Blaze Watch - AI Wildfire Predictor

An AI-driven web tool that predicts wildfire spread across Canada and visualizes risk zones on an interactive map. It helps local communities to stay alert and take early action.

Features

• AI-Powered Risk Prediction Predicts wildfire spread based on weather, satellite, and vegetation data
• Interactive Heatmap Visual heatmap overlay with colour-coded markers
• Detailed Information Clickable map points with risk details
• User-Friendly Fast, simple, and accessible interface

Installation

1. Clone the repo:

   git clone https://github.com/AustinBao/blaze-watch

2. Install pip packages

   pip install -requirements

3. Run Flask

   flask run

How We Trained Our AI

Training the Machine Learning Model

• Trained on satellite, weather, and vegetation data to predict fire spread
• Fire boundaries defined by max/min latitude and longitude points (N/E/S/W edges)
• Model predicts next-day boundary coordinates for each fire area
• Uses an XGBoost multi-output regression model
• Outputs 8 coordinate values representing updated fire edges

Model Evaluation

• Evaluated with Root Mean Square Error (RMSE) in lat/lon degrees
• Applies cross-validation to ensure accuracy and prevent overfitting

Architecture

Backend (Flask)

• Serves pages: / (landing), /map, /about
• API /predict-spread:
  • Takes fire cluster bounds
  • Fetches weather & vegetation data
  • Runs ML model to predict 3-day fire spread
  • Returns prediction JSON

Frontend (JS + Leaflet)

• Map centered on Canada with Esri World Imagery
• Fetches daily fire points from NASA FIRMS (fallback to local CSV)
• Clusters fires using Supercluster at zoom 4
• Cluster colors: yellow (small), orange (medium), red (large)
• Clicking cluster:
  • Expands points
  • Sends bounds to backend for predictions
  • Shows 3-day spread polygons & circles
• Side panel shows risk %, coordinates, and day slider
• Smooth map controls with loading indicators

Data Sources

• NASA FIRMS (hotspot detection via LANDSAT, MODIS, VIIRS)
• Open-Meteo API (Weather features like wind speed, temperature, humidity)
• NASA GISS (Vegetation data)
• Natural Resources Canada (historical wildfire perimeter data)

Tech Stack

Machine Learning

• pandas, geopandas, numpy, matplotlib
• scikit-learn, xgboost, pickle
• openmeteo-requests, retry-requests, requests-cache
• contextily

Backend

• Flask

Frontend

• HTML, CSS, JavaScript
• Bootstrap
• Leaflet.js
• Supercluster
• Turf.js
• noUiSlider