Comprehensive Automobile Research System (CARS) – a 1 Python-based Automobile Emissions Inventory Model

Abstract

methodologies to estimate the hot exhaust, cold start, and evaporative emissions from onroad 23 automobile sources. It can optionally utilize road link-specific average speed distribution (ASD) 24 inputs to reflect more realistic vehicle speed variations by road type than a road-specific single 25 averaged speed approach. Also, utilizing high-resolution road GIS data allows the CARS to 26 estimate the road link-level emissions to improve the inventory's spatial resolution. pollutants, both developed and developing countries' governments have been implementing 51 stringent air pollution abatement control policies to reduce harmful regional air pollutants. 52 Chemical transport models (CTM) are a powerful tool to study and develop an efficient control 53 strategy for local and regional air quality ( strongly rely on precise input data, such as emission inventory, meteorology, land surface 56 parameters, and chemical mechanisms in the atmosphere. The most dominant factor for accurate 57 CTM performance is temporally and spatially high-quality emissions, especially in the worst air 58 quality regions with significant anthropogenic emission sources. 59 The major anthropogenic emission sources in urban areas are from transportation emission 60 sectors. The tailpipe emissions from the vehicle's combustion process contain many air pollutants, 61 including nitrogen oxides (NOx), volatile organic compounds (VOCs), carbon monoxide (CO), 62 ammonia (NH3), sulfur dioxide (SO2), and primary particulate matter (PM) which will participate 63 in the formation of detrimental secondary pollutants like ozone and PM2.5 in the atmosphere. In 64 the Seoul Metropolitan Area (SMA) in South Korea, transportation automobile sources contribute 65 the most to the total NOX and primary PM2.5 emissions across all emission sources. While more 66 than 60% of total ambient PM2.5 are primary PM2.5 directly emitted from the sources, ( processes, such as physical, chemical, and vehicle operation processes from on/off-network roads 74 (Moussiopoulos et al., 2009;Russell and Dennis, 2000). 75 There are two methodologies known in emission inventory development: top-down and 76 bottom-up. The choice of methods is determined by the input data availability. The top-down 77 approach primarily relies on the aggregated and generalized country or regional information, 78 especially in developing countries where only limited datasets and information are available. It has 79 its limitations on representing the vehicle emission process realistically due to the lack of detailed 80 activity and ancillary supporting data. However, the bottom-up approach requires higher-quality 81 spatiotemporal activity datasets like road network information, vehicle composition (vehicle type, The MOVES model has the strength to generate high-quality emissions for up to 16 102 different emission processes (i.e., Running Exhaust, Start Exhaust, Evaporative, Refueling, 103 Extended Idling, Brake, Tire, etc.). It can simulate not only county-level but also road segment 104 level depending on data availability. It can also reflect local meteorological conditions, such as 105 ambient temperature and relative humidity, which can significantly impact both pollutants and 106 emissions processes (Choi et  for CTMs without any 3 rd party emissions modeling system to develop the highest quality CTM-153 ready emissions inputs. All functions are operated by independent modules and can be enabled by 154 users. Details on modularization will be discussed later. The CARS model can be easily adopted 155 and is simple for users to add new functions or modules in the future. The application of the CARS 156 to South Korea will be described in detail later. 157

158
The CARS is an open-source Python-based customizable motor vehicle emissions 159 processor that estimates onroad and offroad emissions for specific criteria and toxic air pollutants. 160 Figure 1

Emission Calculations 201
Automobile emission sources cover motorized engine sources from network (onroad) and 202 off-network (nonroad). Nonroad transportation sources represent any motorized engine vehicle 203 emissions that occurred from off-network roads, such as aviation, railways, construction, and boats. 204 Onroad automobile emissions are ones that occur on the network roads. While nonroad automobile 205 emissions are important, we will focus on the onroad automobile emissions from network roads 206 using their local traffic-related datasets. The following section explains the approach of the onroad 207 automobile emission processes. The onroad emission (Eonroad) in the CARS is defined in Eq.
(2), 208 which includes three major emission processes (Ntziachristos and Samaras, 2000): 209  vehicle speed (Ntziachristos and Samaras, 2000). Figure 3 shows the dependency of NOx emission 253 factors from compact diesel vehicles to vehicle speed (Fig. 3a) and ambient temperature (Fig. 3b). 254 Figure 3a shows a significant decrease of NOx emissions while speed increases. Figure 3b  255 demonstrates the significance of local meteorology on NOx emissions from a compact diesel sedan. 256 Based on these NIER's CAPSS emission factors, the sensitivity to local ambient temperature is 257 limited to NOx pollutant emissions from diesel vehicles. 258 Due to its high sensitivity to the vehicle operating speed, it is important for the CARS to 259 simulate realistic speed patterns for accurate emissions estimates. When a constant single speed is 260 assigned to compute hot exhaust emissions, it won't reflect the emissions under low-speed 261 circumstances, which could cause higher emissions due to its incomplete ICE combustion. To 262 overcome this limitation, the CARS has adopted the 16 average speed bins concepts for a better 263 representation of vehicle speed distribution that varies by road type (i.e., local, highway, 264 expressway). We have implemented a feature for the CARS optionally to apply road-specific 265 average speed distributions (ASD) (Abin,r), which represents the fractions of 16-speed bins (bin) 266 (from 0 to 121 km h -1 defined in Appendix E) for eight different road types (r) (No.101-108, shown 267 in Appendix C) as classified by CAPSS (Fig. 4). Although ASD patterns vary by region, we did 268 not implement the regional variations of ASD due to the lack of region-specific vehicle speed 269 measurements in South Korea. 270 In this study, we developed the most realistic ASDs for eight different road types (No. 101-271 108) in South Korea based on the latest road link-specific average speed and AADT from the GIS 272 road network shapefiles (NIER, 2018) and the U.S. EPA's MOVES ASD datasets (USEPA, 2020). 273 Because a single average speed was assigned to each road link, the ASDs based on South Korea's 274 GSI road shapefiles did not capture the low-speed range (<16 km h -1 ) that occurs in reality. 275 Therefore, we incorporated the ASD developed by U.S. EPA with Georgia state ASD to improve 276 the representation of the low-speed range (speed bin #1 and #2). We modified the total fractions 277 of low-speed bins (the 2:1 ratio of fractions of bin #1 and #2) by adding 2% of distribution for 278 interstate expressways, 3% of distribution for urban expressways, 7% of distribution for all 279 highways, and 15% for all local roads. Further, those increases of low-speed bins reduced the 280 distributions of other higher speed bins homogeneously due to the renormalization of fractions by 281 road type. Figure 4 shows the renormalized ASDs of all road types applied in this study. 282 While 16-speed bins ASD application is critical to computing more realistic hot exhaust 283 emissions, there should be some restrictions on certain road types. Users can adjust the restricted 284 roads control

Cold Start Emissions 298
The cold start emissions occur when a cold-engine vehicle is ignited. The lower 299 temperature of the ICE is not an optimal condition for complete fuel combustion. This process 300 lowers the combustion efficiency (CE) and increases the emissions of hydrocarbon and CO 301 pollutants from the tailpipe exhaust (Jang et al., 2007). The CARS can estimate the cold start 302 emissions for vehicles using gasoline, diesel, or liquefied petroleum gas (LPG) fuel. Besides the 303 vehicle and engine type, road type also plays a critical role in the quantity of cold start emissions 304 because it occurs mostly in parking lots and rarely on highways. 305 The cold start emission, Ecold (g d -1 ), is derived from the hot exhaust emissions, the ratio of 306 hot to cold exhaust emissions (EFcold/EFhot -1.0), and the percentage of the traveled distance with 307 a cold engine (Eq. 6). 308 running loss emissions (R) (Eq. 9). Unlike CAPSS, there is a conversion factor (0.075) applied to 332 Evap for motorcycles to prevent an over-estimation of VOC. 333 ; , = ( ; , + ; , + ; , ) The empirical coefficient α is 0.1 here, which represents that 90% of the running loss is 356 avoided by the newer fuel system. L is the distance traveled (km) by road and is the same one used 357 in hot exhaust emission calculations.  is the same parameter from Eq. (8). The Rh and Rw are the 358 average emission factors from running loss under hot and warm/cold conditions, respectively. 359

Road Link-Level Emissions Calculations 360
In general, district-level automobile emissions calculations are driven by district-level 361 averaged vehicle activity and operating data, which do not reflect realistic spatial patterns of 362 onroad automobile emissions. The CARS model introduces road link-specific traffic data by 363 default to develop spatially enhanced road link-specific emissions that reflect more representative 364 emissions by road link. This high-resolution traffic data is a GIS shapefile that is composed of 365 many connected segments, which are called "road links." All road links hold information such as 366 start/end location coordinates, AADT, road link length, averaged vehicle speed, and road type (No. 367 101-108). 368 The CARS model applies link-level AADT (AADTd,r,l., d -1 ) and road length (Ld,r,l) to 369 compute the road link-specific VKT (VKTd,r,l, km d -1 ) in Eq. (12). The road links are identified by 370 district (d), road type (r), and link (l) labels. The road VKT is a parameter that reflects the traffic 371 activity of each road link and it is different from individual daily vehicle activity data (VKTv,age) 372 in Eq. (1). 373   CTMs. In addition, the CARS also generates various summary reports, graphics, and 399 georeferenced plots for quality assurance. 400 The required Python modules for the CARS are: "geopandas," "shapely.geometry", and 401 "csv" modules to read the shapefiles and table data files. The "NumPy" and "pandas" modules 402 are used to operate the memory arrays and scientific calculations while the "pyproj" module deals 403 with converting the projection coordinate systems. "matplotlib" is for generating any type of 404 figures/plots. Furthermore, the CARS model can also read and write Climate and Forecast (CF)-405 compliant NetCDF-formatted files using "NetCDF4". 406 The first process in the CARS is "Loading_function_path"; it allows users to define and 407 check the input file paths. Once all input files are checked, there are six process modules in CARS 408 to process inputs, compute emissions, and generate various output files, including QA reports. 409 Figure 5 is the schematic of the CARS that consists of six process modules with various functions. 410 The six process modules are (1) "Process activity data", (2) "Process emission factors", (3) 411 "Process shapefile, (4) "Calculate district emissions", (5) "Grid4AQM", and (6) "Plot figures". 412 The main purpose of modularizing the CARS is to meet the needs of various communities, such 413 as policymakers, stakeholders, and air quality modelers. these modules and can simply read the data frame outputs and then run "Grid4AQM" for the 420 modeling dates and domain. The "Grid4AQM" module will not only improve the computational 421 time for CTMs but also eliminate the need for a 3 rd party emissions modeling system like SMOKE 422 (Baek and Seppanen, 2021). 423 The rectangle boxes in Fig. 5  NOx pollutants. District boundary GIS shapefiles and road network shapefiles are processed 431 through "Process shape file" to generate the VKT-based redistribution weighting factors from Eq. 432 (13), (14) and (15) for the "Calculate district emissions" module to compute district-level and 433 road link-level emission rates (metric tons per year, t yr -1 ). 434 The redistributed emission rates (t yr -1 ) from the "Calculate district emissions" module 435 present annual total emission rates until district-level VKTs from the "Process activity data" 436 module are added. Then, the "Grid4AQM" module can generate CTM-ready chemically speciated 437 emissions. The "Read_chemical" function from the "Grid4AQM" module is designed to process 438 the chemical speciation profile that can convert the inventory pollutants such as CO, NOX, SO2, 439 PM10, PM2.5, VOC, and NH3, into the chemically lumped model species that CTM requires for 440 chemical mechanisms, such as SAPRC (L. and Heo, 2012) and Carbon Bond version 6 (CB6) 441 (Yarwood and Jung, 2010). The "Read_temporal" function processes the complete set of monthly, 442 weekly, and hourly temporal allocation profiles that can convert annual total emissions to hourly 443 emissions. "Read_griddesc" defines the CTM-ready modeling domain and computes the gridding 444 fractions for all road link-level emissions by overlaying the modeling domain over the GIS 445 shapefiles. Once annual total emissions are chemically speciated, spatially gridded, and temporally 446 allocated into hourly emissions, the "Gridded_emis" function will combine emission source-level 447 conversion fractions from each function (Read_chemical, Read_temporal, and Read_griddesc) to 448 generate the CTM-ready chemically speciated, gridded hourly emissions in the NetCDF binary 449 format. The "Plot Figures" module is designed for generating various summary reports and 450 graphics to assist users in understanding the estimated automobile emissions inventory computed 451 by the CARS. The following section will describe the detailed processes of the "Grid4AQM" 452 module, which includes chemical, spatial, and temporal allocations. 453

