Assessing the feasibility of using InSAR to improve Landslide Susceptibility Modeling

Abstract

Landslide susceptibility modeling is widely used to support spatial planning and risk mitigation, butits reliability is strongly dependent on the completeness and quality of landslide inventories. In recent years, Interferometric Synthetic Aperture Radar (InSAR) has been proposed as a tool to augment these inventories by detecting surface deformation associated with slope instability. However, the conditions under which InSAR-derived deformation can be effectively integrated into susceptibility modeling remain unclear. This study investigates the feasibility and limitations of incorporating InSAR-derived surface deformation time series into landslide susceptibility modeling for the island of Ischia (Italy), a geomorphologically complex and landslide-prone environment. A traditional susceptibility model was first developed using established static controlling factors, including slope, lithology, soil thickness, terrain wetness index, land use, curvature, aspect, and rainfall. The model achieved satisfactory performance and identified slope as the dominant controlling factor, consistent with the expected behaviour of rainfall-induced landslides. To assess the contribution of InSAR, deformation time series derived from Sentinel-1 data (2019– 2025) were used to generate additional landslide samples based on a statistically defined detectability power threshold. These samples were then incorporated into the landslide inventory to produce an augmented susceptibility model. The analysis focused on comparing three scenarios: a traditional model, an InSAR-augmented model using slope-filtered deformation points, and an augmented model including all InSAR-derived points without additional geomorphological filtering. The results show that the integration of InSAR-derived landslides does not automatically improve model performance. When slope-based filtering is applied, the augmented model preserves the geomorphological consistency of the traditional model, with slope remaining the dominant controlling factor. However, when InSAR-derived points from low-slope areas are included, model performance decreases and land cover becomes the dominant predictor, indicating that the model is capturing non-landsliderelated deformation processes associated with urban areas and observation bias. These findings demonstrate that standard InSAR quality filtering is not sufficient to ensure that detected deformation corresponds to slope instability. The successful integration of InSAR into landslide susceptibility modeling requires the application of explicit geomorphological constraints, particularly slope-based filtering, to exclude deformation signals unrelated to landslide processes. InSAR should therefore be considered a complementary data source whose value depends on careful filtering and interpretation, rather than a direct replacement for traditional landslide inventories.