Floodplain alluvial site inundation and sediment burial risk mapping
Seasonal floods and dam operations bury unexcavated alluvial sites under fresh sediment every year. Sentinel-1 SAR, SRTM elevation models and Landsat multispectral data can map inundation extent, quantify deposition, and rank site risk before the next flood pulse arrives.
Sensors
- Sentinel-1 C-band SAR (ESA): 10 m ground range resolution in Interferometric Wide Swath mode, 6-day repeat at mid-latitudes with both satellites active. C-band (5.405 GHz) backscatter drops sharply over open standing water, giving reliable flood-extent mapping even under partial cloud cover and light vegetation canopy. Coherence between pre- and post-flood acquisitions also captures surface roughness change consistent with fresh sediment deposition.
- SRTM Digital Elevation Model (NASA/NGA): 30 m horizontal posting, absolute vertical accuracy approximately 16 m at 90th percentile globally but considerably better over flat alluvial terrain. Used to derive flood-depth estimates by differencing water-surface elevation (from SAR extent plus a hydraulic slope assumption) against bare-earth topography, and to identify low-lying site locations within the inundation envelope.
- Landsat 8/9 OLI (USGS/NASA): 30 m multispectral, 16-day repeat per satellite (8-day combined). Band 6 (SWIR-1, 1.57–1.65 µm) and band 3 (Green) support MNDWI water-extent mapping in cloud-free conditions. Pre- and post-flood SWIR composites reveal turbid sediment plumes and deposition fans that SAR alone cannot spectrally characterise. Archive extends to 1984 across the full Landsat series.
- Copernicus Emergency Management Service (CEMS) Flood Products: Rapid mapping activations deliver delineated flood extents, grading maps and damage assessments, typically within 24–48 hours of activation. Products are derived from Sentinel-1 and contributing commercial SAR tasking. CEMS Flood grading distinguishes water depth classes and affected infrastructure, providing a validated baseline against which site-specific analysis can be anchored.
- Copernicus DEM (GLO-30, ESA/Airbus): 30 m and 10 m postings derived from TanDEM-X radar stereo, with vertical accuracy generally better than SRTM over vegetated and low-relief terrain. Preferred over SRTM for hydraulic modelling on low-gradient floodplains where sub-metre elevation differences control inundation extent.
Why alluvial sites are the hardest to protect
Alluvial floodplains are archaeologically dense for the same reason they are agriculturally productive: rivers deposit nutrients, concentrate resources and attract settlement across millennia. The Tigris-Euphrates lowlands, the Indus plain, the lower Nile and dozens of smaller river systems contain thousands of unexcavated sites whose surface expression is already subtle. A single flood season can add 5–30 cm of fresh overbank sediment to a low-mound site, deepening burial and compressing the window for non-invasive detection.
Dam operations make this worse in a specific way. Regulated releases flatten the natural flood pulse and can sustain inundation for weeks longer than a natural regime, maximising sediment settling time over site locations. Conversely, dams trap coarse bedload upstream, so downstream reaches receive finer, more cohesive sediment that seals anaerobic preservation conditions but also raises burial rates. Neither outcome is benign for heritage managers trying to prioritise survey effort.
What C-band backscatter actually detects, and where it fails
Open water returns very low C-band backscatter because the specular surface deflects the radar pulse away from the sensor. This contrast against rougher dry land is the physical basis for SAR flood mapping. Sentinel-1 in IW mode achieves 10 m resolution with a 250 km swath, making it practical for basin-scale flood monitoring at useful spatial detail. Studies using Sentinel-1 data over the Mekong and Indus floodplains have demonstrated detection of inundation extents consistent with gauge records to within a few percent of flooded area.
The limits matter. Dense emergent vegetation, particularly reed beds and rice crops at late growth stages, can maintain backscatter levels close to dry-land values even when the surface beneath is flooded. Flood depth below roughly 0.5 m under closed canopy is frequently missed. Double-bounce returns from flooded forest can actually increase backscatter, inverting the detection logic. These ambiguities mean SAR flood maps should always be cross-checked against MNDWI-derived Landsat extents in cloud-free windows, and against CEMS validated products where activations have occurred.
Sediment deposition rates from repeat SAR coherence
Interferometric coherence between two Sentinel-1 acquisitions over the same area decays when the surface changes between passes. Fresh sediment deposition disrupts the phase relationship that a stable surface would maintain. By comparing coherence values over known site locations in pre-flood and post-flood image pairs, it is possible to infer where surface change has been significant, though translating coherence loss to a quantitative deposition thickness requires ground-truth calibration or fusion with a hydraulic sediment transport model.
A more direct approach uses the Copernicus DEM or a locally acquired lidar DEM as a baseline, then differences it against a post-flood photogrammetric DSM derived from commercial stereo imagery. Where stereo archive coverage exists, this can yield deposition estimates at sub-decimetre vertical precision over small areas. Over large floodplains the approach is constrained by stereo availability and cloud persistence during flood events. Honest assessment: for most floodplain heritage contexts, SAR coherence gives a relative ranking of deposition intensity across sites rather than an absolute thickness measurement.
Building a site risk index from open data
The practical workflow begins with a site inventory, typically from national heritage registers, published survey reports or georeferenced excavation records, overlaid on the Copernicus DEM. Each site receives an elevation percentile within its local floodplain unit. Sites below the 10th elevation percentile in a reach with documented annual inundation are, by definition, flooded in most years. That is the first tier of risk.
