Opportunities for improvements in simple models for estimating runway capacity

Arrival rates may be misestimated by Monte Carlo simulation models that assume uniform distributions for landing speeds (V_g) and the inter-arrival distance spacings (δ) between successive arrivals, based on the wake vortex weight classes of the paired aircraft. Analysis of arrival data measured at Memphis International Airport (MEM) shows that the V_g and δ values are not uniformly distributed. Over 100,000 arrivals at MEM from 10/01/2002 to 07/31/2003 were available for analysis from a database developed by Sensis Corporation to support operations research on the airport surface and terminal area. The database was established as part of a NASA/Volpe National Transportation Systems Center-funded project known as the Dynamic Runway Occupancy Measurement System (DROMS). Data on V_g and δ at the runway thresholds were measured under visual meteorological conditions (VMC). The δ data are distributed according to Johnson probability density functions (pdf), with a positive skew and finite lower bounds as a result of minimum safe separation standards. Based on observed distance spacing data, the Johnson pdf implies a greater arrival rate than the uniform distribution due to an increased frequency of small δ values. Use of the uniform distribution with observed spacing data can underestimate the mean arrival rate from other models by more than 7 arrivals per hour. A regression equation for average hourly arrival rate to a single runway is given as a function of the maximum inter-arrival time interval between successive arrivals to the same runway. This research shows that the arrival rate estimates based on independent generation of lead/trail aircraft pairs causes a maximum underestimation of 0.6 arrivals per hour compared to conditional generation of weight pairs. The conditional simulation of weight pairs correctly describes the observed frequencies of the weight pairs. A procedure for collecting independently distributed landing speed and distance spacing data was developed. The landing speed of the trailing aircraft depends on the weight of the leading aircraft and on the meteorological condition when the aircraft landed. Generation of random variates for landing speed and distance spacing in sim- ulation must be on a pair-wise basis.