Road surface condition assessment for maintenance crew access planning
Very-high-resolution optical imagery and SAR coherence change detection combine to flag unpaved access roads degraded by erosion, debris or inundation, so telecoms operators know which tower sites face crew-access delays before dispatching a vehicle.
Sensors
- Pleiades Neo (Airbus): 30 cm panchromatic, 1.2 m multispectral native resolution; tasked revisit of 1–2 days over a given site. Resolves pavement cracking, rut depth shadow, debris scatter and surface ponding at a level of detail sufficient for visual distress classification.
- WorldView-3 (Maxar): 31 cm panchromatic, 1.24 m 8-band multispectral, plus SWIR and CAVIS atmospheric bands. SWIR bands (1.195–2.365 µm) help distinguish wet soil from dry surface, aiding inundation extent mapping on unpaved surfaces.
- Sentinel-1 (ESA): C-band SAR (5.405 GHz), IW mode ground range resolution approximately 5 × 20 m, 6-day repeat at mid-latitudes with both satellites. Interferometric coherence computed between pre- and post-event image pairs detects surface disruption on roads even through cloud cover; coherence loss below roughly 0.3 is a reliable indicator of surface change.
- Planet SuperDove: 3 m multispectral, near-daily revisit globally. Useful for rapid change flagging and temporal confirmation of whether water ponding is persistent or transient, though resolution is insufficient for pavement-crack detection.
Why access-road condition is a network-operations problem, not a civil-engineering one
A tower site that is structurally intact but physically unreachable is, for practical purposes, down. Maintenance crews in remote or rural terrain routinely face unpaved tracks that deteriorate faster than any inspection schedule can track. A single heavy rainfall event can deposit landslide debris across a road, cut an erosion channel deep enough to ground a standard 4x4, or leave standing water that conceals a collapsed culvert. The operational consequence is a delayed repair, an extended outage, and a service-level breach.
Satellite data changes this from a reactive problem to a planned one. The combination of sub-0.5 m optical imagery and SAR coherence analysis can produce a condition flag for every access road in a tower portfolio within 24 to 48 hours of a significant rainfall or seismic event, long before a crew is dispatched. That flag does not replace a ground inspection; it tells the operations centre which sites need one urgently and which can wait.
What a floating roof gives away: reading surface distress in sub-metre imagery
Pleiades Neo at 30 cm and WorldView-3 at 31 cm panchromatic resolution resolve the texture signatures of unpaved road distress directly. Ruts cast narrow shadows detectable as parallel dark lines. Erosion channels crossing the carriageway appear as irregular bright-edged incisions. Debris from slope failure shows as a tonal and textural anomaly against the surrounding road surface. Surface ponding appears as specular bright patches in low-sun geometries or as dark, smooth areas at higher sun angles.
Automated classification uses a combination of object-based image analysis and supervised texture classifiers trained on labelled distress categories. The honest limit here is that classification accuracy degrades on roads narrower than roughly 2 metres, where road pixels mix with roadside vegetation even at 30 cm resolution. Dense tree canopy overhanging the carriageway is a harder problem: the road surface beneath is simply not visible, and no optical sensor resolves that. SAR fills part of this gap.
Coherence loss as a proxy for surface disruption
Sentinel-1 interferometric coherence measures how similar the radar backscatter phase is between two acquisitions over the same area. On a stable, dry, unpaved road surface, coherence between 6-day pairs is typically moderate to high (above 0.5 in C-band). When the surface is disturbed, whether by debris deposition, erosion, or inundation, the coherence drops, often sharply. Studies using Sentinel-1 IW coherence have shown that road-surface disruption after landslide events produces coherence values below 0.3 in affected segments.
The critical caveat: coherence loss is not specific to roads. Vegetation growth, agricultural activity, and even strong winds cause coherence loss in adjacent pixels. Road segments must be masked from a pre-existing road network layer before coherence statistics are extracted, so the analysis is comparing road pixels to their own prior state rather than to surrounding land cover. And a single low-coherence observation after rain cannot distinguish temporary ponding from permanent damage. That is why the workflow requires at least two post-event acquisitions, 6 days apart, to confirm persistence.
Combining the two: a prioritisation score, not a binary pass/fail
The operational output is a ranked list of tower access roads, scored by estimated severity of disruption. The scoring draws on three inputs: coherence change magnitude from Sentinel-1, distress classification confidence from optical imagery where cloud-free acquisitions exist, and a pre-existing vulnerability index derived from slope angle, drainage catchment area, and historical rainfall intensity for the road segment.
High-priority flags trigger a tasking request for fresh Pleiades Neo or WorldView-3 imagery if the most recent optical acquisition predates the event. Medium-priority flags are reviewed against Planet SuperDove's near-daily archive to check whether the anomaly is persistent. Low-priority roads receive no immediate action. This tiered approach is necessary because commercial tasking at sub-0.5 m resolution carries a cost per square kilometre that makes blanket coverage of an entire access-road network impractical for routine operations.