Chemical Speciation 454
To support CTMs applications, the CARS needs to be able to convert inventory pollutants 455 into chemical lumped model species based on the choice of CTM chemical mechanisms. NOx 456 includes nitric oxide (NO), nitrogen dioxide (NO2), and nitrous acid (HONO). VOCs can represent 457 hundreds of different organic carbon species, such as benzene, acetaldehyde, and formaldehyde. 458 These grouped inventory pollutants cannot be directly imported into the chemical mechanism 459 modules in the CTM system and require chemical speciation allocation for CTMs to process them 460 during their chemical reactions. Therefore, the "Grid4AQM" module performs the chemical 461 species allocation step prior to the temporal and spatial allocations to generate the gridded hourly 462 emissions. The "Read_chemical" function in "Grid4AQM" module allows users to assign these 463 emission inventory pollutants to CTM-ready surrogate chemical species (a.k.a lumped chemical 464 species) by vehicle, engine, and fuel type. For example, VOC emissions from diesel busses can be 465 converted into the following composition based on its chemical allocation profile: alkanes (68%), 466 toluene (9%), xylenes (8%), alkenes (4%), ethylene (2%), benzene (1.3%), and unreactive 467 compounds (7%) when CB6 chemical mechanism is selected. Further details on the chemical 468 speciation profile input formats are available in the CARS user's guide. 469

Spatial Allocation 470
The "Calculate district emissions" module calculates not only the total district emissions 471 but also road link-specific emissions based on road link-specific AADT data from road network 472 GIS shapefiles. The "Calculate district emissions" module first gets the district total vehicle 473 emissions (Eq. 2) based on the district-level VKTs, and then the normalized district total emissions 474 by district weight factor, ωd (Eq. 13). Afterwards, the normalized district total emissions are 475 redistributed into every road link using road link-level weight factors (ωd,l) (Eq. 15). The district 476 total emissions from Eq.
(2) and from Eq. (15) remain the same. Then the computed road link-477 level emissions then will be converted into grid cell emissions using the modeling domain grid cell 478 fractions computed in the "Read_griddesc" function in the "Grid4AQM" module. 479

Temporal Allocation 480
Once chemical and spatial allocations are completed, the final step to support CTM 481 application is a temporal allocation that converts the annual total emissions from the "Calculate 482 district emissions" module into hourly emissions. The "Read_temporal" temporal allocation 483 function in the "Grid4AQM" module converts the annual emission rate (t yr -1 ) to the hourly 484 emission rate (mol hr -1 ) using monthly, weekly, and weekday/weekend diurnal temporal profiles. 485 This module processes these temporal profile inputs, which are the monthly (January -December), 486 weekly (Monday -Sunday), and weekday/weekend 24 hour profile tables (0:00-23:00 LST). The 487 users can assign these temporal profiles with a combination of vehicle, engine, fuel, and road types 488 to enhance their temporal representations in detail. 489

Chemical Transport Model Emissions 490
The main goal of the "Grid4AQM" module is to generate temporally, chemically, and 491 spatially enhanced CTM-ready gridded hourly emissions. First, it reads the CTM modeling domain 492 configuration and then overlays it over the road network GIS shapefile and district-boundary 493 shapefile to define the modeling domain. This overlaying process between the road network, 494 district boundary GIS shapefiles, and modeling domain allows the "Grid4AQM" module to 495 compute the fraction of road links that intersects with each grid cell. Figure 6 demonstrates how 496 the district boundary and road network GIS shapefiles are used to perform the spatial allocation 497 processes in CARS. Figure 6a is a native road link shapefile of Seoul with AADT, VKT, district 498 ID, and road type. Figure 6b presents an overlay of two districts's road links (purple and blue) 499 over the selected region. State total emissions will be renormalized into weighed district total 500 emission data and then redistributed into the road link. Figure 6c illustrates how the weighted road 501 link-level emissions get allocated into modeling grid cells for CTMs. The link-level VKT (VKTd,r,l) 502 from Eq. (12) will be used to compute a total of traffic activity fractions by grid cell and then use 503 that to assign the link-level emissions from Eq. (2) into each grid cell. When a road link intersects 504 with multiple grid cells, the "Grid4AQM" module will weigh the emissions by the length of the 505 link that intersects with each grid cell. 506 Through the overlay process, the CARS model can generate various types of output data, 507 such as total district emissions, link-level emissions, and CTM-ready gridded emissions. For 508 example, the CO vehicle emissions from the Seoul metropolitan in South Korea are presented in 509 three different output formats in Fig. 7. Figure 7a shows the annual mobile PM2.5 emissions by 510 district. The road link level annual emissions are presented in Fig. 7b. Furthermore, the CARS 511 applies the link-level emissions from Fig. 7b to generate the hourly grid cell emission data with a 512 1 km × 1 km resolution for the CTM in Fig. 7c. 513

National Control Strategy Application 514
One of the unique features in the CARS compared to other mobile emissions models is that 515 it can promptly develop controlled mobile emissions responding to the national emergency high 516 PM2.5 episodes. It is very common to experience high PM2.5 episodes, especially during the 517 wintertime in South Korea due to domestic and international primary and secondary air pollutants 518 emissions. When the 72 hour forecasted PM2.5 concentration exceeds the average 50 µg/m 3 (0:00-519 16:00 LST), the national PM2.5 emergency control strategy is activated for ten days. It applies a 520 nationwide vehicle restriction policy within 24 hours. It enforces a limit on what kind of vehicles 521 can be operated on a certain date. The restrictions can be applied in the following ways: the 522 closures of public parks and government facilities, and restrictions of certain vehicles based on 523 their fuel type and age, which is a major factor of engine deterioration. This policy will limit the 524 number of vehicles on the network roads significantly, which could reduce primary PM2. To understand the impacts of an even/odd vehicle restriction policy in real-time, we need to 528 quickly develop a rapid control response emissions for the air quality forecast modeling system. 529 The process of generating the controlled mobile emissions can take a long time if we start fresh. 530 Thus, we have implemented this control strategy as an optional "Control Factors" function in the 531 "Calculate district emissions" in the module for users to quickly and easily generate the 532 controlled mobile emissions with consideration of the limited number of vehicles based on the 533 vehicle, engine, fuel, and vehicle manufactured year. A one hundred percent (100%) control factor 534 means that there are no emissions from those selected vehicles. 535 Because of the modularization system in the CARS, we can bypass some computationally 536 expensive data processing modules (i.e., "Process activity data", "Process emission factors", 537 and "Process shape file") and let the "Calculate district emissions" module quickly apply control 538 factors while it computes the district-level mobile emission inventory from Eq. (2). This will allow 539 users to reduce the computational time to generate the controlled mobile emissions under a specific 540 control scenario and develop the controlled CTM-ready gridded hourly emissions using the 541 "Grid4AQM" module. 542

Computational Time 543
While the CARS can generate a high-quality spatiotemporal emission inventory for 544 policymakers, stakeholders, and air quality modelers, it is quite critical for the CARS to generate 545 these complex mobile emissions effectively and accurately without being at the expense of 546 computational time. This is especially important to meet the needs for an air quality forecast 547 modeling system responding to a national emergency control strategy implementation. 548 In this section, we will discuss the details of the CARS computational modeling performance. 549 While the CARS model has been highly optimized, the modularization of CARS has also improved 550 its modeling performance with optional module runs. The breakdown of module-specific 551 computational time estimates based on the benchmark CARS runs are listed in The difference between CAPSS and CARS approaches are caused by three reasons: First, 580 the number of vehicles used in CARS is slightly higher (6%) than CAPSS data (1.3 out of 23 581 million), as well as other key traffic-related activity inputs (i.e., vehicle age distribution, averaged 582 speed distribution, etc). Secondly, the vehicle speed information assigned by vehicle and road type 583 play a critical role in the differences between CAPSS and CARS. The CAPSS calculation was 584 based on the road-specific mean speed value or 80% of the speed limit as an input of vehicle 585 operating speed by three road types (rural, urban, and expressway). In other words, CAPSS only 586 assigns a "single-speed value" for each road type, and does not encounter the variation of vehicle 587 speed during its operation on roads into the emissions calculation. Most running exhaust emissions 588 occur during a vehicle's low-speed operation due to its incomplete combustion of fuel, and it is 589 critical to accurately represent the emissions across various speed bins in order to compute the 590 correct emissions. The CARS model has an option to apply the average speed distribution (ASD) 591 over 16 speed bins for eight road types (Fig. 4). The CARS speed distribution process can better 592 represent the speed variations of vehicle speeds for each road type. A detailed analysis of the 593 impact of vehicle speed will be discussed later in this chapter. Lastly, other advanced processes in 594 the CARS, such as link-level AADT and district-level vehicle data (5,150 districts in South Korea), 595 can reflect more spatial detail and variation than the CAPSS. The CAPSS only considers state-596 level data (17 states in South Korea) and five road types (interstate expressway, urban highway, 597 rural highway, urban local, and rural local). 598

