^{1}

^{1}

^{1}

The purpose of this study is to assess the effects on air pollution that may derive from replacing a signal-controlled intersection with a roundabout, using a before-and-after approach. Based on field data collected with a test car instrumented with a Portable Emission Measurement System, the two intersection configurations were compared in terms of emissions of CO_{2}, CO, and _{2} and CO are generally lower for the roundabout than for the signal-controlled intersection, while an opposite result arises for

The impact on air quality of pollutant emissions produced by road traffic has become a central issue in the context of transportation system analysis and policy. Vehicular emissions depend on traffic, road, and vehicle characteristics, on atmospheric conditions, and on driving behavior, and it is well known that intersections are typically pollution “hotspots” because of the concentration of highly emitting modal activities (in particular vehicle acceleration) in their proximity [

Roundabouts are increasingly used worldwide and are often built to replace intersections previously controlled by traffic signals or stop signs. While several studies have shown that roundabouts can improve safety and, at least in certain flow ranges, the operational performance compared to other types of intersection control [

The main aim of this study is to investigate the effects on air pollution of replacing a signal-controlled intersection with a roundabout, using a before-and-after approach based on field measurements of vehicular emissions. To this end, data have been collected using an instrumented vehicle at a real road intersection where a traffic signal has been converted to a roundabout, so as to allow to compare the two forms of control in terms of emissions of selected pollutants. The research described in this paper represents a development of previous work by the authors: Meneguzzer et al. [

The contribution of the study described in this paper may be valuable for two reasons. First, the question of whether the type of intersection control has a significant impact on vehicular pollution is investigated on the basis of a before-and-after field experiment at a real intersection rather than using simulation tools, the latter being the approach most frequently adopted in previous studies on the subject. Second, in the statistical analysis of the experimental data we propose the use of a method that, to our knowledge, has not been employed by others in this area of research. More specifically, the existence of statistically significant differences in vehicular emissions between the two types of intersection is assessed by means of two-sample biaspect permutation tests, a nonparametric method that allows simultaneously detecting differences in location and variability characteristics of the distributions of the observations for the two forms of control.

The paper is organized as follows. Section

Several studies dealing with the analysis of vehicular emissions at road intersections can be found in the literature, some of them focusing in particular on the effect of the type of traffic control on emissions. Two main approaches have been adopted in these studies: either a modeling approach combining traffic microsimulation and vehicular emission models or an experimental approach based on direct field measurement of emission data.

Hydén and Várhelyi [

Unal et al. [

The relationship between vehicle dynamics and emissions for single-lane roundabouts was analyzed by Coelho et al. [

Mandavilli et al. [_{2},

Ahn et al. [

Hallmark et al. [_{2}, CO,

Papson et al. [

Jackson and Rakha [_{2}, HC, and

Vasconcelos et al. [_{2} and

Gastaldi et al. [_{10}, and total carbon to analyze a real four-leg intersection where a roundabout had replaced a fixed-time traffic signal. They found that the roundabout generally outperformed the fixed-time traffic signal in terms of vehicle emissions but noted that the difference between the two types of control was smaller in terms of environmental impacts than in terms of operational traffic performance.

An empirically supported macroscopic method for comparing vehicular emissions at roundabouts and signalized intersections was proposed by Salamati et al. [_{2}, and HC taking into account several factors, including demand-to-capacity ratio, signal timing, and signal progression characteristics. Results of an application to a real case indicate that roundabouts tend to be less polluting than traffic signals under low demand-to-capacity ratios; however, when demand approaches capacity, signalized intersections with favorable progression generate lower emissions than roundabouts.

