Cleaning airport wind reports

The question. How do we stop garbled METAR winds being counted as storms, without throwing away real ones?

The problem

Airport reports are typed and decoded; a few are wrong. In our data a station reported a mean wind of 214 kt and, half an hour later, 215 kt (EGPI, 23 June 2025); others had single values of 99–130 kt. One such report would count as a 250 mph storm.

What we tried

1. Confirmation: a value counts only if another report from the same airport within ±90 minutes is at least that value less 20 kt. Gusts in a real storm come in runs; a garbled report stands alone. This dropped 79 values of 30 kt or more across 36,822 airport-days, and kept every real storm we looked at (e.g. Storm Floris, 4 August 2025, with 60–72 kt gusts across Scotland). 2. It was not enough: the EGPI 214/215 kt pair confirm each other. A ceiling was added: any wind or gust of 130 kt or more is treated as missing — then lowered to 100 kt after a second pair (EGPI, 114 and 110 kt on 12 November 2026). No airport here is likely to measure that much, and a storm that did would have plenty of other reports above 43 kt (50 mph) the same day, so nothing real is lost. 3. Days without winds: a code review pointed out that a station-day was accepted on 18 reports of any kind, even if none carried a wind — and would then count as calm. Now it needs 18 reports with a wind. This affected 26 of 36,822 airport-days (6 with no wind at all) and moved the coefficients only in the fourth significant figure.

What worked

The three rules together: confirmation for lone spikes, a ceiling for repeated garbage, and a minimum of real readings.

What did not

Confirmation alone. Any filter that relies on neighbouring reports agreeing fails when the garbage repeats, as it does when an instrument or a decoder goes wrong for a while.

← All experiments