The operation

A land-based salmon producer operating a recirculating aquaculture system (RAS) in Norway. Dense sensor coverage across water chemistry, oxygen, CO2, pH, temperature, and flow metrics. Existing SCADA-based threshold monitoring in place.

The challenge

RAS operations are high-consequence environments. Water quality deviations that develop over hours can result in batch mortality events with significant economic impact. The existing monitoring system was threshold-based: it would fire an alarm when a parameter crossed a configured limit, by which point the deviation was already advanced and the intervention window was narrowing.

The operation had no visibility into the early development of water chemistry deviations — the period in which the situation is most recoverable. The data existed in the historian. There was no system reading it for anomalous patterns.

What Eyer found

Eyer was run on the operation's historical time-series data as a Fast Forward analysis. The analysis identified a recurring water chemistry anomaly pattern in the historical data that had preceded operational impact events. The pattern was detectable two to three hours before any operational indicator became visible.

Critically, no threshold alarm fired during the two-to-three-hour window. The individual metrics were deviating from their learned baselines — their normal behaviour under normal operating conditions — but none had crossed the limits that the existing system was configured to watch for. The deviation was in the fingerprint, not in the thresholds.

The correlation structure

The water chemistry anomaly was not a single-metric event. Eyer's correlation analysis identified a consistent pattern: a primary metric deviation, followed by correlated shifts in two related parameters, in a sequence that appeared consistently in the historical data before impact events. The correlation was not visible from the individual alarm events — those were logged as separate alarms. It was visible in the time-series correlation model that Eyer builds from the historical data.

The business case

In a land-based aquaculture operation, a two-to-three-hour early detection window on a water quality event is the difference between an operator intervention that prevents a batch mortality event and a response that manages the aftermath of one. The economic consequence of a mortality event in a dense RAS operation significantly exceeds the cost of the Fast Forward analysis and ongoing Eyer deployment.

The operation now runs Eyer continuously, with the operational fingerprint updated as the system learns new normal behaviour patterns.

Similar results in offshore aquaculture

A comparable early detection window was observed when Eyer was applied to water quality data from an offshore aquaculture operation. Water parameter deviations affecting fish health surfaced hours before any indicator became visible at the surface. See: Early warning for fish health and water quality in offshore aquaculture.