MOVES-Matrix for high-performance on-road energy and running emission rate modeling applications

ABSTRACT

The MOVES model was developed by the U.S. Environmental Protection Agency (U.S. EPA) to estimate emissions from on-road mobile sources and nonroad sources in the United States. Coupling high-resolution on-road vehicle activity data with appropriate MOVES emission rates further advances research efforts designed to assess the environmental impacts of transportation design and operation strategies. However, the complicated MOVES interface and slow performance makes it difficult to assess large, regional scale transportation networks and to undertake analyses of large-scale systems that are dynamic in nature. The MOVES-Matrix system develops an initial Large Matrix of MOVES outputs by running MOVES 146,853 times on the PACE high performance computing cluster to generate more than 90 billion emission rates to populate the matrix for a single area with one fuel regime and one inspection and maintenance program. A total of 117 such Large Matrices would be needed for the entire United States. The MOVES-Matrix system developed can be used to conduct the emissions modeling 200-times faster than using MOVES. The hypothetical case study shows that MOVES-Matrix is able to generate the exact same emission results as the MOVES model to ensure the validity for regulatory analysis. The resulting matrix allows users to link emission rates to big data projects and to evaluate changes in emissions for dynamic transportation systems in near-real-time. MOVES-Matrix does not currently estimate emissions from starts, hoteling or evaporative emissions, and the research team is working on MOVES-Matrix version 2 that supports incorporating off-network modeling.

Implications: MOVES-Matrix should be of interest to a broad readership including those interested in vehicle emission modeling, near-road air quality modeling, transportation conformity analysis. The paper should also interest engineers who are involved in transportation regulatory and conformity analysis, state implementation plan, and who are seeking an efficient way of conducting regulatory emission modeling and air quality analysis in the United States.

Introduction

In the United States, the MOtor Vehicle Emission Simulator (MOVES) model (U.S. EPA 2015) provides significantly improved emission rates compared to the older MOBILE series of models (U.S. EPA 2016a), primarily because MOVES emission rates are more modal in nature, better representing emissions as a function of instantaneous (1 Hz) speed and acceleration. The project-level modeling with MOVES allows the highest resolution of input data. A variety of new fleet activity data are now available for use in emissions modeling in project-level, such as streaming machine vision data (Liu et al. 2015), smartphone location tracking (CRC 2017a; Hilpert, Thoroe, and Schumann 2011), and traffic simulation modeling (Anya et al. 2014; Talbot et al. 2013). Coupling MOVES emission rates with various sources of big data for vehicle activity can further advance research efforts designed to assess the environmental impacts of transportation design and operation strategies. Hot spot analysis and near-road dispersion modeling for environmental impact assessment also benefit from the use of more accurate vehicle activity data in both time and space aspects, and the application of high-resolution emission rates for on-road driving conditions. However, the MOVES interface is complicated, and the structure of input variables and algorithms involved in running MOVES to assess operational improvements makes such analyses cumbersome and time-consuming.

The MOVES interface makes it difficult to assess complicated transportation networks and to undertake analyses of large-scale systems that are dynamic in nature (CRC 2017b). For example, the Atlanta Regional Commission (ARC) Travel Demand Model network includes 74,500 roadway segment links. It is nearly impossible to perform emissions modeling for a dynamic network of this size using individual MOVES emission rates for each link when fleet composition and on-road operating conditions change dynamically over the course of a day. On a typical personal computer (PC), depending on the pollutants and processes to be modeled, MOVES requires around 10–30 s to process emissions for one link for a unique fleet and operating condition. To obtain the composite emission rates for 1,000 roadway links in Atlanta, where the fleet composition and operating conditions vary every hour on every road segment, and where temperatures and humidity values vary by hour of day and month, and for the three Atlanta fuels (summer, winter, and transition), nearly 32 million individual MOVES runs would be required, which would take ten years to run on a typical PC, considering 1,000 road segments, with operations of each hour across 24 h, in 21 temperature bins scenarios (10–110 degrees Fahrenheit in 5-degree bins), 21 humidity bins scenarios (0–100% in 5% bins), and three fuels supply scenarios (summer, winter, and transition fuel supply), which sums to 31,752,000 individual MOVES runs. A lot of shortcuts can be taken to reduce the number of runs required. For example, many runs yield the exact same emission output across a certain temperature or humidity ranges as they are insensitive to several pollutant and processes. In most cases, it is also unrealistic to run emissions with 10 degrees during the summer. However, modeling every operating condition described above is still impractical. A high-performance modeling approach is needed to assess large-scale dynamic networks. Yet, at the same time, regulations require the latest approved regulatory model (i.e., MOVES 2014a) be used in all transportation and air quality planning and assessment work (U.S. EPA 2015).

Previous studies have focused on optimizing model run speed for regulatory emissions models. For example, Guensler et al. (2004) ran MOBILE6, the predecessor of MOVES model, tens-of-thousands of times to generate a matrix of emission rates (known as MOBILE-Matrix) by road class, fleet composition, fuel, inspection and maintenance (I/M), temperature, etc., for Georgia, and applied emission rates in conformity analysis and CALINE4 dispersion model routines. The emission matrix developed for MOBILE6 facilitated rapid analysis via scripts (Guensler et al. 2004). With the release of the more advanced MOVES model as replacement of MOBILE series models, Liu and Frey (2013) developed a simplified MOVES model called MOVES-Lite, based on the proportion of operating mode bin as the cycle adjustment factor, and the resulting discrepancies were within 5% of MOVES outputs. There was also an effort to enable computation of MOVES on the Cloud, while this becomes complicated as the cloud environment changed (CRC 2015).

