Field-Scale Soil Moisture Assimilation: State, Parameter or Bias Estimation?

Observations can be used to constrain model parameters (calibration), model state variables (state updating, initialization), model error (bias estimation, error characterization) or any combination thereof. It is studied how soil moisture profile observations are best exploited with Community Land Model (CLM) simulations to optimize forecasts of the land surface state and fluxes in a small agricultural field (Production Inputs for Economic and Environmental Enhancement field, OPE³). Observations are assimilated to (i) optimize the model parameters with a variational method, (ii) sequentially update the state, or (iii) sequentially correct for forecast bias. The advantages and disadvantages of each technique are described with respect to their impact on the soil moisture and land surface flux estimation. It is shown that calibration only cannot remove all discrepancy between models and observations and bias estimation in data assimilation is necessary.