ModelingAirport Arrival and DepartureUnimpeded Taxi Times
Chad Long and Steven Landry
(Purdue University)
(email: )
The study of taxiway movements and ground operations is becoming more and more important as air travel continues to expand. Unlike many studies, this research examines the stochastic nature of ground operations to include taxi speeds and gate distance from the runway. While conducting our research at a medium size international airport, we found that the unimpeded taxi time for arrivals was well predicted by distance. This was not the case for departure unimpeded taxi times. Distance did impact departure time, but the unimpeded taxi time cannot be predicted by distance alone.
KEY WORDS
- Aviation ground operations. 2. Taxi time.
1. INTRODUCTION. There are currently a number of gaps in the research on taxiway movements. The first gap that this article will address is the lack of a validated model for aircraft unimpeded taxi times. Next, this article will examine the lack of understanding on what factors influence taxi times.
The most obvious gap in the research on airport operations is the lack of a validated model for airport unimpeded taxi times. The majority of models that examine airport taxiway operations primarily focus on queuing operations (Wieland, 2006). While this is very important to large commercial airports, the vast majority of airports do not experience a significant amount of queuing in their ground operations. In our research on the topic of validated models for taxi times, we located an unpublished master thesis from the University of Toronto by Yuval Grinspun (2002). His research on taxi times was the most thorough model on the topic. We ultimately tested his model against empirical taxi data in order to validate his results.
In examining the research on taxi times, there is no clear understanding on the influences of taxi time. As mentioned earlier, some models include aircraft interactions, such as queuing (Anagnostakis, I., Idris, H., Clarke, J.P., Feron, E., Hansman, R., Odoni, A., & Hall, W., 2000; Carr, Evans, Clarke, & Feron, 2002; Idris, Clarke, Bhuva, & Kang, 2002), while other models are just based on the physical characteristics of either the airfield or the taxing aircraft (Atkins, Jung, Brinton, Stell, Carniol & Rogowski, 2004; Carr, 2004; Hebert & Dietz, 1997; Kang & Clarke, 2000; Levy & Rappaport, 2007). Grinspun (2002) used the physical characteristics such as distance from runway to terminal, number of turns on taxi route, and aircraft size in his model. Other models were trajectory-based and required identifying factors, such as queuing, that affected taxi times to make an accurate prediction.
This study was completed to fill some of the current gaps in airport taxi research. To do this, we generated a network model for a medium size airport. This network model accurately depicted the taxiway distances and number of turns from the runway to each gate position. These routes were used in conjunction with actual taxi times collected in a field study. The actual times were compared to the Grinspun model. The results raised questions as to what factors influenced the taxi times. To answer those questions, a second field study was conducted a medium size airport with a different taxi route layout. This test was used to validate the second model.
There are a number of motivational aspectsfor this study. The first aspect is to create a process to determine the unimpeded taxi times at different airports/ runway configurations. This taxi time would be independent of queuing operations. The second portion is to create a model that would predict the time it would take for an aircraft to reach the runway with a departure queue. As ground controller operations become more advanced, the use of unimpeded taxi times could become more significant for queuing-based airports. The ground controller may use the unimpeded taxi time as a factor to coordinate and stagger departure times, ultimately reducing the time spent in the departure queue. This process would save the airlines fuel and reduce customer dissatisfaction. The third feature would be the use of this model to predict taxi times at medium and small airports where queuing is uncommon. The last aspect is to provide information that could be used to improve an airport layout. Changing an airport’s layout will primarily impact unimpeded taxi time, but might not significantly change the airport’s queuing times.
This study provides an empirical analysis of arrival and departure taxi times in an effort to dissect the characteristics of the unimpeded taxi time. The following background section discusses relevant literature giving a detailed discussion on previous ground operation models. The method and results sections review the process used in the study and outcomes of the research. The discussion section interprets the findings provided in the results section.
2. BACKGROUND. The study of taxiway movements and ground operations is becoming more and more important as air travel continues to expand (Evans, Schäfer, & Dray, 2008). Unfortunately, past studies of air transportation systems have not taken ground operations into account (Andersson, Carr, Feron, & Hall, 2000). Most studies have bypassed the stochastic nature of taxiway operations and assumed a constant interval for ground movements, ignoring queuing times, taxi speeds, and distance from runway to terminal. The few studies that examine ground operations have primarily focused on queuing operations (Andretta, 2000; Shumsky, 1995;Wieland, 2006). These queuing studies consider unimpeded taxi times to be a minimal factor compared to the time spent in departure queues. Andersson et al (2000, pg.5) explains this thought process. “Once an aircraft reaches the runway, it usually enters a runway queue, and its position in the queue becomes fixed. The airport throughput is primarily limited by this bottleneck effect at the runway (Idris, Delcaire, Anagnostakis, Hall, Pujet, Feron, Hansman, Clarke, & Odoni, 1998).” These departure queuing studies have been completed primarily at large hub airports such as Boston (BOS), Atlanta (ATL), and Detroit (DTW) where departure delays can be extreme. Most airports throughout the country do not have a lengthy departure delay, so a queuing based model would not be appropriate.
