Fornes, H. & Balakrishnan, H. Title: Analysis of Potential Implementations of Pushback Rate Control at LaGuardia Airport. Number of words: 3536 Number of tables: 1 (250 words) Number of figures: 3 (750 words) Author 1: Name: Hector Fornes Martinez Affiliation: Massachusetts Institute of Technology Mailing address: Massachusetts Institute of Technology. Office 35-217 77 Massachusetts Avenue. 02139 Cambridge, MA USA Telephone number: (+1) 617-308-9877 Email address: hfornes@mit.edu Author 2: Name: Hamsa Balakrishnan Affiliation: Massachusetts Institute of Technology Mailing address: Massachusetts Institute of Technology. Office 33-328 77 Massachusetts Avenue. 02139 Cambridge, MA USA Telephone number: (+1) 617-253-6101 Email address: hamsa@mit.edu 1 Fornes, H. & Balakrishnan, H. 2 ABSTRACT Implementations of surface traffic management strategies at congested airports have the potential to yield significant benefits, but must account for the constraints and objectives of multiple stakeholders. This paper considers the implementation of a pushback rate control strategy at LaGuardia airport in New York. This class of control policies regulate departure push-back rates by holding aircraft at their gates during congested periods, in a manner than maintains the departure throughput of the airport. The pushback rate control strategy accounts for gate conflicts that occur between departures held at the gate and arrivals assigned to those gates, by releasing the departures when the corresponding arrivals land. Another practical concern that arises at congested airports is that of long gate-holds, which can be inconvenient to air carriers. To this end, we analyze the distribution of gate-hold times that would be expected at LaGuardia airport, and explore potential implementations of a pushback rate control policy in which gate-holds would be limited to a maximum of 10 minutes and 15 minutes. The results show that although the total benefits decrease with these constraints, significant reductions in taxi-out times are still possible, and the resultant policies maintain both efficiency (that is, gate-hold times translate to taxi-out time reductions) and fairness (that is, the share of taxi-out savings received by an airline is approximately the share of departures operated by the airline). Fornes, H. & Balakrishnan, H. 3 MOTIVATION Surface congestion at major airports results in excessive taxi-out times, and increased fuel burn and emissions. For example, LaGuardia (LGA) Airport in New York is congested 10-15% of the time, that is, the number of active flights on the surface is higher than needed to sustain departure throughput. The result is taxi-out times of more than 50 minutes, even though the unimpeded (free-flow) taxi-out time is around 15 minutes. Surface congestion management policies have the potential to reduce the taxi-out times, and therefore reduce emissions of pollutants such as Carbon Dioxide, Hydrocarbons, Nitrogen Oxides and Particulate Matter, at congested airports. The implementation of such policies would bring benefits for the different stakeholders, but practical constraints such as gate use conflicts and the length of taxi-out times must be considered. The pushback rate control strategy recommends push-back rates for departures from gates during congested periods, in order to prevent excessive congestion on the surface, while still maintaining the departure throughput of the airport. It accounts for gate conflicts that occur between departures held at the gate and arrivals assigned to those gates, by releasing the departures when the corresponding arrivals land. A practical concern for air carriers is that of long gate-holds, which can be inconvenient and also affect the on-time departure performance. To this end, we analyze the distribution of gate-hold times that would be expected at LaGuardia airport, and explore potential implementations of a pushback rate control policy in which gate-holds would be limited to a maximum of 10 minutes and 15 minutes. We evaluate the impact on the taxi-out times and gate-holds of different airlines, under the different variants of pushback rate control: Unrestricted, 10-minute, and 15-minute maximum limits on the gate-hold times. PUSHBACK RATE CONTROL STRATEGIES Background The policies presented below are variants of the N-control framework based on the virtual queuing of departures (Feron et al. 1997), (Burgain, Feron, and Clarke 2008). This framework makes it possible to define policies that reduce fuel burn and emissions at airports (Simaiakis and Balakrishnan 2009), (Simaiakis and Balakrishnan 2010). At the beginning of each