Case Studies of Truck Activity Data for Emissions Modeling (2019)

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.

Guide to Truck Activity Data for Emissions Modeling C-1 Appendix C. Case Study #3: Truck Starts and Soak Time from GPS Data C.1 Emissions Model Inputs Supported • Starts per vehicle (by source type, hour of day, and weekday/weekend). • Start operating mode distribution (soak time distribution). C.2 Level of Effort for Local Application This data source requires a moderate level of effort for local application. For the case study, the effort to translate start activity metrics from StreetLight into MOVES inputs was minimal, because the project team worked with StreetLight to help define metrics that matched the MOVES inputs as closely as possible. The effort required to convert these aggregated metrics into MOVES inputs centered on mapping to MOVES source types, calculating the pertinent MOVES parameters, and formatting into MOVES input tables. C.3 Overview This case study examines aggregated trip activity data provided by StreetLight Data, Inc. (StreetLight) to develop start activity inputs for trucks, separated by medium-duty and heavy-duty trucks. StreetLight compiles data on vehicle locations gathered from GPS on connected vehicles, mobile phones and commercial fleet management systems to produce trip metrics at very detailed geographic resolution. StreetLight does not provide raw trip data, but rather worked with the 08-101 project team to produce aggregate statistics would could be processed into MOVES inputs for starts per vehicle, and start operating mode (soak time) distribution. Data were obtained for three sample counties. C.3.1 Case Study Approach The following data evaluations and comparisons were made in this case study: • Evaluated starts per truck by source type (single-unit and combination) by hour of day and weekday/weekend. • Evaluated dwell time (soak time) distributions by source type (single-unit and combination) by hour of day and weekday/weekend. • Compared with MOVES defaults. • Evaluated emissions impacts of using StreetLight data for the three counties versus MOVES defaults.

Guide to Truck Activity Data for Emissions Modeling C-2 C.4 Data Sources Data Source Source Agency/Organization Availability and Cost Trip data compiled and aggregated from GPS devices (primarily commercial fleet management systems for trucks) StreetLight Data, Inc. Starts/vehicle and soak time distributions for three counties purchased within project data purchase budget C.5 Data Processing and Analysis StreetLight aggregated GPS trip data into starts/vehicle and total counts by “dwell time” that were roughly equivalent to MOVES soak bins (e.g., start operating mode distribution). These statistics are provided as part of results of the case study, and include underlying data on “trip starts” and “trip ends”; it is important to note the data provided are not absolute trips or absolute vehicles, however. StreetLight uses an index to normalize their database to account for constant shifts in the number of vehicles supplying data over time. Details on the StreetLight trip index are provided in the sidebar. An issue introduced with the use of telematics data is the need to define what constituted a “start” and a “vehicle” for estimation of starts/vehicle. This is not always straightforward when looking at a fixed geographic area (e.g., county), because GPS-based telematics data are tracking vehicles that migrate across county lines, and may just pass through. For a given county, the project team and StreetLight defined “start” as any new trip that originated in the county, regardless of whether the trip stayed in the county of not. “Vehicle” was defined on an hourly basis, as any vehicle whose first start within the given hour was within the county, or that did not take a trip (i.e., “soaked”) in the county for that hour. These vehicles were termed “resident vehicles.” The count of resident vehicles is not related to where the vehicle was actually registered, which is not available in the StreetLight data. Figure C.1 provides a schematic and detailed description of how the starts and vehicles were defined, and how the starts/vehicle metric was calculated. StreetLight Trip Index (provided by StreetLight Data, Inc.) StreetLight provides the results for Travel Projects as "trip index". The trip index values do not represent number of trips or vehicles. Rather, the values represent the relative amount of trips. All personal vehicle trip index values are comparable and all commercial vehicle trip index values are comparable, but you cannot compare between vehicle types. For US projects, the value is normalized by adjusting the number of trips in the data sample to the actual number of trips in a region around Sacramento, CA, as derived from the measurements published by the California Department of Transportation. This allows capture of monthly and seasonal variation more accurately, even as the sample grows. The NCHRP 8-101 project is a custom project that contains a custom StreetLight Trip Index, and cannot be compared with other projects unless specifically stated. The data was normalized and delivered as metrics in the form of an index to protect privacy of data users and allow for comparison between different months of data should there be a follow-up study.

