When the Weather Clears the Airport Recovers Faster Than the Airline Does

A line of storms sits over a hub airport for three hours and then moves east. The ceiling lifts, the runways go back to the configuration they had that morning, and the rate at which arrivals can be accepted returns to normal. The departure boards do not, and a flight scheduled for nine o'clock leaves at eleven for a reason no passenger can see out of the window.

Two kinds of quantity are being confused when that looks like a contradiction. Runway capacity is a rate: counted per hour, and back the moment the constraint that lowered it goes away. The traffic that could not be accepted while the rate was low is not a rate. It is an amount of unflown work, sitting in airframes parked at other airports and in crews whose legal day is already running. A rate recovers instantly. An amount has to be drained, and it drains only at the margin between restored capacity and the demand already scheduled.

Capacity Is a Rate. A Backlog Is Not.

The number the system runs on has a formal name. The FAA's Pilot/Controller Glossary defines the airport arrival rate as "a dynamic input parameter specifying the number of arriving aircraft which an airport or airspace can accept from the ARTCC per hour." The operative words are dynamic and per hour: revised as conditions change, and a rate rather than a stock.

When the rate falls under the demand scheduled against it, the traffic that cannot be accepted waits somewhere, and the tools that decide where are in the same glossary. A ground delay program is "a traffic management process administered by the ATCSCC, when aircraft are held on the ground," whose purpose "is to support the TM mission and limit airborne holding" and which provides "for equitable assignment of delays to all system users." The FAA's traffic flow management site describes the mechanism as controlled departure times issued at the point of departure, with affected flights not permitted to depart until that time. A ground stop "requires aircraft that meet a specific criteria to remain on the ground," and "normally occur[s] with little or no warning."

Neither tool destroys demand; both move it. An aircraft held at its origin still intends to operate and still holds a place in somebody's schedule later in the day. Ground waiting replaces airborne holding, and the quantity survives intact.

The Ceiling Comes From a Separation Standard

An arrival rate is not chosen freely. Underneath it is a spacing requirement in nautical miles, set by FAA Order JO 7110.65, section 5-5-4. In terminal airspace with a single sensor the minimum is 3 miles when less than 40 miles from the antenna and 5 miles at 40 miles or more. On final approach, 2.5 nautical miles is authorized between aircraft established on the final approach course within 10 NM of the landing runway when the equipment and runway occupancy conditions there are met. Wake turbulence and aircraft category move these figures again, each under conditions the section states.

Those are distances, and an arrival rate is a frequency, so turning one into the other requires a speed. That conversion is done here rather than by the FAA, and its assumption has to be stated: a final approach ground speed of 140 knots on a single runway. At that speed, 3 nautical miles of spacing is 77 seconds between arrivals, or 46.7 per hour. Nothing in these minima says what a given constraint does to the accepted rate. A storm might halve it, shave it, or close the field; the separation table does not answer that. What follows is the arithmetic of a halving, whatever produces it.

Put 40 arrivals per hour of schedule against that single runway. At 46.7 per hour the schedule fits with 6.7 per hour to spare. Halve the accepted rate to 23.3 for three hours and the shortfall runs at 16.7 aircraft per hour, so 50 aircraft accumulate. When the constraint lifts the runway is back at 46.7 within minutes, but the surplus available to work the backlog off is still only that 6.7 per hour. Draining 50 aircraft at that surplus takes 7.5 hours, so a three-hour reduction produces a queue that lasts ten and a half hours, three and a half times the event that created it. Every figure there is computed here from one separation minimum and the stated assumptions; a real airport with parallel runways and an hour-by-hour demand curve would produce different numbers. What survives is the structure: the drain is governed by the margin, and at a hub scheduled close to its capacity the margin is thin by design.

