Harmful algal bloom and cyanobacteria detection in inland and coastal waters
Satellite ocean-colour sensors can detect toxic algal blooms days before conventional monitoring networks raise an alarm, but sensor choice, water type and dissolved organics all determine whether the signal is real or an artefact.
Sensors
- Sentinel-3 OLCI: 300 m spatial resolution, 2-day revisit at mid-latitudes (1-2 days globally with both Sentinel-3A and 3B combined). 21 spectral bands purpose-designed for ocean colour, including bands at 620 nm (phycocyanin absorption), 665 nm and 681 nm (chlorophyll-a fluorescence peak), and 709 nm (red-edge). The instrument of choice for large lakes, estuaries and coastal waters; too coarse for reservoirs smaller than roughly 1 km².
- Sentinel-2 MSI: 10 m (visible) and 20 m (red-edge and near-infrared) spatial resolution, 5-day revisit per satellite (2-3 days with both Sentinel-2A and 2B). Four red-edge and NIR bands at 705 nm, 740 nm, 783 nm and 865 nm allow chlorophyll-a index retrieval in small reservoirs and drinking-water catchments. Not designed for ocean colour; atmospheric correction over water is more demanding and band placement misses the 620 nm phycocyanin feature directly.
- MODIS Aqua: 250 m to 1 km spatial resolution depending on band; daily global revisit. NASA's standard chlorophyll-a products (OC3M algorithm) are well-validated for Case-1 open ocean waters but degrade in optically complex inland and coastal waters. Useful for tracking large bloom events and multi-decadal trend analysis given the archive running from 2002.
- Landsat 8/9 OLI: 30 m spatial resolution, 8-day revisit per satellite (16-day per satellite, but the two together give roughly 8 days). Coastal/aerosol band at 443 nm and green/red/NIR bands support empirical chlorophyll retrieval in small water bodies. No dedicated fluorescence band; algorithms rely on band-ratio proxies. Archive depth and free access make it valuable for historical baseline work.
What the sensor is actually measuring
Algal blooms do not emit a signal that satellites read directly. What the sensor records is reflected and emitted light at specific wavelengths that phytoplankton pigments modify in predictable ways. Chlorophyll-a absorbs strongly in the blue (around 443 nm) and red (around 665 nm), and re-emits a fluorescence peak near 681 nm. That fluorescence line height, measured against a baseline interpolated from adjacent bands, is the most physically direct satellite indicator of phytoplankton biomass in open water.
Cyanobacteria add a second signature. Phycocyanin, a pigment found in cyanobacteria but not in most eukaryotic algae, absorbs near 620 nm. A sensor with a band placed there can distinguish a cyanobacterial bloom from a diatom or green-algae bloom of similar chlorophyll concentration. Sentinel-3 OLCI has exactly this band (band 11, centred at 620 nm). Sentinel-2 MSI does not, which is an honest design limitation rather than a deficiency: MSI was built for land. The distinction matters because cyanobacteria include species that produce hepatotoxins and neurotoxins, so a water manager's response to a cyanobacterial bloom is categorically different from the response to a non-toxic diatom event.
Why sensor choice depends on the water body, not preference
Sentinel-3 OLCI's 300 m pixel is adequate for the Baltic Sea, Lake Victoria or a large coastal lagoon. For a 200-hectare drinking-water reservoir, a single OLCI pixel covers the entire surface and mixes bloom signal with shoreline reflectance, making the retrieval unreliable. Sentinel-2 at 20 m (red-edge bands) resolves spatial heterogeneity within the reservoir, identifies where the bloom is densest, and can detect surface scums that OLCI would average away.
The practical rule is roughly this: water bodies larger than about 1 km² in their narrowest dimension are workable with OLCI; smaller bodies need Sentinel-2 or Landsat. Revisit is the trade-off. A bloom can intensify or disperse within 48 hours, and Sentinel-2's 2-3 day combined revisit is adequate for monitoring but not for real-time alert systems. Sentinel-3's near-daily return is better matched to operational early warning, provided the water body is large enough.
