Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands; 5-day revisit at the equator with twin satellites. Used to map road surface extent, adjacent vegetation density via NDVI, and canopy gap width along road corridors.
- VIIRS Day/Night Band (DNB): Approximately 750 m spatial resolution at nadir; nightly acquisition. Radiance values serve as a proxy for traffic intensity and road illumination, distinguishing high-use highways from low-traffic rural tracks. Not a direct vehicle count.
- Maxar WorldView series: 30–50 cm panchromatic resolution with multispectral at 1.2–1.6 m. Used for precise road-width measurement, lane-count confirmation, and detection of existing crossing structures such as culverts or bridges that Sentinel-2 cannot resolve.
- OpenStreetMap-derived road network: Volunteer-contributed vector layer used as a structural baseline and validation check for satellite-extracted road geometry. Coverage and accuracy vary by region; requires field verification in poorly mapped areas.
What a road looks like from orbit, and why that matters
A road is one of the more legible features in satellite imagery. At 10 m resolution, Sentinel-2 resolves most two-lane highways as a spectrally distinct strip: low NDVI, high reflectance in the red and SWIR bands relative to surrounding vegetation. Road width, however, is the critical variable for mortality risk, and a two-lane track at 7 m carriageway sits at the margin of what Sentinel-2 can measure reliably. Width estimates derived from Sentinel-2 carry uncertainty of roughly one to two pixels, meaning a reported 10 m road could be anywhere from 8 to 20 m in practice.
This is where very-high-resolution commercial imagery earns its cost. Maxar WorldView-3, with 30 cm panchromatic imagery, can resolve lane markings, road shoulders, and median barriers. It can also detect existing culverts and underpasses, which are invisible to medium-resolution sensors but are critical inputs to any crossing-structure audit. The trade-off is cost and revisit: WorldView tasking is expensive and non-systematic, so it is best applied to candidate hotspot segments identified first by Sentinel-2 analysis rather than to entire road networks.
Nighttime light as a traffic-intensity proxy: useful, imperfect
VIIRS DNB records emitted radiance at night with roughly 750 m ground resolution. On a major intercity highway, the signal integrates headlight flux, roadside lighting, and petrol-station glow into a composite brightness that correlates broadly with traffic volume. The Colorado School of Mines VIIRS Nightfire programme and NOAA STAR both publish calibrated monthly composites that analysts can use directly.
The proxy has real limits. Rural roads with high wildlife-crossing risk are often unlit and carry modest traffic, producing DNB radiance close to background noise. A lightly used logging road through primary forest may be far more dangerous to large mammals than a brightly lit urban bypass, yet the bypass will dominate the nighttime signal. DNB is therefore most useful for separating major arterials from minor tracks at the network scale, not for fine-grained risk ranking within a single corridor. Analysts should treat it as one layer in a weighted model, not as a stand-alone risk index.
Building the least-cost corridor and locating the pinch points
Least-cost corridor modelling assigns a resistance value to each land-cover cell and finds the path of minimum cumulative resistance between two habitat patches. Roads enter the model as barriers with resistance scaled by width, traffic proxy, and fencing status. The output is a corridor probability surface: cells with low cumulative cost are likely movement paths; cells where a high-resistance road intersects a low-cost corridor are structural mortality hotspots.
The satellite layers feed this model in a specific sequence. Sentinel-2 NDVI composites define habitat quality and canopy connectivity in the matrix around the road. Road geometry from Sentinel-2 or WorldView defines barrier width. DNB radiance scales the traffic-intensity penalty. The resulting resistance raster is then processed through standard circuit-theory or least-cost-path tools, both of which are well-documented in the published literature on landscape connectivity.
One honest caveat: the model identifies structural risk, meaning places where the geometry of road and habitat makes crossing likely and dangerous. It does not predict actual mortality rates, which depend on animal behaviour, seasonal movement patterns, and species-specific road-avoidance responses that satellites cannot observe. Carcass records, camera-trap detections, or GPS-collar crossing data are needed to calibrate and validate the hotspot predictions. Without that ground truth, the model is a prioritisation tool, not a mortality forecast.
Ranking crossing structures by connectivity gain
Not all hotspots are equal. A road segment that bisects the only functional corridor between two large forest patches has far higher conservation value as a crossing-structure site than a segment crossing a marginal habitat strip. Quantifying this difference requires comparing corridor probability surfaces with and without a simulated crossing structure at each candidate location, then ranking by the improvement in effective connectivity.
This comparison can be run as a sensitivity analysis across a grid of candidate locations, producing a ranked list of sites where an underpass or overpass would most improve landscape permeability. The ranking is sensitive to the resistance values assigned to road categories, so analysts should run scenarios with different traffic-penalty weightings and report the range of outcomes rather than a single deterministic rank. Sites that appear in the top tier across all scenarios are the most defensible recommendations for infrastructure investment.
