Impact of General Aviation Operations on Airport Performance Through Fast-Time Simulations at Charlotte Douglas International Airport

Impact of General Aviation Operations on Airport

Performance Through Fast - Time Simulations at

Charlotte Douglas International Airport

Zhifan Zhu

KBR Wyle Services, LLC , NASA Ames Research Center , Moffett Field, CA 94035, USA

Vaishali A. Hosagrahara

Tellus Solutions , NASA Ames Research Center , Moffett Field, CA 94035, USA 3 4

Hanbong Lee , Yoon Jung

NASA Ames Research Center , Moffett Field, CA 94035, USA

Deb orah L. Bakowski

San Jos é State University , NASA Ames Research Center , Moffett Field, CA 94035, USA NASA , in partner ship with the Federal Aviation Administration and commercial airlines, deployed the Integrated Arrival, Departure, and Surface (IADS) traffic management system at Charlotte Douglas International Airport (CLT) in 2017 for field evaluation . The system features new capabilities of data exchange and integration, collaborative surface metering and scheduling , which has demonstrated operational benefits of reducing delay time and fuel consumption . General Avi ation flights , however, are not currently included in the surface metering program s because of their different operational procedures and lack of reliable predictability in departure time . Thus , their impact on the system performance is not well understood . T his paper presents a study of the impact of General Aviation traffic at CLT using fast - time simulations.

I. Introduction In the United States there are more than 200,000 General Aviation (GA) a ircraft in active use (including Part 91 and Part 135 operations) that are estimated to log more than 20 million flight hours annually [1] . They are used for a variety of purposes including personal, business, and instruction . The number and proportion of GA operations varies by airport. A t Charlotte Douglas International Airport (CLT) , for instance, an average of 30 departure and 30 arrival flights per day account for about 4 % of all airport operations .

The numbers are based on a three - month operation data set from July to December in 2019 . As a comparison, the daily number s at Dallas Love Field Airport (DAL) are over 60 each , and accounts for 20% of airport operations . Whereas the overall proportion of GA operations at CLT is small, the potential exists for some of these GA flights to impact surface operations. For example, they may compete w ith other air traffic, including commercial and c argo flights, for airport resources like runways. GA flights have less predictable departure schedules than commercial operations and , at CLT, do not have communication with the ramp tower before taxi, presenting a unique challenge for trajectory - based sur face operations.

A. Integrated Arrival, Departure, and Surface (IADS) Traffic Management System In September of 2017, NASA, in close collaboration with the Federal Aviation Administration (FAA) and industry, deployed the Integrated Arrival, Departure, and Surface (IADS) traffic management system at CLT as part of Airspace Technology Demonstration 2 (ATD - 2 ) project for a three - year field evaluation. T he IADS system has demonstrated operational benefits through its capabilities of data exchange and integration, collaborative surface metering, departure scheduling for overhead stream insertion, and real - time metrics reporting [2] . The objective of the IADS system is to balance demand and capacity to achieve more efficient traffic movement on the surface and overhead stream insertion .

Sr. software en gineer, KBR Wyle Services, LLC Sr. software engineer, Tellus Solutions Airspace engineer, Ames research center, AIAA member Airspace engineer, Ames research center, AIAA senior member Sr. Research Associate, San Josè State University Surface metering is used when t he predicted demand exceeds the airport capacity during departure banks . Holding flights at the gate transfers excess taxi time from the departure runway queue back to the gate , prior to engine start . As of March 31, 2020, it is estimated a total of 3,83 2 hours of engine run time reduction that resulted in 5,075,981 pounds of fuel savings and 15,634,022 pounds of CO 2 emission reduction [3] .

Surface scheduling and metering is enabled by the availability and quality of flight readiness time. Major airlines participa ting in the ATD - 2 field evaluation provide an Earliest Off - Block Time (EOBT) for each departure flight. The EOBT represents the bes t estimate of a flight’s earliest pushback time. A good quality EOBT helps the scheduler achieve accurate demand prediction at each runway and hence, achieve reliable scheduling to keep the surface less congested while maintaining overall airport throughpu t. Along with other data inputs, the surface scheduler uses the EOBTs to predict takeoff times of departure flights and calculate recommended gate holds.

Gate hold advisories are communicated to Ramp Control through the ATD - 2 Ramp Traffic Console (RTC) int erface.

Airlines are motivated to improve the accuracy of EOBT data for the benefit of reduced excess taxi time and fuel savings achi eved through surface metering. A preliminary study of EOBT quality impact on surface metering [4] [5] shows that the EOBT accuracy has a direct impact on the length of the departure gate hold time assigned by ATD - 2’s surface scheduler.

To date, GA operations have not been included in surface metering at CLT. One reason is that, unlike commercial flights at CLT’s main ramp, p ilots departing from the GA ramp area do not contact the Ramp Tower prior to taxi . A t the main ramp , the Ramp Controller communicate s gate hold advisories for surface metering . In contrast, GA flights’ first communication is with the Ground Control ler at the FAA’s Airport Traffic Control Tower (ATCT) . Other challenges in GA operations include less - predictab le departure times and the ability to communicate up - to - date readiness and intent information to the ATD - 2 scheduler . Excluding GA flights before they enter the Airpor t Movement Area (AMA) from surface metering and scheduling may have a potential impact on the efficiency of airport operations that has not been investigated .

B. Mobile App Technology Enables Two - Way Communication with GA Operations A mechanism is needed to enable two - way communicat ion between GA operations and the ATD - 2 scheduler prior to their entry into the AMA . In response to that need, T he MITRE Corporation has developed a prototype Mobile App lication (App) to enable two - way information sharing between GA pilots and NASA’s IADS traffic management system [6] [7] .

MITRE has conducted a field test of their prototype Mobile App technology at CLT with a small group of Corporate Flight Operators.

Using a mobile device, pilots submit a ready time to the ATD - 2 scheduler . This ready time, referred to as a “Ready - to - Taxi Time”, or RTT, represents the pilot’s best prediction of when they will arrive at the edge of the GA ramp area and be ready to contact Ground Control for their taxi clearance to enter the AMA . The RTT submitted by GA pilots is analogous to the EOBT submitted by airlines for departure flights at the main ramp. It provides ATC and the ATD - 2 sched uler with more accurate and up - to - date departure readiness information . In return, the GA pilot receives flight - specific schedule information , generated by the ATD - 2 scheduler , such as assigned runway and expected takeoff tim e, and information about traff ic management restrictions in place . Fig. 1 shows an example timeline of how a GA pilot may incorporate using the Mobile App to submit readiness informa tion into their pre - departure procedures.

Fig. 1 Example t imeline showing how a Corporate GA pilot may incorporate the Mobile App into their pre - departure procedures.

MITRE’s field demonstration has shown that M obile A pp technology can enable two - way communication flow between GA operators and surface schedulers . However, the com munication procedures and requirements related to surface metering for GA flights have not yet been explored . Although not investigated in this fi eld test, Mobile App technology has the potential to support the inclusion of GA flights in surface metering. For example, Mobile App technology might be used to provide metering hold advisories to GA pilots.

Although this research has included primarily business aviation (BA) flights, other GA aircraft, cargo , and military flights have the same communication limitation and similar impacts on airport performance. They are included in the GA flights in this study.

C. Research Objective Because of th e GA flights ’ unique operation al challenges , their impact on surface operations at CLT, particularly during surface metering , has not been fully understood . Research questions include: 1) whether GA flights have an unintended advantage over commercial airl ine flights which adhere to metering advisories at the gates? and 2) within a traffic bank , what GA traffic concentration affects the airline flights the most? Since the grand scheme of surface metering depends on quality of predictability for performance , q uestions exist around whether including GA flights with RTT data in surface metering might improve overall runway throughput, taxi times , delays , and fuel consumption .

Note that the concept of s urface metering “gate holds”, as applied to commercial flights at a parking gate, is that the aircraft is held at the gate, prior to push back. How it would be applied in GA operations has yet to be explored. It may mean that a GA flight holds somewhere in the GA ramp area, and possibly, delays engine start. However, for the purpose of this simulation, the term “gate hold” is applied generally to GA operations to mean they were held prior to entering the AMA. Specific procedures for applying “gate h olds” in GA operations are beyond the scope of this simulation.

This paper investigates the impact of GA flights on the efficiency of airport operations using IADS’s surface metering programs at CLT. The study uses fast - time simulations and explore s two sets of va riables. The first is the GA traffic demand that is quantified by the number of GA flights in the overall flight traffic at the airport and their flight times during traffic peak hours. The other variable is the GA flight readiness information, RTT, includ ing provision and accuracy.

T he paper is organized as follows : Section II describes the simulation environment including system setup, traffic scenarios, metrics and measurements , and the RTT model development. Section III presents the simulation results using the given scenarios and RTT model. The paper concludes with a summ ary in Section IV .

II. Simulation environment description A. A irport The simulations use d the CLT airport in a south simultaneous traffic flow configuration . Fig. 2 shows the airport layout at CLT .

While Runway 18R is for arrival only , Runways 18C and 18L are dual - use runways for both departures and arrivals in this runway configuration . A Fixed Base Operator ( FBO ) , where some GA/BA passengers meet their pilot , is located at the General Aviation Ramp on the east side of the airport , shown in the green box in Fig. 2.

Fig. 2 CLT airport diagram Fig. 3 Fast - time simulation setu p B. Simulation setup The simulations in this paper use the same configuration that was built and validated in the previous study about the impact of EOBT uncertainty [4] with the addition of a newly developed RTT model for GA flights , as shown in Fig. 3 . NASA’s Surface Operations Simulator and Scheduler (SOSS) [8] was used for these simulations . SOSS connects to the ATD - 2 T actical Surface S cheduler through the Surface Modeler. The traffic scenarios are the inputs to the simulator and define the GA tra ffic level and demand density, as well as scheduled airline flights . The EOBT Model provides the EOBT updates for the airline aircraft in the main terminals . The RTT Model provides the estimated readiness times and accuracy for GA flights. The Surface Mo deler used for fast - time simulations is adapted from ATD - 2’s Surface Model used in real - time field demonstration , which is used for data exchanges and integration . One of the core concepts is to group aircraft by their prioritie s and predictability of flight readiness in scheduling [9] . When the predicted traffic demand exceeds the runway capacity and the predicted excess taxi ti me is above a preset threshold, surface metering is triggered. The Tactical Surface Scheduler schedules flights from high - priority group to low - priority group. In the simulations, both a irline departures at gat es that provide EOBT updates and GA aircraft providing a n RTT estimate at stand are placed in the same high - priority group. The GA aircraft with no RTT estimate are placed in the low - priority group for scheduling priority and are exempt from surface metering, and, therefore, start ta xiing from the parking stand whenever they are ready, emulating the current GA ramp operation without Mobil e App communication.

C. Simulation configurations In this study, four simulation configuration groups are considered. They are designed to support two sets of simulation variables as described in the research objectives . The baseline traffic scenario for all groups is derived from actual operation data from Bank 2 traffic at CLT on February 14, 2018. T here is a total of nine traffic banks daily at CLT. Bank 2 is one of the busiest banks in the morning traffic. The typical traffic pattern in Bank 2 is that the departure demand peaks before the arrival demand starts to build up .

In this baseline scenari o, there are 85 airline departures and 83 airline arrivals , respectively. Of the 85 departures, 43 take off from Runway 18C and 42 from 18L , respectively . Of the 83 arrivals, 5, 30 , and 48 land on Runways 18C, 18L and 18R, respectively.

1. Simulation Group 1: GA flight request distribution The purpose of the first simulation group was to investigate the impact of GA flights ’ requested departure time distribution . Four different departure time distribution patterns were examined, with each s cenario having ten GA flights : 1) uniformly distributed over the bank, 2) concentrated in the beginning of the bank, 3) concentrated in the middle of the bank, and 4) concentrated in the end of the bank. GS flights in this simulation group did not provide RTTs and so were exempt for surface metering. It was expected that GA traffic demand at the peak time, i.e., case 3, would have the largest impact on the airport performance in a surface metering situation .

2. Simulation Group 2: Number of GA flights The sec ond simulation group increases t he number of GA flights uniformly distributed in the bank to examine any adverse impact to airline flights, such as increase of taxi times or gate hold times ( when metering is on). As with Group 1, GA flights did not provide RTTs.

3. Simulation Group 3: RTT percentage In the third simulation group , a certain percentage of GA flights were assumed to provide RTT s with similar accuracy as observed in MITRE data described in section E . The percentage of GA flights providing RTTs was varied for ten uniformly distributed GA flights.

Those GA flights submitting RTTs are assigned the same scheduling priority as the airline flights that have EOBTs at main terminals and are expected to comply wi th any assigned metering hold at the parking stands . The performance ( e.g., taxi time ) of the GA flights will be examined together with airline flights.

4. Simulation Group 4: RTT accuracy In the last simulation group , all ten uniformly distributed GA flight s were assumed to provide RTTs with varying accuracy. The RTT model generates the various accuracy levels used in the simulations . The expectation was that as the RTT accuracy improves, taxi time reduction will occur for both airline and GA flights during surface metering .

D. Performance metrics To analyze the impact on the efficiency of airport operations , a set of performance metrics were considered . They include d gate hold time of meter ed GA and airline flights, time series of gate hold time, taxi out time of GA and airline flights, time series of taxi out time, and runway throughput.

E. RTT model In this section, we describe how the RTT accuracy model is developed using a data driven appro ach. For this work, we have used the data collected by MITRE for GA flights that departed from CLT between October 20 th , 2017 and September 17 th , 2019. The proposed RTT model is a combination of two probabilistic quantities: 1) the timing of the updates up to the Actual RTT (ARTT), and 2) the accuracy of the updated RTTs as a function of the timing. The proposed statistical model that generates RTT accuracy i s used in the fast - time simulation for evaluating the impact of RTT accuracy on airport surface oper ations.

1. RTT Data Analysis GA /BA flight data collected by MITRE w as used for the RT T data analysis. Out of 1,656 GA flights in this data set , 301 flights submitted at least one RTT and had AMA e ntry t im es from Airport Surface Detection Equipment, Model X (ASDE - X) . The AMA - entry time represents the GA flight’s Actual RTT, that is, the time they arrived at the edge of the GA ramp area. In this analysis, two main variables were analyzed: 1) RTT accuracy, and 2) RTT update interval. The RTT accuracy is defined as the difference between the A RTT and pilot’s last RTT estimate. The RTT update interval represents the tim e d ifferen c e between two RTT updates of the same flight, or between the last RTT update time and ARTT .

Fig. 4 RTT accuracy changes over time Fig. 4 shows the RTT accuracy changes in the lookahead time window [ - 30 min, 0 min] as time approaches the ARTT . The plot represents the RTT estimate values submitted by the pilot s within the lookahead time window. This plot shows that in general, the RTT errors tend to decrease to negative values as the time progresses , which implies that the RTT prediction becomes more conservative, as it approaches ARTT . The m ajority of the GA flights (85%) submitte d their RTT s once ( dark red markers ), approximately 10% GA flights submitted two RTT updates (red color ), and about 5% provided more than two RTT updates.

2. RTT Model Development For the RTT model development, a two - step approach was used, which was introduc ed in the EOBT modeling in [4] . First, the RTT update times of each flight were modeled within a [ - 30 min, 0 min] lookahead time window as the time approache d ARTT. Then, the RTT accuracy was modelled at each update time ca lculated in the first step. The model produce d RTT accuracy distribution similar to the actual RTT error distribution using the probability distributions provided by commercial software libraries like Apache C ommons Math package [10] and MATLAB [11] .

Fig. 5 Number of RTT updates Fig. 6 RTT update time Fig. 7 RTT accuracy distributio n In the first step , the following two variables were considered : 1) the number of RTT updates per flight (referred to as PD1), and 2) the time distance from the reference time (i.e., ARTT) to the time when the RTT value is updated (referred to as PD2) within the [ - 30 min, 0 min] lookahead time window . To model the num ber of times the RTTs are updated for a single flight, the log - normal probability distribution, which has two parameters ( μ and σ ) , was selected since it best matched with the actual data among various distributions tested . The log - normal distribution is a continuous probability distribution; hence, the values were rounded to the closest integer for computing the number of RTT updates .

For the RTT update time distance modeling, the W eibull distribution was selected. In th e RTT model, the time distance between two updates can be modelled as a ‘size’ in time, given the number of RTT updates obtained from the log - normal distribution. The Weibull distribution, which has only two parameters (A and B), showed the best fit to the RTT data samples among various continuous distributions available in the Apache Commons Math package and MATLAB. From the actual data analysis, probability distributions of these two values were fit , as shown in Figs . 5 and 6. The histogram bar graphs (gr een) and regression curves (red) represent the actual data and the modelled distributions, respectively. The RTT update times, X , for each flight can be modelled using the values k randomly sampled from these two distributions and expressed in Eq. (1).

X = - random(PD2), k = 1,2,…, random(PD1) (1) k In the second step, a linear regression model was developed , as expressed i n Eq. (2), to fit the average accuracy trend along the lookahead time. In (2), x and y represent the lookahead time with respect to the ARTT time and the mean RTT error (ARTT - RTT) at x , respectively, with the regression coefficients, c 0 and c 1 .

y = c 0 + c 1 * x (2) Next, a sequence of probability distributions was fitted to the actual RTT error data (referred to as PD3). The l ogis tic probability distribution model was the closest to the actual RTT error distribution provided by the actual data . Fig. 7 shows the h istogram of actual RTT accuracy value s, and the red line in the chart shows the selected Logistic probability distributio n model. The Kullback - Leiblar divergence (also called relative entropy) [12] was used to measure how much one probability distribution was different from the reference probability distribution. The Kullback - Leibler divergence of the selected model with respect to the a ctual distribution was 0. 4064 , the lowest , or the best , fit of all candidate models.

The RTT accuracy, Y , can be modelled using Eq. (3) : Y = c 0 + c 1 * X k + random(PD3) ( 3 ) where X k is the RTT update time from the previous model i n Eq. (1).

From the data analysis using the actual RTT data at CLT, the following parameters were calculated in TABLE I .

Table I Parameters used in the RTT Model RTT Model Probability Distribution or Regression Model Parameters PD1: Log - normal μ = 0.114848, s = 0.303685 RTT update time PD2: Weibull A = 11.3599 , B = 1.54735 PD3: Logistic μ = - 0,306186 , s = 2.27737 RTT accuracy Linear Regression c = - 2.538, c = - 0.1908 0 1 3. RTT Model Validation With the given parameters in T able I , the RTT values were generated from the proposed RTT model for model validation. The two scatter plots in Fi gs . 8 and 9 show the actual RTT error and the generated RTT erro rs using the model proposed i n Eq. ( 3 ) , respectively , along the lookahead time window [ - 30 min, 0 min] as the time approaches ARTT . The RTT values from the RTT accuracy model cannot be exactly the same as the actual RTTs for an individual GA flight, but it is shown that the distributions are visually quite similar and the Kullback - Leibler divergence is small (0.0389).

Fig. 9 Modeled RTT accurac y Fig. 8 Actual RTT accuracy III. Simulation Results This section prese n ts the simulation results and analyses of the four simulation groups . During Bank 2 at CLT, almost all GA flights use the east runway (18L/36R) which is close st to the GA ramp areas . In the simulations , all GA flights were assigned to Runway 18L . Th e results shown in the analysis are for Runway 18L only.

A. Simulation Group 1: GA flight request distribution In this simulation group , ten GA flights were added to the baseline scenario. This number reflects the Bank 2 non - passenger flight statistical analysis. Their flight ready times from the GA ramp were distributed in the ban k in four ways : • Case 1: uniformly distributed over the whole bank , • Case 2: in the beginning 30 minutes of the bank , • Case 3: in the middle 30 minutes of the bank , and • Case 4: in the last 30 minutes of the bank Twenty Monte Carlo simulation runs were conducted for each case. Figs 10 and 11 show the average total gate hold time s and the average total taxi - out times , respectively, over the twenty simulation runs. Error bars indicate 95% confiden ce level. Since the GA flights provided no RTTs in this group, they were not subject to metering (and zero gate hold time) . In Case 2, where the GA traffic concentration was in the beginning of the bank , airline flights had longer gate hold and taxi - out times , rather than Case 3 as originally expected . It reveals that t he GA traffic was competing with the airline traffic for the runway at the beginning of the bank .

Additionally, because the GA aircraft immediately started taxiing into the AMA , after being designated as “ready”, and joined a departure queue earl ier , they became higher priority in the scheduler than the airline aircraft at the gate . This resulted in longer gate holds and taxi - out times for the airline flights. Case 3, where the GA traffic concentration was in the middle of the bank, had the second largest impact to airline flights. This is because although the airline departure traffic began to decrease , there was sufficient demand on this dual - use runway due to the inbound arrivals . Case 4, where the GA traffic was in the end of the bank, had the least impact on the airline flights. Also, in Case 4, the total GA taxi - out time increased compared to Case s 2 and 3. The probable reason for the increase in taxi out time is that the arrival traffic on the dual - use runway at the end of the bank caused the GA aircraft to wai t longer for the runway use .

Fig. 10 Total g ate h old t ime of g roup 1 Fig. 11 Total t axi o ut t ime of g roup 1 Figs 12 and 13 plot the time series of the gate hold time and taxi - out time, respectively . The time series is made of a sequence of vertical bars in five - minute bins. In t he gate hold time series shown in Fig 12 , the left bars represent the average total gate hold times of aircraft whose readiness time lies in those bins, and the right bars represent the average total gate hold aircraft count. Airline and GA aircraft are s eparated and stacked. In this simulation group , GA aircraft were at the parking stand and can leave freely and not subject to any holds and so only airline aircraft have hold times and visible bar heights as shown in Fig. 12 . For taxi - out time shown in Fi g 13 , each bar shows the average total taxi out time of the aircraft who’s actual off - block time lies in that bin. Airline and GA aircraft taxi - out times are stacked in the plot.

Fig. 12 Gate h old t ime s eries of g roup 1 Fig. 13 Taxi o ut t ime s eries of g roup 1 In the time series for Case 2 , surface metering of the airline flights started early around 20 minutes into the bank , and so the GA flights dominate d the earl y part of the bank as shown in the taxi out time series and ‘pushed’ the airline flights to metering hold. Case 3 shows the similar metering start time to Case 1, but higher metering time bars because of the GA traffic concentration in t he middle of the bank . Case 4 shows the least impact on airline aircraft metering which is consistent with the total numbers in Fig. 11 .

Fig. 14 Runway t hroughput c hart of g roup 1 Fig. 15 Cumulative r unway t hroughput of g roup 1 Figs 14 and 15 show the runway throughput metric. A departure runway operation is registered a t takeoff and an arrival runway operation occurs at landing. The stacked bar charts in Fig. 14 measure the average numbers of runway operations in five - minute bins.

