Backhaul route corridor planning across uncharted terrain

Satellite elevation models and land-cover data let engineers identify viable microwave and fibre-trench corridors before a single survey crew is deployed. This page explains the sensors, methods, honest accuracy limits and atmospheric supplements required.

Sensors

TanDEM-X 12 m DEM: Global digital elevation model at 12 m posting, absolute vertical accuracy better than 10 m (90th percentile), relative accuracy better than 2 m over slopes under 40 degrees. Primary input for Fresnel-zone profiling and slope-gradient analysis. Canopy penetration in X-band is partial: in closed tropical forest the DEM surface sits 5–15 m above true ground, a systematic bias that must be corrected before link profiling.
Sentinel-1 SAR (C-band, 5.4 GHz): 6-day repeat at 10 m resolution in Interferometric Wide swath mode. Supports land-cover classification (distinguishing forest, scrub, open ground, water) for crossing-cost assignment, and coherence-based change detection to flag terrain disturbance along candidate corridors. C-band penetrates light canopy but saturates in dense tropical forest.
ALOS PALSAR-2 (L-band, 1.27 GHz): L-band SAR penetrates forest canopy more deeply than X- or C-band, making it useful for detecting ground topography beneath closed canopy and for mapping forest/non-forest boundaries at 6–10 m resolution. Revisit is 14 days in standard mode. Complements TanDEM-X where canopy-induced DEM bias is the dominant uncertainty.
Sentinel-2 MSI: 13 spectral bands, 10 m resolution in visible and near-infrared, 5-day revisit at the equator. Provides land-cover and vegetation-density layers used to assign cost weights to corridor segments: dense forest, wetland, agricultural land and bare rock each carry different trench or access costs. Cloud cover in tropical zones can limit usable acquisitions to one or two per month.

Why geometry alone is not enough to plan a link

A microwave backhaul link between two nodes is, at first approximation, a geometry problem: is there a clear line of sight, and how much margin exists above the first Fresnel zone? At 7 GHz over a 30 km path, the first Fresnel zone radius at mid-path is roughly 115 m. At 15 GHz over the same distance it narrows to about 80 m. Any terrain or vegetation feature intruding into that zone degrades signal coherence, so the elevation profile must resolve features at roughly that scale or finer.

TanDEM-X at 12 m posting satisfies that resolution requirement comfortably. The practical problem is that the DEM surface in forested terrain is not the ground: it is somewhere between the ground and the canopy top, depending on canopy closure and stand density. In closed tropical forest, published validation studies place the X-band scattering phase centre 5–15 m above ground level. For a link where the Fresnel clearance budget is tight, that bias is operationally significant. Ignoring it can produce a route that appears clear on the model and is obstructed in the field.

Building the corridor cost surface

Route optimisation is not simply finding the shortest path between two nodes. It is minimising a cost surface that combines several independent penalty layers: slope gradient (steeper slopes increase civil-works cost for trench or access road), land-cover type (forest clearance, wetland crossing and agricultural compensation each carry different cost regimes), proximity to existing roads (reducing logistics cost), and geometric suitability for microwave line-of-sight (ridge-following routes may be longer in distance but cheaper in repeater count).

Sentinel-2 MSI provides the land-cover classification. A supervised classifier trained on local spectral signatures can distinguish forest, scrub, grassland, cropland, built-up area and open water at 10 m resolution with overall accuracies typically in the 85–92% range for well-separated classes, though confusion between dense scrub and open forest is common and should be validated with field samples. Sentinel-1 coherence layers add a structural dimension that spectral data alone cannot supply: low coherence between repeat passes often indicates dense, moving vegetation, while high coherence indicates stable bare or built surfaces.

