Future projections of China runoff changes based on CMIP6 and deep learning

Study region: Mainland China, encompassing 185 hydrological stations across climatic zones. Study focus: This study focuses on projecting future runoff changes across China under multiple Shared Socioeconomic Pathway (SSP) scenarios. We applied advanced deep learning (DL) models, including LSTM-SA and GRU-SA, combined with high-resolution climate inputs produced by a DL-based downscaling of CMIP6 data. The models achieved strong performance, with mean Nash–Sutcliffe Efficiency (NSE) values of 0.65 and 0.66, and corresponding medians of 0.72 and 0.73, respectively. These results confirm the robustness of the DL-based rainfall–runoff simulations across diverse climatic and hydrological regimes. Our result further indicated that reliable monthly rainfall–runoff models can be constructed with as few as 500 training samples, highlighting the efficiency of DL approaches in data-limited settings. New hydrological insights for the region: Future runoff projections reveal overall increases at most stations, with particularly strong signals in the central transitional and southern humid regions. While the spatial distribution of changes remains consistent across scenarios, the magnitude intensifies under higher emissions. Seasonal contrasts are pronounced: summer runoff is projected to increase markedly, while winter runoff tends to decline in humid regions and shows spatial heterogeneity in transitional zones. These patterns suggest heightened risks of both floods and droughts, with more frequent extreme events likely in humid areas. The findings provide new insights into regional hydrological responses to climate change and deliver valuable scientific evidence to support adaptive water resource allocation, flood control, and drought mitigation strategies across China.