The environmental impact of a sequence of roundabouts in a corridor was analyzed by Fernandes et al. [

The analysis of pollution at signal-controlled intersections has also been approached from a “design” viewpoint; for example, Li et al. [_{2}, CO,

Among the studies reviewed in the present section, those concerning comparative analyses of emissions under different types of intersection control are of major interest to our research and are summarized in Table

Summary of selected studies comparing vehicular emissions at intersections under different types of control (numbers in first column correspond to cited references) [

Reference | Method for estimation of emissions | Pollutants considered | Main findings |
---|---|---|---|

[ | Emission factors for observed speeds and accelerations | CO, | Emissions for small RBT are lower than for SIG and slightly higher than for UNSIG. |

| |||

[ | Instrumented vehicle | CO, CO_{2}, | Results are dependent on local conditions and time of day; emissions tend to be generally lower for RBT than for SIG and higher for RBT than for UNSIG. |

| |||

[ | aaSIDRA 2.0 software | CO, CO_{2}, HC, | Emissions are reduced following conversion of stop-controlled intersections to RBTs. |

| |||

[ | aaSIDRA 2.1 software | CO, CO_{2}, HC, | Emissions of all considered pollutants are lower for RBT than for AWSC. They are also lower for RBT than for SIG, but only up to a threshold of total intersection demand volume. |

| |||

[ | VISSIM software | CO, | Converting TWSC into RBT can reduce vehicle emissions under most traffic volume scenarios. |

| |||

[ | INTEGRATION and VISSIM software; VT-Micro and CMEM emission models | CO, CO_{2}, HC, | Emissions of all considered pollutants are higher for RBT than for SIG and TWSC. |

| |||

[ | Paramics software; MOVES and CMEM emission models | CO, | Emissions of considered pollutants are higher for RBT than for SIG under both light and congested traffic conditions. |

| |||

[ | Instrumented vehicle | CO, CO_{2}, HC, | Emissions for RBT are often higher than for SIG and AWSC. Strong effect of driver behavior on emissions. |

| |||

[ | INTEGRATION software | CO, CO_{2}, HC, | Emissions of CO, HC, and _{2} are dependent on overall demand and turning ratios. |

| |||

[ | S-Paramics and AIRE software | | Emissions for RBT are slightly lower than for SIG. |

| |||

[ | Vehicle-specific power and observed speed trajectories | CO, CO_{2}, HC, | Results of comparison of emissions for RBT and SIG depend on demand-to-capacity ratio and on quality of signal progression. |

| |||

[ | VISSIM software; vehicle-specific power and EMEP-EEA emissions methodologies | CO, CO_{2}, HC, | For a sequence of intersections along an arterial, emissions are lower for RBT than for SIG and higher for RBT than for TWSC. |

RBT: roundabout; SIG: signal-controlled intersection; UNSIG: unsignalized intersection; AWSC: all-way stop-controlled intersection; TWSC: two-way stop-controlled intersection; EMEP-EEA: European Monitoring and Evaluation Programme–European Environment Agency.

Data were collected at a four-leg road intersection where a roundabout has replaced a traffic signal (Figure

Signalized intersection characteristics [

Signal timing (North-South) | Approach characteristics | ||||||||
---|---|---|---|---|---|---|---|---|---|

Approach | |||||||||

Green | Amber | Red | Cycle | North | East | South | West | ||

Minimum | 49.0 | 3.0 | 31.8 | 85.8 | Entering lanes | 2 | 1 | 2 | 1 |

Median | 50.1 | 4.0 | 46.1 | 100.1 | Exiting lanes | 2 | 1 | 2 | 1 |

Maximum | 54.9 | 4.9 | 62.1 | 116.2 |

Roundabout characteristics [

General characteristics | Approach characteristics | |||||
---|---|---|---|---|---|---|

Approach | ||||||

North | East | South | West | |||

Inscribed circle diameter [m] | 36 | Entering lanes [#] | 2 | 1 | 2 | 1 |

Central island diameter [m] | 20 | Exiting lanes [#] | 1 | 1 | 1 | 1 |

Circulatory roadway width [m] | 8 | Splitter island width [m] | 4.50 | 3.80 | 4.20 | 4.20 |

Lanes in circulatory roadway [#] | 1 | Entry width [m] | 7.25 | 4.50 | 7.25 | 3.75 |

Aerial photograph of the study site: (a) signal-controlled intersection; (b) roundabout [

Field tests were carried out with a Fiat Panda Spark-Ignition (SI) bifuel (gasoline/natural gas) passenger car complying with Euro 4 emission standards. Fiat Panda is Italy’s best-selling car and it can be considered a point of reference of the city-car segment in Europe. The emissive behavior of this car in urban areas has been characterized by Meccariello et al. [_{2} and

On-road measurements of vehicle activity and emissions on a second-by-second basis can be obtained using Portable Emissions Measurement Systems (PEMS). Several studies have shown the effectiveness of PEMS as tools for collecting emission data that are representative of actual and typical vehicle use; see, for example, [

A Semtech PEMS from Sensors Inc. was installed on-board the test car for emission measurement. This device can produce the vehicle’s instantaneous emission profile and estimate the level of emissions produced while the car is running. The equipment consists of a tailpipe attachment, heated exhaust lines, a Pitot tube for measuring the exhaust mass flow and temperature, exhaust gas analyzers, a GPS Garmin 16x, sensors for ambient temperature and humidity, and exhaust pipelines. GPS and weather station were installed outside on the roof of the vehicle. An On-Board Diagnostics (OBD) Matrix from Texa, capable of communicating with the Electronic Control Unit (ECU) of the vehicle, was connected to the car’s OBD socket, in order to collect with 1 Hz frequency the following parameters: vehicle speed, rpm, gear ratio, and engine load. A video camera was also installed on-board to record video images of traffic from the viewpoint of the car driver. All parameters were measured with a time resolution of one second. Emissions were measured only in hot conditions, after a 45-minute preconditioning period necessary to let PEMS reach all the set-points.

Two data collection campaigns were conducted along Viale Mazzini, an urban corridor located in the city of Vicenza, Italy, before (April 1–3, 2014) and after (April 14–16, 2015) the conversion of a signalized intersection to a roundabout (Figure

Information on geometric characteristics of the two intersection configurations (see Tables

Data collected during the field tests were processed so as to obtain a proper time alignment of all signals acquired by the test vehicle (ECU, PEMS, and GPS data), which is essential for the calculation and verification of some test parameters as well as for the computation of pollutant emissions. GPS data were also processed to exclude anomalous and erroneous speed values. Following Rakha et al. [

To isolate the effect of the type of intersection control, vehicular emissions were measured over an influence area that included a 200 m-long segment within the test itinerary, consisting of 150 m upstream the stop/yield line and 50 m downstream the stop/yield line. In order to assess possible effects of the trip direction, test runs were coded as “Trip A” when the vehicle traveled from North to South and “Trip B” when the vehicle traveled from South to North. The itinerary was chosen in such a way that the test vehicle trips could be considered to be representative of the traffic stream having the highest traffic volume for both the N-S and S-N directions.

For each trip, the final dataset includes the following information (1 Hz frequency):

reference time;

vehicle position and speed, rpm, gear ratio, and engine load;

instantaneous CO_{2} (g/s), CO (mg/s), and

estimated fuel consumption (g/s);

exhaust mass flow (kg/h).

As shown in Table

Dataset of trips used for the analysis [

Intersection | Morning | Afternoon | Total | ||||
---|---|---|---|---|---|---|---|

Peak | Off-peak | Peak | |||||

Trip A | Trip B | Trip A | Trip B | Trip A | Trip B | ||

Signalized | 32 | 33 | 25 | 24 | 43 | 45 | 202 |

Roundabout | 39 | 39 | 27 | 28 | 31 | 30 | 194 |

Total | 71 | 72 | 52 | 52 | 74 | 75 | 396 |

Statistical analyses on the data collected at the study site were carried out in order to assess the impact on emissions of replacing signal control with a roundabout. First, a general picture of the distributions of the observations before and after the conversion of intersection control was obtained using descriptive statistics computed for the entire set of trips and for six subsamples defined by time of day, traffic condition, and trip direction. Second, the statistical significance of the differences between traffic signal and roundabout controls in terms of both location and variability indexes of the respective sample distributions was tested. Third, focusing on trips carried out in peak traffic conditions (representing about 75% of the entire dataset), binary logistic regression models were developed in order to quantify the effects of the factors that significantly affect vehicular emissions and hence to estimate the probabilities of emission levels associated with different trip profiles, in particular with different forms of intersection control. All the above analyses, which are described separately in the following, were performed for each of the three pollutants under examination (CO_{2}, CO, and

