Water colour as a proxy for dissolved oxygen and eutrophication state
Hyperspectral and multispectral sensors retrieve chlorophyll-a, phycocyanin and dissolved organics from spectral reflectance, giving water managers an eutrophication signal without field crews. Dissolved oxygen itself is invisible to satellites; the proxy works until stratification or rapid mixing breaks the relationship.
Sensors
- NASA PACE OCI: Hyperspectral ocean-colour imager covering 340–890 nm at ~5 nm spectral sampling, 1.2 km spatial resolution, daily global revisit from a sun-synchronous orbit at 676 km. Designed to resolve phytoplankton community composition and distinguish chlorophyll-a from coloured dissolved organic matter (CDOM) using full spectral shape rather than band ratios alone.
- Sentinel-3 OLCI: 21 bands from 400–1020 nm, 300 m spatial resolution, 2–3 day revisit for mid-latitudes with two satellites (Sentinel-3A and 3B). Purpose-built for ocean colour with dedicated atmospheric correction bands; the primary workhorse for coastal and large-lake eutrophication monitoring in the Copernicus programme.
- Sentinel-2 MSI: 13 bands from 443–2190 nm; red-edge bands at 705 nm and 740 nm (20 m resolution) are particularly useful for chlorophyll-a and phycocyanin retrieval in small inland lakes where OLCI's 300 m footprint smears the signal. 5-day revisit with both satellites. Not designed for water colour; atmospheric correction over dark water surfaces is a significant source of error.
- Landsat 8/9 OLI-2: Coastal/aerosol band at 443 nm plus five visible-to-SWIR bands, 30 m resolution, 8-day revisit per satellite (16-day per sensor, ~8-day combined). Useful for long time-series analysis of eutrophication trends in reservoirs and lakes; archive extends to 1984 for Landsat 5 TM, giving multi-decadal context. Narrower spectral coverage than dedicated ocean-colour sensors limits pigment discrimination.
What the spectrum actually encodes
Water reflects very little light, which is precisely why small spectral shifts carry large biogeochemical signals. Chlorophyll-a absorbs strongly near 443 nm and 675 nm, producing a characteristic reflectance peak around 550 nm (the green of a bloom) and a fluorescence emission peak near 685 nm. Phycocyanin, the accessory pigment that distinguishes cyanobacteria from green algae, has an additional absorption feature near 620 nm. CDOM absorbs exponentially from the UV into the blue, suppressing reflectance at short wavelengths. Total suspended matter scatters broadly across the visible and near-infrared. Separating these four constituents simultaneously is the central retrieval problem, and it requires either many narrow spectral bands or strong prior knowledge about which constituent dominates.
PACE OCI's continuous hyperspectral coverage makes full spectral unmixing tractable for open ocean and large coastal systems. For small inland lakes, Sentinel-2's red-edge bands at 705 nm and 740 nm offer a practical shortcut: the ratio of reflectance at 705 nm to 665 nm correlates with chlorophyll-a concentrations in the range of roughly 5 to several hundred µg/L, a span that covers oligotrophic through hypereutrophic states. Published studies using this band combination have reported retrieval uncertainties of 20–40% depending on the dominance of co-varying CDOM, which is an honest figure to carry into any monitoring programme.
Dissolved oxygen: present in the inference, absent from the signal
Oxygen does not absorb or scatter visible light at concentrations relevant to water quality. There is no spectral fingerprint. What satellites detect is the biological and optical state that correlates with oxygen dynamics: high chlorophyll-a indicates photosynthetic production during the day and potential nocturnal depletion by respiration; dense surface scums of cyanobacteria can block light penetration and accelerate hypolimnetic oxygen loss over days to weeks. The proxy relationship is well-established in well-mixed, shallow, productive systems, and it is unreliable in stratified lakes where surface colour and bottom-water oxygen are effectively decoupled.
Rapid weather events compound the problem. A wind event that mixes a stratified lake can collapse surface chlorophyll concentrations within hours, long before the next satellite overpass. A satellite image taken the day after a mixing event may show apparently healthy water while bottom-water anoxia persists. Any monitoring programme that uses satellite colour as an oxygen proxy must be explicit about these failure modes with the end user, and should pair satellite observations with at least periodic in-situ profiling during the stratified season.
