Tobacco Documents Online

Pergamon INTERLABORATORY COMPARISON OF PM10 AND BLACK SMOKE MEASUREMENTS IN THE PEACE STUDY GERARD HOEK,* HANS WELINDER,~" EVA VASKOVI,~: GIGLIOLA CIACCHINI,§ NICOS MANALIS,¶ ODDVAR ROYSET,[[ AUVO REPONEN,** JOSEF CYRYSt~ and BERT BRUNEKREEF* * Department of Environmental S~iences, Environmental and Occupational Health Unit, Wageningen Agricultural University, P.O. Box 238, 6700 AE Wageningen, The Netherlands; ~" Department of Occupa- tional and Environmental M~:licine, University Hospital, Lurid, Sweden; ~: B Johan National Institute of Public H~lth, Budapest, Hungary; § Multizonal Prevention Service, Pisa, Italy; ¶ Department of Environ- ment, Ministry of Environment, Physical Planning and Public Works, Athens, Gre~m; [{ Norwegian Institut* for Air Research, Kjeller, Norway;, ** National Public Health Institute., Kuopio, Finland; and "f~'GSF Institute of Epid*miology, Neuherberg, Germany (First received 18 July 1996 and in final form 26 March 1997. Published August 1997) Abstract--This paper presents the results of a comparison of ambient particle measurements conducted by 14 European laboratorie~ in the framework of a multi-center epidemiologic study of Pollution Effects on Asthmatic Childr~m in Europe (PEACE). Measurement of the concentration of particles less than 10/zm (PM10) and black smoke in ambient air was an important component of the PEACE study. The Harvard impactor was used by most laboratories for PMI0 sampling, but Hivol samplers,/~ gauge and dichotomous samplers were also used. A field study comparing these sample.rs with the Harvard impactor was conducted at five study sites. Systematic differences between PM10 mSasured with local samplers and a Harvard Impactor were between -- 13 and + 18%. A high correlation between P M 10 from the local sampler and the Harvard Impactor was found for four of the five centers. A field comparison of two identical black smoke samplers using either a membrane filter or the standard Whatman I paper filter showed an almost threefold difference in BS concentration. This illustrates the importance of using the standard filter. A similar comparison tmtw~n two different filter papers show~xi only small differences in BS concentrations. Considerable differences were found in an interlaboratory comparison b~tween two centers and the other enters in reflectance measurements of exposed filter~ This difference could largely tm attributed to the use of different blanks in setting the refl~tometer to 100 before performing the measurements. Systematic differences in weighing Teflon test filters were pre~nt hetweeu the laboratori~. Since these differences were similar for exposed and blank filters for 12 of the 13 centers, only small differences in PM10 concentration ( <5/~gm-3) were expected due to differences in weighing. ~ 1997 Elsevier Science Ltd. Key word index: Particles, PMI0, black smoke, interlaboratory, weighing, refl~ctunce. INTRODUCTION R~c~nt studies conducted in the U.S.A. and Eurotm have indicated that at low levels of outdoor air pollu- tion respiratory health effects still occur (Pop* et al., 199:5). These observations formed the background of the start of the PEACE study, a multi-center study of acute respiratory health effects of outdoor air pollution (Roemer et al., 1995; Brunekre~fet aL, 1995). PEACE is the acronym of Pollution Effects on Asthmatic Children in Eurol~. The PEACE study involved 14 ~nters in I0 European countries. Sin~ a major interest of the PEACE study was to evaluate health effects of ambient particulate matter, the con- c, ntration of particles with a 50% cutoff diameter of 10 #m (PM10) and black smoke (BS) were measured on a dally basis, at both an urban and a non-urban site. Most centers used the Harvard impactor (Marple et al., 1987) for PMI0 sampling. Some centers used other PM 10 samplers. Substantial differences in PMI0 concentrations measured with samplers of dif- ferent deaign have been documented in field studies (Chow, 1995). The use of samplers of different design could thus limit the comparability of concen- trations between different countries or between the urban and non-urban area within one country. Differ- ences in measur~:l concentrations using the same sam- pler may also be introduced by differences in sampler ol~ration and in analysis of filters (Chow, 1995). All centers weighed their own filters and most centers measured the reflectance of their own black smoke filters. Substantial differences in reflectance were found when the same test filters were analyzed by laboratories using different reflcctometers (Paydssat et aI., 1990a). 3341

