Industries / Technology / Mapping and geospatial
Aerial photos and elevation
This work photographs and measures the ground from aircraft and from space and turns the results into orthophotos, pictures corrected to line up with a map, and elevation models, grids that give the height of the ground or of whatever stands on it.
7
Milestones achieved, from 1935 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1935
Farms photographed from the air
The use of aerial photography in US farm programs dates from 1935, by the account of the Farm Service Agency's Aerial Photography Field Office, after the Depression and the Dust Bowl brought federal programs to help farmers. County offices drew field boundaries by hand on paper enlargements of the photos and checked each year whether farmers had kept to program terms by projecting 35 mm slides onto them. The USGS now offers over 6.5 million frames flown by federal agencies from 1937 on as scans for download.
Credit
USDA farm agencies; USGS
2000
Radar maps most land
In February 2000 the Shuttle Radar Topography Mission (SRTM) used an imaging radar aboard Space Shuttle Endeavour, over an 11-day flight, to map the height of the land between 56 degrees south and 60 degrees north, about 80% of Earth's land by NASA's figure. For 14 years its most detailed version, with heights every 30 m, was public only for the United States, and the rest of the world had 90 m data. On 23 September 2014 the White House announced that the 30 m data would be released worldwide over the following year, starting with Africa.
Lessona global survey can wait years to be used at full detail when its release is restricted.
Credit
NASA's Jet Propulsion Laboratory; the US National Geospatial-Intelligence Agency; the German and Italian space agencies
2003
Farm photos every summer
In 2003 the US Department of Agriculture began the National Agriculture Imagery Program (NAIP), which photographs the conterminous United States from aircraft during the growing season, when crops are in leaf, and delivers orthorectified pictures. Pixels were 1 m from 2003 to 2017 and 0.6 m from 2018, and the repeat cycle went from five years to no more than three from 2009. From 2006 its contracts required that well-defined test points fall within 6 m of their true position at 95% confidence, a rule all states met from 2009. NAIP replaced the slides and paper prints that county offices had used to check farm programs.
Credit
USDA Farm Service Agency, Aerial Photography Field Office
2015
England's lidar made free
Lidar measures distance by timing laser pulses, and flown over land it measures the height of the ground and of what stands on it. England's Environment Agency had flown it for 17 years to model floods and track coastal habitats. In September 2015 it began releasing all 11 terabytes free for any use under the Open Government Licence, starting with composite height grids at 25 cm to 2 m that covered over 72% of England. In December 2017 it said it would map all 130,000 km² of England by 2020 and release that data free too.
Credit
Environment Agency
2016
Two satellites map all land
TanDEM-X, launched in 2010, flies a few hundred meters from its near twin TerraSAR-X, launched in 2007, so the pair images the ground by radar from two angles at once, which gives height. By 2015 they had covered all 150 million km² of land several times, and in September 2016 DLR, the German Aerospace Center, completed a global elevation model on a 12 m grid; DLR gives its absolute height error as about 1 m. An edited version named WorldDEM became the source of the Copernicus DEM, whose 30 m and 90 m versions have been free to registered users since 2019, except a small set of 30 m tiles over certain countries. Like any radar model it is a surface model: it shows the tops of trees and buildings, not the bare ground.
dlr.degeoservice.dlr.dedataspace.copernicus.euregistry.opendata.aws
Credit
DLR; Airbus, DLR's partner in the mission; the Copernicus program
2022
Trees and buildings removed
Laurence Hawker, Peter Uhe, Jeffrey Neal and four colleagues used machine learning, trained on reference elevation data from 12 countries, to take forests and buildings out of the Copernicus DEM, and published FABDEM, which they call the first global elevation map with both removed, at about 30 m. They report that it cuts the mean absolute height error from 1.61 m to 1.12 m in built-up areas and from 5.15 m to 2.88 m in forests.
Lessona radar map of height records whatever the signal hits first, and flood models, which need the ground, gave wrong answers until the trees and roofs came off.
Credit
Laurence Hawker and colleagues, University of Bristol and Fathom
2025
A contest ranks terrain maps
DEMIX, a comparison of global elevation models run by Peter Guth with colleagues in Italy, Brazil, Canada and Germany, scored six models at 1 arc-second (about 30 m), Copernicus DEM, AW3D30, EDTM, GEDTM, FABDEM and FathomDEM, against lidar ground models from national mapping agencies over 1,510 test tiles of about 10 × 10 km in the United States and western Europe, using slope and curvature as well as height. Published on 3 December 2025, it found FathomDEM, which Fathom built with a hybrid vision transformer (a type of AI model) and published in February 2025, best against lidar but under a restrictive licence, and the Copernicus DEM the best free choice.