Guide to Truck Activity Data for Emissions Modeling C-3 Figure C.1 Trip Metric Definitions and Example The case study focused on three counties: Denver (Colorado), Fairfax (Virginia), and Norfolk (Virginia). These counties were chosen to represent metro areas with a diversity of truck activity. Denver is a major freight hub for points west, and also is a core county for a major city; it was thought this county would represent long-haul freight activity at warehouses and distribution centers, and extensive local delivery activity common in urban hubs. Fairfax is a large suburban county adjacent to Washington, D.C. (the population of 1.2 million is larger than Denver County) and was chosen to represent typical suburban truck activity. Norfolk County was chosen to include the Port of Virginia; it was expected that heavy truck activity would be predominantly port-related, such as drayage trucks. StreetLight provided starts/vehicle and dwell time metrics for each of these counties by hour of the day and day type (weekend/weekday). C.5.1 Processing Method to Generate MOVES Tables Two MOVES input tables were generated from the StreetLight data: StartsPerVehicle, and ImportStartsOpModeDistribution. Both tables have key fields of source type and hourDayID, which combines hour of the day and day type (weekday or weekend). The processing of the StreetLight data into these tables was minimal, since the data aggregation performed by StreetLight was configured to match these MOVES inputs as directly as possible. Both tables required mapping StreetLight categories of medium duty (MD) and heavy duty (HD) into MOVES source types. The most straightforward approach was to assign MD to Single- Unit Trucks (52 and 53), and assign HD to Combination Trucks (61 and 62). These assignments corresponded with the weight-based regulatory classes dominant in each source type. After translating the

Guide to Truck Activity Data for Emissions Modeling C-4 StreetLight MD data to 52 and 53s, and StreetLight HD to 61 and 62s, the start/vehicle data could be imported directly into StartsPerVehicle. This table is not listed under the “Starts” tab of the County Data Manager, but can be accessed through the “Generic” tab, and read into a custom database specified by the user. This database is then entered in the Manage Input Datasets GUI screen when setting up a runspec. Dwell times were provided by StreetLight, expressed as an indexed trip count in each dwell time bin. These bins were set up to match the MOVES start operating mode distributions as closely as possible. The main differences was that the data for the shortest soak period could only be provide for periods of 0 to 15 minutes, instead of 0 to 6 defined by MOVES. StreetLight bins of 0 to 15 and 15 to 30 were combined to approximate the 6- to 30-minute bin in MOVES. The index trips by bin were normalized to total indexed trips to create the operating mode fraction, by hour and day time. These were then imported into the table ImportStartsOpModeDistribution through the County Data Manager “Starts” tab. Both tables were read into the database 8101_altstartspervehicle_denver_in. C.6 Findings from Sample Data C.6.1 Starts and Soak Time Figure C.2 and Figure C.3 show starts/truck for each combination of county (D = Denver, F = Fairfax, N = Norfolk), truck class (H = Heavy, M = Medium) and day type (Wd = Weekday, We = Weekend). Figure C.2 shows medium trucks, and Figure C.3 shows heavy trucks. The data show the following trends: • HD starts per truck are much higher than MD starts per truck (contradictory to other research findings, as discussed in Section C.7). • Weekend starts are much lower than weekday for MD trucks, but about the same for HD trucks. • Denver shows a sharp increase in HD starts/truck over the course of a day, while Fairfax and Norfolk are more consistent. Starts/truck peak in the early evening (6 – 9 p.m.) for all three cities. • MD truck start patterns follow normal business hours on weekdays, ramping up through 8 a.m. and remaining fairly level until 4 p.m.

