Facility fence-line monitoring using passive samplers.(Technical report)

Journal of the Air & Waste Management Association August 1, 2011 | Thoma, Eben D.; Miller, Michael C.; Chung, Kuenja C.; Parsons, Nicholas L.; Shine, Brenda C.

DOI:10.3155/1047-3289.61.8.834 ABSTRACT In 2009, the U.S. Environmental Protection Agency (EPA) executed a year-long field study at a refinery in Corpus Christi, TX, to evaluate the use of passive diffusive sampling technology for assessing time-averaged benzene concentrations at the facility fence line. The purpose of the study was to investigate the implementation viability and performance of this type of monitoring in a real-world setting as part of EPA’s fence-line measurement research program. The study utilized 14-day, time-integrated Carbopack X samplers deployed at 18 locations on the fence line and at two nearby air monitoring sites equipped with automated gas chro-matographs. The average fence-line benzene concentration during the study was 1075 parts per trillion by volume (pptv) with a standard deviation of 1935 pptv. For a 6-month period during which wind direction was uniform, the mean concentration value for a group of downwind sites exceeded the mean value of a similar upwind group by 1710 pptv. Mean value differences for these groups were not statistically significant for the remaining 6-month time period when wind directions were mixed. The passive sampling approach exhibited acceptable performance with a data completeness value of 97.1% {n = 579). Benzene concentration comparisons with automated gas chromatographs yielded an [r.sup.2] value of 0.86 and a slope of 0.90 (n = 50). A linear regression of duplicate pairs yielded an [r.sup.2] of 0.97, unity slope, and zero intercept (n = 56). In addition to descriptions of technique performance and general results, time-series analyses are described, providing insight into the utility of 2-week sampling for source apportionment under differing meteorological conditions. The limitations of the approach and recommendations for future measurement method development work are also discussed.

INTRODUCTION Development of cost-effective and robust methods for detecting fugitive emissions and monitoring air pollution concentration levels at industrial facility fence lines and remediation site boundaries can yield many benefits. Implementable fence-line and process monitoring systems can enhance protection of public health and worker safety, advance emission inventory knowledge, and realize cost savings by helping reduce product loss. A primary requirement for a fence-line monitoring system is that it provides adequate spatial coverage for determination of representative pollutant concentrations at the boundary of the facility or operation. In an ideal scenario, fence-line monitors would be placed so that any fugitive plume originating within the facility would have a high probability of intersecting one or more sensors, regardless of wind direction. Sufficient measurement coverage can be accomplished using a few open-path instruments (1-6) or through deployment of a larger number of point monitors. With either approach, applications that require high detection sensitivity, chemical speciation, and fast time response demand laboratory-class instrumentation, which comes with significant capital and operational cost. Currently, the expense of high-performance, near-real-time fence-line monitoring systems is likely perceived by industry to outweigh benefits. This is evidenced by the lack of significant voluntary adoption, causing potential benefits to go largely unrealized.

As part of the U.S. Environmental Protection Agency (EPA)’s fugitive emission research program, various cost-effective fence line and process monitoring approaches are under investigation with the aim to improve understanding and facilitate broader access to these technologies. Under the program, time-resolved and time-integrated measurement approaches are being explored. In long-term assessment or screening applications in which sensitivity and speciation are important but time response is not critical, deployment of time-integrated passive diffusive samplers (PSs) with subsequent laboratory analysis is a promising and cost-effective fence-line monitoring approach. This paper presents the results of a year-long field study using PSs to quantify fence-line benzene concentrations at a refinery in Corpus Christi, TX. The objectives of the study were to evaluate the implementation feasibility, cost, and performance of the PS fence-line monitoring approach and to assess the effectiveness of time-integrated sampling for source apportionment under varying meteorological conditions.

