Ingestion diagnostics

World Bank World Development Indicators

• Outside plausible range: 37

Examples:

• outOfRange: withdrawal-municipal-share country-brn 2022-01-01: 164.673913 outside 0–100
• outOfRange: withdrawal-municipal-share country-brn 2021-01-01: 164.673913 outside 0–100
• outOfRange: withdrawal-municipal-share country-brn 2020-01-01: 164.673913 outside 0–100
• outOfRange: withdrawal-municipal-share country-brn 2019-01-01: 164.673913 outside 0–100

Environment Agency Water Quality Archive

• Outside plausible range: 8

Examples:

• outOfRange: nitrate-as-n wq-ea-NW-88000064 2010-01-01: 826.2457333333333 outside 0–200
• outOfRange: orthophosphate-as-p wq-ea-NW-88009737 2011-01-01: 283.44945555555546 outside 0–50
• outOfRange: ammoniacal-nitrogen wq-ea-NW-88000045 2023-01-01: 173.60999999999999 outside 0–100
• outOfRange: ammoniacal-nitrogen wq-ea-NW-88000064 2011-01-01: 434.5722765957446 outside 0–100

Hub'Eau Hydrométrie — French river flow

• Outside plausible range: 6

Examples:

• outOfRange: river-flow-monthly-mean gauge-fr-e4909415 2026-07-01: -0.004 outside 0–250000
• outOfRange: river-flow-monthly-mean gauge-fr-k0436511 2009-09-01: -0.011 outside 0–250000
• outOfRange: river-flow-monthly-mean gauge-fr-k0436511 2017-09-01: -0.171 outside 0–250000
• outOfRange: river-flow-monthly-mean gauge-fr-k2210842 2020-07-01: -0.001 outside 0–250000

SEPA — Scottish river flow, rainfall and groundwater

• Outside plausible range: 7

Examples:

• outOfRange: precipitation-monthly sepa-rain-115353 2007-09-01: 6442451123 outside 0–3000
• outOfRange: precipitation-monthly sepa-rain-115353 2008-11-01: 1610613060.8 outside 0–3000
• outOfRange: precipitation-monthly sepa-rain-115207 1993-11-01: 1610612764.6 outside 0–3000
• outOfRange: precipitation-monthly sepa-rain-15118 2008-10-01: 1610612797.3 outside 0–3000

Hub'Eau Prélèvements — French water abstraction

• Outside plausible range: 1

Examples:

• outOfRange: abstraction-volume-total country-fra 2013-01-01: 1018589.933046 outside 0–1000000

EIA — United States thermoelectric cooling water

• Outside plausible range: 1

Examples:

• outOfRange: facility-water-consumption us-eia-plant-1378 2019-01-01: 3669.1617881133598 outside 0–2000

Met Office HadUK-Grid areal series

• Publisher last updated: 03-Aug-2026 09:28

World Bank World Development Indicators

• Dropped 318 sector-share values across 106 country-years where the three shares did not sum to 100% (e.g. country-alb|2016-01-01, country-alb|2015-01-01, country-atg|1998-01-01). This is an upstream inconsistency, not a Hydrionis one, and the whole triple is withheld rather than a guess published.
• World Bank data is CC BY 4.0. Several water indicators are relayed from FAO AQUASTAT or the WHO/UNICEF JMP; the originating body is recorded on every observation.

Environment Agency Hydrology Data Explorer

• waterFlow: 356 stations ingested of 356 selected; 972,537 daily readings failed quality control; 7,594 months withheld as incomplete
• groundwaterLevel: 409 stations ingested of 409 selected; 127,047 daily readings failed quality control; 11 readings rejected as impossible for their own borehole

