Global–regional integrated subseasonal forecasts of soil moisture drought

Soil moisture is a core element in shaping land–atmosphere interactions, playing a critical role in ecosystem functioning and sustaining water resources for human use. However, existing approaches, including numerical and AI-based methods, still suffer from notable limitations in soil moisture forecasting. In this study, we develop a novel AI-based soil moisture forecasting model (ASM), which is capable of providing low-resolution global forecasts and high-resolution regional forecasts of soil moisture at the subseasonal timescale. ASM consistently outperforms other representative state-of-the-art AI models across all forecast lead times. Compared with ECMWF, ASM is closer to the ground truth, and better preserve finer-scale spatial details. For regional predictions, ASM produces reliable high-resolution subseasonal soil moisture forecasts for two drought-prone regions selected as case studies: Southern Africa and Henan Province, China.