At four in the afternoon, a fishing lake can look reassuring. The water is warm, the fish may still be active, and a dissolved-oxygen reading near the surface may appear entirely acceptable. Yet the lake may already be moving towards the most stressful part of its daily oxygen cycle—at a time when few people are watching.
Heat, dissolved oxygen, algae and cyanobacteria are connected, but not through one simple cause-and-effect chain. The most useful monitoring system is therefore not necessarily the one with the largest number of sensors. It is the one that answers the operator’s actual questions:
The same core measurement problem appears in recreational fishing lakes, inland fish farms, offshore cages and mussel farms. The underlying physical and biological processes are shared; what changes is the degree of water exchange, stocking density, salinity, exposure and the biological objective.
- Is oxygen becoming dangerous for the fish?
- Is biological activity contributing to the change?
- Are cyanobacteria—or another locally relevant bloom group—becoming more prominent?
- Can the trend be recognised early enough to intervene?
This post examines what dissolved oxygen, chlorophyll-a, cyanobacteria, turbidity and salinity measurements can—and cannot—tell us across fishing lakes, inland and offshore fish farms, and brackish or marine mussel farms. The final sections return to the timing question: when is oxygen risk commonly greatest, and why can an afternoon inspection miss it?
Heat Changes the Oxygen Budget
As water warms, its capacity to hold dissolved oxygen decreases. At the same time, respiration by fish and microorganisms continues, and biological oxygen demand may rise. Hot, calm weather can also strengthen thermal stratification, reducing exchange between oxygenated surface water and deeper layers.
These mechanisms do not affect every fishing lake, aquaculture or mussels farm equally. Depth, wind exposure, water clarity, organic loading, fish/mussels density and basin geometry all influence the result. Even relatively shallow water can develop important vertical or spatial differences.
This is the first reason that a single occasional measurement is weak evidence. A reading describes one place and one moment. It does not reveal the direction of travel.
The First Sensor Is Not an “Algae Sensor”
If the management objective is fish or mussels protection, dissolved oxygen is the logical starting point. It is the environmental variable directly connected to the immediate oxygen available to fish.
Temperature should normally be measured with it. Temperature affects oxygen solubility, fish metabolism, mixing and the interpretation of both oxygen concentration and percentage saturation.
At offshore and brackish-water sites, conductivity or salinity belongs in the baseline as well. Salinity affects oxygen solubility and helps identify freshwater pulses, stratification and changing water masses. In mussel farming it also provides important context for filtration conditions and shifts in the phytoplankton community.
A continuous DO and temperature series begins to answer operational questions that a spot measurement cannot:
- How large are the recurring oxygen fluctuations?
- Is the minimum becoming progressively lower?
- How rapidly is oxygen declining?
- Is the water column behaving uniformly?
- Did an aerator produce the expected response?
The appropriate alarm level is site- and organism-specific. Species, life stage, stocking density, water temperature and the duration of exposure all matter. Rate of change can be as important as a fixed threshold: a rapidly falling value may justify intervention before the alarm level is reached.
Can AI Forecast an Oxygen Drop From DO Alone?
Sometimes—provided the claim is kept precise.
A sufficiently long, regular DO time series contains information about recurring cycles and the site’s response to previous conditions. Time-series and machine-learning models can learn those patterns and estimate a future DO trajectory. Eze and Ajmal (2020), for example, reported encouraging short- and longer-term forecasts from an aquaculture-pond DO series using a hybrid EEMD–LSTM approach.
This does not mean that a model trained at one site can simply be transferred to another. A useful operational forecast requires local data, out-of-sample testing and continuing validation as seasons and site conditions change.
DO history alone may also contain patterns consistent with high biological productivity or bloom activity. But it cannot reliably establish the cause. It cannot tell the operator whether algae or cyanobacteria dominate, which species are present, or whether toxins are being produced.
Adding temperature, weather and pigment measurements can make the interpretation more specific. Research has also shown that cyanobacterial dynamics can be forecast from incomplete environmental datasets, although the resulting performance remains dependent on the variables, locations and validation strategy used (Fournier et al., 2024).
The safe operational distinction is this: DO-only models may forecast oxygen risk. They do not directly diagnose a bloom.
