CMIP6 Global Carbon Cycle Model EvaluationNeeds

Forrest M. Hoffman1,2 and James T. Randerson2

1Oak Ridge National Laboratory and 2University of California Irvine

April 30, 2014obs4MIPs–CMIP6 Planning Meeting

NASA HeadquartersWashington, District of Columbia, USA

Observed Carbon Accumulation Since 1850

Year

Tota

l Car

bon

(Pg

C)

1850 1870 1890 1910 1930 1950 1970 1990 2010

0−

5050

150

250

350

0−

5050

150

250

350Anthropogenic Emissions (Andres et al., 2011, 2012)

Atmosphere (Meinshausen et al., 2011)Ocean (Khatiwala et al., 2013)Land (by difference)

Observational estimates of anthropogenic carbon inventories in atmosphere, ocean,and land reservoirs for 1850–2010. Atmosphere carbon is a fusion of Law Dome icecore CO2 observations, the Keeling Mauna Loa record, and more recently the NOAAGMD global surface average, integrated for the purpose of forcing IPCC models. Totalland flux is computed by mass balance as follows:

∆CL =∑i

Fi − ∆CA − ∆CO .

Bias Attribution Process

Atm. CO2

Ocean exchange Net land exchange

Ocean region

Ocean processes -transport -chemistry -gas exchange

Land use change Unmanaged responses

Climate-carbon feedbacks

Concentration-carbon feedbacks

GPP and NPP Responses to CO2

Carbon storage capacity

Model inventory comparison with Khatiwala et al. (2013)

Once normalizedby theiratmosphericcarboninventories,most ESMsexhibit a lowbias inanthropogenicocean carbonaccumulationthrough 2010.

The samepattern holds forthe Sabine et al.(2004) inventoryderived usingthe ∆C*separationtechnique.

Obs

erva

tions

BC

C−

CS

M1.

1B

CC

−C

SM

1.1−

MB

NU

−E

SM

Can

ES

M2

CE

SM

1−B

GC

FG

OA

LS−

s2.0

GF

DL−

ES

M2G

GF

DL−

ES

M2M

Had

GE

M2−

ES

INM

−C

M4

IPS

L−C

M5A

−LR

MIR

OC

−E

SM

MP

I−E

SM

−LR

MR

I−E

SM

1N

orE

SM

1−M

E

Atmosphere (1850−2010)

Ant

hrop

ogen

ic C

arbo

n (P

g C

)

0

50

100

150

200

250

300

Kha

tiwal

a et

al.

(201

3)B

CC

−C

SM

1.1

BC

C−

CS

M1.

1−M

BN

U−

ES

M

Can

ES

M2

CE

SM

1−B

GC

FG

OA

LS−

s2.0

GF

DL−

ES

M2G

GF

DL−

ES

M2M

Had

GE

M2−

ES

INM

−C

M4

IPS

L−C

M5A

−LR

MIR

OC

−E

SM

MP

I−E

SM

−LR

MR

I−E

SM

1N

orE

SM

1−M

E

Ocean (1850−2010)

Ant

hrop

ogen

ic C

arbo

n (P

g C

)

0

50

100

150

200

250

Kha

tiwal

a et

al.

(201

3)B

CC

−C

SM

1.1

BC

C−

CS

M1.

1−M

BN

U−

ES

M

Can

ES

M2

CE

SM

1−B

GC

FG

OA

LS−

s2.0

GF

DL−

ES

M2G

GF

DL−

ES

M2M

Had

GE

M2−

ES

INM

−C

M4

IPS

L−C

M5A

−LR

MIR

OC

−E

SM

MP

I−E

SM

−LR

MR

I−E

SM

1N

orE

SM

1−M

E

Land (1850−2010)

Ant

hrop

ogen

ic C

arbo

n (P

g C

)

−200

−150

−100

−50

0

50

100

Kha

tiwal

a et

al.

(201

3)B

CC

−C

SM

1.1

BC

C−

CS

M1.

