Monitoring fish populations: where accounting breaks and why
Manual bypass of a waterbody works like this: an inspector passes part of the cages, counts fish in a sample and transfers the result to the whole volume.
Already here a counting error is embedded — on 5–10% of the area where fish are distributed unevenly, sampling error doubles when extrapolated to biomass. Then human factor is added: different bypass times, different lighting, fatigue by the end of the shift.
The second source of distortion is blurred cage boundaries. Fish accumulate at feeders and at the net, some silhouettes are lost in turbidity and glare, some double in reflection.
Fish accounting turns into visual estimation, and the discrepancy between two inspectors on the same cage reaches tens of percent.
Fry losses are tied to the window between rounds. Mortality, escape through net damage and crowding are recorded days later, when noticeably less remains from the batch, and the cause can no longer be reconstructed.
Fry losses at early stages break stock control most: young are the most sensitive part, and their loss is visible only in final survival.
In total this distorts all derived figures: stocking density, feed conversion ratio, harvest forecast, survival reporting. Decisions are made on data that lag reality by days and differ from it by times.
Below we analyze how a continuous stream of observations from cameras and sonars removes these failure points in fish population monitoring.
What automatic fish counting provides: measurable effect
An industrial cage or river section is thousands of individuals that cannot be watched by eye. A manual round gives a snapshot at a specific moment: the observer sees only what falls into the field of view, and the rest remains unknown.
FAO in fish stock assessment methodologies explicitly notes the wide error of sample observations — this is where the main part of losses comes from. Below — what exactly automatic fish counting changes and by what metrics this is measured.
- Continuity of observation. Cameras above the cage and hydroacoustic sensors operate 24/7, so the picture does not depend on staff shift, fog or waves. The night period, which previously dropped out of accounting entirely, becomes as measurable as daytime.
- Reduction of manual labor. Counting individuals, splitting by size classes and reviewing recordings move to automatic mode. Instead of watching hours of video, the operator gets a ready summary and deals with decisions, not observation.
- Early detection of fish kill. The system records a drop in motor activity, shift of fish by depth and surfacing of single individuals. The signal comes hours before losses become visually noticeable — this is enough for aeration and feeding correction.
- Stock survival. Fish escape through net damage and losses from predators cease to be invisible: the discrepancy between accounted and actual biomass is visible by dynamics, not by inventory results.
- Biomass reporting. Biomass estimate, growth and mortality form regular reports with the same calculation methodology. They can be compared between cages and seasons, rather than manually consolidated from heterogeneous records.
We analyze what of this can be really measured at your site already at the first stage — with control points where the effect is visible without a long wait.
Comparison of approaches: underwater cameras, sonars, hybrid schemes
The choice of sensor in fish population monitoring is determined by three physical limitations of the environment: water transparency, illumination and stock density. Underwater cameras provide species identification and size range of individuals, but rely on visible light.
Hydroacoustic sonars account for aggregations in turbid water and complete darkness, but see targets as marks — without species and size class.
It is correct to compare approaches not by "image quality", but by shooting mode and target metric: general biomass accounting, size range or species confirmation. Below — working alternatives and their limitations.
| Option | When suitable | Limitations |
|---|---|---|
| Underwater cameras | Clear water, daylight or soft illumination, sparse stock | Turbid water and suspended matter suppress contrast; at night illumination is needed, to which some species react by leaving; in dense stock individuals overlap each other |
| Hydroacoustic sonars | Turbid water, night observation, large aggregations, biomass assessment | No species identification: close species appear as one mark; dense stock reads as a single cloud, not as individual fish |
| Hybrid scheme | Changing conditions by season and day, need both biomass and species confirmation | More expensive in installation and maintenance; requires synchronization of streams and a unified data model |
A hybrid scheme is not "two devices instead of one", but a separation of roles: sonar performs continuous accounting, camera connects for species verification and size refinement.