What a Chlorophyll-a Sensor Actually Sees
Chlorophyll-a is present in oxygen-producing phytoplankton, including eukaryotic algae and cyanobacteria. Chlorophyll-a fluorescence is therefore widely used as a proxy for broader phytoplankton abundance or activity.
The important word is proxy. A fluorometer does not count organisms. It emits light at selected wavelengths and measures light emitted by excited pigments. The relationship between that optical signal and actual biomass can change with:
- Species and community composition.
- Cell size, colonies and pigment content.
- Physiological state and recent light exposure.
- Non-photochemical quenching.
- Temperature and dissolved organic matter.
- Suspended particles.
- Sensor geometry, range and calibration.
Cyanobacteria also contain chlorophyll-a, but conventional CHL fluorometers may not represent them in the same way as green algae or other phytoplankton because accessory pigments and cellular organisation affect the optical response.
USGS field guidance consequently treats in-place fluorescence as an operational proxy requiring objective-specific calibration, interference testing, cleaning and quality assurance rather than as an absolute measure of biomass (Foster et al., 2022). Field evaluation in the Finger Lakes likewise demonstrated the importance of comparison with discrete samples and of understanding sensor-specific performance (Johnston et al., 2024).
DO + CHL is therefore useful when the question is: are broader phytoplankton dynamics changing alongside the oxygen cycle?
What a CYANO Sensor Adds
In freshwater monitoring, a CYANO fluorometer commonly targets phycocyanin, an accessory pigment strongly associated with cyanobacteria. This makes the channel more selective for freshwater cyanobacteria than a general chlorophyll-a measurement.
It does not make the signal exclusive or absolute. Response varies among cyanobacterial species because pigment content, cell arrangement and physiological state differ. Green algae, suspended material, natural organic matter, ambient light and temperature can also influence the reading.
Choo et al. (2018) tested multiple in-situ fluorometers against cyanobacterial cultures, green-algae interference and added turbidity. The instruments performed well in controlled monocultures, but interference could produce either overestimation or underestimation depending on the instrument and conditions.
Ma et al. (2022) found that natural organic matter and increasing temperature could reduce phycocyanin readings. Their field results also showed that the relationship between fluorescence and cyanobacterial biovolume depended partly on the dominant organisms and environmental context.
A CYANO signal is therefore sharper than CHL when freshwater or brackish water cyanobacteria are the specific concern—but it is not a species identification, a cell count or a toxin measurement.
The freshwater qualification matters. A CYANO channel based on phycocyanin fluorescence should not automatically be assumed to represent the bloom community at offshore or brackish sites. Marine and brackish communities may include phycoerythrin-rich cyanobacteria and many harmful eukaryotic microalgae that require a different optical channel, microscopy, molecular analysis or toxin testing. Sensor selection must follow the locally relevant organisms and pigments, not the word “algae” alone (Foster et al., 2022; McKergow, 2025).
DO + CYANO is useful when the question is: is a phycocyanin-associated signal changing at the same time as oxygen risk?
DO + CHL + CYANO is useful when the question is: is the broader phytoplankton community changing, the cyanobacterial component changing, or both?
Turbidity: The Signal That Helps Interpret Other Signals
Turbidity is interesting precisely because fluorescence is an optical measurement.
Suspended particles scatter and absorb light. They can affect both the excitation light sent into the water and the fluorescent light returning to the detector. Depending on particle type, concentration and instrument geometry, the result may be attenuation, additional scattering, loss of linearity, or a positive or negative measurement bias.
In practical terms, an increase in CHL or CYANO can be partly masked—or apparently amplified—when the optical environment changes. A simultaneous turbidity channel can therefore:
- Flag periods in which fluorescence is less trustworthy.
- Help distinguish pigment trends from sediment or resuspension events.
- Provide an input to a locally calibrated correction model.
- Improve automated quality-control rules.
But turbidity does not automatically compensate fluorescence. A bloom can itself increase turbidity, and dense biomass can produce self-shading or optical attenuation that is not fully described by a separate turbidity value. The direction and magnitude of the correction must be established using the actual sensor and representative site samples (Foster et al., 2022; Choo et al., 2018; McKergow, 2025).
