Vegetation transpiration contributions to precipitation recycling in a semi-arid grassland: an isotope-constrained source partitioning framework.

INTRODUCTION: Vegetation transpiration provides an important pathway through which semi‑arid grasslands return plant‑accessible water to the lower atmosphere, but its contribution to precipitation recycling remains difficult to quantify under sparse isotope observations and incomplete satellite records. Here, we developed an isotope‑constrained source partitioning framework to estimate the event‑ to annual‑scale contributions of vegetation transpiration, surface evaporation, and advected moisture in Zhenglan Banner, Inner Mongolia. METHODS: The framework couples event‑scale precipitation isotopes with ERA5 and MODIS descriptors, explicitly encodes missing satellite predictors through value-mask pairs, and enforces a closed three‑endmember partition constrained by isotope consistency. Using 140 precipitation‑isotope events from 2018-2021 for model training and validation, the auxiliary isotope prediction achieved RMSE values of 2.62‰ for δ¹⁸O and 19.8‰ for δD, with near‑nominal 80% interval coverage of 0.83 and 0.82, respectively. RESULTS: The trained framework was then used to reconstruct annual source‑fraction patterns during 2015-2024, including the vegetation‑transpiration fraction (ftr), surface‑evaporation fraction (fev), and advected‑moisture fraction (fadv). The ftr component formed coherent belts over vegetated areas and varied from 0.356 to 0.397, while fadv ranged from 0.321 to 0.365 and fev remained lower and more spatially fragmented. Interannual variability was organized mainly along a recycling-advection axis: years with weaker large‑scale inflow showed expanded transpiration contribution, whereas stronger moisture convergence shifted the mixture toward advection. The retrieved ftr was positively associated with independent vegetation and water‑flux indicators, including growing‑season NDVI, EVI, and MODIS ET_daily, providing indirect consistency with a vegetation‑mediated recycling signal rather than direct validation of transpiration‑derived precipitation recycling. DISCUSSION: These results indicate that isotope‑constrained source partitioning can provide a data‑efficient way to quantify plant‑mediated precipitation recycling in semi‑arid grasslands. Future transfer of the framework should be accompanied by local checks of end‑member signatures and vegetation-atmosphere coupling conditions.