QuakeMap: An AI-Powered Multimodal Earthquake Assessment Platform for Data-Scarce Regions

Khant Nyi Hlaing, Susan E. Hough, Clara E. Yoon, Aarnav Agrawal, S. Mostafa Mousavi, & Salvador Blanco

Submitted September 7, 2025, SCEC Contribution #14635, 2025 SCEC Annual Meeting Poster #TBD

Instrumental seismic data remains limited in many parts of the world, including Myanmar. This makes it difficult to assess the impacts of earthquakes and can delay emergency response. This study presents a cloud-native platform that leverages Large Language Models (LLMs) and a self-corrective Retrieval-Augmented Generation (RAG) framework for automated post-earthquake shaking intensity analysis. The RAG approach allows LLMs to retrieve and consult trusted reference materials before responding to prompts, iteratively reviewing its output for accuracy. We focus on images and videos collected over several months following the 28 March 2025 M7.7 Mandalay, Myanmar, earthquake. In addition to Google’s Gemini model and the RAG framework, we use the official Modified Mercalli Intensity (MMI) scale definitions as a trusted source stored in the Pinecone Vector database. We guide the model’s reasoning with context, prompt-chaining, and Chain-of-Thought prompting techniques to estimate shaking intensities. The system processes visual and auditory cues such as building damage, debris, and ambient sound to estimate shaking intensities using the MMI scale. The system can process dynamic file uploads and generate MMI intensity values, accompanied by confidence metrics and explainable reasoning. The results are stored in the cloud and visualized using Generic Mapping Tools (GMT). For future earthquakes, our workflow could be applied to interpret images and videos posted to social media in the immediate aftermath of a damaging earthquake. Artificial intelligence-driven, crowdsourced macroseismic analysis thus offers an opportunity to complement and enhance traditional systems like Did You Feel It? (DYFI) in underserved regions, improving rapid characterization of damage for emergency response.

Citation
Nyi Hlaing, K., Hough, S. E., Yoon, C. E., Agrawal, A., Mousavi, S., & Blanco, S. (2025, 09). QuakeMap: An AI-Powered Multimodal Earthquake Assessment Platform for Data-Scarce Regions. Poster Presentation at 2025 SCEC Annual Meeting.

Related Projects & Working Groups
Ground Motions (GM)