Reservoir bank erosion and delta sedimentation from multi-temporal imagery
Cyclically fluctuating reservoir levels expose bank materials to wetting, drying and gravity in patterns that satellite time-series can decode. Sentinel-1 InSAR detects centimetre-scale slumping; Sentinel-2 water indices track the moving shoreline that normalises how long each bank zone was exposed.
Sensors
- Sentinel-1 (C-band SAR, ESA): 5.6 cm wavelength, Interferometric Wide Swath mode at 5 x 20 m ground range resolution, 6-day repeat at mid-latitudes with both satellites. Coherence maps reveal surface disturbance on exposed banks; displacement time-series via SBAS or PS-InSAR resolves line-of-sight motion to roughly 5-10 mm per epoch on stable, coherent ground.
- Sentinel-2 MSI (ESA): 10 m resolution in visible and near-infrared bands, 20 m in shortwave infrared. 5-day revisit (combined 2A/2B). NDWI (Green/NIR) and MNDWI (Green/SWIR) accurately delineate the water boundary; SWIR penetrates shallow turbid water better than NIR, improving waterline precision in silt-laden reservoirs.
- Landsat 8/9 OLI (USGS/NASA): 30 m multispectral, 15 m panchromatic, 16-day single-satellite revisit. Archive extends to 1972 across the full Landsat family, making it the primary source for decadal shoreline change and delta progradation or retreat. Band 6 (SWIR-1) supports MNDWI computation consistent with Sentinel-2 workflows.
- Planet SkySat: Sub-metre (0.5 m) panchromatic, 0.8 m colour. Tasked on demand. Provides the ground-truth optical detail needed to classify slump morphology, tension cracks and fresh debris lobes that are below Sentinel-2's detection floor. No free archive; acquisition requires a commercial licence.
- SRTM / Copernicus DEM: 30 m global DEM used to extract bank slope and aspect, which controls how long each zone drains after inundation. The Copernicus GLO-30 product (released 2021) improves on SRTM in vegetated terrain and is freely available via Copernicus Data Space.
What a fluctuating waterline does to a bank
Every reservoir operated for hydropower, irrigation or flood control cycles its level, sometimes by tens of metres across a season. Each drawdown exposes a band of bank material that was recently saturated. Clay-rich soils lose shear strength rapidly when wet and regain it slowly; repeated saturation-desiccation cycles cause progressive weakening even without a single dramatic failure event. The result is a characteristic bathtub ring of erosion scars, slump terraces and fresh debris that is visible in optical imagery and, more usefully for hazard assessment, measurable as pre-failure displacement by SAR interferometry.
The critical variable is exposure time: a bank zone uncovered for two weeks behaves very differently from one uncovered for three months. Mapping the waterline at every available satellite pass, and integrating the result into a per-pixel exposure-duration raster, converts a qualitative observation into a quantitative input for stability modelling. Sentinel-2's five-day revisit makes this practical at reservoir scale.
Mapping the moving shoreline with water indices
NDWI (Green minus NIR, divided by their sum) has been used to delineate open water since McFeeters published the index in 1996. MNDWI substitutes SWIR-1 for NIR and performs better in turbid, sediment-laden water because SWIR is absorbed more completely by water and reflected more strongly by suspended sediment, sharpening the land-water contrast. In reservoir settings where the inflowing river carries a heavy sediment load, MNDWI on Sentinel-2 Band 3 and Band 11 is the preferred choice.
A time-series stack of classified water masks, one per cloud-free acquisition, produces a water-frequency raster: each pixel carries a value from zero (never inundated in the period) to one (always inundated). The gradient zone between 0.1 and 0.9 frequency defines the active fluctuation band. Differencing the waterline position between acquisitions gives a shoreline displacement vector. Over a Landsat archive stretching back to the 1980s, decadal delta progradation or bank recession becomes measurable at 30 m precision, with sub-pixel refinement available through image cross-correlation on the sharper Sentinel-2 record.
