Coverage-gap analysis for universal service fund programmes

Satellite-derived elevation and land-cover data let regulators run independent propagation models to validate or challenge operator coverage claims in USF and rural-broadband subsidy processes, without relying solely on operator-submitted signal maps.

Sensors

SRTM (Shuttle Radar Topography Mission): 1 arc-second (~30 m) global DEM, the standard free-to-use terrain input for radio propagation modelling. Vertical accuracy is nominally ±16 m absolute at 90% confidence over most land surfaces, though forested terrain introduces positive bias because the C-band radar reflects off canopy, not ground.
TanDEM-X: Commercial X-band bistatic DEM at 12 m or 0.4 arcsec horizontal posting, with relative vertical accuracy better than 2 m over open terrain. The improved accuracy matters for line-of-sight calculations in hilly terrain where SRTM error is large relative to Fresnel-zone clearance requirements.
Sentinel-2 MSI: 10 m multispectral imagery with 13 bands and a 5-day revisit at the equator. Used to classify land cover into clutter categories (urban, suburban, forest, open rural, water) that drive the clutter-loss tables in propagation models such as ITU-R P.1546 and the Longley-Rice irregular terrain model.
VIIRS Day/Night Band (DNB): 500 m nighttime radiance at roughly 0.5 nW/cm²/sr detection sensitivity, daily global coverage. Used as a proxy for electrification and settlement presence to cross-check whether claimed coverage areas actually contain addressable population, and to flag lit clusters absent from operator coverage polygons.

What operators submit and why it deserves scrutiny

In most USF and rural-broadband subsidy programmes, operators self-certify coverage using in-house propagation models run on proprietary terrain data, with parameters they select. The regulator receives a polygon or raster asserting, say, outdoor 4G coverage at a minimum signal threshold. What the regulator rarely receives is the model configuration, the clutter-loss assumptions, the antenna tilt settings, or any drive-test validation. The incentive to optimise those parameters toward maximum claimed coverage is obvious: subsidy disbursement often depends on coverage area delivered.

This is not necessarily fraud. Propagation modelling is genuinely uncertain, and two competent engineers using the same underlying terrain data can produce coverage maps that differ by several decibels in shadowed terrain. The problem is that without an independent reference, a regulator cannot distinguish optimistic-but-defensible modelling from deliberate overstatement.

How satellite data builds an independent propagation baseline

An independent coverage model starts with a digital terrain model and a land-cover classification. SRTM at 30 m is the standard public input; TanDEM-X at 12 m is the upgrade when budget allows and terrain relief is sharp enough to matter. Sentinel-2 MSI provides the clutter layer: a supervised classification separating dense urban, low-density suburban, forest, agriculture, and open terrain. Each class carries a median clutter-loss value drawn from ITU-R P.2108 or operator-independent empirical tables.

The propagation engine, typically the Longley-Rice Irregular Terrain Model or ITU-R P.1546, then computes received signal level at each grid cell given the tower location, antenna height, transmit power, and frequency band. Tower locations are taken from the operator's own licence register or from a national infrastructure database, not from the operator's coverage submission. The result is a predicted outdoor coverage surface that is reproducible, documented, and independent of the operator's choices.

Reproducibility is the key word. When a regulator challenges a coverage claim, the dispute must be resolvable by reference to documented inputs and methods. A satellite-derived model with published parameters can be audited. An operator's black-box submission cannot.

The gap between outdoor prediction and indoor reality

Satellite-derived propagation models predict outdoor signal levels. They say nothing reliable about indoor coverage, which is what most USF programmes are actually trying to deliver to rural households. Building penetration loss at 700 MHz is typically 10 to 20 dB for lightweight rural construction; at 3.5 GHz it can exceed 30 dB. A cell that clears the outdoor coverage threshold by 3 dB fails indoors in most building types.

This distinction matters for what satellite analysis can and cannot adjudicate. It can identify geographic areas where the terrain and clutter geometry make even outdoor coverage geometrically implausible given the declared tower configuration. It cannot, without drive-test or stationary measurement data, determine whether a borderline outdoor signal is adequate for the indoor use cases the subsidy is meant to fund. Honest analysis keeps these two claims separate.

Where the gaps actually appear: terrain shadow and clutter accumulation

In practice, the most contested coverage claims cluster in two terrain types. First, valleys and hollows behind ridge lines where a single tower on high ground appears to cover a large area on a flat-earth map but casts deep radio shadows when terrain diffraction is properly computed. SRTM at 30 m resolves most valley floors that matter for rural connectivity; the residual error is in narrow gorges where the DEM smooths the true ridge profile.

Second, forested areas where canopy height adds 10 to 25 dB of clutter loss that a bare-earth DEM ignores. SRTM's C-band phase centre sits 5 to 20 m above ground in closed-canopy forest, meaning the DEM already partially captures canopy height, but inconsistently. Sentinel-2 NDVI and band-ratio classification identifies forested pixels so that an appropriate excess loss can be applied. For high-value disputes, GEDI lidar canopy-height data from the ISS provides direct canopy height at 25 m footprint spacing, published through NASA Earthdata.

VIIRS nighttime light adds a population-presence check. If a cluster of lit pixels falls inside an operator's claimed coverage polygon but the propagation geometry shows terrain obstruction, that is a flagged discrepancy worth field investigation. If a lit cluster falls entirely outside any operator's claimed coverage, that is a candidate for USF-funded intervention regardless of the coverage dispute.

What the analysis delivers and what it cannot replace

The concrete output is a gridded coverage-probability surface, typically at 100 m or 200 m posting, showing cells where the independent model agrees with the operator claim, cells where it disagrees by more than a defined threshold (commonly 6 dB), and cells where the operator claims coverage but the model predicts none. That surface is delivered as a GeoTIFF or vector polygon set, with a companion report documenting every modelling assumption.

Disagreement cells are candidates for drive-test validation, not automatic disqualification. The satellite model has its own errors: antenna configuration uncertainty, DEM vertical error in complex terrain, clutter-loss table scatter. A well-designed USF audit protocol uses the satellite model to triage where drive tests are worth deploying, not to replace them. Satellize has applied this triage logic in its analytics work, including in geographically complex island environments like the Kingdom of Tonga crop-estimation programme, where terrain and clutter heterogeneity make blanket field survey impractical.

The honest limit of the method is this: satellite-derived propagation analysis is a necessary condition for credible USF oversight, not a sufficient one. It closes the information gap between regulator and operator on terrain geometry. It does not close the gap on measured signal quality, device behaviour, or network load.

Running a defensible audit: inputs, sequence, and documentation

A credible satellite-based coverage audit requires five documented inputs: the tower location register (from licence data, not the operator's coverage submission), the declared frequency band and EIRP per sector, the SRTM or TanDEM-X DEM with its known vertical accuracy statement, the Sentinel-2 derived clutter classification with its training-class accuracy matrix, and the propagation model specification with the ITU-R or published empirical source for each loss term.

The sequence is: terrain pre-processing to fill SRTM voids and clip to the study area; clutter classification from Sentinel-2 with a minimum mapping unit of 1 ha; propagation run per tower sector; mosaic and threshold to produce the binary coverage layer; overlay with the operator submission to produce the discrepancy layer; VIIRS overlay to flag lit settlements in discrepancy zones. The entire chain should be version-controlled and reproducible from the same inputs by a third party. That reproducibility is what makes the output usable in a regulatory or legal proceeding.

Typical figures

Terrain model resolution (standard)	SRTM 1 arc-second (~30 m); vertical accuracy ±16 m absolute (90th percentile, open terrain)
Terrain model resolution (upgraded)	TanDEM-X 12 m posting; relative vertical accuracy <2 m over open terrain
Land-cover classification resolution	Sentinel-2 MSI at 10 m; 5-day revisit; 13 spectral bands from 443 nm to 2190 nm
Nighttime light proxy resolution	VIIRS DNB ~500 m; daily global coverage; detection threshold ~0.5 nW/cm²/sr
Propagation model output grid	Typically 100 m or 200 m posting; finer posting increases compute cost without improving DEM accuracy
Frequency bands supported	700 MHz through 3.5 GHz (rural USF typical); ITU-R P.1546 valid 30 MHz to 4 GHz
Archive depth (Sentinel-2)	2015 to present; enables multi-season clutter classification to capture deciduous canopy variation
Minimum detectable coverage gap	Terrain-shadow gaps resolvable to approximately 200 m width at 30 m DEM; narrower gaps require TanDEM-X
Delivery format	GeoTIFF (coverage probability raster), GeoPackage or Shapefile (discrepancy polygons), PDF audit report with documented assumptions

Analytics Satellize can run

Independent outdoor coverage surface	Longley-Rice Irregular Terrain Model or ITU-R P.1546, driven by SRTM or TanDEM-X DEM and Sentinel-2 clutter classification	GeoTIFF raster of predicted received signal level per grid cell, one layer per operator and frequency band
Operator claim discrepancy layer	Cell-by-cell comparison of operator-submitted coverage polygon against independent propagation surface, flagging cells diverging by more than a defined dB threshold	Vector polygon GIS layer of agreed, disputed, and uncontested-gap zones, with discrepancy magnitude attribute
Lit-settlement gap flag	VIIRS DNB radiance thresholding to identify electrified settlements, intersected with coverage discrepancy layer	Point or polygon layer of lit clusters falling in coverage gaps, ranked by estimated population exposure
Clutter-loss classification	Supervised classification of Sentinel-2 MSI (bands 2, 3, 4, 8, 11, 12) into ITU-R P.2108 clutter categories; accuracy assessed by confusion matrix	10 m land-cover raster with per-class clutter-loss lookup table and classification accuracy report
Drive-test triage prioritisation	Spatial scoring of discrepancy cells by population density proxy (VIIRS), road accessibility, and discrepancy magnitude to rank field-validation effort	Ranked site list with coordinates and justification scores, formatted for field-team tasking
Canopy height correction layer	GEDI Level 2A relative height metrics at 25 m footprint, gridded and gap-filled using Sentinel-2 regression, applied as excess clutter loss over forested pixels	Canopy-corrected propagation surface and difference map showing where canopy correction materially changes coverage prediction

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.