Authors

Yutong Wu, Youjia Li, Miaorou Tan

Research Question

How can AI-powered tools help homeowners in wildfire-prone regions in the US assess their property’s wildfire risk and implement targeted, cost-effective mitigation strategies? This project addresses the lack of accessible, data-informed design guidance for individual homeowners by developing an AI agent that integrates geospatial data, risk models, and building codes to provide personalized recommendations and visualizations for wildfire-resilient home and landscape improvements.

Background

As wildfires become increasingly frequent and destructive in the U.S., homeowners in high-risk areas like Los Angeles face urgent pressure to adapt. However, many lack access to the technical guidance needed to assess risks and implement effective mitigation strategies. This project addresses that gap by developing an AI-powered decision support tool that leverages urban data, machine learning, AI, and user-centered design.

Methodology

This project uses a streamlined technology stack, incorporating large language models (LLMs) such as GPT-4 or Claude 3.7 for the chatbot, with Retrieval-Augmented Generation (RAG) technology for pulling relevant information from structured datasets. Students are working across different components of the project, from developing risk assessment models using machine learning to designing the user interface or integrating geospatial data.

This project combines various approaches to create an effective AI-powered solution. Data collection involves acquiring publicly available geospatial datasets, including vegetation, slope, and historical fire data, alongside satellite imagery. These datasets are then processed and analyzed using geospatial tools such as ArcGIS to assess the specific wildfire risk for individual properties. Machine learning techniques are applied to create models that generate personalized risk assessments and recommend mitigation strategies, such as home improvements, based on the specific characteristics of each property. An interactive, user-friendly chatbot interface is being implemented using external commercial AI APIs, such as OpenAI, HuggingFace, Claude, or similar models, to provide homeowners with tailored suggestions and guidance based on their questions. The project focuses on creating basic but effective features, such as personalized risk scoring, visualizations of potential outcomes, and actionable mitigation recommendations, all of which will be developed within the academic timeline using tools like Python and machine learning libraries.

Deliverables

• A wildfire risk assessment tool capable of assessing wildfire risks for individual homes using geospatial data, satellite imagery, and home parcel data will provide personalized risk scores and suggest appropriate mitigation actions, such as material upgrades and defensible space creation.
• An interactive chatbot interface guides homeowners through the planning and implementation of wildfire-resilient improvements. It also provides tailored recommendations, answers questions, and allows users to interact with the system in a simple and intuitive manner.
• A visualization of mitigation strategies displays risk assessments and potential outcomes of different mitigation strategies. This feature allows homeowners to better understand the impact of various actions and make informed decisions to enhance their home’s wildfire resilience.

Data Sources

This project relies on a combination of open-source geospatial datasets and publicly available satellite imagery. All datasets are either public or accessible via institutional research licenses.

1. Vegetation and land cover (e.g., National Land Cover Database (NLCD), Sentinel-2) data are publicly available via the United States Geological Survey (USGS) and Copernicus.
2. Slope and elevation (e.g., USGS Digital Elevation Models – DEM) data are openly accessible through USGS Earth Explorer.
3. Historical wildfire perimeters and risk zones (e.g., CAL FIRE, U.S. Forest Service) are publicly available.
4. Parcel-level data (e.g., LA County Open Data Portal) including lot boundaries, building footprints, and basic property characteristics.
5. Building codes and zoning regulations are available on city and county websites.
6. Satellite imagery (e.g., Google Earth Engine, Sentinel Hub) are accessible for non-commercial research use.