Dataset uses AI and disaster news to fill in knowledge gaps and map interconnected risks
Climate-related disasters such as hurricanes, floods, and wildfires, or geological hazards like earthquakes and landslides generate enormous amounts of news coverage. Yet most of this information remains scattered, unstructured, and too fragmented for scientists, policymakers, or emergency responders to act on quickly.
To bridge the gap, a study by the JRC, carried out in cooperation with researchers from the IT company Engineering Ingegneria Informatica and the Institute of Health and Society (IRSS) of the University of Louvain developed an AI-powered pipeline that reads disaster news and turns them into clear, structured knowledge.
The resulting resource advances data-driven approaches to disaster scenario modelling, impact analysis, and decision support in disaster risk management. The study is published in Nature Scientific Data. All compiled data, code, and processing workflows are openly available. An interactive dashboard lets anyone explore directly the disaster storylines and knowledge graphs.
A global picture that goes beyond headlines
The dataset covers over 3,000 disaster events across 175 countries between 2014 and 2024, spanning 26 disaster types and accounting for around 80% of global economic losses recorded by the Emergency Events Database (EM-DAT) over the same period. Compiled from various sources, and managed by the University of Louvain, the EM-DAT contains data on the occurrence and impacts of over 30,000 disasters worldwide from 1900 to the present day.
Most disaster reporting focuses on high-income countries and sudden events, leaving slower crises like droughts in vulnerable regions largely invisible. In this study, the tool drew on a broader, more systematic news feed to correct that imbalance. But the same approach can be applied to any dataset, making it possible to build a more complete picture of risk in contexts that are typically underrepresented.
Seeing how disasters interact
The most powerful feature of the approach applied by the authors is the ability to capture cascading events. Heavy rainfall doesn't just cause flooding — it can also disrupt transport networks, damage crops, and trigger disease outbreaks.
While traditional databases record each impact in isolation, the knowledge graphs generated in this study help to make visible the full chain. Such information can help make more informed decisions in the future, as understanding how one hazard builds on another helps emergency planners anticipate knock-on effects, allocate resources more strategically, and learn from how past crises unfolded.
From news to knowledge graphs
The system works in two steps. First, AI models scan millions of articles from the Europe Media Monitor (EMM) — a JRC research project tracking hundreds of thousands of news sources worldwide — to identify disaster-related coverage. Then, large language models (LLMs) run on GPT@JRC, the Commission’s own AI service, distil each event into a structured “storyline”: a summary of what happened, who was affected, what caused the crisis, and how it was managed.
The system works in two steps. First, using a technique called Retrieval-Augmented Generation (RAG), the most relevant disaster-related news articles for a given event are automatically extracted from the Europe Media Monitor— a European Commission service tracking hundreds of thousands of articles worldwide. Then, large language models (LLMs) from GPT@JRC, the Commission's own AI service, distil each event into a structured "storyline": a summary of what happened, who was affected, what caused the crisis, and what responses were reported.
Those storylines are converted into visual networks of cause-and-effect relationships between hazards, vulnerabilities, and responses. The result is something that EM-DAT, with its authoritative statistics, does not provide alone: context, narrative, and causal detail.
The outputs were validated at a dedicated workshop in Brussels in June 2025, attended by disaster risk professionals from the Commission’s Directorate-General for Civil Protection and Humanitarian Aid Operations and the Union Civil Protection Mechanism , who confirmed their accuracy and practical relevance.