The circuit is resistant to turbid water and night, while cameras of Teledyne level do not operate in continuous mode and wear out faster.
Implementation stages: from waterbody survey to system handover
Each waterbody differs by transparency, current, depth and daily fish behavior — a standard project without site-specific adaptation gives systematic error in counts.
We go from field survey to handover of a working system so the customer receives a configured monitoring circuit, not a set of equipment.
- Waterbody survey. Depth sounding, assessment of transparency and water background, current speed, daily fish movement at spawning and feeding areas. Based on these data we decide where the acoustic channel is stable by depth, and optics — by species and size.
- Placement scheme. We choose installation points for cameras and sonars so as to cover the water area without dead zones and double counting of one fish in neighboring viewing sectors.
- Installation and connection. Mounts for current and waves, protection of optics from fouling, power and communication channel, time synchronization on all sensors — without it trajectories do not merge into tracks.
- Model calibration. We label frames with local species and size classes, fine-tune recognition on this sample, configure trigger thresholds so ripples, debris and glare do not enter the count.
- Pilot launch. The system operates in real mode in parallel with control fishing; discrepancies are analyzed by causes and counting accuracy is brought to the agreed level.
- System handover. We provide accesses, monitoring panel, maintenance regulations, instructions for observer shift and train the team. Support — according to the agreed schedule.
Set of documentation and artifacts you retain
A fish population monitoring system is maintained by a different team than the one that assembled it: operators change, shooting points shift by season, the model is fine-tuned on new data.
Therefore together with the working circuit we hand over a set by which the system can be maintained independently.
This is not a report for the shelf, but working artifacts: they are consulted when cameras are relocated, a false trigger is analyzed or a shift is accepted.
- System passport — composition of nodes, binding of cameras and sonars to observation points, current model and limits of its applicability.
- Placement schemes — water area plans with viewing angles, overlap zones and uncovered areas, so relocation does not open a hole in observation.
- Access policies — roles of operator, analyst and administrator, rights to archive, data export and change of trigger thresholds.
- Maintenance regulations — frequency of optics cleaning, communication line checks and sonar calibration, actions when signal from a node is lost.
- Operator instructions — how to read events, separate false triggers from real ones, edit labeling and when to involve an engineer.
- Configurations — settings files for cameras, sonars and archive storage rules, with a note on what is adapted to the waterbody and what must not be changed.
The set lies next to configurations in the common directory structure and is updated together with the system: after equipment relocation or model retraining you retain the current version, not a folder of outdated files.
Case: reduced accounting discrepancy at a fish farm
Starting point
A fish farm with cage farming of trout kept records by three independent sources: control fishings, feed conversion ratio and transfer acts.
By the end of the cycle, calculated biomass diverged from actual at sample fishing by 18–22%, so feed, stocking and sales schedule were planned with a "by eye" margin.
An overestimated estimate resulted in feed overconsumption, an underestimated one — underloading of cages and idle volumes.
What was changed
Hydroacoustic sonars Simrad at two horizons and underwater cameras with night illumination on feeding lines were installed in each cage.
The model counts not individual fish, but the density of echo traces and converts the response into size classes, accumulating them into the current biomass estimate.
Cameras work as independent verification: by the approach of fish to the feeder one can see where hydroacoustics diverged from reality, and the conversion coefficient is corrected based on fact, not once per season.
Stock dynamics became daily: growth and mortality for each cage, relation to feed supply. Previously these figures appeared once a month and could no longer influence decisions — only record consequences.
By control fishing, the discrepancy between calculated and actual biomass remains within 4–6% versus previous 18–22%.
Fishing by a team of three people from a boat took about six hours; now the operator spends 20–30 minutes per day on checking the summary and labeling suspicious areas. Freed time went to sorting and cage maintenance.
What to do if cameras are already installed but monitoring is not performed
Cameras are already installed — so there is power, cable routes, mounting points and, as a rule, a video recorder with archive. This is a large part of the circuit that does not need to be built anew.
