STOTEN-15671; No of Pages 7 Science of the Total Environment xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv
Adaptive traffic management in cities — Comparing decision-making methods Sef van den Elshout ⁎, Rinkje Molenaar, Bart Wester DCMR Environmental Protection Agency Rijnmond, PO Box 843, 3100AV Schiedam, The Netherlands
H I G H L I G H T S • Analysis of practical aspects of decision making on adaptive traffic management for environmental purposes. • Real-time traffic management using air quality measurements has few advantages over more generic management strategies. • Measures aimed at peak concentrations only have a small impact on the overall exposure to traffic related air pollution.
a r t i c l e
i n f o
Article history: Received 21 August 2013 Received in revised form 12 November 2013 Accepted 18 December 2013 Available online xxxx Keywords: Adaptive traffic management Real-time decision making Urban air quality Black Carbon
a b s t r a c t Traffic is the dominant source of air pollution in cities. We simulated ‘adaptive traffic management’ (temporary traffic interventions that are invoked based on preset conditions such as high ambient concentrations) aimed at reducing traffic related air pollution. We compared these results with the effect of permanent temporary traffic interventions (measures that are always invoked for a few hours, irrespective of other criteria). The potential impact of the traffic interventions was assessed using Black Carbon and NOx-concentration observations in a busy urban street in Rotterdam, The Netherlands. Results show that generic traffic information (counts, speed, composition) in combination with general knowledge about the atmospheric conditions, provide sufficient information for operational decision making. However, the results also show that the overall net benefits of temporary measures are very small. The impact of permanent measures such as lowering the traffic density during rush hours is higher than measures taken for short time periods when air pollution is high or expected to be high. © 2013 Elsevier B.V. All rights reserved.
1. Introduction Traffic is the dominant source of air pollution in cities. Restricting the amount of traffic in a city is of course the best way to reduce traffic emissions. This, however, is not easy and often politically sensitive. The question arises if the impact of traffic on air quality can be minimized by interventions, given a certain amount of traffic. This is the purpose of environmental adaptive traffic management (ATM) strategies. ATM or Intelligent Transport Systems (ITS) are not only operated for environmental reasons. To manage the flow of traffic through a city, adaptive systems are often already in place to assure that traffic moves as smooth as possible through a city under various conditions (traffic densities). We look at various aspects of the decision-making for environmental ATM. The flow of traffic over a road network or through a city can be optimized for different objectives. If second order effects (smooth traffic with a minimum of interruptions might attract more traffic) are neglected, the technical network optimum – e.g. moving the maximum ⁎ Corresponding author. Tel.: +31 102468368. E-mail address: [email protected] (S. van den Elshout).
number of vehicles with the smallest possible delays (minimum idling times, avoiding stop and go traffic) – is also likely to be the most environmentally efficient way as well. However this can lead to unacceptable solutions where slow traffic or traffic on minor branches faces unacceptable waiting times at intersections (van Baalen et al., 2011; Koning et al., 2011 [Dutch version]). Generally a compromise optimization is chosen that is environmentally still quite good – assuming that most traffic still moves in the best possible way – but not the best solution. They propose to (temporarily) change the optimization criteria based on environmental conditions, e.g. when air pollution is high, the existing traffic management solution (compromise) is changed to a configuration that reduces overall traffic emissions. Similarly, Hodges et al. (2009) describe how the Leicester city ATM system is enhanced with air quality sensors to include this information in decision-making. Interventions can be temporary, manipulating traffic for relatively short (hours) periods with the aim to reduce the traffic emissions when air pollution is high. Alternately they could be permanent short duration optimization measures, e.g. tweaks that always occur on weekdays during rush-hours (thus avoiding a lot of real-time decisionmaking). Temporary traffic measures are needed when: air pollution is high and traffic is the (main) cause. Episodes of for example high
0048-9697/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
2
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx
particulate matter (PM10) concentrations also occur when secondary aerosol is formed or when polluted air masses are transported to a city from surrounding areas. Taking traffic measures on such occasions is not helpful in reducing air pollution levels. Typically traffic related air pollutants are needed as indicator. The two pollutants (regulated by the EU) that pose problems in many cities are nitrogen dioxide (NO2) and PM10. NO2 is typical for traffic (though other sources exist as well) but being a largely secondary pollutant its relation with the emissions is less direct than NOx, the mixture of nitrogen oxides that is actually emitted but not regulated. Black Carbon (BC) is a typically traffic related pollutant and a very health relevant pollutant like PM10. When it comes to traffic related air pollution BC is even the preferred health relevant indicator (Janssen et al., 2011; Keuken et al., 2011). There is no practical absolute limit value for pollutant concentrations with short (hourly) averaging times that could guide short-term decision making. It is mainly the longer averaging time criteria that are hard to meet: daily averages for PM10 year average for NO2. Therefore it was decided to not only look at high concentrations as such. As environmental indicator we looked at the road increment (e.g. ΔBC or ΔNOx). When this Δ is high it is sure that there is a) a problem and b) it is traffic related and a traffic intervention makes sense. Going one step further, the road increment is high when there is a lot of traffic but also when the dispersion conditions are poor. Trying to minimize emissions at these unfavorable conditions for example by shifting them to more favorable times might be an ATM option. Actually making traffic interventions is not easy and many studies rely on modeling to assess the impact of a simulated traffic measure. We demonstrate how, under certain assumptions, the impact of simulated, and existing traffic measures can actually be determined using air quality measurements avoiding the use of emission and dispersion models, each with their inherent uncertainties. We do this for a busy inner urban road in the city of Rotterdam, The Netherlands.
2. Case description and research questions There are four questions to be examined. Firstly does air pollution have to be measured for operational decision making on traffic management? If yes, a dense network of monitoring or sensor equipment is needed to make informed decisions (e.g. as in the MESSAGE project, North et al., 2009). Secondly, is it possible instead to use generic information such as traffic numbers and speed, meteorological data, etc. for management decisions and how does this affect the impact of the measures? In that case, using existing proxy information would be an efficient and cheap alternative. Thirdly, temporary measures (to be activated depending on environmental criteria) are compared to permanent measures that are always invoked at certain times. Lastly the beneficial potential of shifting emissions from unfavorable to more favorable dispersion conditions is examined. In this study we use measured concentrations of BC and NOx, in a busy urban street. In the morning the dominant traffic flow is towards the city center and in the afternoon in the opposite direction. The city employs an intelligent traffic management system (ITS) that manipulates the traffic lights in such a way that a minimum speed of 25 km·h−1 is maintained along a 1.2 km corridor. Seven weeks of traffic observations, evenly spread over the year 2011, resulting in a dataset of 1001 h were analyzed. The data on fleet composition (private cars and vans, medium duty and heavy duty trucks), traffic density and traffic velocity were obtained from the Rotterdam City Traffic Department. The average hourly traffic density was 1056, 62 and 8 vehicles for respectively private cars, medium duty and heavy duty vehicles. The street has two lanes in either direction. The street canyon (roads, parking places, pavement) is approximately 45 m wide. The street is tree-lined. Hourly concentrations of BC and NOx were obtained from a kerbside monitoring station and from a nearby (b 500 m) background station. The meteorological
Fig. 1. The relation between traffic density and the average speed of the private vehicles.
