Estimating biophysical parameters from CO 2 flask and flux
observations Kevin Schaefer 1, P. Tans 1, A. S. Denning 2, J.
Collatz 3, L. Prihodko 2, I. Baker 2, W. Peters 1, A. Andrews 1,
and L. Bruhwiler 1 1 NOAA Climate Monitoring and Diagnostics
Laboratory, Boulder, Colorado 2 Dept. of Atmospheric Science,
Colorado State University, Fort Collins, Colorado 3 Goddard Space
Flight Center, Greenbelt, Maryland
Slide 2
Objective Understand processes driving terrestrial CO 2 fluxes
Technique: estimate model parameters using data assimilation Model:
Simple Biosphere (SiB) Carnegie-Ames-Stanford Approach (CASA)
Observations: CO 2 concentrations from CMDL flask network CO 2
concentrations & fluxes from towers
Slide 3
Status 2-year NAS Postdoc fellowship @ CMDL Joint effort: CMDL
& CSU SibCasa in final testing Switching to EnKF Preliminary
results Offline with SiB2 & TransCom fluxes Single point @
WLEF
Slide 4
Combined SibCasa Model Simple Biosphere (SiB) Biophysical Good
photosynthesis model High time resolution CASA Biogeochemical Good
respiration model Coarse time resolution SibCasa Good GPP Model
Good respiration model High time resolution
Slide 5
Which parameters to estimate? Low High Low High Uncertainty
Influence no botherno problem no way no excuse
Slide 6
WLEF Tall Tower in Wisconsin Hourly and monthly average net CO
2 fluxes WLEF
Slide 7
Monthly Observed vs. SibCasa Fluxes at WLEF Net CO 2 Flux (
mole/m2/s) Date (year) SibCasa Observed
Slide 8
Hourly Observed vs. SibCasa Fluxes at WLEF Net CO 2 Flux (
mole/m2/s) Date (year) SibCasa Observed
Slide 9
SibCasa diurnal cycle too small at WLEF June 2-5, 1997 SibCasa
Observed Net CO 2 Flux ( mole/m2/s) Date (year)
Data Assimilation: Minimize Cost function ( ) Optimize using
Marquardt-Levenberg method (variant of inverse Hessian) No model
adjoint: approximate slope
Slide 12
Q 10 Cost Function at WLEF (no a priori) Hourly Obs: aliasing Q
10 to fix diurnal cycle
Slide 13
Initial Slow Pool Cost Function at WLEF Monthly Obs: aliasing
Slow to fix low GPP in 1998 Equilibrium Pool Size
Slide 14
Conclusions We can estimate model parameters from CO 2 data Be
careful about data assimilation correcting for model flaws
Slide 15
What process information can we extract from CO 2 flask and
flux tower observations? Ocean Processes Net Flux Biosphere
Processes Flux Tower Flask Atmospheric Transport Net Flux Fossil
Fuel
Slide 16
Objectives Use model physics to better understand mechanisms
that drive CO 2 fluxes Optimize model parameters to best match
model output & observations Estimate hard-to-measure
parameters: Q 10, turnover, pool sizes, etc. Joint effort: CMDL
& CSU
Slide 17
Postdoc Plan 6 Months for Software development Add geochemistry
from CASA to SiB2 8 months for simulations and testing Flux towers
first, then flasks 6 months writing papers Status: 3 months into
SiB-CASA development
Slide 18
DAS Setup Combine SiB3 with CASA SiB3: Photosynthesis &
turbulent fluxes CASA: biogeochemistry and respiration Integrate
Sibcasa into TM5 Use Ensemble Kalman Filter (EnKF)
Slide 19
DAS Experiments Single point: Sibcasa & flux tower data
Offline: Sibcasa & Transcom3 fluxes Compare NCEP, ECMWF, GEOS4
reanalysis Integrated: Sibcasa in TM5 & flask data
Slide 20
Problems Parameter Estimation Parameter compensation Model/data
biases EnKF 3-D [CO 2 ] field from sparse flask observations How to
incorporate CO 2 memory How to go from parameter to flask Number
ensemble members
Slide 21
Data Assimilation: Minimize Cost Function ( ) y = observations
f(x) = model output E = uncertainty x = parameter to estimate
Slide 22
Data Assimilation: Minimize Cost function ( ) observed fluxSiB2
fluxparametera priori flux uncertainty a priori uncertainty
Variance between modeled & observed fluxes
Slide 23
Data Assimilation: Minimize Cost function ( ) Iterate using
Marquardt-Levenberg method (variant of inverse Hessian) Approximate
Jacobian:
Slide 24
Data Assimilation: Minimize Cost function ( ) CO 2 Flask
Measurements Transport Models TransCom Inversion Estimated NEE SiB2
Assimilation LAI Weather Modeled NEE T Q 10 Iterate
Slide 25
Ensemble Kalman Filter (EnKF) Use ensemble statistics to
approximate terms in Kalman gain equation Run ensemble ~100 members
No adjoint required Experimental: still under development
Slide 26
History of Kevin 1984: BS in Aerospace Engineering 1984-1993:
NASA Space Shuttle, Space Station Mission to Planet Earth
1994-1997: White House 1997-2004: CSU Atmospheric Science
Slide 27
Kevins Family SusyJason
Slide 28
Simple Biosphere Model, Version 2 (SiB2) TcTc TgTg CO 2 TaTa Rh
a NEE=R-GPP LHSH Snow Canopy Canopy Air Space Soil GPP R W1W1 W2W2
W3W3 T1T1 T2T2 T3T3 T4T4 T5T5 T6T6 10-min time step 11 to 45-year
simulations
Slide 29
SiB2 Input National Centers for Environmental Prediction (NCEP)
reanalysis 1958-2002, every 6 hours, 2x2 resolution European Centre
for Medium-range Weather Forecasts (ECMWF) reanalysis 1978-1993,
every 6 hours, 1x1 resolution Leaf Area Index: Fourier-Adjustment,
Solar zenith angle corrected, Interpolated Reconstructed (FASIR)
Normalized Difference Vegetation Index (NDVI) data 1982-1998,
monthly, variable resolution
Slide 30
NOAAs global flask network Run transport backwards to estimate
CO 2 fluxes Compare estimated & SiB2 regional fluxes
Slide 31
Initial Coarse Woody Debris Pool at WLEF Monthly Obs: aliasing
to fix low GPP in 1998 Hourly Obs: aliasing to fix diurnal cycle
Equilibrium Pool Size
Global Estimated T and Q 10 Global Q 10 = 1.670.04 Agrees well
with published values (1.6-2.4) Q 10 increases with shorter time
scales Global T = 12.7 0.8 months Represents only fast turnover
pools Average between autotrophic & heterotrophic Need more
carbon pools in SiB2