Environment Agency Water Quality Archive

• Nitrate as N: 1,253,837 results read, 33,220 below detection limit; 3,923 points published, 9,776 dropped as too short a record, 40,829 point-years dropped as too few samples, 1,347 dropped as mostly below detection
• Orthophosphate, reactive as P: 939,210 results read, 120,715 below detection limit; 1,899 points published, 11,737 dropped as too short a record, 37,953 point-years dropped as too few samples, 6,384 dropped as mostly below detection
• Ammoniacal Nitrogen as N: 987,568 results read, 339,546 below detection limit; 1,565 points published, 13,076 dropped as too short a record, 35,070 point-years dropped as too few samples, 17,883 dropped as mostly below detection
• Oxygen, Dissolved as O2: 1,080,096 results read, 27 below detection limit; 3,679 points published, 10,257 dropped as too short a record, 36,356 point-years dropped as too few samples, 0 dropped as mostly below detection
• Oxygen, Dissolved, % Saturation: 913,640 results read, 11 below detection limit; 2,365 points published, 12,194 dropped as too short a record, 35,604 point-years dropped as too few samples, 0 dropped as mostly below detection
• pH: 1,380,928 results read, 27 below detection limit; 4,629 points published, 11,759 dropped as too short a record, 49,729 point-years dropped as too few samples, 0 dropped as mostly below detection

USGS National Water Information System

• 13 sites publish the same statistic under more than one USGS time series identifier; the longest series was used for each. 26 year values differed between identifiers by more than 0.1% and were not reconciled — the shorter series was discarded rather than merged.
• 1750 United States gauges ingested from 50 states; 566 sites dropped where the published statistics were shorter than 50 years
• Monthly means are available from the same service and are deliberately not ingested: they would multiply the observation count roughly twelvefold for a marginal gain.

Hub'Eau Hydrométrie — French river flow

• 892 sites retained (360+ months, at most 10 per department); 828 qualifying sites were dropped by the per-department cap, 2532 had shorter records and 5031 had no published monthly series.
• Discharge is published in litres per second and stored in cubic metres per second, with the publisher’s unit recorded on every series.
• 3487 retained sites have been measured by more than one station. The longest series is used for each; the alternative stations are not blended, because their values for the same month frequently differ.

SEPA — Scottish river flow, rainfall and groundwater

• Q: 309 stations retained of 320; 10 had fewer than 120 months of record.
• Precip: 358 stations retained of 382; 23 had fewer than 120 months of record.
• GWLVL: 45 stations retained of 58; 13 had fewer than 120 months of record.
• Scottish coverage uses SEPA under the Open Government Licence. The National River Flow Archive holds a longer and better Scottish record but its licence forbids republishing it, so it is recorded in the source registry as assessed and refused rather than used.

USGS reservoir storage

• 92 reservoirs retained across 14 states, selected as the 12 longest records in each state meeting a 25-year threshold.
• Storage is stored as a volume rather than as a percentage of capacity, because the capacity figures published alongside do not state whether they are gross, usable or conservation storage, and the three differ substantially.
• 0 reservoir records and 4 state listings could not be fetched on this run and are carried forward.

Hub'Eau Prélèvements — French water abstraction

• 1,236,887 abstraction declarations aggregated into 491 annual series across 101 departments and France as a whole.
• 0 declarations were recorded by the water agency as not abstracted and were excluded rather than counted as withdrawals.
• 13,658 declarations carried a purpose outside the five mapped categories. They are included in the total and not attributed to any purpose.
• These are withdrawals, not consumption, and declared volumes, not licensed entitlement.

Hub'Eau Indicateurs des services — French water service performance

• 68,064 SISPEA service-year returns aggregated into 783 series for 101 departments and France.
• 129 returns carried none of the indicators Hydrionis holds and contributed nothing.
• Figures are medians across reporting services, not population-weighted means: the SISPEA served-population field is null on every record this API returns, so no defensible weight exists and none was invented.
• 4,474 returns carried no resolvable commune code and contributed to the national figures only.
• The observatory publishes returns for 2008 to 2019 and has released nothing later through this API, so these series end there rather than being extended. Later years are requested on every run and will appear without a code change if they are published.

Hub'Eau Qualité de l'eau potable — French drinking water

• 732,942 regulatory results for 5 determinands across 101 departments, aggregated into 1387 annual series.
• 162,079 results (22.1%) were reported as below the laboratory detection limit. They are counted, published as a quantified-share series, and excluded from every mean rather than substituted with zero.
• 4,563 results carried no parsable numeric value and are counted in the denominator only.
• Only complete calendar years up to 2025 are aggregated; the current year is excluded rather than published as a partial annual figure.

Hub'Eau Piézométrie — French groundwater levels

• 641 piezometers held from a national network of 23,330, selected as the 10 longest records in each department meeting a 30-year and 1000-reading threshold.
• Capping per department trades some of the very best records for national coverage: a purely national ranking would concentrate in the Paris and Rhône basins and leave large regions unrepresented.
• 17 selected boreholes were read too irregularly to support monthly means and were not stored.
• 1 borehole records and 0 department listings could not be fetched on this run and are carried forward.

