High resolution sediment-specific surface soil moisture retrieval using Sentinel-1 time series and auxiliary data
In this study, we examine the potential of continuous ground moisture monitoring over a mining site using a combination of in-situ soil moisture sensors and multi-sensor SAR images. We focus on assessing and improving methodologies for retrieval of surface soil moisture, i.e. ground moisture, from SAR measurements focusing on detailed in situ reference observations for several key geomaterials, i.e. sediments, typical in the study site. The mining site represents a limestone quarry located in the southeastern Finland. Our hypothesis is that sediment-specific well-calibrated models can be instrumental in improving soil moisture retrieval under different weather conditions to produce spatially explicit soil moisture estimates at high resolution compared to baseline approaches. Studied SAR data are represented by Copernicus Sentinel-1 C-band images, while auxiliary datasets include optical Sentinel-2 data. Reference data were collected using IoT enabled capacitance sensors. The reference data were calibrated based on laboratory analyses conducted on the different sediment types studied, including organic soil, clay, gravel, and flotation sand. The examined machine learning (ML) methods include XGBoost, LightGBM, random forest, support vector regression, linear regression and k-nearest neighbors. The performance of ML algorithms was evaluated using different cross-validation techniques including a five-fold approach as well as Leave-One-Sensor-Out (LOSO) to assess predictive accuracy and transferability to unseen sensors. The best performance under random five-fold cross-validation was achieved with the most comprehensive feature set which combines Sentinel-1 backscatter, time-series based soil moisture indices (SMI), Sentinel-2 optical, topographic, and temperature predictors. SHAP analysis revealed that SAR based features such as VH backscatter, traditional SMI derived from VV backscatter, and incidence angle were among the most important multi-source predictors. Adding sediment information improved Sentinel-1-only retrievals by more than 2 vol-%, but provided little additional benefit when richer multi-source feature sets were used. In the best sediment-area-level configurations, RMSE decreased to 0.037–0.050 m 3 m − 3 (3.7-5.0 vol. % points ), with R 2 values reaching 0.90. Accuracy varied by sediment texture, with the lowest errors for clay and organic soil and higher errors for flotation sand and gravel. However, LOSO validation demonstrated substantially weaker and site-dependent transferability: R 2 ranged from 0.45 to 0.67 for organic soil at a southern site and from 0.07 to 0.21 for organic soil at a northern site. Thus, high within-dataset predictive accuracy did not translate uniformly to unseen sensor locations, highlighting the need for broader spatially independent validation.
Authors
- Pauliina Liwata-Kenttälä (ORCID: https://orcid.org/0009-0006-2100-4125)
- Alireza Hamedianfar (ORCID: https://orcid.org/0000-0003-0377-6405)
- Matthieu Molinier (ORCID: https://orcid.org/0000-0002-2656-001X)
- Maarit Middleton (ORCID: https://orcid.org/0000-0002-9117-7690)
- Lauri Seitsonen (ORCID: https://orcid.org/0000-0002-7783-456X)
- Ulla Salmela
- Hanna Kukkula
- Oleg Antropov (ORCID: https://orcid.org/0000-0001-8576-404X)
Institutions
- Geological Survey of Finland (FI)
- VTT Technical Research Centre of Finland (FI)
Publication Details
- Journal
- International Journal of Applied Earth Observation and Geoinformation
- Published
- 2026-11-13
- DOI
- https://doi.org/10.1016/j.jag.2026.105665
- Citations
- 1
- Primary Topic
- Soil Moisture and Remote Sensing
- Type
- article
- Field-Weighted Citation Impact
- 1.59
Funders
- European Commission
- HORIZON EUROPE Framework Programme