Fig. 15 plots depict the cumulative operations for departure s only to focus on airline and GA departure throughputs. In Case 2, the bar chart shows the GA aircraft took a sign ificant portion of runway takeoff slots in the early part of the bank and pushed airline departure aircraft to later takeoff s , resulting in takeoff delay of airline aircraft. This is also observed in the cumulative throughput plots in Fig. 15 where the ai rline departure (in the middle plot) of Case 2 exhibited lower throughput up to about 70 minutes into the simulation time compared to the other cases. A similar situation can be seen in Case 3, in which the GA aircraft concentrat ion in the middle of the bank caused lower airline flight throughput from 60 to 10 5 minutes. The highest runway through p ut in five - minute bins happened in Cases 2 and 3, at 60 minutes into the simulation time. It is about 5.5 flights combining departures and arrivals , which translate to an average of 55 seconds of separation time , indicating the runway was saturated then .

T he results in this simulation group show tha t GA traffic concentrated in the beginning 30 minutes of the ba nk ha d the largest impact on airline flights, in metering hold time, taxi out time, and throughput . One reason is that airline departure demand starts early in the bank, which is then overlapped by GA traffic. Another reason lies in the unpredictability of GA operation, which tends to cause longer gate hold s from the surface schedul ing algorithm and longer taxi - out time s for airline aircraft .

B. Simulation Group 2: Number of GA flights In this group , six different numbers of GA flights were added to the baseline scenario. Their r eadiness time perturbation was evenly distributed over the entire traffic bank. Table I I shows the numbers of GA flights in each of the six cases. Twenty simulation runs were conducted for each case . Statistically, Case 6 of this group is the same as Case 1 of Simulation Group 1 .

Table II Number of GA flights Case No 1 2 3 4 5 6 Number of GA Flights 0 2 4 6 8 10 Figs 16 and 17 show the average total gate hold times and the average total taxi - out times, respectively, over the Monte Carlo simulation runs. When GA traffic increased, the airline flights experienced longer metering hold time s , as expected . At the same time, the total taxi - out time of the airline flights also showed a n increas ing trend with Case 2 as an outlier , which probably was due to th e demand timing of the two GA flights with respect to the airline flights in the baseline scenario . T he increase of GA traffic and its less predictability had a negative impact on the airline flights in both gate hold and taxi - out times , or on the metering effectiveness for taxi out time reduction . The increased taxi time of GA flights shown in Fig. 17 was due to the increased number of GA flights.

Fig. 16 Total g ate h old t ime of g roup 2 Fig. 17 Total t axi o ut t ime of g roup 2 Fig. 18 Gate h old t ime s eries of g roup 2 Fig. 19 Taxi o ut t ime s eries of g roup 2 The time series of gate hold and taxi out times are depicted in Figs 18 and 19. W hen the number of GA flights increased gradually, the metering started earlier. B ecause the GA traffic was ev enly spread across the bank , however, the system was able to handle the impact over the bank .

The runway throughput compariso ns among the six cases are shown in Figs 20 and 21. The cumulative airline flight throughputs , in the middle graph in Fig. 21 , show delayed throughputs as the number of GA flights increased . This is consistent with the combination of the increased gate hold time and taxi out time seen earl ier . It implies takeoff delay of the airline flights , which can be seen in the time series of bar charts around 100 minutes as more airline flights were pushed to take off late .

Fig. 20 Runway t hroughput c hart of g roup 2 Fig. 21 Cumulative r unway t hroughput of g roup 2 Overall in this simulation group, the simulation results show that the increase of GA traffic affected the airline flights with earlier metering start times , higher hold times , taxi out times and takeoff delays. In particular, because of the less predictability of GA traffic demand, the effectiveness of surface metering to mitigate taxi out times was affected negatively .

C. Simulation Group 3: Percentage of GA flights having RTTs The RTT model was introduced in Simu lation Group 3 and provided the RTT values for GA fligh ts prior to departing the parking stands in the GA hangar. The RTT model was configured to match the accuracy of the actual data from MITRE. Ten GA flights had their ready times perturbed uniformly over the entire bank in the baseline scenario . The percentage of the GA flights that provide d their RTTs were varied in six cases: 0%, 20%, 40%, 60%, 80% and 100%. The 0% case represents the current operations withou t Mobile Applications, whereas the 100% case assumes that all the GA flights provide controllers with the estimate d AMA entry times in advance . The GA flights which provided RTTs were considered in the high priority group in runway scheduling and had to comp ly with the metering hold at parking stand . Forty simulations were run for each case. This doubled number of simulations runs helped expand the statistical sample size to achieve a good accuracy of each target percentage of GA flight s that provided R TTs .

