Coastal turbidity regime analysis for bivalve aquaculture site selection
Satellite time series from Sentinel-2 and Landsat reveal the seasonal turbidity envelope of candidate bivalve sites far more completely than point sampling can. Suspended particulate matter concentration governs both food supply and gill-clogging risk, making its statistical distribution the primary siting criterion.
Sensors
- Sentinel-2 MSI: 10 m visible bands and 20 m red-edge bands enable turbidity and SPM retrieval at the spatial scale of individual lease plots. Five-day revisit at the equator, shorter at higher latitudes with two-satellite constellation. Archive from 2015 (Sentinel-2A) and 2017 (Sentinel-2B).
- Landsat 8/9 OLI: 30 m multispectral with a well-characterised radiometric record extending to 2013 (Landsat 8) and 2021 (Landsat 9). Combined 8-day revisit. Coastal aerosol band (443 nm) improves atmospheric correction in turbid coastal scenes. Landsat 7 extends the archive to 1999 at reduced quality.
- MODIS Aqua/Terra: 250–500 m resolution, daily revisit. Unsuitable for resolving individual lease plots but valuable for capturing event-driven turbidity pulses (storm resuspension, river flood plumes) at synoptic scale. Ocean colour bands calibrated for water-leaving radiance retrieval.
- Sentinel-3 OLCI: 300 m resolution, approximately daily revisit. Designed for ocean colour with 21 spectral bands from 400 to 1020 nm. Better suited to open coastal waters than to narrow estuaries where adjacency effects from land contaminate pixels. Useful for regional SPM context around candidate sites.
Why point sampling produces a dangerously incomplete picture
A turbidity measurement taken on a calm Tuesday in April tells a farm developer almost nothing useful. Bivalve mortality events are driven by episodic peaks: storm resuspension, flood plumes, dredging operations, algal crash sedimentation. A single site visit, or even a monthly monitoring programme, will miss most of them. The distribution of suspended particulate matter (SPM) concentration across seasons and across inter-annual variability is what matters, and building that distribution from in-situ sampling alone requires years of dense field effort that most feasibility studies cannot afford.
Satellite optical imagery solves the sampling frequency problem. Sentinel-2 revisits a given coastal location every five days under cloud-free conditions. Landsat 8 and 9 together add an 8-day cycle. Over a five-year archive, a candidate site accumulates several hundred cloud-free observations. That is enough to characterise the 10th, 50th and 90th percentile SPM concentrations by month, to identify the seasonal window when conditions are most favourable, and to flag sites where high-turbidity events cluster in the post-storm period that coincides with spat settlement.
What a floating roof gives away: the physics of turbidity retrieval
Suspended particulate matter scatters light strongly in the red and near-infrared portions of the spectrum. In clear oceanic water those wavelengths are absorbed quickly and contribute little to the signal reaching a satellite sensor. In turbid coastal water, backscatter from particles elevates the water-leaving radiance measurably at 665 nm, 705 nm and 740 nm, the bands Sentinel-2 carries at 10 and 20 m resolution. Empirical and semi-analytical algorithms relate this elevated reflectance to SPM concentration in milligrams per litre, with published studies reporting reasonable accuracy across a range of estuarine and coastal environments.
The relationship is not universal. Particle size, composition and organic content all shift the reflectance-to-SPM ratio. An algorithm calibrated on mineral-dominated Rhine plume sediments will perform poorly in an estuary dominated by fine organic flocs. This is an honest limit of the method: site-specific or regionally validated algorithms consistently outperform globally generic ones, particularly at high SPM concentrations above roughly 100 mg/L where the signal begins to saturate in shorter wavelengths. Any rigorous siting study should include at least a handful of concurrent in-situ measurements to anchor the satellite retrieval to local optical properties.
Atmospheric correction: the step that most often goes wrong
In open ocean colour remote sensing, atmospheric correction removes the contribution of aerosol scattering using the assumption that water-leaving radiance is zero in the near-infrared. That assumption fails completely in turbid coastal water, where particles scatter strongly at NIR wavelengths. Standard processors applied uncritically to estuarine Sentinel-2 imagery can produce negative reflectance values or SPM retrievals that are off by an order of magnitude.
Alternatives exist. The ACOLITE processor, developed at the Royal Belgian Institute of Natural Sciences and widely used in the research literature, applies a short-wave infrared-based correction that performs better in turbid conditions. The iCOR processor takes a different approach using MODIS atmospheric data. Neither is a solved problem: performance degrades in very shallow water where bottom reflectance contaminates the signal, in scenes with sun glint, and at very high aerosol optical depths following dust events or wildfires. The ESA Sentinel-2 documentation and the published ACOLITE literature are explicit about these limits. Any operational turbidity product for a bivalve siting study should document which correction was applied and provide uncertainty estimates alongside the SPM values.
Building the turbidity climatology a developer actually needs
The analytic output for a siting decision is not a single turbidity map. It is a statistical profile: monthly median SPM, the frequency of exceedance above a biologically relevant threshold (often cited in the literature at around 50–100 mg/L for gill-clogging risk, though species-specific tolerance varies), and the duration of high-turbidity episodes following rainfall events of different magnitudes. Pairing this with a river discharge record from a national gauge network allows the analyst to model how turbidity at a candidate site responds to catchment hydrology.
Spatial comparison across multiple candidate sites within the same estuary is where satellite data adds the most value over point sampling. Two sites 3 km apart can sit on opposite sides of a turbidity gradient that shifts with tidal phase and wind direction. A multi-year time series resolves that gradient statistically, showing which site spends more time in the productive mid-turbidity range where phytoplankton-derived organic particles are present but mineral sediment loading is not yet damaging. That comparison is impossible from a handful of field visits.
