Interpretable machine learning reveals how climate shapes precipitation isotopes across China

In precipitation, the ratios of stable oxygen and hydrogen isotopes (δ 18 O and δ 2 H) serve as important indicators of hydroclimatic processes. However, the spatial distribution of these isotope ratios across Chinese mainland is not well defined because of insufficient observational data. This study employed an ensemble machine learning model to reconstruct monthly precipitation δ 18 O and δ 2 H across Chinese mainland at 0.125° resolution over 1990–2020, with a coefficient of determination of approximately 0.80 for isotopes and root mean square errors of 1.60 ‰ for δ 18 O and 13.17 ‰ for δ 2 H. SHapley Additive exPlanations (SHAP) was used for model interpretation, which showed that predictor importance considerably differs across the region. Precipitation amount is the main factor influencing isotope depletion in monsoonal regions, whereas temperature and vapour pressure are more influential in arid regions, and elevation effects are the strongest on the Tibetan Plateau. SHAP analysis also identified vertical velocity as an important dynamical predictor of isotopic composition, consistent with its documented role in isotopic convection studies, and quantified its contribution within the national-scale monthly reconstruction framework. This modelling framework improves the accuracy of precipitation isotope mapping and supports an interpretable diagnosis of hydroclimate controls. The findings provide a useful resource for the studies of hydroclimate variability and climate change, paleoclimate reconstruction, and water cycle research in East Asia.