Guide to Truck Activity Data for Emissions Modeling C-5 Figure C.2 StreetLight Starts per Truck by County/Class/Day Type: Medium Trucks Figure C.3 StreetLight Starts per Truck by County/Class/Day Type: Heavy Trucks 0 0.05 0.1 0.15 0.2 0.25 0.3 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour of Day Starts Per Truck per Hour D/M/Wd D/M/We F/M/Wd F/M/We N/M/Wd N/M/We 0 0.2 0.4 0.6 0.8 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour of Day Starts Per Truck per Hour D/H/Wd D/H/We F/H/Wd F/H/We N/H/Wd N/H/We

Guide to Truck Activity Data for Emissions Modeling C-6 The corresponding MOVES defaults for starts/truck are shown in Figure C.4 and Figure C.5. Source Type 52 and 53 (Figure C.4) correspond approximately to StreetLight’s medium-duty trucks, and source Type 61 and 62 (Figure C.5) correspond to StreetLight’s heavy trucks. Comparing these defaults to the StreetLight data yields several observations: • For weekdays, the short-haul trucks (52 and 61) default starts are higher than for long-haul trucks (53 and 62). StreetLight does not provide a short-haul/long haul designation. • The single-unit short-haul (52) default is much higher than StreetLight MD, while the single-unit long haul is moderately higher for some peak period hours. • The combination short haul (61) is comparable to StreetLight HD data through 8 a.m., but decreases over the day, while the StreetLight data continues to increase through early evening. The default combination long-haul (62) starts are much lower than StreetLight HD. • For weekends, the single-unit short-haul default is much higher than StreetLight, while single-unit long haul is comparable. The combination truck defaults (same as for weekdays) are much lower than StreetLight. Figure C.4 MOVES Default Starts per Truck: Single-Unit Trucks 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour of Day Starts Per Truck per Hour Default/52/Wd Default/53/Wd Default/52/We Default/53/We

Guide to Truck Activity Data for Emissions Modeling C-7 Figure C.5 MOVES Default Starts per Truck: Combination Trucks Figure C.6 shows average dwell (soak) times for MD trucks. These values represent the average length of time (in minutes) since the last start for a truck starting in hour x. Although MOVES inputs are the full distribution, these average times give a sense of difference between counties, and hour-to-hour. As shown, the average dwell times peak in early morning and late evening. The dip in dwell times during business hours corresponds with increased starts shown in Figure C.2. Figure C.7 shows average dwell (soak) times for HD trucks. The average dwell times mirror the starts/vehicle in Figure C.2; the shorter the dwell time, the more starts, and vice-versa. As shown the HD trends track closely across counties, with Denver having longer dwell times in early hours corresponding with few starts/vehicle. Weekday and weekend patterns are in close agreement as well. 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour of Day Starts Per Truck per Hour Default/61/Wd-We Default/62/Wd-We

Guide to Truck Activity Data for Emissions Modeling C-8 Figure C.6 StreetLight Average Dwell (Soak) Time by County/Day Type Medium Duty Figure C.7 StreetLight Average Dwell (Soak) Time by County/Day Type Heavy Duty 0 50 100 150 200 250 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Average Dwell Time (time since last start), minutes Hour of Day D/M/Wd D/M/We F/M/Wd F/M/We N/M/Wd N/M/We 0 50 100 150 200 250 300 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Average Dwell Time (time since last start), minutes Hour of Day D/H/Wd D/H/We F/H/Wd F/H/We N/H/Wd N/H/We

Guide to Truck Activity Data for Emissions Modeling C-9 Similar to the StreetLight data, MOVES default soak times will mirror the default starts/vehicle shown in Figure C.4 and Figure C.5; when starts/vehicle increase, average soak time decreases. The comparisons of starts/vehicle summarized above will also be mirrored in the comparison of soak times. For example, since default starts/vehicle for single-unit short-haul trucks are much higher than StreetLight MD trucks, it would follow that default soak times are much lower. High soak times translates to an increased proportion of starts when the engine and catalyst (for gasoline equipped trucks) are colder, leading to higher emissions. Figure C.8 and Figure C.9 compare the soak time distributions estimated from StreetLight Denver data with MOVES defaults. The StreetLight data are limited in not breaking out the shortest soak periods (0 to 6 minutes, Operating Mode 101) or the longest soak periods (greater than 12 hours, Operation Mode 108). The “102” bin for StreetLight includes all soaks up to 15 minutes, and is shaded to match the default chart for comparison. The StreetLight soak time distribution shows a high proportion of shorter soak periods (30 minutes or less, modes 102 and 103) throughout the day, peaking over 80 percent for hour 17 (5:00 – 6:00 p.m.). In contrast the default distributions have a similar trend during midday, but are dominated by longer soak durations in off-peak hours. The StreetLight trends appear dominated by the heavy truck data, which had significant start activity over all hours of the day. Figure C.8 MOVES Default Weekday Operating Mode (Soak Time) Distribution Single-Unit and Combination Trucks 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 15 25 35 45 55 65 75 85 95 105 115 125 135 145 155 165 175 185 195 205 215 225 235 245 Operating Mode Fraction HourDayID 101 102 103 104 105 106 107 108

Guide to Truck Activity Data for Emissions Modeling C-10 Figure C.9 StreetLight Denver Weekday Operating Mode (Soak Time) Distribution Medium and Heavy Trucks C.6.2 Emissions Sensitivity Results To evaluate emissions sensitivity to the new start activity data, MOVES was run for Denver County with default inputs, and with StreetLight data (specifying 8101_altstartspervehicle_denver_in in Manage Input Datasets screen) to estimate daily and hourly emissions for source type 52, 52, 61 and 62. Figure C.10 shows emissions for each source type for a default run, separated by start and running. As shown, 52s (single-unit short haul) dominate for HC and CO, and have the largest start contribution. This is because, within MOVES, 52s have a much higher share of gasoline-fueled trucks than the other source types. The other source types have a very low start contribution, because they are dominated by diesel trucks, which in MOVES have very low start emissions. Diesel start emissions are an emerging issue, especially for NOx, as recent research shows cold start emissions are relatively high on trucks equipped with selective catalytic reduction. When future versions of MOVES incorporate this, the importance of start emissions on diesel trucks will increase. For MOVES2014a, however, the start emissions are estimated to be almost negligible. The majority of start emissions sensitivity is therefore traced to the single-unit short-haul trucks (52s). 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 15 25 35 45 55 65 75 85 95 105 115 125 135 145 155 165 175 185 195 205 215 225 235 245 Operating Mode Fraction HourDayID 102 103 104 105 106 107

Guide to Truck Activity Data for Emissions Modeling C-11 Figure C.10 Baseline Denver County Emissions by Truck Source Type Tons per Day Figure C.11 shows the percent change in emissions (start only, and total) when only the StreetLight starts per vehicle are run (with default soak times), and when both the StreetLight starts and soak times are run. As shown, for total emissions the two inputs are offsetting. The StreetLight starts/vehicle reduces start emissions significantly for HC, CO and NOx. This is due to the dominance of 52s in overall truck start emissions, where StreetLight had significantly lower starts than the default. The longer soak times (correlated to fewer starts) then increase starts emissions, so that the overall change is lower than starts/vehicle only. PM2.5 follows a different pattern, increasing with the StreetLight starts/vehicle. This is because the HD trucks account for a larger fraction of PM2.5 emissions than MD trucks compared to the other pollutants, and StreetLight shows a high number of starts/vehicle for HD relative to default. The higher number of HD starts observed in the StreetLight data therefore outweighs the lower number of MD starts, leading to higher PM2.5 emissions, whereas for the other pollutants, the lower number of MD starts outweighs the higher number of HD starts. Overall PM2.5 start emissions still decrease with all StreetLight inputs, however, due the influence of shorter soak periods for HD trucks. Note that the changes in total emissions are quite small, because of the low start contribution to overall truck emissions.

