Precipitation use efficiency (PUE) links ecosystem carbon uptake to precipitation input, but endpoint ratios alone cannot show where carbon–water coupling differs along ecohydrological pathways. This limitation is especially relevant in karst landscapes, where thin soils and heterogeneous hydrological pathways can decouple rainfall, soil moisture, evapotranspiration, and plant carbon gain. Here, we developed a PUE chain decomposition framework based on gross primary productivity (GPP), transpiration (T), evapotranspiration (ET), soil moisture (SM), and precipitation (PRE): PUE = GPP/T × T/ET × ET/SM × SM/PRE. In this framework, GPP/T represents carbon fixation per unit transpiration, T/ET the transpiration fraction of evapotranspiration, ET/SM evapotranspiration output relative to soil moisture, and SM/PRE soil moisture status relative to precipitation input. We used multi-source remote-sensing and reanalysis data from 2003 to 2022 to compare karst and non-karst landscapes in Southwest China, applied variance decomposition to quantify the contributions of chain terms and their interactions, and used Stacking ensemble learning with Shapley additive explanations (SHAP) to interpret model-inferred environmental associations. Mean PUE was 1.16 g C m−2 mm−1 in non-karst areas and 1.08 g C m−2 mm−1 in karst areas, and all four chain components differed significantly between landform types. Variance decomposition identified SM/PRE and its interaction terms as the largest contributors to PUE variability, mainly reflecting a precipitation-sensitive diagnostic signal and soil moisture status relative to precipitation input. Machine learning interpretation showed that solar radiation, leaf area index, aridity, and groundwater storage were associated with different chain components; karst areas showed stronger groundwater-storage signals and lower model-inferred response thresholds. These findings indicate that PUE differences in Southwest China arise from multiple linked diagnostic stages rather than from endpoint carbon uptake or precipitation alone. The framework can help locate water-use constraints and support landform-specific ecological restoration and water management.
He et al. (Fri,) studied this question.