The slope layer comes directly from TanDEM-X. At 12 m posting, gradients steeper than roughly 30 degrees are generally considered prohibitive for conventional fibre trenching without specialist equipment. The corridor optimisation then finds least-cost paths using standard raster cost-distance methods, producing a ranked set of candidate corridors rather than a single deterministic answer. Presenting three or four candidates with their cost breakdowns gives field engineers the information to apply local knowledge that no satellite can supply.

Fresnel-zone profiling at 7 GHz and 15 GHz

Once candidate corridors are identified, each proposed link segment needs a formal Fresnel-zone clearance profile. The standard approach extracts an elevation profile from TanDEM-X along the great-circle path between proposed tower positions, adds a vegetation height correction derived from the PALSAR-2 or Sentinel-1 forest classification, and then computes the first Fresnel zone radius at each point along the path using the standard formula: r1 = sqrt(lambda * d1 * d2 / (d1 + d2)), where lambda is wavelength and d1, d2 are distances from each end.

At 7 GHz (lambda approximately 43 mm) over a 50 km path, the mid-path first Fresnel radius is about 164 m. At 15 GHz (lambda approximately 20 mm) it falls to about 112 m. The ITU-R P.526 recommendation specifies that 60% Fresnel clearance is the minimum for negligible diffraction loss; full clearance is preferable. On ridge-following routes, this is often achievable. In valley crossings it frequently is not, and a repeater or passive reflector must be introduced.

The canopy-correction uncertainty of 5–15 m translates directly into clearance uncertainty. Where the computed clearance margin is less than 20 m, the profile should be flagged for ground-truth survey rather than accepted on the model alone. This is an honest limit of the method, not a deficiency of the DEM.

Where atmospheric physics overrides terrain geometry

ITU-R Recommendation P.530 governs the prediction of fading and outage on terrestrial line-of-sight links, and it introduces effects that no elevation model can capture. Multipath fading arises when atmospheric layering creates multiple propagation paths with different phase delays; the resulting interference can cause deep fades of 30 dB or more on otherwise clear paths. The probability of such fades depends on path length, frequency, terrain roughness, distance from large water bodies and local climate.

Over-water and near-water paths are particularly susceptible. A 40 km link crossing a large lake or coastal inlet in a humid tropical climate may experience multipath outage probabilities an order of magnitude higher than an equivalent inland path over rough terrain. P.530 provides a geoclimatic factor framework that quantifies this, but it requires local refractivity gradient statistics that are not always available for uncharted terrain. In such cases, the conservative approach is to apply the worst-case geoclimatic factor for the climate zone and design fade margin accordingly, typically 30–40 dB for high-availability links.

Rain attenuation is a secondary concern at 7 GHz but becomes significant at 15 GHz in high-rainfall tropical regions. ITU-R P.838 provides specific attenuation coefficients; at 15 GHz, rain rates of 50 mm/hr (common in equatorial climates) produce attenuation of roughly 3–4 dB/km, which is non-trivial on a 20 km path. The satellite-derived corridor analysis should flag paths in high-rainfall climatological zones for explicit rain-fade budget calculation.

Correcting the forest canopy problem

The partial canopy penetration of TanDEM-X X-band radar is the single largest source of systematic error in forested corridor planning. The DEM represents the radar scattering phase centre, which in closed tropical forest sits well above the ground. Published validation work in Southeast Asian and Central African forests places this bias between 5 m and 15 m depending on stand density and moisture content.

Two practical corrections exist. The first uses ALOS PALSAR-2 L-band interferometry, which penetrates deeper into the canopy and produces a scattering phase centre closer to the ground. Differencing the TanDEM-X and PALSAR-2 surfaces gives an approximate canopy height layer, which can then be subtracted from the TanDEM-X DEM to produce a corrected ground model. The accuracy of this approach is limited by temporal mismatch between acquisitions and by the fact that L-band still does not reach the ground in the densest stands. The second approach uses Sentinel-2 canopy-density classification to apply a lookup-table correction by forest type, which is coarser but requires no additional SAR data.