Honest limit: cloud cover over tropical and monsoon-climate reservoirs can reduce usable Sentinel-2 acquisitions to fewer than one per month during the wet season, precisely when the most dramatic level changes occur. Sentinel-1 SAR, being cloud-independent, fills this gap for waterline mapping using backscatter thresholding, though the 5 x 20 m range-azimuth geometry introduces directional bias near steep banks.
InSAR coherence and displacement on exposed banks
Once the bank is above the waterline, Sentinel-1 interferometry can measure surface motion. Two mechanisms matter. First, coherence loss: a slumping or cracking bank surface changes its scattering geometry between passes, decorrelating the interferogram. A coherence map therefore acts as a change detector even when displacement is too chaotic to unwrap cleanly. Second, where the surface remains coherent enough for phase unwrapping, Small Baseline Subset (SBAS) or Persistent Scatterer (PS) processing resolves line-of-sight displacement time-series. Published studies on reservoir bank instability have detected pre-failure motion of a few millimetres per month months before visible slumping, which is the practical value of the technique.
C-band SAR has a known weakness here. Vegetation on the bank, even sparse scrub, decorrelates rapidly between passes and masks the ground motion beneath. L-band SAR (ALOS-2 PALSAR-2, or the forthcoming NISAR mission) penetrates low vegetation more effectively, but ALOS-2 revisit over any given site is typically 14 days at best and coverage is not global by default. For vegetated banks, the honest answer is that C-band InSAR will show you the bare or sparsely covered zones and leave the vegetated zones as a gap.
The geometry of a reservoir bank also creates a layover and shadow problem. Steep banks facing away from the satellite look-direction fall into radar shadow; those facing toward it suffer layover where the radar return from the top of the bank arrives before the return from the base. Careful selection of ascending and descending orbit passes, combined with slope-aspect analysis from the DEM, is needed to identify which bank segments are geometrically observable at all.
The ambiguity at the delta head: compaction or erosion?
Where the inflowing river meets the reservoir, sediment drops out of suspension and builds a delta fan. This fan surface subsides for two reasons that are physically distinct but produce similar InSAR signals. True erosion removes material laterally, which shows as shoreline retreat in optical imagery. Sediment compaction under self-weight causes vertical settlement without any lateral loss of material. Both appear as negative line-of-sight displacement in a descending-orbit interferogram.
Separating them requires combining the optical shoreline time-series with the InSAR displacement field. If the shoreline is advancing (delta progradation) while InSAR shows subsidence, compaction is the dominant process. If the shoreline is retreating while InSAR shows subsidence, both erosion and compaction are active. If the shoreline retreats but InSAR shows no displacement, wave or current erosion is removing material from the subaqueous delta face without disturbing the exposed surface above. Each combination points to a different management response.
A further complication: newly deposited fine sediment on a delta fan has low radar backscatter and poor interferometric coherence. The very zones most likely to be compacting are often the least amenable to InSAR measurement. This is not a solvable problem with current open-constellation C-band data; it is a known limit that any honest analysis must acknowledge.
Integrating the outputs into a bank-stability product
The practical deliverable is a ranked inventory of bank segments, each characterised by exposure-duration frequency, observed InSAR displacement rate, slope angle from the DEM, and optical evidence of past failure morphology. Segments with high exposure frequency, measurable downslope displacement and steep angles above fine-grained lithology are the priority for field investigation or engineering intervention.
Satellize runs this multi-source workflow on open Sentinel and Landsat archives, adding commercial SkySat tasking where sub-metre morphological classification is needed to distinguish tension cracks from drainage gullies. The workflow is structurally similar to the crop-area estimation approach used in the Kingdom of Tonga programme: systematic compositing of a time-series archive, per-pixel statistics, and a ranked output layer rather than a single snapshot classification.
Update cadence matters operationally. A reservoir that drawdowns rapidly in spring may move through its most vulnerable bank-exposure window in six to eight weeks. An analysis updated every 12 days on Sentinel-1 and every five days on Sentinel-2 (cloud permitting) can track that window in near-real time. An annual assessment cannot.
What this method cannot do
Satellite imagery does not measure pore-water pressure, which is the proximate control on bank failure. It measures surface expression: displacement, shoreline position, morphological change. A bank can be at critical pore pressure with no detectable surface motion until the moment it fails. The satellite record is therefore a screening tool that identifies where to deploy in-situ piezometers, not a replacement for them.
