Harmful algal bloom detection and warning for aquaculture sites
Harmful algal blooms can devastate fish farms and shellfish leases within hours of reaching lethal concentrations. Ocean-colour remote sensing maps bloom extent, pigment intensity and drift trajectory before the water turns visibly wrong.
Sensors
- Sentinel-3 OLCI: 21 spectral bands from 400 to 1020 nm at 300 m spatial resolution, global revisit roughly every two days. The dedicated fluorescence baseline band at 681 nm and the red-edge bands around 709 nm make it the primary workhorse for cyanobacterial and dinoflagellate bloom mapping in open and semi-enclosed coastal water. The 300 m floor is a genuine constraint: farm polygons smaller than roughly 9 hectares cannot be resolved as discrete targets.
- MODIS-Aqua: 36 bands, ocean-colour channels at 250 m to 1 km depending on band. Daily to near-daily revisit at most latitudes. Coarser than OLCI for inshore work but the archive extends to 2002, which is invaluable for establishing bloom seasonality and multi-year frequency at a given site.
- PACE OCI: Launched February 2024, NASA PACE carries the Ocean Colour Instrument with continuous hyperspectral coverage from 340 to 890 nm at roughly 1 km resolution and daily global revisit. The hyperspectral sampling enables phytoplankton community composition retrieval beyond what broadband sensors allow, distinguishing cyanobacteria from dinoflagellates by their absorption fine-structure. Operational data products are still maturing as of mid-2025.
- Landsat 8/9 OLI: 30 m multispectral resolution with a 16-day revisit per satellite (8-day combined). Useful for resolving bloom boundaries inside small inshore bays and individual farm lease areas that OLCI cannot separate. Limited to six ocean-relevant bands and no dedicated fluorescence channel, so it works best as a spatial-detail complement to OLCI rather than a primary bloom-detection sensor.
Why pigments betray a bloom before the fish die
Harmful algal bloom taxa carry accessory pigments that shift their spectral reflectance in ways clean water does not. Cyanobacteria contain phycocyanin, which absorbs strongly near 620 nm and produces a fluorescence emission around 650 nm. Dinoflagellates carry peridinin, which shifts absorption into the blue-green. Both taxa also contain chlorophyll-a, which fluoresces at 683 nm when illuminated by sunlight. These signatures are faint relative to the bright background of shallow sediment or foam, but modern ocean-colour algorithms can extract them from calibrated top-of-atmosphere radiance after atmospheric correction.
The practical consequence is that a dense surface bloom becomes detectable from orbit at concentrations well below the point of acute fish mortality. Bloom-forming cyanobacteria can produce detectable chlorophyll-a signals at surface concentrations of a few micrograms per litre, though detection thresholds vary with water clarity, sun angle and sensor noise. That lead time, measured in hours to a day or two depending on bloom growth rate and current speed, is what makes satellite warning operationally useful.
What a 300-metre pixel actually resolves, and what it misses
Sentinel-3 OLCI is the best freely available sensor for this task, and its 300 m resolution is a real limitation that any honest system design must acknowledge. A single OLCI pixel covers nine hectares. Many inshore shellfish leases and fish-farm licence areas are smaller than that. In practice, OLCI is most useful for detecting and tracking the bloom as a regional phenomenon, identifying whether a front or plume is moving toward a farm cluster, and triggering a higher-resolution look.
That higher-resolution look comes from Landsat 8 or 9 OLI at 30 m, or from commercial multispectral satellites available under client tasking agreements. Landsat's 16-day revisit per satellite makes it unreliable as a sole detection system for fast-moving events, but it is well-suited to confirming bloom boundaries once OLCI flags an event. Cloud is the other hard constraint: optical sensors of any resolution see nothing through overcast, which is common in the high-latitude coastal environments where salmon and shellfish farming is concentrated. There is no satellite workaround for this; the practical response is to use the most recent clear-sky observation alongside hydrodynamic drift modelling to project bloom position forward in time.
From spectral signal to farm-gate alert
Detection is only the first step. The operationally useful product is a trajectory forecast: given the bloom's current position, extent and pigment intensity, and given surface current vectors from models such as Copernicus Marine Service NEMO or HYCOM, how likely is it to reach a named farm polygon within 24 or 48 hours, and at what concentration?
The standard analytical chain runs from OLCI or MODIS Level-1 radiance through atmospheric correction (POLYMER or C2RCC are the published algorithms most validated in coastal optically complex water) to chlorophyll-a and phycocyanin index maps, then to a Lagrangian particle advection model seeded with bloom pixels above a threshold index value. The output is a probability surface over the farm network for the next one to two days. Beyond 48 hours, current-model uncertainty degrades the forecast enough that it is better communicated as a directional risk flag than a precise concentration estimate.
Sentinel-3 OLCI data are available through the Copernicus Data Space Ecosystem typically within three hours of acquisition. That latency sets a practical floor on warning time: the satellite overpass, download, processing and alert delivery chain realistically delivers a warning within four to six hours of the satellite seeing the bloom.
