Logo image
Back
Large Language Model-Based Classification of Flash Flood Impacts Across the United States
Journal article   Peer reviewed

Large Language Model-Based Classification of Flash Flood Impacts Across the United States

Jorge A. Duarte, Jonathan J. Gourley, Humberto J. Vergara, Pierre E. Kirstetter, Charles D. Nicholson and Maciej Adamiak
Artificial intelligence for the earth systems
03/06/2026
DOI: 10.1175/AIES-D-24-0112.1

View Online

Abstract

In late 2019, the National Weather Service (NWS) transitioned their warning paradigm into an impact-based warning (IBW) format. This new format applies to flash flood warnings, and intends to provide detailed information about the hazard, its source, an impact narrative , and a flash flood damage threat tag. These damage threat tags and additional warning information aim to elicit different calls to action for the emergency management community and the public (e.g. Wireless Emergency Alerts). With the advent of highly performant, widely available, and affordable access to pre-trained large language models (LLMs) like ChatGPT, in conjunction with the Flash Flood Severity Index (a systematically-conceived framework for assessing and therefore classifying flash flood severity from textual flash flood reports), the present work explores a systematic LLM-based method for classifying flash flood report narratives into damage threat tags. Specifically, this work showcases the classification of flash flood reports into Flash Flood Severity Index (FFSI) impact categories using ChatGPT, and prompts engineered to incorporate textualized forms of FFSI impact definitions based on the published literature. Report classifications were verified and contrasted with Flooded Locations and Simulated Hydrographs (FLASH) product outputs for each reported flood event, within a given spatio-temporal 3D window. This approach was first tested on a reduced dataset of 663 expert-classified reports which enabled us to compare and contrast the method’s performance. Subsequently, it was used to classify over 22 thousand historical Local Storm Reports (LSRs) between May 2018 and June 2022. This unprecedented dataset is the cornerstone that has enabled the research and development of new experimental FLASH products, which look to inform forecasters of the anticipated impacts for any given flash flood forecast, in line with the NWS’ recently implemented IBW framework.

Details

Metrics

1 Record Views
Logo image