Fig. 22 Total g ate h old t ime of g roup 3 Fig. 23 T otal t axi o ut t ime of g roup 3 Figs 22 and 23 show the average total gate hold times and the average total taxi out times, respectively. Fig. 22, show s that th e total hold times of the GA flights increase as more GA flights provided RTTs , but the hold times of the airline flights show no statistically significant difference. This impl ies that the impact on the gate hold tim e of airline flights by GA flights with or without RTTs is small . GA flights that submit RTTs compete with airline flights for slots from the scheduler and hold at park stan d . In contrast, th e GA flights that do not p rovide RTTs directly affect the airline flights by leaving the parking stand freely when ready . Nonetheless, the overall gate hold times of both airline and GA flights increased when more GA flights ha d RTTs. In correlation with the increase in metering hold times, the taxi out time chart in Fig 23 show s a decreasing trend in both airline and GA flights. It indicates that when a greater percentage of GA flights submitted RTTs a nd complied with the metering times , the interruption to the metering ecosystem dropped and benefi tted the taxi out time reduction .

Fig. 25 Taxi o ut t ime s eries of g roup 3 Fig. 24 Gate h old t ime s eries of g roup 3 The time series bar char t s of metering hold and taxi out times are displayed in Figs 24 and 25 , respectively. The metering hold time charts show that the percentage of GA flights submitting RTTs did not accelerate nor delay metering start times , because of the same reasons as described in the total number analysis , i.e., the gate hold tim e of airline flights by GA flights with or without RTTs is small.

The metering hold time distributions of the a irline flights are very similar to each other. The increase of GA flights’ metering hold count and time is evident as more RTTs had been submitted . At the same time, the taxi - out time s of both airline and GA flights decrease d across the bank , consistent with the results in the total numbers.

Fig. 26 Runway t hroughput c hart of g roup 3 Fig. 27 Cumulative r unway t hroughput of g roup 3 The runway throughputs are depicted in Figs 26 and 27 . No obvious differences are found between cases in either plot . That indicate s that the RTT provision of GA flights can increase metering hold time but decrease taxi out tim es of both airline and GA aircraft , helping maintain departure throughput.

In summary, t he results show that overall taxi out times can be reduced when more GA flights provide RTTs and are in compliance with metering hold times . In other words, GA flights submitting RTTs and compl ying to metering times can help improve the airport performance.

D. Simulation Group 4: GA flight RTT accuracy In Simulation Group 4 , four different RTT accuracy levels of GA flights were tested , as shown in Fig. 28 . Case 1 is the ideal case where the pilots’ estimated RTTs are exactly the same as the actual RTTs. Case 3, the baseline case, is the current level of RTT accuracy from the data collected by MITRE. Cases 2 and 4 have the accuracy level s better and worse than the baseline case, respectively , for comparison . The horizontal axis is the lookahead time to wards the actual RTT. The vertical axis is the RTT accuracy measured as the difference of t he actual RTT and the estimated RTT by pilots. The data of the plots were collected from actual simulations.

Fig. 28 RTT a ccuracy c ases In each case, ten GA flights were uniformly added across the baseline traffic scenario with the varied RTT accuracy level. Forty simulations were conducted for each case as well .

Fig. 31 Cumulative runway throughput Fig. 29 Total g ate h old t ime of g roup 4 Fig. 30 T otal t axi o ut t ime of g roup 4 of GA flights of group 4 Figs 2 9 and 30 show the comparison among the four RTT accuracy levels for metering hold and taxi out times, respectively. The metering hold times of GA flights went up steadily when RTT accuracy became worse . However, the additional hold times did not translate to taxi out time reduction. The reason likely lie s in the fact that the added hold times were induced primarily by the inaccuracy of RTT prediction rather than scheduling. In particular, the conservative prediction errors (i.e., estimated RTT is later than ARTT) resulted in unnecessary hold times and consequently takeoff delays as seen in the runway throughput plot, Fig. 31 .

For airline flights , no statistically significant differences are observed in the first three cases . In Case 4, the taxi out times of both airline and GA flights had a small but noticeable amount of increase, compared to the averages of the other three cases, poss ibly due to the increased uncertainty from the RTT prediction errors. The time series of metering hold and taxi - out times, and the runway throughputs , not shown in the paper, had little difference among the four RTT accuracy levels.

The simulation results in this group show that the RTT accuracy affected mainly the GA flights ’ throughput . The RTT accuracy at or better than the baseline from the actual data had no significant impact on the airline flights. On the other hand , RTT error worse than the baseline level may have a negative impact on both airline and GA flights due to the increased prediction errors .

IV. Conclusions In this study, fast - time simulations were used to investigat e the impact of GA flight operations on airport performance . T he ATD - 2 surface metering scheduling algorithm was used for Bank 2 traffic hours at CLT . The objectives of the study are to investigate how the current GA operations may affect surface manageme nt efficiency benefits, such as taxi out time reduction from surface metering, and potential merits of including GA flights in the surface scheduling program . The simulations were conducted in four simulation configuration group s. The first tw o groups focused on the GA departure demand concentration and the number of GA flights in the bank. They were designed to study the impact of current GA operation. In the next two simulation groups, an RTT model was developed based on the actual operational data from MITRE. The model set s percentages of GA flights providing RTT predictions and varies the RTT accuracies in the simulations . Th ese two simulation groups studied how airport performance under surface metering were affected by GA flights submit ting RTT s to the scheduler system , through their mobile devices for example .