3342 G. HOEK et al. Thus, a comparison program was designed to eva- luate differences between measurements conducted by the participating centers• The program involved a field comparison of different PM10 sarnplers. A round robin test of the analysis of filters (weighing, reflectance measurement) was conducted. This paper reports on the results of this comparison program. METHODS Most centers measured ambient PMIO concentrations using filter based methods. Air is sampled for generally 24 h through a size selective inlet, that excludes large particles from the air stream. The remaining particles are next col- lected on a preweighed filter. The exposed filter is then weighed again. Before weighing, the filters were equilibrated at about 20°C and about 44% relative humidity for 24 h. After collection from the field, filters were stored .in the refrigerator at 4°C before weighing. Most laboratories used analytical balances with a reading precision of 10#g, without temperature and relative humidity control of the weighing room. Desiccators were used to ensure a constant relative humidity (between 35 and 55% for the different laboratories). The laboratories in Berlin, Kuopio, Malta6 and Oslo had climate controlled weighing rooms. Micro- balances with a reading' precision of I pg were used in Hettstedt, Kuopio and Oslo. BS measurements were conducted using the method of the Organization for Economic Cooperation and Development (OECD, 1964). The method involves collection of particles on a Whatman I paper filter using a low volume sampler. The reflectance of a filter is measured using a reflectometer. First, the reflectometer reading is set to 100 using a blank Whatman 1 filter. Next, the reflectance of exposed filters~ measured. Higher exposed filters have a lower reflectance. The measured reflectance was next transformed into gg sootcm-2 exposed filter area using the formula describing the Standard Smoke curve (Christolis et aL 1992). PMIO sampler comparison The PEACE study protocol required that all centers that conducted PM10 measurements with samplers other than the Harvard impactor conducted a field comparison of their sampler and a Harvard impactor (Marple et al., 1987). The impactor was operated at 10 ¢ min- 1 and manufactured by Air Diagnostics and Engineering Inc., Naples, Maine, U.S.A. The minimum period of collocation was specified as 10 sampling days. The Harvard impactor including Teflon filters and other consumabIes was supplied by the coordi- nating center, Wageningen Agricultural University (WAU). The operation of both the local sampler and the Harvard impactor and the weighing of the filters was conducted by the center itself. The comparisons were conducted at sites that were not directly influenced by motorized traffic or other sources of air pollution. Comparisons were made in Amsterdam, Teplice, Pisa, Oslo and Budapest (Table 1). A description of the local samplers can be found in Table 2. The mean absolute and percentage difference between two collocated samplers was calculated to evaluate systematic differences. Two types of scatterplots were made. First, PM10 of the Harvard impactor was plotted against PM10 of the local sampler to evaluate correlation in time. Next, the PM10 concentration of the local sampler as a percentage of PMI0 of the Harvard impactor was plotted against the PM10 of the Harvard impactor. The latter plot was used to evaluate whether the difference between the two methods depended on the level of PM10. Next, a linear regression model was specified with PM10 by the Harvard impactor as the dependent variable and PM 10 by the local sampler as the

PM10 and black smoke measurements 3343 independent variable. The squared correlation (R2) was evaluated because the main interest of the PEACE study was to link variation in time of air pollution with variation in time.of health effects. Thus, two samplers have to pick up the same temporal variation in PM10 concentration. Intercept and slope of the regression model were evaluated to detect systematic differences between the samplers. In method-" comparison studies it is not always obvious which variable should be selected as the dependent variable and which as the independent variable. In addition bias of the "true~ regression slope to the null occurs in ordinary least-squares regression when a considerable amount of measurement error is present (Cornbleet and Gochman, 1979). Corubleet and Gochman suggest the use of Deeming regression in this situation. This method simultaneously minimizes the squared distances from the observed data points to the regression line in the horizontal and vertical direction. One regression line is obtained regardless of which of the two methods is considered as the independent variable. The obtained regression slope was shown to be much less biased than the least squares regression slope in case of measurement error (Cornblcet and Gochman, 1979). Slope and intercept were calculated using the formulas given by Corublcet and Gochman (1979). We assumed that the absolute error of the two compared measurement methods was the same. Black smoke sampler comparison Since all centers used samplers that complied with the OECD protocol, no comparison was made for all different samplers. In Athens a BS sampler comparison was made to comply with the requirement of the PEACE study to collect duplicate BS samples for ten days. The Athens center had only one "RIVM" sampler available and thus collocated the RIVM sampler with the sampler that was used in their air quality network (Filtromat, Environnement SA sampler). The RIVM sampler is a sampler used by the National Institute of Public Health and the