Release note
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.
What was released
Version 1.0 of the Hydrionis derivation layer: reference normals over the WMO 1991–2020 period, and decadal means over every sufficiently complete long record the platform holds.
| Family | Values produced | Refused |
|---|---|---|
| Reference normals (1991–2020) | 34 | 0 |
| Decadal means | 59,430 | 0 |
A correction
Methods in this release
| Method | Version | What it must not be read as saying |
|---|---|---|
| Mean over a period | 1.0.0 | A mean over a fixed reference period, not a measurement of any single period within it. Requires the whole span: a mean of a partial span is a different statistic wearing the same name. |
| Anomaly against a reference normal | 1.0.0 | Compares a subject with its own history, never with another subject. A wetter-than-normal desert is still a desert. Meaningless unless a full reference-period normal exists for the same subject and metric. |
| Centred rolling mean | 1.0.0 | A smoothed line is not the data. Values near the ends of a series have no complete window and are not produced rather than being computed from a partial one, because a trend drawn from a half-window is an artefact of the edge. |
| Change between two periods | 1.0.0 | Two endpoints are not a trend. A change computed between two years says nothing about the path between them, and is highly sensitive to whether either endpoint was unusual. Not valid where the series has a documented break. |
| Record extremes of a series | 1.1.0 | An extreme within the record Hydrionis holds, which is not necessarily an extreme within the record that exists. Says nothing about periods before the series begins. Where the record crosses a documented change in how it was measured or reported, that change is stated alongside, and ranking statistics are withheld where the publisher marks the change as one that cannot be compared across. |
| Percentile rank within a series | 1.0.0 | A rank against this subject’s own history only. The percentile shifts as the record lengthens, so it is stated together with the span it was computed over. |
| Seasonal percentile rank | 1.0.0 | Restricted to the same calendar month on purpose, and not interchangeable with a rank against the whole record. A chalk borehole sits near its annual minimum every August, so an August level ranked against all months reads as a drought every summer; ranked against other Augusts it says whether this one is unusual. The rank is against this subject’s own history and says nothing about anywhere else, it shifts as the record lengthens, and it is void across a methodological break marked not comparable — a rank spanning two measurement programmes ranks whichever of them ran hotter. It is a position in a distribution, not a measure of severity: a low rank in a short record is a weaker statement than the same rank in a long one, which is why the number of prior periods compared against is stated with every value. |
| Seasonal snowpack statistic | 1.0.0 | Monthly resolution throughout, because the record is monthly: a peak is the fullest month rather than the fullest day, and a melt a fortnight early is invisible. The water year runs October to September and is named for the year it ends, so a value is a statement about a snow season rather than a calendar year. Published only where the year holds at least nine of its twelve months and every month from February to July, and withheld where the peak falls in the first month held — a maximum at the edge of a record may not be the year’s maximum. Melt-out is withheld where the pack never falls that far inside the year, where a month is missing between the peak and the crossing, or where the peak is zero, because in each case the timing is absent from the record rather than late. A tenth of that year’s own peak is a threshold chosen so a thin year is measured against its own snowpack rather than against a deep one, and it is a convention rather than a physical boundary. A pillow is a point on a mountain: it is not the catchment average, and wind scour or drifting at the instrument is carried into every statistic derived from it. |
| Monthly reference normal | 1.0.0 | Twelve separate statistics, not one: a January normal describes Januaries and says nothing about July. Valid only where the reference period is substantially complete for that month, and only for quantities with a genuine annual cycle. It describes the reference period, not the present — a normal computed over a period that was itself unusual carries that through to every anomaly measured against it. |
| Anomaly against the same month’s normal | 1.0.0 | Compares a month with the same month in the subject’s own reference period, never with another subject and never with a different month. For a managed quantity such as reservoir storage, a negative anomaly may record an operating decision — a flood-control drawdown, a scheduled release — rather than a shortage, and this statistic cannot distinguish the two. |
| Ratio of two metrics | 1.0.0 | Only valid where both inputs cover the same subject and the same period, and where the two metrics are actually commensurable. A ratio of figures collected on different bases is arithmetic, not information. |
| Absolute anomaly against a baseline | 1.0.0 | A difference from this subject’s own baseline, in its own unit. It says how unusual a period was for this place, never how this place compares with another. Requires the baseline to be at least 80% populated, because a normal computed from the years that happened to report describes those years rather than the period. |
| Standardised anomaly (z-score) against a baseline | 1.0.0 | Assumes the baseline distribution is roughly symmetric. Daily rainfall is bounded at zero and heavily skewed, so a z-score of it overstates wet extremes and compresses dry ones; percentiles are the honest choice for skewed quantities. Refused where the baseline has no spread at all. |
| Percentile within a baseline distribution | 1.0.0 | A rank within this subject’s own baseline. A value above every baseline value returns the top percentile and says nothing about how far above it lies, so percentiles compress exactly the extremes that matter most in flood and drought. |
| Indexed to a baseline mean | 1.0.0 | Makes shape comparable between signals and magnitude comparable between none of them: an index of 150 hides whether the underlying quantity rose by a millimetre or a metre. Refused where the baseline mean is at or near zero, because a percentage of nearly nothing is arbitrarily large. |
