Guide to 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.

34 6.1 Overview MOVES uses data regarding the number of starts by time of day and the distribution of soak times in the estimation of engine start emissions and evaporative fuel vapor losses (soak time is defined as the length of time the vehicle engine remained off prior to being started). The base emission rate for engine starts is based on a 12-hour soak period. All engine soaks longer than 12 hours assume the same engine start emission rate as for 12 hours. For all engine soaks shorter than 12 hours, the base engine start emission rate is adjusted based on soak time bins (or start operation modes). MOVES2014 allows users to specify the number of engine starts in each month, day type, and hour of the day, as well as to specify by source type and vehicle age. Start and soak emissions are considered off-network emissions (i.e., not associated with a particular road- way link, as are running emissions). MOVES has eight soak period bins as shown in Table 6.1. 6.1.1 Project-Scale Inputs For project-scale runs, the user is expected to provide data for total start activity and OMDs. Total start activity is specified in the OffNetworkLink table via the Off Network tab of the Project Data Manager. Activity is specified as the fraction of the total vehicle population that starts during the project period being modeled (typically 1 hour). An example is shown in Table 6.2; in this case, there would be 600 total starts in the hour (100 per source type, since one-half of the vehicle population is starting). These starts are then assigned soak time distributions based on user input of the OpModeDistribution table available via the Operating Mode tab of the Project Data Manager. An unpopulated example is shown as Table 6.3. 6.1.2 County-Scale Inputs For county-scale application, MOVES users can enter start activity data in several ways. Using the Starts tab of the Data Importer, users may input total starts per day in the analysis zone and the distribution of these starts by source type, hour, and month. The startsPerDay table can be populated with total starts for all vehicles over an entire day (weekend or weekday). Additional tables then allocate these to source type, month, and hour. These tables are called startsSourceTypeFraction, startsMonthAdjust, and startsHourFraction. Unpopulated versions of these tables are shown as Tables 6.4 to 6.7. The total number of starts per day and related allocation tables are typically more difficult to obtain than the number of starts per vehicle, which is the most direct metric of starts available from instrumented vehicle and telematics data sets. Data on the number of starts per vehicle can S E C T I O N 6 Starts and Soak Times

Starts and Soak Times 35 Nominal Soak Period (Minutes) opMode ID Definition 3 101 Soak Time < 6 minutes 18 102 6 minutes ≤ Soak Time < 30 minutes 45 103 30 minutes ≤ Soak Time < 60 minutes 75 104 60 minutes ≤ Soak Time < 90 minutes 105 105 90 minutes ≤ Soak Time < 120 minutes 240 106 120 minutes ≤ Soak Time < 360 minutes 540 107 360 minutes ≤ Soak Time < 720 minutes 720 108 720 minutes ≤ Soak Time Source: U.S. EPA, 2015. Table 6.1. Start operating modes in MOVES. zoneID sourceTypeID vehiclePopulation startFraction extendedIdleFraction parkedVehicleFraction 120110 21 200 0.5 0 0 120110 31 200 0.5 0 0 120110 32 200 0.5 0 0 120110 42 200 0.5 0 0 120110 61 200 0.5 0 0 120110 62 200 0.5 0 0 Table 6.2. Sample OffNetworkLink table. sourceTypeID hourDayID linkID polProcessID opModeID opModeFraction 52 75 202 101 52 75 202 102 52 75 202 103 52 75 202 104 52 75 202 105 52 75 202 106 52 75 202 107 52 75 202 108 Table 6.3. Sample OpModeDistribution table. zoneID dayID yearID startsPerDay 120110 5 2016 120110 2 2016 …a aEllipsis points indicate that the selected data shown are a sample and there may be many more rows of data in the table. Table 6.4. Sample StartsPerDay table.

36 Guide to Truck Activity Data for Emissions Modeling be entered via the StartsPerVehicle table in the Generic Importer, which was the approach used for Case Study #3 (see Appendix C, available in NCHRP Web-Only Document 261). An example StartsPerVehicle table is shown in Table 6.8. A default StartsPerVehicle is housed in the execution database, MOVES’ repository for intermediate calculations, following a MOVES national-scale run using default inputs. Although it is an intermediate calculation, a custom StartsPerVehicle table can be provided by the user (through the Generic tab on the Data Importer). This user-input data will override the default table, which is generated by MOVES from the SampleVehicleTrip table in the default database. Total starts are then calculated internally as the product of StartsPerVehicle and vehicle popu- lation and used to populate the Starts table. An example excerpt of the Starts table is shown in Table 6.9, also housed in the execution database. A custom Starts table can be provided by the user (through the Generic tab on the Data Importer), which overrides the default table generated from SampleVehicleTrip. A custom Starts table is less likely to be generated by users because calculat- ing starts as a function of age is a challenge for most users and requires extensive preprocessing. The distribution of soak times preceding these starts also is calculated from the Sample VehicleTrip table, and used to populate the StartOpModeDistribution table. The user can sourceTypeID allocationFraction 21 31 32 52 53 …a aEllipsis points indicate that the selected data shown are a sample and there may be many more rows of data in the table. Table 6.5. Sample StartsSourceTypeFraction table. monthID monthAdjustment 1 2 …a aEllipsis points indicate that the selected data shown are a sample and there may be many more rows of data in the table. Table 6.6. Sample StartsMonthAdjust table. zoneID dayID hourID allocationFraction 120110 5 1 120110 5 2 120110 5 3 …a … …. aEllipsis points indicate that the selected data shown are a sample and there may be many more rows of data in the table. Table 6.7. Sample StartsHourFraction table.

Starts and Soak Times 37 supply custom OMDs that override these defaults by populating the ImportStartOpMode Distribution table via the Starts tab of the Data Importer; the structure of this table is shown in Table 6.10. The OpModeIDs correspond to bins that correspond to a range of time periods preceding starts—from a period of 0 to 6 minutes (hot start) to a period of 720+ minutes (cold start), as defined in Table 6.1. 6.2 MOVES Embedded Data At the project scale, estimating off-network emissions requires data characterizing the specific off-network location being modeled. For some inputs, embedded defaults can be used (age distribution, OMD if average speed is used, meteorology, fuels), but a complete off-network analysis cannot be performed without some site-specific data. County-scale, off-network inputs are captured by defining the trip characteristics of vehicles. Default trip characteristics are calculated internally based on data in the sampleVehicleTrip table, which includes individual trip-level data from instrumented vehicle activity studies. The sourceTypeID hourDayID startsPerVehicle startsPerVehicleCV 52 12 0.0344828 0 52 15 0.0446249 0 52 22 0.0344828 0 52 25 0.0446429 0 52 32 0.0344828 0 …a … … … aEllipsis points indicate that the selected data shown are a sample and there may be many more rows of data in the table. Table 6.8. Sample StartsPerVehicle table. hourDayID monthID yearID ageID zoneID sourceType starts startsCV 12 7 2016 0 80310 52 29.0839 0 12 7 2016 0 80310 53 8.5280 0 12 7 2016 0 80310 61 40.1205 0 12 7 2016 0 80310 62 26.0619 0 12 7 2016 1 80310 52 27.9213 0 …a … … … … … … … aEllipsis points indicate that the selected data shown are a sample and there may be many more rows of data in the table. Table 6.9. Sample Starts table. sourceTypeID hourDayID linkID polProcessID opModeID opModeFraction aThe same table structure is used for county-scale and project-scale inputs. For county-scale application, linkID is prepopulated via the Data Importer (when exporting a template) with the CountyID+"1" to denote off network. Table 6.10. ImportStartOpModeDistribution table structure.a

