Unveiling river thermal regimes in the Yangtze river basin, China, with a hybrid deep learning model

River water temperature (RWT) is a crucial indicator of aquatic ecosystems, influencing physical and biogeochemical processes in river systems. Due to limited early monitoring in many regions, addressing the scarcity of RWT data and analyzing spatiotemporal characteristics are essential for effective river management and ecological conservation. In this study, a hybrid deep learning model CNN-LSTM-AT was developed, using historical air temperature (AT), streamflow, and day of year (DOY) as input variables. The CNN-LSTM-AT model outperformed baseline models in terms of predictive accuracy, stability, and computational efficiency, confirming its robustness and reliability for RWT prediction. The RWT of the Yangtze River was reconstructed from 1960 to 2009 to determine the thermal regime. Trend and mutation analyses demonstrated a general warming trend in RWT through the Yangtze River Basin, with an average increase of 0.09 °C/decade. An initial cooling phase followed by a warming trend was observed, with an abrupt shift around 2000. Periodicity analysis indicated a consistent 20-year period in the RWT time series. Additionally, river heatwaves intensified across the basin, with most events having moderate intensity, whereas the frequency of severe and extreme events increased in recent years. The middle reach of the Yangtze River Basin experienced more intense river heatwaves than the upper reach. The river heatwave regime transitioned from a summer-dominated toward a multi-season pattern. This study provides new insights into the historical thermal regimes of the Yangtze River and proposes a practical solution for addressing RWT data scarcity in other regions.