# From UNDRR Reports to Event Records: Schema-Constrained LLM Extraction of Georeferenced Disasters

arXiv: https://arxiv.org/abs/2609.23853 · HTML: https://mlsanity.com/paper/2609.23853/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 7 of its 23 references that are in the map.
- First posted on arXiv 20 Sep 2026.

## Builds on (most cited first, 7 of 7)

- [Grammar-Constrained Decoding for Structured NLP Tasks without Finetuning](https://mlsanity.com/paper/2305.13971/) (2023)
- [Document-Level Event Argument Extraction by Conditional Generation](https://mlsanity.com/paper/2104.05919/) (2021)
- [GoLLIE: Annotation Guidelines improve Zero-Shot Information-Extraction](https://mlsanity.com/paper/2310.03668/) (2023)
- [Instruct and Extract: Instruction Tuning for On-Demand Information Extraction](https://mlsanity.com/paper/2310.16040/) (2023)
- [Harnessing Large Language Models for Disaster Management: A Survey](https://mlsanity.com/paper/2501.06932/) (2025)
- [Improving Toponym Resolution with Better Candidate Generation, Transformer-based Reranking, and Two-Stage Resolution](https://mlsanity.com/paper/2305.11315/) (2023)
- [Large Language Models for Geolocation Extraction in Humanitarian Crisis Response](https://mlsanity.com/paper/2602.08872/) (2026)

## Abstract (arXiv)

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 F1, 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% F1). 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.