Leveraging LLMs for rapid disaster impacts assessment through News Media: A case study of the 2025 Myanmar Earthquake

After a disaster strikes, information starts pouring in. However, beyond the initial seismic analysis, it is primarily through local reporting, distributed media, and testimony that an initial appraisal of the situation is gained. Satellite-based remote sensing, although powerful, can be significantly delayed. This poses a limitation on the information that disaster managers, responders, and decision makers can obtain and act upon meaningfully. Real-time harnessing of distributed media, such as news media, during the acute phase after a disaster can support, focus, and enhance early decision making as well as remote sensing data gathering efforts. The Myanmar Earthquake of March 28, 2025, offers a prime opportunity to test methods that target news media to provide a very early picture of the earthquake’s impacts. The present study proposes a real-time pipeline to harness news media to gain an initial understanding of the disaster and its impact on populations. Moreover, the proposed methods elucidates how reporting evolves in the post disaster window by monitoring the release schedule of media coverage. News articles are processed with Large language models and natural language processing techniques to extract key geographic and impact information. Relative to imagery based remote sensing data and in spite of local impediments to press, news media is more frequent, has a faster return period, has larger coverage, and naturally discriminates between low and high impact areas. The present method provides rapid, actionable intelligence that can be leveraged on its own or to inform and focus other remote sensing based response efforts.