The simulations in the first two groups show that: • The concentration of GA departure traffic at the beginning of the bank had the most significant impact on airline departures.

It result ed in longer metering hold and taxi - out time s for airline flights . It also triggered the scheduler to start surface metering earlier. GA traffic concentration in the middle of the bank had the second largest impact on airline flights ’ performance. GA traffic concentration at the end of the bank and evenly spread out across the bank had the least impact on airline flights. However, GA departure concentration at the end of the bank had the longest taxi - out times because of the arrival traffic demand for the same runway during the second half of the bank.

• The increase of GA traffic affected the airline flights by initiating an earlier surface metering time , longer hold times , longer taxi - out times, and longer takeoff delays. In particular, less predictability of GA flights appear ed to affect the metering effectiveness in reducing taxi - out time for airline flights.

The results from the next two simulation groups show that: • When more GA flights provide d RTTs and followed metering hold advisory , the overall taxi - out times were steadily reduced by a relatively small amount (probably due to the small overall GA traffic level) , and at the same time the runway throughputs and takeoff delays remai ned the same. This finding indicate s readiness information from GA flights via mobile devices, such as MITRE’s proto t ype Mobile App , is beneficial to airport performance with surface metering .

• When the RTT accurac y is the same as or better than the baseline level , no statistically significant impact on airline flights was seen in all performance metrics. When RTT errors were larger than the baseline level, both airline and GA flights showed longer taxi - out times due to the increased uncertainty introduced by the p rediction errors. For GA flights, larger RTT prediction errors caused longer metering hold times that subsequently led to takeoff delay .

Despite the relatively small percentage of GA traffic at CLT, the results showed the clear impact of the current GA - operation levels on airport performance in the surface metering environment . Obtaining readiness information from more GA flights may also impact surface operations. Further research may consider other airports such as DAL where GA operations m ake up a greater percentage of the traffic demand .

Acknowledgement This study is supported by the NASA ATD - 2 project , funded by Airspace Operations and Safety Program (AOSP) . The authors are grateful to The MITRE Corporation for their collaboration and research support.

References [1] Federal Aviation Administration, “General Aviation and Part 135 Activity Surveys – CY 2017,” https://www.faa.gov/data_research/aviation_data_statistics/general_aviation/CY2017 [2] Y. Jung, et al. , “ Airspace Technology Demonstration 2 (ATD - 2) Phase 1 Concept of Use (ConUse), ” NASA/TM - 2018 - 219770, February 28 , 2018 .

[3] NASA ATD - 2 team, “ATD - 2 Benefits Mech anism,” Publications of the ATD - 2 Subproject, 2020 th [4] H. Lee et al. , “Fast - Time Simulation for Evaluating the Impact of Estimated Flight Ready Time Uncertainty on Surface Metering,” 38 Digital Avionics System Conference, San Diego, California, Sep 2019.

[5] S. Badrinath et al. , “Evaluating the impact of uncertainty on airport surface operations,” 2018 AIAA Aviati on Forum, Atlanta, Georgia, Jun 2018.

[6] P. Diffenderfer et al . , “Concepts for De livering IFR Clearances & Exchanging Pre - Departure Data Using Mobile Devices,” 2018 Integrated Communications Navigation and Surveillance (ICNS) Conference, Apr 2018 .

[7] P. Diffenderfer et al. , “ Using Mobile Devices for IFR Clearance Delivery and Release and Data Exchange ,” AIAA Aviation Forum Atlanta, Georgia, June 201 8 .

[8] R. D. Windhorst et al. , “Validation of simulations of airport surface traffic with the Surface Operations Simulator and Scheduler,” AIAA Aviation Technology, Integration and Operations Confer ence, Los Angeles, C alifornia , Aug 2013.

[9] W. Coupe et al . , “Scheduling Lessons Learned During Phase 1 Field Evaluation of the ATD - 2 Integrated Arrival, Departure, Surface Concept,” 13th USA/Europe Air Traffic Management Research and Development Seminar (ATM2019), Vienna, Austria, June 2019 .

[10] Apache Commons Math L ibrary, http s://commons.apache.org/proper/commons - math [11] MATLAB, https://mathworks.com/products/matlab.html [12] S. Kullback, “Information Theory and Statistics,” John Wiley & Son, 1959 , reprinted in 1978: ISBN 0 - 8446 - 5624 - 9.