Flint Hills Resources collaborated with EPA in execution of this study by granting permission to deploy the PSs and by allowing access to their on-site leak detection and repair (LDAR) contractor for sample deployment. The study was performed at the Flint Hills West Refinery in Corpus Christi, TX, which has a nominal crude oil refining capacity of approximately 260,000 barrels per day. The West Refinery includes typical refining operations such as fluid catalytic cracking and distillate hydrocracking, delayed coking, and the associated petrochemical extraction and conversion process units. For the study period, production was relatively consistent, although there were regularly scheduled maintenance and periodic shutdown and startup activities. Because the emphasis for this study was on the use and performance of the PS measurement approach and not on assessment of the actual emissions from the refinery, no attempt was made to gather detailed process or operation information. Because of the relative consistency of production throughout the year and the time-integrated nature of the measurement, it is believed that day-to-day production variability had little impact on observations or the data groupings suggested below.

EXPERIMENTAL METHODS The use of PSs with various designs and sorbent materials for ambient monitoring applications has been documented in the literature (7-13) with much effort related to the development of monitoring protocols for the European Community Directive 2000/69/EC and daughter directives that set limits on ambient concentrations of hazardous air pollutants, including benzene. The study presented here utilized Carbopack X sorbent (Tilde650 mg) in ceramic-lined Perkin-Elmer tubes (Supelco, Inc.), 6 mm in diameter by 90 mm length, with laboratory analysis by a thermal desorption gas chromatograph (GC; TurboMatrix ATD, Perkin-Elmer Instruments, LLC, and Saturn 2000 GC/MS, Agilent). Information on use of Carbopack X sorbent for the determination of the concentration of benzene and other volatile organic compounds in ambient air,(14-17) along with details on the custom Carbopack X Perkin-Elmer tubes and laboratory analytical procedures used in this study, are summarized by McClenny, Mukerjee, and others. (16-19) The PSs were deployed at 18 locations on the fence line of the Flint Hills Refinery West Facility in Corpus Christi, TX, and at two Texas Commission on Environmental Quality (TCEQ) continuous air monitoring station (CAMS) sites: C633, which is south of the facility (Figure 1), and C634, which is located approximately 10 km east of the facility (not shown). The PS locations (loc) are labeled by their approximate angular position as observed from the center of the facility with north representing 0[degrees]. The configurations of the PSs were fixed for the study with the exception of loc 40, 501, and 60, which were added during P12 to help diagnose high observed concentrations in the vicinity of loc 50. Duplicate and field blanks were deployed at loc 180 and C633, with duplicates added to loc 360 during P22. TCEQ CAMS sites C633 and C634 performed automated GC benzene measurements that provided 1-hr average concentration values (Clarus Model 500 GC, Perkin-Elmer) and PSs were placed at both locations for comparison purposes. Also shown in Figure 1 are TCEQ CAMS sites C631 and C632, which did not operate automated GCs but provided total nonmethane organic carbon (TNMOC) measurements and meteorological data useful for future comparisons.

The PS data set produced by the year-long deployment consisted of 579 samples, including 56 duplicates and 49 field blanks. Seventeen samples were excluded from the analysis because of a combination of tube damage or deployment issues (n = 6) and laboratory equipment malfunction (n = 11), yielding a data completeness value of 97.1%.

[FIGURE 1 OMITTED] The benzene concentration data for 10 samples exceeded the demonstrated linearity range of 4071 parts per trillion by volume (pptv) for the analytical system utilized and, as a consequence, these values contain additional uncertainty estimated to be below 20%. Further information on the PS configuration, deployment, and laboratory analysis can be found in the quality assurance project plan and online supplemental information.

RESULTS AND DISCUSSION The objective of this study was to gain information on the implementation feasibility and measurement performance of the 2-week passive samplers in a real-world fence-line deployment scenario. Information such as overall data completeness, duplicate and field blank data, site-to-site trend consistency, and comparisons with automated GCs help form a basis for judging the efficacy of the overall measurement approach. This section begins with a description of meteorological conditions encountered during the study and PS validation data. An overview of the combined fence-line results by location and period are then presented. This information is followed by a discussion of time-series results and upwind (UW) versus downwind (DW) comparisons under uniform and mixed wind direction conditions. The latter data are important because they provide valuable insights on the general utility of time-integrated monitoring for determining a facility’s contribution to the observed fence-line concentrations.