PEGELONLINE — German federal waterway gauges

• W (water level): 737 gauges stored of 737 carrying the series.
• Q (discharge): 92 gauges stored of 92 carrying the series.
• PEGELONLINE serves roughly 31 days of measurement however much history is requested, so this run holds about one month per gauge. The adapter runs in accumulate mode: each subsequent ingestion unions its window with what is already stored, so the German record deepens over time rather than sliding forward. Hydrionis does not hold decades of German history.
• 95 German waterways were recorded from the gauge network. They are held as registry candidates rather than published: the river profile is still the generic entity page, and a page per waterway is worth adding only once there is a profile worth visiting.
• GKD Bayern holds the deep Bavarian record under CC BY 4.0 and is not ingested: its bulk download requires an email address and acceptance of terms through a form, and Hydrionis does not complete access-consent flows programmatically. It is registered as assessed and blocked.

Environment Agency — English water company resources performance

• Read 2 of 2 report editions. Matched 13 of 22 tables to a known measure by header text; the remainder are national summaries or measures Hydrionis does not hold. Collapsed 36 figures the report prints twice at different precision, keeping the finer printing.
• Only two editions of this report exist, so this is seven indicators over two regulatory years — not the long series Ofwat holds and does not serve to automated requests. Two years does not meet the five common years a composite utility index requires, so this deepens the utility layer without triggering an index.
• Figures use the English regulatory year running April to March and the company-reported water balance. They are not comparable with international non-revenue water percentages, which use a different denominator, nor with the French linear leakage index and network efficiency indicators, which normalise by network length rather than by property or person.

Rijkswaterstaat Waterinfo — Dutch national water monitoring

• WATHTE (water level): 328 stations stored of 705; 377 returned nothing for the selected years. Rejected 6,312,728 readings from predicted rather than measured series — Rijkswaterstaat returns astronomical tide predictions alongside measurements at the same timestamps, and only the process type distinguishes them; rejected a further 1,838,461 measured against a reference plane other than NAP; 25 stations offered more than one measured series and the longest was taken rather than interleaving instruments. Rejected 830,132 readings the publisher did not mark quality code 00, including its .999999E+38 no-measurement marker. Collapsed 1,004,534 repeated timestamps from overlapping response blocks, and found 2496 repeats whose values disagreed — the first was kept and the disagreement is reported rather than resolved. Refused 1,223 days with fewer than 108 readings; 186 days held more than one validation state and used the stronger alone rather than averaging across them.
• Q (discharge): 115 stations stored of 199; 84 returned nothing for the selected years. Rejected 69,355 readings from predicted rather than measured series — Rijkswaterstaat returns astronomical tide predictions alongside measurements at the same timestamps, and only the process type distinguishes them; 7 stations offered more than one measured series and the longest was taken rather than interleaving instruments. Rejected 142,942 readings the publisher did not mark quality code 00, including its .999999E+38 no-measurement marker. Collapsed 232,789 repeated timestamps from overlapping response blocks, and found 10734 repeats whose values disagreed — the first was kept and the disagreement is reported rather than resolved. Refused 892 days with fewer than 108 readings; 19 days held more than one validation state and used the stronger alone rather than averaging across them.
• Ingested the 2025 calendar year. Rijkswaterstaat retains its record back to about 2013 at ten-minute resolution and does not discard it, so the depth Hydrionis does not yet hold remains available: one location-year is roughly 16 MB, and the full archive across the network is on the order of 190 GB. Hydrionis does not hold the 2013-present Dutch archive. Set HYDRIONIS_RWS_YEARS to backfill in batches; the adapter accumulates rather than replaces, so a backfill deepens the record instead of overwriting it.

SMHI — Swedish river discharge

• 694 stations stored of 720; 24 publish no corrected archive, almost all of them closed or very short-lived gauges; 2 returned a header with no values. 9,507,623 daily values across 87 main drainage basins.
• Quality as SMHI marks it: 9,109,579 checked and approved, 21,148 roughly checked, 376,896 not yet checked. All three published levels are kept, because the roughly-checked values sit inside continuous records rather than at their ends — dropping them would leave silent holes. Each series states which of its periods are which.
• Stored one partition per main drainage basin. Sweden’s daily archive is roughly 200 MB, which in a single file would be eight times the largest partition Hydrionis holds; split by basin, a page about one Swedish river reads one river’s data.

