The increasing frequency and severity of natural hazards, such as floods, wildfires, land degradation, and ground displacement, pose significant challenges to the protection of urban areas worldwide. While traditional monitoring approaches based on a single-source satellite sensor have proved to be reliable, they often fail to provide a holistic representation of the complexity, scale, and rapid evolution of these phenomena. The recent advancement of artificial intelligence (AI), coupled with the unprecedented availability of multi-source satellite imagery, offers new perspectives for enhancing natural hazard monitoring and susceptibility mapping. In this study, we present a novel approach that leverages state-of-the-art Explainable AI (XAI) techniques, particularly SHAP (SHapley Additive exPlanations), to analyze multi-source satellite imagery for natural hazard monitoring and assessment in urban areas. The framework utilizes globally available, open-source satellite data (Sentinel-1/2, COSMO-SkyMed, SAOCOM) to ensure inherent scalability and transferability. XAI is chosen to move beyond black-box prediction, providing transparent attribution of susceptibility to underlying environmental and infrastructural parameters, which is essential for informed intervention. This interpretability is critical for building stakeholder trust and ensuring that automated predictions align with domain knowledge before deployment. Our approach was developed, applied, and validated in two distinct sites located in the Puglia region, southern Italy: the densely populated Bari Urban Region (BUR) and the diverse settlements and land uses within the Gargano Urban Region (GUR). We combined XAI-based models with optical imagery from Sentinel-2, SAR data from Sentinel-1, COSMO-SkyMed, and SAOCOM to extract the key features explaining the occurrence and magnitude of the following hazards: (1) sediment connectivity; (2) land displacement; (3) urban floods; and (4) urban wildfires. Our results demonstrate that the integration of multi-source satellite imagery through AI not only significantly enhances the accuracy and reliability of hazard detection (e.g., F1 scores consistently above 67.5% for three of the four hazards, and high Recall across all modules) but also enables the identification of subtle spatial patterns and crucial interrelationships.
Lafortezza et al. (Sat,) studied this question.