Abstract:
A fundamental assumption of the Kalman filter is that both the observations and model predictions are unbiased. Bias in observations typically reflects instrumental inaccuracies, representativeness errors, or, in the case of remote sensing observations, errors in the retrieval algorithm. This bias is typically (at best) removed prior to assimilation. Land surface models are usually biased in at least a subset of the simulated variables even after calibration. On-line forecast bias estimation may therefore be needed for data assimilation.

Here, in situ soil moisture observations in a small agricultural field (OPE3) were merged with Community Land Model (CLM2.0) simulations using different algorithms for state and bias estimation with and without bias correction feedback. The different bias correction schemes were tested to study the impact of the state correction on depending model fluxes. The best variant for state and bias estimation depends on the nature of the model bias: an improper bias correction scheme could distort the water balance. The lack of knowledge of the bias `dynamics' in time and space and the approximation of the bias uncertainty structure limit successful bias estimation and correction to directly observed state variables. However, all assimilation schemes including bias correction algorithms yield far improved state analysis results compared to standard state filter analyses.