Crop model data assimilation with the Ensemble Kalman filter for improving regional crop yield forecasts.

摘要

Uncertainty in spatial and temporal distribution of rainfall in regional crop yield simulations comprises a major fraction of the error on crop model simulation results. In this paper we used an Ensemble Kalman filter (EnKF) to assimilate coarse resolution satellite microwave sensor derived soil moisture estimates (SWI) for correcting errors in the water balance of the world food studies (WOFOST) crop model. Crop model simulations with the EnKF for winter wheat and grain maize were carried out for Spain, France, Italy and Germany for the period 1992-2000. The results were evaluated on the basis of regression with known crop yield statistics at national and regional level. Moreover, the EnKF filter innovations were analysed to see if any systematic trends could be found that could indicate deficiencies in the WOFOST water balance. Our results demonstrate that the assimilation of SWI has clearly improved the relationship with crop yield statistics for winter wheat for the majority of regions (66%) where a relationship could be established. For grain maize the improvement is less evident because improved relationships could only be found for 56% of the regions. We suspect that partial crop irrigation could explain the relatively poor results for grain maize, because irrigation is not included in the model. Analyses of the filter innovations revealed spatial and temporal patterns, while the distribution of normalised innovations is not Gaussian and has a non-zero mean indicating that the EnKF performs suboptimal. The non-zero mean is caused by differences in the mean value of the forecasted and observed soil moisture climatology, while the excessive spread in the distribution of normalised innovations indicates that the error covariances of forecasts and observations have been underestimated. These results clearly indicate that additional sources of error need to be included in the simulations and observations.