Assessing Model Structural Uncertainty in Isotope‐Based Moisture Recycling Ratio Estimation Over the Western Loess Plateau

ABSTRACT

Stable‐isotope mixing models are widely used to estimate the moisture recycling ratio ( f _re ), yet uncertainties associated with model structure remain insufficiently evaluated. Using isotopic observations of local ambient vapour, multi‐source upwind vapour, plant xylem water and soil water collected over the western Loess Plateau from April to October 2024, we compared f _re and its component contributions estimated by IsoSource, IsoError, MixSIAR, MixSIR and Simmr. Model performance was assessed using internal reconstruction, inter‐model differences and sensitivity to isotopic inputs. Mean f _re values from Simmr, MixSIAR, MixSIR and IsoSource were 32.3%, 30.1%, 29.0% and 22.9%, respectively, whereas IsoError produced a physically unrealistic value of −3.7% and was excluded from subsequent analyses. All retained models identified advected vapour as the dominant source of local ambient vapour. MixSIR exhibited the best internal reconstruction performance, followed by MixSIAR. Differences among the three Bayesian models were not statistically significant before or after multiple‐comparison correction; after Bonferroni correction, only Simmr differed significantly from IsoSource. Sensitivity analysis showed broadly consistent responses among the three Bayesian models, with f _re sensitivity decreasing in the order of local ambient vapour, advected vapour, transpired vapour and evaporated vapour. Overall, under conditions of end‐member overlap and observational uncertainty, Bayesian models provide a more systematic framework for quantifying f _re uncertainty and offer useful guidance for model selection and uncertainty assessment in future f _re studies. Improving the representativeness of local ambient vapour and advected vapour isotope observations remains an important pathway for reducing uncertainties in f _re estimates.