Laboratoire des Sciences du Climat et de l'Environnement
P. Peylin, C. Bacour, P. Ciais,
H. Verbeek, P. Rayner
Flux data to highlight model Flux data to highlight model
deficienciesdeficiencies
& &
The use of satellite data and flux data The use of satellite data and flux data
to optimize ecosystem model to optimize ecosystem model
parameters parameters
Variational assimilation scheme to improve ORCHIDEE model
Data at the site level
NEE, H, and LE, fluxes fAPAR time series (SPOT – 40m and MERIS – 1 km)
Optimization of the ORCHIDEE vegetation modelOptimization of the ORCHIDEE vegetation model
Scientific issuesScientific issues
What do we learn from the optimisation process ?
Can we combine flux data and satellite fAPAR at the site level ?
objectivesobjectives
LMDZ-GCM«on-line»
anthropogeniceffects
STOMATESTOMATE SECHIBASECHIBAEnergy balanceWater balance
Photosynthesis
Carbon balanceNutrient balances
phenology, roughness, albedo
stomatal conductance, soil temperature and
water profiles
precipitation, temperature, radiation, ...
sensible and latent heat fluxes, CO2 flux,
albedo, roughness, surface and soil temperature
NPP, biomass, litter, ...
BiosphereBiosphere
AtmosphereAtmosphere
dailydaily ½ h½ h
year
lyye
arlyVegetation structure
LAI, Vegetation type,biomass
prescribed Dynamic (LPJ)
Climate data« off line »
The ORCHIDEE vegetation modelThe ORCHIDEE vegetation model
Optimizer BFGSJ(X) and dJ(X)/X
Variational assimilation systemVariational assimilation system
flux tower measurements
PFT compositionecosystem parameters
initial conditions
parameters(X)
J(X)M(X)
Yflux
satellitefAPARYfAPAR
J(X)J(X)
climate NEE, H, LE
Governing processes and parameters to optimizeGoverning processes and parameters to optimize
Carbon assimilation
Autotrophic respiration
Heterotrophic respiration
Plant phenology
Energy balance
Hydrology
Kvmax, Gsslope, LAIMAX, SLA, ThetaLeaf
frac_resp_growth, respm_T_slope, respm_T_ord
Q10, Hc, Kresph
Kgdd, Tsen, Leafage
albedo, capasoil, r_aero
depth_soil_res
J(X) = (Yfluxdaily-M(X))T Rseason
-1 (Yfluxdaily-M(X)) +
(Yfluxdiurnal-M(X))T Rdiurnal
-1 (Yfluxdiurnal-M(X)) +
(YfAPAR-M(X))T RfAPAR-1 (YfAPAR-M(X)) +
(X-X0)T P-1 (X-X0)
Bayesian misfit functionBayesian misfit function
Few technical aspectsFew technical aspects
Gradient of J(X) computed by finite differences ! (adjoint under completion)
How to account for ½ hourly data/model error correlations ?
Relative weight between H, LE, FCO2, Rn ?
How to treat thresholds linked to phenology ? (i.e. GDD,…)
Technical difficultiesTechnical difficulties
daily means
diurnal cycle
fAPAR
prior information
Model – data fit for several forest ecosystems
Highlight of model deficiencies !
• Temperate deciduous forest:HE (96-99), HV (92-96), VI (96-98), WB (95-98)
• Temperate conifers forest:AB (97-98), BX (97-98), TH (96-00), WE (96-99)
• Boreal conifers forest:FL (96-98), HY (96-00), NB (94-98), NO (96-98)
1 year 1 year 1 year 1 year
AB
(97
-98)
BX
(97
-98)
TH
(98
-99)
WE
(98
-99)
FCO2 (gC/m2/Jour) FH2O (W/m2)
a priori model
Optimized model
Observations
Seasonal cycle fit: temperate conifers
Diurnal Cycle
a priori model
Optimized model
Observations
AB
(97
-98)
BX
(97
-98)
TH
(98
-99)
WE
(98
-99)
FCO2 FSENS(μmol/m2/s) (W/m2) (W/m2)
FH2O
Diurnal cycle fit: temperate conifers
Diurnal Cycle Diurnal Cycle
AB
(97
-98)
BX
(97
-98)
TH
(98
-99)
WE
(98
-99)
FCO2 FSENS(μmol/m2/s) (W/m2) (W/m2)
Overestimation of the sensible heat flux during the night
Delay between model and observed FCO2
FH2O
Diurnal Cycle Diurnal Cycle Diurnal Cycle
a priori model
Optimized model
Observations
Diurnal cycle fit: temperate conifers
1 year 1 year 1 year 1 year
HE
(97
-98)
HV
(94
-95)
VI
(97-
98)
WB
(95
-96)
FCO2 (gC/m2/Jour) FH2O (W/m2)
Onset of the growingseason not fully captured !
