USE OF SPEED PROFILE AS SURROGATE MEASURE: EFFECT OF TRAFFIC CALMING DEVICES ON CROSSTOWN ROAD SAFETY PERFORMANCE Ana Tsui Moreno PhD Candidate, Department of Transportation, Universitat Politècnica de València Camino de Vera, s/n. 46022 – Valencia. Spain, e-mail: [email protected] Alfredo García Professor, Department of Transportation, Universitat Politècnica de València Camino de Vera, s/n. 46022 – Valencia. Spain, e-mail: [email protected] Submitted to the 3rd International Conference on Road Safety and Simulation, September 14-16, 2011, Indianapolis, USA ABSTRACT Urban road safety management is usually characterized by both a lack of quantity and quality of crash data and low budgets. However, fifty three percent of road crashes in Spain take place on crosstown roads and urban areas. Moreover, ten percent of fatal crashes on urban areas occur on crosstown roads. In order to reduce both crash frequency and severity, traffic calming measures (TCMs) are often implemented on crosstown roads. The objective of the research is to develop a methodology using continuous speed profile on free- flow conditions to evaluate safety effectiveness of traffic calming measures on crosstown roads. Given the strong relationship between speed and crash experience, safety performance can be related to speed. Consequently, speed can be used indirectly as a surrogate safety measure. Two indexes were defined as surrogate safety measures based on the continuous speed profile: Ra and Ea, related to speed uniformity and speeding, respectively. The indexes were applied to both individual observed speed profiles and aggregated operating speed profile. Twenty four global values of both indexes were obtained. The scenarios with implemented TCMs according to technical criteria, such as a traffic calming density close to nine TCMs per km, presented lower values. Age and gender differences have also been evaluated. More scenarios based on speed predictions over the TCMs will be modeled to propose Ra and Ea safety threshold values. The paper explores continuous speed profiles obtained from naturalistic driving to assess safety performance of crosstown roads with traffic calming devices. With this approach, speed is in effect used as a surrogate safety measure formulated in two new indexes: Ra and Ea. Key words: Surrogate safety measure, traffic calming, speed profile, consistency, speeding. 1 INTRODUCTION Improving road safety in both urban and rural areas is a major objective of the Spanish General Directorate of Traffic. In 2009, fifty three percent of severe road crashes, which includes injury and fatal crashes, took place on urban areas. Despite severe crashes on crosstown roads represent only one point three percent of severe crashes on urban areas, ten percent of fatalities occur on this specific road type (Dirección General de Tráfico, 2010). Crosstown roads are the part of a two-lane rural road which goes through a populated area. Consequently, drivers should adapt their driving from rural road conditions to urban environment. Crosstown roads are common in Europe and they are characterized by low-median traffic volume: annual average daily traffic between 500 and 8,000 vehicles per day, which results in a relatively low number of crashes. Besides, databases have usually lack of reliable data. Thus, traditional urban road safety management based on road crashes may not be the most appropriated approach due to a lack of statistically significance of crash data. Surrogate safety measures based on roadway characteristics are often defined to indirectly assess road safety management where historical crash data are limited or unavailable. In rural highways, the relationship between consistency and safety level was ascertained (Polus and Mattar-Habib, 2004; Polus et al., 2005). Some authors have developed surrogate measures relating speed variation and road safety on rural roads. Given the different purpose of the studies and data collection method, the definition of the surrogate measures in each case was different: from the difference between pre-crash and normal condition traveling speeds (Solomon, 1964), to the standard deviation of speeds (Garber and Gadiraju, 1989; Aljanahi, 1999); and the difference between the operating speed and the mean speed (Lave, 1985). Lamm and Choueiri (1995) proposed two criteria: operating speed difference between two consecutive elements; and difference between operating speed and design speed. This second criterion was incorporated into the Interactive Highway Safety Design Model (IHSDM) on the design-consistency module. Cafiso et al. (2007) used Lamm and Choueiri measures to assess two-lane rural road consistency on the Italian road network. Nevertheless, the former surrogate measures were calculated at one specific location and not along an entire roadway section. Polus and Mattar-Habib (2004) introduced the analysis of operating speed profile to evaluate consistency and safety level. The main hypothesis was that improved speed uniformity along a roadway section resulted in