitation,models tend to simulate too small CDR (Rotstayn,2000). In the cases examined here, however, the CDRare well above this critical threshold value.

55

4.4. Simulations 4. Evaluation of cloud-aerosol parameterizations with ACE-2

Figure 4.6.: Cloud droplet number concentrations ([cm−3], upper panel) and cloud droplet effective radii([µm], lower panel).

4.4.3. Single-step simulations: Method

Single-step simulations were carried out to test themicrophysics parameterization alone. The modelswere initialized by the ECMWF temperature and hu-midity profiles for 12Z on June, 26th and July, 9th.

• EXP-N is the first experiment, in which theability of the parameterizations is tested to pre-dict the cloud droplet number concentrationwith given aerosol concentrations. In case ofthe LMDZ model, which uses the empirical for-mula of Boucher and Lohmann (1995), this isjust a diagnosis of this formula given the aerosolmass concentration.

• In EXP-A simulations, the vertical resolutionof the SCMs was increased to be able to re-solve the five sub-layers of the clouds as in theobservational data from Brenguier et al. (2003).Cloud droplet number concentration, Cloud liq-uid water content, and cloud geometry wereprescribed. With these simulations, the degree

of realism of the parameterization of cloud op-tical properties and radiation could be tested.At the same time, a comparison of rain ratesat the base of the cloud and at the ground ofmodels and observations allowed an evaluationof the autoconversion parameterizations.

• VERT was a third single-step parameterization,which was similar to EXP-A except that thestandard vertical resolution of the SCMs wasapplied.

In addition to the results published in Menon et al.(2003), here, in EXP-N, different parameterizationsfor the link between CDNC and aerosol mass aretested with the LMDZ SCM: The formulae “(A)” and“(D)” from Boucher and Lohmann (1995), and twomodifications of these two formulae: First, the sulfateaerosol mass is replaced by the maximum of the massconcentrations of three different species of hydrophilicaerosols (sulfate, organic matter, and sea salt), andsecond, the sum of the mass concentrations of hy-drophilic aerosols is taken. The idea is that using the

56

4. Evaluation of cloud-aerosol parameterizations with ACE-2 4.4. Simulations

sum or the maximum of the masses of all hydrophilicaerosols instead of sulfate alone could help to over-come the difficulty of using sulfate as a surrogate forall aerosol types in regions where sulfate aerosols aresparse compared to other potential CCN.

4.4.4. Single-step simulations: Results

and Discussion

For the EXP-N experiment (Table 4.4), all modelparameterizations simulate larger CDNC concentra-tions for the polluted compared to the clean case.However, no clear distinction can be made betweenphysically (PNNL and ECHAM) and empiricallybased schemes. While ECHAM overpredicts CDNCin the clean case, it matches the observations exactlyin the polluted case. PNNL, however, matches theobservations exactly in the clean case and largely un-derpredicts the CDNC in the polluted case. Amongthe other models, GISS uses a scheme partially de-rived from the ACE-2 data. This scheme, however,is nevertheless not able to reproduce the observedCDNC in both cases. For the LMD model, two dif-ferent empirical formulae of Boucher and Lohmann(1995) were compared, and each of these was againmodified to also take into account the non-sulfateaerosols. This was done in one study by using themaximum of the masses of sulfate, hydrophilic or-ganic matter, and submicronic sea salt, and in a sec-ond one by using the sum of all these aerosol masses.Using the sum of all masses, the model seems tooverpredict largely CDNC. Both formulae overpre-dict CDNC for the clean case for any of the aerosolmass concentrations tested. Formula “A” is betterin predicting CDNC in the clean case, formula “D”better in the polluted case. It can be deduced thata formula which showed more sensitivity to aerosolconcentration would be needed to simulate the dif-ferences between the two cases well.

Tables 4.6 and 4.7 give the results for the EXP-Asimulations, for the model-simulated cloud opticalthickness, τc, and the SW planetary albedo at TOA.For each of the two clear and polluted cases, foursimulations were carried out: The first time, the liq-

uid water content of the cloud was scaled to matchthe LWP observed by the FSSP instrument, and thesecond time, to match the OVID measurement ofLWP. These two LWPs differ a lot. It is difficult toidentify which estimate is better. On the one hand,as the OVID instrument infers LWP from radiationmeasurements, some uncertainty may be introduced.On the other hand, this method is able to capturea view of all LWP situations, where the FSSP insitu instrument can only collect samples. Therefore,the two different LWPs were used in the simulations.Then, these two simulations were done twice, at theresolution of the dataset, and at the lower standardmodel resolution.

Looking at the model-simulated cloud optical thick-ness, τc, mainly two parameterizations can be exam-ined. The first is the way to derive the cloud opticalthickness from (fixed) LWP, W , and CDR, re. Thisis done using the formula

τc =32

W

ρwre(4.1)

from Stephens (1978), where ρw is the water density,for all models except the MetO model, which uses theparameterization from Slingo (1989)

τc = W

(a +

b

re

)(4.2)

with two empirical constants a and b. However, insome of the models, τc is further scaled to considersubgrid-scale variability. In the GISS model, thisfactor is f

13 (where f is the cloud fraction), which is

explained as a measure of possible vertical subgrid-scale variability (Del Genio et al., 1996). For thePNNL, the scaling factor is f

32 , and the ECHAM

uses an “inhomogeneity factor” of 1.− 0.06 W13 from

Tompkins (2002). Further on, the CDR is calcu-lated using the prescribed liquid water content andCDNC. Therefore, the volume-mean droplet radius,rdv, is calculated assuming spherical particles. Thisis related to the effective radius by a factor k, whichis (for the oceanic conditions) 1.077 in the CSIRO,MetO and ECHAM models, 1.1 in the PNNL andLMD models, and 1.28 in the GISS model, which arein the range of the observations (Pontikis and Hicks,1993; Martin et al., 1994). The second parameteri-zation of interest is how the clouds are assumed to

57

4.4. Simulations 4. Evaluation of cloud-aerosol parameterizations with ACE-2

Date FSSPOBS

OVIDOBS

CSIRO GISS MetO PNNL ECHAM

June 26th 52±16 46 68 51 34 52 84July 9th 256±38 230 358 217 242 138 255

SO4, “D” SO4, “A” max, “D” max, “A” sum, “D” sum, “A”June 26th 92 59 99 64 142 99July 9th 245 186 245 186 300 235

Table 4.4.: The predicted CDNC [cm−3] from EXP-N (with observed aerosols prescribed), from FSSP andOVID observations and from the five SCMs (upper panel), and for six realizations with theLMDZ-GCM (lower panel, “A” and “D” indicate the two respective formulas of Boucher andLohmann (1995), SO4 means sulfate aerosols only, “max”, the maximum of the mass concentra-tions of different hydrophilic aerosols is taken instead of sulfate mass, and “sum”, the sum oftheir mass concentrations is taken).

CSIRO GISS MetO PNNL ECHAM LMD

Opticalthickness

32

Wρwre

32

Wρwre

W(a + b

re

)32

Wρwre

32

Wρwre

32

Wρwre

Scaling of τc none f13 none f

32 1−0.06W

13 none

Factor in re-rd,v relation

1.077 1.28 1.077 1.1 1.077 1.1

Verticaloverlap

random none max-rand random max-rand random

Table 4.5.: Model parameterizations for the derivation of cloud optical thickness

overlap in the vertical, at least in the case of the highresolution, where multiple cloud layers exist. CSIRO,PNNL and LMD use a random overlap assumption,and MetO and ECHAM maximum-random over-lap. In the GISS model, a layer is considered eitherentirely cloudy or clear in the radiation scheme, de-pending on a random number and the cloud fraction.Table 4.5 summarizes the differences in the parame-terizations. For the cloud optical thickness, all mod-els simulate the values within the uncertainty givenfor the FSSP observations. For the high resolution,PNNL shows the smallest optical thicknesses, whichis due to the large scaling factor in τc. Similarly, theGISS model also simulates rather low values. Thedifferent formulation for cloud optical thickness usedin the MetO gives values comparable to those ob-tained from the Stephens (1978) parameterization.This, however, may also be due to the fact that theMetO model uses a maximum-random vertical over-lap assumption, which is different from the othermodels. The fact that the GISS model shows rather

low values may also be due to the relatively largevalue of the scaling factor k. The LMD model simu-lates optical thicknesses within the range of the otherSCMs, but generally somewhat larger values than themean. When initialized with rather low liquid watercontents, this is closer to the observations, while forrelatively larger LWP, this is less realistic.

The models generally show larger optical thicknessesand albedos when coarsening the vertical resolu-tion. For the coarse resolution, the ECHAM modelshows rather low optical thicknesses which may beattributed to the use of the inhomogeneity factor forthe scaling of τc.

Table 4.7 shows the SW planetary albedo. Here,the different radiation parameterizations can be eval-uated against the other ones. Generally, the schemesare able to simulate slightly larger albedos for thepolluted compared to the clean case, however, thisincrease is less in the models than in the OVID obser-

58

4. Evaluation of cloud-aerosol parameterizations with ACE-2 4.4. Simulations

Date Res. LWP OBS CSIRO GISS MetO PNNL ECHAM LMDJune 26th O F 3.99±2.29 3.63 3.34 3.26 3.21 - 3.50

M F 4.32 4.40 - 5.64 3.80 5.50O O 5.09 7.59 8.19 6.44 7.05 - 7.30M O 9.01 9.02 - 12.4 7.00 11.4

July 9th O F 4.23±2.54 4.23 3.49 3.66 3.20 - 4.00M F 4.93 4.73 - 3.33 3.00 6.00O O 4.99 8.85 7.77 7.64 7.20 - 8.50M O 10.4 9.90 - 7.20 8.00 12.5

Table 4.6.: Cloud optical thicknesses simulated by the SCMs in EXP-A. Results for a vertical resolution(column “Res.”) of the observations (O) and the model (M) are given. In one set of simulations,the liquid water path of the FSSP measurements was imposed to the models (column “LWP”, F),and in another one, the LWP of the OVID measurements (O). FSSP deduced an optical thicknessof 3.99±2.29 and 4.23±2.54 for June 26th and July 9th, respectively, and OVID 5.09 and 4.99,respectively.

Date Res. LWP OBS CSIRO GISS MetO PNNL ECHAM LMDJune 26th O F 0.15±0.01 0.26 0.19 0.18 0.17 - 0.14

M F 0.27 0.23 - 0.27 0.21 0.18O O 0.30 0.30 0.27 0.27 - 0.21M O 0.32 0.35 - 0.41 0.27 0.28

July 9th O F 0.18±0.01 0.27 0.20 0.19 0.18 - 0.15M F 0.27 0.24 - 0.18 0.19 0.19O O 0.32 0.30 0.27 0.29 - 0.22M O 0.33 0.37 - 0.29 0.30 0.28

Table 4.7.: As Table 4.6, but for TOA SW albedo.

vations. Models generally simulate too large albedos,except for some cases, as e.g. the LMD model wheninitialized with low LWPs and at high resolution.Coarsening the resolution results in the simulation oflarger albedos, which is consistent with the increasein cloud optical thickness. Also due to the simulateddifferences in COT, the CSIRO model generally sim-ulates rather large albedos, and the PNNL modelrather low ones.

In Table 4.8, the rainfall rates simulated by thedifferent autoconversion schemes are listed. It isastonishing that all models extremely underestimatethe rainfall rates. A reason could be the inherentassumption of homogeneity. When increasing the

vertical resolution from standard model resolution tothe resolution of the observational dataset, somewhatmore precipitation is simulated by some of the mod-els, but still this precipitation is one to three ordersof magnitude too small. The differences among themodels that use an autoconversion scheme similarto the Tripoli and Cotton parameterization (GISS,MetO, CSIRO, LMD) can be used to examine theimpact of the choice of parameters. However, as hasbeen examined already in the analysis of the one-daysimulations, the autoconversion threshold does notplay a role, as the droplets are generally larger thanthe critical values of 7.5 or 8 µm used in the schemes.Testing the Khairoutdinov and Kogan (2000) auto-conversion scheme in the ECHAM, which is similar

59

4.5. Summary and conclusion 4. Evaluation of cloud-aerosol parameterizations with ACE-2

Date OBS Res. CSIRO GISS MetO PNNL ECHAM LMDJune 26th 25.5 O 0.43 0.01 3.33 0.05 - 0.41

M 0.00 0.01 - 0.00 0.00 0.00July 9th 2.22 O 0.00 0.00 0.83 0.01 - 0.00

M 0.00 0.07 - 0.01 0.00 0.00

Table 4.8.: Rain water fluxes at cloud base [mg m−2 s−1] from EXP-A for the clean and polluted cases, forthe vertical resolution of the observations (“O”) and of the models (“M”).

to the Tripoli and Cotton (1980) parameterizationbut does not include a “critical value” reveals thatthis scheme is not able either to produce a realisticdrizzle rate.

4.5. Summary and conclusion

Among single-column versions of six general circula-tion models, the LMDZ including the microphysicalscheme of Boucher et al. (1995a) has been evalu-ated using in situ observations from two cases of theACE-2 CLOUDYCOLUMN field study, a clean anda polluted one. The observational data is availableaveraged over a homogeneous maritime stratocumu-lus cloud on a scale of 60x60 km2, which is a sizecomparable to the GCM resolution.

In all the quantities evaluated, the SCMs show a verylarge scatter. The stratocumulus clouds simulated bythe different models in one day simulations vary infractional cloudiness, liquid water content, CDNC,and cloud optical thickness. However, some resultsare very useful. Models applying a high resolutionin the boundary layer or a vertical subgrid-scale

parameterization are better able to simulate cloudbase height and size. Models forced from a consis-tent three-dimensional field simulated cloud fractionsbetter, indicating that the parameterizations mightdo a better job in the 3D than in the 1D version,possibly due to a canceling of errors.

All the models simulate larger CDNC and smallerdroplet sizes in the polluted compared to the cleancase. However, it is not possible to decide whetherphysically or empirically based schemes are preferablefrom their skill. Precipitation is extremely underes-timated indicating some need for drizzle parameter-ization. The fact that none of the models is ableto simulate drizzle of observed magnitude in neitherthe clean nor the polluted case indicates also that asecond indirect effect cannot adequately be modeledwith current GCMs. As expected, cloud optical prop-erties were simulated much more realistically whenincreasing the vertical resolution. However, none ofthe individual parameterizations for cloud opticalthickness, vertical cloud overlap, or effective radiuscan be distinguished to be preferable compared tothe other ones.

60

4. Evaluation of cloud-aerosol parameterizations with ACE-2 4.6. Resume

4.6. Evaluation des

parametrisations de processus

aerosols-nuages avec des

modeles unicolonnes et des

observations ACE-2

CLOUDYCOLUMN : Resume

Introduction

Les contributions les plus importantes aux effets in-directs des aerosols viennent des nuages stratiformesdes niveaux bas et des epaisseurs optiques moderees.Des nuages bas avec des contenus en eau liquideassez eleves pour former de la bruine dans des con-ditions non-polluees sont les plus influenables par ledeuxieme effet indirect. Ces deux conditions sontremplies pour des stratocumulus marins comme onles observe au large des cotes ouest des continents. Lacampagne de mesures ACE-2 (Second Aerosol Char-acterization Experiment) a ete dediee aux mesuresdes effets des aerosols dans l’environnement des stra-tocumulus marins. Elle a eu lieu en juin/juillet 1997entre le Portugal et les ıles Canaries. Cinq avionsont mesure in situ les proprietes des aerosols et desnuages. De plus, des mesures d’aerosols existent ala station Punto del Hidalgo sur l’ıle Tenerife. Dansle projet CLOUDYCOLUMN, des proprietes micro-physiques des nuages ont ete mesurees dans huit cas(Brenguier et al., 2000a). Deux de ces cas ont eteselectionnes pour cette etude. Le 26 juin, un cycloneau-dessus de l’Europe de l’Ouest advectait des massesd’air propres sur la region d’ACE-2. Le 9 juillet, pourle cas CLOUDYCOLUMN le plus pollue, une zonede haute pression apportait des masses d’air polluedu continent europeen vers la region de mesures.

L’etude presente dans ce chapitre a ete effectueedans le cadre du projet Europeen PACE (Parame-terization of the Aerosol indirect Climatic Effects).Des versions unicolonnes de six differents MCGsont ete utilisees pour etudier les comportements desdifferentes parametrisations dans les deux cas d’ACE-2. Les resultats de cette etude ont ete publies dansMenon et al. (2003). L’evaluation du premier effet

indirect est relativement simple, c’est principalementl’influence de la concentration en aerosols sur la con-centration en nombre des gouttelettes qui doit etreconsideree et ce sont tous les deux des quantites ob-servees. L’evaluation du deuxieme effet indirect estbeaucoup plus compliquee. Le deuxieme effet indi-rect modifie, via des processus microphysiques de laformation de la precipitation, le contenu en eau desnuages, leur duree de vie, et le taux de precipitation.Pour une evaluation parfaite la connaissance desdistributions de toutes ces quantites, qui sont tresvariables temporellement et spatialement, seraitdonc necessaire. Ici, on se focalise sur l’evaluationet comparaison de differentes parametrisations del’autoconversion, le processus le plus important pourla formation de la precipitation dans les modeles degrande echelle.

Donnees et methode de simulations

Les proprietes des aerosols ont ete mesurees a la sta-tion Punto del Hidalgo et avec des avions (Guibert etal., 2003). En selectionnant les cas adiabatiques eten excluant les cas dans lesquels de la bruine a eteobservee, Pawlowska et Brenguier (2003) ont etablita partir des vols d’avions un jeu de donnees valablepour une region de 60 × 60 km2. De plus, Brenguieret al. (2003) deduisent un jeu de donnees pour cinqsous-couches du nuage, pour le rayon effectif, la con-centration en nombre des gouttelettes, et le contenuen eau liquide. Des mesures par teledetection del’albedo et de l’epaisseur optique des nuages viennentcompleter le tableau (Schuller et al., 2003).

Les six MCGs dont les versions unicolonnes ont eteutilisees etaient ceux du CSIRO (Rotstayn, 1997 ;Rotstayn, 2000), du GISS (Del Genio et al., 1996 ;Menon et al., 2002), du Met Office (MetO ; Wilson etBallard, 1999 ; Jones et al., 2001), du PNNL (Ghanet al., 1997a), ECHAM (Lohmann et Roeckner, 1996; Lohmann et al., 1999) et LMDZ (Li, 1999 ; Boucheret al., 1995).

Deux types de simulations ont ete effectues. Dessimulations d’une journee ont ete faites pour les

61

4.6. Resume 4. Evaluation of cloud-aerosol parameterizations with ACE-2

deux jours des 26 juin et 9 juillet 1997. Les con-ditions initiales etaient prescrites par des donnees duCEPMMT, et a l’exception de MetO et PNNL, lesmodeles unicolonnes etaient forces par des donneesdu CEPMMT (CSIRO et GISS) ou a partir des sim-ulations 3D guidees avec des donnees du CEPMMT(ECHAM et LMDZ). Pour LMDZ, la simulation 3Dqui a servi pour initialiser et forcer le modele 1D aete effectuee en mode zoome sur la region d’ACE-2. Un deuxieme type de simulations consistait ensimulations d’un seul pas de temps du modele pourevaluer certaines parametrisations individuellement.Les modeles ont ete initialises avec des profils detemperature et d’humidite du CEPMMT pour le 26juin et le 9 juillet a 12h GMT. Dans l’experiencenommee � EXP-N� , la concentration en nombredes gouttelettes est calculee a partir des aerosolsdonnes. Dans � EXP-A� , la concentration desgouttelettes, le contenu en eau liquide, et la geometriedu nuage etaient prescrits. La parametrisation desproprietes optiques des nuages, du rayonnement et dela precipitation pouvaient ainsi etre evaluees. Danscette experience, la resolution verticale a ete aug-mentee pour resoudre les observations donnees parBrenguier et al. (2003). � VERT� consistait en lameme etude, mais en utilisant la resolution verticalestandard du modele.

Resultats

Les resultats des six modeles varient sous plusieursaspects (voir figs. 4.1 a 4.6). Trois des modeles(CSIRO, PNNL, LMDZ) simulent un nuage en con-tinu, comme c’est le cas dans les observations. Dansle cas non-pollue (26 juin), MetO et ECHAM simu-lent un nuage egalement persistent. La nebulositemoyenne journaliere varie entre moins de 50% et plusde 80% le 26 juin, et encore plus fortement, entre10% et plus de 80%, le 9 juillet, LMDZ montrant desvaleurs plutot elevees dans les deux cas. Des donneessatellitaires ISCCP donnent des valeurs de 80% et82% pour les cas propre et pollue. En comparantles deux situations, les modeles CSIRO, PNNL etLMDZ simulent des cycles diurnes des nuages assezsimilaires pour les deux cas, le nuage etant place unpeu plus bas dans le cas pollue dans le PNNL. Pourle 26 juin, le cycle diurne simule par LMDZ est sim-

ilaire a celui observe par ISCCP, alors que pour le9 juillet, les minima pendant la nuit et le matin nesont pas simules. Dans le cas propre, dans les deuxmodeles qui sont forces par des profils obtenus pardes simulations 3D guidees (ECHAM et LMD), lenuage persistent qui a ete observe est mieux simuleque par les autres modeles. Les deux modeles quiont une resolution verticale plus elevee que les autres(MetO par une meilleure resolution, et GISS parune parametrisation de la fraction nuageuse sous-maille verticale) simulent les fractions nuageuses lesplus petites. Egalement dans ces deux modeles, lesbases des nuages les plus hautes sont simulees, enmeilleur accord avec les observations, et l’epaisseurgeometrique des nuages est plus petite que dans lesautres modeles, ce qui est aussi plus proche des obser-vations. Il est remarquable que le schema simple deparametrisation sous-maille du modele GISS donnedes resultats similaires a ceux du modele MetO quiutilise une resolution plus fine (avec 13 au lieu de 8couches dans la couche limite).

Le taux de precipitation a la surface est fortementsous-estime par tous les modeles. Les deux modelesforces par un champ 3D d’une simulation (ECHAMet LMD) sont pourtant meilleurs que les autres.

Tous les modeles simulent plus de gouttelettes dansle cas pollue, et des gouttelettes plus petites dans lecas propre. Alors que PNNL et ECHAM utilisent desparametrisations physiques pour le lien entre aerosolset nombre des gouttelettes, les autres appliquent desformules empiriques. Le modele PNNL simule desconcentrations trop larges dans les deux cas, et leECHAM des valeurs trop petites dans le cas pollue.Les schemas empiriques sont relativement proches lesuns des autres. Les rayons effectifs des gouttelettessont plutot surestimes par les modeles. Plusieursmodeles sont pourtant bien capables de simuler desvaleurs aux incertitudes des observations pres.

Dans les simulations a pas de temps uniques(tableaux 4.4 a 4.8), tous les modeles sont bien capa-bles de simuler les concentrations de gouttelettes pluselevees dans le cas pollue en comparaison au cas pro-pre. Pourtant, on ne peut pas distinguer clairement si

62

4. Evaluation of cloud-aerosol parameterizations with ACE-2 4.6. Resume

ce sont les parametrisations physiques ou empiriquesqui sont les plus realistes. Alors que ECHAM sures-time la concentration dans le cas propre, il simuleexactement la meme concentration qu’observee dansl’autre cas. PNNL, de l’autre cote, donne la concen-tration exacte dans le cas propre et sous-estime large-ment la concentration dans le cas pollue. Parmi lesparametrisations empiriques, le GISS a ete developpepartiellement a partir des donnees d’ACE-2. Pour-tant, il n’est pas capable de reproduire les con-centrations observees dans les deux cas. Pour leLMD, plusieurs variations des formules de Boucheret Lohmann (1995) ont ete testees, dont leurs for-mules � A� et � D� , en utilisant la masse dessulfates (comme propose par les auteurs), puis lemaximum des masses des differents types d’aerosolshydrophiles (sulfates, sels de mer submicroniques,aerosols carbones hydrophiles), et enfin la somme deces masses. Aucune de ces parametrisations ne simulela variation de la concentration des gouttelettes entreles deux cas comme observee. Pour le cas propre, laconcentration est surestimee par presque toutes lesrealisations, et quand le resultat est proche des obser-vations (avec les sulfates seulement dans la formule� A� et maximum, formule � A� ), la concen-tration est largement sous-estimee dans le cas pollue.

Quand la concentration est bien simulee dans le caspollue (sulfates, formule � D� , maximum, formule� D� , somme, formule � A� ), elle est surestimeedans le cas propre. Donc, une autre parametrisationqui serait plus sensible a la variation de la concentra-tion des aerosols serait necessaire.

Dans les etudes EXP-N, quelques particularitesdans les differents parametrisations pourraient etreevaluees. Plusieurs modeles utilisent des facteurspour ajuster leurs epaisseurs optiques. Ces modelesdonnent des valeurs plus petites. Le modele LMDZ,qui n’inclut pas un tel facteur, simule plutot desvaleurs elevees, ce qui est plus realiste dans certainscas. Quand on utilise la resolution plus grossiere dumodele (etude � VERT� ), tous les modeles simu-lent des epaisseurs optiques et des albedos plus larges,ce qui est moins realiste. La variation de l’albedodes nuages entre les cas propre et pollue est plusgrand dans les observations que dans les modeles.Le taux de precipitation est fortement sous-estimede plusieurs ordres de grandeur par tous les modeles.Evidemment, ils ne sont pas capables de simuler labruine observee dans le cas des contenus en eau liq-uide relativement faible.

63

4.6. Resume 4. Evaluation of cloud-aerosol parameterizations with ACE-2

64

5. Aerosol indirect effects in POLDER satellite

data and in the LMDZ GCM

5.1. Introduction

To understand the aerosol-cloud interactions on aglobal scale, the most promising approach is thecombination of general circulation models (GCMs)and satellite measurements. GCMs include the treat-ment of a multitude of dynamical and physical pro-cesses and are able to simulate interactions betweenmany climate parameters. Comparing the results ofdifferent simulations is a powerful method to under-stand climate processes. In contrast to observations,the impacts of single processes can be isolated. Adisadvantage of GCMs is their coarse resolution.Observations with spaceborne instruments are theonly measurements which cover the whole globe onlong timescales. Satellite observations can there-fore be used to evaluate GCM simulations. Currentsatellite instruments generally measure top-of-the-atmosphere radiances in the shortwave and long-wave spectrum, thus providing a two-dimensional(latitude-longitude) view of cloud properties but withlimited information on the vertical distribution.

Using AVHRR satellite observations, Wetzel andStowe (1999) showed an inverse relationship betweenzonal and seasonal averages of aerosol optical depthand droplet size of stratus clouds over oceans. Theyalso found an increase in seasonal and zonal meancloud optical thicknesses with increasing aerosol opti-cal thickness over oceans. Schwartz et al. (2002) useda combination of a hemispheric model and AVHRRsatellite data to study the correlations between mod-eled aerosol data and satellite-derived cloud proper-ties for a period of one week in April 1987 over theNorth Atlantic Ocean. They found a negative correla-

tion between simulated sulfate aerosol concentrationand observed cloud droplet radius, but no correla-tion between simulated sulfate and observed cloudoptical depth. Again from AVHRR data, Nakajimaet al. (2001) derived a negative correlation betweencloud droplet effective radius and aerosol numberconcentration over oceans. However, their results donot show a correlation between aerosol concentra-tion and cloud optical thickness or cloud liquid waterpath.

Breon et al. (2002) used POLDER data to establisha relationship between quasi co-located aerosol prop-erties and cloud droplet effective radii (CDR). Theyput in evidence the first aerosol indirect effect byidentifying a decrease in cloud droplet effective radiiwith increasing aerosol concentrations. Lohmannand Lesins (2003a) compared these data with resultsfrom simulations with the ECHAM GCM to investi-gate the relative importance of the first and secondAIE in their model. They found a too steep decreasein effective radius with increasing aerosol burden ifthe second indirect effect was excluded. Lohmannand Lesins (2003b) extended this study and furtherfound an increase in LWP with increasing aerosol in-dex in their model. They showed that, independentlyof the LWP, the CDR to aerosol index relationship isdifferent in maritime compared to continental condi-tions.

In contrast to previous studies (with the excep-tion of Breon et al., 2002), we do not average cloudand aerosol parameters over large regions and longtimescales; rather we seek relationships between pa-rameters derived from co-located (within a few degree

65

5.2. Observational data from POLDER 5. Aerosol indirect effects in POLDER and GCM

resolution) and simultaneous measurements. A sec-ond difference from former studies is that we try todistinguish the aerosol impact on clouds for differentcloud liquid water paths. Han et al. [1998, 2002]showed different behaviors of the relationship be-tween cloud droplet column concentration and cloudoptical thickness, τc, for small (τ < 15) and large(τ > 15) cloud optical thicknesses.

In this study, we exploit satellite data from thePolarization and Directionality of the Earth’s Re-flectances (POLDER) instrument and the Labora-toire de Meteorologie Dynamique (LMD-Z) GCM.We establish statistical relationships in both satelliteobservations and results from model simulations tounderstand both the first and second AIE and toisolate the contribution of anthropogenic aerosols.

5.2. Observational data from

POLDER

The POLDER-1 spaceborne radiometer on board theJapanese ADEOS satellite was able to measure vari-ous aerosol and cloud properties (Deuze et al., 1999;Buriez et al., 1997) for the period from November1996 until June 1997. Additionally Breon and Goloub(1998) showed that in some cases multi-directionalpolarized radiance measurements can be inverted toestimate the CDR at cloud top. Their method unam-biguously derives the size of spherical water droplets.This is an advantage compared to other methods toderive CDR which could be contaminated by aerosollayers above clouds (Haywood and Osborne, 2003).This retrieval was performed by Breon and Colzy(2000) who derived cloud top droplet effective radiifor the whole POLDER-1 period in cases of horizon-tally homogeneous liquid clouds on scales of at least150x150 km2.

We use the POLDER aerosol index (AI) derivedover land and ocean (Deuze et al., 1999). The AIis estimated as the product of the aerosol opticaldepth, τa, and the Angstrom coefficient, α, which isa measure of the particle size. It is therefore notedατ and may be interpreted as the load of submicronicaerosols. Recent studies have shown a high correla-

tion between the AI and the column-integrated CCNconcentration (Nakajima et al., 2001). Regardingcloud properties, we use the cloud optical thickness,τc, derived over land and ocean (Buriez et al., 1997)and the CDR described above.

We interpolate all observational data to the coarserGCM grid. In contrast to Breon et al. (2002), whereback-trajectories were used to relate a particularmeasurement of CDR to a clear-sky (upwind) mea-surement of aerosols, we correlate here the aerosoland cloud parameters in the same grid-box. Thisshould be adequate, as the model grid-boxes aremuch larger than the POLDER pixels.

As there is no direct derivation of cloud liquid waterpath (LWP) from POLDER measurements, we esti-mate it from the cloud optical thickness, τc, and theCDR, re, by inverting the equation for τc (Stephens,1978):

τc =32

LWPre ρw

(5.1)

where ρw is the density of liquid water. Thus, we geta measurement of LWP only over grid-boxes wherea measurement of the CDR exists, i.e., for homoge-neous liquid clouds. Note that POLDER estimatesthe cloud optical thickness from the reflectance mea-surements using a radiative transfer model.

5.3. Model description and

derivation of a “satellite-like”

model output

We use the Laboratoire de Meteorologie Dynamique(LMD-Z) GCM (Li , 1999). Condensation of watervapor is calculated in the model using a “hat” prob-ability density function for total water content to al-low for fractional cloudiness (Le Treut and Li , 1991).We apply the microphysical scheme of Boucher et al.(1995a), which includes autoconversion and accretionfor liquid water clouds and a fall-velocity dependentsnow-formation equation for ice clouds. The CDNC(in cm−3) is diagnosed from aerosol mass concentra-tion, ma (in µg m−3), using the empirical formula of

66

5. Aerosol indirect effects in POLDER and GCM 5.3. Model description

Boucher and Lohmann (1995) (formula “D”):

Nd = 10a0+a1 log (ma) (5.2)

where the empirical constants are a0 = 2.21 anda1 = 0.41. The LMDZ model includes a compre-hensive on-line sulfur cycle model (Boucher et al.,2002) as well as atmospheric cycles of sea-salt, or-ganic matter, black carbon, and dust aerosols (Reddyand Boucher , 2003). In contrast to Boucher andLohmann (1995) where the mass of sulfate aerosolswas used as a surrogate for all aerosols, we use hereinstead the maximum of the masses of the three hy-drophilic aerosol species considered in the model (sul-fate, hydrophilic organic matter, and submicronic seasalt). This modification has an influence on cloudproperties in biomass burning regions but a verysmall one in the northern hemisphere, where sulfateconcentrations dominate. The volume-mean clouddroplet radius for liquid water clouds is calculatedassuming spherical particles:

rd = 3

√ql ρair

43 π ρw Nd

(5.3)

where ql is the cloud liquid water mixing ratio, ρair

the air density, ρw the density of liquid water, andNd the CDNC. The volume-mean cloud droplet ra-dius is related to the cloud droplet effective radiusin our model as re = 1.1 rd. While the first AIEcauses the CDR to decrease, the second AIE resultsin an increase in cloud liquid water content, ql. Botheffects cause an increase in cloud optical thickness,which is parameterized as in Eq. 5.1.

We simulate the whole period of the POLDER-1measurements (November 1, 1996 - June 29, 1997),starting two months before to allow for a spin-upof the aerosol concentrations. The SST and sea-iceare imposed using the SST dataset of Reynolds andSmith (1995) and the HADISST1.1 sea-ice data of theHadley Centre (Rayner et al., 2003). We nudge themodel horizontal winds and temperature to ECMWFreanalysis data using relaxation times of 0.1 day for

Figure 5.1.: Comparison of the LWP [g m-2] calcu-lated from cloud optical thickness andcloud droplet effective radius (y-axis)to the LWP directly calculated in themodel. The correlation coefficient is 0.98and the slope of the regression line is1.45.

the winds and 1 day for the temperature in order toget realistic meteorological conditions. We simulateon-line in the model the POLDER swath to sampleaerosols and clouds properties in the model in thesame way the satellite does it. Therefore we do notexpect any bias due to differences in sampling thediurnal cycle in the model and satellite data.

Cloud top quantities are estimated as seen by thesatellite using the random cloud overlap assumption.By doing so we account for the contribution of eachlayer to the two-dimensional distribution seen at eachgridpoint and time-step from above. For the effec-tive radius only liquid water clouds which are notcovered by clouds in layers above are considered. Tomatch the POLDER criterion of horizontally homo-geneous clouds at the 150x150 km2 resolution, weonly perform the estimation when at least a quar-ter of the grid-box is covered by a liquid cloud. Asa model grid-box is 3.75 × 2.5◦ large, this matchesapproximately the resolution of the satellite data atleast near the Equator. The optical thickness of liq-uid clouds is calculated using the ISCCP simulator(Webb et al., 2001).

