Estimating the crop coefficient of meadow ecosystems and its driving factors using machine learning on the Qinghai–Tibet Plateau

Study region: Qinghai–Tibet Plateau (QTP), China Study focus: This study aims to improve crop coefficient (Kc) estimation for alpine meadows on the Qinghai–Tibet Plateau, where data scarcity and complex climate conditions make evapotranspiration (ET) evaluation difficult. We integrated lysimeter observations (2017–2022), meteorological data, and remote sensing vegetation indices using machine learning (ML) and partial least squares structural equation modeling (PLS-SEM) to develop a reliable Kc estimation framework. The random forest (RF) model achieved the best performance (R² = 0.70, RMSE = 0.17). New hydrological insights for the region: The key environmental drivers identified were photosynthetically active radiation (PAR), vapor pressure deficit (VPD), soil temperature, and the Enhanced Vegetation Index (EVI), each showing pronounced nonlinear and threshold-type responses. The PLS-SEM results revealed that these variables influenced Kc both directly and indirectly through interactions among radiation, temperature, and vegetation growth. This study provides the first machine learning–based Kc estimation model for the QTP, improving the understanding of evapotranspiration processes under climate warming. The model performed well at the Haibei station, but it still requires broader multi-site validation to fully assess its spatial generalizability. Overall, this framework offers a practical and scalable approach for advancing water balance and ecohydrological research in data-scarce, high-altitude regions.