Cyanobacterial bloom detection for drinking water toxin risk

Cyanobacteria produce microcystins and other toxins that overwhelm conventional water treatment. Sentinel-2 and Sentinel-3 OLCI can track bloom onset and drift toward drinking-water intakes, though cloud cover and sub-pixel patches remain real operational limits.

Sensors

Sentinel-3 OLCI: 21 spectral bands between 400 and 1020 nm, including a dedicated 620 nm band for phycocyanin absorption and 665 nm for chlorophyll-a. 300 m spatial resolution; full global coverage roughly every two days. Designed for ocean colour; works well on large reservoirs but too coarse for bodies under a few kilometres wide.
Sentinel-2 MSI: 10 m and 20 m multispectral bands. No dedicated 620 nm phycocyanin band, but the red-edge bands at 705 nm and 740 nm support chlorophyll-a indices such as the Maximum Chlorophyll Index (MCI) and 2-band ratio algorithms. Five-day revisit at the equator (2–3 days at mid-latitudes with both satellites). Useful for reservoirs too small for OLCI; the 10–20 m floor still misses sub-pixel bloom patches in early onset.
MODIS Aqua: 250 m to 1 km bands; daily to twice-daily revisit. The 645 nm and 555 nm bands support basic bloom flagging. Coarser than OLCI for spectral bloom discrimination, but the daily cadence is valuable for tracking rapid drift events and provides an archive back to 2002.
Landsat 8/9 OLI: 30 m multispectral resolution. Coastal/aerosol band at 443 nm, blue at 483 nm, green at 561 nm, red at 655 nm, and near-infrared at 865 nm support chlorophyll-a retrieval and floating algae indices. Sixteen-day single-satellite revisit is too slow for operational bloom alerting but useful for historical trend analysis and reservoir-specific algorithm calibration.

What the 620 nm band actually tells you

Phycocyanin is the pigment that gives cyanobacteria their characteristic blue-green colour and distinguishes them from green algae and diatoms. It absorbs strongly at around 620 nm, a feature that most algae lack. Sentinel-3 OLCI was designed with ocean colour science in mind, and its Band 11 at 619 nm sits almost exactly on this absorption trough. Algorithms that combine OLCI Band 11 with adjacent bands at 600 nm and 665 nm can produce a phycocyanin index that is reasonably specific to cyanobacteria rather than total phytoplankton biomass.

Chlorophyll-a is a complementary indicator. All photosynthetic algae contain it, so a high chlorophyll-a signal alone does not confirm a toxic bloom. The combination of elevated chlorophyll-a and a positive phycocyanin index is a much stronger signal. Sentinel-2 lacks the 620 nm band entirely, so analysts working at 10–20 m resolution rely on red-edge ratios and the Maximum Chlorophyll Index, which are proxies for biomass rather than cyanobacterial identity. That is a genuine limitation: a Sentinel-2 bloom flag should be treated as a high-biomass alert, not a confirmed cyanobacterial detection.

Drift toward the intake: the operational question that matters

Detecting a bloom in the middle of a reservoir is scientifically interesting. Detecting it moving toward a water intake is operationally critical. Wind-driven surface transport can move a scum layer at 1–5 cm per second, meaning a bloom two kilometres from an intake can arrive within hours of a wind shift. Sentinel-3's two-day revisit is often sufficient to track large, established blooms on major water bodies, but it will miss rapid drift events between passes.

Sentinel-2's finer resolution helps resolve bloom geometry near intake structures, where OLCI's 300 m pixel would smear the signal. In practice, the two sensors are complementary: OLCI for daily situational awareness on large reservoirs, Sentinel-2 for spatial detail when a bloom is already confirmed and its proximity to infrastructure matters. MODIS Aqua adds a daily check at coarser resolution, useful as a gap-filler between Sentinel-3 passes.

Operational alerting systems typically define a risk polygon around an intake structure and trigger an alert when bloom pixels enter that polygon above a chlorophyll-a or phycocyanin threshold. The threshold needs local calibration: turbidity, dissolved organic matter and water colour vary significantly between reservoirs and confound generic algorithms.

What passive optics cannot see

Cloud cover is the most obvious constraint. Cyanobacterial blooms peak in warm, calm, stratified conditions, which are often associated with high-pressure systems and clear skies. That is fortunate. But a single overcast day during a rapid bloom onset can mean a missed alert.

The deeper problem is vertical structure. Cyanobacteria regulate their buoyancy using gas vesicles and can form dense subsurface layers at depths of 1–5 metres under calm, stratified conditions. Passive optical sensors detect surface reflectance only. A subsurface scum layer can be toxin-rich and completely invisible to Sentinel-2 or OLCI. This is not a minor caveat: some of the most dangerous bloom events occur when surface expression is absent or patchy and the toxin is concentrated below the optical penetration depth, which in productive inland waters may be less than half a metre. Field sampling at the intake remains irreplaceable for toxin confirmation.

Sub-pixel patchiness is the third limit. Early-stage blooms often form as narrow wind-rows or streaks well below the 10 m resolution floor of Sentinel-2, let alone the 300 m pixel of OLCI. The first satellite-detectable signal may lag the actual bloom onset by one to several days.