Meteorology The 14-day average of hourly wind direction (WD), scalar wind speed (SWS), vector resultant wind speed (RWS), temperature (T), and relative humidity (RH) data recorded by the C633 site (20) are summarized in Table 1. The SWS represents a simple average of hourly values of wind speed for the 2-week period. The RWS was determined by first calculating the orthogonal vector components (north = 0[degrees], east = 90[degrees]) for each hourly reading, averaging individual components for the period, and then calculating the magnitude of the resulting vector. The RWS increases from zero with decreasing WD variability, approaching the value of SWS when winds are highly uniform over the 2-week period. The difference in RWS and SWS is one of several ways to quantify the degree of WD uniformity, which is important for time-integrated sampling approaches. Confidence in meteorological data acquired from the TCEQ CAMS site is bolstered by the state’s quality assurance requirements and additionally by well-correlated crosschecks with nearby CAMS sites for important meteorological variables.

In comparison to other parts of the United States, Corpus Christi exhibits strong wind speeds and periods of highly uniform WD that can have a significant effect on comparative analysis with time-integrated monitors. In preparation for discussion on this point, two 6-month groupings of wind data from the study are presented in Figure 2. Figure 2a shows a wind rose for a grouping of P1-P6 coupled with P20-P26 and reflects a period of relatively mixed WDs. Figure 2b shows a grouping of P7-P19 exhibiting a continuous 6-month period when the WD was more uniformly from the southeast. For the grouping of Figure 2a, the average difference in SWS minus RWS is 7 mph, whereas for the more uniform case the difference is 2.8 mph. A 6-month grouping of neighboring periods was chosen for simplicity with period cutoff decided by comparing the SWS and RWS values for individual periods.

Table 1. Summary of PS fence line meteorological data by period.

Period End Date WD([degrees]) SWS RWS T([degrees]F)57 RH(%) (mph) (mph)

P1 December 210 11 1 57 58 17, 2008

P2 December 135 11 3 63 67 31, 2008

P3 January 191 2 60 60 14, 2009

P4 January 166 12 3 61 56 28, 2009

P5 February 157 13 8 62 58 11, 2009

P6 February 133 11 6 65 62 25, 2009

P7 March 11, 169 16 13 69 62 2009

P8 March 25, 167 12 6 64 73 2009

P9 April 8, 190 13 4 71 47 2009

P10 April 22, 154 13 7 75 59 2009

P11 May 6, 138 19 18 80 68 2009

P12 May 20, 141 16 12 81 63 2009

P13 June 3, 173 9 6 80 64 2009

P14 June 17, 158 13 12 84 63 2009

P15 July 1, 151 12 11 86 63 2009

P16 July 15, 151 13 12 87 63 2009

P17 July 29, 151 15 14 87 62 2009

P18 August 12, 150 14 14 87 62 2009 site corpus christi tx

P19 August 26, 146 11 10 87 61 2009

P20 September 175 8 3 84 62 9, 2009

P21 September 191 8 1 82 66 23, 2009

P22 October 7, 163 11 5 80 73 2009

P23 October 132 12 5 75 68 21, 2009

P24 November 198 10 2 69 62 4, 2009

P25 November 191 9 3 68 67 18, 2009

P26 December 166 9 3 61 73 2, 2009 [FIGURE 2 OMITTED] PS Validation Results Benzene concentration data determined by PSs were compared with the 14-day average of 1-hr data from the GCs at the C633 and C634 sites. A linear regression of PS and GC data (Figure 3) shows an unconstrained [r.sup.2] value of 0.86 and slope of 0.90 (n = 50 using average values for 26 duplicate pairs at C633). The method detection limit (MDL) for the PS was determined to be 35 pptv, and the MDL for the GC was found by TCEQ to be 50 pptv for benzene for the study period. Figure 3 utilizes all reported GC data, which include a significant number of hourly values below the MDL (25%) with a disproportionate number occurring in the P7-P19 with winds away from the facility. As an example, for P14-P19, only 1036 of a possible 3427 GC values were above the MDL. For the same time period, the average value of the PS concentrations at the GC locations was 139 pptv with a minimum of 85 pptv, significantly above the MDL, because of the time-integrated nature of the sampling approach. The y-intercept value of 142.2 pptv in Figure 3 is largely determined by the decision to include all available GC data in the comparison. Removing all GC data below the MDL gives a linear regression [r.sup.2] value of 0.78, a slope of 0.87, and a y-intercept of 91.85 pptv. These data comparisons ultimately depend on the definition of minimum quantization limit (MQL) and the choice for assignment of fixed values for below-MQI. entries. This highlights a difficulty in comparing time-integrated and time-resolved approaches that is not of major concern for the current discussion on fence-line monitoring in which levels significantly above MDL are of primary importance.

[FIGURE 3 OMITTED] The range of comparison for the PS and GC data (Tilde100-1000 pptv) is somewhat lower than optimal for validation purposes. The low upper limit in 2-week average concentrations at the TCEQ sites was due in part to their locations that were significantly displaced from the fence line of the facility. Although not optimal, the upper limits of comparison are reasonable when considering that the overall PS study average for fence-line locations was approximately 1000 pptv. To achieve higher ranges of comparison, future field studies should consider co-deployment of PS and GC on the fence line of a facility near areas of high expected concentration. In addition to Figure 3, time-series comparisons of C633 GC and C633 PS providing further validation information are discussed in a subsequent section.

Duplicate PSs were located at C633 and loc 180 for all periods and at loc 360 for P22-P26 (Figure 4). Concentrations of duplicate pairs ranged from approximately 100 to 1200 pptv with a linear regression [r.sup.2] of 0.97, unity slope, and near-zero intercept (n = 56). The average difference in the duplicate values was 8.5% with a maximum of 33% occurring for a low range reading (229 vs. 333 pptv). Duplicates were added to loc 360 during P22 in response to high readings observed in P14 and P15. The delay in deployment of the duplicates was due to the batch processing of samples, which delayed data availability. Field blanks deployed at loc 180 and C633 had an average value of 8 pptv with a standard deviation (SD) of 6.2 pptv (n =49). Supplemental Table S1 contains all duplicate, gas chromatography, and field blank data for the study.

PS Fence-Line Results A total of 454 PSs were deployed at 18 fence-line sites around the facility (Figure 1). Combining all fence-line PS results, the mean benzene concentration was 1075 pptv with a SD of 1935 pptv. The PS median value was 709 pptv with a minimum of 122 pptv and a maximum of 29,280 pptv. With fence-line-deployed duplicates averaged, 8.5% of the readings were above 2000 pptv, 21.2% were between 1000 and 2000 pptv, and 70.3% were below 1000 pptv. Fence-line PS data, along with the off-site C633 PS data, are summarized by location in Figure 5. The C633 PS has a mean benzene concentration value of 318 (SD = 160) pptv, slightly below neighboring fence-line sites, loc 250 ([approximately equal to] 630 m’away) has a mean value of 416 (SD = 183) pptv, and loc 270 (Tilde510 m away) has a mean value of 395 (SD = 121) pptv. The differences in the means of C633 compared with loc 250 and separately with loc 270 are statistically significant at an a of 0.05 with t test P values of 0.042 and 0.048, respectively. The ability to detect differences in PSs deployed on the fence line and at proximate off-site locations is potentially important in future gradient-based source comparison strategies.

[FIGURE 4 OMITTED] PS benzene concentration values on the predominately UW southern fence line, consisting of loc 130-250, show a group mean of 613 (SD = 353) pptv, lower than the northern fence line (loc 310 through loc 50) having a group mean of 1840 (SD = 3169) pptv with group mean difference P value less than 0.001. Excluding the two extreme outliers at loc 360, the northern fence-line group mean is 1512 (SD = 1494) pptv, with similar group mean difference P values.

[FIGURE 5 OMITTED] Figure 6 shows PS benzene concentration data for a subset of fence-line locations by sampling period (loc 40, 501, and 60 are not included because of incomplete sets). The mean values and number of outliers (values that extend beyond 1.5 times the interquartile range) are somewhat higher in the warmer months and lower in the cooler months. It is not known if these differences are because of higher emissions or because of the effects of atmospheric conditions on ground-level concentrations at different times of the year. Because PS comparisons with automated gas chromatography show little seasonal variation (next section), these differences are not believed to be due to measurement bias.

As discussed in the text associated with Figure 2, it can be informative to form two 6-month groupings of PS results from neighboring time periods. This is performed here with the spatially integrated data of Figure 6 and in a subsequent section by resolving the locations into UW and DW subgroups. The mixed WD 6-month group (Figure 2A) contains cooler months (average T = 68 [degrees]F) and has a mean benzene concentration value of 798 (SD = 414) pptv (n = 188). The uniform WD grouping (Figure 2B) has a higher average T (80 [degrees]F) and a mean concentration value of 1288 (SD = 2799) pptv (n > 192). There is a statistically significant difference in group means (P = 0.017) for these sets. For this data comparison, 88.5% of readings above 2000 pptv occur in the P7-P19 uniform wind group. These observed differences were not due to changes in refinery operations because production levels were confirmed to be relatively consistent throughout the study period.

Time-Series Comparisons Techniques for temporal analysis of fence-line monitoring data are determined in large part by the time resolution of the measurement. For example, monitoring schemes with time resolutions less than 1 hr can utilize intraday trend analysis and metrological comparisons to help apportion local source contributions. The time-integrated nature of the PS approach makes it less useful in this context; however, important information on longer-term temporal and spatial trends can be gained through time-series comparisons. Because the implementation cost of the PS approach is lower than similar density deployments of time-resolved monitors, passive sampling has clear advantages for acquisition on longer-term trend information [FIGURE 6 OMITTED] [FIGURE 7 OMITTED] Figure 7 examines the southern fence-line benzene concentrations from the C633 automated GC, the collocated C633 PS, and the average of two nearby sites–loc 250 and loc 270. Similar trends in the concentrations are evident. For example, comparatively lower concentrations in P7 and P9-P19 are observed for the PS and automated GC, although the overall average concentration for the fence-line sites is higher for these periods (Figure 6). Basic WD expressed as the percentage of winds coming out of the southern hemisphere is shown on the secondary y-axis. During periods P7, P11, and P14-P20, winds are directionally toward the north, transporting facility source signal away the from the southern fence-line samplers, resulting in lower observed concentrations. A linear regression comparison of C633 PS data with a percentage of southerly winds yields an unconstrained [r.sup.2] value of 0.70. This relatively high correlation indicates that the PS readings are likely influenced by emissions transported from the facility and that the 2-week time-integrated sampling approach is able to register changes in prevailing wind orientation with respect to the source.

The similarity in the time series for the fence-line PS (loc 250 and loc 270) and the off-site PS (C633) provides some confidence that the mobile sources using Interstate Highway 37, located between the observation points (Figure 1), are not producing significant interfering benzene signal. If this were the case, divergence in the time series with changing prevailing winds would be expected. The time series shown in Figure 7 also provides supporting validation information for the PS by showing similar period-to-period variations of the PS compared with the automated GC. Similar comparative results were also observed for the C634 site. These comparisons, taken over the year-long study, provide some evidence that seasonal changes in T and humidity have little effect in PS performance for the range of conditions encountered in this field campaign.

Ideally, the fence-line PSs should be deployed away from obstructions that can impede wind flow and away from potential interfering sources outside of the fence line. Both of these situations can lead to elevated concentrations measured by the PSs that are not caused by the observed facility. Time-series analysis along with deployment of additional diagnostic samplers can be used to help understand elevated PS readings and to identify PS siting and interfering source issues. Figure 8 shows PS benzene concentrations in the neighborhood of loc 50 (Figure 1), which is positioned in a complicated environment including complex local topography and potential neighboring sources. The ground level to the southwest, near loc 501, is elevated by approximately 2 m compared with the location of loc 50. This local topography could cause complex wind flow (channeling or vortexes) in the neighborhood of loc 50, potentially affecting measured concentrations. Potential sources such as the barge loading operations to the north and the wastewater treatment to the southeast are outside of the defined fence line. Loc 50 exhibited the first notably high concentration during P7, and, in response to this reading, several additional PS sites were implemented (loc 40, 501, and 60) during P12. These samplers also showed somewhat elevated concentrations before P20, but the results are difficult to correlate with loc 50 values or WD. For this fence-line location, the combination of topography and additional potential sources makes it difficult to draw conclusions about emissions from the primary observed target, the facility to the southwest. For example, the PS at loc 501 is inside of the fence line, closer to the potential facility sources (tanks). Because of proximity, higher concentrations at loc 501 would be expected compared with loc 50 if the primary source were the observed facility, all other factors being equal. Additionally, the three highest readings at loc 50 occur during P7, P11, and PI6, with a high percentage of winds from the south (Figure 7) and more specifically from the southeast (Figure 2b). This fact, coupled with the relative response of neighboring PSs, implies that the wastewater treatment area outside of the defined fence line is a likely interfering source.