67

5.4. Cloud droplet radius - aerosol index relationship 5. Aerosol indirect effects in POLDER and GCM

To be consistent with the derivation of LWP fromthe POLDER data, we also invert the LWP in themodel from Eq. 5.1 using the above-derived opti-cal thickness and cloud-top droplet radius of liquidclouds. This way of calculating the LWP is expectedto give an upper bound to the LWP, as τc is an inte-grated value while re is the cloud-top droplet effectiveradius. As in general re increases with height in acloud, the cloud-top value is expected to be largest.For the model output, the LWP calculated this wayis compared to the LWP directly computed from theliquid water content, which is also sampled along thePOLDER swath (Fig. 5.1). Despite the expectedfact that the LWP calculated from τc and re giveslarger values than the directly computed LWP, itis remarkable that there is a very good correlationbetween the two LWPs (with a correlation coefficientof 0.98). We can therefore use this retrieved LWP tocompare with POLDER-estimated LWP and performstatistics at constant LWP (as done in the followingsection).

5.4. Cloud droplet radius - aerosol

index relationship

Fig. 5.2 shows the cloud droplet effective radius(CDR) to aerosol index (AI) relationships. The re-lationship is separated for continental and maritimeconditions. For the POLDER data, a negative cor-relation is found over oceans. Over land, a negativecorrelation is found as well, but it is much less pro-nounced. For relatively large aerosol indices, almostno correlation between aerosol burden and CDR isobserved. The slope of the relationship is calculatedfrom the linear regression (aerosol indices from 0.0125to 0.40) between the logarithms of re and ατ :

s =∂ log re

∂ log ατ(5.4)

We find s=–0.042 for maritime and s=–0.012 for con-tinental conditions. Over oceans, Breon et al. (2002)give a slope of s=–0.085, which is twice steeper thanthe one found in our study.

0.01 0.1ατaer

5

6

7

8

9

10

11

12

13

14

15

r e [µm

]

Land. slope = -0.012Ocean. slope = -0.042

(a) POLDER, all LWP(a) POLDER, all LWP

0.01 0.1ατaer

5

6

7

8

9

10

11

12

13

14

15

r e [µm

]

Land. slope = -0.092Ocean. slope = -0.097

(b) LMDZ, all LWP(b) LMDZ, all LWP

Figure 5.2.: Relationships between cloud dropleteffective radius [µm] and aerosol in-dex for the December 1996 to June1997 period (for all LWP conditions)for POLDER observations (upper panel)and for LMDZ model outputs (lowerpanel). Continental (triangles, red) andmaritime (circles, blue) conditions areseparated. The analysis is restricted tothe region 40◦S–60◦ N, as POLDER ob-servations of cloud optical thickness andthus LWP are not reliable at higher lat-itudes due to an influence of snow cov-ered surfaces. Error bars show the meandeviation within each bin of aerosol in-dex. Dotted lines indicate the numberof points considered in each bin (rightscale). The graphs are slightly shifted tothe left for continental conditions and tothe right for maritime conditions for thesake of better legibility.

This is due primarily to the averaging over a GCMgridbox, where different cloud and aerosol situationsmay be taken into account. Note that over landthe aerosol index retrieval is not as reliable as overoceans.

68

5. Aerosol indirect effects in POLDER and GCM 5.4. CDR to ατ relationship

0 50 100 150 200LWP [gm-2]

0

1

2

3

4

log 10

(N)

LMDZ, oceanLMDZ, landPOLDER, oceanPOLDER, land

Figure 5.3.: Histogram of the LWP in model andobservations. Vertical lines indicate theborders of the LWP bins.

In our model, a clear relationship exists, and the slopeis s=–0.092 and –0.097 for continental and maritimeconditions, respectively. A problem might arise be-cause although we limit the analysis to homogeneousliquid water clouds, all LWP situations are taken intoaccount. A correlation between CDR and AI mighttherefore be fortuitous if large aerosol burdens andlow liquid water paths coincided because sources offine-mode aerosols are located over land where the airis on average dryer. Strictly speaking the first AIEis defined as the CDR to AI correlation at constantcloud liquid water content (or equivalently constantLWP conditions). In order to examine the first AIEin this way, we classify all situations in model re-sults and observations into 20 LWP bins, with LWPsranging between 0 and 200 gm−2, and the numberof observations equally distributed between the bins.Fig. 5.3 shows a histogram of the LWP in model andobservations. The size of the LWP bins used is alsoindicated on the figure. The CDR to AI relation-ship is established for each LWP bin and is shownon Fig. 5.4 for the POLDER data and the model re-sults. In the observational data, the slopes becomeflatter with increasing LWP. Over land, s even be-comes slightly positive for very large LWP values. Areason for the decrease in the negative slope of theCDR to AI relationship may be that for thick cloudsthe size of the droplets at cloud top may be controlledby the available water rather than by the aerosol con-centration.

0 50 100 150LWP [g m-2]

-0.15

-0.1

-0.05

0

slop

e d

log(

r e)/d lo

g(ατ

)

LandOcean

(a) POLDER(a) POLDER(a) POLDER(a) POLDER(a) POLDER

0 50 100 150LWP [g m-2]

-0.15

-0.1

-0.05

0

slop

e d

log(

r e)/d lo

g(ατ

)

LandOcean

(b) LMDZ(b) LMDZ(b) LMDZ(b) LMDZ(b) LMDZ

Figure 5.4.: Slope of the CDR to AI relationship,s, as a function of LWP, for a) thePOLDER observations and b) the LMDZmodel. Ocean and land regions areshown in blue (circles) and in red (tri-angles), respectively.

For the model there is no clear variation of the slopeas a function of LWP. For small LWP (up to 60 gm−2)over the ocean the slope becomes less negative. ForLWPs between 60 and 100 gm−2 the slopes becomesteeper with increasing LWP and approach a ratherconstant value of –0.13 for LWP larger than 100gm−2. Over land a flattening of the slope is foundfor LWPs larger than 50 gm−2 and s approaches avalue similar to that found over the ocean at verylarge LWPs.

In Fig. 5.5, we compare the slopes simulated bythe model with the observed slopes for different situ-ations. We calculate the slopes for the 20 LWP bins(for the whole period and for the whole globe) andthe slopes for 24 regions of 15◦x7.5◦ on the globe (forthe whole period and for all LWP). In general, themodel simulates too negative slopes for the CDR toAI relationship.

69

5.5. Cloud liquid water path - aerosol index relationship 5. Aerosol indirect effects in POLDER and GCM

-0.2 -0.15 -0.1 -0.05 0 0.05slope d log(re)/d log(ατ) (POLDER)

-0.2

-0.15

-0.1

-0.05

0

0.05

slop

e d

log(

r e)/d lo

g(ατ

) (LM

DZ)

LandOcean

(a) LWP bins (a) LWP bins (a) LWP bins (a) LWP bins

-0.2 -0.15 -0.1 -0.05 0 0.05slope d log(re)/d log(ατ) (POLDER)

-0.2

-0.15

-0.1

-0.05

0

0.05

slop

e d

log(

r e)/d lo

g(ατ

) (LM

DZ)

LandOcean

(b) Regions (b) Regions (b) Regions (b) Regions

Figure 5.5.: Scatter plot of the slope s in the modeland observations for a) the 20 LWP binsand b) the twenty-four 15◦x7.5◦ regionsin the globe. Blue circles and red trian-gles are for ocean and land regions, re-spectively.

The model always shows a negative correlation be-tween CDR and AI, while in the POLDER obser-vations, in particular over land, positive correlationsmay occur. There is a considerable scatter in Fig. 5.5,showing that the regional skills of the model are stilllimited.

The slopes are in a range of [–0.15,–0.02] for themodel and [–0.08,+0.01] for the observations. Ex-cept for the positive values observed in POLDERdata for continental conditions, our slopes are com-parable to those given by Feingold et al. (2003), whoused ground-based remote sensing measurements atdifferent sites. In their study, values of [–0.16,–0.02]were observed. Nakajima et al. (2001) found a slopeof –0.17 over oceans on a global scale, which is muchlarger than the mean values of –0.10 and –0.04 thatwe derived for the LMDZ model and POLDER data,respectively.

It is interesting to note that for the model theslope values s are of the order of one third (i.e.,∂ log re/∂ log Nd) of the slope value implied by theconstant a1 = 0.41 in Eq. 5.2 between log Nd andlog ma. This is because of the general linear relation-ship between ma and ατ in the model. A similarrelationship has been proposed by Nakajima et al.(2001) between the aerosol column number concen-tration, Na, and ατ .

0.01 0.1ατaer

0

20

40

60

80

100

LWP

[g m

-2]

Land. slope = 0.01Ocean. slope = 0.1

(a) POLDERGlobe (40oS-60oN)(a) POLDERGlobe (40oS-60oN)

0.01 0.1ατaer

0

20

40

60

80

100

LWP

[g m

-2]

Land. slope = 0.03Ocean. slope = 0.01

(b) LMDZGlobe (40oS-60oN)(b) LMDZGlobe (40oS-60oN)

Figure 5.6.: Cloud liquid water path to aerosol indexrelationships over ocean (blue, circles)and land (red, triangles) for a) POLDERobservations and b) LMDZ model cal-culations. The region is restricted to40◦S–60◦N, and very large LWP values(W >150 gm−2) are excluded. The errorbars indicate the mean absolute devia-tion within each bin.

5.5. Cloud liquid water path -

aerosol index relationship

Examining the LWP to aerosol index relationshipgives us some insight in the behavior of the secondindirect effect. If the second aerosol indirect effect isreal, larger LWP would be expected in regions withlarge aerosol concentrations. Although a correlationis not a proof of the second aerosol indirect effect,one would rather expect a negative correlation in theabsence of second AIE because large AI in dry conti-nental airmasses and low AI in humid maritime air-masses may be associated with thin and thick clouds,respectively.

70

5. Aerosol indirect effects in POLDER and GCM 5.6. Indirect effects of anthropogenic aerosols

0.01 0.1ατaer

0

20

40

60

80

100

LWP

[g m

-2]

Land. slope = 0.1Ocean. slope = 0.1

(a) POLDERNH (30oN-60oN)(a) POLDERNH (30oN-60oN)

0.01 0.1ατaer

0

20

40

60

80

100

LWP

[g m

-2]

Land. slope = 0.14Ocean. slope = 0.1

(b) LMDZNH (30oN-60oN)(b) LMDZNH (30oN-60oN)

Figure 5.7.: As Fig. 5.6 but for midlatitudes of thenorthern hemisphere.

In Figs. 5.6 and 5.7, we plot the dependency of cloudliquid water path on aerosol index. For both theobservations and the model, LWP is calculated byinverting Eq. 5.1, considering large-scale liquid wa-ter clouds only, for which a cloud droplet effectiveradius is defined. LWP values larger than 150 gm−2

are excluded from the analysis, because POLDERmeasurements of cloud optical thicknesses resultingin such large LWPs may be erroneous due to con-tamination by snow-covered surfaces (the LWP to AIrelationship including all LWPs shows a very strongincrease in LWP with increasing AI). While Fig. 5.6shows the LWP to AI relationship for 40◦S–60◦N re-gion, we restrict in Fig. 5.7 the analysis to northernhemisphere (NH) mid-latitudes (30◦–60◦N) where theAIE are expected to be very strong. Both the obser-vations and the model show an increase in LWP as afunction of aerosol index which is statistically signifi-cant at very high confidence levels (>95%) accordingto the Kendall rank correlation test (Conover , 1980).

For the 40◦S-60◦N region in POLDER observations,the relationship is much less steep for continentalcompared to maritime conditions for AI values largerthan 0.1. Although an investigation of this process isbeyond the scope of this study, one could argue thata semi-direct aerosol effect may play a role, which re-

duces cloudiness in regions with large concentrationsof absorbing aerosols (Ackerman et al., 2000a). Afurther analysis shows that the LWP to AI slope isthe smallest in the region 0–30◦N. Other influencesmay play a role as well. For example (Feingold ,2003) shows that the relationship between CDR andaerosol extinction becomes flatter with decreasingupdraft velocity for large aerosol extinctions. It isalso interesting to note, that in the POLDER data,and in some cases in the model simulations, the LWPto AI relationship shows slightly negative correlationat very small AI (for the first two ατ bins). Thiscould be due to the coincidence of low aerosol con-centrations and large LWP values, for example inremote maritime areas.

The positive correlation is even more pronouncedwhen looking at NH mid-latitudes only (Fig. 5.7).The slopes of the LWP to AI relationship in modeland observations are of the same magnitude. How-ever the positive slope occurs for ατ > 0.1 in theobservations whereas it is observed throughout therange of AI in the model.

5.6. Indirect effects of

anthropogenic aerosols

We performed two more experiments to further testthe influence of anthropogenic aerosols. The con-trol simulation is referred to as PD/PD (“present-day/present-day”), indicating that the present-dayaerosol concentrations are taken in both the radiationand precipitation schemes. Anthropogenic aerosolsact on the first and second AIE through these twoparameterizations, respectively. In the PD/PI ex-periment anthropogenic aerosols influence only theradiation parameterization while a pre-calculatedmonthly mean pre-industrial aerosol distribution isused for the precipitation scheme. In the PI/PI ex-periment anthropogenic sources were switched off inthe aerosol model and “pre-industrial” aerosol contri-butions are used for the first and second AIE. Fig. 5.8compares the PD/PD and PD/PI experiments.

71

5.7. Summary and conclusions 5. Aerosol indirect effects in POLDER and GCM

(a)

0.01 0.1ατaer

4

5

6

7

8

9

10

r e [µm

]

PD/PD. slope = -0.111PD/PI. slope = -0.124PI/PI. slope = -0.1

100

101

103

102

104

105

Num

ber o

f poi

nts

(a)(a)(a)(a)(a)(a)

(b)

0.01 0.1ατaer

0

20

40

60

80

100

120

LWP

[g m

-2]

PD/PD. slope = 0.01PD/PI. slope = -0.0

101

103

102

104

105

Num

ber o

f poi

nts

(b)(b)(b)(b)

Figure 5.8.: (a) Relationships between CDR and AIfor the PD/PD control (black), PD/PI(red), and PI/PI (blue) experiments. (b)Relationships between LWP and AI forthe PD/PD (black) and PD/PI (red) ex-periments. The error bars indicate themean absolute deviation within each binand the dotted lines the number of pointswithin each AI bin (right scale). No con-straint on LWP is applied.

The impact of the second AIE on the CDR to ατ re-lationship is relatively small – the slope is steepenedonly from –0.11 to –0.12. As shown in the previoussection, LWP increase with increasing ατ is largestfor ατ values larger than 0.1. Accordingly, we ob-serve an impact of the second AIE on the CDR toατ relationship only at very large ατ values. Thisfinding is in contrast to the conclusion of Lohmannand Lesins (2003a) who found that the slope of theCDR to ατ relationship in their model is stronglyreduced by the impact of the second indirect effect.However they calculate the impact of the second AIEin a different manner, which prevents a more in-depthanalysis of the differences between the two models.

Fig. 5.8a also shows the CDR to AI relationshipin the PI/PI simulation using the pre-industrial AI

value on the x-axis. The PI/PI curve shows thatone should also expect a correlation between CDRand AI in pre-industrial conditions. The slope of therelationship is comparable to those obtained whenusing present-day aerosol conditions. The existenceof a positive slope is therefore by itself not a proof ofan anthropogenic impact; it has to be combined byan increase in aerosol index as well.

Finally Fig. 5.8 shows the LWP to AI relationships.Here, an impact of the second aerosol indirect ef-fect caused by anthropogenic aerosols can clearlybe identified by looking at the difference betweenthe PD/PD and PD/PI curves. While there is noadditional impact of anthropogenic aerosols on thesecond AIE for small aerosol indices, there is a stronganthropogenic contribution at AI larger than 0.1.

5.7. Summary and conclusions

Using satellite-derived data from the POLDER in-strument and model simulations with the LMDZGCM, we investigated the first and second AIE. Weestablished relationships between aerosol index andcloud-top CDR for homogeneous liquid water clouds.In both observations and model, a decrease in CDRwith increasing AI is found. In the observations, thisdecrease is less pronounced at large AI values. Whenestablishing the CDR to AI relationship at fixedcloud liquid water path (LWP) a similar relationshipis found. However, with fixed LWP, the decrease inCDR is also found in the observations at large AI.This difference to the relationship in the case of allLWP conditions is an indication of the impact of thesecond AIE on the CDR to AI relationship. Theslopes of the relationship are –0.01 and –0.04 for theobservations over land and ocean, respectively, and–0.09 and –0.10 for the model over land and ocean,respectively.

The slope of the CDR to AI relationship is foundto become flatter with increasing LWP in the obser-vations, while it approaches an almost constant valueof about –0.13 for the model. In general, the modelsimulates too steep slopes compared to the POLDERdata. The slopes found for different regions, LWPs,

72

5. Aerosol indirect effects in POLDER and GCM 5.7. Summary and conclusions

and periods are, however, comparable to those givenby other authors.

To investigate the second AIE, we established a LWPto AI relationship. An increase in LWP with increas-ing AI is found for both model and observations.This increase is significant in particular for large AI(ατ > 0.1). A decrease in LWP with increasing AI isfound at very small AI, which could be attributed toa geographical coincidence of low aerosol burden andthick clouds in remote regions. When limiting the re-gion to northern hemisphere mid-latitudes, the LWPto AI relationship becomes much more pronounced.The slope of the LWP to AI relationship in the modelmatches well the observations.

In order to investigate the anthropogenic contribu-tion to the aerosol indirect effects, additional exper-iments were carried out with the model, in which apre-industrial aerosol distribution was used for thecalculation of the first and second indirect effect,respectively. We find a small impact of the anthro-

pogenic contribution to the second AIE on the CDRto AI relationship, which is restricted to large AIvalues. Anthropogenic aerosols causes the CDR todecrease by 0.5 µm on average in the CDR to AIrelationship. The slope however is hardly changed.Similarly, a small impact of anthropogenic aerosolson the LWP to AI relationship is found at large AIvalues. The comparison between model and observa-tions provides insight in some skills and deficienciesof the model. Two shortcomings have already beenwidely noticed and are common to many GCMs – theCDR simulated are in general too small compared toobservations by up to 3-4 µm and the LWP is toolarge by 10-20 gm−2. The model simulates too steepslopes for the CDR to AI relationship, in particu-lar at large AI values. In regions with large aerosolburden (the northern hemisphere mid-latitudes), aslightly too steep dependence between LWP and AIis simulated at low AI values. However, the generalshapes of the CDR to AI and the LWP to AI relation-ships are well simulated. The slopes of the LWP toAI relationships match rather well the observations.

73

5.8. Resume 5. Aerosol indirect effects in POLDER and GCM

5.8. Les effets indirects des

aerosols dans les donnees

satellitaires POLDER et dans

le modele LMDZ : Resume

Introduction

Une des approches les plus prometteuses pour lacomprehension des processus climatiques a l’echelleglobale est la combinaison de modeles a grandeechelle et d’observations satellitaires. Les MCGstraitent une multitude de processus dynamiques etphysiques, et ils peuvent simuler les interactions en-tre plusieurs parametres du climat. La compara-ison entre differentes simulations est une methodeimportante pour l’investigation des processus clima-tiques. En contraste aux observations, l’influenced’un parametre particulier peut etre isole. Les obser-vations avec les instruments spatiaux sont les seulesmesures qui couvrent tout le globe avec de longuesseries temporelles. A partir des mesures de lumi-nances, des instruments satellitaires peuvent recon-stituer des distributions a deux dimensions, qui sontvalables pour le sommet des nuages dans le cas desproprietes des nuages.

En se focalisant sur les effets des aerosols sur lesnuages, Han et al. (1994, 2002), Wetzel et Stowe(1999), Nakajima et al. (2001) et Schwartz et al.(2002) ont examine les donnees AVHRR (AdvancedVery High Resolution Radiometer) et ont montredes differences entre continent et ocean, entre lesdeux hemispheres, et des relations statistiques en-tre les aerosols et les proprietes des nuages commele rayon effectif des gouttelettes afin de mettre enevidence les effets indirects des aerosols. Breon etal. (2002) ont montre une correlation negative en-tre l’indice en aerosols et la taille des gouttelettes.Lohmann et Lesins (2002, 2003) ont utilise ces ob-servations pour analyser les effets indirects dans leurmodele de grande echelle. Ils ont trouve des relationsdifferentes au-dessus des continents et des oceans etils ont montre l’importance du deuxieme effet indirectpour reproduire la relation observee entre aerosols etrayon effectif des gouttelettes.

Methodes

Dans cette etude, on utilise les donnees satellitairesderivees de l’instrument POLDER. Pour les aerosols,l’indice en aerosols (ατ ; Deuze et al., 1999) estutilise, et pour les nuages, l’epaisseur optique (τc ;Buriez et al., 1997) et le rayon effectif (re ; Breon etColzy, 2000) sont utilises. Les donnees des observa-tions sont interpolees sur la grille plus grossiere dumodele.

Pour le lien entre aerosols de sulfates et concentra-tion en gouttelettes dans notre modele, nous utilisonsla formule � D� de Boucher et Lohmann (1995)(Eq. 5.2). Ici, on considere la masse maximale destrois especes d’aerosols hydrophiles utilises dans lemodele au lieu de la seule masse des sulfates afin demieux representer les regions de combustion de labiomasse. Le rayon effectif des gouttelettes au som-met des nuages est calcule en faisant l’hypothese derecouvrement aleatoire des nuages, et il n’est calculeque pour les nuages en eau liquide qui couvrent aumoins un quart d’une maille pour reprendre les memeconditions que dans les observations. L’epaisseur op-tique des nuages est calculee en utilisant le simula-teur ISCCP (Webb et al., 2001). Toutes les sorties dumodele sont echantillonnees en simulant la faucheedu satellite dans le modele afin de constituer unedistribution globale journaliere de la meme maniereque pour l’instrument.

Aucune mesure de la colonne d’eau liquide (liq-uid water path, LWP en anglais) n’existe pour lesdonnees POLDER. Le LWP alors est calcule a par-tir des mesures de l’epaisseur optique et du rayoneffectif des gouttelettes, ce qui donne une certainesurestimation du LWP. Pourtant, une comparaisonentre le LWP directement calcule dans le modele etle LWP calcule de cette maniere a partir des donneesdu modele montre que ces deux quantites sont tresfortement correlees et qu’une analyse des LWP ainsiderives des observations est bien raisonnable (fig.5.1).

74

5. Aerosol indirect effects in POLDER and GCM 5.8. Resume

Relation entre le rayon des gouttelettes

et l’indice des aerosols

La relation entre le rayon effectif au sommet desnuages en eau liquide (cloud droplet radius, CDRen anglais) et l’indice des aerosols (aerosol index,AI en anglais) a ete etablie et pour les observa-tions POLDER et pour le modele LMDZ, separementpour les regions continentales et maritimes (fig. 5.2).Premierement, il est important de noter que le modelesous-estime le rayon effectif d’environ 3 µm, ce quiest lie a la parametrisation de l’autoconversion dansle schema de precipitation et une erreur bien connuepour plusieurs modeles a grande echelle (Rotstayn,2000). A la fois dans le modele et dans les obser-vations, le rayon effectif est plus grand au-dessusdes oceans qu’au-dessus des continents. La pente dela relation entre rayon effectif et indice des aerosolspeut etre calculee comme s = ∂ log(re)/∂ log(ατ).Pour les observations, on obtient une pente de –0.012au-dessus des continents et de –0.042 au-dessus desoceans. Pour les conditions maritimes, Breon et al.(2002) donnent une pente de –0.085. La differenceest due au fait que nous interpolons ici les donnesa la grille plus grossiere du modele. Les sorties duLMDZ donnent des pentes de –0.092 dans les condi-tions continentales et de –0.097 dans les conditionsmaritimes. Le fait que ces pentes sont superieuresa celles des observations est d’abord du a une forteanticorrelation dans le modele pour des indices enaerosols eleves, ce qui n’est pas le cas dans les obser-vations.

Le premier effet indirect des aerosols a ete definipar Twomey (1974) a LWP constant, ce qui exclutun impact du deuxieme effet indirect sur la relationCDR-AI. De plus, une correlation entre ces deuxquantites peut etre fortuite, si par exemple des con-centrations en aerosols faibles se trouvent aux memesendroits que des nuages avec des contenus en eau liq-uide eleves comme dans des regions eloignees desoceans. Pour examiner le premier effet indirect aLWP constants, nous classons les situations en 20classes de LWP entre 0 et 200 gm−2 avec un nombrede cas distribue uniformement entre les classes (fig.5.3). La relation CDR-AI est etablie dans chacunedes classes. Dans les observations, la relation de-

vient plus raide, et en particulier pour des AI eleves,l’anticorrelation est plus marquee. L’impact de fixerle LWP est moins clair pour le modele. La relationdevient plus raide au-dessus des continents, mais unpeu plus plate au-dessus des oceans. Le choix dunombre des classes n’a pas d’influence importantesur ces resultats. L’analyse de la pente des relationsen fonction du LWP montre que les relations CDR-AI deviennent plus plates pour de grands AI dansles observations (fig. 5.4). Au-dessus des continents,les pentes peuvent meme devenir positives. Pour lemodele, un tel comportement ne peut pas etre trouve.

En comparant les pentes simulees par le modeledans differentes regions du globe, et pour differentsLWP a celles donnees par les observations dans lesmemes conditions, on observe que le modele simule engeneral des pentes trop fortes, et toujours negatives,alors que dans les observations, on trouve parfois despentes positives (fig. 5.5).

Les valeurs pour les pentes trouvees dans le modeleet les observations sont respectivement dans les in-tervalles de [–0.15,–0.02] et de [–0.08,+0.01]. Al’exception des valeurs positives dans les donneesde POLDER dans les conditions continentales, cespentes sont comparables a celles trouvees par Fein-gold et al. (2003) qui utilisaient des observations deteledetection a partir du sol a differentes stations ettrouvaient des valeurs de [–0.16,–0.02]. Au-dessusdes oceans en moyenne annuelle globale, Nakajimaet al. (2001) donnent une valeur de –0.17, ce qui estgrand compare aux pentes trouvees ici.

Pour le modele, les valeurs trouvees sont prochesd’une valeur de 1/3 (= ∂ log re/∂ log Nd) de la con-stante a1=0.41 de l’equation 5.2 qui reflete la relationentre la concentration en nombre des gouttelettes etma.

Relation entre la colonne d’eau liquide et

l’indice des aerosols

Une correlation positive entre la colonne d’eau liquideet la concentration en aerosols pourrait etre une in-

75

5.8. Resume 5. Aerosol indirect effects in POLDER and GCM

dication du deuxieme effet indirect. Nous regardonsles relations LWP-AI pour tout le globe et pour laregion des moyennes latitudes de l’hemisphere nord(figs. 5.6 et 5.7). A la fois dans les observations etdans le modele, on trouve une correlation positiveentre le LWP et l’indice en aerosols qui est statis-tiquement fiable a des niveaux tres eleves suivantle test de correlation de Kendall (Conover, 1980).Pourtant, cette correlation ne se trouve dans les ob-servations que pour des AI relativement eleves quandon regarde tout le globe, alors que dans le modele,on le trouve aussi pour des petits indices d’aerosols.Quand on se focalise sur les moyennes latitudes del’hemisphere nord, la correlation positive est plusprononcee. Les pentes des relations sont bien simi-laires dans les observations et dans le modele.

Nous avons fait deux simulations de plus avec lemodele pour etudier les impacts des aerosols an-thropiques. Les aerosols peuvent exercer le premiereffet indirect dans le modele via la parametrisation dutransfert radiatif, et le deuxieme effet indirect via laparametrisation de la precipitation. Dans la premiere

des simulations additionnelles, nommee PD/PI, desconcentrations actuelles (� present-day� , PD enanglais) ont ete utilisees dans le schema du rayon-nement et des concentrations pre-calculees pour lasituation pre-industielle (PI) dans la parametrisationde la precipitation. Dans la deuxieme simulation,nommee PI/PI des concentrations pre-industriellesont ete utilisees a la fois dans le rayonnement etdans la precipitation. La simulation de controle,ou les aerosols actuels sont utilises dans les deuxparametrisations est nommee PD/PD. La relationCDR-AI ne change pas beaucoup entre les differentessimulations, parce que la physique derriere la relationne change pas (fig. 5.8). Pourtant, on peut identifierque la relation est legerement plus plate pour desindices en aerosols eleves dans le cas ou le deuxiemeeffet indirect des aerosols anthropiques est inclus(PD/PD) compare au cas ou il ne l’est pas (PD/PI),ce qui montre une (faible) influence du deuxieme effetindirect des aerosols anthropiques sur cette relation.Similairement, on trouve une correlation plus forteentre LWP et indice en aerosols au cas ou le deuxiemeeffet indirect des aerosols anthropiques est inclus.

76

Part III.

Impacts of aerosol forcings on climate

change

77

6. Impacts of greenhouse gases and aerosol

direct and indirect effects on clouds and

radiation in atmospheric GCM simulations of

the 1930 - 1989 period

6.1. Introduction

Looking back at the twentieth century the observedwarming on the global scale has been less than ex-pected from the anthropogenic greenhouse effectalone (Mitchell et al., 2001). Particularly for theperiod of 1940 to 1980, an almost constant tem-perature has been observed. We thus choose aslightly longer period from 1930 to 1989 to investi-gate possible reasons for this behavoir. Since serveralyears, an additional anthropogenic forcing by sul-fate aerosols has been proposed as a negative forcingwhich counterbalanced the warming effect of green-house gases (Wigley , 1989; Charlson et al., 1989). Inthis study, we analyze the impact of greenhouse gasesand aerosols on the Earth’s radiation budget, withfocus on indirect or feedback effects by modificationof cloud properties.

Anthropogenic greenhouse gases consist of carbondioxide from fossil fuel combustion and biomass burn-ing, and other gases like methane, nitrous dioxide,chlorofluorocarbons, and tropospheric ozone. Thesegases are responsible for a warming of the Earth’ssurface because of enhanced atmospheric absorp-tion and emission at lower temperatures of longwave(LW) radiation in the troposphere. The combustionof (sulfur-containing) fossil fuel is also a major sourcefor sulfate aerosols. Aerosols scatter sunlight and en-hance the planetary shortwave (SW) albedo, an effectknown as the “aerosol direct effect” (ADE). In addi-

tion, by their ability to act as cloud condensation nu-clei, (hydrophilic) aerosols change cloud properties,producing essentially an increase in cloud albedo.These processes are called “aerosol indirect effects”(AIE). As it has been shown by field studies (e.g.,Brenguier et al., 2000a) and satellite observations(e.g., Breon et al., 2002), the cloud droplet effectiveradius, re, decreases with increasing aerosol concen-tration. For a constant liquid water content, thisresults in a larger cloud albedo (this effect is calledthe first AIE, Twomey , 1974). Smaller droplets arealso less likely to collide, therefore delaying or sup-pressing rain-forming microphysical processes. Thisresults in a longer cloud lifetime and thus a largercloud liquid water content, which again results in alarger cloud albedo (called the second AIE, Albrecht ,1989). Such an effect has been observed by Rosen-feld (2000) using satellite observations of clouds invarious polluted and unpolluted conditions.

It is useful at least in the framework of modelingof external influences on the climate to introduce theconcept of radiative forcing provoked by an externalperturbation. The radiative forcing, ∆F , is definedas the difference in radiation flux at the tropopausein perturbed and unperturbed conditions, while thetemperature and humidity profiles in the troposphereare fixed. In this definition, e.g., used by the IPCC(2001), any “small” global mean radiative forcingin a GCM is directly related to a change in global

79

6.2. Model description 6. Radiative impacts of greenhouse gases and aerosols

mean surface temperature at equilibrium through amodel-dependent sensitivity parameter λ:

∆Ts = λ ∆F (6.1)

The largest anthropogenic radiative forcing is dueto the greenhouse effect, with a positive annual globalmean of the order of 2.4 Wm-2 for greenhouse gas(ghg) concentrations at the end of the 20th cen-tury compared to pre-industrial conditions (e.g., Ra-maswamy et al., 2001). The annual global meanradiative forcing by the sulfate aerosol direct effectis estimated to be about of –0.5 Wm-2, the forcingby the first AIE to be of the order of –1.0 Wm-2,although both are very uncertain (Boucher and Hay-wood , 2001).

The suitable tools to understand and possibly pre-dict the impact of anthropogenic climate perturba-tions are general circulation models (GCMs) whichsimulate the dynamical and physical processes on aglobal scale. Using an atmospheric GCM, we tryto understand the role greenhouse gases and aerosoleffects have played in the evolution of the climate ofthe last century. We focus on changes in radiationfluxes throughout our simulation period of 1930 -1989. The radiative forcing due to greenhouse gasesand the aerosol direct and first indirect effects arecomputed. The second indirect effect, however, can-not be diagnosed instantaneously as it is related to atime-evolving process. There has been come contro-versy whether it should be considered as a forcing ora feedback mechanism. Rotstayn and Penner (2001)define a “quasi-forcing” for the second AIE as thedifference in radiation flux between two model inte-grations with fixed SST, which are identical exceptfor a change in the aerosol concentration in the pre-cipitation parameterization. They showed that, atleast for the first AIE, the quasi-forcing was a goodestimate of the actual radiative forcing compiled forfixed profiles of temperature and humidity.

In analogy to this, we introduce the “radiative im-pact” of an anthropogenic perturbation. This is de-fined as the difference in radiative fluxes between twosimulations with evolving SSTs, with and without theperturbation, but otherwise identical. We computethe radiative impact due to greenhouse gases and the

total aerosol radiative impacts, including the aerosoldirect and both indirect effects. Thus, the radiativeforcings due to greenhouse gases and aerosol directand first indirect effects can be compared to theradiative impacts of greenhouse gases and aerosols,which include feedback processes.

6.2. Model description

6.2.1. The atmospheric model

The model used is the atmospheric general circulationmodel (GCM) of the Laboratoire de MeteorologieDynamique (LMD). It is a grid-point model whichwe ran in a resolution of 96x73 points on a regu-lar longitude-latitude horizontal grid with 19 hybridsigma-pressure coordinate levels. The timestep is 6min for the dynamical part of the model and 30 minfor the physics. Prognostic variables are tempera-ture, horizontal winds, surface pressure, and totalwater content.

The physics of the model has formerly been describede.g. in Li (1999) and is reported briefly. Radiativetransfer is calculated in the model using an advancedversion of the scheme of Fouquart and Bonnel (1980)for the solar spectrum and an updated version of thecode of Morcrette (1991) for the terrestrial part. Indifference from former studies (e.g., Le Treut et al.,1998), the diurnal cycle is included. Convection isparameterized in the model using the Tiedtke scheme(Tiedtke, 1989). The land surface processes are pa-rameterized through a bucket model, and the sur-face temperature is evaluated in the boundary layerscheme using the surface energy balance equation.Condensation of water vapor is calculated using a“hat” probability density function for total watercontent to allow for fractional cloudiness (Le Treutand Li , 1991). We apply the cloud microphysicsscheme of Boucher et al. (1995a), which includes au-toconversion and accretion for liquid water cloudsand a simple snow-formation equation for ice clouds.

