BRENNEN

Industries / Construction and property / Engineering and architecture / Surveying

Aerial and drone mapping

This work measures the ground from above, from photographs and laser pulses taken from balloons, aircraft and drones, and turns them into maps and elevation models whose accuracy is stated and checked.

7

Milestones achieved, from 1858 to 2025.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1858

    A photograph from a balloon

    Nadar, the Paris photographer Gaspard-Félix Tournachon, photographed the village of Petit-Bicêtre near Paris from a tethered balloon in 1858. Guinness World Records calls him the first person to take an aerial photograph and gives the height as 80 m (262 ft). His wet glass plates had to be developed at once, so he carried a darkroom in the basket. None of those first pictures survive; the oldest aerial photograph still in existence is James Wallace Black's view of Boston, taken from a balloon in 1860.

    guinnessworldrecords.comexpositions.bnf.fr

    Credit

    Nadar (Gaspard-Félix Tournachon)

  2. 1936

    Maps drawn from photo pairs

    A stereoplotter rebuilds the ground in three dimensions from two overlapping aerial photographs, so an operator can trace contours from the model instead of walking the land with a plane table. The US Geological Survey (USGS) bought several of Zeiss's Multiplex plotters, which were small, simple and cheap, and they arrived from Germany late in 1935. Its first mapping with them showed distortions, and Zeiss replied that the Multiplex was meant only for rough reconnaissance maps. The Survey had Russell K. Bean redesign it, with new parts made by Bausch & Lomb in Rochester, New York, and its first bar of nine projectors went to work in April 1936, in a unit set up with the Tennessee Valley Authority.

    Lessonan instrument built for rough maps had to be rebuilt before it could meet the standard of a survey map.

    pubs.usgs.gov

    Credit

    Russell K. Bean, US Geological Survey; Zeiss; Bausch & Lomb

  3. 1947

    A rule for map accuracy

    A committee of the society now called ASPRS (the American Society for Photogrammetry and Remote Sensing) began drafting accuracy rules in 1937, and the US Bureau of the Budget issued the United States National Map Accuracy Standards in 1941, in their current form in 1947. At most 10% of well-defined points tested, such as bench marks, property boundary monuments and road crossings, may be off by more than 1/30 inch on maps at scales larger than 1:20,000, or 1/50 inch at smaller scales, which is 40 ft (12.2 m) on the ground on a USGS 1:24,000 map. At most 10% of tested elevations may be off by more than half the contour interval. Points are checked against field surveys of higher accuracy, and the USGS began testing its maps systematically in 1958.

    pubs.usgs.gov

    Credit

    ASPRS; US Bureau of the Budget; US Geological Survey

  4. 1997

    A country measured by laser

    Lidar measures distance by timing laser pulses. In 1997 the Netherlands began measuring the height of its whole land surface with lidar flown from helicopters and aircraft, for the Actueel Hoogtebestand Nederland (AHN), a national height file made jointly by the water boards, the provinces and Rijkswaterstaat, the Dutch government's water and roads authority, which share responsibility for flood defenses. The program says the ground height of every square meter of the country is known to within 5 cm, and the government publishes the data as open data without conditions.

    ahn.nl

    Credit

    Dutch water boards, provinces and Rijkswaterstaat

  5. 2014

    Why drone maps bulge

    M. R. James of Lancaster University and Stuart Robson of University College London showed that elevation models built from drone photos taken straight down, with the camera's lens distortion estimated during processing, tend to bend into a broad dome, an error that fine surface detail hides. Adding photos taken at an angle cut the error by up to two orders of magnitude when no ground control points were used, and they published flight plans for fixed-wing and rotor drones that do this.

    Lessona sharp-looking drone map can still be bent, and only checkpoints or a better pattern of photos reveal it.

    eprints.lancs.ac.uk

    Credit

    M. R. James; Stuart Robson

  6. 2014

    Accuracy without a map scale

    Digital maps and elevation models can be viewed at any scale, so rules written for paper maps no longer fit. In November 2014 ASPRS published its Positional Accuracy Standards for Digital Geospatial Data, which state accuracy as a root mean square error (RMSE, a kind of average error) measured at surveyed checkpoints, independent of map scale, pixel size or contour interval. Edition 2, approved in August 2023 and revised in June 2024, raised the minimum number of checkpoints from 20 to 30, dropped the 95% confidence figure, added guidance for drones, and stopped using accuracy under vegetation as a pass-or-fail test for lidar, because the ground under trees depends on how many pulses reach it and on the software that sorts ground from vegetation.

    courses.ems.psu.eduaagsmo.org

    Credit

    ASPRS

  7. 2025

    A national lidar baseline

    The US 3D Elevation Program (3DEP) began in 2012 from a recommendation to collect high-quality lidar for the conterminous United States and Hawaii, among other areas, with IfSAR, a radar method, for Alaska. Its first full year of production was 2016, and by the end of fiscal year 2025, by the USGS's count, 99% of the nation had baseline data meeting its specifications available or in progress, free of charge and without use restrictions. From 2015 to 2025 partners such as states and other agencies paid 60% of the cost of collecting data, the USGS 33% and disaster funds 7%.

    usgs.govusgs.gov

    Credit

    US Geological Survey and its partners

Watch for

  1. AI drafts the survey linework

    Linework is the drawing a surveyor delivers from a site's measured points: curbs, pavement edges, walls and other features as lines in a computer-aided design (CAD) file, still largely traced by hand. Mach9 says its Digital Surveyor 2 uses AI to suggest curbs, pavement edges, guardrails and barriers directly from lidar point clouds, with each suggestion editable, and that early users report finishing projects up to 100 times faster than before.

