The Impact of Violations of Bicycles and Pedestrians on Vehicle Emissions at Signalized Intersections

An intersection is a typical emission hot spot in the urban traffic network. And frequent violations such as running the red light have been a critical social problem at signalized intersections in developing countries. ,is article aimed to quantify the impact of violations (behaviors which will block the fleet) on emissions at signalized intersections. Increased emissions of vehicles affected by violations are of two levels: (1) trajectory level for the first four affected vehicles and (2) traffic flow level for the subsequent vehicles. At the trajectory level, the study focuses on the second-by-second activities of the first four affected vehicles. First, the trajectory model of the first affected vehicle is developed. ,en, the trajectory of the other three vehicles is constructed using the Gipps car-following model. At the traffic flow level, a linear emission model is developed to describe the relationship between emission factors and idling time in the one-stop (vehicle stop once) and two-stop (vehicle stop twice) scenarios based on the global position system (GPS)-collected data at 44 intersections in Beijing. Based on the linear emission model, increased emissions at the traffic flow level are calculated as knowing the number of stops and idling time before and after violations. ,e analysis of the subsequent vehicles is divided into unsaturated and saturated conditions. Under the unsaturated condition, the emissions have barely increased due to the increase of idling time for one-stop vehicles caused by the violations. Under the saturated conditions, the emission increment increases sharply as the one-stop vehicle gradually transforms to a two-stop vehicle because of violations, and the maximum emission increment reaches 45% in half an hour in the case. ,e increment of emissions decreases steadily as the proportion of two-stop vehicles reaches 100% after violations, while the proportion before violations keeps increasing.


Introduction
In recent years, urban vehicle pollution has threatened human health. Air pollutants can cause pulmonary and cardiovascular diseases and chronic obstructive pulmonary disease, and decrease lung function [1]. According to the WHO report, air pollution resulted in 3.7 million premature deaths worldwide in 2012 [2]. In the United States, motorized vehicles are responsible for 57% of emissions [3].
An intersection is a typical emission hot spot in the urban traffic network [4], especially in a densely populated metropolis like Beijing. Extensive studies have been carried out on vehicle emissions around intersections, and illustrate that high emissions at intersections mainly resulted from unstable traffic operation and stop-and-go vehicle activity [5][6][7] and lead to high pollutant exposure, particularly to pedestrians and cyclists around the intersections.
When vehicles, pedestrians, and cyclists arrive at the intersection, each traffic subject follows the corresponding rules and passes through the intersection orderly. However, violation behaviors, one of the core factors, are frequently observed in developing countries such as China [8]. Frequent violations will not only aggravate traffic risk but also worsen the stop-and-go condition of vehicles. Four typical violations at intersections are shown in Figure 1 frequent violations of pedestrians and bicycles block the vehicle fleet, aggravating stop-and-go behaviors at intersections. In this article, the critical problem of quantifying emission increments caused by violations has been discussed.
Motor vehicles, bicycles, and pedestrians are three typical traffic components of mixed traffic flow intersections. eir relationships are variable and complex due to the conflict in time and space. Existing studies indicated that violations were frequently observed in developing countries and had a significant impact on driving operation [7]. Researches on the impact of violations on emissions can be summarized into two categories: (1) intersection emission estimation and (2) violation characteristics and impact analysis.
In terms of intersection emission estimation, existing studies indicated that steady speed, shortest delays, and fewest stops are the best operations for emission reduction [9,10]; nonsmooth traffic operations and stop-and-go activity were the main reasons for high emissions at intersections; and much more time was spent in the acceleration mode after highly interrupted movements of vehicles [11][12][13]. During the acceleration period, an engine operates at a high level, which leads to excessive fuel consumption and emissions [14,15]. In existing studies, extensive studies have been carried out on vehicle emission around intersections. Rakha and Ding analyzed the impact of stops on vehicle fuel consumption and emissions and found that vehicle fuel consumption was more sensitive to cruise-speed levels [13]. Papson et al. used a timein-mode model with MOVES to analyze vehicle emissions at congested and uncongested signalized intersections [16]. Zhang et al. developed the SIDRA model to estimate intersection emissions with MOVES, and acceleration mode included constant acceleration and linearly decreasing acceleration (3). Gokhale and Pandian developed a semiempirical box model to predict CO concentration based on the assumed traffic flow pattern at intersections [17]. Braven et al. estimated vehicle emissions at a signalized intersection by using VISSIM [18].
In terms of violation characteristic analysis, it can be summarized into two categories. (1) Violation characteristics mainly include two aspects: (a) for the influencing factors (waiting time, group size of pedestrians, gender, nonmotor vehicle type, and so on), existing studies indicated that these factors had an obvious impact on intersection violations [19][20][21], and (b) for the frequency model, violation probability model [22] and violation waiting time model [23] were developed to predict the frequency of violations. (2) Impact of interference has an impact on vehicles, both at the trajectory level and the traffic flow level. At the trajectory level, Przybyla established a dynamic car-following model for driving distraction [24]. At the traffic flow level, interference has an impact on vehicle speed [25] and further influences intersection capacity [26][27][28]. e mechanism of the trajectory and emission characteristics affected by interference has not been fully understood. Further investigation is thus necessary. Hence, this article was aimed at quantifying the impact of violations on emissions at intersections. e content of this article is divided based on the following two aspects. parameters (number of stops and idling time) of the subsequent vehicles after the first four vehicles. A numerical simulation is conducted to assess the impact of violations on emissions based on the existing study.

