Leveraging large language models for floods mapping and advanced spatial decision support: A user-friendly approach with SATGPT

Flood mapping plays a crucial role in disaster management, but traditional methods often suffer from limitations in speed, accessibility, and real-time data utilization. This paper introduces SATGPT, a novel tool that leverages the code-writing abilities of Large Language Models (LLMs) to automate flood mapping using Google Earth Engine (GEE). Users simply input a prompt with flood duration and location, and SATGPT utilizes LLMs to generate GEE code for either extracting data from historical databases or performing unsupervised classification for flood detection. The resulting flood maps are presented in a user-friendly interface with 3D visualization, enabling detailed analysis of flooded areas and building footprints. Additionally, a self-paced learning course was developed and launched at the UNU-INWEH Water Learning Center to share knowledge and foster wider adoption. This paper presents the architecture and workflow of SATGPT, showcases its performance through flood map examples, and discusses the advantages, limitations, and future directions of LLM-powered flood mapping tools. We believe SATGPT offers a user-friendly, efficient, and scalable approach that can democratize access to flood information and empower informed decision-making in disaster management.