FireGuard: Interactive Fire Prevention Dashboard
Inspired by the recent LA fires of 2024, the team was eager to create a tool that could help communities stay informed about active wild fires and future wild fire risks. It was a tool meant to be leveraged by communities to prepare for such catastrophic events as global warming continues to increase in magnitude.
We have created a one stop dashboard that users can navigate to assess wild fire risk in their areas. This solution currently only supports predictions in California, but future iterations aim to increase the scope to other regions. The dashboard provides an interactive map that users can navigate the find areas of interest/concern, determine additional fire risk details, and general heuristics around the fuel load and availability in the selected area.
The dashboard offers three views. Each view offers different insights.
The default view provides general fire risk assessment. It can be used by the general public assess the risk of wild fires in their immediate area.
The Industrial view overlays the fire risk information with key power infrastructure. Power companies can leverage this information to focus their maintenance efforts on areas that are at highest risk.
Finally the Residential view overkays the fire risk information with areas that are designated to be residential zones. This includes areas that are currently setup as residential zones and have been marked for future development. Community planners can leverage this information to determine ideal, wild fire safe zones for future housing development.
The data for the dashboard was sourced from satellites, NASA's FIRMs API and ERA5 Weather API. We used Lambda to compile the data in batches and we used S3 to store the data. We also used Sagemaker to process the data and to train two machine learning models. The machine learning models were deployed to an endpoint and current weather and climate data is sent to the endpoint to make a prediction on how likely it is that a fire will ignite in the time and area designated.
There are many approaches in academic literature that utilize satellite data to map locations of fires and we learned a lot from these papers and their techniques. The transformation and feature engineering of the data follow multiple approaches that were taken in other papers, however, we have adapted the dashboard to be used to predict fires and also forecast the spread of fires.
Course
Data Science 210. Capstone , Spring 2025More Information
