Methodology & Data Sources
How JeryMotro detects, predicts and alerts on wildfires across Madagascar.
Data Sources — NASA FIRMS
JeryMotro relies on NASA FIRMS (Fire Information for Resource Management System), aggregating active fire detections from two satellite sensors:
- MODIS — 1 km resolution, passes 2×/day.
- VIIRS — 375 m resolution, passes 2×/day.
Data is refreshed approximately every hour.
Prediction Model — XGBoost v2.1
The J+1 prediction engine uses an XGBoost model trained on 3 years of FIRMS historical data for Madagascar (2021–2024), weather data, NDVI, and land classification.
Overall accuracy: 89% on the test set (5-fold cross-validation).
Processing Pipeline
- Hourly automated retrieval from NASA FIRMS API.
- Geocoding and matching each detection to an administrative region.
- Risk score computation by XGBoost model.
- Geographic clustering of nearby detections.
- Alert dispatching via n8n (Email, SMS via Twilio, WhatsApp via WAHA).
Average processing delay: < 2 seconds per detection batch.
Academic Context
JeryMotro is a prototype developed as part of a Bachelor's thesis (L3) in Software Engineering (2026). It demonstrates the feasibility of a low-cost environmental monitoring platform using open satellite data and modern ML techniques.
Limitations & Warnings
- Detections may include false positives (intentional burning, industrial areas).
- Cloud cover can temporarily mask real fires.
- Delay between ignition and satellite detection: approximately 1–3 hours.
- SMS/WhatsApp alerts depend on local mobile network availability.
Contact & Open Source
For methodology or data questions: randriamanantenatsikynyantsa@gmail.com.
