Unpaved road surface condition mapping for rural access and land value
SAR backscatter and optical albedo distinguish compacted gravel from rutted or washed-out unpaved roads, quantifying access degradation that suppresses rural land values even where no road register exists.
Sensors
- Sentinel-1 C-band SAR (ESA): 10 m ground range detected resolution in IW mode, 6-day revisit at mid-latitudes with both satellites active. C-band (5.405 GHz) backscatter responds to surface roughness at centimetre scale, making it sensitive to rutting and aggregate loss on unpaved surfaces, though soil moisture changes can produce similar backscatter shifts and must be disentangled.
- ALOS-2 PALSAR-2 L-band SAR (JAXA): 3 to 10 m resolution depending on acquisition mode, 14-day revisit. L-band (1.27 GHz) penetrates shallow surface layers and responds to larger-scale roughness features and sub-surface moisture, complementing C-band by offering a different roughness sensitivity window. The two frequencies together reduce moisture-roughness ambiguity.
- Planet SuperDove (optical): 3 m resolution, near-daily revisit in most regions. Eight spectral bands including red-edge and near-infrared allow albedo cross-checking: compacted gravel roads have distinctive high visible-band albedo, while wet or organic-rich ruts are darker. Used to validate SAR condition classes and flag cloud-free change epochs.
- Sentinel-2 MSI (ESA): 10 m in visible and NIR bands, 5-day revisit with both satellites. Useful for regional-scale albedo mapping and seasonal baseline construction where Planet tasking is not licensed. Free and globally archived since 2015.
What the physics of rough surfaces gives away
Radar backscatter intensity depends on the ratio of surface roughness to wavelength. A freshly graded gravel track, with aggregate size of a few millimetres, appears relatively smooth to C-band and returns a lower, specular-leaning signal. The same track after six months of heavy rain, now rutted to depths of 5 to 15 cm and strewn with displaced material, presents a roughness scale that couples strongly with C-band wavelength, raising backscatter by several decibels. That shift is detectable. It is not subtle.
L-band sees the same surface differently. Its longer wavelength is less sensitive to millimetre-scale aggregate roughness and more sensitive to bulk surface geometry and shallow sub-surface conditions. A washed-out section where the road crown has collapsed shows strong L-band response even when the surface is dry. Using both frequencies together, analysts can separate the signature of structural road failure from the signature of a temporarily waterlogged surface, which matters enormously when the goal is a stable condition classification rather than a weather snapshot.
The moisture problem, stated plainly
Soil moisture is the main confound. Wet soil raises dielectric constant and therefore backscatter, mimicking roughness. A road that appears degraded in a single SAR acquisition taken after heavy rain may look perfectly serviceable in a dry-season pass. This is not a flaw unique to one vendor or one algorithm; it is a physical constraint of microwave remote sensing.
The practical mitigation is temporal stacking. By building a time series of Sentinel-1 passes across multiple dry-season epochs, analysts construct a moisture-normalised roughness baseline. Persistent high-backscatter anomalies that survive the seasonal filter are structural, not hydrological. The optical albedo layer from Planet or Sentinel-2 adds a second check: a genuinely degraded road surface tends to show lower visible-band reflectance (darker, disturbed material) that persists across acquisition dates regardless of moisture state.
From signal to condition class
The classification output is typically a four-class schema: intact compacted surface, minor surface deterioration (early rutting, edge erosion), moderate degradation (deep ruts, loss of crown, localised washout) and severe failure (impassable or near-impassable). These classes are derived by combining the SAR roughness index, the optical albedo deviation from the road's own dry-season baseline, and where available, topographic slope from SRTM or Copernicus DEM to flag drainage-risk sections.
Minimum detectable feature size is constrained by resolution. At 10 m (Sentinel-1 IW), a washout narrower than roughly 10 to 15 m may not register unless it produces a sufficiently strong backscatter contrast with adjacent road surface. Planet at 3 m captures narrower failures optically but cannot supply the roughness dimension independently. The two sensors are therefore complementary rather than interchangeable. Expecting either alone to produce a complete picture is a common and avoidable mistake.
Road condition as a land value input
Access quality is a direct input to rural land value. Agricultural economists have documented that travel time to market, not straight-line distance, governs the price differential between parcels. An unpaved road in moderate degradation can double effective travel time relative to a maintained track of the same length. That time cost is capitalised into parcel prices, depressing values for land that is nominally accessible but practically difficult to reach during wet seasons.
Satellite-derived condition maps fill a specific gap in regions where road registers are absent, years out of date, or simply do not record surface quality at all. A cadastral database may show a road exists; it will not show that the road has been impassable from June to September for the past three years. The satellite time series shows exactly that, and the information is auditable and reproducible. For rural land valuation, insurance underwriting or agricultural lending, that audit trail has practical value beyond the map itself.
