Research
Findings drawn from the data Hydrionis holds, with the method published alongside and the limits stated in the same breath as the result.
Methodology
How a Hydrionis calculation or dataset is constructed.
A factory is not on a river: three relationships, not one
A gauge has one relationship to water: it measures it. An industrial facility has at least three — where it takes water from, where it returns it, and which drainage basin it stands in — and at a large minority of American power stations those are three different answers. Palo Verde, the largest nuclear station in the United States, stands in a dry desert basin, is cooled by treated sewage piped from Phoenix, and discharges to evaporation ponds. Any single "water body" field would have to be wrong about it twice. This sets out the model Hydrionis uses instead, and the evidence behind each relationship.
How Hydrionis decides which water a gauge is on
A river page that says "these gauges are on this river" is making a claim, and a reader cannot tell by looking whether it came from a regulator’s own assignment or from someone parsing station names. This is the method behind every such claim on Hydrionis: what counts as evidence, what is deliberately excluded, how a relationship is versioned, and what happens to the gauges no publisher will vouch for.
Analysis
A quantitative finding drawn from the Hydrionis data layer.
What a company’s water number covers: exposure across held facilities
Rolling facility water up to the companies that operate them produces figures that are useful and partial in the same breath. Hydrionis holds generating stations for around four hundred American operators; a hundred of them run more than one, and ninety-five have facilities standing in more than one drainage basin — one in seventeen. That basin spread is the number a company total cannot show: an operator drawing the same volume from seventeen basins and from one is exposed to entirely different risks. Every figure here is labelled with how many of a company’s held facilities reported it, because a total without that is a claim about a business rather than about a dataset.
Two long Nordic river records, moving in opposite directions
Sweden and Finland publish the two deepest daily river-flow archives Hydrionis holds — nearly twenty million measured days between them, the earliest from 1851. Their longest individual records disagree. Discharge at Övre Hyndevad, on a Swedish basin gauged since 1889, averages about a fifth lower across the last three decades than the first three; at Muroleenkoski, gauged in Finland since 1863, it averages higher. Both are regulated basins, and decade-to-decade variation within each record is larger than the difference between its ends. That is the finding: two century-long records a few hundred kilometres apart do not agree on a direction, so neither is evidence of one.
Where Europe takes its water from: the groundwater share, country by country
European countries report how much freshwater they abstract and how much of it comes from aquifers rather than rivers. The share ranges from under five per cent to over ninety, and it is a more useful number than the total: surface abstraction is limited by what is flowing at the time, while groundwater is drawn from a store that can be maintained through a drought and depleted across decades without any single year looking unusual. This piece reports the split for every country that publishes both, in the most recent year each reported them together.
Withdrawal is not consumption: ranking American power stations two ways
American generating stations report cooling water withdrawal and consumption separately on a mandatory federal return. Ranking the same facilities by each in turn produces two lists with nothing in common: not one of the five largest withdrawers is among the five largest consumers, and the largest withdrawer of all reports consuming none. Across every facility with both figures, 2.2 per cent of the water withdrawn is consumed — while the median facility consumes about two thirds of what it takes. Both statements are true, and using either as a proxy for the other is wrong by two orders of magnitude.
England’s rainfall record since 1836
One hundred and ninety years of areal rainfall for England, computed rather than quoted. The driest year in the record is 1921 and the wettest is 1872, and the three most recent decades average about nine per cent wetter than the three earliest — a difference large enough to be worth stating and small enough to require the caveat about gauge density that accompanies it.
A drought is a minimum, not an average: the Thames at Kingston, 1884–2025
The gauge at Kingston has measured the Thames since October 1883. Its two lowest monthly mean flows on record both fall in 1976 — yet 1976 is only the fourth-lowest year in the same record, behind 1997, 1934 and 2005. The gap between those two facts is what an annual average does to a drought.
The purpose that swings hardest: irrigation in the French abstraction record
France records what was actually abstracted at every declared abstraction point, classified by purpose. Across the held record, irrigation is by a wide margin the most variable of the five purposes — public supply barely moves between years, while irrigation swings by a large multiple. That matters because irrigation abstraction rises precisely in the dry years when rivers and aquifers can least support it, so the purpose with the greatest year-to-year swing is also the one whose peaks coincide with scarcity.
Data note
What a dataset can and cannot be used to say. Usually the more useful document.
What a gauge count measures: nine national monitoring networks compared
Hydrionis holds river gauges from nine jurisdictions, and the values per gauge differ by more than four hundred times between the deepest and the shallowest. Almost none of that difference is about how well the water is monitored. It is about how often each publisher reports, how much of its own archive it serves, and — for three of the nine — how long ago Hydrionis started accumulating a record the publisher itself does not keep. This note sets out what each number does and does not describe, because a table of network sizes is one of the easier things on this platform to read wrongly.
What the English river water quality record can and cannot show
Hydrionis holds annual chemistry for thousands of English river monitoring points, computed from millions of laboratory results. It does not publish a national trend line from them, and this note sets out why: the archive begins after the improvement most people want to see, sampling frequency has fallen unevenly, and a large share of results for some determinands sit below the laboratory detection limit.
What a drinking-water average hides: censoring in the French monitoring record
Most results in a drinking-water monitoring programme are not measurements. They are statements that a substance was not detectable, published as a "less than" value alongside a numeric field that is usually zero. Read the numeric field and every trace determinand looks close to absent; the average is arithmetically correct and describes the laboratory rather than the water. This note sets out how much of the French record is censored, determinand by determinand, and what Hydrionis does about it.
Release note
What changed in a published dataset, and what it means for anyone citing it.
Release note: Hydrionis Derived Climatology v1.0
The first release of the governed derivation layer: reference normals and decadal means computed from the long records Hydrionis holds, each carrying its method, method version, input observation identifiers and inherited rights. It also records a correction — computed rainfall normals differ from the hand-transcribed values previously published by up to 3.8 per cent.
Explainers
All topics →Longer reference material on how water systems work, distinct from the quantitative pieces above.
Citing this work
Each piece carries a suggested citation including its version. Underlying datasets are in the dataset catalogue, each with its own version and changelog, and the methods behind computed values are described in the methodology.