Descriptive statistics based on the entire data sample are reported in Table _{1}–Q_{3}), and coefficient of variation (CV). The values shown in Table _{2} and CO are higher, on average, for signal control than for roundabout, while the opposite is true for

Descriptive statistics for CO_{2}, CO, and

Signalized | Roundabout | Total | |
---|---|---|---|

Number of trips | 202 | 194 | 396 |

_{ 2 } (g): | |||

Mean | 59.62 | 51.75 | 55.76 |

SD | 19.45 | 18.91 | 19.56 |

Median | 59.88 | 45.46 | 53.05 |

(Q_{1}–Q_{3}) | (44.49–76.22) | (38.53–60.33) | (40.36–69.08) |

CV% | 32.6 | 36.5 | 35.1 |

| |||

| |||

Mean | 132.03 | 98.27 | 115.49 |

SD | 130.73 | 82.53 | 110.95 |

Median | 88.10 | 78.54 | 81.45 |

(Q1–Q3) | (39.48–181.67) | (37.30–133.49) | (39.20–155.69) |

CV% | 99.0 | 84.0 | 96.1 |

| |||

_{ X } (mg): | |||

Mean | 60.81 | 90.79 | 75.49 |

SD | 45.23 | 55.53 | 52.66 |

Median | 54.77 | 79.13 | 63.73 |

(Q1–Q3) | (23.96–86.56) | (46.98–127.64) | (35.94–107.35) |

CV% | 74.4 | 61.2 | 69.8 |

A more complete characterization of the relative environmental performance of the two intersection configurations is provided by Figures _{2} (Figure

Empirical Cumulative Distribution Functions of CO_{2} emissions (g) by intersection type and trip conditions.

Empirical Cumulative Distribution Functions of CO emissions (mg) by intersection type and trip conditions.

Empirical Cumulative Distribution Functions of

In order to determine whether the differences in pollutant emissions between the two types of intersection emerging from the preliminary analysis are statistically significant, we applied two-sample biaspect permutation tests [

In our specific application, we implemented a

Hence, on the one hand, partial tests may provide marginal information for each specific aspect; on the other, they jointly provide information on the global hypothesis. Thus, if a significant departure from

The overall adjusted _{2} and CO are lower for the roundabout than for the signalized intersection, while the opposite is true for

An explanation of these results may be provided by the distribution of total trip time among the four typical driving modes (idle, acceleration, deceleration, and cruise), which, as shown in Table

Driving mode definition and distribution of trip time among modes for signal and roundabout.

Driving mode | Speed [Km/h] | Acceleration [Km/h/s] | Time in mode | |
---|---|---|---|---|

SIG | RBT | |||

Idle | <0.1 | <0.1 and >−0.1 | 12.59 | 0.86 |

Acceleration | >0.1 | >0.36 | 34.14 | 39.93 |

Deceleration | >0.1 | <−0.36 | 25.80 | 41.97 |

Cruise | all other cases | 27.47 | 17.24 |

Two-sample biaspect permutation tests of emissions of CO_{2} (g), CO (mg), and

Trip condition | Signalized intersection | Roundabout | Test for location, | Test for variability, | Overall test, | Overall adjusted | ||
---|---|---|---|---|---|---|---|---|

Number | Mean ± SD | Number | Mean ± SD | |||||

_{ 2 } (g): | ||||||||

Morning/Off-peak/Trip B | 24 | 64.5 ± 16.5 | 28 | 49.1 ± 10.6 | 0.00025 | 0.00025 | 0.00025 | 0.00150 |

Afternoon/Peak/Trip A | 43 | 53.5 ± 19.5 | 31 | 41.3 ± 9.9 | 0.00265 | 0.00105 | 0.00165 | 0.00825 |

Morning/Peak/Trip A | 32 | 56.0 ± 21.4 | 39 | 45.2 ± 9.9 | 0.00595 | 0.00165 | 0.00305 | 0.01220 |

Morning/Off-peak/Trip A | 25 | 48.7 ± 18.9 | 27 | 40.3 ± 8.7 | 0.04275 | 0.01395 | 0.02435 | 0.07305 |