Honest limits and what they mean for programme design
Cloud cover blocks optical sensors for weeks at a time in humid tropical regions. SAR is cloud-independent, which is why it anchors the method, but SAR alone cannot resolve the moisture ambiguity without the temporal stacking described above. A single-date SAR product for road condition is unreliable; a twelve-month stack with seasonal stratification is substantially more defensible.
Very narrow tracks, common in smallholder agricultural areas, may fall below the resolution threshold of freely available SAR data. PALSAR-2 Spotlight mode can reach 3 m, but tasking is not always straightforward to arrange. Planet fills the optical gap at 3 m but does not supply the roughness dimension. For programmes in regions with dense fine-track networks, ground-truth campaigns on a sample of road segments are advisable to calibrate the classification and bound the error rate. Satellize incorporates ground-truth calibration into programme design as a standard step, drawing on the same structured validation approach used in its Tonga crop-estimation work.
Archive depth is an asset. Sentinel-1 data runs back to 2014 in most regions, ALOS-2 to 2014, and Sentinel-2 to 2015. Multi-year condition histories are therefore available without new tasking costs for most of the world.
Putting it into practice
A practical programme begins with road network extraction, either from OpenStreetMap, a client's own cadastral data, or an automated road-detection pass on high-resolution optical imagery. That network becomes the spatial mask within which SAR and optical time series are sampled. Processing then proceeds in three stages: baseline construction from dry-season passes, anomaly detection against that baseline, and condition classification with uncertainty bounds per road segment.
Delivery is typically a GIS vector layer of road segments attributed with condition class, confidence score, date of last assessment and change flag relative to the previous cycle. Update frequency can be monthly using Sentinel-1's free archive or near-weekly with commercial tasking added. The output integrates directly into land valuation models as an access-quality score per parcel, replacing the binary present-or-absent road attribute that most rural cadastral systems carry today.
Typical figures
| Primary SAR spatial resolution | 10 m (Sentinel-1 IW mode); 3–10 m (PALSAR-2 depending on mode) |
| Optical cross-check resolution | 3 m (Planet SuperDove); 10 m (Sentinel-2 visible/NIR) |
| SAR revisit (Sentinel-1, both satellites) | 6 days at mid-latitudes; longer at low latitudes with single satellite |
| Optical revisit (Planet SuperDove) | Near-daily in most regions, subject to cloud and tasking |
| Frequency bands used | C-band 5.405 GHz (Sentinel-1); L-band 1.27 GHz (PALSAR-2) |
| Minimum detectable road failure width | ~10–15 m for SAR at 10 m resolution; ~5 m optically at 3 m resolution |
| Archive depth | Sentinel-1 from 2014; Sentinel-2 from 2015; ALOS-2 from 2014 |
| Condition classes | 4 (intact, minor deterioration, moderate degradation, severe failure) |
| Typical update latency | 2–5 days from SAR acquisition to delivered layer, depending on processing queue |
| Delivery formats | GeoPackage or Shapefile vector layer; GeoTIFF roughness index raster; PDF summary report |
Analytics Satellize can run
| Road segment condition classification | SAR backscatter roughness indexing (time-series stack, seasonal stratification) combined with optical albedo deviation from road-specific baseline | Vector GIS layer: road segments attributed with condition class (1–4), confidence score and assessment date |
| Moisture-normalised roughness baseline | Multi-temporal SAR compositing across dry-season acquisitions; dielectric correction using co-located soil moisture proxy | GeoTIFF raster of normalised roughness index per road buffer, updated seasonally |
| Condition change alert | Bi-temporal differencing of roughness index between consecutive Sentinel-1 cycles; threshold exceedance flagging | Alert feed (GeoJSON or email digest) identifying segments that crossed a condition-class boundary since the previous cycle |
| Access-quality score per land parcel | Network analysis: shortest-path travel time from parcel centroid to nearest paved road, weighted by road condition class | Attribute table joined to cadastral parcel layer; score expressed as effective travel-time multiplier relative to maintained-road baseline |
| Drainage-risk section identification | Slope and flow-accumulation modelling on Copernicus DEM (30 m), overlaid with road network to flag sections in concave or low-order drainage positions | Annotated map layer highlighting high-risk sections; input to seasonal impassability probability estimate |
| Multi-year condition history report | Annual condition classification run retrospectively across full Sentinel-1 archive (2014 to present) for a defined road network | Time-series chart and PDF report showing condition trajectory per road segment; useful for valuation due diligence and insurance underwriting |
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.