Hydrothermal dynamics evolution of dry-hot valleys under multiple impacts

The thermal regime of rivers is related to ecological health, making it crucial to elucidate its evolution under multiple pressures. This study investigated the thermal impacts of climate change and reservoir regulation in the dry-hot valley of the Jinsha River using a daily river surface water temperature (RSWT) dataset (2001-2024) derived from Landsat imagery. An integrated framework combining heat flux analysis with an interpretable machine learning model (XGBoost-SHAP) was developed. Validation against in-situ observations yielded R² values of 0.718-0.848 and RMSE values of 1.17-1.86°C. The reconstructed RSWT series indicated a long-term warming trend of 0.07°C yr^-1. Across the progressive cascade-development phases, fitted annual thermal amplitude decreased by 17-23%, while seasonal phase lags increased by 17-47 days. Reservoir impoundment also significantly reduced annual net heat (Q_net) through enhanced latent heat loss (Q_e). Specifically, the annual mean Q_net declined from 79.2 W m^-2 during the Phase I to 52.4 W m^-2 during the four-reservoir phase. In contrast, the annual mean Q_e nearly doubled from -30.8 to -73.1 W m^-2, indicating that evaporation became the dominant pathway of energy dissipation. Phase-specific XGBoost-SHAP analysis indicated that the increased relative importance of wind speed and relative humidity during Phase III was consistent with the enhanced evaporative cooling identified by the heat-flux analysis. Changes in the relative importance of shortwave and longwave radiation were also consistent with altered seasonal thermal storage under later cascade-development conditions. Overall, the proposed framework provides a transferable approach for evaluating river thermal-regime changes in data-scarce, highly regulated basins.