MASTER Fire Detection Documentation

Detect active fire from NASA MASTER airborne thermal imaging data.

This project processes Level 1B (L1B) calibrated radiance from the MODIS/ASTER Airborne Simulator (MASTER) and identifies fire pixels using physics-based threshold detection, contextual anomaly analysis, multi-pass consistency filtering, and machine learning.

The data comes from the FireSense 2023 campaign – prescribed burns on the Kaibab Plateau in Arizona (October 18–20, 2023).

Contents

• Operator’s Guide
  • Quick Start
  • Flights in the Dataset
  • What Each Script Produces
  • Reading the Outputs
  • Real-Time Simulation
  • Data Requirements
  • Troubleshooting
• Science and Algorithm Guide
  • Instrument Overview
  • Channel Selection
  • Radiometric Processing
  • Fire Detection: Absolute Threshold
  • Fire Detection: Contextual Anomaly
  • Multi-Pass Consistency Filter
  • SWIR for False Positive Discrimination
  • Day/Night Classification
  • Vegetation Mapping (NDVI)
  • Mosaic Gridding
  • Connected Component Fire Zone Analysis
  • ML Fire Detection
  • Assumptions and Limitations
  • References
• Machine Learning Fire Detection
  • Assumptions
  • Feature Engineering
  • Model Architecture
  • Pixel-Wise Weighted BCE Loss
  • Training Data
  • Integration with Real-Time System
  • Comparison: Threshold vs ML
  • Usage
• API Reference
  • lib.constants
  • lib.io
  • lib.fire
  • lib.vegetation
  • lib.mosaic
  • lib.stats
• Script Reference
  • download_data.py
  • mosaic_flight.py
  • detect_fire.py
  • plot_burn_locations.py
  • plot_vegetation.py
  • fire_ml.py
  • realtime_fire.py