Evaluation of Discontinuous Lane Design of Intersection Approach in China

Transcription

1 TRB Evaluation of Discontinuous Lane Design of Intersection Approach in China Maosheng Li Ph.D, Associate Professor, Urban Transport Research Center, School of Traffic and Transportation Engineering, Central South University, Changsha, Hunan 0, P.R. China Linli Chen Master student, Urban Transport Research Center, School of Traffic and Transportation Engineering, Central South University, Changsha, Hunan 0, P.R. China Helai Huang* (Corresponding author) Ph.D, Professor, Urban Transport Research Center, School of Traffic and Transportation Engineering, Central South University, Changsha, Hunan 0, P.R. China huanghelai@csu.edu.cn Submitted for possible presentation and publication in Transportation Research Board Annual Meeting 01. July 0, 01 1

2 ABSTRACT Design of discontinuous connecting lanes between the upstream road lane and approach lane is prevailing in most of Chinese cities. The present paper evaluates the safety and efficiency of the discontinuous lane design in the case that three lanes of the upstream are followed by four lanes of the approach to the intersection. The post-encroachment time are employed as the response variable representing the severity of traffic conflict in an ordered probit model with consideration of a number of latent influencing factors. It is found that the lane used by the vehicle in the upstream road segment has an important effect on traffic safety in the case of discontinuous lane design. The result demonstrates that continuous lane design connecting the upstream road lane and corresponding approach lane improves traffic safety, which could direct a vehicle to its target lane of approach under the relatively stable traffic volume between right and left turning flows. Meanwhile, it is innovatively found that discontinuous lane design is more suitable for improving traffic safety under the condition of dramatic stochastic fluctuation of turning flows across different signal phases. Applicable conditions for continuous lane design before the approach of an intersection are recommended on the basis of the volatility of turning flow. Keywords: Traffic marking design; Surrogate safety analysis; Ordered probit model; Post-encroachment time

3 Introduction A useful means of improving traffic capacity at an intersection is to increase the number of approaching lanes to the intersection. The Chinese national standard Code for planning and design of urban roads (GB00-) specifies the measures to increasing intersection capacity and suggests the adoption of an widened approach of intersection by setting more entrance lanes. Specifically, the standard suggests the approach to the intersection needs to be widened (1) to at least four lanes if there are three lanes upstream of (and leading to) the intersection, () to at least three lanes if there are two lanes upstream or three lanes in total leading to and away from the intersection, and () to at least two lanes if there is only one lane upstream. However, the approaches of many intersections in Chinese cities are not able to be broadened practically in accordance with the design standards, especially in old towns where the boundary lines of roads are generally too narrow and there are no wide median barriers. Increasing the number of lanes approaching the intersection by offsetting the centerline and squeezing the exit lane width would narrow the lane widths in both the entrance lanes and the neighboring exit lanes. In the case that the width of a road profile is limited, a feasible approach is to reduce the lane width of the approach to add a lane (Gai, 0). At present, traffic engineers in China widely adopt this road marking design. The most common application is the road marking design of three lanes upstream, followed by four lanes in the approach to the intersection, in a single direction. In the design, there is a changeover portion between the upstream road segment and the approach, which is blank and without road markings that indicate the driving trajectories of vehicles, as shown in Fig. 1. The purple part is the upstream road segment, the pink part is the changeover portion referred to as discontinuous lanes and the yellow part is the immutable lanes of the approach. Although national standards of road design have no specifications for the road markings of discontinuous lane design, it is employed in most of the Chinese cities. For example, by a field survey, we found that more than half of the intersections with over three lanes in an approach in the urban area of Changsha follow this type of discontinuous lane design. Fig. 1. Schematic diagram of the discontinuous lane design before an approach Given the lack of specifications for discontinuous lane design in the national standards of road marking design, there is a research need to reveal the advantages and disadvantages of a discontinuous lane design by evaluating the effects of this

