CONSTRUCTING REGIONAL CO2 FLUXES USING FLUX-TOWER UPSCALING AND

ATMOSPHERIC BUDGETS

Results from the Chequamegon Ecosystem-Atmosphere Study (ChEAS) and beyond

K.J. Davis1

A.E. Andrews2, J.A. Berry3, P.V. Bolstad4, M.P. Butler1, J. Chen5, B.D. Cook4, A.R. Desai1, A.S. Denning6, F.A. Heinsch7, B.R. Helliker8, N.L. Miles1, A.

Noormets5, D.M. Ricciuto1, S.J. Richardson1, M. Uliasz6, W. Wang9

1Dept. of Meteorology, The Pennsylvania State University; 2Global Monitoring and Division, NOAA; 3Department of Global Ecology, Carnegie Institution of Washington; 4Dept. of Forest Resources,

University of Minnesota; 5Dept. of Earth, Ecological and Environmental Sci, The University of Toledo; 6Dept. of Atmospheric Science, Colorado State University; 7School of Forestry, University of

Montana; 8Department of Biology, University of Pennsylvania; 9Pacific Northwest National Laboratory.

davis@met.psu.edu

LSCE, Gif-sur-Yvette, 7 March, 2006

outline• History and goals• Flux tower measurements

– Tall tower flux measurements– Regional upscaling– Model-data synthesis with flux measurements

• Atmospheric boundary layer continuous CO2 measurements– Atmospheric profiling– Atmospheric budgets

• Future directions– Enhanced regional upscaling and the ‘ring of towers’– Prediction and detection– Continental-scale measurements and inversions

Pho

to c

redi

t:

UN

D C

itatio

n cr

ew,

CO

BR

A

WLEF tall tower (447m)CO2 and H2O flux measurements at: 30, 122 and 396 mCO2 mixing ratio measurements at: 11, 30, 76, 122, 244 and 396 m

WLEF flux and mixing ratio observatory

History

• NOAA tall tower program begins, 1992(?). Pieter Tans and Peter Bakwin.

• WLEF tower instrumented to measure CO2 mixing ratios, 1994.

• WLEF tower instrumented to measure CO2 fluxes, 1995. Davis and Bakwin.

• “You have the perfect site…” Denning.• Many complementary studies are initiated

in the “footprint” of the WLEF tower, 1997.

Goals of the ChEASAt hourly to multi-year time scales, and regional

spatial scales:– Determine ecosystem-atmosphere fluxes of carbon

and water;– Determine the processes governing these fluxes;– Develop the capacity to predict how these fluxes will

change as climate changes.

Characterization of fluxes at regional scales requires advances in methodology.

The development of regional flux measurement methodology is a central focus of the ChEAS.

http://cheas.psu.edu

Methods

Flux of carbon across this plane= tower or aircraft flux approach

-

Change inforest biomassover time = forest inventory approach

Change in atmospheric concentration of CO2 overtime = inversion or ABL budget approach.

Change in CO2 concentration in a smallbox over time = chamber flux approach

Atmospheric approaches to observing the terrestrial carbon cycle

Ci

i

i

i Sx

CU

x

CU

t

C

''

Time rate ofchange (e.g. CO2)

Mean transport

Turbulenttransport (flux)

Source in theatmosphere

Average over the depth of the atmosphere (or the ABL):

0Cz

i

i

CC C

t x

FU

z

F

F0C encompasses all surface exchange: Oceans, deforestation,

terrestrial uptake, fossil fuel emissions.

Inversion study: Observe C, model U, derive FFlux study: Observe F directly

Methodological gap

Methodological gap

Upscaling

Downscaling

Airborne flux

Ch

am

be

r flu

x o

r e

xp p

lot

Tower flux

Forest inventory Inverse study

year

month

hour

day

Tim

e S

cale

Spatial Scale

(1m)2 = 10-4ha

(1000km)2 = 108ha

(100km)2 = 106ha

(10km)2 = 104ha

(1km)2 = 102ha

Rearth

Complementary nature of atmospheric inversions and flux upscaling

Atmospheric inversion Flux upscaling

Excellent spatial Intrinsically local integration measurements.

Strong constraint on Difficult to upscale fluxflux magnitude magnitudes due to

ecosystem complexity.