Cloud cover is an unresolved constraint for all optical sensors. In humid temperate or tropical climates, a bloom can develop, peak and partially dissipate between consecutive clear-sky observations. No amount of algorithm sophistication recovers a signal through thick cloud. Programmes that require continuous monitoring should pair optical bloom detection with in-situ buoy sensors or model forecasts rather than relying on satellite data alone.
The CDOM problem in humic lakes
In lakes with significant inputs of peat-derived or soil-derived organic matter, coloured dissolved organic matter (CDOM) absorbs strongly in the blue and competes with chlorophyll absorption in exactly the wavelength range that standard ocean-colour algorithms use. The OC3M algorithm applied to a brown, humic Finnish lake will return a chlorophyll-a estimate that is substantially too high, because it cannot separate CDOM absorption from phytoplankton absorption without additional spectral information.
Red-edge algorithms partially sidestep this problem. The ratio of reflectance near 709 nm to reflectance near 665 nm (the two-band ratio used in several published inland-water studies) is less sensitive to CDOM because CDOM absorption is much weaker at those wavelengths. Sentinel-2's bands at 665 nm and 705 nm are positioned for exactly this retrieval. The result is more reliable in humic waters, though it still requires site-specific calibration against field measurements to convert band ratios to absolute chlorophyll-a concentrations in mg/m³. Any programme that skips that calibration step is producing relative indices, not concentrations, and should label outputs accordingly.
Detection limits and what they mean operationally
Published studies using OLCI suggest reliable chlorophyll-a detection at concentrations above roughly 5 to 10 mg/m³ in clear to moderately turbid waters, with uncertainty increasing as turbidity rises. The World Health Organisation's first alert level for cyanobacterial risk in recreational water is set at 10 µg/L (equivalent to 10 mg/m³) of chlorophyll-a, so the sensor is in the right order of magnitude for early-warning purposes, but it is not a precision analytical instrument.
Surface scum detection is a different matter. Dense scums can be identified visually in Sentinel-2 true-colour imagery at 10 m resolution and confirmed by near-infrared reflectance anomalies, because floating cyanobacteria cells scatter NIR light strongly. However, a scum visible in imagery may have formed hours before the overpass and drifted from the bloom source. Attributing a scum to a specific upstream input requires hydrodynamic modelling, not just image analysis.
Atmospheric correction is the single largest source of error in water-leaving reflectance retrievals. Over water, the water signal is a small fraction of the total at-sensor radiance; aerosol scattering dominates. Errors in aerosol characterisation translate directly into errors in retrieved chlorophyll-a. OLCI's dedicated atmospheric correction bands and the POLYMER and C2RCC processors have improved this substantially for coastal and inland waters, but the correction is never perfect and should be validated against matchup data for any operational deployment.
Building an operational early-warning system
A workable bloom early-warning pipeline has three components: near-daily imagery processing, a threshold-based alert layer, and a human review step before any public notification is issued. Automating the first two is straightforward with Sentinel-3 OLCI data from the Copernicus Data Space. The third step matters because false positives from atmospheric correction artefacts, sun glint or turbid river plumes are common enough that an unreviewed automated alert will erode trust with water managers quickly.
Satellize runs this kind of multi-sensor optical analytics workflow on open constellations, the same approach used in the Tonga crop-estimation programme adapted to a water-quality context. The output is a georeferenced alert layer, updated on each clear-sky overpass, with a confidence flag that distinguishes spectrally confirmed cyanobacterial signal from generic high-chlorophyll detections.
Historical archive analysis is often more immediately useful than real-time monitoring for a new client. A five-year Sentinel-2 and Sentinel-3 time series over a reservoir system can establish bloom seasonality, identify which sub-basins are most frequently affected, and correlate bloom intensity with upstream land-use or temperature records. That baseline informs where to invest in in-situ sensors and when to increase sampling frequency, which is a more defensible use of a water authority's budget than deploying real-time satellite alerts without prior characterisation.
