Welcome to WildfiresAI

Using ML to help prepare for wildfires

Learn More

Wildfires in California

In 2020, California experienced a record number of wildfires, devastating families and destroying homes. Because of its detrimental impact, we decided to create WildfiresAI, a multi-faceted platform that utilizes Machine Learning to tackle the Wildfires.


  • Supply government officials with a convenient, quick, and accurate app
  • Provide key visualizations of predicted fires to improve resource allocation
  • Utilize 4 different Machine Learning models to improve existing methods

With this app, government officials can be assured that wildfires can be easily contained. Our fire detector (object detection), magnitude predictor (fire size), and put-out time estimator (days) make wildfire management seem like a trivial task.

Wildfires in Numbers

Within the US in 2020 alone

Thousand Wildfires

Million Acres Burned

Thousand Structures Damaged

Billion Dollars Lost

Functions

Fire Localization

Localizing fires using a custom trained Tensorflow2 Object Detection model

Magnitude Prediction

Predicting the size of a fire in acres using a combined ensemble model with Elastic Net, SVM regressor, and XGBoost regressor

Put-Out Time Estimation

Predicting the time it takes to put out a fire in a given area using a Sequential Neural Network

Visualization

Using Folium to visualize fire sizes and conveniently sort by a variety of features

Features

Localizing Fires

Using a custom-trained Tensorflow2 Object Detection Model, this feature accurately uses bounding boxes to surround a fire with a rectangle with the % chance of confidence.

  • Clearly delineates the boundaries of the fire
  • Can be integrated into video to show progression
  • Provides a confidence estimator of the localization

Predicting Magnitude of Fires

Combining Elastic Net, SVM regressor, and XGB regressor into an ensemble model, this feature provides an accurate prediction of the fire size (burn area) in acres.

  • Averages several models' outputs for a final prediction
  • Uses GridSearch to tune the best hyperparameters
  • Surpasses previous basic regression methods for wildfires

Predicting Fire Put-Out time

Using a Sequential Neural Network, we are able to accurately predict the time it takes to put out a fire in a given area.

  • Uses a 3-layered densely connected neural network for predictions
  • Model is compiled with the Adam optimizer and Mean-Squared-Error loss
  • Makes it more covnenient for fire fighters to allocate resources

Visualizing Predicted Fires

In tandem with our ML models, our visualizations page displays predicted fires using Folium.

  • Interactive map that can easily be sorted by various features
  • Markers differ in size and color based on fire size
  • Allows for accurate resource allocation