Submerged archaeological site survey using satellite-derived bathymetry
Multispectral satellite imagery can resolve bottom features through optically shallow water using radiative transfer inversion, revealing submerged harbour moles, inundated field systems and drowned architecture to depths of roughly two to fifteen metres where turbidity allows.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible bands (Band 1 coastal aerosol at 60 m, Bands 2-4 at 10 m); 5-day revisit at mid-latitudes with two satellites. The coastal blue band (Band 1, 443 nm) and blue-green ratio are the primary inputs for radiative transfer depth retrieval. Free archive from 2015.
- WorldView-2 / WorldView-3: WorldView-2 introduced an 8-band multispectral suite including a dedicated coastal blue band (400-450 nm) at 1.85 m GSD (multispectral), enabling finer spatial discrimination of submerged structural edges. WorldView-3 adds a 1.24 m panchromatic band for pan-sharpening. Tasked commercially; archive coverage variable.
- ICESat-2 ATLAS: Photon-counting lidar at 532 nm; published water-penetration depth to roughly 40 m in clear oceanic water, though turbidity in coastal and lacustrine settings typically limits useful returns to 10-20 m. Along-track spacing approximately 0.7 m; cross-track separation of six beams spans roughly 3.3 km. Provides absolute depth control to validate and calibrate passive image-derived bathymetry.
- Landsat 8 / 9 OLI: 30 m resolution across coastal blue through SWIR bands; 16-day revisit per satellite (8-day combined). Coarser than Sentinel-2 for structural archaeology but the archive extends to 1982 (Landsat 4/5 TM), giving multi-decadal change detection useful for tracking inundation progression upstream of dams.
What the water column does to light, and why it matters
Seawater and fresh water attenuate sunlight exponentially with depth, and they do so selectively. Red wavelengths are absorbed within the first metre or two. Blue and green wavelengths travel deeper, scatter off the seabed, and return through the water column to a satellite sensor carrying information about bottom reflectance. This differential attenuation is the physical basis of satellite-derived bathymetry: if you know how the water absorbs and scatters light at each wavelength, you can invert the signal to recover both depth and bottom type.
The standard approach, published by Lyzenga in the early 1980s and extended by Stumpf and colleagues in 2003, uses the log-ratio of two visible bands, typically blue and green, to estimate depth. The ratio is relatively insensitive to bottom albedo variation, which matters enormously in archaeological contexts where you are looking for stone structures against sand or silt. Radiative transfer models such as the semi-analytical approach embedded in tools like HYDROLIGHT go further, explicitly modelling the contributions of phytoplankton, suspended sediment, and coloured dissolved organic matter. The honest limit: turbidity is the enemy. In clear Mediterranean or tropical water, useful depth retrieval reaches 10-15 m. In turbid estuaries or reservoir margins with high suspended load, that ceiling drops to 2-3 m, sometimes less.
What a floating roof gives away: the spectral signature of drowned structures
Submerged masonry, cut stone, and compacted earthworks reflect differently from natural sediment. Stone harbour moles typically appear as linear or rectilinear high-reflectance anomalies against lower-reflectance sand or mud. Inundated field boundaries show as grid-like patterns with slightly elevated backscatter in the coastal blue band. The key discriminant is geometry: natural seabed features, sandbanks and ripple fields are curved or diffuse. A Roman quay is straight.
WorldView-2's dedicated coastal blue band (centred near 425 nm) was designed partly with water-column penetration in mind and gives a meaningful advantage over sensors whose shortest band starts at 450 nm or longer. At 1.85 m multispectral resolution, it can resolve individual stone courses in very shallow, clear water, though this is the optimistic end of the performance envelope. Sentinel-2's 10 m Band 2 (490 nm) is adequate for detecting harbour basins and large inundated field systems but will not resolve individual wall faces.
