How to turn a forecast into an operational decision

The loss nobody blamed on the rain

One winter night it rains hard over a shrimp farm on the coast. The next morning everything looks fine: the ponds are full, the shrimp are eating a little less than usual, nothing worth a line in the logbook. Two or three weeks later the mortality starts: soft shells, cannibalism during the molt, weak animals that infection finishes off. On the farm they check the feed, they suspect the postlarvae, they blame luck. Almost nobody looks as far back as that rain.

The mechanism, though, is well understood. Rainwater is less dense than pond water and sits on top of it like a lid; the sun warms less of the column and dissolved oxygen falls; with less calcium and magnesium in the water, the shell takes longer to harden after each molt, and a soft shrimp is a vulnerable shrimp. Every piece of the damage arrives late. By the time the loss is finally visible, the rain is already a memory, which is why it never even enters the diagnosis.

The story is a shrimp-farming one; the structure is not. In almost any operation exposed to the weather there are losses booked as "bad luck" or "a slow month" because they arrive separated from their cause. The impact of that rain was never in the millimeters: it was in the operation that took it unprepared.

The bridge: decision, threshold, action

This is where the forecast comes in, and where it usually comes in wrong. A forecast on its own is information: it says what the weather may do. It becomes useful the day it is connected to a decision that already exists in your operation and to a threshold that fires it. On the farm in the story the decision had been there all along: faced with heavy rain you can drain the surface water, test the aerators and switch them on, cut the feed ration, keep lime ready. Capacity to act was not the problem; what was missing was the trigger that connects "it is going to rain hard" with "this afternoon we get the ponds ready."

There is a third element, less visible than the decision and the threshold: lead time. Draining surface water and getting aeration running take hours, so the decision only helps if the forecast arrives with those hours to spare. Every action in an operation has its own preparation time, and that time settles which forecast is useful to it and which one arrives too late however good it is. The subject has enough body of its own to deserve a separate guide, and it will get one; for now it is enough to note it alongside the decision and the threshold.

The trigger, then, has a precise shape, and it is worth putting in writing.

The rule that fits on one line

Take illustrative numbers. Preparing the farm ahead of heavy rain (a few hours of pumping and aeration, a couple of day wages, sacks of lime) costs, say, 300 dollars, whether it rains or not. Not preparing it, if the rain comes and the pond was loaded, can cost 10,000 dollars in shrimp lost weeks later. Three hundred over ten thousand: three percent.

The rule fits on one line: it pays to act when the probability of the event exceeds the cost of acting divided by the loss of not acting. In this illustration, a forecast giving barely a 5% chance of heavy rain is already reason enough to get the ponds ready. When the action is cheap and the loss is large, waiting for certainty is the most expensive decision on the table.

Notice which part the forecast plays here and which part it does not. The forecast supplies the probability; the threshold comes out of your own costs and your own losses, numbers your company already has even if it has never divided one by the other. The threshold is not set by the meteorologist: it is set by your operation. Which is why two neighboring companies, looking at the same sky and the same forecast, can decide different things and both be right.

The same structure, in your sector

Cheap action against delayed loss is not the property of one sector. Lowering the greenhouse curtain for a night costs a day's wage; the botrytis that comes in with the humidity charges you export quality weeks later. Postponing a spray because of wind costs a day on the calendar; an application the wind carries off leaves the pest cycle running free. Rescheduling a maintenance crew costs one awkward phone call; a crew standing in the field, unable to work, costs everybody the whole day.

In each case the exercise is the same: name the decision, put a cost on acting too often and on acting too late, and divide. The result almost always surprises whoever does it for the first time.

Acting for nothing is part of the design

One consequence is left, and it should be said to your face, because it is the one that stings. With a 5% threshold you are going to prepare the farm many times over and then watch no rain come. You will lower curtains on nights that end up mild and reschedule crews for mornings that dawn clear. Case by case it looks like waste; seen across the year it is exactly what the arithmetic asked for: many cheap actions in exchange for never eating the big loss.

A false alarm, with a threshold well set, is not a forecast that failed. It is a well-designed decision doing its job. The day your operation can say that sentence without anyone raising an eyebrow, the forecast has stopped being information and become what it should have been all along: decision support.