Winter Wheat-Yield Estimation in the Huang-Huai-Hai Region Based on KNN-Ward Phenological Zoning and Multi-Source Data

Phenology is a key factor influencing the accuracy of regional-scale winter wheat-yield estimation. This study proposes a yield-estimation modeling framework centered on phenological zoning. Based on the remote sensing monitoring results of the heading stage of winter wheat in the Huang-Huai-Hai region from 2016 to 2021, the KNN-Ward spatial constraint clustering method was adopted to divide the Huang-Huai-Hai region into four consecutive wheat phenological zones. The results indicate a consistent spatio-temporal gradient in the phenology of winter wheat across the Huang-Huai-Hai region, characterized by later development in the northern areas and earlier development in the southern areas. The median day of year (DOY) for the heading stage in each zone varies by approximately 4 to 5 days, demonstrating a high degree of interannual stability. Building upon the phenological zoning outcomes, a multi-source data-driven random forest model was developed for wheat-yield estimation by integrating remote sensing data and meteorological variables during the wheat grain filling stage. This model incorporates remote sensing vegetation indices, crop growth parameters, and climatic factors as key input variables. Results show that the phenological zoning strategy significantly improves model prediction performance. Compared with the non-zoning model (R2 = 0.46, RRMSE = 13.02%), the phenological zone model shows strong performance under leave-one-year-out cross-validation, with R2 ranging from 0.54 to 0.68 and RRMSE below 12.50%. The phenological zoning model also exhibits more uniform residuals and higher prediction stability than models based on non-zoning, traditional agricultural zoning, and provincial administrative zoning. These results confirm the effectiveness of phenology-based zoning for regional yield estimation and provide a reliable framework for fine-scale crop yield monitoring. The phenological zoning model also demonstrates superior residual uniformity and prediction stability compared with models based on non-zoning, traditional agricultural zoning, and provincial administrative zoning. These results confirm the effectiveness of the multi-factor-driven modeling framework based on crop phenological zoning for regional yield estimation, providing a robust methodological foundation for fine-scale yield monitoring at the regional level.