80

6. Radiative impacts of greenhouse gases and aerosols 6.2. Model description

6.2.2. Parameterization of the aerosol

indirect effects

The cloud droplet number concentration (Nd, incm-3) is diagnosed from the sulfate aerosol mass con-centration, mSO4 (µg sulfate m-3) using an empiri-cal formula (Boucher and Lohmann, 1995, , formula“D”):

Nd = 102.21+0.41 log10(mSO4) (6.2)

The volume-mean cloud droplet radius for liquid wa-ter clouds assuming spherical particles is

rd = 3

√ql ρair

34 π ρwater Nd

(6.3)

where ql is the liquid water mixing ratio, ρair andρwater the densities of air and water, respectively.From the combination of Eq. 6.2 and 6.3 it can bededuced that an increase in sulfate aerosol mass con-centration directly results in a decrease in the clouddroplet radius for constant liquid water content.The cloud property relevant to the radiation transferin the solar spectrum is the cloud optical thickness,τc. This quantity relates aerosols, clouds, and radi-ation in our model. Cloud optical thickness is pa-rameterized in terms of re and the cloud liquid waterpath, W (Stephens, 1978):

τc =32

W

re ρw

(6.4)

The autoconversion, Rll, of liquid water dropletsto raindrops in the scheme of Boucher et al. dependson the cloud droplet radius

Rll ∝ Ell ql r4d Nd (6.5)

where Ell is the collision efficiency, which is consid-ered to be 0 for droplet radii less than a “criticalradius”, rcrit, and 1 for larger droplets. For con-stant cloud liquid water content, rd ∝ N

−1/3d and

thus Rll ∝ rd for rd > rcrit. The conversion of cloudto rain water is thus less efficient for smaller droplets.

The cloud optical thickness and thus the cloud albedoare increased by the first AIE by decreasing re andby the second AIE by increasing W .

1930 1940 1950 1960 1970 1980 1990

400

500

600

700

[ppt

]

SO4

300

320

340

[ppm

]CO2

1200

1400

1600

[ppb

]

CH4

290

300

[ppb

]

N2O0

100

200

[ppt

]

CFC-110

200

400

[ppt

]

CFC-12

Figure 6.1.: The evolution of the sulfate aerosol at850 hPa (Boucher and Pham, 2002) andgreenhouse gas concentrations (Myhreet al., 2001). Global and annual meansare shown.

6.2.3. Experiment setup

The simulations were done using time-varying ob-served sea surface temperature (SST) and sea-iceextent distributions from Hadley-Centre analyses(HADISST1.1, Rayner et al., 2003)) for the period1930 - 1989. The monthly mean distributions aresmoothed out in time.

Monthly mean sulfate aerosol distributions were com-puted every 10 years of the simulation using histor-ical emission data and applying a comprehensivesulfur cycle model (Boucher et al., 2002; Boucherand Pham, 2002). The monthly distributions arealso smoothed out in time. Fig. 6.1 shows the global

81

6.2. Model description 6. Radiative impacts of greenhouse gases and aerosols

GHG+AER GHG CTL

Greenhouse gases time-varying time-varying pre-industrialAerosols time-varying pre-industrial pre-industrial

Table 6.1.: Summary of the forcings in the three ensembles. Throughout the paper, we will use upper caseabbreviations to name the simulations, and lower case, if the forcing itself is meant.

and annual mean sulfate aerosol mixing ratio at 850hPa.

Greenhouse gases are considered well mixed through-out the troposphere in our model. We use yearlymean averages of CO2, CH4, N2O, CFC-11 and CFC-12 from Myhre et al. (2001) as displayed in Fig. 6.1.

We carried out three ensembles of simulations, eachconsisting of three members for which the initial con-ditions for 1st January 1930 were slightly perturbed.In the first, named GHG+AER, greenhouse gases aswell as sulfate aerosol concentrations were varying asdescribed above. In the second, named GHG, sulfateaerosols were fixed to the calculated pre-industrialdistribution, while the greenhouse gas concentra-tions increased during the simulation. Finally, wedid a control simulation CTL, in which both sulfateaerosols and greenhouse gas concentrations were fixedto their pre-industrial concentrations. A summary ofthe ensembles is given in Tab. 6.1. Throughout thispaper, the abbreviations GHG and AER in capitalswill refer to the scenarios, whereas ghg (“greenhousegases”) and aer (“aerosols”) in lower case refer tothe perturbations of the atmospheric concentrationsthemselves.

6.2.4. Radiation diagnostics

The influence of greenhouse gases and aerosols on theradiative fluxes in our model is examined using twodifferent approaches. Firstly, we calculate the radia-tive forcings, as defined in the introduction. Secondly,we define a radiative impact, a new quantity whichincludes feedback processes of the climate system.The shortwave and longwave spectra are treated sep-arately, as are the perturbations by greenhouse effectand different aerosol effects.

Radiative forcings

For greenhouse effect, the radiative forcing is cal-culated off-line using the temperature and humidityprofiles and the cloud and surface properties for eachmonth of the simulations. The difference in longwaveradiative flux at the tropopause between current andpre-industrial greenhouse gas concentrations is com-puted after adjustment of the stratosphere profiles(Cess et al., 1993, Hansen et al., 1997).

Concerning the aerosol effects, the shortwave ra-diative forcings due to the aerosol direct and firstaerosol indirect effect are calculated on-line in thesimulation. The radiation scheme is applied twice,with the aerosol and cloud optical properties for cur-rent and pre-industrial aerosol distributions, and thedifference in SW radiative fluxes at the top of theatmosphere (TOA) is taken as the radiative forc-ing. This implicitly neglects the adjustment of thestratosphere, which is a valid assumption for the SWspectrum (Hansen et al., 1997). The radiative fluxescomputed with the current aerosol concentrationsare used in the integration, while the fluxes calcu-lated with pre-industrial values are just stored forthe analysis.

Radiative impacts

In order to include the impact of changing cloud prop-erties and (other) feedback processes in our analysis,we define a new quantity, the “radiative impact” ofa given perturbation (Iperturbation). We define thisradiative impact as the difference in radiative fluxes,F , at the top of the atmosphere between two scenar-ios, in which all but one parameter of the simulationare fixed and the internal variability of the model isremoved by averaging over an ensemble of simula-tions.According to this definition, the total sulfate aerosol

82

6. Radiative impacts of greenhouse gases and aerosols 6.3. Results

ISW|aer = < FSW|GHG+AER > – < FSW|GHG >

ISW|ghg = < FSW|GHG > – < FSW|CTL >

ILW|aer = < GHE|GHG+AER > – < GHE|GHG >

ILW|ghg = < GHE|GHG > – < GHE|CTL >

Table 6.2.: Definition of the SW radiative impacts of aerosols and greenhouse gases. The total greenhouseeffect (GHE) is defined in the text. The brackets indicate ensemble averaging.

radiative impact, Iaer, is the difference in radiativefluxes between the GHG+AER and GHG simula-tions. The radiative impact of aerosols is composedof the aerosol direct effect as well as the first and sec-ond aerosol indirect effects, which possibly includesfurther feedback processes due to aerosols. Similarly,we define the radiative impact of greenhouse gases,Ighg (see Table 6.2 for a summary).

In the shortwave spectrum, we compare the net radia-tive fluxes at the top of the atmosphere in two ensem-bles. This is, however, not possible in the longwavespectrum. While over land the surface temperatureadjusts to bring the radiative budget at TOA to equi-librium, this is not the case over the oceans, wherethe SST is fixed. We therefore introduce a new quan-tity, termed the total greenhouse effect (GHE). Thisis defined as the difference in the upward LW radia-tive fluxes at the surface and TOA. The upward LWflux at the surface is the sum of the thermal emissionof the surface and the reflected fraction of the atmo-spheric thermal emission reaching the surface. TheGHE can therefore be written as

GHE =[εσT 4 + (1− ε)Fsfc ↓

]− FTOA ↑

where ε is the surface emissivity, σ the Stefan-Boltzmann constant, and T the surface temperature.The GHE can be seen as the amount of energy in theLW spectrum responsible for a greenhouse effect atthe surface.

Using this radiation quantity, the radiative impact inthe LW spectrum can be defined as the difference inGHE between the ensemble means of two scenarios(Table 6.2).

1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

Rad

iativ

e fo

rcin

g [W

m-2

]

Greenhouse gases (LW)Greenhouse gases clear sky (LW)1st aerosol indirect effect (SW)Aerosol direct effect (SW)Linear sum (GHG+ADE+AIE)

Figure 6.2.: Annual global mean radiative forcings[Wm-2] as diagnosed in the ensemblemean of the GHG+AER simulations:greenhouse gas forcing (LW spectrum,dashed line), aerosol direct forcing (SWspectrum, dot-dashed), and first aerosolindirect forcing (SW, dot-dot-dashed).The solid line corresponds to the sum ofthe three diagnosed radiative forcings.

6.3. Results

6.3.1. Radiative forcings

Figure 6.2 shows the radiative forcings in the scenariosimulation GHG+AER. The greenhouse gas forcingin the model is positive and ranges from 0.74 Wm-2

in 1930 to 2.07 Wm-2 in 1989, which is consistentwith the value of 2.43 Wm-2 given by the IPCC (Ra-maswamy et al., 2001) for all greenhouse gases inyear 2000. The direct forcing by sulfate aerosols de-creases from –0.2 to –0.5 Wm-2 and the first aerosolindirect forcing from –0.6 to –1.3 Wm-2. This is alsoin the range of results compiled by Ramaswamy et al.(2001) (–0.2 to –0.8 Wm-2 for the ADE and –0.3 to–1.8 Wm-2 for the first AIE).

83

6.3. Results 6. Radiative impacts of greenhouse gases and aerosols

If the three forcings add up linearly, the net forc-ing would be slightly negative for the period 1930 to1980 (with a minimum of –0.25 Wm-2 in 1956), andpositive afterwards (up to 0.27 Wm-2 in 1989).

Figure 6.2 also shows the greenhouse gas LW radia-tive forcing, restricted to cloud free conditions. Thisquantity is about 0.1 to 0.2 Wm-2 larger than itsall-sky counterpart. The explanation is that (high-level) clouds themselves cause a greenhouse effectand thus mask part of the greenhouse effect due toanthropogenic greenhouse gases.

6.3.2. Radiative impacts

The longwave spectrum

In Fig. 6.3, the LW radiative impacts of greenhousegases and aerosols are shown. The LW radiative im-pact of the aerosols is negligible. A trend of about0.2 W m-2 decade-1 is simulated for the radiativeimpact of greenhouse gases with very similar valuesfor all-sky and cloudy-sky conditions. The radiativeimpact in cloud free conditions, Icf

ghg, is about 0.2 to0.5 Wm-2 larger. We now seek an explanation as forwhy the radiative impact in cloudy sky, Icc

ghg, and inall sky, Ia

ghg, show the same evolution, which involvesthe role of cloud feedbacks.

The all-sky GHE is composed by the GHE in thecloudy, GHEcc, and in cloud free regions, GHGcf,where each of the contributions is weighted by therespective cloudy and cloud free fractions of the gridcell, f and (1–f), respectively:

GHEa = f GHEcc + (1− f) GHEcf (6.6)

The all-sky radiative impact of greenhouse gases (seeSection 6.2.4), is thus influenced by the difference inGHE in cloudy and cloud free conditions, and by thechange in the cloud cover, ∆f = fGHG − fCTL:

Iaghg = Icf

ghg (1− fGHG) + Iccghg fGHG + (6.7)

∆f ( GHEccCTL −GHEcf

CTL)

The fact that the all-sky radiative impact of green-house gases follows its cloudy-sky counterpart meansthat the third term on the right hand side of Eq. 6.7

1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

Rad

iativ

e im

pact

[W m

-2]

aerghgaer (clear)ghg (clear)aer (cloudy)ghg (cloudy)

LWLWLWLWLWLW

Figure 6.3.: Global annual means of the longwaveradiative impact [Wm-2], as defined inSection 6.2.4, for aerosols and greenhousegases for all-sky (black), cloud free (blue)and cloudy (green) conditions. An 11-year running mean is applied.

compensates for the observed increase in cloud freeradiative impact. Since GHEcc

CTL − GHEcfCTL is a

positive quantity, there must be a decrease in thecloud cover (see Section 6.4).We can summarize thefollowing results:

• The radiative impact of greenhouse gases islarger in clear-sky conditions compared to all-sky conditions.

• The increase in all-sky radiative impact resultsfrom an increase in clear-sky GHE, an increasein cloudy-sky GHE, and a decrease in cloudcover.

The radiative impact of the greenhouse gases variesfrom 0.81 Wm-2 in 1930 to 2.09 Wm-2, which isclose to the corresponding diagnosed radiative forc-ing. Since the radiative impact includes some feed-backs, the two quantities being equal means that thenet radiative impact of these feedback processes issmall in the context of these experiments with fixedSST.

84

6. Radiative impacts of greenhouse gases and aerosols 6.3. Results

1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

Rad

iativ

e im

pact

[W m

-2]

aerghgaer (clear)ghg (clear)aer (cloudy)ghg (cloudy)

SWSWSWSWSWSW

Figure 6.4.: Global annual means of the shortwaveradiative impacts [Wm-2], as defined insection 6.2.4, for aerosols and greenhousegases. An 11-year running mean is ap-plied.

Looking at cloud free conditions however the radia-tive impact of greenhouse gases is larger than thecorresponding radiative forcing. This small differencecan be explained by the existence of a positive feed-back related to an increase in water vapor concentra-tion due to a warmer atmosphere.

The shortwave spectrum

Figure 6.4 shows the radiative impact in the SW spec-trum. A small radiative impact is simulated concern-ing the greenhouse gases. We attribute this to a snowcover feedback (melting of snow due to the warmersurface temperature in the simulation including an-thropogenic greenhouse gases), which is of the or-der of 0.1 Wm-2 and exhibits a gentle trend of 0.002Wm-2 decade −1. In cloudy conditions, greenhousegases cause an increase in radiative impact, which issomewhat smaller than the increase in all-sky condi-tions. We conclude that

1. The increase in SW flux in cloud covered con-ditions corresponds to less reflecting clouds.

2. As the all-sky radiative impact is larger than forboth the clear and cloudy-sky radiative impactthe cloud cover must be reduced by greenhousegases.

These findings will be discussed in the following sec-tion.

For the aerosols a decrease in the clear-sky radia-tive impact is simulated from –0.2 to –0.5 Wm-2.This is almost exactly the evolution diagnosed for theaerosol direct radiative forcing. It can be concludedthat there is almost no net radiative impact due tofeedback processes in cloud free conditions and forfixed SST. In cloudy conditions, the radiative impactis strongly negative and decreases from –0.8 to –1.6Wm-2. The all-sky radiative impact lies in betweenthe two, ranging from –0.7 to –1.4 Wm-2. This is onlya little bit more than the radiative forcing diagnosedfor the first AIE alone and thus much smaller than thesum of first AIE and ADE forcings. We investigatein Section 6.4.2 the feedback processes that introducesome positive radiative impact due to aerosols.

-0.2

0

0.2

Hig

h le

vel [

%] aerosols

greenhouse gases

-0.2

0

0.2

Mid

leve

l [%

]

-0.2

0

0.2

Low

leve

l [%

]

1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985

-0.2

0

0.2

Tota

l [%

]

Figure 6.5.: Temporal evolution of the differences inannual and global mean cloudiness be-tween ensemble means. The evolutionfor high-level (above 450 hPa), mid-level(between 800 hPa and 450 hPa), and low-level clouds (below 800 hPa), and the to-tal cloudiness is shown. An 11-year run-ning mean is applied.

85

6.4. Discussion 6. Radiative impacts of greenhouse gases and aerosols

6.4. Discussion

6.4.1. Trends in cloudiness

Figure 6.5 shows the simulated evolution of thechange in global annual mean cloud cover. Thechanges in total, low-level (below 800 hPa), mid-level(between 800 and 450 hPa) and high-level (above 450hPa) cloud cover are shown. As already noted in theanalysis of the radiative impacts in the previous sec-tion, the anthopogenic greenhouse gases result in adecrease in cloud cover. This reduction is found forclouds at all three levels, and it is strongest for high-level clouds. There is an increase in water vapor con-tent in the atmosphere, qv, in the GHG comparedto the CTL experiments, while the saturation waterwapor mixing ratio, qsat, increases as well. The neteffect is a decrease in cloudiness. As already shown,clouds also contain less water and have thus a smalleralbedo.

-0.1

0

0.1

Hig

h le

vel

ghgaer

-0.1

0

0.1

Mid

leve

l

-0.1

0

0.1

Low

leve

l

EQ30S60SSP 60N NP30N

-0.1

0

0.1

Tota

l [%

dec

ade-1

]

Figure 6.6.: Zonal mean of the linear trends in theannual mean high-, mid-, and low-level,and total cloudiness. The differences be-tween ensemble means are shown, as inFig. 6.5.

Figure 6.6 shows the zonal average of the linear trendsin the high, middle, and low-level, and total cloudi-ness. It shall be noted that in the global mean, thesetrends have the slope of the curves in Fig. 6.5. Forthe aerosol impact on cloudiness, different behaviorscan be noted for clouds at different levels. High-levelcloudiness decreases, which is a result of the coolingof the upper troposphere due to the aerosol radia-tive effects. The reduction in high-level cloudinessdue to aerosols does not show a uniform distributionover the globe. While sometimes even an increasecan be observed (e.g., at 20 ◦S and in the northernhemisphere high latitudes), at most latitudes, thetendency is negative. Aerosols seem always to havean opposite effect of the greenhouse gases.

For low and mid-level clouds, greenhouse gases causea slightly decreasing trend at low and mid latitudes,and an increase at high latitudes. The impact ofaerosols on low and mid-level clouds is different inboth hemispheres. In the southern hemisphere, ex-cept at very high latitudes, almost no trends in lowand mid-level cloudiness are observed. This is ex-pected, as the anthropogenic aerosol concentrationin the SH is rather low. In the northern hemisphere,however, low and mid-level cloudiness increases un-der the influence of aerosols. A reason for this couldbe the second AIE. This question will be addressedin the following Section.

The minima in the trend in low-level cloudiness ob-served at 40 and 60◦N correspond to a decrease incloudiness in limited regions (over Himalaya, Green-land, and Alaska), counterbalancing the general in-crease.

6.4.2. Influence of clouds on the SW

radiative impact

As stated in Section 6.3.2, greenhouse gases have apositive radiative impact in the SW spectrum dueto the reduction in cloudiness, in particular at low-levels.

86

6. Radiative impacts of greenhouse gases and aerosols 6.4. Discussion

EQ30S60SSP 60N NP30N

-0.4

-0.3

-0.2

-0.1

0

0.1

SW

radi

ativ

e im

pact

[W m

-2 d

ecad

e-1]

Iaer

I’aer

Figure 6.7.: Zonal mean linear trends in the an-nual mean SW radiative impact of theaerosols as defined in section 6.2.4 (Iaer,solid line) together with the radiative im-pact of second AIE and cloud feedbacks((I ′aer, dashed line).

Due to the impact of the aerosols, on the other hand,the amount of low- and mid-level clouds increases, es-pecially in northern hemisphere midlatitudes, whilehigh-level cloud cover slightly decreases. To isolatethe effect of aerosols on the cloudiness (or cloud watercontent) from the other aerosol effects acting in theSW spectrum (i.e., the ADE and the first AIE), weintroduce a new quantity, I ′aer, defined as the radia-tive impact of aerosols minus the diagnosed radiativeforcing due to the first AIE and the ADE:

I ′aer = Iaer −∆F1st AIE −∆FADE (6.8)

This quantity I ′aer is a measure of the second aerosolindirect effect and all cloud feedbacks associated toaerosol effects. It is positive, increasing from 0.2to 0.4 Wm-2 in 1989. Since the radiative impact inclear-sky not explained by the direct radiative forcingof aerosols (Icf

aer − ∆FADE) is very close to zero andlimited to high latitudes, we infer that there are nomajor feedbacks involved except processes involvingclouds.

Figure 6.7 shows the zonal mean of the linear trends1930 to 1989 of the Iaer and Iaer′ . The total aerosolradiative impact (Iaer)is negative almost everywhere,with the lowest values in the northern hemispheremidlatitudes.

EQ30S60SSP 60N NP30N

-2

-1

0

1

2

Low

leve

l clo

ud li

fetim

e tre

nds

[min

day

-1 d

ecad

e-1]

ghgaer

Figure 6.8.: Zonal mean linear trends of the annualmean diurnal lifetime of low-level clouds[min day−1 decade−1]. Differences be-tween scenarios are shown as before.

The radiative impact of second AIE and cloud feed-backs (I ′aer) is negative in northern hemisphere mid-latitudes, while it is positive in the region 40◦S -20◦N. Analysis shows that high cloud cover and I ′aertrends are anticorrelated (correlation coefficient –0.38for the globe, and –0.46 for the region mentioned).The increase in I ′aer therefore seems to be linked tothe decrease in high-level cloud cover.I ′aer is negative in northern hemisphere midlatitudesdespite the reduction in high-level cloudiness. This isdue to the increase in low-level cloudiness which weinterpret as the second aerosol indirect effect.

6.4.3. The impact of the second aerosol

indirect effect

There are two possibilities how a time-average cloudcover increase can be generated at a given grid box.Either there is an increase in cloud extent at eachgiven time, or there is a longer persistence of a givencloud amount. The temporal extent of clouds, ortheir lifetime is controlled by the microphysical pro-cesses that convert cloud water into precipitation.The second indirect effect causes these processes tobe slower for liquid water clouds, and thus will in-crease the persistence of clouds. We calculated thecloud lifetime in both scenarios, with and without an-thropogenic aerosols. Cloud lifetime is defined hereas the time per day, in which a grid box is covered

87

6.5. Conclusion 6. Radiative impacts of greenhouse gases and aerosols

with fractional cloudiness above a threshold of 1%(to exclude an impact of possible spurious persistentcloud cover). Cloud lifetime is thus expressed in min-utes per day.

We show in Fig. 6.8 the zonal mean of the 1930to 1989 linear trends in low-level cloud lifetime. Thedifferences between the scenarios are shown to isolatethe impact of the respective perturbations. Green-house gases decrease the cloud lifetime at low lat-itudes. This may be due either to an increase incloud liquid water, thus an increase in cloud dropletsize and an increase in the efficiency of rain-formingmicrophysical processes, or to an intensification ofconvection, which would cause the convective rainformation to become more efficient.

Except at two latitude bands (40◦N and 60◦N, againdue to regionally limited effects of opposite sign overHimalaya, and Greenland and Alaska, respectively),cloud lifetime increases strongly due to aerosols allover the northern hemisphere. The increase in low-level cloud cover is well correlated to the increase incloud lifetime (correlation coefficient of 0.86), whichsuggests that the increase in low-level cloud cover isdue to an increase in the duration of clouds ratherthan an increase in their fractional coverage whenthey are present.

6.5. Conclusion

We investigated the radiative impacts of greenhousegases and aerosols in a simulation of the 20th cen-tury. Using the LMDZ GCM in a version forced byobserved SST and sea ice distributions, we carriedout three ensembles of simulations: a first ensem-ble where both greenhouse gases and sulfate aerosols

were increasing according to historical values, a sec-ond ensemble, where the aerosol distribution wasfixed to pre-industrial values, and finally, a controlensemble, where the greenhouse gas concentrationswere fixed to pre-industrial values as well.

Both greenhouse gases and aerosols impact stronglythe longwave and shortwave radiative fluxes. Whilethe increase in the greenhouse effect due to the green-house gases and the decrease in the SW radiativeimpact due to the aerosol direct effect are straight-forward understandable, the impact on radiation viathe change of cloudiness is more complicated. Thechange in clouds in different atmospheric layers hasa strong impact in both solar and terrestrial compo-nents. The radiative impact of aerosols on the SWradiation is strongly negative and the radiative im-pact beyond the direct radiative forcing of aerosolsand the first AIE is positive. It results from the sec-ond aerosol indirect effect that leads to an increasein cloud water in low-level clouds, particularly inthe northern hemisphere but a decrease in high-levelcloudiness, resulting in a reduction of cloud albedoand thus a positive radiative impact. The latter find-ing may be due to our parameterization which doesnot take into account the impact of aerosols on iceclouds. A simulation with a future complex micro-physics scheme, treating both liquid water and iceclouds, will be needed to confirm this finding.

Our results, although obtained in transient simu-lations, therefore suggest that it may not be appro-priate to diagnose the second AIE as the differencebetween two simulations at fixed SST, as complexfeedbacks involving high-level cloudiness are also atstake.

88

6. Radiative impacts of greenhouse gases and aerosols 6.6. Resume

6.6. Impacts des gaz a effet de

serre et des effets directs et

indirects des aerosols sur les

nuages et le rayonnement

dans des simulations de la

periode 1930 a 1989 avec un

MCG atmospherique :

Resume

Introduction

En regardant l’evolution du climat pendant le 20iemesiecle, le rechauffement global n’a pas ete aussi fortqu’attendu du a l’effet de serre anthropique seul(Mitchell et al., 2001). En particulier dans la periode1940 a 1970, une temperature moyenne annuelle glob-ale presque constante a ete observee. La periode 1930a 1989 est alors choisie pour examiner les raisons pos-sibles a ce comportement. Depuis quelques annees,un forcage radiatif supplementaire du aux aerosolsde sulfates anthropiques a ete propose comme unforcage negatif qui pourrait contrebalancer une par-tie de l’effet rechauffant dus aux gaz a effet de serre(Wigley, 1989 ; Charlson et al., 1989).

Methodes

Trois ensembles de simulations ont ete effectueesavec LMDZ en imposant la temperature de la sur-face de la mer (sea surface temperature, SST enanglais), et l’extension de la glace de mer a par-tir des distributions mensuelles du Hadley-Centre(HADISST1.1, Rayner et al., 2003). Des distribu-tions mensuelles des aerosols soufres ont ete calculeeschaque decennie de la simulation en utilisant desdonnees historiques des emissions et en appliquantun modele complet du cycle de soufre (Boucher etal., 2002 ; Boucher et Pham, 2002). Les gaz a effetde serre sont consideres comme etant bien melangespartout dans l’atmosphere. Les moyennes annuellesdes cinq especes (CO2, CH4, N2O, CFC-11 et CFC-12) de gaz a effet de serre anthropiques de Myhre et

al. (2001) sont utilisees. Chaque ensemble de simu-lations consiste de trois membres. Dans le premierscenario, nomme GHG+AER, les gaz a effet de serreet les sulfates ont ete utilises comme decrit aupara-vant. Dans un deuxieme scenario, la concentration enaerosols a ete fixee a une distribution pre-industrielle(nomme GHG). Dans le troisieme scenario, celui decontrole (CTL), les concentrations des gaz a effetde serre ont ete fixees a des valeurs pre-industriellesaussi.

Dans l’ensemble GHG+AER, le forcage radiatif parles gaz a effet de serre a ete calcule hors-ligne en util-isant les profils de temperature et d’humidite dans latroposphere et les conditions de la surface terrestre.Le forcage radiatif est calcule comme la differencedans le flux du rayonnement a la tropopause entredes situations actuelles et pre-industrielles pour lesgaz a effet de serre apres ajustement des profils dansla stratosphere (Cess et al., 1993 ; Hansen et al.,1997). Pour les aerosols, les forcages par l’effet directet par le premier effet indirect ont ete calcules enligne dans la simulation GHG+AER dans le spectredes ondes courtes. Le schema du rayonnement estapplique deux fois, une fois en utilisant la concentra-tion en aerosols actuels (c’est ce flux de rayonnementqui etait utilise pour l’integration du modele), etune deuxieme fois en utilisant la concentration pre-industrielle. Les deux parametres qui changent sontles proprietes optiques des aerosols et l’epaisseur op-tique des nuages, qui est diagnostiquee avec les deuxconcentrations en nombre de gouttelettes differentes.L’ajustement de la stratosphere peut etre negligedans le spectre des ondes courtes.

Afin d’inclure aussi d’autres effets sur les nuagesdans l’analyse, et en particulier le deuxieme effet in-direct, nous introduisons comme nouvelle quantite� l’impact radiatif� d’une perturbation donnee.Cet impact radiatif est defini comme la differencedu flux de rayonnement au sommet de l’atmosphereentre deux scenarios, dans lesquels tous sauf un desparametres sont fixes, et dont la variabilite internea ete enlevee en moyennant sur l’ensemble. Seloncette definition, l’impact radiatif des aerosols dansle spectre des ondes courtes est la difference des

89

6.6. Resume 6. Radiative impacts of greenhouse gases and aerosols

flux radiatifs entre les scenarios GHG+AER et GHG(tableau 6.1). Pour le spectre des ondes longues, ilfaut pourtant encore modifier un peu cette definitionafin de l’appliquer a nos scenarios, car au-dessus desoceans, la temperature de la surface, qui a ete fixee,ne peut pas s’ajuster pour etablir l’equilibre radi-atif. Au lieu du flux de rayonnement au sommet del’atmosphere, nous utilisons donc l’effet de serre to-tal (greenhouse effet, GHE en anglais), qui est definicomme la difference entre le flux net ondes longuesa la surface terrestre et le flux ondes longues sortantau sommet de l’atmosphere. La difference en GHEentre deux moyennes d’ensembles est ainsi l’impactradiatif dans le spectre des ondes longues.

Resultats

Le forcage radiatif calcule pour les gaz a effet de serreest positif et augmente de 0.74 a 2.07 Wm−2 de 1930a 1989, ce qui est similaire aux valeurs donnees par leGIEC (Ramaswamy et al., 2001) pour les gaz a effetde serre bien melanges en 1990. Le forcage direct parles sulfates decroıt de –0.2 a –0.5 Wm−2, et le forcagepar le premier effet indirect de –0.6 a –1.3 Wm−2,ce qui est aussi dans la marge des valeurs cites parRamaswamy et al. (2001) de –0.2 a –0.8 Wm−2 pourl’effet direct et de –0.3 a –1.8 Wm−2 pour le premiereffet indirect (fig. 6.2).

En regardant les impacts radiatifs sur le spectre desondes longues, des impacts negligeables sont trouvespour les aerosols (fig. 6.3). Pour les gaz a effet deserre, une tendance d’environ 0.2 Wm−2 decade−1

est observee. On peut separer les impacts radiatifspour les conditions claires et nuageuses, et on trouvealors que l’impact radiatif des gaz a effet de serre estpresque aussi grand en ciel nuageux qu’en ciel total,mais qu’en ciel clair il est d’environ 0.2 a 0.5 Wm−2

plus grand. Les gaz a effet de serre augmentent ainsil’effet de serre a la fois dans les conditions claireset dans les conditions nuageuses, mais la couverturenuageuse diminue. Comme l’effet de serre total estplus grand en ciel nuageux qu’en ciel clair, cela ex-plique la concidence trouvee entre l’impact radiatifen ciel nuageux et en ciel total. On peut comparer leforcage radiatif calcule avant avec l’impact radiatif.La difference entre ces deux quantites est que l’une

contient les retroactions, et l’autre non. En ciel clair,on trouve que l’impact radiatif est legerement plusgrand que le forcage radiatif. On peut attribuer celaa la retroaction de la vapeur d’eau, qui pourtant estlimitee aux continents car la temperature de la sur-face oceanique est la meme dans les deux scenariosqui sont compares ici.

Dans le spectre des ondes courtes, on trouve unfaible impact des gaz a effet de serre pour des cielsclairs, qui est du a une retroaction de la couverturede neige dans les hautes latitudes (fig. 6.4). En cielnuageux, on trouve un impact positif d’environ 0.1a 0.3 Wm−2 (1930 a 1989), qui correspond a unediminution de l’albedo des nuages du a une diminu-tion de la couverture nuageuse. Les aerosols ont unimpact negatif en ciel clair qui est tres similaire al’effet direct des aerosols qui a ete diagnostique. Enciel nuageux, ils ont un effet qui est legerement plusgrand que le premier effet indirect.

Discussion

Les retroactions des nuages sont d’un interet par-ticulier (figs. 6.5 et 6.6). On trouve a partir desdifferences entre des ensembles de simulations queles gaz a effet de serre font diminuer la couver-ture nuageuse a tous les niveaux, dans les basses etmoyennes latitudes jusqu’a 0.2 % decennie−1. Lesaerosols, de l’autre cote ont des impacts differentsa differents niveaux. Alors que la couverturenuageuse haute est legerement diminuee, la couver-ture nuageuse dans les niveaux bas et moyens aug-mente jusqu’a 0.2 % decennie−1, avec des tendancesbeaucoup plus fortes dans l’hemisphere nord que dansl’hemisphere sud. En isolant l’impact radiatif desprocessus de retroactions des nuages, et en excluantdonc les impacts radiatifs de l’effet direct des aerosolset du premier effet indirect, nous avons montre lacontribution du deuxieme effet indirect a l’impactradiatif (fig. 6.7). L’impact radiatif des processusde retroaction des nuages combine deux effets. Lalegere reduction de la couverture haute exerce un im-pact positif du a la reduction de l’albedo des cirrus.L’augmentation de la couverture nuageuse dans lesniveaux bas et moyen augmente pourtant l’albedodes nuages, avec un effet net legerement positif en

90

6. Radiative impacts of greenhouse gases and aerosols 6.6. Resume

moyenne globale. L’augmentation de la couverturebasse est liee a une augmentation de la duree devie des nuages plutot qu’a une augmentation de lacouverture nuageuse spatiale a chaque instant donne.C’est donc bien le deuxieme effet indirect qui est re-sponsable de cette retroaction. Le deuxieme effet in-

direct des aerosols est beaucoup plus marque dans lesmoyennes latitudes de l’hemisphere nord qu’ailleursdans le globe. L’impact radiatif des retroactions desnuages total y est negatif alors qu’il est positif dansl’hemisphere sud.

91

6.6. Resume 6. Radiative impacts of greenhouse gases and aerosols

92

7. Relative importance of greenhouse gas and

aerosol forcings for climate trends of the 1950

- 1989 period in atmospheric GCM

simulations.

7.1. Introduction

Analysis of the evolution of the Earth’s climate dur-ing the 20th century indicates an anthropogenic im-pact on several meteorological quantities. One of theobserved features is a trend of up to 0.1 K decade-1

for the annual global mean surface temperature (e.g.,Folland et al., 2001). Trends have been observed aswell in precipitation, cloud cover, and diurnal tem-perature range (Folland et al., 2001), some on globalscales and some only on regional scales. One of thechallenging tasks of climate research is the detectionand attribution of a possible impact of human activi-ties on the Earth’s climate. Two main anthropogenicsources of perturbations to the Earth’s atmospherehave been identified in the last decades, both of themmainly due to energy production, namely greenhousegases and aerosols.

Using an atmospheric GCM forced with transientobserved SST and sea-ice extent distributions, wetry to understand the role the effects of greenhousegases and aerosols played in the evolution of theclimate of the last century. Recently, similar simula-tions have been carried out with the Hadley CentreGCM (Sexton et al., 2003), which identified the im-pact of several anthropogenic forcings on land surfaceair temperatures. Here, we look at concrete impactsof aerosols and greenhouse gases on measurable, andmeasured, climate parameters. As the imposed SSTin our simulation drives the temperature and pre-

cipitation evolution to a large extent, we focus onparameters more sensitive to aerosols such as thecloud cover and the diurnal temperature range.

7.2. Model simulations

The ensemble simulations analyzed in this chapterhave been described in Section 6.2 and are brieflysummarized in Table 7.1. In Fig. 7.1 we show thetemporal evolution of the surface temperature inthe three ensembles. Because the changes in radia-tive forcing are relative to pre-industrial situations(Tab. 7.1), the three ensembles start with differentconditions in 1950. Warming by the greenhouse ef-fect causes the simulation with increasing greenhousegas concentrations to be the warmest and the CTLsimulations, where greenhouse gas concentrations arefixed to pre-industrial values, to be the coolest. TheGHG+AER simulation is in between as the coolingby aerosol effects partially offsets the warming by thegreenhouse effect. Two major findings can be de-duced from Fig. 7.1. Firstly, the time evolution ofthe temperature is largely dominated by the imposedevolving SST, which is the same in all three sets ofsimulations. Secondly, the number of three membersfor each ensemble seems to be sufficient as there areclear differences in the ranges of results correspondingto the three ensembles.

