Yu et al. (2025) Cross-scale soil moisture content monitoring of winter wheat by integrating UAV and sentinel-1/2 data

Identification

• Journal: Agricultural Water Management
• Year: 2025
• Date: 2025-09-23
• Authors: Xingjiao Yu, Qi Yin, Long Qian, Chaoyue Zhang, Lin Shao, Danjie Ran, Wenè Wang, Baozhong Zhang, Xiaotao Hu
• DOI: 10.1016/j.agwat.2025.109831

Research Groups

• Key Laboratory of Agricultural Soil and Water Engineering in Arid and Semiarid Areas, Ministry of Education, Northwest A&F University, Yangling, China
• College of Water Resources and Architectural Engineering, Northwest A&F University, Yangling, China
• State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin, China Institute of Water Resources and Hydropower Research, Beijing, China
• National Center for Efficient Irrigation Engineering and Technology Research-Beijing, Beijing, China

Short Summary

This study developed an innovative framework integrating ground, UAV, and satellite data to accurately estimate and map soil moisture content (SMC) in winter wheat fields across scales, demonstrating significantly improved accuracy compared to traditional ground-satellite models.

Objective

• To develop soil moisture content (SMC) estimation models by combining UAV multispectral data (vegetation indices and texture features) with ground-measured data using a Partial Least Squares Regression (PLSR) model.
• To upscale UAV SMC estimation results to a 10 m × 10 m resolution and integrate them with Sentinel-1 SAR and Sentinel-2 multispectral data to develop an XGBoost-based satellite-scale SMC estimation model for county-level mapping.
• To compare the performance of the integrated ground-UAV-satellite framework with traditional ground-satellite modeling approaches, quantitatively evaluating the advantages of the former in SMC monitoring.

Study Configuration

• Spatial Scale:
  • Study Area: Wugong County, Baojixia Irrigation District, Shaanxi Province, China.
  • Ground Sampling: 180 points across 7 representative plots (0.3–0.7 km² each).
  • UAV Data: Subplot scale, centimeter-level resolution (orthomosaic images).
  • Satellite Data: County scale, 10 m resolution (Sentinel-1/2).
  • Soil Depths: 0–20 cm and 20–40 cm.
• Temporal Scale:
  • Ground Sampling: 20 March to 7 May, 2024.
  • UAV Flights: Conducted between 11:00 and 14:00 under clear skies.
  • Satellite Data: Sentinel-1 and Sentinel-2 images with less than 5% cloud cover, ensuring a maximum 24-hour time difference between UAV imaging, ground sampling, and satellite imagery acquisition.

Methodology and Data

• Models used:
  • Partial Least Squares Regression (PLSR) for UAV-scale SMC estimation.
  • XGBoost for satellite-scale SMC estimation.
  • Support Vector Machine (SVM) for UAV image classification (winter wheat planting areas).
  • SHAP algorithm for XGBoost model interpretability.
  • Random Forest (RF) and SVM were also used for comparative ablation tests.
• Data sources:
  • Ground data: Volumetric soil moisture content (SMC, m³/m³) measured at 0–20 cm and 20–40 cm depths using stainless-steel ring samplers and the standard oven-drying method. Sampling locations geotagged with a differential GPS system (Trimble Geo7X, ≤0.3 m planimetric accuracy).
  • UAV data: Multispectral images acquired by a DJI Mavic 3 Multispectral drone (Green, Red, Red-Edge, Near-Infrared bands). Integrated RTK positioning system (≤3 cm horizontal, ≤5 cm vertical accuracy). Orthomosaic images generated using Pix4D Mapper. Features extracted include 16 vegetation indices, mean values of 4 spectral bands, and 8 gray-level co-occurrence matrix (GLCM) texture features.
  • Sentinel-1 data: C-band Synthetic Aperture Radar (SAR) data (5.404 GHz), dual-polarisation (VV/VH), Ground Range Detected (GRD) level, Interferometric Wide Swath (IW) mode. Pre-processed in SNAP (orbital error correction, noise removal, backscatter calibration, speckle reduction, terrain correction) and resampled to 10 m. 5 polarisation indices calculated.
  • Sentinel-2 data: Level 2A products (surface reflectance) from 13 bands (Bands 2–8, 8A, 9, 11, 12) covering visible to shortwave infrared spectrum, with 10–20 m resolution. Pre-processed in Google Earth Engine (GEE) (orthorectification, atmospheric correction) and resampled to 10 m. Several vegetation indices derived.

