Water quality monitoring in inland aquaculture pond systems
Satellite-derived turbidity, chlorophyll-a and CDOM retrievals can track bloom development and effluent events across intensive fish-pond networks, but only if the sensor resolves individual ponds. Most ocean-colour instruments cannot.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible bands (Band 2–4) and 20 m in red-edge and SWIR; 5-day revisit at the equator with both satellites. The red-edge bands (Band 5, 704 nm; Band 6, 740 nm) are particularly useful for chlorophyll-a retrieval in turbid productive waters. Freely available via Copernicus Dataspace.
- Planet SuperDove: 3 m resolution, daily revisit over most latitudes. Eight spectral bands including a coastal blue and red-edge, enabling basic water-quality index computation. Commercial tasking; archive depth varies by site. Useful for resolving ponds below 0.5 ha that Sentinel-2 pixels straddle.
- Landsat 8/9 OLI: 30 m resolution, 16-day revisit per satellite (8 days combined). The coastal/aerosol band (Band 1, 443 nm) aids CDOM and turbidity separation. Freely available with a long archive back to 1984 for Landsat 5/7 predecessors, useful for establishing baseline pond-water conditions.
- RapidEye archive: 5 m resolution, five bands including a dedicated red-edge at 690 nm. No longer collecting new imagery (constellation decommissioned), but the archive through Planet covers 2009–2020 and remains valuable for retrospective bloom event reconstruction over known pond networks.
Why ocean-colour satellites are the wrong tool here
MODIS, MERIS and their successors were designed for open-ocean and large-lake retrievals. Their pixel footprints run from 250 m to 1 km. A one-hectare pond is roughly 100 m × 100 m, which means it occupies a fraction of a single MODIS pixel. Mixed-pixel contamination from bund soil, surrounding vegetation and adjacent ponds makes any retrieval from coarse sensors essentially meaningless at the individual-pond level.
Sentinel-2 MSI at 10 m is the practical minimum for ponds above roughly 0.25 ha. Below that threshold, Planet SuperDove at 3 m becomes necessary. Neither is a perfect ocean-colour instrument: both were designed primarily for land, which means their atmospheric correction algorithms were not built around the adjacency effects and low water-leaving radiance that dominate small inland water bodies. That distinction matters and is discussed below.
What the spectral signal actually tells you
Three optically active constituents drive water colour in intensive fish ponds. Suspended particulate matter, or turbidity, scatters light strongly in the red and near-infrared, raising reflectance in Sentinel-2 Bands 4 and 8A. Chlorophyll-a absorbs at 443 nm and 675 nm and fluoresces around 700 nm, which is why Sentinel-2's Band 5 (704 nm) is so useful in productive pond water. Coloured dissolved organic matter absorbs strongly in the blue, suppressing Band 2 and Band 3 reflectance in ways that can mimic high turbidity if not separated correctly.
Bloom development follows a recognisable spectral trajectory. Early phytoplankton growth increases the Band 5 / Band 4 ratio (a proxy for the red-edge inflection). Dense blooms, common in intensively fed shrimp and tilapia ponds, push the near-infrared reflectance upward as cell concentrations exceed roughly 50 µg/L chlorophyll-a. Pond turnover events, where stratified bottom water suddenly mixes to the surface, produce a rapid turbidity spike visible as a broadband increase in red reflectance, often within a single overpass. These events can precede mass mortality if dissolved oxygen crashes.
Effluent discharge plumes are detectable when they enter connecting canals or receiving water bodies, typically as a turbidity or chlorophyll anomaly relative to the background. The spatial extent of the plume and its persistence across successive overpasses gives regulators a rough residence-time estimate.
Atmospheric correction: the honest problem
Over open ocean, the standard assumption is that near-infrared water-leaving radiance is negligible, allowing the atmosphere to be characterised from NIR signal alone. Over productive inland ponds, that assumption fails completely: algal-rich water reflects substantially in the NIR. Standard land-surface atmospheric correction products such as Sentinel-2's Sen2Cor also perform poorly over water because they are tuned for vegetated and bare-soil surfaces.
Purpose-built processors exist. ACOLITE, developed by the Royal Belgian Institute of Natural Sciences and documented in peer-reviewed literature, applies a dark-spectrum or exponential extrapolation approach that handles turbid and productive inland waters better than Sen2Cor. iCOR is another option. Neither eliminates error entirely: published validation studies over small water bodies report root-mean-square errors in retrieved turbidity of 20–40% under clear-sky conditions, rising sharply under haze or thin cloud. Adjacency effects, where bright bund soil scatters light into the water pixel, add a further bias that is difficult to correct without site-specific calibration.
Cloud cover is an operational constraint that cannot be engineered away. Sentinel-2 delivers a usable overpass perhaps two to three times per month over cloud-prone tropical aquaculture regions. Planet's higher revisit improves this, but Planet SuperDove's atmospheric correction over water is less mature than Sentinel-2's. Buyers should expect data gaps during monsoon seasons.
