DTM vs DSM vs DEM: What's the Difference?

A DEM (Digital Elevation Model) is the generic umbrella term for any raster representation of terrain elevation. A DTM (Digital Terrain Model) represents the bare earth surface with all vegetation, buildings, and other objects removed. A DSM (Digital Surface Model) represents the top of everything — trees, rooftops, powerlines, vehicles, and the ground where nothing else exists. The difference matters because choosing the wrong model will give you wrong answers: a DSM will overestimate earthworks volumes, a DTM won’t show you the tree canopy you’re trying to measure, and using the terms interchangeably will confuse your clients.

I see these terms misused constantly — in tender documents, client specifications, and even software interfaces. This guide sets the record straight.

Quick comparison table

DEM: the umbrella term

DEM stands for Digital Elevation Model. It is a generic term that refers to any regularly-gridded raster dataset where each pixel stores an elevation value. Both DTMs and DSMs are types of DEM.

In practice, the term “DEM” is used loosely. In the United States, the USGS National Elevation Dataset (now the 3DEP program) uses “DEM” to refer to bare earth models — essentially what we’d call a DTM. In other countries, “DEM” might refer to a surface model. This inconsistency is a source of endless confusion.

My recommendation: avoid using “DEM” when precision matters. Say DTM or DSM instead, so there’s no ambiguity about whether vegetation and structures are included.

A note on terminology by region

Always check the product specification or metadata. Never assume.

DTM: bare earth

What it is

A Digital Terrain Model represents the bare earth surface. All above-ground features — trees, buildings, vehicles, power lines, fences — have been removed. What remains is the natural (or graded) ground surface.

A DTM answers the question: “If I stripped away everything sitting on the ground, what would the terrain look like?”

How it’s created

There are two main workflows:

From LiDAR:

1. Collect a LiDAR point cloud with multiple returns.
2. Classify the point cloud — label each point as ground, vegetation, building, noise, etc. Ground classification is typically done with algorithms like Progressive Morphological Filter or Cloth Simulation Filter.
3. Extract only the ground-classified points.
4. Interpolate a regular grid (raster) from the ground points using IDW, kriging, or TIN-based methods.
5. Export as GeoTIFF.

From photogrammetry:

1. Generate a dense point cloud from overlapping drone imagery.
2. Apply ground filtering algorithms (less reliable than LiDAR because photogrammetry cannot penetrate vegetation).
3. Interpolate the ground surface.
4. Export as GeoTIFF.

LiDAR produces significantly better DTMs than photogrammetry, especially in vegetated areas, because LiDAR pulses penetrate through gaps in the canopy to reach the ground. Photogrammetry can only see what’s visible from above.

Common applications

• Earthworks volume calculations — Cut and fill requires the bare earth surface, not the treetops.
• Drainage and hydrology — Water flows along the ground, not over building rooftops. Flood models need DTMs.
• Contour generation — Site contours should represent the ground, not vegetation height.
• Slope and aspect analysis — For geotechnical assessment, landslide risk, or solar panel placement.
• Road and rail design — Vertical alignment design starts from the existing terrain surface.
• Archaeological prospection — Removing vegetation from LiDAR data reveals subtle terrain features like earthworks, ditches, and mounds.

Accuracy considerations

DTM accuracy depends on:

• Point cloud density — More ground points means better interpolation.
• Classification quality — Misclassified vegetation points will corrupt the DTM.
• Terrain complexity — Steep, variable terrain is harder to model accurately than flat ground.
• Vegetation density — Dense canopy means fewer laser pulses reach the ground, reducing ground point density.

The last row is the critical one. Photogrammetry DTMs under vegetation are unreliable because the camera cannot see the ground. If your site has significant vegetation and you need a DTM, use LiDAR.

DSM: everything on top

What it is

A Digital Surface Model represents the highest surface at each point — the top of buildings, the top of the tree canopy, the top of vehicles, power lines, and the ground where no above-ground features exist.

A DSM answers the question: “If I looked straight down from above, what’s the first thing I’d hit?”

How it’s created

From LiDAR:

1. Collect a LiDAR point cloud.
2. Use only the first returns (the first surface each laser pulse hit).
3. Interpolate a regular grid from the first-return points.
4. Export as GeoTIFF.

From photogrammetry:

1. Generate a dense point cloud from overlapping imagery.
2. Interpolate a grid from the full point cloud (no filtering needed — photogrammetry naturally captures the visible surface).
3. Export as GeoTIFF.

Photogrammetry is actually well-suited for DSM generation because it captures exactly what’s visible from above. The limitation — not seeing through vegetation — is irrelevant when you want the top surface.

Common applications

• Vegetation analysis — Subtract the DTM from the DSM to get a Canopy Height Model (CHM): CHM = DSM - DTM.
• Obstruction mapping — Identifying structures or vegetation that might interfere with a proposed development, flight path, or line of sight.
• Solar analysis — Calculating shadow patterns from buildings and trees for solar panel feasibility.
• Telecommunications — Line-of-sight analysis for antenna placement.
• 3D city modelling — Building height extraction from DSM data.
• Noise modelling — Sound propagation is affected by surface features that the DSM captures.

