Sato et al. (2025) Improving soil moisture estimation in wet soils using L-band Synthetic Aperture Radar (SAR) through polarization and filtering optimization

Identification

• Journal: Science of Remote Sensing
• Year: 2025
• Date: 2025-09-23
• Authors: Naoto Sato, Shinsuke Aoki, Daiki Kobayashi, Yuichi Maruo, Shunsuke Kodaira, Kosuke Noborio
• DOI: 10.1016/j.srs.2025.100290

Research Groups

• School of Agriculture, Meiji University, Japan
• Faculty of Agriculture, Kagawa University, Japan
• NTT Access Network Service Systems Laboratories, Japan
• Innovation Design Institute, Kagawa University, Japan
• Graduate School of Agriculture, Meiji University, Japan

Short Summary

This study optimized L-band Synthetic Aperture Radar (SAR) parameters for soil moisture estimation in wet soils (volumetric water content > 0.3 m³/m³), finding that HH polarization with the Frost filter and VH polarization with Lee Sigma or Refined Lee filters, combined with spatially matched ground-truth data, significantly improved accuracy.

Objective

• To optimize L-band Synthetic Aperture Radar (SAR) parameters (polarization, despeckling filter type, and window size) and ground-truth data determination methods to improve soil moisture estimation accuracy and sensitivity in wet soils (volumetric water content > 0.3 m³/m³).

Study Configuration

• Spatial Scale:
  • Study area: 50 m × 65 m bare ground in Tsukuba, central Japan.
  • SAR image resolution: 6 m × 6 m (original), resampled to 3.88 m × 3.88 m after geometric correction.
  • In-situ sensor spacing: 3 m apart for 64 time-domain reflectometry (TDR) sensors.
  • Averaged ground-truth spatial extents: 6 m, 12 m, 18 m, and 24 m.
  • Despeckling filter window sizes: 3 × 3, 5 × 5, and 7 × 7 pixels (a 3 × 3 window corresponds to approximately 11.6 m × 11.6 m on the ground).
• Temporal Scale:
  • Experiment period: March 25, 2022, to June 30, 2022.
  • SAR data acquisition: 10 full-polarimetric images between April 1 and May 10, 2022.
  • In-situ soil water content measurement interval: 30 minutes.

Methodology and Data

• Models used:
  • Despeckling filters: Boxcar (Mean), Lee, Refined Lee, Lee Sigma, Frost, and Gamma Map.
  • Statistical analysis: Linear regression for correlation coefficient (R), sensitivity (slope of σ-θ relationship), and root mean square error (RMSE).
• Data sources:
  • SAR data: L-band ALOS-2/PALSAR-2 in High Beam Quad polarization (HBQ) mode (HH, HV, VH, VV), Level 1.1 images.
  • Ground-truth soil moisture: 64 time-domain reflectometry (TDR) sensors (true TDR-315H, Acclima, Inc.) inserted diagonally from the soil surface to 5 cm deep.
  • Digital Elevation Model (DEM): 5 m spatial resolution from the Geospatial Information Authority of Japan (GSI).
  • Software: Sentinel Application Platform (SNAP), QGIS.

Main Results

• Volumetric water content in the study area varied between 0.3 m³/m³ and 0.7 m³/m³, consistently exceeding 0.4 m³/m³ for a significant portion of the study.
• HH and VH polarizations demonstrated favorable performance for soil moisture estimation in wet soils.
• The optimal combination was HH polarization with the Frost filter, yielding high correlation coefficients (R > 0.55, p < 0.1), high sensitivity, and low root mean square error (RMSE).
• VH polarization, when filtered with Lee Sigma or Refined Lee, also achieved high correlation (R = 0.57 and R = 0.55, respectively, p < 0.1) and low RMSE (approximately 0.04 m³/m³), despite exhibiting lower sensitivity compared to HH.
• HV polarization consistently showed low sensitivity, low correlation, and high error, indicating its unsuitability for soil moisture estimation in this context.
• The application of despeckling filters significantly improved correlation coefficients across all polarizations, underscoring their necessity for accurate soil moisture estimation.
• An optimal despeckling filter window size of 3 × 3 or 5 × 5 pixels provided the best balance between noise reduction and detail preservation, enhancing estimation accuracy. Larger window sizes (e.g., 7 × 7) degraded accuracy.
• Correlation coefficients improved when ground-truth soil moisture was averaged over spatial extents matching the effective resolution of the despeckling filter window (up to approximately 15 m). However, overly large averaging extents (e.g., 21 m) reduced correlation due to spatial mismatch.
• The low dry bulk density (< 1.0 g/cm³) of the Kanto loam soil at the experimental site likely enhanced microwave penetration, contributing to the observed meaningful backscatter-soil moisture relationships under wet conditions.

Contributions

• Systematically evaluated the combined effects of multiple polarizations, despeckling filters, and varying spatial extents of in-situ ground-truth measurements on L-band SAR soil moisture estimation in high-moisture environments.
• Identified optimal combinations of L-band SAR polarization (HH, VH), despeckling filter type (Frost, Lee Sigma, Refined Lee), and window size (3 × 3 or 5 × 5 pixels) for improved soil moisture estimation in wet soils.
• Demonstrated the critical importance of matching the spatial scale of in-situ ground-truth measurements with the effective resolution of SAR observations (influenced by despeckling filter window size) for maximizing retrieval accuracy.
• Provided insights into SAR sensitivity under high-moisture conditions, suggesting that specific soil properties, such as low dry bulk density, can mitigate conventional limitations of SAR-based retrieval in wet soils.

Funding

• NTT Access Network Service Systems Laboratories

Citation

@article{Sato2025Improving,
  author = {Sato, Naoto and Aoki, Shinsuke and Kobayashi, Daiki and Maruo, Yuichi and Kodaira, Shunsuke and Noborio, Kosuke},
  title = {Improving soil moisture estimation in wet soils using L-band Synthetic Aperture Radar (SAR) through polarization and filtering optimization},
  journal = {Science of Remote Sensing},
  year = {2025},
  doi = {10.1016/j.srs.2025.100290},
  url = {https://doi.org/10.1016/j.srs.2025.100290}
}