A high-precision catalogue of landslide events in China based on news text mining with large language model

Résumé fourni par la source

Landslides are a major geological hazard causing significant casualties and economic losses. Reliable risk assessment requires high-quality spatiotemporal event data, yet no publicly available landslide catalogue with fine-grained precision exists for China. To address this, we developed a landslide event catalogue for mainland China from 2008-2024 based on news reports. The dataset was generated via large-scale web crawling, information extraction using an open-source large language model (LLM), event deduplication, geocoding, and multi-stage validation. It contains 1,582 events with detailed spatiotemporal attributes, some with minute-level temporal precision and spatial resolution down to the county, village, or specific reported sites. Evaluation shows that, while casualty-related information is less accurate, the LLM reliably captures key attributes such as time, location, and triggering factors. This demonstrates the feasibility of using LLMs to extract critical landslide data from news reports. Compared with existing catalogues, our dataset offers more events and improved spatiotemporal accuracy, providing a valuable resource for landslide hazard assessment, early warning model development, and disaster risk management in China.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.