Improving Ground Support Equipment Operational Data for Airport Emissions Modeling (2015)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

26 This section presents a consistent and systematic protocol to collect GSE operational and inven- tory data to satisfy emissions inventory input requirements identified in Chapter 3. Special atten- tion is also given to safety and security measures and regulations specific to airports, statistical methods with which to calculate a sufficient sample size for data collection, and quality assurance checks and procedures necessary to ensure the collected data are appropriate for use. 4.1 Data Requirements The target data for an airport GSE emissions inventory is that which is required as input(s) to the computer modeling tools that will be used to com- pute the emissions inventory. As described in Chapter 3, FAA’s AEDT are currently the recommended models for this purpose. For airport GSE emissions estimates, these models use data on GSE activ- ity by type and activity from GSE assigned to an aircraft gate (i.e., aircraft/ gate GSE assignment option) and/or GSE as an overall population (i.e., GSE population-based option). Three alternative approaches (basic, intermediate, and advanced) are also discussed in Chapter 3. 4.2 Health, Safety, and Security A large component of the overall approach outlined within this protocol requires access to air- port airfields and apron areas. It is imperative that all regulations pertaining to airport safety and security are adhered to, and that the safety of personnel is ensured at all times. For any data col- lection activity involving field presence on airport, a Health, Security, and Safety Plan (HASSP) is suggested to be developed hand-in-hand with the data collection work plan. Appendix A outlines applicable federal and airport-specific regulations in these areas and how these regulations interface with the development of a HASSP specific to the collection of airport GSE operational data. In addition, an example of a HASSP is also included in Appendix A. 4.3 Data Collection Methods Data collection methods evaluated for development for this protocol included (1) the apron survey; (2) acquisition of airline/ground handler equipment inventories and/or property lists; and (3) remote sensing methods. Of these, the apron survey is the most complicated to plan, organize, and implement and is thus described in detail. By comparison, the acquisition of GSE GSE Data Collection Protocol C H A P T E R 4 Reminder—Chapter 3 provides listings and explanations of the data requirements for computing GSE emission inventories using three alternative approaches: basic, intermediate, and advanced.

GSE Data Collection Protocol 27 inventories/property lists is more straightforward. As for the remote sensing options evaluated in this research, they were determined to be not practicable at this time for most airport applications. 4.3.1 Apron Survey An apron survey focuses on GSE activity at terminal gates or in areas where aircraft are otherwise being parked and serviced during turnaround or en route operations (Figure 8). A turnaround operation signifies that the aircraft departs from Airport A, arrives at Airport B to exchange passengers and/or cargos, and then returns back to Airport A. An en route operation represents an aircraft departing from Airport A, stopping over at Airport B, and continuing on to its final destination at Airport C. Aircraft services requiring GSE in these apron areas include aircraft cabin ser- vice and climate control; baggage/cargo handling; passenger loading/unloading; aircraft fueling, starting, and maneuvering; deicing; and occasional service and maintenance. Utilization of either fixed-gate or semi-fixed-gate GSE (e.g., a GPU or a belt loader) as well as mobile GSE (e.g., a baggage tractor or hydrant truck) are commonplace; however in the case of mobile GSE, only the portion of activity as they approach, egress, or dwell in the gate area is required as data for the emissions inventory. A number of different types of apron areas can exist within the same airport, defined for the purposes of this protocol as: • Terminal area aprons—Areas where commercial air carrier and air taxi/commuter aircraft are serviced for the purposes of enplaning and deplaning passengers. Inclusive of close-in contact and non-contact gates. Belly cargo handling and gate deicing is considered part of terminal area apron activity. • Deicing aprons—Areas where aircraft are transported or directed to a remote location from the passenger gates (i.e., to a deicing pad) for either fluid or alternative (e.g., infrared) deicing. • Cargo aprons—Areas where dedicated cargo air carrier operations (i.e., the exclusive loading/ unloading/transfer of cargo) occur at separate facilities from the terminal area. • GA aprons—Areas where private pilot, air charter, corporate jet, and other miscellaneous commercial aircraft park and receive service. Reminder—Chapter 2 provides a comprehensive listing and descrip- tions of types and functions of GSE commonly found at U.S. airports. Figure 8. Airport GSE apron survey viewpoint.

28 Improving Ground Support Equipment Operational Data for Airport Emissions Modeling Other apron types also exist (e.g., maintenance, remote); however, terminal area and cargo aprons are suggested to be assigned priority in an inventory because they typically account for a majority share of aircraft GSE gate service events, and safety/logistical considerations may make some apron areas (e.g., deicing) more difficult to effectively sample. 4.3.1.1 Survey Team To conduct an apron survey, deployment of a number of survey members (Figure 9) or team(s) may be necessary, based on airport-specific operational information and the desired level of sampling. The survey team should comprise the following: • Dispatch—The teams’ activities are suggested to be coordinated by a dispatch whose primary responsibilities include: (1) relaying information among teams, airport staff, or other person- nel; (2) directing back-up or contingency sampling regimes in the event aircraft operations significantly deviate from the flight schedule; and (3) making decisions to ensure worker health and safety (e.g., if safety, security, ambient conditions, or other airport-specific con- siderations preclude the research team from being positioned within the apron area, the dis- patcher may instruct that the apron survey operations shift to the inside of a vehicle, also referred to as a windshield survey. In advance of the field work, the dispatcher should assign each team a series of turn- arounds to observe on a given sampling day, based on sampling requirements and the day’s flight schedule. Each team can be allocated to a specific, non-overlapping series of gates/ apron areas where they can conduct assigned observations. Survey teams are suggested to be assigned apron area(s) for which they are responsible (e.g., Gates 12 through 20), but should not be charged with observ ing more than one terminal gate at a time (e.g., Gate 12). Each survey team should ideally consist of two to three members, filling the roles described below and depicted on Figure 10: • Controller—Responsible for aircraft turnaround activity observation, monitoring status of upcoming flights, and communicating to escort personnel and team dispatch any necessary mod- ifications to the transportation and sampling plans as flight arrival/departure status is updated. • Primary Observer—Responsible for communicating to the survey team GSE approach, egress, and dwelling events. • Scribe—Responsible for data recording as well as acting as a secondary observer. Figure 9. GSE apron survey team member. Figure 10. GSE survey team hierarchy.

GSE Data Collection Protocol 29 In cases where budget, staffing resources, airport regulations or other considerations pre- clude three-person teams, the primary observer may also fulfill the responsibilities of controller. Responsibilities of the scribe should not deviate from those described as this may impact the consistent and reliable recording of operational data. AEDT equipment emissions characteristics and data that can potentially be captured using an apron survey include aircraft utilization (e.g., cargo versus passenger, narrow-body versus wide- body), operating times (e.g., 30 minutes per enroute arrival), and GSE equipment parameters (e.g., equipment type, fuel type, horsepower). The following sections present a series of best practices on (1) designing and executing a sam- pling strategy for an apron survey; (2) establishing a sampling domain for apron observations; (3) creating and utilizing data collection field resources; and (4) recording aircraft GSE utiliza- tion and operating time data. 4.3.1.2 Designing a Sampling Strategy Shown on Figure 11, the first step in sampling design is evaluation of air- craft fleet operational data and anticipated flight schedules to identify the most prevalent carriers (e.g., Delta, UPS, etc.) and their most frequently flown operations (e.g., B752). The operational data are suggested to be obtained and assessed during early coordination steps (see Chapter 6). Reviewing these data offers a two-fold benefit in that it (1) increases the sampling success by maximizing opportunities to observe a particular operation during the sam- pling period and (2) ensures that the largest share possible of the airport fleet operations are being populated with airport-specific GSE operational data for emissions modeling. Depending on which carriers/aircraft represent the majority share of the airport fleet mix, and depending on how many observations would need to be captured to represent that major- ity share, the protocol user can then make decisions on subsampling, such as to what extent narrow-body and wide-body aircraft should be sampled, or to what extent should passenger or cargo operations be sampled. Once decided, preliminary sampling is suggested to be conducted with the aim of collecting at least 10 preliminary observations (e.g., observe 10 aircraft turnarounds). This enables the calcu- lation of sample size sufficiency based on the user’s acceptable levels of statistical uncertainty or confidence in the sample data. Once preliminary samples are collected, calculate the mean (x–) and stan- dard deviation (sx) of the data (e.g., GSE operating times) from a set of gate service observations and verify that the data to be tested are normally distrib- uted. If so, define acceptable levels of uncertainty (dr) as well as the minimum acceptable confidence interval (a) with which to determine whether the spec- ified uncertainty is statistically valid. Note, a is expressed as a fractional value, so for instance if the desired confidence is 95%, a would be expressed as (1 - 0.95 = 0.05). Each selected value of a is then related to a Z-score (Z1 - a/2) on the standard normal distribution and applied to Equation 3 to obtain an estimate of sample size (n).18 Idea—Airport operators can aid in reducing preparation times and costs for conducting airport GSE emissions inventories by collecting supporting data and information ahead of time. Reminder—When conducting airport GSE surveys, accurate and airport-specific data are preferred over generalized or presumed information. However, other real- world factors and considerations may impede or moderate this objective. 18 For a standard list of z-scores: http://www.stat.ufl.edu/~athienit/Tables/Ztable.pdf