Poor temporal Excellent temporalresolution resolution

Limited process More processunderstanding. understanding

ChEAS observations

Tall tower with Fco2, [CO2] Radar and ceilometer ABL profiling[CO2] tower network Airborne and satellite remote sensingFlux tower network Chamber and sap flux measurementsAirborne [CO2] profiles Biometric measurementsFTIR column [CO2]

Powered parachute photograph: M. Jensen

View from 396m on the WLEF tower: OK Tower Service

Region: Flat, heterogeneous, forested, managed, rich in wetlands, low in humans

4 meter 30 meter 1 kilometer

I. Flux tower resultsTall tower flux measurements

Flux measurement method

' '

0

' '

0

0

zsC

zs

zs

Cstorage turbulent advection

Cstorage turbulent

CNEE dz wC

t

C C u CU W dz

x z x

NEE F F F

NEE F F

CCz NEEFcw 00''

Yi et al, 2000

Eddy-covariance methods summary

• Sonic axes are rotated into the long-term mean wind direction.

• Fast-response CO2 and H2O measurements calibrated from slow-response profile measurements.

• Long tubes are used to sample CO2 and H2O. Lag-time correction applied.

• Spectral correction for high-frequency loss applied. It is substantial for H2O fluxes.

• Integral of the cospectrum indicates 1 hour averaging time needed for 396 m flux measurement.

• Hourly random sampling errors are large.

One mustcapture the large and small eddies

Berger et al, 2001

Random errors – a finite number of eddies are counted in one hour

Random sampling errors for any one hour can be as large asthe magnitude of the measured flux!

Berger et al, 2001, following Lenschow and Stankov, 1986.

Radar ABL depth

WLEF fluxes

CO2 profile

Davis et al, 2003

Daily cycle of ABL depth, and CO2 fluxes and mixing ratios

“Preferred” NEE• Data is taken from 30m at night and 122 or 396m during the day (the

highest level where there is turbulent flow) when all data are available.

• If data are missing, any existing flux measurement is used.• Data are screened out when the level of turbulence is very low. CO2

is probably draining down hill.• Early in the morning upper level data from WLEF is replaced with

30m data (Yi et al, 2000) because the flow appears to be systematically 2-D.

• Thus from 3 NEE measurements, one “preferred” flux measurement is save for each hour. (But all flux levels and components are reported.)

• Some bias exists among flux measurement levels. Contribution to annual NEE is of the order of a few tens of gC m-2 yr-1 (Ricciuto et al, in review).

Nighttime drainage flows?

Coo

k et

al,2

004;

Dav

is e

t al

, 20

03

Loss of flux at low turbulence levelsat the Willow Creek tower.

WLEF morning advection?

• Compute Del-NEE among levels.

• Find a persistent signature of advection during the morning transition.

• Loss of storage is not offset by turbulent flux.

• Hypothesis: Venting of the nocturnal pool occurs elsewhere, at a persistent location in the landscape?

Yi et al, 2000, JGR.Due to storage term.

Storage terms are small.

Flux measurement method

' '

0

' '

0

0

zsC

zs

zs

Cstorage turbulent advection

Cstorage turbulent

CNEE dz wC

t

C C u CU W dz

x z x

NEE F F F

NEE F F

CCz NEEFcw 00''

Yi et al, 2000

Differences among levels at WLEF

• Grassy clearing is a significant part of daytime 30m footprint but not much of the 122m or 396m footprints.

• Difference between 30m and 122m implies that using 30m may cause daytime fluxes to be underestimated by 8-10%

• Daytime 396m fluxes 33% larger than 30m. Can’t explain 122-396 meter difference.

Davis et al, 2003; Wang et al, in press A and B; Ricciuto et al, in review

U* screening bias at the WLEF tower

• We use a u* cutoff value of 0.2 ms-1

• This screens about 50% of nighttime growing season data.

• Data indicates modest flux loss even with 0.2 cutoff.

Average annual NEE as a function of u* cutoff

U* cutoff (ms-1)

Ann

ual N

EE

(gC

m-2yr

-1)

1997-2001 average

Ricciuto et al, in review

Hourly fluxesat WLEF for1997, observedand filled.

Davis et al, 2003.

Net ecosystem-atmosphere exchange of CO2 in northern

Wisconsin

A net source of CO2 to the atmosphere!

…

year after year!

I. Flux tower resultsRegional upscaling

WLEF tall tower

wetland

mature hardwood

old growth

• Large differences in growing season fluxes among sites.

• Net annual source of CO2 to the atmosphere observed at WLEF, caused by large respiratory fluxes.

Desai et al, in press; Wang et al, in press A and B; Ricciuto et al, in review.

ChEAS observations

Tall tower with Fco2, [CO2] Radar and ceilometer ABL profiling[CO2] tower network Airborne and satellite remote sensingFlux tower network Chamber and sap flux measurementsAirborne [CO2] profiles Biometric measurementsFTIR column [CO2]

 

NEE (gC m-2)

Respiration (gC m-2)

Photosynthesis (gC m-2)

WLEF 1997 27 991 964

WLEF 1998 48 986 938

WLEF 1999 100 1054 954

WLEF 2000 74 1005 931

WLEF 2001 141 1067 926

WLEF average 78 1021 942

Willow Creek 2000 -347 762 1109

Willow Creek 2001 -108 741 849

Willow Creek 2002 -437 648 1085

Willow Creek average -297 717 1014

Lost creek 2001 1 759 758

Lost Creek 2002 -58 631 689

Lost Creek average -30 695 724

NEE and gross fluxes at ChEAS sites: 1997-2002

The difference appears to be large respiratory fluxes at WLEF

(evident in nighttime flux data)

Drying wetlands?Disturbance/logging?

Regional flux estimates• Upscaling

1. Aggregate stand-level flux tower measurements. Desai et al, in press, AgFMet.

2. Flux footprint decomposition using the WLEF tall tower.

Wang et al, in press, JTech, JGR.

• Atmospheric budget1. Traditional ABL budget using WLEF tall tower [CO2]

dataWang et al, submitted and in preparation

2. ABL-free troposphere CO2 mixing ratio differenceHelliker et al, 2004; Bakwin et al, 2004

3. Ring of towers, mesoscale inversionUliasz et al, under construction

WLEF 2003 May-Sept fluxes were “decomposed” using a flux footprint model, simple ecosystem model, and a six stand-type vegetation map.

-Forested wetlands, mature deciduous and young aspen are implicated as strong respiratory sources.

-Comparison of WLEF “mature deciduous” and Willow Creek fluxes suggest differences exist within this vegetation class.

(Wang et al, in press JTECH, JGR)

Stand-level tower upscaling• Twelve stand-level

flux measurements are matched to vegetation categories.

• Stand age since disturbance is a primary control on the long-term net carbon flux in the region

Desai et al, in press.

Landsat-based land cover map (WISCLAND) used for upscaling – 40x40 km2

4 meter 30 meter 1 kilometer

WLEF region bottom-up comparisons Jun-Aug 2003

0

100

200

300

400

500

600

700

800

NEE * -1 ER GEP

gC

m-2

Tall-tower Footprint weighted decomposition Multi-tower aggregation

•Comparison of two independent upscaling approaches is promising.•Uncertainty in each aggregate flux, however, is fairly large and difficult to quantify.

(Desai et al, in press, AFM; Wang et al, in press JTech, JGR)

Regional upscaling appears to work? Independent methods and data!

Forested wetlands and young aspen implicated as strong respiratory sources

Uncertainties in region upscaling:Flux footprint accuracy

Land cover classificationRepresentativeness of stand-level flux measurements

Systematic errors in eddy-covariance flux measurements

WLEF tall tower

wetland

mature hardwood

old growth

• Interannual variability in WLEF fluxes are statistically significant, and strongly correlated with climate (and, for 2001, insects).

• Multi-year record begins to suggest degree of coherence in interannual variability among sites.

Desai et al, 2005, in press; Ricciuto et al, in review.

Hypotheses: Temporal variability in NEE of CO2 is governed primarily by

climate and weather.

Site-to-site variability in NEE of CO2 is governed primarily by ecosystem properties.

Temporal variability in NEE is easier to upscale than, say, the summer regional value of NEE?

Plans:Apply parameter estimation to multiple towers over multiple

years. Test hypotheses.Assess applications to inverse modeling, carbon cycle

prediction.