1−M

BN

U−

ES

M

Can

ES

M2

CE

SM

1−B

GC

FG

OA

LS−

s2.0

GF

DL−

ES

M2G

GF

DL−

ES

M2M

Had

GE

M2−

ES

INM

−C

M4

IPS

L−C

M5A

−LR

MIR

OC

−E

SM

MP

I−E

SM

−LR

MR

I−E

SM

1N

orE

SM

1−M

E

Ocean/Atmosphere (1850−2010)

Ant

hrop

ogen

ic C

arbo

n R

atio

0.0

0.2

0.4

0.6

0.8

1.0

(Hoffman et al., 2014)

(a) Ocean inventoryestimates have afairly persistentordering during thesecond half of the20th century.

(b) ESMs have awide range of landcarbon accumulationresponses toincreasing CO2 andland use change,ranging from a netsource of 170 Pg C toa sink of 107 Pg C in2010.

ESM Historical Ocean and Land Carbon Accumulation

Year

Oce

an C

Acc

umul

atio

n (P

g C

)

050

100

150

200

250

050

100

150

200

250

ObservationsBCC−CSM1.1BCC−CSM1.1−MBNU−ESM (units corrected)CanESM2 (x3) (units corrected)CESM1−BGCFGOALS−s2.0GFDL−ESM2GGFDL−ESM2MHadGEM2−ESINM−CM4IPSL−CM5A−LRMIROC−ESMMPI−ESM−LR (x3)MRI−ESM1 (units corrected)NorESM1−ME

a)

Year

Land

C A

ccum

ulat

ion

(Pg

C)

1850 1870 1890 1910 1930 1950 1970 1990 2010

−17

5−

125

−75

−25

2575

−17

5−

125

−75

−25

2575

b)

(Hoffman et al., 2014)

Contemporary CO2 Tuned Model (CCTM)

Year

CO

2 (pp

m)

1850 1875 1900 1925 1950 1975 2000 2025 2050 2075 2100

300

500

700

900

1100

300

500

700

900

1100Observations

RCP 8.5Multi−Model Mean and95th percentile rangeCCTM and95% confidence interval

We used this regression to create acontemporary CO2 tuned model(CCTM) estimate of the atmosphericCO2 trajectory for the 21st century.

The width of the probability densityis much smaller for the CCTM, byalmost a factor of 6 at 2060 andalmost a factor of 5 at 2100,indicating a significant reduction inthe range of uncertainty for theCCTM prediction.

Probability Density of Atmospheric CO2 Mole Fraction

CO2 Mole Fraction (ppm)

Den

sity

520 540 560 580 600 620 640 660 680 700

0.00

0.02

0.04

0.06

02

4

Fre

quen

cy

Multi−model PDMulti−model MeanCCTM PDCCTM Prediction

a) 2060

CO2 Mole Fraction (ppm)D

ensi

ty

750 800 850 900 950 1000 1050 1100 1150

0.00

00.

005

0.01

00.

015

0.02

00.

025

02

46

Fre

quen

cy

b) 2100

Best estimate tuned using Mauna LoaCO2 data:

At 2060: 600 ± 14 ppm, 21 ppm belowthe multi-model mean

At 2100: 947 ± 35 ppm, 32 ppm belowthe multi-model mean

(Hoffman et al., 2014)

Total Carbon Column Observing Network

• Not all ESMs carry a 3-D atmospheric CO2 tracer

• Transport of land and ocean fluxes using an offline atmospheric model may be needed for some comparisons

Keppel-Aleks et al. (J. of Climate, 2013)

• TCCON • CESM 1

Todd-Brown et al. (2014)

21st century gains and losses: (2090-2099)-(1997-2006)

Todd-Brown et al. (2014)

Chan

ges i

n

Soil C gain Soil C loss

I We co-organized inaugural meeting and ∼45 researchers participatedfrom the United States, Canada, the United Kingdom, the Netherlands,France, Germany, Switzerland, China, Japan, and Australia.

I ILAMB Goals: Develop internationally accepted benchmarks for modelperformance, advocate for design of open-source software system, andstrengthen linkages between experimental, monitoring, remote sensing,and climate modeling communities. Initial focus on CMIP5 models.

I Provides methodology for model–data comparison and baseline standardfor performance of land model process representations (Luo et al., 2012).