Accuracy considerations

DSM accuracy from photogrammetry is typically good (3-10 cm) because the camera can clearly see the surfaces being modelled. DSM accuracy from LiDAR depends on the pulse rate and scan pattern — some thin features like power lines may not be fully captured if the point density is insufficient.

DTM vs DSM: visual comparison

Consider a cross-section through a site with a building and some trees:

DSM profile:    ____|````|____/````\____
                     bldg      trees

DTM profile:    _________________________
                    (bare ground only)

The DSM follows the tops of everything. The DTM follows the ground beneath. The vertical difference between them at any point tells you the height of whatever is sitting on the ground at that location.

When to use each

Use a DTM when:

• Calculating earthworks cut/fill volumes
• Modelling water flow or flood risk
• Designing roads, platforms, or foundations
• Generating ground-level contours
• Performing slope stability analysis
• Measuring terrain change over time

Use a DSM when:

• Measuring canopy height or vegetation density
• Assessing building heights
• Performing line-of-sight or viewshed analysis
• Modelling shadow patterns
• Mapping obstructions for aviation or telecommunications
• Creating a realistic 3D surface for visualisation

Use both when:

• Calculating a Canopy Height Model (DSM minus DTM)
• Assessing vegetation clearance around infrastructure
• Comparing terrain change vs. surface change between two survey epochs
• Delivering a complete survey dataset to a client

File format: GeoTIFF

Both DTMs and DSMs are typically delivered as GeoTIFF files. A GeoTIFF is a standard TIFF image with embedded geospatial metadata — coordinate reference system, pixel size, origin coordinates, and no-data values.

Key parameters in a GeoTIFF elevation model:

For a 50-hectare site at 10 cm pixel size, the GeoTIFF will be approximately 200-500 MB uncompressed. LZW or DEFLATE compression within the TIFF reduces this significantly.

How to create DTMs and DSMs

From LiDAR point clouds

Most LiDAR processing software can generate both:

• CloudCompare (free) — CSF plugin for ground classification, rasterize tool for grid export.
• LAStools (commercial/open) — lasground for classification, las2dem or blast2dem for rasterization.
• PDAL (free) — SMRF or PMF ground classification, writers.gdal for GeoTIFF output.
• Global Mapper — Built-in ground classification and raster export.
• TerraScan (Terrasolid) — Industry standard for airborne LiDAR classification.

From photogrammetry

• Agisoft Metashape — Generate DEM from dense cloud, with option to use ground-classified points only (DTM) or all points (DSM).
• Pix4Dmapper — Generates DSM automatically; DTM requires point cloud classification.
• OpenDroneMap — Free, generates DSM and DTM.
• DJI Terra — Generates DSM from DJI drone imagery.

From Swyvl

If you’re delivering DTM or DSM data to clients, you can upload the GeoTIFF directly to Swyvl. Clients view it in the browser without needing GIS software — see How to Share a GeoTIFF Online.

Common mistakes

Mistake 1: Using a DSM for earthworks volumes. If there are trees on the site, the DSM includes the canopy. Your volume calculation will be wildly inflated. Always use a DTM for cut/fill.

Mistake 2: Expecting a good DTM from photogrammetry under vegetation. Photogrammetry cameras cannot see through trees. The “ground” surface in a photogrammetry DTM under canopy is an interpolation at best, a guess at worst.

Mistake 3: Assuming “DEM” means “DTM.” Always check. If a client asks for a DEM, clarify whether they want bare earth (DTM) or top surface (DSM). This one conversation can save you a redo.

Mistake 4: Ignoring the coordinate reference system. A DTM in one CRS overlaid with a DSM in another CRS will give you nonsensical CHM values. Always verify that both datasets share the same horizontal and vertical datum.

Mistake 5: Delivering at the wrong resolution. A 1 m pixel DTM from 100 pts/sq m LiDAR is throwing away data. A 5 cm pixel DTM from 10 pts/sq m LiDAR is over-interpolated. Match pixel size to source data density.

Summary

• DEM is the generic term for any elevation raster. Avoid using it when precision matters.
• DTM is bare earth — vegetation and structures removed. Use it for earthworks, drainage, and terrain analysis.
• DSM includes everything on top — canopy, buildings, vehicles. Use it for obstruction mapping, vegetation analysis, and line-of-sight studies.
• LiDAR produces better DTMs than photogrammetry because it penetrates vegetation. Photogrammetry produces good DSMs.
• Deliver as GeoTIFF — the universal format for raster elevation data.
• Always clarify which model your client actually needs. DTM and DSM answer very different questions.

For more on the underlying data formats, see What is a GeoTIFF?, What is LiDAR?, and What is an Orthomosaic?.