DSM

A Digital Surface Model (DSM) represents the elevation of the Earth's surface including all objects above bare ground — buildings, trees, bridges, power lines, and any other structures. Every pixel records the height of the topmost visible surface as seen from above. This is the fundamental distinction from a DEM, which strips away surface objects to reveal only the bare terrain beneath. A DSM is what a sensor actually measures when it looks down at the Earth — it's the "first return" surface. Most radar-derived elevation datasets (like SRTM) are technically DSMs, not DEMs, because the radar signal bounces off treetops and rooftops rather than penetrating to the ground.

DSMs are essential for any application that needs to model the real-world surface as it actually appears: urban 3D modeling, line-of-sight analysis, solar irradiance estimation, aviation obstacle clearance, and telecommunications tower placement. When combined with a DEM, the difference between the two surfaces reveals the height of above-ground features — this derived product is called a normalized DSM (nDSM) or, when focused on vegetation, a Canopy Height Model (CHM).

DSMs can be generated from several remote sensing technologies. Photogrammetry using stereo satellite imagery (e.g., WorldView, Pléiades) or aerial photographs computes surface elevation through parallax measurement — because the images overlap from different viewpoints, software can triangulate the height of each visible surface point. This naturally produces a DSM since cameras see the top of objects, not the ground beneath them.

LiDAR generates DSMs from first-return points — the first laser reflections, which come from the highest surface. Airborne LiDAR typically achieves 5–15 cm vertical accuracy for DSMs. Radar interferometry (InSAR), as used by SRTM and TanDEM-X, produces DSMs because microwave energy reflects from the first surface encountered — treetops in forests, roof surfaces in cities. The vertical accuracy is typically 2–10 m depending on the mission.

Post-processing can include filtering, void filling, and mosaicking. Unlike DEM production, DSM generation does not require classification of ground vs. non-ground points — the raw surface is the desired output.

DSMs are affected by the objects present at acquisition time — a construction site will show partially built structures, and seasonal vegetation changes alter the surface. Radar-derived DSMs have difficulty in dense urban canyons where multipath effects and layover distort height measurements. Photogrammetric DSMs cannot resolve heights beneath dense tree canopy because the camera only sees the canopy top. DSMs from optical stereo are affected by cloud cover, shadow, and low-texture areas (water, sand) where stereo matching fails. The distinction between DSM and DEM is sometimes blurred in product documentation — always verify whether an "elevation model" includes or excludes surface objects before using it.

DSMs have been produced since the early days of photogrammetry — stereoscopic aerial photographs inherently capture the visible surface, not bare earth. The SRTM mission (2000) produced the first near-global DSM from radar interferometry. TanDEM-X (2010–present) improved this to 12 m resolution globally, and a reprocessed version is freely available as the Copernicus DEM (which is actually a heavily edited DSM with some vegetation removed). Commercial stereo satellite imagery from WorldView and Pléiades enabled DSM generation at sub-meter resolution in the 2010s. Today, AI-assisted photogrammetric pipelines and drone-based LiDAR have made high-quality DSMs accessible for projects of any size.