Harmful algal bloom genus discrimination using hyperspectral remote sensing
Broadband ocean-colour sensors flag bloom presence but cannot resolve genus-level pigment signatures. Hyperspectral retrievals, led by NASA PACE OCI since 2024, open the first operational path to distinguishing toxic genera such as Karenia, Alexandrium and Pseudo-nitzschia from space.
Sensors
- NASA PACE OCI: Ocean Colour Instrument launched February 2024; continuous spectral coverage from 340 to 890 nm at approximately 5 nm spectral resolution, 1 km spatial resolution, daily global revisit. The first publicly available sensor with the spectral density needed to resolve phycoerythrin absorption near 495–565 nm and phycocyanin near 620 nm at ocean-colour signal levels.
- ASI PRISMA: Italian Space Agency hyperspectral imager; 239 bands across 400–2500 nm at 12 nm sampling, 30 m spatial resolution, swath of 30 km. Revisit is irregular and tasked, not systematic. Spatial detail is valuable for nearshore blooms but atmospheric correction over water remains more demanding than for land, and signal-to-noise at ocean-colour levels is a documented limitation.
- DESIS on ISS: DLR Earth Sensing Imaging Spectrometer; 235 bands from 400 to 1000 nm at approximately 2.55 nm sampling, 30 m resolution, 30 km swath. ISS orbital inclination limits coverage to roughly 51.6° latitude. Tasked acquisition only; no systematic revisit. Useful for validation and nearshore case studies rather than operational monitoring.
- Sentinel-3 OLCI: 21 discrete bands from 400 to 1020 nm, 300 m resolution, 1–2 day revisit. Not hyperspectral, but band placement was designed for ocean colour and includes bands at 620 nm and 665 nm that provide partial phycocyanin sensitivity. Sufficient for cyanobacteria dominance flags in freshwater and some coastal settings, but cannot resolve the finer pigment features that separate marine toxic genera.
Why a red tide is not just a red tide
A bloom of Karenia brevis off Florida and a bloom of Pseudo-nitzschia off California can look nearly identical in a true-colour image and produce similar chlorophyll-a anomalies in broadband sensors. Their consequences are entirely different. Karenia produces brevetoxins that kill fish and hospitalise people who inhale sea spray. Pseudo-nitzschia produces domoic acid, which accumulates in shellfish and has caused fatal amnesic shellfish poisoning. Alexandrium species produce saxitoxins responsible for paralytic shellfish poisoning. A binary bloom alert tells a shellfish farmer that something is happening. It does not tell them whether to close harvest beds.
The physical basis for discrimination lies in accessory pigments. Each phytoplankton functional type carries a characteristic pigment package alongside chlorophyll-a. Peridinin is a marker for dinoflagellates including Karenia and Alexandrium. Fucoxanthin dominates in diatoms such as Pseudo-nitzschia. Phycocyanin is a cyanobacteria marker. These pigments absorb at wavelengths that are close together, with features typically 10–30 nm wide. Resolving them requires contiguous spectral sampling at 5–15 nm intervals across the visible range. No broadband sensor, including the standard Sentinel-3 OLCI band set, has the spectral density to do this reliably in open-ocean or coastal water.
What PACE OCI actually changes
PACE OCI, launched in February 2024, is the first publicly available sensor designed from the outset for phytoplankton functional type (PFT) retrieval at global scale. Its continuous spectral coverage from 340 to 890 nm at approximately 5 nm resolution means that absorption features previously resolvable only by in-water spectrophotometry or airborne campaigns can, in principle, be retrieved from orbit daily. NASA's published pre-launch science objectives explicitly include PFT and inherent optical property retrieval using algorithms such as GIOP (Generalised Inherent Optical Property) and matrix factorisation approaches that decompose the water-leaving radiance spectrum into component pigment contributions.
