Exploiting the potential of multi-frequency satellite synthetic aperture radar data for ground deformation and soil moisture monitoring of unstable ground

Monitoring slow-moving landslides is challenging because the principal precursors—sub-canopy soil-moisture fluctuations and millimeter-scale ground motion—occur at mismatched spatial and temporal scales. Multi-frequency satellite synthetic aperture radar (SAR) datasets were employed to monitor both processes across the Petacciato landslide in the Molise region of Italy, linking soil-moisture dynamics with deformation patterns within a spatially and temporally consistent, physically interpretable framework. This research advances by applying radiative transfer models, which are widely used for soil moisture retrieval; however, their application to L-band SAR data remains comparatively limited. This study presents a comprehensive retrieval framework based on SAOCOM L-band dual-polarization observations (VV–VH) and the bistatic first-order radiative-transfer model (RT1), previously validated with ASCAT scatterometer and Sentinel-1 C-band SAR. The RT1 model was applied to SAOCOM acquisitions over the Petacciato landslide spanning January 2021 to December 2023. Soil-moisture estimates derived from L-band measurements (λ ≈ 23 cm) were statistically evaluated against regional reference products (ASCAT, ERA5-Land, and SMAP) using time-series comparisons and standard performance metrics. The analysis additionally incorporated the Antecedent Precipitation Index (API) to represent soil wetness carry-over from preceding rainfall. The RT1-based retrievals demonstrated strong consistency with reference datasets, achieving correlations of up to r ≥ 0.67 (e.g., relative to ASCAT). To evaluate high-resolution performance, in-situ soil-moisture sensors will be deployed on 23 March 2025, with data acquired between 25 March and 25 September 2025 using multi-frequency SAR observations from SAOCOM L-band, Sentinel-1 C-band, and CosmoSkyMed X-band. RT1 parameterization incorporated leaf-area-index (LAI) inputs at two spatial resolutions: a 50 m product and a 300 m Copernicus product, with parameter settings optimized independently for the L-, C-, and X-bands. Relative to in-situ observations, SAOCOM L-band data exhibited the strongest agreement at both 50 m and 300 m resolutions, yielding a maximum Pearson correlation of r = 0.76 and an RMSE = 0.23m3 m−3. At 300 m spatial resolution, SAOCOM L-band soil moisture estimates exhibited the consistent correlation with in-situ measurements, outperforming both Sentinel-1 C-band and CosmoSkyMed X-band retrievals. The Bayesian dual-frequency fusion (L + C) further enhanced accuracy r = 0.63 and generated uncertainty-aware soil-moisture estimates. The resulting soil-moisture fields provide reliable inputs for shallow-landslide numerical modelling of a representative test slope, enabling enhanced spatial and temporal analysis in landslide-prone terrain, including agricultural and other heterogeneous land covers. Monitoring slow-moving landslides is essential for effective risk prevention and mitigation. Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) provides precise measurements of ground deformation in landslide-prone terrain. In this study, PS-InSAR line-of-sight displacement time series from CosmoSkyMed (ascending and descending orbits) were analysed for the Petacciato slow-moving landslide over the period 2011–2022. API derived from cumulative rainfall, was used as a proxy for antecedent wetness to evaluate its relationship with landslide reactivation. Sequential Turning Point Detection (STPD) was applied to the PS-InSAR time series to identify statistically significant trend reversals, and their co-occurrence with API threshold exceedances was assessed within a two-month window. A correspondence of 38% was observed for the ascending track and 52% for the descending track. Consistency between satellite- and ground-based precipitation estimates was confirmed using Global Precipitation Measurement (GPM) data and local rain-gauge records, yielding strong correlations (r ≥ 0.85). Notably, prominent API peaks preceded major STPD-identified reversals during 2015–2019; in March 2015, a pronounced reversal coincided with the highest API values (≥ 90 mm). Independent soil-moisture retrievals from Sentinel-1 using the RT1 algorithm showed a maximum on 25 March 2015, aligning with the detected turning point. Integrating PS-InSAR deformation metrics with antecedent wetness indicators enhances the identification of time windows conducive to landslide reactivation and supports the development of operational risk-management strategies in unstable terrain. Ground deformation was further examined using Sentinel-1 C-band and SAOCOM L-band data. In the heavily vegetated study area, coherence at X- and C-band was often reduced, limiting the spatial density of persistent scatterers and distributed scatterers outside urban settings or sites with corner reflectors. By contrast, the longer wavelength of SAOCOM L-band provided improved canopy penetration and a higher PS density; ascending and descending acquisitions collected between 2021–2025 were processed accordingly. The L-band phase-to-displacement conversion factor was approximately 0.94 cm per radian, indicating centimetre-scale sensitivity in the line of sight. The LOS velocity observed ranging from -10 mm/year to -40 mm/year in a active landslide zones. Overall, SAOCOM L-band demonstrated superior sensitivity to ground deformation beneath vegetation canopies, capturing centimetre-scale displacements, whereas CosmoSkyMed X-band performed best in built-up areas, resolving millimetre-scale motion. The research further advances through the development of the state-of-the-art PS–SMaRT (Persistent Scatterer–Soil Moisture Analysis for Risk and Triggering), an automated processing pipeline that integrates PS–InSAR deformation data with hydro-geomorphic indicators to detect unstable slopes and derive corresponding hazard indices. Line-of-sight (LOS) velocities and displacement time series are projected onto the local downslope direction using slope, aspect, and sensor geometry, and subsequently filtered by slope and displacement-magnitude thresholds. Spatially coherent instabilities are delineated using the DBSCAN density-based clustering algorithm, vectorized into polygons, and characterized through descriptive statistics. Optional analytical modules quantify correspondence with wet-anomaly rasters using Pearson’s χ2 and the Matthews correlation coefficient, and compare topographic wetness index (TWI) values inside versus outside unstable polygons using Welch’s t-test. A normalized composite of available layers (e.g., slope, wet anomaly, TWI) is used to derive a hazard index and generate a categorical hazard map, with polygon-level zonal summaries and tabular outputs. The system ensures full provenance tracking and logging, exports outputs in raster, vector, and spreadsheet formats, and incorporates a Streamlit-based user interface for interactive execution and rapid visualization of products. The methodological advances in this thesis demonstrate how open-source, reproducible workflows can transform multi-frequency satellite data into actionable insights for landslide monitoring, and infrastructure resilience. These approaches provide a scalable foundation for future SAR missions such as NASA–ISRO’s NISAR and ESA’s ROSE-L, enabling long-term, high-resolution assessment of soil–vegetation–slope interactions.