Llm extracts detailed disaster events with location and hazard tags
From UNDRR Reports to Event Records: Schema-Constrained LLM Extraction of Georeferenced Disasters
Computation and LanguageArtificial IntelligenceInformation Retrieval
Summary
Disaster reports are often written like stories and hard for computers to understand directly. The authors created a computer method using a new language model that reads these stories and turns them into organized records describing disaster events, their types, and exact locations. They tested this method on thousands of reports from a global disaster database, finding it much better at identifying disaster details and places than previous tools. This work helps turn messy disaster descriptions into clear facts that people can review and use.
What this means in practice
- •For emergency management teams: Create structured, location-linked disaster event databases from large volumes of unstructured reports to improve situational awareness and decision support.
- •For geospatial data analysts: Automatically convert textual disaster descriptions into georeferenced records compatible with spatial analysis and mapping tools.
Authors
Camilla Andreozzi, Phuong-Anh Nguyen-Le, Zhijing Jin, Revati Mani
Abstract
Disaster-risk-reduction archives describe hazard events in prose that databases such as EM-DAT (Delforge et al., 2025) cannot ingest directly. We present an LLM pipeline that generates candidate georeferenced event records using a controlled hazard vocabulary and fixed schema, retaining evidence for review. Applied to 10,000 documents from PreventionWeb, the knowledge hub managed by UNDRR, it produced 3,572 records from 1,913 documents across 24 hazard types and resolved 81% of location mentions to OpenStreetMap geometries. On 171 human-positive document windows from a stratified 217-document reference set, GPT-5 achieved 86.0% pooled attribute $F_1$, versus 44.2% for the spaCy-gazetteer baseline. Evaluation pools hazard families, location strings, and event years within documents, without assessing their assignment to individual events. GPT-5.4 ranked highest among ten LLMs (86.6% $F_1$). Verbatim evidence occurrence was 72.0% for GPT-5 and 47.2% for GPT-5.4, measuring textual traceability without establishing attribute support. We report production failure modes and automated label and location-rule compliance checks. Prompts, schema, and outputs will be released for adaptation to national reporting archives.