Satellite + Drone Fusion: TerraXplore's Mining Intelligence Platform

The Intelligence Layer for African Mining

Africa holds an estimated 30% of the world's mineral reserves, yet mining operations across the continent often rely on outdated survey methods, manual site inspections, and fragmented data. TerraXplore was built to change that -- providing mining companies, governments, and environmental agencies with a unified intelligence platform that fuses satellite imagery, drone surveys, and ground-truth sensors into actionable insights.

This post covers the technical architecture behind our multi-spectral imaging pipeline, 3D terrain modeling system, mineral deposit prediction models, and environmental monitoring capabilities.

Multi-Spectral Imaging Pipeline

The foundation of TerraXplore is our ability to ingest, process, and analyze imagery from multiple sources at multiple spectral bands.

Satellite Data Ingestion

We consume data from several satellite constellations:

• Sentinel-2 (ESA): 13 spectral bands at 10-60m resolution, 5-day revisit. Our primary source for broad-area mineral mapping and vegetation health monitoring.
• Landsat 8/9 (NASA/USGS): 11 bands at 15-100m resolution. Provides thermal infrared data essential for detecting subsurface heat signatures associated with certain mineral deposits.
• Commercial high-resolution: Planet SkySat and Maxar WorldView for sub-meter resolution when site-specific detail is needed.

Our ingestion pipeline processes approximately 4 terabytes of satellite imagery daily, automatically tiling, georeferencing, atmospheric-correcting, and storing it in a cloud-optimized GeoTIFF (COG) format for efficient random access.

class SatelliteIngestionPipeline:
    def __init__(self, config: PipelineConfig):
        self.sources = [Sentinel2Source(), Landsat9Source(), PlanetSource()]
        self.preprocessors = [
            AtmosphericCorrection(method="6S"),
            CloudMasking(model="cloud_unet_v3"),
            Georeferencing(crs="EPSG:4326"),
            TileGenerator(tile_size=512, overlap=64),
        ]
        self.storage = COGStorage(bucket="terraxplore-imagery")

    async def ingest(self, aoi: GeoJSON, date_range: DateRange):
        for source in self.sources:
            scenes = await source.search(aoi, date_range)
            for scene in scenes:
                raw = await source.download(scene)
                processed = self.run_preprocessors(raw)
                await self.storage.write(processed, metadata=scene.metadata)

Spectral Analysis for Mineral Detection

Different minerals reflect and absorb electromagnetic radiation at characteristic wavelengths. We exploit this with spectral analysis techniques:

• Band ratio indices: Simple ratios like iron oxide index (Red/Blue) and clay mineral index (SWIR1/SWIR2) provide fast, interpretable mineral indicators.
• Spectral angle mapping (SAM): Compares pixel spectra against a library of known mineral signatures, classifying each pixel by the mineral it most closely resembles.
• Deep spectral unmixing: A neural network that decomposes each pixel's spectrum into fractional abundances of constituent minerals, handling mixed pixels where multiple minerals coexist.

Our mineral signature library contains over 800 entries validated against laboratory spectroscopy of samples from mines across 22 African countries.

Drone Survey Integration

While satellite imagery provides broad coverage, drone surveys deliver the centimeter-level resolution needed for operational mining decisions.

Flight Planning and Execution

TerraXplore's drone module generates optimized flight plans based on:

• Terrain complexity: Areas with steep topography require more overlap between images to ensure complete 3D reconstruction.
• Target resolution: Users specify their desired ground sampling distance (GSD), and the system calculates optimal altitude and flight speed.
• Battery constraints: Multi-battery missions are automatically planned with safe return-to-home waypoints.
• Airspace compliance: The flight planner integrates with national aviation authority databases to flag restricted zones and generate required flight permits.

Photogrammetric Processing

Drone images are processed through our photogrammetric pipeline to produce:

• Orthomosaics: Geometrically corrected, seamlessly stitched aerial maps
• Digital Surface Models (DSM): High-resolution elevation data including structures and vegetation
• Digital Terrain Models (DTM): Bare-earth elevation with vegetation and structures removed
• 3D point clouds: Dense point clouds with millions of points per hectare, colorized from imagery

The processing pipeline uses a distributed structure-from-motion (SfM) algorithm that scales horizontally across our GPU cluster, processing a 500-hectare survey in under 2 hours.