Satellize's approach to this kind of analysis draws on the same open-archive processing pipeline used in its Tonga crop-estimation programme, adapted for coastal water optics rather than vegetation indices. The core methodology is publicly documented; what changes is the calibration and the site-specific validation layer.
Cloud cover and the limits of optical time series in high-rainfall coasts
The most turbid events in a coastal system tend to follow the heaviest rainfall. Those are also the days most likely to be cloud-covered. This is not a coincidence; it is a systematic bias in optical satellite archives that any honest turbidity climatology must acknowledge. In persistently cloudy regions, such as the west coasts of Scotland, Norway or British Columbia, cloud-free observations may cluster in the drier, calmer seasons, precisely when turbidity is lowest. The resulting archive underrepresents the high-turbidity tail of the distribution.
There is no clean solution. Partial mitigations include using MODIS or Sentinel-3 data to characterise storm-period turbidity at coarser resolution, modelling the missing observations using rainfall-runoff relationships, and being explicit in the site report about which percentile estimates are well-constrained by observations and which are extrapolated. A siting study that presents a smooth seasonal curve without confidence intervals has almost certainly not addressed this problem.
From analysis to a licence-ready site dossier
Regulatory bodies in most jurisdictions require environmental baseline data as part of an aquaculture site licence application. A satellite-derived turbidity climatology, properly documented with sensor metadata, atmospheric correction method, validation measurements and uncertainty ranges, can form a significant part of that baseline. It demonstrates multi-year monitoring at a cost and spatial coverage that in-situ programmes cannot match.
The practical output is a set of GIS layers covering the candidate area: seasonal median SPM maps, exceedance frequency maps, and a ranked comparison table if multiple sites are under evaluation. Paired with tidal flushing analysis (covered separately in the sibling page on tidal current assessment) and harmful algal bloom risk (covered in the HAB detection page), the turbidity climatology becomes one component of a complete environmental siting dossier rather than a standalone product.
Typical figures
| Spatial resolution (primary) | 10 m (Sentinel-2 visible), 20 m (Sentinel-2 red-edge), 30 m (Landsat 8/9 OLI) |
| Spatial resolution (synoptic context) | 250–500 m (MODIS), 300 m (Sentinel-3 OLCI) |
| Revisit frequency | 5 days (Sentinel-2 two-satellite), 8 days combined (Landsat 8+9), daily (MODIS, Sentinel-3) |
| Key spectral bands for SPM retrieval | Red (665 nm), red-edge (705, 740 nm), NIR (842 nm), SWIR (1610, 2190 nm for atmospheric correction) |
| Archive depth | Sentinel-2 from 2015; Landsat 8 from 2013, Landsat 7 from 1999 (scan-line corrector failure from 2003 reduces coverage) |
| Typical SPM detection range | Approximately 1–1000 mg/L; retrieval uncertainty increases above ~100 mg/L due to NIR saturation |
| Cloud contamination | Optical only; cloudy acquisitions excluded. High-rainfall coasts may yield fewer than 30 usable Sentinel-2 scenes per year |
| Minimum resolvable feature | Lease-plot-scale gradients visible at 10 m; narrow tidal channels (<20 m wide) affected by mixed-pixel land adjacency |
| Atmospheric correction dependency | SWIR-based correction (e.g. ACOLITE) required for turbid coastal water; standard NIR-based OC correction not appropriate |
| Deliverable latency | Climatology products: weeks (archive processing); near-real-time SPM maps: 1–3 days after acquisition |
Analytics Satellize can run
| Multi-year SPM climatology maps | Semi-analytical or empirical SPM retrieval (red/NIR band ratio or Nechad-type algorithms) applied to full Sentinel-2 and Landsat archive for candidate area | GIS layers (GeoTIFF) of monthly median, 10th and 90th percentile SPM concentration per pixel, with pixel-count metadata indicating observation density |
| High-turbidity exceedance frequency | Threshold exceedance counting on per-pixel SPM time series; threshold set at biologically relevant concentration specified by client or drawn from published bivalve physiology literature | Raster map of exceedance frequency (fraction of cloud-free observations above threshold) per month and annually, exported as GeoTIFF and summary CSV |
| Storm-event turbidity response profiles | Event compositing: satellite acquisitions within 3–10 days of rainfall events above a defined quantile, extracted from national gauge or ERA5 reanalysis precipitation; SPM anomaly relative to seasonal baseline computed per event | Time-series chart and tabular summary of mean SPM anomaly magnitude and recovery duration per site, delivered as PDF report section |
| Candidate site ranking table | Multi-criteria scoring across seasonal median SPM, exceedance frequency, storm-response amplitude and inter-annual variability; scoring weights agreed with client prior to processing | Ranked comparison table in PDF and spreadsheet, with per-site radar charts showing performance across criteria |
| Atmospheric correction quality assessment | Comparison of ACOLITE SWIR-based and standard Sen2Cor outputs against any available in-situ Secchi depth or turbidity meter records; flagging of scenes with suspected sun glint or high aerosol load | Correction method selection report and per-scene quality flags included in data package |
| Near-real-time SPM monitoring feed | Automated SPM retrieval on new Sentinel-2 acquisitions over defined AOI, with alert trigger when 5-day mean exceeds agreed threshold | Email or API alert with SPM map attachment; suitable for operational monitoring after site commissioning |
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.