关键词: 流域土壤水分, 时间稳定性, 关联路径, 随机森林, 结构方程模型, 下垫面因子

Abstract: Temporal stability of soil moisture is crucial for optimizing hydrological monitoring and understanding eco-hydrological processes. However, the driving forces from underlying surfaces are complex. Large-scale ground-based measurements are costly and difficult to implement, whereas the applicability of remote sensing products in this field remains to be validated. To address those challenges, based on in-situ soil moisture observations from stations in seven typical experimental watersheds in the United States, we utilized high-resolution (1 km) remote sensing soil moisture products and multi-source underlying surface attribute data, and comprehensively integrated the random forest (RF) and structural equation model (SEM) to analyze the spatial pattern of soil moisture temporal stability at the watershed scale, identify the dominant influencing factors, and analyze their linkage pathways. The results showed that: (1) In-situ station measurements verified the pronounced spatial heterogeneity in the temporal stability characteristics of watershed soil moisture, while the spatial wetness pattern from remote sensing products agreed well with ground-based measurements (Spearman’s rank correlation coefficients 0.50-0.90). This means that remote sensing produces can reliably characterize the spatial distribution of the mean relative difference (MRD) of temporal stability index at the watershed scale, though the temporal variation amplitude is relatively dampened. (2) Dominant controls on MRD showed a clear hierarchical structure across the underlying surface. Topography (especially elevation) exerted the most extensive and dominant influence, followed by soil properties (sand content, clay content, and bulk density), while vegetation (leaf area index) played a minor role. The dominant factors and their contributions varied among watersheds. (3) The response of MRD to underlying surface factors generally exhibited nonlinearity and threshold effects, with distinct response patterns across watersheds. MRD prediction models based on dominant influencing factors achieved favorable accuracy (R²=0.53-0.78). (4) Structural equation modeling further quantified the pathways of dominant factors. Elevation exerted a significant and dominant negative direct effect on watershed MRD. The results highlighted regional specificity in linkage pathways, and reflected strong modulation by local topography-soil-vegetation configurations. This study confirms the feasibility of using remote sensing data for watershed soil moisture temporal stability research, clarifies the hierarchical, nonlinear, and regionally differentiated nature of multi-factor synergistic effects. It provides scientific basis for optimizing soil moisture monitoring networks and developing mechanism-oriented hydrological models.

Key words: watershed soil moisture, temporal stability, linkage pathway, random forest, structural equation modeling, underlying surface factor