Environment (RIVM) in the National Air Quality Network in the Netherlands. This sampler was used by most centers in the PEACE study. The Pisa center used a 0.8/~m pore size Sartorius cellulose nitrate membrane filter instead of the commonly used Whatman ! paper filter for which the formula transforming reflectances into #gsootcm-2 exposed filter area was derived. A field comparison using the same samplers for 10 days in.the summer of 1994 comparing the two filter types was conducted to evaluate potential differences. In Oslo a Whatman 40 paper filter was used. In March-April 1996 a field study was conducted for 28 days using identical samplers to compare BS concentrations using Whatman 1 and Whatman 40 filters. Reflectance measurement comparison To compare the reflectance measurement of filters between centers, three sets of 10 Whatman 1 paper filters were prepared by the coordinating center. Each set consisted of six exposed and four blank filters. The exposed filters were 24 h samples in the urban area of Rotterdam, the Netherlands in the winter of 1992/1993. All sets were first measured by the laboratory of WAU. Next, test series one was sent to the first lab (Oslo) for measurement. After reflectance measurement the series was sent back to WAU laboratory where the reflectance was measured again on two different days to document potential transport effects. Then the test series was sent to the second lab (Athens). This procedure was repeated until four labs had measured the test series. To limit the total time period involved in conducting the study, three other labs measured another test series. WAU laboratory was included in both test series. A third set of 10 filters was kept in WAU laboratory and measured when a transported set was measured, to document potential variability in reflectance measurement of WAU laboratory. The centers received specific instructions to treat the samples carefully, to analyze the samples as soon as possible and to send back the filters in the same way as they had received the filters. All transport was conducted by express mail in the filter cassettes (preventing contact of the filter surface with the petri dish) used for field sampling, in para- film sealed petfi dishes. Results were reported to one of the authors (GH) who was unaware of the analysis results in WAU laboratory. The WAU laboratory technieiaus were unaware of the results reported by the other centers. The mean refiectance of a specific filter measured by different centers was considered as the "true" value. For each center, deviations from that value for a specifio filter were calculated and averaged for exposed and blank filters separately. The first measurement made by WAU was used as the reflectances for WAU laboratory in the comparison. Filter weighing comparison We have chosen to use blank and exposed filters for the comparison study instead of certified, metal mass pieces because we felt that this would result in a more realistic assessment of weighing differences between laboratories. Temperature and particularly humidity effects on weights are most likely different for (exposed) filters and metal mass pieces. Procedures for comparing filter weighing were similar to filter reflectance measurements, only the differences are dis- cussed here. Three sets of 10 Andersen 37 mm diameter 2 pore size Teflon filters were prepared by the coordinating center, since this was the filter used by most centers. Each set consisted of five exposed and five blank filters. The exposed filters were collected in the rural area of Wageningen in the winter of 1993/1994. After collection the exposed filters were heatefl overnight to evaporate volatile components of par- ticles. Weighing of the complete third set (that remained in WAU laboratory) was started only after filters returned from the second center they were sent to. Weighing of exposed filters from this set was started simultaneously with the first two test series. A second shortcoming was that no post weighing of the filters sent to the last two centers was conducted, thus leaving uncertainty about transport effects in these centers. All 13 centers that conducted low volume PMI0 measurements received one of the two prepared sets. The mean weight of a specific filter measured by different centers was considered as the "true" value. In both sets one center reported clearly different weights than the other centers. The values from these centers were not used for the calculation of the "true" weight. Deviations from the "true" value for a specific filter were calculated for each center and averaged for exposed and blank filters separately. Filter weights determined on the first measurement day by WAU were used as the weights for WAU laboratory. RESULTS PMIO sampler comparison Figure la-e shows the comparison between the Harvard impactor and other samplers used in the PEACE study. Table 2 presents the systematic difference and the correlation between the PM10 concentration measured with these samplers and the Harvard impactor at the different sites. Except for Teplice, PM 10 concentrations measured with all sam- plots differed significantly from the Harvard impactor. However, the mean percentage differences were mode- rate, ranging from - 12.5 to + 17.7%. A high correla- tion is present for all sites---except Pisa--with little indication of nonlinearity. The comparison period in

