Post-earthquake road network accessibility assessment
After a major earthquake, knowing which roads are passable can determine whether aid reaches survivors within hours or days. This page explains how sub-metre optical and commercial SAR imagery, combined with road-network graph analysis, produces actionable accessibility maps.
Sensors
- Maxar WorldView-3: Panchromatic resolution of 0.31 m, multispectral at 1.24 m. The panchromatic band is the workhorse for road-damage detection: at this resolution a blocked two-lane road (typically 6–7 m wide) spans roughly 20 pixels, enough to distinguish rubble texture from pavement. Off-nadir tasking up to 45° enables same-day collection over a target, though steep off-nadir angles worsen shadow occlusion on narrow roads in terrain.
- Airbus Pleiades Neo: Native panchromatic resolution of 0.30 m, four-band multispectral at 1.20 m, with a daily revisit per satellite (four-satellite constellation as of 2023). Pleiades Neo's stereo and tri-stereo tasking modes allow a surface model to be derived from a single pass, which helps distinguish rubble height from flat debris and supports volume estimation of blockages.
- ICEYE SAR (X-band): Spotlight mode delivers approximately 0.5 m ground resolution; strip mode 3 m. SAR is insensitive to cloud cover, which is critical because post-earthquake weather is often poor. X-band backscatter changes over road surfaces reveal coarse rubble and displaced material, though narrow roads under tree canopy remain difficult to assess. ICEYE can task and deliver imagery within hours of an event.
- Capella Space SAR (X-band): Spotlight resolution to approximately 0.35 m (published Spotlight Ultra mode), with rapid tasking turnaround. At this resolution, SAR can detect large debris piles and surface rupture scarps crossing roads, though interpreting backscatter on narrow unpaved tracks requires careful comparison against a pre-event baseline collected under similar incidence angles.
- Copernicus Sentinel-2 (baseline context only): 10 m multispectral resolution, free and open, two-to-five day revisit. Useful for mapping the broader landslide and surface-rupture footprint that informs which road corridors are likely affected, but wholly insufficient for detecting individual road blockages. It sets the spatial context; the commercial sensors do the damage detection.
The resolution floor below which damage disappears
A standard two-lane rural road in an earthquake-affected country is 5.5 to 7 metres wide. A blockage may be a rubble pile covering 3 metres of that width, or a tension crack 0.2 metres across running diagonally across the surface. To detect the crack reliably, you need pixels smaller than 0.5 m. To detect the rubble pile, 1 m imagery is marginal and 0.5 m is workable. Anything coarser than 1 m per pixel, including all free Sentinel and Landsat imagery, cannot reliably identify individual road blockages. It can identify the landslide that caused them, but not whether the road beneath is passable.
This is not a conservative estimate. It follows directly from the Nyquist principle applied to spatial data: a feature must span at least two pixels to be detected with confidence. A 3-metre rubble pile on a 6-metre road spans three pixels at 1 m resolution, which is detectable in ideal conditions. Add radiometric noise, compression artefacts, and the fact that rubble and road surface can have similar spectral signatures, and the practical detection floor for blockages sits at 0.5 m or better for reliable automated classification.
What shadow and geometry hide
Shadow is the underappreciated enemy of road-damage assessment. A narrow road running through a gorge or alongside a multi-storey building may be in shadow for most of the day. WorldView-3 and Pleiades Neo both collect in the visible spectrum, so a shadowed road surface loses most of its diagnostic information. The problem is worst in the first hours after an event, when off-nadir tasking is often necessary to achieve rapid coverage, and steep off-nadir angles push building shadows further across road surfaces.
SAR partly solves this because it is its own illumination source, but SAR introduces layover and foreshortening in steep terrain. A road running along a hillside facing away from the satellite is in radar shadow. No single orbital geometry solves all terrain configurations, which is why multi-sensor and multi-pass approaches matter. Practically, this means combining a nadir-to-moderate off-nadir optical collect for urban and flat terrain with an ascending and descending SAR pair for mountainous corridors.
