Hackathon project · Team member, first place · 1st of 14 teams
Wildfire Spread Prediction
A model and a live map of how fast an Alberta wildfire spreads and which pipeline assets it reaches first.
- 1st
- of 14 teams at the Databricks × Enbridge hackathon
- 20,000+
- past Alberta wildfires to learn from
- 0.78
- R² on fires the model hadn’t seen
What it is
A model that predicts how fast an Alberta wildfire will spread, and a map that shows which Enbridge pipelines and facilities it would reach first. We built it for the Databricks × Enbridge hackathon and took first place out of 14 teams.
How it works
- 19 years of fire and weather data
- Delta tables
- Spread-rate model
- Map simulation
- Time until each asset is reached
Predicting the spread rate
A gradient-boosted model trained on more than 20,000 past Alberta fires. It takes wind speed, temperature, humidity, fuel type, slope and the fire’s starting size, and predicts how fast the fire spreads. On held-out fires it reached an R² of 0.78.
Putting it on a map
The predicted rate drives a spread simulation on a map grid. Each step turns the rate into a chance of catching, weighted toward where the wind blows. Enbridge’s pipelines and facilities sit on the same map, so from any ignition point you can read how long each one has.
Built on Databricks
Everything ran on the Databricks Lakehouse. We cleaned and joined 19 years of fire and weather records into Delta tables in notebooks, and deployed the interactive simulation as a Databricks App. That app is what we demoed live to the Enbridge and Databricks judges.
What’s next
We proposed a next step to the judges: pull reported fires from the provincial API, so risk alerts go out on their own.