To improve modeling speed, but at the same time ensure that regulatory requirements for use of MOVES are met, the research team developed MOVES-Matrix. The MOVES model was run hundreds of thousands of times to generate a matrix for all combinations of MOVES input variables (i.e., MOVES runspecs and user csv supplied data, see U.S. EPA 2015). This concept of iterative model processing and matrix generation was applied with the MOBILE model many years ago (Guensler et al. 2004; Guensler and Leonard 1995; Guensler et al. 2000). The MOVES-Matrix emission rates described in this paper can be queried for any analytical purpose that can be conducted by MOVES, without ever having to launch MOVES or transfer MOVES modeling output files into the analyses. Obtaining regulatory approval for any modeling approach is predicated on the approval of the U.S. Environmental Protection Agency (U.S. EPA), which requires that the latest MOVES model must be employed. The MOVES-Matrix modeling approach is demonstrated to yield exactly the same emission rates as MOVES when run for the same conditions. As will be demonstrated in this paper, because MOVES-Matrix is the comprehensive set of rate outputs from the MOVES model, the application of MOVES-Matrix emission rates yields exactly the same results as MOVES.

MOVES and MOVES-Matrix

MOVES background

Historically, regulatory emissions models, such as the MOBILE series of models, defined emissions as a function of average speed, essentially irrespective of acceleration (Carlson and Austin 1997). In the MOVES model, emissions are now defined as a function of speed and vehicle-specific power (VSP) for light-duty vehicles, or speed and scaled-tractive power (STP) for heavy-duty vehicles, which better reflects the impact of acceleration on engine load and work (U.S. EPA 2016b). These speed-power combinations are broken into bins called operating modes. The U.S. EPA’s MOVES model employs a “binning” approach in modeling emissions for different on-road fleets and on-road operating conditions, where activity that falls into the same operating mode bin receives the same emission rate for a given vehicle type and environmental condition. In MOVES, driving cycles (speed-acceleration activity) can be decomposed into operating mode bins and modeled as a function of time spent operating in each bin. This design enables MOVES to provide common emission rates for all modeling scales (macroscale, mesoscale, and microscale) (U.S. EPA 2016b). MOVES requires refined input data, including meteorology, calendar year, fuel type, I/M program elements, traffic volume, operating speed, fleet age distribution and vehicle type distribution (U.S. EPA 2015). Base emission rates for specific operating modes are also adjusted in the model to account for the impacts of temperature, humidity, fuel composition, vehicle aging, and other factors on the emission rates. Figure 1 presents the data processing flow of the MOVES model.

Figure 1. Data processing overview of MOVES.

Because emissions are a function of the energy required to move the vehicle, which depends upon power demand, vehicle weight, and on-road operating conditions, the MOVES model employs surrogates for engine load: vehicle-specific power (VSP) for light-duty vehicles, and scaled tractive power (STP) for heavy-duty vehicles. VSP and STP are a function of vehicle speed, acceleration, and vehicle mass. Second-by-second VSP and STP are calculated as shown in Equation (1) (U.S. EPA 2016b):

(1) $\begin{array}{rl}& \mathrm{V}\mathrm{S}\mathrm{P}{\left(\mathrm{S}\mathrm{T}\mathrm{P}\right)}_{\mathrm{t}}\left(\mathrm{k}\mathrm{W}/\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{c}\mathrm{t}\mathrm{o}\mathrm{n}\right)\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\\ & =\left(\frac{\mathrm{A}}{\mathrm{M}}\right){\mathrm{v}}_{\mathrm{t}}+\left(\frac{\mathrm{B}}{\mathrm{M}}\right){{\mathrm{v}}_{\mathrm{t}}}^2+\left(\frac{\mathrm{C}}{\mathrm{M}}\right){{\mathrm{v}}_{\mathrm{t}}}^3\\ & +\left(\frac{\mathrm{m}}{\mathrm{M}}\right)\left({\mathrm{a}}_{\mathrm{t}}+\mathrm{g}\ast sin{\theta }_{\mathrm{t}}\right){\mathrm{v}}_{\mathrm{t}}\end{array}$(1)

Where${\mathrm{v}}_{\mathrm{t}}$is the velocity at time $\mathrm{t}$ (m/sec), ${\mathrm{a}}_{\mathrm{t}}$is acceleration at time $\mathrm{t}$ (m/sec²), ${\theta }_{\mathrm{t}}$ is road grade (radians or degrees), $\mathrm{g}$ is acceleration due to gravity (9.81 m/sec²), $\mathrm{m}$ is vehicle mass (metric tons), $\mathrm{A}$ is rolling resistance (kW-sec/m), $\mathrm{B}$ is rotating resistance (kW-sec²/m²), $\mathrm{C}$ is aerodynamic drag (kW-sec³/m³), $\mathrm{M}$ in $\mathrm{V}\mathrm{S}\mathrm{P}$ is the fixed mass factor for the source type (metric ton), $\mathrm{m}=\mathrm{M}$ for $\mathrm{V}\mathrm{S}\mathrm{P}$ calculations, and $\mathrm{M}$ in $\mathrm{S}\mathrm{T}\mathrm{P}$ is the scaling factor to scale $\mathrm{S}\mathrm{T}\mathrm{P}$ ranges to within the same range as $\mathrm{V}\mathrm{S}\mathrm{P}$ (metric ton).

