Policy Extraction and Mapping for Wildfire Resilience

Bernette Chan

This project develops a machine learning pipeline to help city planners identify fire-related policies in municipal planning documents for the PIRS scorecard, a wildfire resilience assessment tool. The policy extraction combines large language model querying with rule-based methods to analyze both structured and unstructured documents.

Policies are classified using Latent Dirichlet Allocation (LDA), enabling efficient evaluation of relevant policy categories.

Poster summarizing the wildfire policy extraction pipeline, initial classification results, and spatial analysis efforts.

I contributed to the spatial tagging of extracted policies using named entity recognition (NER), which is detailed on pages 7–10 of the full report. This work will enhance the ability to map and analyze wildfire risk geographically.

Download Poster (PDF) | Download Report (PDF)

← Back to Portfolio