[FIGURE 8 OMITTED] A strength of PS-based fence-line monitoring is in providing cost-effective, high spatial density, long-term monitoring capability. As evidenced by the results of Figure 8, a weakness of time-integrated monitoring lies in its inability to apportion contributions to the measured concentration in complex source and micrometeorological conditions. In cases in which additional source apportionment capability is required, the PS approach can be selectively augmented through the use of time-resolved fence-line monitoring coupled with WD analysis.

The time series of Figure 9 illustrates the elevated nature of data acquired DW of the facility and provides some perspective on extreme outlier readings. In comparison to UW site loc 130, the DW sites register consistently higher benzene concentrations, especially during P7-P19, when winds are uniformly from the southeast. The two highest readings were recorded at loc 360 during P14 and P15 (29,280 and 20,007 pptv, respectively). These outlier values are significantly elevated compared with the third-highest reading of 8891 pptv (loc 20, P11) and are approximately 6 SDs displaced from the northern fence-line group average of 1840 pptv. At approximately 5 times the demonstrated analytical linearity range for this study, the accuracy of the P14 and P15 outlier readings is uncertain. However, the occurrence of significantly elevated values at loc 360 during these periods is not unexpected when considering the elevated neighboring observations in the time series of Figures 8 and 9.

Two-week integrated PS readings at the 30,000-pptv level are somewhat difficult to understand because if these readings are not due to gross analytical error, they are a result of sustained elevated concentrations transported by the wind to the sampling location or a shorter time duration intense spike in local concentration in close proximity to the PS caused by a transient source. Because the two similar outlier values were produced from separate sorbent tubes and analyzed during different laboratory runs, it is unlikely that analytical error is the cause of the elevated readings. The explanation of sustained elevated benzene concentrations at the PS is somewhat unlikely considering loc 360 is over 300 m distant from the nearest above-ground facility structure (tanks to the southeast, Figure 1). These observations do not preclude the presence of a below-ground emission or a temporary source not obviously part of the facility fence-line observation. An unknown or temporary source (such as a rail car) could have been located in very close proximity to the loc 360 PS, thereby causing a large integrated response.

[FIGURE 9 OMITTED] A local benzene spike could include an actual emission or could be attributed to a sample handling issue. As an example of the latter, the field operators were instructed not to refuel their vehicles before sample handling to reduce the chance that gasoline vapor entrained on their hands could provide an intense concentration spike to the PS when uncapped during deployment. This type of sample corruption could affect two successive sampling periods because the pick-up of the deployed PS and the placement of the new PS occur at similar times. This type of deployment error can be investigated by looking at neighboring samples (loc 20 and loc 330), which in this case were deployed within 7 min before and after the loc 360 PI5 PS and did not show abnormally high results. In future PS protocol development work, placement of secondary samplers with offset deployment schedules could assist in diagnosing outlier issues. These secondary samplers could be analyzed only when abnormal results arise to minimize cost for this diagnostic.

UW versus DW Comparisons A primary objective of any fence-line monitoring strategy is to positively identify the observed facility’s contribution to the measured concentrations. One way to accomplish this is by comparing the concentrations registered by the monitors stationed UW of the facility with DW sampling locations with the difference indicative of emissions. Because meteorological conditions change over time, the designation of UW and DW monitors is mutable. When using time-resolved fence-line monitoring instrumentation, this UW versus DW comparison is accomplished by reviewing measured concentration data in conjunction with simultaneously acquired meteorological data. When using 2-week time-integrated PSs, this approach is less viable, especially in cases in which the WD is mixed.

