*4.1. Data Collection*

The navigation data was acquired in December 2017, and among the data on 13 December (Wednesday), the driving data on the morning peak hours (07:00–09:00), non-peak hours (13:00–15:00), and afternoon peak hours (17:00–19:00) for the northbound and southbound parts of each road section was extracted. Table 1 summarizes the number of vehicle trajectories collected for each analysis unit, showing that the vehicle trajectories corresponding to approximately 2–7% of the traffic volume were collected. After extracting the vehicle position in seconds from the navigation data, the data was organized into vehicle-specific driving trajectory data to calculate the speed per second and acceleration per second. Because most of the navigation data was provided from passenger cars, the analysis was conducted by considering the passenger car as the vehicle type.

### *4.2. Cluster Analysis Results*

Cluster analysis was performed for each of the 12 groups listed in Table 1, and Table 2 summarizes the results. Among the collected vehicle trajectories, the data showing outliers in terms of the travel time was removed before the cluster analysis. Most of the removed vehicle trajectories were from vehicles that did not pass through all sections of the corresponding road. Table 2 shows the criteria for outlier removal and the number of selected vehicle trajectories. The cluster analysis results reveal that the best number of clusters for highway data is 7 to 8 and the ratio of goodness of fit is about 0.6 to 0.8. The results also show that the best number of clusters for the arterial road data is 6 to 8 and the ratio of goodness of fit is about 0.8 to 0.9.


1 07:00–09:00, 2 13:00–15:00, 3 17:00–19:00, 4 Estimated traffic volume provided by the Korean Traffic Database.


**Table 2.** Summary of trajectory clustering.

### *4.3. Emission Estimation Results*

The emissions of air pollutants (CO, NOx, PM10, PM2.5, and CO2) were estimated based on the method described in Section 3. The estimated emissions were compared with the results obtained by estimating emissions using each individual vehicle trajectory through the micro-level emission estimation method of MOVES. As shown in Table 3, the difference in the emissions calculated by the two methods is insignificant, at 1–4% for the highway and 1–6% for the arterial road. The results prove that the proposed method has acceptable accuracy in estimating emissions.


**Table 3.** Comparison of link emission estimation results.

1 Estimating emissions using each individual vehicle trajectory through the micro-level emission estimation method of MOVES, 2 Estimating emissions using emissions of cluster centers through the proposed method, 3 B/A.

### *4.4. Micro-Level Link Emission Factors*

The micro-level link emission factors are calculated by emissions from vehicles in all clusters, which are B columns in Table 3 by the number of vehicles (the number of vehicle trajectories), and summarized in Table 4. These values are the micro-level emission factors of the passenger car at the analysis time interval on the analysis link for each pollutant. As shown in Figure 3, plotting these values through bar graphs can determine whether the estimated micro-level link emission factors can appropriately reflect the emission characteristics of the link. The emission factors show the highest trend in the morning peak hours and the lowest in the non-peak hours, indicating the change in emissions according to time periods of the link. Even though the driving length of the highway (about 1 km) is longer than that of the arterial road (about 600 m), the emissions are lower on the highway, which indicates that the emission factors appropriately reflect the driving characteristics by the road characteristics (uninterrupted flow and interrupted flow) of the link.


**Table 4.** Estimated link emission factors (g/veh).

**Figure 3.** Estimated Link Emission Factors by Pollutant Type.

### **5. E**ff**ects on Clustering and Micro-Level Link Emission Factors by Using Accumulated Vehicle Trajectories**

This study has confirmed that the proposed method can be applied to estimate the total emissions from vehicles traveling on a road section with the actual vehicle trajectory data through the case study. Micro-level link emission factors for links were derived through the proposed emission estimation process using vehicle trajectories collected from one day. This method may be applied by using accumulated vehicle trajectory data collected over several days to increase the micro-level link emission factors' representativeness. Use of an accumulated dataset can offset the limitations of vehicle trajectory data, which have a low rate of acquisition, and enable the analysis time period to be divided into shorter periods to increase detail.

To analyze the e ffect of using accumulated vehicle trajectory data, the vehicle trajectory data of the freeway northbound link selected for the case study was additionally acquired. The trajectory data of the vehicles traveling through the road section during the 2-h morning peak on weekdays (Tuesday, Wednesday, and Thursday) in December 2017 was used. A total of 12 datasets were made by increasing the number of data collection days from one day to 12 days. For example, the first dataset included one day of data, the second dataset included an additional day of data, and so on. The proposed method was applied to each of those datasets.

In this analysis, changing patterns in cluster analysis results and estimated micro-level emission factors were investigated by accumulating daily data. Table 5 shows the number of remaining vehicle trajectories after removing outliers among the vehicle trajectories collected on each date. Table 6 summarizes the cluster analysis results subject to daily accumulation of the data, which shows that the number of vehicle trajectories used for cluster analysis increases as the daily data is accumulated, reaching 1029 after the 12th day of accumulation. The average travel times of the vehicle trajectories used for cluster analysis show a pattern that converges to about 101 s. The optimal number of clusters is 7, and the ratio of goodness of fit is 0.7 or more.


**Table 5.** Vehicle trajectory selection results.

**Table 6.** Summary of trajectory clustering by increasing number of data collection days.


Figure 4 is a diagram comparing the OpMode distributions of seven cluster centers derived from each dataset. It can be observed that the shapes of the OpMode distribution of the seven cluster centers are similar after Day 4. This graph o ffers useful information for determining how many days are required to accumulate vehicle trajectories for estimating micro-level link emission factors.

**Figure 4.** OpMode Distributions of Cluster Centers.

Table 7 summarizes the micro-level emission factors for each pollutant estimated by adding the number of data collection days. The bar graphs plotted in Figure 5 show that the micro-level emission factors tend to converge as the number of days of data accumulation increases. It means that the estimated micro-level link emission factors' representativeness increases as the data is accumulated.


**Table 7.** Estimated link emission factors (g/veh) by adding number of data collection days.

**Figure 5.** Changes in Estimated Link Emission Factors by Increasing Number of Data Collection Days.
