Crop yield prediction using MODIS LAI, TIGGE weather forecasts and WOFOST model: A case study for winter wheat in Hebei, China during 2009–2013

Abstract

Timely and reliable in-season forecasting of crop yield is crucial for regional and national agricultural management. To improve the winter wheat yield prediction accuracy at the regional scale, we developed a data assimilation scheme that assimilated the MODIS leaf area index (LAI) into the WOrld FOod STudies (WOFOST) model. The meteorological data of the WOFOST model include current weather data, 15-day THORPEX Interactive Grand Global Ensemble (TIGGE) forecast dataset and weather forecast data generated by TIGGE forecast and historical meteorological data (1979–2008) using the vectorial angle method. The WOFOST model was calibrated within 10 subregions of Hebei province based on field measured winter wheat growth data from each corresponding agrometeorological station to account for the spatial variability of crop and soil parameters to some extent, and the Savitzky-Golay (S-G) filtered MODIS LAI was then assimilated into the WOFOST model for regional winter wheat yield forecasting. We constructed a four-dimensional variational data assimilation (4DVar) cost function to account for the observations and model errors from Feb. 10th to Apr. 30th, and the Shuffled Complex Evolution-University of Arizona (SCE-UA) algorithm was used to minimize the cost function by reinitializing three WOFOST parameters. The winter wheat yield forecasting date was started from Apr. 30th, and the results showed that assimilating MODIS LAI into the WOFOST model substantially improved the accuracy of regional wheat yield predictions (R = 0.60, CCC = 0.53, RMSE = 619.73 kg/ha) compared with the unassimilated results (R = 0.35, CCC = 0.24, RMSE = 857.32 kg/ha), and the relative error (RE) between the averaged predicted yield and official statistics for most cities decreased after data assimilation. This demonstrated that assimilating MODIS LAI can optimize the simulation of LAI in the WOFOST model, thereby further reduce the uncertainty of yield forecasting. These promising results highlighted the potential of integrating remotely sensed data, crop model and weather forecasts for in-season prediction of crop yield at the regional scale.