time-period, the departing surface traffic is evaluated, and further pushbacks as regulated if needed. In this manner, the control policy prevents the presence of idling aircraft that do not lead to additional departure throughput; the number of taxiing aircraft and the average taxi-out times are thereby reduced. Figure 1 depicts the relationship between the number of departures active on the surface (referred to as N) and the takeoff throughput. The relation indicates that lower values of active departing aircraft on the surface lead to increments of departure throughput; however, this is only valid up to a saturation threshold after which the system achieves the departure runway capacity, and no additional flights can takeoff, regardless of how many are ready to do so. Such a saturation threshold is referred as N*, and it defines a border between the number of aircraft that will contribute to takeoff throughput and those aircraft that will not, and therefore will idle burning fuel and emitting gases. This traffic classification allows the possibility to take action when the number of departing aircraft on the surface (N) at a particular moment exceeds a threshold value related with N*. This threshold value is Nctrl, which is always larger or equal than N* (Nctrl≥ N*) and when N>Nctrl, the system requires those aircraft requesting pushback, in excess of Nctrl, to wait at the gate until the surface traffic diminishes and therefore N becomes smaller or equal than Nctrl. Fornes, H. & Balakrishnan, H. 4 Figure 1: Departure throughput as a function of the number of aircraft taxiing out, for the (VMC; 31|4) configuration at LGA. Source: (Simaiakis 2013) In the literature cited above there are two policies to go about limiting pushbacks based on Nctrl; the first requires controllers to release as many aircraft as required to reach the threshold whereas the second involves recommending a pushback rate to controllers. There are several implementation challenges associated with the former because it involves more active control intervention on the controller side compared to the latter (Sandberg et al. 2014). Definition of pushback rates The recommended pushback rates are determined based on the conservation principle applied to the taxiway system. At the beginning of each time period, the model estimates the number of aircraft active in the system until that moment (current surface traffic, curN), the number of expected aircraft that will leave the system during that time period (expected takeoff throughput), and the desired number of aircraft on the ground. The current traffic is an observed variable, the takeoff throughput is determined following the guidelines below (regression trees), and the desired level of traffic is given will by the Nctrl values. The conservation principle provides a recommended pushback rate as indicated in the following equation: Push rate= Nctrl + throughput – curN This equation is evaluated at the beginning of every time period, and the pushback rate is communicated to the air traffic controllers. Estimation of parameters In order to determine the recommended pushback rates, one needs to predict the takeoff throughput and the unimpeded taxi-out times, and decide on the length of the time-period. The takeoff throughput is determined from regression trees (Simaiakis 2013) as a function of the runway configuration, the number of arrival rates, visibility conditions (IMC or VMC) and the weather, using the Route Availability Planning Tool (RAPT) as the weather indicator (DeLaura et al. 2008). When building the model, the runway configuration, the arrival rate, and the visibility conditions are obtained from the Federal Aviation Administration (FAA) Aviation System Performance Metrics (ASPM) database; the RAPT weather Fornes, H. & Balakrishnan, H. 5 archives were provided by Lincoln Labs. During implementation, all these parameters can be either found in the monitors of the control tower (runway configuration, visibility conditions, and RAPT) or estimated based on the tower equipment (arrival rate). The availability of these variables in the tower allow for the use of regression trees during implementation. The unimpeded taxi-out time is defined by the FAA as the taxi-out time under optimal operating conditions, when neither congestion, weather nor other factors delay the aircraft during its movement from gate to takeoff. (Simaiakis 2013) did a further detailed analysis of all the factors that need to be included in the unimpeded taxi-out time calculations. This time was obtained from a convex curve-fit correlating the taxi-out time as a function of the