Industries / Technology / Mapping and geospatial
Satellite imagery
This work turns pictures taken from orbit into maps and map data: it keeps the archives, corrects the pictures so they line up with the ground, sets how sharp a picture may be sold, and draws maps of land cover from them.
7
Milestones achieved, from 1984 to 2021.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1984
Landsat sold, users priced out
In 1984 the Land Remote Sensing Commercialization Act let the US privatize its land-imaging satellites, and NOAA, then in charge of Landsat, chose the Earth Observation Satellite Company (EOSAT), a joint venture of General Electric and Hughes Aircraft, to run Landsats 4 and 5 and sell their data. By NASA's account prices rose from $650 to as much as $4,400 a scene, many users turned to free, coarse pictures from weather satellites, and from 1984 to 1999 the satellites collected mostly where there was an obvious buyer, leaving gaps in the record. In a survey of 118 US academic geographers who had used Landsat, 83% of those who answered had bought no Landsat data after 1985 (Eisenbeis, 1992). Landsat 6, the program's only commercial launch, failed to reach orbit on 5 October 1993. In the Land Remote Sensing Policy Act of 1992 Congress found that "the cost of Landsat data has impeded the use of such data for scientific purposes" and that full commercialization "cannot be achieved within the foreseeable future", and set out federal management of Landsat data.
Lessonselling a public archive at a price cut its use, and the scenes that were never collected cannot be taken now.
Credit
US Congress; NOAA; EOSAT
1986
Satellites that look sideways
France's space agency, CNES, launched SPOT-1 on an Ariane 1 rocket on 22 February 1986. Its pictures covered squares 60 km on a side at a ground resolution of 10 m, by CNES's figures. Unlike Landsat, SPOT carried steerable mirrors that let it look to either side of its track, over a strip 950 km wide, and photograph the same ground from different angles on different passes. Two such pictures make a stereo pair, from which mapmakers measure relief and build digital terrain models, grids of ground height. NASA notes that SPOT broke the US monopoly on moderate-resolution land data.
Credit
CNES; Spot Image, which sold the data until 2015
1995
Spy photos opened to all
From 1960 to 1972 the US photo-intelligence satellites code-named CORONA, ARGON and LANYARD took more than 860,000 pictures of the Earth's surface and returned the film in capsules. The pictures were used for reconnaissance and to produce maps for US intelligence agencies. After a 1992 task force looked at their value for environmental studies, Executive Order 12951, signed by President Bill Clinton, declassified them in February 1995, and the film went to the US National Archives and the USGS. The University of Arkansas's Center for Advanced Spatial Technologies (CAST) says the last CORONA cameras, the KH-4B of 1967 to 1972, resolved about 1.8 m (6 ft) straight below the satellite and took stereo pairs, which makes them a detailed picture of the world as it was in the 1960s.
Credit
US government (Executive Order 12951); USGS EROS Center; US National Archives
2010
Archives moved next to computers
On 2 December 2010, at the UN climate conference in Cancún, Mexico, Google launched Earth Engine, which put current and historical satellite images online together with Google's computers to analyze them. It started with 25 years of Landsat data for most of the developing world, with the full global archive promised, and tools to remove clouds and haze. Google said the scale of the archives and the computing needed had meant "many of these images have never been seen, much less analyzed", and it donated 10 million CPU-hours a year for two years to help developing countries track their forests. Earth Engine is free for noncommercial use and sold as a Google Cloud product for commercial work.
Credit
Rebecca Moore and the Google Earth Engine team
2014
Sharper pictures for sale
Until 2014 DigitalGlobe, of Longmont, Colorado, could sell pictures sharper than 50 cm only to the US government. On 11 June 2014 it announced that the US government, through the Commerce Department, which licenses US imaging satellites, had approved sale of its sharpest pictures to all customers. For WorldView-3, then due to launch in August with a ground sampling distance (the width of ground one pixel covers) of 31 cm, black-and-white pictures at 25 cm could be sold six months after the satellite was declared operational. DigitalGlobe had argued that aerial photography, which faced no such limits, was taking its customers.
Credit
US Department of Commerce; DigitalGlobe
2015
Free 10 m pictures
The European Space Agency (ESA) launched Sentinel-2A on 23 June 2015 for the European Union's Copernicus program, followed by Sentinel-2B on 7 March 2017 and Sentinel-2C on 5 September 2024. Each satellite photographs a swath 290 km wide in 13 spectral bands, at up to 10 m a pixel, and two together cover all land every five days at the equator. Its first pictures showed individual buildings in Milan and field plots along the Po River. EU law gives everyone "free, full and open access" to the data.
Credit
ESA; European Commission
2021
A world land map, checked
A land-cover map shows what covers each patch of ground: trees, crops, buildings, water. In October 2021 ESA's WorldCover project released a free global land-cover map for 2020 at 10 m, made from Sentinel-1 radar and Sentinel-2 pictures, with 11 classes. It was independently validated at a global overall accuracy of about 75%. A 2021 version, released on 28 October 2022, scored 76.7%.
Credit
ESA; the WorldCover project consortium
Watch for
Ten-centimeter pictures from orbit
Albedo, a US company, launched Clarity-1 on 14 March 2025 to take visible pictures with 10 cm pixels from very low Earth orbit (VLEO), an altitude where thin air drags hard on a satellite and which Albedo says was long considered too harsh for commercial satellites. Albedo says the satellite proved it could hold a VLEO orbit and "validated 98%" of the technology for 10 cm pictures, but it did not reach that sharpness before Albedo lost contact with it about nine months into the mission.