Viewing geometry also affects road width in the image. A road at 30° off-nadir appears narrower in the cross-track direction than the same road imaged at nadir. On narrow mountain tracks of 3 to 4 metres, the difference between nadir and 40° off-nadir can push the road below reliable detection width. Analysts must account for this when interpreting imagery collected under operational time pressure.
Change detection against pre-event baselines
The fastest route to a blockage map is pixel-level or object-level change detection between a pre-event image and the post-event collect. For optical imagery, this means co-registering both images to sub-pixel accuracy, normalising for illumination differences, and then applying a classifier to flag areas where surface texture or spectral signature has changed significantly. Rubble has a characteristic rough texture and mixed spectral signature; fresh landslide debris has a specific soil or rock colour depending on geology. Both differ measurably from the smooth, spectrally consistent signature of intact asphalt or compacted gravel.
Pre-event baseline quality matters enormously. A baseline collected six months earlier under different sun angle and seasonal vegetation cover will produce false positives. Commercial archive depth helps: WorldView has been collecting since 2009, Pleiades since 2012, so for most populated areas a recent, seasonally matched baseline exists. Where it does not, SAR coherence change (covered in the sibling page on SAR coherence loss) can substitute, though at coarser spatial resolution for road-specific features.
Automated change detection at very high resolution is computationally intensive over large areas. A practical workflow tiles the road buffer (typically 50 m either side of the centreline from OpenStreetMap or a national road dataset) and runs classification only within that mask, which reduces processing time by an order of magnitude compared to full-scene analysis.
From blocked pixels to a routable network
Detecting a blockage is step one. Converting that detection into a logistics decision requires integrating the spatial results with a road-network graph. The standard approach assigns each detected blockage as an impassable edge in the graph, then runs shortest-path or accessibility algorithms to compute which settlements, hospitals, or warehouses can be reached from a given entry point and which cannot.
The output is typically an accessibility zone map: areas within X minutes or kilometres of an open route versus areas that have become isolated. This is operationally more useful than a raw damage map because it answers the question a logistics coordinator actually asks: can I get a truck from the staging area to this village? The graph computation is fast once the blockage layer is ready. The bottleneck is always the imagery acquisition and damage classification, not the network analysis.
A critical honest caveat: the road-network graph is only as good as its input data. In many earthquake-prone regions, OSM coverage of rural tracks is incomplete, and national road datasets may be outdated or unavailable. Gaps in the network graph produce false accessibility readings. Pre-event investment in road-network data quality pays dividends that no amount of post-event satellite tasking can substitute.
Tasking and latency under operational pressure
The International Charter on Space and Major Disasters (covered in its own sibling page) can activate free tasking from multiple operators within hours of a qualifying event. Commercial operators including Maxar and Airbus can task independently with delivery timelines of 12 to 48 hours depending on orbital geometry and cloud conditions. ICEYE and Capella, operating X-band SAR, are largely cloud-agnostic and have published tasking-to-delivery windows of under 12 hours in emergency mode.
Cloud cover is the dominant operational risk for optical sensors. The 2023 Kahramanmaraş earthquakes in Turkey illustrated this: cloud cover over parts of the affected area delayed useful optical collection by 24 to 48 hours in some zones, while SAR imagery provided the first usable damage indicators. A pre-planned sensor hierarchy, optical first where clear, SAR as fallback, is standard practice.
Satellize integrates commercial tasking on client licence alongside open-constellation analytics. For road-network assessments, the workflow we run for clients follows the same tiling and graph-integration approach described here, with delivery as GIS-ready vector layers and a summary accessibility report. Our Tonga crop-estimation programme demonstrated the same underlying pipeline logic, applied to a different feature class, in a similarly data-sparse environment.
