Satellite Imagery Analytics

The size of the smallest discernible object on the ground, measured in meters per pixel. High-resolution sensors (30cm–50cm) can identify individual vehicles and lane markings, while medium-resolution (10m–30m) is sufficient for crop health monitoring. This metric directly determines the granularity of the economic signal—counting cars in a retailer's parking lot requires sub-meter resolution, whereas estimating a country's wheat yield can be achieved with coarser data.

The specific ranges of the electromagnetic spectrum captured by the sensor beyond standard RGB. Key bands include:

• Near-Infrared (NIR): Essential for calculating NDVI vegetation indices to assess crop biomass and health.
• Short-Wave Infrared (SWIR): Penetrates haze and differentiates between mineral types, useful for monitoring mining activity.
• Synthetic Aperture Radar (SAR): An active sensor that emits its own signal, allowing it to image through cloud cover and at night—critical for tracking maritime vessel traffic and oil storage tank levels regardless of weather.

The frequency at which a satellite can capture an image of the same location. High-cadence constellations like PlanetScope offer daily revisits, enabling the detection of rapid changes such as construction progress or weekly retail foot traffic trends. Low-cadence systems with 7–16 day revisits are suitable for slower-moving indicators like deforestation or seasonal agricultural cycles. The revisit rate defines the maximum frequency of the derived trading signal.

The process of correcting raw satellite imagery for terrain distortion, sensor angle, and Earth curvature to align it precisely with map coordinates. Sub-pixel accuracy ensures that a detected object at coordinates (lat, lon) corresponds to the exact real-world location. This is non-negotiable for overlaying imagery with other geospatial datasets—such as property boundaries or pipeline routes—and for performing accurate time-series analysis of a specific asset.

The algorithmic removal of atmospheric interference—aerosols, water vapor, and haze—that distorts the spectral signature of ground objects. Without this preprocessing step, the same soybean field would appear to have different health characteristics on different days. Bottom-of-atmosphere (BOA) reflectance normalizes imagery to a consistent physical baseline, making quantitative indices like NDVI comparable across time and across different satellite sensors.

The computer vision models that transform pixels into countable, structured data. Modern pipelines use convolutional neural networks (CNNs) and vision transformers (ViTs) to identify and enumerate specific assets:

• Car detection: Counting vehicles in retailer parking lots to estimate foot traffic and quarterly revenue.
• Ship detection: Identifying vessel type and draft depth from SAR imagery to infer cargo weight.
• Oil tank measurement: Using shadow geometry from floating-roof tanks to calculate fill levels and global crude supply.

Non-traditional datasets sourced from outside standard financial filings and market data. Satellite imagery is a premier example, providing objective, real-world observations of economic activity—such as counting cars in retailer parking lots or measuring oil tank shadows—to generate alpha before traditional metrics are reported. Key characteristics: uncorrelated with market data, high acquisition cost, requires specialized engineering.

A historical data snapshot preserving the exact state of a dataset as it was known on a specific past date. For satellite analytics, this means storing the imagery and derived metrics exactly as they existed before cloud cover corrections or reprocessing. Critical for backtesting: using a later, cleaned version of an image introduces look-ahead bias, making a strategy appear prescient when it was not.

The prediction of the present or very near future state of an economic indicator using high-frequency, real-time data sources before official statistics are released. Satellite imagery enables nowcasting of:

• Retail sales: via parking lot fullness indices
• GDP components: via nightlight luminosity and factory activity
• Agricultural yields: via NDVI vegetation health indices
• Oil inventories: via floating-roof tank shadow measurements

The computational process of identifying and merging disparate records that refer to the same real-world entity across multiple datasets. In satellite analytics, this involves geocoding a corporate headquarters address, matching it to a specific building polygon in an image, and linking that to a ticker symbol. Without precise entity resolution, a parking lot count cannot be attributed to the correct security.

The end-to-end tracking of data's origin, transformations, and movement through pipelines. For satellite imagery, lineage captures:

• Sensor provenance: which satellite and instrument captured the raw image
• Processing steps: atmospheric correction, orthorectification, cloud masking
• Model versioning: which neural network generated the car count or crop mask This auditable map is essential for regulatory compliance and debugging signal degradation.

The gradual erosion of a trading signal's predictive power over time as the market adapts to the inefficiency. A satellite-derived parking lot count may generate alpha for months, but as more funds acquire the same imagery and build similar models, the edge compresses. Mitigation strategies include increasing observation frequency, combining with orthogonal datasets, and moving to less-crowded geographies or asset classes.