information was obtained from the National Meteorological Institute from their site at Rotterdam Airport just north of the city. 2.1. Assumptions • The increment between the concentrations measured at the roadside monitoring station and the background monitoring station (ΔBC or ΔNOx) is due to traffic emissions. If the observed increment was b0 it was set to 0. • Using traffic emission factors (Velders et al., 2012) weights were given to the number of medium and heavy duty trucks while analyzing BC and NOx. Medium and heavy duty trucks are equivalent to 4.7 and 6.8 private vehicles respectively for BC and 19.0 and 29.4 for NO x. Using these factors a weighted traffic density Dw (expressed as private vehicle equivalent/h — pve·h− 1), is calculated. Note that at a given hour Dw is different while analyzing BC and NOx. • Vehicle emissions depend, to some extent, on the velocity and the velocity depends amongst others on the traffic density. Traffic interventions therefore tend to have multiple effects that each cause a change in emission per km. However in this case the Traffic Department succeeds in maintaining average speeds between 25 and 50 km·h− 1 at all times (see Fig. 1). We therefore assume that there are no major changes of the emission/vehicle/km as a function of the traffic speed and density Dw. Hence we assume that the observed street increment normalized by traffic density (ΔBC/Dw) at a given hour is mainly the result of the atmospheric dispersion conditions at that hour and rather independent of the traffic conditions. • If the traffic numbers do not actually change the normalized street increment ΔBC/Dw (or ΔNOx/Dw), the impact of an intervention in the traffic density can be assessed by simply multiplying the change of the traffic density Dw with the street increment per unit of traffic ΔBC/Dw at the given hour. This seems to be the case: the increment per unit of traffic remains rather constant, independent of the traffic numbers. Fig. 1 shows that removing/adding 500 vehicles hardly affects the speed so it is a fair assumption that ΔBC/Dw and ΔNOx/Dw remain constant upon changes in Dw. In this way we can use measured air quality data to estimate the impact of hypothetical interventions in Dw. 2.2. Traffic scenarios studied Most of the ATM measures and ITS systems are used to minimize stop and go traffic and idling times by assuring that the traffic density doesn't exceed the road network capacity. This can be done by optimizing traffic light cycle times in real-time, by gating (Tate and Bell, 2000) e.g. if necessary creating congestion elsewhere where it is environmentally
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx
3
less important, or by reducing maximum speed (typical on motorways) to homogenize the traffic flow. Several of the scenarios simulated in this study actually remove (temporarily) part of the traffic. This is not a common intervention and their practical feasibility is not always obvious.1 However, the example can easily be used to study the impact of the different decision making processes. For the environmentally prompted measures the simulated intervention consists of the removing of all medium and heavy duty trucks. This traffic ban is lifted as soon as the criteria that trigger the measure are no longer met and the previously avoided emission occur at a hopefully more favorable moment. The temporary adaptive traffic management interventions are invoked using the following criteria: 1. Poor dispersion indicated by a mixing height ≤ 200 m. AND Dw ≥ 2000 pve·h−1 2. Poor dispersion indicated by a wind speed ≤ 2 m·s−1. AND Dw ≥ 2000 pve·h−1 3. Air quality observations are used to decide on a measure. The top 10% of the hourly observed road increment for NOx and BC can be used to invoke a measure. By studying historic data a threshold value can be set. AND Dw ≥ 2000 pve·h−1 4. NO2 has a distinct diurnal and yearly concentration pattern. A concentration reference pattern was established that takes into account the typical concentration variation. If an hour has a concentration of ≥20 (μg·m−3) above the reference concentration for that hour, it is considered as too high and a measure is needed (see van den Elshout et al., 2009a). In addition to the temporary measures that are invoked based on environmental criteria, it is also possible to create permanent short duration traffic measures irrespective of environmental criteria. These traffic interventions are either truly permanent (numbers 6 and 7) or active at a fixed hour of the day/day of the week (number 5). Measures 6 and 7 have actually been implemented at the road under consideration. 5 On working days between 6.00 and 8.00 h Dw is reduced: 500 pve·h− 1 are removed in this busy period either by banning medium and heavy duty trucks or by an imaginary road pricing scheme. The vehicles are allowed back-in between 09.00 and 11.00 h when average dispersion conditions are better. 6 Using an ITS the Traffic Department manages the traffic lights in such a way that a minimum velocity of 25 km/h is maintained. The likely impact of this measure is estimated assuming that stagnating traffic would otherwise occur on working days for at least 30 min during both the morning and evening rush-hour. Emission factors for stagnating traffic are 1.5 and 1.4 times higher respectively for NOx and BC (Velders et al., 2012). So the gain of the measure is 0.5 ∗ ΔNOx and 0.4 ∗ ΔBC at the given hour. 7 A park-and-ride facility near a subway station next to this route. On working days this removes 350 cars during the morning and evening peak hours (7.00–8.00 h and 17.00–18.00 h). Because of this measure only the number light vehicles is reduced. This reduces the average pve·h−1 from 2681 to 2331 during the hours considered. Using the average ΔBC/Dw and ΔNOx/Dw ratios during these hours the impact can be calculated.
Fig. 2. The diurnal pattern of traffic density and the fraction of medium and heavy duty vehicles.
neutral atmospheric stability) strongly resembles the traffic pattern so both are good indicators for traffic related air pollution. Fig. 4 shows ΔBC/Dw for three atmospheric stability classes. It shows that ΔBC/Dw (or ΔNOx/Dw not shown here) remains rather constant throughout the day between 06.00 h and 18.00 h despite considerable variation in traffic numbers (Fig. 2). Fig. 4 also shows that the road increment per unit of traffic depends on the prevailing atmospheric stability: it shows the influence of dispersion conditions on concentrations given the same average emissions. Neutral stability is the only atmospheric stability class that occurs throughout the day. Stable and unstable conditions lead to higher concentrations with the same amount of emissions. Stable conditions often occur in the morning and evening. The Monin–Obukov length (calculated by the KEMA Stacks model, Erbrink et al., 2012) was used to determine the stability classes. Atmospheric stability determines the concentrations resulting from an emission and explains a considerable amount of the variation in the road increment (van den Elshout et al., 2009b). Note that the impact of atmospheric stability on the concentrations is less when the emissions occur in narrow and deep street canyons. If stable atmospheric conditions can be avoided (often occurring in the morning), in theory this can reduce the concentrations and be the basis of a temporary traffic measure: reducing Dw when ΔBC/Dw is above average. This is the logic behind measure 5 (and 7).
3. Results and discussion 3.1. Concentrations and dispersion conditions Fig. 2 shows some traffic characteristics. It appears that more vehicles enter than leave the town using this road. Fig. 3 shows the close relation between ΔBC and ΔNOx. The diurnal pattern (during 1 Banning trucks based on unpredictable environmental criteria, as occurs in some of the examples studied is not practical. In a permanent (hence predictable) measure such as banning trucks during the morning hours when pollution is likely to be high is feasible.
Fig. 3. Diurnal pattern of the road increment for BC and NOx on working days.
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
4
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx
Fig. 4. Diurnal pattern of ΔBC/1000 pve for the three atmospheric stability classes on working days (3 hour averages due to scarcity of data per hour/per class).
3.2. Impact and net-impact If a measure is taken during an hour with unfavorable conditions there obviously is a positive impact. However, if traffic is banned from
an area at a given hour, it will have to enter at another time. So either the same amount of traffic has to enter at a later hour, or via different roads. This causes additional pollution partly offsetting the gains at the site of, and during the measure. We therefore calculate the impact during the measure and the net impact, taking into account the emissions shifted to another time or place (not studied here). The net impact is particularly important for BC. As mentioned, this is a health relevant pollutant with no safe threshold (similar to PM10/PM2.5). Hence it doesn't matter when or where the concentrations occur. Only if traffic could be channeled through a less populated area an exposure benefit could be achieved. For NO2 the situation is slightly different. If the traffic measure is devised for the purpose of meeting the limit values (and not to reduce exposure) spreading traffic over the network in such a way that in all streets the limit values are met, could be an option. The ethics of increasing someone's exposure to the benefit of others are not considered in this case. Spreading traffic over different roads, however, has practical limitations. In Rotterdam (and in most cities?) the road network operates at near full capacity during peak hours and the practical possibility to reroute traffic in the face of adverse air quality conditions is very limited. Fig. 5 shows the road under study (thick line) and a theoretically alternative route to the city center. Both are equally busy and trying to shift traffic from one route to the other is unlikely to yield net benefits. Gating (Tate and Bell, 2000), not studied in this paper, might have some
Fig. 5. Study area (thick line) and routes into the city center coming from the southern ring road.