a priori model
Optimized model
Observations
Seasonal cycle fit: temperate deciduous
1 year 1 year 1 year 1 year
FL
(97
-98)
HY
(98
-99)
NB
(96
-97)
NO
(96
-97
)FCO2 (gC/m2/Jour) FH2O (W/m2)
a priori model
Optimized model
Observations
Instabilities because of snow falls
Seasonal cycle fit: boreal conifers
Complementarity between fAPAR and flux data ?
First test for the Fontainebleau “OAK” forest
Data at the Fontainebleau forest site Data at the Fontainebleau forest site
gap-filled half-hourly measurements (LE, H, FCO2)
year 2006
Flux tower measurementsFlux tower measurements
Neural Network estimation algorithm SPOT- 40m: temporal interpolation with a 2-
sigmoid model
MERIS - 1km:
Remotely sensed fAPAR Remotely sensed fAPAR
Deciduous Broadleaf forest (Oak )
SPOTSPOTMERISMERIS
RMSE = 0.17RMSE = 0.31
RMSE = 0.054
RMSE = 64.96
RMSE = 33.66
ORCHIDEE simulationsORCHIDEE simulations
80% Temperate Broadleaf Summergreen
20% C3G
local meteorological (30’ time step)
previous spinup of the soil carbon pools
SPOTSPOTMERISMERIS
obsprior
Data at the Fontainebleau forest site Data at the Fontainebleau forest site
diurnal cycles (July)diurnal cycles (July)daily datadaily data
improvement of the seasonal fit
obspriorposterior
Assimilation of flux data onlyAssimilation of flux data only
SPOT-fAPARSPOT-fAPAR
Assimilation of fAPAR data onlyAssimilation of fAPAR data only
potential unconsistency of the phasing between NEE flux and
fAPAR observations
obspriorposterior
SPOT-fAPAR onlySPOT-fAPAR only fluxes & SPOT-fAPARfluxes & SPOT-fAPAR
Assimilation of flux data + fAPAR data Assimilation of flux data + fAPAR data
obspriorposterior
Estimated ORCHIDEE parametersEstimated ORCHIDEE parameters
flux onlyflux + SPOTflux + MERIS
Are the differences on the retrieved parameters induced by the use of SPOT or MERIS fAPARs significant?
Still need to quantify the uncertainties on the parameters!
Conclusion
ResultsResults
ORCHIDEE simulates quite well the seasonal, synoptic, and diurnal flux variations at Fontainebleau; this is even better after assimilation!
Lesser agreement with remotely sensed fAPAR
We learned on deficiencies of the model:
spatial heterogeneity leads to smooth increase of observed fAPAR
unconsistency between NEE and fAPAR timing ?
need for high temporal resolution / high resolution fAPAR data to conclude on potential deficiencies of ORCHIDEE
PerspectivesPerspectives Technical improvements:
improve the convergence performances thanks to ORCHIDEE adjoint model
analyze the posterior on the estimated parameters
Application to other sites!
Experimental Validation Kvmax
Leaves Age
Observations (Porté et al., 98)
Vc,jmax optimized
Vc,jmax a priori
Vcm
ax (μ
mol
m-2 s
-1)
Vjm
ax (μ
mol
m-2 s
-1)
Dependency of the carboxylation rates wrt leaves age
Optimized values: variabilitiesK
vmax
βK
HR
KC
sol
AB BX TH WE HE HV VI WB FL HY NB NO
Temperate conifers
Temperate deciduous
Boreal conifers
Parameters optimizedevery year
Optimized Values strongly variable amongst:
1) the different years of a same site.
2) between sites of a same PFT
Constant parameters :
Optimized values follow the same trends amongst the different sites and PFT.
Mea
n u
ncer
tain
ties
a posteriori uncertainties
β
Kvm
ax
KT
opt
KT
min
KT
max
KM
R
QM
R
FR
c
KH
R
Q10
Kra
Kz0
Kal
b
KC
sol
SL
AA
gef
Temperate conifers
Temperate deciduous
Boreal conifers