Systematic Evaluation of Interpolation Strategies for Soil Moisture Retrieval Using Sentinel-1/2 Data: A Case Study in Shandian River Basin

Microwave remote sensing is widely used for soil moisture (SM) monitoring due to its sensitivity to dielectric properties. However, microwave backscattering is affected by surface roughness, vegetation structure, and soil texture, which limits high-resolution SM retrieval accuracy and necessitates multi source data integration. Optical remote sensing provides complementary vegetation information but suffers from data gaps caused by cloud cover. Although various gap-filling methods for vegetation indices have been compared, their impacts on multi source SM retrieval remain underexplored. This study integrates Sentinel-1 and Sentinel-2 data through random forest (RF) regression in northern China's Shandian River Basin, comparing univariate and multivariate interpolation methods for SM retrieval. Comprehensive evaluation includes spatial cross validation, leave-one-station-out (LOSO) validation, Station Based Spatial Error Assessment and feature importance assessment. Results demonstrate that interpolation methods significantly improve SM retrieval accuracy. The Savitzky-Golay (SG) filter achieves highest accuracy (R = 0.849), while Multivariate Singular Spectrum Analysis (MSSA) demonstrates superior spatial generalization. This study provides quantitative guidance for interpolation method selection in agro-pastoral transitional zones.