Credit
Peter L. Guth, Sebastiano Trevisani, Carlos H. Grohmann, John B. Lindsay and Hannes I. Reuter
Watch for
A fused open terrain map
Yu-Feng Ho and colleagues at OpenGeoHub combined the Copernicus DEM, ALOS World 3D and other models, trained on more than 30 billion ICESat-2 and GEDI laser heights, into GEDTM30, an open 30 m model of bare ground. They say it cuts the Copernicus DEM's root mean square error by about 27.3% where tree cover exceeds 50%, and that it has the lowest errors of the models they tested when checked against GNSS (satellite positioning) stations, though FathomDEM did better against reference lidar models. The DEMIX comparison of December 2025 found GEDTM v1.2 poor for slope and curvature and advised against it for those uses.
Achieved whena later version matches or beats the Copernicus DEM in an outside lidar-referenced test such as DEMIX.
Claimed by
Yu-Feng Ho and colleagues, OpenGeoHub, July 2025
Open
01
Terrain maps tested in tropics
A DEMIX-style ranking of the global terrain models (Copernicus DEM, FABDEM, FathomDEM, GEDTM30, AW3D30) in tropical forest, scored against ground models from open airborne lidar, with the reference grids and code published.
The briefCredit · unclaimed02
Building heights checked on lidar
An outside test of global building-height maps, GlobalBuildingAtlas and Microsoft's Global Building Footprints heights, against roof heights measured from open airborne lidar in cities on at least three continents, reported by building height band.
The briefCredit · unclaimed
Open 01 · the brief
Terrain maps tested in tropics
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A DEMIX-style ranking of the global terrain models (Copernicus DEM, FABDEM, FathomDEM, GEDTM30, AW3D30) in tropical forest, scored against ground models from open airborne lidar, with the reference grids and code published.
Why it is still open
DEMIX's 2025 ranking used tiles only from the United States and western Europe, and its authors list as future work to "evaluate tiles with tropical forests" and to "keep looking for desert areas in the great belt from North Africa through the Middle East into Central Asia". Meadows, Jones and Reinke (2024) tested five free models against 65 airborne lidar surveys chosen to represent flood-prone places worldwide and ranked FABDEM first; their test came before FathomDEM and GEDTM30 were published and was not built around tropical forest. Forests are where height errors are largest: FABDEM's authors measured 5.15 m mean absolute error there before correction. Laser heights from space do not fill the gap: DEMIX notes that after seven years ICESat-2 has returns in fewer than 10% of 30 m pixels.
Built with what exists
NASA's archive of airborne lidar surveys over forest sites in Brazil, 2008 to 2018, 3,154 files with terrain models, shared without restriction; ICESat-2 ground heights as spot checks; the global models above; GDAL and Python.
First step
Download the Brazilian lidar for five survey sites, average each ground model to 1 arc-second, and compute height differences for the Copernicus DEM, FABDEM and GEDTM30.
Done looks like
Scores for at least 100 tropical-forest tiles of about 10 × 10 km, computed the way DEMIX computes them, with a ranking that can be set beside DEMIX's, published with the code and reference grids.
Open 02 · the brief
Building heights checked on lidar
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An outside test of global building-height maps, GlobalBuildingAtlas and Microsoft's Global Building Footprints heights, against roof heights measured from open airborne lidar in cities on at least three continents, reported by building height band.
Why it is still open
GlobalBuildingAtlas's authors (June 2025) give their own height errors for 2.75 billion buildings made from PlanetScope pictures. The only outside lidar check found is a USGS and Ohio State University study published in July 2026, which compared USA Structures and Microsoft's heights with USGS lidar at two US sites, Centennial, Colorado and Houston, Texas.
Built with what exists
Open lidar from the USGS 3D Elevation Program, England's Environment Agency and New Zealand's LINZ, whose point clouds are free to download; building outlines from the same maps; PDAL, a free point-cloud library; Python with GeoPandas.
First step
Take one English city covered by Environment Agency lidar, compute roof heights for 1,000 buildings, and compare them with GlobalBuildingAtlas heights.
Done looks like
For at least 10,000 buildings in at least six cities on three continents, each dataset's median and 90th-percentile height error in meters, split by building height, published with the code.