Terrain and clutter model generation for radio propagation
Accurate DTMs and clutter classification layers are the foundation of any credible propagation study. This page explains how SAR interferometry and stereo optical imagery produce them, and where each method breaks down.
Sensors
- TanDEM-X bistatic SAR: X-band (9.65 GHz) bistatic interferometry at 0.4 m slant-range resolution. The global TanDEM-X DEM is delivered at 0.4 arcsecond (~12 m) posting with 2 m LE90 vertical accuracy over flat, vegetated terrain, degrading to 4–10 m LE90 on steep slopes and in dense forest where penetration is partial.
- Pleiades Neo stereo optical: 30 cm panchromatic imagery acquired in same-pass stereo or tri-stereo mode. Photogrammetric DSMs derived from matched stereo pairs typically reach 0.5–1 m horizontal posting and 0.3–0.5 m RMSE vertical accuracy over open terrain with good contrast. Clouds are a hard blocker; a single overcast pass yields nothing.
- SRTM (Shuttle Radar Topography Mission): C-band (5.6 GHz) single-pass interferometry acquired in 2000. Available globally at 1 arcsecond (~30 m) posting. Vertical accuracy is approximately 5–9 m LE90 depending on region. Because C-band partially penetrates vegetation, SRTM elevations in forested areas sit above bare earth and below canopy top, making it a surface model of ambiguous phase centre.
- Sentinel-1 repeat-pass InSAR: C-band SAR with 6-day repeat at mid-latitudes (12-day global). Suitable for relative elevation change detection and for coherence-based clutter classification (low coherence correlates with vegetation and water). Absolute elevation retrieval requires a reference DEM and careful unwrapping; standalone Sentinel-1 InSAR is not a substitute for TanDEM-X for first-pass terrain modelling.
What a propagation tool actually needs from terrain data
Propagation models such as ITU-R P.1812 and Longley-Rice do not consume raw elevation grids directly. They require two distinct inputs: a bare-earth DTM describing the ground surface beneath buildings and trees, and a clutter classification layer that assigns each cell a land-cover category carrying an associated additional loss term. The distinction matters enormously. A DSM that includes a 15 m forest canopy will produce a path profile that looks like a ridge where none exists. Feed that to a 700 MHz coverage prediction and you will over-predict loss in the forest and under-predict it everywhere the canopy is absent.
The clutter categories used in P.1812 Annex 1 (open, suburban, urban, dense urban, forest, water) each carry empirically derived height and loss coefficients. Generating those categories from satellite data is a classification problem, not just an elevation problem. The two tasks, separating ground from surface, and labelling surface type, are solved by different sensors and different methods.
TanDEM-X: the closest thing to a global bare-earth standard
The TanDEM-X global DEM, produced by DLR from bistatic X-band acquisitions between 2010 and 2015, remains the most consistent global elevation dataset available for propagation work. Its 2 m LE90 specification over flat terrain is well-documented and independently validated. For rural and suburban propagation studies at frequencies below 6 GHz, where the first Fresnel zone radius at mid-path can exceed 30 m, a 12 m posting with 2 m vertical accuracy is generally adequate.
The caveats are real. On slopes steeper than roughly 20 degrees, layover and shadow artefacts degrade accuracy substantially, and LE90 figures of 4–10 m have been reported in mountainous regions. In dense tropical forest, X-band penetrates only partially, so the DEM phase centre sits somewhere between the ground and the canopy top, not reliably at either. For millimetre-wave planning in forested terrain, this ambiguity is not academic: a 5 m error in a 200 m path profile segment can shift a predicted diffraction loss by several decibels.
Stereo optical imagery: surface detail where SAR struggles
Pleiades Neo stereo pairs fill the gap TanDEM-X leaves in urban and peri-urban areas. Same-pass stereo acquisition eliminates the temporal decorrelation that plagues repeat-pass InSAR in cities, and 30 cm imagery resolves individual building edges cleanly enough for photogrammetric DSM generation at sub-metre posting. The resulting surface model captures rooftop geometry, tree crowns, and wall faces with a fidelity that no spaceborne SAR currently matches in horizontal resolution.
The practical constraint is cloud. Pleiades Neo is a push-broom optical sensor. A single cloud layer during the acquisition window produces a gap that cannot be filled from the same pass, and re-tasking adds days to weeks of latency depending on orbital geometry and tasking queue. For a project covering a mountainous or tropical corridor, planners should budget for multiple acquisition attempts and consider whether a cloud-tolerant SAR baseline DEM with optical refinement is more reliable than an optical-only workflow.
From elevation data to clutter classification
Clutter classification is derived by combining the DSM, the DTM, and a land-cover signal. The height difference between DSM and DTM (the normalised surface height, or nDSM) separates bare ground from elevated objects. Thresholding nDSM values against optical spectral indices, particularly NDVI from Sentinel-2 at 10 m resolution, separates vegetation from built structures. Urban density metrics, building footprint coverage and mean building height within a cell, then distinguish suburban from urban and dense urban categories.
Sentinel-1 coherence adds a useful discriminant. Persistent high coherence over multiple 6-day pairs indicates hard, stable surfaces: roads, rooftops, bare soil. Low coherence indicates vegetation or water. This coherence layer, combined with the nDSM and NDVI, produces a classification that is more stable than optical alone in areas with frequent cloud cover, because the SAR acquisitions accumulate regardless of weather.
The output is typically a raster at 10–20 m posting, with each cell assigned one of the P.1812 clutter categories and an associated representative clutter height. That raster is the direct input to the propagation engine. Accuracy of the classification, validated against ground truth in published studies, is typically 85–92% at category level, with the main confusion occurring at the suburban-urban boundary.
Honest limits and the cases where satellite data is not enough
Satellite-derived terrain and clutter data has a well-defined accuracy ceiling. For macro-cell planning at sub-3 GHz, that ceiling is usually sufficient. For small-cell and millimetre-wave planning at 26 GHz or above, where path loss is sensitive to centimetre-scale obstruction geometry, it is not. At those frequencies, individual parapets, HVAC units, and billboard structures matter. No current spaceborne sensor resolves them. That problem is addressed in the sibling page on urban canyon geometry characterisation.
Temporal currency is a separate concern. TanDEM-X data was acquired between 2010 and 2015. A city that has added a dense high-rise district since then will have an outdated clutter layer in that area. Sentinel-2 change detection can flag where the optical signature has shifted, prompting a targeted re-acquisition of Pleiades Neo stereo over the changed area. For rapidly developing markets, a workflow that combines a stable SAR baseline with periodic optical updates is more defensible than a single-epoch DSM treated as permanent.
Satellize applies this layered workflow operationally, running Sentinel-1 coherence stacks and Sentinel-2 NDVI composites on open-constellation data and adding commercial Pleiades Neo tasking where clients need sub-metre DSM accuracy over specific corridors.
Typical figures
| TanDEM-X DEM posting | 0.4 arcsecond (~12 m globally) |
| TanDEM-X vertical accuracy (flat terrain) | 2 m LE90 (DLR specification) |
| TanDEM-X vertical accuracy (steep slopes / dense forest) | 4–10 m LE90 (published validation range) |
| Pleiades Neo DSM posting | 0.5–1 m (same-pass stereo or tri-stereo) |
| Pleiades Neo vertical RMSE (open terrain) | 0.3–0.5 m |
| SRTM posting and vertical accuracy | ~30 m posting; 5–9 m LE90 (region-dependent) |
| Sentinel-1 revisit (mid-latitudes) | 6 days (combined Sentinel-1A and -1B where available) |
| Clutter classification output posting | 10–20 m typical |
| Clutter classification accuracy | 85–92% at P.1812 category level (published validation studies) |
| Delivery formats | GeoTIFF (DTM, DSM, nDSM, clutter raster), Shapefile or GeoPackage (clutter polygons), ASCII grid for direct propagation tool import |
Analytics Satellize can run
| Bare-earth DTM | TanDEM-X bistatic InSAR phase unwrapping with hydrological correction and void filling | GeoTIFF at 12 m or resampled posting, referenced to WGS84 ellipsoid or national vertical datum on request |
| Surface DSM | Pleiades Neo stereo photogrammetry (SGM dense matching) or TanDEM-X where optical cloud-blocked | GeoTIFF at 0.5–1 m posting covering tasked corridor |
| Normalised surface height (nDSM) | DSM minus DTM, with noise filtering and minimum-height thresholding | GeoTIFF layer indicating height of above-ground objects per cell |
| P.1812-compatible clutter classification raster | Decision-tree fusion of nDSM, Sentinel-2 NDVI composite, and Sentinel-1 coherence stack | Raster at 10–20 m posting with P.1812 Annex 1 category codes and representative clutter heights, plus confidence layer |
| Clutter update change mask | Bi-temporal Sentinel-2 spectral change detection and Sentinel-1 coherence difference to flag areas requiring re-acquisition | GeoPackage polygon layer with change-flagged zones and recommended re-tasking priority |
| Path profile extraction | DTM and clutter raster sampled along defined link bearings using standard great-circle or projected geometry | CSV or XML path profile files formatted for direct import into Longley-Rice or P.1812 propagation engines |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.