Bivariate assimilation of Sentinel‐2 data into the CERES‐Wheat model for winter wheat yield estimation

Abstract Accurate early yield estimates of winter wheat ( Triticum aestivum L.) are essential for effective management and regional food security. To determine the planting area of winter wheat in the Loess Plateau in regions, supervised classification methods were employed. Based on the decision support system for agrotechnology transfer framework, Sentinel‐2 remote sensing data and the crop estimation through resource and environment synthesis (CERES)‐Wheat crop growth model were calibrated and validated at both the point and regional scales. In the analysis, bivariate leaf area index (LAI) and biomass (β) were inverted from Sentinel‐2 derived normalized difference vegetation index (NDVI) using a linear regression model; NDVI was selected after evaluating 14 spectral indices, and the resulting models achieved R 2 values of 0.63 and 0.62, respectively. This was followed by assimilating the Sentinel‐2 data into the CERES‐Wheat model using the ensemble Kalman filter (EnKF) algorithm. The results showed that the accuracy of the winter wheat planting area extracted from remote sensing reached 96.9%. Additionally, compared to remote sensing estimation methods, the EnKF‐LAI and EnKF‐β models, which were assimilated using the EnKF algorithm, showed superior performance, with mean relative error values of 7.24% and 6.91%, respectively. Furthermore, the wheat yield estimation model obtained using bivariate assimilation of LAI and β ( R 2 = 0.80, root mean square error RMSE = 525.84 kg hm −2 ) outperformed models based on univariate assimilation. The findings of this study provide a reliable method for improving winter wheat yield estimation in the Loess Plateau region of China, offering a novel bivariate assimilation framework that integrates Sentinel‐2 observations with the CERES‐Wheat model.