Water turbidity and suspended sediment concentration mapping
Suspended particles scatter red and near-infrared light in ways satellites can measure and calibrate to concentration. This page explains which sensors work at which scales, where the physics breaks down, and what the data can honestly tell a water manager.
Sensors
- Sentinel-2 MSI: 10-metre resolution in Band 4 (red, 665 nm) and Band 8 (NIR, 842 nm); 5-day revisit at the equator with both satellites. Best choice for estuaries, river plumes and small reservoirs where spatial detail matters more than frequency.
- Sentinel-3 OLCI: 21 spectral bands from 400 to 1020 nm at 300-metre resolution; near-daily global revisit. Designed for ocean colour and large water bodies. OLCI's dedicated water-leaving reflectance products (Level-2 WFR) are operationally processed by EUMETSAT.
- MODIS Aqua/Terra: 250-metre resolution in Bands 1 and 2 (red and NIR); one to two passes per day per sensor. Long archive from 2000 onwards makes it the standard tool for multi-decadal sediment trend studies, though spatial detail is limited for anything narrower than a large river.
- Landsat 8/9 OLI: 30-metre resolution across visible and NIR bands; 8-day combined revisit. Radiometric consistency between OLI and earlier Landsat sensors allows trend analysis back to the 1980s. Useful for medium-sized water bodies where Sentinel-2 coverage is intermittent.
What a muddy river tells the sensor
Suspended sediment changes the optical behaviour of water in a predictable direction. Clean water absorbs red and near-infrared light strongly, returning very little to a satellite sensor. Add silt, clay or fine sand and those particles scatter light back upward. The more sediment, the higher the water-leaving reflectance in the red band (roughly 620 to 700 nm) and, at higher concentrations, into the near-infrared. This scattering relationship is the physical foundation of every satellite turbidity algorithm.
The practical consequence is that a calibrated ratio or regression between band reflectance and in-situ measurements of either nephelometric turbidity (in NTU or FNU) or gravimetric total suspended matter (TSM, in mg/L) can be applied across an image. Published semi-analytical models such as the Nechad et al. single-band approach use the red band reflectance and two tunable coefficients to retrieve TSM. Empirical approaches fit a polynomial or exponential curve directly to field measurements. Both require in-situ calibration data from the specific water body to perform well, because particle size, composition and colour vary between systems.
Choosing the right sensor for the water body in question
Scale drives sensor selection more than any other factor. Sentinel-3 OLCI's 300-metre pixels are appropriate for open coastal waters, large lakes and shelf seas, where the signal is spatially homogeneous enough to fill a pixel cleanly. Its dedicated ocean-colour processing chain, maintained by EUMETSAT, delivers atmospherically corrected water-leaving reflectance operationally, which reduces the pre-processing burden considerably.
Sentinel-2 MSI changes the problem for narrower environments. A 10-metre red-band pixel can resolve a tidal creek, a river 30 metres wide, or the sharp gradient at the edge of a sediment plume entering an estuary. That spatial precision comes at a cost: MSI was not designed as an ocean-colour instrument, so its signal-to-noise ratio in the blue bands is lower than OLCI's, and atmospheric correction over water requires careful treatment. The ACOLITE processor, developed at the Royal Belgian Institute of Natural Sciences and documented in peer-reviewed literature, is widely used for this purpose with both Sentinel-2 and Landsat.
MODIS remains the instrument of choice for long time-series work. Its archive spans more than two decades and its twice-daily overpass frequency captures tidal and storm-driven variability that a 5-day revisit sensor will miss. Landsat 8 and 9 OLI bridge the gap between MODIS coarseness and Sentinel-2 detail, and their cross-calibrated archive enables decadal change analysis at 30-metre resolution.
Where the algorithms saturate and the adjacency effect bites
Standard single-band TSM algorithms begin to saturate above roughly 1000 mg/L. At those concentrations, the red band is already near-saturated and adding more sediment produces diminishing reflectance returns. Switching to a longer NIR band (such as Sentinel-2's Band 8A at 865 nm or Band 9 at 945 nm) extends the dynamic range, but NIR atmospheric correction is harder because water-leaving radiance at those wavelengths is very small and the aerosol signal dominates. Some published switching algorithms use the red band at moderate loads and the NIR at extreme loads, with a crossover threshold tuned to the local sediment regime.
The adjacency effect is a separate problem near shorelines. Bright land pixels, sand beaches or concrete infrastructure scatter light laterally in the atmosphere, contaminating the water signal in pixels within roughly one to several hundred metres of the boundary. The effect is worst in the blue and green bands and less severe in the red and NIR, which is one reason turbidity retrieval is more reliable than chlorophyll retrieval close to shore. Correcting for adjacency requires either spatial deconvolution in the atmospheric correction step or a conservative exclusion buffer around land. For Sentinel-2, a 60-metre buffer is a common operational compromise, though it sacrifices information in the narrowest channels.