~0- 10- 0 0 (c) ~. HOEK ~t aL PM~O local ~ampler (ug/m3) 10¸ (b) o 10 20 ~0 40 50 60 PMIO Io~al =ampler (ug/m3) 20 ~0 40 50 80 70 PMIO Iomd mzmp~er (ug/m3) 0 o lO 20 3o 4o (d) PMIO local ~ampl~ (ug/m3) o 10 2O 30 4O 5o 60 70 8o 9O (e) PMIO local lam~e~ (ug/m3) Fig. 1. Field comparison of local sampler with Harvard impactor. (a) Amsterdam; (b) Teplice; (c) Pisa; (d) Oslo; (e) Budapest~

PM I0 and black smoke measurements 3345 Table 2. FieldcomparisonofPMl0samplersin the PEACE study. Resuits ofcollocated sampling in the winter of 1993/1994 Absolute Percentage difference difference Center Local sampler PMI0aA =a +bPM10f~ SDb R2 (~gm-~)~ (%)~ na Amsterdam Dichot inlet* - 1.4 + 0.93 5.2 0.93 - 4.6 (I.1) -- 12.5 (3.2) 45 Teplice Hivol - 1.3 + 1.19 3.3 0.87 + 2.2 (1.3) + 4.7 (3.5) 15 Pisa Hivol - 19.3 + 1.30 3.6 0.65 - 3.8 (1.7) - 8.2 (3.4) 10 Oslo Dichot - 0.9 + 1.23 0.7 0.98 + 3.4 (0.5) + 16.4 (1.6) 15 Oslo~ Dichot - 2.0 + 1.30 0.7 0.93 + 3.1 (0.4) + 16.3 (1.7) 14 Budapest )8 gauge 6.0 + 1.06 1.4 0.99 + 7.6 (0.7) + 17.7 (6.6) 9 Budapestt fl gauge 2.4 + 1.23 1.3 0.91 ~ 7.4 (0.8) + 19.6 (7.2) 8 • PMI0~.~ ~ PM10 ~gm-3) measured with the local sampler. PMI0aA ,- PMI0 (~gm-3) measured with the Harvard impactor. a m intercept, b ~, slope of the linear regression model. b Standard error of y estimate. • Mean with standard error of mean in parentheses (PMI0 by Harvard impactor minus PMI0 by local sampler). d Number of measurement days. • Amsterdam: inlet as Sierra Andersen 241 dichotomous sampler (Hock and Brunekreef, 1993). Teplice: Strohlein Andersen PMI0 High Volume sampler. Pisa: Andersen PM10 high volume sampler. Oslo: Sierra Andersen 245 dichotomous sampler. Budapest: MPSI 100 Environnemeat // gauge monitor with PM10 inlet. ~ Excluding day with highest concentration. Budapest and Oslo involved one day with consider- ably higher concentrations. Exclusion of these observations from the regression analysis did not sub- stantially affect the regression coefficients (Table 2). The plots of PM10 of the local sampler as a percent- age of PMI0 of the Harvard impactor vs PM10 of the Harvard impactor confirmed the results (specifically intercepts) of the regression analysis. Percentage difference b~twven samplers was constant for Amster- dam and Oslo and decreasing with increasing PM10 concentration for Teplice and especially Pisa. However, due to the small range in PM10 concentra- tions during the comparison study period in Pisa, it was more difficult to evaluate the agreement at this site. Estimates of intercepts and slopes using Deeming regression were almost identical to estimates obtained with the least squares method for the Oslo, Budapest and Amsterdam. For the Teplice and particularly Pisa data, substantially different regression lines were obtained. The small variability of the PM10 concen- tration p~obably explains why especially in Pisa this phenomenon occurred. It is consistent with the obser- vation by Cornbleet and Gochman (1979) that the bias in least-squares estimates depends on the ratio of the absolute error in the independent variable and the standard deviation of the independent variable. B$ sampler comparison In Athens the sampler used in the network (Filtromat sampler) was collocated with the "RIVM" sampler. BS concentrations measured with the RIVM sampler during the 10 comparison days averaged 96/zgm-3, with a range from 61 to 150/zgm-s. BS concentrations measured with the two samplers were highly correlated (R2 =0.99). The mean difference (RIVM minus Filtromat) was 3.4/~gm-3 (standard error of the mean difference 1.6 #gin-a). There was thus a small significant difference of about 3% between the concentrations measured with the two typeg of samplers. Collocation for ten days of the same type of sampler with different filters in Pisa documented large differences in BS concentrations. The mean BS con~ntration using a Whatman 1 paper filter was 18/~gm-3 (range 10-24 ggm-S). Mean and range for BS measurements with the cellulose nitrate membrane filter were 49 pgm-3 (range 19-78/zg m-S). The linear regression model that best fitted these data was: BSwhatman ~5.88 (1.56) 4"0.245 (0.030)BSm,m~ .... (~ =0.90) BSwhatman = BS measured with a Whatman 1 paper filter (gg m- 3) BS~mb~, = BS measured with a cellulose nitrate membrane filter ~g m-s) All BS concentration data conducted by the Pisa center were recalculated using this empirical formula. This introduces some additional uncertainty in these measurements, e.g. for concentration data out of the range of the collocation study. The same experiment conducted for 28 days in Oslo showed only minor differences in BS concentrations measured with two different paper filters. Mean and range of BS concentration with Whatman 1 and Whatman 40 filters were respectively 12.5 (range 4-32) and 12.1 (3-35)/~gm-~. The linear regression model that best fitted these data was BS(Whatman 1)= 1.5 + 1.07BS(Whatman 40), with an R~ of 0.95. 0 O~ "~0 0

3346 G. HOEK et al. Reflectance measurement comparison Repeated measurements of the two transported test series of filters by WAU documented that the trans- port procedures had been sufficient to avoid changes of reflectance of the filters. The mean difference in reflectance of the measurement before any transport and the measurement when the test series had been received from the last center was 0.3 and 0.1 reflec- tance units for series 1 and 2, respectively. This implies a negligible difference, since the OECD specifies that integer numbers are sufficient to measure reflectances. The stability of the reflectance of the test filters is consistent with observations in the 1st European Quality Assurance Program. In this program stability of the reflectance of BS filters was observed for as much as 12 months (Payrissat et al., 1990a). Its is also consistent with the thermal stability of elemental carbon, the main component in determining darkness of a filter (Chow, 1995). Table 3 shows the results of the comparison of reflectance measurements. The range of reflectances was 80-100, typical for current ambient air samples in the Netherlands. Minimal differences for the four blank filters were found. WAU and Athens measured lower reflectances for exposed filters (higher soot con- centrations) than the other six centers. Differences between the other centers were minimal. WAU and Athens were the only centers that had followed the recommendation of the study manual to measure the reflectance of a filter on a stack of five blank Whatman 1 filters, after setting the reflectance to 100 on the same stack of five blank filters. The rationale for this procedure was that this approaches the measurement of absolute reflectances. We observed that if the reflectometer had been set to I00 using one blank Whatman 1 paper filter, an additional blank filter on top of that filter resulted in a reflectance reading above 100. After a stack of five filters, the reflectance increase with an additional filter was less than one unit. Only after a stack of twenty filters, no increase with an additional filter could be observed anymore. In this case the reflectance measured is "truly" 100. The other centers had used the OECD method in which the reflectance of a filter is measured directly on top of a white tile, after setting the reflec- tometer to 100 with one blank Whatman 1 filter. The relation between the two measurement procedures can be derived using assumptions of linearity of the reflectance scale, same difference of"true" reflectances between exposed and blank filters for the two procedures and experimental determination of the difference in reflectance of one blank filter and a stack of five blank filters after setting the reflectometer to I00 using one procedure. A more detailed derivation can be found in Roemer et al. (1995). The derived formula to transform reflectances measured on top of five blank filters (Rs) to reflectances measured according to the OECD protocol (RoEc~) was RoEc~ = (Rs +11)/1.11. This formula takes into account that the results of the two methods converge at reflectances close to 100. Because the reflectometer is set to 100 using either one or a set of five blank filters, (near) blank filters give (near) identical results. The theoretical equation was checked by WAU laboratory in a set of 49 samples with a large range of BS concentrations, collected during the main PEACE study phase in Budapest. The resulting (empirical) regression model was Ro~c~ ==(R5 + 10)/1.09, close to the theoretical formula. The theoretical formula was used to transform reflectances measured by Wageningen and Athens. After transformation, the mean difference (standard error'of mean in paren- theses) between WAU/Athens and the other centers was --2.2 (0.9) for the exposed filters of control series 1 and -2.2 (0.4) for control series 2. The original differences were -3.9 (1.2) and -3.8 (0.6) reflectance units. In addition to a significant reduction of the mean difference, the'difference also did not depend on the magnitude of the reflectance anymore. The effect of the remaining systematic difference of 2 reflectance Table 3. Comparison of reflectance measurement (R in %) of 10 Whatman 1 paper test filters of laboratories conducting reflectance measurements in the PEACE study Mean deviation of R Mean deviation of R Center Reflectomer type Time series blank filters (n =4)* exposed filters (n =6)* O~ Amsterdamb EEI.A3 1 + 0.4 (0.05) - 1.8 (0`3) Oslo EEL43 1 -0.4 (0`10) + 1.2 (0.3) Athens EEIA3 1 - 0.0 (0.03) - 2.9 (0.3) Pisa EEL43 1 + 0`1 (0.02) + 2.0 (0.4) Katowice RM2* 1 -0.0 (0.03) + 1.4 (0.6) Amsterdam EEL43 2 + 0.3 (0.05) - 2.8 (0.2) Berlin EEL43 2 - 0`1 (0.02) + 0.9 (0.2) Cracow ARM1/ARM2c 2 -0.I (0.02) + 1.3 (0.2) Sweden't EEL43 2 -- 0`1 (0.02) + 0.6 (0.1) • Deviation from "true" value per filter (Methods). Standard error of the mean in parentheses.' bCenters are shown in order of receiving the test filters. c Polish reflectometers. d Measurements for the two Swedish centers (Umea and Malm/~) made by the same, external Swedish laboratory.

