Advantages of Classification Strategies over Deterministic Wind Forecasts for Reduced Wake Separation Opportunities

It is well known that during periods of inclement weather (precipitation, high winds and/or low visibility), air traffic control may have difficulty maintaining throughput. Studies show that in a typical year, weather is involved in approximately half of aircraft delay [1]. One contributor to these delays is the need for safely spacing aircraft to avoid hazardous encounters with wake vortices produced by other aircraft in the vicinity. Significant science is involved in computing safe distances for spacing aircraft, which has shown that these distances vary according to the impact of the meteorological environment on the wakes. This opens up possibilities of safely reducing spacing when environments are conducive, but such measures require forecasting to ensure the environments adequately persist for reliable wake impacts. To take advantage of this, the Federal Aviation Administration (FAA) developed Wake Turbulence Mitigation for Departures (WTMD), where the presence of crosswind provided opportunity by advecting any wakes out of the path of subsequent aircraft departing on a closely-spaced parallel runway, effectively removing the need for a wake separation for such operations. To support this procedure, a wind forecast algorithm (WFA) was built to predict the weakest (i.e., least effective) crosswind expected in the ensuing 20 minutes [2]. Note that for WTMD, only a single meteorological variable was forecast utilizing a statistical model to generate the deterministic forecast. While the procedure was successful, low capture rates limited availability for spacing reduction. An alternative to seeking times when a wake separation is not required, is to seek times when the existing separation can be decreased. This will occur during predominant headwinds of impactful strength; thus both the wind speed and some aspect of direction must be forecast to infer the appropriateness of a spacing reduction. Unfortunately, utilizing the same forecast strategy as WTMD (a deterministic forecast) suffers from additive errors of the multiple forecast quantities. To overcome this, a classification algorithm based on eight features derived from surface wind observations was successfully developed to identify appropriate opportunities for spacing reduction.

This presentation covers the building of the classification model, with key areas being: feature importance and selection, target development, and error analysis. The discussion will include how human expertise combined with feature importance provided the best performing model, and how capture rates were increased by: idealizing the training target times, identifying patterns of errors that could be avoided, and leveraging the complementary nature of the errors between the wind speed- and direction-related forecasts. Development was done at a variety of sites across the country (EWR, JFK, LAX, MEM, and ORD) with a focus on the major parallel runways. Wind speed thresholds of 4, 6, and 8 kts were considered over a two-year period. Results showed that with an error rate less than 1%, average and median capture rates were approximately 100%, with a slight increase in rate corelated with the target wind speed.

[1] https://www.transtats.bts.gov/OT_Delay/ot_delaycause1.asp?6B2r=I&20=E

[2] Robasky, Frank M., and David A. Clark. "A Wind Forecast Algorithm to Support Wake

Turbulence Mitigation for Departures (WTMD): Baseline algorithm description and performance

summary", 2008, MITLL Memo 43PM-Wx-0105.