The honest position is that PACE OCI is new. Validated, operationally mature PFT products at genus level are not yet standard catalogue items. The sensor data are publicly available through NASA Earthdata, and the science community is actively developing and validating retrieval algorithms. Early published results are promising for broad functional type separation (diatoms versus dinoflagellates versus cyanobacteria), but sub-genus discrimination and detection of low-biomass toxic patches remain active research problems. Spatial resolution of 1 km also means that thin nearshore bloom filaments narrower than a few kilometres will be missed or averaged out.
The atmospheric correction problem over water
Ocean-colour remote sensing is harder than land surface reflectance retrieval by roughly an order of magnitude in signal terms. The water-leaving radiance that carries pigment information is typically less than 10% of the total signal at the sensor; the rest is atmospheric path radiance and sun glint. Errors in atmospheric correction propagate directly into retrieved pigment concentrations. For hyperspectral sensors not optimised for ocean colour, such as PRISMA and DESIS, this is a significant operational constraint. Published studies using PRISMA over coastal waters report that standard atmospheric correction schemes developed for land are inadequate and that dedicated ocean-colour corrections are required, with residual uncertainty that can exceed the pigment signal itself in oligotrophic or low-biomass conditions.
PACE OCI was designed with this in mind, including a dedicated polarimetry package (SPEXone and HARP2) to constrain aerosol properties and improve atmospheric correction. This is one reason PACE represents a qualitative step rather than an incremental one. Even so, sun glint, adjacency effects near coastlines, and bottom reflectance in shallow water all introduce retrieval uncertainty that must be flagged in any operational product.
Practical detection limits and what they mean for aquaculture managers
Published detection thresholds for HAB-relevant pigments from hyperspectral ocean-colour retrievals vary considerably by water type, bloom density and atmospheric conditions. For phycocyanin in cyanobacteria-dominated freshwater blooms, retrievals from DESIS and PRISMA have been demonstrated at concentrations above roughly 10–20 µg/L in relatively clear water. Marine toxic dinoflagellate blooms present a harder problem: cell concentrations that cause shellfish toxicity can be below the biomass threshold at which spectral pigment features become detectable above noise. Karenia brevis blooms typically become detectable in ocean-colour imagery at cell densities above approximately 100,000 cells/L, but shellfish harvest closures in some jurisdictions are triggered at 5,000 cells/L. Satellite detection will therefore always lag behind the earliest risk threshold.
This is not a reason to dismiss the technology. It is a reason to position it correctly. Hyperspectral PFT retrieval is most valuable for tracking the spatial extent and movement of established blooms, for distinguishing toxic-genus-dominated patches from non-toxic biomass anomalies at regional scale, and for providing environmental context that guides where to deploy in-water sampling effort. It does not replace water sampling; it makes sampling programmes more efficient and spatially informed.
Integrating hyperspectral retrieval into a monitoring programme
An operational HAB genus-discrimination service has several components. The foundation is a near-real-time PACE OCI data feed, processed with atmospheric correction and a PFT retrieval algorithm to produce daily maps of dominant phytoplankton functional type and accessory pigment index values across the area of interest. These maps are overlaid with chlorophyll-a anomaly fields and sea-surface temperature to provide bloom development context. Where PRISMA or DESIS tasking is available, higher-resolution acquisitions over priority sites (fish farms, shellfish beds, recreational beaches) can be scheduled to resolve finer spatial structure.
Satellize integrates open-constellation data including Sentinel-3 OLCI and, as PACE products mature, OCI retrievals, with commercial hyperspectral tasking where clients hold the appropriate licences. The analytics workflow for genus discrimination draws on published matrix factorisation and spectral unmixing methods applied to water-leaving reflectance spectra. Outputs are delivered as GIS layers with per-pixel dominant PFT classification, pigment index rasters, and alert flags where spectral signatures are consistent with known toxic genera. Confidence intervals are included; the system does not report a genus identification without also reporting the retrieval uncertainty. In-water validation data, where clients can provide it, are used to calibrate and ground-truth the spectral retrievals for the specific water optical properties of each site.