PM I0 and black smoke measurements 3347 units on the soot concentration depends on the abso- lute value of the reflectance since the transformation from reflectance to ggm-3 is nonlinear. At reflec- tances of 90, 80, 70 and 60, BS concentrations that are calculated for a nominal sample volume of 2.05 ms and an exposed filter arcs of 5.067 cm2 are 14 vs 18, 36 vs 41, 66 vs 74 and !11 vs 123 itgm-3 for the other centers vs corrected WAU/Athens, respectively. Filter weight comparisons Repeated weighing of the three test series by WAU laboratory showed a small upward trend of the weight of the test filters (Fig. 2). In Fig. 2 the check measure- ments made on two consecutive days by WAU have been averaged. Measurement day I refers to the measurements made before filters were transported. Measurement day 2 refers to the measurements when the filters were reweighed at WAU after returning from the first center. During two of the measurement days at WAU the mean weight of a test series was more than 100 #g different from the consecutive day. During the first of these days, quality control filters routinely weighed before samples are weighed, had too low weights as well. Thus, the results of this day were not used. For the second day no explanation could be found and thus the results have been retained. Adjustment for the upward trends has been con- ducted using a linear regression model, with filter weight as the dependent variable and number of times the filter had been weighed already (at WAU and other centers) as the independent variable. This model is consistent with filter handling adding a small amount of weight to the filter. In all of the three series 123,50 123,45 IJ~3,dO 123,3S 123,30 123,25 • • • • • 123,20 - ~r~---'~---'-~t ~ r 0 2 4 6 8 10 12 Measurement day • Test sedes 1 • Testsed~2(+0.1 mgforsesling) • Check slides 3 (not transported) Fig. 2. Trend in weight of test filters of weighing comparison • study. there was a tendency towards a stronger upward trend for blank compared to exposed filters. The mean regression slope (standard error in parentheses) describing the increase of filter weight with test day in filter series one was 8.4 (1.4) for blank filters vs 6.5 (0.9) Itgweighingday. For series two the increases for blank and exposed were 1.21 (0.99) vs 2.17 (2.92); for series three 5.1 (0.8) vs 2.8 (1.1). An explanation for the difference in trend may be that the exposed filters lose some (coarse) particles due to handling of the Teflon filter (Chow, 1995). Transport of the filters by mail did not seem to affect the filter weights, since the trends of series three, which remained in WAU laboratory, was not different from those of the two transported series. The calculated regression slopes were next used to adjust the deviation from a center from the "true" filter weight. Specifically, the deviation of the reported weights by a center from the "true" weight was adjusted by subtracting the product of the calculated regression slopes and the number of times the filter was weighed already minus the average number of times a filter was weighed. Significant systematic differences in absolute filter weights between the laboratories are apparent from Table 4 (third column). In the calculations of the mean deviations, one exposed filter from Teplice and one blank filter from Cracow were left out. The weights of these filters differed respectively + 1164 and" + 9864 ltg from the "true" weights of these filters. The latter may be due to a recording error, however when the center was asked to check whether the data reported to us were correct (without specifying for what filter a problem existed), no problem was reported. If'the systematic difference is the same for exposed and blank filters, no bias in the measured PMI0 concentration occurs. Thus, column four of Table 4 shows the difference of the mean deviations between exposed and blank filters. This difference is generally smaller than the deviation of absolute filter weights (column three). For example, the large systematic difference for Cracow almost disappears, indicating that no bias in PM10 concentrations is expected due to weighing. The adjustment for weight changes of the test filters did not substantially affect the interpretation of the differences between laboratories. The unadjusted differen.ce between exposed and blank filters was at most--for the first and last centers in each series-- 25/zg different from the results reported in Table 4. The unadjusted deviation of the 10 test filters was at most 70 and 5 #g different from the results reported in Table 4 for series 1 and 2, respectively. The adjust- ment was an extrapolation for the centers Prague and Katowice (series 1) and Budapest (series 2), since post weighing of the test series was not conducted. The implications of differences in analysis between centers are illustrated in Fig. 3. For most centers the magnitude of the differences is small, considering that a typical sample volume for the Harvard impactor for

