A provincial risk score is an average over an area that can hold a reservoir, a rain shadow, an irrigated plain and a rain-fed slope at once. The number is usually correct. It can still be useless for the field you actually farm, and the gap between those two facts is where most disappointment with risk maps comes from.
What a province is, as a unit of measurement
We publish at province level because that is the unit administrations plan in, budget in and declare emergencies in: Thailand has 77, Vietnam 63, Malaysia 16. Those boundaries were drawn for governing, not for water. A catchment crosses them, a rain shadow ignores them, and a canal network ends somewhere inside one.
So a province score answers a question with a real owner — how does this administrative area compare with its own history — and it is the right answer to that question. What it cannot do is tell two neighbours on opposite sides of a ridge that their seasons will differ.
The average is not the experience
Take any province with irrigation on one side and rain-fed land on the other. In a dry season the irrigated half stays green and the rain-fed half does not. The provincial mean lands between them, describing a place where nobody lives, and both halves read the map as wrong: one because nothing happened, the other because far more happened than the colour suggested.
This is not noise to be averaged away with a better statistic. It is the actual distribution, and the moment you need it is the moment the map is most confident.
Why a finer grid is not automatically the answer
The obvious fix is smaller cells, and it helps only where the underlying data has something to say at that size. A satellite index computed on a coarse grid and then cut into district shapes looks more precise and carries no more information; it has moved the boundary, not the measurement.
There is a second cost that is easy to miss. A finer map invites a decision at a scale nobody validated — a district colour read as a statement about one farm, when the thing that was checked was the province.
So the honest sequence is: ask what resolution the input actually resolves, publish at that, and say what it is. Resolution you cannot defend is a claim, not a service.
What to ask instead
If a provincial score matters to a decision, the useful follow-up is not "can you go finer" but "what is the spread inside this province, and which side of it am I on". That question can be answered from the same data without pretending to a resolution nobody has.
In practice: ask whether the province contains irrigation, and whether your plot is served by it. Ask where the rain actually fell, not what the monthly total was. And when you escalate, say which part of the province you mean — a score with no location inside the polygon is a number waiting to be misread.
None of that requires new instruments. It requires treating the province as the coarse first cut it is, and not as a description of a place.
What we are not saying
This is not an argument that provincial data is wrong or that we plan to stop publishing it. It is the unit the decisions we serve are made in, and a coarse answer that states its own scale is more useful than a fine one that cannot.
What we will not do is let a polygon's colour stand in for a place. If a score is going to carry a decision about one location, someone has to say how the location relates to the polygon — and that sentence is part of the answer, not a caveat underneath it.