The MOVES model uses a binning approach in emissions modeling. VSP and STP bins are established for three types of operations: braking, idle, and cruise-acceleration. Bins for cruise-acceleration are further separated into three average speed groups (0–25 mph, 25–50 mph, 50+ mph), and then into VSP ranges within each average speed group. Higher VSP and STP values within specific operating speed ranges are linked with higher fuel consumption, CO₂ emission rates, and criteria pollutant emission rates. Figure 2 describes and defines each MOVES operating mode bin (opMode bin) by speed and VSP ranges, and presents an example of the MOVES CO₂ emission rates for model year (MY) 2016 passenger truck in each operating mode bin. High speeds, moderate accelerations at high speed, and quick accelerations at moderate or high speed that push on-road activity into higher VSP bins, which then use higher fuel consumption and emission rates in energy and emissions calculations.

Figure 2. Definition of running operating mode bin and example CO₂ emission rates for passenger trucks (model year 2016).

MOVES-Matrix conceptual approach

Because emissions are a complex function of many locally dependent variables, and because MOVES integrates a number of aggregation functions for use in emission estimation at state and county levels, the interface is complex and requires numerous inputs to properly characterize any specific emission scenario modeled by a user. Extensive labor is required to prepare MOVES input files. In addition, running MOVES is time-consuming, because emission calculations begin with base emission rates, which will be internally adjusted by various correction factors such as temperature, humidity, fuel property, etc. This also makes MOVES difficult to use for large-scale transportation networks that experience dynamic changes in on-road fleet composition and operating conditions that affect corrections factors during the day.

MOVES-Matrix is a gigantic lookup table composed of the outputs from a tremendous number of MOVES2014a model runs. The MOVES-Matrix development process is to run MOVES across all variables that affect output emission rates, where each MOVES run yields a pollutant emission rate for single vehicle source type, model year, vehicle fuel type, on-road operating condition (average speed and road type, or a single on-road VSP/STP operating mode bin), calendar year, temperature, humidity condition, and other applicable regional regulatory parameters (fuel properties, I/M program characteristics). After conducting 146,853 runs, the resulting MOVES-Matrix can be queried to obtain the exact same emission rates that would be obtained for any MOVES model run, without ever having to launch MOVES again, or transfer MOVES outputs into the analyses.

Figure 3 provides an overview of MOVES-Matrix application process. Users first identify the subset of the MOVES-Matrix needed, by specifying calendar year, fuel month, and meteorology data. Then, the user can access each cell that contains an emission rate for a specific vehicle type and model year from MOVES-Matrix and weight each emission rate by on-road activity to reassemble the fleet emission rate. Because the weighting process is exactly the same as used in MOVES to generate a fleet composite emission rate for a link, the MOVES-Matrix process yields the exact same emission rates as a direct MOVES run, but in a fraction of the time to use the MOVES run. Because each MOVES run already performed the complex emission rate calculations and adjustments for temperature, humidity, fuel composition, I/M program, etc., MOVES-Matrix exclusively contains the resulting emission rates. For the user, applying MOVES-Matrix is significantly faster than running MOVES. The MOVES-Matrix emission rate assembly process is efficient enough to enable large scale and real-time emission estimation.

Figure 3. MOVES-Matrix conceptual flow.

Moves-Matrix generation based on super-computing system

To develop the MOVES-Matrix emission rate database for each region of interest, a total of 146,853 (21 years × 3 months × 111 temperature bins × 21 humidity bins) MOVES runs were prepared by the research team (calendar years 2010–2025, 2030, 2035, 2040, 2045, 2050; winter, summer, and transition fuel months, 10 degree-110 degree temperatures in 1 degree intervals, 0–100% relative humidity in 5% intervals). Three emission rate processing steps are performed: (1) develop the set of input files and RunSpecs files to support MOVES runs across all relevant input variables; (2) run the MOVES input files in an advanced computing cluster to obtain multi-dimensional emission rates outputs; and, (3) design algorithms and a MOVES-Matrix user interface that can be used to pull applicable emission rates from the matrix for use in regional emissions inventory modeling, traffic simulation modeling, corridor-monitored second-by-second activity analysis, and microscale dispersion modeling. In addition to the csv input tables, each MOVES modeling run employs an import xml file and an execution mrs file. Table 1 outlines the model inputs used to create MOVES-Matrix.

Table 1. Content of input file for one MOVES run.

Input	Description
Link	23 links with each assigned operating mode bin Set volume for each link as 13(source types)×31(model years) = 403
Age distribution	Uniform age distribution (1/31 for each age group from age 0 to 30 years) for each source type
Source type	Uniform source type distribution (1/13 for each source type) for each link
I/M strategy	Default from MOVES, determined by calendar year (CY) and region
Fuel supply	Default from MOVES, determined by calendar year, month and region: November to March: winter fuel April and October: transition fuel May to September: summer fuel
Fuel formulation	Default from MOVES, determined by fuel supply
Fuel usage fraction	Default from MOVES
Alternative fuel technology (AVFT)	Default from MOVES
Meteorology	Temperature: 0–110° F with 1° F-bin interval, 111 bins in total Humidity: 0%–100% with 5%-bin interval, 21 bins in total
Operating mode distribution	Single 100% fraction of a specific operating mode bin for each link.
Year	Each year in 2010–2025, 5-year intervals in 2030–2050, input in xml and mrs file

The research team has priority access to the Partnership for an Advanced Computing Environment (PACE) high performance computing (HPC) cluster. PACE is a collaboration between Georgia Tech faculty and the Office of Information Technology and was established for the primary purpose of providing an environment for distributed high-performance computing. Participating researchers benefit from the large-scale computing and storage infrastructure, which is organized in the form of shared queues and distributed computational runs. Dedicated technical services are provided to manage the hardware and software infrastructure for the cluster. Users submit jobs to PACE from select head nodes and the cluster assigns jobs to available cores (or CPU). On its largest shared queue, PACE manages around 35,000 cores, with 90 terabytes of memory, 2 petabyte of online commodity storage, and nearly 300 terabytes of high-performance scratch storage.

When a MOVES job is launched on a cluster machine, the scripts first install MOVES on the machine by unzipping the MOVES source files onto the disk. The script then proceeds with installing a thin version of MYSQL server, by unzipping its files onto the disk, and starts the SQL server on an available port. MOVES command line java processes are then launched to create input and output database files, respectively. The output files are zipped and stored on PACE persistent storage. More details on launching MOVES in PACE can be found in Liu et al. (Liu et al. 2016). Depending on the application and research scope, MOVES-Matrix was prepared with different sizes (resolution). Table 2 lists two typical sizes of MOVES-Matrix (i.e., with temperature of 1-degree interval and 5-degree interval), and corresponding amounts of time to generate MOVES-Matrix using PACE. It takes 18 days to generate emission rates of MOVES-Matrix with 1-degree temperature resolution. The preparation time for MOVES-Matrix with temperature of 5-degree interval can be as quick as 3.5 days.

Table 2. Time for generating MOVES-Matrix*.

MOVES-Matrix Size	Time to prepare
30,429 files, 121 GB 21 calendar years × 3 fuel months × 23 temperature bins (0–110 degrees with 5-degree interval) × 21 humidity bins (0–100% with 5% interval)	3.5 days
146,853 files, 605 GB 21 calendar years × 3 fuel months × 111 temperature bins (0-110 degrees with 1-degree interval) × 21 humidity bins (0–100% with 5% interval)	18 days

Notes. *Energy rate, greenhouse gas emissions (CO₂, GHG, CH₄, N₂O), criteria pollutants (NOx, THC, VOC, CO, SO₂, PM₁₀, PM_2.5)

Energy and emissions modeling using Moves-Matrix

MOVES-Matrix is generated based upon 146,853 MOVES runs (21 calendar year × 3 fuel months × 111 temperature bins × 21 humidity bins = 146,853) for Atlanta region. The scripts repeated MOVES runs where each parameter in Table 1 was incremented. To support varied levels of detail for on-road fleet composition and operating conditions that may be available to modelers, the research team prepared three MOVES-Matrix versions. (1) Average Speed and Facility Type Matrix (Speed-Matrix): In the speed matrix version, users provide average speed and road type (arterial versus freeway) as inputs. The emission results are a function of internal MOVES default driving cycles. The fleet is specified by source type and model year, and MOVES default fuel type distributions (% of gasoline and diesel vehicles) are applied by source type. (2) Operating Mode Matrix (OpMode-Matrix): In the operating mode bin matrix version, users need to provide a driving schedule, or operating mode distribution for on-road operations. The fleet is specified by source type and model year, with MOVES default fuel type distribution (% of gasoline and diesel vehicles) applied to each source type. (3) Matrix of Operating Mode and Fuel (OpMode&Fuel-Matrix): In the operating mode bin and fuel matrix version, users need to provide a driving cycle (second-by-second speed schedule), or operating mode distribution for on-road operations. The fleet is then specified by source type, model year, and fuel type.

The MOVES-Matrix application consists of three modules: (1) input files and RunSpecs files, (2) emission database, and (3) output. Input modules are created for each of the three versions of the modeling approach described above and emission databases are structured for query by average speed and facility or operating mode bin. In designing MOVES-Matrix, it was important to first assess model user preferences. Real-world applications of MOVES for emission inventory development or project-level conformity analyses currently use a variety of simplified approaches to limit the number of MOVES runs that will be required. For example, analysts often assume that fleet composition does not vary (using a default regional registration mix for model years and technology groups) with heavy-duty truck fractions quantified in specific percentages by road class (0% or 1% on certain local roads and arterials and 3% or 5% on certain freeways). Planning inventories may also assume a single temperature, humidity, and fuel supply. Every time another transportation scenario needs to be assessed, a new set of emission rates applied with new meteorology or fuel scenario generally needs to be developed from MOVES and connected with the activity data. Table 3 shows the key fields to identify emission rate from three versions of MOVES-Matrix.

Table 3. Key fields to identify emission rate in MOVES-Matrix.

MOVES-Matrix version	Key fields to identify Sub-Matrix	Key fields to identify emission rate in the Sub-Matrix
Average Speed and Facility Type Matrix (Speed-Matrix)	Calendar year, month, temperature, humidity	Vehicle source type, model year, pollutant type, facility type, speed
Operating Mode Matrix (OpMode-Matrix)	Calendar year, month, temperature, humidity	Vehicle source type, model year, pollutant type, operating mode bin
Matrix of Operating Mode and Fuel (OpMode&Fuel-Matrix)	Calendar year, month, temperature, humidity	Vehicle source type, model year, fuel type, pollutant type, operating mode bin

To support typical applications, in each region, the MOVES-Matrix emission database was grouped into 146,853 subset of matrices, with each subset storing emission rates for all source types, all source model years, all on-road operations (average speed bins or operating mode bins), for one specific calendar year, one fuel month, one temperature bin, one relative humidity bin, one fuel supply (by year and month), and one I/M strategy (by year). This way, a small portion of emission rates from subset of matrix can be extracted from the matrix based on the user-specified year, month, and meteorology inputs. This structure helps support emission control strategy analysis, given that users tended to assume a single temperature, humidity, and fuel, when exploring the impacts of strategies on traffic activity and emissions. Using a subset of matrix is significantly faster than extracting data from the full emission rate in MOVES-Matrix.

After the subset of MOVES-Matrix is identified and accessed, the emission rate processing algorithm is the same as used by MOVES in project-level modeling. The emission rates in the subset of matrix are connected to vehicle activity data through MOVES-Matrix algorithms. MOVES-Matrix weights the emission rates from individual source types to generate the composite emission rate. The weighting combines on-road vehicle activity, as defined by combined source type and model year distribution (newer vehicles typically represent a larger share of the on-road fleet than older vehicles) and the amount of on-road activity by operating mode bin or average speed bin to calculate a composite emission rate for each link. If second-by-second driving cycle is used to describe vehicle activity, VSP/STP for each second will be calculated first, and then converted to operating mode bin. Operating mode distribution is then calculated by aggregating operating mode bin of each second for each vehicle source type in each link. The emission rate weighting function is as follows in Equations (2) and (3).

(2) $\begin{array}{l}\text{Fleet}\text{ER}={\sum }_{\text{ST}}{\sum }_{\text{MY}}{\sum }_{\text{OM}\left(\text{V\&F}\right)}\text{ST},\%\\ \times \text{MY }{\%}_{\text{ST}}\times \text{OM}\left(\text{V\&F}\right),{\%}_{\text{ST},\text{MY}}\\ \times {\text{ER}}_{\text{ST},\text{MY},\text{OM}\left(\text{V\&F}\right)}\end{array}$(2)

(3) $\mathrm{T}\mathrm{E}\mathrm{M}=\mathrm{V}\mathrm{H}\mathrm{T}\left(\mathrm{o}\mathrm{r}\mathrm{V}\mathrm{M}\mathrm{T}\right)\times \mathrm{F}\mathrm{l}\mathrm{e}\mathrm{e}\mathrm{t}\mathrm{E}\mathrm{R}$(3)

where $\mathrm{F}\mathrm{l}\mathrm{e}\mathrm{e}\mathrm{t}\mathrm{E}\mathrm{R}$ is the fleet comprehensive emission rate (mass/mile or mass/hr), $\mathrm{T}\mathrm{E}\mathrm{M}$ is the total emissions, $\mathrm{V}\mathrm{H}\mathrm{T}$ is the vehicle hours traveled (used if $\mathrm{F}\mathrm{l}\mathrm{e}\mathrm{e}\mathrm{t}\mathrm{E}\mathrm{R}$ is in mass/hr), $\mathrm{V}\mathrm{M}\mathrm{T}$ is the vehicle miles traveled (used if $\mathrm{F}\mathrm{l}\mathrm{e}\mathrm{e}\mathrm{t}\mathrm{E}\mathrm{R}$ is in mass/mile), $\mathrm{S}\mathrm{T}$ is the vehicle source type, $\mathrm{M}\mathrm{Y}$ is the model year, $\mathrm{O}\mathrm{M}$ is the operating mode bin, $\mathrm{V}\mathrm{\&}\mathrm{F}$ is the average speed bin and facility type, $\mathrm{S}\mathrm{T}\mathrm{\%}$ is the proportion of one source type (from source type distribution input), $\mathrm{M}\mathrm{Y}{\mathrm{\%}}_{\mathrm{S}\mathrm{T}}$ is the proportion of one model year by one source type (from age distribution input), $\mathrm{O}\mathrm{M}\left(\mathrm{V}\mathrm{\&}\mathrm{F}\right){\mathrm{\%}}_{\mathrm{S}\mathrm{T},\mathrm{M}\mathrm{Y}}$ is the time proportion of one operating mode bin (or proportion of average speed bin and facility type) by one source type and one model year (from operating mode distribution input or drive schedule), and $\mathrm{E}{\mathrm{R}}_{\mathrm{S}\mathrm{T},\mathrm{M}\mathrm{Y},\mathrm{O}\mathrm{M}\left(\mathrm{V}\mathrm{\&}\mathrm{F}\right)}$ is the emission rate of one source type, model year, and operating mode bin.

Benefits of Moves-Matrix

Figure 4 compares MOVES with MOVES-Matrix in terms of overall working mechanisms. MOVES starts with a set of base emission rates, and these base emission rates are adjusted during each run before they are connected to activity data. Instead of base emission rates, MOVES-Matrix stores adjusted emission rates for all scenarios, and for the scenario of interest. MOVES-Matrix filters the emission rates for the specific scenario, rather than doing adjustment calculations.

Figure 4. MOVES vs. MOVES-Matrix working mechanism.

