Change-Point Detection of Gaseous and Particulate Traffic-Related

Oct 13, 2006 - the introduction of congestion charging, the increment in. NO2 above ... estimates show that the traffic volume has reduced by 18% comp...
0 downloads 0 Views 304KB Size

Environ. Sci. Technol. 2006, 40, 6912-6918

Change-Point Detection of Gaseous and Particulate Traffic-Related Pollutants at a Roadside Location DAVID C. CARSLAW,* KARL ROPKINS, AND MARGARET C. BELL Institute for Transport Studies, University of Leeds, Leeds LS2 9JT, U.K.

An 8-year (1998-2005), hourly data set of measurements of NOX, NO2, PM10, PM2.5, and PMcoarse (defined as PM2.5-10) from a busy roadside location in central London has been analyzed to identify important change-points in the time series using a cumulative sum (CUSUM) technique. Randomization methods were used to estimate the uncertainty level associated with the change-points with uncertainty intervals derived using a bootstrap approach. The results show that there is a clear change-point increase for NO2 coinciding with the introduction of the London congestion-charging in February 2003 (95% confidence interval from January-March 2003). At this time there was both an increase in bus numbers and buses fitted with catalyzed diesel particulate filters, which increase direct emissions of NO2. A highly statistically significant changepoint was also observed for PMcoarse (95% confidence interval from December 2002-February 2003), which also occurred close to the time of the congestion charge introduction and is most closely related to the increase in bus flows. The increase in PMcoarse at this time has largely compensated for reductions in the concentration of PM2.5, such that the concentration of PM10 has remained almost constant. Comparing the 2 years before and after the introduction of congestion charging, the increment in NO2 above background increased from 22 to 34 ppb and PMcoarse increased from 4 to 9 µg m-3. These results could have important implications for meeting European air quality standards that currently set limits for PM10 rather than PM2.5.

Introduction Over the past decade emissions from European vehicles have been progressively reduced due to ongoing improvements in vehicle technology to meet increasingly stringent emissions legislation. These improvements take time to affect total vehicle fleet emissions because of the time it takes for new vehicles to enter the fleet. Calculated emission trends therefore tend to be relatively smooth and do not show abrupt changes. There are some conditions, however, where a more abrupt change in emissions may be observed. If, for example, there is a step-change improvement in vehicle emissions brought about by the rapid introduction of new technology (e.g., a particle filter with an efficiency of over 90%), it is possible that this change could introduce an observable step change in total vehicle fleet emissions and observed concentrations. Another example is the introduction of a traffic * Corresponding author e-mail: [email protected] 6912



management change that changes the vehicle fleet emissions sufficiently to result in a measurable change in concentration. In London, the recent introduction, in February 2003, of the London congestion charging zone (CCZ) is one example of such a change. The CCZ levies a charge of £8 (≈$14) for certain vehicle types (principally cars, vans, and trucks) entering the zone during weekday periods from 07:00 to 18.30 h. Buses and taxis are exempt from the charge. Recent estimates show that the traffic volume has reduced by 18% compared with precongestion charge levels (1). Within the charging zone itself, which currently covers 22 km2 (1.3%) of Greater London, it is estimated that vehicular emissions of nitrogen oxides (NOX) and particles 95% were recorded for NOX, NO2, and PM10 at Marylebone Road, NOX and NO2 from North Kensington, and PM2.5 from Harwell. Lower data capture rates were recorded for PM2.5 at Marylebone Road (91.1%) and PM10 at Harwell (93.7%). For all NOX and NO2 measurements, data were available from January 1998-December 2005. PM10 and PM2.5 data were available from May 1998 at all sites except PM10 at Harwell, which were available from August 1998. PM10 data from Rochester (also a rural site ∼50 km from London) were therefore substituted for Harwell data in the period May-July 1998. Traffic flow was measured using a Golden River Marksman 660 Counter classifier loop induction system, located approximately 5 m from the monitoring site. The system provides continuous 15-min mean traffic count and vehicle speed estimates for each of the six traffic lanes and classifies vehicles into motorcycles, cars, cars with trailers, rigid trucks, articulated trucks, and buses. In the bus lanes, articulated buses are detected as articulated trucks. However, because the bus lane is strictly enforced, it is known with high confidence that these articulated vehicles are buses and not trucks. Cumulative Sum (CUSUM) Approach. The CUSUM approach was initially developed for industrial control purposes (6). More recently the technique has been adapted and used for a range of environmental problems where the detection of a change is important (7-9). A CUSUM chart is constructed by plotting the accumulated differences between a measured variable and a fixed value, such as the mean of the time series. Let x1, x2,...,xn be a time series of n data points with a mean xj , from which cumulative sums S0, S1,...,Sn are calculated, where S0 ) 0 and the ith CUSUM, Si, is given by

Si ) Si-1 + (xi - xj )

for i ) 1, 2,...,n


A randomization technique has been used to quantify the mostly likely timing of the change (10). A series of CUSUMs can be constructed by randomly reordering the time series and calculating a CUSUM for each. A test statistic can be derived that tests whether the observed CUSUM falls in the highest R % of the distribution of all CUSUMs. The test statistic used here considers the difference between the maximum and minimum CUSUM: i)n




∆S ) max (Si) - min (Si)


The ∆S for each randomized CUSUM can then be compared with the actual (measured) ∆Smeas, to determine where it lies in the distribution of randomized values. The confidence level that a change has occurred can be calculated by 100‚X/n, where X is the total number of randomizations where ∆S < ∆Smeas. Therefore, ∆Smeas is considered to be statistically significant at the 95% confidence interval if it lies in the top 5% of all randomized values of ∆S. It should be noted that a large number of randomizations are needed to describe the distribution. However, for the 8 years of monthly data, 10 000 randomizations were found to be adequate. The time that a change took place is the value of Si furthest from zero:


Smax ) max |Si|



Smax is the last point before a change occurred, and Smax+1 is the first point after the change occurred. These calculations therefore provide a best estimate of when a change occurred and the confidence level associated with that change. Bootstrap Estimate of Change-Point Uncertainty. A bootstrap approach has been used to estimate the uncertainty associated with a change-point, θ. Hinkley and Schechtman (11) describe several methods for estimating the confidence interval associated with a change in a mean. The method used presently uses the bootstrap approach in its basic form. Let µ1 be the mean of time series values x1 to xθ and µ2 be the mean from xθ+1 to xn, where n is the number of records in the time series. The estimation of the errors begins with the assumption that the errors are distributed according to the empirical distribution function based on the residuals of the observed fit:

et ) xt - µ1 (t ) 1,...,θ) et ) x t - µ 2


(t ) θ+1,...,n)


Many simulated time series were constructed by randomly sampling, with replacement, errors (e*1,...,e*n):

x*t ) µ1 + e*t

(t ) 1,...,θ)


x*t ) µ2 + e*t (t ) θ+1,...,n)


For each simulation, a new estimate of θ is made using the CUSUM estimate (eq 3). A total of 10 000 simulations were run. A normalized histogram of these simulations was derived, showing the probability that a change occurred for a particular month.

Results and Discussion Estimated Trend in Vehicle Emissions. Trends in vehicle emissions have been calculated using hourly data from the automatic traffic counter and U.K. emission factors for NOX and PM10 (12). Vehicle age distributions from the London Atmospheric Emissions Inventory (LAEI) were used to determine the different vehicle technology profiles (13). A comprehensive description of how these data were processed can be found elsewhere (3). Figure 1a shows that the total emissions of NOX declined by approximately 38% over the period 1998-2004. From July 1998 to December 2002, total NOX emissions declined by 36%, but from January 2003 to December 2005 they were only reduced by 8%. Most of the reduction in NOX has been due to cars, where emissions were reduced by a factor of 2 from July 1998-December 2005. There was a step-change reduction in car NOX emissions when the bus lane was installed in August 2001. NOX emissions from trucks have remained almost constant over the period, although there was a noticeable reduction that occurred in December 2001, which was sustained until around December 2004. Emissions of NOX from buses have increased over the period due to the increase in bus flows associated with congestion charging, where a step-change can be seen at the beginning of 2003. Exhaust emissions of PM10 (not shown) declined by approximately 50% over the same period and show similar features to the trend in NOX, e.g., the effect of the bus lane. Application to Marylebone Road. The monthly time series for Marylebone Road traffic flows and different pollutant species, with background concentrations subtracted, are shown in Figure 2. A visual inspection of these plots shows there is considerable variation in the monthly concentration of all species. It is also apparent at Marylebone Road that VOL. 40, NO. 22, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY



FIGURE 1. (a) Monthly trend in emissions of NOX from major categories of vehicle (July 1998-December 2004) and (b) vehicle flows at Marylebone Road recorded by the automatic traffic counter. The total vehicle flow is that for all 6 lanes. The bus and articulated bus flows are those for the bus lane.

FIGURE 2. Plots in the left panel show the monthly mean flow of cars, trucks, and buses and the increment in concentration above a background site for NOX, NO2, PM10, PM2.5, and PMcoarse. The red line is a locally weighted regression smoothing fit (15). The plots on the right panel show the corresponding CUSUMs. The vertical dashed lines show change-points that are statistically significant at the 95% confidence level. The histograms highlight the uncertainties associated with the timing of the change-point, normalized such that the area under them is equal to 100. there is very little seasonal variation in the concentrations of most species. There is a decrease in the concentration of 6914



NOX from 1998 to 2005 and an increase in NO2. The PM10 time series does not show any obvious trend but does display

TABLE 1. Estimated Change-Point Dates with Associated Confidence Levels and Confidence Intervals for Monthly Concentrations of NOX, NO2, PM10, PM2.5, and PMcoarsea species NOX NO2 PM10 PM2.5 PMcoarse


confidence e level

confidence interval





Feb 2001 Nov 2001 Nov 2002 Aug 2001 Feb 2003 Jul 1999 Feb 2001 Nov 2001 Aug 1999 Nov 2001 Jun 2002 Mar 200200 Jan 2003

100.00 99.79 100.00 100.00 100.00 99.70 99.47 99.79 100.00 100.00 100.00 100.00 100.00

Nov 2000-Jun 2001 Jul 2001-May 2002 May 2000-Apr 2001 Oct 2000-Nov 2001 Jan 2003-Mar 2003 Jun 199-Jun 2000 May 2000-Dec 2001 Nov 1999-Jun 2002 Aug 1999-May 2000 Jun 2001-Jun 2002 Apr 2002-Aug 2002 Dec 1999-Apr 2000 Dec 2002-Feb 2003

167 162 25 26 22 14 19 19 10 14 14 6 4

127 120 22 24 34 19 15 16 14 9 7 4 9

-40 -42 -3 -2 +12 +5 -4 -3 +4 -5 -7 -2 +5

ppb ppb ppb ppb ppb µg m-3 µg m-3 µg m-3 µg m-3 µg m-3 µg m-3 µg m-3 µg m-3

a Only change-points identified with a confidence level of >95% are shown. b The ‘from’ column is the mean concentration 24 months before the change-point, and the ‘to’ column is the mean concentration in the 24 months after the change-point.

a large peak during 1999. Previous work had attributed elevated concentrations of PM10 to building work close to the Marylebone Road site from June-December 1999 (14). The PM2.5 time series shows an increasing trend to 2002 and then decreases. Figure 2 (right panel) shows the calculated CUSUMs for the traffic flows and the different pollutants. For clarity, CUSUMS are only shown for the changes in bus and truck flows. These results show that a single change-point was detected for truck flows on November 2001 (95% confidence interval from July 2001 to February 2002). For buses, two candidate change-points were detected, both with narrow confidence intervals in August 2002 and January 2003, with the January change-point associated with the highest confidence (89% of the bootstrap uncertainty). A clear changepoint was also detected for cars in August 2001, with 100% of the bootstrap uncertainty in that month. The NOX and PM2.5 CUSUMs have some of the characteristics of that expected of a decreasing trend in concentration. The CUSUM for NO2 and PMcoarse shows the opposite behavior to NOX, i.e., a line that reflects a concentration that increases with time. The CUSUM for PM10 is different from the other species, and no clear upward or downward trend behavior is discernible. These results suggest there is evidence that concentrations of PM2.5 and NOX have decreased, while NO2 and PMcoarse have increased over the period 1998-2005. Based on a consideration of expected changes in emissions, the CUSUMs for NO2 and PMcoarse appear anomalous. The presence of a trend in the time series has the potential to result in the false detection of change-points. To overcome this potential problem, tests were also carried out by applying the CUSUM technique to time spans less than the full 96 month time series to check whether consistent change-points were identified to reduce the effect of the trend component. It is also important that the time-span considered is not too short: a time-span of less than 1 year could be susceptible to seasonal and other meteorological variations. Furthermore, because the aim of this work is to identify changes that are significant and persistent, short-term changes such as over 1 year are not of interest. A moving average time period of 48 months was therefore chosen for the analysis, which yielded 49 overlapping periods for analysis from January 1998-December 2005. This approach was used to determine whether consistent change-points were identified regardless of the start point of the time series. From these 49 sets of results, change-points that were significant at the 95% confidence level were identified, and the confidence intervals associated with them estimated. Frequently, the same change point was identified many times, and thus the point with the least uncertainty associated with it was selected as the best

estimate of change point time. Time spans of 24 and 36 months identified change-points that were consistent with the 48-month time span (i.e., the change-point associated with the least uncertainty was identified with all time spans). Change-Point Estimates in Relation to Traffic Flow and Composition. Table 1 summarizes significant change-points and their uncertainties for all pollutants. Also highlighted in Table 1 is the mean concentration of a species in the 24 months up to the change-point and the 24 months after it. These before-after concentrations provide an indication of the magnitude of the change. For NO2, the change in February 2003, coinciding with the introduction of congestion charging, was the most prominent, with 83% of the bootstrap uncertainty distribution in that month. There were several important changes in traffic flows that occurred at this time, which would have affected concentrations of NO2. First, there was a considerable increase in bus flows along Marylebone Road when congestion charging was introduced, as shown in Figure 1b and the CUSUM plot for buses in Figure 2. The increase in buses would have increased the emissions of NOX. However, no significant change in the concentration of NOX was identified at this time, as shown in Figure 2. Another important change that has occurred in recent years in London is the use of catalytic diesel particulate filters (CDPF) fitted to Transport for London (TfL) buses. By the end of 2005 there were more than 8000 TfL buses and >95% were fitted with CDPF (13). As highlighted by Carslaw (16), the use of these filters has progressively increased in London, such that by the end of 2005 all TfL buses were fitted with CDPF. It is known that these filters lead to much higher NO2/ NOX ratios than diesel vehicles not fitted with these devices. Typical diesel vehicle NO2/NOX emission ratios without CDPF are 10-15% by volume (17). Reported NO2/NOX ratios (by volume) are in the range of 30-50% (18-20). It seems likely, therefore, that the timing of the change-point in NO2 concentration is associated with both the increase in bus numbers and the proportion of buses fitted with CDPF. Despite the likelihood that buses are responsible for much of the observed increase in NO2 concentration since February 2003, it is unlikely they are the only contributor. This is because other vehicle types known to emit a higher proportion of their NOX in the form of NO2, e.g., diesel cars with oxidation catalysts, have also increased in recent years (21). Nevertheless, it is likely that bus emissions have a disproportionate effect on ambient concentrations of NOX and NO2 because the emissions are released from the lane closest to the monitoring site. If buses fitted with CDPF resulted in detectable increases in NO2 concentrations, it might also be expected that reductions in PM2.5 concentrations would be detectable. VOL. 40, NO. 22, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY



Elsewhere, CDPF systems are reported to reduce particle mass with an efficiency of over 90% (22-24). However, the CUSUM plots give no indication of a change in PM2.5 concentrations around the time of the start of congestion charging. There are two effects of importance here: the retrofitting of existing buses in the fleet with CDPF and the addition of new buses to the fleet fitted with CDPF. No significant increase in PM2.5 emissions would be expected from the addition of new buses (with CDPF) because they would not appreciably increase PM2.5 emissions, and thus it would, therefore, be difficult to detect a change. The most significant change in PM2.5 detected by the CUSUM occurred in June 2002 with a 95% confidence interval from AprilAugust 2002. The TfL program of fitting retrofitting CDPF was phased in over a period from 2000 to 2005. One explanation for a detected change in PM2.5 in 2002 could be that sufficient bus numbers were fitted with CDPF at this time that the concentration reduction became detectable. It is estimated that 52% of the bus fleet was fitted with CDPF in December 2002 (16). However, it is not known how this proportion has changed over time at specific locations such as Marylebone Road. The analysis of the PM10 time series does not indicate similar downward trend, which can be seen by comparing the CUSUM plots in Figure 2. The most significant changepoint identified for PM10 was in July 1999 (confidence interval from June 1999-June 2000) is associated with building work close to the site (14). Although not shown in Figure 2, a change-point was also identified in April 2003 with a confidence level of 91% (95% confidence interval from November 2002 to September 2004), where PM10 concentrations begin to increase. This change-point is close to the implementation of congestion charging. Further insights can be gained into the change in particle mass concentrations by considering PMcoarse concentrations. PMcoarse particles are derived from many potential road trafficrelated sources including exhaust, abrasion, and resuspension processes. However, it is estimated that only 10% of the mass of PM10 emitted by catalyst-equipped gasoline cars and diesel exhausts are in the coarse fraction (25). In certain locations it has been shown that particle resuspension can significantly affect PMcoarse concentrations (26). For Marylebone Road, Charron, and Harrison (5) have shown that trucks are largely responsible for PMcoarse particles and that emission factors suggest a much lower contribution from abrasion processes. These considerations imply that PMcoarse is unlikely to be affected much by the ongoing reductions in vehicular exhaust emissions but are much more likely to depend on traffic volume and vehicle type. These results were based on the use of rural colocated PM10 and PM2.5 measurements, which were used for background subtraction. It is likely that a clearer contribution from road traffic on Marylebone Road would have been apparent for these species (and PMcoarse) if colocated measurements had been available at an urban background location such as North Kensington with sufficient data capture. An increase in PMcoarse was estimated to occur in January 2003 with a 95% confidence interval from December 2002February 2003. Together with the change in NO2 concentration around this time, the timing of the change in PMcoarse has a high certainty with narrow confidence intervals, with 80% of the bootstrap uncertainty distribution in January 2003. PMcoarse concentrations increased from 4 µg m-3 in the 24 months prior to January 2003 to 9 µg m-3 in the 24 months after January 2003, which represents a substantial increase in concentration. Furthermore, as shown in Figure 2, this increase has been sustained up until December 2005. This increase in PMcoarse has mostly compensated for the comparatively large reductions observed in PM2.5, such that there is weaker evidence of an increase in PM10 at the beginning 6916



FIGURE 3. Diurnal variation in weekday bus flows and PMcoarse at Marylebone Road. The data represent the mean difference in values for the time period 2001-2002 subtracted from the period 20032004. All data are expressed in local time. of 2003. The timing of the change close to the introduction of congestion charging and increases in the concentration of NO2, which appear likely to be due to buses, might indicate that increased bus flows are associated with increased in PMcoarse. To investigate the likelihood that the increase in PMcoarse observed at Marylebone Road could be due to buses, the weekly cycle of traffic flow and PMcoarse concentration has been considered. The change in PMcoarse from 2001 to 2002 compared with 2003-2004, i.e., 2-year periods either side of the estimated change-point for PMcoarse in January 2003, has been analyzed. Data were filtered for weekdays only because of the consistent diurnal variation in PMcoarse and traffic flow for those days. Furthermore, data were corrected for British Summer Time (BST) to local time to ensure the diurnal change in emissions was on a consistent time-basis. Figure 3 shows the diurnal increment in PMcoarse and total bus flow. In both cases there is a clear increase for each hour. Moreover, the pattern of change shows similarities. The Pearson correlation test for these data yielded a coefficient of 0.67, which was significant at the p ) 0.01 level. Neither the correlation between the flow of cars and PMcoarse nor trucks and PMcoarse was significant (cars ) -0.10 and trucks ) 0.27). The flow of cars decreased in all but 1 h of the day (05:00), which was probably due to the CCZ. Although there was little change in the daily flow of trucks post- and precongestion charging, there is evidence to suggest flows of these vehicles have been redistributed throughout the day. For example, flows of trucks increased by 48 vehicles per hour at 06:00 (more than double the increase of any other hour), which could be the result of vehicles entering the zone before the congestion charge is levied at 07:00. The increase in trucks at 06:00 might also contribute to the PMcoarse profile shown in Figure 3. These results therefore appear to show that the increase in PMcoarse observed from January 2003 onward is most closely associated with an increase in bus flows. The traffic data in Figure 1b show that there has been a large increase in the flow of articulated buses along Marylebone Road. The flow of these vehicles also increased markedly when congestion charging started, such that they now account for approximately one-third of the total flow of buses. Articulated buses are considerably heavier than the other types of buses (typically 18 tons vs 10 tons unladen weight). As shown by Gillies et al. (27), there is a strong linear relationship between vehicle weight and particle resuspension rates. The increased number of heavier buses in the traffic lane closest to the sampling inlets may be an important contributory factor affecting the observed increase in PMcoarse.

Unfortunately, there are no long-term measurements of particle composition at Marylebone Road, which have been shown to be useful in source apportionment studies (2729). In particular, the measurement of metals related to tire and brake wear (e.g., Ba, Zn, Cu, Sb, Fe) and elements associated with crustal material (e.g., Si) would have been beneficial. Measurements of these species in the fine and coarse modes at this location would help confirm the factors that have led to increased concentrations of PMcoarse. The CUSUM results also indicate a change in NOX, NO2, PM10, and PM2.5 that occurred around November 2001. There is more uncertainty in the timing of this change compared with the changes identified for NO2 and PMcoarse at the beginning of 2003. For all of these species, a reduction in concentration was observed around this time. Flows of total traffic decreased by 11% after the bus lanes were operating compared with the period before their operation, but as shown in Figure 1a, the change in total emissions of NOX is similar in magnitude to the variability in emissions on a monthly basis. Of more importance was a reduction in truck traffic in December 2001 (i.e., coinciding with a changepoint in November 2001), which is shown by the clear reduction in estimated emissions of exhaust NOX and PM10 shown in Figure 1a and the CUSUM plot in Figure 2 for trucks. Comparing the total emissions of NOX and PM10 in the 3 months pre- and postchange indicates that emissions were reduced by 14 and 20%, respectively. It is not known, however, what caused this change in truck traffic at this time. This work highlights several issues of relevance to policy makers. For the 8-year record of atmospheric measurements at Marylebone Road it has been shown that concentrations of NO2 have increased and concentrations of PM10 have changed very little, despite improvements in vehicle emissions technology. In Europe, the atmospheric concentration of these two pollutants is regulated through air quality directives. The increased provision of buses has been central to congestion charging, by providing an alternative transport mode to the private motor vehicle. This work highlights that it is very likely that the use of CDPF filters fitted to buses have made a major contribution to the observed increase in NO2 concentrations. However, such an increase might be seen as an acceptable tradeoff for controlling fine particle emissions, which have been shown to be more injurious to health (30). It is likely that reduced concentrations of PM2.5 due to the increased use of CDPF have been detected in this study, albeit with less certainty than the increase in NO2. However, increases in PMcoarse concentrations have offset the reduction PM2.5, although more work is required to confirm whether an increase in bus flows has led to increased concentrations of PMcoarse. Currently, EU air quality limits are set for PM10 and not PM2.5, and it is, therefore, a concern that PM10 concentrations have remained stable.


(6) (7) (8) (9)

(10) (11) (12) (13) (14)

(15) (16) (17)





(22) (23)



David Carslaw would like to thank the University of Leeds for funding his University Research Fellowship.


Literature Cited


(1) TfL. Central London Congestion Charging: Impacts Monitoring. Third Annual Report; Transport for London: 2005. (2) Beevers, S. D.; Carslaw, D. C. The impact of congestion charging on vehicle emissions in London. Atmos. Environ. 2005, 39, 1-5.


(3) Carslaw, D. C.; Beevers, S. D. Investigating the Potential Importance of Primary NO2 Emissions in a Street Canyon. Atmos. Environ. 2004, 38, 3585-3594.


(4) AQEG. Particulate Matter in the United Kingdom; Air Quality Expert Group. Report prepared by the Air Quality Expert Group for the Department for Environment, Food and Rural Affairs;