effective surface traffic for each departing flight, defined as the sum of the aircraft taxing out at the time of the pushback of that specific flight plus the number of aircraft that push back while the specific flight is travelling to the departure runway (Simaiakis 2013). Four departure terminals and ten important runway configurations at LGA were modeled using ASPM data. The unimpeded taxi-out times obtained ranged from 11 minutes to 17 minutes. Based on an analysis of the typical unimpeded times, the length of the time-interval over which the pushback rate is determined was set equal to 15 minutes. Model structure The overall structure of the model is as follows: The model first analyzes the baseline case with empirical data, and then simulates the performance of the different metering policies proposed. The model re-simulates each metering policy resolving those additional gate conflicts caused by the metering policy. One of the main consequences of metering is an increase in the time spent at the gate. There is a chance that the next arriving aircraft to use that gate touches down while a departure is being held at the gate, which causes a gate conflict. In order to resolve such gate conflicts, the departure involved in a gate conflict are cleared for pushback immediately after the arriving aircraft lands. In this case, the model does not consider Nctrl, because the maximum average frequency of additional conflict occurrence is 0.4 conflicts per hour, which is small compared to Nctrl values. The analysis and simulations carried out rely on three datasets. First, ASPM provides data on actual arrival and departure (call for pushback) times, “wheels-on” time (arriving aircraft touch down), “wheelsoff” time (departing aircraft rotate-and-lifts), air carrier, year, month, day, runway configuration and visibility conditions. These data is complemented with terminal and gate data from Flightstats.com; these data is important to carry out the gate conflict simulation. Finally, as indicated previously, the model also uses RAPT data to have a better accuracy in accounting for weather. These models have been previously tested in field-trials at Boston Logan Airport, as presented in (Sandberg et al. 2014) and (Simaiakis et al. 2011). IMPACT OF DEPARTURE METERING ON GATE HOLDS The developed model makes it possible to evaluate the length of gate holds. Figure 2 depicts the percentage distribution of holding times for each airline. It indicates the magnitude and the distribution of the impacts of the metering policy on gate operations. This is one of the key indicators for airlines because this represents the most noticeable change in the airline surface operations. Fornes, H. & Balakrishnan, H. 6 Figure 2: Duration of gate holds by airline at LGA airport during the period July 1 2013 to August 31 2013 Figure 2 shows that for all the airlines, more than 75% of the flights do not receive any gate-holds. For the 1 to 10-minute range, between 1% and 5% of flights are held for each airline. Less than 1% of flights are held for more than ten minutes. These results suggest that the percentage of flights severely affected by metering is small, and that most of the flights receive less than a 10-minute gate-hold. Although the percentage of flights held at the gate for more than ten minutes is small, those flights that do sit at the gate longer represent an operational inconvenience for airlines. One reason is that the Department of Transportation calculates on-time performance in terms of the pushback times from the gate, and a flight is considered delayed if its actual pushback time exceeds its scheduled pushback time by more than 15 minutes. This paper therefore considers variants of pushback rate control that can enable the benefits of regulating departing demand, while minimizing the adverse impacts of long gate-holds on the airlines. The three implementation variants considered are: Policy 1: Departure metering with no limits on gate-holds, namely, the metering strategy previously presented. Policy 2: Metering with a maximum gate hold of 15 minutes. This policy keeps track of the duration of gate holds for each flight and in those circumstances when a 15-minute wait is fulfilled, that flight pushes back and starts the taxi-out process. This restricted version of metering avoids those situations represented to the right of the dotted lines for 15 minutes in Figure 2, making them move exactly to the 15-minute mark. The fifteen-minute restriction reduces the number of flights with long gate holds, which help the airline operations on the ground as they become more predictable. There