better quality and less strain in driving, thus improving safety. Two consistency measures were defined: the relative area bounded by the speed profile and the average weighted speed (Ra); and the standard deviation of operating speeds (σ). As design consistency increased, crash rates decreased significantly. Both consistency measures provided a similar assessment of consistency as Lamm and Choueiri measures. However, Polus and Mattar-Habib measures were calculated for the entire segment under investigation. The Polus consistency model was based on operating speed prediction models on curves and tangents; and estimates acceleration and deceleration rates. In urban areas, such as crosstown roads, developed operating speed models are fewer than in two-lane rural roads. In fact, only a few studies have developed operating speed models in low-speed urban streets (Poe et al, 1996;; Poe et al., 1998; Bonneson, 1999; Poe and Mason, 2000; Fitzpatrick et al., 2003; Wang et al., 2007). Poe et al. (1996) concluded that access and land use characteristics influenced on operating speed. A regression model carried out by Poe et al. (1998) showed that alignment and traffic control explained a large portion of the speed variation, although a high correlation 2 between both variables was detected. Fitzpatrick et al. (2003) found that posted speed limits were the most significant variable for both curve and tangent sections. Wang et al. (2007) used in- vehicle GPS technologies for the first time to determine operating speed on urban streets. They found that operating speed was influenced by number of lanes, roadside objects density, the density of T-intersections, raised curb presence, sidewalk presence, on-street parking, and land uses. Considering the previous models, operating speed profile of a crosstown road could be developed. However, the speeding problem along roads running through populated areas is usually handled by using traffic calming measures (TCMs), which were not included in the previous research. Given that TCMs involve traffic control at one location, specific operating speed models should be considered. Several studies have been conducted to evaluate effectiveness on speed reduction and operating speed over individual TCMs; and their results have been summarized on several publications (Department for Transport, 2004; Elvik et al., 2009; Ewing and Brown, 2010; Federal Highway Administration, 2009; Transportation Research Board, 2011). TCMs’ acceleration and deceleration rates have also been assessed. Barbosa et al. (2000) studied mean acceleration and deceleration rates by types of measures based on continuous speed profiles. Deceleration rates varied from -0.25 to -0.82 m/s2, while acceleration rates were set between 0.24 and 0.50 m/s2. Therefore, TCMs implementation usually produced an irregular speed profile with frequent decelerations and accelerations. The studies showed that spacing between TCMs was a key factor on speed reduction (Ewing et al, 1996; Ewing, 1999; Barbosa et al., 2000; Cottrell et al, 2004; Bassani et al, 2011). Hence, most of the guidelines and recommendations propose geometry and spacing of traffic calming devices to reduce speeding in urban areas. Then, the assessment of TCMs implemented along a segment is often characterized by average speed reduction rather than consistency of the resulted speed profile or accumulated speeding along the segment. Moreover, uniformity of a speed profile on a calmed crosstown road has never been assessed. Similar to rural roads, crosstown roads with adequate TCMs and optimal spacing would result in a more uniform speed profile and; therefore, more consistent design; which would likely lead to a safer crosstown road. This paper explores continuous speed profiles obtained from naturalistic driving to assess safety performance of calmed crosstown roads. OBJECTIVES The aim of the research was to develop a methodology using continuous speed profile to evaluate safety effectiveness of traffic calming measures (TCMs) on crosstown roads. The main objectives of the research were: to observe drivers’ behavior and characteristics on twelve different scenarios by using GPS trackers; to define two indexes as surrogate safety measure; to apply the measures to both individual observed speed profiles and operating speed profiles; and to analyze the measures’ values depending on the crosstown road characteristics, such as speed limit, operating speed or traffic calming density. Moreover, driver’s age and gender influence on the proposed indexes were also evaluated. It should be noted that TCMs were considered as an integrated system along an entire crosstown road rather than isolated or segregated measures. 