4 empirical method on both safety and efficiency. Such evaluation can also help improve discontinuous lane design and provide a scientific, theoretical foundation for standardizing road marking design at signalized intersections. Although the irregular design is prevailing in China where the urban traffic system is in developing with unbalanced traffic flows, it is rarely found in developed countries. To our knowledge, there is no comprehensive study on this topic. In China, Liu et al. (00) briefly described discontinuous lane design in an analysis of the safety of the design of the left-turn lane. They pointed out that compressing the lane width to add a left-turn lane might result in non-correspondence between lanes of the upstream road segment and lanes of the approach, and might bring traffic into conflict when vehicles choose a target lane of approach. Because of the difficulty of connecting lane markings between the upstream road segment and the approach, there is a blank portion without road markings that is a disadvantage to the safe driving of vehicles. However, their research object was the left-turn lane, and they did not evaluate discontinuous lane design, though they mentioned its drawbacks. This study presents a quantitative evaluation of the safety and efficiency of the discontinuous lane design by analyzing the behavior of vehicles driven in a discontinuous lane design before the approach of an intersection, and studying the severity of vehicle conflict and its influencing factors, by employing an ordered probit model. Upon the evaluation results, the paper presents recommendations for the improvement of such lane design and the applicable conditions for continuous lane design based on the volatility of turning flow.. Safety evaluation model Due to the lack of credible and detailed intersection crash data, this study proposes a surrogate safety analysis by using traffic conflict observation. Surrogate safety analysis commonly employs two primary indexes to measure the safety risk, namely the time to collision (TTC) and the post-encroachment time (PET). The TTC is the time to collision if two vehicles continue at their present speeds and directions (Sayed et al., 1), whereas the PET is the elapsed time between the end of a preceding vehicle departing and the front of a trailing vehicle arriving at the point of conflict (Pirdavani et al., 0; Souleyrette and Hochstein, 0). Both indexes can reflect the level of conflict severity, and a lower TTC or PET means a higher possibility of collision. The PET is adopted as the analysis index in the present study because it is convenient and accurate to take PET values from traffic flow videos, while measuring the TTC requires knowledge of the instantaneous speeds of vehicles at each moment, which are difficult to estimate accurately. There are three levels of conflict severity potential conflict, slight conflict, and serious conflict and the PET values are ordinal in nature. When the dependent variable takes more than two values, and these values have discrete and natural ordering, the ordered probit model is appropriate for analysis (Abdel-Aty, 001). Hence, this statistical model is applied here to explore the conflict severity of a discontinuous lane design and its potential influencing factors. The dependent variable representing the level of conflict severity is unobserved, and is expressed as, (1)

5 where is a variable depending on the explanatory variable, which is an observed variable, and is the vector of coefficient corresponding to the explanatory variable. is the number of observed variables and is an error term assumed to follow a standard normal distribution. Therefore, the relationship between the unobserved dependent variable and the observed variable can be stated as: if the unobserved dependent variable takes a value between two fixed thresholds, then it corresponds to a certaint choice of the observed dependent variable. This paper uses three levels of the observed dependent variable: i.e., a value of 1 denotes potential conflict, a value of denotes slight conflict and a value of denotes serious conflict. The relationship of the unobserved dependent variable and observed variable can be specifically expressed as 1, If, If, (), If where threshold values, are parameters to be estimated and they satisfy. The ordered probit model can provide threshold values representing the levels of conflict severity. The conditional distribution of the conflict severity of the discontinuous lane design and explanatory variables is decided by the threshold values ( 1,) combined with the distribution of the free error term, and the probabilities of the conflict severity of the discontinuous lane design are thus determined as 1 Φ, () Φ Φ, () 1Φ, () where Φ. is the standard normal cumulative distribution function. Parameters,, and can be estimated by maximum likelihood estimation.. Data collection and processing.1. Data collection To quantitatively study the safety and efficiency of discontinuous lane design, field data were collected for four intersections on major arterials of the city of Changsha in China. These intersections satisfy four conditions: (1) the intersections with one leg adopting discontinuous lane design have a similar design of approach, () traffic flow is unsaturated so that the drivers can randomly choose the two through lanes of the approach when they travel toward the intersection, () there is little interference from non-motorized vehicles and pedestrians, and () the intersections have the two upstream road conditions and the discontinuous lane design between them, described as three lanes upstream followed by four lanes of approach before the intersection, as shown in Fig.. Details of the selected intersections are given in Table 1. In Table 1, all the four intersections investigated are numbered for convenience. Nos. 1,, and represent the intersection of Tongzipo Rd and Gufeng South Rd, Tongzipo Rd and

