Skip to main content

Satellite & Geospatial Data

Updated Jun 24, 2026 at 2:35pm

Research Draft High 1,256 words

Satellite and geospatial data is an alternative-data category in which investors buy commercial Earth-observation imagery — and the analytics derived from it — to measure real-world economic activity directly, ahead of official reports or company disclosures. The canonical use cases are counting cars in retailers' parking lots, gauging crude-oil inventory from the shadows cast by floating storage-tank roofs, and tracking crop health, shipping, mining, and construction. The core tension is structural: the signal is genuinely informative about the physical economy, but the edge is expensive, perishable, and concentrated in a few large sophisticated buyers — which makes it as much a story about informational inequality as about alpha.

How it's formed

Three layers turn pixels into a tradable signal:

  • Capture. Constellations of imaging satellites photograph the Earth on a revisit cadence. Planet Labs operates a large fleet of small "Dove" cubesats giving near-daily medium-resolution coverage; Maxar provides high-resolution (sub-meter) optical imagery; synthetic-aperture radar (SAR) operators (e.g. Capella, ICEYE) image through cloud and at night, which matters for assets like oil tanks where optical imagery fails under cloud cover.
  • Computer vision. Object-detection models count cars, ships, or tanks; for oil, analysts estimate the fill level of a floating-roof tank from the length of the shadow the roof casts inside the tank wall. Vendors such as Orbital Insight, RS Metrics, Kayrros, and Ursa Space Systems sell the processed counts/indices rather than raw images.
  • Normalization and mapping to a security. Raw counts must be adjusted for sun angle, weather, holidays, store openings/closings, and base-rate seasonality, then aggregated across a company's geolocated store footprint into a year-over-year change that proxies a reported metric (same-store sales, throughput, inventory).

A critical, non-obvious constraint runs through all of this: when key sites are under cloud, optical satellites cannot "see," so the data simply goes missing — a property researchers have exploited as a natural experiment (clear vs. cloudy weeks) to measure the signal's price impact.

How it's used in practice

The earliest commercial use traces to RS Metrics, which began selling satellite parking-lot car counts to hedge funds in the early 2010s; UBS analyst Neil Currie famously incorporated such counts into Walmart research around 2010 (Berkeley Haas). The dominant pattern is nowcasting a number the market is waiting for: estimate a retailer's quarter before the earnings print, or estimate U.S. crude inventories before the weekly EIA report, then position ahead of the release and trade the reaction. In commodities the same logic extends to global oil afloat, agricultural yields ahead of USDA reports, and Chinese industrial activity. Buyers are predominantly systematic and discretionary hedge funds, commodity trading houses, and large asset managers; the data is one input blended with fundamentals and other alt-data, not a standalone system.

Adoption, debate & evidence

This is one of the better-evidenced alt-data categories, with peer-reviewed academic support — which separates it from much alt-data folklore.

  • Retail / parking lots. Zhu (2019) and the Berkeley team (Katona, Painter, Patatoukas, Zeng) studied roughly 4.8 million RS Metrics image observations covering ~67,000 unique store locations across 44 major U.S. retailers, 2011–2017. They confirm year-over-year car-count changes reliably predict reported sales, and that traders with access earned on the order of 4–5% in the three days around earnings announcements — a real, documented edge (Berkeley Haas).
  • The distributional finding. The same research frames this as a transfer: gains accrue to large investors who can afford the data, partly at the expense of retail investors who are net buyers of the very names hedge funds short ahead of bad prints. The authors explicitly expect the edge to erode as the technology democratizes.
  • Government macro data. Mukherjee, Panayotov & Shon, "Eye in the Sky" (Journal of Financial Economics, 2021), use the cloudy-vs-clear instrument for U.S. crude and Chinese manufacturing and find satellite estimates now reduce the price surprise of official announcements — markets are sometimes no longer surprised because the satellite already told them (ScienceDirect).
  • Oil and weather. A Humanities & Social Sciences Communications (Nature, 2023) study finds cloud cover over U.S. storage hubs predicts oil returns — higher cloudiness (less satellite visibility) one week is followed by lower oil returns the next, indirect evidence the imagery is being priced (Nature).

The contested part is durability and cost-effectiveness, not whether a raw signal exists. The market is large (commonly cited at hundreds of millions of dollars annually for finance) but feeds are expensive — industry estimates put meaningful high-resolution feeds in the tens to hundreds of thousands of dollars per year, and comprehensive monitoring far higher. As coverage and buyers proliferate, the documented edge is widely expected to compress; magnitudes vary by study and many vendor-cited backtests are not independently verified.

Strengths & limitations

Strengths. Direct, near-real-time, physical-world measurement that front-runs both lagging official statistics (EIA monthly, USDA annual) and quarterly corporate disclosure. SAR overcomes the cloud/night gap. The signal is causally grounded — cars and oil are not sentiment.

Limitations and the #1 misuse. The signal measures only the physical slice of a business — e-commerce, services, pricing, mix, and margins are invisible, so parking counts can be right about footfall and wrong about earnings. Cloud cover, sensor calibration, store openings, and footprint coverage inject noise; survivorship and look-ahead bias plague vendor backtests. The single biggest misuse is treating a car-count beat as an automatic earnings/stock-up signal — ignoring that the number may already be priced (the "Eye in the Sky" result), that the read covers only part of revenue, and that data and trade-execution costs can consume a modest, decaying edge. It is an input, not a thesis.

Sources

Dispute flags: return magnitudes (4–5% earnings-window) come from a specific 2011–2017 retailer sample and may not generalize or persist; cost figures are industry estimates, not audited; vendor backtests (long/short alpha claims) are not independently verified and likely overstate live, decay-adjusted returns.