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx
potential as a temporary adaptive management strategy by queuing traffic south of the road fork. Besides, since this road is still in the built environment the ethics are questionable. A permanent, kind of gating situation was implemented on the northern part of town on the motorway leading into town. Before this intervention, traffic entering into town from the north at a speed of 100 km·h−1 often led to highly dynamic traffic situations and/or stagnation with relatively high emissions per km in the approach of the T-junction with the northern ring-road. When the maximum speed was reduced to 80 km·h− 1, stagnation was shifted to the entrance of the town, at the point where the change from 100 to 80 km.h−1 took place. Within the city limits, traffic moving relatively smoothly at 80 km·h−1, saw their NOx emissions reduced by 20–30% (Wesseling et al., 2003). Though this measure was presented as a speed limit, it operates in a similar way as the gating technique: assuring a smooth traffic flow where it matters. Reducing traffic dynamics by making personnel vehicles drive as fast as heavy duty vehicles (speed limit at 80 km·h−1) on a motorway is an effective way of reducing the emissions per km. This was further confirmed by Keuken et al. (2010).
3.3. Results and discussion The traffic related air pollution characteristics of the study area are shown in Table 1. It shows the average weighted traffic density, the road increment and the road increment per 1000 pve. These average characteristics can be compared to the situations when the traffic interventions are invoked (Table 2). The results of the seven interventions are summarized in Table 2. Looking at the impact during a traffic measure it shows that those based on the actual occurrence of high concentrations (3 and 4) do pick, as expected, the hours where traffic has the highest impact (where ΔBC/Dw is high). See the two right most columns in Table 2. If the measure was based on meteorological parameters like mixing height and wind speed (like in measures 1 and 2), hours with a lower concentration increment per unit of traffic are picked. Nevertheless, the measures are timed approximately at the right moment: the ratios of ΔBC/Dw that were selected are much higher than the sample median (Table 1). The permanent measures during the (morning) rush-hours (5–7) have ΔBC/Dw ratio's slightly higher than the median. The net benefits show that the temporary restriction measures are not very successful. The hours where the measure is invoked apparently occurs on unfavorable days where the dispersion remains quite poor for prolonged periods. Allowing the banned vehicles to move again still creates substantial concentration rises, almost nullifying the previous gains or making the net situation worse. Sometimes it was even impossible to allow all the cars banned during the day to start to move again (the adverse environmental conditions remain throughout the day). The theory that traffic emissions during hours with unfavorable dispersion can be traded for emissions during better conditions, e.g. later on the day, hardly seems to work in reality in our case. The “road pricing”/morning rush-hour restriction scheme (intervention 5) shifting cars from the rush hours to a more quiet time a few hours later gives a lower impact during the measure than the temporary
Table 1 Summary statistics, whole sample studied (1001 h). Average Dw (pve·h−1)
Whole Sample Private cars
Average road increment (μg·m−3)
Median road increment per 1000 pve (μg·m−3)
BC
NOx
BC
NOx
BC
NOx
1400 75%
2445 43%
1.63 1.23
46 20
1.08
14.7
5
measures based on environmental criteria, but the net impact is slightly higher. As can be seen from the road increment/1000 pve during this measure, these morning hours are only marginally worse than the median conditions. However the “payback” hours are not unfavorable either (and on average better than the intervention hours) so a net benefit remains. The ITS scheme (6) that keeps cars moving at a reasonable speed gives the highest net impact. Without this scheme the concentrations would have been substantially higher during the hours affected. Flow optimizations that assure smoothly running traffic are by far the best interventions as rapid speed changes lead to extra emissions (e.g. van Driel, 2007; Servin et al., 2006; Gense et al., 2001; all in Mahmod, 2011). The measures taken to optimize the road network performance in general also seem to serve the environment quite well. Similarly the park and ride scheme that removes 350 cars (in the morning as well as the afternoon rush-hour), is quite effective in reducing concentrations during these hours. Scenarios 6 and 7 are the ones with hardly any or no “payback effects”. In scenario 6 the emissions are reduced by avoiding stop and go traffic (reducing emissions per km) and in scenario 7 traffic numbers are actually reduced. Looking at the decision making questions: if environmental criteria are used and if these are based on actual air quality measurements (the interventions are done during the hours with the highest concentrations). If other environmental criteria are used to estimate when air quality is high the gross impact of the intervention (e.g. measure 2) would be on average 65% of the measure with the highest gross impact (measure 3). On the other hand the hours with high concentrations seem to occur in batches: they tend to last for the major part of the day or even days, resulting in substantial “payback” and very small net benefits. Considering the paybacks, it is doubtful that there will be net advantages of decision making based on actual air pollution measurements over estimating peak situations based on traffic numbers and a meteorological parameter. In our simulated case they were virtually absent. The third question examines whether day of the week and time of the day are sufficient to identify hours when an intervention is needed. It turns out that scenario 5 directed at the morning rush-hour during weekdays, is about as effective as the scenarios 1 and 2 where the intervention was timed on generic environmental information (if the same amount of traffic equivalents would be involved). The net results are even slightly better. This suggests that the added advantage of decisionmaking based on measurements and on proxy environmental indicators is limited. If simply the rush-hours are identified, at least in our case, the net air quality results would be similar while avoiding a lot of complicated real-time decision making. Looking at scenario 5 where emissions are shifted from morninghours with on average less favorable dispersion, to a moment later on the day it turns out that there is a small net benefit, e.g. environmentally there is a (small) benefit from allowing medium and heavy duty traffic into town only after 09.00 h or 10.00 h when the mixing layer is higher and the impact of the same emissions on the ambient concentrations is smaller. A restricting scheme such as scenario 5 would have a higher impact if second order effects did occur, e.g. if it also contributed to reduce stop and go traffic and excess idling. This didn't occur in the first place in our experimental set-up. The permanent measures with no real-time decision-making and hardly any payback emissions have the highest net impact (scenarios 6 and 7). 3.4. Health effects short-term and long-term exposure/measures Health effects of traffic related air pollution occur on two time scales, short-term (day/days in case of PM) and more importantly long-term. It is the permanent exposure to any level of particulate air pollution that causes most health effects (WHO, 2013). The report mentions evidence that very short (1–2 h) exposure to very high concentrations cause health effects but notes that the effects of the sum of all short-term
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
6
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx
Table 2 Results of the different traffic measures. case
Average # pve involved in measurea
1 2 3 4 5 6 7
641 645 718 585 500 2331 350
# hours when measure is invoked
Average impact of the measure during the measure (μg·m−3)
hours
%
BC
104 73 71b 184 70 35 70e
10% 7% 7% 18% 7% 3% 7%
0.85 0.89 1.37 0.90 0.59 1.04 c) 0.39
Average net impact per hour of measure (μg·m−3)
NOx
BC
88 61 9 37c 6
b0 0.03 0.04 b0 0.06 0.94d 0.39
Observed average road increment per 1000 pve when measure is invoked (μg·m−3)
NOx
BC
NOx
b0 b0 1.7 33d 6
1.33 1.38 1.91 1.54 1.18 1.12 1.12
30 26 18 18 18
a
Weighting shown based on BC emisison factors (NB: for measures 3–7, the NOx results are calculated with the NOx weighting factors). The 90 percentile of ΔBC/pve·h−1 would suggest 100 out of 1001 h in the sample. However 29 h were lost as the criterion of having sufficient traffic was not met, e.g. early morning hours with very poor dispersion and also a low traffic density. c The impact is calculated (example BC, see measure 6) as 0.4 ∗ # pve·h−1 ∗ ΔBC/pve·h−1. d If improving the flow on the main road causes extra stagnation on the side roads the net benefit is smaller than the benefit during the measure. According to the Traffic Department, the cycle times on the branches are hardly affected and no negative impact of measure is expected. A maximum trade-off of 10% was included. e We assume that the park and ride facility fills in 1 h during the morning peak hours and empties in the evening rush-hour so the benefit occurs twice a day. b
exposures is smaller than the long-term exposure effects. Unless more evidence is emerging that it is useful to avoid very short (hours) exposure peaks the efforts to particularly address peak concentration hours with ATM seem complicated with only a small contribution to the reduction of overall exposure. Looking at absolute improvements (net effect during the measure ∗ fraction of time) of the year average concentrations, even the most effective measure considered (ITS) has a small overall impact. The year average BC road increment is reduced by 0.03 μg·m− 3. For NOx the road increment is reduced by a maximum of 1.2 μg·m− 3 (by the ITS measure). This amounts to net year average road increment concentration reductions of some 2%. This would of course be higher if without this measure there would have been substantial stop and go traffic. In this study we have conservatively estimated this at one hour per day only. Apart from the net impact of the measure, the time that it can have an impact is, by its temporary nature, also quite short, hence a small overall impact. For another example, look at the Flanders smog-day scheme (Lefebvre et al., 2011). During predicted smog episodes (a few days a year) the maximum speed on the motorway is reduced from 120 to 90 km·h−1. During the measure the Elemental Carbon (EC) concentrations are reduced by 30% immediately adjacent to the motorway. However, like in the measures analyzed in this study, the fraction of time the measure is applied is small, limiting the overall exposure reduction impact. The permanent speed limit on the northern access motorway in the study from Wesseling et al. (2003) reduces the year average traffic contribution by 20–30% (for NOx). If a constant speed reduction is unacceptable a permanent speed reduction during the busiest hours (e.g. morning rush hour) would likely have a bigger impact.