Materials and Methods
In this study, violations are defined as the behavior that will block the vehicle fleet. Pedestrians and cyclists are the objectives of violations. Increased emissions of vehicles affected by violations include two levels (see Figure 2): (1) trajectory level for the first four affected vehicles and (2) traffic flow level for the subsequent vehicles. At the trajectory level, the study focused on second-by-second activities of the first four affected vehicles. First, the trajectory model of the first affected vehicle is developed. en, the trajectory of the other three vehicles is constructed by using the Gipps car-following model. At the traffic flow level, a linear emission model that can describe the relationship between emission factors and idling time in the one-stop and two-stop scenarios is developed by using Global Position System (GPS)-collected data at 44 intersections in Beijing. Increased emissions are calculated by the number of stops and idling time before and after violations based on the linear emission model. Finally, a case study is conducted to assess the impact of violations on vehicle emissions at signalized intersections.

Data Collection.
is article includes two types of data.  (c) Idling time of the first affected vehicle. e time precision is 0.02 s, and 66 groups of data are collected.

Emission Data.
e emission data includes two parts.
(1) Emission rate. Vehicle emission data are derived from the local emission rate model for light-duty gasoline vehicles [29,30]. e emission standard of China IV was selected to provide the emission rates for LDVs, which is more common in Beijing compared with other emission standards. (2) Global positioning system (GPS) trajectory data.
1666 valid trajectories of data, whose range is 200 meters of the intersection radius, were selected from 44 arterial and collector intersections in Beijing. Vehicle specific power (VSP, kW/ton) is estimated after data quality control [31]. Other details about the data are listed as follows:  Figure 4. When the violation occurs, the headway of the first affected vehicle will increase obviously. From the first affected vehicle to the fourth, the headways gradually return to be stable. e impact of violations on the vehicle fleet disappears after the fourth vehicle. As a result, the first four vehicles are determined as vehicles affected by violations in the trajectory level.
where t lost (s) is the total lost time of violation behaviors. i is the i th vehicle after the violation location. t head,after (s) is the headway after the violation, and t head,before (s) is the time headway before the violation.
Vehicles trajectories under the violation are shown in Figure 5. e x-axis is time, and the y-axis is the distance. e negative y-axis is the location of the queuing vehicles, and the positive y-axis is downstream of the intersection. e red point represents the location and time of the violation. τ is the reaction time plus braking time, and S τ is the corresponding distance. e first affected vehicle will Journal of Advanced Transportation return to normal speed after the process of decelerationidling-acceleration. e second vehicle starts to decelerate after the reaction time of D r since the first affected vehicle began to decelerate. e deceleration of the second vehicle is less than the first vehicle as the interference decreases as the impact of the violation spreads in the vehicle fleet. e second vehicle begins to accelerate after the reaction time of A r since the first affected vehicle accelerated. When the yaxis of D r and A r is the same, the second affected vehicle will have a second stop; the process of the second vehicle is deceleration-idling-acceleration; if not, the process is deceleration-acceleration. e interference is gradually transmitted to the rear of the vehicle fleet, as described above. When the speed is low enough, the subsequent vehicles will extend the idling time to eliminate the interference as the fifth vehicle. en, the subsequent vehicles will pass the intersection with the saturated time headway.

