**2. Literature Review**

To apply the micro-level emission estimation method to on-road mobile sources, it would be ideal to collect the trajectories of all vehicles running on the road and put the data into the micro-level emission calculation program. However, doing so is highly impractical, and even if it were not, it would be ine fficient because it takes a large amount of time to calculate the micro-level emissions from all driving vehicles in the transportation network. To overcome this limitation, the U.S. EPA has developed MOVES [13], and several studies have incorporated vehicle trajectories (called driving cycles) into MOVES, which represent driving characteristics by vehicle type, road type, and level of service (LOS) [19–22]. However, it is not suitable to use the MOVES driving cycle in countries other than the United States because the road geometry, vehicle type composition ratio, and driver's driving characteristics that determine the driving characteristics of a vehicle vary among countries [23,24]. Therefore, some studies conducted in countries other than the United States have applied MOVES with adjusted base emission rates to analyze vehicle emissions in those countries [19–21].

Another solution is to collect vehicle trajectory data from on-road mobile sources in the target area to be analyzed. The individual vehicle trajectories used in most studies were collected from a GPS device installed in a driving vehicle [7–9] or extracted from a microscopic tra ffic simulation model [25,26]. The vehicle trajectory data is collected or extracted from several vehicle types, including cars, trucks, and buses. Therefore, distinguishing the data by vehicle type is crucial for calculating vehicle emissions. Several studies have applied cluster analysis methods [27,28] to extract representative vehicle trajectories. Moreover, variables used to distinguish similar types of vehicle trajectories have included average speed, average acceleration, average deceleration, time proportion of idling mode/cruising mode/acceleration mode/deceleration mode/creeping mode, frequency of vehicle stops, mean length of driving period, average number of acceleration–deceleration changes, and root mean squared acceleration, which are aggregates representing the corresponding characteristics [29].

Previous studies associated with cluster analysis were performed to classify the vehicle trajectories into the representative driving cycles by vehicle type and road type. This study utilizes cluster analysis to assemble the vehicle trajectories of the same vehicle type on the same road section at the same time intervals into several similar vehicle trajectory groups and then uses the representative patterns of each group for estimating the corresponding highway link-based vehicle emissions. This process is expected to reflect the various driving situations that can occur in the same context. If aggregated characteristics, such as average speed and maximum acceleration, are used as variables for clustering, these measures cannot be used for micro-level emission estimation. Thus, the MOVES OpMode distribution, which can be used immediately for estimating micro-level vehicle emissions in MOVES, is applied as a variable for cluster analysis.
