Data-driven geological multi-hazard risk analysis at urban scale

This research work explores the intricate domain of geological multi-hazard risk assessment in urban environments, with a particular focus on the Municipality of Rome, Italy. By leveraging data-driven methodologies, advanced Machine Learning techniques, and open-access data, this study addresses the challenges posed by ground instability hazards, including landslides, subsidence, and sinkholes. The research offers valuable insights, innovative models, and practical applications, contributing to urban resilience enhancement and risk mitigation. As the global trend of urbanization continues to transform landscapes and increase population concentration in cities, the understanding and management of geological hazards in urban areas have never been more critical. This research encompasses three key articles published within the thesis, to delve into geological multi-hazard risk analysis in the urban context. In Chapter 2, we tackle the challenge of integrating fragmented, incomplete, and often unreliable data sources for landslide hazard assessment. Recognizing the crucial role of hazard evaluation in risk mitigation policies, we focused on making the most of available data. The methods employed include landslide inventory review, boosting the training dataset, and optimizing Machine Learning-based susceptibility analysis. The research explores temporal recurrence by analysing multi-temporal landslide inventories and historical rainfall databases, offering valuable insights into hazard probability. The integration of spatial and temporal attributes led to the creation of Rome's first large-scale landslide hazard product. This product, designed for statutory purposes, represents a significant milestone in improving the resilience of urban areas. Chapter 3 delves into the world of Machine Learning techniques for landslide susceptibility mapping, emphasizing the significance of data quality. The study compared various models, with the ExtraTreesClassifier emerging as the most reliable choice. However, recognizing the rarity of high-quality, site-specific landslide inventories, the research examines the reliability of open-source landslide susceptibility maps at different spatial scales. By conducting a thorough statistical and spatial analysis, the research highlights the impact of mapping unit and analysis scale on prediction accuracy. Furthermore, the fusion of multiple low-resolution susceptibility maps is introduced, offering a promising approach to overcome data limitations. This milestone not only improves our understanding of landslide susceptibility mapping but also provides practical tools for large-scale assessments in urban environments. The development of a novel model for multi-risk assessment takes centre stage in Chapter 4. The model's goal is to support decision-makers and risk managers in prioritizing buildings exposed to ground instability hazards, including landslides, subsidence, and sinkholes. By integrating spatial hazard assessments, satellite interferometric data, building characteristics, and asset market values, the model calculates a multi-risk score, enabling the ranking of urban buildings. The research offers a detailed explanation of how each dataset contributes to the overall risk assessment, from spatial hazard to potential economic damage and activity rates. The data harmonization process leads to a practical, semi-quantitative approach for multi-risk assessment, fostering a more proactive stance in mitigating ground instability hazards. This Ph.D. work could bring relevant insights for geological multi-hazard risk assessment and urban resilience. Its findings lay the foundation for further research, fostering safer, more resilient cities worldwide.