Afternoon/Peak/Trip B | 45 | 68.8 ± 16.5 | 30 | 76.6 ± 23.6 | 0.09964 | 0.05964 | 0.07704 | |

Morning/Peak/Trip B | 33 | 63.2 ± 17.0 | 39 | 57.2 ± 18.4 | | | | |

| ||||||||

Afternoon/Peak/Trip A | 43 | 178.3 ± 161.4 | 31 | 89.9 ± 64.9 | 0.00475 | 0.00415 | 0.00435 | 0.02610 |

Morning/Peak/Trip A | 32 | 184.5 ± 166.5 | 39 | 108.3 ± 74.6 | 0.00885 | 0.00315 | 0.00515 | 0.02575 |

Afternoon/Peak/Trip B | 45 | 128.4 ± 118.9 | 30 | 70.5 ± 59.0 | 0.01225 | 0.03005 | 0.01735 | 0.06940 |

Morning/Off-peak/Trip A | 25 | 65.2 ± 64.0 | 27 | 92.2 ± 89.8 | | | | |

Morning/Off-peak/Trip B | 24 | 67.5 ± 57.5 | 28 | 86.2 ± 57.8 | | | | |

Morning/Peak/Trip B | 33 | 123.4 ± 95.4 | 39 | 129.2 ± 115.2 | | | | |

_{ X } (mg): | ||||||||

Morning/Peak/Trip A | 32 | 38.4 ± 26.1 | 39 | 66.7 ± 40.0 | 0.00095 | 0.00085 | 0.00075 | 0.00450 |

Afternoon/Peak/Trip B | 45 | 94.0 ± 52.4 | 30 | 135.1 ± 53.4 | 0.00205 | 0.00225 | 0.00185 | 0.00925 |

Morning/Off-peak/Trip A | 25 | 44.2 ± 34.9 | 27 | 86.0 ± 50.2 | 0.00165 | 0.00315 | 0.00205 | 0.00820 |

Morning/Off-peak/Trip B | 24 | 61.5 ± 32.1 | 28 | 100.9 ± 57.6 | 0.00345 | 0.00325 | 0.00335 | 0.01005 |

Morning/Peak/Trip B | 33 | 45.0 ± 38.8 | 39 | 81.3 ± 64.1 | 0.00645 | 0.00865 | 0.00735 | 0.01470 |

Afternoon/Peak/Trip A | 43 | 64.2 ± 45.4 | 31 | 85.1 ± 41.2 | 0.04584 | | 0.08024 | 0.08024 |

In order to evaluate possible associations between emission levels, intersection control type, and other significant explanatory variables, we developed a binary logistic regression model for each of the three pollutants considered in the study. Logistic regression is a statistical technique that can be especially useful when trying to identify the existence (and quantify the strength) of relationships involving a categorical response variable. In this analysis, only trips occurring during peak traffic conditions (292 observations, approximately equal to 75% of the entire sample) were considered. This choice is consistent with the idea of incorporating the assessment of the environmental impact of intersections into the process of their operational analysis (or design), for which peak traffic volumes are commonly considered. A binary response variable was obtained by classifying each trip into one of two mutually exclusive categories, based on whether or not the emissions of the given pollutant exceeded the respective mean value computed over all 292 trips.

The general form of the logistic regression model is

For each model, the following information is provided in table format: the coefficient

Basic statistics describing the emissions of CO_{2} for the sample of trips carried out in peak traffic conditions are presented in Table

Descriptive statistics for CO_{2} emissions (g) by intersection type (traffic condition = peak).

Intersection | Number of trips | Mean | SD | Median | (Q_{1}–Q_{3}) | CV% |
---|---|---|---|---|---|---|

Signalized | 153 | 60.63 | 19.44 | 60.43 | (45.42–78.21) | 32.1 |

Roundabout | 139 | 54.49 | 20.72 | 48.14 | (39.75–63.79) | 38.0 |

| ||||||

Total | 292 | 57.70 | 20.26 | 53.98 | (41.18–72.12) | 35.1 |

Using binary logistic regression with CO_{2} emissions as the response variable, we identified intersection type and trip direction as statistically significant predictors. A cutoff value equal to the mean (58 g per trip) was assumed in order to categorize the response into “low” and “high” emissions of the pollutant under consideration. The results reported in Table _{2} per trip are about 2.6 times as likely to exceed 58 g under signal control than with the roundabout, and about 3.8 times as likely to exceed the above threshold for direction B than for direction A. Therefore, trip direction is seen to be a very influential predictor of CO_{2} emissions in this case.