4. Conclusions Temporary interventions to reduce traffic related air pollution at peak moments can be appropriately timed, either by air quality measurements or by using generally existing environmental data (such as wind speed) and traffic counts. The latter simplifies the decision making on such interventions. If one is interested in such measures there is no need to first create elaborate monitoring or sensor networks. One could alternatively opt for permanent short duration adaptations (at fixed times) or include environmental criteria in the overall optimization of the existing traffic management systems. This avoids complex decision making and is likely to have similar or even larger effects. As long as epidemiological evidence continues to suggest that longterm exposure is more important than short-term peaks permanent
measures rather than short-term peak-shaving interventions are needed to achieve substantial exposure reductions. Acknowledgments Rotterdam Traffic Department, dS + V, kindly supplied detailed traffic data. References Erbrink Hans, Kokmeijer Esther, de Wolff Joost. An extended OSPM model for pollutant concentrations in street canyons with detached houses and specified traffic lane width. ITM–NATO/SPS International Technical Meeting on Air Pollution Modelling and its Application: Proc; 2012. Gense NLJ, Wilmink IR, van den Burgwal HC. Emissions and congestion: Estimation of emissions on road sections and the Dutch motorway network. TNO report, TNO Automotive; 2001. Hodges N, Bell M, Galatioto F, Hill G, Neasham J, Suresh V. Sustainable network management — the integration of intelligent transport systems and “grid enabled” pervasive sensors. Newcastle University; 2009 [No. CS-TR-1167. www.cs.ncl.ac.uk/publications/trs/papers/1167.pdf. (Accessed July, 2013)]. Janssen NAH, Hoek G, Lawson-Simic M, Fischer P, Bree van L, Brink van H, et al. Black carbon as an additional indicator of the adverse health effects of combustion particles compared to PM10 and PM2.5. Environ Health Perspect 2011;119(12):1691–9. Keuken M, Jonkers S, Wilmink IR, Wesseling J. Reduced NOx and PM10 emissions on urban motorways in The Netherlands by 80 km/h speed management. Sci Total Environ 2010. http://dx.doi.org/10.1016/j.scitotenv.2010.03.008. Keuken Menno, Zandveld Peter, van den Elshout Sef, Janssen Nicole, Hoek Gerard. Air quality and health impact of PM10 and EC in the city of Rotterdam, The Netherlands in 1985–2008. Atmos Environ 2011;45(30):5294–301. Koning Anne de, van Baalen Janneke, Voogt Marita, Stelwagen Uilke, Turksma Siebe. EcoFLEX: Het verbeteren van de luchtkwaliteit met behulp van Groen Dynamisch VerkeersManagement, gebaseerd op real time luchtkwaliteitsmetingen. www. verkeerskunde.nl/Uploads/2011/11/Bijdrage105.pdf, 2011. [(Accessed July, 2013)]. Lefebvre W, Fierens F, Trimpeneers E, Janssen S, Van de Vel K, Deutsch F, et al. Modeling the effects of a speed limit reduction on traffic-related elemental carbon (EC) concentrations and population exposure to EC. Atmos Environ 2011;45: 197–207. Mahmod M. Using co-operative vehicle-infrastructure systems to reduce traffic emissions and improve air quality at signalized urban intersections [TRAILThesis series. http:// doc.utwente.nl/76481/1/thesis_M_Mahmod.pdf. (Accessed July, 2013)]; 2011. North R, Cohen J, Wilkins S, Richards M. Hoose Neil, et al. Field deployments of the MESSAGE system for environmental monitoring. Traffic Engineering & Control 2009;50(11):484–8. (Hemming Information Services, ISSN 0041-0683). Servin O, Boriboonsomsin K, Barth B. An Energy and Emissions Impact Evaluation of Intelligent Speed Adaptation. In Intelligent Transportation Systems Conference. Toronto, Canada: ITSC’06. IEEE; 2006. Tate J, Bell MC. Designing, implementing and evaluating a gating strategy to reduce pollution levels in sensitive areas of the urban environment. 10th IEE International Scientific Symposium on Transport and Air Pollution, Avignon, France; 2000. van Baalen J, De Koning A, Voogt M, Stelwagen U, Turksma S. ECOFLEX: improving air quality with green dynamic traffic management based on real time air quality measurements. 18th Intelligent Transport Systems World Congress, Orlando, United States, 2011; 2011.
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx van den Elshout Sef, Mahmod Mohamed, Ingrid Arts, van Arem Bart. Decision making on short term traffic measures to influence traffic related air pollution. ETTAP 2009. Actes INRETS n°122; 2009a. van den Elshout Sef, de Gier Cornelis, Arts Ingrid. PM10, PM2.5 and black smoke concentrations in urban background and traffic situations — what is a good traffic indicator? Conference Proceedings ETTAP 2009. Actes INRETS n°122; 2009b. van Driel CJG. Driver Support in Congestion: An assessment of user needs and impacts on driver and traffic flow. PhD thesis, TRAIL Research School; 2007. Velders GJM, Aben JMM, Jimmink BA, Geilenkirchen GP, van der Swaluw E, de Vries WJ, et al. Grootschalige concentratie- en depositiekaarten Nederland: Rapportage 2012.
7
RIVM Rapport 680362002. www.rivm.nl/Bibliotheek/Wetenschappelijk/Rapporten/ 2012/juni/Grootschalige_concentratie_en_depositiekaarten_Nederland_Rapportage_ 2012, 2012. [(Accessed 04-07-2012)]. Wesseling JP, Hollander K, Teeuwisse S, Keuken MP, Spoelstra H, Gense R, et al. Onderzoek naar effecten van de 80 km/u maatregel voor de A13 op de luchtkwaliteit in Overschie. TNO rapport 2003/258; 2003. WHO. Review of evidence on health aspects of air pollution — REVIHAAP Project. http://www.euro.who.int/__data/assets/pdf_file/0004/193108/ REVIHAAP-Final-technical-report-final-version.pdf, 2013. [(Accessed July, 2013)].
