Implications of Driving Style and Road Grade for Accurate Vehicle Activity Data and Emissions Estimates

Transcription

1 Sentoff, Aultman-Hall & Holmén Page 1 of Implications of Driving Style and Road Grade for Accurate Vehicle Activity Data and Emissions Estimates Paper No.: Originally Submitted: August 1, 2014 Karen M. Sentoff, corresponding author University of Vermont Transportation Research Center Farrell Hall 210 Colchester Avenue Burlington, VT ksentoff@uvm.edu (802) Lisa Aultman-Hall University of Vermont Transportation Research Center Farrell Hall 210 Colchester Avenue Burlington, VT laultman@uvm.edu (802) Britt A. Holmén University of Vermont School of Engineering Votey Hall 33 Colchester Avenue Burlington, VT bholmen@uvm.edu Word Count: 5013 words plus 4/5 figures/tables = = 7263 Abstract: Real-world vehicle operating mode data (2.5 million 1Hz records), collected by instrumenting the vehicles of 82 volunteer drivers with OBD datalogger and GPS while they drove their routine travel routes, were analyzed to quantify vehicle emissions estimate error due to road grade and driving style in rural, hilly Vermont. EPA MOVES2010b was used to estimate running exhaust emissions associated with measured vehicle activity. Changes in vehicle specific power (VSP) and MOVES operating mode (OpMode) due to proper accounting for real-world road grade indicated emission rate errors between 10-48%, depending on pollutant, chiefly because grade-related changes in VSP could shift activity by as many as six OpModes, depending on road type. The correct MOVES OpMode assignment was made only 33-55% of the time when road grade was not included in the VSP calculation. Driving style of individual drivers was difficult to assess due to unknown traffic operations data, but the largest differences between individual drivers was observed on rural restricted roads. The study results suggest the importance of (a) measuring and incorporating real-world road grade in order to correctly assign MOVES emission rates; and (b) developing a driving style typology to account for error in the MOVES emissions estimates due to driver variability.

2 Sentoff, Aultman-Hall & Holmén Page 2 of INTRODUCTION Motor vehicle emissions are a primary concern because of their potential detriment to local air quality and global atmospheric condition. In an effort to improve the fidelity of emissions estimates from mobile sources, the Environmental Protection Agency developed the Motor Vehicle Emission Simulator (MOVES). MOVES is capable of emissions estimates at a range of spatial (national, county, and projectlevel) and temporal scales and is a required tool for state implementation plans, transportation project conformity, and air quality compliance. At the project-level scale, MOVES inputs consist of speed-time trajectories including vehicle specific power (VSP), a surrogate for the resistances a vehicle must overcome for propulsion. Combined with speed, the VSP activity is classified, or binned, into 23 running operating modes for which corresponding emissions rates by vehicle type, vehicle age, fuel type, emission control technology, and atmospheric conditions are provided. In the real-world, these speed-time trajectories would be expected to vary based on the individual s driving style. In addition, the vehicle trajectory approach is generally applied without regard for an important term in the calculation of vehicle specific power: road grade. Little research has been conducted to date to quantify the uncertainty introduced to vehicle activity by the real-world factors of driving style and road grade and the associated implications for emissions estimates. This paper aims to broaden understanding and answer the following questions based on on-road data collected from 82 drivers during routine travel in real world conditions: (a) how important is road grade in estimating accurate, real-world vehicle activity and emissions, especially for areas with hilly terrain?, and (b) what is the uncertainty in vehicle activity and emissions estimates due to individual driving style? LITERATURE REVIEW There have been attempts to link an individual s driving style to changes in tailpipe emissions or fuel consumption. Comparison of multiple drivers along a replicate route is one method used to better understand the variability in driving style. Previous work studying 24 drivers in a single vehicle indicated that frequency of activity in seven, coarse operating modes was similar across the driver participants, but the intensity of emissions events in these operating modes varied between drivers (1). Authors suggested that, in this case, the route was more indicative of the mode distribution than the individual driving style (1). A later study of 20 different drivers on a single route in one test vehicle (model year 1999) resulted in a wide range of tailpipe emissions across drivers, identifying the importance of interaction between driver and road type (2). In essence, a driver with wide variability of a particular pollutant on one road type may have narrow variability on another road type (2). Another methodology utilized for driving style has been to disaggregate driving style at the trip level. For instance, one approach looked at the driving styles of three different drivers controlling for route, vehicle, travel direction, and time of day (3). Driving style across the different drivers on average introduced variability for different pollutants, with 5% for CO 2 and up to 108% for CO (3). In a similar study, an aggressivity factor was defined based on speed and acceleration, with the goal of capturing the range of driving styles and the upper and lower bounds of tailpipe emissions (4). In this case, aggressive driving produced increased carbon monoxide and hydrocarbon emissions by almost one order of magnitude, as well as slightly increased fuel consumption, carbon dioxide, and oxides of nitrogen (4). Standard on-board diagnostics (OBD) data loggers and relatively low-cost GPS units have made real-world, onboard data collection of vehicle activity increasingly feasible, but little attention has been paid to real-world road grade. Some methods used to link high temporal resolution (1-hz) vehicle activity data with the real-world road grade include GPS measurements, gyroscope equipped vehicles, and digital elevation models (DEM) derived from orthophotos or light detection and ranging (LIDAR). Although the spatial data (latitude and longitude) from low-cost GPS units is robust, elevation data can be inaccurate, making it difficult to estimate road grade (5). The Automatic Roadway Analyzer (ARAN) van, like that used by the Vermont Agency of Transportation to collect roadway geometries, collects accurate (+/-

3 Sentoff, Aultman-Hall & Holmén Page 3 of %) road grade data using a series of gyroscopes and other onboard equipment. These data may only be available for a selection of roads and is relatively expensive to collect. However, DEM data are becoming more widely available at higher spatial resolution due to the increasing use of LIDAR for many different purposes. It has been shown that a 0% to 6% increase in road grade corresponds to a 20 kw/ton increase in VSP over operating speeds of 0 to 120 kph (5). This sensitivity of VSP to road grade has prompted the development of empirical relationships of tailpipe emissions and fuel consumption to real-world road grade (3, 5 7). Model estimates of fuel consumption rate decreased about 19% when positive road grades were neglected and increased about 23% when negative road grades were neglected (3). When compared to a flat route, a downhill route (road grade < 0%) corresponded to nearly two times improved mile-pergallon fuel economy and an uphill route (road grade > 0%) fuel economy was reduced by 1.5 to 2 times (6). This is consistent with fuel consumption increases of 40 to 100% as road grade increased 0% to 5% (3). According to the literature, steeper road grades and more aggressive driving style are generally associated with increased tailpipe emissions. Although some work has been conducted to develop empirical relationships between emissions and road grade or individual driving style, the uncertainty or error these factors introduce across hilly terrains and for a full range of drivers are not yet well defined. This research aims to quantify the uncertainty in vehicle activity and resulting emissions estimates due to real-world road grade and individual driving style. METHODOLOGY Data Collection Volunteer drivers were recruited for this study via postings at gas stations throughout Vermont and advertisements on a community-based online forum, Front Porch Forum. Drivers were excluded from the study if their primary vehicle was (a) a hybrid vehicle, as