Onroad Emissions Analysis 608
The CARS is a bottom-up emissions model, which utilizes local individual vehicle activity 609 data, detailed local emission factors for every vehicle and fuel type, and localized inputs such as 610 average speed distribution by road type and deterioration factor. It allows users to assess the 611 detailed breakdown of localized emission contributions. The IF of the CNG bus is 320 kg yr -1 and emits 19.5% of the total VOC. Comparing the IFs of 635 buses across fuel types, the CNG bus emits less NOx but higher VOC than a diesel vehicle. Each 636 CNG bus has about 33 times higher IF of VOC (320 kg yr -1 ) than a diesel bus (9.51 kg yr -1 ), and 637 CNG buses released slightly lower NOx (248 kg yr -1 ) than diesel buses (340 kg yr -1 ) (Table 3a and  638 3b). It indicates that a CNG bus is better for rural areas and a diesel bus is better for urban areas to 639 control ozone, because the rural area is usually NOx limited for ozone formation and urban areas 640 are VOC limited. 641 The current South Korea CAPSS onroad emissions inventory does not consider the PM2.5 642 emissions from tire and brake wear, which are the highest contributors of PM2.5 emissions from 643 vehicles on roads. For that reason, diesel vehicles become the major source of PM2.5 emissions, 644 which contributes over 98.5% (9,959 t yr -1 ) of the PM2.5 emissions based on the CARS 2015 645 emissions (Table 3c). The diesel truck, SUV, and van are the three major sources, and their 646 contributions of total PM2.5 are 53.6%, 21.4%, and 11.2%, respectively. Although over 52% of the 647 vehicles are gasoline vehicles, their primary PM2.5 contribution is limited to 1.44%. The diesel 648 bus has the highest IF (2.83 kg yr -1 ), which is caused by the largest average daily VKTs. 649 Similar to VOC emissions, CO is mostly emitted through the tailpipe due to incomplete 650 internal combustion of fuel and share similar emissions distributions across vehicle and fuel types 651 (Table 3d). Gasoline vehicles contribute most of the CO (220,390 t yr -1 , 59.0%), and sedan vehicles 652 are the primary source (178,121 t yr -1 , 47.6%) of this out of all gasoline vehicles. Across vehicle 653 types, bus shows the highest IF of CO (81.2 kg yr -1 ) due to its largest daily VKT. CO is the most 654 abundant pollutant released from vehicles (373,864 t yr -1 ) across all pollutants from onroad 655 automobile sources. Although CO is much less reactive than other vehicle VOCs (Rinke and  656 Zetzsch, 1984; Liu and Sander, 2015), the majority of CO emissions from onroad automobile 657 sources plays a critical role in generating 30% of hydroperoxyl radicals (HO2) and causing ozone 658 formation in urban areas (Pfister et al., 2019). Thus, CO is also another crucial precursor to ozone 659 formation in urban areas. 660 SOx emissions are related to the sulfur content within the fuel component; diesel has a 661 higher sulfur content than any other fuels. Most SOx is contributed by diesel vehicles (93.8 t yr -1 , 662 54.5%) (  The CARS can also optionally apply the average speed distribution (ASD) by road type to 680 compute more realistic mobile emissions on the road network when compared to using a current 681 single average speed value for each road type (Appendix E). Applying the ASD will generate a 682 much better representation of actual traffic patterns from each road type. To understand the impacts 683 of ASD application, we performed sensitivity runs between using a single-speed to the ASD 684 application (Appendix F). The ASD data was described in Fig. 4, and the road-specific average 685 single-speed values were developed based on the weighted average method using the same ASD 686 data. Appendix E and S6 describes the details of ASD as well as road-specific speed values. 687 Figure 9a shows the differences in total emissions between two scenarios and is organized 688 by pollutant. The single-speed scenario largely underestimates the emissions across all pollutants 689 compared to the ones from the ASD scenario. NOx (16%), VOC (40%), and CO (30%) were 690 especially underestimated. The difference is caused by the lack of low-speed bins (<16 km h -1 ) 691 representation when a single average speed approach was used. Higher emissions are emitted while 692 vehicles are operated with low-speed bins, which decreases the combustion efficiency of ICE and 693 releases more pollutants. 694 Figure 9b shows the road-specific breakdown between the ASD and single speed scenarios 695 to understand the impacts of vehicle operating speeds on onroad automobile emissions. In this 696 figure, each color indicates the emissions percentage differences by road types. Other than NH3, 697 significant discrepancies happened between local urban roads (5.8%), highways (3.9%), and urban 698 highways (3.0%). Other pollutants, VOC, PM2.5, CO, and SOx, have similar fractions of road types. 699 This phenomenon is caused by low-speed conditions (<16 km h -1 ) and the fractions of road VKT 700 contributions (Appendix C). The lower speeds cause the incomplete combustion of ICE and 701 increase the emission rate. Also, local urban roads, highways, and urban highways have higher 702 road VKT contributions at 17%, 18%, and 12%, respectively (Appendix C) than rural roads. 703 Higher emissions from low speed conditions from these high contribution roads (urban local, urban 704 highway, and highway) caused these significant differences between the ASD and single-speed 705 approaches. Although the interstate expressway has the largest VKT contribution (41%), it also 706 has the lowest fraction of low-speed bins (2%). That is why the difference between the ASD and 707 single speed scenarios on interstate expressways is less than 1%. In general, NH3 emission factors 708 do not change by vehicle operating speed, so the ASD impact is quite minimal. 709       Taxi   --Compact  -------Fullsize  -------Midsize  -----Special   -Tow  ------Wrecking  Wrecking  Wrecking  Wrecking  ----Others  Others  Others  -----Motorcycle   Compact  -------Midsize  -------Fullsize  ------