93

7.3. Observations 7. Climate trends in GCM simulations

GHG+AER GHG CTLGreenhouse-gas concentration current current 1750Sulfate aerosol concentration current natural sources only natural sources only

Table 7.1.: Characterization of the three ensembles.

1955 1960 1965 1970 1975 1980 1985

8.6

8.8

9

9.2

9.4

9.6

9.8

10

Tem

pera

ture

[°C

]

CTLGHGGHG+AER

Figure 7.1.: Evolution of the annual and global (60◦S- 80◦N) mean surface temperature overland in the three ensembles [◦C]. An11-year running mean is applied. Theshaded area corresponds to the rangecovered by the members of each ensem-ble. In green, the GHG+AER simula-tion, in red, the GHG simulation, and inblue, the CTL simulation.

7.3. Observations

Folland et al. (2001) summarize the identified ob-served trends in climate parameters since the be-ginning of the industrialization. On a global scale,the most important change observed is the generalincrease in surface temperature (Jones and Moberg ,2003; Peterson et al., 1998). This increase is duemainly to an increase in nighttime temperatures. Thedaily minimum temperature has risen almost twiceas much as the maximum temperature during the20th century. Thus, the diurnal temperature range(DTR), defined as the difference between monthlymean daily maximum and minimum temperatures,has decreased (e.g., Karl et al., 1993; Easterling et al.,1997; Hansen et al., 1995). Other climate parame-ters show trends as well, but we will limit the presentstudy to these two features which are considered to be

well proven on a global scale. However, while study-ing the trends in DTR, it should be noticed, that atleast for several regions an increase in cloud cover hasbeen observed and that over some regions precipita-tion has increased (Easterling et al., 1997).

7.4. Results

7.4.1. Land surface temperature

In Fig. 7.2 we show the evolution of the global annualmean land surface temperature in the three ensem-bles and the HadCRUT2 observations dataset (Jonesand Moberg , 2003). As already shown in Fig. 7.1, toa large degree, the evolution in the model scenariosis dominated by the imposed SST. Generally, themodel captures well the interannual variability aswell as the overall trend. The ensemble means andthe observations are correlated with correlation coef-ficients of 0.81, 0.80 and 0.78 for GHG+AER, GHG,and CTL scenarios, respectively. The linear trend intemperature over the whole period (1950 - 1989) is0.29 ◦C decade-1 in the observations, and 0.74, 0.84,and 0.70 ◦C decade-1 for the GHG+AER, GHG, andCTL experiments, respectively. To identify the dif-ferent impacts on climate parameters introduced bythe forcings by greenhouse gases and aerosols, respec-tively, we now look at the geographical distributions.

In the observations and the model output, we calcu-late the linear trend at each grid point as the slope ofthe linear regression of the time series. We isolate theregions where the trend is significant above the 90%confidence level according to the rank correlation testof Kendall (Conover , 1980, pp. 256f.).

94

7. Climate trends in GCM simulations 7.4. Results

1950 1955 1960 1965 1970 1975 1980 1985 1990

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

Tem

pera

ture

[°C

]

CTLGHGGHG+AEROBS

Figure 7.2.: The evolution of the departure of theannual global (60◦S-80◦N) mean landsurface temperature from the 1950 to1989 mean, in the GHG+AER (greenline), GHG (red), and CTL (blue) simu-lations, and in the HadCRUT2 observa-tions (black).

These trends are plotted in Fig. 7.3 for the period1950 - 89 for the northern hemisphere summertime(June-July-August, JJA). In JJA, aerosol radiativeeffects are most marked, because they are concen-trated in the northern hemisphere and impact on theSW radiation. Figure 7.3 shows the trends in JJAsurface temperature over land in the three ensem-bles and observations. For many parts of the globe,the trends in surface temperature are controlled bythe impact of the imposed SST on the continentsvia dynamical teleconnections. For most parts ofthe southern hemisphere, the three simulations donot differ much. Some differences in the geograph-ical distributions of the temperature increase overAustralia can be noted. The CTL scenario shows acooling over South America, which is not the casefor the two other ensembles. Also the high latitudesof the northern hemisphere (north of 60◦N) show asimilar pattern of increasing temperature. This isdue to a decreasing trend in the imposed sea ice ex-tent, which is the same in all three simulations. InNorth Africa and the midlatitudes of the Eurasianand North American continents, however, some dif-ferences between the three scenarios can be identified.

Figure 7.4.: Differences in the linear trend (1950 -1989) in JJA surface temperature overthe continents [K decade-1] between thesimulations. (a), the impact of bothgreenhouse gases and aerosols (differenceGHG+AER–CTL), (b), the impact ofgreenhouse gases (GHG–CTL), and (c),the impact of the aerosols (GHG+AER–GHG). Only regions with trends statis-tically significant above the 90% confi-dence level are shown.

The deceasing SST of the North Atlantic causes

95

7.4. Results 7. Climate trends in GCM simulations

Figure 7.3.: The linear trend (1950 - 1989) in JJA surface temperature over land [K decade-1] as simulatedby the LMDZ GCM. (a) for HadCRUT2 observations, (b) the GHG+AER simulation, (c) theGHG simulation, and (d) the CTL simulation. Only regions with trends statistically significantabove the 90% confidence level are shown.

∆∂Tsfc∂t |ghg+aer = <

∂Tsfc∂t |GHG+AER > – <

∂Tsfc∂t |CTL >

∆∂Tsfc∂t |ghg = <

∂Tsfc∂t |GHG > – <

∂Tsfc∂t |CTL >

∆∂Tsfc∂t |aer = <

∂Tsfc∂t |GHG+AER > – <

∂Tsfc∂t |GHG >

Table 7.2.: Definition of the impacts of greenhouse gases (“ghg”), aerosols (“aer”) and the combination ofboth (“ghg+aer”) on the trends in surface temperature (Tsfc). The brackets indicate ensembleaveraging.

the surface temperature to decrease in the simula-tion without other imposed forcings CTL in WesternEurope. This decrease is offset by warming dueto increased GHE in the GHG simulation. In theGHG+AER simulation, the same tendency can be

observed for the northern part of western Europe,while the southern part shows a strong cooling. TheGHG+AER simulation shows a persistent cooling inthe midlatitudes (20 - 50◦N), which is not observedin the other two simulations. This cooling is stronger

96

7. Climate trends in GCM simulations 7.4. Results

than in the two other ensembles also for the midlati-tudes of the North American continent.

Comparing observations and the three ensembles,it seems difficult to identify one of the scenarioswhich would unambiguously simulate more realisticresults that the other ones, mainly due to the strongimpact of the imposed SST. However, when lookingat the Eurasian continent, the GHG scenario is notable to simulate the observed cooling trend. TheCTL ensemble shows some cooling but the geograph-ical distribution over Europe and over East Asia arerather different from the observed patterns. On av-erage, the cooling simulated by the CTL scenariois not as strong as in the HadCRUT2 observations.Particularly, a strong cooling trend is observed inthe measurements and in the GHG+AER scenarioover regions with increasingly high aerosol burdensin the period of interest, i.e., over East Europe andEast Asia. Generally, the order of magnitude of thetrends in surface temperature is the same in both,model simulation and observations. Figure 7.4 showsthe differences in the surface temperature trends be-tween the three simulations. The impact of aerosolson surface temperature is defined as the differencein the trends between the GHG+AER and the GHGensembles. The impacts of greenhouse gases and thecombination of greenhouse gases and aerosols aredefined accordingly (see Table 7.2 for a summary).The decreasing temperature trend observed in theGHG+AER simulation in the NH midlatitudes is dueto the aerosol forcing. Via dynamical effects, aerosolscause a warming in South Africa. This warming isamplified when both forcings, greenhouse gases andaerosols, are combined. This warming trend is alsoshown by the observations.

7.4.2. Diurnal temperature range

As pointed out by several authors (e.g., Karl et al.,1993; Easterling et al., 1997; Hansen et al., 1995),the diurnal temperature range (DTR) (defined asthe difference between monthly mean daily maxi-mum and minimum temperatures at a given loca-tion) has decreased in the second half of the 20thcentury over large parts of the continents. The mini-mum nighttime temperatures have risen much faster

than the daytime maximum temperatures. Easter-ling et al. (1997) shows decreasing trends for mostcontinental regions for the 1950 to 1993 period. Inthe midlatitudes of the North American continent,a consistent decrease by –0.1 to –0.3◦C decade−1

is given. For Europe the trend is not clear includ-ing an increase in DTR in central Europe, slightdecreases over the British Islands and the Mediter-ranean region, and a decrease over Eastern Europe.

Figure 7.5.: As for Fig. 7.3, but for the diurnal tem-perature range over land [K decade-1].

Over Asia, although at some locations a slight in-crease in DTR is observed, generally the DTR has

97

7.5. Discussion 7. Climate trends in GCM simulations

decreased by –0.1 to –0.3◦C decade−1. Similarily,over Australia most regions show a decrease, and overSouth America and Africa, also a rather strong de-crease of up to –05 ◦C is found. Recently, it hasalso been shown on a regional scale for the case ofFrance that a clear increasing trend can be detectedfor summer nighttime minimum temperatures whichis not obvious in daytime maximum temperatures

Figure 7.6.: As for Fig. 7.4, but for the diurnal tem-perature range over land [K decade-1].

(Spagnoli et al., 2002). This behavior in the re-gion of France has been attributed to variations inevapotranspiration using a modeling study (Planton

and Spagnoli , 2003). Figure 7.5 shows the distribu-tions of the linear trends in DTR in the three sce-narios for northern hemisphere summertime (June-July-August; JJA). For South and North America,all three simulations show similar results - a reduc-tion of the DTR of the order of –0.1 to –1 K decade-1.Particularly, this reduction is concentrated over thewestern half of the continents. Over Alaska, however,an increase in DTR is observed. These findings arein good agreement with the observational data shownby Folland et al. (2001). Over Australia, both scenar-ios show a stronger reduction in DTR than the con-trol ensemble. For Africa and the Eurasian continent,both perturbations, greenhouse gases and aerosols,lead to a decrease in DTR. The GHG+AER scenariois the only one which reproduces a reduction in DTRof the same order of magnitude as observed. Fig-ure 7.6 shows the differences between the ensemblesto identify the impacts of the two forcings. Clearly,aerosols are needed to explain the strong reductionof the DTR over the Eurasian continent. In particu-lar, aerosols are responsible for the reduction in DTRover Europe and over Southeast Asia.

7.5. Discussion

A reduction in DTR might be due to a negative ra-diative forcing in the solar spectrum such as expectedfrom aerosol direct and indirect effects. While a forc-ing in the terrestrial spectrum as the greenhouseeffect is active both night and day, an increase inSW albedo has a cooling effect on the Earth’s surfaceonly at daytime (solar heating during the daytimeis reduced). This would reduce the daily maximumtemperature. The warming effect due to the green-house gases increases the nighttime temperature, butthe daytime temperature as well, and therefore wouldnot directly reduce the DTR. However, a possible in-crease in cloud cover as a feedback to the increasedgreenhouse effect would increase the SW albedo andmight thus explain the decrease in DTR.

Other explanations for the reduction in DTR ex-ist. For example, the increase in air-traffic inducedcontrail coverage of the globe may be responsible forthe reduction in DTR (Travis et al., 2002; Meerkotter

98

7. Climate trends in GCM simulations 7.5. Discussion

Figure 7.7.: Linear trends (1950 - 1989) in JJA cloud cover (I) and SW cloud radiative forcing (CRF, II),(a) for GHG+AER, (b) for GHG, and (c), for CTL scenarios [Wm-2 decade-1].

et al., 1999). Dai et al. (1997) show a negative cor-relation between precipitation increase and DTRdecrease in timeseries for several regions. They ar-gue that precipitation increase is correlated with anincrease in cumulonimbus and nimbostratus cloudi-ness. They explain that assuming the existence of thesecond aerosol indirect effect, a forcing different fromaerosol forcing would be responsible for this increasein precipitation, as the second AIE rather would tend

to decrease the precipitation. Soil moisture variationmay have an important influence on DTR changes.Increased evaporation due to larger soil moisturewould lead to a lowering of temperatures during day-time with less stable boundary layers, while duringnighttime with stable boundary layers, this effectwould be much smaller (Dai et al., 1997; Dai et al.,1999; Stone and Weaver , 2003).

99

7.6. Conclusions 7. Climate trends in GCM simulations

Dai et al. (2001) and Stone and Weaver (2003)studied the trends in DTR in the twentieth andtwenty-first century in simulations performed withtwo different GCMs. Both studies argue that it is anincrease in low-level cloud albedo and in soil mois-ture that causes the decrease in DTR. They do notfind an influence of aerosol direct effects. However,they did not examine the aerosol indirect effects.

Figure 7.8.: Linear trends (1950 - 1989) in JJA TOA(a) net SW flux, and (b) SW forcing bythe first aerosol indirect effect and theaerosol direct effect [Wm-2 decade-1].

Folland et al. (2001) show a correlation between thedecrease in DTR and the increase in cloud cover.Hansen et al. (1995) deduct from a modeling study,that a combination of aerosol and cloud cover in-crease particularly over land causes the decrease inDTR, which they attribute to an aerosol indirect ef-fect.

In Fig. 7.7, we show the linear trends in cloudinessand in SW cloud radiative forcing (CRF). Compar-ing the GHG and CTL scenarios, a largely similarpattern of change in cloudiness and CRF is observed.The slight differences include a smaller increase in

cloud cover and CRF over the South and NorthAmerican continents in the GHG compared to theCTL ensemble. The GHG scenario does not show theincrease in CRF as observed in the CTL ensembleover the equatorial region over Africa.

Some interesting differences are found comparingthe GHG+AER scenario to the two others. A strongincrease in cloud cover is observed over the mid lati-tudes of the northern hemisphere, and accordingly astronger decrease in (negative) CRF than in the GHGand CTL ensembles. This is observed as well overthe Eurasian as over the North American continents.Figure 7.8 shows in addition the trends in net SW ra-diation flux at the top of the atmosphere (TOA) andthe diagnosed forcing by the aerosol direct and firstindirect effects in the GHG+AER scenario. Trends inSW radiation are largely corresponding to the trendsin CRF, except for the high latitudes of the northernhemisphere, where the surface albedo changes due tosnow cover feedbacks play an important role. Com-paring the patterns of diagnosed forcings by aerosolsto the geographical distribution of the trends in SWradiation flux, it can be noted that the strongly nega-tive trends in the NH midlatitudes are correspondingto strongly negative forcing trends due to aerosols.In Fig. 7.9 we show the JJA linear trends of surfaceevaporation in the three scenarios. For some regions,the anticorrelation between evaporation and DTRtrends postulated by Dai et al. (1997) and Plantonand Spagnoli (2003) can be observed, e.g. for centralAfrica and, in the GHG simulation, over Australia.However, as these regions are rather limited, the cor-relation may be fortuitous. Similarily, the trendsin surface humidity and DTR are not found to becorrelated (not shown).

7.6. Conclusions

We showed results of three ensemble simulationswith an atmospheric GCM (the LMD-Z) usingimposed observed SST distributions. In one ofthese scenarios GHG+AER, anthropogenic green-house gas and sulfate aerosol perturbations wereimposed according to historical evolutions. In asecond one (GHG), only greenhouse gases were

100

7. Climate trends in GCM simulations 7.6. Conclusions

used, and in the third one (CTL), greenhouse gasesand aerosols were fixed to pre-industrial conditions.

Figure 7.9.: The linear trend in JJA surface evapo-ration [mm day−1 decade−2]. For (a)GHG+AER, (b) GHG and (C) CTLscenarios. Only statistically significanttrends are shown.

The forcing by the imposed SST controls the evolu-tion of temperature also over continental surfaces toa large extent. Interesting impacts of the two an-thropogenic forcings are observed. Looking at pat-terns of linear trends in surface temperature, it canbe noted that the scenario including both forcings is

closest to observations (HadCRUT2 dataset, Jonesand Moberg , 2003).

90◦S-90◦N 40◦S-55◦N 40◦N-55◦NDTR-CRF 0.24 0.68 0.70DTR-F SW 0.83 0.73 0.82DTR-E 0.50 –0.08 –0.02Ts-CRF –0.37 0.46 0.11Ts-F SW 0.87 0.66 0.57Ts-E 0.73 –0.22 –0.11

Table 7.3.: Correlation between trends in cloud ra-diative forcing (CRF), SW radiation flux(F SW), and surface evaporation (E),and trends in diurnal temperature range(DTR) and surface temperature (Ts), fordifferent latitude belts.

In particular over the Eurasian continent, a coolingtrend has been observed, which can only be simulatedwith the model when aerosol forcings are included.Similarly, while all three ensembles simulate com-parable patterns of decreasing diurnal temperaturerange (DTR) for most parts of the globe (South andNorth America, Australia), the scenario includingthe aerosol forcing is the only one that is able toreproduce the reduction in DTR in the magnitudeobserved in the northern hemisphere. We analyzedtrends in cloudiness and SW cloud radiative forcing(CRF) in the three scenarios. CRF changes largelycontrol trends in SW radiation flux at the top ofthe atmosphere, except for NH high latitudes, wheresurface albedo changes due to snow cover feedbacksplay a role. Aerosol forcings contribute strongly tostronger CRF and thus SW forcings particularly inNH midlatitudes. Table 7.3 lists the correlation co-efficients between trends in CRF and SW radiationflux and surface temperature and DTR, respectively,for the GHG+AER ensemble.

The correlation coefficients in the two other en-sembles are similar to the ones given for this sce-nario. Trends in surface temperature are correlatedto changes in SW forcing, but only slightly con-nected to trends in CRF. However, the DTR changesare strongly correlated to both the changes in CRF

101

7.6. Conclusions 7. Climate trends in GCM simulations

and SW radiation flux. We do not find a consistentanticorrelation between evaporation and DTR trendsas was the case in previous studies.

The overall conclusion is that in our model study,DTR decreases as observed particularly over con-tinental surfaces of the northern hemisphere arestrongly correlated to changes in SW cloud radia-tive forcing. Among the three ensembles, the onlyone which is able to simulate the observed magnitude

and geographical distribution of the decrease in DTRis the scenario including the two forcings.

For the trends in surface temperature, the resultsare more ambiguous. However, the reduction incontinental surface temperature over the Eurasiancontinent is resulting in the model study only inthe scenario including both anthropogenic forcings,greenhouse gases and aerosols, where aerosols are theforcing responsible for the cooling trends.

102

7. Climate trends in GCM simulations 7.7. Resume

7.7. Importance relative des gaz a

effet de serre et des forages

d’aerosols pour des tendances

climatiques de la periode

1950 a 1989 a partir des

simulations de MCG

atmospheriques : Resume

Introduction

Outre l’augmentation de la temperature moyenneannuelle globale de 0.1 K decennie−1 (Folland etal., 2001), des tendances ont ete observees pour laprecipitation, la couverture nuageuse, et l’ecart jour-nalier de la temperature (diurnal temperature range,DTR en anglais), quelques-unes a l’echelle globale,d’autres a l’echelle regionale (Folland et al., 2001).

Nous utilisons les simulations presentees dans ledernier chapitre pour examiner les impacts relat-ifs des gaz a effet de serre et des differents foragespar les aerosols sur les tendances de temperature dela surface continentale et sur le DTR. Nous limitonscette etude a la periode 1950 a 1989 car c’est danscette periode que les tendances observees ont ete lesplus claires.

Resultats

Les trois ensembles de simulations sont capables dereproduire l’evolution observee de la temperature dela surface continentale, meme si des differences entreles scenarios existent (fig. 7.1). Pourtant, le scenarioqui inclut les deux forages anthropiques reproduit lesobservations de maniere legerement meilleure que lesdeux autres (coefficient de correlation de 0.81 com-pare a 0.80 et 0.78 pour les ensembles GHG et CTL).

Nous regardons les distributions geographiques destendances lineaires de quantites physiques, qui sontdefinies comme la pente de la regression lineairede la serie temporelle a chaque point de la grillehorizontale. L’impact des aerosols est le plus fortdans l’ete de l’hemisphere nord, parce que les for-ages des aerosols sont actifs dans le spectre solaire

et les aerosols anthropiques sont concentres dansl’hemisphere nord. Nous nous focalisons alors surla saison juin-juillet-aot (JJA). De plus, seulementdes tendances qui sont statistiquement fiables a desniveaux superieurs a 90% sont considerees.

Pour plusieurs parties du globe, les tedances de latemperature sont controlees par la SST imposee, etce meme au-dessus des continents (fig. 7.3). Dansla plupart de l’hemisphere sud ainsi qu’aux hauteslatitudes de l’hemisphere nord, les trois ensemblesne different donc que peu. C’est en particulier dansles moyennes latitudes de l’hemisphere nord et enAfrique du Nord que se trouvent les differencesentre les trois scenarios. La SST decroissante del’Atlantique du Nord fait diminuer la temperaturedans la simulation CTL au-dessus de l’Europede l’Ouest. Cet effet est contrebalance par lerechauffement par l’effet de serre dans l’ensembleGHG. Dans les simulations GHG+AER, on trouvele meme effet pour la partie nord de l’Europe del’Ouest, alors que la partie sud montre un fort re-froidissement. Le scenario GHG+AER donne unrefroidissement persistent dans les moyennes lati-tudes de l’Amerique du Nord et au-dessus du conti-nent eurasiatique qu’on ne trouve pas dans les deuxautres. Il est difficile d’identifier lequel des ensemblesest le plus proche des observations, en particulier acause de l’impact fort de la SST imposee. Pourtant,en regardant le continent eurasiatique, l’ensembleGHG ne simule pas la tendance de refroidissementobservee. Le scenario CTL montre des regions de re-froidissement, mais la distribution spatiale au-dessusde l’Europe et de l’Asie de l’Est est differente decelle observee. De plus, le refroidissement n’est pasassez fort en moyenne. Ce n’est donc que le scenarioGHG+AER qui montre l’amplitude et la distribu-tion des tendances de refroidissement comme dansles observations. En etablissant les differences entreles moyennes d’ensembles, on peut isoler les impactsrespectifs des deux forages anthropiques. Cette anal-yse montre que c’est le forage par les aerosols quiest responsable tendances de refroidissement dansla simulation GHG+AER. Comme il l’a ete montrepar plusieurs auteurs (par exemple, Karl et al., 1993; Easterling et al., 1997 ; Hansen et al., 1995) la

103

7.7. Resume 7. Climate trends in GCM simulations

temperature minimale de nuit a augmente beaucoupplus rapidement que la temperature maximale dejour, et donc l’ecart journalier de la temperaturea diminue pendant la deuxieme moitie du 20iemesiecle. Dans nos simulations, des tendances tres sim-ilaires ont ete trouvees dans le DTR pour la plupartde l’hemisphere sud, et pour les hautes latitudesde l’hemisphere nord (fig. 7.5). Pour les basseset moyennes latitudes de l’hemisphere nord, pour-tant, des differences entre les scenarios peuvent etreidentifiees. La reduction du DTR est plus fort dansl’ensemble GHG compare a CTL, et encore beaucoupplus fort dans l’ensemble GHG+AER. Les deux for-ages anthropiques, les gaz a effet de serre ainsi queles forages par les aerosols, augmentent donc la ten-dance negative du DTR. C’est pourtant seulement lescenario GHG+AER qui inclut les deux forages quiest capable de simuler cette reduction et en ampli-tude et en distribution.

La reduction du DTR observee peut etre expliqueepar plusieurs processus differents. L’une des possi-

bilites serait un forage negatif actif dans le spectresolaire qui agit donc seulement pendant la journee.Un tel forage pourrait venir des effets directs ou in-directs des aerosols (Hansen et al., 1997), ou encored’une augmentation de l’albedo des nuages bas, d soita une augmentation de la couverture nuageuse soita une augmentation de leur contenu en eau liquide(Easterling et al., 1997). D’autres etudes montrentune correlation entre une augmentation du taux deprecipitation et la reduction du DTR (Dai et al.,1997) ou de l’evapotranspiration de la surface et leDTR (Planton et Spagnoli, 2003).

Dans notre etude, nous trouvons une correlationdes tendances du DTR avec les tendances du fluxdu rayonnement solaire au sommet de l’atmosphere(coefficient de correlation de l’ordre de 0.8), avec leforage radiatif des nuages (coefficients de correlationentre 0.24 et 0.70), mais pas avec l’evaporation au sol(coefficients de correlation entre –0.1et 0.50 ; tableau7.3).

104

Part IV.

Parameterization of ice phase processes

105

8. Evaluation of the cloud thermodynamic phase

parameterization in the LMDZ GCM by using

POLDER satellite data

8.1. Introduction

Clouds cover the Earth’s surface by about 60% to70% and have a very strong effect on the radiationbalance in both the terrestrial and the solar spec-trum. Unfortunately, the representation of clouds inclimate models is still unsatisfying. As most cloudproperties cannot be resolved by global-scale mod-els, they have to be parameterized in terms of themodel variables. Cloud parameterizations are com-plicated by the fact that clouds cover a large rangeof scales (from microphysics to mesoscale systems).Moreover they must treat the liquid and ice phasesseparately. Liquid and ice clouds have quite differenteffects on the Earth radiation balance. Precipitation-forming microphysical processes in ice and mixed-phase clouds are different from those in liquid waterclouds. In the absence of explicit liquid and ice cloudmicrophysics there is no simple parameterization forthe partitioning of condensed water into liquid andice phase. The local temperature is therefore oftenused to define the liquid and ice fractions.

Giraud et al. (2001) used combined data fromPOLDER-1 and ATSR-2 satellite instruments to es-tablish a relationship between cloud top brightnesstemperature and cloud top thermodynamic phase fora day in June, 1997. They found a clear relationshipbetween the two quantities. A sharp transition be-tween ice and liquid clouds in the temperature rangeof 240 to 260 K was observed.

Satellites observe clouds with a global coverage at

approximately the spatial scale of GCMs. Thus,satellite observations are well suited for the eval-uation of GCM parameterizations. Here, we usethe data from the POLDER-1 instrument in order toinfer a statistical relationship between cloud top tem-perature and cloud phase and to test the partitioningbetween liquid and ice water as a function of the localtemperature in the atmospheric General CirculationModel (GCM) of the Laboratoire de MeteorologieDynamique (LMDZ).

8.2. Tools and methodology

8.2.1. POLDER-1 satellite observations

POLDER (POLarization and Directionality of theEarth’s Reflectances) is a radiometer which was onboard the Japanese polar orbiting ADEOS-1 plat-form from August 1996 until June 1997. POLDERis a wide field-of-view camera (with a swath of2400x1800 km2 and a resolution of 6x7 km2) whichobserves the Earth’s surface and atmosphere through15 filters and polarizers in the visible and near in-frared. It was the first space instrument to simultane-ously observe the polarization and the multi-spectraland directional signatures of the reflected radiation(Deschamps et al., 1994). We use cloud top thermo-dynamic phase and cloud top temperature retrievedfrom POLDER data (Buriez et al., 1997). The cloudtop Rayleigh pressure is retrieved from the polar-ized signal at 443 nm based on molecular scatter-ing by the atmosphere above the cloud (Vanbauceet al., 1998). Using the corresponding interpolated

107

8.2. Tools and methodology 8. Evaluation of cloud phase parameterization

temperature profile from 6-hourly ECMWF meteo-rological analyses we convert the cloud top pressureinto a cloud top temperature. The cloud thermo-dynamic phase is also retrieved from POLDER dataas described in Goloub et al. (2000) and Riedi et al.(2000). The retrieval is based on the angular andpolarized signatures of cloud reflected radiances atscattering angles near 140◦. These cloud parametersare extracted at the pixel resolution. The POLDERlevel-2 products are then computed at the scale of a“super-pixel” composed of 9x9 POLDER level-1 pix-els (about 0.5◦x0.5◦ at the Equator). In POLDERthe level-2 cloud phase parameter can take three dif-ferent values: liquid if all cloud pixels are liquid, ice ifall cloud pixels are ice, and mixed if both liquid andice pixels coexist within a super-pixel. The satelliteis used at its original resolution of 9x9 pixels as wellas at the GCM resolution after an appropriate regrid-ding.

8.2.2. The LMD GCM

We use here the Laboratoire de Meteorologie Dy-namique (LMD) GCM. The resolution considered is96 points evenly spaced in longitude and 72 points inlatitude. The model is vertically discretized on 19 hy-brid σ-pressure levels (σ = p/ps where p and ps arethe atmospheric and surface pressures). An impor-tant characteristic of the LMD GCM version used inthis study is the definition of the cloud and precipita-tion parameterizations. Cloud water is predicted inthe model by a budget equation where the ice and liq-uid phases are considered jointly. The condensationscheme uses a “hat” probability density function fortotal water in a grid box, which allows for fractionalcloudiness (Le Treut and Li , 1991). Depending onthe local temperature, the total condensed water ispartitioned between liquid water and ice. Clouds arecomposed entirely of ice crystals or liquid droplets ifthe local temperature is lower than Tice or larger thanT0, respectively. In between T0 and Tice, the fractionof liquid water is given by

xliq =

0 T < Tice(

T−TiceT0−Tice

)nx

T ∈ [Tice, T0]

1 T > T0

(8.1)

-80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30Local temperature [°C]

0

0.2

0.4

0.6

0.8

1

Liqu

id w

ater

frac

tion

nx = 0.1, Tice = -70.0

nx = 0.5, Tice = -70.0

nx = 1.0, Tice = -70.0

nx = 2.0, Tice = -70.0

nx = 10.0, Tice = -70.0

nx = 0.1, Tice = -40.0

nx = 0.5, Tice = -40.0

nx = 1.0, Tice = -40.0

nx = 2.0, Tice = -40.0

nx = 10.0, Tice = -40.0

Figure 8.1.: Liquid water fraction as a function ofthe local temperature for Tice = −70◦Cand Tice = −40◦C, with values for nx ∈[0.1, 0.2, 1, 2, 10].

In the standard version of the model, Tice =−15◦C, T0 = 0◦C, and nx = 6. The choice ofT0 = 0◦C is rather straightforward as ice crystalsin the atmosphere melt rapidly above the meltingpoint. The two other parameters, however, are moredifficult to define. Observations suggest Tice to be inthe range of –35◦C to –40◦C (Houze, 1993). Valuesfor Tice in GCMs include –40◦C (Del Genio et al.,1996), –35◦ (Lohmann and Roeckner , 1996), –20◦

(Fowler et al., 1996), –15◦C (Smith, 1990). Schemesusing the a formula similar to Eq. 8.1 use nx = 2 (DelGenio et al. (1996); Smith (1990)). The impact ofTice and nx on the xliq to T relationship is illustratedin Fig. 8.1. We will examine the sensitivity of themodel results to these parameters and compare themto the satellite observations.

The period simulated is that of POLDER-1 fromNovember 1996 to June 1997. Reynolds sea surfacetemperature (SST) and HadISST1.0 sea-ice extentare imposed (Rayner et al., 2003). Winds and tem-perature are nudged to ECMWF reanalyses in orderto get realistic meteorological conditions.

8.2.3. Simulation of satellite observations

from the model results

In order to compare the model results with the satel-lite parameters, we produce a 2D “satellite-like” fieldfrom the 3D model results. For this study, we assumethat clouds are opaque for the quantities we are in-

108

8. Evaluation of cloud phase parameterization 8.3. Results

terested in and we use a random overlap assumption.The column cloud fraction, f , can be expressed as:

f = 1−n∏

i=1

(1− fi)

where fi is the cloud fraction at model layer i and n

is the uppermost layer. Under the same assumptionthe cloudy-sky mean of a variable x (cloud-top tem-perature, fraction of liquid clouds, ...) as seen fromthe top of atmosphere can be computed as:

xcloudy =∑n

i=1 xifi

∏nk=i(1− fk+1)f

(8.2)

using the convention fn+1 = 0.

For each cloudy layer we approximate the Rayleightemperature as the temperature of the interface be-tween the highest cloudy layer and the clear layerabove it. We then derive the cloud Rayleigh tem-perature at the top of the atmosphere (TOA) fromEq. 8.2. In the same way a TOA cloud liquid frac-tion is computed from the cloud liquid fractions atthe different levels. Finally the GCM output is sam-pled along the satellite overpass which is calculatedon-line in the model. Because the physical timestepis 30 minutes in our model, the sampled cloud fieldsare within ±15 minutes of the actual satellite obser-vations. Therefore we do not expect any bias in theGCM due to the diurnal cycle.

8.3. Results

8.3.1. Cloud phase to cloud temperature

relationship in the satellite

observations

We estimate the average liquid cloud fractionfor each 10 K bin of cloud Rayleigh tempera-ture in the POLDER retrievals. Fig. 8.2 showsthe phase to temperature relationships for the 8-month POLDER period. Looking at shorter peri-ods or particular regions gives rather similar results.

-50 -40 -30 -20 -10 0 10 20 30Cloud top Rayleigh temperature [°C]

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

Liqu

id w

ater

frac

tion

100

106

101

103

102

104

105

Num

ber o

f poi

nts

-50 -40 -30 -20 -10 0 10 20 30Cloud top Rayleigh temperature [°C]

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

Liqu

id w

ater

frac

tion

100

106

101

103

102

104

105

Num

ber o

f poi

nts

Figure 8.2.: Fraction of liquid phase at cloud top as afunction of the cloud Rayleigh tempera-ture in the POLDER observations at (a)the POLDER resolution and (b) at theGCM resolution. The dashed line showsthe number of points in each bin (rightscale).

It is noteworthy, however, that the relationship ismuch less smooth if POLDER data are not regriddedto the GCM resolution due to the larger heterogene-ity in the cloud properties. It is therefore importantto first decrease the resolution of satellite data beforecomparing with the GCM results. It is due to thecloud sampling that even for temperatures above themelting point small fractions of ice clouds exist. Thiswould be the case e.g. in tropical regions with deepconvection where a cumulonimbus may exist alongwith low level trade cumuli. Our results differ slightlyfrom those of Giraud et al. (2001), who found less iceclouds at temperatures close to 0◦C.

109

8.3. Results 8. Evaluation of cloud phase parameterization

-50 -40 -30 -20 -10 0 10 20 30Cloud top Rayleigh temperature [°C]

0

0.2

0.4

0.6

0.8

1

Liqu

id w

ater

frac

tion

100

106

101

103

102

104

105

Num

ber o

f poi

nts

LMDZLMDZLMDZLMDZ

Figure 8.3.: Same as Fig. 8.2 but for standard LMDZmodel results.

8.3.2. Model relationships

Fig. 8.3 shows the same relationships but for model-simulated quantities. An obvious deficiency is the toosteep transition between 0% and 100% liquid waterfraction in a temperature range of only about 50◦C.The simulated phase to temperature relationship isvery robust and does not vary much when looking atdifferent regions or periods. Particularly, the resultof a short simulation of a few days is not differentfrom that from a 8-month simulation.