Turbidity is best treated as context and quality control: it can reveal when fluorescence may be masked or distorted, but it cannot convert CHL or CYANO into a direct biomass measurement.
Choosing the Smallest Useful Monitoring Set
Scenario 1: The Priority Is Preventing Acute Fish Oxygen Stress
Start with DO + temperature. This combination measures the immediate oxygen condition, its trend and the response to aeration. It does not identify the cause of a decline.
Scenario 2: The Priority Is Earlier Oxygen Warning
Use DO + temperature + sufficient local history. A validated forecasting model can estimate the likely trajectory and the time remaining before a management threshold is reached.
Scenario 3: Broader Algal Dynamics Matter
Add CHL. This provides a proxy for changes in the wider phytoplankton community. Site-specific samples are needed to relate fluorescence to biomass or chlorophyll concentration.
Scenario 4: Freshwater Cyanobacteria Are a Specific Concern
Add CYANO. This provides a more selective phycocyanin-associated signal, but not species, cell-count or toxin information.
Scenario 5: The Operator Needs to Separate General Algal and Cyanobacterial Trends
Use DO + temperature + CHL + CYANO. The two pigment channels can be interpreted together while DO records the condition directly relevant to fish.
Scenario 6: Sediment, Runoff or Resuspension Changes Frequently
Add turbidity to the relevant configuration. It can identify changing optical conditions and support locally validated corrections.
Scenario 7: Toxins or Public-Health Decisions Are Involved
Use continuous sensors to guide representative sampling. No CHL or CYANO fluorometer replaces species identification or toxin analysis.
Scenario 8: The Site Is Offshore or Brackish
Add conductivity or salinity and select pigment channels based on the locally relevant phytoplankton community. In these environments, water-mass changes, tides and currents can dominate the signal, while a freshwater phycocyanin channel may not be the most informative bloom indicator.
How the Framework Changes by Production System
Inland Fish Farming
Inland ponds and tanks share much of the fishing-lake oxygen cycle, but stocking density, feeding and accumulated organic matter can make oxygen demand more intense and operational intervention more immediate. DO + temperature remains the baseline; CHL, CYANO and turbidity are added when bloom dynamics or changing optical conditions affect management. In intensive systems, pH and ammonia may also be important even though they are outside the narrower sensor comparison developed here.
Forecasting is particularly actionable because aerators, pumps or supplementary oxygen can often be activated directly. The pond study used by Eze and Ajmal (2020) illustrates the potential for site-specific DO forecasting, while Shoko et al. (2014) demonstrates why measurement time matters in stocked earthen ponds.
Offshore Fish Farming
Offshore cages usually experience more exchange than an inland pond, so oxygen dynamics may be driven by currents, tides, water masses, cage biomass, feeding and vertical fish distribution rather than by one repeatable pond cycle. The baseline should normally include DO, temperature and salinity or conductivity, with sensors placed at biologically relevant depths. Measurements upstream, within or downstream of a cage answer different questions.
CHL can provide broad phytoplankton context, but a freshwater CYANO channel is not a universal marine harmful algal bloom (HAB) sensor. Local bloom history and organism-specific sampling remain essential. Environmental conditions including oxygen, temperature, salinity, currents and light are recognised drivers of fish behaviour and distribution in sea cages (Oppedal et al., 2011).
Mussel Farming, Especially in Brackish Water
For mussels, phytoplankton is both food and potential hazard. CHL can provide a broad indicator of phytoplankton availability, while turbidity helps distinguish changing optical conditions and suspended inorganic material from a simple pigment trend. Neither signal describes food quality by itself.
Brackish sites are especially dynamic because freshwater inputs, salinity gradients, tides and stratification can rapidly change oxygen conditions and community composition. A practical baseline is therefore temperature + conductivity/salinity + DO, with CHL and turbidity added for phytoplankton and particle context. A pigment-specific channel should be chosen only after identifying the locally relevant organisms.
Some harmful microalgae can contaminate shellfish even when DO, CHL and visible water conditions appear unremarkable. Continuous sensors can direct attention and sampling, but they cannot replace species identification, marine-biotoxin analysis or the applicable shellfish-safety programme (Shumway, 1990).
