Reservoir sediment turbidity and siltation monitoring for hydropower asset management
Optical satellites retrieve surface turbidity and total suspended matter concentrations across reservoir surfaces at 10–300 m resolution, giving hydropower operators a cost-effective complement to infrequent bathymetric surveys and early warning of post-flood siltation pulses.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible bands (B4 red, B8 NIR), 20 m in red-edge and SWIR; 5-day revisit at mid-latitudes with twin satellites. The B4/B3 ratio and the Normalised Difference Suspended Sediment Index are the primary turbidity proxies. Useless under cloud or at night.
- Landsat 8/9 OLI: 30 m resolution across red (Band 4) and NIR (Band 5); 16-day single-satellite revisit, 8-day combined. Calibrated surface reflectance products (Collection 2 Level-2) allow multi-decadal TSM time series back to 1984 via the Landsat archive, enabling long-term siltation trend detection.
- MODIS Aqua/Terra: 250–500 m resolution; daily revisit makes it useful for tracking fast-moving post-flood turbidity plumes in large reservoirs where spatial detail is secondary to temporal frequency. Detection of fine spatial gradients inside small reservoirs is not feasible at this resolution.
- Sentinel-3 OLCI: 300 m resolution; daily revisit for large water bodies. Designed primarily for ocean colour, but published studies have applied it to large inland reservoirs. Its 21-band configuration from 400–1020 nm supports more sophisticated bio-optical inversion than the four-band Sentinel-2 approach, though spatial detail is coarser.
What suspended sediment looks like from orbit
Water scatters and absorbs light in ways that shift predictably with particle load. Clean water absorbs strongly in the red and near-infrared; suspended mineral sediment backscatters strongly in those same bands. The ratio of red-band reflectance to green-band reflectance, or the Normalised Difference Suspended Sediment Index (NDSSI, computed from NIR and blue bands), therefore rises monotonically with total suspended matter (TSM) concentration across a useful working range, typically from around 5 mg/L up to several hundred mg/L before the signal saturates.
Sentinel-2 Band 4 (665 nm, red) and Band 8 (842 nm, NIR) at 10 m resolution resolve plume boundaries, river-entry fans, and stratified turbidity gradients that a bathymetric survey boat would take weeks to map. Landsat OLI Band 4 and Band 5 at 30 m resolution provide the same physics with a longer archive. Both sensors require atmospheric correction before band ratios are physically meaningful; the Sentinel-2 Sen2Cor processor and the Landsat Collection 2 Level-2 surface reflectance product apply that correction operationally.
Calibrating pixels to sediment concentrations
A pixel value is not a sediment concentration. Converting one to the other requires a site-specific or regionally validated empirical relationship, typically a power-law or exponential fit between in-situ TSM measurements and the corresponding atmospherically corrected band ratio. Published relationships for different reservoir types span a wide range of coefficients; applying a relationship calibrated on a lowland river reservoir to a glacially fed alpine reservoir will produce meaningless numbers.
Where in-situ data are sparse, published bio-optical model families, such as the Nechad et al. (2010) single-band algorithm or the QAA (Quasi-Analytical Algorithm) family, offer a physically grounded starting point. These have been validated against field data across dozens of water bodies. The honest position is that satellite-derived TSM concentrations carry uncertainties of roughly 20–40% relative to in-situ reference values under good atmospheric conditions, and larger uncertainties when thin cloud, aerosol haze, or sun-glint contaminate the scene. Operators should treat satellite TSM as a relative indicator and spatial pattern detector, not a laboratory-grade measurement.
Seasonal patterns, flood pulses, and the value of the archive
A single image shows where sediment is now. A time series shows where it is going. With Sentinel-2 providing clear-sky observations every five days and the Landsat archive extending to 1984, it is possible to reconstruct seasonal turbidity cycles, identify which tributary catchments contribute the largest sediment loads during monsoon or snowmelt events, and detect step-changes in siltation rate following upstream land-use change or extreme rainfall.
Post-flood sediment pulses are particularly informative. A large storm event can deliver more sediment to a reservoir in three days than the preceding twelve months of baseflow. Satellite imagery acquired within days of the flood peak captures the spatial extent and intensity of that pulse, information that is otherwise lost before the next scheduled bathymetric survey. MODIS, with its daily revisit at 250–500 m, is often the right sensor for tracking the pulse in real time across large reservoirs; Sentinel-2 then provides the spatial detail once the flood recedes.
Multi-year turbidity time series can also proxy sedimentation trends. If the turbid zone near the dam face expands progressively across dry-season images, that is a signal worth investigating with bathymetry. The satellite record does not replace the survey; it tells the operator when and where to commission one.
What satellites cannot see, and why that matters
Optical sensors measure the top optical depth of the water column, roughly the first metre or two in turbid conditions, less in clear water. Sediment deposited on the reservoir bed is invisible. Density currents carrying sediment along the thermocline are invisible. The satellite record is therefore a surface signal, not a volumetric one. Inferring bed-level change from surface turbidity alone is an extrapolation that requires supporting assumptions about settling dynamics and reservoir hydrodynamics.