6 Jinxing Center Rd, Yinshuang Rd and Jinxing North Rd, and Dujuan Rd and Jinxing North Rd respectively. Fig. (a) shows the road marking design for a limited road width, while Fig. (b) shows the road marking design without a limitation of road width. (a) (b) Fig.. Road marking design of three lanes upstream followed by four lanes of approach before an intersection. (a) Road marking design for a limited road width, (b) road marking design without a limitation of road width. Table 1 Characteristics of the four intersections Intersection Condition of road width Width of intersecti on (m) Length of discontin uous zone (m) Speed limit (km/h) Width of lanes of upstream road segment (m) Width of lanes of approach (m) No. 1 limited.. No. limited.. No. unlimited No. unlimited As shown in Fig., the intersections are not broadened and the center line is not offset to guarantee a standard lane width of the approach. Instead, to improve traffic capacity, the number of lanes in the approach is increased by merely compressing

7 1 1 the lane width, with no road markings in a blank portion of road with a length of approximately m connecting the upstream road segment and approach. The traffic flow characteristics were captured on video cameras for the whole observation period. During data collection, a camera shot video in the direction of the approach to satisfy the study requirements. The coverage of each camera was from the stop line of the approach to the upstream road segment, approximately 0 m further away. Fig. illustrates one observation site (the No. 1 intersection), with the green part indicating the camera coverage area. Videos of the No. 1 and No. intersections were taken with an elevated camera, giving a tilted perspective, whereas videos of the No. and No. intersections were obtained by an intelligent transportation system maintained by traffic police, which filmed the approach with a surveillance camera installed at the top of a building near each intersection. Furthermore, to guarantee the condition of unsaturated traffic flow, data were collected during off-peak hours; that is, :00 :00 and 1:00 1:00 from January (Monday) to January (Thursday), 01. A 1 0m B Fig.. Simplified configuration of one observation site.. Feature parameter extraction In the design as shown in Fig., two of the four lanes in the approach are through lanes, and the volumes of through traffic are far more than those of turning traffic. Almost all turning vehicles changed lanes in the upstream road segment under the guidance of road signs, and drove in their target lane before entering the discontinuous zone. As a result, the possibility of crossing conflict between turning and through traffic was relatively low in the discontinuous zone. Therefore, we focused on conflict among vehicles in though traffic when extracting the feature parameters, i.e. PETs. As through traffic volumes are far greater than those of turning traffic, a single straight-through lane in the upstream road segment cannot satisfy the through traffic demand. Consequently, vehicles travelling straight ahead usually take over the turning lanes in the upstream road segment, such that three arrays of vehicles travelling straight ahead merge into two arrays when the vehicles enter the approach. The possible driving trajectories and conflict points are shown in Fig..

