Guidance for Quantifying the Contribution of Airport Emissions to Local Air Quality (2012)

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.

19 Modeling Guidance and Findings 3.1 Overview The modeling guidance provided in this chapter is a mixture of findings from the literature review and the airport modeling assessments that were conducted during this project. Although there may be some overlap, the guidance does not seek to duplicate existing guidance materials, including the measurement and modeling protocols mentioned in the previous sections. As a result, the guidance is intended for experienced modelers. The guidance mainly focuses on air quality (as opposed to emissions) modeling capabilities of EDMS and CMAQ with supplemental discussions on the potential use of receptor models as well. (These models were selected based on the reviews presented in Appendix A. The assess- ments conducted using the models are presented in Appendixes E through G.) In addition to some of the modeling protocol-type issues, the guidance helps to better under- stand the accuracy and impacts of using different datasets and options in the models. In this way, the guidance helps to better allocate resources so that, for example, 95% of the effort is not wasted on 5% of the work. These guidance points as well as other conclusions are summarized in Chapter 6. Although guidance is provided, the information also adds to the growing knowledge base on airport air quality contributions. As such, it is expected that the guidance and the results pre- sented in the various appendices should be used as a basis for further studies. 3.2 EDMS and AERMOD 3.2.1 Modeling Accuracy and Applications Based on the modeled-versus-measured results presented in Appendix E, there appears to be much room for improvement with the models themselves (EDMS and AERMOD) but perhaps more importantly with the input data. This includes not only correctly representing the sources and their activity information, but using representative meteorological data, especially wind data. When modeling ambient concentration contributions from airports, the level of accuracy for receptor locations that are relatively far away from sources, especially runways, may not be very high. For example, under this project, the accuracy at the first and second base stations which were about 1,500 ft and 2,600 ft away, respectively, from the closest runway was low. How- ever, for those receptor locations that were only a few hundred feet from a runway, the accuracy was generally higher. This appears to support the understanding that at closer distances, the potential for influence by erroneous input data decreases. That is, the closer the receptors are to sources, the greater the influence of each source’s emissions on the concentrations. C h a p t e r 3

20 Guidance for Quantifying the Contribution of airport emissions to Local air Quality For regulatory-based modeling, receptor locations may be placed at relatively far locations (e.g., locations accessible by the public) in the vicinity of the airport. As such, concentration- prediction errors may be up to a magnitude (i.e., a factor of 10) for each pollutant based on the maximum and minimum errors witnessed as part of the assessments. However, using CO concentrations as an example, the average error as represented by the modeled-versus-measured comparisons conducted for the base station is approximately 120%. In contrast, the average error at Sites K and M (very close to a runway) was about 23%. In addition to these accuracy issues, the differences between the application of AERMOD for generally well-defined stationary sources and airports need to be understood. AERMOD was developed with a focus on meeting regulatory requirements. Although its ability to model line, area, and volume sources may imply the potential to model mobile sources, some source char- acteristics may be difficult to capture with these sources types. More importantly, the input data necessary to accurately represent the sources may be difficult to implement. For example, the release heights of the exhaust from aircraft main engines, APUs, and GSEs may not be accurate. Also, the movements of these sources may not be accurately represented by the area sources currently used in EDMS. That is, the user may not be able to correctly identify the size of the GSE activity area around a gate and may just be guessing at which taxiways are used in EDMS by specific aircraft types. More detailed protocols need to be developed to improve and make consistent the development of source characterization data. Therefore, the errors in the mod- eled concentrations is partly based on the underlying modeling methodologies (e.g., dispersion methods in AERMOD), but is likely based in large part on the application of the models to the varied sources found at an airport. This perspective is necessary; otherwise, both EDMS and AERMOD could be wrongly criticized as performing poorly. 3.2.2 Pollutants and Chemical Transformations AERMOD allows the modeling of CO, NOx, sulfur dioxide (SO2), particulate matter with an aerodynamic diameter of 10 um or less (PM10), and total suspended particulate matter (TSP). Each of these pollutants can be modeled using the assumption that no chemical transformations occur, even for NOx and SO2. The optional ozone-limiting methods used for NOx and the SO2 half-life decay parameter allow some semblance of modeling first-order chemical transforma- tion effects. Although these options exist, AERMOD is mainly considered a steady-state dispersion model with sophisticated characterizations of the atmosphere’s dispersive potential. This is indicative of the “OTHER” keyword that can be used in AERMOD to specify other nonreactive pollutants. The non-chemistry aspects of AERMOD are relatively common among most other Gaussian plume models. This does not mean that reactive pollutants cannot completely be modeled in AERMOD. Three categories of chemical transformation can be defined: (1) pollutants such as ozone which are formed from other “input” pollutants (e.g., NOx and VOCs); (2) relatively fast reacting pollutants (e.g., nitric oxide [NO]); and (3) relatively slower reacting pollutants (i.e., those with lower reaction rate constants). Except for NOx, the first two categories of pollutants can- not be modeled in AERMOD. However, if a pollutant’s reaction rates are low enough, they can be assumed to be nonreactive and modeled in AERMOD. The potential to do this will depend largely on the location of the receptor with respect to the source(s), wind speed, and wind direction. In simple terms, if the dispersing pollutant can reach a receptor before the pol- lutant experiences any significant transformations, then AERMOD can be used to predict its concentrations.

Modeling Guidance and Findings 21 3.2.3 Modeling Options Fortunately, EDMS already includes data and methods for representing certain plume charac- teristics and behavior such as the initial dispersion parameters (initial s values) and plume rise (e.g., through thermal buoyancy). Although the specifications for these effects can be modified in the AERMOD input file(s) created by EDMS, it is not recommended, unless the new data can be justified technically and the responsible agency consents to the changes. This is also true for most of the other modeling options, including those visible through the EDMS interface used to both create and run the AERMOD input file. The default SO2 decay option in AERMOD (and as presented in the EDMS interface) is based on a nominal 4-hour half-life. In EDMS, the default setting is not to use this option. Similar to the previous discussions on chemical transformations, the use of this option should mainly be based on a combination of receptor location in relation to the source(s), wind speed, and wind direction. For the receptor locations modeled within the airport under this study, the decay option produced virtually no difference in concentrations. For regulatory purposes, such decay options are normally not used. Similarly, the non-regulatory options for modeling NO to NO2 conversion were also exercised. Given that the EDMS interface does not provide access to these options, they were exercised directly through the AERMOD input file. Based on the understanding (see Appendix E) that the Plume Volume Molar Ratio Method (PVMRM) is more accurate than the Ozone-Limiting Method (OLM), the PVMRM option was used with a field-measured NO2/NOx ratio of 0.7 (derived from the IAD measurements). As this value was specific to the measurement condi- tions and location of receptors at IAD, it should not be considered to be representative of other airports and is only presented here to add to the existing knowledge base. The modeling results showed little difference between using and not using the PVMRM option. The option to use terrain data was also exercised. The flat terrain-modeled data were com- pared to those from using 1-degree Digital Elevation Model (DEM) data covering the region surrounding IAD. The resulting comparisons showed virtually no difference in modeled con- centrations. It is suspected that because most of the receptors were placed within the boundary of the airport which is relatively flat, the additional fidelity provided by the DEM data made little difference. As such, if receptors are placed outside, but still relatively close to the airport, the assumption of flat terrain may be justified. To gain a better understanding of the impact of elevated plumes, a sensitivity assessment was conducted to determine the impact of using different receptor locations, including flag pole receptor heights. In order to account for thermal buoyancy, EDMS elevates the taxiway and run- way area sources to a height above the average aircraft engine exhaust height. A potential concern here is that any errors associated with the estimated plume rise can cause errors in the modeled concentrations, which are usually conducted at 1.8 m, which approximates human breathing height. Based on the assessment conducted under this project, the impacts caused by an elevated plume may have noticeable effects for distances within about 10 m from a runway source. How- ever, for receptor locations about 60 m away from the runway (and probably closer), the impact of the elevated plume appears to be minimal (i.e., concentration at 1.8 m height is similar to the concentration at the centerline). Although these assessments were specific to the runway at IAD, the overall conclusions are likely to be similar at other airports. 3.2.4 Spatial and Temporal Protocols In developing a modeling plan, the modeling protocols from EPA should be followed as much as possible (USEPAd 2011 and CFR 2005). These protocols are intended to help guide the user

22 Guidance for Quantifying the Contribution of airport emissions to Local air Quality in making technically sound and consistent decisions that focus on meeting regulatory require- ments, including criteria for placement of receptors and modeling periods used to identify the appropriate concentrations for use in addressing regulations. The decisions for receptor place- ment should not be based on the number of receptors, but on other criteria such as public access, topography, and weather. Modeling periods should be based on justifying adequate coverage of variability (e.g., 5-years’ worth of meteorological data). In general, these protocols should be applicable to most airport studies, but the local respon- sible regulatory agency should be consulted to ensure that all local or regional practices are followed and requirements met. These requirements include the use of state-specific emissions and air quality standards. 3.2.5 Input Data Fidelity Although some of the modeling options may not have significant impacts under the situations discussed, the overall fidelity of the input data can have significant impacts. To illustrate this, two different fidelity levels were modeled (as discussed in Appendix E). While both cases used the same aircraft flight schedule to provide fleet mix and operating information, the lower fidelity case involved the use of default EDMS data for other sources such as GSEs, APUs, and GAVs. Although not all pollutants showed noticeable differences, the accuracy level for CO concen- trations showed noticeable improvements in modeled-to-measured comparisons. Part of this is likely because most receptors were relatively far away from the impacted sources (i.e., GSEs, APUs, and GAVs). But more importantly, the fact that the aircraft fleet mix and operations were the same for both cases seems to corroborate the significant influence aircraft can have on airport air quality. 3.2.6 Source Contributions While it is common knowledge that aircraft are generally the largest source (or one of the largest sources) of emissions at an airport, the study also demonstrated that aircraft generally appear to be the biggest contributors to ambient concentrations as well. The magnitude of air- craft emissions appears to override any effects of plume rise and other dispersion-related fac- tors. While this appears to be true for receptor locations that are relatively far or in-between the various sources, it will depend on location of receptors with respect to the sources. For example, a receptor located very close to a nearby highway will likely be heavily influenced by on-road vehicle emissions—possibly more than aircraft emissions. Depending on how the boundaries are specified, passenger vehicles (GAVs) entering and exit- ing the airport may provide a significant amount of emissions, perhaps even more than aircraft. Therefore, it is important that roadways and portions thereof are properly identified in an air- port air quality assessment. In general, roadways should be included to the point where the destination of the associated traffic is clearly the airport. Because aircraft and GAVs tend to be the biggest airport contributors to local ambient con- centrations, resources should be appropriately applied to focus on these sources first. It should also be recognized that airport contributions to air quality in locations that the public typically has access will tend to be small. In addition to sources, meteorological data should also receive significant attention as the use of wrong or non-representative meteorological data could result in erroneous concentrations, no matter how accurate the source fleet mix and activity data are. Therefore, the “significant” emissions from airport sources need to be tempered with the realistic locations of the public and the dispersive characteristics of the local atmosphere when planning the assessment of airport contributions.

Modeling Guidance and Findings 23 If source contribution modeling assessments are to be conducted using EDMS, it is recom- mended that a complete study first be developed that includes all sources. Using this study, EDMS allows the specification of individual or a group of sources when creating an AERMOD input file. Also, multiple copies of the study can be made, and then each study can be dedicated to a single or group of sources by deleting the unwanted sources (e.g., if various sensitivity-type studies are to be conducted with a subset of the sources). These approaches are recommended rather than starting with individual studies for each source. The reason is that aggregation of sep- arate studies is difficult. Other than inputting the data for one source into the study of another source through the EDMS interface, the only other way would be to copy the appropriate dataset from one study’s system data tables to the other. Neither of these approaches is recommended as they are tedious and error prone. In addition to modeling ambient concentration contributions from airport sources, proper background concentrations should also be developed to ensure proper reporting of total con- centrations. USEPA protocols specify the use of nearby monitor data to determine background concentrations. Consistent with this protocol, more specific guidance is provided in the next section using IAD as an example. Developing accurate background concentrations is very important because airport contribu- tions for most pollutants are likely to be smaller than those from surrounding sources unless a receptor is located very close to an airport source. As such, the accuracy of the total concentra- tion may depend in large part on the accuracy of the estimated background concentration. The modeled-versus-measured comparisons conducted under this project showed that varying the fidelity of background concentrations can have noticeable impacts on overall accuracy. 3.3 CMAQ 3.3.1 Overview As discussed in Appendix F, the application of CMAQ for the IAD region was demonstrated at a horizontal resolution of 12 and 4 km to model both primary and secondary pollutants. CMAQ, being a one-atmosphere model, was used to model various gas-phase pollutants–CO, NO, NO2, O3, SO2, particulate matter of size 2.5 (PM2.5) and 10 (PM10) microns and their various chemical constituents, as well as various hazardous air pollutants or air toxic species such as formaldehyde, acetaldehyde, acrolein, 1,3-butadiene, benzene, toluene and xylene. These HAPs were chosen based on a health risk prioritization study of aircraft emissions related air pollutants (Levy 2008). 3.3.2 Modeling Domains For modeling air quality contributions from a single airport, it is recommended that a nested model application of 12 and 4 km be developed. While the 12-km modeling domain is intended to capture the broad regional background, the 4-km domain will provide adequate spatial resolution to model the impacts of airport sources. Furthermore, based on results from this work as well as previous airport air quality studies, the secondary components of PM2.5 due to airport emissions change at distances of up to 300 km away from the airport. Thus, the spatial extent of the modeling domain should be at least 500 km at the coarse resolution (12 km), and at least 200 to 300 km at the fine resolution (4 km). However, if multiple airports are being modeled in a single study, and these airports are at least 200 to 300 km apart, a single 12-km model application can be developed. The vertical resolution of the model should be comparable to what was used in this study, i.e. 22 layers from the surface to 50 millibars (about 18 km), with about 15 layers spanning the lowest 10,000 feet, where aircraft emissions are modeled during the Landing and Takeoff (LTO) cycle.

24 Guidance for Quantifying the Contribution of airport emissions to Local air Quality 3.3.3 Input Requirements CMAQ requires the following types of major inputs: • Emissions (from non-airport sources): A complete inventory of all anthropogenic and natu- ral sources in the modeling domain needs to be created for use in CMAQ. Such datasets can be obtained from the USEPA’s National Emissions Inventories (NEI) from http://www.epa. gov/ttnchie1/eiinformation.html. Appendix F provides a complete list of source sectors from which emissions were estimated. Given that the USEPA develops and publishes the NEI on a 3-year cycle (e.g., NEI-2005 and NEI-2008), usually, the emissions need to be adjusted for the modeling period using control factors that take into account both emissions increases due to increases in economic activity and emissions decreases due to the effects of various regulatory control programs. These control factors are available as part of projection year inventories created by EPA for various regulatory needs. After acquiring all the emissions inventories from the EPA or any other state/local agency that has jurisdiction over the airport region, one should use the Sparse Matrix Operator Kernel Emissions (SMOKE) modeling system to cre- ate CMAQ-ready emissions. SMOKE performs three main functions—chemical speciation, temporal allocation and spatial allocation, and the outputs are usually hourly estimates for each hour of the modeling period for each grid-cell in the modeling domain. In processing the NEI emissions, care must be taken to identify emissions from the study airport and remove them, to avoid double counting of these sources. • Emissions from airport sources: These can be based on the EDMS modeling. The user can use the EDMS2Inv utility (Baek 2007) to process the emissions in a format that SMOKE can use and then use SMOKE to prepare them for CMAQ. Care must be taken to use the appropriate chemical speciation profiles for airport sources in SMOKE. EDMS estimates PM2.5 emissions from aircraft engines based on the First Order Approximation (FOA) and provides non- volatile PM (PEC) and volatile PM broken down into primary organic carbon (POA) and primary sulfate emissions (PSO4). EDMS provides total volatile organic compounds (VOCs) from aircraft engine as a single number. The FAA and EPA have jointly developed a specia- tion profile for total organic gases (TOG) (USEPAb,c 2009), and this profile varies depending on whether the aircraft engine is piston-driven or turbine-driven. The user must choose the appropriate TOG speciation profile depending on the engine type in the fleet mix. SMOKE uses default chemical speciation and temporal profiles for all other airport sources (based on previous studies) which are built into the modeling system. • Meteorology: CMAQ needs meteorological inputs from a prognostic model such as MM5 or WRF. Sample WRF model configuration with preferred physics options are provided in the Technical Memorandum. The outputs from MM5 or WRF are processed through the MCIP utility, and the outputs are hourly for each hour of the modeling period for each grid-cell in the modeling domain. As a mesoscale model, WRF relies on data from a model run on a larger domain for boundary conditions. To create meteorological inputs that coincide with a measurement period, it is beneficial to take advantage of the real-time forecasted meteorologi- cal model outputs from the National Center for Environmental Prediction’s (NCEP) 40-km North American Model (NAM). This model is run over the entire North American continent four times daily at 0, 6, 12, and 18 Z. Each of these model runs has an analysis field created by blending the results for that time from the previous model run with observations to create an updated set of initial conditions. These fields can be considered as the best representation of the atmosphere at a given point in time. For the demonstration under this study, an automated download of outputs from the NCEP ftp site (ftp://ftp.ncep.noaa.gov) was set up and created tools to downscale from NAM to drive WRF at 12-km resolution. In the WRF simulations, these NAM data were also used for initial conditions and for grid-based nudging. • Initial and Boundary Conditions (IC/BC): To obtain regionally representative initial and background air quality conditions, one can use either outputs from another regional-scale

Modeling Guidance and Findings 25 air quality model application or from climatological data. Under this study, real-time experi- mental CMAQ forecasts from NCEP for the nation were relied on. An automated download of outputs from the NCEP ftp site was set up and tools were developed to create CMAQ-ready IC/BC files for each hour of the modeling period for the 12-km resolution. The 4-km resolu- tion IC/BCs are created using outputs from the 12-km CMAQ outputs using CMAQ’s icon and bcon utilities. • Ocean-mask File: This is a file that defines whether each grid-cell in the domain is on land or on the ocean. All grid-cells that either border or completely span the ocean are needed to create surf-zone information necessary to predict coarse mode (particulate matter of size between 2.5 and 10 microns) aerosols from sea surf. This file is created using the Spatial Allo- cator tool developed and supported by CMAS. • Photolysis Rates: CMAQ also needs a photolysis lookup table for modeling certain species that undergo photolytic reactions. This table is developed through a standard CMAQ preprocessor called phot. 3.3.4 CMAQ Model Configuration Given that the CMAQ code is modular, the user has flexibility in configuring the model by choosing various modules and science process algorithms as necessary. For example, for each science process listed in Table 3-1, multiple module choices are available. Some combination of modules makes the overall model configuration incorrect and leads to unexpected or even erroneous results in the model application that are not easily trackable. As with any model, it is recommended that the user review the documentation thoroughly, understand the pros and cons of each module, and adopt recommended configurations for urban-scale studies by con- sulting with the EPA or other expert model users. A sample CMAQ model configuration based on this study is provided in Table 3-1. The same configuration was used for both the 12-km and 4-km resolutions. 3.3.5 Resource Requirements Unlike the use of “simpler” plume-based models, it is worthwhile to discuss resources and runtimes necessary to conduct grid model runs. In the tables below, the computational resource requirements for the CMAQ modeling are provided. Table 3-2 presents the daily input and out- put file sizes for each of the two modeled grid resolutions and the totals for the month-long data- sets. Recent versions of CMAQ provide an option to perform inline processing of both biogenic and major elevated point sources. Using this option will reduce the size of emissions input files, but will increase the computational time modestly. In Table 3-3, the CPU-hours taken for each Table 3-1. CMAQ Science configuration. Science Process Module Name ModDriver Ctm_yamo ModInit Init_yamo ModHadv Hyamo ModVadv Vyamo ModHdiff multiscale ModVdiff Acm2_inline_txhg ModPhot Phot_table ModChem Ebi_cb05cltx_ae5 ModAero Aero5_txhg ModAdepv Aero_depv2 ModCloud Cloud_acm_ae5_txhg Mechanism Cb05cltx_ae5_aq PAOpt Pa_noop

26 Guidance for Quantifying the Contribution of airport emissions to Local air Quality modeled grid resolution for each modeled scenario are presented. These numbers are based on model simulations performed on a Linux cluster with Xeon 2.0 or 2.8 or 3.0 GHz quad-core pro- cessors and with each execution using eight CPUs in parallel. More information about the Linux cluster and the technical specifications can be found at: http://help.unc.edu/6020. The estimates in both Tables 3-2 and 3-3 need to be scaled for additional months/seasons that will be modeled. 3.3.6 Evaluation and Accuracy Evaluation of WRF and CMAQ is performed through the use of the Atmospheric Model Eval- uation Tool (AMET). WRF is evaluated against observations from the Meteorological Assimila- tion Data Ingest System (MADIS) observational data, available from http://madis.noaa.gov/. The MADIS data includes NWS surface and Rawindsonde Observations (RAOBs), MESONET surface monitors, and the wind profiler networks. The AMET toolkit allows for comparisons of wind speed and direction, temperature, and specific humidity. Since this study was conducted in a near real-time forecast mode, WRF outputs for both a 24-hour (D1) and a 48-hour (D2) forecast were developed and evaluated independently, and no significant differences in the model outputs between the D1 and D2 forecast outputs were found. Furthermore, for all three seasons modeled, the 4-km model application was found to consistently give better performance than the 12-km outputs, emphasizing the need for finer grid resolution to capture local-scale features at the urban scale. CMAQ is evaluated against observations from several routine monitoring networks in the United States [e.g., the Air Quality System (AQS), Speciated Trends Network (STN), Clean Air Status and Trends Network (CASTNet), and the Interagency Monitoring of Protected Visual Environments (IMPROVE) network]. Most of these data are available for download from http://epa.gov/ttn/airs/airsaqs/detaildata/downloadaqsdata.htm. However, it usually takes a few months because the measurements are taken for EPA to make these data available on this website. During the near real-time modeling, the model was evaluated on a daily basis using data from EPA’s AIRNOW system (http://airnow.gov/) which is used for providing air quality guidance for the nation. This includes only O3 and total PM2.5 from the AQS monitors. Data for the previous day are usually available within 24 hours on this website, and a system was created to acquire and ingest these data in near real-time for the CMAQ model evaluation. However, although the availability of these data from AIRNOW are timely and assist in obtaining an initial evaluation of the model performance and help identify weaknesses, they are subject to further screening and QA before they are made final and posted to the first link mentioned earlier; hence, the data quality is not as robust as the former. Table 3-2. Data requirements (in Gigabytes) for CMAQ inputs and outputs. 12-km 4-km Both Grids Per Month Inputs - Meteorological 4.7 0.6 5.3 186.3 Inputs - Emissions (Base Case) 3.4 1.4 4.8 169.4 Inputs - Emissions (Sens Case) 0.3 0.1 0.4 13.7 4.62 8.0 3.0 5.0 rehtO - stupnI Outputs (Base Case) 17.4 7.3 24.7 863.2 Outputs (Sens Case) 17.4 7.3 24.7 863.2 Total 43.6 17.1 60.6 2122.2 Table 3-3. CPU requirements for CMAQ simulations. Total CPU-hours / model month Average CPU-hours / model day 12-km 4-km 12-km 4-km Each scenario 148.4 132.6 3.9 3.5

Modeling Guidance and Findings 27 Besides evaluating against routine monitoring, this study also gave an opportunity for an integrated measurement and modeling-based assessment of air quality impacts due to airport emissions. Specifically, CMAQ was evaluated against additional sets of on-site measurements of both gas-phase and aerosol species. In addition to total PM2.5 mass, speciated measurements of sulfate, EC and OC, and size-segregated measurements from the rotating drum impactor (RDI) were performed. Although evaluations of CMAQ against RDI measurements are not routine, this gave an opportunity to evaluate the relevant size fraction of PM2.5 that is more critical from the perspective of AQ impacts of aviation emissions. However, given CMAQ’s modal treatment of the formation of particulate matter, it takes additional effort to convert CMAQ’s predictions of PM to sectional values so they can be compared. (The details of this additional effort along with the results of the model evaluation are presented in Appendix F.) The overall goal of the model evaluation was to ensure that CMAQ model performance is within the goals and criteria recommended by Boylan (2006) and by the USEPA for such urban- to-regional scale applications (USEPAb 2007). Although this application for assessing air quality impacts from an airport is not a regulatory application for SIP purposes, the USEPA’s modeling guidance presents a good overview of considerations necessary to develop a CMAQ application, and procedures for model evaluation. 3.3.7 Limitations Given that CMAQ is an urban-to-regional scale model, the 4–12 km resolution used above does not capture the fine scales of spatial variability that may be observed in and around an airport. Although a preferred option would be to develop an even finer resolution model application (e.g., 1-km or less), it is usually computationally intensive, and developing meteorological inputs using a prognostic model at that resolution is even more challenging. Thus, depending on the layout of the airport, the use of three or four grid-cells that span the airport property at the 4-km grid resolution is typically used. This limitation is further addressed in the next section. 3.4 CMAQ and AERMOD Hybrid Modeling The overall goal of hybrid modeling is to use CMAQ outputs at a relatively coarser resolution of 4–12 km to provide secondary components of PM2.5 and then use AERMOD to predict con- centrations of primary pollutants at a much more spatially resolved network of receptors. The rationale for this is that secondary pollutants tend to be more spatially diffused, while primary pollutants have a much more localized signature, especially when considering airport sources. While details of the algorithm for performing hybrid calculations are provided in Appendix F, the key point is that there are two different options—one meant for evaluating model outputs against observations, and the other for performing source attributions. Airport operators plan- ning to perform air quality impact assessments can adopt a hybrid modeling approach, where a combination of CMAQ and AERMOD are used so as to take advantage of the respective model- ing system’s features to come up with a comprehensive set of outputs for all pollutants at the finest spatial and temporal scales that are desirable. For this study, a nested grid of AERMOD receptors was used. The outer coarse grid had recep- tors spaced uniformly every 1 km in a 15- by 15-km area, while the inner finer grid had receptors spaced uniformly every 500 m in a 5- by 5-km area—both centered on the airport. This nested approach gave the ability to obtain air quality impacts at a highly resolved scale in the immediate vicinity of the airport and a relatively coarser scale at downwind distances. The inputs for AERMOD are exactly the same as those used if AERMOD was used alone (or as part of EDMS), except for the network of receptors at a finer resolution. The outputs of the

28 Guidance for Quantifying the Contribution of airport emissions to Local air Quality hybrid model provide an estimate of pollutant concentrations at a very highly resolved spatial scale (compared to what CMAQ alone can provide at the 4–12-km resolution). Based on the analyses of the hybrid modeling outputs under this project, for PM2.5, temporal variability domi- nates. However, for NOx, spatial variability is more important than temporal variability. 3.5 Receptor Modeling 3.5.1 Overview Receptor modeling is the application of data analysis methods to elicit information on the sources of air pollutants. Receptor-oriented or receptor models are focused on the behavior of the ambient environment at the point of impact as opposed to the source-oriented dispersion models that focus on the transport, dilution, and transformations that occur beginning at the source and following the pollutants to the sampling or receptor site. The fundamental principle of receptor modeling is that mass conservation can be assumed and a mass balance analysis can be used to identify and apportion sources of airborne particulate matter in the atmosphere. This methodology has generally been referred to within the air pollution research community as receptor modeling (Hopke 1985). The approach to obtaining a data set for receptor modeling is to determine a large number of chemical constituents such as elemental concentrations in a number of samples. Alternatively, automated electron microscopy can be used to characterize the composition and shape of particles in a series of particle samples. In either case, a mass balance equation can be written to account for all m chemical species in the n samples as contributions from p independent sources. x g f eij p P ip pj ij= + = a Eq. 3-1o 1 where xij is the measured concentration of the jth species in the ith sample, fpj is the concentration of the jth species in material emitted by source p, gip is the contribution of the pth source to the ith sample, and eij is the portion of the measurement that cannot be fit by the model. When the source profiles are not known, factor analysis methods that are quite different from traditional Principal Components Analysis and related techniques have been developed over the past decade. An explicit least-squares approach, Positive Matrix Factorization (PMF) (Paatero 1997 and 1999), has been found to be very useful when applied to particulate matter (e.g., Zhao 2006, Lee 2006, and Kim 2007) and volatile organic compound (VOC) (Kim 2005) composi- tional data. In these factor analysis methods, the problem is expanded to the solution of the source profiles and contributions over a set of samples. An advantage of the PMF approach is that it can be expanded to the analysis of more com- plex data that require advanced data models such as the size-composition data produced by the University of California at Davis RDI sampler coupled with synchrotron x-ray fluorescence. Pere-Trepat et al. (2007) demonstrated such a model that takes into account that sources produce particles whose composition vary by size and require a model different from that presented in Equation 3-1. Thus, the methodology can provide a mechanism to fit conceptual models of the nature of the sources for the particular type of data available. 3.5.2 Data Requirements and Usage Factor analysis methods like PMF can be applied to a variety of data. The critical issue is hav- ing sufficient variability in the contributions of the various sources (Henry 2003). This variabil- ity can be obtained through short-duration samples like those obtained using the rotating drum

Modeling Guidance and Findings 29 impactor (RDI) or other semi-continuous sampling methods (Chow 2008). Variation can also be obtained through the spatial arrangement of the samplers relative to the source areas (e.g., runways) knowing the prevailing wind directions. The combination of spatial and temporal variation can provide a very effective sampling strategy that provides data with specific source contributions that will be zero or close to it. The data need to include species that have relative concentrations that are sufficiently different from one another so that there is limited colinearity among the source profiles. These species can include elements, ions, or organic compounds. A critical issue is that the species have to have sufficient stability in the atmosphere that the pattern of relative concentrations for a given source does not change substantially over the spatial and temporal scales of the measured region. The methodology can handle data with a substantial number of below-detection-limit values and a wide range of measurement uncertainties. However, there needs to be sufficiently high signal-to-noise ratio variables that will permit the analysis to succeed. A careful sampling and analysis design based on the understanding of this data analysis method can provide the data required to assess the major source types arising from multiple sources around an airport. PMF as provided by the USEPA only permits the solution of the simple mass balance equation shown in Equation 3-1. However, sampling systems like the RDI provide more complex data with samples segregated by time and particle size. It is then necessary to build conceptual models of the source emissions to ambient concentration processes. For example, for the RDI data, the data should be considered as a three-dimensional array of size by elements by time. The sources emit particles whose compositions vary with size. Thus, the source profile is a matrix of elements by size. The model can be expressed as shown in Figure 3-1. 3.5.3 Model Application and Results It is necessary to explore a range of solutions. The choice of the number of factors is generally difficult. A number of possible diagnostics are available, but the primary criterion for selecting solutions is the examination of the distribution of the scaled residuals. A residual is the difference between the measured and modeled values. e x g fij ij p P ip pj= − = a Eq. 3-2o 1 These residuals are then weighted by the estimated data point uncertainty. e s x g f s ij ij ij p P ip pj ij = − = a Eq. 3-3 o 1 Figure 3-1. Schematic view of the new three-way model proposed for the analysis of the DRUM data. It is possible to fit these more complex models such as is shown in Appendix G.

30 Guidance for Quantifying the Contribution of airport emissions to Local air Quality The distributions for each variable should be symmetric and generally the values should lie in the range of –3 to +3. Such results suggest that the data has been adequately fit by the model. The model produces estimates of the source profiles that have to be interpreted as to what sources that they represent. The interpretation of the results can be difficult if new or unexpected sources are derived. The assignment of a source name to each profile depends on an understand- ing of the physical and chemical processes that result in the release of particles into the air. Differ- ent mechanisms produce particles of different size and composition. Prior examples of measured source profiles are available from the USEPA in its SPECIATE database (USEPAc 2010) and the scientific literature reporting source profiles. There should be consistent results in the pattern of the time-series of source contributions. For example, the RDI analysis provided in Appendix G identified a profile with high Na and Cl in particle sizes greater than 1 µm that only was observed during the winter sampling campaign. This profile can be assigned to salt applied on icy surfaces that is then suspended by the move- ment of vehicles over the salt-laden surface. The PM-focused receptor modeling and assessments presented in Appendix G demonstrate the difficulties in identifying the exact sources for each pollutant, but also serve to illustrate the usefulness in categorizing airport versus non-airport sources. Also, the assessments showed the potential to help identify “gaps” in the source modeling work (e.g., emissions from brake and tire abrasion). Although these gaps were demonstrated, the assessments also showed the need for detailed data such as those from an XRF analysis in the context of relatively higher temporal resolutions and longer time spans. In general, the more spatially and temporally itemized sets of data are available, the higher the fidelity in predicting source speciation. Although receptor modeling can provide useful results, its use should generally be consid- ered supplementary. Considering the resources available at an airport, source-oriented modeling should be considered first along with any necessary measurements. If receptor modeling is to be conducted, airport operators will need to carefully consider whether the work can be conducted using ambient data that is already expected to be collected (i.e., as part of the overall air qual- ity project) or whether resources are available to develop a concerted monitoring-and-receptor modeling program. In addition, consideration will need to be given to the expertise necessary to conduct such assessments. In general, there are far fewer individuals at airports as well as consul- tants (including universities) who have the expertise necessary to both conduct and understand the results from receptor modeling assessments.