Ancient irrigation canal and field-system network mapping

Extinct irrigation networks leave moisture, texture and elevation signatures that persist for millennia. Combining declassified CORONA photography, Sentinel-1 SAR, ALOS-2 PALSAR-2 and Sentinel-2 spectral indices can reconstruct field systems across thousands of square kilometres before modern agriculture erases them.

Sensors

CORONA (KH-4 / KH-4A / KH-4B, USGS declassified archive): Panchromatic film imagery from 1960–1972, ground resolution approximately 1.8–7.5 m depending on mission generation. Captured landscapes before large-scale mechanised agriculture and urban expansion, providing the single most valuable pre-disturbance baseline available for the Middle East, South Asia and Central America. Accessible via USGS Earth Explorer.
Sentinel-1 C-band SAR (ESA): 6-day repeat at mid-latitudes (12-day for single satellite), 5 × 20 m IW mode resolution. C-band backscatter is sensitive to surface roughness and dielectric contrasts caused by residual soil-moisture differences above buried or filled canals. Interferometric coherence products can enhance subtle linear micro-topography.
ALOS-2 PALSAR-2 L-band SAR (JAXA): L-band (23.6 cm wavelength) penetrates dry sandy soils to depths of roughly 1–2 m, increasing sensitivity to subsurface moisture and compaction anomalies associated with ancient earthworks. Fine-beam mode delivers 3 m single-look resolution; standard mode 10 m. Revisit approximately 14 days.
Sentinel-2 MSI (ESA): 13 spectral bands, 10 m resolution in visible and near-infrared, 20 m in red-edge and shortwave infrared (SWIR). SWIR bands (1610 nm and 2190 nm) are particularly sensitive to soil-moisture and clay-mineral content differences that mark former canal beds. Five-day revisit enables multitemporal compositing to reduce noise.

What a filled canal leaves behind

An irrigation canal dug into alluvial sediment and then abandoned does not simply disappear. The fill accumulates organics, retains slightly higher clay content than surrounding soil, and compacts differently under agricultural traffic. These differences persist in the soil column for centuries to millennia, producing a consistent set of physical signatures: a marginally higher dielectric constant when dry conditions are transitional, a subtly distinct surface roughness, and a micro-topographic depression or levee that may be only 20–50 cm in amplitude.

Each of those properties is detectable by a different part of the electromagnetic spectrum. That is why single-sensor approaches consistently underperform against multi-sensor fusion. No one image type finds everything; the question is which combination is most efficient for a given landscape.

CORONA as a pre-disturbance baseline, not a historical curiosity

The declassified CORONA reconnaissance imagery, available through the USGS Earth Explorer archive, covers much of the Middle East, South Asia and parts of Central Asia from 1960 to 1972. In that window, the Mesopotamian alluvial plain, the Indus floodplain and the piedmont zones of Iran and Afghanistan had not yet been transformed by deep-ploughing, centre-pivot irrigation and urban sprawl. Canal networks visible in CORONA frames have since been obliterated at the surface.

KH-4B imagery reaches approximately 1.8 m ground resolution, sufficient to trace individual field boundaries and secondary canal branches. Published work by researchers including those associated with the Ur Region Archaeology Project has used CORONA systematically to map canal hierarchies across tens of thousands of hectares in southern Iraq. The honest caveat: CORONA is panchromatic only, stereo coverage is inconsistent, and geometric correction of the panoramic film frames requires careful handling of scan-line distortion. Accuracy after orthorectification is typically 5–15 m without ground control, better with it.

SAR backscatter and the moisture-contrast signal

Synthetic aperture radar does not care about cloud cover, which matters in regions where the optimal detection window, after light rain that briefly amplifies soil-moisture contrasts, coincides with convective cloud. Sentinel-1 C-band backscatter over arid alluvial plains shows measurable sigma-nought differences of 1–3 dB above filled canal features compared with adjacent undisturbed soil in published studies from the Fertile Crescent. That is a real signal, but it is not large. Single-date SAR is unreliable; multitemporal averaging over 10–20 acquisitions suppresses speckle enough to make linear features traceable.

ALOS-2 PALSAR-2 L-band adds a different capability. At 23.6 cm wavelength, the radar energy penetrates dry sandy or loamy soils to depths of roughly one to two metres before being scattered back. Subsurface moisture contrasts associated with former canal beds, which would be invisible to C-band, become detectable. The limitation is spatial: standard PALSAR-2 mode delivers 10 m pixels, which means canals narrower than about 20–30 m are at or below the reliable detection floor. Fine-beam mode at 3 m helps, but coverage swaths are narrower and acquisition planning is required.

Sentinel-2 SWIR and the clay-mineral argument

The shortwave infrared bands of Sentinel-2, centred at 1610 nm (Band 11) and 2190 nm (Band 12), respond to clay-mineral absorption features and to soil-moisture content in the top few centimetres. Former canal beds that filled with fine-grained sediment often have a measurably different Band 11 / Band 12 ratio compared with surrounding coarser alluvium. Indices such as the Clay Mineral Ratio (Band 11 / Band 12) and the Normalised Difference Soil Index have been applied in published remote-sensing archaeology literature to map these contrasts.

At 20 m resolution, Sentinel-2 SWIR cannot resolve individual secondary canals, which were often only 5–15 m wide. It is most useful for mapping the coarser hierarchy: main supply channels, distributary branches, and the boundaries of irrigated field blocks. The signal is also strongly affected by surface roughness from recent tillage, so timing acquisitions to coincide with fallow periods or light rainfall is important. Multitemporal composites built from 20–30 cloud-free scenes over two to three years consistently outperform single-date analysis.

Integrating DEM analysis: where micro-topography confirms the network

Shuttle Radar Topography Mission (SRTM) data at 30 m and TanDEM-X at 12 m are the most widely used global DEMs for landscape archaeology. Neither reliably resolves canal features below about 0.5 m in relief. Copernicus DEM GLO-30, derived from TanDEM-X, performs similarly. For finer work, commercial stereo imagery from Pleiades or WorldView can generate DSMs at 0.5–1 m resolution over targeted areas, resolving levee remnants and field-boundary banks that are invisible in global products.

The analytic workflow typically runs as a convergence test: a linear feature that appears in CORONA photography, shows a backscatter anomaly in multitemporal Sentinel-1, has a SWIR spectral contrast in Sentinel-2, and aligns with a micro-topographic ridge or depression in a high-resolution DEM is almost certainly a real ancient hydraulic feature. Features that appear in only one data layer should be treated as candidates, not conclusions. Published work from the Indus Valley and from pre-Columbian Amazonia, where geoglyph-associated canal systems have been mapped using similar convergence approaches, supports this multi-evidence standard.

What this analysis can and cannot tell you

Satellite-derived canal mapping produces spatial extent and network topology. It does not directly establish date of construction, water-flow volume, or the social organisation behind the system. Those questions require ground-truthing, ideally targeted excavation at nodes identified from orbit. The satellite layer is most valuable as a prioritisation tool: it tells you where to look across a region that would take decades to survey on foot.

Detection limits are real constraints. Canals narrower than roughly 10–15 m are below the reliable detection threshold of all open-access sensors discussed here. Features buried beneath more than 1–2 m of aeolian sand are beyond L-band penetration in most soil conditions. Dense vegetation cover, common in Amazonian contexts, suppresses optical signals entirely and degrades SAR performance. Cloud cover is not a problem for SAR but is a serious constraint on the SWIR compositing approach in humid regions.

Satellize applies this multi-sensor convergence workflow on open Sentinel and JAXA archives, with commercial tasking added where fine-resolution DEM or sub-metre optical coverage is needed. The same spectral-index pipeline that underpins the Tonga crop-estimation programme can be adapted for soil-contrast mapping across arid archaeological landscapes. Outputs are delivered as georeferenced GIS layers with confidence classifications, suitable for direct integration into heritage management plans or research publication workflows.

Typical figures

Best optical resolution (CORONA archive)	~1.8 m (KH-4B); panchromatic only; 1960–1972 coverage
Sentinel-2 SWIR resolution	20 m (Bands 11 and 12); 10 m visible/NIR; 5-day revisit
Sentinel-1 SAR resolution	5 × 20 m (IW mode); 6-day repeat at mid-latitudes
ALOS-2 PALSAR-2 resolution	3 m (fine-beam) to 10 m (standard); 14-day revisit; L-band 1.27 GHz
Minimum detectable canal width (optical/SAR)	~10–15 m reliably; narrower features detectable only with sub-metre commercial imagery
L-band subsurface penetration depth	~1–2 m in dry sandy/loamy soil; degrades rapidly with moisture or clay content
Archive depth (CORONA)	1960–1972; freely accessible via USGS Earth Explorer
Archive depth (Sentinel-1/2)	2014/2015 to present; full global archive freely accessible
Typical DEM vertical accuracy	SRTM/Copernicus GLO-30: ~4–6 m RMSE; TanDEM-X: ~2 m; commercial stereo DSM: 0.3–1 m
Delivery format	GeoTIFF, GeoPackage, Shapefile or PostGIS-compatible vector layers with confidence attributes

Analytics Satellize can run

Multitemporal SAR backscatter composite	Mean and standard-deviation stacking of 15–30 Sentinel-1 IW scenes; speckle filtering with refined Lee filter; sigma-nought anomaly extraction along linear features	GeoTIFF anomaly layer with linear-feature vectors and per-feature backscatter contrast value
SWIR clay-mineral contrast map	Sentinel-2 Band 11 / Band 12 Clay Mineral Ratio composited over fallow-season acquisitions; percentile compositing to suppress tillage noise	Classified raster showing high-contrast zones aligned with candidate canal corridors, exported as GeoTIFF
CORONA baseline network tracing	Orthorectification of KH-4B frames using USGS-supplied camera models; manual-assisted vectorisation of canal and field-boundary features; co-registration to modern coordinate frame	Georeferenced vector network (Shapefile / GeoPackage) with feature-type classification and confidence flag
Multi-sensor convergence confidence layer	Weighted overlay of SAR anomaly, SWIR contrast, DEM micro-topography and CORONA trace; features scored 1–4 by number of independent sensor confirmations	Polygon and polyline GIS layer with convergence score, suitable for field-survey prioritisation
L-band subsurface anomaly map	ALOS-2 PALSAR-2 HH/HV ratio analysis; comparison with Sentinel-1 C-band to isolate features visible only at L-band, indicating subsurface rather than surface origin	Raster anomaly map with candidate subsurface linear features extracted as vectors
Network topology and catchment reconstruction	Hydrological flow-direction analysis on high-resolution DEM; canal vectors snapped to flow network; upstream catchment delineation for each identified intake point	GIS network dataset with flow hierarchy, estimated command areas and intake-point locations; PDF interpretive report

Sources

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.