3 FIELD STUDY Site selection For the research, six crosstown roads were selected. Five of the sites had TCMs installed; while the sixth location had no TCMs. The first five cross-town roads were selected according to the recommendations of a previous road safety study, taking into account: annual average daily traffic (AADT); length of the cross-town road (L); and type of existing traffic calming measures. The selected towns were: Albalat de la Ribera; Chelva; Genovés; Quatretonda; and Llutxent. TCMs included speed tables, speed humps and one roundabout. The sixth location was Belgida. No TCMs were initially installed on Belgida’s crosstown road. Two pedestrian crossings were located on a tangent section. The community asked the responsible agency to install TCMs to reduce speeding along the roadway. The road safety project was implemented by stages on Belgida, which allowed deducing individual effect of diverse TCMs on the continuous speed profile: one speed hump; two speed tables; one speed bump; one chicane; and one set of dragon’s teeth. A total of seven scenarios were considered: (0) no TCMs; (1) Southern speed table and speed hump construction; (2) Northern speed table and speed bump installation; (3) chicane construction; (4) dragon’s teeth construction; (5) Northern speed table removal; and (6) stage 5 after one year. Consequently, a total of twelve different scenarios with different TCMs type and location were observed. Crosstown road characteristics are summarized on Table 1. The posted speed limit was 40 or 50 km/h; while the AADT varied from 650 to 4,230 vehicles per day. Belgida’s AADT reduction from 1,920 veh/day to 1,180 veh/day was caused by the construction of a highway segment near the area, not the TCMs implementation because the new highway segment diverted traffic from the rural road which goes through Belgida. The length of the crosstown roads was between 560 and 945 m, considering only the urban area of the rural road. Thus, the length was obtained from the beginning to the end of the town. In order to classify the crosstown roads, TCMs density (TcD) was calculated. TcD was defined as the percentage of traffic calming devices per unit length. TcD equal to 1 meant that TCMs were spaced 100 m on average. Entrance gates were also considered as one TCM on the parameter, as well as curves with radius lower than 150 m, which also controlled speed below 50 km/h. Table 1 Scenarios characteristics Speed limit AADT Length Curves Number of Entrance TCM density (ud/m/%) Scenario (km/h) (veh/day) (m) (ud) TCMs (ud) gate (ud) Bound 1 Bound 2 Albalat 40 4,230 765 1 5 0 0.78 (E) 0.78 (W) Chelva 40 2,490 885 1 4 0 0.56 (W) 0.56 (E) Genoves 40 4,550 945 2 4 1 0.74 (E) 0.63 (W) Quatretonda 50 3,250 680 0 4 0 0.59 (E) 0.59 (W) Llutxent 40 2,930 690 0 4 0 0.58 (E) 0.58 (W) Belgida 0 50 2,650 560 1 0 0 0.18 (NE) 0.18 (SW) Belgida 1 50 1,920 560 1 2 0 0.54 (NE) 0.54 (SW) Belgida 2 50 1,920 560 1 4 0 0.89 (NE) 0.89 (SW) Belgida 3 50 1,180 560 1 4 1 1.16 (NE) 0.89 (SW) Belgida 4 50 1,180 560 1 4 2 1.16 (NE) 1.16 (SW) Belgida 5 50 1,180 560 1 3 2 0.98 (NE) 0.98 (SW) Belgida 6 50 1,180 560 1 3 2 0.98 (NE) 0.98 (SW) 4 All the crosstown roads presented good pavement conditions and their lane width varied from 3.10 to 3.25 m; which is common on Spanish crosstown roads. As crosstown roads were in urban areas, they were sidewalks on both sides of the road and on-street parking was allowed. Grade was nearly horizontal in all scenarios. Data collection To collect drivers’ behavior, passive GPS trackers were used. The available passive GPS tracking equipment recorded GPS location information and vehicle speed for one-second intervals. Consequently, an individual continuous speed profile and acceleration profile could be deduced. To collect the information at each scenario, two road controls were placed before drivers’ approach the town from each direction. The road controls were separated at least 1 km from the town to enable drivers to adapt their desired speed before entering the town. On each road control, drivers were asked to collaborate in a road safety study. Drivers were only told that a device had to be fitted on their vehicles and they were encouraged to drive in a normal way. Only passenger cars were taken into account. A survey was conducted at the first control to collect age, gender and vehicle type. At the other road control, drivers were stopped to return the device and were asked whether they had been influenced or not on their speed by another vehicle or pedestrian. The methodology was proven not to influence drivers’ speed selection with spot speeds verification measures before and during the observed time. Thus, drivers were not induced to reduce their usual speeds or behave differently. Tests were performed during morning period between 9:00 a.m. and 2:00 p.m., on a working day and with good weather conditions. In Belgida, data collection took place at least 14 days after the TCM implementation. Data reduction The data collected by the GPS trackers contain latitude, longitude, altitude, heading, time and date, every 1 second. After importing data from the devices, a coordinate’s conversion to UTM (x, y) was carried out before a successive data debugging process. Firstly, the data storage errors were found by analyzing the recorded time sequence. Secondly, a transversal positioning debugging was carried