8.3.3. Parameter fit

Keeping the parameterization for the repartition ofcondensed water into liquid water and ice (Eq. 8.1),we try to find the values of the two parameters Tice

and nx which match best the POLDER observations.The upper bound of 0◦C is kept as larger values forthe melting point would not be physically reasonable.We vary Tice between -70◦C and 0◦C and nx between0.1 and 10. We carry out three-day simulations forthe period of June 10-12, 1997 for each pair of param-eters. The resulting phase to temperature relation-ships can well be fitted with an hyperbolic tangentfunction of the form:

xliq = (1 + tanh(a1T + a2))/2 (8.3)

with a1 and a2 as fitting parameters determining thedegree of flatness of the curve and a shift to lowertemperatures, respectively. The fitted hyperbolictangent functions for the POLDER observations andthe standard model parameterization are superim-posed to Figs. 8.2 and 8.3.

0 1 2 3 4 5 6 7Parameter a2 (shift)

0

0.1

0.2

Par

amet

er a

1 (fla

tteni

ng)

nx 0.1 0.2 0.5 1 2 5 10l n u s + x ∗

Tice -70oC -60oC -50oC -40oC -35oC -30oC -25oC -20oC -15oC -10oC -5oC 0oCn n n n n n n n n n n n

nx 0.1 0.2 0.5 1 2 5 10l n u s + x ∗

Tice -70oC -60oC -50oC -40oC -35oC -30oC -25oC -20oC -15oC -10oC -5oC 0oCn n n n n n n n n n n n

nx 0.1 0.2 0.5 1 2 5 10l n u s + x ∗

Tice -70oC -60oC -50oC -40oC -35oC -30oC -25oC -20oC -15oC -10oC -5oC 0oCn n n n n n n n n n n n

nx 0.1 0.2 0.5 1 2 5 10l n u s + x ∗

Tice -70oC -60oC -50oC -40oC -35oC -30oC -25oC -20oC -15oC -10oC -5oC 0oCn n n n n n n n n n n n

Figure 8.4.: Scatter plot of the a1 and a2 parame-ters fitting the top-of-atmosphere cloudliquid fraction to temperature relation-ships. space. On the x-axis, parametera2, and on the y-axis, parameter. Theopen circle is for the POLDER observa-tions while the other symbols depict therelationships obtained in the GCM sim-ulations with varying Tice and nx.

Fig. 8.4 shows the results of the different simula-tions in the parameter space of a1 and a2. Firstly,a temperature below which all cloud water is ice ofrather low value is needed. The fit corresponds to atemperature of Tice = −32◦C. It is interesting thatin Fig. 8.2 considerable liquid fractions are shownfor temperatures below –35◦C, which is due to thesampling at the model grid resolution. The freezingtemperature found is slightly larger but rather closeto the values found by observational studies (Sassenand Dodd , 1988) or used in high resolution models(Spice et al., 1999). Secondly, a concave function hasto be chosen to match the observations, but the ex-ponent is smaller than so far used in the model.

110

8. Evaluation of cloud phase parameterization 8.4. Conclusions

-80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30Rayleigh temperature [°C]

0

0.2

0.4

0.6

0.8

1Li

quid

wat

er fr

actio

n

POLDERLMDZLMDZ, new formula

Figure 8.5.: Regressions of the phase to temperaturerelationship as from Figs. 8.2 (blackcurve) and 8.3 (red) for the 8-monthPOLDER period, and for a simulationusing the new pair of parameters (green)

8.4. Conclusions

Using data derived from the spaceborne instrumentPOLDER-1, we establish a relationship betweencloud top temperature and cloud top thermodynam-ical phase. We simulate the same period with theLMDZ GCM by nudging the winds and temperatureto ECMWF reanalysis data. The model’s parameter-ization for the thermodynamical phase of condensedwater, a distribution according to local temperature,is tested. We carried out multiple simulations varyingtwo parameters and identified the best fit of modelto POLDER data. The standard parameterizationfor the liquid and ice fraction clearly is too steep andcan not capture the fraction of liquid water cloudsobserved even at very low temperatures. The curveresulting from the “best fit” to the observationaldata obtained from short simulations using differentparameters is able to reproduce well the observa-tions (Fig. 8.5). The temperature where all cloudwater consists of ice is found to be Tice = −32◦C,which is close to observed values. The exponentof the transition function is determined as nx=1.9.

EQ30S60SSP 60N NP30NLatitude

-100

-90

-80

-70

-60

-50

-40

-30

-20

-10

0

SW

CR

F [W

m-2

]

ScaRaBStandardNew parameters(a) SW(a) SW(a) SW

EQ30S60SSP 60N NP30NLatitude

0

10

20

30

40

50

60

LW C

RF

[W m

-2]

ScaRaBStandardNew parameters

(b) LW(b) LW(b) LW

Figure 8.6.: Annual cloud radiative forcing [Wm−2].(a) Long-wave and (b) Short-wave forScaRaB observations (black, solid), theStandard model scheme (blue, dash-dotted), and the new pair of parameters(red, dashed).

These two values are different from those arbitrar-ily chosen in the standard model parameterizationof the LMDZ, but are rather close to values usedin other GCMs. We carried out simulations of oneyear for both, the old and new pairs of parameters(Tice and nx). Fig. 8.6 shows the zonal annual meancloud radiative forcings (CRF) in the shortwave andlongwave spectra, comparing the two model versionswith ScaRaB satellite measurements for March 1994to February 1995 (Kandel et al., 1998). While in theSW spectrum, the new parameterization simulates acloud radiative forcing closer to the observations ex-cept for southern hemisphere midlatitudes, a generalbias of the model to simulate too strong CRF in thelongwave spectrum is slightly enhanced at midlati-tudes when using the new parameters.

111

8.5. Resume 8. Evaluation of cloud phase parameterization

8.5. Evaluation de la

parametrisation de la phase

thermodynamique dans le

MCG LMDZ en utilisant des

donnees satellitaires

POLDER : Resume

Introduction

Les processus de nuages et de precipitation sont assezdifferents suivant la phase thermodynamique danslesquels ils ont lieu. De plus, le transfert radiatifest influence differemment dans les nuages de glaceet dans les nuages d’eau liquide. Parmi d’autres,l’etude presentee dans les derniers deux chapitres ade nouveau montre les reactions differentes des nu-ages de glace et d’eau liquide aux forcages externes.

Ce sont la quelques-unes des raisons pour lesquelles laparametrisation de la phase thermodynamique dansdes modeles de grande echelle est d’une importanceparticuliere. En absence d’une representation ex-plicite de la distribution de l’eau condensee en eauliquide et glace, l’indicateur le plus important pourla phase thermodynamique de l’eau condensee estla temperature locale. La parametrisation la plussimple consiste a fixer une temperature au-dessus delaquelle l’eau est liquide et au-dessous de laquelle elleest solide. Pourtant, des observations montrent quetoute l’eau est bien liquide au-dela de 0◦C, alors quede l’eau liquide peut etre trouvee a des temperaturesaussi froides que –40◦C (Houze, 1993). Beaucoupde modeles a grande echelle definissent donc unetemperature, au-dessus de laquelle toute l’eau est liq-uide (dans la plupart des cas, 0◦C), une temperatureTice, au-dessous de laquelle toute l’eau est congelee,et ils partagent l’eau condensee dans la maille dela grille entre eau liquide et glace en suivant uneformule exponentielle entre ces deux temperatures.Dans cette etude, nous cherchons a fixer les deuxparametres de cette parametrisation, qui sont Tice etl’exposant de la fonction de transition.

Methode

Nous utilisons pour cela des observations spatialesde l’instrument POLDER de la temperature et laphase thermodynamique au sommet des nuages (voirchapitre 3). Dans une etude similaire avec desdonnees combinees de la phase thermodynamiquederivee par POLDER et la temperature de bril-lance donnee par l’instrument ATSR-2, Giraud etal. (2001) ont montre une forte correlation entre latemperature au sommet des nuages et la phase ther-modynamique comme elle est derivee par POLDER.Nous simulons avec le modele LMDZ la periodede mesures de POLDER (Novembre 1996 a Juin1997) en mode guide et avec des SST observeesimposees. La temperature et la phase thermody-namique au sommet des nuages sont echantillonneesen utilisant l’hypothese de recouvrement aleatoire surla fauchee du satellite. La relation statistique entre latemperature et la phase thermodynamique au som-met des nuages est etablie a la fois dans les sortiesdu modele et dans les observations POLDER. Nousajustons ensuite une tangente hyperbolique a deuxparametres a cette relation. Comme nous l’avonsmontre, cette relation est tres stable dans le modele,et elle est tres similaire que l’on l’etablisse a par-tir des sorties de quelques jours, plusieurs jours, outoute la periode. Nous effectuons alors plusieurs sim-ulations courtes en variant les deux parametres dela parametrisation de la phase thermodynamique enfonction de la temperature locale dans le modele etdeduisons les deux parametres de la tangente hyper-bolique ajustee.

Resultat

Avec cette methode, nous trouvons un tres bon ac-cord entre modele et les observations pour le choixde –32◦C comme Tice et une transition suivant unefonction exponentielle avec un exposant de 1.7. Cesdeux valeurs sont differentes de celles choisies parhasard dans la formule standard du LMDZ (–15◦Cet 6), mais sont proches des parametres utilises dansd’autres modeles (Tice = −35◦C ; Del Genio etal., 1996 ; Lohmann et Roeckner , 1996 ; nx=2,Smith et al., 1990). Nous avons effectue deux sim-ulations en imposant la SST, une avec le jeu de

112

8. Evaluation of cloud phase parameterization 8.5. Resume

parametres � standard� , l’autre avec les deux nou-veaux parametres. Comme nous avons pu le montreren comparant les sorties du modele aux observationsde ScaRaB (Kandel et al., 1998), le forcage radiatif

des nuages est beaucoup mieux represente dans lespectre des ondes courtes en utilisant les nouveauxparametres, mais il reste tres similaire dans le spec-tre des ondes longues (fig. 8.6).

113

8.5. Resume 8. Evaluation of cloud phase parameterization

114

9. A new liquid- and ice phase microphysical

scheme for the LMDZ GCM: Description and

preliminary results

9.1. Introduction

The term “microphysics” in the global atmosphericmodeling context refers to the physical processes con-nected with clouds and precipitation, beginning fromnucleation of cloud particles and condensation, toformation of different forms of precipitation, to thefall and possible evaporation of precipitation. For allthese processes, individual particles have to be con-sidered, having a typical size scale of some microm-eters. Nevertheless, it is by these processes that allclouds and precipitation appear, evolve, and disap-pear.Clouds play a fundamental role in the atmosphericenergy balance. They increase the shortwave plane-tary albedo, a cooling effect, and on the other hand,they increase the longwave greenhouse effect, warm-ing the surface. In addition, they distribute latentheat in the atmosphere. The different representationof clouds is a major source of uncertainty in climatechange scenario simulations with different general cir-culation models (Cess et al., 1990; Le Treut and McA-vaney , 2002).Precipitation is one of the main climate variables.Nevertheless, climate models are so far not able tocoherently simulate the response of precipitation toanthropogenic climate change. This, however, wouldbe one of the most important predictions to provideto the public.One of the reasons for these shortcomings may be theinsufficient parameterization of microphysics in themodels. There has been some experience in modelingof liquid water microphysics in GCMs with simple for-

mulations for ice-phase microphysics (e.g., Boucheret al., 1995a; Del Genio et al., 1996). Some globalmodels also contain microphysical formulations forice- and mixed-phase clouds (Lohmann and Roeck-ner , 1996; Fowler et al., 1996; Rotstayn, 1997; Ghanet al., 1997a; Wilson and Ballard , 1999). In thisstudy, we develop a comprehensive liquid- and ice-phase cloud microphysical scheme and apply it inthe Laboratoire de Meteorologie Dynamique (LMD-Z) GCM. We replace thus the former microphysicalscheme of Boucher et al. (1995a), which consisted offormulations for autoconversion and accretion of liq-uid clouds, and a simple fall-velocity dependent snowformation equation. Liquid water droplet numberconcentration so far was linked to sulfate aerosol massusing an empiric equation (Boucher and Lohmann,1995).The new scheme is described in Section 9.2. Finally,in Section 9.3, we show results from simulations usingthe new scheme compared to the former one.

9.2. Description of the scheme

In the new microphysical scheme, water is dividedinto categories of vapor, cloud water, and precipitat-ing water. Cloud water is further split in liquid waterand ice. For precipitating water, liquid water (rain),slowly-falling (snow) and fast-falling (graupel) icewater are separated. The mixing ratios of these sixwater species are treated prognostically in the model.For cloud water, in addition, the number concentra-tions of liquid droplets and ice crystals are treated as

115

9.2. Description of the scheme 9. A new microphysical scheme

Activa

tion

Evapo

ratio

n

Conde

nsati

on

Evaporation

NucleationSublimation

Freezing

MeltingE

vapo

ratio

n

rain

vapor

liquid iceclouds

precipitation

collisions

Freezing

Melting

Melting

Hallett−

Mossop

Hal

lett−

Mos

sop

�������������������������

�������������������������

�����������������������������������

�����������������������������������

rl�������������������������

�������������������������

�������������������������

�������������������������

ll����������

����������

�����������������������������������

�����������������������������������

ri � � � � � � � � � �

�������������������������

�����������������������������������

�������������������������

gl

vapor

snow

�������������������������

�������������������������

�������������������������

�������������������������

ii

������������������������������

������������������������������

�������������������������

�������������������������

si

�����������������������������������

�����������������������������������

�����������������������������������

�����������������������������������

rs

graupel

������������������������������

������������������������������

������������������������������

� � � � � � � � � � � �

sl

Figure 9.1.: The six water categories of the model and transition processes between them.

ql Cloud liquid water mixing ratio [kg kg-1]qi Cloud ice mixing ratio [kg kg-1]qr Rain water mixing ratio [kg kg-1]qs Snow mixing ratio [kg kg-1]qg Graupel mixing ratio [kg kg-1]Nl Cloud droplet number concentration [m-3]Ni Ice crystal number concentration [m-3]

Table 9.1.: Prognostic variables for water species in the model.

prognostic variables. Precipitating water, however,is assumed to follow a Marshall-Palmer distribution(Marshall and Palmer , 1948; Gunn and Marshall ,1957). While water vapor and cloud water are trans-ported by the large scale dynamics, precipitation isconsidered to remain in the grid column of its origin.This is a valid assumption because the time for pre-

cipitation (fall speed times height) is small comparedto the time for advection to an adjacent grid box(horizontal wind speed times horizontal resolution).In future studies, however, a possible impact of trans-portation of precipitating water shall be examined.Table 9.1 summarizes the prognostic variables of themodel.

116

9. A new microphysical scheme 9.2. Description of the scheme

Process Description Variables con-cerned

Condensation / Evapora-tion / Sublimation

Transformation of water vapor to liquid waterand ice and vice versa

qv, ql, qi, Nl, Ni.

Activation/Nucleation Creation of new droplets and new crystals Nl, Ni, qv, ql, qi

Collisions/Coalescence Transformations between liquid water, ice, andprecipitating species

ql, qi, qr, qs, qg, Nl,Ni

Freezing / Melting Conversions between liquid water and ice and be-tween rain and snow

ql, qi, qr, qs, Nl, Ni

Evaporation of precipita-tion

Conversion of precipitating species to water vapor qv, qr, qs, qg

Table 9.2.: Processes in the new microphysical scheme.

The scheme treats five types of processes. Condensa-tion and evaporation are the transitions between wa-ter vapor and cloud liquid water, and sublimation thetransition between water vapor and cloud ice. Newcloud particles may form via activation of aerosolsor nucleation for liquid droplets and ice crystals.Cloud and precipitation particles may collide witheach other and coalesce, which yields a transitionto another water category. When rime is produced,new ice splinter are created by the secondary ice pro-duction mechanism. Finally, precipitation particlesmay fall, and when falling through sub-saturated airmasses, they may evaporate. Table 9.2 lists the fiveprocesses in the order they are treated in the model,and Fig. 9.1 schematically visualizes the processesconnecting the different water categories. The mi-crophysical processes are described in more detail inthe following section.

9.2.1. Condensation/Evaporation

To calculate condensation and evaporation of cloudliquid water and cloud ice, a scheme based on LeTreut and Li (1991) is applied. The scheme applies aprobability density function (PDF) to the total wa-ter content (i.e., cloud liquid and ice water are re-evaporated before applying the PDF). The satura-tion water vapor mixing ratio is calculated using theTetens-Formula (Tetens, 1930):

qsat =Rd

Rvesat(T )

p−(1− Rd

Rv

)esat(T )

(9.1)

where

esat(T ) = a1 exp[a3

(T − T0

T − a4

)](9.2)

with the empirical constants a3 and a4 being differentfor T < T0 and T > T0. A parameter, ∆qs

, deter-mines the width of the PDF of the total water mixingratio distribution. This parameter varies vertically inthe range [0.4-0.99], but is constant horizontally andtemporally. Figure 9.2 shows schematically the PDFin a grid box. The cloud fraction of the grid box iscalculated as

f =qt(1 + ∆qs

)− qs

2 ∆qsqt

(9.3)

and is restricted to f ∈ [0, 1]. The total water mixingratio in the cloudy part of the grid box is

qt,cloud =qt(1 + ∆qs

) + qs

2(9.4)

The fraction of qt that exceeds the saturation watervapor mixing ration in the cloudy part, is condensed.The grid-box mean condensed water is thus:

qcond = (qt,cloud − qs)1

1 + Lv

cp

∂qsat

∂T

f (9.5)

=(qt(1 + ∆qs

)− qs)2

4 ∆qs qt

11 + Lv

cp

∂qsat

∂T

(9.6)

where the last factor on the right hand side of thisequation accounts for the latent heating in the con-densation process. The derivative of saturation watervapor mixing ration w.r.t. temperature is calculated

117

9.2. Description of the scheme 9. A new microphysical scheme

PD

F(q

) t

Dqsqt½

Dqs(1− )qt qt, clear qt qs qt, cloud qt Dqs(1+ )

������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

���

���

Figure 9.2.: The PDF of total water mixing ratio.

from the Clausius-Clapeyron equation. Lv is replacedby Ls for T < T0 and qcond is restricted to values be-tween 0 and qt. The condensed water is divided intocloud liquid water and cloud ice depending on thelocal temperature. The ice fraction is defined as

xliq =

(

T−TiceT0−Tice

)nxT ∈ [Tice, T0]

0 T < Tice

1 T > T0

(9.7)

where Tice and nx are parameters. Their values re-cently have been fixed using POLDER satellite datato Tice = −32◦C and nx = 1.7 (Boucher and Quaas,2003, see Chapter 8).

9.2.2. Activation of droplets

The formulation of cloud droplet activation fromaerosol is of major importance to simulate the aerosolindirect effects, i.e., the impact of aerosols on clouds.We use the Twomey (1959) formula

dNl

dt

∣∣∣act

= c2

2+k

(1.63× 10−3 w1.5

k B(

32 , k

2

) ) kk+2

(9.8)

where B(x1, x2) is the Beta-function, k the Twomeyconstant, and c is the concentration of aerosols ac-tivated at 1% supersaturation. We set c to 90% ofthe hydrophilic aerosol number concentration. Tak-ing a constant value is a valid limitation, accordingto Junge and McLaren (1971) and Fitzgerald (1973).The parameter k is chosen to be k = 0.2, which issomewhat smaller than the values given by Twomey(1959). However, he looked at convective clouds withrather strong updraft velocities, so a smaller valueis supposed to be valid for the large scale. New

cloud droplets are activated in the part of the gridcell which is supersaturated, i.e., the cloudy fractionas defined earlier (Section 9.2.1). The grid-box meanvertical wind, w, generally is close to zero and of-ten slightly negative. Thus, the subgrid scale verticalwind in the cloud has to be evaluated. To do this,we assume subsidence to take place in the cloud-freefraction of the grid cell. When w is negative, thesubsidence in the cloud free part, w ↓cf, is calculatedfrom the mean velocity by multiplying it with a ran-dom number between 1 and 2 – an approach similarto that of Donner et al. (1997). When the mean ver-tical velocity is very slightly negative or positive, theclear-sky subsidence is assumed to be w ↓cfmin= –1.2cm s-1 (which is a typical value for subsidence accord-ing to Suhre et al., 2000):

w ↓= min(w × (1 + r), w ↓min) (9.9)

where r is a random number between 0 and 1. Theupdraft in the cloudy part of the grid cell is deter-mined as the residual:

wcc ↑= w − (1− f)wcf ↓f

(9.10)

where indices “cf” and “cc” refer to the “cloud-free”and “cloud-covered” parts of the grid cell, respec-tively.

9.2.3. Nucleation of ice crystals

Little is known about the process of nucleation of icecrystals from aerosol. In general, two different waysare distinguished (Young , 1974b; Vali , 1985)):

• Homogeneous nucleation, which is the freezingof supercooled liquid water droplets. Homoge-neous nucleation occurs at temperatures below–35◦C (Sassen and Dodd , 1988; Spice et al.,1999).

• Heterogeneous nucleation, including the actionof a solid aerosol. Three modes of functioningexist:

– Contact nucleation, a collision between asupercooled drop and an aerosol particle.

– Deposition nucleation, where water vapordeposits on a dry or wetted aerosol.

118

9. A new microphysical scheme 9.2. Description of the scheme

– Immersion nucleation, the freezing of adroplet including an aerosol with an in-soluble component.

The nucleation processes where ice crystals form byfreezing of liquid droplets are treated in the freezingpart of the scheme. From the nucleation processesgiven above, the nucleation by deposition nucleationis the only one explicitly treated by the nucleationscheme in the parameterization presented here. Ac-cording to Meyers et al. (1992), ice crystal numberconcentration can be described in dependence on thesupersaturation:

NID = exp (a + b Si) (9.11)

where Si is the supersaturation over ice in %, anda = −0.639 and b = 0.1296 are constants. The su-persaturation of the cloudy part of the grid is used,calculated from the grid-box mean supersaturationand an assumed sub-saturation in the clear part ofthe grid box, which varies between a minimum valueand twice the grid-box mean value. Similarity to thesubgrid updraft velocity calculated for the droplet ac-tivation scheme, the supersaturation in the cloudypart of the grid box is thus given as

Scci =

Si − (1− f)Scfi

f(9.12)

9.2.4. Collisions/Coalescence

Particles may collide and potentially coalesce. As-suming a constant size distribution for all particles,the number of particles that collide per volume andtime is

rpq = Epq Np Nqπ

4(Dp + Dq)

2∣∣∣Vp − Vq

∣∣∣ (9.13)

where all quantities are averaged over the size distri-bution, and p and q refer to the particle species (liquidwater, ice, rain, snow, or graupel), and Epq are thecollision efficiencies (see Table 9.4; the efficiencies aretaken from Rutledge and Hobbs (1983) except for theautoconversion cases Ell and Eii). The terminal fallvelocities V of the different species are listed in Sec-tion 9.2.6. While number concentrations N are cal-culated prognostically for droplets and ice crystals,the precipitation species are assumed to follow the

Marshall-Palmer distribution, so their number con-centration can be estimated from the precipitatingwater mixing ratio. As mentioned above, droplets,raindrops, and graupel are considered to be spher-ical, and thus the effective diameter is determinedfrom the mixing ratio with the fixed density of theparticles, while ice crystals and snow flakes are con-sidered as plates (Rutledge and Hobbs, 1983), with adiameter [m] of

Dp = 16.3√

ρair qp

Np(9.14)

Similarly to other schemes, Eq. 9.13 is changed forp=q (i.e., for autoconversion of droplets and crystals)to become

rpp = H(Dp −Dp0) N2p

π

4D2

p Vp (9.15)

In this particular case, the collision efficiency is aHeaviside function, which is 0 below a threshold “crit-ical diameter” and 1 above. The critical diameteris chosen as 14 µm for liquid droplets and 70 µmfor ice crystals. Secondary ice particle productionis accounted for (Hallett and Mossop, 1974; Mossop,1985). When rime is formed, i.e., when coalescencebetween graupel and liquid droplets, or between rainand snow takes place, a new ice crystal of initial massof mi0 = 10−12 kg is produced per kg of new rime.

9.2.5. Freezing/Melting

Where the temperature exceeds 0◦C, ice crystals andsnow flakes melt to form droplets and raindrops, re-spectively.For temperatures below 0◦C, we treat the heteroge-neous freezing by relaxing the ice crystal number con-centration to the empirically temperature-dependentvalue from Meyers et al. (1992):

NIC = exp (a− b T ) (9.16)

where T is the temperature in ◦C, and a = −2.80 andb = 0.262 are empirical constants. On the expense ofdroplets and raindrops, the number concentration ofice crystals and snowflakes, respectively, is then re-laxed towards the maximum concentration possiblegiven by the freezing nuclei. For temperatures be-low –35◦C, liquid droplets and rain particles freezeimmediately.

119

9.2. Description of the scheme 9. A new microphysical scheme

droplet crystal raindrop snowflake graupeldroplet raindrop N/A raindrop snowflake graupelcrystal - snowflake graupel snowflake N/A

raindrop - - N/A graupel N/A

Table 9.3.: Particles that may collide. Coalescence for a particle in a row with the one in the column gives aparticle of the form mentioned. “N/A” indicates that a collision between the two particles cannotoccur in the model.

droplet crystal raindrop snowflake graupeldroplet H(r − rl,0) N/A 1. 1. 1.crystal - H(D −Di,0) 1. 0.1 N/A

raindrop - - N/A 1. N/A

Table 9.4.: Coalescence efficiencies for the collisions considered. “H(x − x0)” is the Heaviside function,which is 0 for x < x0 and 1 otherwise. Values are from Rutledge and Hobbs (1983) except for thetwo autoconversion cases.

9.2.6. Falling of precipitation

Different from cloud droplets and ice crystals, thethree precipitation species may fall. The terminalfall velocity of raindrops is calculated as

Vr = C2 r − C3 (9.17)

where C2=86.2 and C3=0.155 ms−1 are constantsand r is the raindrop radius (valid for 35.9µm< r <

300µm, Johnson (1982). For snowflakes consideredas “sideplanes”, Locatelli and Hobbs (1974) give thefall-speed

Vs = a (D × 103)b (9.18)

with the empirical constants a=0.82 ms−1 andb=0.12. Finally, graupel falls at a speed of

Vg = a (D × 103)b (9.19)

with the constants a=1.2 ms−1 and b=0.65 (“Cone-shaped graupel”; Locatelli and Hobbs, 1974). Allquantities are given in SI units. For the explicitfall scheme to be numerically correct, a sub-timestepis introduced for the microphysical scheme. Thetimestep in this scheme is 90 s.

9.2.7. Precipitation fractions

To somehow take into account the subgrid-scale vari-ability, the given fractional cloudiness is exploited

to calculate a precipitation fraction for each of theprecipitation species. When precipitation originatesfrom autoconversion of droplets or ice crystals, thisprecipitation is assumed to be homogeneously dis-tributed throughout the cloud. The precipitationfraction changes when new rain is formed in a layer,when the respective precipitation species is totallyevaporated, or when due to a conversion between twodifferent precipitation species one of them is entirelyconsumed. When rain falls, random overlap betweenthe precipitation and cloud fractions is assumed con-sistently with the cloud overlap assumption in themodel.

9.2.8. Evaporation

The evaporation scheme is adapted from Boucheret al. (1995a):

Ep = −α

2

(qcft

qs− 1

)qp

ρwater r2p

(9.20)

where the index p refers to the respective precipita-tion species (rain, snow, or graupel), and α = 0.5as in Boucher et al. (1995a). In contrast to theirscheme, however, the size of precipitation particles iscalculated using the Marshall-Palmer distribution.

120

9. A new microphysical scheme 9.3. Results of 1D case studies

-80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40Temperature [oC]

100

200

300

400

500

600

700

800

900

1000

Hei

ght [

hPa]

T (26th June)

T (9th July)

0 1 2 3 4 5 6 7 8 9 10 11 12Water vapor mixing ratio [g kg-1]

100

200

300

400

500

600

700

800

900

1000

Hei

ght [

hPa]

qv (26th June)

qv (9th July)

Figure 9.3.: Initial profiles of temperature and wa-ter vapor mixing ratio for the two ACE-2cases (blue, June 26th; red, July 9th).

9.3. Results of 1D case studies

We test this new parameterization in two test casesusing a single column model (SCM) version of theLMDZ GCM. This approach has formerly been usedin Chapter 4.

9.3.1. Test cases for stratocumulus

clouds: ACE-2

To test the liquid water part of the new microphysi-cal scheme, the ACE-2 CLOUDYCOLUMN cases arechosen (Brenguier et al., 2000a). Two of the ACE-2cases have formerly been used to test the parame-terization of the aerosol-cloud interactions in SCMversions of several GCMs (Chapter 4; Menon et al.,2003).

0 6 12 18 2402468

Pr [m

m d

ay-1

]

0

10203040

τ c [-]

0

100

200

W [g

m-2

]

0

5

10

r e [µm

]

0

30

60

90

120

Nd [c

m-3

]

020406080

100

f c [%]

Sfc precipitation flux [mm day-1]Sfc precipitation flux [mm day-1]Sfc precipitation flux [mm day-1]Sfc precipitation flux [mm day-1]

COT [-]

Sfc precipitation flux [mm day-1]

COT [-]

Sfc precipitation flux [mm day-1]

COT [-]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Cloudiness [%]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Cloudiness [%]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Cloudiness [%]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Cloudiness [%]

Figure 9.4.: Results of a SCM simulation for theclean ACE-2 case (June 26th, 1997) us-ing the new microphysical scheme (redsolid line), and using the microphysi-cal parameterization of Boucher et al.(black, dotted line). Superimposedcrosses show the observations. Forcloudiness, ISCCP satellite observationsare shown as blue dashed line.

Aerosol properties have been measured along withcloud microphysical properties and radiation (Bren-guier et al., 2000b). Observational values for theaerosols are summarized in Table 4.1 (Guibert et al.,2003), and for microphysical cloud properties in Table4.2. Figures 4.1 to 4.6 summarize the results from thesingle column versions of six different GCMs, includ-ing the LMDZ GCM with the microphysical scheme(Boucher et al., 1995a) and the empirical formulato link CDNC to sulfate aerosol mass (Boucher andLohmann, 1995).

121

9.3. Results of 1D case studies 9. A new microphysical scheme

0 6 12 18 24012345

Pr [m

m d

ay-1

]

01020304050

τ c [-]

0

100

200

W [g

m-2

]

0

2

4

6

8

r e [µm

]

050

100150200250300

Nd [c

m-3

]

020406080

100

f c [%]

Sfc precipitation flux [mm day-1]Sfc precipitation flux [mm day-1]Sfc precipitation flux [mm day-1]Sfc precipitation flux [mm day-1]

COT [-]

Sfc precipitation flux [mm day-1]

COT [-]

Sfc precipitation flux [mm day-1]

COT [-]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Cloudiness [%]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Cloudiness [%]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Cloudiness [%]

Sfc precipitation flux [mm day-1]

COT [-]

LWP [g m-2]

CDR [µm]

CDNC [m-3]

Cloudiness [%]

Figure 9.5.: As Fig. 9.4, but for the polluted ACE2case (July 9th, 1997).

Figures 9.4 and 9.5 show the results of the LMDZGCM in its single column version including the newmicrophysical scheme for the ACE-2 CLOUDYCOL-UMN clear (June 26th, 1997), and polluted cases(July 9th, 1997), respectively. As aerosol data inEq. 9.8, the total number concentration of sulfate,organic carbon, and sea salt is taken. The observa-tions of the ACE-2 field study are shown as crosses,and for the cloud amount, also ISCCP cloud cover isdisplayed (Rossow and Schiffer , 1991). Along withthese data, we show the results of the LMDZ SCMusing the microphysical scheme of Boucher et al. asin Chapter 4. The results are very encouraging, asseveral of the deficiencies of the old microphysicalscheme and of schemes of other GCMs identified inthe study reported in Chapter 4 seem to be improvedby the new parameterization.

-80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40Temperature [oC]

100

200

300

400

500

600

700

800

900

1000

Hei

ght [

hPa]

T (GCSS cold-neutral)T (GCSS warm-neutral)

0 1 2 3 4 5 6 7 8 9 10 11 12Water vapor mixing ratio [g kg-1]

100

200

300

400

500

600

700

800

900

1000

Hei

ght [

hPa]

qv (GCSS cold-neutral)qv (GCSS warm-neutral)

Figure 9.6.: Initial profiles of temperature and wa-ter vapor mixing ratio for the two GCSSworking group II cases (blue, warm-neutral; red, cold-neutral).

The new parameterization is very well able to cap-ture the differences between the clean and pollutedcases. Cloud droplet number concentration is a fac-tor of 5 larger in the polluted compared to the cleancase, and the simulated CDNC is in close agreementwith the observations in both cases, where the oldscheme overestimated CDNC in the clean case andunderestimated it in the polluted case. The dropletsize is very well simulated in the clean case and some-what overestimated in the polluted case. However,the scheme still simulates slightly smaller droplets inthe polluted compared to the clean case. Cloud liquidwater path and subsequently cloud optical thicknessis much better simulated by the new scheme than bythe former one, which overestimated both quantitiesby almost an order of magnitude. Now, the LWP iseven underestimated in the clean case, and slightlyoverestimated in the polluted case.

122

9. A new microphysical scheme 9.3. Results of 1D case studies

0 1 2 3 4 5 60

0.5

1

1.5

τ c [-]

0

20

40

IWP

[g m

-2]

0

0.2

0.4

0.6

0.8

1

Ni [c

m-3

]

30

35

40

45

50

55

f c [%]

COT [-]COT [-]

IWP [g m-2]

COT [-]

IWP [g m-2]

COT [-]

IWP [g m-2]

Ni [cm-3]

COT [-]

IWP [g m-2]

Ni [cm-3]

Cloudiness [%]

Figure 9.7.: Results of a SCM simulation for the“warm-neutral” GCSS WG2 case forcloudiness, ice crystal number concentra-tion, ice water path, and cloud opticalthickness. In red, the LMDZ GCM, andfor the ice water content, in black, theMeso-NH model.

Cloud optical thickness is very well simulated by thescheme in both cases. An important improvementof the new parameterization is its ability to simu-late the drizzle precipitation observed. Consistentlywith the observations, the model simulates a slightlylower precipitation rate in the polluted than in theclean case.

9.3.2. Test cases for cirrus clouds: GCSS

WG2

In the framework of the Global Energy and WaterCycle Experiment (GEWEX) Cloud System Studies(GCSS), one of the working groups (working group2, WG2) investigates cirrus cloud systems.

0 1 2 3 4 5 60

0.2

0.4

0.6

0.8

τ c [-]

0

5

10

IWP

[g m

-2]

0

0.2

0.4

0.6

Ni [c

m-3

]

40424446485052545658

f c [%]

COT [-]COT [-]

IWP [g m-2]

COT [-]

IWP [g m-2]

COT [-]

IWP [g m-2]

Ni [cm-3]

COT [-]

IWP [g m-2]

Ni [cm-3]

Cloudiness [%]

Figure 9.8.: As Fig. 9.7, but for the “cold-neutral”GCSS WG2 case.

