Developing a multi-hazard impact and response dataset for the Global South

Multi-hazard global disaster and impact datasets are often biased towards the Global North, resulting in significant data gaps for developing countries. To address this imbalance, we developed a new dataset by automatically analyzing the reports from the International Federation of Red Cross and Red Crescent Societies (IFRC). These reports document immediate aid, recovery, and resilience-building in the aftermath of disasters, targeting mainly countries in the Global South. From the 1,664 reports spanning 1996 and 2024 years, we identified 620 unique disasters affecting 143 different locations (39% in Asia, 16% in Africa, 18% in the Americas, 7% in Europe, 4% in Oceania). Using natural language processing, large language models, and machine learning, we extracted structured information on (i) the direct and indirect societal and environmental impacts and (ii) the response measures taken to address these disasters. Our approach captures a broad range of impacts, from traditional metrics like fatalities and economic losses to displacement, health, and well-being. Using guided topic modelling, we developed a typology of response measures, categorized into ten main classes (e.g., Healthcare and Medical Response, Shelter and Infrastructure Support, and Community Engagement and Communication). Our results show that hazard impacts in the Global South are much more diverse than previously reported in global databases. Moreover, preliminary results on the response measures characterization reveal notable geographical and hazard-specific biases. Our approach bridges critical data gaps, providing a more nuanced understanding of disaster impacts and responses, which is particularly valuable for informing and enhancing disaster risk reduction efforts in the Global South.