What is Satellite Imagery Data?
Remote-sensing satellite or aerial imagery, with multispectral bands and geographic annotation, used for land-cover classification, change detection, and agricultural monitoring.
Satellite imagery data is remote-sensing imagery with geographic and spectral annotation, used for land-cover classification, change detection and agricultural monitoring. Its distinguishing feature is that every image carries a location and a time.
The rights chain is longer than in any other category. The satellite operator, the distributor and the processor are typically different parties, and the terms each imposes have to be traced through to the buyer.
The technical specification is spectral and spatial: resolution, band count and time span. A single-date corpus and a multi-temporal one are different products, and the difference determines whether change detection is possible at all.
What buyers get wrong
The rights chain for remote sensing data is long. The rights holder, the distributor, and the processor of satellite imagery are all different parties, so provenance has to be traceable.
The specification detail that decides everything
Spatial resolution, band count, and time span (whether multi-temporal coverage is included).
This is the line item that most often separates a dataset that works from one that gets re-ordered. It belongs in the specification before production starts, not in the delivery review.
How we source it
We do not hold inventory in this category. A requirement comes in, we match it to a producer who can meet the specification, and we agree the quality bar with both sides before recording begins. A pilot batch goes first so misalignments surface early.
Scenarios this covers
Buyers rarely ask for a category by name. They describe a situation, and the situation is what tells us whether this category is the right one. If yours is in the list, name it in the brief and it goes straight into the specification. If it is not, describe it anyway, because it usually means the requirement sits in a neighbouring category.
What good Satellite Imagery Data looks like
Satellite data is judged on its rights chain and on the completeness of its spatial and temporal metadata.
-
Rights chain documented through each party
Operator, distributor and processor, with the terms each imposes, because the chain is where commercial use fails.
-
Spatial resolution and band count stated
These set what can be distinguished, and a corpus at one resolution cannot be resampled into another for classification purposes.
-
Acquisition dates and time span recorded per image
Change detection requires at least two dates, and the interval between them is a specification item.
-
Geographic metadata complete
Coordinates, projection and coordinate reference system, because imagery without a defined projection cannot be combined with other layers.
-
Cloud and quality masking applied and labelled
Optical imagery is frequently cloud-covered, and whether cloud masking was applied determines how much of the corpus is usable.
Public datasets in this category
These are openly documented resources, not products we sell. Each one covers part of the category, and knowing where each stops is usually what decides whether a project needs a collection of its own.
-
Sentinel-2
Free optical imagery from the European Copernicus programme, at ten to sixty metre resolution with a five-day revisit. Global, open, and the resolution limits what it can be used for.
-
Landsat
A decades-long optical archive at thirty metre resolution. The length of the record is unique and makes it the standard source for change detection over time.
-
SpaceNet
High-resolution imagery with building footprints and road labels, released for challenges. The labels are the valuable part and they cover selected areas rather than the globe.
-
BigEarthNet
Sentinel imagery with multi-label land-cover annotation across many classes. Useful for classification and it inherits the resolution ceiling of the Sentinel programme.
Questions to ask a supplier
Questions that establish the rights chain and the temporal coverage before any technical evaluation.
- Who are the parties in the rights chain, and what terms does each impose?
- What is the spatial resolution and the band count?
- What is the time span, and how many dates per location?
- What projection and coordinate reference system is used?
- Was cloud masking applied, and how is it labelled?
The vague answer in this category is about the sensor. A supplier who says the corpus contains satellite imagery without naming the sensors has described a family of products with different resolutions, bands and radiometry. Mixing them without labels produces a corpus that cannot train a consistent model.
A verifiable answer names the sensors, the acquisition dates, the cloud filtering method and the geolocation format. Ask for the per-scene cloud measure, because optical imagery over many regions is cloudy for much of the year, and a corpus without that measure contains unusable scenes mixed in.
Technical attributes
The items below are the ones a specification can hold a producer to, and the ones that decide whether a delivery is usable for training rather than only present.
| Rights chain | Operator, distributor and processor named, terms documented |
|---|---|
| Spatial resolution | Stated per source |
| Spectral bands | Count and wavelengths listed |
| Time span | Dates per location, and the interval |
| Projection | Coordinate reference system named |
| Delivery format | Imagery with per-scene metadata including date, location and quality mask |
The geolocation and projection column decides whether the corpus can be joined to anything, and it is the one most often stripped to save space. Imagery without coordinates or projection cannot be linked to labels, boundaries or other layers. The metadata is what makes the pixels useful.
The values come from the application. Change detection needs a long time series at consistent resolution, land cover classification needs labelled scenes, and object detection needs high resolution over small areas. The three pull toward different sensors and different dates, so the application has to be named first.
A mistake here produces a corpus that is unusable rather than merely imperfect. Where the projection was dropped, the imagery cannot be aligned with anything else, and re-projecting it requires the original metadata that no longer exists. This is the category where the metadata matters more than the pixels.
How Satellite Imagery Data is collected
For remote sensing, collection is licensing and metadata assembly rather than capture.
- Coverage defined against the use case. The area, the time span and the revisit frequency are derived from what the model has to detect.
- Rights chain assembly. Each party in the chain is identified and its terms reviewed, and the chain is documented as a chain rather than as a single license.
- Scene selection and quality filtering. Scenes are selected against a quality criterion, and cloud masking is applied and labelled so the usable extent is visible.
- Metadata assembly. Date, location, projection, resolution and band information are assembled per scene, because imagery without them cannot be combined with anything else.
- Validation and delivery. The geographic and temporal metadata is validated for consistency, and the per-scene metadata is delivered with the imagery.
After acquisition, scenes are filtered for cloud and quality, geolocation and projection are verified, and the files are packaged with per-scene metadata covering the sensor, the date and the measured cloud cover. The verification step is what makes the imagery joinable to other data.