Within this group, a model intercomparison projectis undertaken. A range of models covering parcelmodels to general circulation models are comparedagainst each other. Results have been published forthe parcel models by Lin et al. (2002). It shouldbe noted that even such small-scale models show avery large scatter in results, often covering severalorders of magnitude. Among the meso-scale mod-els, Meso-NH has been evaluated using the GCSSWG2 cases. Meso-NH includes a comprehensive cir-rus clouds microphysical scheme (Thouron, 2003).The two GSCC WG2 cases consist of a “warm-neutral” and a “cold-neutral” test case. In the in-tercomparison, a constant aerosol concentration of200 cm−3 is considered, and for the first four of sixhours of integration, a vertical ascending motion of3 cm s−1 is applied. No radiation is used in thesingle-column test integration. Figure 9.6 showsthe initial profiles of temperature and humidity.

123

9.4. Preliminary 3D results 9. A new microphysical scheme

EQ30S60SSP 60N NP30NLatitude

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

Clo

udin

ess

[%]

ISCCPNew schemeBoucher et al.

(a) Cloudiness(a) Cloudiness(a) Cloudiness

EQ30S60SSP 60N NP30NLatitude

-240

-220

-200

-180

-160

-140

-120

-100

-80

-60

-40

-20

0

SW

CR

F [W

m-2

]

ScaRaBNew schemeBoucher et al.

(b) SW CRF(b) SW CRF(b) SW CRF

EQ30S60SSP 60N NP30NLatitude

-60

-50

-40

-30

-20

-10

0

LW C

RF

[W m

-2]

ScaRaBNew schemeBoucher et al.

(c) LW CRF(c) LW CRF(c) LW CRF

Figure 9.9.: DJF zonal mean (a) cloudiness [%], (b)SW and (c) LW cloud radiative forcing[Wm−2]. Model results with the schemeof Boucher et al. (dashed, red) and thenew scheme (blue, dotted), and observa-tions (solid, black; (a) ISCCP, and (b-c)ScaRaB.

Figures 9.7 and 9.8 show the main results forthe warm-neutral and cold-neutral cases, respec-tively. Along with the results from the LMDZmodel using the new microphysical scheme, re-sults of an integration of Meso-NH are shown forice water path. The GCM simulated a thin cir-rus in both cases, with rather small optical thick-ness. Ice water content is very well simulatedwhen comparing the GCM to the mesoscale model.

EQ30S60SSP 60N NP30NLatitude

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

Clo

udin

ess

[%]

ISCCPNew schemeBoucher et al.

(a) Cloudiness(a) Cloudiness(a) Cloudiness

EQ30S60SSP 60N NP30NLatitude

-240

-220

-200

-180

-160

-140

-120

-100

-80

-60

-40

-20

0

SW

CR

F [W

m-2

]

ScaRaBNew schemeBoucher et al.

(b) SW CRF(b) SW CRF(b) SW CRF

EQ30S60SSP 60N NP30NLatitude

-60

-50

-40

-30

-20

-10

0

LW C

RF

[W m

-2]

ScaRaBNew schemeBoucher et al.

(c) LW CRF(c) LW CRF(c) LW CRF

Figure 9.10.: As Fig. 9.9, but for JJA.

9.4. Preliminary 3D results

Using the new parameterization applying the sameset of parameters as in the one dimensional simula-tions shown in the previous section, we carried outthree-dimensional model simulations with the atmo-spheric GCM, forced with observed SST for 1997(Reynolds and Smith, 1995) and using aerosol sourcedata for ≈ 1990.

In Figs. 9.9 and 9.10 we show the zonal meansfor northern hemisphere summertime (June-July-August, JJA) and wintertime (December-January-February, DJF) for cloudiness and cloud radiativeforcing in the SW and LW spectra. Observationaldata from the International Satellite Cloud Clima-tology Project (ISCCP; Rossow and Schiffer , 1991),averaged over the 1984 to 1992 period is shown forcloudiness and from the Scanner of the Earth Radi-ation Budget (ScaRaB; Kandel et al., 1998) for 1994

124

9. A new microphysical scheme 9.5. Summary and discussion

and 1995 for cloud radiative forcings. The simulationusing the new scheme is compared to a simulationwith the scheme of Boucher et al. (1995a). The re-sults are encouraging, even though of course severaldeficiencies still exist. Cloudiness is well simulatedat most latitudes, slightly overestimating the cloudcover in the southern hemisphere and in the tropics,and underestimating it in the northern hemisphere.This behavior is related to a generally too small lowlevel cloudiness over land and a slightly too large lowlevel cloud cover over oceans. Shortwave cloud radia-tive forcing is overestimated with the new scheme,probably due to the too large low level cloud coverover the oceans. LW cloud radiative forcing how-ever, is captured quite well, and in some cases thenew scheme already does a better job than the oldmicrophysical scheme of the model (e.g., at midlati-tudes in the JJA case). The main deficiency of thenew scheme so far is the strong underestimation ofprecipitation (not shown). As the observations areavailable only over land, only for continental casesthe precipitation is evaluated here. Over oceans,the precipitation simulated by the new scheme isalso somewhat smaller than the one given by the oldscheme, but this is much less pronounced than overland (not shown). Probably, the underestimationin precipitation is linked to the underestimation incontinental low level cloud cover.

9.5. Summary and discussion

The new scheme has been tested on the ACE-2CLOUDYCOLUMN clean and polluted cases, andcompared to observations and to the scheme ofBoucher et al.. The new scheme successfully pre-dicts larger CDNC, smaller CDR, and larger cloudoptical thickness in the polluted compared to theclean case, consistently with the two aerosol indirecteffects. However, due to the much larger LWC in thepolluted case, it does not capture well the observedand expected decrease in precipitation.

To evaluate the parameterizations in the ice phase,

two test cases of the GCSS working group 2 have beenused. The new scheme has been evaluated using asingle-column version of the GCM and comparingthe results to a simulation with the meso-scale modelMeso-NH. The new scheme underpredicts the ice wa-ter content in the cirrus in both cases by an orderof magnitude. Ice crystal number concentration isunderestimated in the warm case, and overestimatedin the cold case. The results are strongly sensitiveto several model parameters as well as to the initialtemperature profile.

A preliminary study of the new scheme in the globalmodel has been presented. Although some deficien-cies of the new parameterization still exist, the resultsare encouraging. Precipitation is strongly underesti-mated by the scheme, and shortwave cloud radiativeforcing is overestimated. However, the cloud cover issimulated rather well, and longwave cloud radiativeforcing is sometimes already better simulated withthe new scheme compared to the previous one. Thisis a particularly interesting result as it suggests thatthe ice phase is simulated well, and that major de-ficiencies still present may rather be linked to liquidphase processes.

Certainly, the degree of evaluation for the new schemeso far is not sufficient. The introduction of this newscheme opens a variety of possibilities to evaluate themodel thoroughly. Quantities to be evaluated includethe column droplet number concentration, ice crystalnumber concentration and size, and different precip-itation species besides the quantities which could al-ready be investigated in the old scheme (as presentedin Chapter 2). Such evaluation and development isplanned for the near future, in particular using along time series of a comprehensive set of ground-based remote sensing measurements from the SiteInstrumental de Recherche par Teledetection Atmo-spherique (SIRTA) near Paris in France as well as us-ing satellite data, particularly for the ice phase froma new TIROS-N Vertical Sounder (TOVS) dataset(see also Chapter 10).

125

9.6. Resume 9. A new microphysical scheme

9.6. Un nouveau schema de

microphysique de phase

liquide et de glace pour le

MCG LMDZ : Description et

resultats preliminaires:

Resume

Introduction

Comme nous l’avons explique dans les dernierschapitres, ainsi que dans l’introduction de la these,la phase glace joue un role essentiel dans les pro-cessus microphysiques des nuages. Il est importantde representer ces processus de maniere realiste dansun modele a grande echelle. De plus, des processusjusque-la peu connus comme des impacts possiblesdes aerosols sur les nuages en glace pourraient etreetudies une fois qu’une representation de la phaseglace dans un schema de microphysique existe. Laprecipitation devrait etre simulee de maniere plusrealiste, car les processus microphysiques de forma-tion de la precipitation dans les nuages de glaceet de phase mixte sont importants pour le taux deprecipitation.

Description du schema

Dans le nouveau schema, l’eau est partagee entre lesclasses de vapeur d’eau, d’eau nuageuse, et d’eau deprecipitation. L’eau des nuages est de plus divisee eneau liquide et glace, et l’eau precipitante est diviseeen eau liquide (de la pluie) et eau solide tombant avitesse lente (de la neige) et rapide (des grelons). Lesrapports de melange sont traites de maniere pronos-tique dans le modele. Pour l’eau nuageuse, les con-centrations en nombre des gouttelettes et des cristauxsont de plus traitees de maniere pronostique. L’eauprecipitante est supposee suivre la distribution deMarshall-Palmer (Marshall et Palmer, 1948 ; Gunnet Marshall, 1957). Alors que la vapeur d’eau etl’eau nuageuse sont transportees par la dynamique agrande echelle, la precipitation est consideree resterdans la colonne de la grille ou elle est creee.

Cinq types de processus sont pris en compte dans

le schema. La condensation/evaporation ainsi que lasublimation sont pris en compte en gardant le schemade Le Treut et Li (1991). Des nouvelles particulespeuvent etre formees par les processus d’activationd’aerosols en gouttelettes, et de la nucleation pourles cristaux. Des particules peuvent tomber en colli-sion et coalescer. Cela forme de la precipitation parautoconversion et peut causer des transformationsentre les differentes especes d’eau nuageuse et deprecipitation. Si des particules givrees sont formees,de nouveaux cristaux sont produits par la formationsecondaire de cristaux (Hallett et Mossop, 1974). Laprecipitation tombe et elle peut s’evaporer quand elletombe a travers de l’air sous-sature.

Comme mentionne, la condensation/evaporation esttraitee suivant Le Treut et Li (1991). L’activationdes gouttelettes a partir des particules d’aerosols sefait avec la formule de Twomey (1959) qui dependde la concentration en nombre des aerosols en faisantune hypothese sur la fraction d’aerosols activee a 1%de sursaturation, et de la vitesse du vent vertical.La derniere doit etre prise en compte a l’echelle dunuage. Dans le schema, nous partageons la vitesseverticale en une subsidence dans la partie claire dela maille, et un panache ascendant dans la partienuageuse, en considerant une subsidence minimale,et en multipliant le vent ascendant par un facteurentre 1 et 2 suivant un chiffre aleatoire. Cette ap-proche est similaire a celle de Donner et al. (1997).La nucleation heterogene des cristaux est prise encompte suivant Meyers et al. (1992), et au-dessousdes temperatures locales de moins de –35◦C, toutesles gouttelettes gelent de maniere homogene dans lemodele.

Le taux de collision entre deux particules depend dela difference des vitesses de chute des deux particules,du carre de la somme des diametres et des concen-trations en nombre des deux especes. La probabilitede coalescence est prise en compte avec un facteurd’efficacite suivant Rutledge et Hobbs (1993), et pourl’autoconversion de gouttelettes et des cristaux, ladispersion de la distribution en taille des particulesest prise en compte en mettant la vitesse de chuted’une des particules en collision a zero et en modifiant

126

9. A new microphysical scheme 9.6. Resume

le facteur d’efficacite en une fonction Heaviside. Pourles particules liquides et les grelons, on considere desspheres, alors que les cristaux de glace et les floconsde neige sont consideres comme etant hexagonaux.Des particules solides fondent immediatement quandla temperature depasse 0◦C. La congelation d’uneparticule liquide depend de la presence des noyauxde congelation au-dessus de –32◦C, dont la concen-tration depend de la temperature suivant Meyers etal. (1992).

L’evaporation de la precipitation est parametriseesuivant Boucher et al., 1995. Pour prendre en comptel’information sous-maille qui vient de la fractionnuageuse a l’origine de la precipitation, nous esti-mons une fraction sous-maille de chaque espece deprecipitation qui est modifiee quand de la nouvelleprecipitation est formee, quand toute la precipitationest evaporee ou transformee ou quand elle tombe.

Resultats

Comme tests unidimensionnels, deux cas de stratocu-mulus marins et deux cas de cirrus ont ete choisis,dont les premiers etaient les cas d’ACE-2 deja decritsdans le chapitre 4, et les dernieres etaient des casde cirrus du groupe de travail 2 du GCSS, un caschaud-neutre, et un cas froid-neutre. Pour les casliquides, le nouveau schema ameliore beaucoup lemodele (figs. 9.4 et 9.4). Le schema est bien ca-pable de simuler les differences entre le cas pollueet le cas propre. La concentration en nombre desgouttelettes est plus large d’un facteur 5 dans lecas pollue compare au cas propre, les deux etant enbon accord avec les observations, alors que l’ancienschema surestimait la concentration dans le cas pollue

et la sous-estimait dans le cas propre. La taille desgouttelettes est bien simulee dans le cas propre etun peu surestimee dans le cas pollue ou le rayonest quand-meme plus petit que dans le cas propre.La colonne d’eau liquide et donc l’epaisseur optiquedu nuage sont beaucoup mieux simulees par le nou-veau schema que dans l’ancien, qui surestimait lesdeux quantites de presqu’un ordre de grandeur. Uneamelioration importante est la capacite du nouveauschema de simuler la bruine observee. En accord avecles observations, le schema simule moins de taux deprecipitation dans le cas pollue que dans le cas pro-pre.

Concernant les cas de cirrus, nous avons pu com-parer les colonnes de glace simulees par le nouveauschema a celles simulees par un modele meso-echelle,Meso-NH, qui contient un schema complet de micro-physique des phases liquide et glace (Thouron, 2003).Dans les deux cas, le nouveau schema du MCG esten bon accord avec le modele a haute resolution (figs.9.7 et 9.8).

Des resultats tres preliminaires ont ete montres pourle cas 3D (figs. 9.9 et 9.10). La couverture nuageuseest bien simulee en comparaison avec l’ancien schemaet compare aux observations ISCCP a l’exceptiond’une surestimation dans l’hemisphere sud et dans lestropiques, et une sous-estimation dans l’hemispherenord, ce qui est du a une couverture nuageuse bassetrop faible au-dessus des continents. Le forcage ra-diatif des nuages est surestime dans le spectre desondes courtes, alors qu’il est bien simule dans lespectre des ondes longues. Le taux de precipitationest largement sous-estime, en particulier au-dessusdes continents.

127

9.6. Resume 9. A new microphysical scheme

128

10. Conclusions and perspectives

10.1. Summary and conclusions

A study of the indirect effects of aerosols on radiationthrough an altering of cloud properties was presented.The main tool used in the study is the general cir-culation model of the Laboratoire de MeteorologieDynamique (LMDZ GCM). A large part of the studyconcerned the evaluation of parameterizations in thismodel using in situ measurements and satellite obser-vations. New parameterizations have been developedand evaluated.

The microphysical scheme of Boucher et al. (1995a)along with the empirical link between sulfate aerosolmass and cloud droplet number concentration follow-ing Boucher and Lohmann (1995) has been imple-mented in the recent version of the LMDZ model.Comparison with the standard model scheme ofclouds and precipitation and an evaluation with satel-lite data shows the skills of the new scheme in realisticsimulation of cloud and precipitation fields.

An intercomparison project, which analyzed resultsof single-column versions of six different GCMs fo-cused on the evaluation of different parameteriza-tions for the aerosol indirect effects, thereby consid-ering also parameterizations of cloud schemes, micro-physics, and radiation. The results of the SCMs werecompared to observations from the ACE-2 CLOUDY-COLUMN field study. Concerning the LMDZ SCM,it was noted that the scheme compares well withthe other models. Major deficiencies common to themodels and evident also for the LMDZ scheme aretoo large cloud geometrical thicknesses, liquid wa-ter paths, and optical thicknesses. From the modelintercomparison, it could be concluded that it isprobably the too coarse vertical resolution which

is responsible for this deficiency. A model with asimple parameterization of vertical subgrid fractionalcloudiness gives results somewhat closer to the obser-vations, which suggests that such an approach mightbe promising. Thus, the development and evaluationof a more comprehensive and realistic scheme forthe parameterization of subgrid variability of wateris a major part of planned future work. All modelsseverely underestimate the drizzle precipitation byseveral orders of magnitude. Thus, a microphysicsscheme able to simulate drizzle at low liquid watercontents would be of importance. The models not be-ing able to realistically simulate the drizzle in neitherpolluted nor clean cases may imply that the secondaerosol indirect effect cannot be well captured by cur-rent GCMs. Several sensitivity experiments with theLMDZ scheme reveal the inability of the examinedBoucher and Lohmann formulae to simulate realisticCDNC for both cases, mainly because it is not sen-sitive enough to changes in aerosol mass. However,none of the other GCM schemes can unambiguouslybe identified to be more skillful either, and in partic-ular, even the most comprehensive physically basedapproaches do not give more realistic results than therather simple Boucher and Lohmann formulae.

A centerpiece of the study was the evaluation ofaerosol indirect effects in satellite observations of thePOLDER instrument and in the LMDZ GCM. Sta-tistical relationships between the POLDER aerosolindex (AI) and effective droplet radius at the top ofhomogeneous liquid water clouds (CDR) have beenestablished in both model and observations. Whilethere is a systematic underestimation of CDR in themodel, the CDR to AI relationship in model and ob-servations compare well. However, a saturation effectat large AI values is found in the observations but not

129

10.1. Summary and conclusions 10. Conclusions and perspectives

in the model. The relationship has been establishedfor several different fixed liquid water path (LWP)conditions. While the observations show flatter CDRto AI relationships for larger LWPs, the model doesnot show a coherent dependence of this relationshipon LWP. Although some correlation can be found formodel-simulated and observed slopes of the CDR toAI relationship for different LWPs and over differ-ent geographical regions, the model is not yet tooskillful. In a second part of the study a relationshipbetween AI and LWP was established. LWP is foundto increase with increasing AI in both model andobservations, which is an indication of the existenceof the second AIE. This relationship is more pro-nounced when looking at the northern hemispheremidlatitudes only. Major shortcomings of the modelinclude a too steep increase in LWP for large AI overland, which might be an overestimate of the secondAIE or be due to the fact that the model does notinclude a semi-direct effect of aerosols. The modelshows a too steep increase in LWP with increasingAI for small AI values when looking at NH midlati-tudes, which might indicate that the autoconversionscheme is too sensitive to changes in CDNC. A mod-eling study further reveals that the general behaviorsof the CDR to AI and LWP to AI relationships arevery similar for pre-industrial and present-day con-ditions. However, the influence of the second AIE issomewhat stronger under present-day conditions.

Having thus validated to a certain degree the mod-eling of the aerosol indirect effects using the micro-physical scheme, the question arises of the impactof aerosol forcings on climate change. A main resultof early modeling studies of anthropogenic climatechange due to greenhouse gas forcing was that sim-ulated warming was much stronger than observedin the 20th century. In particular, for a period ofabout 30 years from 1940 to 1970, the global meansurface temperature has been constant rather thanincreasing. Scientists proposed direct forcing by sul-fate aerosols and indirect forcing as an additionalanthropogenic forcing counterbalancing part of thegreenhouse gas warming. In this study, an analysisof the impacts of both greenhouse gases and sul-fate aerosols on the climate of the 1930-1989 period

is presented. Three different ensemble simulationswere carried out with the LMDZ GCM in a modeforced with observed SST. The first ensemble in-cludes both, greenhouse gas and aerosol forcing, thesecond, greenhouse gas forcing only, and the third, acontrol experiment using just the time-varying SSTwithout additional anthropogenic impacts. The im-pact of greenhouse gases and aerosols on radiationis analyzed, including the indirect impacts via alter-ing of cloud properties. It has been shown that theresults for the radiative forcings compare well withresults of other modeling studies. We defined in ad-dition to the radiative forcing a “radiative impact”of greenhouse gases and aerosols which we calculateas the difference in radiation fluxes between two sce-narios. With this quantity, we analyze the radiativeimpacts of greenhouse gases and aerosols in the solarand terrestrial spectra, and in clear and cloudy skyconditions. The radiative impact of aerosols is foundto be of almost equal magnitude compared to thegreenhouse gas radiative impact. However, while thegreenhouse gas radiative impact is stronger in cloudfree conditions, aerosols have a much larger radiativeimpact in cloudy conditions. For greenhouse gases,it is found that the LW radiative impact is composedof an increase in greenhouse effect in both, clear andcloudy sky, but a decrease in high-level cloudinessreduces the radiative impact in cloudy conditions.Aerosols have a negligible radiative impact in the LWspectrum. In the solar spectrum, greenhouse gasesexert a slightly positive radiative impact due to areduction in cloudiness. Aerosols cause a largely neg-ative radiative impact in clear sky conditions due tothe aerosol direct effect, and in cloudy sky conditionsdue to the first aerosol indirect effect. Additionally,cloud feedback processes are active, as well as thesecond aerosol indirect effect. The sum of these twomechanisms is evaluated to be positive in the globalmean because of a reduction in high-level cloudi-ness due to a cloud feedback. For low-level clouds,however, in particular in northern hemisphere mid-latitudes, clearly a second aerosol indirect effect canbe identified. An increase in low-level cloud lifetimeof up to 2 min day−1 decade−1 can be observed inthe zonal mean for the northern hemisphere.

130

10. Conclusions and perspectives 10.1. Summary and conclusions

Impacts of greenhouse gases and aerosols on ob-served trends are evaluated for the different scenar-ios. Comparing the evolution of global mean landsurface temperatures of the three simulations withobservations we find a rather good agreement for thetrend and variability. Looking at distributions of thelinear trends for the 1950 to 1989 period, we findthat the scenario including both forcings agrees bestwith the observations. In particular for the midlati-tudes of the Eurasian continent, a cooling is observed,which is not simulated except for the scenario whereaerosols are present. A clear trend has been observedfor the diurnal temperature range (DTR), which isdefined as the difference between daily maximum andminimum temperatures. It has formerly been arguedthat this may be due to a negative forcing active inthe SW spectrum, which thus limits the greenhousegas warming on daytime but is not active at night.From the scenarios, we find that both greenhousegases and aerosols result in a decrease in DTR, butthat only the simulation including aerosols is ableto simulate a DTR decrease of the distribution andmagnitude indicated by observations. Further analy-sis shows that the DTR trend is explained to a largedegree by increasing cloudiness due to both forcings,and that for the ensemble simulation including theaerosols, the aerosol direct and indirect effects playan important role as well at least in the midlatitudesof the northern hemisphere. A general conclusion ofthis study is that aerosols played an important roleduring the 1930 to 1989 period, having a radiativeimpact of comparable magnitude as the greenhousegases. Climate trends in the model are altered byaerosol forcings, being in better agreement with theobservations.

The important impacts of both greenhouse gas andaerosol forcings on high-level cloudiness and the re-sulting implications for climate change identified inthe ensemble simulations of the last century highlightonce more the importance of a realistic representa-tion of ice cloud processes in GCMs. Further more,recent research indicates that aerosol impacts oncirrus clouds may be important. Finally, the pre-dominant role of ice and mixed-phase clouds cloudsfor the formation of precipitation is another hint

for the importance of ice cloud parameterizations.This makes the determination of the cloud ther-modynamic phase a particularly important param-eterization of the model. Like most GCMs, LMDZdiagnoses whether condensed water consists of liquidwater or ice by local temperature, which is in prin-ciple a physically reasonable approach. The schemeuses two parameters, the temperature Tice, belowwhich all cloud water immediately freezes, and theshape of the transition between entirely liquid andice phases for the range in between Tice and the melt-ing temperature. Observations of cloud phase andcloud top temperature from POLDER satellite datahave been averaged to the coarser model grid. Inthe model, “satellite-like” cloud top properties werecalculated using nudged simulations, sampling thePOLDER swath, and estimating cloud top proper-ties using a random overlap of fractional clouds inthe vertical. Statistical relationships between ther-modynamical phase and cloud top temperature havebeen established in both model and observations.From a multitude of model simulations, the pair ofparameters is found for which the phase to tem-perature relationship matches best the observations.The values found differ from those used previouslyin LMDZ, but are similar to values in use in otherGCMs. Testing the cloud radiative forcing usingboth the new and old parameterizations, reveals thatthe cloud fields agree better with the observations forthe new parameters than for the arbitrarily chosenvalues used before.

A microphysical scheme has been developed in thepresent study which consistently treats microphysi-cal processes of liquid and ice clouds. The schemeuses mixing ratios of water vapor, liquid and icecloud water, and three different precipitation cate-gories as model variables. For liquid cloud dropletsand ice crystals, the number concentrations are usedas prognostic variables as well. For the activationof liquid droplets from aerosols, a physically basedscheme has been adapted, which uses number con-centrations of sulfate, hydrophilic organic carbon,and submicronic sea salt aerosols and the verticalwind speed using some assumptions on the subgridvariability to determine the cloud droplet number

131

10.2. Perspectives 10. Conclusions and perspectives

concentration (CDNC). For ice clouds, heterogeneousnucleation of crystals is calculated using an empiricalsupersaturation-dependent formulation. Collectionand coalescence processes forming liquid and solidprecipitation, and collection and accretion processescausing increase of precipitation particles and tran-sitions between different precipitation categories aretreated. Secondary ice production by the Hallett-Mossop process is included by a simple parameteri-zation. Precipitation falls and may evaporate whenfalling through clear air. The new scheme is evalu-ated for one-dimensional cases for marine stratocu-mulus clouds (using the ACE-2 data) and for cirrusclouds (using the GCSS WG2 cases). Very encourag-ingly, the new scheme improves several shortcomingsthat were identified for the previous scheme in thestudy reported in Chapter 4. Cloud droplet numberconcentration is very well simulated in both cases.Cloud liquid water path and cloud optical thicknessare much better simulated by the new scheme, beingnow in close agreement with the observations. Per-haps most important is the ability of the new schemeto simulate the drizzle observed in the ACE-2 cases.The parameterization simulates very well the differ-ence between the clean and polluted cases, with muchmore droplets in the polluted case of slightly smallersize, and producing less drizzle. For the cirrus cases,we could compare the ice water path with resultsfrom the mesoscale model Meso-NH. It is very wellsimulated in both cases. Preliminary results of aglobal scale simulation reveal that still several short-comings of the new parameterization exist. However,cloudiness already is well simulated at most latitudes,showing not enough cloud cover for northern hemi-sphere mid and high latitudes, which is due to the toosmall low level cloud cover over land. Precipitationover continents is strongly underestimated comparedto GPCP observations and to a simulation using themicrophysical scheme of Boucher et al.. Shortwavecloud radiative forcing is overestimated by the newscheme. Longwave cloud radiative forcing, on theother hand, is already very well simulated. This isparticularly interesting as it is controlled by the highlevel ice clouds, which were of main interest by theintroduction of the new scheme.

10.2. Perspectives

While some progress has been made since the earliestattempts to quantify the aerosol indirect effects in themid 1990s, plenty of uncertainties remain. In its lat-est report, the IPCC (2001) still assesses the level ofscientific understanding of the aerosol indirect effectsas “very low”, giving a range in the estimate of theglobal mean radiative forcing due to the first aerosolindirect effect as large as 0 to –2 Wm−2, without fix-ing a best estimate. For the second aerosol indirecteffect, no radiative impact is given at all. Aerosolclimate impacts are still very poorly taken into ac-count in climate change simulations (Anderson et al.,2003). The studies of Chapter 4 reveals large uncer-tainties in aerosol indirect effect modeling in GCMs.

Current research work on the issue of aerosol indi-rect effects is done using observations and modelingapproaches and investigating processes ranging fromthe microscale to global extension. A large part ofthis range has been addressed in the present study,while the interest was centered on the large scale andfocused on the modeling part. Satellite data from avariety of instruments with a large contribution ofdata from the innovative POLDER instrument havebeen used to evaluate the model (Chapters 2, 3 and8) and to analyze processes related to aerosol indirecteffects (Chapter 5). Data from a field study has beenused to study the cloud and aerosol processes for aspecific case (Chapter 4). Large scale historical ob-servations have been used to investigate the climatechange during the 20th century (Chapter 7). Thedevelopment of the new microphysical parameteriza-tion included the interaction with mesoscale modelsfor both adaptation of parameterizations existentin those models and validation of the GCM scheme(Chapter 9).

In future work, the approach which consists of us-ing a variety of observational data, particularly ofsatellite instruments, and smaller-scale models willbe continued.

132

10. Conclusions and perspectives 10.2. Perspectives

Short term: Improvement of the new scheme

A long timeseries of ground-based remote sensingradar, lidar, and radiometer measurements of aerosol,cloud and precipitation properties is readily avail-able from the Site Instrumental de Recherche parTeledetection Atmospherique (SIRTA) near Paris inFrance. The data will be used to evaluate and im-prove the new microphysical scheme presented in thisstudy. This will be done firstly by establishing sta-tistical distributions in the observations to be able tocompare the one dimensional measurements to thelarge scale model. Secondly, the mesoscale modelMeso-NH which includes a comprehensive liquid andice microphysical scheme will be used as an inter-mediate tool to translate the measurements to thescale of a GCM grid box. The focus of this part ofthe evaluation and development studies will be toevaluate the simulated physical quantities related tothe microphysical processes as cloud water, particleconcentrations, and sizes. The mesoscale model willalso be used to evaluate particular microphysical pro-cesses.

A new, comprehensive, and so far unique datasetfor cirrus cloud observations from TIROS-N VerticalSounder (TOVS) satellite measurements has recentlybeen established (Stubenrauch et al., 2003; Radelet al., 2003). These data will be used to evaluatethe ice cloud part of the new microphysical scheme.Particularly, radiative properties will be evaluatedusing this satellite data, and new approaches for theparameterization of radiative properties of ice cloudswill be tested.

Mid-term: A new subgrid-scale scheme using

PDFs, use of new satellite data

The development of a new, very promising ap-proach to improve microphysics parameterizationsis planned. Tompkins (2002) introduced the conceptof a prognostic probability density distribution of to-tal water content, with width and skewness treatedas prognostic variables in the model and predictedbased on turbulence, convection, and microphysics.The use of such a knowledge of subgrid scale distri-bution of water in the microphysical scheme shouldlargely improve the skill of the parameterizations.

Again, the use of mesoscale models and satellite datawill be a central tool to develop this new approach,and to evaluate the new parameterizations.

Future work implies as an important component theuse of upcoming new satellite measurements. Newdata are now available for aerosol and cloud prop-erties from the POLDER-2, MODIS, and ENVISATsatellite instruments. More complete data cover-ing larger timescales will soon be available. The datafrom the first spaceborne lidar and radar instrumentson the CALIPSO and CLOUDSAT satellites arealso very promising. For the first time global three-dimensional data will be available. A multitude ofsimultaneous measurements of vertical distributionsof aerosol and cloud properties will be available. Thisis a very promising approach to evaluate and improvethe parameterizations of cloud processes and aerosolindirect effects in global scale models.

Long term: Climate change studies

The goal of improvements of global scale modeling isa better understanding of the climate system and aprediction of climate change. Many shortcomings stillexist due to which simulations of climate change stillgive somewhat ambiguous results (see also Chapters6 and 7). However, the approach of a comprehen-sive description of cloud microphysics including iceand mixed-phase processes and aerosol indirect ef-fects as developed in the present study and improvedby the methods outlined above is promising to reducethe uncertainties related to these processes. Furtheron, it is expected that the performance of computinghardware will improve. Simulations with a fully cou-pled Earth System Model including the comprehen-sive microphysical scheme can thus be carried out. Abetter resolution of the GCM in both horizontal andvertical directions will further improve the results.Regional scale predictions of climate change becomepossible, which is of particular importance for thestudy of aerosol indirect forcings as they are concen-trated near the source regions. Aerosol impacts onthe hydrological cycle imply feedback processes withother components of the Earth System as vegetationand oceans. Such processes will be studied in simu-lations of future climate change.

133

10.3. Resume 10. Conclusions and perspectives

10.3. Conclusions et

perspectives : Resume

Resume et conclusions

Nous avons presente une etude des effets indirectsdes aerosols dans le modele de grande echelle LMDZet dans les observations satellitaires. L’outil prin-cipal utilise dans cette etude est le modele de cir-culation generale du Laboratoire de MeteorologieDynamique (MCG LMDZ). Le schema de micro-physique de Boucher et al. (1995) a ete implementedans ce modele, et une comparaison avec la versionstandard et avec des observations satellitaires mon-tre des tres bons resultats pour cette parametrisation.

Dans un projet de comparaison entre des versionsunidimensionnelles de differents modeles de grandeechelle et d’evaluation avec des donnees in-situ dela campagne de mesures ACE-2, le modele LMDZavec le schema de microphysique de Boucher et al.donne d’assez bons resultats. Des problemes majeurscommuns a tous les modeles et evidents aussi pourLMDZ sont les epaisseurs geometriques et optiqueset la colonne d’eau liquide trop grande. Une desraisons pour cette insuffisance est la resolution verti-cale trop grossiere. Deux modeles avec une meilleureresolution verticale, soit par une parametrisationsous-maille, soit par un nombre plus eleve de couches,donnent des meilleurs resultats compares aux autresmodeles. Ceci implique que l’approche d’introduireune representation sous-maille de la variabilite ver-ticale des nuages serait prometteuse pour ameliorerle modele. Un tel travail fait partie des plans pourle futur. Tous les modeles montrent une incapacitea simuler la bruine pour les contenus en eau liquideobserves. Ceci peut impliquer que les modeles sontincapables de simuler adequatement le deuxieme effetindirect qui est considere etre le plus fort justementsur les nuages bas qui produisent de la bruine enmasses d’air pur, et non plus dans des conditionspolluees. Pour l’activation des gouttelettes, on amontre qu’aucun des modeles ni des differentes vari-antes des formules de Boucher et Lohmann (1995)peuvent bien simuler simultanement les deux cas.En particulier, parmi les parametrisations testees,les schemas physiques n’etaient pas meilleurs que les

schemas empiriques.

Une etude centrale de la these etait l’evaluation deseffets indirects a partir des relations statistiques en-tre proprietes des nuages et aerosols dans les obser-vations satellitales de POLDER et dans le modeleLMDZ. La relation entre le rayon effectif des gout-telettes (CDR en anglais) au sommet des nuageset l’indice des aerosols (AI) est une indication dupremier effet indirect. La taille des gouttelettesest systematiquement sous-estimee dans le modele.Pourtant, la relation CDR-AI dans le modele est sim-ilaire a celle des observations, a l’exception du faitque dans les observations, on trouve une correlationmoins forte pour les AI eleves que pour les petitsAI, alors que dans le modele, la relation reste lameme pour tous les AI. En separant les situations decolonne d’eau liquide (LWP en anglais) a peu presfixes, nous avons pu nous rapprocher plus stricte-ment la definition de l’effet indirect des aerosols parTwomey (1974). Pourtant, la relation calculee enmoyennant sur des differents relations a LWP fixen’est pas tres differente comparee a celle qui con-sidere toutes les situations. En correlant la pente dela relation CDR-AI au LWP pour lequel elle a eteetablie, on trouve dans les observations des relationsplus fortes pour les petits LWP. Au-dessus des con-tinents, des pentes legerement positives sont memestrouvees pour des LWP eleves. Dans le modele, unetelle dependance n’est pourtant pas identifiable. Lespentes de la relation CDR-AI et dans les observationsPOLDER et dans le modele pour des differents LWPet pour des differentes regions sont dans la gammedes valeurs donnees par d’autres auteurs. Dans unedeuxieme partie, la relation entre la colonne en eauliquide et l’indice des aerosols a ete etablie. A lafois dans les observations et dans le modele, unecorrelation positive a ete trouvee, qui est pourtantplus marquee au-dessus des moyennes latitudes del’hemisphere nord qu’en considerant tout le globe.La pente de la relation est un peu trop forte dans lemodele, ce qui pourrait indiquer une surestimationdu deuxieme effet indirect ou etre du au fait que lemodele n’inclut pas l’effet semi-direct des aerosols.Les relations CDR-AI et LWP-AI sont tres similairesdans des conditions pre-industrielles et dans les con-

134

10. Conclusions and perspectives 10.3. Resume

ditions actuelles dans le modele, ce qui montre lapersistance des principes physiques sous jacentes.Pourtant, on trouve un faible effet du deuxieme ef-fet indirect additionnel du aux aerosols anthropiques.

