Answers are grounded in this paper and cite their sources.
Accurate soil moisture information in the root zone is critical for precision agriculture and water resource management, yet remote sensing can only observe surface conditions and existing prediction models are largely limited to point scales. To address these gaps, this study proposes an integrated framework that combines satellite retrieval, progressive vertical estimation, and machine learning forecasting to generate spatially continuous, multi-layer soil moisture predictions over a one-week horizon. Surface soil moisture (0–10 cm) was first retrieved using a coupled Water Cloud-Random Forest model (WC-RFM) based on Sentinel-1 and Sentinel-2 imagery (R 2 = 0.908, RMSE = 0.013 m 3 /m 3 ). Then, a progressive estimation strategy incorporating meteorological factors was developed to estimate deep-layer soil moisture at 20 cm, 30 cm, and 40 cm. Among multiple linear regression, random forest, and XGBoost, XGBoost achieved the highest accuracy at all depths, with test R 2 values of 0.952 (20 cm), 0.615 (30 cm), and 0.562 (40 cm) when meteorological factors were included. The contribution of meteorological factors increased with depth (R 2 gain from 0.010 at 20 cm to 0.147 at 40 cm). Third, a recursive multi-step XGBoost forecasting model was constructed to predict surface soil moisture for the next seven days, achieving day 1 R 2 = 0.9088 and maintaining reasonable skill through day 7 (R 2 = 0.5301). The model successfully captured the moisture rise following a 12.8 mm rainfall event, with forecast errors generally within 0.05 m 3 /m 3 at four monitoring stations. Finally, the point-scale forecasting model was extended to the regional scale to generate daily, 10-m resolution forecast maps. The proposed framework bridges the gap between remote sensing observations and operational agricultural water management by providing actionable, high-resolution spatiotemporal soil moisture information.
No takes yet. Share an insight, caveat, or question.
Gao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: