ESTIMATION OF ARRIVAL CAPACITY AND UTILIZATION AT MAJOR AIRPORTS

Transcription

1 ESTIMATION OF ARRIVAL CAPACITY AND UTILIZATION AT MAJOR AIRPORTS Antony D. Evans, Husni R. Idris (PhD), Titan Corporation, Billerica, MA Abstract Airport arrival capacity has become a primary constraint to air traffic flow in the National Airspace System (NAS). Increased airport capacity may be achieved through decision support tools (DSTs) designed to increase operations under current runway configurations and air traffic procedures. In order to assess the potential benefits of such DSTs it is important to measure the degree to which airport arrival capacity is utilized by current operations. This paper presents an approach to estimate the arrival capacity achievable at an airport and the degree to which it is utilized by current operations. In this approach the degree to which airport arrival throughput saturates under high demand is identified and quantified under specific runway configurations and meteorological conditions. Capacity envelopes representing the maximum achievable airport arrival capacity were estimated conservatively as the 99 th percentile of throughput for runway configurations that exhibited throughput saturation, and as the th percentile of throughput for those that did not exhibit throughput saturation. The effective arrival capacity for each runway configuration and meteorological condition was also identified by simulating arrival operations and calibrating the arrival capacity to result in the current observed level of delay under high demand. The ratio of effective to achievable capacity was used as a measure of the underutilization of arrival capacity due to inefficiency in operations under high demand. This ratio reflects the potential benefits of DST application, which may increase utilization through improved operation efficiency. The ratio of arrival throughput to effective arrival capacity was used as a measure of the utilization of the capacity by the current level of demand. This ratio reflects the degree to which the demand at an airport is constrained by capacity. Along with the demand growth rate this ratio can indicate when DST application may be needed in the future at unconstrained airports. This approach was applied to BOS, and the secondary airports in the region: BDL, PVD and MHT. Utilization due to demand at BDL, PVD and MHT was found to be on average 5% of that at BOS, suggesting that demand is well below capacity at these airports. This suggests that metering is not necessary at these airports under current demand levels. Those configurations at BOS that saturate show an average ratio of effective to achievable capacity of.83. This indicates significant potential throughput benefit from DSTs that increase operation efficiency. Introduction Airport arrival capacity has become a primary constraint to air traffic flow in the National Airspace System (NAS), particularly under high demand and under adverse weather conditions. The resulting increase in delays has pushed airlines and other operators to increase scheduled traffic at secondary airports in an attempt to offer lower delay alternatives. Increasing airport capacity may be achieved through constructing additional runways and reducing safety separation requirements. However, these are long term, costly investments. Decision support tools (DSTs) designed to increase operations under current runway configurations and air traffic procedures offer beneficial alternatives. One successful example of such a tool is the Traffic Management Advisor (TMA)T which applies time based metering of arrivals into constrained runways and terminal area gates [1]. TMA is already implemented at a number of airports and its extension to multiple center environments (TMA-MC) [2,3,4], is under deployment. In order to assess the potential benefits of such DSTs and thus make strategic decisions about their implementation, it is important to measure the degree to which airport arrival capacity is utilized by current 1

2 operations. This requires estimation of the maximum arrival capacity that is achievable under current runway configurations, safety procedures, and weather conditions. This achievable capacity reflects an upper bound on the potential benefits from increased throughput. The achievable capacity may be underutilized due to either lack of demand or inefficiency in operations at the airport. Since DSTs improve the efficiency of operations it is important to separate these underutilization factors. This requires estimation of the inefficiency of the current operations in reaching the achievable capacity and the degree to which demand is constrained by capacity at the airport, as explained in the next section. This paper serves to present a methodology for the estimation of airport arrival capacity and utilization, and its application to Boston Logan International Airport (BOS), which has 5 crossing runways, and the secondary airports in the region: Hartford Springfield (Bradley) Airport (BDL), which has 3 crossing runways; Providence T.F. Green Airport (PVD); and Manchester Airport (MHT), each of which have 2 crossing runways. This estimation was in support of assessing the need for DSTs such as TMA and TMA-MC for these airports, currently and as demand increases in the near future. Approach Arrival capacity represents the primary arrival flow constraint of an airport, which is measured by the airport arrival acceptance rate (AAR). The AAR depends mainly on runway configuration, meteorological conditions such as visibility and ceiling, runway conditions, and airport departure rate. The AAR reported by the FAA generally represents a conservative estimate. Thus, airport AARs must instead be estimated by data analysis and modeling. Airport capacity may be underutilized due to a lack of demand or inefficiency in operations, which results in missing available runway landing slots. An airport that is underutilized due to lack of demand is unlikely to benefit from implementation of a DST such as TMA. Airports with high demand that are underutilized due to inefficient operations, however, may benefit significantly from implementation of a DST. It is therefore essential to identify different measures of capacity utilization: 1. The ratio of effective arrival capacity to achievable arrival capacity, which provides a measure of underutilization due to inefficiency in operations only; where: Effective arrival capacity refers to the capacity operated at the airport under high demand, that results in current measured delay levels [5]. Achievable arrival capacity refers to the maximum capacity that can be attained at the airport [5]. Both these measures of arrival capacity are dependent on runway configuration and meteorological conditions. 2. Utilization calculated as the ratio of throughput to effective capacity, which measures underutilization due to lack of demand only. 3. Utilization calculated as the ratio of throughput to achievable capacity, which measures underutilization due to both lack of demand and inefficiency in operations. These measures of utilization and capacity were estimated for the commonly operated runway configurations at BOS, BDL, PVD, and MHT, under visual and instrument meteorological conditions (VMC and IMC). The approach by which these measures of arrival utilization and capacity were calculated is presented in Figure 1. Actual Delay Delay model Vary capacity Actual Demand Historic Throughput Effective Capacity Airport Saturation Saturates? No Historic Demand Yes Achievable capacity from simulation or historical throughput (e.g. th percentile) Achievable Capacity Airport Utilization Airport Utilization Achievable capacity from historic throughput (e.g. 99 th percentile) Figure 1. General analysis approach Achievable Capacity As described in Figure 1, the achievable arrival capacity was estimated directly from historical throughput data for those configurations for which arrival demand was high enough that arrival throughput saturated. The degree to which arrival throughput saturates was identified by comparing historical arrival throughput and historical arrival demand. For those configurations for which arrival demand was not high enough to saturate arrival throughout, achievable arrival capacity may be 2

3 underestimated by historical data, and should instead be estimated by other means, such as simulation. For this analysis no simulation was developed. Instead a higher percentile ( th ) of the historical throughput data was used for the configurations that did not saturate, and a lower percentile (99 th ) was used for those that did saturate, as explained in the section entitled Arrival Capacity Modeling. Effective arrival capacity was estimated by calibrating an arrival flow delay model by varying the applied capacity to match actual and modeled arrival delays. Utilization was then estimated according to modeled demand, the achievable configuration capacities estimated, and the effective configuration capacities estimated, as defined above. Modeling Methodologies Identification of the degree to which runway configurations saturate requires the comparison of arrival throughput and arrival demand. Arrival throughput was extracted directly from historical data, while arrival demand was modeled as described in detail in the section entitled Arrival Demand Modeling. Details of the arrival capacity modeling required to estimate the achievable arrival capacities are presented in the section entitled Arrival Capacity Modeling. Arrival demand is also required for estimation of the effective arrival capacities by simulating arrival delays resulting from constrained arrival demand. Details of the arrival delay modeling required to estimate the effective arrival capacities are presented in the section entitled Delay Modeling. Arrival Demand Modeling Airport arrival demand is driven by the scheduling of flights into the airport. The demand on the airport is represented by a series of Estimated Times of Arrival (ETAs) at the runway threshold. Aircraft 1 (unimpeded) System boundary t 1 TT 1 1 t x i Aircraft x actual time of arrival at fix i ETA x i Aircraft x estimated time of arrival at fix i Nominal unimpeded transition time from fix i to j TT x ij Airport ETA 1 1 = t 1 + TT 1 1 Figure 2. Demand modeling based on estimated times of arrival As illustrated in Figure 2, ETAs were calculated by adding unimpeded transition times (TT) to the flight entry time into the system (t ). This entry time into the system was taken as the actual time that the aircraft crossed the outer boundary of the system. The unimpeded transition times to the airport represent aircraft flight times from the system boundary to the airport unimpeded by factors such as other aircraft or ATC restrictions. In this manner each flight s ETA at the runway threshold is calculated. Arrival flows into each airport were identified by plotting Enhanced Traffic Management System (ETMS) tracks for all arrivals in May 4. A total of 8 primary arrival flows were identified and modeled at BOS, as illustrated in Figure 3 by the arrows. Also illustrated is the location of BOS and the system boundary. Similarly, 4 primary arrival flows were identified at BDL, and 5 at each of PVD and MHT. Latitude Longitude BOS 8 nm Figure 3. Major arrival flows into BOS Each flight s entry time into the system was extracted from ETMS track data by interpolating crossing times on the system boundary, which was modeled as an 8nm radius circle surrounding the airport, as shown in Figure 3. 8nm was used for the analysis because it is large enough to capture runway approach queues, and thus large enough to capture holding and delay due to arrival flow metering into the airport. It is not so large, however, as to include the majority of other impacts on arrival delay such as ground delay programs (impacting delays at the departure airport) and en-route weather (impacting delays en-route). Each flight s landing time was extracted directly from ETMS data. Unimpeded transition times were estimated according to a statistical analysis of actual historic transition times for flights on each arrival flow. In this analysis unimpeded transition times were estimated from historical transition times for flights encountering only small arrival queues. The actual transition times for all flights on each arrival flow from May 1 to July 21, 4 were extracted from 3

4 ETMS data and plotted against the queue size encountered, for 2 different weight classes, and for commonly operated runway configurations. Queue size encountered was calculated as the number of aircraft that landed at the airport from the time when the flight under question crossed the 8nm boundary to when it landed at the airport. The weight classes plotted were Small and Other, incorporating Large, B757, and Heavy (unimpeded transition times were not found to differ significantly over these classes). Figure 4 below shows an example of the plots generated, and the observed queuing dynamics. Transition time can be seen to increase with increasing queue size. The increase in transition time is of a higher order than one, suggesting that there is a queuing effect within the 8nm upstream boundary. Transition time [min] Queuing Model for Large, Heavy and B757 weight class aircraft on the west arrival flow into BOS, landing on runways 33L and 33R Queue size [ac] Figure 4. Queuing model Unimpeded transition times were estimated by fitting a curve to the data, and estimating the average unimpeded transition time as the zero queue intercept of the curve fit. In each case the better fit of a 2 nd order parabolic curve fit and an exponential curve fit was used to estimate the average unimpeded transition time. A normal distribution was fitted around this average with a standard deviation equal to that of a sample of data points with low queue size. Airport ETAs were calculated by adding unimpeded transition times sampled from the distributions generated (according to aircraft weight class and runway configuration), to the flight entry times into the system (from ETMS). These ETAs represent the baseline relative to which arrival delay is accumulated. Arrival Capacity Modeling In order to determine the degree to which the runway system saturates under current demand levels the throughput of the runway system was plotted as a function of demand, as shown in Figure 5 for two different runway configurations at BOS. Runway configuration is referred to in the form: arrival runway designations departure runway designations. Throughput [a/c per hr] Throughput [a/c per hr] Average with Std. Dev. error bars Average with std. dev. error bars -- Hyperbolic fit to Average Hyperbolic fit to average Dem and [a/c per hr] 5 4 Low Capacity Configuration at BOS (22L 15R) VMC (a) High Capacity Configuration at BOS (4L, 4R 4L, 4R, 9) VMC 7 - Average with Std. Dev. error bars Average with std. dev. error bars -- Hyperbolic fit to Average 6 Hyperbolic fit to average Dem and [a/c per hr] (b) Figure 5. Saturation analysis (ASPM, April 3 to Sept. 4) Such plots were generated using throughput and demand data per half hour 1 from the Aviation System Performance Metrics (ASPM) database from April 3 to September 4, excluding periods during ground delay programs (GDPs) and night time; and from the Aircraft Situation Display to Industry (ASDI) database, from May 1 to July 21, 4, excluding night time. 2 Two databases were used because ASPM data was not available for PVD and 1 Because of binning errors the FAA recommend that capacity be identified conservatively using half hour periods, and not quarter hour periods [4]. 2 GDP periods excluded because ASPM demand during GDPs is based on original scheduled departure times, and does not account for delayed departure times. Night time excluded because of low demand. 4

5 MHT. The throughput plotted represents the number of aircraft that landed at the airport per half hour, multiplied by 2 to be specified as an arrival rate per hour. As defined for the ASPM database, the demand plotted represents the number of aircraft that intend to land at a specific airport per half hour, also multiplied by 2 to be specified as an arrival rate per hour. A flight is included in the demand count in those periods starting from its estimated time of arrival, and ending at its actual wheels-on time. Average throughput is also plotted against corresponding demand with error bars representing one standard deviation in each direction. Plots of throughput versus demand using ASPM data and ASDI data were found to correspond closely, with mean arrivals per day differing by only 7.7%. Because of the larger dataset available, ASPM data was used where possible (i.e. BOS and BDL). Considering Figure 5, throughput increases linearly with demand at low demand levels. However, as illustrated in Figure 5(a), as demand increases, throughput plateaus to a maximum above which it does not increase further. As the number of aircraft waiting to land increases, throughput is expected to increase because of the increased pressure applied to the system. However, as throughput approaches the capacity of the system, further increase is limited, regardless of additional demand pressure. This saturation level is set mainly by the safety separation requirements between aircraft and by controller workload. If demand is never high enough to saturate the airport, throughput should continue to increase linearly with demand as in Figure 5(b). In order to quantify whether arrival throughput saturates or not, a hyperbolic curve was fitted to average throughput as a function of demand (dashed lines in Figure 5). This hyperbolic curve fit was forced to asymptote to demand equaling throughput as demand tends to zero, and to constant throughput as demand tends to infinity. The constant throughput to which the curve fit asymptotes at high demand then represents the arrival rate at which the runway configuration throughput saturates. By fitting the curve to average throughput, it represents an average arrival rate over factors such as departure throughput. Because the arrival service rate is a function of runway configuration, meteorological conditions, and runway conditions, throughput was plotted against demand for each runway configuration independently, and for VMC and IMC independently. The effect of this filtering is presented in Figure 6 and Figure 7, where Figure 6 shows a plot of all configurations combined, while Figure 7 separates three different configurations one that does not saturate (4L, 4R 4L, 4R, 9), and two that saturate at different levels (22L 22L, 22R and 27 27). It is clear in Figure 7 that each different configuration has a different trend, and that much of the data scatter and drop-off in throughput in Figure 6 at high demand is as a result of the differences in configuration saturation capacities. Throughput [a/c per hr] Average - with with std. Std. dev. Dev. error bars Hyperbolic -- fit fit to to average Average All configurations Demand [a/c per hr] Figure 6. BOS saturation analysis all configurations (ASPM April 3 to Sept. 4) Throughput [a/c per hr] All configurations at BOS BOS Filtering by Runway Configuration Average with std. dev. error bars Hyperbolic fit to average 4L, 4R 4L, 4R, 9 22L 22L, 22R Demand [a/c per hr] Figure 7. BOS saturation analysis by runway configuration (ASPM, April 3 to Sept. 4) For runway configurations that saturate under current demand levels, the hyperbolic curve fit horizontal asymptote is likely to be equal to or lower than the maximum throughput observed. However, for configurations that do not saturate under current demand levels, this hyperbolic curve fit horizontal asymptote is likely to be higher possibly significantly higher than the maximum throughput observed. The ratio of the hyperbolic curve fit horizontal asymptote to a measure of the observed throughput thus serves as an indicator of the degree of throughput saturation. The lower the measure of observed throughput selected for the ratio, the stricter 5

