Research Experiences for Undergraduates (REU), NSF NHERI 2025: Evaluating the Efficacy of AI Vision-Language Models in Aerial Image Wildfire Damage Assessments

This project analyzes building damage in the Pacific Palisades, based on aerial imagery and the proposed Combustion Hierarchy Scale (CHS Index) damage levels. Damage assessment is analyzed for both human manual classification as well as automated classification through the use of various vision-language models including Google Gemini 2.5 Pro, Claude Sonnet 4, and ChatGPT-5 Plus. Data discussed in this project can be used to understand the applications of AI in hazard modeling research, with applications to future damage assessment, modeling analyses, and recovery planning. Examples include debris removal planning, burn intensity analysis, probabilistic fire spread modeling, and chemical contamination detection. This work uniquely analyzes large-scale wildfire building damage with multi-level damage classifications. Although results in this study are focused on a limited study area, the methodology proposed is applicable to the entire Pacific Palisades, Altadena (Eaton wildfires), and beyond.