Assimilation of remote sensing data and crop model simulations for improved rice yield prediction under controlled irrigation and extreme precipitation in Northeast China

Heilongjiang Province in Northeast China is a major single-cropping rice production area, where controlled irrigation (CI) is widely promoted for water conservation. However, frequent extreme rainfall events threaten rice production, underscoring the need for accurate in-season yield predictions to support water management and ensure food security. This study integrates remote sensing data with crop model simulations through data assimilation (DA) to improve yield prediction based on different extreme weather scenarios under CI. In 2022, a CI experiment was conducted at Youyi farm and a regional investigation was carried out in Heilongjiang Province. An ensemble Kalman filter approach was developed, assimilating leaf area index (LAI) data, combining remote sensing and the WOFOST model. Various DA strategies were tested, adjusting temporal step sizes and background ensemble sizes to identify the optimal approach for rice yield simulation in Heilongjiang Province. Using meteorological data from 1968 to 2022, we created a hindcast dataset for rice yield prediction under normal and extreme conditions. By setting forecast nodes throughout the season, we identified the optimal node with minimal error and applied this method across Heilongjiang. Results show that: (1) State updating via DA effectively integrates remote sensing data and the WOFOST model, enabling accurate simulation of rice yield; (2) The optimal configuration, with a temporal step of 4 days and 75 background ensembles, achieved the highest accuracy; (3) Through the analysis of yield prediction accuracy across different growth stages, September 1 is identified as the optimal hindcast node, which minimizes meteorological uncertainty within the hindcast framework; (4) The application of this integrated approach achieves a high hindcast accuracy of 97% for regional-scale rice yield under the constructed meteorological dataset. This study underscores the critical role of integrating crop growth models with DA techniques, remote sensing, and meteorological forecasts in supporting in-season rice yield estimation in CI systems, thereby providing a scientific basis for adaptive management in rain-prone regions.

Hali tarjima qilinmagan