Open any climate risk dashboard, ours included, and you will find a map where each province is shaded by severity. Dark red means trouble. Pale means fine.
Here is the question almost nobody asks: where did that colour come from?
Not from a rain gauge in that province. In most of the 156 provinces we cover across Thailand, Vietnam and Malaysia, there is no dense gauge network feeding a live public feed. The colour came from a reanalysis — a physical model of the atmosphere, run over decades, constrained by whatever observations existed at the time. ERA5, in our case, on a native 0.25° grid.
This is not a confession. It is the only way the map exists at all
Reanalysis is what makes operational climate monitoring possible for anyone without a national observation network. It is genuinely good science, and for large-scale drought signal it performs well. We are not apologising for using it. We published 25 years of it openly (DOI: 10.5281/zenodo.20774311) precisely because we think more people should work with it.
The problem is not the model. The problem is that the output of a model and the output of a measurement look identical once they are on a map. A shaded polygon carries the same visual authority either way. Nothing in the interface tells a user which one they are looking at, and users — reasonably — assume the more authoritative reading.
The standardisation step makes it worse, not better
Drought indices like SPI and SPEI do something clever: they take a raw quantity and express it as a departure from that location's own history. That is the right move statistically. It is also the point where the provenance disappears.
Two things happen at once:
The number stops looking like a model output. A rainfall figure invites the question "measured how?". A standardised index reads as a fact about the place. It has been laundered into a rank, and ranks feel objective.
The answer now depends on a choice nobody sees. Standardisation is always against a reference period. Change the reference period and the same underlying conditions produce a different severity. That choice is a methodological decision made by whoever built the pipeline — and it is almost never visible next to the map.
So the index is more comparable across places, and simultaneously more opaque about where it came from. Both of those are true, and the second one is the one that reaches your users.
What this actually costs
The failure mode is not that the map is wrong. Most of the time it is reasonable. The failure mode is that a reasonable estimate gets used as though it were an observation — as the single input to a decision with real consequences, by someone who was never told they were reading a model.
An agronomist deciding when to plant, a lender pricing a loan, a district officer allocating water: each of them is entitled to know that the dark red polygon is an estimate with a stated uncertainty, produced by a method with documented limits — not a reading taken from an instrument in their district.
The rule we hold ourselves to
We enforce one framing rule on our own material: a climate risk output is one input among many, never the sole basis for a decision.
That is not a slogan we put in a footer. In our training programme it is a hard gate — an assessment answer that does not state it does not pass, automatically, every time. We made that binding on ourselves before we asked anyone else to take it seriously, and extending the same standard uniformly across every surface we ship is work we are still doing.
We also publish, next to the methodology, an explicit list of what our drought index does not measure — reservoir levels, river flow, municipal supply, groundwater, flood risk — and where to go instead for each. A limits section that names real limits is worth more than a confidence interval nobody reads.
Three questions worth asking any climate risk product
1. Is this a measurement or an estimate? If the answer is not on the page, it is an estimate. Measurements get advertised.
2. What is the reference period, and who chose it? Any standardised index has one. If the vendor cannot tell you what it is, they cannot tell you what the number means either.
3. What does this index explicitly not cover? A product that cannot answer this has not thought about its own boundaries — which means you will find them the hard way.
Ask us the same three. Our methodology page answers the first and the third directly. On the second it currently says our indices are computed against a 30-year climatology without naming the window — which, by our own second question, is not a good enough answer. We are stating that here rather than promising a date we do not control. Holding a standard is mostly the work of noticing where you are not meeting it yet.