3348 (3. HOEK et aL Table 4. Comparison of weighing 10 Teflon PMI0 test filters by laboratories participating in the PEACE study Differenc¢ in deviation Test Mean deviation of between exposed and Center series" 10 test filters (pg)b blank filters ~g)~ Amsterdam" 1 + 10 (7) - 16 Oslo 1 + 38 (6) + 10 Athens 1 + 46 (10) + 42 Umea 1 - 12 (13) - 64 Kuopio 1 - 32 (4) 0 Teplice 1 + 275 (52) + 237 Prague 1 + 30 (6) + 27 Katowice 1 - 47 (7) -- 9 Amsterdam 2 - 40 (9) - 28 Berlin 2 -- 48 (6) -- 16 Cracow 2 - 154 (10) - 29 Hettstedt 2 + I (2) + 10 Maim~ 2 + 30 (6) + 21 Budapest 2 + 51 (7) + 19 "Centers arc shown in order of receiving the set of filters. b Deviation from "true" value ger filter (Methods). Standard error of mean in parentheses. Adjusted for trend observed in test filter weight. 120 ~70 - ~a~-ted mean PM~O Fig. 3. Mean PM10 concentration corrected for differences in sampler and weighing vs uncorrected mean. a~4 h sample is 14.4 m3. For Teplice a small differ- ence is found because a high volume sampler was and thus the large difference in weighing does not affect the concentration much. This assumes that the absolute weighing differences for this center would have been the same if larger filters had been compared. DISCUSSION Significant differences of the PMI0 concentra- tion ranging from -13 to +18% were found be- tween four different PM10 samplers and the Harvard impactor in a field comparison study. Differences be- tween PMI0 samplers of different design up to about 50% have been found in early comparison studies conducted in the U.S.A. (Chow, 1995). These differ- ences were partly explained by differences in sampler operation procedures--such as greasing of the in- letmand differences in inlet design. A later study showed mean differences between 4.5 and 14% between cyclone type and impactor type PM10 inlets (Chow, 1995). In a collocation study of the dichotomous sampler and the Harvard impactor for ten days, no systematic differences (PMI0-HI minus PM10-dichot on average -1%) and a high cor- relation (r = 0.99) of P M 10 concentrations were found (Lioy et al., 1988). Wedding and Weigand (1993) found no systematic difference between a B gauge sampler and a Wedding high-volume sampler. The small systematic difference in BS concentra- tion measured with two types of BS samplers in Athens is in agreement with the second European Quality Assurance Programme for sulphur dioxide and suspended particulates measurements (Payrissat et al., 1990b). In that study a field comparison was made between a "reference sampler" operated by a central laboratory and local samplers operated by the local laboratory in air quality networks for one day. The majority of the samplers agreed well with the reference sampler. However, twofold differences were observed for some samplers. These differences could be explained by errors in flow measure- ment, leaks in the sampler, use of different reflec- tometers, differences in setting the reflectometer to 100 and use of a different curve to convert reflectanees to/~gcm-2. BS concentrations measured with Whatman 1 paper filters were almost three times lower than measurements with a cellulose nitrate membrane

PMI0 and black smoke measurements 3349 filter. In contrast measurements made with two different paper filters showed only minor differences in BS concentrations. Explanations for the higher soot concentrations observed with the membrane filter probably include the following factors: less deep penetration of particles in the membrane filter and more efficient sampling of fine particles. Probably the surface structure of the two paper filters is very similar. It has been documented that the collection efficiency of Whatman 1 paper filters is considerably less than 100% for submicron particles (ISO, 1993). This is a significant problem since a large proportion of soot particles are in the submicron range (Horvath etal., 1988). The comparison of reflectance readings of BS filters showed differences related to the setting of the reflectometer to 100. These results stress that if the concentrations are to be compared with concentra- tions using the original OECD protocol, this protocol should be followed in detail. After adjustment, the differences between reflectance readings between centers was comparable to differences found between different European laboratories in the 1st European quality assurance program (Payrissat etal., 1990a). This study showed that the standard deviation of reflectance readings of test filters was about 0.7 units, independent of the darkness level. The range between the laboratories with the lowest and highest reading