Crop-mark and soil-mark site prospection from orbit
Buried archaeological features alter soil moisture and nutrients, forcing differential crop growth that multispectral and hyperspectral satellites can read. Timing is everything: a single misplaced acquisition can miss the window by a full year.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and NIR bands (B4, B8), 20 m in red-edge (B5, B6, B7) and SWIR; 5-day revisit at mid-latitudes with twin satellites. The red-edge bands are the primary diagnostic channel for crop-stress anomalies over buried features.
- Planet SuperDove: 3 m resolution across 8 bands including two red-edge channels (at roughly 705 nm and 740 nm); daily revisit globally. Fine spatial detail resolves narrow ditches and wall-lines that Sentinel-2 blurs, though the radiometric depth of the red-edge bands is shallower than Sentinel-2's.
- PRISMA (ASI): Hyperspectral imager covering 400–2505 nm in 239 contiguous bands at roughly 30 m ground sampling distance; on-demand tasking. Contiguous spectral coverage allows full red-edge inflection-point mapping and detection of subtle chlorophyll-content gradients invisible to broad-band sensors.
- WorldView-3 SWIR: 8 SWIR bands at 7.5 m resolution alongside 1.24 m panchromatic. SWIR channels (1195–2365 nm) are sensitive to soil moisture differences in bare-soil conditions, extending prospection into periods when no crop is present.
Why a buried ditch shows up in a wheat field
The physics is straightforward, even if the timing is not. A ditch filled with humic sediment over centuries retains more moisture and releases more nutrients than the surrounding plough-soil. A buried wall or metalled road does the opposite: it impedes root penetration and drains quickly. These differences in water and nutrient availability produce measurable differences in chlorophyll concentration, leaf area index, and canopy water content.
Those physiological differences are what satellites detect. Healthy, nitrogen-rich plants reflect strongly in the near-infrared (roughly 750–900 nm) and absorb strongly in the red (around 670 nm). Plants under mild stress shift the red-edge inflection point, the steep rise in reflectance between 680 nm and 750 nm, toward shorter wavelengths. That shift is small, typically a few nanometres, which is why broad-band sensors only catch it when the contrast between stressed and unstressed plants is large enough to produce a visible tonal difference across a 10 m or 30 m pixel.
The narrow window: when prospection is actually viable
Crop-marks appear only during a specific phenological stage. For temperate cereals, the critical period is grain-fill through to ripening, roughly late May to mid-July in northern Europe, depending on variety and season. Before grain-fill, the canopy is still building and stress signals are masked by general vigour variation. After harvest, the field is bare and the crop-mark mechanism no longer operates.
Soil-marks, by contrast, appear on bare or very sparsely vegetated ground. They reflect differences in soil colour, texture, and moisture rather than plant physiology. They are best captured in spring before sowing or after autumn ploughing, and in dry conditions when moisture contrast between feature fills and matrix soils is greatest. A wet spring equalises soil moisture and suppresses the signal entirely.
This means that a single-year acquisition strategy is genuinely risky. A cloudy June, an early harvest, or an unusually wet spring can eliminate the opportunity for twelve months. Multi-year stacking of Sentinel-2 time series, selecting only acquisitions within the viable phenological window and ranking by vegetation index contrast, significantly improves detection rates. Published work using multi-annual Sentinel-2 stacks over known sites in the UK and Central Europe has confirmed features that single-date imagery missed entirely.
Which spectral indices carry the signal
NDVI (Normalised Difference Vegetation Index, using red and NIR) is the starting point but often not the end point. Over buried ditches in cereal crops, NDVI differences between mark and background can be as small as 0.02–0.05, well within the noise from variable soil background and atmospheric correction residuals at 10 m. The red-edge chlorophyll index (CIre), computed from Sentinel-2 bands B7 and B5, is generally more sensitive to chlorophyll concentration differences and suppresses soil background effects more effectively.
PRISMA's contiguous spectral coverage allows derivation of the red-edge position itself, not just a proxy ratio, and permits mapping of canopy water content via the 970 nm and 1200 nm water-absorption features. These additional dimensions help separate genuine archaeological anomalies from agronomic variation caused by drainage tiles, field tramlines, or soil type boundaries, which are the main sources of false positives in broadband analysis. Honest caveat: even with hyperspectral data, separating an archaeological ditch from a modern land-drain requires contextual knowledge of the site. Spectral data alone rarely closes that ambiguity.
Resolution floors and what they mean for feature detection
A Roman field boundary ditch might be 1–2 m wide. At Sentinel-2's 10 m NIR pixel, such a feature is sub-pixel and will only be detectable if the differential stress signal bleeds into surrounding vegetation strongly enough to affect the mixed-pixel value. In practice, Sentinel-2 reliably detects features wider than roughly 5–8 m or features that produce strong contrast over larger areas, such as enclosure ditches tens of metres across.
Planet SuperDove at 3 m resolves narrower linear features, but its red-edge bands sit at 20 m resampled in some pipeline configurations; users should verify the native band resolution for their specific tasking contract. WorldView-3 at 1.24 m panchromatic can image individual furrows but lacks the spectral depth to compute meaningful vegetation stress indices on its own. The practical approach for high-priority sites is a two-stage workflow: Sentinel-2 or PRISMA to identify candidate anomalies across a landscape, then commercial high-resolution tasking to characterise the geometry of confirmed marks.