ILAMB 1.0 Benchmarks Now Under Development

Annual Seasonal InterannualMean Cycle Variability Trend Data Source

Atmospheric CO2Flask/conc. + transport X X X NOAA, SIO, CSIRO

TCCON + transport X X X CaltechFluxnet

GPP, NEE, TER, LE, H, RN X X X Fluxnet, MAST-DCGridded: GPP X X ? MPI-BGC

Hydrology/Energyrunoff ratio (R/P) river flow X X GRDC, Dai, GFDL

global runoff/ocean balance X Syed/Famigliettialbedo (multi-band) X X MODIS, CERES

soil moisture X X de Jeur, SMAPcolumn water X X GRACE

snow cover X X X X AVHRR, GlobSnowsnow depth/SWE X X X X CMC (N. America)

Tair & P X X X X CRU, GPCP and TRMMGridded: LE, H X X MPI-BGC, dedicated ET

Ecosystem Processes & Statesoil C, N X HWSD, MPI-BGC

litter C, N X LIDETsoil respiration X X X X Bond-Lamberty

FAPAR X X MODIS, SeaWIFSbiomass & change X X Saatchi, Pan, Blackard

canopy height X Lefsky, FisherNPP X EMDI, Luyssaert

Vegetation Dynamicsfire — burned area X X X GFED3

wood harvest X X Hurttland cover X MODIS PFT fraction

Example Benchmark Score Sheet from C-LAMP

Models

BG

C D

atasets

Uncertainty Scaling TotalMetric Metric components of obs. mismatch score Sub-score CASA′ CN

LAI Matching MODIS observations 15.0 13.5 12.0• Phase (assessed using the month of maximum LAI) Low Low 6.0 5.1 4.2• Maximum (derived separately for major biome classes) Moderate Low 5.0 4.6 4.3• Mean (derived separately for major biome classes) Moderate Low 4.0 3.8 3.5

NPP Comparisons with field observations and satellite products 10.0 8.0 8.2• Matching EMDI Net Primary Production observations High High 2.0 1.5 1.6• EMDI comparison, normalized by precipitation Moderate Moderate 4.0 3.0 3.4• Correlation with MODIS (r2) High Low 2.0 1.6 1.4• Latitudinal profile comparison with MODIS (r2) High Low 2.0 1.9 1.8

CO2 annual cycle Matching phase and amplitude at Globalview flash sites 15.0 10.4 7.7• 60◦–90◦N Low Low 6.0 4.1 2.8• 30◦–60◦N Low Low 6.0 4.2 3.2• 0◦–30◦N Moderate Low 3.0 2.1 1.7

Energy & CO2 fluxes Matching eddy covariance monthly mean observations 30.0 17.2 16.6• Net ecosystem exchange Low High 6.0 2.5 2.1• Gross primary production Moderate Moderate 6.0 3.4 3.5• Latent heat Low Moderate 9.0 6.4 6.4• Sensible heat Low Moderate 9.0 4.9 4.6

Transient dynamics Evaluating model processes that regulate carbon exchange 30.0 16.8 13.8on decadal to century timescales• Aboveground live biomass within the Amazon Basin Moderate Moderate 10.0 5.3 5.0• Sensitivity of NPP to elevated levels of CO2: comparison Low Moderate 10.0 7.9 4.1

to temperate forest FACE sites• Interannual variability of global carbon fluxes: High Low 5.0 3.6 3.0

comparison with TRANSCOM• Regional and global fire emissions: comparison to High Low 5.0 0.0 1.7

GFEDv2Total: 100.0 65.9 58.3

(Randerson et al., 2009)

Biogeochemistry–Climate System Feedbacks

comparisonsModel

experimentsSensitivity

campaignsMeasurement Manipulation

experiments sensingRemote

Measurements & Experiments Communityfor Understanding Fundamental Processes

Earth System Modeling Communityfor Predicting Impacts of Environmental Change

AmeriFlux

TestbedsModel

Modelinggroups

BenchmarkingPackage

Biogeochemistry−ClimateFeedbacks Project Earth

SystemGrid

Federation

New

mod

el im

prov

emen

tsN

ew m

easu

rem

ent c

ampa

igns

E

D C

A

B

Prototype System • Current variables:

– Above Ground Live Biomass (North America FIA, pan tropical from Saatchi et al.), Soil carbon (HWSD, NCPSCD), Burned area (GFED3), Albedo (MODIS, CERES), LAI (LAI3g), Gross Primary Production (Max Plank MTE), Net Ecosystem Exchange (Ameriflux)

• Near term targeted variables: – Terrestrial water storage (GRACE), Atmospheric CO2 (TCCON, NOAA

GMD, HIPPO, CARVE, GCP), and runoff (GRDC). • Graphics and scoring systems

– Bias, RMS, seasonal phase, interannual variability, long-term trend scores

– Global maps, variable-variable comparisons • Software

– Open source, designed to be user friendly and to enable easy entrainment of new variables

Mu, Hoffman, Riley, Koven, Lawrence, Randerson

Model Evaluation Needs

I Clear definitions of model variables and units (e.g., land usechange, net biosphere productivity)

I Additional model variables needed to diagnose process-levelbehavior, e.g.,

I FAPAR or NDVI (models simulating observations)I canopy heightI above- and below-ground litterI wood harvest and other land-use-related fluxes

I Process-response benchmarks

I Model simulators (e.g., run the land model in MODIS orA-Train mode)

I Realistic and usable uncertainty estimates on all observations

In CMIP5, landcarbon uptakecomputed as theresidual did notalways match thereported netbiosphere productivity(nbp).

Some fluxes werereported in units ofg C m−2 s−1 and somein units ofg CO2 m−2 s−1.

INM−CM4 r1i1p1 Carbon Accumulation (1850−2100)

c(1850, 2100)

Tota

l Car

bon

(Pg

C)

050

010

0015

0020

00

050

010

0015

0020

00

Emissions: 2236.77 Pg CHistorical + RCP 8.5 Atmosphere: 1386.45 Pg CESM Atmosphere: 1345.82 Pg CESM Ocean: 860.66 Pg CESM Land (from residual): 30.30 Pg CESM Land (from nbp): 201.23 Pg CESM Land (from nbp and luc): 16.13 Pg C

Year

Tota

l Car

bon

(Pg

C)

1850 1900 1950 2000 2050 2100

−20

0−

100

010

020

030

0

−20

0−

100

010

020

030

0

ESM Land (from residual): 30.30 Pg CESM Land (from nbp): 201.23 Pg CESM Land (from nbp and luc): 16.13 Pg C

Global LAI for 47 CMIP5 Simulations Compared to MODIS

CMIP5 Model Monthly Mean Leaf Area Index (2000−2009)

Year

Leaf

Are

a In

dex

(m2 m

−2)

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 20100.

50.

91.

31.

72.

12.

52.

93.

33.

7

0.5

0.9

1.3

1.7

2.1

2.5

2.9

3.3

3.7

MODIS−level−3 v5BCC−CSM1.1 r1i1p1BCC−CSM1.1−M r1i1p1BNU−ESM r1i1p1CanESM2 r1i1p1CanESM2 r2i1p1CanESM2 r3i1p1CanESM2 r4i1p1CanESM2 r5i1p1CCSM4 r1i1p1CCSM4 r2i1p1CCSM4 r3i1p1CCSM4 r4i1p1CCSM4 r5i1p1CCSM4 r6i1p1CESM1−BGC r1i1p1CESM1−CAM5 r1i1p1CESM1−CAM5 r2i1p1CESM1−CAM5 r3i1p1CESM1−WACCM r2i1p1GFDL−CM3 r1i1p1GFDL−ESM2G r1i1p1GFDL−ESM2M r1i1p1HadGEM2−CC r1i1p1HadGEM2−CC r2i1p1HadGEM2−CC r3i1p1HadGEM2−ES r1i1p1HadGEM2−ES r2i1p1HadGEM2−ES r3i1p1HadGEM2−ES r4i1p1INM−CM4 r1i1p1IPSL−CM5A−LR r1i1p1IPSL−CM5A−LR r2i1p1IPSL−CM5A−LR r3i1p1IPSL−CM5A−LR r4i1p1IPSL−CM5A−MR r1i1p1IPSL−CM5B−LR r1i1p1MIROC5 r1i1p1MIROC5 r2i1p1MIROC5 r3i1p1MIROC−ESM−CHEM r1i1p1MIROC−ESM r1i1p1MPI−ESM−LR r1i1p1MPI−ESM−LR r2i1p1MPI−ESM−LR r3i1p1MPI−ESM−MR r1i1p1NorESM1−ME r1i1p1NorESM1−M r1i1p1

We need “observations” and uncertainties useful for comparisonwith model results.