Atmospheric correction: the error that dominates over dark water
Over open ocean, the atmosphere contributes roughly 90% of the top-of-atmosphere radiance measured by a satellite sensor. Over dark inland lakes, that fraction is even higher, because the water-leaving signal is so faint. A 1% error in aerosol optical depth retrieval can translate into a 20–50% error in retrieved water-leaving reflectance, which propagates directly into chlorophyll-a estimates. This is not a solvable problem with better sensors alone; it requires accurate aerosol characterisation at the time and location of the image.
Sentinel-3 OLCI and PACE OCI carry dedicated short-wave infrared and UV bands specifically to constrain aerosol type and loading before correcting the visible channels. Sentinel-2 was not designed with this in mind, and standard atmospheric correction processors (Sen2Cor, ACOLITE, C2RCC) produce meaningfully different chlorophyll-a retrievals over the same lake on the same date. ACOLITE and C2RCC are generally preferred for inland water applications based on published inter-comparisons, but neither eliminates the problem. Users should treat Sentinel-2 chlorophyll-a retrievals as indicative rather than metrological, particularly for lakes smaller than roughly 1 km² where adjacency effects from surrounding land add a further bias.
Eutrophication indices you can actually compute from orbit
Several operational indices translate multi-band retrievals into management-relevant outputs. The Floating Algae Index (FAI), computed from the near-infrared and red bands of Landsat or Sentinel-2, detects surface scums at spatial resolutions fine enough to map bloom extent within a reservoir. The Cyanobacterial Index (CI) used in NOAA's operational harmful algal bloom products exploits the phycocyanin absorption feature at 620 nm and is available from OLCI at 300 m resolution. The Trophic State Index (TSI) derived from Secchi depth equivalents estimated from diffuse attenuation coefficients is computable from PACE OCI retrievals for large water bodies.
None of these indices is a direct regulatory measurement. They are spatial and temporal indicators that tell a water manager where to send a boat, not what the laboratory will find when the boat arrives. That framing is the correct one: satellite eutrophication monitoring is a prioritisation and early-warning tool, not a substitute for the in-situ sampling that underpins legal compliance.
Archive depth and what it reveals about long-term trends
Landsat's archive from 1984 onward is the most valuable asset for understanding whether a lake's eutrophication state is worsening, stable or recovering. Consistent retrieval of a simple band ratio such as the green-to-red reflectance across Landsat 5 TM, 7 ETM+, 8 OLI and 9 OLI-2 can reveal multi-decadal chlorophyll-a trends in lakes large enough to resolve at 30 m. Cross-sensor calibration is non-trivial and requires careful harmonisation, but the USGS Collection 2 processing baseline has reduced inter-mission biases substantially.
Sentinel-3 OLCI provides a denser time series from 2016 onward at 300 m, suitable for lakes above roughly 5–10 km². PACE OCI, launched in February 2024, adds hyperspectral capability but has no historical archive. Combining PACE for pigment discrimination with Landsat for historical context is a reasonable analytical strategy for large-lake monitoring programmes. Satellize applies this kind of multi-mission fusion in its analytics work, including the Tonga crop-estimation programme, where harmonising sensors across different eras is a routine part of the methodology.
What a monitoring programme needs to get right from the start
Define the water bodies by size. Lakes smaller than 1 km² are poorly served by any current operational sensor; Sentinel-2 at 20 m is the practical floor, and atmospheric correction uncertainty will dominate the error budget. Lakes above 10 km² are tractable with OLCI or PACE. Reservoirs with complex shorelines introduce adjacency effects regardless of size.
Agree in advance on which outputs are decision-relevant. Bloom extent maps, peak chlorophyll-a estimates, trophic state classifications and anomaly alerts relative to a historical baseline are all achievable at different confidence levels. A system that flags a 50% increase in chlorophyll-a relative to the five-year mean for the same calendar week is a useful operational product. A system that claims to report absolute dissolved oxygen to 1 mg/L from orbit is not. The honest version of this service is more useful than the overclaimed one, because water managers can act on it without being burned by false precision.