Cloud cover is the other hard constraint. Tropical and monsoon-climate reservoirs, which are often the most heavily silted, are also the most persistently cloud-covered during the wet season, precisely when sediment loads are highest. A site at 10 degrees latitude may yield fewer than four usable Sentinel-2 scenes per month during the monsoon. Operators in these regions should plan for data gaps and weight their analysis accordingly. Combining optical imagery with radar-derived water-surface change from Sentinel-1 SAR (covered in a separate page on reservoir water-level monitoring) partially compensates, though SAR does not retrieve turbidity.
From spectral index to operational decision
The analytic chain runs from atmospherically corrected reflectance through TSM retrieval, spatial aggregation by reservoir zone (inlet fan, mid-reservoir, forebay), and temporal trending. The output that matters to an asset manager is not a colour map but a number: the rate at which the turbid inlet zone is expanding, the frequency with which TSM concentrations near the intake exceed a threshold associated with accelerated turbine abrasion, and the cumulative sediment-pulse exposure since the last bathymetric survey.
Satellize runs this chain on open Sentinel and Landsat data, with commercial tasking added where revisit frequency needs to increase around a specific event. The Tonga crop-estimation programme demonstrated that open-constellation time-series analytics, calibrated against sparse in-situ data, can produce operationally useful outputs in data-sparse environments. The same principle applies here. A hydropower operator with two bathymetric surveys per decade and a satellite-derived turbidity record covering the intervening years is in a materially better position than one relying on bathymetry alone.
The practical starting point is a cloud-frequency audit of the target reservoir, followed by a calibration data review. If the operator holds any historical water-quality measurements, even a handful of grab samples with timestamps and coordinates, those are enough to anchor an empirical TSM relationship and begin extracting value from the existing archive.
Typical figures
| Spatial resolution (primary) | 10 m (Sentinel-2 red/NIR bands); 30 m (Landsat 8/9 OLI) |
| Spatial resolution (high-frequency option) | 250–500 m (MODIS); 300 m (Sentinel-3 OLCI) |
| Revisit period | 5 days (Sentinel-2 twin); 8 days combined (Landsat 8+9); daily (MODIS, Sentinel-3 OLCI) |
| Spectral bands used | Red (660–670 nm), NIR (840–865 nm), green (560 nm), blue (490 nm) for NDSSI and single-band TSM algorithms |
| TSM retrieval range | Approximately 5–500 mg/L before signal saturation; site-specific calibration required |
| Typical retrieval uncertainty | 20–40% relative to in-situ under clear-sky, low-aerosol conditions; higher under haze or thin cloud |
| Sensing depth | Surface optical depth only, roughly 0.5–2 m depending on turbidity; bed sediment not detectable |
| Archive depth | Landsat: 1984 to present; Sentinel-2: 2015 to present; MODIS: 2000 to present |
| Cloud limitation | No data under cloud or at night; tropical sites may lose 60–80% of acquisitions during monsoon months |
| Deliverable formats | GeoTIFF TSM maps, zonal time-series CSV, PDF trend reports, GIS-ready vector zone overlays |
Analytics Satellize can run
| Surface TSM concentration maps | Atmospherically corrected red/NIR band-ratio retrieval calibrated against published empirical algorithms (e.g. Nechad single-band, NDSSI); site-specific power-law fit where in-situ data are available | Per-scene GeoTIFF with TSM concentration (mg/L) and a quality mask flagging cloud, shadow, and sun-glint contamination |
| Seasonal turbidity climatology | Multi-year clear-sky composite stacking by calendar month; percentile-based summary statistics per reservoir zone | Monthly median and 90th-percentile TSM rasters; PDF report with annotated seasonal cycle chart |
| Post-flood sediment pulse detection and quantification | Automated anomaly detection against seasonal baseline; MODIS for near-real-time pulse tracking, Sentinel-2 for spatial detail post-recession | Alert with estimated pulse arrival date, peak spatial extent, and TSM exceedance area relative to site-specific threshold |
| Inlet fan and forebay siltation trend index | Time-series regression of turbid-zone area (pixels above TSM threshold) against date; Mann-Kendall trend test for monotonic change detection | Annual trend summary GIS layer and statistical significance report; flags accelerating siltation for bathymetric survey prioritisation |
| Cloud-gap-filled turbidity time series | Harmonic or LOESS temporal interpolation across cloud-affected acquisitions; uncertainty bounds widened proportionally to gap length | Continuous daily TSM index time series per reservoir zone, delivered as CSV with per-date confidence flag |
| Tributary sediment-load ranking | Spatial attribution of turbidity plumes to individual inflow channels using plume-tracing on high-resolution Sentinel-2 imagery; relative load ranking by plume area and intensity | Ranked tributary contribution table and annotated image mosaic; supports catchment management prioritisation |
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.