II. Atmospheric boundary layer continuous CO2 measurements

Atmospheric profilingVTTs

seasonal phase lagsynoptic cycles

Atmospheric budgets

Diurnal cycle of CO2 in the ABL

Bak

win

et a

l, 19

98

Daily Mixing Ratio Profile from a Tall Tower

Midday difference in CO2 between the mid-CBL and the surface layer

If You Prefer Numbers…

Month CO2 (ppm) at 30m, midday

CO2 (ppm) at 396m, midday

CO2 (ppm) 30m-396m,

midday σ(396m-30m)

CO2 (ppm) at 396m, entire day

CO2 (ppm)396m(pm)

396m(entire)

1 373.15 372.53 0.62 1.60 372.46 0.07

2 375.34 374.57 0.78 1.72 373.96 0.60

3 372.91 372.69 0.22 0.85 372.73 -0.04

4 370.75 370.91 -0.16 0.22 371.23 -0.32

5 363.91 364.68 -0.77 0.94 366.21 -1.53

6 358.49 359.96 -1.47 1.58 362.54 -2.59

7 352.50 353.63 -1.13 0.98 354.59 -0.97

8 355.95 356.72 -0.77 0.93 356.85 -0.13

9 364.19 364.45 -0.26 0.64 364.76 -0.31

10 371.18 370.49 0.69 2.38 370.49 -0.0005

11 373.55 373.06 0.49 0.85 372.89 0.17

12 374.25 373.55 0.70 1.26 373.26 0.30

AnnualMean 367.18 367.27 -0.09 367.67 -0.40

Monthly Summary for 1998

Synoptic variability in CO2

What Is This Correction?

Following the mixed layer similarity theory of Wyngaard & Brost [1984] and Moeng & Wyngaard [1989], the vertical gradient of a scalar in the boundary layer:

0

* *

izb t

i i i i

wcwcC z zg g

z z w z z w z

where

gb and gt are bottom-up and top-down gradient functions scaled by boundary layer depth zi

w* is the convective velocity scale

wc0 and wczi are the surface and entrainment fluxes of the scalar C

The Gradient Functions

The LES gradient functions are from a study by Patton et al. [2003].

The observed gradient functions will be in Wang et al., [in prep].

Hourly 396 – 30 m CO2 difference Spring/Summer

What a surface layer observation is missing:

• Unlike a tall tower:– Nocturnal boundary layer profile is missing

(see Wang et al, in prep, budget estimates)

– Midday observations only (night very hard to interpret, though some are trying)

– Limited number of species observed (add flasks to VTTs?)

– Surface layer measurement adds bias, variance

(but not a great deal!)

Net ecosystem-atmosphere exchange of CO2 in northern

Wisconsin

Flux-CBL-FT phase lagCBL mixing ratio leads local NEE. (Davis et al, 2003; Yi et al, 2004)

Evidence of large-scale transport. (Hurwitz et al, 2004)

ABL-FT CO2 difference used to compute regional fluxes (Helliker et al, 2004).

Applied successfully to 4 flux tower sites by Bakwin et al (2004).

Regional-Scale InventoriesRegional-Scale Inventories

Simple observational approaches show signs of convergence

Uncertainties in methods are large

Plan:Enhance the upscaling approach.Deploy additional CO2 sensors.Reduce uncertainty in both approaches and

intercompare again.

“ring” of towers inversion

Tall tower with Fco2, [CO2] Radar and ceilometer ABL profiling[CO2] tower network Airborne and satellite remote sensingFlux tower network Chamber and sap flux measurementsAirborne [CO2] profiles Biometric measurementsFTIR column [CO2]

1200 UTCApril 29, 2004

CO2 from 5 sites, April 29, 2004

The Richardson-Miles Package

For more information, see www.amerifluxco2.psu.edu

Performance Testing

100 120 140 160 180 200 220 240

0

3

6

Day of Year

[CO

2] PSU

[C

O2] C

MD

L (ppm

)

a

100 120 140 160 180 200 220 2400.4

0.2

0

0.2

0.4

Day of Year

[CO

2] PSU

[C

O2] C

MD

L (ppm)

b

Difference between the PSU system and WLEF 76m CO2 measurements in a test from April-August 2004. [Miles/Richardson/Uliasz, in prep.]