Acknowledgments

This research was sponsored by the Climate and Environmental SciencesDivision (CESD) of the Biological and Environmental Research (BER) Programin the U. S. Department of Energy Office of Science and the National ScienceFoundation (AGS-1048890). This research used resources of the NationalCenter for Computational Sciences (NCCS) at Oak Ridge National Laboratory(ORNL), which is managed by UT-Battelle, LLC, for the U. S. Department ofEnergy under Contract No. DE-AC05-00OR22725.

We acknowledge the World Climate Research Programme’s Working Group onCoupled Modelling, which is responsible for CMIP, and we thank the climatemodeling groups for producing and making available their model output. ForCMIP the U. S. Department of Energy’s Program for Climate Model Diagnosisand Intercomparison provides coordinating support and led development ofsoftware infrastructure in partnership with the Global Organization for EarthSystem Science Portals.

References

R. J. Andres, J. S. Gregg, L. Losey, G. Marland, and T. A. Boden. Monthly, global emissions of carbon dioxide fromfossil fuel consumption. Tellus B, 63(3):309–327, July 2011. doi: 10.1111/j.1600-0889.2011.00530.x.

F. M. Hoffman, J. T. Randerson, V. K. Arora, Q. Bao, P. Cadule, D. Ji, C. D. Jones, M. Kawamiya, S. Khatiwala,K. Lindsay, A. Obata, E. Shevliakova, K. D. Six, J. F. Tjiputra, E. M. Volodin, and T. Wu. Causes andimplications of persistent atmospheric carbon dioxide biases in Earth System Models. J. Geophys. Res.Biogeosci., 119(2):141–162, Feb. 2014. doi: 10.1002/2013JG002381.

S. Khatiwala, T. Tanhua, S. Mikaloff Fletcher, M. Gerber, S. C. Doney, H. D. Graven, N. Gruber, G. A. McKinley,A. Murata, A. F. Rıos, and C. L. Sabine. Global ocean storage of anthropogenic carbon. Biogeosci., 10(4):2169–2191, Apr. 2013. doi: 10.5194/bg-10-2169-2013.

M. Meinshausen, S. Smith, K. Calvin, J. Daniel, M. Kainuma, J.-F. Lamarque, K. Matsumoto, S. Montzka,S. Raper, K. Riahi, A. Thomson, G. Velders, and D. P. van Vuuren. The RCP greenhouse gas concentrationsand their extensions from 1765 to 2300. Clim. Change, 109(1):213–241, Nov. 2011. doi:10.1007/s10584-011-0156-z.

J. T. Randerson, F. M. Hoffman, P. E. Thornton, N. M. Mahowald, K. Lindsay, Y.-H. Lee, C. D. Nevison, S. C.Doney, G. Bonan, R. Stockli, C. Covey, S. W. Running, and I. Y. Fung. Systematic assessment of terrestrialbiogeochemistry in coupled climate-carbon models. Global Change Biol., 15(9):2462–2484, Sept. 2009. doi:10.1111/j.1365-2486.2009.01912.x.

C. L. Sabine, R. A. Feely, N. Gruber, R. M. Key, K. Lee, J. L. Bullister, R. Wanninkhof, C. S. Wong, D. W. R.Wallace, B. Tilbrook, F. J. Millero, T.-H. Peng, A. Kozyr, T. Ono, and A. F. Rios. The oceanic sink foranthropogenic CO2. Science, 305(5682):367–371, July 2004. doi: 10.1126/science.1097403.

Extra Slides

Cox et al. 2013

Diagnostics of the phase of the annual cycle of atm. CO2

Keppel-Aleks et al. (J. of Climate, 2013)

NOAA GMD Observations CESM 1

Todd-Brown et al. (2014) doi: 10.5194/bgd-10-18969-2013

Arora et al. (2013)