What One Measurement Point Can Miss
A sensor measures the water surrounding it—not the whole water body or farm.
A surface sensor can miss lower oxygen near the bed. A sensor next to an aerator may record the aerator’s local effect rather than the wider fish habitat. Sheltered bays, inflows and deeper areas may follow different trajectories.
At sea, placement must also be interpreted relative to the prevailing current, cage geometry and stock depth. At a mussel longline or raft, conditions can differ between the surface and the lower culture depth, while a reversing tide can reverse which side of the farm is upstream and which is downstream.
Deployment design should therefore consider:
- Where the fish or cultured organisms are located during warm conditions.
- Depth and the likelihood of stratification.
- Inflows, outflows and sheltered zones.
- The position and influence of aerators.
- Whether more than one depth or location is needed.
Biofouling and deposited sediment can also create false trends. Cleaning, inspection, calibration checks and paired grab samples are not optional extras: they are part of the measurement system. USGS and NIWA guidance both emphasise matching sensor operation and quality assurance to the intended use of the data (Foster et al., 2022; McKergow, 2025).
By using multiple sensors, with extended time series, to train a Machine Learning model, one can expect to achieve significantly superior forecasts with respect to the bare DO forecasting models.
The Timing Trap: Why the Lowest Oxygen Often Arrives Before Dawn
We can now return to the lake that looked reassuring in the afternoon.
During daylight, algae, cyanobacteria and aquatic plants photosynthesise, adding oxygen to the water. In productive systems, this biological production can temporarily outweigh oxygen consumption and create an impressive afternoon reading.
After sunset, photosynthesis stops. Respiration does not: fish, plants, plankton and microorganisms continue consuming oxygen. The concentration can therefore decline through the night until light returns and photosynthetic production restarts.
That is why the daily minimum commonly occurs around dawn or shortly before it—not during the hottest and most visible part of the day. Measurements in earthen fish ponds have found significantly lower DO from midnight through the pre-dawn period than during the afternoon (Shoko et al., 2014). High-frequency monitoring of eutrophic Taihu Lake similarly recorded recurring minima around 05:00–06:00, alongside strong seasonal and day–night dynamics (Xie et al., 2025).
This pattern is especially relevant to productive, relatively enclosed lakes and inland ponds. It should not be imposed blindly on a well-flushed offshore cage, where tides, currents and passing water masses may control the minimum, or on a brackish lagoon where both diel biology and tidal exchange can be important. Continuous local data reveal which clock actually governs the site.
Blooms can widen this daily swing. A dense population may produce substantial oxygen in daylight and consume it through respiration at night. If the bloom collapses, decomposition can add a further oxygen demand.
The afternoon reading was not necessarily wrong. It simply answered the wrong question.
The operational question is not “What is the oxygen level while I am standing beside the lake?” It is “What will the lowest oxygen level be before the next recovery begins?”
From Prediction to Intervention
A useful warning system can combine the measured DO value, its current rate of decline, the expected minimum, forecast uncertainty and a site-specific intervention level.
In a fishing lake or inland fish farm, that warning may provide time to activate aeration, circulation or supplementary oxygen before visible fish distress. Offshore, it may inform feeding, oxygenation or other cage-management decisions. In mussel farming, it may trigger targeted sampling, inspection or harvest-related decisions rather than an attempt to control the surrounding water body.
Monitoring must then continue. Post-intervention data show whether conditions actually improved, how quickly they changed and whether the selected sensor position represents the organisms being managed.
Operational Tools
For operational monitoring tools, have a look at our netH₂O buoys: the B100-5 for inland sites and the B400-10 for more demanding coastal and offshore installations.
Conclusions
- DO + temperature is the foundation for managing fish oxygen risk.
- DO history can support site-specific forecasts of oxygen decline, but DO alone does not diagnose blooms.
- CHL is a proxy for broader phytoplankton dynamics.
- A freshwater CYANO channel is more selective for cyanobacteria but does not measure species, cells or toxins directly.
- Offshore and brackish sites also require salinity or conductivity context, and their locally relevant pigment channel may not be phycocyanin.