The Chain

1. A constraint lowers the rate at which arrivals can be accepted per hour.
2. Scheduled demand exceeds that rate, and the difference accumulates as unflown aircraft.
3. Traffic management moves the waiting to the ground, relocating the queue without shortening it.
4. The constraint lifts and the rate returns to normal.
5. The backlog can leave only at the surplus between the restored rate and the demand already scheduled.
6. Each delayed aircraft carries its lateness into its next leg, the airframe being one physical object.
7. Ground time between legs absorbs only the slack scheduled into that particular turn.
8. Crew duty limits are fixed by rule, so at the ceiling the remaining leg is cancelled rather than flown late.
9. The reporting system assigns the downstream legs to a category naming the aircraft, not the constraint.

Delay Is Stored in Airframes

The first store is physical. An aircraft that lands 90 minutes late is a 90-minute-late object at a gate, and the flight booked on that tail number cannot go until it has arrived, been unloaded, serviced, boarded and pushed back. The Bureau of Transportation Statistics keeps a category for exactly this: late-arriving aircraft covers the case where "a previous flight with same aircraft arrived late, causing the present flight to depart late."

This store empties slowly because the only thing that absorbs lateness is scheduled ground time in excess of the minimum turn. A turn scheduled for 60 minutes against a 35-minute minimum absorbs 25 minutes and no more; everything above that is handed forward intact. A four-leg day gives an airframe three chances to shed time, each worth only the slack built into that turn, and mid-day slack is what a schedule spends to keep the aircraft productive. The reliable reset is the overnight, which is why an afternoon disruption reads as an evening of late departures and a normal morning.

The Crew Is a Second Store, and It Has a Hard Ceiling

Airframes bend. Crews do not. Under 14 CFR 117.3, a flight duty period begins when a flightcrew member reports for duty intending to conduct a flight and ends "when the aircraft is parked after the last flight and there is no intention for further aircraft movement." Section 117.13(a) then bars a certificate holder from assigning, and a flightcrew member from accepting, "an assignment for an unaugmented flight operation if the scheduled flight duty period will exceed the limits in Table B of this part," except as provided in 117.15.

Table B is a grid, keyed to the acclimated local time the duty period starts and the number of flight segments it contains. A duty period beginning between 0700 and 1159 allows 14 hours for one or two segments, 13 hours for three or four, and 11.5 for seven or more. Beginning between 1700 and 2159 those same three figures are 12, 11 and 9; beginning between 0000 and 0359, every column reads 9. Under 117.13(b), if the flightcrew member is not acclimated, the Table B value is reduced by 30 minutes.

Flight time is capped separately from duty, and the cap has two sources. Under 117.11(a)(1), a flight operated with the minimum required flightcrew takes its limit from Table A, which allows 9 hours for a duty period reporting between 0500 and 1959 and 8 hours for one reporting between 2000 and 0459. Augmented crews are not in that table at all: the limits sit in the section text, where 117.11(a)(2) sets 13 hours for a three-pilot flightcrew and 117.11(a)(3) sets 17 hours for a four-pilot flightcrew. Flight time beyond the limits must be reported to the Administrator within 10 days.

There is give, and it is bounded and countable. Section 117.19(a) permits a flight duty period to be extended by up to 2 hours for unforeseen operational circumstances arising before takeoff, and an extension of more than 30 minutes "may occur only once prior to receiving a rest period." After takeoff, 117.19(b) allows extension only "to the extent necessary to safely land the aircraft."

So the crew store does not merely slow as it fills. It stops. A leg that would push a duty period past its Table B ceiling cannot be flown by that crew, so the schedule finds another crew or cancels. Cancellation is one way a schedule discharges what it cannot absorb, and one reason a disruption's tail can include cancellations of flights that were never near the weather.

Where the Cause Gets Reassigned

The public record is a set of categories, each describing the flight being reported rather than the event that started the sequence. Since June 2003, carriers filing on-time data have also filed causes. Extreme weather covers significant meteorological conditions "that, in the judgment of the carrier, delays or prevents the operation of a flight such as tornado, blizzard or hurricane." National aviation system delay covers non-extreme weather, airport operations, traffic volume and air traffic control. Air carrier delay covers circumstances within the airline's control; late-arriving aircraft covers the previous-flight case.

