Ecosystem service flows are fundamental to the generation, delivery, transformation, and maintenance of ecosystem services. Spatially quantifying soil conservation service flows (SCSF) is critical not only for elucidating the intrinsic mechanisms governing the complete trajectory of soil and water conservation—from “provision” to “retention”—but also for providing a scientific foundation to identify benefit transfer pathways and optimize watershed management strategies. Focusing on the Northern Shaanxi Loess Plateau, this study integrated the InVEST model with the D8 algorithm to simulate the spatial distribution of SCSF, and applied the XGBoost–SHAP model to analyze its driving factors and interaction effects. The results indicate that: (1) Both potential and actual soil erosion in the study area showed an upward trend from 2000 to 2020, with erosion intensity transitioning from slight to light classifications. (2) The total soil conservation service capacity fluctuated between 523 and 992 million tons (Mt), characterized by a “low-northwest, high-southeast” spatial distribution, whereas the soil conservation service retention rate remained spatially stable with marginal growth. (3) The SCSF volume ranged from 67 to 127 Mt., generally converging from the northwest to the southeast, characterized by an “East-West dominance and small-watershed aggregation” pattern. (4) The impacts of different driving factors on SCSF showed significant differentiation, with natural geographical factors playing an absolutely dominant role. Specifically, the driving effect of slope demonstrates a typical inverted “U”-shaped non-linear threshold characteristic, while precipitation, despite showing a generally positive correlation, exhibits significant interactive differentiation effects in high-rainfall areas. This study clarifies the direction, volume, and drivers of SCSF, providing scientific evidence to support ecosystem protection and comprehensive watershed governance in the Loess Plateau. • Spatially quantified Soil Conservation Service Flows (SCSF) using coupled InVEST-D8 models. • XGBoost–SHAP identified non-linear drivers and thresholds for ecological management. • Slope and precipitation were the most influential drivers of SCSF dynamics in this region. • Flows exhibit a “northwest-to-southeast convergence” pattern.
Sang et al. (Fri,) studied this question.