There are four design characteristics that contribute to the accuracy and fast processing speed of MOVES-Matrix: (1) MOVES-Matrix emission rates are obtained directly from MOVES runs carried out by the research team. There are no code modifications, no correction factors, and no approximations involved, which ensure that the emission results obtained from MOVES-Matrix are exactly the same as the MOVES model. (2) MOVES-Matrix allows users to assess impacts of changes in on-road operating conditions and on-road fleet composition. Rather that running MOVES repeatedly, MOVES-Matrix directly employs emission rates that have already been adjusted by fuel, meteorology and I/M strategy. Repeating MOVES calculations are not needed. The matrix structure also facilitates sensitivity analysis of MOVES algorithms. (3) In MOVES-Matrix, the emission rates database is pre-organized by calendar year, fuel specification, I/M program, temperature, and humidity. Hence, the emission rate is ready to apply to any specific scenarios of interest. This significantly increases the speed of the emission assignment process. 4) MOVES-Matrix is open source and collaborative. Python, Java, Perl, or any other regular language scripts can be used to link MOVES-Matrix emission rates with travel demand models, traffic simulation, monitored data, and dispersion models.

Moves-Matrix applications

Hypothetical case study: MOVES-Matrix performance test

To demonstrate the effectiveness and efficiency of MOVES-Matrix, the researchers developed a set of test runs to compare the performance of MOVES-Matrix with the MOVES batch mode for emission rate values and run speeds. Table 4 lists the iteration variables and increments for the test runs. All three methods of inputting activity information were verified, including average speed which uses default driving cycle embedded in MOVES model (average speed method), operating mode distribution (opMode method), as well as second-by-second driving cycle (driving cycle method) for each link. Speed-Matrix was used for applying average speed method, and OpMode-Matrix was used for applying opMode and driving cycle method. There were a total of 4,050 test runs, with 14 links representing different average speed or operating conditions, and 5 emission types covered in each link, which totaled 283,500 emission results.

Table 4. Iteration scenarios for test runs.

Variables	Iteration Increment	Number of Iterations
Facility type	Freeway, Local	2
Fleet (MY & type)	Freeway fleet, Local fleet
Calendar Year	2010–2050 in 5-year interval	9
Fuel month	January (winter), April (transition), July (summer)	3
Temperature	20–100 F in 20-F interval	5
Humidity	20–100% in 20% interval	5
Operation input	Average speed, OpMode distribution, Driving cycle	3
Total Scenarios (Number of Runs): 2 × 9 × 3 × 5 × 5 × 3		4,050
Number of links in each run	Average speed method: 5–70 mph in 5-mph interval OpMode distribution method: 5–70 mph in 5-mph interval with each applied with operating mode distribution Driving cycle method: 14 links with each applied with customized second-by-second driving schedule	14
Total Number of Link Scenarios: 4,050 × 14		56,700
Pollutant type	THC, CO, NOX, PM2.5, CO2	5
Total Number of Emission Results: 56,700 × 5		283,500

Result verification

Because MOVES is the official regulatory emissions model, it is important that results from implementing MOVES-Matrix match the results obtained from the MOVES model. Emission results from MOVES-Matrix and from MOVES batch mode were compared to assess MOVES-Matrix performance. The MOVES-Matrix outputs were within 0.0005% of the results obtained with the conventional MOVES batch mode, as shown in Figure 5. Any potential round-off errors associated with any weighting calculations in MOVES were insignificant.

Figure 5. Emission results (Grams per Mile) of (a) HC, (b) CO, (c) NO_x, (d) PM_2.5 and (e) CO₂ from MOVES and MOVES-Matrix.

Model run time

A model performance comparison for run times was also conducted between the MOVES batch mode and MOVES-Matrix for average speed, operating mode distribution, and driving cycle methods. The results of interest included THC, CO, NO_x, PM_2.5, and CO₂. The two models were set and run in the same computer with configuration of Intel(R) Xeon(R) CPU W3550 at 3.07 GHz, Windows 7 64-bit, RAM: 6 GB.

Based on the same data input, the research team recorded the total run time, including input loading time and model calculation time for these runs in MOVES batch mode and MOVES-Matrix for all three methods. The comparison is shown in Table 5. MOVES-Matrix saves a tremendous amount of computer run time relative to using the MOVES batch mode.

Table 5. Model run time comparison.

Model	Process	Method [total runs = 4,050]
		Average speed method (1,350 runs)		OpMode distribution method (1,350 runs)		Driving cycle method (1,350 runs)
		Time (hr)	Speed (sec/link)	Time (hr)	Speed (sec/link)	Time (hr)	Speed (sec/link)
MOVES	Load Input	3.66	0.69	4	0.76	4	0.76
	Calculation	206.3	39.29	152	28.95	191	36.4
	Total	210	40	156	29.71	195	37.1
MOVES-Matrix	Load Input	0.77	0.14	0.15	0.029	0.54	0.10
	Calculation	0.06	0.01	0.02	0.004	0.16	0.03
	Total	0.83	0.15	0.17	0.032	0.70	0.13
Ratio of Run Time (MOVES Batch Mode/MOVES-Matrix)		253		917		278

Comparing the run times across three methods with MOVES-Matrix, we found that running Matrix with opMode method (0.032 sec/link) is faster than average speed (0.15 sec/link) method and driving cycle method (0.13 sec/link) for two reasons: (1) opMode-Matrix that stores emission rate of 23 operating mode bins, is smaller than Speed-Matrix which stores emission rates of 73 highway average speed bins and 73 local average speed bins, and that makes query process faster in opMode-Matrix than using Speed-Matrix; and, (2) compared with opMode method, driving cycle method requires one more process that converts speed and acceleration of each second to operating mode bin.