6 the requirement for throughput to show saturation. For this analysis the 99 th percentile of the observed throughput was selected for this measure consistently across all airports and configurations, for the purpose of comparison. Any ratio below 1 was taken to indicate saturation, and above 1, non-saturation. For runway configurations that saturate according to the criteria described above, the achievable arrival capacity was estimated by the 99 th percentile of the observed throughput. The th percentile is likely to be a rare occurrence, and may represent ideal wind and fleet mix conditions and human error in meeting separation requirements. Percentiles below the 99 th were considered too conservative based on qualitative analysis of the distribution of observed throughput. For runway configurations that do not saturate, even the th percentile of the observed throughput does not represent ideal conditions, and may underestimate the achievable capacity. Therefore, the achievable arrival capacity should be estimated by other methods such as explicitly simulating runway arrival operations, or using a suitable hyperbolic curve fit asymptote. Such methods will be investigated in future research. In this analysis, the th percentile of the observed throughput was used to conservatively estimate the achievable capacity for runway configurations that do not saturate. The arrival service rate of a runway system is also dependent on the departure service rate when the arrivals and departures share the same or intersecting runways. As the arrival and departure rates increase a tradeoff develops where serving more arrivals is accomplished at the expense of serving less departures, and vice versa, forming a capacity envelope [5], such as those presented in Figure 9 to Figure 12. While arrival throughput saturation was estimated for average departure rates, a capacity envelope is essential for isolating the effect of departures by modeling the tradeoff between maximum arrival and departure rates. The effect of departures was modeled by calculating percentiles of arrival throughput during periods with a range of actual departure throughputs. The percentiles of arrival throughput were then plotted against corresponding departure throughput, yielding capacity envelopes, as shown in Figure 9 to Figure 12. A range of percentiles (from the 75 th to the 99 th percentile) were calculated in order to visualize the distribution of the data. In order to smooth the envelopes the percentiles were calculated from beta distributions fitted to the throughput data, instead of the raw data. Delay Modeling A delay model was calibrated in order to identify the effective airport capacity that results in current observed delay levels. The delay modeling approach is presented in Figure 8 below. AAR ATAs 8nm upstream of Airport Computation of ETAs using statistical Unimpeded Transition Times ETAs at Runway Arrival Delay Model (Time Slot Allocation) ATAs at Runway Figure 8. Delay modeling The system modeled was from the 8nm radius boundary to the runway threshold. By modeling this system only, arrival delays incurred within the 8nm radius boundary are modeled, while all delays incurred outside the system are excluded. Inputs to the delay model come from the actual arrival traffic through the 8nm boundary, and the runway capacity. The output of the model is actual times of arrival (ATAs) at the runway, including incurred delays. Unimpeded transition times between the system boundary and runway are added to the ATAs at the system boundary to derive ETAs at the runways. The arrival delay model then adds any delay needed to satisfy the applied AAR, calculating runway ATAs. The arrival flows into the airport were delayed to meet the applied AAR using time slots. This includes first come first serve allocation of time slots based on ETAs, such that the applied AAR is not exceeded over a -minute moving window. Each flight is delayed just enough to fit it into a time slot calculated from its ETA, the demand from all other traffic, and the applied AAR. This delay represents the delay that must be incurred by the flight within the system for the applied AAR to be satisfied. An error was added to ATAs at the runway to model uncertainty in meeting runway arrival slots. The error was sampled from a normal distribution centered at zero with a standard deviation of 9 seconds. A 9 second standard deviation was chosen based on the average standard deviation of the distributions of unimpeded transition time. The delay model was calibrated by adjusting the AAR applied to the model to constrain the simulated arrival traffic, so as to equate the distributions of actual and modeled delay as closely as possible. The 6