8 Fig.. Driving trajectories and conflict points of vehicles travelling straight ahead Vehicles travelling straight ahead in each lane of the upstream road segment have two target lanes of the approach from which to choose. Drivers must make selection decisions before driving into the discontinuous lane segment, and these decisions are affected by the road traffic volume, the queue length of the target lane in the approach, the queue length of the neighboring target lane, the lane used by the vehicle in the upstream road segment, the speed of the vehicle before entering the discontinuous zone, and other factors. If two drivers in different lanes of the upstream road segment choose the same target lane of the approach at the same time, or choose target lanes of the approach that are not connected to their driving lanes, then a traffic conflict may occur. In the case of the driving trajectories of vehicles travelling straight ahead, discontinuous lane design may generate eight decision conflict points: three diverging conflict points, two merging conflict points and three crossing conflict points. For convenience, road lanes and velocity measuring points are numbered (see Fig. ). The driving trajectories of all vehicles can be classified according to the lane combination of the vehicle before and after the discontinuous zone, which can be seen in the videos. Vehicles having a driving trajectory from one of lanes 1//, entering the discontinuous zone and finishing in one of lanes ///, belong to one category (i.e., set), and there are thus a total of trajectory sets (i.e., a vehicle from lane 1 moving to lane, a vehicle from lane 1 moving to lane, a vehicle from lane 1 moving to lane, a vehicle from lane 1 moving to lane,, a vehicle from lane moving to lane ). We first recorded the number of vehicles in each trajectory set. Then, the trajectory sets were divided into three subsets according to the lane used in the upstream road segment by each vehicle, and the number of vehicles from the same lane of the upstream road segment was calculated as a cluster-class number. There are in total three cluster-class numbers corresponding to the three lanes of the upstream road from which the vehicles came. The total number of vehicles travelling along the road is the sum of the three cluster-class numbers. Each number of vehicles with the same lane in the upstream road segment and the same driving trajectory was divided by the corresponding cluster-class number to give the decision percentage of drivers whose vehicle came from the same lane of the upstream road segment; e.g., the

9 decision percentage of drivers who drive into different target lanes of the approach from lane 1 of the upstream road segment. We calculated the sum of the traffic volumes of all lanes in the upstream road segment, and multiplied the decision percentage of a single lane by the proportion of the traffic volume on the road that was in that single lane. A total decision percentage for drivers deciding to move from one lane of the upstream road segment to another lane of the approach was obtained (e.g., the total decision percentage of vehicle drivers in lane 1 moving to lane ). We then multiplied the total decision percentages for two trajectories where vehicles generate a conflict, and the decision conflict probability of two trajectories was obtained. Finally, adding the decision conflict probabilities of the same type of conflict, the possible decision conflict probability of each type of conflict was obtained. Decision conflict refers to the latent conflict result for vehicles in different lanes of the upstream road segment choosing to enter the same target lane of the approach, or the latent conflict result for vehicles in different lanes of the upstream road segment choosing to cross to target lanes of the approach that are not connected with their current driving lanes. Table Decision conflict probability Intersection Decision conflict probability (in No. lane) Decision conflict probability (in No. lane) Decision conflict probability (inside of discontinuous zone) No No No No The field data given in Table reveal that the decision conflict probability for vehicles entering lanes and of the approach is higher than other decision conflict probabilities. The present paper thus only focuses on recording the conflicts at the entrance of through-lanes and of the approach as the feature parameter represented by the PET.. Safety analysis The No. and No. intersections were chosen as control sites. The difference between the time that the preceding vehicle left the conflict point and the time that the trailing vehicle reached the conflict point was estimated as the PET. A total of 10 PET data points were obtained; the maximum value was. s and the minimum value was 0. s. As shown in Table, it can be intuitively seen that the overall safety level of continuous lane design is higher than that of discontinuous lane design; the mean PET value for continuous lane design is.01 s, while the mean PET value for discontinuous lane design is. s. Furthermore, the standard deviation demonstrates that the dispersion degree of discontinuous lane design is relatively larger than that of continuous lane design.

