DSM vs DTM vs orthomosaic: what each one is for

A drone survey, once processed, comes back as three files with names that look like alphabet soup. They are three views of the same measurement, each answering a different question, and most mistakes with survey data come from asking one of them a question meant for another. This guide says what each file holds, which to measure on, and what to hand to the people downstream.

The three files

The orthomosaic

The aerial photographs stitched into one image and corrected so that every pixel is seen straight down, at a constant scale, in map coordinates. "Ortho" means that correction: a plain photo from a drone is taken at a slight angle and over ground at different heights, so near things look larger than far things and a vertical pole leans. The orthomosaic undoes that, pixel by pixel, using the surface model, so that a distance measured on it is a distance on the ground. It is the file people mean when they say "the map", and the one you can point at in a meeting.

It holds no height. A measurement of a pile's footprint on it is correct; a measurement of the pile's volume has to come from somewhere else.

The digital surface model

A grid of heights, one per pixel, of whatever the cameras saw first from above. On bare ground it is the ground. Over a stockpile it is the top of the pile; over a loader, the roof of its cab; over a tree, the canopy; over a building, the roof. It is the raw product of the photogrammetry, the dense cloud of matched points turned into a surface, and it is the honest record of what the site looked like at the moment of the flight.

This is the file for anything that stands on the ground and that you want measured: stockpile volumes, the height of a bench face, the top of a wall. It is also the file a contour map or a cut and fill must not be drawn on, unless the site is bare, because every truck and tree becomes relief.

The digital terrain model

The same grid of heights with everything that is not ground taken out. Processing software classifies points as ground or not-ground (by slope, by height above neighbours, by colour) and fills the gaps under what it removed by interpolating from the ground around. The result is the bare earth: the surface an engineer designs on and a grader shapes.

It is derived, so it can be wrong in two directions. Over dense vegetation, where the cameras never saw the ground, the fill is a guess, and the terrain under a wood is a smooth invention. On a quarry, an aggressive filter can decide a stockpile is "not ground" and flatten it, so a stockpile measured on the DTM reads as nothing. Know which you are holding before you measure.

Which file for which job

• Stockpile volume: the surface model, against a base that describes the ground under the pile. (How the base is chosen.)
• Cut and fill against a design: the terrain model, on the same datum as the design. Trees in the "existing ground" become cut nobody will dig. (The methods.)
• Contours, slope, drainage: the terrain model. Contours on a surface model wrap every parked machine. (Contours from a survey.)
• Progress between two flights: usually the surface model, because the thing that changed (a pile drawn down, a pad built up) is what you want to see. Compare terrain to terrain when the question is the ground itself. (Comparing two flights.)
• Lengths, areas, positions, the plan: the orthomosaic, with the heights read from the elevation model when a tool needs them.
• A height at a point: depends on the question. The top of the pile is DSM; the ground at the toe is DTM; on bare ground they agree.

Reading the files: a few things to check

Resolution

All three are usually delivered at the ground sample distance of the flight, 2 to 5 cm per pixel for a normal mapping flight. The elevation models are sometimes delivered coarser, at 2 to 4 times the ortho's pixel, because the dense cloud is noisier than the imagery. A terrain model at 10 cm per pixel is fine for earthwork; a surface model at 10 cm is still fine for a 6 m pile. (What sets the resolution.)

Coordinate system and vertical datum

A GeoTIFF carries its horizontal coordinate system in the file; check it is the one the rest of the project uses (a UTM zone or a state plane). The vertical datum is the one that bites. Heights straight from a GPS are ellipsoid heights; heights on a plan are orthometric (above sea level, via a geoid model), and the two differ by tens of metres. A terrain model in ellipsoid heights compared to a design in orthometric gives a site that is entirely cut or entirely fill. The software should tell you which it has; if it does not, measure the height of something you know.

No-data holes

Water, glass, deep shadow and anything that moved between photos leave holes in the elevation models that the software either leaves empty or fills by interpolation. A filled hole over a pond is a flat invention at some height. Check the edges of the site and any water before measuring across them.

What to send downstream

Clarity reads the three GeoTIFFs you already have (or produces them from your photos on Enterprise), keeps the elevation models under every tool so a volume, grade or contour comes from the right surface, and exports the terrain as LandXML, contours as DXF at true elevation, and measurements as CSV, GeoJSON or DXF. (What Clarity produces from your photos.)