Planet SuperDove's 3 m resolution is not fine enough to classify pavement distress, but it is sufficient to detect whether a road corridor is inundated and whether that inundation has cleared. It serves as a low-cost screening layer between SAR alerts and expensive optical tasking.
Honest limits of the method
Cloud cover is the most persistent operational constraint. Persistent cloud in tropical or monsoon climates, precisely the conditions that cause road damage, can block optical acquisitions for days or weeks. SAR penetrates cloud, which is why Sentinel-1 coherence is the primary trigger layer rather than a secondary one. But SAR at 5 × 20 m resolution cannot see a 1-metre erosion channel. The two sensors are complementary, not interchangeable.
Revisit is the second constraint. Sentinel-1's 6-day repeat means that a road damaged on day one may not appear in a coherence pair until day six, and confirmation of persistence requires day twelve. For emergency response this can be too slow; for routine maintenance planning it is generally adequate. Pleiades Neo and WorldView-3 can be tasked within 24 to 48 hours, but cloud may prevent acquisition. There is no sensor combination that guarantees cloud-free sub-metre imagery on demand.
Finally, the method produces access-difficulty flags, not engineering assessments. A satellite-derived score of 'high disruption' tells the operations centre to send a scout vehicle or request a drone inspection before committing a heavy maintenance crew. It does not specify repair method, load-bearing capacity, or safe vehicle type. Satellize applies this same principle in its analytics work, including the Kingdom of Tonga crop-estimation programme, where satellite outputs inform decisions rather than replace field verification.
From alert to dispatch: what the deliverable actually looks like
The standard deliverable is a GIS layer, updated after each Sentinel-1 acquisition pair and after each cloud-free optical tasking, showing road segments coloured by condition tier. Each segment carries an attribute table with coherence delta, optical distress confidence score, date of last cloud-free observation, and a recommended action flag. This feeds directly into a field operations management system or a network operations centre dashboard.
For operators with large tower portfolios across difficult terrain, a weekly condition report summarising new high-priority flags and any changes in status from the previous cycle is a practical complement to the live GIS feed. The report format is designed to be read by an operations manager, not a remote-sensing analyst.
Typical figures
| Optical resolution (panchromatic) | 30 cm (Pleiades Neo), 31 cm (WorldView-3) |
| Optical resolution (multispectral) | 1.2 m (Pleiades Neo), 1.24 m (WorldView-3), 3 m (Planet SuperDove) |
| SAR resolution (Sentinel-1 IW mode) | Approximately 5 m range × 20 m azimuth ground range |
| SAR revisit (Sentinel-1, dual satellite) | 6 days at mid-latitudes; up to 12 days near equator depending on orbit geometry |
| Optical tasking revisit | 1–2 days (Pleiades Neo, WorldView-3); near-daily (Planet SuperDove) |
| Coherence-change detection latency | 6–12 days post-event (two Sentinel-1 acquisitions required for confirmation) |
| Minimum detectable road disruption (optical) | Features wider than approximately 0.6 m on roads wider than 2 m carriageway |
| Coherence disruption threshold | Coherence drop below approximately 0.3 (C-band, 6-day pair) used as alert trigger |
| Archive depth | Sentinel-1: from 2014; Pleiades/WorldView-3: varies by prior tasking; Planet: from approximately 2016 |
| Delivery format | GeoPackage or Shapefile road-segment layer, GeoTIFF coherence difference raster, PDF condition report |
Analytics Satellize can run
| Post-event road condition alert layer | Sentinel-1 interferometric coherence change detection, road-mask extraction from OpenStreetMap or operator GIS | GIS vector layer of flagged road segments with coherence delta attributes, delivered within 24 hours of SAR pair availability |
| Pavement distress classification | Object-based image analysis and supervised texture classification on Pleiades Neo or WorldView-3 imagery | Raster distress map and vector segment summary at per-100 m road section granularity |
| Inundation persistence confirmation | Multi-temporal NDWI and visual change detection on Planet SuperDove daily archive | Temporal inundation profile per flagged segment, distinguishing transient ponding from persistent blockage |
| Access-priority ranking | Weighted scoring combining coherence delta, optical distress confidence, slope-derived vulnerability index | Ranked tower-site list with recommended action tier (immediate scout, scheduled inspection, no action) |
| Seasonal access-risk baseline | Historical Sentinel-1 coherence time-series analysis correlated with rainfall climatology for road network | Annual risk calendar per road segment identifying highest-probability disruption windows |
| Weekly maintenance-planning report | Automated aggregation of GIS layer changes with plain-language condition summaries | PDF or HTML report formatted for operations managers, updated on each Sentinel-1 acquisition cycle |
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.