In general, MOVES-Matrix can finish the emissions computation tasks over 200 times faster than using the MOVES batch mode. The fast calculation speed of MOVES-Matrix provides a user platform that can be employed with newer and larger datasets, such as INRIX GPS data, traffic simulations, smartphone data, etc., and supports dynamic, real-time emission modeling.

Other real applications

The research team has implemented MOVES-Matrix in a variety of emission modeling research, including: emission impacts of HOV-HOT conversion (Xu et al. 2017a), transit eco-driving (Xu et al. 2017b), emissions benefits of transit deadheading reduction (Li et al. 2016), individual vehicle emission modeling (Guensler et al. 2017), MOVES sensitivity analysis (Liu et al. 2015), near-road dispersion modeling (Liu et al. 2017), connection with travel demand model (Xu et al. 2018), and connection VISSIM model (Xu et al. 2016). For each assessment, the research results demonstrated the results from MOVES-Matrix were the same as using MOVES directly. Figure 6 shows four examples of MOVES-Matrix applications in connection with Atlanta Travel Demand Model (TDM), connection with traffic microsimulation (VISSIM), individual vehicle emission modeling, and dispersion modeling with AERMOD.

Figure 6. Examples of MOVES-Matrix applications: (a) Connection with Atlanta TDM; (b) Connection with traffic microsimulation; (c) Vehicle emission calculator in smartphone; and (d) Dispersion modeling with AERMOD.

For regional scale scenarios involving large number of (e.g., 74,500 in Atlanta) roadway links, the research team recommends users manage fleet composition by road type and traffic analysis zones. Link speeds and volumes can be obtained from travel demand models, and/or dynamic traffic assignment. MOVES-Matrix supports batch mode processing and enable multitask runs, just as MOVES does. Each task specifies a single calendar year, meteorology, fuel supply, and fleet model year distribution. At the link level, links that have the same fleet composition could be grouped in the same task, allowing users to obtain an emission rate for all speeds and for fleet compositions for multiple calendar years and meteorology scenarios. These emission rates can then be mapped back to specific links based on traffic analysis zone and link speed, and multiplied by link volumes to obtain fuel consumption and mass emissions for each link. The research team is currently implementing a MOVES-Matrix connection with the Atlanta Regional Commission’s travel demand model, which will serve as a guide for MOVES-Matrix application at regional scale.

For project-level emission analysis, users can link MOVES-Matrix emission rates with traffic simulation model outputs. The simulated vehicle driving schedules (speed-time traces) for individual vehicles yield second-by-second on-road operating conditions (which translate to speed and VSP bin) that can be linked with operating mode emission rates in the MOVES-Matrix. For example, the research team linked MOVES-Matrix with VISSIM microsimulation software and predicted emissions as a function of VISSIM-simulated second-by-second vehicle trajectories (Xu et al. 2016). To accomplish the linkage, a local fleet composition (fleet composition for 13 source types and their on-road model year distributions) was developed for use in the VISSIM simulation and in emissions modeling. The VISSIM model was coded and calibrated to represent on-road traffic conditions. A Component Object Model (COM) interface was applied to collect network information and second-by-second speed profiles for the simulated vehicles on network. Second-by-second vehicle traces data were post-processed to obtain second-by-second operation mode bins. Finally, the applicable MOVES-Matrix operating mode bin emission rates (by county, fuel formulation, I/M strategy, and meteorology) were obtained from the MOVES-Matrix emission rate table. Emission results were calculated by matching the operation conditions for each second from simulation model with applicable MOVES-Matrix emission rates for the vehicle source type, model year, operating mode bin, and pollutant.

MOVES-Matrix also links monitored on-road operating conditions, such as observed driving cycles from smartphones. The development of MOVES-Matrix has simplified the use of large scale of traffic activity data in emission modeling, as is currently being demonstrated in a Department of Energy ARPA-E project (DOE 2015) in Atlanta, making real-time MOVES energy consumption and emissions modeling feasible. The research team has applied MOVES-Matrix in individual vehicle modeling to predict second-by-second fuel consumption and emissions, and incorporated the emissions modeling approach in Commute Warrior®, an Android® travel survey application, to predict real-time fuel consumption and emissions given second-by-second speed data concurrently collected by the smartphone GPS. When a vehicle make, model, and model year is selected by the user, the vehicle source type by MOVES is identified from a lookup table and the applicable VSP vehicle parameters are identified for VSP calculations. Information for individual vehicles is easily converted to corresponding vehicle source type in MOVES for emission rate connection if the make, model, and model year are obtained from a registration database. The subset of corresponding emission rates for the vehicle source type (associated with vehicle make and model), fuel type, model year, is extracted from MOVES-Matrix for all operating mode bins, temperatures, and humidity combinations. When a vehicle speed trace is recorded in Commute Warrior®, second-by-second VSP is calculated as a function of source-type dynamics parameters (i.e., rolling resistance, rotating resistance, and aerodynamic dragging coefficients), speed, acceleration, and road grade. The operating mode bin for each second is allocated according to VSP and speed values. The second-by-second fuel and emission rate can be assigned by extracting the emission rate of corresponding operating mode bin, for the temperature and humidity specified. The system also allows researchers to directly assess strategies designed to change individual travel behavior to increase efficiency, and to evaluate the potential impacts of major transportation design and operation strategies. Energy and emission analysis tools coupled with simulation allows short-time prediction, and feedback to travelers to support more efficient decision-making. Users will be able to track fuel economy, carbon footprint, and emissions, and playback vehicle speed and fuel consumption rates along trip routes, and generate trip summary reports by time period or trip purpose.

