CarboEurope IP Integration Meeting, 22–24 August 2005
The Carbon Cycle Data
Assimilation System (CCDAS)
Wolfgang Knorr
QUEST/U Bristol, formerly Max-Planck Institute for Biogeochemistry, Jena
with contributions from: Marko Scholze (QUEST), Jens Kattge (MPI Jena),
Nadine Gobron (JRC/IES, Ispra), Thomas Kaminski, Ralf Giering (FastOpt) and Peter
Rayner (LSCE)
Overview
• Carbon Cycle Observations
•
•
•
•
Assimilation of Eddy Covariance Data
Assimilation of Satellite "Greenness"
Assimilation of Atmospheric CO2 Data
Outlook
Fluxnet
Eddy
Covariance
Network
Key Remotely Sensed
Variables
I
TOC
ITOC
FAPAR:
[(ITOC+IS)–(ITOC+IS)]
/ ITOC
canopy
I S
I S
soil
Atmospheric CO2
Measurements
CCDAS inverse modelling period
... and more stations in CCDAS
Carbon Cycle Data Assimilation
System (CCDAS)
atm. CO2
satellite
FAPAR
eddy flux
CO2 & H2O
soil water
LAI
veg. distr.
params
& error cov.
Monte Carlo
Param. Inversion
full BETHY
CCDAS Step 1
full BETHY
collaborators:
T. Kaminski, R. Giering (FastOpt);
P. Rayner (CSIRO)
B. Pinty, N. Gobron, M. Verstraete
(JRC, Ispra)
CCDAS Step 2
BETHY+TM2
energy balance/
photosynt.
CO2 and water
fluxes + uncert.
2°x2°
Overview
• Carbon Cycle Observations
• Assimilation of Eddy Covariance Data
• Assimilation of Satellite "Greenness"
• Assimilation of Atmospheric CO2 Data
• Outlook
Carbon Cycle Data Assimilation
System (CCDAS)
atm. CO2
satellite
FAPAR
eddy flux
CO2 & H2O
soil water
LAI
veg. distr.
params
& error cov.
Monte Carlo
Param. Inversion
full BETHY
CCDAS Step 1
full BETHY
CCDAS Step 2
BETHY+TM2
energy balance/
photosynt.
CO2 and water
fluxes + uncert.
2°x2°
The Cost Function
Measure of the mismatch (cost function):
model diagnostics
measurements
1
1
-1
T
T
J(m)  [m  m0 ]Cm 0 [m  m0 ]  [y (m)  y0 ]C-1
[y
(m
)

y
]
y
0
2
2
assumed
model parameters
a priori
parameter values
error covariance matrix
of measurements
a priori error covariance
matrix of parameters
met. data
aim: sample exp{–J(m)}
BETHY
=probability density function
parameters
J
eddy flux
CO2 & H2O
(7 selected days)
Convergence of parameters (BETHY model)
Convergence of Cost Function, diagnostic vs. parameter (=Bayes) space
Fig. 1, Knorr & Kattge, GCB 2005
Fig. 4, Knorr & Kattge, GCB 2005
C4 grassland [FIFE]
1–sopt/sprior
conifer forest [Loobos]
photosynth.
respiration
Fig. 3, Knorr & Kattge 2005
stomata
energy balance
Fig. 5, Knorr & Kattge, GCB 2005
Overview
• Carbon Cycle Observations
• Assimilation of Eddy Covariance Data
• Assimilation of Satellite "Greenness"
• Assimilation of Atmospheric CO2 Data
• Outlook
Carbon Cycle Data Assimilation
System (CCDAS)
atm. CO2
satellite
FAPAR
eddy flux
CO2 & H2O
soil water
LAI
veg. distr.
params
& error cov..
Monte Carlo
Param. Inversion
full BETHY
CCDAS Step 1
full BETHY
CCDAS Step 2
BETHY+TM2
energy balance/
photosynt.
CO2 and water
fluxes + uncert.
collaborators:
B. Pinty, N. Gobron, M. Verstraete
(JRC, Ispra)
2°x2°
The Cost Function
Measure of the mismatch (cost function):
model diagnostics
measurements
1
1
-1
T
T
J(m)  [m  m0 ]Cm 0 [m  m0 ]  [y (m)  y0 ]C-1
[y
(m
)

y
]
y
0
2
2
assumed
model parameters
a priori
parameter values
error covariance matrix
of measurements
a priori error covariance
matrix of parameters
aim: minimize J(m)
at each grid cell:
m: relative contributions of
vegetation types
met. data
BETHY
parameters
J
FAPAR
Step 1: FAPAR Assimilation
prior
cover fraction of PFT:
evergreen coniferous tree
optimized
Step 1: FAPAR Assimilation
relative cover fraction: tropical evergreen trees
prior
optimized
deforestation?
Overview
• Carbon Cycle Observations
• Assimilation of Eddy Covariance Data
• Assimilation of Satellite "Greenness"
• Assimilation of Atmospheric CO2 Data
• Outlook
Carbon Cycle Data Assimilation
System (CCDAS)
soil water
LAI
veg. distr.
params
& error cov.
Monte Carlo
Param. Inversion
full BETHY
CCDAS Step 1
full BETHY
Background
CO2 fluxes*
*ocean: Takahashi et al. (1999), LeQuere et al. (2000);
emissions: Marland et al. (2001), Andres et al. (1996);
land use: Houghton et al. (1990)
atm. CO2
satellite
FAPAR
eddy flux
CO2 & H2O
CCDAS Step 2
reduced BETHY
+TM2
CO2 and water
fluxes + uncert.
Uses adjoint and Hessian
generated by TAF of
T. Kaminski, R. Giering (FastOpt);
2°x2°
The Cost Function
Measure of the mismatch (cost function):
model diagnostics
measurements
1
1
-1
T
T
J(m)  [m  m0 ]Cm 0 [m  m0 ]  [y (m)  y0 ]C-1
[y
(m
)

y
]
y
0
2
2
assumed
model parameters
a priori
parameter values
a priori error covariance
matrix of parameters
aim: minimize J(m):
m: 58 BETHY parameters
error covariance matrix
of measurements
met. data
BETHY+TM2 J
parameters
atm. CO2
Prior/Optimized Fluxes
Table 4, Rayner et al., GBC 2005
Error Covariances in Parameters
J(x)
Second Derivative
(Hessian) of J(m):
∂2J(m)/∂m2
yields curvature of J,
provides estimated
uncertainty in mopt
Figure taken from
Tarantola '87
Space of m (model parameters)
relative error reduction
1–sopt/sprior
CCDAS
photosynth.
plant resp.
from Table 1, Rayner et al., GBC 2005
soil resp.
Error Covariances in Diagnostics
Error covariance of diagnostics, y,
after optimisation (e.g. CO2 fluxes):
yi (mopt)  y i (mopt) 
Cm 

Cy (mopt)  
 m j   m j 
T
error covariance
of parameters
adjoint or
tangent linear
model
gC m-2 yr -1
gC m-2 yr -1
mean NEP 1980–2000, CCDAS
uncertainty in mean NEP 1980–2000, CCDAS
Fig. 9/10, Rayner et al., GBC 2005
Outlook
• More data: inventories, regional inversions
and budgets, satellite CO2 columns,
isotopes, O2/N2
• More components: ocean (“free”
optimization indicates no big changes)
• More processes: fire (under construction)
• Prognostic step...