Earth Observation Research

Earth Observation for Trading

Satellite data, commodity prices, and the information advantage.

What Earth Observation Data Is

Earth observation (EO) encompasses all data acquired by sensors on orbiting platforms. The modalities differ in what they measure, when they can measure it, and what they reveal.

ModalitySensors / MissionsResolutionKey Property
Optical (multispectral)Sentinel-2, Landsat 8/9, Planet SuperDove3–30 mReflectance in visible + NIR bands; cloud-dependent
Synthetic Aperture Radar (SAR)Sentinel-1, ICEYE, Capella Space1–20 mMicrowave backscatter; day/night, cloud-penetrating
Thermal infraredLandsat TIRS, ECOSTRESS, VIIRS60–375 mSurface temperature; detects industrial activity, fires
HyperspectralPRISMA, EnMAP, Pixxel Fireflies5–30 m100+ narrow bands; mineral/crop species identification
AIS (ship tracking)Spire, exactEarth, OrbcommShip-levelMaritime vessel identity, course, speed, draft

How Satellite Data Informs Commodity Prices

The connection between orbital imagery and price discovery is direct. Every physical commodity has a spatial footprint that satellites can measure before the market knows.

Crop Yield Estimation

NDVI (Normalized Difference Vegetation Index) time series from Sentinel-2 track photosynthetic activity across growing seasons. A fund monitoring NDVI for the U.S. corn belt at 10m resolution sees drought stress 4–6 weeks before USDA crop reports. The signal is in the deviation from the historical NDVI curve for a given county and growth stage.

Oil Storage Monitoring

Floating-roof crude oil tanks cast shadows proportional to the volume stored. SAR imagery measures these shadows regardless of cloud cover. Orbital Insight and Kayrros process imagery of Cushing, Oklahoma and the SPR to estimate U.S. crude inventories before EIA reports. The edge: weekly global coverage vs. monthly national reports.

Shipping and Maritime Intelligence

AIS data from satellite constellations (Spire operates 100+ nanosats) tracks every commercial vessel globally. Combine with SAR for dark-vessel detection (ships that disable AIS transponders). Applications: crude oil tanker tracking for supply flow estimation, iron ore fleet monitoring, grain shipment routing from Black Sea ports.

Mining Output and Industrial Activity

Change detection on mine sites reveals expansion/contraction of operations. Thermal imagery detects smelter activity. Nighttime light intensity (VIIRS) correlates with industrial output. Chinese factory activity was tracked via NO2 emissions (Sentinel-5P) during COVID lockdowns before official PMI data.

Deforestation and Carbon Credits

Forest cover change is the basis for REDD+ carbon credits. Monitoring deforestation with SAR (cloud-penetrating, critical in tropics) allows verification of carbon credit validity and detection of fraudulent offsets. The voluntary carbon market trades ~$2B/year, and satellite verification is becoming mandatory.

Hedge Fund and Private Equity Usage

FirmApproachEdge
Two SigmaIn-house satellite data team; automated ingestion pipelinesProcessing speed, cross-signal integration
CitadelSatellite + alternative data divisionScale of data fusion across modalities
Point72Cubist Systematic (quant arm) uses EO dataSystematic signal extraction, backtesting infrastructure
Man GroupAHL quant fund integrates satellite-derived featuresTime-series modeling of physical signals

The alpha is not in access. Sentinel data is free. Planet offers daily global coverage at commercial rates. The alpha is in processing speed and domain-specific interpretation. A fund that can go from raw imagery to a tradeable signal in hours has an edge over one that takes days. A team that understands the agronomic meaning of a NDVI anomaly has an edge over one that treats it as a generic feature.

Private Equity Due Diligence

Satellite data verifies claims that management cannot obscure:

Connection to Embedding Research Satellite imagery features are high-dimensional and domain-specific. The spectral signature of wheat stress (low NDVI in band 8, elevated reflectance in band 4) occupies different embedding dimensions than corn stress (different chlorophyll absorption profile). PCA compression to 16 dimensions would collapse these into a single "crop stress" cluster, destroying the distinction that determines whether you trade wheat futures or corn futures. This is the same phenomenon documented in our dimensionality research: the fine-grained features that matter live in the long tail of variance.

Technical Pipeline

The end-to-end flow from satellite to trading signal follows a consistent architecture:

StageProcessTools
1. IngestDownload imagery from provider APIs; manage tile cataloguesSTAC API, Planet SDK, Copernicus Data Space
2. PreprocessAtmospheric correction, orthorectification, cloud maskingSen2Cor, ACOLITE, s2cloudless
3. Feature extractionCompute indices (NDVI, NDWI, BSI), segment objects, detect changesrasterio, scikit-image, custom CNN/ViT
4. EmbedGenerate vector representations of image patches or time seriesSatMAE, Prithvi, domain-specific encoders
5. Vector searchFind similar historical patterns; anomaly detection against baselineQdrant, FAISS, pgvector
6. Signal generationConvert anomalies to directional signals with confidence scoresCustom models, ensemble methods
7. BacktestValidate signal against historical price data; measure Sharpe, drawdownzipline, vectorbt, custom frameworks

Key Providers

ProviderCoverageSpecialty
Planet LabsDaily global at 3mHighest temporal cadence; SuperDove constellation
Maxar30cm resolutionHighest spatial resolution commercial optical
Spire GlobalGlobal AIS + weatherMaritime tracking, GNSS radio occultation
Orbital InsightAnalytics platformOil storage, economic activity indices
Descartes LabsAnalytics platformCrop forecasting, supply chain monitoring
RS MetricsAnalytics platformRetail foot traffic, metal inventory monitoring
KayrrosAnalytics platformEnergy market intelligence, methane monitoring
ICEYESAR constellationSub-daily SAR revisit; flood, infrastructure monitoring

Related Research

The Dimensionality Illusion
Why PCA compression to 16 dimensions destroys domain-specific features. The same principle applies to satellite imagery embeddings.
Embedding Research
GPS/PNT Authentication
Location as evidence, timing as proof. How authenticated positioning prevents fraud in financial transactions and legal proceedings.
Earth Observation
Space-Based Data Processing
Edge compute on orbit, bandwidth constraints, and why embedding dimensionality is a design parameter for satellite ML systems.
Earth Observation