Calibration, validation and the limits of remote sensing alone
A satellite turbidity map without in-situ calibration is a relative product, not an absolute one. The reflectance-to-TSM relationship shifts with particle size distribution: coarse sand reflects differently from fine clay at the same mass concentration. Seasonal changes in organic matter, algal content or industrial discharge can alter the relationship further. Any programme that needs TSM in mg/L rather than a qualitative turbidity index must invest in field sampling campaigns timed to satellite overpasses.
Validation is equally non-negotiable. Published studies on the Yangtze, the Amazon and the Rhine suggest that well-calibrated algorithms achieve root-mean-square errors in the range of 20 to 40 per cent of the measured TSM value under moderate sediment loads. Performance degrades at the extremes and in optically complex waters where multiple constituents co-vary. Cloud cover is the other hard limit: optical sensors see nothing through cloud, and in humid tropical catchments where sediment loads are highest during monsoon events, cloud fraction can exceed 80 per cent during the periods of greatest interest. Combining satellite observations with hydrological modelling or radar-derived discharge proxies is the standard response to this gap.
Operational applications and what the data can honestly support
Dredging monitoring is one of the most straightforward applications. A port authority can use Sentinel-2 imagery to track sediment plume extent during dredge operations, verify that plumes remain within permitted zones, and provide a documented record for regulators. The 10-metre resolution is sufficient to distinguish the plume from background turbidity in most port environments, provided atmospheric correction is applied consistently.
Reservoir sedimentation surveys use multi-temporal Landsat or Sentinel-2 composites to track the spatial pattern of high-turbidity inflow zones, which correlate with delta formation and loss of storage capacity. River basin authorities in sediment-rich catchments use MODIS time series to monitor whether upstream land-use change or dam construction is altering the long-term sediment flux to the coast, a question with direct implications for delta subsidence and coastal erosion. Satellize applies this class of analysis on open Sentinel and Landsat archives; the Tonga crop-estimation programme is a different domain, but the underlying workflow of calibrated reflectance retrieval against field measurements is structurally similar. For clients who need higher revisit or tasked imaging during specific hydrological events, commercial optical satellites can be added to the stack.
What the data cannot support, honestly stated: sub-daily tidal dynamics at high spatial resolution, reliable TSM retrieval through cloud, or absolute concentration values without field calibration. Those limits should be part of any project design from the outset.
Typical figures
| Spatial resolution (best available) | 10 m (Sentinel-2 red band); 30 m (Landsat 8/9 OLI); 300 m (Sentinel-3 OLCI); 250 m (MODIS bands 1–2) |
| Revisit frequency | 5 days (Sentinel-2 dual satellite, equator); 8 days combined (Landsat 8+9); ~1 day (Sentinel-3 OLCI); 1–2 passes/day (MODIS Aqua + Terra) |
| Spectral bands used | Red (620–700 nm) primary; NIR (740–900 nm) for high-load extension; blue-green for ancillary water quality context |
| Typical TSM retrieval range | ~1 mg/L to ~1000 mg/L with standard red-band algorithms; NIR switching extends to several thousand mg/L with reduced accuracy |
| Algorithm accuracy (published range) | RMSE roughly 20–40% of measured TSM under moderate loads in calibrated deployments; degrades at extremes |
| Cloud limitation | Full optical blockage; no retrieval through cloud. Tropical monsoon periods may yield <20% usable scenes |
| Archive depth | MODIS from 2000; Landsat from 1984 (with caveats on early radiometric consistency); Sentinel-2 from 2015; Sentinel-3 OLCI from 2016 |
| Adjacency effect buffer | Typically 60–200 m exclusion zone from shoreline depending on sensor and correction method |
| Deliverable formats | GeoTIFF reflectance and TSM maps; time-series CSV; WMS/WMTS tile service; PDF monitoring reports |
Analytics Satellize can run
| Calibrated TSM concentration map | Semi-analytical single-band retrieval (Nechad-type) or empirical regression against in-situ gravimetric samples, with ACOLITE or Sen2Cor atmospheric correction | GeoTIFF layer per scene, ingested into client GIS or delivered via WMS feed |
| Turbidity anomaly alert | Scene-to-baseline difference in atmospherically corrected red-band reflectance; threshold exceedance flagged automatically | Email or API alert with scene date, affected area in km², and map thumbnail |
| Sediment plume extent polygon | Thresholded classification of TSM map; plume boundary vectorised and attributed with area and centroid | Shapefile or GeoJSON per overpass, suitable for regulatory reporting |
| Multi-year TSM trend analysis | Seasonal Mann-Kendall trend test on MODIS or Landsat time-series pixel stacks; change-point detection for dam or land-use events | PDF report with trend maps, significance statistics and annotated time-series plots |
| Reservoir inflow turbidity monitoring | Sentinel-2 10-metre TSM mapping of inflow zones on each cloud-free overpass; spatial statistics per defined zone polygon | Monthly summary table and map pack; optional dashboard integration |
| Dredge compliance plume tracking | Per-scene plume delineation against permitted boundary polygons supplied by client; exceedance flagged with area and duration | Structured log file and GIS layer per event, formatted for submission to environmental regulator |
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.