conventional methods of vehicle idling were critical to other study objectives, or (b) a pre-model year 1996 vehicle, as compatibility with onboard diagnostics (OBD) was essential to retrieving data directly from each vehicle s computer during operation. Two phases of data collection were conducted through 2011 to The first phase included a two-week data collection for 26 drivers in each of summer 2011 and winter The second phase included a 10-day data collection across 86 volunteer drivers. Volunteers were asked to drive their vehicles as they normally would and were informed only that the study was evaluating general travel behavior (i.e. origins and destinations, number of trips, etc.). Each volunteer s own vehicle was instrumented with an EASE Diagnostics MiniDL Onboard Diagnostics (OBD) logger to collect operation data (vehicle speed, engine speed, etc.) directly from the vehicle s engine control unit (ECU) and a GeoStats GeoLogger Global Positioning System (GPS) to collect spatial location. In addition to the data collected directly from each volunteer s vehicle, a questionnaire was used to collect information on vehicle age, make, model, and type; individual driver age, gender, education, and employment status; and household size, number of vehicles, and income. Although volunteers traveled as they normally would during the study period, the study area for this analysis was limited to travel undertaken in the state of Vermont. Of the valid trip data, less than 7% of the data (188,919 records) were classified as travel outside of the state. The road network in Vermont is comprised of just over 14,000 miles of road, 11.5% of which is urban and 2.2% of which is interstate highway (8). Vermont is also home to the Green Mountains and known for its relatively steep terrain ( to 12.2 % road grade observed for this particular study). Data Tabulation Methods for post-processing of the data and alignment were detailed elsewhere (9). Data were considered usable if the OBD and GPS data streams could be temporally aligned; valid trips were determined by high

4 Sentoff, Aultman-Hall & Holmén Page 4 of correlation coefficients between aligned speed from the GPS and OBD. Some equipment failures occurred such that the valid data set contained data for 14 and 68 volunteers for each data collection period, respectively, or 82 volunteers total. The Vermont Agency of Transportation s road layer was the source for assignment of the MOVES road types (Table 1) to each one-second record. Proposed roadways were removed to avoid assignment of second-by-second driver data to non-existing infrastructure. The 97.5 percentile of distance between each data record and the closest road was used as the threshold for activity on the transportation network. Valid data that fell outside of this 63 meter from centerline buffer was not used for analysis. The functional class attribute was used to determine the road type for the matched 2,565,690 records (Table 1). Functional classes 1 through 9 are rural and 11 through 19 are urban and their classifications match with the MOVES road types (Table 1). TABLE 1 MOVES Road Type Assignment based on Functional Class Definitions MOVES Road Type Off-Network (1) Rural Restricted (2) Rural Unrestricted (3) Urban Restricted (4) Urban Unrestricted (5) Functional Class (Fix) Definition 0 Not part of Functional Classification System 1 Rural, Principal Arterial - Interstate 22 Selection of FUNCL 2 with no at grade intersections recoded 2 Rural, Principal Arterial 4 Rural, Principal Arterial - Other (not other freeway) 6 Rural, Minor Arterial 7 Rural, Major Collector 8 Rural, Minor Collector 9 Rural, Local 11 Urban, Principal Arterial-Interstate 32 Selection of FUNCL 12 with no at grade intersections recoded 12 Urban, Principal Arterial-Other Freeway 14 Urban, Principal Arterial-Other 16 Urban, Minor Arterial 17 Urban Collector 19 Urban Local A digital elevation model (DEM), published by the U.S. Geological Survey for the state of Vermont at a 10 meter resolution (3 meter resolution where available), was used to assign the elevation (in feet) for every record in the data set. Road grade was calculated from consecutive seconds during an individual driver s trip, where the previous (t-1) and current (t) record speed determined the distance traveled and the elevation of the previous (t-1) and current (t) record determined the change in elevation. These values were filtered for reasonable distance traveled in a second and reasonable elevation change in a second. The maximum allowable distance traveled was 90 meters over two seconds (~ 100 mph) and maximum allowable change in elevation was 30 meters over two seconds (~ 30% grade). The fractional grade was then computed based on trigonometric principles. Even with the maximum allowable limits, fractional road grade outside of the 99 th percentile was considered erroneous and removed from further analysis. Calculation of VSP and MOVES OpMode The power required from a vehicle s drivetrain to achieve a desired speed and acceleration while overcoming the forces resisting forward motion is known as vehicle specific power (VSP). The equation for VSP (Equation 1) was adapted from the EPA (10) and accounts for the kinetic energy (speed and

5 Sentoff, Aultman-Hall & Holmén Page 5 of acceleration), potential energy (road grade and speed), rolling resistance, rotational resistance, and aerodynamic drag acting on the vehicle. VSP t 2 t A vt B v 3 C vt at m g sin t vt m (1) where: A = coefficient of rolling resistance (kw-sec/m) B = coefficient of rotational resistance (kw-sec 2 /m 2 ) C = coefficient of aerodynamic drag (kw-sec 3 /m 3 ) v t = speed at time t (m/sec) a t = acceleration at time t (m/sec 2 ) m = mass (metric tons) sinθ t = road grade (fractional) Vehicle speed was acquired from the OBD data and acceleration was calculated based on the previous (t-1) and current (t) vehicle speed, as reported for each record. VSP was binned for analysis by rounding each record to the closest 1 kw/ton. Information collected directly from each volunteer driver was used to determine the vehicle type as defined by MOVES and the associated coefficients to be used in calculation of VSP. The equation coefficients used are average fleet values for passenger cars and trucks according to the MOVES default database and assumed to be sufficient for this analysis. These were assumed to be passenger car values if no vehicle type was reported, as was the case for 6 volunteers in the valid data set. As previously described, MOVES uses vehicle activity bins or categories defined by speed and instantaneous VSP to define 23 MOVES running exhaust operating modes (OpModes). The operating modes are defined by (3 or 6 kw/ton) increments of instantaneous VSP and four speed bins (idle, low, moderate, and high speed), with additional considerations of acceleration rates to define the braking OpMode. Table 2 details the VSP and speed definitions for each OpMode. TABLE 2 MOVES Operating Mode Definitions (adapted from the EPA (11)) Description Definition and OpMode a t -2.0 Braking 0 (a t < -1.0 & a t-1 < -1.0 & a t-2 < -1.0) Idle -1 v t < v t < v t < v t Coast VSP t < Cruise/Acceleration 0 VSP t < VSP t < VSP t < VSP t < VSP t < VSP t < VSP t < VSP t Where: v t = vehicle speed (mph) a t = acceleration (m/s 2 ) VSP t = vehicle specific power (kw/ton) t = time (seconds)

6 Sentoff, Aultman-Hall & Holmén Page 6 of Database Description The final analysis data set was comprised of second-by-second trip data with variables including volunteer ID, trip ID, and road type, as well as instantaneous VSP and operating mode calculated both with estimated road grade and assuming road grade was zero. A second-by-second data set of 82 different drivers on valid, in state trips resulted. Data for each vehicle type and road type were broken down by frequency of occurrence in the second-by-second data set, as well as by number of volunteers. Records for passenger cars represented 65.7% of the sample, while trucks were 27.2% (7.2% unspecified). Total travel on rural restricted and unrestricted roads represented 3.7% and 59.0% of the travel time. Urban restricted and unrestricted were 3.7% and 33.6% respectively. Vehicle activity data for each road type (Table 3) based on mean speed, acceleration and grade showed that the restricted access facilities had greater mean speed than the unrestricted facilities, as expected. The resulting restricted urban and rural mean VSP was about two to three times that of rural unrestricted VSP, respectively. The urban unrestricted road type had the lowest average speed and corresponding VSP, as would be expected for stop-and-go, lower speed operation. Due to the prevalence of speed-time trace data from laboratory dynamometer-based studies