Vegetation cover on the approaches matters too. An underpass flanked by bare cleared shoulders is far less effective than one with continuous canopy or dense shrub cover leading to the entrance. Sentinel-2 NDVI at 10 m resolution can characterise approach vegetation within roughly 50 to 100 m of each candidate structure, flagging sites where revegetation would be a necessary co-investment.
Archive depth and change detection
Sentinel-2 data runs back to 2015, and Landsat extends the optical archive to 1972. This depth allows analysts to detect road widening events, new road construction through previously intact corridors, and vegetation clearance along road margins, all of which change the risk profile of a corridor over time. A road that appears in a current hotspot analysis may have been a low-risk track a decade ago; the archive reveals the trajectory.
Change detection on road networks is straightforward for major construction events but becomes ambiguous for smaller changes such as shoulder widening or vegetation trimming. Sentinel-2 at 10 m struggles to distinguish a 2 m shoulder expansion from seasonal vegetation variation at the road edge. For precise before-and-after measurement of road geometry changes, WorldView archive imagery is more reliable, though archive availability at specific locations is not guaranteed.
From analysis to infrastructure brief
The practical output of this workflow is a prioritised list of road segments with supporting spatial evidence: corridor probability maps, resistance surfaces, DNB-derived traffic classification, vegetation approach scores, and existing-structure inventory. Satellize packages this as a GIS layer set with an accompanying technical brief that a transport ministry or conservation agency can take directly into a feasibility study for crossing-structure design.
The workflow draws on the same open-constellation infrastructure that Satellize uses in its Tonga crop-estimation programme, with commercial tasking added where road-width precision demands sub-metre imagery. The analysis identifies where to invest in crossing structures; it does not replace the ecological fieldwork needed to confirm which species are at risk and whether a proposed structure is the right design for them. That distinction is worth stating plainly to any client who receives the outputs.
Typical figures
| Road-mapping spatial resolution | 10 m (Sentinel-2); 30–50 cm panchromatic (WorldView-3) |
| Sentinel-2 revisit | 5 days at equator (Sentinel-2A + 2B combined) |
| VIIRS DNB resolution | ~750 m at nadir; nightly acquisition |
| Sentinel-2 spectral bands used | Band 4 (Red), Band 8 (NIR) for NDVI; Band 11/12 (SWIR) for surface discrimination |
| Minimum detectable road width (Sentinel-2) | ~10 m carriageway with confidence; narrower tracks detectable but width uncertainty ±1–2 pixels |
| Sentinel-2 archive depth | 2015 to present; Landsat extends optical record to 1972 |
| VIIRS DNB archive depth | 2012 to present (Suomi NPP); monthly composites available via NOAA STAR and EOGDATA |
| Crossing-structure detection | Culverts and underpasses not reliably detectable below ~2 m aperture at any freely available resolution; WorldView required |
| Deliverable formats | GeoTIFF resistance surfaces, vector hotspot layers (GeoPackage/Shapefile), PDF technical brief |
| Coverage | Global for Sentinel-2 and VIIRS; WorldView tasking subject to acquisition geometry and licensing |
Analytics Satellize can run
| Road-network extraction and classification | Spectral unmixing and object-based image analysis on Sentinel-2 10 m composites; WorldView panchromatic edge detection for width measurement | Vector road layer with carriageway-width estimates and surface-type classification (paved/unpaved) |
| Traffic-intensity classification | VIIRS DNB monthly composite radiance binned into low/medium/high classes; cross-checked against OSM road hierarchy | Raster traffic-proxy layer at 750 m; tabular classification per road segment |
| Vegetation approach-quality scoring | NDVI computed from Sentinel-2 within 50 m and 100 m buffers of each candidate crossing-structure site; seasonal compositing to reduce cloud artefacts | Per-site approach-vegetation score appended to hotspot vector layer |
| Resistance surface for least-cost modelling | Multi-layer weighted combination of road width, traffic class, fencing presence, and habitat quality; weights documented and adjustable | GeoTIFF resistance raster at 10 m; scenario variants for different traffic-penalty weightings |
| Corridor probability surface and hotspot delineation | Circuit-theory or least-cost-path analysis on resistance raster between user-defined habitat patch centroids | Corridor probability GeoTIFF; vector hotspot polygons ranked by crossing-risk score |
| Crossing-structure prioritisation ranking | Connectivity-gain sensitivity analysis: corridor probability compared with and without simulated structure at each candidate site across resistance-weight scenarios | Ranked site table with connectivity-gain range, approach-vegetation score, and existing-structure flag; PDF infrastructure-brief summary |
| Road-change detection | Multi-temporal Sentinel-2 differencing to detect new road construction or carriageway expansion; WorldView archive comparison for precise width-change measurement | Change-detection map with flagged segments and before/after imagery chips |
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.