11.1 DEM, DSM, DTM — What's the Difference?

Key takeaways

• DEM is a generic term; DSM includes surface features (buildings, canopy); DTM is the bare-earth surface.
• Global free DEMs (SRTM, Copernicus, ALOS) cover the world at 10–90 m.
• Higher-resolution DEMs come from LiDAR, drone photogrammetry, or commercial aerial surveys.

Introduction

Elevation data is fundamental to GIS: it underpins hydrology, visibility, infrastructure, agriculture, and much more. Three closely related terms — DEM, DSM, DTM — get used loosely and interchangeably, but they mean different things. This short lesson clears that up.

Definitions

• DEM (Digital Elevation Model) — a generic term for any raster of elevation values. Could be DSM or DTM depending on source and processing.
• DSM (Digital Surface Model) — the elevation of the first reflective surface — typically tree canopy, building rooftops, bridges, and bare ground where nothing's on top.
• DTM (Digital Terrain Model) — the bare-earth elevation — trees and buildings removed, sometimes with additional structure information (breaklines, contours).

A rough mnemonic: DSM = what you see; DTM = what you'd see if you stripped everything off.

Canopy height model

If you have both DSM and DTM for the same area, subtracting gives you a Canopy Height Model (CHM):

$$CHM = DSM - DTM$$

Forest height, building height, crane height — all emerge from this difference.

Data sources

Free global

• SRTM (2000) — 30 m. DSM-like (canopy-top in forests, ground in open areas).
• ASTER GDEM — 30 m global.
• Copernicus DEM GLO-30 — 30 m, DSM-ish, newer than SRTM.
• ALOS PALSAR — 12.5 m via radar, limited coverage.
• OpenTopography — aggregator + cloud access.

Free national

Many countries publish LiDAR-derived DTMs. Examples:

• USA: USGS 3DEP 1 m LiDAR DTMs for most populated areas.
• Denmark: DHM 40 cm nationwide.
• UK: Environment Agency 1–2 m nationwide.
• Netherlands: AHN 25 cm nationwide.

Commercial high-res

• WorldDEM (Airbus) — 12 m DSM globally.
• NextMap — 1 m DSM.
• Vexcel / Nearmap — aerial photogrammetry at 10–15 cm.

Resolution and accuracy

Elevation rasters have three kinds of accuracy:

• Horizontal (x, y) — how precisely cells are geolocated.
• Vertical (z) — the accuracy of the elevation values.
• Data source — LiDAR → cm accuracy; radar / photogrammetry → metres.

A 1 m raster does not imply 1 m accuracy — it's the cell size, not the error.

Pixel area vs real surface area

A DEM's cell area is flat (e.g., 30 × 30 m = 900 m²). The real surface area of that cell is larger if the slope is steep. Some analyses (e.g., runoff coefficient by surface area) correct for this.

Processing from raw data

Creating a DTM from a LiDAR point cloud involves:

1. Classify ground vs non-ground points.
2. Interpolate bare-earth surface from ground returns.
3. Grid to a raster at chosen resolution.

Tools: PDAL, LAStools, CloudCompare, QGIS processing.

Working with DEMs

Standard operations:

• Slope and aspect — Module 11.2.
• Hillshade — Module 11.3.
• Contours — Module 11.4.
• Watershed / hydrology — Module 11.5.
• Viewshed — Module 11.6.
• Elevation profile — elevation along a path.

GDAL ships a suite (gdaldem slope, gdaldem hillshade, gdal_contour) for all of these.

Choosing the right DEM

Match the question to the data:

• Continental climate modelling — SRTM 30 m, Copernicus DEM.
• Regional hydrology — Copernicus 30 m, national 10 m.
• Urban flood modelling — national 1 m LiDAR DTM.
• Forest canopy — LiDAR DSM + DTM → CHM.
• Construction site — drone orthophoto + DSM at 3–5 cm.

Common pitfalls

• Treating SRTM as DTM — it's closer to DSM in forested areas; don't use it to model runoff under canopy.
• Mixed DSM / DTM in one analysis — subtle biases accumulate.
• Vertical datum — is the elevation ellipsoidal (GNSS) or orthometric (sea-level)? Module 4.1 covers this.
• Nodata — check for nodata values (often -9999) before statistics.

Self-check exercises

1. You need to model water flow through a forested catchment. Which elevation product?

A DTM — bare-earth — so water is routed correctly along the ground, not over tree canopy. SRTM and similar global DSM-like products overestimate elevation in forested areas, diverting flow unrealistically. National LiDAR DTMs are the gold standard.

2. How do you compute building heights across a city from open data?

Acquire DSM and DTM rasters at the same resolution and extent. Subtract: building_height_raster = DSM - DTM. Mask to building footprints (from OSM or a municipal dataset). Compute zonal max / mean per building. For best results, use a DSM with ≤1 m resolution.

3. A 30 m DEM has a "vertical accuracy of 16 m at 90% confidence". What does this tell you?

The elevation value for any cell is within ±16 m of the true height 90% of the time. This limits precision — you can't distinguish features with elevation differences smaller than ~16 m, and slope calculations at 30 m horizontal with 16 m vertical error are noisy on gentle terrain. For small-scale hydrology you'd want better data.

Summary

• DEM is generic; DSM is surface (first reflection); DTM is bare earth.
• Subtracting DSM − DTM gives canopy / building heights.
• Resolution and accuracy are different things.
• National LiDAR is the gold standard where available; SRTM / Copernicus cover the rest.

Further reading

• OpenTopography — free DEM aggregator with tutorials.
• Florinsky, I. V. — Digital Terrain Analysis in Soil Science and Geology.
• NOAA — Digital Elevation Models overview.
• USGS 3DEP program documentation.