Multi-scale Remote Sensing geomorphological applications for updating landslide inventories supported by Artificial Intelligence

Landslides are acknowledged as one of the most critical geo-hazards at the global scale, as they threaten human safety, impact social and economic aspects, and cause instantaneous and drastic changes in the landscape depending on the volume of material involved in the displacement. Considering the landslide typologies and magnitudes variability, along with the evidence that past events may likely represent the preferential path for the activation or reactivation of future instabilities, the need to record the spatial occurrence and geomorphological characteristics of such phenomena is undeniable. Landslide inventories fulfil this purpose by recording the extent and the specific morphographic and morphodynamic peculiarities of slope processes at different spatial scales. As such, they stand as a pivotal tool not only for the morphoevolution analysis of the landscape but, above all, as a fundamental step for an accurate evaluation of hazards and risks associated with these phenomena, thus playing a crucial role for authorities and decision-makers to implement development plans and prioritise mitigation measures implementation. To ensure appropriate land use planning, information on landslide spatial and temporal distribution requires frequent and extensive updates and should be easily exploitable by government authorities. The inventory maps currently suffer from outdated, fragmented and partial records, imputable to the fact that the updating process often depends on catastrophic triggering event occurrences (e.g., earthquakes and prolonged or intense precipitations) or scientific projects, limited in time. The adequate implementation, updating and use of slope processes distribution data for urban planning holds significant contemporary relevance for scientific and practical applications. Furthermore, directing planning and economic efforts in areas where predisposing factors pervasively influence the slope morpho-evolution is still challenging. This work aims to investigate the application of remotely sensed data, with specific regard to advanced satellite interferometry analysis, for updating landslide inventories from a morphographic and morphodynamic point of view on a multi-scale-basis, thus providing a clear direction on identifying high-priority areas warranting in-depth investigations at regional to local scales. Considering the challenge of investigating millions of measuring points (i.e., Persistent Scatterers-PS) derived from the InSAR analysis, the support of artificial intelligence facilitated the effective management of a vast amount of data, which is hampered when applying traditional inventory methods. The implemented methodology, benefitting from the support provided by user-friendly artificial intelligence models, can be incorporated into a multi-scale workflow, offering support to authorities responsible for updating local, regional, or even national-scale landslide inventories. On the regional scale, the landslide prioritisation was based on a clustering analysis that considered the effective interdependence of the single measuring point with its neighbours. The aggregation algorithm, namely Density-Based Spatial Clustering of Applications with Noise (DBSCAN), has allowed a robust, statistically-driven aggregation of millions of measuring points due to their density and the similarity in inherent displacement characteristics. Topological constraints incorporated in the clustering workflow favoured the correct discrimination of moving PS related to a specific process from those representing adjacent ground deformation. Moreover, outliers were effectively filtered and excluded from further analysis. The resulting PS clusters, accurately representing the extension of areas experiencing active and homogeneous deformational behaviour, were then integrated into a ranking process. The integration of this clustering procedure in a semi-automatic selection workflow proved its efficacy in the quick identification and ranking of slope processes that require a thorough evaluation of their morphographic boundaries. At the local scale, along with a thorough examination of the morphographic and morphodynamic characteristics of the single slope process, the analysis of the displacement temporal evolution (related to the interplay of preparatory and triggering factors on the rock mass) was also considered. Specifically, the study was targeted at investigating landslide systems, where differential morphodynamics complicates the assessment of a comprehensive deformational scenario. To better depict the internal displacement heterogeneity, a post-processing analysis was conducted on the PS data to derive the bidimensional displacement vector, resulting from the combination of the measurements obtained from each orbital satellite geometry. The statistical analysis carried out on the resultant decomposed displacement vector provided substantial results in entangling the superimposition of the kinematics associated with the different sectors composing the landslide system. Furthermore, a Bayesian model was applied to the time series of the resultant vector to highlight anomalous trends and acceleration periods for each sector of the system, then compared with site-specific meteo-climatic factors. This approach allowed for a more robust correlation between accelerations in disruption rates and factors destabilising the slope (i.e., preparatory and triggering). Eventually, it was possible to identify specific sectors within the landslide system that required monitoring or prioritised mitigation measures. When nationwide transportation corridors are involved, providing a quick and accurate indication of the slope sectors susceptible to the most severe instabilities enables the subsequent planning of detailed analysis and monitoring surveys tailored to the specific criticality degree. For those cases where the interferometric technique is not feasible enough to investigate slope instability (e.g., fast landslide kinematics and topographical limitations), a multi-scale approach is best suited. The first screening phase of a km-wide highway corridor was conducted through the application of interferometric analysis, which can identify active ground deformations related to slow phenomena or the displacement of debris talus derived from rockfalls and debris detachments. After selecting a few specific areas of investigation, multitemporal UAS-based high-resolution products were acquired to further analyse the slope at the local and microscale. Object-Based Image Analysis and change detection techniques were applied to multispectral and optical products: several early instability precursors and micro-landforms were identified, particularly those linked to rill erosion, that already started impacting and damaging the road infrastructure and the operating mitigation measures. The applied two-phase approach provided effective outcomes in identifying the areas of the slope most susceptible to disruption, supporting the selection of specific critical scenarios that demand immediate attention to safeguard network security and functionality and allowing for the planning of preventive measures instead of mitigation ones alone.