Flame Wacth

Flame Watch is an advanced wildfire prediction system powered by machine learning, built and trained entirely by our team to help detect fire risks before they escalate.

Comment

What the App Does

FlameWatch is a wildfire prediction web application that provides early warnings about potential wildfire risk based on real-time environmental conditions and historical fire occurrence data.

The application allows users to search or click on any location on an interactive map and instantly view the wildfire risk level categorized as Safe, Moderate, High, or Critical. The risk predictions are generated by a machine learning model trained on trusted satellite data.

This tool is intended for use by local authorities, fire departments, researchers, and individuals living in high-risk areas. It empowers them to take preventive action before fires occur, helping protect lives, property, and the environment.

How the ML Model Was Developed and Trained

The core of the system is a Random Forest classifier developed using public datasets from NASA.

Datasets Used:

  • NASA FIRMS (Fire Information for Resource Management System): Provided labeled wildfire occurrences (with date, latitude, and longitude).
  • NASA POWER (Prediction Of Worldwide Energy Resource): Provided environmental features such as temperature, humidity, wind speed, solar radiation, precipitation, etc.

Development Steps:

  1. Data Collection: Fire event data from FIRMS was paired with environmental data from NASA POWER for the same locations and timestamps.
  2. Feature Engineering: Key variables known to influence wildfire behavior were extracted and standardized.
  3. Labeling: Each data point was labeled as either “fire occurred” or “no fire”.
  4. Model Selection and Training: A Random Forest classifier was trained using the scikit-learn library. An 80/20 train-test split and k-fold cross-validation were used to validate performance.
  5. Model Export and Deployment: The final model was saved using joblib and integrated directly into the Django backend for inference.

Model Performance:

  • Accuracy: 93%
  • Precision: 92%
  • Recall: 91%

These metrics reflect the model’s ability to accurately identify wildfire-prone conditions, balancing false positives and false negatives effectively.

How It Integrates Into the App

The application consists of two primary components: a vanilla JavaScript frontend and a Django backend.

Frontend (HTML, CSS, JavaScript, Leaflet.js):

  • Built using plain HTML, CSS, and JavaScript for speed and simplicity.
  • Displays a live map using Leaflet.js, with color-coded overlays representing risk levels.
  • Allows users to search or select any location on the map.
  • Sends geographic coordinates to the backend via AJAX to receive the current wildfire risk score.

Backend (Django):

  • Receives location input from the frontend and fetches real-time environmental data from the NASA POWER API for that location.
  • Prepares feature vectors based on the environmental conditions.
  • Loads the trained Random Forest model and generates a wildfire risk probability.
  • Returns the risk level (Safe, Moderate, High, Critical) to the frontend for display on the map.

This full-stack setup ensures a fully dynamic application where predictions are updated in real time based on current environmental data.

The Problem It Solves and Our Approach

The Problem: Wildfires are increasingly frequent and destructive, threatening ecosystems, communities, and economies. Most existing solutions only detect wildfires after they have started, leaving little time for evacuation or preventive action.

Our Approach: FlameWatch-OpenML is designed to shift from reactive fire detection to proactive fire risk prediction. The system enables communities to assess danger levels in advance and take timely actions such as clearing vegetation, creating defensible space, or preparing for evacuation.

Key Principles:

  • Preventive Intelligence: By forecasting risk instead of detecting existing fires, users gain valuable time.
  • Real Machine Learning: The model is fully custom-trained—no language models, wrappers, or external ML APIs.
  • Open Source and Transparent: All code, model training processes, and data references are available in a public GitHub repository.
  • Lightweight and Accessible: Built with minimal dependencies for maximum compatibility and ease of deployment.

FlameWatch provides data-driven insight where it matters most before a wildfire becomes unmanageable.

Share this project:

Updates