Logistic regression relating risk of “high” CO_{2} emissions to intersection type and trip direction.

Number of trips with CO_{2} ≥ 58 g | Number of total trips | Coefficient | Odds ratio | 95% CI | | |
---|---|---|---|---|---|---|

Intersection: | ||||||

| 45 | 139 | 1 | |||

| 82 | 153 | 0.958 | 2.61 | (1.57, 4.32) | 0.0002 |

Trip direction: | ||||||

| 41 | 145 | 1 | |||

| 86 | 147 | 1.331 | 3.78 | (2.28, 6.27) | <0.0001 |

(Likelihood ratio: ^{2} = 41.96, DF = 2,

We hypothesize that this effect, which is specific to the study site under consideration, is mainly attributable to unbalanced traffic volumes and, for the roundabout, also to differences in geometric and functional characteristics between the two directions (see Figure

Similar conclusions about the effects of the two predictors are suggested by the risk chart displayed in Figure

Probabilities of CO_{2} (g) ≥ 58 g (traffic condition = peak).

Basic statistics describing the emissions of CO for the sample of trips carried out in peak traffic conditions are presented in Table

Descriptive statistics for CO emissions (mg) by intersection type (traffic condition = peak).

Intersection | Number of trips | Mean | SD | Median | (Q_{1}–Q_{3}) | CV% |
---|---|---|---|---|---|---|

Signalized | 153 | 153.10 | 140.10 | 106.30 | (46.80–208.00) | 91.5 |

Roundabout | 139 | 101.89 | 85.46 | 81.09 | (39.24–137.86) | 83.9 |

| ||||||

Total | 292 | 128.70 | 119.86 | 92.18 | (42.06–174.65) | 93.1 |

Logistic regression relating risk of “high” CO emissions to intersection type and test car driver.

Number of trips with CO ≥ 129 mg | Number of total trips | Coefficient | Odds ratio | 95% CI | | |
---|---|---|---|---|---|---|

Intersection: | ||||||

| 38 | 139 | 1 | |||

| 70 | 153 | 0.662 | 1.94 | (1.16, 3.24) | 0.0114 |

Driver: | ||||||

| 19 | 108 | 1 | |||

| 89 | 184 | 1.398 | 4.05 | (2.27, 7.23) | <0.0001 |

(Likelihood ratio: ^{2} = 35.92, DF = 2,

Also in this case the likelihood ratio test and the

Probabilities of CO (mg) ≥ 129 mg (traffic condition = peak).

Table

Descriptive statistics for

Intersection | Number of trips | Mean | SD | Median | (Q_{1}–Q_{3}) | CV% |
---|---|---|---|---|---|---|

Signalized | 153 | 63.42 | 48.01 | 58.22 | (23.14–93.01) | 75.7 |

Roundabout | 139 | 89.67 | 56.23 | 79.69 | (44.26–125.86) | 62.7 |

| ||||||

Total | 292 | 75.92 | 53.62 | 65.47 | (34.13–109.97) | 70.6 |

Logistic regression relating risk of “high”

Number of trips with | Number of total trips | Coefficient | Odds ratio | 95% CI | | |
---|---|---|---|---|---|---|

Intersection: | ||||||

| 46 | 153 | 1 | |||

| 74 | 139 | 0.883 | 2.42 | (1.44, 4.05) | 0.0008 |

Trip direction: | ||||||

| 48 | 145 | 1 | |||

| 72 | 147 | 0.831 | 2.30 | (1.36, 3.87) | 0.0019 |

Driver: | ||||||

| 52 | 184 | 1 | |||

| 68 | 108 | 1.447 | 4.25 | (2.50, 7.23) | <0.0001 |

(Likelihood ratio: ^{2} = 54.84, DF = 3,

Also in this case the likelihood ratio test and the _{2} model, it is interesting to observe that the introduction of the “driver” predictor tends to “mitigate” the effect of trip direction, which is now similar (in terms of coefficient value and odds ratio) to the effect of the type of control. Finally, the risk chart displayed in Table

