Phd in EARTH SCIENCES

Thesis title: Ground Deformation Characterization using Machine Learning and Multi-Sensor PS-InSAR

Satellite-based interferometric monitoring through Persistent Scatterer (PS) techniques has become a fundamental tool for ground deformation assessment. The growing availability of PS data from different satellites, while offering unprecedented opportunities for comprehensive monitoring, raises fundamental questions about efficient data integration and interpretation strategies. This research develops automated methodologies for enhancing ground deformation characterization through the integration of multi-sensor PS data and Machine Learning techniques, with a particular focus on supporting post-earthquake reconstruction planning in the Central Apennines (Italy). The research progressively advances from spatial to temporal analysis approaches. Initial developments focus on a multi-sensor prioritization framework that combines PS measurements with socio-economic indicators to assess landslide impacts on urban areas systematically. This is followed by the implementation of a data fusion methodology based on a weighted least squares approach with an adaptive weighting scheme that generates synthetic measurement points capturing integrated displacement contributions. The temporal evolution of ground movements is then investigated through the analysis of displacement time series and their relationship with environmental drivers. Finally, a multivariate clustering framework is developed to analyze the spatiotemporal evolution of ground deformation processes through the integrated analysis of two-dimensional displacement patterns and their spatial context. The results demonstrate that multi-sensor integration significantly enhances ground deformation characterization by overcoming single-sensor limitations in terms of spatial coverage and information content. The developed prioritization framework enables systematic identification of high-risk areas, revealing that 30% of analyzed landslides exhibit displacement beyond their mapped perimeters. This significant percentage indicates potential expansion or underestimation of landslide boundaries, providing evidence-based insights for directing reconstruction planning efforts toward areas requiring detailed investigation. Temporal analysis established clear correlations between extreme rainfall events and acceleration periods in displacement time series, particularly evident in cases where intense precipitation followed extended dry periods. In the context of observed climate trends, the understanding of trigger-response relationships becomes crucial for implementing preventive and adaptive measures to reduce potential impacts. Finally, the multivariate time series analysis successfully classified distinct deformation patterns and delineated unstable areas in regions with complex displacement combinations, while revealing the underlying logic of the clustering process. This interpretable spatiotemporal characterization provides new insights into the evolution and extent of ground deformation dynamics, offering quantitative evidence to support the development of site-specific mitigation strategies based on the identified deformation behaviors. These methodological developments bridge the gap between monitoring capabilities and operational needs in post-event reconstruction scenarios, supporting efficient resource allocation and risk mitigation planning. The research demonstrates how automated post-processing of multi-sensor interferometric data advances the technical capabilities and practical applications of ground deformation monitoring, providing decision-makers with quantitative evidence for implementing targeted intervention strategies. However, while the application of current Machine Learning techniques to this context represents a significant step forward, the intrinsic complexity of deformation phenomena and their multi-parametric nature suggest the need to develop more sophisticated and specific approaches, beyond the mere adaptation of existing methodologies, to characterize the dynamics of deformation processes.