Integrating AI and Geospatial Data for Vulnerability Analysis
Natural hazards such as earthquakes, floods, and landslides pose a persistent threat to the built environment, causing large economic losses and loss of life worldwide. The vulnerability of buildings to these hazards is strongly influenced by construction materials, structural design, age, maintenance, and their interaction with local environmental conditions. Despite the availability of hazard maps and engineering guidelines, detailed and up-to-date information on building vulnerability is often limited, especially at urban and regional scales. This lack of information hampers effective risk assessment, emergency planning, and the design of targeted mitigation measures.
Recent advances in Artificial Intelligence (AI) and data availability provide new opportunities to assess building vulnerability more efficiently and consistently. AI techniques can integrate diverse data sources, such as remote sensing imagery, street-level images, LiDAR data, building footprints, and historical damage records, to infer vulnerability characteristics of buildings exposed to earthquakes, floods, and landslides. Machine learning models can identify patterns linking building attributes and environmental factors to observed damage, enabling large-scale vulnerability mapping that would be impractical using traditional field-based surveys alone.
The objective of this MSc research is to develop and evaluate AI-based methods for detecting and mapping building vulnerability to natural hazards. The research will focus on one or more hazard types (earthquake, flood, or landslide) and will involve the collection and preprocessing of geospatial and damage data, feature extraction from imagery and ancillary datasets, and the training of machine learning or deep learning models to classify or score building vulnerability. Model outputs will be validated against existing vulnerability assessments or post-disaster damage data. Depending on the student’s interests, the research may emphasize multi-hazard vulnerability assessment, transferability of models across regions, or the interpretability of AI results to support risk reduction and urban planning.