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
Contents lists available at ScienceDirect
Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv
Adaptive traffic management in cities — Comparing decision-making methods Sef van den Elshout ⁎, Rinkje Molenaar, Bart Wester DCMR Environmental Protection Agency Rijnmond, PO Box 843, 3100AV Schiedam, The Netherlands
H I G H L I G H T S • Analysis of practical aspects of decision making on adaptive traffic management for environmental purposes. • Real-time traffic management using air quality measurements has few advantages over more generic management strategies. • Measures aimed at peak concentrations only have a small impact on the overall exposure to traffic related air pollution.
a r t i c l e
i n f o
Article history: Received 21 August 2013 Received in revised form 12 November 2013 Accepted 18 December 2013 Available online xxxx Keywords: Adaptive traffic management Real-time decision making Urban air quality Black Carbon
a b s t r a c t Traffic is the dominant source of air pollution in cities. We simulated ‘adaptive traffic management’ (temporary traffic interventions that are invoked based on preset conditions such as high ambient concentrations) aimed at reducing traffic related air pollution. We compared these results with the effect of permanent temporary traffic interventions (measures that are always invoked for a few hours, irrespective of other criteria). The potential impact of the traffic interventions was assessed using Black Carbon and NOx-concentration observations in a busy urban street in Rotterdam, The Netherlands. Results show that generic traffic information (counts, speed, composition) in combination with general knowledge about the atmospheric conditions, provide sufficient information for operational decision making. However, the results also show that the overall net benefits of temporary measures are very small. The impact of permanent measures such as lowering the traffic density during rush hours is higher than measures taken for short time periods when air pollution is high or expected to be high. © 2013 Elsevier B.V. All rights reserved.
1. Introduction Traffic is the dominant source of air pollution in cities. Restricting the amount of traffic in a city is of course the best way to reduce traffic emissions. This, however, is not easy and often politically sensitive. The question arises if the impact of traffic on air quality can be minimized by interventions, given a certain amount of traffic. This is the purpose of environmental adaptive traffic management (ATM) strategies. ATM or Intelligent Transport Systems (ITS) are not only operated for environmental reasons. To manage the flow of traffic through a city, adaptive systems are often already in place to assure that traffic moves as smooth as possible through a city under various conditions (traffic densities). We look at various aspects of the decision-making for environmental ATM. The flow of traffic over a road network or through a city can be optimized for different objectives. If second order effects (smooth traffic with a minimum of interruptions might attract more traffic) are neglected, the technical network optimum – e.g. moving the maximum ⁎ Corresponding author. Tel.: +31 102468368. E-mail address: [email protected] (S. van den Elshout).
number of vehicles with the smallest possible delays (minimum idling times, avoiding stop and go traffic) – is also likely to be the most environmentally efficient way as well. However this can lead to unacceptable solutions where slow traffic or traffic on minor branches faces unacceptable waiting times at intersections (van Baalen et al., 2011; Koning et al., 2011 [Dutch version]). Generally a compromise optimization is chosen that is environmentally still quite good – assuming that most traffic still moves in the best possible way – but not the best solution. They propose to (temporarily) change the optimization criteria based on environmental conditions, e.g. when air pollution is high, the existing traffic management solution (compromise) is changed to a configuration that reduces overall traffic emissions. Similarly, Hodges et al. (2009) describe how the Leicester city ATM system is enhanced with air quality sensors to include this information in decision-making. Interventions can be temporary, manipulating traffic for relatively short (hours) periods with the aim to reduce the traffic emissions when air pollution is high. Alternately they could be permanent short duration optimization measures, e.g. tweaks that always occur on weekdays during rush-hours (thus avoiding a lot of real-time decisionmaking). Temporary traffic measures are needed when: air pollution is high and traffic is the (main) cause. Episodes of for example high
0048-9697/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
2
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx
particulate matter (PM10) concentrations also occur when secondary aerosol is formed or when polluted air masses are transported to a city from surrounding areas. Taking traffic measures on such occasions is not helpful in reducing air pollution levels. Typically traffic related air pollutants are needed as indicator. The two pollutants (regulated by the EU) that pose problems in many cities are nitrogen dioxide (NO2) and PM10. NO2 is typical for traffic (though other sources exist as well) but being a largely secondary pollutant its relation with the emissions is less direct than NOx, the mixture of nitrogen oxides that is actually emitted but not regulated. Black Carbon (BC) is a typically traffic related pollutant and a very health relevant pollutant like PM10. When it comes to traffic related air pollution BC is even the preferred health relevant indicator (Janssen et al., 2011; Keuken et al., 2011). There is no practical absolute limit value for pollutant concentrations with short (hourly) averaging times that could guide short-term decision making. It is mainly the longer averaging time criteria that are hard to meet: daily averages for PM10 year average for NO2. Therefore it was decided to not only look at high concentrations as such. As environmental indicator we looked at the road increment (e.g. ΔBC or ΔNOx). When this Δ is high it is sure that there is a) a problem and b) it is traffic related and a traffic intervention makes sense. Going one step further, the road increment is high when there is a lot of traffic but also when the dispersion conditions are poor. Trying to minimize emissions at these unfavorable conditions for example by shifting them to more favorable times might be an ATM option. Actually making traffic interventions is not easy and many studies rely on modeling to assess the impact of a simulated traffic measure. We demonstrate how, under certain assumptions, the impact of simulated, and existing traffic measures can actually be determined using air quality measurements avoiding the use of emission and dispersion models, each with their inherent uncertainties. We do this for a busy inner urban road in the city of Rotterdam, The Netherlands.
2. Case description and research questions There are four questions to be examined. Firstly does air pollution have to be measured for operational decision making on traffic management? If yes, a dense network of monitoring or sensor equipment is needed to make informed decisions (e.g. as in the MESSAGE project, North et al., 2009). Secondly, is it possible instead to use generic information such as traffic numbers and speed, meteorological data, etc. for management decisions and how does this affect the impact of the measures? In that case, using existing proxy information would be an efficient and cheap alternative. Thirdly, temporary measures (to be activated depending on environmental criteria) are compared to permanent measures that are always invoked at certain times. Lastly the beneficial potential of shifting emissions from unfavorable to more favorable dispersion conditions is examined. In this study we use measured concentrations of BC and NOx, in a busy urban street. In the morning the dominant traffic flow is towards the city center and in the afternoon in the opposite direction. The city employs an intelligent traffic management system (ITS) that manipulates the traffic lights in such a way that a minimum speed of 25 km·h−1 is maintained along a 1.2 km corridor. Seven weeks of traffic observations, evenly spread over the year 2011, resulting in a dataset of 1001 h were analyzed. The data on fleet composition (private cars and vans, medium duty and heavy duty trucks), traffic density and traffic velocity were obtained from the Rotterdam City Traffic Department. The average hourly traffic density was 1056, 62 and 8 vehicles for respectively private cars, medium duty and heavy duty vehicles. The street has two lanes in either direction. The street canyon (roads, parking places, pavement) is approximately 45 m wide. The street is tree-lined. Hourly concentrations of BC and NOx were obtained from a kerbside monitoring station and from a nearby (b 500 m) background station. The meteorological
Fig. 1. The relation between traffic density and the average speed of the private vehicles.
