Texas 24/7 Rank

How a station is ranked

The 247 Rank answers one question about a weather station: can its readings be relied on? A station has to deliver: its reports arrive when due, soon, without long silences. And its readings have to be right: close to a trusted station nearby, passing their checks, and in step with the weather the forecast and its neighbours confirm.

The 247 score

Storm Station's poller asks every network on its own schedule, around the clock, and keeps the time of every report it hears; the reference comparison and the forecast engine keep the rest. From the last seven days of that record each station is scored out of 100 on six parts, and its 247 score is their weighted mean:

1. Reports receivedweight 35The share of its scheduled reports received. A station that sends every 5 minutes is due 2,016 reports in a week; one that reports once a day is due 7. Each report is counted in a slot centred on the station's own report time, so a report a minute early or late is still on time, and the week ends an hour back, so a report still on its way is never counted missed.Full marks: 100% received. No marks: none received.
2. Arrivalweight 10The median time its reports took to arrive from the moment a reading was taken to the moment it reached Storm Station.Full marks: within 10 minutes, or a tenth of its schedule for a station that reports less often. No marks: an hour, or its whole schedule.
3. Continuityweight 10Its longest silence against its own schedule, so a daily gauge is not marked down for a day between reports.Full marks: a silence of 3 reports or less. No marks: 72 reports, falling evenly on a log scale between.
4. Accuracyweight 20How closely its temperature follows its reference station the nearest airport or climate reference station that reported within half an hour, up to 15 miles away, reading by reading. A steady offset is taken out, and where both heights are known only the offset beyond 3 F that the height difference does not explain is counted.Full marks: within 1.5 F, allowing 0.1 F more for each mile to the reference. No marks: 6.5 F beyond that.
5. Instrument checksweight 15The share of its readings that pass a check. For an airfield, the check the forecast engine runs the next day: each reading against the stations around it, as weather services check their own. For any other station, a reading 20 F or more from its reference fails, unless the two stand 3,000 ft or more apart.Full marks: none failed. No marks: 5% failed.
6. Forecast checkweight 10How closely its temperature matched Storm Station's forecast for the same hours, checked the next day, for every airfield the forecast engine scores. A station whose readings stray far from a forecast its neighbours confirm is one to doubt.Full marks: within 2 F on average. No marks: 8 F.

A part is scored once the week holds 24 readings or comparisons for it. A part that cannot be measured where a station stands counts at the middle of its state's measured stations, so the station is neither lifted nor sunk by it. A part measured nowhere in the state is left out, and the others carry its weight.

From score to rank

Each state's stations are ranked by their score, to the tenth. At an equal score the longer record comes first: a station that sent 2,016 of 2,016 reports has shown more than one that sent 7 of 7. Then the faster arrival, then the shorter silence. Ties on all of them share a rank, as a race is scored: three stations tied for third are all third, and the next is sixth.

A station is ranked once its week holds enough reports to measure: 12, or 3 days for a gauge that reports once a day. The rank is made again every day, so a station that falls silent or starts to read wrong drops, and one that recovers climbs back. Every page shows the same ranking: a station's rank here, in its state's list and on its own page come from one run, and a new run reaches them together. Each station's page shows its score part by part, with the figures behind each.

The same figures as data

/api/ranks for a state and /api/rank for one station, its score in parts; the developer page has every field.