Difference of daily averages

See http://www.amerifluxco2.psu.edu for calibrated CO2 at AmeriFlux sites

Ring of towers,Summer 2004

Ideas for research next fall

• Consider how (when?) to merge flux and mixing ratio measurements in a single inversion.

• Study the spatial and temporal coherence of flux (and mixing ratio?) observations, and consider the implications for inverse flux estimates and observational network design.

• Evaluate the ability of forwards models of CO2 transport to resolve, e.g., seasonal and synoptic events.

– Region?– Continent?

Acknowledgements

Department of Energy Terrestrial Carbon Processes Program

National Institutes for Global Environmental Change Midwestern Regional Center, DoE

National Oceanic and Atmospheric Administration Office of Global Programs

National Science Foundation Division of Environmental Biology

National Aeronautics and Space Administration Terrestrial Ecology Program

Instruments at WLEF

Ber

ger

et a

l, 20

01

Instruments at WLEF• Two “profiling” LI-CORs in the trailer, one sampling

396m, one cycling among all 6 levels. “Slow” time response. High-precision and accuracy calibration (Bakwin et al, 1998). C-bar.

• Vaisala humidity and temperature sensors at 3 levels (30, 122 and 396m). “Slow” Q-bar, T-bar.

• Three sonic anemometers (30, 122 and 396m). w’, T’• Three LI-CORs in the trailer, one for each sonic level.

“Fast” time response. Long tubes, big pumps. Measure CO2 and H2O. c’, q’

• Two LI-CORs on the tower (122 and 396m). “Fast” time response. Short tubes, smaller pumps.

Calibration of “fast” CO2 and H2O sensors at ChEAS towers

• Calibration occurs using the fluctuations in the ambient atmospheric CO2 and H2O mixing ratios.

• “Slow” sensors provide absolute values of these mixing ratios used to calibrate the “fast” LI-CORs.

• Ideal gas law corrections to LI-COR cell temperature, pressure and humidity are applied.

• Calibration slope and intercept are derived every 2 days. These values are smoothed (monthly running mean) to derive the long-term calibration factors used for the “fast” LI-CORs.

Calibration of “fast” CO2 and H2O sensors

Ber

ger

et a

l, 20

01

What’s up? (Sonic rotations)

• Sonic anemometers are oriented perfectly in the vertical, (and the wind’s “streamlines” aren’t always perpendicular to gravity).

• Data is collected over a long time (about a year) and we define “up” by forcing the mean vertical wind speed to be zero.

Sonic rotations

Ber

ger

et a

l, 20

01

Lag time calculation• We must correct for the delay between the CO2 and H2O

measurements and the vertical velocity measurements.• Lag time is determined by finding the maximum in the lagged

covariance between vertical velocity and CO2/H2O for every hour.

Level (m) IRGA position

Tube length (m)

Lag time (s)

Tube inner diameter (m)

Flow rate (L min-1)

Reynolds number

396 Trailer 406 87 0.009 17.8 2640

122 Trailer 132 23 0.009 21.9 3250

30 Trailer 40 16 0.009 9.5 1420

396 Tower 5 1.7 0.0032 1.4 592

122 Tower 5 1.1 0.0032 2.2 915

Ber

ger

et a

l, 20

01

Lag time calculation

Ber

ger

et a

l, 20

01

Spectral corrections• Flow through tubes smears out some of the atmospheric

fluctuations, especially the small (high frequency) eddies.– Obvious for H2O. Much worse than theory predicts.– Not directly observed for CO2. Small effect.

• The sonic anemometer (virtual) temperature measurement is not smeared out, so we use similarity between the virtual temperature spectrum and the water vapor spectrum to correct for the loss of high frequency eddies in H2O.

• We use past studies of flow in tubes to correct for the loss of high frequency eddies in CO2.

Spectral corrections

Ber

ger

et a

l, 20

01

CO2

H2O

Tv

Spectral corrections

Level (m)

IRGA position

CO2 (day)

CO2 (night)

H2O

396 Trailer 1 7 16

122 Trailer 1.5 9 19

30 Trailer 5 12 21

396 Tower <0.1 1 13

122 Tower <0.1 1 11

Table shows the typical % of flux lost due to smearing of small eddies.

Ber

ger

et a

l, 20

01