Deep Learning-based Multi-scale Monitoring of Drought in China with High Spatial Resolution

Abstract

Under the background of global warming, the impacts of extreme climate events are becoming increasingly severe, with drought posing particularly significant threats to both human society and the natural environment. Drought is recognized as the second most devastating natural disaster globally. Characterized by its complexity and variability, identifying, assessing, and predicting drought features remains challenging. The Standardized Precipitation Evapotranspiration Index (SPEI), known for its multi-temporal scale characteristics, can represent various drought types and better reflect changes in drought dynamics. It has been increasingly applied in climatological and hydrological studies. However, using SPEI data with a 0.5-degree resolution to assess drought conditions in localized regions of China yields relatively low accuracy, hindering precise evaluation and prediction of drought severity and trends. Therefore, enhancing the spatial resolution of SPEI data is critically important. This study proposes the High Spatial-Resolution SPEI Network (HSR-SPEINet), which integrates environmental factors and remote sensing reflectance data to generate a 1 km resolution SPEI dataset. Experimental results demonstrate its strong accuracy.