Improving regional wheat yields estimations by multi-step-assimilating of a crop model with multi-source data

摘要

Assimilating multi-source data into crop models is a promising way to improve crop growth simulations and yield estimations over a large area. Most of previous studies have mainly assimilated one of the observed/retrieved variables such as leaf area index (LAI) or soil moisture. However, assimilating multi-source data into a model and evaluating their respective contributions to improvements in model simulations have been rare. In this study, we proposed a novel Multi-Step-Assimilating of a crop model with Multi-source Data (MSAcmMD) and further demonstrated it with the MCWLA-Wheat model in improving the simulations of crop development, soil moisture dynamics, and grain yield for winter wheat in the North China Plain. The MSAcmMD, based on the calibrating assimilation strategy, followed the logical links among sub-modules of the crop model. It includes four assimilation steps: (i) calibrating crop model parameters; (ii) assimilating crop phenology; (iii) assimilating soil moisture; and (iv) assimilating crop LAI. The results showed that MSAcmMD can improve substantially the simulations of crop development, soil moisture dynamics, grain yields, and their spatiotemporal patterns over a large area and during a relative long-term period. During 2001-2008, across the study areas, the coefficient of determination (R-2) of the simulated yields was increased from 0.39 to 0.75, and root-mean-square-error (RMSE) was reduced from 1096 to 467 kg/ha, relative to the initial model estimates. An additional validation for the year of 2009 further substantiated the robustness of MSAcmMD, with average R-2 of 0.65 and RMSE of 500 kg/ ha. Further analyses showed that assimilation of soil moisture contributed most to the improvement of yield estimations, with R-2 increasing by 43% and RMSE reducing by 47%. Our findings demonstrated a reliable and promising assimilation system in improving crop growth simulations and yield predictions over a large area. MSAcmMD and the related methods provided a large potential in applying to other crops and regions.