information was obtained from the National Meteorological Institute from their site at Rotterdam Airport just north of the city. 2.1. Assumptions • The increment between the concentrations measured at the roadside monitoring station and the background monitoring station (ΔBC or ΔNOx) is due to traffic emissions. If the observed increment was b0 it was set to 0. • Using traffic emission factors (Velders et al., 2012) weights were given to the number of medium and heavy duty trucks while analyzing BC and NOx. Medium and heavy duty trucks are equivalent to 4.7 and 6.8 private vehicles respectively for BC and 19.0 and 29.4 for NO x. Using these factors a weighted traffic density Dw (expressed as private vehicle equivalent/h — pve·h− 1), is calculated. Note that at a given hour Dw is different while analyzing BC and NOx. • Vehicle emissions depend, to some extent, on the velocity and the velocity depends amongst others on the traffic density. Traffic interventions therefore tend to have multiple effects that each cause a change in emission per km. However in this case the Traffic Department succeeds in maintaining average speeds between 25 and 50 km·h− 1 at all times (see Fig. 1). We therefore assume that there are no major changes of the emission/vehicle/km as a function of the traffic speed and density Dw. Hence we assume that the observed street increment normalized by traffic density (ΔBC/Dw) at a given hour is mainly the result of the atmospheric dispersion conditions at that hour and rather independent of the traffic conditions. • If the traffic numbers do not actually change the normalized street increment ΔBC/Dw (or ΔNOx/Dw), the impact of an intervention in the traffic density can be assessed by simply multiplying the change of the traffic density Dw with the street increment per unit of traffic ΔBC/Dw at the given hour. This seems to be the case: the increment per unit of traffic remains rather constant, independent of the traffic numbers. Fig. 1 shows that removing/adding 500 vehicles hardly affects the speed so it is a fair assumption that ΔBC/Dw and ΔNOx/Dw remain constant upon changes in Dw. In this way we can use measured air quality data to estimate the impact of hypothetical interventions in Dw. 2.2. Traffic scenarios studied Most of the ATM measures and ITS systems are used to minimize stop and go traffic and idling times by assuring that the traffic density doesn't exceed the road network capacity. This can be done by optimizing traffic light cycle times in real-time, by gating (Tate and Bell, 2000) e.g. if necessary creating congestion elsewhere where it is environmentally
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx
3
less important, or by reducing maximum speed (typical on motorways) to homogenize the traffic flow. Several of the scenarios simulated in this study actually remove (temporarily) part of the traffic. This is not a common intervention and their practical feasibility is not always obvious.1 However, the example can easily be used to study the impact of the different decision making processes. For the environmentally prompted measures the simulated intervention consists of the removing of all medium and heavy duty trucks. This traffic ban is lifted as soon as the criteria that trigger the measure are no longer met and the previously avoided emission occur at a hopefully more favorable moment. The temporary adaptive traffic management interventions are invoked using the following criteria: 1. Poor dispersion indicated by a mixing height ≤ 200 m. AND Dw ≥ 2000 pve·h−1 2. Poor dispersion indicated by a wind speed ≤ 2 m·s−1. AND Dw ≥ 2000 pve·h−1 3. Air quality observations are used to decide on a measure. The top 10% of the hourly observed road increment for NOx and BC can be used to invoke a measure. By studying historic data a threshold value can be set. AND Dw ≥ 2000 pve·h−1 4. NO2 has a distinct diurnal and yearly concentration pattern. A concentration reference pattern was established that takes into account the typical concentration variation. If an hour has a concentration of ≥20 (μg·m−3) above the reference concentration for that hour, it is considered as too high and a measure is needed (see van den Elshout et al., 2009a). In addition to the temporary measures that are invoked based on environmental criteria, it is also possible to create permanent short duration traffic measures irrespective of environmental criteria. These traffic interventions are either truly permanent (numbers 6 and 7) or active at a fixed hour of the day/day of the week (number 5). Measures 6 and 7 have actually been implemented at the road under consideration. 5 On working days between 6.00 and 8.00 h Dw is reduced: 500 pve·h− 1 are removed in this busy period either by banning medium and heavy duty trucks or by an imaginary road pricing scheme. The vehicles are allowed back-in between 09.00 and 11.00 h when average dispersion conditions are better. 6 Using an ITS the Traffic Department manages the traffic lights in such a way that a minimum velocity of 25 km/h is maintained. The likely impact of this measure is estimated assuming that stagnating traffic would otherwise occur on working days for at least 30 min during both the morning and evening rush-hour. Emission factors for stagnating traffic are 1.5 and 1.4 times higher respectively for NOx and BC (Velders et al., 2012). So the gain of the measure is 0.5 ∗ ΔNOx and 0.4 ∗ ΔBC at the given hour. 7 A park-and-ride facility near a subway station next to this route. On working days this removes 350 cars during the morning and evening peak hours (7.00–8.00 h and 17.00–18.00 h). Because of this measure only the number light vehicles is reduced. This reduces the average pve·h−1 from 2681 to 2331 during the hours considered. Using the average ΔBC/Dw and ΔNOx/Dw ratios during these hours the impact can be calculated.
Fig. 2. The diurnal pattern of traffic density and the fraction of medium and heavy duty vehicles.
neutral atmospheric stability) strongly resembles the traffic pattern so both are good indicators for traffic related air pollution. Fig. 4 shows ΔBC/Dw for three atmospheric stability classes. It shows that ΔBC/Dw (or ΔNOx/Dw not shown here) remains rather constant throughout the day between 06.00 h and 18.00 h despite considerable variation in traffic numbers (Fig. 2). Fig. 4 also shows that the road increment per unit of traffic depends on the prevailing atmospheric stability: it shows the influence of dispersion conditions on concentrations given the same average emissions. Neutral stability is the only atmospheric stability class that occurs throughout the day. Stable and unstable conditions lead to higher concentrations with the same amount of emissions. Stable conditions often occur in the morning and evening. The Monin–Obukov length (calculated by the KEMA Stacks model, Erbrink et al., 2012) was used to determine the stability classes. Atmospheric stability determines the concentrations resulting from an emission and explains a considerable amount of the variation in the road increment (van den Elshout et al., 2009b). Note that the impact of atmospheric stability on the concentrations is less when the emissions occur in narrow and deep street canyons. If stable atmospheric conditions can be avoided (often occurring in the morning), in theory this can reduce the concentrations and be the basis of a temporary traffic measure: reducing Dw when ΔBC/Dw is above average. This is the logic behind measure 5 (and 7).
3. Results and discussion 3.1. Concentrations and dispersion conditions Fig. 2 shows some traffic characteristics. It appears that more vehicles enter than leave the town using this road. Fig. 3 shows the close relation between ΔBC and ΔNOx. The diurnal pattern (during 1 Banning trucks based on unpredictable environmental criteria, as occurs in some of the examples studied is not practical. In a permanent (hence predictable) measure such as banning trucks during the morning hours when pollution is likely to be high is feasible.
Fig. 3. Diurnal pattern of the road increment for BC and NOx on working days.
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
4
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx
Fig. 4. Diurnal pattern of ΔBC/1000 pve for the three atmospheric stability classes on working days (3 hour averages due to scarcity of data per hour/per class).
3.2. Impact and net-impact If a measure is taken during an hour with unfavorable conditions there obviously is a positive impact. However, if traffic is banned from
an area at a given hour, it will have to enter at another time. So either the same amount of traffic has to enter at a later hour, or via different roads. This causes additional pollution partly offsetting the gains at the site of, and during the measure. We therefore calculate the impact during the measure and the net impact, taking into account the emissions shifted to another time or place (not studied here). The net impact is particularly important for BC. As mentioned, this is a health relevant pollutant with no safe threshold (similar to PM10/PM2.5). Hence it doesn't matter when or where the concentrations occur. Only if traffic could be channeled through a less populated area an exposure benefit could be achieved. For NO2 the situation is slightly different. If the traffic measure is devised for the purpose of meeting the limit values (and not to reduce exposure) spreading traffic over the network in such a way that in all streets the limit values are met, could be an option. The ethics of increasing someone's exposure to the benefit of others are not considered in this case. Spreading traffic over different roads, however, has practical limitations. In Rotterdam (and in most cities?) the road network operates at near full capacity during peak hours and the practical possibility to reroute traffic in the face of adverse air quality conditions is very limited. Fig. 5 shows the road under study (thick line) and a theoretically alternative route to the city center. Both are equally busy and trying to shift traffic from one route to the other is unlikely to yield net benefits. Gating (Tate and Bell, 2000), not studied in this paper, might have some
Fig. 5. Study area (thick line) and routes into the city center coming from the southern ring road.
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx
potential as a temporary adaptive management strategy by queuing traffic south of the road fork. Besides, since this road is still in the built environment the ethics are questionable. A permanent, kind of gating situation was implemented on the northern part of town on the motorway leading into town. Before this intervention, traffic entering into town from the north at a speed of 100 km·h−1 often led to highly dynamic traffic situations and/or stagnation with relatively high emissions per km in the approach of the T-junction with the northern ring-road. When the maximum speed was reduced to 80 km·h− 1, stagnation was shifted to the entrance of the town, at the point where the change from 100 to 80 km.h−1 took place. Within the city limits, traffic moving relatively smoothly at 80 km·h−1, saw their NOx emissions reduced by 20–30% (Wesseling et al., 2003). Though this measure was presented as a speed limit, it operates in a similar way as the gating technique: assuring a smooth traffic flow where it matters. Reducing traffic dynamics by making personnel vehicles drive as fast as heavy duty vehicles (speed limit at 80 km·h−1) on a motorway is an effective way of reducing the emissions per km. This was further confirmed by Keuken et al. (2010).
3.3. Results and discussion The traffic related air pollution characteristics of the study area are shown in Table 1. It shows the average weighted traffic density, the road increment and the road increment per 1000 pve. These average characteristics can be compared to the situations when the traffic interventions are invoked (Table 2). The results of the seven interventions are summarized in Table 2. Looking at the impact during a traffic measure it shows that those based on the actual occurrence of high concentrations (3 and 4) do pick, as expected, the hours where traffic has the highest impact (where ΔBC/Dw is high). See the two right most columns in Table 2. If the measure was based on meteorological parameters like mixing height and wind speed (like in measures 1 and 2), hours with a lower concentration increment per unit of traffic are picked. Nevertheless, the measures are timed approximately at the right moment: the ratios of ΔBC/Dw that were selected are much higher than the sample median (Table 1). The permanent measures during the (morning) rush-hours (5–7) have ΔBC/Dw ratio's slightly higher than the median. The net benefits show that the temporary restriction measures are not very successful. The hours where the measure is invoked apparently occurs on unfavorable days where the dispersion remains quite poor for prolonged periods. Allowing the banned vehicles to move again still creates substantial concentration rises, almost nullifying the previous gains or making the net situation worse. Sometimes it was even impossible to allow all the cars banned during the day to start to move again (the adverse environmental conditions remain throughout the day). The theory that traffic emissions during hours with unfavorable dispersion can be traded for emissions during better conditions, e.g. later on the day, hardly seems to work in reality in our case. The “road pricing”/morning rush-hour restriction scheme (intervention 5) shifting cars from the rush hours to a more quiet time a few hours later gives a lower impact during the measure than the temporary
Table 1 Summary statistics, whole sample studied (1001 h). Average Dw (pve·h−1)
Whole Sample Private cars
Average road increment (μg·m−3)
Median road increment per 1000 pve (μg·m−3)
BC
NOx
BC
NOx
BC
NOx
1400 75%
2445 43%
1.63 1.23
46 20
1.08
14.7
5
measures based on environmental criteria, but the net impact is slightly higher. As can be seen from the road increment/1000 pve during this measure, these morning hours are only marginally worse than the median conditions. However the “payback” hours are not unfavorable either (and on average better than the intervention hours) so a net benefit remains. The ITS scheme (6) that keeps cars moving at a reasonable speed gives the highest net impact. Without this scheme the concentrations would have been substantially higher during the hours affected. Flow optimizations that assure smoothly running traffic are by far the best interventions as rapid speed changes lead to extra emissions (e.g. van Driel, 2007; Servin et al., 2006; Gense et al., 2001; all in Mahmod, 2011). The measures taken to optimize the road network performance in general also seem to serve the environment quite well. Similarly the park and ride scheme that removes 350 cars (in the morning as well as the afternoon rush-hour), is quite effective in reducing concentrations during these hours. Scenarios 6 and 7 are the ones with hardly any or no “payback effects”. In scenario 6 the emissions are reduced by avoiding stop and go traffic (reducing emissions per km) and in scenario 7 traffic numbers are actually reduced. Looking at the decision making questions: if environmental criteria are used and if these are based on actual air quality measurements (the interventions are done during the hours with the highest concentrations). If other environmental criteria are used to estimate when air quality is high the gross impact of the intervention (e.g. measure 2) would be on average 65% of the measure with the highest gross impact (measure 3). On the other hand the hours with high concentrations seem to occur in batches: they tend to last for the major part of the day or even days, resulting in substantial “payback” and very small net benefits. Considering the paybacks, it is doubtful that there will be net advantages of decision making based on actual air pollution measurements over estimating peak situations based on traffic numbers and a meteorological parameter. In our simulated case they were virtually absent. The third question examines whether day of the week and time of the day are sufficient to identify hours when an intervention is needed. It turns out that scenario 5 directed at the morning rush-hour during weekdays, is about as effective as the scenarios 1 and 2 where the intervention was timed on generic environmental information (if the same amount of traffic equivalents would be involved). The net results are even slightly better. This suggests that the added advantage of decisionmaking based on measurements and on proxy environmental indicators is limited. If simply the rush-hours are identified, at least in our case, the net air quality results would be similar while avoiding a lot of complicated real-time decision making. Looking at scenario 5 where emissions are shifted from morninghours with on average less favorable dispersion, to a moment later on the day it turns out that there is a small net benefit, e.g. environmentally there is a (small) benefit from allowing medium and heavy duty traffic into town only after 09.00 h or 10.00 h when the mixing layer is higher and the impact of the same emissions on the ambient concentrations is smaller. A restricting scheme such as scenario 5 would have a higher impact if second order effects did occur, e.g. if it also contributed to reduce stop and go traffic and excess idling. This didn't occur in the first place in our experimental set-up. The permanent measures with no real-time decision-making and hardly any payback emissions have the highest net impact (scenarios 6 and 7). 3.4. Health effects short-term and long-term exposure/measures Health effects of traffic related air pollution occur on two time scales, short-term (day/days in case of PM) and more importantly long-term. It is the permanent exposure to any level of particulate air pollution that causes most health effects (WHO, 2013). The report mentions evidence that very short (1–2 h) exposure to very high concentrations cause health effects but notes that the effects of the sum of all short-term
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
6
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx
Table 2 Results of the different traffic measures. case
Average # pve involved in measurea
1 2 3 4 5 6 7
641 645 718 585 500 2331 350
# hours when measure is invoked
Average impact of the measure during the measure (μg·m−3)
hours
%
BC
104 73 71b 184 70 35 70e
10% 7% 7% 18% 7% 3% 7%
0.85 0.89 1.37 0.90 0.59 1.04 c) 0.39
Average net impact per hour of measure (μg·m−3)
NOx
BC
88 61 9 37c 6
b0 0.03 0.04 b0 0.06 0.94d 0.39
Observed average road increment per 1000 pve when measure is invoked (μg·m−3)
NOx
BC
NOx
b0 b0 1.7 33d 6
1.33 1.38 1.91 1.54 1.18 1.12 1.12
30 26 18 18 18
a
Weighting shown based on BC emisison factors (NB: for measures 3–7, the NOx results are calculated with the NOx weighting factors). The 90 percentile of ΔBC/pve·h−1 would suggest 100 out of 1001 h in the sample. However 29 h were lost as the criterion of having sufficient traffic was not met, e.g. early morning hours with very poor dispersion and also a low traffic density. c The impact is calculated (example BC, see measure 6) as 0.4 ∗ # pve·h−1 ∗ ΔBC/pve·h−1. d If improving the flow on the main road causes extra stagnation on the side roads the net benefit is smaller than the benefit during the measure. According to the Traffic Department, the cycle times on the branches are hardly affected and no negative impact of measure is expected. A maximum trade-off of 10% was included. e We assume that the park and ride facility fills in 1 h during the morning peak hours and empties in the evening rush-hour so the benefit occurs twice a day. b
exposures is smaller than the long-term exposure effects. Unless more evidence is emerging that it is useful to avoid very short (hours) exposure peaks the efforts to particularly address peak concentration hours with ATM seem complicated with only a small contribution to the reduction of overall exposure. Looking at absolute improvements (net effect during the measure ∗ fraction of time) of the year average concentrations, even the most effective measure considered (ITS) has a small overall impact. The year average BC road increment is reduced by 0.03 μg·m− 3. For NOx the road increment is reduced by a maximum of 1.2 μg·m− 3 (by the ITS measure). This amounts to net year average road increment concentration reductions of some 2%. This would of course be higher if without this measure there would have been substantial stop and go traffic. In this study we have conservatively estimated this at one hour per day only. Apart from the net impact of the measure, the time that it can have an impact is, by its temporary nature, also quite short, hence a small overall impact. For another example, look at the Flanders smog-day scheme (Lefebvre et al., 2011). During predicted smog episodes (a few days a year) the maximum speed on the motorway is reduced from 120 to 90 km·h−1. During the measure the Elemental Carbon (EC) concentrations are reduced by 30% immediately adjacent to the motorway. However, like in the measures analyzed in this study, the fraction of time the measure is applied is small, limiting the overall exposure reduction impact. The permanent speed limit on the northern access motorway in the study from Wesseling et al. (2003) reduces the year average traffic contribution by 20–30% (for NOx). If a constant speed reduction is unacceptable a permanent speed reduction during the busiest hours (e.g. morning rush hour) would likely have a bigger impact.
4. Conclusions Temporary interventions to reduce traffic related air pollution at peak moments can be appropriately timed, either by air quality measurements or by using generally existing environmental data (such as wind speed) and traffic counts. The latter simplifies the decision making on such interventions. If one is interested in such measures there is no need to first create elaborate monitoring or sensor networks. One could alternatively opt for permanent short duration adaptations (at fixed times) or include environmental criteria in the overall optimization of the existing traffic management systems. This avoids complex decision making and is likely to have similar or even larger effects. As long as epidemiological evidence continues to suggest that longterm exposure is more important than short-term peaks permanent
measures rather than short-term peak-shaving interventions are needed to achieve substantial exposure reductions. Acknowledgments Rotterdam Traffic Department, dS + V, kindly supplied detailed traffic data. References Erbrink Hans, Kokmeijer Esther, de Wolff Joost. An extended OSPM model for pollutant concentrations in street canyons with detached houses and specified traffic lane width. ITM–NATO/SPS International Technical Meeting on Air Pollution Modelling and its Application: Proc; 2012. Gense NLJ, Wilmink IR, van den Burgwal HC. Emissions and congestion: Estimation of emissions on road sections and the Dutch motorway network. TNO report, TNO Automotive; 2001. Hodges N, Bell M, Galatioto F, Hill G, Neasham J, Suresh V. Sustainable network management — the integration of intelligent transport systems and “grid enabled” pervasive sensors. Newcastle University; 2009 [No. CS-TR-1167. www.cs.ncl.ac.uk/publications/trs/papers/1167.pdf. (Accessed July, 2013)]. Janssen NAH, Hoek G, Lawson-Simic M, Fischer P, Bree van L, Brink van H, et al. Black carbon as an additional indicator of the adverse health effects of combustion particles compared to PM10 and PM2.5. Environ Health Perspect 2011;119(12):1691–9. Keuken M, Jonkers S, Wilmink IR, Wesseling J. Reduced NOx and PM10 emissions on urban motorways in The Netherlands by 80 km/h speed management. Sci Total Environ 2010. http://dx.doi.org/10.1016/j.scitotenv.2010.03.008. Keuken Menno, Zandveld Peter, van den Elshout Sef, Janssen Nicole, Hoek Gerard. Air quality and health impact of PM10 and EC in the city of Rotterdam, The Netherlands in 1985–2008. Atmos Environ 2011;45(30):5294–301. Koning Anne de, van Baalen Janneke, Voogt Marita, Stelwagen Uilke, Turksma Siebe. EcoFLEX: Het verbeteren van de luchtkwaliteit met behulp van Groen Dynamisch VerkeersManagement, gebaseerd op real time luchtkwaliteitsmetingen. www. verkeerskunde.nl/Uploads/2011/11/Bijdrage105.pdf, 2011. [(Accessed July, 2013)]. Lefebvre W, Fierens F, Trimpeneers E, Janssen S, Van de Vel K, Deutsch F, et al. Modeling the effects of a speed limit reduction on traffic-related elemental carbon (EC) concentrations and population exposure to EC. Atmos Environ 2011;45: 197–207. Mahmod M. Using co-operative vehicle-infrastructure systems to reduce traffic emissions and improve air quality at signalized urban intersections [TRAILThesis series. http:// doc.utwente.nl/76481/1/thesis_M_Mahmod.pdf. (Accessed July, 2013)]; 2011. North R, Cohen J, Wilkins S, Richards M. Hoose Neil, et al. Field deployments of the MESSAGE system for environmental monitoring. Traffic Engineering & Control 2009;50(11):484–8. (Hemming Information Services, ISSN 0041-0683). Servin O, Boriboonsomsin K, Barth B. An Energy and Emissions Impact Evaluation of Intelligent Speed Adaptation. In Intelligent Transportation Systems Conference. Toronto, Canada: ITSC’06. IEEE; 2006. Tate J, Bell MC. Designing, implementing and evaluating a gating strategy to reduce pollution levels in sensitive areas of the urban environment. 10th IEE International Scientific Symposium on Transport and Air Pollution, Avignon, France; 2000. van Baalen J, De Koning A, Voogt M, Stelwagen U, Turksma S. ECOFLEX: improving air quality with green dynamic traffic management based on real time air quality measurements. 18th Intelligent Transport Systems World Congress, Orlando, United States, 2011; 2011.
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
S. van den Elshout et al. / Science of the Total Environment xxx (2014) xxx–xxx van den Elshout Sef, Mahmod Mohamed, Ingrid Arts, van Arem Bart. Decision making on short term traffic measures to influence traffic related air pollution. ETTAP 2009. Actes INRETS n°122; 2009a. van den Elshout Sef, de Gier Cornelis, Arts Ingrid. PM10, PM2.5 and black smoke concentrations in urban background and traffic situations — what is a good traffic indicator? Conference Proceedings ETTAP 2009. Actes INRETS n°122; 2009b. van Driel CJG. Driver Support in Congestion: An assessment of user needs and impacts on driver and traffic flow. PhD thesis, TRAIL Research School; 2007. Velders GJM, Aben JMM, Jimmink BA, Geilenkirchen GP, van der Swaluw E, de Vries WJ, et al. Grootschalige concentratie- en depositiekaarten Nederland: Rapportage 2012.
7
RIVM Rapport 680362002. www.rivm.nl/Bibliotheek/Wetenschappelijk/Rapporten/ 2012/juni/Grootschalige_concentratie_en_depositiekaarten_Nederland_Rapportage_ 2012, 2012. [(Accessed 04-07-2012)]. Wesseling JP, Hollander K, Teeuwisse S, Keuken MP, Spoelstra H, Gense R, et al. Onderzoek naar effecten van de 80 km/u maatregel voor de A13 op de luchtkwaliteit in Overschie. TNO rapport 2003/258; 2003. WHO. Review of evidence on health aspects of air pollution — REVIHAAP Project. http://www.euro.who.int/__data/assets/pdf_file/0004/193108/ REVIHAAP-Final-technical-report-final-version.pdf, 2013. [(Accessed July, 2013)].
Please cite this article as: van den Elshout S, et al, Adaptive traffic management in cities — Comparing decision-making methods, Sci Total Environ (2014), http://dx.doi.org/10.1016/j.scitotenv.2013.12.084
