Regional Winter Wheat Maturity Date Prediction Using Remote Sensing-Crop Model Data Assimilation and Numerical Weather Prediction

Optimizing harvesting schedules requires a method for maturity date prediction, to avoid the influence of adverse weather and prevent the decline of crop yield or quality due to inappropriate harvest schedule. However, most prediction models are statistical-based thus are not suitable for regional application, and remote sensing-based models lacked predictability. We presented a framework that assimilated leaf area index (LAI) derived from MODerate Resolution Imaging Spectroradiometer (MODIS) into WOrld FOod Studies (WOFOST) crop growth model, and forecast meteorological data from THORPEX Interactive Grand Global Ensemble (TIGGE) was used as weather data input for the future periods. We selected the winter wheat planting area in Henan Province as study area and recalibrated WOFOST model based on observation data from agrometeorological sites. A cost function based on normalization was constructed to quantify the difference between simulated LAI and MODIS LAI products. First the MODIS LAI profile was smoothed by Savitzky-Golay (S-G) filter, and then these two LAI profiles were normalized to keep their trend information. Then we selected parameters in WOFOST model that are sensitive to maturity date as optimization parameters, such as emergence date (IDEM), effective temperature sum from emergence to anthesis (TSUM1) and effective temperature sum from anthesis to maturity (TSUM2). These parameters have significant differences between years and no obvious spatial and temporal patterns. By means of Shuffled Complex Evolution method developed at the University of Arizona (SCE-UA) algorithm, we simulated in each pixel in the study area and retrieved the optimal parameters set of this pixel. Then we run WOFOST by this optimal parameter set to simulate the growth and development of winter wheat. Moreover, we transformed TIGGE data into the CABO-format weather file to drive WOFOST simulating winter wheat growth in the next 16 d and obtained a spatial distribution of winter wheat maturity date in the study area. Comparing the forecasting date with the observed date from agrometeorological sites, it demonstrated that this method had substantial accuracy in predicting regional maturity date with correlation coefficient (R2) of 0.90 and the root mean square error (RMSE) was 1.93 d. Besides that, the distribution map of maturity prediction showed obvious spatial variability. This method can remedy the shortages of poor predictability and lacking regional differences in most previous methods, and it provides a reference for the future study of crop maturity prediction at a regional scale with longer forecast period.