Revisit frequency is the remaining operational constraint. Daily global coverage from PACE at 1 km is the best currently available from a public sensor. Cloud cover over productive coastal seas can reduce effective clear-sky observations to fewer than ten per month in some regions. Combining PACE with Sentinel-3 OLCI (which provides complementary coverage and a longer archive back to 2016) improves temporal sampling but at the cost of spectral resolution. No current constellation provides both daily revisit and hyperspectral resolution over the same area simultaneously.
Typical figures
| Best available spatial resolution (ocean colour) | 1 km (PACE OCI); 300 m (Sentinel-3 OLCI); 30 m (PRISMA, DESIS, tasked) |
| Spectral resolution for PFT discrimination | ~5 nm continuous (PACE OCI); ~12 nm (PRISMA); ~2.55 nm (DESIS); discrete bands only (Sentinel-3 OLCI) |
| Spectral range relevant to pigment retrieval | 400–750 nm core window; PACE OCI covers 340–890 nm continuously |
| Revisit (systematic) | Daily global (PACE OCI); 1–2 days (Sentinel-3 OLCI); tasked/irregular (PRISMA, DESIS) |
| Effective clear-sky revisit (mid-latitude coastal) | Typically 8–15 usable observations per month depending on cloud regime |
| Minimum detectable bloom biomass (approximate) | Karenia brevis detectable in ocean colour above ~100,000 cells/L; shellfish closure thresholds can be as low as 5,000 cells/L — a gap that satellite cannot close alone |
| Latency from acquisition to processed product | PACE OCI Level-2 products available via NASA Earthdata typically within hours of acquisition; near-real-time processing pipelines under active development |
| Archive depth | PACE OCI: from February 2024; Sentinel-3 OLCI: from April 2016; PRISMA: from 2019 |
| Primary retrieval uncertainty | Atmospheric correction error; sun glint; adjacency effects nearshore; bottom reflectance in water shallower than ~15 m |
| Delivery formats | NetCDF (standard ocean-colour); GeoTIFF per-band or classified; GIS vector alert polygons; JSON alert feeds |
Analytics Satellize can run
| Dominant phytoplankton functional type classification map | Spectral unmixing and matrix factorisation of water-leaving reflectance spectra (e.g. GIOP framework, published NASA PACE science algorithms) | Daily GeoTIFF raster with per-pixel PFT class label and posterior confidence score, covering client-defined coastal area of interest |
| Accessory pigment index rasters | Band-ratio and spectral deconvolution retrievals targeting peridinin (dinoflagellates), fucoxanthin (diatoms) and phycocyanin (cyanobacteria) absorption features | Per-pigment concentration index rasters (relative units with uncertainty bounds) delivered as GeoTIFF and ingested into client GIS |
| Toxic-genus-consistent bloom alert | Threshold exceedance on PFT classification combined with chlorophyll-a anomaly; spectral signature match against published pigment reference libraries for Karenia, Alexandrium and Pseudo-nitzschia | Polygon alert layer with genus-consistency flag, confidence tier and recommended in-water sampling priority; delivered via GIS feed or email alert within agreed latency window |
| Bloom trajectory and extent time series | Multi-date compositing of PACE OCI and Sentinel-3 OLCI retrievals; cloud-gap-filled using temporal interpolation where clear observations permit | Weekly animated extent map and tabular bloom area statistics for aquaculture licensing or regulatory reporting |
| High-resolution bloom structure mapping (priority sites) | PRISMA or DESIS hyperspectral retrieval with dedicated ocean-colour atmospheric correction applied over nominated farm or shellfish bed coordinates | 30 m resolution PFT classification scene delivered within agreed scheduling window; requires commercial tasking licence held by client |
| In-water sampling prioritisation report | Spatial clustering of high-confidence toxic-genus-consistent pixels to identify highest-risk zones; integrated with vessel transit cost model if client provides vessel positions | PDF and GIS report recommending ranked sampling stations for the following 48–72 hours, updated on each clear-sky overpass |
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.