Sentinel-1 multitemporal composites from the Copernicus Data Space archive, which extends back to 2014, allow flood frequency mapping: the fraction of flood-season acquisitions in which a given pixel shows low-backscatter water signal. Combining flood frequency with site elevation and an estimate of local sediment load (proxied from Landsat-derived turbidity in post-flood imagery) produces a ranked risk index. This is not a substitute for field survey, but it tells a heritage authority which sites to prioritise for ground inspection before the next wet season, and which to flag for emergency recording if dam release schedules change.
Honest limits of the satellite approach
Resolution is a persistent constraint. At 10 m, Sentinel-1 resolves a site footprint of roughly 0.01 ha as a single pixel cluster. Many alluvial sites, particularly those expressed only as low spreads of ceramic scatter or shallow occupation horizons, have no topographic expression detectable in a 30 m DEM. The method works best for sites with at least modest mound relief, say 0.5 m above the surrounding plain, which is a fraction of the total alluvial site population.
Cloud cover during monsoon and post-monsoon periods limits Landsat utility precisely when flood dynamics are most active. SAR fills much of this gap but cannot replace spectral characterisation of sediment composition. Archive depth for Sentinel-1 starts in 2014, which is sufficient for trend analysis but misses the dam-filling periods of major reservoirs commissioned before that date. For those, declassified CORONA imagery and Landsat TM archives from the 1970s onward provide the only satellite baseline, at coarser resolution and without SAR.
Satellize runs this class of analysis on open Sentinel and Landsat archives, with optional commercial SAR tasking for higher-frequency monitoring during active flood seasons. The Tonga crop-estimation programme demonstrated the same core workflow of multitemporal change detection over agricultural land, adapted here to heritage contexts.
From risk map to management decision
A flood-risk layer for alluvial heritage sites is only useful if it connects to a decision. The most immediate application is triage for emergency recording: sites ranked highest on the composite index become candidates for rapid drone photogrammetry or geophysical survey before the next flood pulse. A second application is input to environmental impact assessment for proposed dam or irrigation infrastructure, where regulators need quantified evidence of heritage exposure rather than qualitative concern.
Longer term, annual repeat analysis of flood extent and coherence change over a site inventory creates a documented record of cumulative burial. That record matters for two reasons. It supports the case for protective designation or buffer zone extension. And it provides the baseline against which any future mitigation, such as site capping or drainage engineering, can be evaluated. Satellite data does not stop a floodplain from flooding. It does make the argument for intervention harder to dismiss.
Typical figures
| SAR spatial resolution (Sentinel-1 IW) | 10 m ground range × 10 m azimuth |
| SAR revisit (Sentinel-1 A+B combined, mid-latitudes) | 6 days |
| DEM vertical accuracy (Copernicus GLO-30) | Typically ±4 m absolute; better over flat alluvial terrain |
| Optical resolution (Landsat 8/9 OLI) | 30 m multispectral; 15 m panchromatic |
| Optical revisit (Landsat 8+9 combined) | 8 days, cloud-permitting |
| Minimum detectable inundation patch (SAR) | ~0.01 ha at 10 m resolution; smaller features below detection threshold |
| Sentinel-1 archive depth | From 2014 (Copernicus Data Space) |
| Landsat archive depth | From 1972 (Landsat 1); consistent multispectral from 1984 (TM) |
| CEMS flood product latency | 24–48 hours post-activation for rapid mapping products |
| Flood frequency mapping output format | GeoTIFF raster, percentage of acquisitions showing inundation signal |
Analytics Satellize can run
| Flood extent time series (per flood season) | Thresholded Sentinel-1 VV backscatter change detection, cross-validated against Landsat MNDWI in cloud-free windows | Annual GeoTIFF flood-extent rasters and site-intersection table (CSV) showing inundation frequency per registered site |
| Site elevation risk index | Copernicus GLO-30 or SRTM DEM percentile ranking within local floodplain hydrological units, combined with flood frequency layer | Ranked site risk table (GIS layer, GeoPackage or Shapefile) with elevation percentile, flood frequency score and composite risk tier |
| Post-flood surface change map | Sentinel-1 interferometric coherence differencing between pre- and post-flood acquisition pairs to identify pixels with significant surface disturbance | Coherence-change GeoTIFF with site-overlay annotations; flagged sites exported as priority inspection list |
| Turbidity and sediment plume mapping | Landsat OLI Band 3 (Green) and Band 4 (Red) reflectance ratios calibrated against published turbidity proxies for post-flood imagery | Sediment load proxy raster per flood event, with site-location overlay indicating relative exposure to high-turbidity water |
| Multitemporal flood frequency composite (2014–present) | Pixel-wise count of low-backscatter acquisitions across the full Sentinel-1 archive, normalised to flood-season acquisition count | Flood frequency GeoTIFF (0–100 % scale) for the study reach, delivered as a single-band raster with accompanying metadata report |
| Dam-operation impact scenario layer | Hydraulic inundation modelling using Copernicus DEM and published or client-supplied dam release schedules, intersected with site inventory | Scenario comparison report (PDF and GIS layers) showing site exposure under current versus alternative release regimes |
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.