Avec un modele comprenant le schema de micro-physique et le lien entre aerosols et nuages ainsivalide, on est capable d’examiner les impacts desaerosols sur le changement climatique. Nous avonseffectue des simulations pour la periode historiquede 1930 a 1989, pour laquelle des observations duchangement climatique suggerent un impact proba-ble des forcages par les aerosols anthropiques. Nousavons fait trois ensembles de simulations dont cha-cun contenait trois membres. Tous les simulationsetaient forces par des temperatures de la surface dela mer (SST en anglais) et extension de la glacede mer observees. Le premier scenario, nommeGHG+AER, contenait des gaz a effet de serre etdes distributions des aerosols de sulfates changeantsselon des emissions historiques. Dans le deuxieme,GHG, les aerosols etaient fixes a une distribution pre-industrielle, et dans le troisieme, CTL, les gaz a effetde serre etaient fixes a des valeurs pre-industriellesaussi. L’impact radiatif des gaz a effet de serre esttrouve etre plus fort dans les conditions de ciel clairque de ciel nuageux. Une reduction de la couverturenuageuse due a l’effet de serre anthropique est ob-servee, qui implique une reduction du forcage dansle spectre des ondes longues, et un impact radiatifpositif dans le spectre solaire. L’impact radiatif desaerosols est aussi fort que celui par les gaz a ef-fet de serre, mais de signe oppose. L’impact desaerosols est pourtant plus fort en ciel nuageux qu’enciel clair, ou seulement l’effet direct des aerosolsest actif, et il joue sur le spectre solaire seulement.En ciel nuageux, outre le premier effet indirect, ledeuxieme effet indirect est un forcage negatif im-portant. Pourtant, dans l’hemisphere sud et dans lestropiques, il est contrebalance par une retroaction desnuages qui consiste en une reduction de la couverturenuageuse haute qui implique un forcage positif dansle spectre solaire due a une reduction de l’albedo desnuages hauts. Pour la periode 1950 a 1990, nousavons compare les impacts des differents forcages surles tendances de la temperature des surfaces conti-

nentales et sur l’ecart journalier de la temperature(DTR en anglais) au-dessus des continents pour l’etede l’hemisphere nord ou les impacts des aerosols an-thropiques sont estimees etre les plus forts. Alorsque pour la plupart des regions du globe, les ten-dances de temperatures et du DTR sont controleespar la SST imposee, on trouve que les tendancesnegatives de temperature au-dessus des moyenneslatitudes du continent eurasiatique ne peuvent etreexpliquees par le modele que dans le scenario quicontient les forcages anthropiques des gaz a effet deserre et des aerosols. Pareillement, les tendances deDTR different le plus parmi les scenarios au-dessusdes moyennes latitudes de l’hemisphere nord, ou aussic’est le forcage par les deux, gaz a effet de serre etaerosols sulfates anthropiques combines, qui expliquele mieux les tendances observess. En correlant lestendances des differentes quantites dans les sortiesdu modele, on a pu montrer que dans le modele,la diminution du DTR est expliquee par les ten-dances dans le flux du rayonnement dans le spectredes ondes courtes, qui sont largement controles parles tendances dans le forcage radiatif des nuages etpar les forcages des aerosols au cas ou ils sont inclus(scenario GHG+AER). Contrairement a d’autresetudes, l’evaporation ne semble pas jouer pas un roledeterminant dans notre modele pour la reduction duDTR.

Comme il a ete montre par ces etudes, et commeil est suggere par d’autres raisons comme le realismephysique, la representation de la phase glace dansun modele de grande echelle est d’une grande im-portance. Nous avons etudie la parametrisation quidistribue l’eau condensee dans le modele en eau liq-uide et en glace a partir de la temperature locale enutilisant des observations satellitaires de POLDER.Dans ce but, nous avons etabli des relations statis-tiques entre la temperature et la phase thermo-dynamique au sommet des nuages dans la mememaniere dans les observations et dans plusieurs simu-lations avec le modele en variant les deux parametresqui controlent la temperature au-dessous de laque-lle l’eau liquide n’existe plus (Tice) et l’exposant dela fonction de transition entre cette temperature etcelle, au-dessus de laquelle toute l’eau est fondue. Les

135

10.3. Resume 10. Conclusions and perspectives

deux parametres trouves sont de –32◦C et 1.7. Ilsdifferent de ceux utilises dans la version standard dumodele, mais sont proches des valeurs utilisees dansd’autres modeles. Le forcage radiatif des nuagesn’est pas change dans le spectre des ondes longues enchoisissant le nouveau jeu de parametres au lieu del’ancien, et il est beaucoup ameliore dans le spectredes ondes courtes.

Un nouveau schema microphysique a ete developpequi traite les phases liquides et glace de maniereconsistante. Il utilise les rapports de melange de lavapeur d’eau, de l’eau liquide et glace des nuagesainsi que de trois differentes especes de precipitation,et les concentrations en nombre des goutteletteset des cristaux de glace comme des variables dumodele. Pour l’activation des gouttelettes a partirdes aerosols, une parametrisation physique a ete in-troduite. Le nouveau schema donne des resultatsencourageants dans des tests sur les cas unidimen-sionnels. Pour les cas de stratocumulus marinsd’ACE-2, il simule beaucoup mieux les concentra-tions des gouttelettes, les epaisseurs optiques desnuages et la bruine observes que l’ancien schema. Ilest ainsi sous beaucoup d’aspects meilleurs que lesautres parametrisations testees et presentees dans lechapitre 4. Dans le cas de cirrus, le nouveau schemaest capable de simuler le contenu integreen glace desnuages aussi bien qu’un modele de meso-echelle avecun schema complet de microphysique. Des resultatspreliminaires tridimensionnels relevent pourtant en-core des deficiences. Alors que la couverture nuageuseet le forcage radiatif des nuages dans le spectre desondes longues sont assez bien simules, il y a encoredes problemes avec le forcage radiatif dans le spectredes ondes courtes et dans la precipitation.

Perspectives

Alors qu’il y a eu des avancees scientifiques depuisles premiers essais de quantifier les forcages indi-rects des aerosols dans les annees 1990, une mul-titude d’incertitudes persistent. Dans son dernierrapport, le GIEC (IPCC en anglais) jugent toujoursle niveau de la comprehension scientifique des ef-fets indirects des aerosols d’etre � bas� ou meme� tres bas� . Pour le premier effet indirect, des

valeurs sont donnees dans un ecart aussi grand que 0a –2 Wm−2, alors que pour le deuxieme effet indirectaucune valeur n’est donnee du tout.

Le travail de recherche recent dans le domaine deseffets indirects des aerosols se fait en s’appuyantsur des approches de modelisation et d’observations,etudiant des processus de la micro-echelle jusqu’a desextensions globales. Une grande partie de ce domainea ete traitee dans cette these, alors que l’interet etaitcentre sur la modelisation et sur la grande echelle.Des donnees satellitaires de plusieurs instrumentsavec une importante contribution de l’instrumentPOLDER ont ete utilisees pour l’evaluation dumodele et pour des etudes de processus. Des donneesde campagnes de mesures ont ete utilisees pourevaluer des processus de nuages et d’aerosols dansdes cas specifiques. Des observations historiquesa grande echelle ont ete utilisees pour etudier lechangement climatique pendant le 20ieme siecle. Ledeveloppement d’une nouvelle parametrisation de lamicrophysique des nuages a implique l’interactionavec des modeles de meso-echelle afin d’adapter desparametrisations existantes dans de tels modeles etpour l’evaluation du MCG.

Cette approche d’utiliser une variete de donneesd’observations, en particulier des instruments satel-litaires, ainsi que des modeles de plus petite echelle,doit etre poursuivie dans les travaux futurs.

A court terme, le nouveau schema doit etre evalueet ameliore en utilisant des donnees de teledetectiona partir du sol ainsi que des donnees satellitaires.Le Site Instrumental de Recherche par TeledetectionAtmospherique (SIRTA) a Palaiseau a proximite deParis fournit des donnees d’instruments radar, li-dar, et radiometres desquels des quantites physiquesd’aerosols, nuages et precipitation peuvent etrededuites. Des distributions statistiques des quan-tites vont etre etablies pour utiliser les donneesunidimensionnelles pour l’evaluation du modele agrande echelle, et puis un modele intermediaire ameso-echelle va etre introduit pour pouvoir utiliserles donnees du SIRTA. Cette etude sera focaliseesur l’evaluation et le developpement des quan-

136

10. Conclusions and perspectives 10.3. Resume

tites physiques simulees liees aux processus micro-physiques comme le contenu en eau des nuages ou laconcentration et la taille de particules. Le modele ameso-echelle sera egalement utilise pour evaluer lesprocessus microphysiques.

Un nouveau, complet, et jusqu’a present uniquejeu de donnees d’observations de cirrus de TOVSa ete recemment etabli. Ces donnees pourront etreutilisees afin d’evaluer la partie glace du nouveauschema. En particulier, les proprietes radiatives peu-vent etre evaluees, et des nouvelles approches pourleur parametrisation seront examinees.

A moyen terme, une parametrisation de la variabilitesous-maille de l’eau doit etre developpee en poursuiv-ant l’approche de Tompkins (2002) qui a introduitune fonction de probabilite de densite, dont la vari-ance et l’inclinaison sont predites en fonction de laturbulence, de la convection et de la microphysique.La connaissance de la variabilite sous-maille de ladistribution de l’eau pourrait largement ameliorer ledegre de realisme des processus microphysiques. Pourle developpement et l’amelioration d’un tel schema,de nouveau, des observations satellitaires ainsi quedes modeles a meso-echelle vont etre utilises.

Concernant les donnees satellitaires, des nouveaux in-struments vont fournir des jeux de donnees plus com-

plets et pour une periode plus longue, comme MODISet les instruments d’ENVISAT. De plus d’ici quelquetemps, les observations d’un radar sur CLOUDSATet d’un lidar sur CALIPSO vont pouvoir fournir enquelque sorte des donnees tridimensionnelles, ce quiva largement ameliorer les possibilites d’evaluationet de developpement des modeles a grande echelle.

A long terme enfin, un modele avec desparametrisations des nuages, de la precipitation etdes interactions aerosols - nuages ainsi amelioreesaccompagne d’un developpement de la puissance desordinateurs et des techniques de calcul, permettrontdes etudes plus poussees du changement climatique.Avec un modele du systeme terrestre comprenant ladynamique de l’ocean et des surfaces continentalescomme il est developpe a l’IPSL, des simulations descenarios pour le changement climatique des 20iemeset 21iemes siecles vont pouvoir etre effectuees. Lesresultats vont etre examines en se focalisant sur lesaspects regionaux du changement climatique pourlesquels les forcages par les aerosols jouent un roleimportant, ainsi que sur le cycle hydrologique. Ledernier point implique une etude du couplage entreles differents sous-systemes du systeme climatique,dont l’atmosphere mais aussi l’ocean, les surfacescontinentales et la vegetation sont des composantesimportantes.

137

10.3. Resume 10. Conclusions and perspectives

138

A. Abbreviation and notation list

A.1. Abbreviations

ACE Aerosol Characterization Experiment

ADE Aerosol Direct Effect

ADEOS ADvanced Earth Observation Satellite

AIE Aerosol Indirect Effect

AOT Aerosol Optical Thickness

AVHRR Advanced Very High Resolution Radiometer

BL Planetary Boundary Layer

CCM Community Climate Model

CCN Cloud Condensation Nuclei

CCSR Center for Climate System Research

CDNC Cloud Droplet Number Concentration

CDR Cloud Droplet Radius

CGT Cloud Geometrical Thickness

CNES Centre National d’Etudes Spatiales

C.N.R.S. Centre National de Recherche Scientifique

COT Cloud Optical Thickness

CRM Cloud Resolving Model

CSIRO Commonwealth Scientific and Industrial Research Organization

CSU Colorado State University

CTM Chemistry Transport Model

DJF December-January-February

ECHAM ↑ECMWF-HAMburg ↑GCM

ECMWF European Centre for Medium-Range Weather Forecasting

139

A.1. Abbreviations A. Abbreviation and notation list

ERBE Earth Radiation Budget Experiment

ESM Earth System Model

FIRE.ACE First ↑ISCCP Regional Experiment - Arctic Cloud Experiment

GCM General Circulation Model

GCSS ↑GEWEX Cloud System Study

GEWEX Global Energy and Water Cycle Experiment

GHE GreenHouse Effect

GHG GreenHouse Gases

GISS Goddard Institute for Space Studies

IDRIS Institut de Developpement et des Resources en Informatique Scientifique

IFS Integrated Forecasting System

INCA INteraction with Chemistry and Aerosols

INDOEX Indian Ocean Experiment

IPCC Intergovernmental Panel on Climate Change

IPSL Institut Pierre Simon Laplace

ISCCP International Satellite Cloud Climatology Project

ITCZ Inner Tropical Convergence Zone

JJA June-July-August

LES Large Eddy Simulation model

LGGE Laboratoire de Glaciologie et Geophysique de l’Environnement

LMD Laboratoire de Meteorologie Dynamique

LMDZ ↑LMD-Zoom ↑GCM

LOA Laboratoire d’Optique Atmospherique

LODyC Laboratoire d’Oceanographie Dynamique et du Climatologie

LSCE Laboratoire des Sciences du Climat et de l’Environnement

LW Longwave

LWP Liquid Water Path

MAST Monterey Area Ship Track experiment

NCAR National Center for Atmospheric Research

NH Northern Hemisphere

140

A. Abbreviation and notation list A.1. Abbreviations

OPA Ocean Parallelise

ORCHIDEE Organizing Carbon and Hydrology In Dynamic EcosystEms

PACE Parameterization of the Aerosol Climatic Effects

PDF Probability Density Function

PNNL Pacific Northwest National Laboratory

POLDER POLarization and Directionality of the Earth’s Reflectances

PPH Plane-Parallel Homogeneity

PW Precipitable Water

RACE Radiation, Aerosol, and Cloud Experiment

RAMS Regional Atmospheric Modeling Section

RH Relative Humidity

SA Service d’Aeronomie

ScaRaB Scanner of the Earth Radiation Budget

SCM Single-Column Model

SH Southern Hemisphere

SST Sea Surface Temperature

SW Shortwave

TIROS Television Infrared Observation Satellite

TOA Top Of the Atmosphere

TOVS ↑TIROS-N Operational Vertical Sounder

TRMM Tropical Rainfall Measuring Mission

141

A.2. Notation list A. Abbreviation and notation list

A.2. Notation list

α Albedo (often also used for parameter constants)

ατ POLDER aerosol index

Al Activation rate of cloud droplets, [m−3 s−1]

Ai Nucleation rate of ice crystals, [m−3 s−1]

cp Specific heat of dry air at constant pressure, 1004.64 J kg−1 K−1

Dj Effective diameter of cloud or precipitation species j, [m]

Dv Coeff. of vapor diffusion in air, [m2 s−1]

ε Emissivity

Ejk Collision efficiency between species j and k

esat Saturation vapor pressure, [hPa]

∆F Radiative forcing, [Wm−2]

f Cloud fraction of grid cell

fact Fraction of NCCN activated

Frs Freezing rate of rain drops to form snow, [m−3 s−1]

Fli Freezing rate of liquid water droplets to form ice crystals, [m−3 s−1]

g Asymmetry parameter

H Cloud geometrical thickness, [m]

I Radiative impact, [Wm−2]

k Exponent in droplet activation

λ Climate sensitivity parameter, [K W−1 m2]

λj Slope of size distribution of species j, [m−1]

Lv Latent heat of vaporization of water, 2.501× 106 J kg−1

Ls Latent heat of sublimation of water, 2.834× 106 J kg−1

142

A. Abbreviation and notation list A.2. Notation list

Lf Latent heat of freezing of water, 0.333× 106 J kg−1

mj Mass of particle of species j, [kg]

ma Initial mass of activated droplets, [kg]

m0 Initial mass of nucleated ice crystals, 10−12 kg

Nj Number concentration of species j, [m−3]

NCCN Number concentration of CCN, [m−3]

NIN Number concentration of IN, [m−3]

NIC Number concentration of contact-freezing nuclei, [m−3]

Na Number concentration of aerosols, [m−3]

nHM Rate of formation of ice splinter by Hallett-Mossop-process, [kg−1]

ω0 Single scattering albedo

Pj Rate of precipitation of species j, [m s−1]

∆q Parameter in Condensation scheme (“ratqs”), [kg kg−1]

qt Total water mixing ratio, [kg kg−1]

qc Condensed water mixing ratio, [kg kg−1]

qj Mixing ratio of species j, [kg kg−1]

qsat,j Saturation vapor mixing ratio w.r.t. species j, [kg kg−1]

ρair Air density, [kg m−3]

ρj Density of precipitation species j, [kg m−3]

r Correlation coefficient

r0 Minimum radius for autoconversion, 0.8× 10−5 m

rdv Volume-mean cloud droplet radius, [m]

re Cloud droplet effective radius, [m]

rjk Collision rate between species j and k, [m−3 s−1]

σ Stefan-Boltzmann constant, 0.67 10−8 J K−4 m−2 s−1

143

A.2. Notation list A. Abbreviation and notation list

Sj Supersaturation w.r.t. species j

T Temperature, [K]

Ts Surface emperature, [K]

T0 Freezing temperature of water, 273.15 K

Ti Temperature, below which no supercooled water exists, 258.15 K

t Time, [s]

∆t Time step [s]

Vj Terminal fall speed of cloud or precipitation species j, [m s−1]

w Updraft speed, [m s−1]

xice Ice mass fraction in a cloud

∆z Layer thickness, [m]

Indexes

v Water vapor

l Liquid water cloud droplets

i Ice crystals

r Rain water

g Graupel

s Snow

144

Bibliography

Abdul-Razzak, H., and S. J. Ghan, A parametriza-tion of aerosol activation: 2. Multiple aerosol types,J. Geophys. Res., 105, 6837–6844, 1999.

Abdul-Razzak, H., S. J. Ghan, and C. Rivera-Carpio,A parametrization of aerosol activation: 1. Sin-gle aerosol type, J. Geophys. Res., 103, 6123–6131,1998.

Ackerman, A. S., O. Toon, D. E. Stephens, A. J.Heymsfield, V. Ramanathan, and E. J. Welton, Re-duction of tropical cloudiness by soot, Science, 288,1042–1047, 2000a.

Ackerman, A. S., O. B. Toon, J. P. Taylor, D. W.Johnson, P. V. Hobbs, and R. J. Ferek, Effects ofaerosols on cloud albedo: Evaluation of Twomey’sparameterization of cloud susceptibility using mea-surements of ship tracks, J. Atmos. Sci., 57, 2684–2695, 2000b.

Albrecht, B. A., Aerosols, cloud microphysics, andfractional cloudiness, Science, 245, 1227–1230,1989.

Anderson, T. L., R. J. Charlson, S. E. Schwartz,R. Knutti, O. Boucher, H. Rodhe, and J. Heintzen-berg, Climate forcing by aerosols - a hazy picture,Science, 300, 1103–1104, 2003.

Bartlett, J. T., The growth of cloud droplets by coa-lescence, Q. J. R. Meteorol. Soc., 92, 93–104, 1966.

Beheng, K. D., A parametrization of warm cloud mi-crophysical conversion processes, Atmos. Res., 33,193–206, 1994.

Bergeron, T., On the physics of cloud and precipi-tation, Proces-verbaux des seances de l’associationde meteorologie, Union Geodesique et GeophysiqueInternationale, Lisbon, 1933.

Bony, S., and K. A. Emanuel, A parameterization ofthe cloudiness associated with cumulus convection;evaluation using TOGA COARE data, J. Atmos.Sci., 58, 3158–3183, 2001.

Boucher, M., and J. Quaas, Testing and developingcloud parametrisations in the LMDZ GCM by us-ing POLDER satellite data, Geophys. Res. Lett.,submitted, 2003.

Boucher, O., Aircraft can increase cirrus cloudiness,Nature, 397, 30–31, 1999.

Boucher, O., and T. L. Anderson, General circula-tion model assessment of the sensitivity of directclimate forcing by anthropogenic sulfate aerosols toaerosol size and chemistry, J. Geophys. Res., 100,26,117–26,134, 1995.

Boucher, O., and J. Haywood, On summing thecomponents of radiative forcing of climate change,Clim. Dyn., 18, 297–302, 2001.

Boucher, O., and U. Lohmann, The sulfate–CCN–cloud albedo effect - a sensitivity study with twogeneral circulation models, Tellus, 47B, 281–300,1995.

Boucher, O., and M. Pham, History of sulfate aerosolradiative forcings, Geophys. Res. Lett., 29, 1308,doi:10.1029/2001GL014048, 2002.

Boucher, O., H. Le Treut, and M. B. Baker, Precip-itation and radiation modeling in a general circu-lation model: Introduction of cloud microphysicalprocesses, J. Geophys. Res., 100, 16,395–16,414,1995a.

Boucher, O., Z.-X. Li, and H. Le Treut, Introductiondu schema de convection de Tiedtke dans les MCGsdu LMD, Tech. rep., Laboratoire de MeteorologieDynamique, Paris, 20pp, 1995b.

Boucher, O., M. Pham, and C. Venkataraman, Sim-ulation of the atmospheric sulfur cycle in the Lab-oratoire de Meteorologie Dynamique general circu-lation model. Model description, model evaluation,and global and European budgets, in Note scien-tifique de l’IPSL, vol. 23, edited by J.-P. Boulangerand Z.-X. Li, IPSL, Paris, 26pp., 2002.

145

References

Brenguier, J.-L., Parameterization of the condensa-tion process: A theoretical approach, J. Atmos.Sci., 48, 264–282, 1991.

Brenguier, J.-L., P. Chuang, Y. Fouquart, D. W.Johnson, F. Parol, H. Pawlowska, J. Pelon,L. Schuller, F. Schroder, and J. Snider, Anoverview of the ACE–2 CLOUDYCOLUMN clo-sure experiment, Tellus, 52B, 815–827, 2000a.

Brenguier, J.-L., H. Pawlowska, L. Schuller,R. Preusker, J. Fischer, and Y. Fouquart, Radia-tive properties of boundary layer clouds: Dropleteffective radius versus number concentration, J.Atmos. Sci., 57, 803–821, 2000b.

Brenguier, J.-L., H. Pawlowska, and L. Schuller,Cloud microphysical and radiative properties forparameterization and satellite monitoring of the in-direct effect of aerosol on climate, J. Geophys. Res.,108, 8632, doi:10.1029/2002JD002682, 2003.

Breon, F.-M., and S. Colzy, Global distribution ofcloud droplet effective radius from POLDER po-larization measurements, Geophys. Res. Lett., 27,4065–4068, 2000.

Breon, F.-M., and P. Goloub, Cloud droplet effec-tive radius from spaceborne polarization measure-ments, Geophys. Res. Lett., 25, 1879–1882, 1998.

Breon, F.-M., D. Tanre, and S. Generoso, Aerosol ef-fect on cloud droplet size monitored from satellite,Science, 295, 834–838, 2002.

Buriez, J. C., C. Vanbauce, F. Parol, P. Goloub,M. Herman, B. Bonnel, Y. Fouquart, P. Couvert,and G. Seze, Cloud detection and derivation ofcloud properties from POLDER, Int. J. RemoteSensing, 18, 2785–2813, 1997.

Calahan, R. F., Bounded cascade clouds: albedo andeffective thickness, Nonlin. Proc. Geoph., 1, 156–167, 1994.

Carlin, B., Q. Fu, U. Lohmann, G. D. Mace,K. Sassen, and J. M. Comstock, High-cloud hor-izontal inhomogeneity and solar albedo bias, J.Clim., 15, 2321–2339, 2002.

Cess, R. D., G. Potter, J. P. Blanchet, G. Boer,A. Genio, M. Deque, V. Dymnikov, V. Galin,W. Gates, S. Ghan, T. Kiehl, A. Lacis, H. Le Treut,Z. Li, X.-Z. Liang, B. McAvaney, V. Meleshko,J. Mitchell, J.-J. Morcrette, D. Randall, L. Rikus,E. Roeckner, J. Roer, U. Schlese, D. Sheinin,A. Slingo, A. Sokolov, K. E. Taylor, W. Washing-ton, R. T. Wetherald, I. Yagai, and M.-H. Zhang,Intercomparison and interpretation of climate feed-back processes in nineteen atmospheric general cir-culation models, J. Geophys. Res., 95, 16,601–16,615, 1990.

Cess, R. D., M.-H. Zhang, G. L. Potter, H. W.Barker, R. A. Colman, D. A. Dazlich, A. D. Del Ge-nio, M. Esch, J. R. Fraser, V. Galin, W. Gates, J. J.Hack, W. J. Ingram, J. T. Kiehl, A. A. Lacis, H. LeTreut, Z.-X. Li, X.-Z. Liang, J.-F. Mahfouf, B. J.McAvaney, V. P. Meleshko, J.-J. Morcrette, D. A.Randall, E. Roeckner, J.-F. Royer, A. P. Sokolov,P. V. Sporyshev, K. E. Taylor, W.-C. Wang, andR. T. Wetherald, Uncertainties in carbon dioxideradiative forcing in atmospheric general circulationmodels, Science, 262, 1252–1255, 1993.

Chameides, W. L., C. Luo, R. Saylor, D. Streets,Y. Huang, M. Bergin, and F. Giorgi, Correlationbetween model-calculated anthropogenic aerosolsand satellite-derived cloud optical depths: Indica-tion of indirect effect?, J. Geophys. Res., 107, 4085,doi:10.1029/2000JD000208, 2002.

Charlson, R. J., J. E. Lovelock, M. O. Andreae, andS. G. Warren, Sulphate aerosols and climate, Na-ture, 340, 437–438, 1989.

Charlson, R. J., J. H. Seinfeld, A. Nenes, M. Kul-mala, A. Laaksonen, and M. C. Faccini, Reshapingthe theory of cloud formation, Science, 292, 2025–2026, 2001.

Cheinet, S., The parameterization of clear and cloudyconvective boundary layers, Ph.D. thesis, EcolePolytechnique, Palaiseau, 139pp., 2002.

Cheinet, S., A multiple mass-flux parameterizationfor the surface-generated convection. Part I: Dryplumes, J. Atmos. Sci., 60, 2313–2327, 2003.

146

References

Cheinet, S., and J. Teixara, A simple formulationfor the eddy-diffusivity parameterization of cloud-topped boundary layers, Geophys. Res. Lett., 30,1930, doi:10.1029/2003GL017377, 2003.

Chuang, C. C., J. E. Penner, K. E. Taylor, A. S.Grossman, and J. J. Walton, An assessment of theradiative effects of anthropogenic sulfate, J. Geo-phys. Res., 102, 3761–3778, 1997.

Coakley, Jr., J. A., and C. D. Walsh, Limits to theaerosol indirect radiative effect derived from obser-vations of ship tracks, J. Atmos. Sci., 59, 668–680,2002.

Coakley Jr., J. A., R. L. Bernstein, and P. A. Durkee,Effect of ship-track effluents on cloud reflectivity,Science, 237, 1020–2022, 1987.

Coakley Jr., J. A., P. A. Durkee, K. Nielsen, J. P.Taylor, S. Platnick, B. A. Albrecht, D. Babb, F.-L.Chang, W. R. Tahnk, C. S. Bretherton, and P. V.Hobbs, The appearance and disappearance of shiptracks on large spatial scales, J. Atmos. Sci., 57,2765–2778, 2000.

Conover, J. H., Anomalous cloud lines, J. Atmos.Sci., 23, 778–785, 1966.

Conover, W. J., Practical Nonparametric Statistics,2nd ed., John Wiley and Sons, New York, 493pp.,1980.

Cusack, S., J. M. Edwards, and R. Kershaw, Esti-mating the subgrid variance of saturation, and itsparameterization for use in a GCM cloud scheme,Q. J. R. Meteorol. Soc., 125, 2057–3076, 1999.

Dai, A., A. D. Del Genio, and I. Y. Fung, Clouds,precipitation and temperature range, Nature, 368,665–666, 1997.

Dai, A., K. E. Trenberth, and T. R. Karl, Effects ofclouds, soil moisture, and water vapor on diurnaltemperature range, J. Clim., 12, 2451–2473, 1999.

Dai, A., T. M. L. Wigley, B. A. Boville, J. T. Kiehl,and L. E. Buja, Climates of the twentieth andtwenty-first centuries simulated by the NCAR cli-mate system model, J. Clim., 14, 485–519, 2001.

de Rosnay, P., J. Polcher, M. Bruen, and K. Laval,Impact of a physically based soil water flow andsoil-plant interaction representation for modelinglarge scale land surface processes, J. Geophys. Res.,107, 4118, doi:10.1029/2001JD000634, 2002.

Del Genio, A. D., M.-S. Yao, W. Kovari, and K. K.-W. Lo, A prognostic cloud water parameterizationfor global climate models, J. Clim., 9, 270–304,1996.

DeMott, P. J., K. Sassen, M. R. Poellot, D. Baum-gardner, D. C. Rogers, S. D. Brooks, A. J. Prenni,and S. M. Kreidenweis, African dust aerosols as at-mospheric ice nuclei, Geophys. Res. Lett., 14, 1732,doi:10.1029/2003GL017410, 2003.

Deschamps, P., F. Breon, M. Leroy, A. Podaire,A. Bricaud, J. Buriez, and G. Seze, The POLDERmission: Instrument characteristics and scientificobjectives, IEEE Trans. Geosc. Rem. Sens., 32,598–615, 1994.

Deuze, J.-L., M. Herman, P. Goloub, D. Tanre, andA. Marchand, Characterization of aerosols overocean from POLDER/ADEOS-1, Geophys. Res.Lett., 26, 1421–1424, 1999.

Deuze, J. L., P. Goloub, M. Herman, A. Marchand,G. Perry, S. Susana, and D. Tanre, Estimate of theaerosol properties over the ocean with POLDER,J. Geophys. Res., 105, 15,329–15,346, 2000.

Deuze, J. L., F. M. Breon, C. Devaux, P. Goloub,M. Herman, B. Lafrance, F. Maignan, A. Marc-hand, F. Nadal, G. Perry, and D. Tanre, Re-mote sensing of aerosols over land surfaces fromPOLDER- ADEOS-1 polarized measurements, J.Geophys. Res., 106, 4913–4926, 2001.

Di Giuseppe, F., The effect of unresolved verticalcloud structure on radiative transfer calculations,EGS/AGU/EUG Joint Assembly, 6 - 11th April2003, Nice, France, 2003.

Donner, L. J., C. J. Seman, B. J. Soden, R. S. Hem-ler, J. C. Warren, and J. S. K.-N. Liou, Large-scaleice clouds in the GFDL SKYHI general circulationmodel, J. Geophys. Res., 102, 21,745–21,768, 1997.

147

References

Doutriaux-Boucher, M., J.-C. Buriez, G. Brogniez,L. C.-Labonnote, and A. J. Baran, Sensitivity ofretrieved POLDER directional cloud optical thick-ness to various ice particle models, Geophys. Res.Lett., 27, 109–112, 2000.

Dufresne, J.-L., C. Gautier, P. Ricchiazzi, andY. Fouquart, Longwave scattering effects of min-eral aerosols, J. Atmos. Sci., 59, 1959–1966, 2002.

Durkee, P. A., R. E. Chartier, A. Brown, E. J. Tre-hubenko, S. D. Rogerson, C. Skupniewicz, K. E.Nielsen, S. Platnick, and M. D. King, Compositeship track characteristics, J. Atmos. Sci., 57, 2542–2553, 2000.

Easterling, D. R., B. Horton, P. D. Jones, T. C. Pe-terson, T. R. Karl, D. E. Parker, M. J. Salinger,V. Razuvayev, N. Plummer, P. Jamason, andC. K. Folland, Maximum and minimum temper-ature trends for the globe, Science, 277, 364–367,1997.

Emanuel, K., A scheme for representing cumulus con-vection in large–scale models, J. Atmos. Sci., 48,2313–2335, 1991.

Feichter, J., U. Lohmann, and I. Schult, The atmo-spheric sulfur cycle in ECHAM-4 and its impact onthe shortwave radiation, Clim. Dyn., 13, 235–246,1997.

Feingold, G., Modeling of the first indirect effect:Analysis of measurement requirements, Geophys.Res. Lett., 30, 1997, doi:10.1029/2003GL017967,2003.

Feingold, G., L. A. Remer, J. Ramaprasad, and Y. J.Kaufman, Analysis of smoke impact on clouds inBrazilian biomass burning regions: An extensionto Twomey’s approach, J. Geophys. Res., 106,22,907–22,922, 2001.

Feingold, G., W. L. Eberhard, D. E. Veron,and M. Previdi, First measurements of theTwomey indirect effect using ground-based re-mote sensers, Geophys. Res. Lett., 30, 1287, doi:10.1029/2002GL016633, 2003.

Fitzgerald, J. W., Dependence of the supersaturationspectrum of CCN on aerosol size distribution andcomposition, J. Atmos. Sci., 30, 628–634, 1973.

Flatau, P. J., G. J. Tripoli, J. Verlinde, and W. R.Cotton, The CSU-RAMS cloud microphysics mod-ule: General theory and documentation, Atmos.Sci. Paper, vol. 451, Dept. Atmos. Sci., ColoradoState University, Fort Collins, Colorado, 88pp.,1989.

Folland, C. K., T. R. Karl, J. R. Christy, R. A.Clarke, G. V. Gruza, J. Jouzel, M. E. Mann,J. Oerlemans, M. J. Salinger, and S.-W. Wang,Observed climate variability and change, in Cli-mate change 2001 – The scientific basis. Contri-bution of working group I to the Third AssessmentReport of the Intergovernmental Panel on ClimateChange, edited by J. T. Houghton, Y. Ding, D. J.Griggs, M. Noguer, P. J. van der Linden, X. Dai,K. Maskell, and C. A. Johnson, pp. 99–182, Cam-bridge University Press, Cambridge, 2001.

Fouquart, Y., and B. Bonnel, Computations of solarheating of the Earth’s atmosphere: A new param-eterization, Contr. Atmos. Phys., 53, 35–62, 1980.

Fowler, L. D., D. A. Randall, and S. A. Rutledge,Liquid and ice cloud microphysics in the CSU gen-eral circulation model. Part I: Model descriptionand simulated microphysical processes, J. Clim.,9, 489–529, 1996.

Georgii, H.-W., Uber die Eiskeimbildung in un-terkuhlten Wolken in ihrer Abhangigkeit von at-mosharischem Aerosol, Berichte des DeutschenWetterdienstes, vol. 58, Deutscher Wetterdienst,Offenbach, 43pp., 1959.

Ghan, S. J., L. R. Leung, and R. C. Easter, Predic-tion of cloud droplet number in a general circula-tion model, J. Geophys. Res., 102, 21,777–21,794,1997a.

