The image is a calculation.
Ocean-colour products are assembled from a chain of corrections and assumptions, most of them made before anyone looked at the water. This is what is in that chain, and where it bends.
A satellite bloom map looks like a photograph and is not one. It is the output of a processing chain in which the largest single correction has nothing to do with the water at all.
1. What arrives at the sensor
An orbiting radiometer records radiance at the top of the atmosphere in a set of visible and near-infrared bands. Over a typical ocean scene, the overwhelming majority of that signal is atmosphere: molecular scattering, aerosol, and specular reflection off the sea surface. The part that came out of the water is a small residual.
So the first and hardest step is removing everything that is not water-leaving radiance. Get the aerosol model wrong over a dusty coastline and the residual is wrong by a proportion that no downstream algorithm can recover.
This is why two providers can publish different chlorophyll fields from the same overpass and both be defensible. They made different atmospheric assumptions.
2. From radiance to a pigment estimate
Standard chlorophyll algorithms use ratios between blue and green bands, calibrated against a global set of shipboard measurements. In clear open ocean, where phytoplankton dominate the optical signal, that relationship is strong and well tested.
Coastal water is not that. Suspended sediment scatters, dissolved organic matter absorbs in the blue, and in shallow water the bottom reflects. All three push a band ratio in directions that look like chlorophyll and are not.
Regional algorithms and machine-learned retrievals exist to handle this, and they work better where they were trained. Ask any supplier where their coastal algorithm was tuned, and whether it was tuned on water resembling yours.
3. The pixel problem
A pixel is an average. In a narrow inlet, a marina or a lagoon fringed by mangrove, a single pixel may contain water, wet sand, vegetation and a jetty, and the retrieval will return one number for all of it. Adjacent bright land also scatters light into nearby water pixels, which biases the retrieval near exactly the shoreline that people care about.
This is not solved by asking for finer resolution alone, because finer pixels usually mean a narrower swath, a longer revisit and a noisier signal. It is a set of trade-offs, and the right point on the curve depends on the size of the water body you are watching.
4. Revisit, cloud, and the only coverage figure that matters
Procurement documents quote revisit interval. Operations care about clear-sky revisit, which is a different and much worse number.
A sensor with a nominal daily revisit over a coast that is overcast, hazy or dust-laden for a fortnight delivers nothing for a fortnight. In several of the regions where blooms do the most damage, the worst atmospheric conditions coincide with the bloom season, which is not a coincidence — both follow the same seasonal wind.
Before buying anything, pull two years of archive imagery for your own area and count the days with usable water pixels. That number is the honest specification of what earth observation will give you, and it is often half of what the brochure implies.
5. What a well-specified operational index looks like
The clearest example in marine management comes from an adjacent problem. NOAA Coral Reef Watch converts satellite sea-surface temperature into Degree Heating Weeks, a cumulative index of accumulated thermal stress, then maps that index onto named alert levels and states in plain language what each one is expected to produce on a reef.
Alert Level 1 begins at four degree-Celsius-weeks and carries a risk of reef-wide bleaching. Level 2 begins at eight, and adds mortality among heat-sensitive corals. Levels 3, 4 and 5 were added in December 2023 after the heat stress that year outran the old scale: multi-species mortality from twelve, severe multi-species mortality above half of corals from sixteen, near-complete mortality above eighty per cent from twenty.
What makes it good is not the physics. It is that the index is defined, the levels are published, each one is stated in outcome terms rather than in units, and anyone can check the product afterwards against what actually happened on the reef. Note also what happened in 2023: the scale was extended because reality exceeded it, in public, with a date. Bloom products should be judged against that standard and most are nowhere near it.
6. A regional blank space
Coverage is not uniform, and the gaps are not where you would guess. A 2021 study in the Journal of Marine Science and Engineering, working from MODIS chlorophyll-a data, observed that very few harmful algal bloom studies had been carried out in the Red Sea — a basin lined with desalination plants supplying drinking water to millions of people, and exposed to biofouling risk from high-biomass blooms.
That is a striking asymmetry. Some of the world's most exposed water infrastructure sits on one of its least studied bloom regimes, and the satellite record is the only long-baseline observation most of that coast has.
The practical consequence for anyone building a programme there: you will be establishing a baseline, not consulting one. Budget for that.
7. Reading a supplier's claim
Four questions separate a real product from a demonstration.
Which sensor and which processing level. Where the coastal algorithm was validated, and against how many matched in-situ samples. What clear-sky coverage the product achieved over the client's own area last year. And what the false-alarm rate was when it was checked against sampling.
A supplier who answers all four quickly is worth talking to. One who answers with imagery is showing you a picture of the sea.
Sources cited on this page
- NOAA Coral Reef Watch — the Degree Heating Week product itself, and the bleaching alert levels, which is where levels 3 to 5 are defined. The tutorial page describes only the first two.
- Journal of Marine Science and Engineering, MDPI, 2021 — MODIS chlorophyll-a assessment of harmful algal bloom risk in the Red Sea.
- Alarivean, Inc. — data services and full-cycle monitoring, for how these products are used to route intervention.
Asked in procurement, and afterwards
Five questions about what a satellite can do
Can a satellite tell us which species is blooming?
Not on its own, and not reliably in coastal water. Multispectral ocean-colour sensors estimate pigment concentration, and pigment is shared across almost all phytoplankton.
Hyperspectral instruments can sometimes separate pigment assemblages well enough to suggest a dominant group, which usefully narrows the candidate list. Confirming a species still needs a sample under a microscope, or a molecular assay.
Can it see a bloom below the surface?
Only the upper part of the water column contributes to the signal, so a subsurface layer sitting beneath clearer water is largely invisible from orbit.
That matters more than it sounds. Thin subsurface layers are common, and they are exactly what damages a deep intake or a net pen. Depth structure is a job for a profiler, a glider or a moored string.
What spatial resolution do we actually need?
It depends on the size of your water body, not on what is being marketed. A pixel of several hundred metres tracks a large offshore feature perfectly well and is useless inside a marina.
Ask how many usable water pixels the product yields inside your area at your typical tide. That is the specification. Nominal resolution is not.
Why does the satellite bulletin disagree with our buoy?
Because they measure different quantities at different scales. One is an estimate averaged across a pixel and across the first optical depth. The other is fluorescence at a single point at a single depth.
Occasional disagreement is normal and informative. Persistent disagreement in one direction usually points at atmospheric correction over turbid water — or at a fluorometer that needs cleaning.
Should we buy imagery or buy a service?
Buy imagery if you employ people who can maintain a processing chain and defend its outputs under scrutiny. Buy a service if you do not.
The expensive failure is the middle path: a programme licenses data, builds a chain around one enthusiastic analyst, and quietly loses the capability the month that person changes job.
Before the procurement closes
Have someone check the clear-sky number.
Send the area, the season and the products you are being offered. Alarivean will say what earth observation can realistically contribute at that site, and where an in-situ layer would do more for less.