Palaeochannel and ancient route mapping using multitemporal spectral analysis
Abandoned channels and ancient caravan routes leave grain-size, moisture and compaction signatures readable in SWIR and thermal bands. Multitemporal PCA and band-ratio stacks reveal them where the eye sees only desert.
Sensors
- Landsat 8/9 OLI-TIRS: 30 m multispectral, 100 m thermal (resampled to 30 m). SWIR bands 6 and 7 (1.57 µm and 2.11 µm) are diagnostic for soil moisture and clay-mineral content. 16-day repeat per satellite; combined 8-day revisit. Archive from 1972 (Landsat 1 MSS) enables multi-decade change stacks.
- Sentinel-2 MSI: 10 m visible, 20 m SWIR and red-edge. Bands 11 and 12 (1.61 µm and 2.19 µm) resolve finer channel morphology than Landsat at comparable cost. 5-day revisit at mid-latitudes. Free archive from 2015; useful for seasonal compositing to suppress vegetation noise.
- ASTER: 15 m VNIR, 30 m SWIR (6 bands between 1.6 µm and 2.43 µm), 90 m thermal infrared (5 bands). The SWIR suite is unmatched on a freely archived sensor for clay and carbonate discrimination. Thermal inertia mapping from day/night TIR pairs reveals compaction contrasts from buried or compacted route surfaces. SWIR detector failure in 2008 limits post-2008 SWIR use, but the archive to 2000 remains valuable.
- WorldView-3 SWIR: 8 SWIR bands at 7.5 m resolution, covering 1.195 µm to 2.365 µm. Resolves individual channel levees and track ruts at sub-10 m scale. Commercial tasking only; cost limits systematic coverage but suits targeted confirmation of anomalies flagged by medium-resolution surveys.
Why the ground still holds the shape of a river that dried up three thousand years ago
When a river abandons a channel, it leaves behind a distinct sedimentary legacy. Point bars and channel floors accumulate coarser, better-sorted sand; overbank deposits are finer-grained silts and clays. That grain-size contrast governs two things detectable from orbit: soil moisture retention and thermal inertia. Coarser channel sands drain faster and warm quickly by day, while adjacent fine-grained floodplain soils retain moisture longer and buffer temperature swings. The contrast is sharpest in arid and semi-arid environments, where overlying vegetation is sparse enough not to mask it.
Compacted surfaces left by millennia of foot and caravan traffic behave differently again. Compaction reduces porosity, which suppresses moisture infiltration and increases bulk density. The thermal inertia of a compacted track surface is measurably higher than surrounding unconsolidated soil, producing a persistent cool anomaly in daytime thermal imagery and a warm anomaly at night. ASTER day/night TIR pairs have been used in published literature to map exactly this contrast across desert terrain in the Near East and Central Asia.
The spectral toolkit: which bands carry the signal and which carry noise
SWIR reflectance between roughly 1.5 µm and 2.5 µm is where soil moisture and clay mineralogy express themselves most clearly. Water absorption features at 1.45 µm and 1.94 µm reduce reflectance in moist soils; the depth of these features is proportional to moisture content. Clay minerals, including kaolinite, montmorillonite and illite, show hydroxyl absorption near 2.2 µm. Palaeochannel fills frequently contain higher clay fractions than surrounding deflated desert surfaces, making band ratios such as ASTER bands 5/3 or 6/4 effective discriminators.
Principal-component analysis applied to a multitemporal SWIR stack exploits the fact that palaeochannel signatures are persistent across seasons while surface moisture from recent rainfall is transient. The first few PCs absorb seasonal variation; later components tend to isolate stable soil-property contrasts. This is not a guaranteed separation: in areas with shallow alluvial mantles or aeolian sand sheets, the channel signal can be buried beyond spectral reach. The method works best where the palaeosurface is within the top 10 to 20 cm, which is the approximate penetration depth of SWIR radiation in dry soil.
What the Silk Road and Mesopotamia have taught us about what this actually finds
Published work on Silk Road route detection has used Landsat TM and ETM+ band composites alongside ASTER TIR to trace track networks across the Taklamakan margins and the Karakum desert. Researchers have identified linear compaction anomalies consistent with known historical routes, corroborated by ground survey and historical cartography. In Mesopotamia, Landsat-based studies have mapped abandoned irrigation canal networks of the Sasanian and earlier periods by exploiting the moisture-retention contrast between canal fills and the surrounding alluvial plain, particularly in late-summer imagery when the contrast is at its seasonal maximum.
The honest qualification is that neither palaeochannel nor route signatures are unambiguous at 30 m resolution. Natural geomorphic lineaments, fault traces, deflation corridors and modern agricultural boundaries all produce similar spectral and thermal contrasts. Discrimination requires convergence of evidence: consistent orientation relative to historical geography, corroboration from DEM analysis, and ideally ground-truth or archival documentary sources. A spectral anomaly is a hypothesis, not a discovery.