10 1 Table Comparison of discontinuous lane design and continuous lane design through PET values PET values Maximum Minimum Mean Variance Std. Deviation Continuous lane design.0 s 1.0 s.01 s Discontinuous lane design. s 0. s. s Fig. presents the frequency histogram and cumulative distribution of the PET values. The histogram shows that most PET values are much lower than a threshold of s, and the values are mainly concentrated around.0 s. Alhajyaseen (0) proposed a new safety measure called conflict index that considered crash probability as well as severity, which is evaluated using a combination of speed and PET. It was found that conflict indices above the th percentile have significant exponential relationships with severe crash numbers. Based on the cumulative distribution of the PET, the 1th and th percentiles of PET values were set as critical values of conflict severity; these percentiles corresponded to PET values of 1.1 and. s respectively. According to the division of the conflict severity level, statistical results are presented in Table Fig.. Frequency histogram and cumulative distribution of the PET Table Statistical results for conflict severity level Level of severity PET range (second) Sample size (%) Risk of collision Potential PET>. (1.) Low risk Minor 1.1<PET<. (.) Moderate risk Serious PET<1.1 0 (.) High risk Conflict severity for the discontinuous lane design is not only related to driving behavior but is also affected by the road traffic volume, the queue length of the target lane in the approach, the queue length of the neighboring target lane, the lane used by the vehicle in the upstream road segment, the vehicle speed before the vehicle enters the discontinuous zone, and other factors. Thus, the conflict severity generated by vehicles travelling straight ahead is taken as the dependent variable, and the road traffic volume, the queue length of the target lane in the approach, the queue length of the neighboring target lane, the lane used by the vehicle in the upstream road segment

11 and the vehicle speed before entering the discontinuous zone (at point B in Fig. ) are regarded as explanatory variables. The definitions and values of influencing factors are given in Table. Table Definitions and values of explanatory variables associated with the PET Explanatory variables Description of variables Road traffic volume Queue length of target lane in approach Queue length of the neighbor target lane Speed Lane used by vehicle in upstream road segment The total traffic volume in a signal period when a conflict happened The queue length of target lane of approach which existed when a conflict happened The queue length of through lane next to the target lane which existed when a conflict happened The current speed of vehicle before entering discontinuous zone Whether vehicles in upstream road segment are in the straight lane or not (0 if yes, 1 if no) The descriptive statistics of the relevant variables are as shown in Table. The standard deviation of road traffic volume is larger than that of the other variables, at., while the standard deviations of the others are less than., suggesting that the variables include different characters of dispersion degree. Here, dispersion degree relates to whether the data points are clustered or spread out, and is indicated by standard deviation. It is defined as that the larger standard deviation denotes the higher dispersion degree of data. Besides, the differences in queue length of the target lane and the neighbor target lane of the approach are relatively small; their minimums and maximums are the same, and they have similar standard deviations. Furthermore, the dispersion degree of speed is similar to that of the queue length of the target lane and the neighbor target lane of the approach. Also, the standard deviation of PET values is close to 1, which shows the data are clustered. The dispersion degree of PET values coincides with the data shown in Fig. which demonstrates that the PET values are concentrated on the range of 1- s.. Table Descriptive statistics of relevant variables Variables Minimum Maximum Mean Statistic Std. Error Std. Deviation PET (s) Road traffic volume (per unit) Queue length of target lane in approach (per unit) Queue length of the neighbor target lane (per unit) Speed (m/s)

12 Table Preliminary analysis of conflicts Conflict number Traffic volume Proportion of conflicts in total traffic volume Number of vehicles which are not aimed at the target lanes of approach Proportion of vehicles which are not aimed at the target lane of approach in total conflicts If the observation data recorded by the camera are divided into several sets based on the observation time of every 1 minutes, then the number of PET values which are in the range of minor conflict and serious conflict can be defined as the conflict number according to the standard of classification in Table. The number of conflicts and data for relevant influencing factors are given in Table. This preliminary analysis illustrates that the number of conflicts between vehicles travelling straight ahead is relatively large, exceeding 0% of the total traffic volume during the observation period. Moreover, vehicles that did not drive in the lane of the upstream road segment that aligns with the target lane of the approach account for a large portion of the obtained conflicts between vehicles travelling straight ahead; this proportion exceeded 0% and even reached % during the observation period. Table Parameter estimation of the ordered probit model Variables Coefficients % Confidence interval Lower bound Upper bound p-value Road traffic volume <0.001 Queue length of target lane in approach Speed <0.001 Lane used by vehicle in upstream road segment < < Log Likelihood 1.1 Chi-square. Sig. <0.001 Further to this, the ordered probit model was adopted to analyze the relationship between the set of explanatory variables and the conflict severity of discontinuous