Below-water bank geometry is invisible to all optical and SAR systems. Subaqueous erosion undermining an apparently stable bank above the waterline is a real failure mode that remote sensing cannot detect. Bathymetric survey, repeated by boat or drone sonar, remains necessary for any site where subaqueous undercutting is a credible mechanism. Satellite data narrows the list of sites that need that survey.
Typical figures
| Optical waterline resolution | 10 m (Sentinel-2 MNDWI); 30 m (Landsat OLI); 0.5 m (SkySat, tasked) |
| SAR displacement resolution | 5 x 20 m (Sentinel-1 IW mode, ground range x azimuth) |
| Minimum detectable line-of-sight displacement | Approximately 5-10 mm per epoch on coherent bare ground (Sentinel-1 C-band SBAS); degrades to undetectable on vegetated or freshly deposited surfaces |
| Revisit cadence | 6 days SAR (Sentinel-1A+B combined, mid-latitudes); 5 days optical (Sentinel-2A+B); 16 days (Landsat 8 or 9 individually) |
| Archive depth | Sentinel-1 from 2014; Sentinel-2 from 2015; Landsat from 1972 (MSS), 1984 (TM), 2013 (OLI) |
| Spectral bands used | Sentinel-2: Band 3 (Green, 560 nm), Band 8 (NIR, 842 nm), Band 11 (SWIR-1, 1610 nm) for MNDWI; Sentinel-1: C-band 5.405 GHz VV/VH polarisation |
| Cloud sensitivity | Optical (Sentinel-2, Landsat) fully blocked by cloud; SAR (Sentinel-1) cloud-independent |
| Typical analysis latency | 2-5 days from satellite acquisition to delivered layer, depending on processing queue and cloud cover |
| Delivery formats | GeoTIFF displacement and water-frequency rasters; GeoPackage or Shapefile shoreline vectors; PDF ranked-segment report |
| DEM input | Copernicus GLO-30 (30 m, global, released 2021) for slope-aspect and shadow-layover masking |
Analytics Satellize can run
| Water-frequency raster and active fluctuation band | MNDWI thresholding on Sentinel-2 and Landsat time-series stacks; per-pixel inundation frequency calculation | GeoTIFF raster (0-1 frequency values) plus shapefile of fluctuation-band boundary; updated each quarter or after major drawdown events |
| Shoreline displacement vectors | Waterline position extracted per acquisition, differenced between epochs; sub-pixel refinement by image cross-correlation where archive density allows | Polyline shapefile with displacement magnitude and direction per segment per epoch; summary statistics table in PDF report |
| InSAR displacement time-series on exposed banks | SBAS processing of Sentinel-1 IW SLC pairs; coherence masking to exclude vegetated and inundated zones; LOS-to-vertical projection using DEM slope | GeoTIFF displacement velocity map (mm/year) plus per-point time-series CSV for flagged segments; updated every 12-day Sentinel-1 cycle |
| Coherence-change detection map | Sentinel-1 interferometric coherence computed per 12-day pair; anomalous coherence loss flagged against a rolling baseline of stable-season coherence | Binary change raster and alert email when coherence loss exceeds threshold in a user-defined zone of interest |
| Delta-head compaction vs. erosion classification | Joint interpretation of InSAR LOS displacement sign and magnitude against optical shoreline change direction; per-segment process attribution | Classified polygon layer (compaction-dominant / erosion-dominant / mixed / unobservable) with confidence rating; included in quarterly PDF report |
| Ranked bank-segment stability inventory | Multi-criteria scoring combining exposure frequency, InSAR displacement rate, DEM slope, and optical failure-morphology classification | Ranked GeoPackage table with per-segment scores; PDF summary identifying top-priority segments for field investigation or bathymetric survey |
| Sub-metre slump morphology classification (tasked) | SkySat imagery interpreted for tension crack lineaments, debris-lobe boundaries and fresh scarp faces; manual and semi-automated object-based classification | Polygon shapefile of failure morphology classes; delivered within 5 days of tasked acquisition |
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.