Neither correction eliminates the uncertainty entirely. In the densest forest, residual vertical error of 3–8 m is realistic after correction. For link planning, this means that clearance margins below about 15 m in forested terrain should always trigger a field survey recommendation. Satellize flags these segments automatically in its corridor-planning outputs, drawing on the same analytic pipeline used in the Tonga crop-estimation programme to classify vegetation density from multi-sensor inputs.

What the output looks like and what it cannot replace

A completed corridor analysis delivers a ranked set of candidate routes as GIS vector layers, each annotated with slope statistics, land-cover crossing costs, Fresnel clearance profiles at the specified frequency bands, flagged uncertainty zones requiring ground survey, and a summary of atmospheric risk factors by climate zone. Repeater positions are suggested where clearance deficits are unavoidable. The output is designed to brief a field survey team, not to replace one.

The satellite analysis compresses the pre-survey phase from months of ground reconnaissance to days of desktop analysis. It cannot resolve ambiguities below its resolution floor, cannot see through cloud on optical passes, and cannot substitute for the local knowledge of communities and landowners whose consent determines whether a corridor is actually buildable. Those limits are real. The value is in arriving at the field with three credible options rather than none, and with a clear map of which uncertainties the ground team needs to resolve.

Typical figures

Primary DEM spatial resolution	12 m posting (TanDEM-X global DEM)
DEM vertical accuracy (open terrain)	Absolute < 10 m, relative < 2 m (90th percentile, slopes < 40°)
DEM vertical bias in closed tropical forest	5–15 m above ground (X-band scattering phase centre offset)
Land-cover classification resolution	10 m (Sentinel-2 MSI); 10 m (Sentinel-1 SAR coherence)
SAR revisit for change detection	6 days (Sentinel-1); 14 days (ALOS PALSAR-2)
Fresnel-zone profiling frequency bands	7 GHz and 15 GHz (configurable to other bands on request)
Maximum modelled link distance	Up to 50 km per segment; longer paths require repeater insertion analysis
Atmospheric fading model	ITU-R P.530 multipath and rain fade (ITU-R P.838 at 15 GHz)
Corridor output format	GeoPackage / Shapefile / KMZ with annotated profile CSVs
Archive depth for land-cover baseline	Sentinel-2 from 2015; ALOS PALSAR-2 from 2014; TanDEM-X DEM circa 2010–2015 acquisition epoch

Analytics Satellize can run

Least-cost corridor ranking	Weighted cost-surface raster analysis combining slope, land-cover penalty and geometric suitability layers derived from TanDEM-X and Sentinel-2	Ranked corridor GIS layer (top 3–5 routes) with per-segment cost breakdown
Fresnel-zone clearance profiles	Standard Fresnel radius calculation (ITU-R P.526) applied along extracted TanDEM-X elevation profiles, with canopy-correction layer applied in forested zones	Per-link clearance profile PDF and CSV; flagged obstruction points; suggested tower heights
Canopy-corrected ground model	TanDEM-X minus PALSAR-2 phase-centre differencing or Sentinel-2 density lookup-table correction to reduce systematic forest bias	Corrected DEM GeoTIFF with uncertainty band raster; residual-error flag layer
Land-cover crossing-cost map	Supervised classification of Sentinel-2 MSI with Sentinel-1 coherence as additional feature; class-specific cost weights applied per client tariff schedule	10 m resolution cost-weight raster; class accuracy report with confusion matrix
Atmospheric risk summary	ITU-R P.530 geoclimatic factor computation by path, supplemented by ITU-R P.838 rain-attenuation estimate using climate-zone rainfall statistics	Per-link fade margin recommendation table; high-risk path flags for detailed propagation study
Ground-survey priority map	Uncertainty propagation from DEM bias estimates and classification confidence scores; segments with clearance margin below 15 m in forested zones flagged automatically	Survey-priority GIS layer with justification notes; field-team briefing document

Sources

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.