Building a time-series stack that is worth analysing
A single scene is almost never sufficient. Cloud, seasonal moisture variation and recent disturbance (ploughing, flooding, dust deposition) all introduce noise that mimics or masks the target signal. A practical workflow stacks 10 to 20 cloud-free Landsat or Sentinel-2 scenes distributed across multiple dry seasons, computes per-pixel medians or percentile composites to suppress transient effects, then applies PCA or minimum-noise-fraction transformation to the composite stack. The stable low-variance components that emerge from this process are where palaeochannel and route signatures tend to concentrate.
Temporal depth matters. Landsat's archive back to 1972 is genuinely useful here, not because ancient channels changed between 1972 and 2024, but because land-use change has. Agricultural expansion, irrigation and urban growth progressively obscure palaeosurfaces. Scenes from the 1970s and 1980s often show signatures that are now buried under fields or settlements. Comparing early-archive Landsat with current Sentinel-2 also quantifies how much of the archaeological landscape has been lost to modern development, which is a distinct and important output.
Where the method fails, and what to do about it
Vegetation is the primary masking problem. Even sparse scrub over a palaeochannel suppresses the SWIR soil signal significantly. Timing acquisitions to the driest part of the year reduces but does not eliminate this. Red-edge bands on Sentinel-2 can help separate vegetation from bare-soil signal at 20 m, but where channel fills support denser vegetation than surrounding terrain, the vegetation itself becomes a proxy indicator rather than a mask, and that is worth exploiting.
Medium-resolution sensors cannot distinguish a 15 m wide irrigation canal from a 15 m wide natural distributary channel. WorldView-3 SWIR at 7.5 m begins to resolve morphological detail, such as levee asymmetry and bifurcation geometry, that helps with this discrimination. The cost of systematic WorldView-3 coverage over large survey areas is prohibitive for most programmes, so the practical approach is to use Landsat and Sentinel-2 for regional screening and task commercial SWIR only over high-priority anomalies. Satellize applies this tiered approach in its analytics workflows, combining open-constellation screening with targeted commercial tasking where client licences permit.
Deeply buried channels, covered by more than roughly 20 to 30 cm of aeolian sand or alluvial overburden, are beyond the reach of optical and SWIR sensors entirely. That is a different problem requiring SAR or active microwave methods, which are covered elsewhere in this library.
Typical figures
| Typical spatial resolution (screening) | 10 to 30 m (Sentinel-2 SWIR at 20 m; Landsat OLI at 30 m) |
| Typical spatial resolution (confirmation) | 7.5 m (WorldView-3 SWIR) |
| Revisit for time-series compositing | 5 days (Sentinel-2 combined); 8 days (Landsat 8+9 combined) |
| Thermal inertia mapping | ASTER TIR 90 m; Landsat TIRS 100 m (resampled 30 m); day/night pair required |
| Key diagnostic spectral bands | SWIR 1.5 to 2.5 µm (soil moisture, clay minerals); TIR 8 to 12 µm (thermal inertia) |
| Archive depth | Landsat from 1972; Sentinel-2 from 2015; ASTER from 2000 |
| Minimum detectable channel width (optical SWIR) | Approximately 30 to 60 m at Landsat resolution; approximately 15 to 20 m at Sentinel-2 SWIR |
| Cloud and atmospheric limits | Optical and SWIR blocked by cloud; dry-season compositing essential in seasonally humid zones |
| Subsurface penetration depth (SWIR) | Approximately 10 to 20 cm in dry bare soil; near zero under vegetation or moist soil |
| Delivery formats | GeoTIFF band composites, PCA output rasters, GIS vector anomaly layers (GeoPackage or Shapefile) |
Analytics Satellize can run
| Multitemporal SWIR composite | Percentile compositing across 10 to 20 dry-season scenes to suppress transient moisture and cloud artefacts | GeoTIFF stack (Landsat bands 6, 7 or Sentinel-2 bands 11, 12) ready for PCA or band-ratio analysis |
| Principal-component anomaly layer | PCA or minimum-noise-fraction transformation of multitemporal SWIR stack; stable low-variance components extracted | Raster layer of persistent soil-property anomalies with candidate palaeochannel and route lineaments highlighted |
| Clay-mineral and moisture band-ratio map | Published band-ratio indices (e.g. ASTER bands 5/3 for clay, 4/6 for moisture) applied to seasonal composites | Classified GeoTIFF with spectral units keyed to soil-type interpretation, supplied with uncertainty notes |
| Thermal inertia contrast map | Day/night ASTER TIR differencing or Landsat TIRS diurnal modelling to identify compaction anomalies | GeoTIFF of apparent thermal inertia, with linear anomalies vectorised as candidate route segments |
| Historical land-cover change comparison | Co-registration of 1970s/1980s Landsat MSS or TM with current Sentinel-2; change detection to identify obscured palaeosurfaces | Change map and report quantifying area of palaeochannel signatures lost to agricultural or urban expansion since first available imagery |
| Targeted WorldView-3 SWIR confirmation | Commercial tasking over high-priority anomalies flagged by medium-resolution screening; 7.5 m SWIR morphological analysis | High-resolution SWIR scene with annotated morphological interpretation (levee geometry, bifurcation, track ruts) |
| Candidate feature vector dataset | Automated lineament extraction from PCA and band-ratio outputs, filtered by orientation, length and spectral consistency | GeoPackage of ranked candidate features with confidence scores and recommended ground-truth priorities |
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.