OPW waterlevel.ie — Irish hydrometric network

• 450 stations stored of 450 carrying an OD sensor. Rejected 17,208 unparseable readings; refused 500 partial days.
• Every sensor’s parameter is confirmed against the publisher’s own station page before its values are read. The interface identifies a series only by a code, and the codes are not self-explanatory: sensor 0003 returns values around 12.9 that read perfectly as a water temperature and are the station’s battery voltage.
• The OPW serves about five weeks per station here; its longer record is on the separate Hydro-Data site, which Hydrionis does not scrape. This adapter accumulates, so the Irish record deepens with each run rather than sliding forward. Hydrionis does not hold Ireland’s historical archive.

Syke — Finnish river discharge

• 3 basins exceeded 1,000,000 values and were split on Syke's own sub-basin boundaries: kokemaenjoki, kymijoki, vuoksi. A page about one gauge parses the whole partition its series lives in, so a basin is divided before that becomes expensive rather than after.
• 671 stations stored of 711 listed; 40 hold fewer than 365 usable days. Basins: 62.
• Refused 622 values that Syke itself flags as possibly erroneous, and 0 that were empty or unparseable. Stored 35,586 interpolated values, which are a stated method rather than a doubt, and which each series describes.
• Daily discharge is taken exactly as Syke publishes it. Every station record also names the main drainage basin it sits in, so Finnish gauges join the water graph at ingestion rather than through a separate linkage pass.

Eurostat — European freshwater abstraction

• 405 series across 40 countries, from 23,447 cells; skipped 459 cells for aggregates and units Hydrionis does not hold as countries; skipped 14,710 for sector and source combinations without a metric; deferred 131 to a national register Hydrionis already holds in more detail.
• Every value is reported actual abstraction. Eurostat publishes no licensed or authorised volumes, so nothing here can be confused with an entitlement. Abstraction is also not consumption, and no return flow is estimated.
• The sector codes double-count by design — manufacturing cooling is part of manufacturing, and irrigation and aquaculture are parts of agriculture. Only the leaves Hydrionis holds a metric for are stored, so a page cannot sum a country’s sectors to more than its own total.

EIA — United States thermoelectric cooling water

• 838 generating stations with at least one complete reported year, from 160,585 monthly rows across 11 annual workbooks. 441 distinct named water sources.
• Refused 488 plant-years in which the operator reported fewer than twelve months. A year assembled from part of itself is a different quantity with a year's name on it.
• Refused 2 plant-years reporting negative consumption. EIA derives the figure as a difference, so a plant returning more than it withdrew reports below zero; that is a real outcome and not one this metric can hold, since consumption is the part of a withdrawal that did not come back.
• Refused a further 104 plant-years reporting more water consumed than withdrawn, which cannot happen. Both figures are dropped for those years: one of the two is wrong, nothing says which, and correcting an operator's statutory return is not Hydrionis's to do.
• Withdrawal, consumption and reclaimed water are stored as three separate metrics and are never combined. For cooling water the difference between withdrawal and consumption is a factor of a hundred at a once-through plant, so substituting one for the other is not a rounding matter.
• 788 of 838 facilities carry an operator name, a position and an ownership record from the companion EIA-860 return. Ownership is a percentage a named company filed about a specific generator, not a corporate hierarchy Hydrionis assembled.
• These are facility figures. A utility operating fifteen plants appears fifteen times and the fifteen are never summed here: a company-wide total is assembled on a different basis and belongs to the company, not to any of its sites.

IMGW — Polish hydrological network

• 898 stations stored of 913 listed: 875 water levels, 646 discharges, 104 water temperatures. Refused 71 readings older than 14 days — a rolling snapshot repeats an old value indefinitely, and storing it on every run would fill the record with one ancient point per station.
• 898 of 898 stations name the watercourse they gauge, so Polish gauges join the water graph at ingestion rather than through a separate linkage pass.
• IMGW serves the latest reading per station and nothing older. This adapter accumulates, so the Polish record deepens with each run rather than sliding forward. Hydrionis does not hold Poland’s hydrological archive, which IMGW releases on request.