3D Terrain Modeling

Mining operations need precise volumetric data. TerraXplore fuses satellite-derived elevation models with drone photogrammetry to produce multi-scale 3D terrain models.

Volumetric Analysis

Our volumetric tools allow mining engineers to:

• Calculate stockpile volumes by defining a base plane and computing the volume above it from the DSM. Accuracy is within 2% of ground-truth LIDAR measurements.
• Track excavation progress by differencing terrain models from successive surveys. The system automatically highlights areas of material removal and deposition.
• Estimate reserves by combining surface terrain models with subsurface geological models imported from drilling data.

interface VolumetricReport {
  surveyDate: string;
  areaOfInterest: GeoJSON;
  stockpiles: {
    id: string;
    location: [number, number];
    volumeCubicMeters: number;
    materialType: string;
    confidence: number;
  }[];
  excavationDelta: {
    periodStart: string;
    periodEnd: string;
    materialRemovedM3: number;
    materialDepositedM3: number;
    netChangeM3: number;
  };
}

Change Detection

By comparing terrain models over time, TerraXplore automatically detects and classifies changes:

• Authorized mining activity: Matches expected excavation patterns from submitted mine plans
• Unauthorized mining: Detects unplanned excavation that may indicate illegal artisanal mining
• Structural changes: New roads, buildings, or infrastructure appearing in the concession area
• Erosion and subsidence: Gradual terrain changes that may indicate geotechnical risk

Mineral Deposit Prediction

Beyond analyzing what is visible on the surface, TerraXplore uses machine learning to predict what lies beneath.

Prospectivity Mapping

Our prospectivity model integrates multiple data layers to produce probability maps for mineral deposits:

• Remote sensing features: Spectral indices, lineament maps (fault and fracture traces), and alteration zone classifications
• Geological maps: Digitized geological survey data including rock type, formation age, and structural geology
• Geophysical data: Magnetic and gravity anomaly surveys where available
• Geochemical samples: Stream sediment and soil sampling results from national geological surveys
• Known deposits: The locations and characteristics of known deposits serve as positive training examples

The model is a gradient-boosted ensemble trained on a continental-scale dataset of known mineral occurrences. For gold deposits specifically, our model achieves an AUC of 0.87 on held-out test regions, meaning it successfully identifies prospective areas that contain known deposits while maintaining a low false-positive rate.

Environmental Monitoring

Responsible mining requires continuous environmental monitoring. TerraXplore provides automated tools for:

• Vegetation health tracking: NDVI time series detect deforestation, vegetation stress from dust or chemical contamination, and reforestation progress in rehabilitated areas.
• Water quality indicators: Spectral analysis of water bodies detects turbidity, algal blooms, and sediment plumes that may indicate tailings dam seepage or runoff contamination.
• Tailings dam monitoring: Automated change detection on tailings storage facilities flags volumetric changes, seepage indicators, and structural anomalies. Given the catastrophic consequences of tailings dam failures, this feature alone justifies the platform for many clients.
• Carbon stock estimation: Using LIDAR-fused vegetation height models and allometric equations, we estimate above-ground carbon stocks for carbon offset reporting.

Regulatory Compliance Reporting

TerraXplore automatically generates environmental compliance reports aligned with:

• IFC Performance Standards for environmental and social sustainability
• Equator Principles for project finance environmental assessment
• National environmental regulations for the specific jurisdiction

Reports are generated quarterly, combining satellite-derived metrics with any ground-truth sensor data integrated into the platform.

Deployment and Scale

TerraXplore currently monitors over 180,000 square kilometers of mining concessions across 9 African countries. Our platform processes satellite data continuously and supports on-demand drone survey integration for over 200 active mining sites.

The combination of continental-scale satellite monitoring with site-level drone precision gives mining companies an intelligence layer they have never had before. And by making environmental monitoring automatic and continuous rather than periodic and manual, we are helping raise the bar for responsible mining across the continent.