The short answer
Geospatial intelligence, often shortened to GEOINT, is the analysis of satellite imagery and other location-based data to measure what is happening at specific places on Earth. For a business, that means answering physical questions from orbit: how far a construction project has progressed, how full a set of storage tanks is, how many vehicles are in a lot, whether crops are under stress, and what has changed at a site since the last pass.
Good geospatial intelligence software does more than show pictures. It turns imagery from several kinds of sensor into measurements, states the uncertainty of each one, and lets you trace every number back to the satellite pass it came from.
What geospatial intelligence means in business
The term comes from defense and government, where it describes the analysis of imagery and geographic information to understand physical features and activity on the ground. The same discipline now serves commercial questions, because more satellites, cheaper computing and better models have made it practical to watch specific places regularly.
In a business setting it answers questions that would otherwise need a site visit, a phone call or a company's own disclosure:
- Operations and supply chains. Whether suppliers' plants, mines, ports and warehouses are running, growing or slowing.
- Real estate and infrastructure. Construction progress, new development and changes in land use.
- Commodities and energy. Storage levels, stockpiles, plant activity and shipping.
- Insurance and risk. Exposure before an event and damage after it.
- Agriculture and environment. Crop condition, vegetation stress, water and land change.
The common thread is that the answer is a measurement of a place, repeated over time, rather than a single image.
The sensors and what each one reveals
No single sensor answers every question. The main types complement each other:
| Sensor | What it reveals | Limits |
|---|---|---|
| Optical | What a person would see from above: buildings, vehicles, ships, construction and land cover. Extra bands beyond visible light add vegetation health and water. | Needs daylight and a clear view; cloud, haze and smoke block it. |
| Synthetic-aperture radar (SAR) | Structure, surface change, ships and flooding, and with repeated passes, ground movement as small as millimeters. It sends its own microwave pulses and records the echo, so it sees at night and through cloud. | Hard to read without training; grainy texture and geometric distortion; heavy rain can affect some wavelengths. |
| Thermal | Heat: whether a plant, furnace or engine is running, fires, and warm water discharged into rivers or the sea. Works at night. | Usually much coarser than optical from orbit; measures surface temperature only; blocked by cloud. |
| Hyperspectral | What things are made of, from tens to hundreds of narrow bands: minerals, plant species and health, some gases such as methane, water quality. | Coarser detail, very large data volumes, fewer satellites, and careful correction for the atmosphere. |
Two practical points follow. Revisit matters as much as resolution: a sharp image once a month misses what a coarser image every day would catch. And combining sensors closes the gaps: radar fills in when cloud blocks optical, and thermal shows whether a site that looks unchanged is actually running.
Typical business questions
- Construction progress. Ground cleared, foundations poured, structure rising, roof closed. Measured over successive passes, progress becomes a timeline that can be set against the schedule a developer or contractor has published.
- Storage levels. Many oil tanks have floating roofs that sink as the tank empties, and the shadow the tank wall casts on the roof shows how full it is. Stockpiles of coal, ore or grain can be measured as volumes from elevation data.
- Vehicle counts. Cars in a retail or factory lot, trucks at a distribution center, ships at berth or at anchor. Counts over time show activity rising or falling.
- Vegetation stress. Healthy plants reflect strongly in near-infrared light and stressed plants reflect less, which shows crop condition across whole regions and vegetation encroaching on power lines or pipelines.
- Structural change. New buildings, demolitions and expansions, damage after a storm or flood, and ground subsidence under assets.
- Thermal output. Whether an industrial plant is running, and roughly how hard, from the heat it gives off.
Why provenance and validation matter
A number from orbit is the end of a long chain: a satellite, a pass on a particular date and time, processing that corrects for the atmosphere and the viewing angle, a model that turns pixels into a count or a volume, and a version of that model. Any link in the chain can shift the answer. Cloud can be mistaken for change, shadows for vehicles, and a pass at a steep angle can make a tall structure appear to lean across its neighbors.
That is why every figure should be traceable to its inputs: which sensor, which acquisition dates, which processing and which model, with a stated uncertainty. When a number drives an investment or a compliance decision, someone will ask where it came from, and a black box cannot answer.
Validation is the second half. Before a layer is used in earnest, it should be tested against an independent reference that played no part in building it: published statistics, reported production, ground surveys or measurements taken on site. A model that has never been checked against something outside itself is a hypothesis, not a measurement.
How to start
- Define the ground. Draw the exact boundaries of the sites or region you care about, and fix them, so every measurement over time covers the same area in the same way.
- Define the question. Decide which number would change a decision: activity at a plant, months to completion, the share of storage in use. A precise question picks the method.
- Pick the layers. Choose sensors for the question and the place: radar where cloud or darkness would blind optical, thermal for whether a plant is running, hyperspectral for materials, optical for most counting and construction work. Set the revisit you need.
- Build a history. Pull past imagery to learn the normal pattern, including the seasons, before reading meaning into a change.
- Validate against an independent reference. Check each layer against figures you can verify elsewhere before relying on it, and record how close it comes.
- Put it into the workflow. Decide who receives the readings, how often, and what size of change calls for a closer look.
Pitfalls
- Pictures instead of measurements. A striking image of a site is not evidence of a trend. Measure the same thing, the same way, over time.
- Gaps that bias the series. If cloud hides a site every winter, an optical-only time series quietly loses the winter months.
- Comparing unlike images. Different resolutions, viewing angles, seasons and sun angles make change appear where there is none.
- Ignoring uncertainty. A count with a wide margin of error is a different statement from an exact count. Tools that drop the margin invite overconfidence.
- Revealing your interest. The places you watch say a great deal about your plans. Sending your sites and boundaries to an outside platform shares that interest with whoever runs it.
- Treating experimental models as finished. A layer that has not passed validation should be labeled experimental and used as such.
For how investors put these measurements to work, see the guide to satellite data for investors.
How MITHRIL does it
MITHRIL: any coordinate on Earth, every number with a receipt
MITHRIL is a geospatial intelligence workstation. Draw a boundary over any point on the planet, whether a site you run, one a competitor runs or ground no one on your team has stood on, and that boundary locks the extent, the projection and the compute grid for everything that follows. Every layer reads the same ground the same way, so the numbers compare exactly, not approximately.
It carries a library of proprietary analytical layers, each a cited, validated model: construction progress, vehicle counts, storage levels, vegetation stress, thermal output and structural change since the last pass. Switch on one or forty; each declares what it depends on and carries that record into the result. Click any point to read the value, its unit, its uncertainty and the chain back through inputs and acquisition dates to the original pass. Where that record cannot be produced, the value is not drawn.
Imagery comes from optical, synthetic-aperture radar, thermal and hyperspectral sensors on the commercial and civil constellations in orbit today, with revisit measured in hours over the ground you care about. Every layer is benchmarked against an independent public reference before release; one reproduces an external figure to within 0.24 percent against a 5 percent tolerance, and anything short of the bar ships badged experimental. The workstation runs inside your own entity, so your ground and your decisions are never handed to anyone else's model.
Questions
What is geospatial intelligence?
Geospatial intelligence, or GEOINT, is the analysis of satellite imagery and other location-based data to measure and understand physical features and activity on Earth. In business it answers questions such as construction progress, storage levels, vehicle counts and whether a site is running.
What does geospatial intelligence software do?
It turns imagery from different sensors into measurements over places you define and tracks them over time. Good software also shows the uncertainty and the source of every number, so each result can be checked.
Can satellites see through clouds?
Optical satellites cannot. Synthetic-aperture radar can, because it sends its own microwave pulses, which pass through cloud and need no sunlight, so it works at night and in bad weather.
What is the difference between GEOINT and satellite imagery analysis?
Satellite imagery analysis is one part of geospatial intelligence. GEOINT combines imagery from several sensors with maps, location data and other sources to answer a question about a place, usually as a measurement tracked over time.
How accurate is satellite imagery analysis?
It depends on the sensor, the resolution, the conditions during the pass and the model used. The reliable approach is to validate each measure against an independent reference and to read every value together with its stated uncertainty.