Achieved whena satellite delivers pictures with 10 cm ground sampling and an outside party measures that resolution on a ground test target.
Claimed by
Albedo, in a post by Topher Haddad, January 2026
Landsat 10 by 2031
NASA says Landsat 10, with a "superspectral" set of instruments and finer resolution than earlier Landsats, will launch no later than 2031 and continue the record of land that began in 1972. It released the request for proposals to build the spacecraft on 14 July 2026 and expects to award the contract on 30 December 2026.
Achieved whenLandsat 10 is in orbit and its data are released at no cost.
Claimed by
NASA and the USGS, July 2026
Open
01
A 1960s base map
An open mosaic of 1960s CORONA KH-4B photographs for one large region outside the Middle East, orthorectified (corrected for camera angle and terrain so that each pixel sits where it belongs on a map), with every frame's position error measured at checkpoints against today's Sentinel-2 pictures, and the processing scripts published so the next region costs less.
The briefCredit · unclaimed02
Landsat's first decade, corrected
Surface reflectance, the share of sunlight the ground reflects once the haze of the air is removed, and a cloud mask for every usable Landsat Multispectral Scanner (MSS) scene of 1972 to 1992 over one continent, checked against the Thematic Mapper (TM) that flew beside the MSS on Landsats 4 and 5.
The briefCredit · unclaimed
Open 01 · the brief
A 1960s base map
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open mosaic of 1960s CORONA KH-4B photographs for one large region outside the Middle East, orthorectified (corrected for camera angle and terrain so that each pixel sits where it belongs on a map), with every frame's position error measured at checkpoints against today's Sentinel-2 pictures, and the processing scripts published so the next region costs less.
Why it is still open
The USGS holds more than 860,000 declassified pictures from 1960 to 1972. The main open collection of corrected CORONA pictures found, CAST's CORONA Atlas, lists 2,214 images, and its initial focus was the Middle East. Iacone, Allington and Engstrom (2022) georeferenced four CORONA images in Inner Mongolia, China, to an accuracy of 11.77 m and wrote that the existing literature "lacks the necessary detail required to replicate these processes easily". Hou and colleagues (2023) published 2OC, a method that oriented and orthorectified KH-4B frames automatically in places from the United States and Russia to Burkina Faso and western China, matched against Google Earth pictures; it is a method, not a published mosaic. NASA's Ames Stereo Pipeline, free software for building maps from satellite stereo pairs, calls its own KH-4B support "very experimental".
corona.cast.uark.eduapi.crossref.orgapi.crossref.orgstereopipeline.readthedocs.io
Built with what exists
KH-4B scans from USGS EarthExplorer; the Ames Stereo Pipeline's optical bar (panoramic) camera model; LightGlue, an open learned feature matcher from ETH Zurich (2023), to find the same points in an old photo and a Sentinel-2 picture; the Copernicus DEM for terrain; GDAL, the standard open library for raster maps.
First step
Download four overlapping KH-4B frames for a test area, run the Ames Stereo Pipeline's KH-4B workflow, and measure 30 checkpoints against a Sentinel-2 picture.
Done looks like
At least 50 KH-4B frames mosaicked over at least 50,000 km², with root mean square position error under 15 m at 200 checkpoints not used in the fitting, published with the code.
Open 02 · the brief
Landsat's first decade, corrected
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Surface reflectance, the share of sunlight the ground reflects once the haze of the air is removed, and a cloud mask for every usable Landsat Multispectral Scanner (MSS) scene of 1972 to 1992 over one continent, checked against the Thematic Mapper (TM) that flew beside the MSS on Landsats 4 and 5.
Why it is still open
The USGS's Collection 2 Level-2 products, its analysis-ready surface reflectance, cover Landsats 4 to 9 with the TM, ETM+ and OLI instruments, not the MSS, so the record from Landsat 1 in July 1972 to Landsat 4's TM in July 1982 has no official surface reflectance. Open pieces exist but do not join up: LandsatLinkr, free software whose guide lists MSS surface reflectance and an MSS-to-TM calibration among its outputs, works through one scene footprint at a time; msslib, an open Earth Engine library, computes top-of-atmosphere reflectance and the MSScvm cloud mask; and Lin Yan's code on GitHub registers MSS scenes to Landsat 8 and 9. Zhao, Wu, Qin and Ye (2022) built a framework for MSS surface reflectance over China and checked it against same-day TM scenes, but this research found no collection published from it.
usgs.govgithub.comgithub.comscience.nasa.govjdbcode.github.ioapi.crossref.org
Built with what exists
USGS Collection 2 MSS Level-1 scenes in Earth Engine; LandsatLinkr; msslib and MSScvm; Lin Yan's registration code; same-day Landsat 5 TM Level-2 scenes for calibration; Python.
First step
Pick 20 same-day MSS and TM scene pairs from Landsat 5 over one region, compute MSS top-of-atmosphere reflectance with msslib, and fit a band-by-band conversion to TM surface reflectance.
Done looks like
For one continent, a public MSS collection for 1972 to 1992 with, for at least 200 same-day MSS and TM scene pairs from Landsat 5, the mean difference and root mean square error per band in reflectance units, and the cloud mask scored against 1,000 hand-labeled points.