BTX measurements with diffusive samplers in the vicinity of a cokery:

Comparison between ORSA-type samplers and pumped sampling

Hans-Ulrich Pfeffer* and Ludger Breuer

Landesumweltamt Nordrhein-Westfalen, Wallneyer Str. 6, D-45133 Essen, Germany.E-mail: ulrich.pfeffer@lua.nrw.de; Fax: z49 201 7995 1575; Tel: z49 201 7995 1264

Received 29th March 2000, Accepted 17th July 2000First published as an Advance Article on the web 17th August 2000

In the framework of new European directives in the ®eld of ambient air quality assessment speci®c

requirements for measurement methods for benzene will be de®ned and have to be met in the future. In a

residential area in the vicinity of a cokery a comprehensive monitoring programme for BTX compounds

(benzene, toluene, xylenes) was carried out with diffusive samplers. Measurement results from 20 sites show a

distinct spatial distribution of benzene concentrations which is caused by the emissions of the plant.

Comparisons with two pumped measurement methods reveal excellent agreement in the case of benzene. An

assessment of measurement uncertainty for the diffusive method is presented in terms of standard deviations

from parallel measurements (precision) and by comparison with a reference method according to national and

international standards. Data quality objectives required by an upcoming European directive for benzene can

be met by diffusive sampling.

1. Introduction

On the basis of EC Directive 96/62/EC on ambient air qualityassessment and management,1 the so-called `EU-FrameworkDirective', the monitoring of air pollution in Europe has beenthoroughly reorganized. Speci®c requirements for the measure-ment of benzene (and carbon monoxide) will be laid down in adaughter directive.2 For the implementation of these directives,standardized and well documented measurement methods areneeded. These methods have to meet speci®c data qualityobjectives and should be as cost effective as possible.

The method of diffusive sampling plays an important role inthis ®eld, especially in the neighbourhood of emission sources,where a high variability of ambient concentrations in space andtime is to be expected and therefore several monitoring sitesmay be necessary.

ORSA-type diffusive samplers were used for BTX measure-ments for 2.5 years (October 1996 to April 1999) at 20monitoring sites in a residential area in the vicinity of acokery in the city of Duisburg. In addition, pumpedhydrocarbon measurements were carried out for most of thetime at a ®xed monitoring station. The site of this station wasidentical with one of the sites of the diffusive samplers.

The aims of this study were: (i) the investigation of the spatialdistribution of ambient benzene concentrations in the residen-tial area adjacent to the cokery, (ii) a comparison betweendiffusive and pumped methods for BTX measurements, and(iii) an assessment of the measurement uncertainty of thediffusive method.

These objectives are essential steps in the process ofimplementation of the upcoming European directive forbenzene.

2. Methods and materials

The ORSA samplers (DraÈger Company, LuÈbeck, Germany)were exposed over periods of approximately 15 days using asimple rain roof as a protective shelter. At all sites and for allmeasuring periods two samplers were used in parallel. Afterexposure, the 400 mg coconut shell carbon of the tubes wereextracted with 2 ml of blank tested carbon disul®de andanalyzed with an HP5890 gas chromatograph (Hewlett-

Packard Company, Palo Alto, USA) and a 60 m DB1 capillarycolumn (J&W Scienti®c Products GmbH, Cologne, Germany).The hydrocarbon concentrations were calculated using dose-dependent uptake rates as described earlier in detail.3,4

The pumped measurements were done with a commercialinstrument (HC 1010, Airmotec GmbH, Essen, Germany)based on thermodesorption and by a manual pumped samplingtechnique according to a draft German standard,5 performedon a continuous daily basis (24 h sampling every day). For eachsample, approximately 60 l of air were sucked through a tubecontaining 40 mg of activated carbon. The tubes were extractedwith 200 ml carbon disul®de and analyzed with the samemethod as described above for the diffusive samplers. Thismanual method was also used for the calibration of thecontinuous BTX-monitor and as a reference method for theestimation of measurement uncertainty of the ORSA sampler.More analytical and technical details relevant for this study aredescribed elsewhere.3,4

For the assessment of measurement uncertainty an addi-tional data set generated in a previous project3,4 was used. Thestatistical evaluation was performed using national andinternational standards (see Section 5).

3. Spatial distribution of benzene concentrationsnear the cokery

One of the aims of the study was the investigation of the spatialdistribution of the benzene concentrations in a residential areaadjacent to the cokery plant.

Fig. 1 shows a map of the average benzene distribution,obtained by measuring benzene concentrations at 20 sites nearthe cokery, using a regular grid of 200 m6200 m. The gradientof concentrations from the source to the urban background canclearly be recognized. The annual mean of 3.2 mg m23 at themeasuring point in the east (south of the motorway A42) can betaken as a typical background value in highly industrializedareas. This site may also be in¯uenced by road traf®c on themotorway.

In addition to the spatial structure in the residential area, thebenzene to toluene ratio also points to the cokery being thedominating benzene emission source. At typical urban back-

DOI: 10.1039/b002514n J. Environ. Monit., 2000, 2, 483±486 483

This journal is # The Royal Society of Chemistry 2000

Publ

ishe

d on

17

Aug

ust 2

000.

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity o

n 25

/10/

2014

08:

27:1

4.

View Article Online / Journal Homepage / Table of Contents for this issue

ground stations or traf®c-related sites in Germany, thebenzene : toluene ratio is about 0.3 to 0.4.6 At the site of the®xed station near the cokery this ratio is 1.2, averaged over thewhole period indicated above. The mean value for toluene(6.6 mg m23) is a typical concentration in urban backgroundareas, whereas the value for benzene (8.0 mg m23) is about 3±4times higher.6

Fig. 2 shows the dependence of the benzene and toluene valueson the wind direction. This evaluation is based on thecontinuously registered, half hourly concentration values of theautomated monitor. Wind data were taken from a nearbyrepresentative monitoring station in Duisburg±Walsum, about3 km north of Bruckhausen. The year 1997 serves as an example;data from the other time periods revealed very similar results.

High average benzene concentrations were only observed,when the wind came from the cokery (west to north). This isnot the case for toluene. Nearly all wind directions areconnected with toluene concentrations in the range from 5 to10 mg m23. The highest values were measured during episodeswith easterly wind directions which may be partly due to roadtraf®c on the motorway A42 (please note that the concentra-tion ranges in the ®gure are different for benzene (0±40 mg m23)and for toluene (0±15 mg m23)).

It should be noted that such a survey with simultaneous BTXmeasurements at 20 sites would be impracticable and veryexpensive with conventional methods and active sampling. Inthis way, it is a good example for a source related monitoringprogramme claimed by the European directives.1,2

4. Comparison of diffusive and pumped sampling

For the site with the continuous monitoring station the resultsof the different BTX measuring methods were compared bylinear regression analysis. The results are shown in Table 1.

The agreement of the respective methods for benzene isexcellent, but not as good for toluene and especially for mzp-xylene. However, slope and R2 for toluene are acceptable.Relatively high intercepts on the ordinates found for toluene andmzp-xylene are astonishing and cannot be explained easily,because no signi®cant blank values for any method wereobserved. The following factors may contribute to these effects.(i) The time coverage of the diffusive measurements is practically100% due to the parallel measurements with two samplers. Thedata series of the pumped methods show several gaps (missingvalues). In view of the often fast varying concentrations at themonitoring site this may lead to considerable differences in somecases. (ii) The observed concentration levels for toluene andmzp-xylene were relatively small compared with those used forthe calibration for these components (up to approximately75 mg m23 for toluene and 35 mg m23 for mzp-xylene, see ref. 3).

For benzene, there are no signi®cant intercepts on theordinate for the regression function. At least in this case, it isjusti®ed to force the regression lines through the origin of theco-ordinate system leading to the results shown in Table 2.

In both cases, the correlation of the respective methods isvery good with R2¢0.95. Also the slopes of the regression lines(0.96 and 0.98) are in excellent agreement, bearing in mind thatthe pumped methods are quite different (automated thermo-desorption monitor, adsorption on charcoal with solventextraction (CS2)). Deviations from the ideal function with aslope of 1.0 are not higher than 4%.