38 Guide to Truck Activity Data for Emissions Modeling data in this table are used as the basis for number of vehicle starts (based on source type population) and source hours parked (SHP), two fundamental activity elements of off-network start and evaporative emissions. Default starts per vehicle are estimated directly from individual trips in sampleVehicleTrip, which are time- and day-stamped to provide a precise accounting for the number of trips per day taken by each vehicle included in sampleVehicleTrip, along with the exact start and end times. This allows an internally consistent method to integrate the start and soak activity necessary for calculating start and evaporative emissions. The individual truck trips used for MOVES defaults are estimated from GPS data from a sample of 120 trucks collected from 1997 to 1998 (U.S. EPA, 2016). The EPA designed this approach so users could repopulate sampleVehicleTrip with custom individual trip data from other instrumented vehicle studies, or telematics; in practice, however, this feature is rarely used. This guide focuses on the intermediate trip activity inputs that are more readily accessible to users. 6.3 Sensitivity/Importance Emissions from starts tend to be less significant for diesel vehicles (the large majority of trucks) than for gasoline vehicles and show only modest sensitivity to start and soak period inputs. The NCHRP Project 08-101 team performed a sensitivity analysis using a project-level MOVES run (starts, soak, and extended idling) composed of two 0.5-mile rural, restricted- access, on-network links and one off-network link. The vehicle population was split evenly between gasoline-fueled light commercial trucks and diesel-fueled combination long-haul trucks. The off-network vehicle populations were set at 10 percent of the on-network popula- tions. Increasing the off-network number of starts by a factor of three increased total emis- sions (on and off network) by up to 3 percent. Changing the soak period from a hot start to cold start increased emissions by 5 percent for hydrocarbons (HCs) and 1 percent for NOx and PM2.5. The changes were negligible for diesel trucks only, but substantial for gasoline trucks only, as shown in Table 6.11.8 The NCHRP Project 08-101 team also performed a sensitivity test on start emissions using starts data estimated from StreetLight, as described in Case Study #3 (see Appendix C, avail- able in NCHRP Web-Only Document 261). Use of the StreetLight-estimated starts and/or soak times reduced start emissions for Denver County by up to 50 percent for NOx and VOC and changed PM2.5 by −9 to +25 percent, compared to default MOVES parameters. Hourly start emissions varied by ±50 to 100 percent. However, the effect on total daily emissions, as shown in Table 6.11, is small (5 percent or less for a truck population that is one-half gasoline-fueled trucks and one-half diesel-fueled trucks) due to the small contribution of start emissions to total emissions. For gasoline trucks only, the impact was greater, particularly on HC, for which total daily emissions decreased by 17 to 24 percent using the StreetLight data, as compared to using the default data. The excessive weighting of diesel truck emissions as compared to gasoline truck emissions in the combined results is due mainly to extended idle emissions from source type 62 vehicles.9 8 The EPA has proposed updating the start emission rates for 2010 and later heavy-duty diesel trucks incorporating recent data showing high cold-start emissions from trucks equipped with selective catalytic reduction. This update is expected to significantly increase the contribution of start emissions for HC and NOx. 9 Analysis of data in Case Study #4 (see Appendix D, available in NCHRP Web-Only Document 261) suggests that MOVES defaults may significantly overestimate extended idling, meaning that the start contribution to total emissions could be larger than shown here.

Starts and Soak Times 39 6.4 Generating Local Data Options for generating local data on truck start patterns may include • Conducting field surveys (instrumented vehicles) with portable data loggers. • Purchasing GPS/locational + ECU data collected from truck fleets. • Purchasing only GPS/locational data and using these data to infer start patterns. Field surveys allow the analyst the most control over the vehicle population; however, field surveys will be the most expensive option and therefore the most unlikely option to be taken, except in the case of very large projects. GPS data from third-party providers will likely be the easiest and cheapest data to obtain. However, 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. This limitation can be overcome by purchasing GPS data linked with ECU data from third-party providers, but the availability of such data currently is very limited. Purchased telematics data (GPS only or GPS/ECU) may contain limited or no information on the characteristics of the truck (e.g., vocation, idle reduction technology, carrier operational pol- icies), meaning that it may be difficult to verify that the activity of the purchased sample is repre- sentative of the entire truck population. Recent research suggests that starts, idling, and hotelling activity vary substantially by vocation and geography (e.g., urban versus rural) (Boriboonsomsin Parameter Varied Scenario Change Delta Composite HC Delta Composite NOx Delta Composite PM2.5 Combined Gasoline and Diesel Trucks Project-Scale Scenario, Calendar Year: 2020 Number of Starts 125 375 3% 1% 1% Soak Period (operating mode) Hot Start Cold Start 5% 1% 1% Denver County, Default versus StreetLight Start Data Starts per Vehicle Default StreetLight −5% −1% 0% Starts per Vehicle and Soak Times Default StreetLight −3% −3% 0% Gasoline Trucks Project-Scale Scenario, Calendar Year: 2020 Number of Starts 125 375 43% 33% 24% Soak Period (operating mode) Hot Start Cold Start 78% 26% 36% Denver County, Default versus StreetLight Start Data Starts per Vehicle Default StreetLight −24% −7% −5% Starts per Vehicle and Soak Times Default StreetLight −17% −5% −4% Diesel Trucks Project-Scale Scenario, Calendar Year: 2020 Number of Starts 125 375 0% 0% 0% Soak Period (operating mode) Hot Start Cold Start 0% 0% 1% Denver County, Default versus StreetLight Start Data Starts per Vehicle Default StreetLight 1% 0% 0% Starts per Vehicle and Soak Times Default StreetLight 1% 0% 0% Source: Analysis by Eastern Research Group, Inc. Table 6.11. Sensitivity analysis on starts: change in total truck emissions.

40 Guide to Truck Activity Data for Emissions Modeling et al., 2017; Kotz et al., 2018). Furthermore, it is difficult to measure vehicle population with a locally bounded telematics data set in a way that is consistent with MOVES definitions, making it difficult to esti- mate certain inputs, such as starts per vehicle per hour. The representa- tiveness of any particular data source (including those described here), therefore, needs to be considered before generalizing the results to the entire truck population. Currently, purchased telematics data may be most useful for looking at temporal patterns of activity (start fractions by hour). Table 6.12 summarizes the various current start inputs in MOVES and which data sources may be applicable for developing local values for each input. 6.4.1 Field Data Collection Source, Availability, and Cost Project-scale field data on start patterns can be collected by instru- menting a sample of the vehicles to be modeled. Low-cost data loggers slightly larger than a thumb drive can be plugged into OBD ports to record when the engine is turned on and off. The option of instru- menting sample vehicles is more realistic in situations where there is a locally based fleet (e.g., drayage trucks serving a port) than in a general highway situation with vehicles originating in many locations. While the data loggers are easy to put on the vehicles, an effort to instrument vehicles for this purpose does require an upfront effort to recruit participants and oversee instrumentation, data collection, and analysis needs. The costs of such a program depend on the support needed to recruit vehicles, manage device installation, and compile and analyze data. Low-overhead programs (requiring under $100,000 for a targeted program) can mail devices to owners who install them in the OBD or ECU port, from which the loggers automatically transmit data to a central server. At the other end of the spectrum are programs that require drivers to bring vehicles to a central location for installation and removal of loggers by technicians, with data manually compiled from individual devices. These programs may cost hundreds of thousands of dollars or more. MOVES Input Instrumented Vehicle Surveys Purchased Telematics— GPS Only Purchased Telematics— GPS/ECU OffNetworkLink (project scale—start, extended idle, and parked fractions) Possiblya No No OpModeDistribution Possiblya No No startsPerDay (total starts) Possiblyb Possiblyb Possiblyb startsSourceTypeFraction No No No startsMonthAdjust Yes Yes Yes startsHourFraction Yes Yes Yes StartsPerVehicle Possiblyb Possiblyb Possiblyb Starts (total starts in domain) No No No aOnly for a project where a vehicle population can be clearly defined and instrumented in a representative way (e.g., on-port activity). bRequires a sample representative of a clearly defined vehicle population and an external estimate of total vehicle population. Table 6.12. Applications of data sources for estimation of local starts and soak times. Example: Instrumented Heavy-Duty Diesel Vehicle Survey in California Researchers in California recruited and instrumented 100 heavy-duty diesel trucks and buses to examine operational and emissions patterns. Trucks were recruited from 20 vocations in four cate- gories: long haul, short haul, pick-up and delivery, and service-oriented. Vehicles were instrumented with a combined GPS/ECU data logger for at least 1 month. Recruitment was done through project team, consultant, and public agency contacts. The research team applied a variety of quality assurance procedures, such as comparing GPS and ECU speeds. The research team compared parameters such as operating time, distance, and speed; operating mode fractions; drive cycles; and start patterns. Significant differences were found across vocations, for in-state versus out-of-state operation, and for urban versus rural conditions. Source: Boriboonsomsin et al., 2017.

Starts and Soak Times 41 The cost of an instrumented vehicle study will depend on the sample size. The required sample size depends on the data of interest and also will be constrained by the available budget. Greater specificity in sampling criteria increases the recruitment effort. A small sample may be sufficient for measuring traffic conditions on a facility, where vehicles tend to move in simi- lar patterns. Obtaining drive cycles may require more vehicles, because of variation in vehicle power, vocation, and drivers. Trip start and hotelling data vary more with individual behavior and vocation, requiring a larger sample still. A recent California study aimed to gather data on as many vocations as possible within the available budget, so five trucks were selected per vocation for 20 vocations (Boriboonsomsin et al., 2017). The study team felt that five trucks would be enough to reveal major trends in activity by vocation type. Level of Data-Processing Effort Field data collection will require a high level of effort due to the need to recruit participants, implement the instrumentation, and process all data, including quality control. Recruitment of participants also may require financial incentives. Data-Processing Steps The steps for an instrumented vehicle study include the following: 1. Determine instrumentation, develop a sampling and recruitment plan focused on the target population of trucks, and identify candidate vehicles. 2. Recruit sample. 3. Install loggers. 4. Gather data centrally or manually. 5. Conduct quality control and analyze data. The MOVES input easiest to generate from instrumented vehicle data is starts per vehicle (see Table 6.8), which can be estimated on a sample-wide basis. MOVES’ other start inputs are focused on areawide start activity, which instrumented data (and even telematics data, yet) cannot estimate. For each source type, the MOVES analyst could multiply starts per vehicle per day by total vehicle population (as developed for other MOVES inputs) to estimate total starts per day. However, this would require an estimate of starts per vehicle for every vehicle type in the modeling domain (as would updating the start source type fraction shown in Table 6.5). OMDs (see Table 6.10) can be generated based on the sample of vehicles, regardless of the size of the sample. Generating an OMD requires compiling the distribution of time before all starts in a given hour, for the entire sample (the OMD input varies by hour). Monthly and hourly adjustments (see Table 6.6 and Table 6.7) could be determined from instrumented vehicle data, but again, would be specific to the vehicle types instrumented, and MOVES inputs currently do not provide for these adjustments by source type. Applicability and Limitations Compared to acquiring telematics data, the main limitation of collecting field data is sample size. It is difficult to recruit owners, and financial incentives are usually required. If data are not relayed automatically, manual collection is cumbersome, and errors will not be detected until after the device is removed from the vehicle. 6.4.2 Third-Party Data Logging/Telematics (GPS/ECU) Source, Availability, and Cost GPS-based data on truck movements can be acquired from multiple sources. Start and soak time patterns can be inferred from GPS data by assuming that trucks that stopped for a time

42 Guide to Truck Activity Data for Emissions Modeling longer than some threshold represent an engine-off condition. However, it is preferable to have ECU data so that actual cases of engine on/off can be identified and also because this is likely to provide more accurate locational and movement data than GPS alone. The ECU data also must be linked with GPS coordinates if the activity is to be linked with a specific project or study area. The NCHRP Project 08-101 team identified only one provider, Vnomics, which was able to provide ECU data linked with GPS data. While this is likely to change in the future, analysis of sample data from this provider gives insights into the opportunities and potential limitations of this data source. As described in Case Study #6 (see Appendix F, available in NCHRP Web- Only Document 261), the NCHRP Project 08-101 team was able to purchase a sample of vehicle movements occurring within a specified county. To simplify the data purchase request, the team specified a bounding box of latitude/longitude coordinates that encompassed the county and obtained data for all the trips with any data point inside the bounding box. The team then performed its own spatial queries to isolate activity within the county boundaries. Purchase costs typically are negotiated on a case-by-case basis and cannot typically be dis- closed due to contractual limitations with the provider. The scope and costs of these data vary widely, as they are custom data pulls for data providers, and cost structures are not yet in place to support emission inventory development. For this research, the NCHRP Project 08-101 team was able to acquire GPS and GPS/ECU data sets from multiple providers for a total cost of less than $50,000. The data purchases included GPS-based data for three counties; GPS/ECU data for 3 county-months; and a national sample of GPS/ECU data for 300 trucks for 1 month. The sample size will be determined by how many trucks are in the fleets from which the pro- vider collects data. In the example discussed here, there were 443 trucks in a 2-month Denver County, Colorado, sample and 52 trucks in a 1-month Norfolk County, Virginia, sample. All the trucks were Class 7 (26,000–33,000 lb) or Class 8 (> 33,000 lb), which are equivalent to the heavy- duty category. These trucks were from fleets that operate a less-than-truckload model with home domiciles in the respective areas. Most of the truck operation was pickups and deliveries within the area, which can be considered short haul in the context of MOVES. Level of Data-Processing Effort Analysis of a local sample of GPS/ECU data is likely to require significant effort due to the large data files that are involved and multiple processing steps needed. Data-Processing Steps The raw data in the sample data set included a number of data files containing second- by-second time stamp, latitude, longitude, vehicle speed, accelerator pedal position, fuel rate, engine revolutions per minute (RPM), total distance (vehicle odometer), and reference torque. Also included was the make, model, and year of the vehicle. Vnomics’ data logger starts recording data after the key is switched on (key-on event), and it stops recording data after the key is switched off (key-off event). This entire period makes up a data file. On the other hand, a trip can be defined as the period from when the engine is turned on (engine-on event) to when the engine is turned off (engine-off event), as shown in Figure 6.1. Figure 6.1. Illustration of events in truck activity.

Starts and Soak Times 43 Thus, one data file may consist of more than one trip if the engine is turned off and then on without a key-off event. On the other hand, a data file may not consist of any trip if the key is switched on and then off without an engine-on event. Since the recorded data include engine speed, engine-on and engine-off events can be identified. Due to possible noise in the engine speed data, an engine speed of 300 RPM is used as the threshold for determining whether the engine is running. This threshold is consistent with the one used in a previous study (Boriboonsomsin et al., 2017). The following steps can be undertaken to analyze this data set: 1. Remove the data files where one or more of the following essential data fields is missing— time stamp, latitude, longitude, vehicle speed, and RPM. 2. Remove data points located outside the domain boundary (if the acquired data set covers a larger area than the MOVES modeling domain). 3. Identify trip starts and ends based on an engine speed threshold of 300 RPM. Compile all the trips into a trip table containing information about vehicle ID, trip ID, trip duration, trip distance, trip average speed, start and end times, start and end locations, and day of week. 4. Calculate number of vehicles active in domain (i.e., at least one trip record) by time period as the vehicle population. 5. Using the trip table, compile a soak table containing information on vehicle ID, trip ID, soak start and end times, soak location, and day of week. 6. Analyze the distributions of number of starts per vehicle by hour of day. 7. Analyze the distributions of start operating mode (i.e., number of starts by soak period) by hour of day. Applicability and Limitations The sample data sets analyzed for this project, as well as any similar samples obtained by local analysts, will only include trucks in the fleets that are customers of the vendor. These trucks may or may not be representative of the entire fleet of heavy-duty vehicles in the geographic area of analysis. The data set also may be representative only of a limited time period. However, the truck samples in Denver did not exhibit significant seasonal differences in start and soak activity patterns between the 2 months studied (January and May 2017). Finally, the measure- ment of a resident vehicle population, based on active vehicles in the observation time period, may be inconsistent with the MOVES definition of a vehicle population, so the starts per vehicle metric may not be comparable. The sample data only included heavy trucks, most of which are likely used in short- haul operation. Start patterns may not be representative of smaller trucks or of long-haul operation. Specific vocations of the sample are not known, and activity has been found to vary substantially by vocation. Some comparisons of the Vnomics data with StreetLight data and MOVES defaults are also made in the “applicability and limitations” subsection of Section 6.4.3. 6.4.3 Third-Party Data Logging/Telematics (GPS Only) Source, Availability, and Cost The NCHRP Project 08-101 team acquired sample data for three counties from StreetLight Data, Inc. (StreetLight): Fairfax and Norfolk Counties in Virginia and Denver County in Colorado. StreetLight compiles data on vehicle locations gathered from GPS on connected vehicles, mobile phones, and commercial fleet management systems to produce trip metrics at a very detailed geographic resolution. StreetLight does not provide raw trip data, but rather worked with the project team to produce aggregate statistics that could be processed

44 Guide to Truck Activity Data for Emissions Modeling into MOVES inputs for starts per vehicle and start operating mode (soak time) distribution. The data set distinguishes medium-duty and heavy-duty trucks. Medium-duty trucks were assumed to correspond to the single-unit truck type in MOVES, and heavy-duty trucks to the combination truck type. No distinction could be made between short-haul and long-haul operation. GPS data for individual trucks could potentially be obtained from other vendors in the future. Case Study #6 (see Appendix F, available in NCHRP Web-Only Document 261) evaluates this possibility by comparing start estimates produced using Vnomics GPS data only with estimates produced from the same truck fleets using GPS and ECU data. While purchase costs typically are negotiated on a case-by-case basis and cannot be disclosed due to contractual limitations with the provider, the NCHRP Project 08-101 team was able to acquire GPS and GPS/ECU data sets from multiple providers for a total cost of less than $50,000. The data purchases included GPS-based data for three counties; GPS/ECU data for 3 county- months; and a national sample of GPS/ECU data for 300 trucks for 1 month. Level of Data-Processing Effort The analysis of the sample StreetLight data required modest effort since the vendor performed the data processing. A moderate to significant effort would be required if raw data were pur- chased and processed by the analyst. Data-Processing Steps StreetLight aggregated GPS trip data into starts per vehicle and total counts by “dwell time” that were roughly equivalent to MOVES soak bins (e.g., start OMD). These statistics include underlying data on “trip starts” and “trip ends”; however, it is important to note that the data provided are not absolute trips or absolute vehicles. StreetLight uses an index to normalize its database to account for the constant shifts in the number of vehicles supplying data over time. An issue introduced with the use of telematics data is the need to define what constitutes a “start” and a “vehicle” for estimation of starts per vehicle. This is not always straightforward when looking at a fixed geographic area (e.g., a county) because GPS-based telematics data are tracking vehicles that migrate across county lines and may just pass through. For a given county, the NCHRP Project 08-101 team and StreetLight defined “start” as any new trip that originated in the county, regardless of whether the trip stayed in the county or not. “Vehicle” was defined on an hourly basis, as any vehicle whose first start within the given hour was within the county or any vehicle 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 6.2 provides a schematic and detailed description of how the starts and vehicles were defined and how the starts per vehicle metric was calculated. Two MOVES input tables were generated from the StreetLight data: StartsPerVehicle, and ImportStartsOpModeDistribution. Both tables have key fields of source type and hourDayID, which combine 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 Street- Light was configured to match these MOVES inputs as directly as possible. Both tables required mapping StreetLight categories of medium duty and heavy duty into MOVES source types. The most straightforward approach was to assign medium duty to single-unit trucks (52 and 53), and assign heavy duty to combination trucks (61 and 62). The starts per vehicle data could then be imported directly into StartsPerVehicle.

Starts and Soak Times 45 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 OMDs as closely as possible. The main difference was that the data for the shortest soak period could only be provided for periods of 0 to 15 minutes, instead of 0 to 6 minutes as in MOVES. StreetLight bins of 0 to 15 minutes and 15 to 30 minutes were combined to approximate the 6- to 30-minute bin in MOVES. The minimum time before a trip was set at 5 minutes to help ensure that the next movement is a trip start, rather than a brief stop (e.g., at a stoplight). 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. If an analyst obtained coordinate data for individual trucks, the processing steps would be similar to the steps required to process collected field data, as described in Section 6.4.1. The final (optional) step in using GPS-only data to estimate start inputs is to adjust the estimated number of starts and soak distributions for the difference between estimates based purely on vehicle activity/movement and estimates based on actual engine-on/-off events to account for the fact that the engine will remain on during some stopped events. The sample Vnomics GPS/ECU data from Denver and Norfolk suggest that if GPS-only data is being used, (blue) (blue) (yellow) (blue) (green)(blue) (red) (red) Figure 6.2. Trip metric definitions and example.

46 Guide to Truck Activity Data for Emissions Modeling the number of estimated starts should be increased by about 19 percent and the fraction of start operating mode with soak time less than 6 minutes increased by about 65 percent, to account for the undercounting of starts when using only a GPS unit that records key-on events rather than engine-on events. The increased fraction in this start operating mode should then decrease the fraction of the other start operating modes proportionally. The ratio observed here may be specific to the data set and could depend on other factors, such as the vocational makeup of the instrumented fleet. Applicability and Limitations The GPS-based data provided by StreetLight are likely to be a reasonable source of distri- butional information (distribution of starts by hour and potentially by soak time, accounting for missing short-duration soak time). These distributions will be less affected by the defini- tion of the vehicle population or by differences between vehicle activity and engine activity patterns. These distributions are more likely to be representative of short-haul activity than they are of long-haul activity in which vehicles spend only a small part of the day in the analysis domain. Confidence in the ability of the StreetLight data to represent the total number of starts per vehicle is lower. This is due not only to engine activity that does not correspond precisely with vehicle activity, but also to the problem of defining a vehicle “population” on which to base the number of starts because of the way this particular vendor gathers and reports data. The definition used in this example (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) can produce a more accurate measure of total starts per vehicle per hour or day, but not necessarily for small domain areas since the vehicles are moving among domains. The sample results shown in the case studies (see Appendices A through G, available in NCHRP Web-Only Document 261) may or may not be transferable. Within a given data source, broad patterns across sample counties were similar; however, the patterns of starts by hour dif- fered greatly among the StreetLight, Vnomics, and MOVES data (see Figure 6.3). Based on the StreetLight sample data, start activity patterns for Fairfax and Norfolk Counties were generally similar within a given data set, 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. This was incon- sistent with the Vnomics data, however, which showed higher numbers of overnight starts in Denver than in Norfolk (note that only 52 trucks were sampled in Norfolk and no starts were observed between 1:00 a.m. and 7:00 a.m.). For combination trucks, the Vnomics data showed a much lower number of starts per truck per day than the MOVES defaults, which in turn showed lower numbers than estimated using the StreetLight data (see Table 6.13). Even when Street- Light starts are reduced to account for the difference between GPS-based and ECU-based start estimates (as observed in Vnomics data), total StreetLight starts are still significantly higher than MOVES defaults or starts observed in Vnomics data. It cannot be determined with certainty how much the variations are due to differences in data definitions, differences in the fleets sampled, location-specific differences, or possibly even changes in operational characteristics since the MOVES data were collected. Large variability in trip activity between the StreetLight, Vnomics, and MOVES data sets may reflect the wide varia- tion of truck vocations in the underlying truck samples, in addition to different measurement and definitional methods. Package delivery trucks may have dozens of starts per day, while long- haul trucks have only a few starts. It is therefore critical to understand the makeup of the truck fleet on which data is collected, and, for future data collection efforts, to consider truck vocation as a stratifying variable that is deliberately accounted for.

Starts and Soak Times 47 aWeekend results are shown in Case Study #6 (see Appendix F, available in NCHRP Web-Only Document 261). Note that almost no weekend starts were observed in the Vnomics data. bWd=weekday 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Vnomics/Denver/Wdb StreetLight/Denver/Wd MOVES_61/Wd Vnomics/Norfolk/Wd StreetLight/Norfolk/Wd Starts per Truck per Hour Hour of Day (1 = midnight - 00:00:59) Figure 6.3. Starts per truck (heavy duty) by hour of day during weekday.a Data Source Day of Week Single Unit Short Haul, 52 Single Unit Long Haul, 53 Combination Short Haul, 61 Combination Long Haul, 62 MOVES2014 dataa Weekday 8.1–8.4 5.1–5.2 7.4–7.9 5.3–5.7 U.S. EPA, 2016b 7.0 4.3 5.9 4.3 StreetLightc—Denver 2.6 16.0 StreetLight—Fairfax 2.8 21.0 StreetLight—Norfolk 2.5 20.5 Vnomics—Denver 3.6 Vnomics—Norfolk 2.0 MOVES2014 data Weekend 1.9–2.0 2.0 1.9–2.0 1.9 U.S. EPA (2016) 1.3 1.3 1.2 1.3 StreetLight—Denver 1.1 16.3 StreetLight—Fairfax 1.1 20.8 StreetLight—Norfolk 0.9 21.2 Vnomics—Denver N/A Vnomics—Norfolk N/A aMOVES2014 data computed by Eastern Research Group, Inc. Ranges encompass slightly different values for Denver, Fairfax, and Norfolk Counties. bTable 12-8, “Starts per day by source type.” cStreetLight data are not specific to short-haul trucks. Table 6.13. Estimated starts per day.

48 Guide to Truck Activity Data for Emissions Modeling 6.4.4 Other Resources Case Study #3 (see Appendix C, available in NCHRP Web-Only Document 261) describes the estimation of starts data using StreetLight GPS data. Case Study #6 (see Appendix F, available in NCHRP Web-Only Document 261) describes the estimation of starts data using Vnomics GPS/ECU data. One of the MS Excel files that supplements this guide (NCHRP08-101_Data_CS3+4+6_Starts- Idle-Hotelling.xlsx) contains the StreetLight and Vnomics data used in the case studies, as well as comparison with MOVES data. This file is available on the NCHRP Research Report 909 web page (http://www.trb.org/Main/Blurbs/178921.aspx).