Optimising ORCHIDEE simulations at tropical sites

1
Optimising ORCHIDEE simulations at tropical sites

• Hans Verbeeck

LSCE, Laboratoire des Sciences du Climat et de
l'Environnement - FRANCE
LSM/FLUXNET meeting June 2008, Edinburgh
2
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsOutline

• Introduction
• Model ORCHIDEE model
• Assimilation system ORCHIS
• Temperate sites results from Santaren et al.
• Tropical sites first results
• Conclusions

3
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsPOLICE
Marie Curie project Parameter Optimisation of a
terrestrial biosphere model to Link processes to
Inter annual variability of Carbon fluxes in
European forest Ecosystems
4
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsPOLICE goals

• Increase knowledge about parameters
• Variation between and within species (PFTs)
• Spatio-temporal variability of parameters
• Validation of the model, model deficiencies
• Improve the models performance
• ...

5
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsORCHIDEE

• ORganizing Carbon and Hydrology In Dynamic
  EcosystEms
• Process-driven global ecosystem model
• Spatial Developed for global applications ?
  grid point mode
• Time scales 30 min 1000s years

6
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsORCHIDEE
Model Parameters
Output variables
Meteorological forcing
7
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsORCHIDEE

• 13 Plant Functional Types (PFTs)
• Standard parameterisation
• Specific phenology
• Initial carbon pools
• Spinup runs (e.g. 500 years), until pools and
  fluxes are at equilibrium

How to deal with spinup runs when optimising a
model? New spinup run for each new parameter
combinantion? Using forest inventory data to
optimise spinup runs?
8
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsOrchidee Inversion
System
Forward approach
Obs.Errors Y, R
Modeled flux M(X)
Meteorological drivers Initial conditions
FCO2 (µmol/m2/s)
Model ORCHIDEE M
Parameters and uncertainties X, P
1 DAY
1 DAY
9
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsOrchidee Inversion
System

• Bayesian optimisation approach
• Prior info on parameters (standard values
  uncertainties PDF)
• Data uncertainties
• Cost function
• BFGS algorithm

10
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsData

• Fluxes
• Carbon
• Latent Heat
• Sensible Heat
• Net Radation
• Only real data
• Errors on the data (PDF)
• Gaussian
• s15 (day),
• 30 (night)

11
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsCost Function

• Mismatch between model and observed fluxes
• Mismatch between a priori and optimised
  parameters
• Covariance matrices containing a priori
  uncertainties on parameters and fluxes and error
  correlations

12
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsBFGS algorithm

• Gradient based calculates gradient at each time
  step (method of finite differences)
• Takes into account lower and upper bound of each
  parameter
• Minimum reached curvature, sensitivity,
  uncertainties and correlations between parameters
  are calculated

13
Introduction ORCHIDEE ORCHIS Temperate
sites Tropical sites ConclusionsSantaren et
al. GBC 2007
FCO2 (gC/m2/Day)
FH2O (W/m2)
AB (97-98)
A priori Model
Optimised Model
BX (97-98)
Observations
TH (98-99)
WE (98-99)
1 year
1 year
1 year
1 year
14
Introduction ORCHIDEE ORCHIS Temperate
sites Tropical sites ConclusionsResults
problems

• Preliminary results show that this is a promising
  aproach
• Assimilating 3 weeks of summer data
• Improves diurnal fit
• Diurnal fit for rest of growing season is not so
  good ? seasonality

Should we vary parameters with time? Yearly,
monthly, ...
15
Introduction ORCHIDEE ORCHIS Temperate
sites Tropical sites ConclusionsResults
problems

• Same results could be obtained when only NEE and
  ?E observations were included
• Photosynthesis parameters are well constrained
• Respiration parameters can not be robustly
  determined. High dependence on initial carbon
  pools.

Assimilate NEE, ?E, GPP, Reco, ...? How to
constrain the pools?
16
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsGuyana
17
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsSantarem km 67
Parameter optimisation vs. Model structure
improvement?
18
Saleska et al. Science, 2003
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsSantarem km 67
Unexpected seasonality dominated by moisture
effects on respiration
Drought response GPP weak R strong
Wet Dry
19
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsSantarem km 67 GPP
and Reco
Should we only use real measured fluxes or also
GPP and Reco? Equifinality?
20
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsSantarem km 67 soil
depth
21
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsSantarem km 67 soil
water stress
22
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsConclusions

• Possibilities to include forest inventory data
  multiple constraint approach? (C pools, spinup
  runs,...)
• How to modify the cost function to assimilate
  data on different time scales?
• How much data are needed?

23
Introduction ORCHIDEE ORCHIS Temperate sites
Tropical sites ConclusionsConclusions

• Temporal variation of parameters?
• Optimal parameter value vs. biological
  significance? Model structure?
• How to deal with uncertainty on the measured
  fluxes? Should we take correlation between
  uncertainties into account?
• Use of GPP and Reco?

24
Thank you!

• Thanks to
• Philippe Peylin, Diego Santaren, Cédric Bacour,
  Philippe Ciais
• Data at tropical sites PIs from Guyana and
  Brazilian sites