out. The diverted points were discarded. After, a longitudinal positioning debugging was done by taking into account abnormal speeds, accelerations or decelerations. Finally, vehicles which had left the track were discarded. Only free-flow conditions were considered; so, stopped vehicles and vehicles conditioned by other vehicles or pedestrians were removed from the sample. Therefore, traffic flow was not conditioning drivers’ speed; and, drivers were individually selecting their own desired speed. Three second headway is often considered enough to determine if a driver is driving at free-flow speeds. However, GPS did not record traffic conditions or headway. To overcome this interrupted flow problem, individual speed profile and 15th, 30th, 50th, 70th, and 85th percentile were plotted. Abnormal speed profiles were removed. Besides, the response on the second survey about conditioning was considered. Nearly 30 % of the initial sample was discarded due to non free- flow conditions, detour or stopping. The total sample was 925 drivers. Figure 1 shows an 5 example of 25 vehicles’ observed speed profile obtained from the GPS trackers in Belgida’s sixth scenario. Figure 1 25 Individual speed profiles of Belgida 5 -NW SURROGATE MEASURES Two indexes are defined as surrogate safety measures based on the continuous speed profile: Ra and Ea. The first measure evaluates uniformity of the speed profile; while the second measure assesses speeding along the entire crosstown road. Uniformity of the speed profile is not enough to consider a good design quality if speed level is higher than the speed limit. Both measures should be accomplished to determinate design quality. Ra is defined as the normalized relative area (per unit length) bounded between the speed profile and the average speed line. The measure can be applied to individual speed profiles or to the operating speed profile; considering operating speed as the speed which is only exceeded by 15% of drivers. The first step is to calculate the average speed of the speed profile along the crosstown road. Then, the areas bounded between the speed profile and the average speed lines (Ar) are i obtained (Figure 2a). The consistency measure is given as the sum of the absolute value of the areas divided by the length of the segment (L). Therefore, the inconsistency of speeds increases as Ra increases. Equation 1 can be applied. (1) where: Ra: relative area measure of uniformity (m/s) 6 ΣAr: sum of areas (absolute values) bounded between the speed profile and the average i speed (m2/s) L: entire crosstown road length (m). Ea is the normalized relative area (per unit length) bounded between the speed profile values higher than the speed limit and the speed limit line. The measure can be also applied to individual speed profiles or to the operating speed profile. Given the speed limit, the areas bounded between the speed profile and the speed limit lines are determined. Only the areas over the speed limit line (Ae) are considered in the measure (Figure 2b). The consistency measure is calculated applying i Equation 2 as the sum of the value of the areas divided by the length of the segment (L). Consequently, the higher the Ea, the higher speeding magnitude along the crosstown road. (2) Where: Ea: relative area measure of speeding (m/s) ΣAei: sum of areas bounded between the speed profile and the speed limit where speed is higher than speed limit (m2/s) L: entire crosstown road length (m). (a) 7 (b) Figure 2 Speed profile and surrogate measures of Belgida 5 - NW The proposed measures are calculated based on continuous speed profile rather than individual speed differentials between consecutive elements. They provide an evaluation of speed profile uniformity and accumulated speeding along the whole crosstown road under study. RESULTS AND ANALYSIS Each direction was analyzed separately, as TCMs type and quantity varied. At each scenario and direction (site), operating speed profile was obtained. Then, average operating speed, uniformity and speeding were calculated. At each individual speed profile, average speed, uniformity and speeding were calculated along the crosstown road. After, 85th percentile of each site was deduced. Consequently, two values were considered: (1) a global value from the operating speed profile; and (2) an individual value as the 85th percentile of the individual values’ distribution. Table 2 summarizes global and individual values of the analyzed variables. According to Poe et al. (1998), the use of aggregate statistics fails to recognize the probability distribution of the individual observed values. The apparent improvement in explaining variation of the parameter is given by the aggregation speed data; which may reduce the individual extreme values. Given that the methodology allows studying individual speed profiles, individual values were examined. However, practitioners may not have individual observations to determine design quality; and, only global estimates from the literature could be used to approximate the operating speed profile. Consequently, global values should also be analyzed. Thus, both individual and global values are considered on the following analyses. 