Vegetation stress index mapping for buried feature detection beyond crop-marks
Buried walls, ditches and pits alter soil moisture and depth in ways that show up as differential plant stress, even where no crops grow. Red-edge and SWIR spectral indices can resolve those stress patterns from orbit, extending systematic site prospection into grassland, scrub and managed woodland.
Sensors
- Sentinel-2 MSI: 10 m visible and NIR bands, 20 m red-edge bands (B5 704 nm, B6 740 nm, B7 783 nm) and SWIR bands. Five-day revisit at mid-latitudes from the two-satellite constellation. Free archive from 2015. Red-edge bands are the primary input for the Red-Edge Chlorophyll Index (CIre) and PSRI calculations.
- WorldView-3: 31 cm panchromatic, 1.24 m multispectral, 3.7 m SWIR (eight bands, 1195–2365 nm). SWIR bands resolve lignin, cellulose and moisture absorption features invisible to Sentinel-2, sharpening stress boundaries to sub-metre scale. Tasked commercially; not free.
- PlanetScope: 3 m, near-daily revisit. Four to eight spectral bands depending on generation; newer SuperDove instruments include a red-edge band at 705 nm. Best used for dense time-series stacking to separate phenological background from persistent stress anomalies tied to buried features.
- Landsat 8/9 OLI: 30 m resolution, 16-day revisit per satellite (8-day combined). No red-edge band, but the coastal aerosol, NIR and SWIR-2 bands support PSRI and NDVI anomaly detection. Archive depth to 1972 (earlier sensors) enables multi-decade change context.
Why grass tells a different story from wheat
The classic crop-mark depends on a cereal crop responding to soil depth and moisture over a buried feature during a single dry season. Grassland, scrub and managed woodland behave differently: root systems are shallower and more plastic, species composition is mixed, and the stress response is subtler and more persistent across seasons rather than acute in a single window.
That subtlety is actually an advantage. Because the stress signal over buried masonry or a filled ditch is chronic rather than episodic, it can be extracted statistically from a time-series stack rather than hunted in a single lucky overpass. The physics is the same as for crop-marks: a buried wall reduces soil volume and water retention above it, stressing plants during dry periods. A filled pit or ditch retains moisture, producing a vigour anomaly. The spectral signature of either state is detectable in chlorophyll fluorescence proxies and senescence indices, provided the sensor has adequate spectral resolution in the red-edge region.
What the Red-Edge Chlorophyll Index and PSRI actually measure
The Red-Edge Chlorophyll Index (CIre) is defined as (NIR / Red-Edge) minus one. It is sensitive to chlorophyll concentration in the canopy, which declines before visible yellowing occurs. Sentinel-2 band B7 (783 nm) divided by B6 (740 nm) gives a practical CIre approximation at 20 m. Published work on Romano-British and Iron Age sites in upland Britain, including studies in the Cheviot Hills and on Dartmoor, has demonstrated that CIre anomalies of 0.1 to 0.3 index units above or below local background are detectable over known buried features when imagery is composited across multiple dry-season acquisitions.
The Plant Senescence Reflectance Index (PSRI) uses the ratio of red reflectance to green and NIR reflectance to track carotenoid-to-chlorophyll ratios during stress and senescence. It responds earlier in the stress cycle than NDVI and is less saturated in dense swards. WorldView-3 SWIR bands add a further dimension: liquid water absorption at 1450 nm and 1940 nm, and dry-matter absorption at 2100 nm, allow direct mapping of canopy water content gradients that correlate with soil moisture differences over buried features. The combination of CIre from Sentinel-2 and SWIR moisture indices from WorldView-3 is more diagnostic than either alone.
Separating archaeology from phenology: the time-series problem
The principal technical difficulty is that grassland stress varies spatially for entirely non-archaeological reasons: slope aspect, drainage class, grazing pressure, soil type transitions, and species composition all produce index variation that can mimic or mask a buried-feature signal. A single image is almost never sufficient.
The standard mitigation is to build a multi-year, multi-season composite from PlanetScope or Sentinel-2, then compute the coefficient of variation across the stack. Pixels that show consistent anomaly relative to their immediate neighbourhood across multiple dry seasons, regardless of the prevailing phenological state, are candidates for archaeological investigation. Pixels that vary with the regional phenological wave are not. This approach requires at least three to five dry-season composites to produce a reliable anomaly map; with the Sentinel-2 archive running from 2015, eight or more years of data are now available for most of Britain and Europe. Cloud cover in upland Britain is a genuine constraint: usable acquisitions in any given summer may number only four to eight per year, making the full archive essential rather than optional.
Managed woodland adds a further complication. Canopy closure limits understorey signal, and the canopy itself introduces a confounding stress response to soil variation. Deciduous woodland in early spring, before full leaf-out, offers a narrow window where the red-edge signal from low vegetation reaches the sensor with less canopy interference. That window is typically two to four weeks wide and varies by latitude and elevation.
Resolution floors and what they mean for feature size
Sentinel-2 red-edge bands at 20 m will not resolve a single Roman wall course. What they resolve is the stress halo around a buried structure: the zone of altered soil moisture that may extend two to ten metres beyond the physical feature. A buried enclosure ditch of 2 m width may produce a detectable anomaly footprint of 20 to 40 m if the soil contrast is sufficient, which is within Sentinel-2's capability. Smaller features, isolated post-holes or narrow gullies, are below the detection threshold at 20 m and require WorldView-3 or airborne hyperspectral follow-up.
WorldView-3 SWIR at 3.7 m narrows the gap considerably and can resolve individual wall lines where soil depth contrast is strong. The trade-off is cost and coverage: WorldView-3 tasking is priced per square kilometre and is practical for targeted confirmation of Sentinel-2 anomalies rather than landscape-scale prospection. PlanetScope at 3 m with a red-edge band sits between the two in both resolution and cost, and its near-daily revisit makes it the best option for dense time-series work over areas of tens to hundreds of square kilometres.
From anomaly map to field priority list
An index anomaly map is not an archaeological site map. Every flagged pixel requires cross-referencing against known geology, drainage, land management records and existing heritage inventories before it earns a field visit. The workflow that has proven most productive in published British upland studies runs in three stages: landscape-scale screening with Sentinel-2 CIre composites to identify candidate zones; targeted WorldView-3 or high-resolution PlanetScope acquisition over the top-ranked candidates; and finally, comparison with LiDAR-derived DTMs (where available) to test whether the spectral anomaly coincides with a micro-topographic feature. Agreement between spectral stress and micro-relief substantially increases confidence that the anomaly is structural rather than pedological.
Satellize runs this three-stage workflow as a configurable analytics pipeline, drawing on open Sentinel-2 archive data and adding commercial tasking where clients hold appropriate licences. The approach is directly analogous to the crop-estimation methodology developed for the Kingdom of Tonga programme, where multi-temporal index compositing was used to separate signal from background variation across a heterogeneous landscape. Output is a ranked GIS layer with anomaly confidence scores, delivered alongside a plain-language interpretation note for each priority zone.
Honest limits of the method
This technique works best in temperate climates with seasonal drought stress, on sites where buried features create measurable soil depth or moisture contrasts, and over vegetation communities with a meaningful chlorophyll signal. It performs poorly on deep alluvial or colluvial sequences where burial depth exceeds roughly one metre, on heavily improved or irrigated grassland where management overrides natural moisture variation, and on sites where post-depositional soil mixing has homogenised the contrast between feature fill and surrounding matrix.
Cloud cover in upland Britain, precisely where many Iron Age and Romano-British upland sites are found, limits usable Sentinel-2 acquisitions and extends the time needed to build a reliable composite. Atmospheric correction quality also matters: the red-edge bands are sensitive to aerosol and water vapour variation, and poorly corrected imagery introduces artefacts that can resemble stress patterns. The method produces leads, not conclusions. Ground-truthing remains non-negotiable.
Typical figures
| Primary spectral resolution (red-edge) | Sentinel-2: 20 m at B5/B6/B7; WorldView-3 SWIR: 3.7 m; PlanetScope red-edge: 3 m |
| Revisit frequency | Sentinel-2: 5 days (two-satellite); PlanetScope: near-daily; WorldView-3: tasked, typically 1–4 days |
| Usable dry-season acquisitions (upland Britain, cloud-limited) | Typically 4–8 per year per sensor; full archive compositing essential |
| Archive depth | Sentinel-2: 2015 to present; Landsat: 1972 to present; PlanetScope: 2016 to present |
| Minimum detectable anomaly (CIre) | ~0.1–0.3 index units above local background, based on published upland British studies |
| Minimum feature footprint detectable | ~20–40 m stress halo at Sentinel-2 20 m; individual wall lines at WorldView-3 3.7 m SWIR where soil contrast is strong |
| Key spectral bands | Red-edge: 704–783 nm; SWIR moisture: 1195–1940 nm; SWIR dry matter: 2100–2365 nm |
| Delivery format | GeoTIFF anomaly rasters, ranked vector point/polygon GIS layer, PDF interpretation note |
Analytics Satellize can run
| Multi-year CIre anomaly composite | Red-Edge Chlorophyll Index time-series compositing with local neighbourhood z-score normalisation | GeoTIFF raster and ranked candidate polygon layer, with per-polygon confidence score |
| PSRI senescence anomaly map | Plant Senescence Reflectance Index calculated from Sentinel-2 B4/B3/B8 or WorldView-3 multispectral; dry-season composite | GeoTIFF raster overlaid on OS or national basemap; anomaly zones flagged for cross-referencing |
| WorldView-3 SWIR moisture gradient map | Canopy water content index from SWIR bands 1450 nm and 1940 nm; comparison with Sentinel-2 anomaly footprints | 3.7 m resolution GeoTIFF; boundary shapefiles for priority zones |
| Phenological separation layer | Coefficient of variation across multi-year PlanetScope stack; persistent anomaly pixels isolated from seasonal phenological wave | Binary persistent-anomaly mask as GeoTIFF; summary statistics per candidate zone |
| Heritage inventory cross-reference report | Spatial join of anomaly layer against national or regional heritage register polygons; geology and drainage filtering | PDF report listing ranked candidates, known site overlaps, and recommended field-visit priority order |
| LiDAR-spectral coincidence analysis | Overlay of spectral anomaly centroids on client-supplied or open LiDAR DTM; micro-topographic agreement scoring | GIS layer with coincidence score per candidate; flagged high-confidence targets for ground survey |
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.