The question is not what to purchase, but which of the existing streams is suitable for counting, and which will need refinement.
Equipment check takes one visit and divides everything into three groups. Suitable as is: stable stream, working view, distinguishable fish at the point of interest. Suitable after refinement: shift angle, replace lens, clean optics, add illumination.
To be replaced: at night the frame is unreadable, the point does not cover the area, the stream breaks.
| Option | When suitable | Limitations |
|---|---|---|
| Keep cameras, add counting module | View and stream stable, coverage zones match task | Accuracy limited by quality of existing optics |
| Modernize part of circuit | Part of cameras gives suitable frame | Reconfiguration of angles and recalibration |
| Supplement cameras with sonars | Turbid water, night, large depths | Separate binding to circuit required |
| Replace circuit entirely | Equipment outdated, coverage does not match task | Full installation and setup from scratch |
Integration with an existing circuit is built so as not to break what works. The recorder continues writing the archive, while the stream goes in parallel to the recognition module; if communication breaks, frames are buffered, so the gap does not turn into a data gap.
Where water is turbid or shooting is in darkness, cameras are supplemented with sonars — they provide size and depth that optics lacks.
Common technical doubts before starting a project
Engineering doubts before start usually concern not the idea of monitoring itself, but its behavior in a real waterbody: turbid water after rain, night fish passages, camera maintenance in winter. Below — answers to the most frequently asked questions.
Let us immediately define the framework: counting accuracy is not a constant, but a measurable value. We fix it at the pilot site by control surveys and consolidate it in acceptance criteria, so the result does not depend on weather outside.
How accurate is counting in turbid water and after flood?
Optics lose contrast with suspended matter, but hydroacoustics does not depend on transparency. Therefore the channels work in pair: in turbid water the weight shifts to sonar, and species classification is refined by cameras in areas with transparency. The final error by size classes is fixed in advance — by it we accept the work.
Does the system work at night?
Yes, and at night most passages are usually recorded. We use invisible-range illumination and hydroacoustics, which do not affect fish behavior. Frames and echograms are stitched into a single event series, so the daily schedule is not interrupted.
How is the system maintained on water?
Regulations are simple: cleaning optics from fouling, checking mounts and tightness, testing communication channels. Frequency depends on the waterbody — somewhere once a month, somewhere less often. The system itself signals a drop in frame quality, so the visit is based on fact, not calendar.
How is model fine-tuning on our waterbody carried out?
Initial setup is done on your recordings, disputed fragments are labeled together with your ichthyologists. Then the model is fine-tuned by season: feeding movement or water level changed — we update weights. Each model version is saved, rollback to the previous takes minutes.
Who owns the data and how is it protected?
Data remains yours and is stored in your circuit or in isolated storage with encryption. Access is distributed by roles, all actions are logged. Export is in open formats linked to your equipment.
Let's discuss your task and come back with an estimate
To discuss the task substantively and not collect inputs in circles, one message with source data is enough. Below — what is needed for calculation: this is enough to determine the composition of cameras and sonars, installation scheme and scope of integration work.
- map of the waterbody or site: area, depths, relief changes, spawning and feeding zones;
- shooting conditions: water transparency, seasonal turbidity, ice period, illumination by hours;
- installation points for cameras and sonars, access to them and mounting method — pier, platform, pontoon, vessel;
- power and communication channel at the site: wired line, radio channel, autonomous sources;
- reporting requirements: which indicators are needed — abundance, size range, distribution by depths, dynamics by months;
- existing systems and observation archives, if accounting is already performed, with data storage formats;
- access regime to the site and approval procedure for installation and commissioning works.
In response you will receive a project estimate with breakdown by stages, implementation plan and proposal for a pilot at the waterbody: deploy equipment on a limited site, verify automatic accounting against control fishings and manual observations, and only after results converge scale to the entire water area.