13 1 1 lane design, employing SPSS software to estimate the parameters. Table presents the final resultss of the ordered probit model, showing only the statistically significant variables (p-value < 0.0).. The ordered probit model identifies four factors, for a confidence interval of %, with positive coefficients estimated, all of whichh have a negative effect on the safety of discontinuous lane design. The factor which has the maximum estimated influence coefficient is the lane used by the vehicle in the upstream road segment. In other words, if theree are more vehicles needing to choose a target lane before entering the approach, then the risk of traffic conflict is higher.. Efficiency analysis The speed and acceleration of 0 cars for the No. intersection were primarily analyzed so as to observee effects on traffic efficiency due to the discontinuous lane design. A video processing software was used to record the speed and acceleration of each car that smoothly travels through the two vertical sections during the green signal phases. By counting the number of cars at each level, the histogram in Fig. presents the distribution of speed reductions beforee vehicles entering the discontinuous zone. In addition, the varying acceleration of each straight-ahead vehicle is captured as shown in Fig.. NO.(per) 1 0 straight- turning- vehiclee vehiclee Acceleration(m/s) Deceleration degree(%) Fig.. Distribution of deceleration of vehicles 1 11 Serial-number of vehicle position Fig.. Varying acceleration of each straight-ahead vehicle Vehicle 1 Vehicle Vehicle Vehicle Vehicle Vehicle Vehicle Vehicle Vehicle Vehicle Vehicle 1 Vehicle Vehicle Vehicle Vehicle 1 Vehicle 1

14 Fig. demonstrates that vehicles passing through the intersection during the green signal phase are still decelerating strongly as they reach the discontinuous zone. There are a highest probability by 0 0% of deceleration rate for straight-ahead vehicles, and the highest deceleration rates of turning vehicles appears in deceleration degree interval 0% 0%. Comparing the deceleration rates of turning vehicles with the deceleration rates of straight-ahead vehicles, it could be inferred that the net difference may be the impact of discontinuous zone, since turning vehicles are seldom impacted by the discontinuous zone. Furthermore, Fig. demonstrates that most straight-ahead vehicles decelerated upstream of the discontinuous zone when selecting their target lane of the approach. The positions of vehicle deceleration and the driving trajectories infer that drivers sharply decelerate when selecting their target lane of the approach. This result greatly counteracts the expected improvement of traffic efficiency from increased lanes.. Suggestions of Traffic Engineering.1. Suggestions for the improvement of discontinuous lane design For the purpose of reducing conflict probability and efficiency loss, it is better to adopt continuous lane design. A continuous lane reminds drivers in the upstream road segment to choose their target lane of the approach ahead of time, and thus allows drivers to smoothly enter the target lanes of the approach. If the right-turn volume is large while the left-turn volume is relatively small, then the road marking design shown in Fig. (a) can be adopted; i.e., the lane distribution of the upstream road segment is a combination of left-turn straight-ahead, straight ahead, and right-turn lanes. The straight-ahead lane of the upstream road segment connects with the straight-ahead lane of the approach that is closest to the right-turn lane. In contrast, if the left-turn volume is large while the right-turn volume is relatively small, then the road marking design shown in Fig. (b) can be adopted; i.e., the lane distribution of the upstream road segment is the combination of left-turn, straight-ahead, and right-turn straight-ahead lanes. The straight-ahead lane of the upstream road segment connects with the straight-ahead lane of the approach that is closest to the left-turn lane. When both the left turn and right turn volumes are large and the queuing lengths reach the discontinuous zone and block vehicles travelling straight ahead, it is better to adopt the road marking design shown in Fig. (c); i.e., the lane distribution of the upstream road segment is a combination of left-turn, straight-ahead, and right-turn lanes. The straight-ahead lane of the upstream road segment connects with the two straight-ahead lanes of the approach, such that vehicles travelling straight ahead choose the target straight-ahead lanes of the approach before entering the approach.