30 Improving Ground Support Equipment Operational Data for Airport Emissions Modeling 1 2 2 n Z x d x r = σ   −α Equation 3 where x– = sample mean sx = sample standard deviation dr = defined level of uncertainty a = minimum acceptable confidence Z1 - a/2 = Z-score on the standard normal distribution table Transform DataYes Oponal Then No Evaluate Fleet Op- erations and Flight Schedules Are the Data Normally Distrib- uted? Enough Samples Col- lected? Determine Overall Sampling Scheme (e.g., 10 narrow-body and five wide-body flights) Specify Sub-sampling Strategy (e.g., five SWA narrow-body, five EGF narrow-body) Conduct Sampling Sampling Complete Yes Figure 11. Sampling strategy development.

GSE Data Collection Protocol 31 Shown in Table 5, for instance, after 16 observations of a lavatory service truck servicing a narrow-body aircraft, an average running time of 05:29 minutes with a standard deviation of 01:54 minutes was computed. According to Equation 3, 21 samples are required to accept 20% uncertainty in the computed average running time. Because only 16 samples were initially collected, five more samples would need to be obtained to accept a 20% uncertainty on the sample average, and even more samples would be suggested to further reduce the accepted level of uncertainty. Table 5 further illustrates that, with an assigned statistical confidence of 80%, additional sam- pling would be suggested to further reduce the uncertainty in the data averages to 10% (with the exception of the hydrant truck). The exercise demonstrates that there is a trade-off between estimated sample size, statistical confidence, and acceptable uncertainty/error. Given the inherent variability in GSE operations, and how this variability is impacted by airport-, aircraft-, airline-, and potentially climate-related variables, the protocol user might find it sufficient to accept 20% uncertainty in an estimated average, especially if there are time or budget constraints that would preclude additional sampling. Reducing the level of averaging in the data set (e.g., computing averages and sample sizes for each individual air carrier) may decrease the observed variance and uncertainty but would also require additional samples from each subsampling parameter to form a sufficient basis of statistical comparison. 4.3.1.3 Sampling Domain The event sampling domain (ESD) for each gate service event can be segregated into two areas (see Figure 12). The direct ESD is suggested to be generally demarcated using the apron/taxiway edge, boundary of responsibility line/stand safety line, restricted stand-by area lines, equipment parking lines, and vehicle limit lines on the airfield pavement. In addition, a conditional ESD is suggested to be included, encompassing adjacent gates to the left and right of the direct ESD so long as these gates are not occupied and it is obvious that equipment traversing or dwelling in the conditional ESD are servicing the aircraft turnaround occurring in the direct ESD. The only activity outside of this area that are suggested to be included in the ESD is aircraft tractor activity, as it is a typical operation for the tractor to leave to escort the aircraft and then return to the area to park. This is not necessarily the case for other equipment, so activities for these other assorted equipment outside of the designated sampling domain will be captured separately using other data collection methods (see Chapter 5). GSE Type Time per LTO Observed n Estimated n, = 0.05 (% uncertainty) 20% 15% 10% Aircraft tractor 00:05:29 00:01:54 16 5 9 21 Baggage tractor 00:19:56 00:08:17 17 7 13 30 Belt loader 00:44:09 00:26:08 17 11 20 44 Cabin service truck 00:31:10 00:09:19 14 4 7 15 Hydrant truck 00:08:38 00:02:18 14 3 5 12 Lavatory service 00:06:57 00:04:43 7 20 35 79 Bolded font signifies that additional samples are suggested. LTO = landing/takeoff operation. = sample average; = sample standard deviation; and n = sample size. Table 5. Example sample size, 80% confidence.

32 Improving Ground Support Equipment Operational Data for Airport Emissions Modeling 4.3.1.4 Field Resources Appendix B provides example apron survey data collection resources to be provided to field observers, such as the GSE Operating Time Observation Form shown as Figure 13. These resources are intended to facilitate logistics, ensure adherence to a HASSP, and familiarize the workers where necessary with the appearance of GSE and typical operations at an aircraft apron. Also provided in Appendix B are suggested templates for recording equipment utiliza- tion, operating time, and equipment data, including the following: • Team Sampling Itinerary—A matrix denoting the sampling order, scheduled time of arrival/ departure, scheduled gate, flight operator, flight number, and aircraft type. • Airport Planning Manual Terminal Servicing Guide—A graphical display of typical GSE ser- vice positions and approach order during aircraft arrival/departure services; one each per aircraft included in the team’s sampling itinerary. The intent of this information is to provide a general understanding of the likely locations and times GSE will be present at the aircraft. • Survey Team Gate Assignment Map—Indicates to a team which gate(s)/area(s) are under their responsibility and which are assigned to other team(s). • Visual Guide for Typical Apron GSE—Provides representative pictures of each type of GSE (e.g., belt loader), in the event an observer encounters a piece of equipment that they cannot identify with certainty or are otherwise unfamiliar. Figure 12. Apron survey gate service event sampling domain (ESD).

Figure 13. Airport GSE survey form. 10:56:10 11:01:30 Belt Loader Aircraft Type: Apron Type: Turnaround: Enroute: Operator: Tail: Time In: Time Out: AAL N585 10:17:09 11:10:20 Date: Airport: Gate: Flight: 7/10/2014 DFW C33 132 Team: Scribe: MD82 Terminal C B Joe Smith GSE Start Stop Start Stop Start Stop Forward Aft 10:20:01 11:02:35 10:18:53 10:24:54 10:20:32 10:20:53 10:25:27 10:28:31 10:19:29 10:19:57 11:08:18 11:13:12 10:19:19 10:38:50 10:19:20 10:41:04 10:30:08 10:41:21 10:40:45 10:47:32 10:39:50 10:43:10 10:18:06 11:03:17 10:20:12 11:03:50 10:28:09 10:30:22 11:05:05 11:08:59 10:37:41 10:43:41 Aft Aft Aft Aft Aft Aft Aft Aft Aft Aft Aft Aft Aft Aft Aft Aft Air Conditioner Heater Truck Truck Van Van Van Pickup GatePowered Pickup Pickup Cart Cart Power In/Push Back Tug In/Push Back Forward Forward Forward Forward Forward Forward Forward Forward Forward Forward Forward Forward Forward Forward Forward Forward Baggage Tractor Aircraft Tractor Air Start Cabin Service Truck Catering Truck Air Conditioners/Heaters Deicer Forklift Fuel Truck Ground Power Unit Hydrant Fueling Lavatory Service On-road Vehicle Water Service Passenger Stairs Other: 400 Hz gate power Other: PCA Other: Notes: Pick-up truck observed to be associated with refueling activities. Water service observed but not timed.