and/or from PEMS on road studies, VSP is often calculated excluding the grade term. This makes the assumption that road grade is zero, and was calculated here as such for the purposes of comparison. The zero road grade case had slightly lower mean VSP than the VSP calculated using measured road grade for all four road types (Table 3). It is also important to note that the standard deviation of VSP was greater by margins of 40% to 180% for the estimated road grade scenario compared to the zero grade scenario. TABLE 3 Vehicle Activity Data (Mean ± Standard Deviation) by Road Type Road Type Rural Restricted Rural Unrestricted Urban Restricted Urban Unrestricted Speed (mph) 67.3 ± ± ± ± 16.5 Acceleration (m/s 2 ) 0.09 ± ± ± ± 0.48 Deceleration (m/s 2 ) ± ± ± ± 0.50 Uphill Grade (%) 3.5% ± 4.8% 3.0% ± 4.1% 2.6% ± 3.4% 1.9% ± 3.3% Downhill Grade (%) -3.4% ± 4.7% -2.9% ± 4.1% -2.4% ± 3.2% -1.9% ± 3.4% VSP Grade (kw/ton) 14.7 ± ± ± ± 8.5 VSP Zero Grade (kw/ton) 14.4 ± ± ± ± 6.1 RESULTS How Much Does Road Grade Matter? Figure 1 shows the typical VSP-based activity across all volunteers for a particular road type and the redistribution of activity across VSP bins when road grade is included in the calculation compared to the base case where road grade is assumed to be zero. It is important to note that this is not a comparison of steeply graded sections to flat sections of the volunteer/trip data base, but rather a comparison of the entire data set when grade is and is not accounted for in the calculation of second-by-second VSP.

7 Sentoff, Aultman-Hall & Holmén Page 7 of 13 Ndrivers = 24 Nrecords = Ndrivers = 73 Nrecords = Ndrivers = 62 Nrecords = Ndrivers = 77 Nrecords = FIGURE 1 VSP distributions by road type for the two road grade cases: with estimated real-world grade (blue) and assuming road grade was zero (orange). Restricted access road types had peak activity at a slightly higher VSP than unrestricted road types, with rural restricted having the highest peak of activity at 16 kw/ton VSP. Although the peak activity for urban restricted road types was higher than its unrestricted counterparts, there was a significant peak at 0 kw/ton. For unrestricted road types, peak activity occurred at a slightly higher VSP for the rural compared to the urban road type, but the urban road type had more 0 kw/ton VSP activity, due to the stop-and-go, more heavily trafficked nature of the urban driving. Figure 1 also demonstrates that including road grade dampens the peak activity and redistributes activity across VSP to either more positive or more negative bins, depending on whether the added road grade was uphill or downhill. Although the change to any specific VSP bin was not any more than about 10% of the data (with the exception of 0 kw/ton activity for the urban unrestricted road type), there could be significant implications on vehicle emissions if the shift is more than 3 kw/ton (the MOVES OpMode VSP bin resolution), effectively changing MOVES operating mode assignment and associated emissions estimates. As with VSP, the frequency of data occurring in each of the MOVES operating modes (Figure 2) was distinctive for each road type and also shifted as a function of including road grade. It is clear in Figure 2 that the restricted road types have most of the activity in the high speed bins (50 mph and above), with only 2 to 5% of the urban restricted data, and less than 1% of the rural restricted data occurring in bins with speed of less than 50 mph. Unrestricted road types had more evenly distributed data across the three speed regimes, with the greatest frequency of activity at idle (OpMode 1) for the urban unrestricted road type and at the moderate speed coasting operating mode (OpMode 21) for the rural unrestricted road type. For the unrestricted road types, data were redistributed from operating modes with VSP of 0 to 6 kw/ton to either coasting operating modes (VSP < 0 kw/ton) or to operating modes with higher VSP (> 6 kw/ton) when road grade was included in VSP. For restricted road types, activity was redistributed away from 0 to 12 kw/ton VSP operating modes, towards either coasting (VSP < 0 kw/ton) or high power operating modes with VSP greater than 12 kw/ton. This is particularly evident for rural restricted roads, where almost 30% of the data were redistributed away from OpMode 37 and over 10% were redistributed to OpMode 40.

8 Sentoff, Aultman-Hall & Holmén Page 8 of 13 Ndrivers = 24 Nrecords = Ndrivers = 73 Nrecords = Ndrivers = 62 Nrecords = Ndrivers = 77 Nrecords = FIGURE 2 Change in operating mode distributions between road grade scenarios where estimated grade is included in the equation of VSP or road grade is assumed to be zero. The shift in operating mode due to inclusion of road grade resulted from an increase or decrease in VSP. The magnitude of the shift was dependent on the road grade magnitude and direction of the shift was dependent on whether uphill or downhill operation was occurring (Figure 3). The degree to which the shift occurred (i.e. how many operating modes was the shift) would be expected to have implications in terms of emissions estimates: shifts to higher OpModes within a speed category are associated with higher emission rate estimates in MOVES. The histogram in Figure 4 indicates how many operating modes to the left or right (increase or decrease in VSP) the data were shifted. The majority of the data did not shift OpMode because the grade change was too small to change VSP by greater than 3kW/ton, the smallest VSP resolution in MOVES OpModes. An OpMode bin shift of as much as +/- 8 was observed in the Vermont dataset, with shifts ranging from -5 to +6 representing 99% of the data.

9 Sentoff, Aultman-Hall & Holmén Page 9 of FIGURE 3 Histogram of the change in operating mode between road grade scenarios, where delta OpMode is the magnitude of the shift in consecutive operating modes to the right (+) or to the left (-) (i.e. a shift of +2 could represent a shift from OpMode 35 to OpMode 38). The emissions implications of redistributing OpMode as a result of including road grade in the calculations were explored by modeling the two grade scenarios in MOVES. A base case project-level MOVES scenario for passenger cars was evaluated for an hour at noon in May of 2013 in Vermont. This assumes a seasonally appropriate temperature and relative humidity for the area. The actual age distribution of vehicles in the study was used as input because only vehicles post-model year 1996 were observed, this prevented overly conservative estimates from the presence of older, pre-emission control device era vehicles in the simulated fleet. Each of the grade scenarios were evaluated from the base case for six simulated outputs: carbon dioxide (CO 2 ), hydrocarbons (HC), carbon monoxide (CO), oxides of nitrogen (NO x ), volatile organic compounds (VOC), and total energy consumption. The percent error due to neglecting road grade was significant for the five simulated pollutants and estimated total energy consumption (Table 4). TABLE 4 Percent Error for MOVES Estimates Associated with Including Road Grade as Compared to Neglecting Road Grade Road Type Parameter Rural Rural Urban Urban Restricted Unrestricted Restricted Unrestricted CO 2 10% 16% 10% 15% HC 30% 35% 25% 24% CO 48% 44% 37% 30% NO x 17% 28% 17% 26% VOC 28% 34% 24% 23% Energy Consumption 10% 16% 10% 15%

10 Sentoff, Aultman-Hall & Holmén Page 10 of Overall, the road grade comparison indicates that emission rate errors ranging from 10-50% occur if real-world road grade is not accounted for in computing VSP used to assign MOVES operating mode. The error varies by pollutant and road type. How Much Does Driving Style Matter? The implications of driving style were also evaluated in the context of the MOVES framework. Driving style is challenging to evaluate because some variation in VSP would be expected due to traffic operation and control, such as traffic signals. Keeping this in mind, the vehicle activity across individual drivers was summarized as a function of VSP, with the normalized frequency of operation for a particular driver in a given VSP bin represented by the mean and standard deviation across all drivers. The standard deviation on the normalized activity was greater across all volunteer drivers on the restricted road types compared to the unrestricted road types, suggesting that driver style, not traffic control or operations, makes up a significant portion of driver to driver variation in the VSP distribution. It should be noted that restricted rural roads are rarely heavily congested in rural states like Vermont; this may enable distinguishing driver style on this road type. The normalized frequency of activity across all drivers as a function of OpMode (Figure 4) showed operating mode distributions that were representative of the type of operation typically encountered for each road type: restricted roads typically have high speeds of operation that confine most of the activity to the high speed bins (OpModes 33 through 40), rural unrestricted roads had most activity in the moderate and high speed bins, whereas urban unrestricted activity was confined mostly to the low to moderate speed bins. The uncertainty due to the aggregate of individual driving style and operations is represented by the standard deviation in Figure 4, with as much as plus or minus 9% around the mean volunteer activity. Ndrivers = 24 Nrecords = Ndrivers = 73 Nrecords = Ndrivers = 62 Nrecords = Ndrivers = 77 Nrecords = FIGURE 4 Normalized frequency of vehicle activity data across all individual volunteer drivers binned by OpMode. Mean (bar) and standard deviation (error bar) represents data across drivers for each road type. The emissions implications of driver to driver differences were evaluated using the upper and lower bounds of driver activity, defined by the standard deviations around the mean from Figure 4. The percent difference between the upper bounds and the average activity are summarized in Table 5 for the case of CO 2. The percentages in Table 5 represent the percent difference in CO 2 emission rate in a particular

11 Sentoff, Aultman-Hall & Holmén Page 11 of operating mode for a driver at the upper bound versus a driver at the mean. Large errors exceeding 100% would be expected for some drivers in the lower OpModes on restricted facilities and in the higher OpModes in urban unrestricted facilities due to the relatively low frequency of activity in these OpModes for these particular road types. It is of course impossible to say whether these differences balance out; the data indicate only that drivers CO 2 emissions (and associated fuel use) may differ from each other in certain conditions to a very large extent. TABLE 5 Percent Difference of Upper Bound to Mean of MOVES CO 2 Estimates based on Driving Style Road Type OpMode Rural Rural Urban Urban Restricted Unrestricted Restricted Unrestricted 0 77% 30% 59% 19% 1 46% 62% 89% 48% 11 88% 52% 80% 40% 12 86% 52% 78% 41% % 53% 85% 41% % 50% 78% 33% % 47% 88% 26% 16 97% 46% 79% 23% 21 99% 28% 59% 22% 22 85% 23% 81% 22% 23 79% 25% 88% 24% % 22% 92% 23% 25 88% 21% 73% 24% 27 94% 21% 89% 26% 28 83% 27% 58% 34% 29 81% 32% 57% 43% 30 96% 39% 69% 50% 33 19% 45% 26% 80% 35 45% 46% 29% 75% 37 30% 47% 31% 81% 38 27% 50% 32% 94% 39 33% 52% 43% 106% 40 27% 52% 69% 125% Shading indicates MOVES speed categories CONCLUSIONS There is no doubt that driving style varies across space and time, with vehicle type, operating environment, and traffic conditions. This in turn impacts emissions along with environment and vehicle factors. It is a challenging endeavor to model this system in either traffic simulation models that replicate the second-by-second movements of vehicles or EPA s MOVES which estimates emissions based on vehicle activity in OpMode categories of speed and VSP. This study is based on a uniquely large dataset

12 Sentoff, Aultman-Hall & Holmén Page 12 of of 2.5 million one-second records of in-vehicle real-world driving activity data. We cannot assess whether this is representative of overall driving in Vermont but it is the routine driving of the 82 drivers over 10-day periods. The results indicate that differences in person-to-person driving style have a significant impact on the level of emissions estimated. It is similarly obvious that road grade varies over space and affects the VSP of vehicles. For this sample, omission of the grade resulted in overestimating the amount of operating time in the mid-range MOVES bins and underestimating time in both the higher and lower OpModes. This translates into a large number of misallocations of MOVES OpMode assignment. Depending on road type, the correct bin allocation was only made 33-55% of the time when grade was excluded. This translates into a total emissions estimate error of 10-48% depending on the road type and pollutant due to grade alone. The implications of these results related to grade are straightforward: the error level from omitting grade in hilly areas is very high. It is currently feasible to include grade in emissions models such as MOVES and steps to incorporate grade data into activity estimation should be undertaken using widely available GPS and DEM data. Next steps related to driving style are more complicated to assess. This relatively large dataset was not large enough to estimate the true level of significance for differences in driving style. It was only possible to say that driving style could potentially impact emission estimates by large percentages. A method to consider driving style while controlling for operational conditions and traffic control that also varies over time and space is necessary. A very large dataset would be needed to compare distributions and variances for a single driver compared to differences between drivers. However, such large datasets are becoming readily available as GPS and cell phone datasets continue to proliferate. These results suggest the use of such data to define a driving style typology would be productive in creating more accurate microsimulation and emissions models especially for smaller project-level applications. Understanding the true distribution and extent of different driver styles will be useful beyond MOVES and its OpModes, it will be critical to evaluate the efficacy of programs such as eco-driving and to assess the cost-benefits of in-vehicle technologies to adapt second-by-second vehicle operations. ACKNOWLEDGEMENTS The authors acknowledge funding through the University Transportation Center program of the US DOT at the University of Vermont. The contributions of Jon Dowds and James Sullivan to data collection originally funded by the Vermont Agency of Transportation are gratefully acknowledged.

13 Sentoff, Aultman-Hall & Holmén Page 13 of REFERENCES 1. Holmén, B. A., and D. A. Niemeier. Characterizing the effects of driver variability on real-world vehicle emissions. Transportation Research Part D: Transport and Environment, Vol. 3, No. 2, Mar. 1998, pp Jackson, E., Y. Qu, B. A. Holmén, and L. Aultman-Hall. Driver and Road Type Effects on Light- Duty Gas and Particulate Emissions. Transportation Research Record: Journal of the Transportation Research Board, No. 1987, Frey, H. C., K. Zhang, and N. M. Rouphail. Fuel Use and Emissions Comparisons for Alternative Routes, Time of Day, Road Grade, and Vehicles Based on In-Use Measurements. Environ. Sci. Technol., Vol. 42, No. 7, 2008, pp Nam, E. K., C. A. Gierczak, and J. W. Butler. A Comparison Of Real-World and Modeled Emissions Under Conditions of Variable Driver Aggressiveness. Presented at the 82nd Annual Meeting of the Transportation Research Board, Washington, DC, Zhang, K., and H. C. Frey. Road Grade Estimation for On-Road Vehicle Emissions Modeling Using Light Detection and Ranging Data. Journal of the Air & Waste Management Association, Vol. 56, No. 6, 2006, pp Boriboonsomsin, K., and M. Barth. Impacts of Road Grade on Fuel Consumption and Carbon Dioxide Emissions Evidenced by Use of Advanced Navigation Systems. Transportation Research Record: Journal of the Transportation Research Board, Vol. 2139, No. -1, Dec. 2009, pp Cicero-Fernândez, P., J. R. Long, and A. M. Winer. Effects of Grades and Other Loads on On-Road Emissions of Hydrocarbons and Carbon Monoxide. Journal of the Air & Waste Management Association, Vol. 47, No. 8, 1997, pp Vermont Agency of Transportation. 2012: Statewide VMT by Functional Class Dowds, J., J. Sullivan, and L. Aultman-Hall. Comparisons of Discretionary Passenger Vehicle Idling Behavior by Season and Trip Stage with Global Positioning System and Onboard Diagnostic Devices. Transportation Research Record: Journal of the Transportation Research Board, Vol. 2341, No. -1, Dec. 2013, pp Koupal, J., L. Landman, E. Nam, J. Warila, C. Scarbro, E. Glover, and R. Giannelli. MOVES2004 Energy and Emission Inputs Draft Report. Publication EPA420-P EPA Office of Transportation and Air Quality, Mar EPA. Development of Emission Rates for Light-Duty Vehicles in the Motor Vehicle Emissions Simulator (MOVES2009). Publication EPA-420-P Assessment and Standards Division Office of Transportation and Air Quality U.S. Environmental Protection Agency, Aug