Consistency and divergence in drivers of soil erosion risk: Insights from heterogeneous watersheds

Abstract

Watersheds are shaped by a complex interplay of natural geographic conditions and socioeconomic factors, resulting in marked heterogeneity. However, the drivers and mechanisms underlying specific ecological risks, such as soil erosion (SE), may exhibit greater consistency than expected. Most existing studies focus on isolated risks within individual watersheds, often overlooking comparative insights across heterogeneous regions. To address this gap, we propose a multi-model watershed ecological risk analysis framework that integrates clustering algorithms, geostatistical models, and multivariate statistical methods. The framework is applied to three typical heterogeneous sub-watersheds in the Pearl River Basin, namely the Bei River Basin (BRB), the North and South Pan Basin (NSPB), and the Pearl River Delta (PRD), to analyze the spatiotemporal dynamics, key multidimensional driving factors, and the mechanisms driving SE risk from 2000 to 2020. The findings reveal that high-risk SE hotspots are primarily located in cropland, karst regions, and urbanized areas. Land use change is identified as the dominant driver of SE risk, exerting a stronger influence than climatic factors. Among the key indicators, fractional vegetation cover (FVC), cropland area coverage degree (CACD), and nighttime lights (NLT) consistently show the greatest explanatory power. Dominant pathways include soil–vegetation coupling in karst areas, climate–vegetation regulation in agricultural zones, and human activity–vegetation disturbance in urbanized landscapes. Structural equation modeling (PLS-SEM) reveals consistent latent variable structures across the three sub-watersheds, while highlighting marked differences in the strength and direction of effects. Vegetation emerged as the strongest determinant of SE risk, with total effects of FVC being highest in the BRB (β = −0.685, P < 0.01), followed by the PRD (β = −0.594, P < 0.01), and lowest in the NSPB (β = −0.494, P < 0.01). The relatively weaker vegetation effect in the NSPB may be attributed to its high agricultural intensity and extensive karst coverage, which jointly reduce the stabilizing capacity of vegetation. This heterogeneous watershed analysis framework is transferable and scalable, offering a robust foundation for advancing ecological risk assessments and informing precision watershed management under changing environmental conditions.