青藏高原湖泊表层水温的非线性协同驱动机制:基于深度学习+SHAP融合分析框架OA

The Tibetan Plateau,a region highly sensitive to global climate change,exhibits significant evolution in lake surface wa-ter temperature(LSWT),which has profound implications for regional ecological security.Investigations into the drivers of LSWT changes involve multiple factors,including meteorological conditions and topographic features.However,conventional approaches possess limited capability to quantitatively resolve nonlinear interactions among these drivers.This study examined 106 large and medium-sized lakes across the Tibetan Plateau,employing a deep learning model based on long short-term memory(LSTM)net-works combined with SHapley Additive exPlanation(SHAP)interpretability analysis.This framework quantitatively disentangles the individual and interactive contributions of seven drivers-air temperature,precipitation,downward longwave radiation,down-ward shortwave radiation,air pressure,specific humidity,and wind speed-to LSWT variations at both regional and individual lake scales,thereby systematically elucidating driving mechanisms and synergistic patterns.Key findings include:(1)Longwave and shortwave radiation were identified as the dominant drivers,collectively accounting for over 80.0％of global SHAP values across scales and showing strong positive correlations with LSWT.Air temperature and specific humidity exerted secondary influ-ences,whereas precipitation and wind speed had minimal effects.(2)Widespread interactive effects revealed four primary syner-gistic modes:a linear pattern(e.g.,downward longwave radiation and air temperature,affecting 67.92％of lakes),an inverted U-shape pattern(e.g.,specific humidity and air temperature,51.89％of lakes),an effect cross-driven pattern(e.g.,wind speed and specific humidity,70.75％of lakes),and a threshold-constrained pattern(e.g.,precipitation and air pressure,100％of lakes).(3)The SHAP methodology effectively quantified nonlinear synergistic behaviors,highlighting the heightened sensitivi-ty of plateau lakes to radiative factors due to high solar radiation permeability under thin atmospheric conditions.This study innova-tively integrates deep learning with interpretability analysis to establish a quantitative framework for unraveling complex driving mechanisms behind high-altitude LSWT dynamics.The results offer critical insights for predicting thermal responses under ongoing climate change and for developing differentiated management strategies,thereby holding substantial scientific and practical rele-vance.