MOVES-Matrix can be linked with dispersion models for transportation conformity and hot-spot analysis. The research team has successfully connected MOVES-Matrix with AERMOD and CALINE4 model in an automated way through Python scripts (Liu et al. 2017). At the beginning of each model run, the system extracts a subset of matrix containing emission rate and energy consumption rates applicable to the scenario of interest. This extraction from MOVES-Matrix is based on the calendar year and month of the analysis, and the temperature and humidity range of the analysis. Hourly emission rate data can be calculated through MOVES-Matrix based on hourly traffic volume, on-road operating speeds and meteorology, and can be aggregated to any time-scale based on index of National Ambient Air Quality Standards (NAAQS) (U.S. EPA 2016c). The fleet average emission rate and meteorology data then serves as the emission rate input for CALINE4 and AERMOD modeling. Static input parameters can be prepared in advance, including link geometry, geographic data, and receptor coordinates, and normally do not change within any single analysis. Because the MOVES emission rates outputs are contained in MOVES-Matrix, and no approximations or corrections are employed, the emission results from the MOVES-Matrix, and the modeled air pollutant concentration are exactly the same as the traditional ways of using MOVES model and dispersion models recommended by U.S. EPA (U.S. EPA 2013), with a speed 200 times faster. This means that the MOVES-Matrix model obtains the same results as the standard regulatory dispersion analysis with significant efficiency.

Moves-matrix availability

By inventorying the fuel specification and I/M programs scenarios across the United States from 2010 to 2050, there are a total of 22 fuel regions and 89 I/M scenarios implemented across the 3,228 counties in the United States, which ends up with 117 unique fuel and I/M program combinations. This implies that by preparing 117 MOVES-Matrix runs, all scenarios in every county will be completed across the United States. Since it takes 3.5 days to run a full set of scenarios on the PACE cluster (at temperature of 5-degree intervals). It is feasible to conduct national implementation of MOVES-Matrix if computing time is available. Figure 7 shows geographic areas that have been calculated in MOVES-Matrix up to April 2019. 2,885 counties have been calculated in MOVES-Matrix.

Figure 7. Counties calculated in MOVES-Matrix.

The Georgia Tech team is preparing MOVES-Matrix setup for modeling region that meets users’ needs. The team can generate emission rates for a new region for users if it is not covered in Figure 7. There is no cost associated with generating MOVES-Matrix from the users’ side. The only cost will be data delivery (for example, delivery of external drive may be required). The MOVES-Matrix package includes MOVES-Matrix database, a python script linking the MOVES-Matrix data for users’ application, and documents describing the MOVES-Matrix database and how it can be used based on the Python script. The Georgia Tech team will maintain the MOVES-Matrix script, and notify all users when an updated version of MOVES-Matrix Python script is available.

Conclusion

This study introduced the MOVES-Matrix modeling approach; a high-performance emission modeling system that uses emission rates pre-generated by MOVES, rather than performing MOVES modeling runs repeatedly for transportation scenarios of interest. The scenario runs demonstrate that MOVES-Matrix can finish the emissions computation tasks over 200 times faster than using the MOVES batch mode and the generated results are exactly the same. In addition to its high performance in calculation speed, there are other benefits in applying MOVES-Matrix: (1) MOVES emission rates are employed directly in MOVES-Matrix (there are no code modifications, no use of correction factors, nor any approximations employed). (2) In project-level emissions analysis, users typically assume a single temperature, humidity, and fuel, and estimate the emissions impact of the changes in vehicle operations and fleet composition. Hence, the database is organized into sub-matrices that fit the users’ work schemes, and allows users to conveniently and rapidly assess impacts of changes in on-road operating conditions and fleet composition. (3) MOVES-Matrix emission rates can be called in Java, Python, Perl, or any similar scripting program to link MOVES emission rates with travel demand models, simulation models, monitored vehicle data, and dispersion modeling. (4) Because the emission database of MOVES-Matrix is composed of MOVES outputs, and the model achieves the exact same results as MOVES, the model is ready for regulatory review and approval. (5) MOVES-Matrix is an open source system that anyone can use. (6) The research team is currently building an online version of MOVES-Matrix that will allow users to implement emission analysis online without having to run MOVES.

MOVES-Matrix is currently developed for on-road running emission modeling only. The research team is exploring incorporating nonroad emission modeling in MOVES-Matrix for further comprehensive analysis. MOVES-Matrix does not currently estimate emissions from starts, hoteling or evaporative emissions, the research team is working on MOVES-Matrix version 2 that supports incorporate off-network modeling.

Disclaimer

The information, data, or work presented herein were funded in part by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof.

Additional information

Funding

This work was supported by the National Center for Sustainable Transportation.

Notes on contributors

Haobing Liu

Haobing Liu is affiliated with the School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA.

Randall Guensler

Randall Guensler is affiliated with the School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA.

Hongyu Lu

Hongyu Lu is affiliated with the School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA.

Yanzhi Xu

Yanzhi Xu is affiliated with the Technology at Center for Advancing Research in Transportation Emissions, Energy, and Health, Texas A&M Transportation Institute, College Station, Texas, USA.

Xiaodan Xu

Xiaodan Xu is affiliated with the School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA.

Michael O. Rodgers

Michael O. Rodgers is affiliated with the School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA.