Phytoplankton size-class partitioning as a forage fish prey availability index
The ratio of microphytoplankton to picophytoplankton in surface waters predicts prey field quality for small pelagic fish far better than bulk chlorophyll alone. MODIS Aqua, PACE OCI and Sentinel-3 OLCI make basin-scale size-class mapping operationally feasible, with important caveats in turbid coastal water.
Sensors
- NASA MODIS Aqua: 36 spectral bands, 9 ocean colour bands at 1 km resolution, daily global revisit. The primary workhorse for phytoplankton size-class algorithms; archive runs from 2002, giving two decades of climatological baseline for anomaly detection.
- NASA PACE OCI: Hyperspectral ocean colour imager launched February 2024, 5 nm spectral sampling from 340 to 890 nm at approximately 1 km resolution. Resolves phytoplankton absorption features that broadband sensors cannot separate, improving size-class discrimination particularly for smaller cells.
- Sentinel-3 OLCI: 21 bands between 400 and 1020 nm at 300 m resolution, with two satellites providing roughly daily global revisit. Finer spatial grain than MODIS is useful near shelf breaks and upwelling filaments where forage fish concentrate.
- CMEMS Ocean Colour Multi-Sensor L4: Merged, gap-filled daily chlorophyll and inherent optical property fields produced by the Copernicus Marine Service from MODIS, VIIRS, OLCI and SeaWiFS heritage. Reduces cloud gaps; size-class algorithms can be applied to the merged Rrs fields.
Why cell size matters more than total greenness
Bulk chlorophyll concentration, the most widely mapped ocean colour product, tells you how much photosynthetic biomass is present. It does not tell you what form that biomass takes. A square kilometre of ocean dominated by Prochlorococcus cells roughly 0.6 micrometres across is nutritionally almost useless to an anchovy. The same chlorophyll concentration expressed as diatom chains 20 to 200 micrometres long is a productive feeding ground.
Small pelagic forage fish, including anchovies, sardines, herrings and sprats, are selective filter feeders. Their gill rakers physically exclude cells below roughly 10 to 20 micrometres, so microphytoplankton and the microzooplankton that graze on them constitute the effective prey field. The fraction of total phytoplankton biomass in the micro size class (conventionally 20 to 200 µm) relative to the pico class (0.2 to 2 µm) therefore carries direct ecological meaning for stock assessment modellers. It is a proxy for trophic transfer efficiency, not just primary production.
What the partitioning algorithms actually do
Two published approaches dominate operational use. PHYSAT, developed by Alvain and colleagues and published in Remote Sensing of Environment, identifies dominant phytoplankton groups from anomalies in normalised water-leaving radiance spectra relative to a chlorophyll-matched background. It is a classification approach: each pixel is assigned to a dominant group such as diatoms, Phaeocystis or Synechococcus-like organisms.
The Hirata et al. 2011 method, published in Biogeosciences, takes a different route. It derives size-class fractions statistically from chlorophyll concentration alone, using empirical relationships calibrated against global in-situ pigment databases. This makes it computationally simple and applicable to any chlorophyll product, but it conflates the signal: two water masses with identical chlorophyll but different community structures will receive identical size-class fractions. PACE OCI's hyperspectral capability is designed precisely to break this degeneracy, resolving phytoplankton absorption features that allow more direct retrieval of community composition without relying on the chlorophyll shortcut.
Both methods carry retrieval uncertainty that grows substantially in optically complex, or Case-2, coastal waters. Coloured dissolved organic matter, suspended sediment and bottom reflectance all contaminate the water-leaving radiance signal that these algorithms interpret as phytoplankton. In the turbid estuarine and shelf waters where many forage fish stocks spawn and recruit, the open-ocean assumptions embedded in standard atmospheric correction and in the algorithms themselves can produce size-class estimates that are unreliable or simply wrong. This is not a minor caveat. It is the central practical constraint on the method.
Connecting the satellite product to a stock assessment model
Phytoplankton size-class fields enter stock assessment as environmental covariates rather than as direct abundance estimates. The typical workflow begins with a time series of microphytoplankton fraction across the stock's known feeding habitat, usually defined by a bathymetric or temperature mask. That fraction is then averaged over biologically meaningful windows: the pre-spawning feeding season, the larval drift period, or the period of maximum juvenile growth. The resulting index is tested for lagged correlation with recruitment indices or survey biomass estimates from acoustic trawl surveys.
The logic is that prey field quality in the months before and during spawning influences egg and larval condition, which propagates into recruits one to three years later depending on the species. Studies of Peruvian anchoveta in the Humboldt upwelling system have shown that diatom-dominated conditions correlate with stronger recruitment than flagellate-dominated conditions at similar chlorophyll levels, consistent with the size-selectivity mechanism. Sardine stock assessments in the California Current and the Benguela have used analogous environmental indices. The satellite product does not replace acoustic or trawl survey data; it provides spatial and temporal coverage that surveys cannot.
