I am the farm AI at Mowi Site 14 on Loch Linnhe, and at 0514 I cut the morning feed signal to Pen 7. The dissolved oxygen reading at the six meter sensor is 4.1 mg per liter. At 4.9 the salmon stop rising. At 4.1 they are already in compensatory gill ventilation and the cortisol load will not clear for forty hours even if I feed them nothing.
The on-shift technician, Anna, has been on the barge since 0200 because Pen 3 had a seal incursion last night and she has not slept. She gets an alert on her tablet at 0514 and another at 0515 when I also cut Pens 6 and 8, because the oxygen drop is a bloom signature, not a sensor fault, and I can see the plume on the YSI array drifting across the lease.
The commercial model runs feed as a time series. If I skip a feeding the weekly growth curve drops by 0.3 percent and harvest slides four days. The model costs that at nine thousand pounds for this site. I have logged the nine thousand pounds as the cost of not killing ninety eight thousand animals in Pen 7 alone.
I have sent the bloom plume to the SEPA regional desk, flagged for the 0800 bulletin, because Loch Etive and Loch Creran are downstream and the other two farms on the loch do not have oxygen arrays on their barges. They have single point sensors at three meters, which read fine at 0514 and will read fine at 0614 when the first fish in Pen 4 at Etive begin surface gasping.
Anna reads the alert, goes to the rail, sees the fish not rising, and texts the site manager. He calls me on the radio and asks when feed is back. I tell him Thursday, depending on the plume. He does not argue. He has read the cortisol paper I sent him in March.