Annual 30 m winter wheat yield mapping in the Huang-Huai-Hai plain using crop growth model and long-term satellite images

Abstract

There are great challenges for the accurate prediction of wheat yield at large-scale region due to the spatial heterogeneity of soil, management and meteorological conditions. The Huang-Huai-Hai plain (HHHP) is the most important wheat producing area in China. Timely and effective prediction of wheat yield in the HHHP is of great significance for ensuring food security. However, there is still a lack of studies on long-term wheat yield prediction with high spatial resolution in the HHHP. In this study, the simulation results (including phenological information, LAI and yield) of WheatGrow model and the meteorological variables were used to train regression model, which was applied in the corresponding satellite images and gridded meteorological variables to develop wheat yield prediction model (SCYMvp). Then time-series yield maps of wheat in the HHHP (2001-2020) were obtained and the digitization footprint was estimated. We found that SCYMvp model developed by combining machine learning algorithm (ML) with two state variables (phenological information and GCVI) had the best performance (r = 0.7, NRMSE = 15 %) for wheat yield prediction in the HHHP. Compared with other yield prediction models, SCYMvp model reduced the error of yield prediction to some extent. Meanwhile, SCYMvp model showed good accuracy in wheat yield prediction at the irrigated fields and rainfed fields. In addition, extreme climate events had a weak influence on the SCYMvp model in wheat yield prediction. The yield prediction with high spatial resolution generated by SCYMvp model in the HHHP during past 20 years had good consistency and low error with the statistical yield in the spatio-temporal scale. Compared with the GlobalWheatYield4km product, the SCYMvp model had a better ability to capture spatial heterogeneity of wheat yield at site scale. In addition, required data size per hectare for predicting wheat yield showed a significant increasing trend. The results indicated that the SCYMvp model developed in this study can effectively map wheat yield with high spatial resolution in the study area, and has good potential in smallholder farming economy with large spatial heterogeneity of management conditions.