An AI-powered web app that predicts and visualizes wildfire risk across Ontario based on user-uploaded weather data.

1. Upload a .csv file containing weather data for Ontario stations.
2. A machine learning model predicts wildfire occurrence tiers (Low / Moderate / High / Extreme) for the next month.
3. Results are visualized on an interactive Leaflet map with color-coded markers and a visible legend:
   • 🟢 Low Risk — Fewer than 50 fires — Routine local response
   • 🟡 Moderate — 50–150 fires — Regional resource prep
   • 🟠 High — 150–300 fires — Full deployment
   • 🔴 Extreme — Over 300 fires — National/international coordination
4. Each marker includes a GPT-generated explanation and prevention tip.

 ```mermaid graph TD A[User Uploads CSV] --> B[FastAPI Backend] B --> C[ML Model (.pkl)] C --> D[Predicted Fire Count] D --> E[Frontend Receives Prediction] E --> F[Animated Number Display] E --> G[Map Color Update] G --> H[CircleMarker Color & Tooltip] ``` 

forest-fire-predictor/
├── ml-model/              
│   ├── train_model.py
│   ├── merge_weather_fire.py
│   ├── forest_fire_merged.pkl
│   └── fire_counts_clean.csv
├── web-app/
│   ├── frontend/   # React + Tailwind + Leaflet
│   └── backend/    # FastAPI
│       ├── main.py
│       ├── model_loader.py
│       └── requirements.txt
├── README.md

cd web-app/backend
pip install -r requirements.txt
uvicorn main:app --reload

cd web-app/frontend
npm install
npm run dev

Set .env in frontend/:

VITE_API_URL=http://localhost:8000

Station,TempMax,TempMin,TempMean,Precipitation,CoolingDegreeDays
Pickle Lake,32,5,18,65,45
Red Lake,30,6,17,70,50
Sioux Lookout,33,4,19,80,60

Each row = 1 Ontario station 📈 Output = wildfire tier + natural language explanation

“This location is at high risk due to elevated temperatures (34°C), low humidity, and low rainfall. Avoid open flames and clear dry brush near your property.”

• Leaflet map renders all Ontario fire stations

• CircleMarkers color-coded by predicted fire count:

  • 🟢 0–49
  • 🟡 50–99
  • 🟠 100–149
  • 🔴 150+

• Hover tooltip shows station name + predicted fire count

• Legend explains color scale

• Smooth color fade transitions using CSS

Trained with a Random Forest Regressor to predict monthly fire counts from historical weather.

A fire count (rounded) → Mapped to Risk Tier:

import pandas as pd
import pickle

# Load model
with open("forest_fire_model_aggregated.pkl", "rb") as f:
    model = pickle.load(f)

# Predict
X_new = pd.DataFrame([{
    "Tm": 17.5,
    "Tx": 32.0,
    "Tn": 3.0,
    "S": 0.0,
    "P": 60.0,
    "CDD": 25.0,
    "Weather_Month_Num": 6,
    "Year": 2024
}])

y_pred = model.predict(X_new)
print("Predicted fires:", y_pred)

✔️ Core pipeline complete:

• ✅ Cleaned weather + fire count datasets (2000–2023)
• ✅ Merged and aggregated per station per month
• ✅ Trained ML model on aggregated dataset
• ✅ Backend integration and prediction API
• ✅ CSV upload + file parsing working
• ✅ Map color visualization and tooltips

• Animated splash loader (~1.5s)
• Prediction button disables on click
• RingLoader spinner during model run
• Fire count animates from 0 to predicted value

• 🔁 Integrate live weather APIs (e.g., OpenWeatherMap)
• 🛰️ Add satellite/vegetation data
• 🔄 Monthly model retraining
• 📊 Add dashboard for trend analysis

• Frontend: https://fire-predict.vercel.app](#)

MIT License