Satellite detection of ancient roads, causeways and processional ways
Roman roads, Mayan sacbeob and Andean qhapaq ñan leave measurable signatures in high-resolution DEMs, SAR backscatter and multispectral imagery. Satellite workflows now detect linear anthropogenic features that ground survey alone would take decades to map.
Sensors
- TanDEM-X (DLR): X-band bistatic SAR interferometry; global DEM at 12 m posting, with 0.2 m relative vertical accuracy over flat terrain. Resolves raised causeways of 30 cm or more above surrounding ground. Canopy penetration is partial at X-band, limiting utility in dense closed-canopy forest.
- ALOS World 3D (JAXA): 5 m posting DEM derived from PRISM tri-stereo optical data. Effective for open and semi-open terrain; no SAR canopy penetration. Useful for mapping qhapaq ñan segments across Andean grassland and scrub where optical stereo is unobstructed.
- Sentinel-1 SAR (ESA): C-band, 10 m IW mode ground range resolution, 6-day repeat at mid-latitudes with both satellites. Backscatter texture analysis separates compacted engineered surfaces from surrounding soil and vegetation. Multi-temporal coherence stacking reduces speckle and improves linear feature contrast.
- WorldView-3 (Maxar): 0.31 m panchromatic, 1.24 m multispectral, 8 shortwave infrared bands. Resolves road surface width down to roughly 1 m in pan-sharpened imagery. SWIR bands detect moisture and mineralogical contrasts at road margins that are invisible in visible wavelengths.
- ALOS-2 PALSAR-2 (JAXA): L-band SAR at 3–10 m resolution depending on mode. Longer wavelength gives greater canopy penetration than X- or C-band, improving detection of sub-canopy earthworks in tropical forest. Published studies of Cambodian and Mesoamerican sites have used L-band backscatter for exactly this purpose.
What an ancient road looks like from 500 km up
Engineered roads differ from natural terrain in three measurable ways: they are deliberately straight or geometrically regular; their surface material is compacted or imported; and they sit either proud of or below the surrounding ground level by a consistent margin. Each property leaves a distinct signal in satellite data.
In a high-resolution DEM, a Roman agger or a Mayan sacbé appears as a narrow, continuous positive relief anomaly, typically 0.3 to 2 m above surrounding ground, running for kilometres without the sinuous deviation of a natural ridge. Directional slope-aspect filtering, applied along candidate azimuths, amplifies these features against the background noise of natural topography. The technique was formalised in published LiDAR-based landscape archaeology workflows and transfers directly to satellite DEMs where vertical precision is sufficient.
SAR backscatter adds a second, independent line of evidence. Compacted gravel, limestone or clay surfaces scatter radar energy differently from surrounding organic soils and vegetation. At C-band, Sentinel-1 multi-temporal averaging over 12 or more acquisitions suppresses speckle enough to make road-width linear anomalies detectable at widths of roughly 5 to 10 m, depending on surface contrast. At L-band, ALOS-2 PALSAR-2 penetrates closed forest canopy sufficiently to image earthwork surfaces that optical sensors cannot see at all.
Three road traditions, three detection challenges
Roman roads in Britain and Europe are among the most studied cases. Many are already mapped, but satellite DEM analysis continues to extend known networks into areas where surface expression is subtle. The agger survives under pasture and arable land as a low ridge, often 4 to 8 m wide and 0.3 to 0.6 m high. At TanDEM-X 12 m posting, the feature is detectable but marginal; 5 m ALOS World 3D data resolves it more cleanly in open farmland. The principal ambiguity is field boundaries and drainage ditches, which produce similar linear signatures and must be cross-checked against cadastral records.
Mayan sacbeob, the white-plastered causeways of the lowland Maya, present a different problem. They cross dense tropical forest in the Yucatán and Petén. Optical sensors are useless under closed canopy. L-band SAR offers partial penetration, and published research on sites including Caracol in Belize has demonstrated that PALSAR data can trace causeway networks invisible to optical survey. TanDEM-X provides complementary texture data at the canopy surface, which correlates with sub-canopy relief in areas of thinner forest.
The qhapaq ñan, the Inca road system designated a UNESCO World Heritage Site in 2014, traverses terrain ranging from high-altitude puna grassland to coastal desert. In open puna, 5 m DEMs resolve the road clearly. In desert, the surface is often just a cleared and levelled strip distinguishable from surrounding terrain by its lower surface roughness, detectable in SAR backscatter. In steep Andean terrain, the road is cut into slope faces or built on retaining walls, making it a negative feature on the upslope side and positive on the downslope, a signature that slope-aspect analysis picks up reliably.
The lineament problem: separating engineering from geology
The central analytical difficulty is that natural geology produces linear features too. Fault scarps, dyke intrusions, stream captures and aeolian dunes all generate linear signals in DEMs and SAR imagery. Distinguishing these from anthropogenic roads requires a combination of geometric, spectral and contextual tests.
