Urban canyon geometry characterisation for multipath modelling
Street-canyon aspect ratios and facade orientations are the primary geometric inputs to urban ray-tracing propagation models. Very-high-resolution stereo satellites can supply most of that geometry, but not all of it.
Sensors
- WorldView-3 (Maxar): 0.31 m panchromatic GSD, 1.38 m multispectral. Stereo and tri-stereo collection modes yield dense photogrammetric point clouds; nominal absolute horizontal accuracy around 3.5 m CE90 without ground control, improving to sub-metre with GCPs. Revisit approximately 1 day at mid-latitudes.
- Pleiades Neo (Airbus): 0.30 m panchromatic GSD. Four-satellite constellation gives same-day stereo acquisition over a single city. Stereo DSMs typically achieve 0.5 m vertical RMSE in open areas; facade occlusion in canyons narrower than roughly 15 m degrades that figure substantially.
- WorldView Legion (Maxar, in deployment): Six-satellite constellation targeting 0.30 m GSD with revisit rates up to 15 times daily over major cities. Intended to reduce the temporal gap between stereo collections, which matters when construction changes canyon geometry between tasking epochs.
- Airborne oblique photogrammetry (e.g. Vexcel UltraCam Osprey, Leica CityMapper): Typically 5 to 10 cm GSD at 1,000 m AGL. Five-camera oblique systems directly image facade surfaces, recovering parapet heights, setbacks and recessed floors that nadir stereo misses. Used as the authoritative bridge between satellite DSMs and terrestrial lidar.
What a street canyon actually tells a propagation model
Ray-tracing engines for dense urban 4G and 5G deployments, such as those implementing the 3GPP TR 38.901 channel model, require three geometric primitives per city block: the aspect ratio of each canyon (height-to-width, H/W), the horizontal setback of each building face from the kerb, and the azimuth orientation of each facade plane. These inputs determine which reflected and diffracted ray paths reach a receiver, at what delay and with what angle of arrival. Get the H/W ratio wrong by 20 percent and predicted median path loss can shift by several decibels in the 3.5 GHz band, enough to misplace a small-cell antenna by an entire storey.
The geometry is not static. Infill construction, facade cladding changes and rooftop plant additions all alter the scattering environment. A model calibrated on a five-year-old building dataset will drift. That is the practical argument for periodic satellite re-acquisition rather than a one-time survey.
What the 0.30 m satellite limit actually means for facades
WorldView-3 and Pleiades Neo both sit at approximately 0.30 m panchromatic GSD, which is the current commercial optical floor set by US licensing under NOAA Part 960 regulations. At that resolution, a stereo-derived DSM can reliably resolve building outlines, ridge heights and gross rooftop geometry. For canyon characterisation that is genuinely useful: you can extract mean facade height to roughly plus or minus 1 m RMSE in open-frontage streets, and derive H/W ratios for canyons wider than about 10 to 12 m.
The problems start below that width. In a canyon of 8 m or less, the nadir and off-nadir stereo images both look nearly straight down the slot. Facade surfaces are occluded, point-cloud density drops sharply, and the derived wall height at street level can be off by several metres. Oblique airborne photogrammetry partially solves this by imaging facades at 40 to 45 degrees from vertical, but even airborne oblique systems struggle with the lowest one or two floors in very deep canyons. This is an honest geometric limit, not a processing failure.
The photogrammetric pipeline from stereo collect to canyon model
A standard WorldView-3 stereo pipeline runs through image orientation using rational polynomial coefficients supplied with the imagery, dense image matching (semi-global matching or patch-based methods are both common in commercial photogrammetry software), and point-cloud filtering to remove vegetation and moving vehicles before building a digital surface model. The DSM is then differenced against a terrain model, or against a prior bare-earth model, to produce a normalised DSM of building heights.
Canyon geometry extraction follows. Street centrelines from OpenStreetMap or a municipal GIS are buffered to the kerb edge; the normalised DSM is sampled along perpendicular transects at regular intervals, typically every 5 to 10 m, to derive H/W profiles. Facade orientation is taken from the azimuth of each street segment. The output is a polygon layer where each building face carries height, setback and orientation attributes, formatted for direct ingestion into ray-tracing tools such as Wireless InSite or similar commercial propagation solvers.
Accuracy degrades predictably with canyon depth, facade material (glass reflects specularly and confuses dense matching), and the convergence angle of the stereo pair. A convergence angle below about 20 degrees produces shallow parallax and poor height discrimination; above 45 degrees, occlusion increases. Most tasking specifications for urban stereo target 25 to 35 degrees.
Where terrestrial lidar still wins
No current commercial satellite, and no airborne oblique system at typical survey altitude, recovers sub-metre facade detail: window reveals, balcony soffits, recessed ground-floor arcades, or the precise edge geometry of parapet copings. For millimetre-wave deployments above 26 GHz, where Fresnel zone radii shrink to centimetres and specular reflection off individual facade panels matters, these details are not cosmetic. They affect predicted received power by several decibels.
Mobile mapping systems, mounting lidar scanners on a vehicle driving the street, routinely achieve 2 to 5 cm point spacing on facades at kerb distance. Static terrestrial lidar achieves better. The practical workflow for a dense urban 5G deployment is therefore layered: satellite stereo DSM for the city-wide building stock, airborne oblique for canyon-level facade heights and setbacks, and terrestrial lidar for the specific micro-cells where millimetre-wave links will be engineered. Satellite data reduces the terrestrial survey to a targeted campaign rather than a city-wide one. That is where the cost argument sits.
Honest limits and common misreadings
Satellite-derived DSMs are often described as if they were building models. They are not. A DSM is a surface; it does not know whether a height anomaly is a parapet, a rooftop HVAC unit, a billboard or a tree. Post-processing to classify and regularise building outlines introduces its own errors, typically 0.5 to 2 m on facade height depending on urban morphology and the density of the photogrammetric point cloud.
Cloud cover is a real operational constraint. A single stereo collect requires a cloud-free window over the entire target area, which in tropical cities or high-latitude winters may mean waiting weeks for a usable acquisition. Archive stereo imagery can substitute, but building stock changes; using a DSM that is more than two or three years old in a rapidly developing district introduces systematic error into the canyon model.
Satellize can run the DSM-to-canyon-geometry pipeline on WorldView-3 or Pleiades Neo stereo collects acquired under client licence, delivering attributed facade polygon layers in standard GIS formats. The Tonga crop-estimation programme is a different domain, but the underlying workflow, tasking commercial sensors and running structured extraction pipelines, is the same.
Specifying a canyon survey: what to tell your data provider
The most common mistake in commissioning satellite stereo for propagation work is under-specifying the convergence angle and the off-nadir limit. Requesting 'stereo imagery' without constraining these parameters can result in pairs with poor height sensitivity or excessive occlusion. A reasonable specification for canyon work: convergence angle 25 to 35 degrees, maximum off-nadir 30 degrees for either image in the pair, sun elevation above 30 degrees to limit shadow occlusion, and cloud cover below 10 percent over the area of interest.
For the canyon-model output itself, propagation engineers typically need facade polygons attributed with mean height, height standard deviation along the facade, setback from kerb centreline, and facade azimuth, delivered as a GeoPackage or shapefile with a defined coordinate reference system. Agreeing that schema before acquisition avoids a rework cycle that can cost more time than the survey itself.
Typical figures
| Best commercial optical GSD (nadir) | 0.30 m (WorldView-3, Pleiades Neo) |
| Stereo DSM vertical accuracy (open streets, with GCPs) | 0.5 to 1.0 m RMSE typical; degrades to 2 to 5 m in canyons narrower than ~10 m |
| Stereo convergence angle (recommended for canyon work) | 25 to 35 degrees |
| Revisit (WorldView-3 single satellite) | Approximately 1 day at mid-latitudes |
| Revisit (WorldView Legion, target) | Up to 15 times daily over major cities |
| Minimum canyon width reliably characterised from nadir stereo | Approximately 10 to 12 m; narrower canyons require oblique or terrestrial survey |
| Airborne oblique GSD (typical city survey) | 5 to 10 cm at 1,000 m AGL |
| Archive depth (WorldView-3) | Imagery available from 2014; stereo pairs subject to prior tasking availability |
| Standard delivery formats | GeoTIFF DSM, LAS/LAZ point cloud, GeoPackage or shapefile polygon layer |
| Spectral bands used for canyon extraction | Panchromatic (stereo matching); multispectral used for facade material classification |
Analytics Satellize can run
| City-wide normalised DSM | Semi-global matching on WorldView-3 or Pleiades Neo stereo pairs; vegetation and vehicle filtering by height and NDVI thresholding | GeoTIFF raster at 0.5 m pixel spacing, clipped to licence area |
| Building-footprint height attribution | DSM-minus-terrain differencing; footprint regularisation using morphological operators against municipal or OSM building polygons | GeoPackage polygon layer with mean height, max height and height standard deviation per building |
| Street-canyon aspect ratio profiles | Perpendicular transect sampling of normalised DSM along street centrelines at 5 m intervals; H/W ratio computed per transect | CSV and GeoJSON of transect points with H/W, canyon width and azimuth attributes |
| Facade orientation and setback layer | Building edge extraction from regularised footprints; setback measured from kerb centreline buffer; facade azimuth from edge vector | Line geometry GeoPackage attributed with height, setback and azimuth, formatted for Wireless InSite or equivalent ray-tracer import |
| Change detection: canyon geometry update | DSM differencing between two stereo epochs; change polygons where height delta exceeds 2 m threshold | Change polygon layer with height-delta values and acquisition dates, for targeted re-survey prioritisation |
| Facade material classification | WorldView-3 eight-band multispectral classification (glass, concrete, brick, metal cladding) using published spectral libraries; output used to flag specular-reflection risk zones for ray-tracing | Classified facade polygon layer with dominant material class and confidence score |
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.