Scottish Executive; Welsh Assembly Government; and Department of the Environment in Northern Ireland; 2005. Charron, A.; Harrison, R. M. Fine (PM2.5) and coarse (PM2.5-10) particulate matter on a heavily trafficked London highway: Sources and processes. Environ. Sci. Technol. 2005, 39, 77687776. Page, E. S. Cumulative Sum Charts. Technometrics 1961, 3, 1-19. Manly, B. F. J.; MacKenzie, D. A cumulative sum type of method for environmental monitoring. Environmetrics 2000, 11, 151166. MacNally, R.; Hart, B. T. Use of CUSUM methods for waterquality monitoring in storages. Environ. Sci. Technol. 1997, 31, 2114-2119. Newchurch, M. J.; Yang, E. S.; Cunnold, D. M.; Reinsel, G. C.; Zawodny, J. M.; Russell, J. M. Evidence for slowdown in stratospheric ozone loss: First stage of ozone recovery. J. Geophys. Res., [Atmos.] 2003, 108. Manly, B. F. J. Randomization, bootstrap and Monte Carlo methods in biology, 2nd ed.; Manly, B. F. J., Ed.; Chapman & Hall: London, 1997. Hinkley, D.; Schechtman, E. Conditional Bootstrap Methods In The Mean-Shift Model. Biometrika 1987, 74, 85-93. Barlow, T. J.; Hickman, A. J.; Boulter, P. G. Exhaust emission factors 2001: database and emission factors; Report number PR/SE/230/00; Transport Research Laboratory: 2001. LAEI. The 2001 London Atmospheric Emissions Inventory; The Greater London Authority: 2003. Fuller, G. W.; Green, D. The impact of local fugitive PM10 from building works and road works on the assessment of the European Union Limit Value. Atmos. Environ. 2004, 38, 49935002. Cleveland, W. S. Robust Locally Weighted Regression And Smoothing Scatterplots. J. Am. Stat. Assoc. 1979, 74, 829-836. Carslaw, D. C. Evidence of an increasing NO2/NOX emissions ratio from road traffic emissions. Atmos. Environ. 2005, 39, 4793. Latham, S.; Kollamthodi, S.; Boulter, P. G.; Nelson, P. M.; Hickman, A. J. Assessment of Primary NO2 Emissions, Hydrocarbon Speciation and Particulate Sizing on a Range of Road Vehicles; Transport Research Laboratory (TRL): 2001. Ayala, A.; Kado, N. Y.; Okamoto, R. A.; Homen, B. A.; Kuzmicky, P. A.; Kobayashi, R.; Stiglitz, K. E. Diesel and CNG heavy-duty transit bus emissions over multiple driving schedules: Regulated pollutants and project overview. SAE 2002, 2002-01-1722. Guo, G.; Xu, N.; Laing, P. M.; Maricq, M. M.; Hammerle, R. H. Performance of a catalyzed diesel particulate filter system during soot accumulation and regeneration. SAE 2003, SAE 2003-010047. Shorter, J. H.; Herndon, S.; Zahniser, M. S.; Nelson, D. D.; Wormhoudt, J.; Demerjian, K. L.; Kolb, C. E. Real-time measurements of nitrogen oxide emissions from in-use New York City transit buses using a chase vehicle. Environ. Sci. Technol. 2005, 39, 7991-8000. Carslaw, D. C.; Beevers, S. D. Development of an urban inventory for road transport emissions of NO2 and comparison with estimates derived from ambient measurements. Atmos. Environ. 2005, 39, 2049-2059. Richards, P.; Kalischewski, W. Retrofitting of diesel particulate filters - particulate matter and nitrogen dioxide. Soc. Automot. Eng. 2003, 2003-01-1883. Su, Q.; Huang, Y.; Weiguo, L.; Li, T.; Chen, J. Research on the continuous regeneration of diesel particulate filter by NO2. Soc. Automot. Eng. Jpn. 2001, 2001-08-0146. Mayer, A.; Evequoz, R.; Wyser-Heusi, M.; Czerwinski, J.; Matter, U.; Graf, P. Particulate traps used in city-buses in Switzerland. SAE 2000, 200-01-1927. APEG. Source apportionment of airborne particulate matter in the United Kingdom; Department of the Environment, Transport and the Regions: London, 1999. Abu-Allaban, M.; Gillies, J. A.; Gertler, A. W.; Clayton, R.; Proffitt, D. Tailpipe, resuspended road dust, and brake-wear emission factors from on-road vehicles. Atmos. Environ. 2003, 37, 52835293. Gillies, J. A.; Etyemezian, V.; Kuhns, H.; Nikolic, D.; Gillette, D. A. Effect of vehicle characteristics on unpaved road dust emissions. Atmos. Environ. 2005, 39, 2341-2347. Gillies, J. A.; Gertler, A. W.; Sagebiel, J. C.; Dippel, W. A. On-road particulate matter (PM2.5 and PM10) emissions in the Sepulveda Tunnel, Los Angeles, California. Environ. Sci. Technol. 2001, 35, 1054-1063. Janssen, N. A. H.; VanMansom, D. F. M.; VanDerJagt, K.; Harssema, H.; Hoek, G. Mass concentration and elemental




composition of airborne particulate matter at street and background locations. Atmos. Environ. 1997, 31, 1185-1193. (30) WHO. Health aspects of air pollution with particulate matter, ozone and nitrogen dioxide; World Health Organization: 2003.




Received for review March 8, 2006. Revised manuscript received August 18, 2006. Accepted September 11, 2006. ES060543U