Downscaling multilayer soil moisture using parameterized depth profiles associated with environmental factors

Abstract

High-resolution multilayer soil moisture is valuable for water resources management, drought monitoring, and crop optimization. However, integrating process-based models into statistical downscaling methods to estimate multilayer soil moisture can increase both prediction uncertainty and computational complexity of parameterization. Here, three parameterized equations (Peak, Trough, and Balance) were proposed to quantify soil moisture profiles associated with environmental factors, and downscale multilayer soil moisture. Global (0–40 cm) and local (0–10, 10–20, and 20–40 cm) fitting strategies represented the vertical extent of environmental influences. Three machine learning models were employed to downscale satellite soil moisture and predict depth parameters, which were then used to generate sub-surface soil moisture. The results indicated that the local fitting strategy consistently outperformed the global strategy across all depths, reducing the mean absolute error (MAE) and root mean square error (RMSE) by 81.8 % and 80.4 %, and increasing the coefficient of determination (R²) by 22.5 %. Furthermore, local fitting revealed distinct nonlinear and linear patterns within the 0–20 cm and 20–40 cm intervals, respectively. The prediction accuracy of sub-surface soil moisture was improved compared to surface results, with MAE and RMSE reduced by 30.2 %–60.7 % and 34.9 %–56.7 %, respectively. The modified local fitting strategy further enhanced prediction performance, with the Peak–Trough–Balance combination recommended as the optimal configuration. Future studies should account for variations in soil and vegetation cover, refine interval division by incorporating soil-forming processes, and adopt advanced uncertainty quantification methods to enhance its adaptability and robustness. This study provides a reference for multilayer soil moisture downscaling in large-scale regions with depth variability and environmental heterogeneity.