Main Results

• UAV-scale SMC estimation (PLSR):
  • Optimal performance achieved by combining vegetation indices and texture features:
    • 0–20 cm: R² = 0.775, RMSE = 0.018 m³/m³, MAE = 0.012 m³/m³.
    • 20–40 cm: R² = 0.723, RMSE = 0.021 m³/m³, MAE = 0.015 m³/m³.
  • Vegetation indices alone yielded R² of 0.624 (0–20 cm) and 0.606 (20–40 cm).
  • Texture features alone showed lower accuracy, with R² reduced by 7.69–11.22% compared to vegetation indices.
  • Key features (VIP scores) included Red, Green Index (GI), Green Normalized Difference Vegetation Index (GNDVI) for vegetation indices, and Variance (NIR), Correlation, Homogeneity (Green) for texture features.
• Satellite-scale SMC estimation (XGBoost) using upscaled UAV samples:
  • Achieved high accuracy:
    • 0–20 cm: R² = 0.901, RMSE = 0.0071 m³/m³.
    • 20–40 cm: R² = 0.884, RMSE = 0.011 m³/m³.
  • Average performance over 100 model runs: R² = 0.881, RMSE = 0.0074 m³/m³, MAE = 0.0056 m³/m³ for 0–20 cm; and R² = 0.868, RMSE = 0.013 m³/m³, MAE = 0.009 m³/m³ for 20–40 cm.
  • SMC values below 0.15 m³/m³ at 20–40 cm depth were systematically overestimated.
  • Coefficient of Variation (CV) for predictions was predominantly below 0.2, with greater uncertainty observed at 20–40 cm depth.
• Feature Importance (SHAP for XGBoost):
  • For 0–20 cm SMC, Normalized Difference Water Index (NDWI) was the most influential feature, followed by Shortwave Infrared 1 (SWIRI).
  • For 20–40 cm SMC, SWIRI was predominant, followed by Red Edge 2 (RE2) and NDWI. Sentinel-1 SAR features (MVVVH, SVVVH) showed increased importance for deeper SMC estimation.
• Comparison with traditional ground-satellite models:
  • The integrated ground-UAV-satellite approach significantly improved accuracy:
    • R² increased by 9.53–10.52 %.
    • RMSE decreased by 11.11–31.25 %.
    • MAE decreased by 18.19–25.00 %.
• Spatial Distribution:
  • SMC in Wugong County showed moderate levels (0–20 cm: 0.176–0.197 m³/m³; 20–40 cm: 0.173–0.183 m³/m³).
  • High SMC zones were primarily located in the Qishui River influence zone and along the Wei River main canal irrigation belt, consistent with in situ observations.
• Sensitivity Analysis:
  • PLSR model (UAV scale) demonstrated robustness to feature reduction (R² decline within 3.6% for 20% feature removal).
  • XGBoost model (satellite scale) showed exceptional robustness to input data noise (standard deviation of R² fluctuations within ±0.009 for ±20% feature value perturbations).

Contributions

• Developed an innovative and robust cross-scale framework for soil moisture content (SMC) estimation in winter wheat, integrating ground measurements, high-resolution UAV data, and satellite remote sensing (Sentinel-1/2).
• Demonstrated that utilizing UAV-derived high-precision SMC maps as an intermediate upscaling step significantly enhances the accuracy of satellite-scale SMC retrieval compared to traditional direct ground-to-satellite calibration methods.
• Effectively addressed the scale discrepancy and mixed-pixel problem inherent in satellite remote sensing data for SMC monitoring.
• Provided a practical and efficient methodology for large-scale agricultural drought monitoring and precision irrigation management, offering high-precision input parameters for decision-making systems.

Funding

• National Natural Science Foundation of China (Project codes: 52079113, U2243235, 52409069)
• National Key Research and Development Program of China (Project codes: 2022YFD1900404–01, 2024YFC3211800)

Citation

@article{Yu2025Crossscale,
  author = {Yu, Xingjiao and Yin, Qi and Qian, Long and Zhang, Chaoyue and Shao, Lin and Ran, Danjie and Wang, Wenè and Zhang, Baozhong and Hu, Xiaotao},
  title = {Cross-scale soil moisture content monitoring of winter wheat by integrating UAV and sentinel-1/2 data},
  journal = {Agricultural Water Management},
  year = {2025},
  doi = {10.1016/j.agwat.2025.109831},
  url = {https://doi.org/10.1016/j.agwat.2025.109831}
}