Early-season estimation of winter wheat sowing date: Integration of dynamic climate windows and phenological indicators into machine learning models

Climate change and the drawbacks of traditional monitoring techniques pose challenges to efficient crop phenology management, making accurate winter wheat sowing date estimation crucial for agricultural optimization. We present a machine learning framework for estimation of winter wheat sowing dates using high-resolution early-season remote sensing. It uses the Normalized Difference Greenness Index (NDGI) from Sentinel-2 data to detect crop emergence. A dynamic climate window extracts pre- and post-emergence environmental variables, and machine learning models estimate sowing dates. Evaluated for Henan province, China, during the 2024 growing season, the framework achieved an R² of 0.82, supporting high-resolution spatial mapping. This approach provides a reliable and scalable tool for large-scale sowing date monitoring, supporting climate-resilient agricultural management and data-driven farming decisions.