Guide to Truck Activity Data for Emissions Modeling C-12 Figure C.11 Total Change in Denver County MD/HD Emissions Using StreetLight Data versus Default Hourly differences are also important. Figure C.12 shows the change in start emissions by hour for HC and CO (the most consequential pollutants for start emissions) by hour. As shown, start emissions in off-peak hours are significantly increased from the StreetLight data, while emissions during peak/business hours are significantly lower. This doesn’t reflect the relative importance of each hour to daily emissions; because most emissions are generated during peak/business hours, the reduction in start emissions show during these hours drive the overall daily trends seen in Figure C.10. Figure C.12 Hourly Change in MD/HD Start Emissions using StreetLight Data versus Default C.6.3 Summary of Findings Conclusions drawn from this case study include: • It is possible to derive estimates of starts per vehicle and dwell times with GPS data, which is much more readily available than GPS linked with engine control unit data as demonstrated in Case Study #4. • A drawback of relying on GPS data only is that trip starts must be used as a proxy for engine starts. Multiple starts per stopped event, or continuous idling that does not result in a start when the trip begins, are not accounted for.

Guide to Truck Activity Data for Emissions Modeling C-13 • A challenge that will be encountered with any dataset is defining a vehicle “population” (for the denominator of the starts/vehicle metric) comparable to the MOVES definition. The idea of vehicle population specific to a small geographic area (e.g., a county) is somewhat of a synthetic construct for trucks, especially for long-haul trucks that move all over the country and may be registered anywhere. The definition used here (the number of trucks with their first start in the domain or entirely dwelling in that domain within the hour) was unavoidably arbitrary. Using a vehicle-based data source (such as Vnomics—see Case Study #4) can produce an accurate measure of total starts per vehicle per hour or day, but not for small domain areas since the vehicles are moving among domains. Alternative heavy vehicle indicators (such as starts per vehicle-mile) might be considered for future versions of MOVES since vehicle “population” often is already estimated from vehicle- miles of travel, especially for long-haul trucks. • Because of the lack of engine on/off data, GPS-only sources may be more reliable for looking at temporal patterns (distributions) of starts than for estimating total starts per truck or soak times. • The impact on emissions of alternative start and soak time distributions compared to the MOVES defaults is quite small, except for the short-haul single-unit truck category which includes a high fraction of gasoline vehicles. In the future, as MOVES data change, start emissions may become more important for diesel vehicles. C.7 Transferability • Start activity patterns for Fairfax and Norfolk Counties are generally similar despite these counties having potentially very different industrial structures and truck trip activity generators. Patterns differed somewhat in Denver County, with a pattern of starts skewed towards later in the day than the Virginia counties. • The StreetLight data also shows somewhat different hourly patterns than the MOVES default data, although it cannot be determined with certainty how much this is due to data definitional issues as compared to actual differences in truck patterns in different locations. • Medium trucks in the StreetLight data have fewer starts per truck than heavy trucks. This is inconsistent with findings from NREL’s analysis of FleetDNA (Kotz et al, 2018) that shows much single-unit short-haul trucks making 36 starts per day compared with 12 for combination short-haul and less than one for combination long-haul trucks. Vocationally-based sampling differences between the two data sets could be at least partly responsible for the difference; the NREL research found very high start rates (over 50 per day) for parcel delivery trucks, with most other short-haul vocations in the range of 11 to 21 starts per day. Vocational representation in the StreetLight data is unknown, but the estimated starts per day for medium trucks is quite low compared to almost any short-haul vocational category in FleetDNA.