What is GEOINT?
GEOINT (Geospatial Intelligence) is intelligence extracted from imagery and mapping data. The US IC (Intelligence Community) defines it as "intelligence about physical features and events on Earth derived from analyzing imagery and geospatial information."
Three pillars: imagery (satellite, aerial photos), geospatial (maps, digital elevation), temporal (change over time).
Bellingcat's MH17 investigation (2014) pioneered open-source GEOINT: using Google Earth, satellite metadata, and geolocation to pinpoint missile launch coordinates with 200m accuracy.
Geospatial Data Sources (2026)
Free/Open Satellite Imagery
Sentinel-1: EU space program. Synthetic Aperture Radar (SAR). All-weather (clouds don't block SAR), 10m resolution, revisit every 12 days. Free. Good for detecting construction, water bodies, deforestation.
Sentinel-2: Optical imagery. 10-60m resolution (depends on band), multispectral (13 bands). Revisit every 5 days (2 satellites). Free. Best for NDVI (vegetation indices), land-use classification.
USGS Landsat: 30m resolution, multispectral, free archive dating to 1972. Lower resolution than Sentinel-2 but longer history.
Commercial/Premium Imagery
Planet Labs: ~200 daily satellites, 3m resolution, daily revisit capability. Cost: $100-1000/month for API access. Real-time monitoring of cities, ports, facilities.
MAXAR (formerly DigitalGlobe): Military-grade, 30cm resolution, classified contracts. Civilian imagery ~$3-5 per km².
Google Earth: Free, accessible. Imagery age varies (6 months to 3 years old). Resolution 15m-1m depending on location.
Analysis Techniques
Change Detection
Compare two timestamps. Identify pixel changes > threshold. Applications: construction sites, deforestation, urban sprawl, military deployments.
Simple approach: normalize images (same sensor, sun angle) → calculate NDVI (Normalized Difference Vegetation Index) → diff between dates → threshold (e.g., NDVI drop > 0.3 = deforestation).
Object Detection
Use CNN (YOLO, Faster R-CNN) trained on satellite imagery. Detect vehicles, buildings, ships, aircraft. Typical accuracy: 85-95% for large objects (vehicles, buildings), 60-80% for small objects (people, vehicles at distance).
Challenge: training data imbalance (millions of background pixels, few objects). Solution: data augmentation, focal loss.
Temporal Analysis
Track object movement over weeks/months. Build dense optical flow to detect population movement, supply line logistics, military formations. Example: counting vehicles entering/leaving military bases weekly.
Much GEOINT analysis requires scraping location-based data: building permits, business registrations, infrastructure projects. Reliable proxies enable cross-region scraping of municipal databases.
Real-World Applications
Conflict Monitoring
Satellite imagery of Ukraine (2022-2026) tracked military movements, supply lines, damage. Bellingcat, Planet Labs, ESI (European Stability Initiative) publish weekly GEOINT reports.
Environmental Monitoring
Deforestation tracking (Amazon, Southeast Asia). NDVI drop = forest loss. Temporal data (Sentinel-2 archive since 2015) enables trend analysis.
Urban Planning
Track urban sprawl, construction. Estimate housing density, infrastructure gaps. Cities use GEOINT for planning.
Supply Chain Intelligence
Monitor port activity (ship traffic), refinery output (flare detection at night), semiconductor fab utilization (traffic patterns).
GEOINT Tools & Platforms (2026)
Open source: GDAL (geospatial raster/vector processing), Rasterio (Pythonic raster I/O), PostGIS (spatial database), QGIS (desktop GIS).
Cloud platforms: Google Earth Engine (free, petabyte-scale analysis), Sentinel Hub (Copernicus imagery API), Planet API (Planet Labs subscription).
ML frameworks: TensorFlow with geospatial datasets (SAT-4, SAT-6, Eurosat), PyTorch with satellite imagery augmentation.
Typical pipeline: Fetch tiles from Sentinel Hub → preprocess (normalize) → run CNN inference → postprocess (aggregate, spatial filter) → store in PostGIS → query/visualize in Mapbox.