Detecting Wildfires
Before They Spread
AI-powered satellite monitoring system that analyzes NASA FIRMS data across Indonesia using machine learning anomaly detection and H3 spatial indexing
Why This Matters
Indonesia loses millions of hectares of forest each year to wildfires. Early detection is critical — every hour of delay means thousands more hectares at risk. Current monitoring systems struggle with the scale and complexity of Indonesia's 17,000-island archipelago.
Burned annually across Indonesian forests and peatlands
Detected per year via NASA FIRMS satellite instruments
From satellite capture to actionable alert delivery
How It Works
From raw satellite data to actionable wildfire alerts in four automated steps
NASA FIRMS VIIRS/MODIS satellite data pulled via automated daily pipeline
H3 hexagonal grid system groups hotspots into 23 km² resolution cells
Isolation Forest ML model identifies unusual hotspot patterns and severity levels
Top-K daily alerts ranked by anomaly severity score delivered to monitoring dashboard
Key Features
Everything you need to monitor and detect wildfire anomalies in real-time
Hexagonal grid visualization with color-coded anomaly severity and 3D elevation based on hotspot intensity
Isolation Forest model trained on temporal and spatial features to identify unusual fire patterns automatically
Daily top-K alert ranking with spatial coherence scoring (High/Medium/Low/Isolated) and severity classification
Real-time ingestion of VIIRS and MODIS satellite data with automated daily pipeline processing
30-day rolling window trend charts with daily aggregation, peak detection, and neighboring cell correlation
Full pipeline observability with status tracking, manual score triggers, and auto-refresh health monitoring
Built With Modern Tools
A robust stack designed for real-time satellite data processing and visualization
By The Numbers
Actual results from our Indonesia deployment — Nov 2025 to Jan 2026
FIRMS records ingested from VIIRS & MODIS satellites
H3 hexagonal cells grouped by date for spatial analysis
9.7% of cell-days flagged as anomalous by Isolation Forest
Daily top-K alerts after spatial coherence validation