Whether any of it is counted turns on 14 CFR 234.2, which makes a flight late when it arrives at the gate 15 minutes or more after its published arrival time. Section 234.4(f) measures times "at which the aircraft arrived at and departed from the gate or passenger loading area," so time between gate and runway sits inside the measurement.

The reassignment happens at the seam between legs. When a constraint delays one leg and the same airframe then flies three more legs late, the first is a weather or system delay and the next three are late-arriving aircraft. BTS states the limitation on the page built to work around it: "Airlines do not report the cause of delay for the first late flight that caused the second delay." That page, Aircraft Arriving Late: Causes of the Original Delay, estimates what stood behind the bucket.

In June 2026, the most recent month published there when this was written, 66,761 operations were delayed by a late-arriving aircraft, accounting for 5,309,882 delay minutes. Traced back, air carrier delay accounts for 53.02 percent of those minutes, national aviation system weather 21.90, extreme weather 11.67, volume 7.13, other 5.00, closed runway 0.98, security 0.19 and equipment 0.11. The two weather rows are listed separately; adding them gives 33.57 percent of late-aircraft minutes traceable to weather and 66.43 percent traceable to something else. That sum is computed here.

What the Two Columns Cannot Separate

The page states its method, and the method sets how far the table can be read. Because airlines file no cause for the first late flight, the calculations "use the percentages of delay minutes reported by the airlines in the air carrier, national aviation system, security and weather categories and assign them proportionately to the late arriving aircraft category." One set of shares, drawn from delay minutes reported elsewhere, is laid over the late-aircraft totals.

That set of shares lands in both columns. Air carrier delay is 53.020 percent of the minutes and 53.020 percent of the operations, national aviation system weather is 21.898 percent of each, and across all eight rows the two shares never differ by more than 0.001 of a percentage point. Dividing one column by the other therefore returns the same figure every time: 79.54 minutes in five of the eight rows, and 79.50, 79.28 and 79.21 in the other three. The spread of 0.33 minutes is rounding on the smallest counts. Those quotients and that comparison are computed here.

So the table holds one estimate printed twice, in minutes and in operations. It supports reading the estimated share of late-aircraft delay attributable to each original cause. It does not support treating the operations column as an independent count, nor ranking causes by how long their delays run, because that quotient is fixed by construction rather than measured. The totals the apportionment starts from are a different matter: 5,309,882 divided by 66,761 is 79.5 minutes per affected operation. The page notes its numbers are rounded and may not add to the total, and names its sources as BTS Airline Service Quality Performance 234 data and FAA OPSNET.

Where This Doesn't Apply

• Where demand is well under capacity. The drain arithmetic depends on a surplus existing. A field scheduled far below its restored rate absorbs a backlog quickly; one scheduled at or above it does not drain until demand falls, usually overnight.
• Where the network is not tightly coupled. Spare airframes and reserve crews at the affected station break the chain at step six or seven. Propagation is a property of thin schedules.
• The rate arithmetic is illustrative. The 3 NM figure is one minimum in section 5-5-4 under the conditions stated there, converting it to arrivals per hour needed an assumed 140-knot ground speed on one runway, and the halving is a stipulation rather than a measured effect of any condition. Published arrival rates depend on far more, including runway configuration, aircraft mix and staffing.
• Part 117 has more cases than the ones quoted. The ceilings above are the unaugmented Table B values; split duty periods and the rest of the augmented rules fall under other sections not covered here.
• The cause categories are carrier judgments. BTS defines extreme weather as conditions that, "in the judgment of the carrier," delay or prevent operation, so the line between it and non-extreme weather is a reporting decision, not a meteorological threshold.
• Cancellations leave the delay statistics. A network that cancels early shows a shorter delay tail than one that flies its schedule late, without recovering any faster.

The Board Is Showing an Inventory

A departure board late in the evening after a clear afternoon is not reporting the weather. It is reporting a level: how much unflown work is still in the system, held in airframes in the wrong places and crews whose legal days are partly spent. The constraint that produced that level was a rate, and rates return to normal as soon as the condition setting them ends. The sky outside the window has stopped being evidence of anything.