From raw satellite pixels to operational fire intelligence — anywhere on Earth
Fork of earthlab/firedpy — extended with VIIRS burned area processing, FIRMS cross-validation, land cover classification, and a real-time web dashboard for operational fire monitoring worldwide.
Release: 2026-03-19 21:00 UTC+2
🇺🇦 Made in Ukraine with love
This system transforms NASA VIIRS VNP64A1 burned area rasters (500m monthly composites) into classified, confidence-scored fire event polygons — ready for analysis, reporting, and operational decision-making. Works for any region on Earth where VNP64A1 data is available.
Compared to using raw VNP64A1 rasters directly, this pipeline adds:
- Spatial-temporal event grouping — clusters individual burned pixels into coherent fire events
- Multi-window union merge — combines Win08 + Win15 for maximum detection coverage
- QA bit-mask filtering — removes water, cloud artifacts, and low-quality observations
- Confidence scoring — area + duration + patch count + uncertainty weighting
- FIRMS cross-validation — independent active fire confirmation
- Land cover classification — fire type identification (agricultural / wildfire / grassfire / peatfire)
- Perimeter refinement — morphological smoothing for irregular fire boundaries
| Year | Raw Patches | Events | Total Area |
|---|---|---|---|
| 2014 | 58,429 | 26,149 | 14,953,895 ha |
| 2015 | 54,078 | 23,990 | 15,969,695 ha |
| 2016 | 38,512 | 16,813 | 9,137,181 ha |
| 2017 | 50,411 | 21,952 | 21,071,475 ha |
| 2018 | 28,828 | 13,477 | 8,222,237 ha |
| 2019 | 49,847 | 21,447 | 15,983,538 ha |
| 2020 | 39,915 | 18,018 | 10,380,973 ha |
| 2021 | 27,963 | 11,910 | 10,592,786 ha |
| 2022 | 22,207 | 10,197 | 8,635,026 ha |
| 2023 | 24,634 | 11,133 | 6,923,100 ha |
| 2024 | 27,599 | 12,093 | 5,879,824 ha |
| Total | 422,423 | 187,179 | 127,749,729 ha |
VNP64A1 TIFFs (500m burned date rasters, Win08 + Win15)
|
QA filtering (VNP64A1 bit-mask: land + valid + special condition codes)
|
Multi-window union merge (earlier DOY for ignition dating)
|
Vectorization (raster -> polygon with per-patch DOY attribution)
|
Quality pipeline:
-> Area filtering (>=25 ha, removes single-pixel noise)
-> Spatial merging (1km buffer, 5-day calendar date tolerance)
-> Confidence scoring (area + duration + patch count + uncertainty)
-> Perimeter refinement (morphological smoothing)
|
FIRMS cross-validation (NASA active fire hotspots)
|
Land cover classification (ESA WorldCover 2021)
|
Export: GeoPackage + GeoJSON + Statistics CSV
|
Web dashboard (Leaflet) + REST API (FastAPI)
| Feature | Description |
|---|---|
| Sensor-agnostic core | Supports both MODIS MCD64A1 and VIIRS VNP64A1 via ProductConfig |
| VNP64A1 QA decoder | 3-mode bit-mask filtering (strict/standard/permissive) with bits 5-7 special condition handling |
| Union-merge strategy | Combines Win08 and Win15 temporal windows for maximum fire detection |
| Calendar date merging | Correctly handles year-boundary fires (Dec+Jan) using calendar dates, not raw DOY |
| FIRMS cross-validation | NASA active fire hotspots confirm events — boosts confidence |
| Land cover classification | ESA WorldCover 10m classifies fires as agricultural, wildfire, grassfire, peatfire |
| Cross-sensor validation | Compare VIIRS vs MODIS events with IoU, area bias, and temporal metrics |
| Web dashboard | Dark-themed Leaflet map with filters by year, sensor, confidence, fire type |
| REST API | FastAPI endpoints for programmatic access |
| Equal-area geometry | All area calculations in EPSG:6933 — correct hectares regardless of input CRS |
git clone https://github.com/d-grabovets/firedpyVIIRS-Modis.git
cd firedpyVIIRS-Modis
pip install -e .# Required: NASA FIRMS API key (free)
# Get yours at: https://firms.modaps.eosdis.nasa.gov/api/area/
export FIRMS_API_KEY="your_key_here"
# Required: path to VNP64A1 burn TIFs
export VIIRS_DATA_ROOT="/path/to/VNP64A1_TIFF"
# Optional: ESA WorldCover raster for land cover classification
export WORLDCOVER_PATH="/path/to/worldcover_6933_500m.tif"# Single year