5. Assessment of measurement uncertainty

An important aspect of measurements according to theupcoming EU daughter directive2 for benzene (and carbonmonoxide) is the measurement uncertainty of the data. For thepurpose of checking compliance with the limit values of thisdirective measurement, data have to ful®l speci®c data qualityobjectives laid down in Annex VI (25% for ®xed measurements;30% for indicative measurements).

In this study, two methods were used for the estimation ofmeasurement uncertainties of the diffusive method, oneestimating the precision from parallel measurements, theother one assessing the uncertainty with regard to a referencemethod.

5.1. Standard deviations from parallel measurements (precision)

As mentioned above, all measurements in Duisburg±Bruck-hausen with the diffusive method were performed in parallelwith two samplers. Accordingly, standard deviations, sD, from

Fig. 1 Spatial distribution of mean benzene concentrations (mg m23) inthe vicinity of a cokery in Duisburg±Bruckhausen (30.10.1996 to14.04.1999, 59 measurement periods).

Fig. 2 Dependence of mean benzene and toluene concentrations(mg m23) in Duisburg±Bruckhausen on the wind direction (see text).

Table 1 Regression parameters for the function c(ORSA)~b6c(pumped sampling)za. Number of measurement periods: 55

CompoundConcentrationrange/mg m23 b a R2

Airmotec HC 1010ÐBenzene ¡26 0.93 0.29 0.97Toluene ¡16 0.93 1.00 0.85mzp-Xylene ¡8 0.78 1.78 0.82

Manual methodÐBenzene ¡26 0.97 0.05 0.95Toluene ¡16 0.88 0.91 0.87mzp-Xylene ¡8 0.80 1.50 0.86

Table 2 Regression parameters for the function c(ORSA)~b'6c(pumped sampling). Number of measurement periods: 55

Compound: benzene b' R2

Airmotec HC 1010 0.96 0.97Manual method 0.98 0.95

484 J. Environ. Monit., 2000, 2, 483±486

Publ

ishe

d on

17

Aug

ust 2

000.

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity o

n 25

/10/

2014

08:

27:1

4.

View Article Online

parallel measurements were calculated as described in theGerman standard VDI 2449 Part 17 as follows:

sD~

����������������������������Pni~1

(y1i{y2i)2

2n

vuuutn is the number of pairs of the measured values y1i and y2i.Taking these values leads to the results shown in Table 3.

The standard deviations determined are nearly independentof the respective concentrations. For this reason the data aregiven as absolute values in mg m23. The standard deviation forbenzene of 0.14 mg m23 corresponds to a relative value of 2.8%at the level of the foreseen EU limit value of 5 mg m23. It shouldbe born in mind that the standard deviations from parallelmeasurements contain only random contributions of thediffusive method to measurement uncertainty (precision).

5.2. Comparison with a reference method

The second approach for the estimation of measurementuncertainty was an evaluation according to the internationalstandard ISO 13752, which describes the assessment ofuncertainty of a measurement method under ®eld conditionsusing a second method as reference.8 In this standard, which isin line with the ISO Guide to the Expression of Uncertainty inMeasurement,9 the values of the reference method are assumedto be `true values'. All differences to the test method underevaluation are attributed to the test method. This may lead toan overestimation of the uncertainty of the test method.

For this evaluation of the uncertainty of the ORSA diffusivesampler an extensive data set generated in an earlier project3, 4

was used. The reasons for this decision are as follows. (i) Thedata set represents the calibration basis of the ORSA diffusivesampler and was used for the evaluation of the dose-dependentuptake rates. (ii) The data set was produced over a long period

at three sites with different air quality (two traf®c-related sites,one urban background site). (iii) The data set covers largeconcentration ranges, especially also for toluene and mzp-xylene (benzene: ¡25 mg m23; toluene: ¡75 mg m23; mzp-xylene: ¡35 mg m23). (iv) All diffusive measurements wereperformed with four parallel samplers. (v) The data set consistsof more than 540 data pairs for each compound (exposuretimes of 2 and 4 weeks), compared with only 55 data pairsgained in this study.

The active method taken as reference was the same as in thisproject.

ISO 13752 uses the following model equation:

y~b0zb1x

where xi are the values of the reference method, yà are thepredicted values of the model equation and b0 and b1 are theregression equations.

In addition, the following abbreviations are used: n, numberof measurements; y, values of the test method; Dy, systematicdeviation (bias) of the test method, Dy~b0z(b121)x; s2,variance as a function of x (variance function with coef®cientsa0, a1, a2; U, expanded uncertainty of the test method for a 95%con®dence level, describing a single measurement.

U~2

��������������������s2z(Dy)2

qThe results of the treatment are given in Table 4.In the case of benzene and toluene the coef®cients a1 of the

general variance function were found to be statistically notdifferent from zero. According to the recommendations of ISO13752 the simpler variance model was applied for thesecompounds.

For all compounds the slopes b1 and the intercepts b0 arestatistically not different from their ideal values (95% level ofcon®dence).

The assessed values of expanded uncertainty are dominatedby the contribution of the variance function, whereas thein¯uence of the bias can be more or less neglected. Theuncertainties found are much higher than the standarddeviations calculated from parallel measurements (seeTable 3). It is assumed that this is essentially due to theuncertainty of the calibration of the diffusive method asdescribed earlier in detail.3

For benzene and toluene the relative expanded uncertainties

Table 3 Standard deviations from parallel measurement with ORSAsamplers

Compound Concentration range/mg m23 sD/mg m23 n

Benzene ¡26 0.14 1324Toluene ¡16 0.28 1324mzp-Xylene ¡8 0.24 1324

Table 4 Estimation of measurement uncertainty of the ORSA diffusive sampler using the pumped charcoal method as reference (ISO 13752)

Regression parameter Value Standard deviation Concentration/mg m23 (Dy)2 s2 U (p~0.95)/mg m23 U (p~0.95) (%)

BenzeneÐb0 20.009 0.030 5 0.002 0.43 1.31 26.2b1 1.011 0.008 10 0.009 1.71 2.62 26.2n 544 Ð 20 0.041 6.85 5.25 26.3

Variance function: s2~a02za2

2x2

TolueneÐb0 20.063 0.083 5 0.000 0.52 1.44 28.7b1 1.015 0.009 10 0.007 2/06 2.88 28.8n 547 Ð 20 0.054 8.25 5.76 28.8

30 0.144 18.57 8.65 28.850 0.454 51.59 14.43 28.975 1.087 116.08 21.65 28.9

Variance function: s2~a02za2

2x2

mzp-XyleneÐb0 0.075 0.086 5 0.021 1.46 2.43 48.7b1 1.014 0.012 10 0.046 3.93 3.99 39.9n 545 Ð 20 0.126 11.88 6.93 34.6

30 0.244 23.85 9.82 32.7Variance function: s2~a0

2za12xza2

2x2

J. Environ. Monit., 2000, 2, 483±486 485

Publ

ishe

d on

17

Aug

ust 2

000.

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity o

n 25

/10/

2014

08:

27:1

4.

View Article Online

for single measurements on the 95% level of con®dence are 26%and 29%. These relative values do not depend on theconcentration level.

On the other hand, for mzp-xylene the relative expandeduncertainty increases at lower levels. At 10 mg m23 a value of40% was found.

The expanded uncertainty for benzene of 26% meets clearlythe requirement of Annex VI of the EU daughter directive forindicative measurements (30%) and exceeds only slightly thatfor ®xed measurements (25%).

At this point a weakness of the ISO 13752 procedure has tobe mentioned. This method does not allow the uncertainty ofthe reference method to be taken into account. As aconsequence, the uncertainty calculated for the test methodcan be regarded as an upper bound.

In reality, all reference standards and reference methodsreveal considerable uncertainties which should be included inthe statistical treatment as described by Beier and Kordecki10,11

for similar applications, e.g., calibration experiments.

6. Summary and conclusions

BTX hydrocarbon measurements were performed with ORSA-type diffusive samplers from October 1996 until April 1999 inthe vicinity of a cokery in the city of Duisburg.