Probabilities of

Probability of | Intersection | Driver | Trip direction |
---|---|---|---|

14.2% | Signalized | D1 | A |

27.6% | Signalized | D1 | B |

28.7% | Roundabout | D1 | A |

41.4% | Signalized | D2 | A |

48.0% | Roundabout | D1 | B |

61.8% | Signalized | D2 | B |

63.1% | Roundabout | D2 | A |

79.7% | Roundabout | D2 | B |

A comparative analysis of vehicular emissions at a road intersection under different types of control has been described in this paper. The study has adopted a before-and-after approach based on field measurements of three major pollutants (CO_{2}, CO, and

The existence of statistically significant differences between emissions produced by the test vehicle in the “before” and “after” intersection configurations has been assessed using two-sample biaspect permutation tests, a method that can provide more robust evidence as compared to traditional parametric tests, as it allows simultaneously detecting differences in location and variability characteristics of the distributions of the observations in the situations being compared. Relationships between emission levels, intersection control type, and other significant explanatory variables have then been identified using logistic regression models separately for each of the three pollutants under consideration.

Our results are clear-cut: the above analyses show that vehicular emissions of CO_{2} and CO are generally lower for the roundabout than for the signal-controlled intersection, while an opposite finding emerges for

A comparison of these findings with the results reported in the existing literature is not straightforward, mainly because most of the previous studies on the subject are based on a modeling/simulation approach rather than on direct field measurements. Despite these difficulties in establishing meaningful comparisons, we note that our results are in agreement with at least some of the previous works found in the literature.

It is believed that results obtained in field studies such as the one described in this paper may be very useful for the calibration and validation of microscopic models relating pollutant emissions to vehicular traffic conditions. These models, in turn, can become very effective tools in the context of procedures for the environmental impact assessment of road projects.

The complexity of the problem under consideration suggests that further research is needed. A possible future development of the study described in this paper is a detailed analysis of the effect on emissions of the composition of trips in terms of vehicle operating modes (idle, acceleration, cruise, and deceleration) in relation to the type of intersection control.

Another issue that could be explored in future work is the effect on emissions of driving style, for which we plan to design and implement a controlled experiment using a driving simulator. This will allow testing the behavior of a considerable number of subjects, which is an essential condition to support the validity and statistical significance of the results. In the study described in this paper, only two drivers were involved, and this did not allow exploring the full range of behaviors that one can expect to find in the real world.

Since the signal-controlled intersection considered in this study did not necessarily operate with optimal timing parameters, further research could also be devoted to comparing pollutant emissions produced by vehicles under an optimized traffic signal plan and under roundabout intersection control. A microsimulation approach appears to be particularly suitable to investigate this issue.

Another aspect deserving consideration and analysis in future developments of this study is the fact that geometric layout (e.g., number of approach lanes) of the intersection is generally modified when switching from signal control to roundabout, with possible significant impacts on pollutant emissions. Moreover, the research described in this paper could be extended by considering emissions produced by light and heavy goods’ vehicles, in consideration of the growing importance of environmental aspects in the context of urban freight delivery systems [

Finally, with specific reference to the roundabout, it would be interesting to assess the impact on emissions of conflicting pedestrian flows on intersection crosswalks. This would require a new campaign of data collection, in which pedestrian flows should be measured during the intervals in which the test vehicle crosses the roundabout. Such data on pedestrian volumes could then be included in the logistic regression models as an additional explanatory variable and tested for statistical significance.

The authors declare that there are no conflicts of interest regarding the publication of this article.

This work was supported by the University of Padova [Grant no. CPDA 128393/12].

_{2}vehicle emissions under alternative forms of intersection control

_{2}emissions at urban arterials

_{2}emissions, cost, and service quality trade-offs in an urban delivery system case study