7 AAR was adjusted by scaling the 99 th percentile capacity envelopes developed. The delay model was calibrated by comparing the means of the distributions of actual and modeled delay using the standard t-test. The capacity that resulted in the best test performance was selected. The standard t-test measures the probability that the means of the distributions of actual and modeled delay are equal. Only flight delays during 15 minute time periods in which arrival queues were simulated were considered in these distributions. In this way the model was calibrated according to periods of high demand only. Delay modeled for different runway configurations was calibrated independently, except for those with small data sample sizes. These were grouped according to similarities, including number of runways operated, relative geometry of runways operated, and runway length and equipage. The t-test results for the selected capacities were found to range from 18% to 89% for the different airports, runway configurations, and meteorological conditions, with an average of 55%. The standard t- test is generally regarded to pass if this probability is greater than 5%. Results The approach described above was applied to estimate the following parameters for BOS, BDL, PVD and MHT: Airport saturation, illustrated by the degree to which arrival throughput saturates with increasing arrival demand on the airport. Airport arrival rate capacity (both achievable and effective capacity) presented in the form of capacity envelopes relating airport arrival rate capacity to departure throughput. Airport utilization, measuring under-utilization due to lack of demand only, under-utilization due to inefficiency in operations only, and underutilization due to lack of demand and inefficiency in operations. For BOS, these parameters were estimated for each of the commonly operated runway configurations, under VMC and IMC. Because runway configuration data was not available for BDL, its parameters were estimated for all runway configurations combined, but under VMC and IMC independently. Because neither runway configuration data nor meteorological conditions were available for PVD and MHT, these parameters were estimated for PVD and MHT for all runway configurations and meteorological conditions combined. Airport Saturation Arrival throughput was plotted against arrival demand, and airport saturation analyzed for each airport by calculating the ratio of the asymptote of the fitted hyperbolic curve to the 99 th percentile of the arrival throughput. These saturation ratios are presented in Table 1. For this analysis any ratio below 1 was taken to indicate saturation, and above 1, non-saturation. Table 1. Ratio of hyperbolic curve fit asymptote to 99 th percentile arrival throughput Airport Runway Configuration Met. Cond. Ratio: Asymptote/ 99 th Percentile 22L 15R VMC.86 BOS 4R 9 IMC.96 33L,33R 27,33L VMC.91 22L,27 22L,22R VMC 1.1 4L,4R 4L,4R,9 VMC 1.3 BDL All VMC 1.5 All IMC.85 PVD All All 1.6 MHT All All 2.1 At BOS arrival throughput for the lower capacity configurations 22L 15R, and 4R 9 was found to saturate, with ratios of.86 and.96 respectively. Figure 5 (a) shows the plot of arrival throughput versus arrival demand for configuration 22L 15R under VMC. In this figure average throughput clearly reaches a saturation level at high demand. Arrival throughput at the higher capacity configuration 33L, 33R 27, 33L at BOS was also found to saturate, with a ratio of.91. The throughput at the highest capacity configurations operated at BOS, (22L, 27 22L, 22R and 4L, 4R 4L, 4R, 9), was not found to saturate, however, with the ratios of 1.1 and 1.3 respectively. Figure 5 (b) shows the plot of arrival throughput versus arrival demand for configuration 4L, 4R 4L, 4R, 9 under VMC. In this figure arrival throughput does not appear to approach a saturation level at current demand levels. At BDL arrival throughput was not found to saturate under VMC, with a ratio of 1.5. Arrival throughput was found to saturate under IMC, however, with a ratio of.85. Winter operations, which are more often under IMC, are included in this analysis. Saturation under IMC may thus represent predominantly winter operations at BDL. At PVD and MHT arrival throughput was not found to saturate, with ratios of 1.6 and 2.1 respectively. 7

8 Airport Arrival Capacity A range of airport capacity envelopes were plotted for each of the commonly operated configurations at BOS under VMC and IMC, at BDL under VMC and IMC, and at PVD and MHT for all configurations and meteorological conditions combined. Achievable (99 th or th percentile for configurations that saturate or do not saturate, respectively) and effective (based on calibrated delay model) capacity envelopes were identified for each. Figure 9 and Figure present the envelopes for a low capacity configuration, 4R 9 under IMC, and for a high capacity configuration, 22L, 27 22L, 22R under VMC, at BOS, respectively. Arrival Capacity [ac/hr] Arrival Capacity [ac/hr] th Percentile Achievable Capacity (99 th Percentile) 95th, 9th, 