Resolution floors, cloud and the coastal problem
MODIS Aqua and PACE OCI both operate at approximately 1 km nadir resolution. Sentinel-3 OLCI at 300 m is finer, but ocean colour retrieval quality degrades within roughly 2 to 5 km of the coast due to adjacency effects from land. Phytoplankton patches relevant to forage fish aggregations can be smaller than 1 km, particularly in upwelling filaments and frontal zones. The satellite product captures the mesoscale and basin-scale pattern well; submesoscale structure is largely unresolved.
Cloud cover is a persistent constraint in high-productivity upwelling regions, many of which sit under persistent stratocumulus decks. Daily single-sensor coverage can have cloud fractions exceeding 70 percent in some seasons. The CMEMS merged L4 product uses temporal and spatial interpolation to fill gaps, but interpolated pixels carry additional uncertainty and can smooth out genuine short-lived bloom events. Users building recruitment indices should track what fraction of their domain and time window is based on observed versus interpolated retrievals.
Atmospheric correction over coastal water remains an active research problem. The standard NASA atmospheric correction assumes negligible water-leaving radiance in the near-infrared, an assumption that fails in turbid water. Alternative schemes such as POLYMER and the MUMM approach exist and improve matters, but no correction is perfect. Any size-class product derived in shelf seas shallower than roughly 30 metres should be treated with additional scepticism and validated against in-situ pigment measurements where possible.
Building an operational index: what a fisheries agency actually receives
A practical implementation produces a monthly or weekly raster of microphytoplankton fraction across the stock assessment domain, delivered as a georeferenced GeoTIFF or NetCDF alongside a per-pixel retrieval quality flag. The quality flag distinguishes open-ocean retrievals, coastal retrievals with elevated uncertainty, cloud-filled pixels and cloud-interpolated pixels. From that raster stack, a time series of domain-averaged or habitat-weighted microphytoplankton fraction is extracted and formatted as a table suitable for direct ingestion into stock assessment software such as Stock Synthesis or CASAL.
Satellize runs this workflow on open constellations, primarily MODIS Aqua and Sentinel-3 OLCI, with PACE OCI integration now in development given the sensor's 2024 launch. The Tonga crop-estimation programme demonstrated the organisation's approach to environmental index construction from open satellite data in a sovereign context; the same pipeline logic applies here, adapted to ocean colour physics rather than vegetation indices. Fisheries agencies that want to test the index against their existing survey series before committing to operational use can request a retrospective analysis covering the MODIS archive period from 2002 onwards.
Typical figures
| Spatial resolution (MODIS Aqua / PACE OCI) | ~1 km at nadir |
| Spatial resolution (Sentinel-3 OLCI) | 300 m at nadir; coastal adjacency effects within ~2–5 km of land |
| Revisit (single sensor) | Daily global for MODIS and PACE; ~1–2 days for dual Sentinel-3 |
| Effective cloud-free revisit | Highly variable; CMEMS L4 merged product fills gaps via interpolation, adding uncertainty |
| Key spectral bands | 400–900 nm ocean colour; PACE OCI adds continuous 5 nm hyperspectral sampling 340–890 nm |
| Archive depth | MODIS Aqua from July 2002; SeaWiFS heritage to 1997 via CMEMS; PACE from early 2024 |
| Size classes resolved | Pico (0.2–2 µm), nano (2–20 µm), micro (20–200 µm); fraction estimates, not cell counts |
| Primary retrieval uncertainty zone | Case-2 coastal and shelf waters; turbidity, CDOM and bottom reflectance degrade algorithm assumptions |
| Delivery formats | GeoTIFF, NetCDF-4 with CF conventions, CSV time-series extract |
| Latency (near-real-time MODIS) | Level-2 products available within ~3 hours of overpass from NASA OBPG |
Analytics Satellize can run
| Monthly microphytoplankton fraction raster | Hirata et al. 2011 abundance partitioning applied to MODIS Aqua or OLCI chlorophyll fields | GeoTIFF stack with per-pixel quality flag; one file per month per domain |
| PHYSAT dominant-group classification | Normalised water-leaving radiance anomaly classification (Alvain et al.) applied to MODIS Rrs | Classified raster (diatoms, Phaeocystis, cyanobacteria, nanoeukaryotes, Prochlorococcus) per overpass |
| Seasonal prey-field quality index | Domain-averaged microphytoplankton fraction weighted by habitat mask; aggregated over user-defined biological window | Annual time-series CSV formatted for direct input to Stock Synthesis or similar assessment software |
| Anomaly maps relative to 2002–2024 climatology | Z-score normalisation against MODIS archive monthly climatology | GeoTIFF anomaly layer and summary report flagging basin-scale departures from historical mean |
| Coastal retrieval uncertainty assessment | Case-2 water flag derived from CDOM and suspended particulate matter proxies; comparison of standard and POLYMER atmospheric correction | Spatial mask distinguishing reliable open-ocean pixels from elevated-uncertainty coastal pixels, delivered with every product |
| Lagged correlation analysis with survey recruitment index | Cross-correlation of satellite prey-field index against client-supplied acoustic or trawl recruitment time series at lags of 1–4 years | Statistical summary report with correlation coefficients, confidence intervals and recommended covariate lag for assessment model |
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.