Zhu et al. (2025) Multi-source remote sensing retrieval and spatiotemporal distribution characteristics of soil moisture content in typical karst farmlands of southwestern China

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2025
• Date: 2025-12-17
• Authors: Zan Zhu, Dongdong Liu, Xuyang Guo, Ya Yang, Shimei Yang, Lianrui Wang
• DOI: 10.1016/j.ejrh.2025.103052

Research Groups

• College of Resource and Environmental Engineering, Key Laboratory of Karst Geological Resources and Environment, Guizhou University, Guiyang, China.
• School of Aeronautics and Astronautics, Guilin University of Aerospace Technology, Guilin, China.

Short Summary

This study establishes an optimized machine learning framework using the XGBoost algorithm and eight key environmental variables to accurately retrieve soil moisture in complex karst farmlands. The resulting model significantly outperforms standard ERA5-Land reanalysis data and reveals distinct spatiotemporal moisture patterns influenced by monsoon cycles and proximity to water systems.

Objective

• To develop an optimal selection mechanism for input variables and machine learning algorithms to improve the accuracy of surface soil moisture content (SMC) prediction in karst regions characterized by high vegetation coverage and complex topography.

Study Configuration

• Spatial Scale: Regional scale focusing on Qiannan Prefecture, Guizhou Province, China (~26,200 km²), with transferability tests conducted in Guangxi, China, and Baena, Spain.
• Temporal Scale: Monthly resolution spanning April 2022 to February 2024 for model training/validation, and a three-year analysis period (2022–2024) for spatiotemporal characterization.

Methodology and Data

• Models used: XGBoost (identified as optimal), Random Forest (RF), Convolutional Neural Network (CNN), and Backpropagation Neural Network (BPNN). Feature selection was performed using the regression ReliefF algorithm.
• Data sources:
  • Remote Sensing: Sentinel-1 (SAR backscattering coefficients), Sentinel-2A (NDVI, EVI, SAVI, FVC, VWC, NDWI).
  • Topography: SRTM Digital Elevation Model (Altitude, Slope, Aspect, Water Convergence Line Distance).
  • Meteorology: ERA5-Land monthly averaged data (Temperature, LST, Runoff, Evapotranspiration, Rainfall).
  • Observations: Field-measured SMC at 0.2 m depth (SMC20) from 62 monitoring stations using the gravimetric method.

Main Results

• Algorithm Performance: XGBoost was the most effective algorithm, maintaining superior $R^2$ values in both training (>0.9) and testing sets across all seasons compared to CNN, RF, and BPNN.
• Optimal Predictors: The most effective model utilized 8 variables: Rainfall, Aspect, Altitude, Slope, Runoff, Water Convergence Line Distance (WLD), Air Temperature, and Land Surface Temperature.
• Model Accuracy: The optimized model achieved high validation accuracy ($R^2 = 0.85$, $RMSE = 0.057$ m³/m³, $Bias = -0.013$, $KGE = 0.887$), significantly outperforming ERA5-Land data ($R^2 = 0.19$).
• Spatiotemporal Patterns: SMC20 peaked in June (monsoon influence) and reached its minimum in January. Spatially, SMC20 increased within the first 150 m of river systems (transition zone) before gradually decreasing with further distance.
• Transferability: Model accuracy declined as the distance from the training region increased, with $R^2$ dropping to 0.51–0.22 in Guangxi and failing ($R^2 = 0.089$) in the Mediterranean climate of Spain.

Contributions

• Systematic evaluation of 18 environmental factors to identify the dominant drivers of soil moisture in karst landscapes, highlighting the primary importance of topographic and meteorological variables.
• Development of a dynamic selection framework for machine learning algorithms and input variables that can be adapted to different karst sub-regions.
• Quantitative characterization of the spatial relationship between water systems and farmland moisture, identifying a 150 m hydrological transition zone specific to karst basins.

Funding

• Guizhou Provincial Science and Technology Foundation (Reference code: ZD [2025]068).

Citation

@article{Zhu2025Multisource,
  author = {Zhu, Zan and Liu, Dongdong and Guo, Xuyang and Yang, Ya and Yang, Shimei and Wang, Lianrui},
  title = {Multi-source remote sensing retrieval and spatiotemporal distribution characteristics of soil moisture content in typical karst farmlands of southwestern China},
  journal = {Journal of Hydrology Regional Studies},
  year = {2025},
  doi = {10.1016/j.ejrh.2025.103052},
  url = {https://doi.org/10.1016/j.ejrh.2025.103052}
}

Generated by BiblioAssistant using gemini-3-flash-preview (Google API)