Ground operation studies have traditionally classified airports into two types. The first has minimal departure queuing when taxiing from the terminal to the runway and the second has departure queuing as the primary factor for taxi times (Andersson et. al., 2000). There are very few airports in the United States that would be classified as departure queue limited airports. These airports are predominately only the large airports with heavy commercial traffic. (Pujet, Delclaire, & Feron, 2000). The taxi time for the majority of airports in the United States is determined by unimpeded taxi time. There currently is no single model for determining unimpeded taxi times. In an effort to create a general model, this study examined the unimpeded taxi times for arrival and departure at a moderately trafficked, international airport. The airport was selected in part because it exhibited minimal departure queuing.
2.1. Current Methods for Determining Unimpeded Taxi Time.When studies need to determine taxi times, two methods are commonly used (Grinspun, 2002). The first method uses the Consolidated Operations and Delay Analysis System (CODAS) database. The Federal Aviation Administration (FAA) computes unimpeded taxi time using a statistical analysis of existing data. The regression equation used to determine unimpeded taxi time is (Boeing 1997, pg. 26):
TOa,c,s = b1 TOQa,c,s + b2 TIQa,c,s + c
Where
TO =Taxi-out time
TOQ = number of aircraft in taxi-out Queue
TIQ=number of aircraft in taxi-in Queue
a=airport
c=carrier
s=season
b1=coefficient for TOQ
b2=coefficient for TIQ
c=constant
It is apparent that the CODAS taxi-out time is primarily dependent on the number of aircraft in queue. For example, if there are 3 aircraft taxiing to the runway (pushed back from the gate, but not yet taken off) and there are no landing aircraft (wheels-on deck), then the TOQ would be 3 and the TIQ would be 0. The unimpeded taxi-out time would be captured for a TOQ of 1 and TIQ of 0. The number of aircraft in a queue is derived through a minute-by-minute analysis of actual flights in the Airline Service Quality Performance (ASQP) database. The actual taxi times are adjusted before being used in the regression.
In estimating the unimpeded taxi-out times, the highest 25 percent of the values of actual taxi time were excluded from the regression. This step was taken to remove the influence of extremely large taxi-out times from the estimation of expected taxi time under normal operating conditions (Boeing, 1997, pg.27).
The estimation of unimpeded taxi-out times has three important limitations. The first is the impact of the use of different runway configurations. There is no data collected for the runway configuration for each flight, so the actual taxi-out times and unimpeded taxi-out times are a compilation of all configurations. The second limitation is that OOOI data is not available for all aircraft. OOOI times are the collection of each flight movement: gate departure (gate-out or OUT), takeoff (wheels-off or OFF), landing (Wheels-on or ON), and gate arrival (gate-in or IN). Since unimpeded taxi times are based on a TOQ of 1 and TIQ of 0, for OOOI data a number of aircraft may be in the queue and the taxi may be impeded. The third limitation is that ATC delays, such as receiving an IFR clearance, are not accounted for in the CODAS calculations. The unimpeded taxi time may include 10 minutes of ATC delay. It is possible that the removal of the highest 25 percent of the actual taxi times also removes all times with ATC delays, but this is difficult to determine and probably airport specific.
The second method to determine unimpeded taxi time is to use aircraft taxi speed and distance travelled. This is a straight forward method, and there are many other variables that could impact the final taxi speed such as weather, number of actively moving aircraft on the ramp, number of turns, and airframe size. One of the earlier studies on average taxi speed was completed at Tokyo International airport in 1975. This study showed that there was a relationship between the distance travelled and the average taxi speed.The study’s results were that the average taxi speed is “16 times 6 times the square root of the taxi distance, the taxi distance being the distance from the point of entry on the taxi-way to the end” (Nagoaka, Muto, & Yoshioka, 1978, pg 494). While the results are interesting, they are not relevant to the current aviation fleet of aircraft. Nagoaka’s study assumed that the maximum taxi speed for an aircraft was 20 knots. This 1978 study maximum speed is 4.5 knots under the average speed found in Grinspun’s 2002 study.
