Flight-Time Forecast Accuracy
How we measured a 3.47-minute mean absolute error
Internal frozen-test results for the WeatherIQ® airborne flight-time model. The test population was held out during development and evaluated once, after model selection. These are internal results, not an independent audit.
3.47 min
mean absolute error
2.41 min
median absolute error
79.8%
within ±5 minutes
95.1%
within ±10 minutes
+0.02 min
mean signed error
Frozen test
42,428 US domestic jet flights, October 2025 through April 2026, held out during model development and evaluated once after model selection. No tuning against this population.
Target
Airborne flight time: actual wheels-off to touchdown. Mean absolute error (MAE) is the average of |predicted − actual| across flights; median absolute error is the middle value of the same distribution.
Inputs
GFS forecast winds published before wheels-off; actual wheels-off time as the prediction anchor; the filed flight route; the filed cruise altitude; runway prediction from forecast surface winds; aircraft-type performance.
Not included
Taxi-out, departure ground stops, gate-to-gate block time, or any ground segment. Those are modelled separately by the operational delay intelligence and are not covered by this figure.
Tested vs production
A website user may enter only an airport pair, aircraft and time; the system must then estimate routing and cruise altitude. That airport-pair workflow has not been evaluated on the frozen population at this scale. The 3.47-minute figure applies to the tested model under the inputs above.
Airline-filed ETE
On 42,036 comparable flights the airline-filed ETE had MAE 5.22 min, median 3.73 min, 63.0% within five minutes. We do not publish a percentage improvement until WeatherIQ® error is confirmed on that identical 42,036-flight subset; MAEs from different populations are not combined.
International
A separate set of approximately 17,198 international flights produced an overall MAE of about 8.9 minutes. The US domestic headline does not apply to international operations.
What the headline figure does not cover
Taxi-out, ground stops, departure queues, block time, the full door-to-door journey, international flights, and the public airport-pair workflow in which routing and cruise altitude are estimated rather than filed. Operational delay intelligence is a separate model and is reported separately.