8 Table 2 Average speed, uniformity and speeding results Speed Tc Vm (km/h) Ra (m/s) Ea (m/s) Site limit density (km/h) (m-1) Global Individual Global Individual Global Individual Albalat-W 40 0.78 40.68 38.82 1.31 1.68 0.75 0.56 Albalat-E 40 0.78 41.71 40.27 1.08 1.68 0.81 0.89 Belgida 0-SW 50 0.18 60.43 58.10 2.98 3.68 3.21 2.82 Belgida 0 -NE 50 0.18 64.53 63.62 3.18 3.64 4.23 4.24 Belgida 1-SW 50 0.54 53.98 54.79 3.72 4.18 2.39 2.41 Belgida 1-NE 50 0.54 55.95 53.26 4.04 4.05 2.73 2.22 Belgida 2-SW 50 0.89 48.33 44.58 2.94 3.36 1.30 1.17 Belgida 2-NE 50 0.89 48.42 45.98 3.27 3.72 1.50 1.50 Belgida 3-SW 50 0.89 44.66 43.56 2.17 2.53 0.65 0.68 Belgida 3-NE 50 1.16 43.42 41.80 1.30 1.86 0.11 0.16 Belgida 4-SW 50 1.16 45.27 43.49 2.33 2.67 0.77 0.63 Belgida 4-NE 50 1.16 44.43 41.88 1.29 1.93 0.13 0.16 Belgida 5-SW 50 0.98 48.51 46.77 2.06 2.45 0.90 0.93 Belgida 5-NE 50 0.98 46.79 45.05 0.98 1.88 0.18 0.22 Belgida 6-SW 50 0.98 50.94 49.74 2.33 2.59 1.55 1.29 Belgida 6-NE 50 0.98 48.84 47.83 1.08 1.79 0.42 0.49 Chelva-E 40 0.56 42.97 41.35 1.33 1.72 1.15 1.10 Chelva-W 40 0.56 43.10 41.30 1.54 2.06 1.28 1.13 Genoves-W 40 0.63 40.06 38.47 1.51 2.01 0.76 0.73 Genoves-E 40 0.74 42.35 40.95 1.69 2.13 1.14 1.09 Llutxent-W 40 0.58 45.50 43.90 0.98 1.52 1.56 1.44 Llutxent-E 40 0.58 43.61 43.14 1.25 1.63 1.22 1.15 Quatretonda-W 50 0.59 51.40 50.00 2.02 2.41 1.23 1.07 Quatretonda-E 50 0.59 48.06 45.82 2.10 2.66 0.80 0.68 Average speed Average speed reduction is a usual performance measure of traffic calming plans. The traditional analysis is to compare average speed before and after the TCMs implementation. A first analysis was carried out to determine the overall traffic performance on the evaluated sites. Average global and individual speed of the different sites was calculated from the beginning to the end of the crosstown road. 85th percentile of individual average speed was lower than average operating speed in all the sites. Aggregation of individual speed data into operating speed resulted in a higher value of average speed, as individual speed profiles distribution was not considered. Individual speed profiles may result higher at one location; however, the distribution along the entire road segment showed lower average values. Consequently, average operating speed was higher than 85th percentile of individual average speeds. However, global and individual values were close to each other and no statistical differences were found between them. 9 The analyzed sites presented the same number of lanes, roadside objects density, intersections density, raised curb, sidewalks, on-street parking, and residential land use. Therefore, according to Wang et al. (2007), the operating speed would be equal on all the sites. Nevertheless, the global and individual average speeds were different among the sites. The main difference between the sites was the posted speed limit, which differed from 40 to 50 km/h; and the implemented TCMs. Considering TCMs as traffic control devices, Poe et al. (1998) models application would result in different operating speeds; which agreed with the results. A Chi-square test was executed to verify if average operating speed had equal means among the sites. The resulted p-value indicated that the null hypothesis had to be rejected. Thus, average operating speed was statistically different between sites. A multiple regression analysis was performed to determine the variables that influenced on global and individual average speed. The multiple regression analysis showed that the values should be grouped depending on the speed limit. Speed limit was usually defined by crosstown road alignment. Moreover, traffic calming measures type and geometry has often been determined depending on speed limit. Given this common relationship among speed limit, crosstown road alignment and TCMs type, average speed analysis depending on speed limit also included crosstown road alignment and TCMs type. Hence, two groups were created: crosstown roads with speed limit equal to 40 km/h and crosstown roads with speed limit equal to 50 km/h (Figure 3). A linear regression was calculated within the two groups. The regression analysis of the crosstown with speed limit of 50 km/h showed that traffic calming density (TcD) explained 73% of the variability of the average operating speed. The coefficient of determination of the second group was only 31%. Consequently, average operating speed depended on TcD; which was an important parameter on average operating speed and average individual speed. Figure 3 Average operating speeds depending on speed limit and traffic calming density On the other hand, average operating speed was compared to speed limit. Only crosstown roads with TcD higher than 0.9 % m-1, which is equivalent to 9 traffic calming measures per km, presented an average operating speed lower than the speed limit. Thus, traffic calming density 10