85th, 8th & 75th Percentiles Effective Capacity Departure Throughput [ac/hr] Figure 9. Capacity envelope: BOS runway configuration 4R 9 IMC Achievable Capacity ( th Percentile) 99th, 95th, 9th, 85th, 8th & 75th Percentiles Effective Capacity Departure Throughput [ac/hr] Figure. Capacity envelope: BOS runway configuration 22L, 27 22L, 22R VMC In Figure 9 both the achievable capacity envelope and effective capacity envelope are lower than the corresponding envelopes in Figure, as expected. However, the difference between the effective and achievable capacities in Figure 9 is significantly less than that in Figure. Thus, the effective capacity under the low capacity configuration is closer to the maximum achievable capacity than under the high capacity configuration. This suggests that current operations under the low capacity configuration are more efficient than under the high capacity configuration. For BDL the envelopes under VMC and IMC are presented in Figure 11 (a) and (b) respectively, plotted to the same scale as Figure 9 and Figure. Arrival Capacity [ac/hr] 4 th Percentile Achievable Capacity Effective Capacity 4 4 Departure Throughput [ac/hr] (a) (b) Figure 11. Capacity envelopes: BDL under a) VMC, b) IMC The BDL capacity envelopes presented in Figure 11 show significantly lower capacities than those presented for BOS in Figure 9 and Figure, because of fewer runways. Under VMC (Figure 11(a)), the estimated effective capacity is as high as the estimated achievable capacity ( th percentile), indicating that BDL is particularly efficient under VMC, noting that the achievable capacity may be underestimated because the airport does not saturate under VMC. Under IMC (Figure 11(b)), the estimated effective capacity is also as high as the estimated achievable capacity (99 th percentile), indicating that BDL, which saturates under IMC, is also particularly efficient under IMC. 3 For PVD and MHT the envelopes are presented in Figure 12(a) and (b) respectively, also plotted to the same scale as Figure 9 and Figure. Arrival Capacity [ac/hr] 4 99th, 95th, 9th, 85th, 8th & 75th Percentiles 99th, 95th, 9th, 85th, 8th & 75th Percentiles 4 4 Departure Throughput [ac/hr] (a) Achievable Capacity Effective Capacity (b) Figure 12. Capacity envelopes: a) PVD, b) MHT 3 The th percentile in Figure 11(b) is significantly higher than the achievable capacity (99 th percentile), which may be due to high throughput under non-constraining IMC. 8

9 Like Figure 11, the capacity envelopes presented in Figure 12 show significantly lower capacities than those presented for BOS in Figure 9 and Figure. However, as in Figure 11(a) the estimated achievable capacity envelopes presented in Figure 12(a) and (b) may underestimate the actual achievable capacities because the airports do not saturate. In both cases the effective capacity envelopes are lower than the achievable capacity envelopes, but higher than the 99 th percentile envelopes. Airport Arrival Utilization Airport arrival utilization is presented in Table 2 below, averaged for all periods excluding night times (8pm to 7am). In the table, τ refers to throughput, µ to capacity, λ to demand, and η to efficiency. Three measures of utilization are presented: The ratio of throughput to effective capacity, which measures underutilization due to lack of demand only; The ratio of effective capacity to achievable capacity, which measures underutilization due to inefficiency of operations only; and The ratio of throughput to achievable capacity, which measures underutilization due to both lack of demand and inefficiency of operations. The measure of utilization due to demand only (first results column) shows the highest utilization for configurations that show greatest saturation (such as.78 for BOS configuration 22L 15R under VMC), and the lowest utilization for configurations that show least saturation (such as.29 for BDL under VMC). Airport Runway Configuration Table 2. Airport arrival utilization Met. Cond. τ / µ effective (Due to λ only) This is expected as higher demand is likely to result directly in greater saturation. All configurations at BOS show significantly higher utilization than BDL, PVD or MHT, suggesting that demand at these secondary airports is particularly low relative to capacity. The measure of underutilization due to inefficiency of operations only at BOS (second results column) shows the highest utilization for the configurations that show the greatest saturation and lowest capacity (.85 for configuration 22L 15R under VMC), and the lowest utilization at the configurations that show the least saturation and highest capacity (.64 for configuration 4L, 4R 4L, 4R, 9 under VMC). This suggests that BOS operates more efficiently when it is closer to saturation. This is expected as throughput increases when the pressure applied to the system is increased. A DST such as TMA or TMA-MC may increase the efficiency of operations further, and increase the effective arrival