Trajectory Level.
At the trajectory level, the objective is the first four affected vehicles after the violation. It is necessary to divide the model into two parts: (1) the trajectory model for the first vehicle and (2) the car-following model for the other three vehicles.   For the idling time, sixty-six groups of idling time data in violation situations are collected. e average idling time is 2.08 s (standard deviation is 3.03).
For the acceleration and deceleration data, a corresponding relationship between vehicle speed and acceleration is developed in the low-speed range based on GPS trajectory data (see Figure 6).
As shown in Figure 6, based on the GPS trajectory data of 44 intersections, a total of 18,709 pieces of valid data of the first 10-s deceleration and 10-s acceleration around the idling time are extracted. At low speed, each speed corresponds to a group of accelerations. In other words, the process of acceleration and deceleration in violation conditions can be totally described after knowing the initial speed before violations. e acceleration of the first affected vehicle can be calculated as equations (2)-(3): a s,t � random a range v s,t , when acceleration or deceleration, where V s,t are the speeds in the acceleration and deceleration states at time t; the states are acceleration and deceleration when s is a and d, respectively. a range (V s,t ) are the corresponding acceleration range of V a,t and V d,t . Each corresponding acceleration range has a database in which random values are taken by numerical simulation.

e Rear ree Vehicles of the First Four Affected Vehicles.
e rear three vehicles decelerate after the sudden stop of the first affected vehicle. At the intersection, acceleration of the rear vehicle is closely related to the speed of the preceding vehicle, the speed of the rear vehicle, and the distance between the two vehicles. e Gipps car-following model can not only fulfill the above requirement but also generate rather realistic VSP distributions among the car-following models [33], which are calculated as equations (4)

Journal of Advanced Transportation
where a n (m/s 2 ) is the maximum desired acceleration of the following vehicle; d n (m/s 2 ) is the maximum desired deceleration of the following vehicle; d n−1 (m/s 2 ) is the estimation of the maximum desired deceleration of the leading vehicle; T (s) is the apparent reaction time; S n−1 is the effective length of a vehicle, which consists of the vehicle length and the minimum distance between the vehicles; θ is the additional delay for braking, which is 0.5; and α, β, and c are parameters, which are 2.5, 0.025, and 0.5, respectively.

Emission Increment at Trajectory
Level. At the trajectory level, the study focused on second-by-second activities of the first four affected vehicles. Two different models are developed: one is the trajectory model of the first vehicle, while the other is the Gipps car-following model of the three rear vehicles. e trajectory model of four vehicles affected has been developed above; thus, increased emission factors of the four vehicles can be calculated as follows:   where EF increased,trajectory,x (g/km) is the increased emission factors of the x th affected vehicle, VSP Bin i,trajectory,x is the time fraction in the i th VSP bin of trajectory after violations, and V effected,x (km/h) is the average speed after violations.   (30,31). It is hypothesized that each of these speed profiles generates different levels of emissions, with type three generating the highest amount of emission due to longer idling time and more acceleration and deceleration circles. e equation for calculating VSP is provided by equation (8). VSP values are binned at an interval of 1 kW/ton. is article analyzes the VSP bins of −20 kW/ton to 20 kW/ton. More than 98% of the VSP data are in this range.

Traffic Flow
where a t (m/s 2 ) is the acceleration, g (m/s 2 ) is the acceleration due to gravity, which is 9.8 m/s 2 ; and sin θ is the road grade. A (kw·s/m), B (kw·s 2 /m 2 ), and C (kw·s 3 /m 3 ) are road load coefficients, representing rolling resistance, rotational resistance, and aerodynamic drag, respectively; the values of A, B, and C are 0.156461, 0.0020002, and 0.000493, respectively. m (ton) is the vehicle weight, and the value is 1.4788 ton. Based on GPS trajectory data, idling time rarely exceeds 150 s for one-stop vehicles, or 275 s for two-stop vehicles.
us, 85 VSP distributions are developed based on the number of stops, idling time, and division of upstream and downstream (see Table 1).
(2) Emission Rates. Vehicle emission rate data were derived from the local emission rate model for light-duty gasoline vehicles (see Figure 8), and the type of gasoline emission standard is China IV. e emission factors are calculated as follows: where EF v (g/km) is the emission factor at the average speed of v (km/h). ER i (g/s) is the mean emission rate of the i th VSP Bin. VSP Bin i is the fraction in the i th VSP Bin.  e emission factors increase linearly with the increase in idling time, as shown in Figure 9. e difference in emission factors between one-stop and two-stop vehicles is small under the same idling time. It is because the two-stop vehicle keeps a low speed and low acceleration, while the corresponding VSP value is concentrated around zero, and the corresponding emission rate is also small. But the range of the idling time of one-stop and two-stop vehicles is obviously different. e emission factor model is shown as follows: where EF i , EF j (g/km) are emissions of types i and j. When i � 1, 2, 3, 4, EF i is the emission factor of CO 2 , CO, NO X , and HC of one-stop vehicles, respectively. When c � 1, 2, EF j is the emission factor of one-stop and two-stop vehicles, respectively. t idling is the idling time of the vehicle. a i and b i are parameters.

Variations of Vehicles Operating at Traffic Flow Level.
At the traffic flow level, the study focused on the number of stops and idling time for the subsequent vehicles after the first four vehicles, which is divided into two scenarios (see Figure 10).
(1) Unsaturated condition of the vehicle fleet after being disturbed by violations; (2) Saturated condition of the vehicle fleet after being disturbed by violations.
As shown in Figure 10, the total lost time of the first four affected vehicles is 4 s in the unsaturated condition. e idling time of subsequent vehicles increases by 4 s, and the number of stops is still one. Under the saturated condition, two vehicles will transform from one stop to two stops due to the increased idling time. And the idling time of two-stop vehicles increases by 4 s. Suppose that the maximum number of stops for all vehicles is two. Subsequent vehicles are divided into three types: (type A) one-stop vehicle, the increase in idling time is equal to the total lost time, (type B) transformation from one-stop vehicle to two-stop vehicle, and (type C) two-stop vehicle, the increase in idling time is equal to the total lost time of cycles.
Under the unsaturated condition, the subsequent vehicles are all type A. Under the saturated condition, vehicles that passing through the intersection are types A, B, and C at the beginning. As cycles pass, type A vehicles will disappear first, followed by type B vehicles. After all vehicles transform to two-stop vehicles, vehicles passing through the intersection are all type C.
As shown in Figure 11, VSP distributions of vehicles with and without violations are divided into three conditions, which corresponds to types A, B, and C (see Figure 11). For types A and C, the number of stops of affected vehicles is constant. us, the difference in VSP distributions of vehicles with and without violations is small. For type B, the number of stops transforms from one to two, and the idling time increment is the sum of the lost time and the red-light time. e frequency of VSP distributed above 0 kW/ton will be much highly affected by violations.
Take equation (10) as a reference. e emission model has been developed based on the number of stops and idling time. e operation change of subsequent vehicles, which are after the first four affected vehicles, is reflected in the number of stops and idling time.
erefore, increased emissions of the subsequent vehicles can be obtained by the emission model above. Based on equation (11), the numerical simulation method is used to calculate the emissions under violation conditions. EF increased,flow,y � a cAfter,i · t after + b cAfter,i − a cBefore,i · t before + b cBefore,i , (11) where EF increased,flow,y (g/km) is the increased emission factors of subsequent vehicles. y is the y th vehicle of vehicles going through the intersection in the cycle. t after is the idling time after violations, and t before (s) is the idling time without violations. cAfter and cBefore are stopping numbers before and after violations, respectively.

Increased Emissions Affected by Violations.
e analysis of emissions affected by violations includes two levels. At the trajectory level, the study focused on second-by- where AE (%) is the increased emission increment of the cycle. e is the e th vehicle, which is the location of the first affected vehicle. n is the vehicle number of the cycle. EF increased,trajectory,x (g/km) is the increased emission factors of the first four affected vehicles. EF increased,flow,y (g/km) is the increased emission factors of the subsequent vehicles after the first four affected vehicles. EF i (g/km) is the normal emission factors of vehicles. EF downstream (g/km) is the emission factors downstream of the intersection. D (km) is the distance of the study range, which is 200 meters.

Case Study
Numerical simulations are designed in the case. e impact of violations on intersection emissions can be quantified based on the model built at the trajectory and traffic flow levels. e numerical simulation object of the case is the arterial (north-south) direction of the intersection. e seven simulation conditions are listed as follows: (1) e average lost time is 5.52 s according to the violation data.

Case Intersection.
A typical intersection in Beijing (Anli and Huizhong North) is chosen as a reference for the case intersection, which is the arterial and collector protected intersection. e simulation object is the arterial (northsouth) direction of the intersection. e signal information and channelization information are shown in Table 2.   From stop once to twice    -south straight  74  3  2  Straight  Left  Right  North-south turn left  42  3  2  North  3  1  1  East phase  34  3  2  South  3  1  Straight-right  West phase  28  3  2  East  0  1  Straight-right  Cycle time  198  West  1  1 Straight-right    (2) Under the saturated condition, the emission increment increases sharply as the one-stop vehicle gradually transforms to the two-stop vehicle because of violations. As the proportion of two-stop vehicles after violations reaches 100%, and as the proportion before violations keeps increasing, the increment of emissions will decrease steadily. At the ninth cycle (half an hour), the emission increment is changed from 45% to 33% as the volume is changed from 36 to 42 veh/lane/cycle. (3) As the cycle passes, the increment emissions of violations will have another increasing-decreasing process when three-stop vehicles appear. And a more increasing-decreasing process will appear when multiple-stop vehicles appear.

Conclusions
is paper studies the impact of violations on emissions at intersections based on the real-world driving trajectory data. e emission increment is constant under the unsaturated condition and is 1.08% when the volume approaches saturated flow. e emission increment increases sharply as the one-stop vehicle gradually transforms to the two-stop vehicle because of violations under the saturated condition, and the maximum emission increment reaches 45% in half an hour in the case. e increment of emissions decreases steadily as the proportion of two-stop vehicles after violations reaches 100%, and the proportion before violations keeps increasing.
More emission numerical simulations for different types and frequencies of violations are recommended for further studies. In addition, three-stops and above should be considered in high-frequency violation scenarios.

Data Availability
Previously reported emission data were used to support this study and are available at https://doi.org/10.3141/2627-08 and https://doi.org/10.3141/2570-09. ese prior studies (and datasets) are cited at relevant places within the text as references [30,32]. e manual investigation data used to support the findings of this study are included within the article.

Disclosure
is paper was submitted for presentation at the annual meeting of the 98 th Transportation Research Board.

Conflicts of Interest
e authors declare that there are no conflicts of interest regarding the publication of this paper.