Sun angle matters considerably. High solar elevation reduces surface glint and maximises the proportion of the signal attributable to bottom reflectance. Acquisition planning for archaeological bathymetry should target solar elevation above 50 degrees and sea-state below Beaufort 2. Both conditions are achievable in the Mediterranean between May and September.
Published results: Caesarea, Pavlopetri, and the Nile Valley reservoirs
The submerged harbour at Caesarea Maritima, built by Herod the Great and documented by underwater archaeological surveys since the 1980s, has been used repeatedly as a validation target for satellite bathymetry methods. The harbour's outer breakwaters lie at depths of 4-8 m in relatively clear eastern Mediterranean water, within the reliable retrieval window of Sentinel-2 and WorldView imagery. Published studies using Stumpf-ratio and semi-analytical inversion have recovered the principal mole alignments at horizontal accuracies consistent with the 10 m Sentinel-2 pixel, useful for site-extent mapping even if not for architectural recording.
Pavlopetri, off the Laconian coast of Greece, is among the best-documented submerged prehistoric settlements in the world. Lying at 3-4 m depth in clear water, its street plan and building outlines are detectable in high-resolution commercial imagery. The site has been surveyed with multibeam sonar and photogrammetric diving surveys; satellite imagery serves a complementary role, providing synoptic coverage and enabling change detection after storm events.
The Aswan High Dam, completed in 1970, inundated roughly 500 km of the Nile Valley, displacing Nubian communities and submerging hundreds of archaeological sites. Many lie in Lake Nasser at depths of 5-30 m. In the shallower margins, particularly during low-water periods when the reservoir level drops, Landsat and Sentinel-2 imagery has been used to map exposed and near-surface site distributions. ICESat-2 passes over the reservoir offer depth transects that can be combined with passive imagery to extend coverage into the 10-20 m zone.
ICESat-2 as depth control: what photon-counting lidar adds
Passive image-based bathymetry is always relative unless anchored to known depths. ICESat-2's ATLAS instrument fires 532 nm green laser pulses at 10 kHz and detects individual returned photons, distinguishing water-surface and seabed returns by their travel time. Published retrieval depths in clear coastal water reach 30-40 m; in turbid water the practical limit is closer to 10-15 m. The six-beam configuration provides along-track depth profiles at roughly 0.7 m point spacing.
For archaeological bathymetry, ICESat-2 transects serve two functions. First, they provide absolute depth calibration for the image-derived bathymetry, converting relative band-ratio values into metres. Second, in areas where image retrieval fails due to depth or turbidity, the lidar transects themselves may detect bottom relief from submerged structures, provided the structure has sufficient vertical relief (published estimates suggest features of 0.5-1 m height are detectable in favourable conditions). The limitation is spatial: ICESat-2 transects are lines, not grids, and their orbital repeat is 91 days. You get excellent depth accuracy along a narrow swath, not wall-to-wall coverage.
Honest limits: when satellite bathymetry will not work
Turbidity is the primary failure mode. Reservoirs receiving high sediment loads, such as those behind dams on silt-laden rivers, can have Secchi depths below one metre, rendering the water column opaque to satellite sensors at archaeologically relevant depths. Kelp, seagrass, and algal bloom cover confound bottom-reflectance retrieval by introducing spectrally variable canopy signals that mimic or mask structural features. Depth is a hard ceiling: no passive optical method recovers reliable bottom information below roughly 15 m in the clearest natural waters, and the practical limit in typical coastal conditions is closer to 8-10 m.
Structural detection also depends on contrast. A limestone wall against a limestone sand seabed may produce insufficient reflectance difference to discriminate. Organic-rich sediment draped over inundated earthworks reduces contrast further. Where these conditions apply, multibeam sonar or sub-bottom profiling remain the appropriate primary tools; satellite data can still contribute change detection, site-extent mapping, and acquisition planning.