The 2.5 year survey at 20 monitoring sites in a residentialarea adjacent to the cokery demonstrates a distinct spatialdistribution of average benzene concentrations. It can beshown by the benzene : toluene ratios as well as by plots of theconcentrations for different wind directions that the benzeneburden is mainly caused by the plant. The study can be taken asa good example of a monitoring campaign related to industrialsources as claimed by the upcoming EU daughter directive forbenzene.

The comparison of the results with those of pumped methods(automated thermodesorption monitor, manual method basedon adsorption on charcoal and solvent extraction) at one of thesites shows excellent agreement for benzene.

Measurement uncertainty of the ORSA diffusive samplerwas estimated in terms of standard deviations from more than1300 parallel measurements (precision) and according to ISO13752.

The precision of the ORSA diffusive sampler is below0.3 mg m23. Expanded relative uncertainties (95% level ofcon®dence) of 26% and 29% were found for benzene andtoluene, respectively. The level of uncertainty was constantover the range 5±20 mg m23 for benzene, and from 5±75 mg m23

for toluene. The values calculated for mzp-xylene, on theother hand, increase at low concentrations. For a concentra-tion of 10 mg m23 an expanded relative uncertainty of 40% wasfound. All these values describe single measurements.

The uncertainty of the ORSA diffusive sampler meets the

requirements of the EU benzene daughter directive, at least forindicative measurements. Probably, also the requirements formandatory measurements can be met. The statistical procedureof ISO 13752 used in this study does not take into account anyuncertainty of the reference method, which is not in line withreality and may lead to an overestimation of the uncertainty ofthe diffusive technique.

Internationally accepted statistical procedures which canclose this gap, are urgently needed.

The campaign has shown that diffusive samplers can be anadequate and cost effective tool for benzene measurements inthe framework of European directives, especially when a highspatial resolution is required.

Acknowledgements

The authors thank Dr Reinhold Beier, LandesumweltamtNRW, for fruitful discussions on the treatment of measurementuncertainty.

References

1 Council Directive 96/62/EC of 27 September 1996 on ambient airquality assessment and management, Off. J. Eur. Communities,No. L 296/55-63.

2 Common position adopted by the Council with the view to theadoption of a Directive of the European Parliament and of theCouncil relating to limit values for benzene and carbon monoxidein ambient air.

3 H.-U. Pfeffer, L. Breuer and K. Ellermann, Materialien No. 46,Landesumweltamt Nordrhein-Westfalen [North Rhine-Westpha-lia State Environment Agency], Essen, 1998, ISSN 0947-5206,92 pp.

4 H.-U. Pfeffer and L. Breuer, ESPR - Environ. Sci. Pollut. Res.,1998, 5(3), 157.

5 VDI Guideline 2100, Part 2 (Draft), VDI, DuÈsseldorf, November1999.

6 LuftqualitaÈt in Nordrhein-WestfalenÐLUQS-Jahresbericht 1998[Air Quality in North Rhine-WestphaliaÐLUQS Report on AirQuality 1998], Landesumweltamt Nordrhein-Westfalen, Essen,1999, ISSN 1438-8448, 84 pp.

7 VDI Guideline 2449, Part 1, VDI, DuÈsseldorf, February 1995.8 International Organisation for Standardization, Air qualityÐ

Assessment of uncertainty of a measurement method under ®eldconditions using a second method as reference, ISO, Geneva, 1998,ISO 13752.

9 International Organisation for Standardization, Guide to theExpression of Uncertainty in Measurement, ISO, Geneva, 1st edn.,1993, ISBN 92-67-10188-9.

10 R. Beier and R. Kordecki, in Neuere Entwicklungen bei derMessung und Beurteilung der LuftqualitaÈt [Recent Developments inMeasurement and Assessment of Air Quality], VDI, DuÈsseldorf,1999, Report 1443, pp. 171±181.

11 R. Beier and R. Kordecki, Gefahrstoffe - Reinhaltung der Luft,2000, 60, 101.

486 J. Environ. Monit., 2000, 2, 483±486

Publ

ishe

d on

17

Aug

ust 2

000.

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity o

n 25

/10/

2014

08:

27:1

4.

View Article Online