was about 2 reflectance units. All laboratories used the EEL 43 reflectometer. In our study different types of reflectometers were used. In a second study of the 1st European quality assurance program, large reflectance differences (up to 6 reflectance units) were found with certain models of (French) Photovolt reflectometers. The comparison of filter weighing showed statisti- cally significant but limited differences between 12 of the 13 participating laboratories. This was partly a result of similar systematic differences for exposed and blank Teflon filters. Differences between labora- tories in relative humidity (RH) used for conditioning filters, did not explain the differences in weighing. For example while Umea used the lowest RH (36%) and reported lower weights for exposed filters, Kuopio and Malmib applied the highest RH (55%) and repot-ted weighs near the mean value. The results for the deviating center do not necessarily mean that a substantial positive bias in PM10 concentration occurred, since this center used a Hivol sampler for PM10 sampling in the main study phase. In conclusion, statistically significant but small differences of PM10 concentrations were found between different samplers. Large differences were found in BS concentration between samplers using a membrane filter instead of the standard Whatman 1 paper filter. Differences in the procedure of setting reflectometers to 100%, resulted in significantly differ- ent BS concentrations. Significant differences in filter weighing were found, that would result in modest differences in PM10 concentration. AcknowledgementsqThe PEACE study was funded in the framework of the Commission of the European Communi- ties Environment Programme, contract~ EV5V-Cr92-0220, CIPD-CT-92-5052 and ERBCIPDCT930046. Preparation of the manuscript and data analysis have been performed during a Research Fellowship of the Netherlands Organiza- tion for Scientific Research for G Hock at the Harvard School of Public Health. The authors thank Kees Meliefste and Marieke Oldenwening (Department of Epidemiology and Public Health, WAU) for their contribution to the measurements. Brunekreef, B., Hock, G., Reemer, W., Kalandidi, A., Baldini, G., Vondra, V. and Kotesove~ F. (1995) PEACE study. background and design. Epidemiology 6, $65. Chow, J. (1995) Measurement method~ to determine compli- ance with ambient air quality standards for suspended particles. Journal of Air Waste Management Association 45, 320-382. Christolis, M, Clayton, P., Hecq, P., Payrissat, M. and Petit-Covianx, F. (1992) Instruction manual for air pollution monitoring. Volume II: Black Smoke monitor- ing. Report EUR 14550/11 EN, Joint Research Centre, Commission for the European Communities. Cornbleet, P. J. and Gochman, N. (1979) Incorrect least- squares regression coefficients in method-comparison analysis. Clinical Chemistry 25, 432-438. Hook, G. and Brunekreef, B. (1993) Aente effects of a winter air pollution episode on pulmonary function and respirat- ory symptoms of children. Arch Environment Health 48, 328--335. Horvath, H., Kreiner, I., Norek, C. and Preining, O. (1988) Diesel emissions in Vienna~ Atmospheric Environment 22, 1255-1269. ISO (1993) Ambiem air-determination of a black smoke index. ISO 9835. Lioy, P. J., Wainman, T., Turner, W. and Marple, V. A. (1988) An intercomparison of the indoor air sampling impactor and the dichotomous sampler for a 10-/tin cut si~e. Journal of Air Pollution Control A~sociation 38, 668-670. Marple, V. A., Rubow, K. L., Turner, W. and Spengler, J. D. (1987) Low flow rate sharp cut impactors for indoor air sampling: design and calibration. Journal of Air Pollution Control Association 37, 1303-1307. OECD (1964) Methods of measuring air pollution. Report of the working group on methods of measuring air pollution and survey techniques, Paris. Payrissat, M., Biising. J, Ran, H, Pfeiffer, W. and Zierock, K. H. (1990a) 1st European quality assurance programme for sulphur dioxide and suspended particulates measure- ments. Report EUR 12333 EN. Luxembourg, Payrissat, M., Biising. J, Ran, H, Pfeiffer, W. and Zierock, K. H. (1990b) 2rid European quality assurance programme for sulphur dioxide and suspended particulates measurements. Report EUR 12334 EN, Luxembourg. Pope, C. A. III, Dockery, D. W. and Spongier, J. D. (1995) Re, dew of epidemiologic evidence of health effects of particulate air pollution. Inhalation Toxicology 7, 1-18. Roomer, W, Hock, G. and Brunekreef, B. (1995) Study. manual PEACE project. A study on effects of short-term changes in urban air pollution on the respiratory health of children with chronic respiratory symptoms. University of Wageningen. Wedding, J. B. and Weigand, M. A. (1993) An automatic particle sampler with Beta gauging. Journal of Air Waste Management Association 43, 475-479.