From imagery to site map: the analytical workflow
A standard workflow begins with atmospheric correction, specifically the removal of surface glint and aerosol effects, using tools such as ACOLITE (developed at RBINS, Belgium) or the C2RCC processor in the Copernicus SNAP toolbox, both of which are designed for optically shallow water. Band-ratio depth retrieval or semi-analytical inversion follows, producing a relative or absolute depth grid. ICESat-2 transects are then used to calibrate and validate the depth grid against photon-counted returns.
Bottom-type classification separates sand, rock, vegetation and anomalous high-reflectance features. The anomalous features are then filtered by geometry: linear, rectilinear, or curvilinear shapes above a minimum length threshold are flagged as candidate anthropogenic structures. These candidates are cross-referenced with existing site records, historical charts, and, where available, aerial photography from low-water periods.
Satellize runs this workflow on Sentinel-2 open data combined with commercial WorldView tasking on client licence, and can integrate ICESat-2 transect data from the NASA Earthdata archive. The Tonga crop-estimation programme demonstrated the organisation's ability to operationalise multi-source remote sensing pipelines in sovereign contexts; the same data infrastructure applies here. Output is typically a georeferenced GIS layer of depth-classified bottom features with candidate archaeological anomalies flagged for field verification.
Typical figures
| Spatial resolution (passive optical) | 10 m (Sentinel-2 visible bands); 1.85 m multispectral / 0.46 m pan (WorldView-2/3) |
| Depth retrieval range (passive optical) | 0.5-15 m in clear water; typically 2-8 m in turbid coastal or lacustrine settings |
| ICESat-2 depth accuracy | Published vertical accuracy approximately 0.1-0.3 m in calm, clear water; along-track point spacing ~0.7 m |
| Revisit (Sentinel-2 two-satellite) | 5 days at mid-latitudes; usable cloud-free acquisitions depend on local climate |
| Spectral bands used | Coastal blue (400-450 nm), blue (490 nm), green (560 nm); red used for shallow-limit masking |
| Minimum detectable structural feature | Linear features >10-20 m length at Sentinel-2 resolution; >2-4 m at WorldView-2 resolution, subject to contrast and depth |
| Archive depth (Sentinel-2) | 2015 to present; Landsat archive extends to 1972 at 30-60 m resolution |
| ICESat-2 orbital repeat | 91-day exact repeat; transect coverage is linear, not areal |
| Delivery formats | GeoTIFF depth grids, GeoPackage or Shapefile anomaly layers, PDF site-extent report with uncertainty estimates |
Analytics Satellize can run
| Satellite-derived bathymetry (SDB) depth grid | Log-ratio band inversion (Stumpf 2003 method) or semi-analytical radiative transfer inversion (ACOLITE / C2RCC processors), calibrated against ICESat-2 photon-counted depths | GeoTIFF raster of water depth at 10 m or 2 m pixel resolution with per-pixel uncertainty band |
| Bottom-type classification map | Supervised spectral classification of depth-corrected bottom reflectance into sand, rock, vegetation, and anomalous high-reflectance classes | GeoPackage polygon layer with class labels and confidence scores |
| Candidate anthropogenic anomaly layer | Geometric filtering of classified anomalies by linearity, rectilinearity and minimum length threshold; cross-referenced against published site registers | Shapefile of flagged candidate features with coordinates, estimated depth, orientation and priority ranking for field verification |
| Multi-epoch change detection report | Differencing of depth-normalised bottom-reflectance mosaics across Sentinel-2 or Landsat archive acquisitions to detect sediment accretion, exposure, or structural change | PDF report with annotated change maps and time-series plots for selected sites |
| Acquisition optimisation plan | Solar elevation and sea-state modelling to identify optimal tasking windows for commercial WorldView acquisitions at target sites | Tasking brief with recommended acquisition dates, solar angles, and cloud-probability forecasts |
| ICESat-2 transect extraction and integration | Download and filtering of ATL03 global geolocated photon data from NASA Earthdata; water-surface and seabed photon separation; co-registration with passive image depth grid | CSV and GIS layer of calibrated depth transects with residual statistics against image-derived bathymetry |
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.