Turning retrievals into management signals
A single chlorophyll-a map is interesting. A time series of chlorophyll-a across 200 ponds on a single farm, flagging which ponds crossed a threshold of 100 µg/L in the past fortnight, is actionable. The operational value of satellite water-quality monitoring in aquaculture comes from systematic comparison against pond-specific baselines, not from absolute retrieval accuracy alone.
Bloom alerts can be structured as simple threshold exceedances on the red-edge ratio, calibrated against in-pond sensor data where available. Pond turnover detection works well as a change-detection product: a sudden increase in turbidity index relative to the preceding clear-sky overpass, flagged with a confidence score based on the magnitude of change and the quality of the atmospheric correction. Effluent plume mapping requires a slightly different approach, tracing connected-water anomalies downstream from known discharge points.
Satellize structures these retrievals as recurring analytic feeds rather than one-off maps, drawing on the same open-constellation pipeline used in the Tonga crop-estimation programme and adding commercial tasking where revisit frequency is insufficient.
Resolution, revisit and the limits of what satellites can see
Pond water quality can change within hours during a bloom crash or a storm-driven turnover. No current freely available satellite revisits a given site more than once per day, and cloud typically reduces usable frequency further. Satellite monitoring is therefore a surveillance tool, not a real-time sensor. It catches events that persist for at least one to two days and that produce a spectral signal large enough to survive atmospheric correction uncertainty.
Sub-hectare ponds push the edge of what Sentinel-2 can reliably resolve. A 10 m pixel covering a 50 m × 50 m pond will include bund edges in at least some pixels, contaminating the retrieval. Planet SuperDove at 3 m resolves these ponds cleanly but at commercial cost and with less-mature water-quality correction. The practical answer for most large pond networks is a tiered approach: Sentinel-2 for routine surveillance of ponds above 0.5 ha, with Planet tasked selectively when an anomaly needs confirmation or when ponds are too small for Sentinel-2 to resolve.
Typical figures
| Spatial resolution (routine) | 10 m (Sentinel-2 visible bands); 30 m (Landsat 8/9 OLI) |
| Spatial resolution (high-frequency tasking) | 3 m (Planet SuperDove); 5 m (RapidEye archive) |
| Revisit frequency | 5 days at equator (Sentinel-2 twin satellites); 16 days per satellite, 8 days combined (Landsat 8+9); near-daily (Planet SuperDove, commercial) |
| Minimum resolvable pond size | Approx. 0.25 ha at 10 m resolution (Sentinel-2); approx. 0.01 ha at 3 m (Planet SuperDove) |
| Key spectral bands for water quality | Blue 443–490 nm (CDOM, turbidity); Green 560 nm (turbidity); Red 665 nm (chlorophyll absorption); Red-edge 704–740 nm (chlorophyll fluorescence peak); NIR 833 nm (turbidity, bloom density) |
| Chlorophyll-a retrieval uncertainty (clear sky, inland water) | Typically 20–40% RMSE after purpose-built atmospheric correction (e.g. ACOLITE); higher under haze or adjacency contamination |
| Cloud impact on usable revisit (tropical regions) | Effective clear-sky overpasses typically 2–4 per month during wet season; up to 15 per month in dry season |
| Archive depth | Sentinel-2: 2015–present; Landsat: 1984–present (OLI from 2013); RapidEye: 2009–2020 (archive only) |
| Delivery formats | GeoTIFF raster (per-overpass retrieval), GeoJSON polygon alerts, tabular time-series CSV per pond, optional WMS feed |
| Latency (open-constellation processing) | Sentinel-2 L1C available within 3 hours of acquisition; processed water-quality retrievals typically within 24 hours of clear-sky overpass |
Analytics Satellize can run
| Chlorophyll-a concentration map (per overpass) | Red-edge band-ratio algorithm (e.g. two-band or three-band model using Sentinel-2 Bands 4, 5, 6) applied after ACOLITE atmospheric correction | GeoTIFF raster layer per clear-sky overpass, clipped to pond network extent |
| Turbidity index time series per pond | Normalised difference turbidity index or single-band red/NIR ratio, aggregated per pond polygon from pond-boundary mask | Tabular CSV with pond ID, date, mean turbidity index, quality flag; updated each usable overpass |
| Bloom development alert | Threshold exceedance on red-edge chlorophyll proxy relative to pond-specific 90-day rolling baseline; confidence scored by atmospheric correction quality flag | GeoJSON alert layer with pond ID, exceedance magnitude, date, confidence tier; optional email or API push |
| Pond turnover event detection | Change detection comparing turbidity index between consecutive clear-sky overpasses; flags ponds with step-change above two standard deviations of historical variability | Incident report with affected pond IDs, estimated event date window, and before/after false-colour imagery |
| Effluent discharge plume mapping | Connected-water anomaly tracing downstream from registered discharge points using turbidity and CDOM proxies; plume extent delineated by region-growing on anomaly threshold | Polygon GIS layer showing plume extent and intensity, timestamped per overpass, suitable for regulatory reporting |
| Seasonal water-quality baseline report | Statistical aggregation of all clear-sky retrievals over a defined season or year; percentile distribution per pond, trend analysis, identification of chronically poor-quality ponds | PDF or interactive HTML report with per-pond summary statistics and ranked anomaly list |
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.