python scripts/build_burned_area.py \
--year 2024 \
--burn-dir /data/VNP64A1/raw/2024 \
--qa-dir /data/VNP64A1/raw/2024 \
--qa-mode standard \
--merge-tolerance 5
# Multi-year with FIRMS and land cover
python scripts/run_production.py --year 2023 --year 2024 \
--download-firms --classify-landcoveruvicorn scripts.api_server:app --host 0.0.0.0 --port 8000
# Open http://localhost:8000/web/firedpyVIIRS-Modis/
+-- firedpy/ # Core library
| +-- product_config.py # Sensor-agnostic MODIS/VIIRS config
| +-- firms_integration.py # NASA FIRMS download & cross-validation
| +-- event_quality.py # Confidence scoring & quality pipeline
| +-- landcover_classify.py # ESA WorldCover fire type classification
| +-- export_geojson.py # GeoJSON export for web
| +-- qa/
| | +-- viirs_qa.py # VNP64A1 QA bit-mask decoder (3 modes)
| +-- validation/
| | +-- cross_sensor.py # VIIRS vs MODIS event matching
| +-- data_classes.py # Data access (Earthaccess, parametrized)
| +-- model_classes.py # EventGrid space-time clustering
+-- scripts/
| +-- build_burned_area.py # Core VIIRS burned area builder
| +-- run_production.py # Production pipeline orchestrator
| +-- api_server.py # FastAPI REST API server
| +-- validate_cross_sensor.py # Cross-sensor validation CLI
+-- web/
| +-- index.html # Leaflet web dashboard
+-- examples/
| +-- ukraine_2014/ # Example output: 6,069 events
+-- config/ # Configuration templates
+-- ARCHITECTURE_PLAN.md # Technical architecture document
| Source | Resolution | Coverage | Usage |
|---|---|---|---|
| VIIRS VNP64A1.002 | 500m, monthly | Global, 2012+ | Primary burned area detection |
| NASA FIRMS VNP14IMG | 375m, near-real-time | Global | Active fire cross-validation |
| ESA WorldCover 2021 | 10m | Global | Land cover classification |
| MODIS MCD64A1.061 | 500m, monthly | Global, 2000+ | Alternative/comparison product |
| Endpoint | Description |
|---|---|
GET /health |
Service health check |
GET /burned/events?year=2024 |
Fire events GeoJSON |
GET /burned/bbox?year=2024&west=30&east=36&south=49&north=52 |
Spatial query |
GET /burned/stats?year=2024 |
Monthly statistics |
GET /burned/summary?year=2024 |
Annual summary |
GET /docs |
Interactive Swagger UI |
| # | Improvement | Impact |
|---|---|---|
| 1 | VIIRS VNP64A1.002 support | Higher-resolution burned area (500m grid, 375m sensor) |
| 2 | ProductConfig architecture | Sensor-agnostic — switch MODIS/VIIRS via --product flag |
| 3 | QA bit-mask filtering | 3 modes with bits 5-7 special condition codes |
| 4 | Multi-window union merge | Win08+Win15 combined for maximum detection |
| 5 | Calendar date merging | Fixes year-boundary bug (Dec+Jan fires correctly merged) |
| 6 | Equal-area CRS | All geometry operations in EPSG:6933, not geographic degrees |
| 7 | FIRMS integration | Active fire hotspot cross-validation |
| 8 | Land cover classification | ESA WorldCover fire type identification |
| 9 | Confidence recalibration | Balanced 13/43/44% (high/med/low) with uncertainty weighting |
| 10 | Cross-sensor validation | VIIRS vs MODIS event matching with IoU and area bias |
| 11 | Web dashboard | Leaflet map with sensor/year/confidence/type filters |
| 12 | REST API | FastAPI for programmatic access |
This repository uses a dual license model:
| Component | License | Commercial Use |
|---|---|---|
| Original firedpy code (MODIS) | MIT License | Allowed |
| VIIRS extension (new modules) | AGPL-3.0 | Must open-source your code, or buy a commercial license |
The VIIRS extension is licensed under AGPL-3.0 — free to use, but you must open-source any modifications and derivative works. Companies that want to keep their code proprietary can purchase a Commercial License — see LICENSE_VIIRS.md for details.
- Original firedpy: Earth Lab, University of Colorado Boulder
- VIIRS Extension & Global Adaptation: Dmytro Grabovets
- NASA FIRMS: Fire Information for Resource Management System
- ESA WorldCover: European Space Agency land cover product
🇺🇦 Developed in Ukraine with love 🇺🇦
For firefighters, researchers, and anyone who cares about our planet