| Lagged association between two signals | 1.0.0 | An association, not a mechanism and never a cause. Both series are usually strongly seasonal, and two seasonal series correlate whether or not either affects the other, so a correlation computed on raw values mostly measures the shared calendar. The lag at which r peaks is a property of these two records over this period, not a physical travel time. Refused below 30 shared periods. |
| Fish density from a depletion population estimate | 1.0.0 | Both inputs are the publisher’s. The estimate comes from successive electrofishing runs over the same reach, where the fall in catch between runs is used to infer how many were there — which assumes the population is closed for the duration and that every fish has the same chance of capture. Neither assumption holds exactly, and both fail worst in deep, wide or turbid water, so estimates are least reliable where rivers are largest. The published standard error is not carried: a Hydrionis series point holds a value and a date and has nowhere to put an uncertainty, so a density of 14 and a density of 14 ± 9 are stored identically. Density is comparable only between surveys using the same method and strategy. Dividing by a very small fished area extrapolates a whole reach from a few square metres, so surveys below 50 m² are not used at all. |
| Burned area within a radius of a measurement point | 1.0.0 | A radius is not a catchment. A fire fifty kilometres away may be in another basin entirely, and one just outside the radius may sit directly upstream — the circle knows nothing about which way water flows. The centroid stands in for the whole perimeter, so a long fire straddling the boundary is counted wholly in or wholly out. Above all, burned area near a gauge is not evidence of any effect on the water at it: erosion, turbidity and nutrient response depend on slope, soil, burn severity and what rain fell afterwards, none of which this counts. A zero means no mapped perimeter fell within the radius, which is not the same as no fire. |
| Total over a period | 1.0.0 | A total is the one statistic a missing reading corrupts silently. A mean of 340 days is a mean; a total of 340 days is a year that appears drier than it was, and nothing about the number shows it. So this is produced only where the period is close to fully covered, the days actually read are stated, and a shortfall is never made up — an absent day is not a dry day. Coverage is counted in days held, so a period met by a long unbroken run and one met by scattered readings around a month-long outage are not distinguished by the count alone, which is why a maximum run of consecutive absent days is applied as well. |
| Percentile of a flow duration curve | 1.0.0 | A property of the record held, not of the river in perpetuity: the curve moves as the record lengthens, and one drawn from a short or unrepresentative run of years describes those years. The span it was computed over is stated with it for that reason. It assumes the days held are a fair sample of the days that occurred, which fails where a gauge was out of service through a drought or a flood — precisely the conditions the tails of the curve describe. Rating curves are extrapolated at extreme discharges, so the highest percentiles carry the largest measurement uncertainty. Not valid at a station whose discharge is signed or managed — through a sluice, lock or pumping station — because the curve would then describe operating decisions rather than hydrology. |
| Annual maximum of a daily series | 1.0.0 | The maximum of the daily means, which is lower — often much lower — than the instantaneous peak on the same day, because a flood peak lasting hours is averaged across twenty-four. It is not a flood peak and must not be read as one, and a flood frequency analysis needs the publisher’s own instantaneous annual maximum series rather than this. A maximum is also the statistic most damaged by absence: the largest day of a year the gauge missed is simply not in the record, so this is withheld where the year is not substantially covered, and the days held are stated with every value. |
| Discharge as a depth over its catchment | 1.0.0 | The whole result rests on the denominator, which Hydrionis does not measure: the catchment area is the station publisher’s own delineation, and a topographic area used where a contributing area was meant rescales every value by the ratio between them. A runoff depth is what left the catchment past this gauge and is not the catchment’s water balance — the difference between the rain that fell and the runoff that emerged is not evaporation alone, but also change in groundwater storage, abstraction taken elsewhere, and water imported by transfer schemes. A catchment yielding more runoff than its rainfall is usually being fed from outside rather than measured wrongly. Valid only where the discharge is a mean over the whole period being converted: applying it to an instantaneous or peak discharge produces a depth that was never reached. |
| Period a published index spent above a threshold | 1.0.0 | A run above a threshold is not a declaration and not a statement that anything was declared. It is a property of a published series, and both the threshold and the minimum run length are Hydrionis choices a reader may reasonably disagree with — moving either changes which periods appear and how long each lasts. A missing period ends a run rather than being bridged, so a gap in the record can split one long condition into two or hide it altogether. A run still open where the record ends carries no end date: the record stopped, which is not the same as the condition stopping. |
Rights
Derived values inherit the rights of every observation that fed them. A computed figure is never more freely usable than its most restrictive input: doing arithmetic on someone else’s data does not change whose data it is. Each value records the identifiers of its inputs and of their sources, so the inheritance can be checked rather than trusted.
Data behind this
Every figure above is computed from these datasets rather than written into the text. If the underlying record changes, this page changes with it.
Citation
Hydrionis (2026), Release note: Hydrionis Derived Climatology v1.0, version 1.0. Retrieved from https://hydrionis.com/research/derived-climatology-v1-release-note/