Evaluation of data assimilation strategies on improving the performance of crop modeling based on a novel evapotranspiration assimilation framework

Recently, data assimilation (DA) has garnered significant attention. Integration of DA approaches and crop models could diminish model uncertainties and improve the precision of model simulations. While previous research extensively focused on assimilating leaf area index (LAI) or soil moisture (SM), the feasibility and effectiveness of assimilating evapotranspiration (ET) have been rarely explored. In this study, we proposed a novel framework of ET assimilation. Then, together with commonly assimilated LAI and SM, we evaluated the performance of this new method in simulating the key indicators (i.e., LAI and ET in daily and interannual scales, and the crop yield) based on the long-term eddy covariance observations and well-calibrated crop model. DA strategies we utilized to evaluate consist of two approaches (i.e., Ensemble Kalman filter (EnKF) and EnKF with simultaneous state-parameter estimation (EnKF-SSPE)) and combinations of three assimilated observations (i.e., LAI, SM, and ET). Our results demonstrate that joint assimilation of LAI and ET with EnKF-SSPE performs best for wheat while joint assimilation of SM and ET with EnKF-SSPE is the best for maize. For a single observation, LAI and ET play a dominant role in DA for wheat and maize, respectively. This is because the interannual variability of wheat growth is primarily influenced by agricultural management (e.g., cultivar change) and can be represented by LAI. For maize which is mostly rainfed, water stress usually occurs. Therefore, ET, with its ability to reflect the water stress status, proves to be effective. EnKF-SSPE outperforms EnKF, exhibiting potential in revealing the parameter evolution during long-term crop modeling, especially when crop cultivars are regularly renewed. This study evaluates different observations and methods through DA based on a newly proposed sequential ET assimilation framework, which might be illuminating for future applications of DA.