The pilot for imagery is a small set of scenes over the areas the specification names, checked for geolocation, projection and cloud cover before the full acquisition. Its purpose is to confirm that the metadata survives the delivery, because imagery without coordinates or projection cannot be joined to anything and the loss is not recoverable from the pixels.
Acceptance is checked on the per-scene cloud measure and on the geolocation and projection fields, both against the specification. The specification should name the sensors and the date range before acquisition starts. Where the pilot shows the scenes cannot be joined, the delivery format is changed rather than the corpus accepted, since this is the category where the metadata matters more than the pixels.
How a delivery goes wrong
The defects below are the ones that recur across this category. Each is decided at collection or annotation time, which is why none of them can be repaired in delivery review.
-
Cloud cover was not filtered or reported
Optical imagery over many regions is cloudy for much of the year. A corpus without a per-scene cloud measure contains unusable scenes mixed in, and the buyer cannot filter them out.
-
Geolocation and projection were dropped
Imagery without coordinates or projection cannot be joined to anything else. Stripping the metadata makes the tiles unlinkable, and it is often done to reduce file size.
-
Sensors were mixed without labels
Different satellites have different spectral bands, resolutions and radiometry. A corpus that mixes them without a per-tile sensor label cannot be used to train a consistent model.
Why this matters to buyers
Land-cover classification, change detection and agricultural monitoring all depend on multi-temporal coverage, and a single-date corpus cannot support any of them. The time span is therefore the specification item that decides what the corpus is for.
The second driver is rights. Remote sensing data passes through more hands than most data types, and a corpus with a documented chain is the difference between imagery that can be used commercially and imagery that cannot be traced.
A model trained on mixed sensors without labels learns the sensor rather than the ground. It performs well on more imagery from the same satellites and poorly on imagery from others, and the limitation appears as soon as the product expands to a new source. The corpus cannot be filtered to fix this, because the sensor was never recorded.
The second cost is coverage. A corpus assembled without per-scene cloud measures contains scenes that carry no usable information, and the effective size is smaller than reported. Teams that plan compute against the reported figure find themselves short, and the shortfall is discovered during training rather than at delivery.
Imagery with verified geolocation and per-scene cloud measures gives a corpus whose effective size is known and whose scenes can be joined to labels and boundaries. The team avoids a manual alignment pass, and the compute plan is built on usable scenes rather than reported ones.
The second gain is that the corpus can take a second sensor. Where the sensor and the date are recorded per scene, imagery from a new satellite can be added without the model learning the sensor instead of the ground. That record is what lets coverage expand.
Specification at a glance
| Category | Satellite Imagery Data |
|---|---|
| Group | Non-speech data |
| Critical spec | Spatial resolution, band count, and time span (whether multi-temporal coverage is included). |
| Included by default | Pilot batch, metadata schema, source and consent documentation |
| Pricing | Quoted from specification. No published rates |
Related categories
-
LLM Training Data
Text corpora for LLM pretraining and fine-tuning: instruction data, preference data, and multi-turn dialogue data.
-
Computer Vision Data
Annotated image and video data, with annotation types including classification, bounding boxes, segmentation masks, and keypoints.
-
NLP Data
Annotated text for NLP tasks: named entities, dependency syntax, sentiment polarity, and relation extraction.
-
Multilingual Language Data
Parallel or non-parallel text corpora spanning multiple languages, used for multilingual models and low-resource language research.
-
Translation Data
Parallel corpora pairing source and target language, used for machine translation training and evaluation.
-
Dictionary Data
Structured dictionary entries with definitions, parts of speech, pronunciation, and example sentences, used for dictionary products and lexical understanding tasks.
Questions about Satellite Imagery Data
Why is the rights chain more complicated for satellite data?
Because the operator, the distributor and the processor are usually different organizations, and each imposes its own terms. The buyer needs the terms of all of them, and a single license statement does not cover the chain.
What is the minimum for change detection?
At least two dates for the same location, with the interval stated. The interval determines what kind of change is detectable, since seasonal variation and permanent change look different at different intervals.
How much of an optical corpus is unusable because of cloud?
It depends on the region and the season, which is why cloud masking is applied and labelled rather than estimated. The usable extent is a property of the corpus and has to be reported with it.
Can imagery at one resolution be resampled to another?
It can be resampled, and the resampling does not create detail that was not captured. For classification purposes a corpus at a coarser resolution cannot substitute for a finer one.
Do you have Satellite Imagery Data available now?
No. We do not hold inventory. Everything is produced against a specification. That means the timeline starts when the spec is agreed rather than immediately, and it also means the data matches your requirement instead of approximately matching something already built.
Can this be combined with other categories in one delivery?
Yes, and it is common. A single specification can cover several categories across several languages, with one pilot batch and one quality bar. It usually reduces total cost because contributor recruitment and project setup are shared.
What does the consent documentation cover?
Signed contributor consent stating the intended use, the collection methodology, and the annotation guideline the data was produced under. For projects involving EU data subjects the chain is built to satisfy GDPR, including the transfer mechanism. The terms for moving the data across a border.
Request Satellite Imagery Data
Tell us the language, the hours, and what the data needs to look like. You will get a real number and a real timeline, not a range. If we cannot source it well, we will tell you that instead.
- Pilot batch before the full run, so problems surface early.
- Consent documentation delivered with the data.
- No medical or clinical data. No recorded telephone calls.
- You send the spec Language, hours, recording conditions, and the deadline.
- We reply within two business days A real number and a real timeline, not a range.
- You approve a pilot batch A small run first, so problems surface before the full order.
- Full run, then delivery with documents Audio, transcripts, and the consent record for every speaker.