Honest limits of the method
Sub-metre optical imagery cannot see through cloud, smoke, or dust. It cannot assess subsurface damage: a road that looks intact from orbit may have lost its foundation to liquefaction and will collapse under vehicle load. SAR detects surface change but cannot distinguish a road blocked by debris from one that has been deliberately cleared and re-blocked, or from a road where the debris has been pushed to the verge.
Narrow unpaved tracks, which are often the only access routes to remote mountain settlements, are the hardest targets. At 3 to 4 metres width, even 0.3 m imagery is marginal for automated classification, and manual interpretation by a trained analyst remains necessary. Analyst capacity is a real constraint in the first 24 to 72 hours of a response.
Finally, a road-accessibility map is a snapshot. Roads are cleared, re-blocked by aftershock-triggered slides, and rerouted continuously in the days after a major event. A single post-event collect becomes stale within 24 to 48 hours in active response conditions. Repeated tasking, ideally daily, is required to keep the network graph current, and that has a cost that should be planned for before the event, not negotiated during it.
Typical figures
| Minimum useful spatial resolution for road blockage detection | 0.5 m (reliable automated); 1 m (marginal, analyst-dependent) |
| Typical optical sensor resolution (WorldView-3 / Pleiades Neo) | 0.31 m / 0.30 m panchromatic; 1.24 m / 1.20 m multispectral |
| Typical SAR resolution (ICEYE Spotlight / Capella Spotlight Ultra) | ~0.5 m / ~0.35 m ground range |
| Post-event tasking to delivery latency | 12–48 h optical (cloud-dependent); under 12 h SAR (emergency mode) |
| Revisit frequency (commercial constellation) | Pleiades Neo: daily per satellite; WorldView constellation: 1–4 collects per day over target; ICEYE / Capella: multiple passes per day |
| Spectral bands used | Panchromatic (450–800 nm) for texture; VNIR multispectral for surface type; X-band SAR (9.6 GHz) for cloud-penetrating change detection |
| Road buffer width processed | Typically 50 m either side of network centreline |
| Minimum detectable blockage width | ~1 m at 0.3 m resolution (automated); ~3 m at 1 m resolution |
| Commercial archive depth | WorldView: from 2009; Pleiades: from 2012; Sentinel-2: from 2015 (free) |
| Deliverable formats | GeoJSON / Shapefile blockage layer; routable network graph (GeoPackage); PDF accessibility report with zone maps |
Analytics Satellize can run
| Road blockage detection layer | Object-based image analysis and pixel-level change detection against pre-event baseline; texture and spectral classifiers applied within road-buffer mask | GeoJSON vector layer of confirmed and probable blockage locations, attributed with blockage type (rubble, debris flow, surface rupture) and confidence score |
| Road accessibility zone map | Shortest-path and isochrone analysis on OSM or client-supplied road graph with blockage edges set to impassable; NetworkX or pgRouting graph engine | GIS polygon layer showing accessible and isolated zones from specified logistics hubs, with estimated travel-time contours |
| Isolated settlement priority list | Spatial join of accessibility zones against settlement population layer (WorldPop or national census); settlements with no passable route ranked by estimated population | Tabular report of isolated settlements with coordinates, estimated population, and nearest open road endpoint |
| Multi-temporal blockage update | Repeated change detection on daily or near-daily tasked imagery; delta comparison between successive blockage layers to track clearance progress or new blockages from aftershock activity | Time-stamped GeoJSON updates and a summary clearance-progress dashboard (PDF or web feed) |
| SAR-optical fusion damage confidence layer | Logical combination of SAR backscatter change and optical texture change; features flagged by both sensors assigned high confidence; single-sensor detections flagged as probable | Merged blockage layer with dual-sensor confidence attribute, suitable for prioritising field verification |
| Logistics corridor risk rating | Overlay of blockage layer with slope, landslide susceptibility index (published national or global datasets), and aftershock probability contours to rate corridor stability | Line-attributed road network GeoPackage with per-segment risk score; input-ready for humanitarian logistics planning tools |
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.