is, however, a cost associated to this increase in predictability in terms of the taxi-out times. Indeed, as more aircraft pushback “unnecessarily” – from the metering perspective – the surface traffic increases to levels above the saturation threshold, Ncontrol, and therefore, there are aircraft on the surface that do not contribute to departing throughput. Hence, the taxi-out reductions are expected to be smaller than in the Policy 1 case. Policy 3: Metering with a maximum gate hold of 10 minutes. This policy is equivalent to Policy 2, with a 10-minute limit, but it represents less of a change from the baseline case and it is expected to longer taxi- Fornes, H. & Balakrishnan, H. 7 out times. In this case, the restricted metering approach would lead to avoiding those flights to the right of the dashed line in Figure 2. These three implementation policies present an interesting range of metering alternatives to evaluate. RESULTS Before proceeding to the simulation results, it is worth understanding the implications to the model of each one of the implementation policies. Figure 3 provides the reader with a graphic representation of the main conceptual outputs of the simulation as well as a visual representation of how such model outputs vary by implementation policy. Figure 3: Results of the model for July 9 2013. The first set of curves (top subplot) represent departure surface traffic under different policies. The second set of curves (positive side of the bottom subplot) depict average taxi-out time. The third set of curves (negative side of the bottom subplot) illustrates the average duration of gate holds. Figure 3 illustrates, by 15-minute period of a day, the departing surface traffic, the average taxi-out time, and the gate-holding time for the baseline and the three policies. The first observation is the clear correspondence between the three variables, as was expected. We note that there are no gate holds – represented by gate holding time different from zero – unless the surface traffic is above the N-control value, which varies by runway configuration and visibility conditions. For surface traffic values below Ncontrol, the three metering implementation policies track the baseline case. The surface traffic for any of the three implementation policies is, as expected, always equal or smaller than the baseline case. As a consequence, average taxi-out times are always equal or lower than the baseline case for every policy. Furthermore, Policy 3 (by definition is the most similar to the baseline) leads to both surface traffic and taxi-out times closer to those in the baseline; Policies 1 and 2 differ most significantly from the baseline, but also experience shorter taxi-out times. Figure 3 helps explain the conceptual relations between the departing surface traffic, the taxi-out time, and the gate holding time. However, this figure is just a snapshot of one specific date, which may not necessarily be representative of a set of days. Another aspect worth being cautious about while extrapolating these figure results is the use of averages, particularly after knowing of the existence of long distribution Fornes, H. & Balakrishnan, H. 8 tails of up to 30 minutes of taxi-out time. Indeed, averages do a better job than medians to include outliers, but there might still be a lack of representation of those extreme values. In order to carry out sound comparisons between implementation policies, it is helpful to compare the total taxi-out time reduction time by airline, the percentage taxi-out time reduction by airline, and the percentage of the aggregate time reduction each airline takes advantage of; along with the total duration of gate holding times by airline, as well as the percentage of the aggregate gate holding time that each airline encounters. Table 1 presents all these indicators for simulations of the three different policies from July 1 2013 to August 31 2013, in addition to the share of departures corresponding to each airline. Table 4: Simulation results by airline and by pushback policy. Taxi-out time reductions Airline Policy Minutes % reduction % share 1 2 3 4 5 6 7 Others 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 18,299 13,659 11,724 8,894 6,599 5,702 4,520 3,260 2,844 1,792 1,313 1,116 794 594 527 1,956 1,364 1,158 1,066 710 594 9,849 7,447 6,497 4.9% 3.6% 3.1% 5.0% 3.7% 3.2% 5.3% 3.8% 3.4% 4.8% 3.5% 3.0% 4.6% 3.4% 3.1% 4.6% 3.2% 2.7% 4.2% 2.8% 2.3% 5.1% 3.9% 3.4% 38.8% 39.1% 38.9% 18.9% 18.9% 18.9% 9.6% 9.3% 9.4% 3.8% 3.8% 3.7% 1.7% 1.7% 1.7% 4.1% 3.9% 3.8% 2.3% 2.0% 2.0% 20.9% 21.3% 21.5% Gate holding time Minutes % share 18,294 13,752 