Species matter, and sensors have limits in telling them apart
Not all harmful blooms are equally dangerous, and the risk profile depends on the taxon. Alexandrium species produce paralytic shellfish toxins at concentrations that can close shellfish harvesting areas at cell densities invisible to any satellite sensor. Pseudo-nitzschia produces domoic acid. Karenia brevis produces brevetoxins. None of these toxins produce a spectral signature detectable from orbit. What satellites detect is the biomass accumulation, not the toxin itself.
This is a fundamental limit of the method. A satellite bloom alert is a probabilistic trigger for ground-truth sampling, not a substitute for it. The value of the satellite layer is in directing where and when to sample, reducing the cost of blanket monitoring programmes and shortening the time between a bloom forming offshore and a sampler reaching the water. PACE OCI's hyperspectral capability offers a genuine step forward in distinguishing cyanobacteria from dinoflagellates by absorption shape, but even PACE cannot identify species below the genus level, and its operational products for this application are still being validated against in-situ data.
Building a site-specific early-warning system
An effective system for a farm operator or a national aquaculture regulator combines three layers: a daily satellite-derived bloom index map for the regional water body, a hydrodynamic drift model configured for local tidal and wind forcing, and an alert threshold calibrated against historical bloom events at the specific site. The historical calibration is where archive depth matters. MODIS-Aqua's record back to 2002 and Landsat's archive back to the early 1970s allow a statistically meaningful characterisation of bloom seasonality, dominant wind directions during bloom transport events, and the lag between offshore detection and inshore arrival.
Satellize runs this type of analytics pipeline on open constellations including Sentinel-3 and MODIS, with commercial tasking added for high-resolution confirmation passes. The firm's analytical work to date has been concentrated in Pacific island contexts, including the Tonga crop-estimation programme, and the same data-pipeline architecture applies directly to coastal aquaculture monitoring in island and coastal-state settings where national monitoring capacity is limited. A useful first step for any farm operator or regulator is a retrospective bloom frequency analysis for their specific site polygon, using the available archive, before committing to a live alert service.
Typical figures
| Primary sensor spatial resolution | 300 m (Sentinel-3 OLCI); 250 m to 1 km (MODIS-Aqua); ~1 km (PACE OCI); 30 m (Landsat 8/9 OLI) |
| Revisit frequency | ~2 days (Sentinel-3 OLCI at mid-latitudes); daily (MODIS-Aqua); daily (PACE OCI); 8 days combined (Landsat 8+9) |
| Data latency from acquisition | Sentinel-3 NRT products typically within 3 hours of overpass via Copernicus Data Space |
| Key spectral bands for bloom detection | 620 nm (phycocyanin absorption), 665 nm (chlorophyll-a absorption), 681 nm (chlorophyll fluorescence), 709 nm (red-edge reflectance peak in bloom water) |
| Minimum detectable bloom concentration (chlorophyll-a) | Approximately 2 to 5 µg/L in clear coastal water under favourable sun angle; higher in turbid or shallow-sediment water |
| Cloud limitation | Optical sensors are fully blocked by cloud cover; no satellite optical workaround exists |
| Archive depth | MODIS-Aqua from 2002; Sentinel-3 OLCI from 2016; Landsat multispectral from 1972 (with band limitations for older missions) |
| Trajectory forecast horizon | Operationally useful to 24 to 48 hours; beyond 48 hours, current-model uncertainty reduces output to directional risk flag |
| Typical deliverable formats | GeoTIFF bloom index maps, GeoJSON farm-polygon alert layers, JSON alert feeds, PDF event reports |
Analytics Satellize can run
| Daily bloom index map | Chlorophyll-a and phycocyanin index retrieval from OLCI or MODIS after POLYMER or C2RCC atmospheric correction | GeoTIFF raster, updated each clear-sky overpass, covering the client's defined coastal zone |
| Farm-polygon risk alert | Lagrangian particle advection from bloom pixels above threshold, using Copernicus Marine Service surface current fields | JSON alert feed or email notification, per named farm polygon, with estimated arrival probability and time window |
| Bloom trajectory animation | Time-series composite of daily bloom index maps with overlaid drift vectors | Animated GIF or MP4 for operational briefings, covering the preceding 5 to 10 days |
| Retrospective bloom frequency analysis | Statistical analysis of MODIS-Aqua and Sentinel-3 archive for site polygon, identifying seasonal patterns and dominant transport directions | PDF report with bloom frequency calendar, percentile concentration maps, and historical event log |
| High-resolution bloom boundary confirmation | Landsat 8/9 OLI band-ratio analysis (red-edge normalised difference) triggered by OLCI bloom flag | 30 m GeoTIFF bloom extent polygon for the event day, suitable for regulatory reporting |
| PACE hyperspectral community index | Phytoplankton absorption spectral decomposition using published PACE OCI Level-2 products to partition cyanobacteria versus dinoflagellate contribution | Experimental GeoTIFF layer with taxon-group index, flagged as indicative pending further PACE product validation |
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.