A long-term (1950–2024) machine learning-corrected land surface temperature dataset over the Tibetan Plateau

However, current ERA5-Land reanalysis data exhibit substantial biases across this complex terrain. Here, we fuse high-density meteorological observations (1980–2020) with ERA5-Land using an ensemble machine learning approach, generating a corrected, high-resolution (0.1°) monthly LST dataset spanning 1950–2024.

Validation demonstrates transformative improvements: the coefficient of determination (R²) increases from 0.79 to 0.96 and the root-mean-square error (RMSE) decreases from 9.26°C to 2.33°C. Critically, the corrected dataset uncovers a significantly stronger warming trend (0.32°C per decade, p < 0.05) than indicated by ERA5-Land (0.24°C per decade, p< 0.05). Warming is most pronounced in winter (locally exceeding 0.80°C per decade) and exhibits marked spatial heterogeneity, with maxima in western Ngari and minima in the southeastern TP.