Time-Window Mismatch in Reactive Windshear Alerting During Approach and a False-Alarm Suppression Algorithm Based on a Physical-Consequence Criterion

Low-level windshear is among the most hazardous weather on approach, and the reactive alerting chain fixed by certification standards such as TSO-C117 is the last line of defense. This paper asks whether the fixed five-second decision window of the certified F-factor criterion leaves a systematic false-alarm space in compact-microburst encounters, and whether a criterion based on physical consequence can suppress this space without modifying certified behavior; the question is examined on a Boeing 747-400 approach-phase simulation platform, with the TSO-C117 alerting chain reproduced in the loop as the certification baseline. Replaying the standard’s six reference window lengths (5–10 s) shows that false-alarm suppression and hazard detection trade off along the window axis, and none of the tested windows satisfies both. A three-channel short-window (0.5 s) neural detector with force-residual auxiliary tasks from Newton’s second law therefore runs in parallel with the untouched certified chain and arbitrates the alert level. On 14,825 Monte Carlo scenarios, the false-alarm-zone alert rate drops from the 99.2% certification baseline to 8.0%, with residual triggers concentrated near the upper edge of the divergence band, while at the classifier level (a complementarity analysis, not a deployed behavior) the union of the two criteria misses none of the 6888 genuinely hazardous scenarios; re-adjudication under four third-party consequence criteria confirms the suppressed scenarios are benign. With zero real data in training, the detector covers 68.4% of 329 GPWS windshear alert events in real-flight QAR records, responding a median 14.5 s before the alert trigger. The physical-consequence criterion extends the classical predictive/reactive complementarity into the reactive layer.