Calibrating algorithms to a specific reservoir

Published bloom detection algorithms, including those from the European Space Agency's Inland Waters thematic group and USGS Landsat water-quality work, are designed for general application. They perform poorly when applied without adjustment to reservoirs with high suspended sediment, coloured dissolved organic matter, or unusual optical depth profiles. A eutrophic, tea-coloured reservoir in a peat catchment will confound a generic chlorophyll-a retrieval tuned for clear alpine water.

Calibration requires concurrent field measurements: in-situ chlorophyll-a, phycocyanin fluorometry, Secchi depth and ideally hyperspectral above-water reflectance to match satellite overpass timing. Even a small campaign of ten to twenty matched field-satellite observations can substantially improve retrieval accuracy for a specific site. Without it, bloom flags carry large uncertainty bounds that erode the confidence of water utility operators.

Landsat's 30 m archive back to 1984 (and Landsat 8 OLI from 2013 at improved radiometric quality) is useful for establishing baseline seasonal patterns and identifying which months carry the highest historical bloom risk, before any real-time monitoring system is deployed.

Turning detections into a water-utility decision

A satellite bloom detection is a probabilistic flag, not a toxin measurement. Microcystin concentrations, the most common hepatotoxic concern, can vary by two orders of magnitude within a single bloom depending on the dominant Microcystis or Planktothrix strain, nutrient conditions and cell lysis state. The World Health Organisation's drinking water guideline of 1 µg/L microcystin-LR is a concentration threshold that no satellite can measure directly.

The practical output of a satellite monitoring programme is a tiered risk signal: bloom absent, bloom present at low biomass, bloom present at high biomass within a defined distance of the intake. Each tier maps to a pre-agreed operational response, from increased sampling frequency to intake shutdown. Satellize can configure this kind of alert pipeline on open Sentinel and MODIS data streams, with reservoir-specific thresholds and delivery via GIS feed or daily report. The Tonga crop-estimation programme demonstrated that operationally useful analytics can be built on open satellite data for clients without existing earth-observation infrastructure.

The honest summary for a water utility: satellite monitoring substantially reduces the chance of a bloom reaching an intake undetected on a clear day. It does not replace field sampling, it does not detect subsurface toxin layers, and it cannot confirm microcystin concentration. Used correctly, it compresses the reaction window from days to hours.

Typical figures

Best spatial resolution (bloom mapping)	10 m (Sentinel-2 MSI, visible bands); 20 m (Sentinel-2 red-edge); 300 m (Sentinel-3 OLCI)
Revisit frequency	~2 days (Sentinel-3 OLCI); 2–5 days (Sentinel-2 dual satellites); daily (MODIS Aqua); 16 days (Landsat 8/9 single satellite)
Key spectral bands	619 nm phycocyanin (OLCI Band 11); 665 nm chlorophyll-a (OLCI/MSI); 705 nm, 740 nm red-edge (Sentinel-2 MSI); 645 nm, 555 nm (MODIS)
Minimum detectable bloom area	~1 ha practical floor for Sentinel-2 at 10 m; ~10–20 ha for OLCI at 300 m; sub-pixel patches are undetectable
Latency (open data)	Sentinel-2 Level-2A typically available within 3–5 hours of acquisition; Sentinel-3 OLCI NRT within 3 hours via EUMETSAT
Optical penetration depth (productive inland water)	Often <0.5 m in eutrophic conditions; subsurface scum layers below this depth are not detectable
Cloud cover constraint	Total blockage under cloud; no SAR workaround for optical bloom detection
Archive depth	Sentinel-2 from 2015; Sentinel-3 OLCI from 2016; MODIS Aqua from 2002; Landsat OLI from 2013 (Landsat TM/ETM+ from 1984 at lower radiometric quality)
Delivery formats	GeoTIFF bloom-extent rasters, vector polygon shapefiles, time-series CSV, GIS-compatible alert feed (GeoJSON)

Analytics Satellize can run

Daily bloom presence/absence flag	Phycocyanin index (OLCI Band 11/Band 9 ratio) and Maximum Chlorophyll Index on Sentinel-2 red-edge bands; cloud-masked composite from available passes	Daily GeoJSON alert layer with bloom confidence class (absent / low / high biomass)
Bloom extent polygon and area estimate	Threshold-based classification of chlorophyll-a retrieval using two-band red-edge ratio; morphological filtering to remove noise pixels	GeoTIFF and shapefile of bloom extent with area in hectares, timestamped per acquisition
Intake proximity risk score	Spatial intersection of bloom polygon with user-defined intake buffer zones (configurable radii); risk tier assigned by biomass class and distance	Automated alert report (PDF or API push) triggered when bloom pixels enter intake buffer
Bloom drift trajectory estimate	Multi-date centroid tracking combined with ERA5 reanalysis wind vectors to project probable surface transport direction over 24–48 hours	Annotated map showing observed drift vector and projected bloom position, issued with each detection event
Seasonal bloom climatology	Time-series analysis of Landsat OLI and MODIS chlorophyll-a retrievals over the historical archive; percentile-based anomaly detection	Annual risk calendar showing historically high-probability bloom months and inter-annual trend chart
Reservoir-specific algorithm calibration report	Regression of satellite-derived reflectance against client-supplied field measurements of chlorophyll-a and phycocyanin; site-specific retrieval coefficients derived	Calibration report with retrieval accuracy statistics (RMSE, bias) and updated algorithm coefficients for operational deployment

Sources

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.