The sensor you should not buy
A sensor proposal usually arrives with a specification and a price. Accuracy, drift, sampling interval, installation cost. What it almost never arrives with is the number that decides the question: how much better would the decision be, if you had it.
That number is computable, and it is not a matter of opinion. Take the model, take the decision you make on the basis of it, and compare two worlds. In one, you decide as you do now. In the other, the measurement exists, and you decide with it. Run both across the range of situations that actually occur, and the difference in outcome is what the measurement is worth per season.
The results are unintuitive often enough to be worth doing every time.
A measurement can be highly accurate and worth nothing, because the quantity it reports does not change what you would do. If you would feed the same either way, a better number is a better number and no more.
A measurement can be crude and worth a lot, because it arrives early. Half the value of a detection is in its timing, not its precision, and a rough signal two weeks earlier routinely beats a certain one two weeks later. This shows up sharply in disease work, where the cost of waiting compounds and the cost of a false alarm does not.
And a measurement can be worth a lot in one part of the season and nothing in the rest. Sensitivity is not constant in time. A parameter that dominates during a heat week can be irrelevant in spring, which means the right answer is often a hired sensor for six weeks rather than an installed one for ten years.
None of this is an argument against measuring. It is an argument for deciding what to measure with the same model you will use to act on it, instead of buying the instrument first and looking for the question afterwards.