Geometric regularity is the first filter. Engineered roads maintain consistent width and tend to ignore local topographic obstacles or cross them with deliberate engineering such as cuttings or embankments. Natural lineaments follow structural geology and are rarely consistent in width. Spectral anomaly detection in WorldView-3 SWIR bands can identify imported surface materials, limestone or compacted clay, that differ from the local geological substrate. Contextual proximity to known settlement sites, river crossings or resource locations provides a final probabilistic check. No single test is definitive; the workflow is convergent evidence.
Honest limits of the method
Satellite DEM analysis cannot replace ground survey or LiDAR in high-priority areas. TanDEM-X at 12 m posting misses road features narrower than roughly 10 to 15 m, and its 0.2 m relative vertical accuracy is a best-case figure over flat, low-vegetation terrain. In steep or forested terrain, vertical errors increase and small earthworks become undetectable. ALOS World 3D at 5 m is better for narrow features in open country but has no canopy penetration at all.
Cloud cover is a persistent problem for optical confirmation in tropical regions. Sentinel-1 SAR is cloud-independent, but C-band canopy penetration is limited to perhaps the top 1 to 2 m of vegetation in most forest types. L-band ALOS-2 penetrates further but is not freely available at the same cadence as Sentinel-1, and commercial tasking adds cost and scheduling lead time.
Multispectral anomaly detection at WorldView-3 resolution is powerful for surface characterisation but requires clear skies and works only where the road surface is exposed or very shallowly buried. Features under more than a few centimetres of sediment or vegetation mat are not detectable by any optical method. For those cases, the sibling page on SAR subsurface penetration for desert sites covers the relevant physics.
From detection to heritage management
A detection workflow produces candidate linear features as GIS vector layers, ranked by confidence score derived from the convergence of DEM, SAR and optical evidence. These layers feed directly into heritage management systems, enabling national agencies to prioritise ground-truth surveys, assess development threat corridors and submit documentation for protection or inscription.
Satellize has built similar convergent-evidence pipelines for agricultural applications, including the Kingdom of Tonga crop-estimation programme, and the same multi-source fusion logic applies directly to archaeological feature extraction. For road-network detection specifically, a typical project delivery includes a ranked candidate-feature layer in GeoPackage or Shapefile format, a confidence-score raster, and a methodology report suitable for submission to heritage authorities.
The practical value is scale. A ground survey team walking transects across a 500 km² study area takes years. A satellite-derived candidate map covering the same area, produced in weeks, tells the survey team exactly where to look. That is not a replacement for fieldwork; it is a way to make fieldwork count.
Typical figures
| Best DEM spatial resolution (open terrain) | 5 m (ALOS World 3D); 12 m (TanDEM-X global product) |
| Relative vertical accuracy (TanDEM-X, flat terrain) | ~0.2 m (1-sigma); degrades in forest and steep terrain |
| SAR resolution (Sentinel-1 IW mode) | 10 m ground range; 6-day repeat at mid-latitudes (both satellites) |
| Optical resolution (WorldView-3) | 0.31 m panchromatic; 1.24 m multispectral; 3.7 m SWIR |
| Minimum detectable road width (DEM method) | ~10–15 m at TanDEM-X 12 m; ~5 m at ALOS World 3D 5 m |
| Minimum detectable road relief (open terrain) | ~0.3 m above surrounding ground (TanDEM-X, low vegetation) |
| SAR archive depth (Sentinel-1) | 2014 to present; multi-temporal stacking improves feature contrast |
| Cloud independence | SAR (Sentinel-1, ALOS-2) fully cloud-independent; optical requires clear sky |
| Delivery formats | GeoTIFF (rasters), GeoPackage or Shapefile (vector features), PDF methodology report |
| Canopy penetration | L-band (ALOS-2): partial penetration of closed tropical forest; C-band: top 1–2 m only; X-band: surface only |
Analytics Satellize can run
| Candidate road-feature vector layer | Directional slope-aspect filtering of DEM combined with SAR backscatter texture analysis; lineament extraction and geometric regularity scoring | Ranked GeoPackage of candidate linear features with confidence score per segment, covering client-defined study area |
| Confidence-score raster | Convergent evidence fusion: DEM relief anomaly, SAR texture anomaly and multispectral surface-material anomaly combined in weighted overlay | GeoTIFF confidence surface at native sensor resolution, suitable for overlay in QGIS or ArcGIS |
| Natural lineament discrimination report | Geometric regularity analysis, geological substrate cross-check and contextual proximity scoring against known site locations | PDF report classifying candidate features as high, medium or low archaeological probability, with per-feature rationale |
| Multi-temporal SAR coherence stack | Sentinel-1 coherence averaging across 12-plus acquisitions to suppress speckle and enhance persistent linear surface anomalies | GeoTIFF coherence composite with road-candidate overlay, cloud-independent, covering repeat-pass archive from 2014 |
| WorldView-3 SWIR surface-material anomaly map | Spectral unmixing of SWIR bands to detect imported or compacted surface materials differing from local geological substrate | GeoTIFF anomaly layer at 3.7 m resolution, flagging pixels consistent with limestone, compacted clay or gravel surfaces |
| Heritage authority submission package | Aggregation of vector layers, confidence rasters and methodology documentation into a format consistent with UNESCO and national heritage body requirements | Zipped GIS package plus PDF methodology report with sensor provenance, processing chain and uncertainty statements |
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.