11,873 8,572 6,460 5,576 4,957 3,360 2,869 1,776 1,266 1,096 668 569 503 1,851 1,323 1,142 1,000 718 628 10,053 7497 6476 38.8% 39.4% 39.4% 18.2% 18.5% 18.5% 10.5% 9.6% 9.5% 3.8% 3.6% 3.6% 1.4% 1.6% 1.7% 3.9% 3.8% 3.8% 2.1% 2.1% 2.1% 21.3% 21.5% 21.5% % share of departures 38.0% 19.8% 8.2% 4.3% 2.0% 5.1% 3.1% 19.6% There are several observations worth noting. First, as expected, airlines obtain the most significant taxi-out time reductions with Policy 3, at the cost of having the longest taxi-out times. Policy 1 leads to the opposite results; this is valid for all airlines. The taxi-out time reductions from Policy 1 range between 4.2% and 5.3%, those from Policy 2 are between 2.8% and 3.9%, and from Policy 3 are between 2.3% and 3.4%. Second, one minute of gate-hold leads to approximately one minute reduction in taxi-out time. This result is important because it ensures that no additional departure delay is added, and that instead of idling on the taxiway, the waiting time occurs at the gate. Third, the advantages (taxi-out reductions) and impacts (gate-holding times) of the model are approximately distributed in a proportional way to the market share of each airline (in terms of the number Fornes, H. & Balakrishnan, H. 9 of operations). Additionally, these proportions are rather well-maintained for the three policies (+/- 2%), illustrating the fairness of the approach. CONCLUSIONS This paper considers modifications to a pushback rate control strategy previously field-tested at Boston, in order to ease the practical implementation challenges that arise at a more chronically congested airport such as LGA. Policy 1, which does not limit gate-hold times, achieves the largest reduction in taxi-out times, and therefore fuel burn and emissions. In addition to this, fewer aircraft on the ground reduces the number of interactions between arriving and departing aircraft on the surface, potentially improving safety. Airlines will see fuel burn reduction from this type of implementation policy, but at the same time may be concerned about operational challenges on the ramp due to lack of predictability. Policies 2 or 3 help alleviate some of these practical challenges for airlines by limiting the gate hold times. Simulations of all three variants and the baseline operations show that all three variants can be implemented in such a manner that the gate-hold times are commensurate with the reductions in taxi-out times. In addition, gate-holds and taxi-out time savings in all three variants are allocated in proportion to the number of departure operations amongst the different air carriers, thereby ensuring fairness. REFERENCES Burgain, Pierrick, Eric Feron, and John-Paul Clarke. 2008. “Collaborative Virtual Queue: Fair Management of Congested Departure Operations and Benefit Analysis.” arXiv Preprint arXiv:0807.0661. http://arxiv.org/abs/0807.0661. DeLaura, Rich, Michael Robinson, Russell Todd, and Kirk MacKenzie. 2008. “Evaluation of Weather Impact Models in Departure Management Decision Support: Operational Performance of the Route Availability Planning Tool (RAPT) Prototype.” In 13th Conference on Aviation, Range, and Aerospace Meteorology, AMS, New Orleans, LA. https://ams.confex.com/ams/pdfpapers/132880.pdf. Feron, Eric R., R. John Hansman, Amadeo R. Odoni, R. B. Cots, B. Delcaire, W. D. Hall, H. R. Idris, A. Muharremoglu, and N. Pujet. 1997. “The Departure Planner: A Conceptual Discussion”, December. http://dspace.mit.edu/handle/1721.1/34944. Sandberg, M., I Simaiakis, H. Balakrishnan, T.G. Reynolds, and R.J. Hansman. 2014. “A Decision Support Tool for the Pushback Rate Control of Airport Departures.” IEEE Transactions on HumanMachine Systems 44 (3): 416–21. doi:10.1109/THMS.2014.2305906. Simaiakis, Ioannis. 2013. “Analysis, Modeling and Control of the Airport Departure Process”. Massachusetts Institute of Technology. http://dspace.mit.edu/handle/1721.1/79342. Simaiakis, Ioannis, and Hamsa Balakrishnan. 2009. “Queuing Models of Airport Departure Processes for Emissions Reduction.” In AIAA Guidance, Navigation and Control Conference and Exhibit. http://arc.aiaa.org/doi/pdf/10.2514/6.2009-5650. Simaiakis, Ioannis, and Hamsa Balakrishnan. 2010. “Impact of Congestion on Taxi Times, Fuel Burn, and Emissions at Major Airports.” Transportation Research Record: Journal of the Transportation Research Board 2184 (-1): 22–30. doi:10.3141/2184-03. Simaiakis, Ioannis, Hamsa Balakrishnan, Harshad Khadilkar, Tom G. Reynolds, R. John Hansman, Brendan Reilly, and Steve Urlass. 2011. “Demonstration of Reduced Airport Congestion through Pushback Rate Control.” http://18.7.29.232/handle/1721.1/60882.