Various ORCHIDEE Data Assimilation Systems have been developed to optimize the ORCHIDEE model parameters within prescribed physical range (and given some uncertainties), using various data streams, in order to improve the simulation of the land surface fluxes of energy, water and carbon, as well as the carbon and water stocks, from regional to global scales.

The different data assimilation systems rely on:

  • the process-based global vegetation model, ORCHIDEE, that can be applied at various spatial (local, regional, or global) and temporal (going from hours to centuries) scales, and also possibly on
  • an atmospheric tracer transport model, LMDz, that allows to assimilate atmospheric concentration data (it relates the surface fluxes to measured atmospheric concentrations).

The main data streams that have been assimilated so far are:

  • satellite measurements of vegetation activity (for instance MODIS NDVI),
  • in-situ eddy-covariance flux measurements (FLUXNET database),
  • atmospheric CO2 concentration measurements,
  • in-situ carbon stock measurements (above ground biomass, soil carbon stocks, etc.),
  • surface temperature measurements (from space and in situ).

Different numerical implementations of the model-data fusion have been implemented. They combine possibly several observational data streams to optimize the ORCHIDEE model parameters, using either:

  • a variational approach, that follows the gradient of the misfit function (J) to be minimzed,
  • a Monte Carlo approach (Genetic Algorithm or Particle Filter method), that performs a random search in the parameter space.

Note that for the variational approach the gradient of J is estimated with the tangent linear model of ORCHIDEE, derived with TAF software (automatic differentiation tool), or for one application with YAO software (tool to facilitates the calculation of the adjoint model).