Ghan, S. J., L. R. Leung, and Q. Hu, Application ofcloud microphysics to NCAR CCM2, J. Geophys.Res., 102, 16,507–16,527, 1997b.

Ghan, S. J., L. R. Leung, and J. McCaa, A com-parison of three different modeling strategies forevaluating coud and radiation parameterizations,Mon. Weather Rev., 127, 1967–1984, 1999.

148

References

Ghan, S. J., R. C. Easter, E. G. Chapman, H. Abdul-Razzak, Y. Zhang, L. R. Leung, N. S. Laulainen,R. D. Saylor, and R. A. Zaveri, A physically basedestimate of radiative forcing by anthropogenic sul-fate aerosol, J. Geophys. Res., 106, 5279–5393,2001.

Giraud, V., O. Thouron, J. Riedi, and P. Goloub,Analysis of direct comparison of cloud top temper-ature and infrared split window signature againstindependant retrievals of cloud thermodynamicphase, Geophys. Res. Lett., 28, 983–986, 2001.

Goloub, P., and O. Arino, Verification of the consis-tency of POLDER Aerosol Index over land withASTR-2/ERS-2 fire product, Geophys. Res. Lett.,27, 899–902, 2000.

Goloub, P., M. Herman, H. Chepfer, J. Riedi,G. Brogniez, P. Couvert, and G. Seze, Cloudthermodynamical phase classification from thePOLDER spaceborne instrument, J. Geophys.Res., 105, 14,747–14,759, 2000.

Guibert, S., J. Snider, and J.-L. Brenguier,The aerosol activation in marine stratocumu-lus clouds. 1. Measurement validation for a clo-sure study, J. Geophys. Res., 108, 8628, doi:10.1029/2002JD002678, 2003.

Guilyardi, E., and G. Madec, Performance ofthe OPA/ARPEGE–T21 global ocean-atmospherecoupled model, Clim. Dyn., 13, 149–165, 1997.

Gunn, K., and J. Marshall, The distribution with sizeof aggregate snowflakes, J. Meteorol., 15, 452–461,1957.

Hallett, J., and S. Mossop, Production of secondaryice particles during the riming process, Nature,249, 26–28, 1974.

Han, Q., W. B. Rossow, and A. A. Lacis, Near–global survey of effective droplet radii in liquid wa-ter clouds using ISCCP data, J. Clim., 7, 465–497,1994.

Han, Q., W. B. Rossow, J. Chou, and R. M. Welch,Global survey of the relationship of cloud albedoand liquid water path with droplet size using IS-CCP, J. Clim., 11, 1516–1528, 1998.

Han, Q., W. B. Rossow, J. Zeng, and R. Welch, Threedifferent behaviours of liquid water path of waterclouds in aerosol-cloud interactions, J. Atmos. Sci.,59, 726–736, 2002.

Hansen, J., M. Sato, and R. Ruedy, Radiative forc-ing and climate response, J. Geophys. Res., 102,6831–6864, 1997.

Hansen, J. E., M. Sato, and R. Ruedy, Long-termchanges of the diurnal temperature cycle: Implica-tions about mechanisms of global climate change,Atmos. Res., 37, 175–209, 1995.

Hartmann, D. L., and M. L. Michelsen, No evidencefor iris, Bull. Am. Meteorol. Soc., 83, 249–254,2003.

Hauglustaine, D. A., F. Hourdin, L. Jourdain, M.-A.Filiberti, S. Walters, and J.-F. Lamarque, Interac-tive chemistry in the Laboratoire de MeteorologieDynamique general circulation model: Descriptionand background tropospheric chemistry evaluation,J. Geophys. Res., submitted, 2003.

Haywood, J. M., and O. Boucher, Estimates of thedirect and indirect radiative forcing due to tropo-spheric aerosols: A review, Rev. of Geophysics, 38,513–543, 2000.

Haywood, J. M., and S. R. Osborne, The effect ofoverlying absorbing aerosol layers on remote sens-ing retrievals of cloud effective radius and cloudoptical depth, Q. J. R. Meteorol. Soc., submitted,2003.

Heymsfield, A. J., and L. J. Donner, A scheme forparameterizing ice–cloud water content in generalcirculation models, J. Atmos. Sci., 47, 1865–1877,1990.

Hocking, L. M., The collision efficiency of smalldrops, Q. J. R. Meteorol. Soc., 85, 44–50, 1959.

Hogan, R. J., and A. J. Illingworth, Deriving cloudoverlap statistics from radar, Q. J. R. Meteorol.Soc., 126, 2903–2910, 2000.

Houze, R. A., Cloud Dynamics, 573 pp., AcademicPress, San Diego, 1993.

149

References

Hudson, J. G., and S. S. Yum, Maritime-continentaldrizzle contrasts in small cumuli, J. Atmos. Sci.,58, 915–926, 2001.

Huffman, G., R. Adler, P. Arkin, A. Chang,R. Ferraro, A. Gruber, J. Janowiak, A. McNab,B. Rudolf, and U. Schneider, The Global Precipi-tation Climatology Project (GPCP) combined pre-cipitation data set, Bull. Am. Meteorol. Soc., 78,5–20, 1997.

Jakob, C., and S. A. Klein, The role of vertically vary-ing cloud fraction in the parametrization of micro-physical processes in the ECMWF model, Q. J. R.Meteorol. Soc., 125, 941–965, 1999.

Johnson, D. B., The role of giant and ultragiantaerosol particles in warm rain initiation, J. Atmos.Sci., 39, 448–460, 1982.

Jones, A., and A. Slingo, Predicting cloud–droplet ef-fective radius and indirect sulphate aerosol forcingusing a general circulation model, Q. J. R. Meteo-rol. Soc., 122, 1573–1595, 1996.

Jones, A., D. Roberts, and A. Slingo, A climatemodel study of indirect radiative forcing by anthro-pogenic sulphate aerosols, Nature, 370, 450–453,1994.

Jones, A., D. L. Roberts, M. J. Woodage, and C. E.Johnson, Indirect aerosol forcing in a climate modelwith an interactive sulphur cycle, Tech. Rep. 25,Hadley Centre, Bracknell, 34pp., 2001.

Jones, P. D., and A. Moberg, Hemispheric and large-scale surface air temperature variations: An exten-sive revision and an update to 2001, J. Clim., 16,206–223, 2003.

Junge, C., and E. McLaren, Relationship of cloudnuclei spectra to aerosol size distribution and com-position, J. Atmos. Sci., 28, 382–390, 1971.

Kandel, R., M. Viollier, P. Raberanto, J.-P. Du-vel, L. A. Pakhomov, V. A. Golovko, A. P. Tr-ishchenko, J. Mueller, E. Raschke, R. Stuhlmann,and the International ScaRaB Scientific WorkingGroup (ISSWG), The ScaRaB earth radiation bud-get dataset, Bull. Am. Meteorol. Soc., 79, 765–783,1998.

Karcher, B., and U. Lohmann, A parameterization ofcirrus cloud formation: Homogeneous freezing ofsupercooled aerosols, J. Geophys. Res., 108, 4698,doi:10.1029/2002JD003220, 2003.

Karl, T. R., P. D. Jones, R. W. Knight, G. Kukla,N. Plummer, V. Razuvayev, K. P. Gallo, J. Linde-say, R. J. Charlson, and T. C. Peterson, Asymmet-ric trends of daily minimum and maximum tem-perature, Bull. Am. Meteorol. Soc., 74, 1007–1023,1993.

Kaufman, Y., and R. S. Fraser, The effect of smokeparticles on clouds and climate forcing, Science,277, 1636–1639, 1997.

Kaufman, Y. J., and T. Nakajima, Effect of Amazonsmoke on cloud microphysics and albedo - analy-sis from satellite imagery, J. Appl. Meteorol., 32,729–744, 1993.

Kawamoto, K., and T. Nakajima, Seasonal variationof cloud particle size as derived from AVHRR re-mote sensing, Geophys. Res. Lett., 30, 1810, doi:10.1029/2003GL017437, 2003.

Kessler, E., On the Distribution and Continuity ofWater Substance in Atmospheric Circulation, Me-teorological Monographs, vol. 10, Mass, Boston,84pp., 1969.

Khairoutdinov, M., and Y. Kogan, A new cloudphysics parameterization in a large-eddy simula-tion model of marine stratocumulus, Mon. WeatherRev., 128, 229–243, 2000.

Kiehl, J. T., and K. E. Trenberth, Earth’s annualglobal mean energy budget, Bull. Am. Meteorol.Soc., 78, 197–208, 1997.

Krinner, G., Impact of lakes and wetlands on bo-real climate, J. Geophys. Res., 108, 4520, doi:10.1029/2002JD002597, 2003.

Kristjansson, J. E., Studies of the aerosol indirect ef-fect from sulfate and black carbon aerosols, J. Geo-phys. Res., 107, 4246, doi:10.1029/2001JD000887,2002.

150

References

Kristjansson, J. E., J. M. Edwards, and D. L.Mitchell, A new parametrization scheme for theoptical properties of ice crystals for use in generalcirculation models of the atmosphere, Phys. Chem.Earth, 24, 231–236, 1999.

Le Treut, H., and Z.-X. Li, Sensitivity of an atmo-spheric general circulation model to prescribed SSTchanges: Feedback effects associated with the sim-ulation of cloud optical properties, Clim. Dyn., 5,175–187, 1991.

Le Treut, H., and B. McAvaney, A model intercom-parison of equilibrium climate change in responseto CO2 doubling, in Note scientifique de l’IPSL,vol. 18, edited by J.-P. Boulanger and Z.-X. Li,IPSL, Paris, 27pp., 2002.

Le Treut, H., M. Forichon, O. Boucher, and Z.-X.Li, Sulfate aerosol indirect effect and CO2 green-house forcing: Equilibrium response of the LMDGCM and associated cloud feedbacks, J. Clim., 11,1673–1684, 1998.

Le Van, P., Description de directives pour l’utilisationde la nouvelle dynamique du modele de circu-lation generale, Tech. Rep. 147, Laboratoire deMeteorologie Dynamique du C.N.R.S., Paris, 1989.

Li, Z.-X., Ensemble atmospheric GCM simulation ofclimate interannual variability from 1979 to 1994,J. Clim., 12, 986–1001, 1999.

Liepert, B. G., and U. Lohmann, A comparison ofsurface observations and ECHAM4–GCM experi-ments and its relevance to the indirect aerosol ef-fect, J. Clim., 14, 1078–1091, 2001.

Lin, B., B. A. Wielicki, L. H. Chambers, Y. Hu, andK.-M. Xu, The iris hypothesis: A negative or pos-itive cloud feedback?, J. Clim., 15, 3–7, 2002.

Lindzen, R., M.-D. Chou, and A. Hou, Does the earthhave an adaptive infrared iris?, Bull. Am. Meteorol.Soc., 82, 417–432, 2001.

Liu, J. Y., and H. D. Orville, Numerical mod-eling of precipitation and cloud shadow effectson mountain-induced cumuli, J. Atmos. Sci., 26,1283–1298, 1969.

Locatelli, J. D., and P. V. Hobbs, Fallspeed andmasses of solid precipitation particles, J. Geophys.Res., 79, 2185–2197, 1974.

Lock, A. P., A. R. Brown, M. R. Bush, G. M. Mar-tin, and R. N. B. Smith, A new boundary layermixing scheme. Part I: Scheme description andsingle-column model tests, Mon. Weather Rev.,128, 3187–3199, 2000.

Lohmann, U., A glaciation indirect aerosol effectcaused by soot aerosols, Geophys. Res. Lett., 29,1052, doi:10.1029/2001GL014357, 2002a.

Lohmann, U., Possible aerosol effects on ice cloudsvia contact nucleation, J. Atmos. Sci., 58, 647–656,2002b.

Lohmann, U., and B. Karcher, First interactive sim-ulations of cirrus clouds formed by homogeneousfreezing in the ECHAM GCM, J. Geophys. Res.,107, 4105, doi:10.1029/2001JD000767, 2002.

Lohmann, U., and G. Lesins, Stronger constraints onthe anthropogenic indirect aerosol effect, Science,298, 1012–1015, 2003a.

Lohmann, U., and G. Lesins, Comparing con-tinental and oceanic cloud susceptibilities toaerosols, Geophys. Res. Lett., 30, 1791, doi:10.1029/2003GL017828, 2003b.

Lohmann, U., and E. Roeckner, Design and perfor-mance of a new cloud microphysics scheme devel-oped for the ECHAM general circulation model,Clim. Dyn., 12, 557–572, 1996.

Lohmann, U., J. Feichter, C. C. Chuang, and J. E.Penner, Prediction of the number of cloud dropletsin the ECHAM GCM, J. Geophys. Res., 104, 9169–9198, 1999.

Lohmann, U., J. Feichter, J. Penner, and R. Leaitch,Indirect effect of sulfate and carbonaceous aerosols:A mechanistic treatment, J. Geophys. Res., 105,21,193–12,206, 2000a.

Lohmann, U., G. Tselioudis, and C. Tyler, Why isthe cloud albedo - particle size relationship differ-ent in optically thick and optically thin clouds?,Geophys. Res. Lett., 27, 1099–1102, 2000b.

151

References

Marshall, J. S., and W. M. Palmer, The distributionof raindrops with size, J. Meteorol., 5, 165–166,1948.

Martin, G. M., D. W. Johnson, and A. Spice, Themeasurement and parameterization of effective ra-dius of droplets in warm stratocumulus clouds, J.Atmos. Sci., 51, 1823–1842, 1994.

Maynard, K., and J. Polcher, Impact of land-surfaceprocesses on the interannual variability of tropicalclimate in the LMD GCM, Clim. Dyn., 20, 613–633, 2002.

Meerkotter, R., U. Schumann, D. R. Doelling,P. Minnis, T. Nakajima, and Y. Tsushima, Ra-diative forcing by contrails, Ann. Geophysicæ, 17,1080–1094, 1999.

Menon, S., A. D. Del Genio, D. Koch, and G. Tse-lioudis, GCM simulations of the aerosol indirecteffect: Sensitivity to cloud parameterization andaerosol burden, J. Atmos. Sci., 59, 692–713, 2002.

Menon, S., J.-L. Brenguier, O. Boucher, P. Davison,A. D. Del Genio, J. Feichter, S. Ghan, S. Guib-ert, X. Liu, U. Lohmann, H. Pawlowska, J. E.Penner, J. Quaas, D. L. Roberts, L. Schuller, andJ. Snider, Evaluating cloud-aerosol process param-eterizations with single column models and ACE-2cloudy column observations, J. Geophys. Res., inpress, 2003.

Meyers, M. P., P. DeMott, and W. Cotton, New pri-mary ice–nucleation parameterizations in an ex-plicit cloud model, J. Appl. Met., 31, 708–721,1992.

Mitchell, J. F. B., D. J. Karoly, G. C. Hegerl, F. W.Zwiers, M. R. Allen, and J. Marengo, Detection ofclimate change and attribution of causes, in Cli-mate change 2001 – The scientific basis. Contri-bution of working group I to the Third AssessmentReport of the Intergovernmental Panel on ClimateChange, edited by J. T. Houghton, Y. Ding, D. J.Griggs, M. Noguer, P. J. van der Linden, X. Dai,K. Maskell, and C. A. Johnson, pp. 695–738, Cam-bridge University Press, Cambridge, 2001.

Morcrette, J.-J., Evaluation of model–generatedcloudiness: Satellite–observed and model–generated diurnal variability of brightnesstemperature, Mon. Weather Rev., 119, 1205–1224,1991.

Mossop, S. C., The origin and concentration of icecrystals in clouds, Bull. Am. Meteorol. Soc., 66,264–273, 1985.

Myhre, G., A. Myhre, and F. Stordal, Historical evo-lution of radiative forcing of climate, Atmos. Env-iron., 35, 2361–2373, 2001.

Nakajima, T., A. Higurashi, K. Kawamoto, and J. E.Penner, A possible correlation between satellite-derived cloud and aerosol microphysical parame-ters, Geophys. Res. Lett., 28, 1171–1174, 2001.

Novakov, T., and J. E. Penner, Large contribution oforganic aerosols to cloud–condensation–nuclei con-centrations, Nature, 365, 823–826, 1993.

Parol, F., J. Descloitres, and Y. Fouquart, Cloud op-tical thickness and albedo retrievals from bidirec-tional reflectance measurements of POLDER in-struments during ACE-2, Tellus, 52B, 888–908,2000.

Pawlowska, H., and J.-L. Brenguier, Microphysicalproperties of stratocumulus clouds during ACE–2,Tellus, 52B, 868–887, 2000.

Pawlowska, H., and J.-L. Brenguier, An observa-tional study of drizzle formation in stratocumu-lus clouds for general circulation model (GCM) pa-rameterizations, J. Geophys. Res., 108, 8630, doi:10.1029/2002JD002679, 2003.

Pawlowska, H., J.-L. Brenguier, and L. Schuller, Mi-crophysical and radiative properties of stratocumu-lus, Phys. Chem. Earth, 24, 927–932, 1999.

Peng, Y., U. Lohmann, R. Leaitch, C. Banic, andM. Couture, The cloud albedo-cloud droplet ef-fective radius relationship for clean and pollutedclouds from RACE and FIRE.ACE, J. Geophys.Res., 107, 4106, doi:10.1029/2000JD000281, 2002.

Peterson, T. C., T. R. Karl, P. F. Jamason,R. Knight, and D. R. Easterling, First difference

152

References

method: Maximizing station density for the calcu-lation of long-term global temperature change, J.Geophys. Res., 103, 25,967–25,974, 1998.

Pincus, R., and S. A. Klein, Unresolved spatial vari-ability and microphysical process rates in large–scale models, J. Geophys. Res., 105, 27,059–27,065,2000.

Planton, S., and B. Spagnoli, Quand la simulationnumerique reproduit et explique le rechauffementdes nuits d’ete, La Meteorologie, 42, 4–5, 2003.

Pontikis, C. A., and E. M. Hicks, Droplet activa-tion as related to entrainment and mixing in warmtropical maritime clouds, J. Atmos. Sci., 50, 1888–1896, 1993.

Pruppacher, H. R., and J. D. Klett, Microphysicsof Clouds and Precipitation, 954 pp., Kluwer Aca-demic Press, Dordrecht, 1997.

Quaas, J., Kopplung eines Atmospharenmodells undeines Ozeanmodells auf einem Meta-Computer,Master’s thesis, Institut fur Geophysik und Mete-orologie der Universitat zu Koln, 90pp., 2000.

Quaas, J., O. Boucher, and F.-M. Breon, Aerosol-cloud interactions in POLDER satellite data andin the LMDZ GCM, J. Geophys. Res., submitted,2003a.

Quaas, J., O. Boucher, J.-L. Dufresne, and H. LeTreut, Impacts of greenhouse gases and aerosol di-rect and indirect effects on clouds and radiationin atmospheric GCM simulations of the 1930-1989period, Clim. Dyn., submitted, 2003b.

Quaas, J., O. Boucher, J.-L. Dufresne, and H. LeTreut, Relative importance of greenhouse gas andaerosol forcings for climate trends of the 1950-1989period in atmospheric GCM simulations, J. Clim.,submitted, 2003c.

Radel, G., C. J. Stubenrauch, R. Holz, and D. L.Mitchell, Retrieval of effective ice crystal sizein the infrared: Sensitivity study and globalmeasurements for TIROS-N Operational Verti-val Sounder, J. Geophys. Res., 108, 4281, doi:10.1029/2002JD002801, 2003.

Ramanathan, P. Crutzen, J. Lelieveld, A. P. Mi-tra, D. Althausen, J. Anderson, M. O. Andreae,W. Cantrell, G. R. Cass, C. E. Chung, A. D.Clarke, J. A. Coakley Jr., W. D. Collins, W. C.Conant, F. Dulac, B. J. Heintzenberg, A. J. Heyms-field, B. Holben, S. Howell, C. J. Hudson, A. Ja-yaraman, J. T. Kiehl, T. N. Krishnamurti, D. D.Lubin, G. McFarquhar, T. Novakov, J. Ogren,E. I. A. Poddorny, K. Prather, K. Priestly, J. M.Prospero, F. P. K. Quinn, K. Rajeev, P. Rasch,S. Rupert, R. Sadourny, G. S. K. Satheesh, G. E.Shaw, P. Sheridan, and F. P. J. Valero, IndianOcean Experiment: An integrated analysis of theclimate forcing and effects of the great Indo-Asianhaze, J. Geophys. Res., 106, 28,371–28,398, 2001a.

Ramanathan, V., P. J. Crutzen, J. T. Kiehl, andD. Rosenfeld, Aerosols, climate, and the hydrolog-ical cycle, Science, 294, 2119–2124, 2001b.

Ramaswamy, V., O. Boucher, J. Haigh, D. Hauglus-taine, J. Haywood, G. Myhre, T. Nakajima, G. Y.Shi, and S. Solomon, Radiative forcing of climatechange, in Climate change 2001 – The scientificbasis. Contribution of working group I to the ThirdAssessment Report of the Intergovernmental Panelon Climate Change, edited by J. T. Houghton,Y. Ding, D. J. Griggs, M. Noguer, P. J. van derLinden, X. Dai, K. Maskell, and C. A. Johnson, pp.349–416, Cambridge University Press, Cambridge,2001.

Rasch, P. J., and J. E. Kristjansson, A comparisonof the CCM3 model using diagnosed and predictedcondensate parameterizations, J. Clim., 11, 1587–1614, 1998.

Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Fol-land, L. V. Alexander, D. P. Rowell, E. C. Kent,and A. Kaplan, Global analyses of SST, sea ice andnight marine air temperature since the late nine-teenth century, J. Geophys. Res., 108, 4407, doi:10.1029/2002JD002670, 2003.

Reddy, M. S., and O. Boucher, A study of the globalcyle of carbonaceous aerosols in the LMDZT gen-eral circulation model, J. Geophys. Res., submitted,2003.

153

References

Reid, J. S., P. V. Hobbs, A. L. Rangno, and D. A.Hegg, Relationshiphs between cloud droplet effec-tive radius, cloud liquid water content, and dropletconcentration for warm clouds in Brazil embeddedin biomass smoke, J. Geophys. Res., 104, 6145–6153, 1999.

Reynolds, R. W., and T. M. Smith, A high-resolutionglobal sea surface temperature climatology, J.Clim., 8, 1571–1583, 1995.

Riedi, J., M. Doutriaux-Boucher, P. Goloub, andP. Couvert, Global distribution of cloud top phasefrom POLDER/ADEOS I, Geophys. Res. Lett., 27,1707–1710, 2000.

Rosenfeld, D., TRMM observed first direct evidenceof smoke from forest fires inhibiting rainfall, Geo-phys. Res. Lett., 26, 3105–3108, 1999.

Rosenfeld, D., Suppression of rain and snow by urbanand industrial air pollution, Science, 287, 1793–1796, 2000.

Rosenfeld, D., and I. M. Lensky, Satelite–base in-sights into precipitation formation processes incontinental and maritime convective clouds, Bull.Am. Meteorol. Soc., 79, 2457–2476, 1998.

Rossow, W. B., and R. Schiffer, ISCCP cloud dataproducts, Bull. Am. Meteorol. Soc., 72, 2–20, 1991.

Rotstayn, L. D., A physically based scheme for thetreatment of stratiform clouds and precipitation inlarge–scale models. I: Description and evaluationof the microphysical processes, Q. J. R. Meteorol.Soc., 123, 1227–1282, 1997.

Rotstayn, L. D., On the “tuning” of autoconversionparameterizations in climate models, J. Geophys.Res., 105, 15,495–15,507, 2000.

Rotstayn, L. D., and J. E. Penner, Indirect aerosolforcing, quasi–forcing and climate response, J.Clim., 14, 2960–2975, 2001.

Rutledge, S., and P. Hobbs, The mesoscale andmicroscale structure and organization of cloudsand precipitation in midlatitude cyclones. VIII: Amodel for the ”seeder–feeder” process in warm–frontal rainbands, J. Atmos. Sci., 40, 1185–1206,1983.

Sassen, K., and G. C. Dodd, Homogeneous nucleationrate for highly supercooled cirrus cloud droplets, J.Atmos. Sci., 45, 1357–1369, 1988.

Schuller, L., J.-L. Brenguier, and H. Pawlowska, Re-trieval of microphysical, geometrical, and radia-tive properties of marine stratocumulus from re-mote sensing, J. Geophys. Res., 108, 8631, doi:10.1029/2002JD002680, 2003.

Schwartz, S. E., Harshvardhan, and C. M. Benkovitz,Influence of anthropogenic aerosol on cloud opticaldepth and albedo shown by satellite measurementsand chemical transport modeling, Proc. Nat. Acad.Sci. USA, 99, 1784–1789, 2002.

Sexton, D. M. H., H. Grubb, K. P. Shine, and C. K.Folland, Design and analysis of climate model ex-periments for the efficient estimation of anthro-pogenic signals, J. Clim., 16, 1320–1336, 2003.

Slingo, A., A GCM parameterization for the short-wave radiative properties of water clouds, J. At-mos. Sci., 42, 1419–1427, 1989.

Slingo, J. M., The development and verification of acloud prediction scheme for the ECMWF model,Q. J. R. Meteorol. Soc., 113, 899–927, 1987.

Smith, L. D., and D. A. Randall, Parameterizationof Cloud Microphysical Processes in the CSU Gen-eral Circulation Model, Atmos. Sci. Papers, vol.491, Department of Atmospheric Sciences, Col-orado State University, Colorado, 58pp., 1992.

Smith, R. N., A scheme for predicting layer cloudsand their water content in a general circulationmodel, Q. J. R. Meteorol. Soc., 116, 435–460, 1990.

Snider, J. R., and J.-L. Brenguier, Cloud condensa-tion nuclei and cloud droplet measurements duringACE–2, Tellus, 52B, 828–842, 2000.

Sommeria, G., and J. W. Deardorff, Subgrid-scalecondensation models of non-precipitating clouds,J. Atmos. Sci., 34, 344–355, 1977.

Spagnoli, B., S. Planton, M. Deque, O. Mestre, andJ.-M. Moisselin, Detecting climate change at a re-gional scale: The case of France, Geophys. Res.Lett., 29, 1450, doi:10.1029/2001GL014619, 2002.

154

References

Spice, A., D. W. Johnson, P. R. A. Brown, A. G.Darlison, and C. P. R. Saunders, Primary ice nuce-leation in orographic cirrus clouds: A numericalsimulation of the microphysics, Q. J. R. Meteorol.Soc., 125, 1637–1667, 1999.

Squires, P., An estimate of the anthropogenic produc-tion of cloud nuclei, J. Rech. Atmos., 2, 297–308,1966.

Stephens, G. L., Radiation profiles in extended wa-ter clouds. II: Parameterization schemes, J. Atmos.Sci., 35, 2123–2132, 1978.

Stone, D. A., and A. J. Weaver, Factors contribut-ing to diurnal temperature range trends in twen-tieth and twenty-first century simulations of theCCCma coupled model, Clim. Dyn., 20, 435–445,doi:10.1007/s00382-002-0288-y, 2003.

Stubenrauch, C., A. D. Del Genio, and W. B. Rossow,Implementation of subgrid cloud vertical structureinside a GCM and its effect on the radiation bud-get, J. Clim., 10, 273–287, 1997.

Stubenrauch, C. J., F. Eddounia, and G. Radel, Cor-relations between microphysical properties of large-scale semi-transparent cirrus and the state of theatmosphere using satellite TIROS-N OperationalVertical Sounder observations and ECMWF ERA-40 reanalyses, Atmos. Res., submitted, 2003.

Suhre, K., V. Crassier, C. Mari, R. Rosset, D. John-son, S. Osborne, R. Wood, M. Andreae, B. Bandy,and T. Bates, Chemistry and aerosols in the ma-rine boundary layer: 1-D modelling of the threeACE-2 Lagrangian experiments, Atmos. Environ.,34, 5079–5094, doi:10.1016/S1352-2310(00)00221-1, 2000.

Sun, B., and R. S. Bradley, Solar influences on cosmicrays and cloud formation: A reassessment, J. Geo-phys. Res., 107, 4211, doi:10.1029/2001JD000560,2002.

Sundqvist, H., A parameterization scheme for non–convective condensation including prediction ofcloud water content, Q. J. R. Meteorol. Soc., 104,677–690, 1978.

Svensmark, H., and E. Friis-Christensen, Variation ofcosmic ray flux and global cloud coverage - a miss-ing link in solar-climate relationships, J. Atmos.Sol.-Terr. Physics, 59, 1225–1232, 1997.

Takemura, T., H. Okamoto, A. Numaguti, A. Hig-urashi, and T. Nakajima, Global three-dimensionalsimulation and radiative forcing of various aerosolspecies, in Current Problems in Atmospheric Ra-diation, edited by W. L. Smith and Y. M. Timo-feyev, IRS 2000, pp. 733–736, A. Deepak Publish-ing, 2000.

Tanre, D., F.-M. Breon, J. L. Deuze, M. Herman,P. Goloub, F. Nadal, and A. Marchand, Global ob-servation of anthropogenic aerosols from satellite,Geophys. Res. Lett., 28, 4555–4558, 2001.

Taylor, J. P., M. D. Glew, J. A. Coakley Jr., W. R.Tahnk, S. Platnick, P. V. Hobbs, and R. J. Ferek,Effects of aerosols on the radiative properties ofclouds, J. Atmos. Sci., 57, 2656–2670, 2000.

Tetens, O., Uber einige meteorologische Begriffe,Zeitschrift fur Geophysik, 6, 297–309, 1930.

Thouron, O., Developpement d’un schema mi-crophysique adapte a la modelisation des cir-rus a meso-echelle. Apport des instruments deteledetection active pour sa validation, Ph.D. the-sis, Universite des Sciences et Technologies (LilleI), Lille, 130pp., 2003.

Tiedtke, M., A comprehensive mass flux scheme forcumulus parameterization in large-scale models,Mon. Weather Rev., 117, 1779–1800, 1989.

Tompkins, A., A prognostic parameterization for thesubgrid-scale variability of water vapor and cloudsin large-scale model and its use to diagnose cloudcover, J. Atmos. Sci., 59, 1917–1942, 2002.

Travis, D. J., A. M. Carleton, and R. G. Lauritsen,Contrails reduce daily temperature range, Nature,418, 601, 2002.

Trenberth, K. (Ed.), Climate System Modeling, 788pp., Cambridge University Press, 1995.

Tripoli, G. J., and W. R. Cotton, A numerical in-vestigation of several factors contributing to the

155

References

observed variable intensity of deep convection overSouth Florida, J. Appl. Meteorol., 19, 1037–1063,1980.

Twomey, S., The nuclei of natural cloud formation.Part II: The supersaturation in natural clouds andthe variation of cloud droplet concentration, Ge-ofis. Pura Appl., 43, 243–249, 1959.

Twomey, S., Pollution and the planetary albedo, At-mos. Environ., 8, 1251–1256, 1974.

United Nations, United nations frameworkconvention on climate change, Available onhttp://unfccc.int/resource/docs/convkp/conveng.pdf,25pp., 1992.

United Nations, Report of the world summiton sustainable development, Available onhttp://www.un.org/jsummit/, 173pp., 2002.

Vali, G., Atmospheric ice nucleation: A review, J.Rech. Atmos., 19, 105–115, 1985.

Vanbauce, C., J. C. Buriez, F. Parol, B. Bonnel,G. Seze, and P. Couvert, Apparent pressure de-rived from ADEOS–POLDER observations in theoxygen a–band over ocean, Geophys. Res. Lett., 25,3159–3162, 1998.

Verant, S., K. Laval, J. Polcher, and M. Castro, Sen-sitivity of the continental hydrological cycle to thespatial resolution over the Iberian peninsula, J. Hy-drometeorology, submitted, 2003.

Webb, M., C. Senior, S. Bony, and J.-J. Morcrette,Combining ERBE and ISCCP data to assess cloudsin the Hadley Centre, ECMWF and LMD atmo-spheric climate models, Clim. Dyn., 17, 905–922,2001.

Wetzel, M. A., and L. L. Stowe, Satellite-observedpatterns in stratus microphysics, aerosol opticalthickness, and shortwave radiative forcing, J. Geo-phys. Res., 104, 31,287–31,299, 1999.

White, P. W., IFS Documentation. Part IV: Physicalprocesses (CY25R1), Tech. rep., ECMWF, Read-ing, 166pp., 2002.

Wigley, T. M. L., Possible climate change due to SO2-derived cloud condensation nuclei, Nature, 339,365–367, 1989.

Williams, K. D., A. Jones, D. L. Roberts, C. A.Senior, and M. J. Woodage, The response of theclimate system to the indirect effects of anthro-pogenic sulfate aerosol, Clim. Dyn., 17, 845–856,2001.

Wilson, D., and S. Ballard, Microphysically basedprecipitation scheme for the UK meteorological of-fice unified model, Q. J. R. Meteorol. Soc., 125,1607–1636, 1999.

Wobrock, W., A. I. Flossmann, M. Monier, J.-M. Pi-chon, L. Cortez, J.-F. Fournol, A. Schwarzenbock,S. Mertes, J. Heintzenberg, P. Laj, G. Orsi,L. Ricci, S. Fuzzi, H. T. Brink, P. Jongejan, andR. Otjes, The Cloud Ice Mountain Experiment(CIME) 1998: experiment overview and modellingof the microphysical processes during the seedingby isentropic gas expansion, Atmos. Res., 58, 231–265, doi:10.1016/S0169-8095(01)00094-1, 2001.

Wood, R., P. R. Field, and W. R. Cotton, Autocon-version rate bias in stratiform boundary layer cloudparameterizations, Atmos. Res., 65, 109–128, 2002.

Young, K. C., A numerical simulation of wintertime,orographic precipitation: Part I. Description ofmodel microphysics and numerical techniques, J.Atmos. Sci., 31, 1735–1748, 1974a.

Young, K. C., The role of contact nucleation in icephase initiation in clouds, J. Atmos. Sci., 31, 768–776, 1974b.

Ziegler, C. L., Retrieval of thermal and microphys-ical variables in observed convective storms. Part1: Model development and preliminary testing, J.Atmos. Sci., 42, 1487–1509, 1985.

156

Remerciements

Le soutien par mon directeur de these Herve Le Treut m’a ete precieux pour tout mon travail ainsi quepour l’etablissement des contacts par les ecoles d’ete, les collaborations avec des scientifiques externes et lesconferences.

Je tiens a remercier mon co-directeur de these Olivier Boucher qui a inspire presque toutes les etudes,qui m’a repondu dans innombrables mels ou pendant nos reunions a toutes mes questions, et qui m’a apprisle fonctionnement du monde scientifique.

Je suis reconnaissant a Jean-Louis Brenguier pour tout son soutien.

Avec Marie Boucher, Surabi Menon, Jean-Louis Dufresne et Francois-Marie Breon j’ai collabore fructueuse-ment, et je leur en remercie.

Pour avoir eu la gentillesse d’etre les rapporteurs de ma these je remercie Ulrike Lohmann et Yves Fouquart,et egalement Serge Planton et Michael Schulz d’avoir participe au jury.

L’ensemble de l’equipe de modelisation du LMD m’a accueilli chaleureusement, et je suis tres reconnais-sant pour tous les conseils pratiques et scientifiques, et pour toute leur amitie.

157