SourcesThe data family
satellitedata.net
A free map and API for Sentinel-2 and Landsat satellite imagery: search scenes for a place and date, view them as map tiles, and download an area as a GeoTIFF.
What it is
satellitedata.net gives you one way into two big public archives of satellite images: Sentinel-2, the optical mission of Europe’s Copernicus programme, documented by the European Space Agency (ESA), and Landsat 8 and 9, processed by the US Geological Survey (USGS). You can search for the scenes over a place and time, view them on a map in true colour, false colour or a vegetation or water index, compare two dates, and download an area as a GeoTIFF. It reads the files from the public copies of both archives on Amazon Web Services and finds scenes through Element 84’s Earth Search catalog.
It’s related to the data .dev family but works differently. The .dev sites are keyless downloads of tables. This one renders images on demand, and most of it needs a free API key: you fill in a short form, confirm your email, and the key arrives by email. A demo on the map needs no key. Like this site, it’s run by Eric Ness.
What it measures
Four collections are available now:
| Collection | What it is | Pixel size |
|---|---|---|
| Sentinel-2 L2A | Surface reflectance, corrected for the atmosphere | 10 to 60 m |
| Sentinel-2 L1C | Reflectance at the top of the atmosphere | 10 to 60 m |
| Landsat 8–9 Level-2 | Surface reflectance and surface temperature | 30 m |
| Landsat 8–9 Level-1 | Top-of-atmosphere reflectance, plus a 15 m panchromatic band | 15 to 30 m |
Sentinel-2 scenes start in June 2015 and Landsat 8–9 scenes in March 2013. Each scene keeps its native id, such as S2A_T31TCF_20260829T105044_L2A, so you can find the same scene in the agencies’ own archives.
Processing levels matter more than they sound. Level-1 is what the sensor saw from orbit, haze included. Level-2 is an estimate of what the ground reflects, after a model removes the atmosphere. For comparing places or dates, Level-2 is usually what you want. Some bands exist only at Level-1: Sentinel-2’s cirrus band and Landsat’s panchromatic band.
What the site adds: one set of band names across both missions (so nir08 means the same near-infrared band on each), ten ready-made indices such as NDVI for vegetation and NDWI for water, charts of an index over a field through time, and email alerts when a new clear image of an area arrives. It also documents each mission’s scale and offset, which you need to turn stored numbers into reflectance.
What it misses
Clouds. These are optical sensors: they see reflected sunlight, so they can’t see through cloud. Each scene has a cloud cover percentage, but the docs warn that it’s for the whole scene, so a scene at 5 percent cloud can still be cloudy right over your area. For that you need a pixel mask, and the current version of the API doesn’t apply one for you; you download each mission’s own classification layer and mask the pixels yourself. Sentinel-2’s layer has no buffer around clouds, so their edges can leak through.
Revisit. A satellite only sees a place when it passes over. ESA gives a 5-day revisit at the equator with two Sentinel-2 satellites, more often at higher latitudes. Landsat 8 and 9 each pass every 16 days, 8 days together. Add clouds, and the gap between usable images can be much longer. Anything that happens and clears up between passes, such as a short flood, may never appear.
Resolution. Each pixel is one value for a square of ground: 10 m on a side for Sentinel-2’s sharpest bands, 20 or 60 m for the rest, and 30 m for Landsat. Anything smaller than a pixel blends with its surroundings.
Level-2 isn’t always made. USGS only makes Landsat Level-2 products for scenes with the sun high enough (a solar zenith angle under 76 degrees) and with the extra data the corrections need. Some scenes get surface reflectance but no surface temperature.
Gaps in the copy it reads. The site serves Sentinel-2 Collection 1, ESA’s reprocessing of the archive, as Earth Search indexes it. In its own check over two small areas of Indonesia in October 2026, many scenes from 2017 to 2019 and 2022 were missing compared with the earlier index. It says these gaps may be smaller elsewhere, and advises counting what a search returns before you rely on a period. Values for scenes before 2022 can differ from the first-published products by up to about 0.02 reflectance.
Limits. The free key allows 20,000 requests a month and 20 downloads a day. A map view can use 20 to 60 requests. Synchronous downloads are capped at 2,500 pixels a side. Unlike the .dev sites, the data has no version ids: the API is versioned, but the images come straight from the archives.
How to access it
1Try the map
Open map.satellitedata.net. The demo runs without a key. With one, you get every collection, downloads, fields and watched areas.
2Check the service and the collections
These calls need no key:
curl -s https://api.satellitedata.net/v1/health
curl -s https://api.satellitedata.net/v1/collections
When we tested them on 14 November 2026, the health check returned "status":"ok", and the collection list returned the four collections with their start dates and credit lines. Looking up one scene also works without a key:
curl -s https://api.satellitedata.net/v1/collections/sentinel-2-l2a/items/S2A_T31TCF_20260829T105044_L2A
That returned the 29 August 2026 Sentinel-2A scene, with a cloud cover of about 0.6 percent and processing baseline 05.12. Its asset links point at the public image files, which you can read directly.
3Get a key and search
Request a free key. Send it in the X-API-Key header, never in the URL. Searching needs it; without it, our test search returned 401. The docs show a search as a POST to https://api.satellitedata.net/v1/search with a collection, a bounding box, a date range and a cloud filter such as "eo:cloud_cover": {"lte": 20}. The answer is a list of scenes in the STAC format, which tools like pystac-client can read.
4Download an area
POST /v1/process cuts an area out of the scenes in a time range and returns a GeoTIFF of the bands you ask for. The quickstart has a worked example, and larger areas go through background jobs.
5Read the collection pages
Each collection has a page in the docs with its bands, scale and offset, cloud layers and known gaps. Read the one for your collection before you analyse anything.
How to cite it
The site doesn’t publish versioned copies of the images, so cite the agency’s scene and the archive, not a satellitedata.net URL. Give the mission, the processing level, the scene id and the date, and use the credit line the site gives: “Contains modified Copernicus Sentinel data” plus the year for Sentinel-2, and “Landsat data courtesy of the U.S. Geological Survey” for Landsat. USGS also asks you to cite the Landsat Level-2 dataset by its DOI. For example:
Contains modified Copernicus Sentinel data 2026. Sentinel-2A MSI Level-2A, Collection 1, scene S2A_T31TCF_20260829T105044_L2A (29 August 2026). Accessed through satellitedata.net, 14 November 2026.
If your result depends on settings, such as the cloud threshold, the index or the bands, say what they were, so someone else can get the same pixels.
Sources & further reading
- satellitedata.net
- satellitedata.net docs: Sentinel-2 L2A
- satellitedata.net docs: Landsat 8-9 Level-2
- satellitedata.net docs: cloud mask
- satellitedata.net docs: limits and quotas
- ESA SentiWiki: the Sentinel-2 mission
- ESA SentiWiki: Sentinel-2 processing levels
- USGS: Landsat Collection 2 Level-2 science products
- Element 84: Earth Search
End of page · Back to top