Typical figures
| Spatial resolution (bloom mapping) | 300 m (Sentinel-3 OLCI); 20 m red-edge bands (Sentinel-2 MSI); 30 m (Landsat 8/9 OLI); 250 m–1 km (MODIS Aqua) |
| Revisit frequency | ~1 day (Sentinel-3A+3B combined at mid-latitudes); 2–3 days (Sentinel-2A+2B combined); ~8 days (Landsat 8+9 combined); daily (MODIS Aqua) |
| Key spectral bands for cyanobacteria | 620 nm (phycocyanin, OLCI band 11); 665 nm (chlorophyll-a absorption); 681 nm (fluorescence peak, OLCI); 705–709 nm (red-edge chlorophyll index, OLCI and MSI) |
| Minimum detectable chlorophyll-a | Approximately 5–10 mg/m³ in clear to moderately turbid water (OLCI, published range); higher in turbid or humic conditions |
| Minimum water body size (reliable retrieval) | ~1 km² narrowest dimension for OLCI; ~1–4 ha for Sentinel-2 red-edge bands |
| Cloud limitation | All optical sensors: no retrieval through cloud or dense aerosol; effective clear-sky revisit in humid climates may be 1–3 weeks |
| Archive depth | Sentinel-3 OLCI from 2016; Sentinel-2 MSI from 2015; Landsat from 1972; MODIS Aqua from 2002 |
| Typical data latency (NRT products) | Sentinel-3 NRT: ~3 hours after acquisition; Sentinel-2 L2A: typically 1–3 hours via Copernicus Data Space |
| Delivery formats | GeoTIFF chlorophyll-a concentration maps; georeferenced alert polygons (GeoJSON/Shapefile); time-series CSV; WMS/WMTS tile feeds |
Analytics Satellize can run
| Chlorophyll-a concentration map | Red-edge two-band ratio (665/709 nm) or fluorescence line height retrieval from OLCI, with C2RCC or POLYMER atmospheric correction | Georeferenced GeoTIFF per overpass, with concentration in mg/m³ and per-pixel uncertainty flag |
| Cyanobacteria likelihood index | Spectral decomposition using 620 nm phycocyanin band (OLCI) relative to adjacent chlorophyll bands; outputs a dimensionless index separating cyanobacterial from non-cyanobacterial bloom signal | Raster layer with three-class output (low/medium/high cyanobacterial likelihood), updated per clear-sky overpass |
| Bloom extent polygon and area estimate | Threshold segmentation on chlorophyll-a or NIR reflectance anomaly; bloom boundary vectorised and area computed | GeoJSON polygon with area in km², bloom centroid coordinates and peak concentration value |
| Surface scum detection (Sentinel-2) | NIR reflectance anomaly and floating algae index applied to 10 m MSI bands; distinguishes dense surface scum from subsurface bloom | Binary scum mask at 10 m, with confidence score based on spectral consistency across NIR and red bands |
| Seasonal bloom climatology | Time-series analysis of Sentinel-2 and Sentinel-3 archive (2015/2016 to present); percentile compositing to establish baseline and anomaly detection | Annual report with bloom frequency maps, peak-season timing, and year-on-year trend summary per water body |
| Early-warning alert feed | Automated threshold trigger on chlorophyll-a or cyanobacteria index exceeding configurable WHO or national guideline levels, with human review gate before dispatch | Email or API alert with bloom location, estimated concentration, sensor source, and confidence flag; issued within 6 hours of clear-sky overpass |
| CDOM-corrected chlorophyll retrieval (humic lakes) | Site-calibrated red-edge algorithm using field matchup data to separate CDOM and phytoplankton absorption; applied to Sentinel-2 MSI | Calibrated concentration map with documented uncertainty bounds and matchup validation statistics |
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.