Why some values are missing

An annual rainfall total for a year where three months are missing is not a slightly uncertain annual total. It is a nine-month total wearing an annual label, and it will be the driest year in the record for reasons that have nothing to do with the weather. The same is true of a reference normal: the 1991–2020 average is only computed where the whole reference period is held, because a thirty-year normal computed from nineteen years is a different statistic with the same name.

A rolling mean follows the same rule. A ten-year mean is drawn only where all ten years are present, rather than averaging whatever the window happens to contain — which is how a chart comes to show a trend that is an artefact of its own edges.

A groundwater level is a height above a datum, and boreholes do not share a datum. Two levels that read 42 m and 61 m are not nineteen metres apart in any meaningful sense; they are two measurements against two different reference surfaces. So Hydrionis will not average them, will not rank them against each other, and will not draw them as one series. Each borehole is compared against its own history, which is a question the datum does not affect.

This is also why the 5 contextual signal pairings are drawn on separate panels rather than sharing a vertical axis.

“Percentage full” is the figure a drought is discussed in, and it is a division — stored volume over capacity. Capacity is published on several different bases: gross, usable, live, and the figure at a particular spillway or licence condition. Dividing a stored volume by the wrong one produces a percentage that is confidently wrong by the ratio between them, at exactly the moment the number matters most.

Where Hydrionis holds a storage volume and no compatible capacity basis, it publishes the volume and no percentage. A volume in cubic metres is less immediately readable and it is the number that was actually measured.

River flow becomes comparable with rainfall when it is divided by the land draining to the gauge, turning cubic metres per second into a depth in millimetres. The whole result rests on that denominator, which Hydrionis does not measure — it is the station publisher’s own delineation, and there are two of them: the topographic basin, and the contributing area that excludes land not delivering water to the gauge in an ordinary year.

Where a published area and a published flow imply a runoff depth no catchment produces, the two cannot both describe the same place, and no runoff is derived for that gauge at all — not the years that happen to fall under the ceiling. A series wrong by a constant factor throughout is worse than no series, because it no longer says so.

The largest daily mean flow in a year is not the largest flow in that year. A flood peak lasting four hours is averaged across twenty-four, and the published maximum is often much lower than the instantaneous peak on the same day. Hydrionis holds both where publishers provide both, keeps them as separate metrics, and records that a flood frequency analysis needs the instantaneous series rather than this one.

A maximum is also the statistic most damaged by absence: the largest day of a year the gauge did not record is simply not in the record. So an annual maximum is withheld where the year is not substantially covered, and the number of days actually held is published with every value that survives.

Every value carries how it came to exist — measured, reported, calculated, derived, aggregated, estimated, modelled or inferred — and the last three are not observations. They are shown with that basis attached, are excluded from statistics that claim to describe measurements, and are never presented as gauge readings. A projection is not held as a value at all: a figure for 2050 is a scenario, and a scenario in an observations table would be read as a record of 2050.

Where Hydrionis relates two series it refuses to produce a coefficient at all below 30 overlapping pairs of values, and refuses to offer the comparison at all where the two series overlap for fewer than 10 years. Hydrological series are seasonal and strongly autocorrelated, so a short window contains far fewer independent observations than it appears to, and any pattern in it is as likely to be the particular run of wet and dry years that happened to be sampled.

Both series also have their own seasonal cycle removed before correlating, and the result records that it was. Two seasonal series correlate because they are seasonal; rainfall peaks in winter and flow peaks in winter, and reporting that as a relationship without saying so is the commonest way this analysis misleads.

There is no machine-readable feed of official drought status, hosepipe restrictions or drought permits for England: the authoritative record is prose in a monthly PDF and press notices. Hydrionis will not derive its own declaration to fill that gap, because an unofficial declaration presented beside official indices is read as official.

What it will do is arithmetic over a published sequence, clearly labelled as such. Where the U.S. Drought Monitor — whose categories and weekly assessment are its own — held a state above a stated line for a stated run of weeks, Hydrionis marks that period and titles it after the index rather than the place. The rule is published on every one:

Several of the richest sources Hydrionis has assessed publish only as PDF, and extracting series from them was considered and refused. The numbers carry statutory weight, the layout can change without notice, and a series whose accuracy depends on a page template keeps its authority in a reader’s eyes long after it has quietly started returning something else. A wrong reservoir level during a drought is not a cosmetic error.

Those sources stay recorded, unheld, with the reason — which is what the coverage and gaps report is for. Each names what would have to exist for Hydrionis to hold it.

A gap on Hydrionis is a statement, not an oversight. If a value you need is absent, the honest answer is usually that no defensible version of it exists in the sources Hydrionis has access to — and the gaps report will normally say so explicitly and name what would close it. If you think a gap is closeable and Hydrionis has missed the source, that is a genuinely useful thing to send.

What Hydrionis does when a published value turns out to be wrong is on accuracy and corrections.