34 Improving Ground Support Equipment Operational Data for Airport Emissions Modeling • Equipment Data Plate Examples—Illustrates to the user what an equipment data plate looks like in the event that any information therefrom can be gleaned from parked GSE in the assigned areas (e.g., horsepower, model year). In addition, Appendix A presents an example Health, Safety, and Security Plan (HASSP) that provides a general description of the levels of personal protection and safe operating guidelines expected from all personnel performing the GSE surveys. 4.3.1.5 Aircraft GSE Utilization Aircraft utilization relates to the types of GSE required to service an aircraft, their service positions, and assigned duties based on an aircraft’s service purpose (e.g., passenger, commuter, cargo, mixed cargo); aircraft type (e.g., narrow-body, wide-body, regional jet); and trip configu- ration (e.g., turnaround versus en route). It is also advantageous for the survey team to record which airline or carrier is operating the aircraft to allow for the potential distinction of GSE utilization among carriers (e.g., some carriers practice “quick” turnarounds). To collect this information, field observers are suggested to be issued a GSE Operating Time Observation Form (Appendix B). The form has multiple formats tailored as to whether the obser- vation is occurring at a terminal apron, a cargo apron, or some other area. The form also provides opportunities to record aircraft tail number, type, flight number, carrier, gate, and trip configu- ration. During data post-processing, this information can be used to research additional infor- mation about the aircraft’s operation as well as to combine any secondary data (e.g., number of passengers, weight of baggage/cargo, trip distance) acquired for that flight via coordination with airlines/ground handlers to the utilization data collected during the apron observations. The GSE Operating Time Observation Form also allows for the identification of service posi- tions for select GSE (e.g., whether a baggage tractor is servicing the forward or aft loading door, or whether a cargo loader is moving cargo into the upper or lower deck of an applicable aircraft) (Figure 14). Further, the form provides the ability for the observer to differentiate aircraft tractor operat- ing times based on the aircraft maneuvering method observed. For example, aircraft maneuver- ing as they approach/depart the gate can be categorized and defined as follows: • Power-in/Pushback—The aircraft taxis in to the gate area under its own engine power until it is parked. The aircraft tractor is attached once flight safety checks and aircraft service have been completed, and the tractor powers the aircraft out of the gate area to the taxiway. The aircraft tractor is considered running from the time it leaves with the aircraft to the time it returns to its parking position at the gate. • Tug-in/Pushback—The aircraft is escorted to its gate stop line under the power of the aircraft tractor. The tractor operating time begins once the aircraft is in visual sight of the sampling domain, stops during gate service, and resumes once flight safety checks and aircraft service have been completed, when the tractor powers the aircraft out of the gate area to the taxiway. The aircraft tractor is considered running from the time it leaves with the aircraft to the time it returns to its parking position at the gate. 4.3.1.6 Operating Times (Gate Turnaround) Again, the main utility of an apron survey includes the recording of GSE operating times as they approach, egress, and dwell at an aircraft while performing the variety of passenger move- ment, cargo movement, and aircraft servicing functions. As mobile equipment encroach upon the sampling domain for a gate service event, observers should mark the time (HH:MM:SS) and assume the equipment has its engine running unless there is a visual signal that the equipment operator turns off the ignition. Equipment with multiple

GSE Data Collection Protocol 35 approaches within a gate service event, such as baggage tractors, are suggested to be assigned start and stop times (HH:MM:SS) each time they leave and return through the boundary of the sampling domain. The only exception is the aircraft tractor. Fixed or semi-fixed equipment, such as GPUs, are suggested to be considered operational once there is a clear signal that the device is being activated by an operator or attached to the aircraft, and considered non-operational if visual cues indicate it’s been deactivated. 4.3.2 GSE Inventory/Property Lists The preferred method for collecting information on GSE equipment parameters such as make/ model, engine make/model, fuel type, and power rating is via acquisition of airline/ground handler equipment inventories or property lists. These data are best obtained from the air and cargo carriers, ground support providers, and/or the airport (see Chapter 6). However, to supplement the preferred method, one can also attempt to collect as much of this information as possible from parked and/or dwelling equipment in the terminal, cargo, and/or GA areas. In this case, observers should, to the extent permissible, inspect the equipment cabins, engine compartments, and/or fuel compartments to locate the equipment data plate and record information that either directly fills out the fields on the Equipment Parameter Observation Form or provides the opportunity to look the information up at a later time (e.g., recording the serial number). Figure 14. Typical aircraft GSE positions. Idea—As discussed in Chapter 6, coordination with airlines, cargo carriers and other airport ground support providers may enhance the quantity and quality of the GSE database for computing a GSE emissions inventory.

36 Improving Ground Support Equipment Operational Data for Airport Emissions Modeling 4.3.3 Remote Sensing As mentioned at the beginning of this section, the initial draft protocol included ideas for testing the efficacy and utility of remote sensing techniques in GSE field surveys. The techniques identified for testing in the research included thermal imaging (e.g., to determine if GSE power is on/off), global position system (GPS) tracking of GSE, and video recognition of GSE activ- ity. During the protocol testing completed at several airports during the research, these remote sensing options were found to be unreliable, not cost-effective, and/or not acceptable within the security requirements of the airports. For these reasons, remote sensing techniques for collecting GSE data in support of computing emissions inventories are not suggested at this time. 4.3.4 Data Collection Time The amount of time required to collect GSE data is dependent on the airport’s arrival and departure schedule and the type of aircraft being surveyed. For example, cargo aircraft typically have a longer turnaround time than passenger aircraft. As a result, the amount of time needed to sample cargo aircraft may be longer than the amount of time needed to sample passenger aircraft. During the planning phase of sampling, the dispatcher is assigned the task of organiz- ing a series of turnarounds for each team to observe. If possible, in order to reduce the time needed to collect results, it is highly suggested that the dispatcher select successions of quick turnarounds. As a means of providing an example of data collection time that is needed to survey, the days and number of samplers for each airport surveyed for ACRP Project 02-46 are shown in Table 6. 4.4 Quality Assurance The purpose of a quality assurance routine is to both qualitatively and quantitatively assess collected data to ensure that the data are useful, reliable, and valid before the data are used in follow-on analyses. Figure 15 offers a systematic approach to assuring data quality collected using this protocol. 4.4.1 Data Cleaning The process of data cleaning is also referred to as code and value cleaning. The purpose is to determine whether each variable in a data set contains only legitimate codes or values and whether these seem reasonable. For example, if GSE fuel mix data are collected and are coded Airport Type of Sam- pling Number of Days Sampled Number of Samplers Number of Turnaround Observations Logan International (BOS) Boston, MA Passenger only 2 2 16 Portland International (PDX) Portland, OR Passenger only 2 1 11 Dallas-Fort Worth International (DFW) Dallas, TX Passenger only 2 2 18 Oakland International (OAK) Oakland, CA Cargo only 3 1 11 T.F. Green (PVD) Providence, RI Passenger only 3 2 11 Table 6. Airports sampled, survey days, and number of samplers.

GSE Data Collection Protocol 37 (e.g., D = diesel, G = gasoline, L = LPG, C = CNG, and E = electric), and all data under the fuel type variable are coded with only these letters, then the variable is considered clean. In contrast, if there is a value coded with X or a value that is not a character but a number, then that value would need to be verified. In terms of GSE operating times, if the modal GSE service time during a turnaround was 30 min- utes with a range of between 20 and 40 minutes, and it was discovered that there was a turnaround time of 12 minutes (which could be a legitimate value), then the authenticity of that value would need to be verified, possibly through other sources. The cleaning process helps flag these illegiti- mate and unreasonable values. If warranted, they can be considered outliers to be eliminated, sub- stituted with the correct value if recorded incorrectly, or treated as missing values to be replaced. Yes Perform Code and Value Cleaning Then No Are there missing values in the data? Are data MAR/MCAR? Are there outliers in the data? Are the Data Normally Distributed? Are the data unbi ased? Screen Outliers, if Desired Transform Data No Idenfy and Address Bias No Quality Assurance Complete Perform Imputaon of Missing Values Yes Yes Yes Yes Yes Then Then Then No Perform Listwise or Pairwise Deleon Then No Then No Do the data contain any illegimate codes and/or values? Figure 15. Quality assurance routine.

38 Improving Ground Support Equipment Operational Data for Airport Emissions Modeling 4.4.2 Missing Data Encountering missing data may happen for a number of reasons, including having limited access to information or missing values due to random processes. Randomly missing data can be (1) missing completely at random (MCAR) or (2) missing at random (MAR). Data are MCAR when the missing data are in no way related to any other variables or other missing information in a given data set, and there is no pattern to the missing data, which rarely occurs. On the other hand, missing data are MAR when their absence can be explained, or is dependent on, other variables for which one has full information. For example, missing data for fuel truck operating times in the range of 50 to 60 minutes would not be MAR if most of the operating times fell in the 50 to 60 minutes range in the first place. If the missing data conform to MCAR or MAR criteria, then there is no explicit problem with ignoring or deleting the missing observations, but with the caveat that there may be complications if there is a significant amount of missing data. Options for the deletion of MCAR/MAR data include: • List-wise deletion—In this instance, if an arrival time for a baggage tug is recorded and the depar- ture time is missing, both observations can be discounted despite the validity of the recorded arrival time. Concerns with this method are that resulting sample size reduction may increase the estimate of measurement error, and the survey may not meet the sample size requirement. • Pair-wise deletion—In this case, if an observation has missing data for any one variable, but there are data present for other variables, the observation can still be used. For example, if a series of eight aircraft turnarounds were observed and only six of them recorded a baggage tractor, then summary statistics for the baggage tractor would be computed on the six available measure- ments. Meanwhile, if an aircraft tractor was observed for all eight of these turnarounds, the sum- mary statistics for the aircraft tractor would be computed based on all eight measurements; that is, the sample set would not be reduced to six just because of the missing baggage tractor values. If data do not conform to MAR or MCAR criteria, missing data cannot be ignored and the data set is subject to imputation—referring to the practice of substituting a missing value with one that is a reasonable approximation or surrogate. One method is called mean substitution, where all missing values of a variable are replaced with the average of that variable. This is reasonable when the sample mean is a good estimate of the total population mean. Another imputation method is via multivariate regression analysis, whereby a regression equation is generated with only complete data, using several independent variables to predict a dependent variable (the one with missing data). Some statistical software and tools available for data screening include SPSS, NCSS, and Microsoft Excel. SPSS is capable of executing the aforementioned data screening techniques. NCSS performs data screening in a database and reports the type of data (discrete or continuous), normality of each variable, missing value patterns, and presence of outliers. Furthermore, Microsoft Excel has an add-in functionality called Data Analysis. Some useful tools include creating histo- grams, obtaining descriptive statistics, and doing regression on data. Other statistical software that may be useful includes SAS, Stata, and R. 4.4.3 Outliers and Data Distribution Analyzing the distribution of data allows one to identify outliers, verify the normality of the distribution, and find trends among variables. Some visual tools for displaying data include frequency tables, histograms and bar graphs, box and whisker plots, and scatterplots. The first four tools display data for one variable at a time (i.e., univariate), while the last tool looks at the relationship between two or more variables (i.e., multivariate).

GSE Data Collection Protocol 39 With respect to an airport GSE survey, the univariate case is suggested to be primarily consid- ered because emissions calculations for each type of GSE is additive and some variables are inde- pendent of one another. Upon data collection, post-processing, and quality assurance screening, the potential for multivariate analysis can be additionally considered. Options for displaying the distribution of data are briefly discussed below. • Frequency table—A frequency table provides counts of how many times a given variable is observed in a data set, and is useful in that it easily reveals data entry errors (e.g., using non- existent or erroneous values for fuel type). • Histograms/bar graphs—These represent a frequency table in graphic form and are useful to visually inspect the general shape of the distribution. Histograms are suggested to be used for frequency counts of continuous variables (e.g., turnaround time) and bar graphs for fre- quency count of categorical variables (e.g., fuel type). Descriptive statistics give more detailed information on how close the distribution is to a normal distribution, conveyed as either skewness or kurtosis (Figure 16). As shown, skewness relates to the symmetry of a distribution. A data set that is not approximately symmetric about its sample median is skewed. “Skewed to the left” is when the histogram has a long tail to the left, whereas “skewed to the right” signifies a long tail to the right. Kurtosis refers to the clustering of scores toward the center of a distribution. Data can be distributed close to the mean (small standard deviation) or far from the mean (large standard deviation). • Box and whisker plot—A box and whisker plot shows the spread or distribution of a data series based on calculated statistical quartiles and percentiles and is a good visual tool to identify extreme values in numerical data. An important feature of a box and whisker plot is the interquartile range (IQR), which is a measure of statistical dispersion and is defined as the difference between the third quartile (Q3, or the 75th percentile of the data) and the first quartile (Q1, or the 25th percentile of the data). Conventionally, an outlier is defined as a data point that is farther than ±1.5 IQRs from the median value (i.e., the 50th percentile) but less than ±3.0 IQRs. An extreme outlier exceeds ±3.0 IQRs. Figure 17 provides an example box and whisker plot and denotes the demarcations for Q3, Q1, IQR, median, 1.5 IQR, and outliers. • Scatterplot—It may also be useful to use scatterplots to compare two or more variables at a time (Figure 18). For example, for a belt loader, the forward position operating time can be plotted against the aft position operating time, or vice versa. In examining the interrelationship of two or more variables, unusual patterns of variability may be found among the variables. For instance, such plots may reveal extreme values that can be labeled as outliers. Furthermore, trends may be found among variables that were otherwise undiscernible. Figure 16. Skewness/Kurtosis about the normal distribution.

40 Improving Ground Support Equipment Operational Data for Airport Emissions Modeling 4.5 Sampling Error and Bias Sampling error occurs when a value derived to represent a large population is estimated from a subset of the population; it is the difference between the estimated value and its true popula- tion value due to random selection. It can be estimated using statistical confidence intervals and probability values. Because of the nature of the data collection methods, especially windshield surveys and remote sensing and measurement, it is important to know the estimated error of the calculations. Sampling bias occurs when a sample is not truly random and representative of the population. Figure 17. Example box and whisker plot. Figure 18. Example scatterplot.

GSE Data Collection Protocol 41 For each data variable to be collected, the sampling error can be calculated using statistical soft- ware and tools. For instance, one could compute a 95% confidence interval for a set of normally distributed data and perform hypothesis testing on the data using a t-test. A two-sided t-test at a 95% confidence level tests a hypothesis on the computed sample average. If the hypothesis is not statistically rejected, it signifies that one could be 95% confident that the sample average lies within the true population average, given a margin of error assigned specific to the 95% confi- dence interval. Similarly, a one-sided t-test would test whether the sample average is significantly less than or greater than the true population average. A test for statistical significance, using statistical values called p-values, is also commonly conducted in conjunction with these types of hypothesis tests. Because every airport is different, there are different standards and practices for GSE utiliza- tion. Each airline may have different practices for GSE as well. Moreover, there may be limitations on the selected airports and aprons for the study, so GSE utilization may not be representative. Therefore, to some degree sampling bias is unavoidable. For example, it may be that only certain airlines allow apron surveys to be done, and these airlines’ GSE procedures have a greater turn- around time on average than other airlines. Then the operating times for GSE would be over- estimated. Although a larger sample size would increase the precision of the results, this may not reduce sampling bias.