- For mussel farming, continuous sensors can guide attention and sampling but cannot replace marine-biotoxin surveillance.
- Turbidity can reveal optical interference and support correction only after local validation.
- In freshwater systems, DO + CHL + CYANO is justified when separating wider algal and cyanobacterial trends will change a management decision.
- Placement, cleaning and representative samples remain essential.
- The most informative oxygen measurement may be the one taken when almost nobody is on site.
References
- Choo, F.; Zamyadi, A.; Newton, K.; Newcombe, G.; Bowling, L.; Stuetz, R.; Henderson, R.K. (2018). “Performance evaluation of in situ fluorometers for real-time cyanobacterial monitoring.” H2Open Journal, 1(1), 26–46. https://doi.org/10.2166/h2oj.2018.009.
- Eze, E.; Ajmal, T. (2020). “Dissolved Oxygen Forecasting in Aquaculture: A Hybrid Model Approach.” Applied Sciences, 10, 7079. https://doi.org/10.3390/app10207079.
- Foster, G.M.; Graham, J.L.; Bergamaschi, B.A.; Carpenter, K.D.; Downing, B.D.; Pellerin, B.A.; Rounds, S.A.; Saraceno, J.F. (2022). Field Techniques for the Determination of Algal Pigment Fluorescence in Environmental Waters—Principles and Guidelines for Instrument and Sensor Selection, Operation, Quality Assurance, and Data Reporting. U.S. Geological Survey Techniques and Methods, Book 1, Chapter D10. https://doi.org/10.3133/tm1D10.
- Fournier, C.; Fernandez-Fernandez, R.; Cirés, S.; López-Orozco, J.A.; Besada-Portas, E.; Quesada, A. (2024). “LSTM networks provide efficient cyanobacterial blooms forecasting even with incomplete spatio-temporal data.” Water Research, 267, 122553. https://doi.org/10.1016/j.watres.2024.122553.
- Johnston, B.D.; Finkelstein, K.M.; Gifford, S.R.; Stouder, M.D.; Nystrom, E.A.; Savoy, P.R.; Rosen, J.J.; Jennings, M.B. (2024). Evaluation of Sensors for Continuous Monitoring of Harmful Algal Blooms in the Finger Lakes Region, New York, 2019 and 2020. U.S. Geological Survey Scientific Investigations Report 2024-5010. https://doi.org/10.3133/sir20245010.
- Ma, L.; Moradinejad, S.; Guerra Maldonado, J.F.; Zamyadi, A.; Dorner, S.; Prévost, M. (2022). “Factors Affecting the Interpretation of Online Phycocyanin Fluorescence to Manage Cyanobacteria in Drinking Water Sources.” Water, 14, 3749. https://doi.org/10.3390/w14223749.
- McKergow, L. (2025). Algal Fluorescence Sensor Selection. High Frequency Water Quality Monitoring Guidance, NIWA Client Report 2025313HN. Official guidance PDF.
- Oppedal, F.; Dempster, T.; Stien, L.H. (2011). “Environmental drivers of Atlantic salmon behaviour in sea-cages: A review.” Aquaculture, 311(1–4), 1–18. https://doi.org/10.1016/j.aquaculture.2010.11.020.
- Shumway, S.E. (1990). “A review of the effects of algal blooms on shellfish and aquaculture.” Journal of the World Aquaculture Society, 21(2), 65–104. https://doi.org/10.1111/j.1749-7345.1990.tb00529.x.
- Shoko, A.P.; Limbu, S.M.; Mrosso, H.D.J.; Mgaya, Y.D. (2014). “A comparison of diurnal dynamics of water quality parameters in Nile tilapia monoculture and polyculture with African sharp tooth catfish in earthen ponds.” International Aquatic Research, 6, 56. https://doi.org/10.1007/s40071-014-0056-8.
- Xie, D.; Chen, X.; Qian, Y.; Feng, Y. (2025). “Prolonged Summer Daytime Dissolved Oxygen Recovery in a Eutrophic Lake: High-Frequency Monitoring Diel Evidence from Taihu Lake, China.” Water, 17(22), 3221. https://doi.org/10.3390/w17223221.