Typical figures
| Spatial resolution (inland water focus) | 20–60 m (Sentinel-2 red-edge bands); 30 m (Landsat 8/9 OLI-2); 300 m (Sentinel-3 OLCI); 1.2 km (PACE OCI) |
| Revisit frequency | ~1 day (PACE OCI, single satellite); 2–3 days (Sentinel-3A+B combined, mid-latitudes); 5 days (Sentinel-2A+B combined); ~8 days (Landsat 8+9 combined) |
| Spectral coverage | 340–890 nm continuous at ~5 nm sampling (PACE OCI); 443–1020 nm in 21 bands (OLCI); 443–2190 nm in 13 bands (Sentinel-2 MSI); 443–2200 nm in 9 bands (Landsat OLI-2) |
| Minimum lake size (practical floor) | ~1 km² for Sentinel-2; ~5–10 km² for Sentinel-3 OLCI; ~50 km² for PACE OCI without significant adjacency or mixed-pixel error |
| Chlorophyll-a retrieval range (published) | Approximately 1–500 µg/L for red-edge band ratio methods; retrieval uncertainty typically 20–40% depending on CDOM co-variation |
| Cloud cover limitation | Optical sensors only; cloudy pixels are unusable. Persistent cloud cover in humid tropical regions can reduce clear-sky observations to fewer than 5–10 per month |
| Atmospheric correction error (dark water) | Dominant error source; 1% aerosol optical depth error can translate to 20–50% error in water-leaving reflectance; ACOLITE and C2RCC preferred over Sen2Cor for inland water |
| Archive depth | Landsat: 1984 to present (Collection 2); Sentinel-2: 2015 to present; Sentinel-3 OLCI: 2016 to present; PACE OCI: February 2024 to present (no historical archive) |
| Latency (open data streams) | Sentinel-2 and Sentinel-3: typically 3–24 hours after acquisition via Copernicus Data Space; Landsat: same-day to next-day via USGS EarthExplorer; PACE: near-real-time products under development |
| Dissolved oxygen observability | Not directly observable at any resolution. Proxy relationship via chlorophyll-a and trophic state; breaks down in stratified or rapidly mixed systems |
Analytics Satellize can run
| Chlorophyll-a concentration map | Red-edge band ratio (705/665 nm) for Sentinel-2; semi-analytical bio-optical inversion for OLCI and PACE OCI; atmospheric correction via ACOLITE or C2RCC | GeoTIFF raster layer per overpass, classified into trophic state bins (oligotrophic, mesotrophic, eutrophic, hypereutrophic), delivered to client GIS or web dashboard |
| Cyanobacteria presence and bloom extent | Phycocyanin index from 620 nm absorption feature (OLCI CI product); Floating Algae Index from NIR-red-SWIR combination (Sentinel-2/Landsat) | Polygon shapefile of bloom extent with area estimate (km²) and peak pigment intensity; email or API alert when bloom exceeds defined threshold |
| Multi-decadal trophic state trend | Harmonised Landsat Collection 2 time-series analysis; cross-sensor calibration using published USGS surface reflectance products; Mann-Kendall trend test on annual peak chlorophyll-a | Annual trend report with time-series chart, statistical significance, and change-point detection; suitable for regulatory baseline documentation |
| Anomaly alert relative to historical baseline | Z-score or percentile ranking of current-week chlorophyll-a against same-week historical distribution from Sentinel-2 or Landsat archive | Weekly automated alert layer flagging water bodies where current state exceeds the 90th percentile of the historical record; delivered as GIS layer or structured JSON feed |
| CDOM and total suspended matter co-retrieval | Semi-analytical or machine-learning inversion of multi-band reflectance; trained on published in-situ datasets for optically complex inland waters | Separate GeoTIFF layers for CDOM absorption coefficient (m⁻¹) and TSM concentration (g/m³), flagged with retrieval confidence score per pixel |
| Eutrophication risk score for reservoir network | Composite index combining chlorophyll-a, bloom frequency, CDOM loading and historical trend gradient; weights configurable to client regulatory framework | Ranked table of monitored water bodies by risk score, updated monthly, with supporting evidence layers; formatted for water authority reporting cycles |
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.