After the Weather Clears

Summary

In the first normal hour after an episode of bad weather, 4.9% of departures were cancelled, 6.2 times the expected rate. The rate came within 1.5 times the expected rate in hours 25 to 36. Measured on 1,626,222 departures after episodes of 3 or more bad-weather hours at the covered airports.

Every departure from a covered airport is placed in that airport’s weather for its scheduled hour. A run of consecutive bad-weather hours at one airport is an episode, and the normal-weather hours after it are numbered by how long ago the last bad hour was, from 1 to 36. 1,626,222 departures fall in those hours after an episode of 3 or more hours. Each is compared with departures from the same airport, calendar month and scheduled hour when the weather was normal and had been for 36 hours. Hours with another bad hour within 8 hours ahead are left out of both, so the figures follow a recovery and not the next storm.

Departures from covered airports in normal-weather hours that follow a run of 3 or more bad-weather hours, with no bad hour within 8 hours ahead. Expected is the rate among departures in the same airport, calendar month and scheduled hour when the weather was normal and had been for 36 hours. A dash is a row under 120 departures.

In the first normal hour after an episode 4.9% of departures were cancelled, against 0.8% expected: 6.2 times the usual rate, or 41.1 extra cancellations in every 1,000. In hours 4 to 6 it was 2.3% against 0.7%. The rate came within 1.5 times the expected rate in hours 25 to 36, 1.06% against 0.73%. Departures that left 15 or more minutes late followed the same shape: 40.0% against 19.7% in the first hour, and within 3 points of the expected share in hours 25 to 36.

Snow or ice was the slowest to recover: its cancellation rate had not come within 1.5 times the expected rate by hour 36, 2.18% against 0.71% in hours 25 to 36. The fastest were low ceiling, low visibility or fog and high wind, within that range in hours 13 to 24. Across kinds, the extra cancellations in the 36 hours after an episode came to 37 for every hundred extra in the bad-weather hours themselves, so counting only the bad hours leaves out about 27% of what an episode cost. Longer episodes took longer: 12 hours or more of bad weather were followed by 8.0 times the expected cancellation rate in hours 1 to 3, against 3.0 times after an episode of under 3 hours.

An episode is sorted by the first of these that any of its bad hours had: snow or ice, thunderstorm, low ceiling or visibility or fog, high wind. The last column sums the extra cancellations in the 36 hours after and sets them against the extra cancellations in the bad hours themselves, for episodes of 3 hours or more.

Hours of bad weather in the last run before the departure. The first row includes episodes under 3 hours, which the other tables leave out.

Across the 61 covered airports with at least 500 departures in hours 4 to 24 after an episode, DFW lost the most, with 32.9 extra cancellations per 1,000 departures (3.84% against 0.55%). SEA had the lowest figure, -3.2 (0.64% against 0.96%). The weather that visits an airport is not the weather that visits another, and the airlines at each are different, so this describes the departures from each airport and does not rate it.

Covered airports with at least 500 such departures and a cancellation rate at least 2 standard errors from its expectation. The first rows are the airports where the cancellations last longest after the weather has gone and the last rows the ones where they end soonest.

The aircraft that left an airport during bad weather fly on to other airports. 449,928 departures from airports in normal weather had an aircraft whose previous flight left an airport in bad weather, and 33.9% of them left 15 or more minutes late, against 22.7% for departures from the same airport, calendar month and scheduled hour whose aircraft had been only in normal weather (+11.2 points, 12.3 minutes of average delay). The gap shrank with each flight, +7.1 points when the bad weather was two flights back and +4.3 three back, and was still above zero. The delay does not stay at the airport where the weather was.

Departures from an airport in normal weather whose aircraft’s earlier flights, up to 3 back on the same operating day (a scheduled turn under 6 hours at each stop), left an airport in bad weather. Expected is the rate among departures from the same airport, calendar month and scheduled hour whose aircraft had left only normal-weather airports on its last 3 flights. A stratum with fewer than 120 such comparison departures gives no expectation, and 253,509 departures whose previous flight left bad weather fall in strata like that and are left out.

The weather is the hourly observation at the airport and says nothing of the weather elsewhere. A cancellation after the weather has gone may be the result of weather at another airport, a crew or aircraft out of position, or a schedule the airline chose to thin, and the data does not say which. The figures describe the hours after an episode and not the hours of the next one: an hour with another bad hour within 8 hours ahead is left out. The data has no announcement time, so it cannot say when a cancellation was decided. An episode runs through consecutive hours with a report, so a missing hour ends it early. An hour that two weather hours map to when clocks go back, and an hour skipped when they go forward, are left out. The aircraft section follows a tail number through its day and uses the same links as the late-aircraft analysis; a flight with an unknown or ambiguous previous flight is not in it, and cancelled flights are not part of a chain.