Regional winter wheat yield prediction by integrating MODIS LAI into the WOFOST model with sequential assimilation technique

In this study, a regional winter wheat yield prediction method was developed by integration of time series of Moderate-Resolution Imaging Spectroradiometer MODIS LAI products (MOD15A2) with WOrld FOod STudies (WOFOST) model through Ensemble Kalman Filter (EnKF) algorithm at the regional scale in the Hengshui District, Hebei province in China. WOFOST model was selected as the crop growth dynamic model, calibrated and validated by the field measured data of Gucheng Ecological-Meteorological Experiment Station in order to accurately simulate the state variables and the growing process of winter wheat. We integrated Landsat TM LAI in three phenological stages with time series of Savitzky–Golay filtered MODIS LAI to generate scale-adjusted MODIS LAI. The theoretically optimal time series of LAI was obtained through the EnKF algorithm to reduce the errors, which existed in both MODIS LAI and WOFOST model. The winter wheat yield was obtained based on WOFOST model with EnKF optimized LAI for each wheat grid of 1 km resolution. The simulated wheat grain yield at each 1 km grid was aggregated to the county-level using the actual wheat fraction for each grid derived from wheat crop map. Comparisons with official statistical yield at county level show that the proposed method with assimilation of scale-adjusted MODIS LAI significantly improves winter wheat yield estimation accuracy. After EnKF data assimilation, in the potential mode, R increases from 0.22 to 0.41 and the root mean square error (RMSE) decreases from 2480 to 880 kg/ha at the county level. Meanwhile, in the water limited mode, high correlation (R increases from 0.20 to 0.50, RMSE decreases from 1320 to 860 kg/ha) with statistical data was achieved. Our results show that assimilating scale-adjusted MODIS LAI into WOFOST model with EnKF algorithm is a reliable and promising method for winter wheat yield estimation at the regional scale.