Grinspun’s 2002 study found the average taxi speed for heavy or medium sized aircraft to be approximately 24.5 knots. To determine this speed, Grinspun collected actual unimpeded taxi time data, route distances, and number of turns at Toronto International Airport. He then completed a regression stratifying the data according to aircraft size. His study found the estimated taxi speed and approximate time to turn for heavy, medium, light and prop aircraft.
Grinspun (2002) completed an in-depth study on the relationship between somenon-queuing variables and their impact on taxi times. He found that the number of turns an aircraft must take from runway to gate impacts the final taxi time. His study also found that different size aircraft taxi at different speed.For example, he found the average taxi speed for a light or propeller plane was 26 km/h while a heavy or medium size aircraft travel at approximately 45.5 km/h (Grinspun, 2002, pg 43). Grinspun’s model is one of the most recent and in-depth study on unimpeded taxi times. His model can easily be adapted to any airport by creating a network of the airport layout.
3. METHOD: APPLICATION OF GRINSPUN MODEL. A field study was conducted at a medium-sized international airport on October 9, 2008 from 9:30 AM to 2:00 PM to obtain data to validate the Grinspun model. The observation location gave the observers an unobstructed view of the runways and most gate traffic. The gates that were unobservable used an intersection of taxiways as their starting or ending points. The weather during this observation was VFR (visual flight rules) and there were no abnormal weather conditions noted around the country. The data collected from the time study included: taxi route, distance travelled, runway usage, gate usage, number of turns, arrival/departure, aircraft type, airline type, and time stopped for queuing.
The first step in validating the model was to create a network diagram of the airport. The network diagram is composed of a number of nodes and connecting arcs. The nodes are positioned at every intersection so the taxi route to the different terminals and runways can be identified. Each taxi path is identified by a collection of links between nodes. From this, a taxi route’s distance and number of turns can be calculated. Figure 1 is the network diagram for this study.
Figure 1. Airport network diagram (end-field terminal location).
The data for unimpeded taxi times was divided into arrivals and departures. For arrivals, the time was documented from the runway exit point until the aircraft stopped moving at the gate. For departures, the taxi time was documented from aircraft movement after pushback to runway entrance. There were a number of potential delays that could impact departing aircraft, including air traffic control (ATC) delays, departure (takeoff) queue, and taxiway congestion. All delays were removed to determine unimpeded taxi times.
The Grinspun model is essentially a regression using aircraft type, taxi distance and number of turns, where the coefficient for taxi distance is the average taxi speed. Given the input set obtained from the field study and airport network model, the regression was applied to determine the unimpeded taxi time. The results of the regression were compared to the actual unimpeded taxi time in two ways: examining the prediction errors and comparing coefficients with a linear regression computed from the actual data.
4. RESULTS. The study observed taxi times for 44 arriving and 42 departing aircraft. This included 52 regional jets and 34 mid-size jets. The aircraft arrived on 23R and departed on 23L. All aircraft took common taxi routes once they were on the main taxiways.
The data for unimpeded taxi times were divided into arrivals and departures. For arrivals, it was noted that on four occasions that aircraft were required to wait for gate employees before it could approach the gate. These minimal delay times were subtracted out to get the unimpeded taxi times. There were no delays caused by an aircraft occupying the gate of an arriving aircraft.
For departures, the taxi time was documented from aircraft movement after pushback to runway entrance. There were a number of delays that impacted departing aircraft. Nine of the 42 departing aircraft were delayed by air traffic control (ATC) prior to departure. ATC delays were obvious to the observers because the aircraft stopped in a specific ATC parking area. This delay ranged from as little as 162 seconds to as much as 1,625 seconds. The average ATC delay for all departing aircraft (including those with a zero second delay) was 174 seconds. In addition to the ATC delay, three additional delays were imposed on some of the departing aircraft. Five aircraft were delayed in a departure (takeoff) queue, four aircraft stopped on the taxiway for unknown reasons, and one aircraft was delayed for taxi congestion. All of these departure delays were of short duration and were removed to determine unimpeded taxi times.
Grinspun’s model uses three variables to determine unimpeded taxi time: aircraft size, aircraft speed, distance travelled, and number of turns. The regional and midsize jets observed in the field study were classified as medium sized aircraft in Grinspun’s study due to similar taxi speeds. The airport studied did not have a mid-field terminal, so there was a significant difference in distances for arriving and departing aircraft.
Figure 2 is a scatter plot of distances and actual unimpeded taxi times. The arrival distances were much longer than the departure distances, which is apparent on the plot. Another apparent observation is that there appears to be linear relationship between arrival taxi times and distance taxied. This relationship does not seem to exist for departure taxi times and distance.