15 (a) (b) (c) Fig.. Suggestions for the improvement of the continuous lane design. (a) Road marking design of a continuous lane design when the right-turn volume is larger, (b) road marking design of a continuous lane design when the left-turn volume is larger, (c) road marking design of a continuous lane design when both turning volumes are large... Applicable conditions of continuous lane design To intensively analyze the effect of flow structure characteristics on the road marking design adopted, the present study also collected data on the left-turn and right-turn traffic volumes in each signal phase for the four intersections. The total statistical time period was one hour. In total, sets of left-turn and right-turn volume data were obtained. The turning rate, which is the ratio of the right-turn volume to the left-turn volume, was calculated for each signal period, as shown in Table. It can be 1

16 observed that the dispersion degree of turning rates at No., No. and No. intersections are relatively higher than at No. 1 intersection, with standard deviations of.0,.0 and. respectively. The dispersion degrees of the left-turn traffic volumes of these intersections are higher than that of the right-turn traffic volumes, while the No. 1 intersection has relatively small variation across the total sample data. Table Descriptive statistics of the turning flow for four intersections Intersection Category Minimum Maximum Mean Std. Std. Statistic Deviation Error Left-turn volume No. 1 Right-turn volume Turning rate Left-turn volume No. Right-turn volume Turning rate Left-turn volume No. Right-turn volume Turning rate Left-turn volume No. Right-turn volume Turning rate Though the descriptive statistics of the turning flow demonstrate that No. 1 intersection shows the least deviation in right/left turns, since the reported measures are closer to 1, the variation between left-turn volume and right-turn volume is the main concern. Turning flow volatility is used to describe the varying relationship between the left-turn and right-turn traffic volumes. When the volume of one of left-turn or right-turn traffic is always more than the other in each signal period, then the turning flow volatility is relatively small; otherwise, the turning flow volatility is large. Table Analysis results of average turning rates Intersections No. 1 No. No. No. Average turning rate (per signal period) Average turning rate (per 1 minutes) 1 Total average turning rate

17 Average turning rate (per one hour) From the perspective of turning flow volatility, No.1 intersection shows large fluctuations, with the maximum turning flows in different signal phases located in different directions. Comparatively, the distributions of the No., No. and No. intersections are relatively stable. The final results of the average turning rate are given in Table, which presents the average turning rates for 1 minutes and one hour, from which we obtain the volatility difference of the average turning rates for different statistical periods. Table shows that the average turning rates during one signal period fluctuate more than those in the other two statistical periods, especially at No. intersection. It illustrates that the volatility of turning volumes is sensitive to the statistical period. When considering the suggestion made in this paper to use a continuous road marking design, the performance of which is affected by the turning rate of vehicles, a traditionally-used statistical period of 1 minutes or one hour would insufficiently reflect the volatility of the turning volume that is used in the planning of signal control. As a result, to guarantee an optimal design, the total average turning rate in a signal period could be adopted as the threshold; i.e.,.. The suggestion of a continuous lane design in this paper is suitable only if the average turning rate calculated in each signal period exceeds the threshold. Thus, among the four intersections, the No., No. and No. intersections suit the use of the continuous lane design. It is better to maintain the original discontinuous lane design for No. 1 intersection as the turning flow in the two directions fluctuated significantly, so as to increase the intersection traffic capacity. The saturation flow is then calculated on the basis of the basic saturation flow of the turning lane and its relative correction coefficients. Using the average turning rate threshold obtained above from the sample survey, a diagram for selecting different designs can be drawn (see Fig. ). When the average turning rate of the right-turn volume and left-turn volume during a signal period is larger than. or less than 0. throughout the total observation time, the suggestion of a continuous lane design is suitable. In contrast, if the average turning rate is in the range of 0.., it is better to use the discontinuous lane design that is widely used at present, because fixed continuous lane design is unsuitable for fluctuating flow demands of turning vehicles per signal period. Note that the average turning rate threshold used here is obtained from the sample survey of four intersections, and may not be applicable to other intersections. The turning rate threshold proposed in this paper is thus not strictly fixed. In Fig., the straight line boundary is the threshold of the average turning rates of. and 0. obtained by sample survey, while the curve boundaries near the both sides of the straight line boundary define a fuzzy region, which is suitable for either of the design plans. The analysis of different turning flow rates in one signal period shows that it is better to adopt the continuous lane design only if the average turning rate satisfies a condition, i.e., a stable turning flow structure, which is defined as above or below the thresholds of average turning rates of. and 0. respectively in this paper. On the basis of a stable demand of turning flow and straight flow, the lanes of the upstream road segment can be made continuous with the lanes of the approach and a definite right of way for vehicles is determined. 1

18 Fig.. Selection diagram of different designs. Conclusions This paper presents a quantitative analysis of the road safety and efficiency of the engineering practice of applying a discontinuous lane design between lanes of an upstream road segment and lanes of an approach to an intersection that is widely used at present in China. The main findings of the study are as follows. (1) Discontinuous lane design gives rise to the phenomenon that drivers have to make lane selection decisions, which may bring about decision conflict. Based on the field research, the decision conflict probability for vehicles entering straight-through lanes of the approach is higher than other decision conflict probabilities, i.e. the latent conflict probabilities of vehicles in different lanes of the upstream road segment choosing to enter the same target lane of the approach is higher. () The lane used by the vehicle in the upstream road segment is one of the main influencing factors of conflict severity for discontinuous lane design. The primary reason is the ambiguous right of way of each lane in the upstream road segment associated with the discontinuous lane design, requiring vehicles to merge into two target lanes of the approach from three lanes of the upstream road segment. () Comparison of speed variation between straight-ahead vehicles and turning vehicles suggests that discontinuous lane design negatively affects capacity efficiency, as most straight-ahead vehicles decelerated by 0 0% while the highest deceleration of turning vehicles was 0% 0%. Besides, the acceleration variation of each straight-ahead vehicle also demonstrates that discontinuous lane design is to the disadvantage of capacity efficiency improvements. () Continuous lane design is suggested according to the average turning rates of turning vehicles during a signal period. When the average turning rate of the right-turn and left-turn traffic volumes during a signal period in the total observation time is clearly greater than., or clearly less than 0., the continuous lane design is suitable. In contrast, if the average turning rate is in the range of 0.., the discontinuous lane design widely used at present is better. However, in an oval range near the thresholds of average turning rates of. and 0., the performance

19 difference between the two designs is comparatively small, and either of the two designs would be suitable. Acknowledgements This work was jointly supported by Natural Science Foundation of China (No. ) and National Social Science Foundation of China (BTJ01). References Abdel-Aty M., 001. Using ordered probit modeling to study the effect of ATIS on transit ridership. Transportation Research Part C: Emerging Technologies, 0, -. Alhajyaseen W., 0. The development of conflict index for the safety assessment of intersections considering crash probability and severity. Procedia Computer Science., -1. Gai C.Y., 0. Rethink of intersection red line broadening in Beijing. In: Urban Age, Collaborative Planning 0 Urban Planning Conference in China, Qingdao, Shandong, China. Liu H.Q., Wu Y., Shen T., Zhang W.H., 00. Research on left-turn lane safety design of intersection. The Chinese and Foreign Road, 0, -0. Pirdavani A., Brijs T., Wets G., 0. Evaluation of traffic safety at un-signalized intersections using microsimulation: a utilization of proximal safety indicators. Advances in Transportation Studies,, -0. Sayed T., Brown G., Navin F., 1. Simulation of traffic conflicts at un-signalized intersections with TSC-Sim. Accident Analysis and Prevention,, -0. Souleyrette R., Hochstein, J., 0. Development of a conflict analysis methodology using SSAM. Center for Transportation Research and Education, Ames, Iowa State University, In Trans Project -. 1