From pixel anomaly to site candidate: the analytic chain
Raw spectral indices are not deliverables. The analytic chain starts with atmospheric correction (Sen2Cor for Sentinel-2, or the PRISMA Level-2 product), followed by topographic normalisation where terrain variation exists, then index computation and change detection across multiple acquisitions. Anomalies that appear in more than one independent acquisition within the viable phenological window, and that align with no modern agricultural or drainage infrastructure, are elevated to site candidates.
Those candidates are then cross-referenced against existing heritage inventories, historical maps, and where available, lidar DEMs. A crop-mark that aligns with a known earthwork on a lidar hillshade model is a strong candidate for ground investigation. One that sits in an area of intensive modern drainage with no corroborating evidence warrants lower confidence. Satellize applies this multi-source cross-referencing workflow across open Sentinel-2 archives and, for the Kingdom of Tonga crop-estimation programme, has refined the phenological windowing logic that underpins similar time-series approaches.
The output for a heritage client is typically a ranked GIS layer of anomaly polygons, each attributed with the number of independent detections, the vegetation index contrast value, the acquisition dates, and a confidence tier. That layer goes directly into a field-survey prioritisation workflow.
Honest limits and what ground-truth cannot replace
Satellite prospection finds anomalies. It does not confirm archaeology. The false-positive rate in any broadband crop-mark survey is non-trivial: modern field drains, geological stripes, and variable soil parent material all produce tonal patterns that resemble archaeological features in spectral indices. Hyperspectral data reduces but does not eliminate this problem.
Cloud cover over temperate Europe during the critical June-July window is a genuine operational constraint. In a typical UK summer, usable Sentinel-2 acquisitions over a given field during the grain-fill window may number two or three across a five-year archive, not per season. Sites in seasonally arid climates, such as the Mediterranean or the Near East, offer longer and more reliable windows for soil-mark detection on bare ground, and the archive depth of Sentinel-2 (from 2015) and Landsat (from 1972 for lower resolution) provides multi-decadal context unavailable to any ground-based survey. That archive is one of the most underused resources in landscape archaeology.
Typical figures
| Best spatial resolution (vegetation stress) | 3 m (Planet SuperDove red-edge); 10 m (Sentinel-2 NIR/red) |
| Best spatial resolution (soil marks, bare ground) | 1.24 m panchromatic (WorldView-3); 3 m multispectral (SuperDove) |
| Revisit frequency | Daily (Planet); 5 days (Sentinel-2 twin); on-demand (PRISMA, WorldView-3) |
| Key spectral bands | Red-edge 705–740 nm, NIR 750–900 nm, SWIR 1200–2400 nm, Red 660–680 nm |
| Viable prospection window (temperate cereals) | Late May to mid-July (grain-fill to ripening); varies by latitude and variety |
| Minimum detectable feature width (practical) | ~5–8 m with Sentinel-2; ~2–3 m with SuperDove in favourable contrast conditions |
| Archive depth | Sentinel-2 from 2015; Landsat from 1972 (30 m); Planet from ~2016 |
| Cloud-cover constraint | Acquisitions with >10% cloud over area of interest are typically discarded; temperate summers may yield only 2–4 usable scenes per season |
| Hyperspectral spectral range (PRISMA) | 400–2505 nm in 239 contiguous bands at ~30 m GSD |
| Delivery format | GeoTIFF index layers, GeoPackage anomaly polygons, attributed shapefile with confidence tiers |
Analytics Satellize can run
| Phenological window selector | Time-series NDVI and CIre profiling across Sentinel-2 archive to identify per-field grain-fill onset and peak stress dates | Per-field acquisition calendar with ranked usable scene dates; GIS layer |
| Multi-annual crop-mark composite | Median stacking of red-edge chlorophyll index anomaly rasters across all viable-window acquisitions, weighted by cloud-free pixel count | Anomaly heatmap GeoTIFF showing persistent tonal deviations across multiple seasons |
| Soil-mark bare-ground index | SWIR-NIR soil adjusted vegetation index (SAVI/MSAVI) computed on post-harvest or pre-sowing acquisitions; WorldView-3 SWIR or Sentinel-2 Band 11/12 | Ranked soil-mark candidate polygons with moisture-contrast values; shapefile |
| Red-edge position mapping (hyperspectral) | PRISMA Level-2 reflectance; linear four-point interpolation of red-edge inflection across 680–750 nm to derive per-pixel red-edge position in nm | Red-edge position raster; anomaly polygons where position deviates >3 nm from field median |
| False-positive filter: infrastructure cross-check | Spatial intersection of spectral anomaly candidates with OS/cadastral drainage records, modern field-tile registers, and geological survey soil-type boundaries | Filtered candidate layer with infrastructure-conflict flags; confidence-tiered GeoPackage |
| Heritage inventory cross-reference report | Spatial join of anomaly candidates against national heritage inventory polygons and lidar hillshade derivatives; Jaccard overlap scoring | PDF site-candidate report with spectral evidence, acquisition metadata, and inventory cross-reference per anomaly |
| Survey prioritisation ranking | Multi-criteria scoring combining detection frequency, spectral contrast magnitude, geometric regularity, and heritage inventory proximity | Ranked field-survey target list with map; exportable to standard heritage management GIS formats |
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.