Altitude driven mechanisms and machine learning prediction of the evaporation paradox reversal in the Nyainqentanglha Mountains

The Tibetan Plateau is a major regulator of Asian hydroclimate, yet rapid changes in atmospheric demand are reshaping its water balance. Here, we examine whether the evaporation-paradox pattern has reversed by analysing meteorological records from 2000 to 2024 across the Nyainqêntanglha Mountains. Potential evapotranspiration (ET 0 ) increased across the region, but the mechanisms were strongly elevation dependent. In low-elevation valleys, ET 0 rebound was driven mainly by recovering wind speed and an expanding vapour pressure deficit, indicating increasing aerodynamic control. By contrast, high-altitude areas showed a cloud-brake effect, in which topographically enhanced cloudiness reduced radiation input and constrained ET 0 growth despite pronounced warming. Explainable machine-learning analyses showed that the dominant predictors identified by XGBoost and Random Forest were broadly consistent with Penman–Monteith sensitivity. Long Short-Term Memory models further reproduced ET 0 dynamics with strong predictive skill, supporting their use as virtual observation tools in data-sparse alpine environments. These findings indicate a transition from an energy-limited regime towards more dynamically controlled evaporative demand, with important implications for plateau water resources.