- Every 15 minutesReadings arrive from over 1,000 EA, NRW and SEPA gauges.
- Eight checksLimits, spikes, neighbours, radar and tipping-bucket fault patterns.
- Quality indexStarts at 1.0; each failed check takes some off.
- AI reviewEvery flagged reading is judged genuine, uncertain or suspect.
What the four states mean
- Likely faulty
- The AI review judged the reading suspect, or it was above anything physically possible in 15 minutes (60 mm or more). These are the readings to leave out of a rainfall total.
- Uncertain
- The evidence points both ways. Worth checking against radar and nearby gauges before relying on it.
- Likely genuine
- At least one check failed, but radar, the gauge's own history and its neighbours support the reading. Usually a real, local downpour that nearby gauges missed.
- Not yet reviewed
- Flagged, with no AI verdict yet. In a widespread event only the 40 heaviest readings are reviewed each run; the rest follow on later runs.
Why most flags are genuine rain
The checks are deliberately sensitive, so a lone shower that the neighbouring gauges missed trips them too. The review weighs radar, the gauge's recent history and its neighbours to tell that apart from a fault. Start with anything marked likely faulty.
Using the page
The counts at the top of the list give the verdict for the run on screen and filter the list and the map. Move through earlier runs with the slider under the map: red and amber marks show when gauges were judged faulty or uncertain. Flag history lists each episode, a gauge flagged for several runs in a row, over the chosen window; open one to see every run and go to it. Runs are kept for 14 days and the history for 90.
Gross error QI 0 or −0.50
Physically implausible readings in a single 15-minute slot, usually a telemetry glitch or counter reset. 60 mm or more is rejected outright: beyond any credible UK 15-minute total (the UK 30-minute record is about 80 mm). 35 mm or more is very suspicious, around a 1-in-50 to 1-in-100-year total for most sites.
Seasonal range −0.25
Totals unusual for the time of year, even if possible: 25 mm in 15 minutes from April to September, 15 mm from October to March. UK-wide limits, as per-gauge 30-year climatology is not available.
Sudden spike −0.25
A reading at least 8 times the median of the previous seven (1 hour 45 minutes), or a jump from nothing, and at least 5 mm.
Spatial outlier −0.10, −0.25 or −0.50
Compares the gauge with every gauge within 100 km, using the median and the median absolute deviation (MAD) so a few wet gauges do not skew it. The deviation index is the gap from the median divided by the MAD: above 4 is weak, above 8 medium, above 15 strong. Needs 5 or more gauges and a reading of 1 mm or more. A lone real shower often trips the weak level.
Stuck at zero −0.75
A gauge reporting nothing while it clearly rains around it, typically a blocked funnel or frozen sensor: 0.2 mm or less in the last 24 hours while at least 2 gauges within 20 km, and most of those reporting, recorded 3 mm or more, and the radar within 5 km totals 5 mm or more (3 wet gauges if there is no radar history). Gauges with fewer than 2 reporting neighbours are not judged. A gauge that fails this is marked likely faulty whatever the review says.
Radar mismatch −0.25
Compares the gauge with the Met Office 1 km radar composite within 5 km: gauge 5 mm or more, radar under 0.5 mm, and the gauge more than 15 times the radar. Radar can miss heavy rain, so this is weak evidence on its own.
Drip or stuck counter −0.25 or −0.50
The signature of a leaking or blocked tipping bucket: over 6 hours, at least 18 of 24 readings above zero, averaging under 1.5 mm with very little variation, unless nearby gauges are in light rain too. Or 12 identical readings in a row: a stuck counter.
24-hour total anomaly −0.25
Steady over-recording that only shows over a whole day: a 24-hour total of 15 mm or more and over 5 times the average of at least 3 neighbours that each recorded 5 mm or more.
Every reading starts with a quality index (QI) of 1.0. Each failed check takes off a fixed amount, down to 0. It follows the RainGaugeQC scheme; see Method.
0 to 0.24Reject
0.25 to 0.49Significant
0.50 to 0.74Moderate
0.75 to 0.99Minor
1.0Trusted
| Gross error, 60 mm or more | set to 0 |
| Gross error, 35 to 60 mm | −0.50 |
| Seasonal range, sudden spike, radar mismatch, 24-hour anomaly | −0.25 each |
| Spatial outlier: weak, medium, strong | −0.10, −0.25, −0.50 |
| Stuck at zero | −0.75 |
| Drip, stuck counter | −0.25, −0.50 |
The QI is mechanical: it counts failed checks and knows nothing of context. That is why the status on this page comes from the AI review, which does, and not from the QI.
When it runs
When a gauge is first flagged: at least one check fails, or the reading is 20 mm or more even if every check passes. While it stays flagged the verdict carries forward, and it is asked again every hour, or sooner if another check starts failing or the reading jumps. A reading already judged is never sent twice. At most 40 readings are reviewed per run, heaviest first.
What it sees
- The gauge's location, the season and month
- The reading, how long the gauge has been flagged, and its 1, 3, 6, 12 and 24-hour totals
- Every check result with its numbers, and which failures are new this run
- Met Office radar rainfall within 5 km and 30 km, and in the runs either side
- The Met Office UKV forecast's 1-hour rainfall at the gauge, as soft support only
- How many gauges within 50 km are recording rain, and the three nearest gauges' recent readings
- UK rainfall records, so it knows what is physically plausible
Advice, not correction
Readings are never changed or removed. The verdict, two or three sentences of reasoning with it, only says how far to trust them. It is written by a language model and can be wrong, which is why the checks and the evidence behind it are shown beside it.
The checks and quality index follow the RainGaugeQC framework: Ośródka, K., Otop, I. and Szturc, J. (2022). Automatic quality control of telemetric rain gauge data providing quantitative quality information (RainGaugeQC). Atmos. Meas. Tech., 15, 5581–5597. doi:10.5194/amt-15-5581-2022
UK records used as sanity limits: 5 minutes 32 mm (Preston, 1893); 30 minutes 80 mm (Eskdalemuir, 1953); 60 minutes 92 mm (Maidenhead, 1901).
UK adaptations
- No paired sensors. The paper compares heated and unheated sensors at each site. UK telemetry gauges have one tipping bucket, so the temporal check only looks for spikes.
- No per-gauge climatology. UK-wide seasonal limits replace 30-year per-station range limits.
- One spatial domain. The paper adds four offset 25 km sub-domains; only the 100 km domain is used.
- Flat radar deductions, where the paper varies them by terrain.
- Stuck at zero softened from an outright reject to −0.75, as radar quality is not formally assessed.
- Two extra checks, drip or stuck counter and the 24-hour anomaly, for common tipping-bucket faults.