Earth SAT Data
Home › Technical Guides › Elevation & 3D
Elevation & 3D

Digital Elevation Models: DSM, DTM, Optical Stereo, and InSAR

Understand vertical accuracy metrics, base-to-height stereo geometry, bare-earth DTM filtering, and geoid datum conversions for 3D terrain.

DSM, DTM, and Normalized Height Models Explained

A Digital Surface Model (DSM) captures the top reflective envelope including trees, roofs, and stockpiles, whereas a Digital Terrain Model (DTM) removes above-ground structures to represent bare-earth topography for hydrology and grading.

Subtracting DTM from DSM yields a normalized Digital Surface Model (nDSM) isolating object heights. Using a DSM inside flood routing causes artificial damming along highway overpasses and tree lines.

  • DSM: First-surface heights including canopy and buildings.
  • DTM: Bare-earth surface required for watershed and drainage models.
  • nDSM = DSM - DTM: Direct canopy or structure height above ground.

Optical Stereo Photogrammetry Versus InSAR Elevation

Optical stereo derives 3D terrain by matching conjugate pixels across two or three along-track views using Semi-Global Matching. Tri-stereo adds a near-nadir look that fills urban street canyons hidden in two-view pairs.

InSAR DEMs (such as TanDEM-X) derive elevations from radar phase differences across separated antennas or passes, penetrating clouds cleanly while exhibiting layover on cliff faces.

  • Optical Stereo (0.5–2 m): Crisp mine stockpiles and 3D urban models.
  • InSAR DEM (10–30 m): Cloud-free regional topography.
  • Airborne LiDAR (0.2–1 m): Highest bare-earth DTM beneath dense forest.

Base-to-Height Ratio and Vertical Accuracy (LE90 / RMSEz)

Vertical precision scales with the Base-to-Height (B/H) ratio—viewpoint separation divided by orbital altitude. Optimal B/H sits near 0.35–0.65; smaller ratios inflate height noise while excessive angles break stereo correlation. Vertical accuracy is reported as RMSEz or LE90 (1.6449 x RMSEz).

Vertical Datums: WGS84 Ellipsoid Versus EGM2008 Geoid

Satellite RPCs output heights above the WGS84 ellipsoid, whereas civil engineering requires gravity-referenced orthometric elevations above a geoid like EGM2008. Ignoring a 20–100 m geoid offset creates severe errors against local survey benchmarks.

Elevation Data Sources and Typical Accuracy Benchmarks

Source TechnologyTypical Grid SpacingAbsolute Vertical LE90Primary Engineering Applications
VHR Optical Tri-Stereo (w/ GCPs)0.5 m – 1.0 m0.5 m – 1.5 mOpen-pit stockpiles, 3D city twins, site grading
VHR Optical Stereo (RPC-only)1.0 m – 2.0 m2.0 m – 4.0 mCorridor planning, remote feasibility studies
Commercial InSAR (X-band)5 m – 12 m2.0 m – 4.0 mCloud-prone watershed modeling, ortho-DEM
Global Public DEMs (Copernicus 30)30 m4.0 m – 10.0 mSAR terrain flattening, regional screening
Airborne LiDAR0.25 m – 1.0 m0.10 m – 0.25 mSub-canopy bare-earth DTM, levee design

Absolute LE90 improves with Ground Control Points (GCPs); relative intra-scene precision is typically 2x tighter.

Worked Illustration: Stereo B/H Vertical Precision and Geoid Conversion

A 0.50 m stereo pair has B/H = 0.50 and matching precision sigma_px = 0.30 pixels. A site reports ellipsoidal height h = 142.80 m where EGM2008 undulation is N = -28.40 m.

  1. Ground parallax error: 0.30 px x 0.50 m = 0.15 m.
  2. Relative vertical RMSEz: 0.15 m / 0.50 = 0.30 m.
  3. Convert RMSEz to 90% confidence (LE90): 1.6449 x 0.30 m = 0.49 m.
  4. Convert ellipsoidal to EGM2008 orthometric height: H = 142.80 - (-28.40) = 171.20 m.

Calculated Result & Takeaway: Relative stereo precision is 0.30 m RMSEz (0.49 m LE90), and geoid correction shifts height from 142.80 m to 171.20 m orthometric.

3D Elevation & DEM Quality Assurance Checklist

  • Confirm project specs require a DSM (first-surface) or bare-earth DTM.
  • Verify horizontal and vertical EPSG datum codes (WGS84 vs EGM2008).
  • Target B/H ratio of 0.35–0.65 when tasking new optical stereo.
  • Request tri-stereo over dense urban cores or steep terrain.
  • Match vertical datums before differencing multi-date earthwork volumes.
  • Inspect shaded relief renders for water spikes or cloud artifacts.

Frequently Asked Questions

Why does water look spiked in optical stereo DSMs?

Moving waves lack persistent texture between views, causing stereo matching errors unless hydro-flattened.

Can optical stereo produce a bare-earth DTM under dense forest?

No. Optical sensors only see the top canopy; airborne LiDAR or long-band radar is required beneath closed trees.

How does Copernicus DEM GLO-30 compare to SRTM?

Copernicus GLO-30 uses newer TanDEM-X X-band InSAR with tighter accuracy and hydro-editing than year-2000 SRTM.

Do stockpile volume calculations require Ground Control Points?

Relative volume differences work well with RPC-only stereo, while tying grades to civil designs requires GCPs.

Primary References & Standards

  • ASPRS Positional Accuracy Standards for Digital Geospatial Data (2024).
  • NGA EGM2008 Earth Gravitational Model & Vertical Datum Transformation Standard.
  • ESA Copernicus DEM Product Handbook (TanDEM-X Derived Specifications).

Related Technical Guides