Inspiration
Wildfires constantly ravage Canada and the United States every year; a low cost offline simulation model can help prevent property loss or loss of life.
What it does
Tracks wildfires from both FIRMS and allows users to place custom fires, and creates a simulation of how the fire might spread.
How we built it
We spent a great deal of time collecting data ranging from terrain elevations to weather and wind speeds to road and city infrastructure in order to make the fire behave as realistic as possible. For a tech stack, we used traditional ML models such as histgradientboostingclassifier from scikit-learn, FastAPI with Uvicorn for a webpage, and Pyside6 for the desktop app.
Challenges we ran into
It was difficult to optimize the model; a lot of data needs to be fetched real time. To account for this, we pre-fetch some data during the simulation time, while the app is loading, once we know where the user has placed or the FIRMS locations. We were able to pull simulation down over 10x from the first iteration of the model.
Accomplishments that we're proud of
We were able to get a model with over 70% accuracy to historical wildfires, and create an efficient, lightweight, and portable app with offline capability.
What's next for Wildfire Atlas
Support for responses (dropping water on fires, allowing the user to simulate weather such as rain, drought), and allowing an interactive environment (i.e. the user can interact with and set variables in the map).
Log in or sign up for Devpost to join the conversation.