capacity even closer to the estimated achievable capacity for all configurations, including those that saturate less. The measure of underutilization due to inefficiency of operations only at BDL, PVD, and MHT (second results column) shows a range of utilization values which are generally high (from.7 to 1.). The achievable capacities estimated for these airports may, however, be underestimated, which would result in overestimated utilization. The measure of underutilization due to both lack of demand and inefficiency of operations (third results column) is the product of the other two measures of utilization. Utilization µ effective / µ achievable (Due to η only) τ / µ achievable (Due to λ & η) 22L 15R VMC R 9 IMC BOS 33L,33R 27,33L VMC L,27 22L,22R VMC L,4R 4L,4R,9 VMC All All All VMC BDL All IMC All All PVD All All MHT All All Throughput for these configurations was not found to saturate. This means that the estimated achievable capacities ( th percentile), may underestimate the actual achievable capacities, meaning that these measures of utilization may be overestimated. 9

10 Conclusions This paper presented an approach to estimate the arrival capacity achievable at an airport and the degree to which it is utilized by current operations. This approach was applied to BOS, and the secondary airports in the region: BDL, PVD and MHT. According to the results presented in this paper, arrival throughput at BDL, PVD and MHT does not saturate under current demand levels (except BDL under IMC, which may include predominantly winter weather operations). BOS, however, does saturate under all but the two highest capacity configurations. This suggests that demand at BDL, PVD and MHT is not currently high enough for throughput to saturate, while that at BOS it is. Utilization due to demand at BDL, PVD and MHT is on average 5% of that at BOS, suggesting that demand is well below capacity at these airports. This suggests that metering is not necessary at these airports under current demand levels. Those configurations at BOS that saturate under current demand levels show an average ratio of effective to achievable capacity of.83. This indicates the magnitude of potential throughput benefits from DSTs or improved procedures that increase operation efficiency. For runway configurations that exhibit throughput saturation, an estimate of the achievable arrival capacity was determined from historical throughput data. However, for those that do not exhibit throughput saturation, such as BDL under VMC, PVD, MHT, and some configurations at BOS, the achievable arrival capacity may be underestimated even by the th percentile throughput, which may increase under increased demand. Analytical or simulation techniques may be investigated in the future to determine the achievable capacity in these conditions more accurately. Acknowledgements The authors would like to thank Todd Farley and Daniel Kozarsky of the NASA Ames Research Center for supporting this research. References [1] Swenson, HN, T Hoang, S Engelland, D Vincent, T Sanders, B Sanford, K Heere, 1997, Design and Operational Evaluation of the Traffic Management Advisor at the Fort Worth Air Route Traffic Control Center, Saclay, France, 1st USA/Europe Air Traffic Management R&D Seminar. [2] Farley, TC, JD Foster, T Hoang, KK Lee, 1, A Time-Based Approach to Metering Arrival Traffic to Philadelphia, AIAA , Los Angeles, CA, AIAA Aviation Technology, Integration, and Operations (ATIO) Forum. [3] Hoang, T, T Farley, J Foster, T Davis, 2, The Multi-Center TMA System Architecture and Its Impact on Inter-Facility Collaboration, Los Angeles, CA, AIAA Aviation Technology, Integration, and Operations (ATIO) Forum. [4] Idris H, A Evans, S Evans, D Kozarsky, 4, Refined Benefits Assessment of Multi-Center Traffic Management Advisor for Philadelphia and New York, AIAA , Chicago, IL, AIAA Aviation Technology, Integration, and Operations (ATIO) Forum. [5] De Neufville, R, A Odoni, 3, Airport Systems, Planning, Design, and Management, New York, NY, McGraw-Hill Companies, Inc. Key Words Airport Arrival, Airport Capacity, Capacity Envelope, Demand, Airport Saturation, Utilization, Delay, BOS, BDL, PVD, MHT. Biographies Antony Evans is currently working for the Titan Corporation, where he is doing system analysis, benefit assessment and concept development in air traffic systems. Mr. Evans received an S.M. in Aeronautics and Astronautics (2), and an S.M. in Technology and Policy (2), both from the Massachusetts Institute of Technology. Husni Idris is currently working for the Titan Corporation leading research in air traffic systems. His research has focused on concept development, decision support, human factors and benefit assessment, applied to airline and air traffic operations. Dr. Idris received an S.B. (1989) and an S.M. (1992) in mechanical engineering, an S.M. (1) in operations research and a Ph.D. (1) in human factors and automation, all from the Massachusetts Institute of Technology.