    Achieved whenan outside test reports the share of features the tool found and their position error against points surveyed on the ground.

    geoweeknews.com

    Claimed by

    Mach9, as reported by Geo Week News, April 2026

Open

  1. 01

    Ground under trees, scored

    A public scoreboard that reruns ground-finding programs on US 3DEP lidar point clouds and scores each one where it counts: at the checkpoints that surveyors measured on the ground under forest, shrub and tall grass. Each program's ground surface is compared with the surveyed heights and with the surface the contractor delivered.

    The brief
    Credit · unclaimed
  2. 02

    Linework checked on the ground

    An outside test of the AI tools that draft survey linework from drone photos and lidar. The same site data goes through each tool, and every curb, pavement edge and painted line it draws is compared with points measured on the ground by GNSS (satellite positioning) or a total station. The test also times how long a person takes to finish what each tool leaves undone.

    The brief
    Credit · unclaimed

Open 01 · the brief

Ground under trees, scored

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public scoreboard that reruns ground-finding programs on US 3DEP lidar point clouds and scores each one where it counts: at the checkpoints that surveyors measured on the ground under forest, shrub and tall grass. Each program's ground surface is compared with the surveyed heights and with the surface the contractor delivered.

Why it is still open

ASPRS dropped accuracy under vegetation as a pass-or-fail test in 2023 partly because it depends on the software that separates ground from vegetation. Published comparisons of that software score it against points labeled as ground in the data, not against the ground itself: OpenGF, the largest benchmark (Qin and colleagues, 2023), took its labels from the data providers' own classifications. The USGS's own 2025 study by Sampath, Irwin and Stoker checks the delivered 3DEP data, not alternative programs. Meanwhile the USGS has gathered 145,299 survey-grade checkpoints from 551 lidar projects into one database (version 2.0, December 2025), "tens of millions of dollars" of field data in its words, and no comparison of ground-finding programs against it was found.

aagsmo.orguwaterloo.cacatalog.data.govusgs.gov

Built with what exists

The USGS 3DEP checkpoint database and the per-project checkpoint files on the USGS rockyweb server; 3DEP point clouds; PDAL, a free point-cloud library, with its SMRF, cloth-simulation and morphological ground filters; deep networks trained on the OpenGF dataset; Python.

  1. First step

    Take the Jackson County, Oregon 2022 project, whose delivered surface the USGS scored at 15.91 cm (95th percentile) over 46 vegetated checkpoints, rerun PDAL's SMRF filter on the tiles around ten forest checkpoints, and compare the two.

    rockyweb.usgs.gov

  2. Done looks like

    For at least 20 projects and 1,000 vegetated checkpoints, a public table of each program's 95th-percentile height error in cm by land cover (forest, shrub, tall grass), set beside the delivered surface's figure, with the code published.

Open 02 · the brief

Linework checked on the ground

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An outside test of the AI tools that draft survey linework from drone photos and lidar. The same site data goes through each tool, and every curb, pavement edge and painted line it draws is compared with points measured on the ground by GNSS (satellite positioning) or a total station. The test also times how long a person takes to finish what each tool leaves undone.

Why it is still open

AirWorks says its automation extracts about 85% of the expected linework and the rest is drawn by people; Mach9's speed figures (see Watch for) come from its users. Published research on extracting curbs scores the results against lines a person digitized from the same scan, not against the ground: in Miraliakbari, Hahn and Sok's 2015 study in Stuttgart, "ground truth measurements are digitized manually from the laser scanner data." No test of the commercial tools against field-surveyed points was found.

xyht.comisprs-archives.copernicus.org

Built with what exists

A car park or campus lot closed to traffic, used with the owner's permission; a mapping drone flown under the local drone rules with nobody beneath it, or the area's public lidar; a GNSS RTK rover or total station borrowed from a college surveying program; trial accounts of Mach9 Digital Surveyor, AirWorks, PIX4Dsurvey and Trimble Business Center; QGIS and CloudCompare, free programs for maps and point clouds.

  1. First step

    Get permission for one closed car park, measure 100 curb points on a quiet weekend morning, and run that site through one free trial.

  2. Done looks like

    500 field-measured points on curbs and pavement edges at three sites; for each tool, the share of features found, the median and 95th-percentile position error in cm, and the minutes of hand finishing per hectare, all published.