To investigate this effect, uniform and mixed WD scenarios can be compared using the two 6-month periods defined in Figure 2. An UW group, consisting primarily of southern fence-line sites (loc 90-270), and a DW group (loc 290 through loc 20) can be formed for the P1-P6, P20-P26 mixed WD case (Figure 2a) and the P7-P19 uniform WD case (Figure 2b). For the mixed WD case, the UW mean is 768 (SD 366) pptv and the DW mean is 718 (SD = 297) pptv (Figure 10a). Differences in these group means are not statistically significant (P = 0.330). For the uniform WD case, the UW mean of 487 (SD = 277) pptv is significantly lower than the DW mean, 2197 (SD = 4361) pptv, with a P value of 0.003 (Figure 10b). Excluding the two high outliers at loc 360, the DW mean becomes 1473 (SD = 1371) pptv with an improved P value of less than 0.001 for a mean difference comparison with the UW samples. Comparing across the mixed and uniform WD cases, the UW means are also statistically different from each other at the 99% confidence interval. For multimonth groupings of 2-week PSs, it is possible to resolve UW and DW differences in the case of highly uniform WD. However, it is difficult to draw conclusions on facility contributions to the measured fence-line concentrations using a simple UW-DW approach for the mixed WD case that may be more typically encountered in other areas of the United States.

[FIGURE 10 OMITTED] To further investigate the effects of WD, Figure 11 plots the percentage difference in UW and DW benzene concentrations with a metric indicative of WD uniformity formed by calculating the percentage difference in the scalar and resultant vector wind speeds (SWS and RWS in Table 1). For periods with uniform WD, SWS and RWS are similar, so emissions from the facility are transported toward the DW sampling locations a high percentage of the time, resulting in a larger difference in the UW and DW concentrations. As the difference in SWS and RWS increases, the temporal overlap of the facility-generated plume is more equally shared between the UW and DW samplers, so their percentage difference decreases and actually becomes slightly negative during the winter months as the UW leg experiences higher concentration on average. The relationship expressed in Figure 11 depends on the definition of UW and DW sites, which may change throughout the year on the basis of site-specific metrological conditions. site corpus christi tx

The ability to resolve statistically significant differences in PS concentrations using fence-line-deployed PSs depends on the degree of WD uniformity and on factors such as sampling time integration, wind speed (degree of stagnation), and the offset distance of PSs from facility sources (dilution effects). At 14 days, the time duration of sampling used for this study was judged to be an optimal tradeoff between time resolution and cost. To help increase the diagnostic capability of the 2-week PS approach, future protocol development could include a significant number of off-fence-line sites set back from the primary monitors so as to allow concentration gradient analysis to aid in deciphering facility contributions under mixed WD cases. The reduction in concentration by atmospheric dispersion along the gradient will help provide source apportionment information. Additionally, ways to systematically define UW and DW site groups on the basis of statistical comparisons of concentrations on a rotating sector basis should be explored.

CONCLUSIONS This field demonstration provides first-level validation data for the PS fence-line monitoring approach while explaining future method development needs. With high data completeness rates, the year-long study provides evidence that the approach is relatively robust and implementable by modestly trained personnel. On the basis of cost figures from the study presented here, the expense for commercial application of a standardized method is projected to be below $200 per sample for a single component analysis. The implementation factors for the PS approach are attractive in comparison to similar density deployments of time-resolved monitoring technologies, which can come at much higher capital and operational costs.

The PS fence-line concept can provide useful information on overall concentration levels and potential problem areas on the facility fence line using simple source identification techniques such as UW-DW comparisons, temporal trend investigation, and gradient analysis. A weakness of the time-integrated approach is found when attempting source apportionment in complex environments. In this event, elevated concentration areas found in the PS screen can be further investigated with selective use of time-resolved monitoring where deemed necessary. The use of PSs alone or in combination with optimally deployed time-resolved monitors can form the basis for cost-effective and flexible fence-line monitoring strategies.

[FIGURE 11 OMITTED] Remaining method development questions center on establishment of PS performance with a wider concentration range and the expansion to compounds other than benzene. Future validation work should consider GC placement at DW fence-line locations to expand the range of comparison and potentially include the deployment of spike duplicate samples to investigate off-gassing effects. New deployment strategies must also be developed to allow effective source apportionment for the observed facility in areas with mixed WDs and higher percentages of stagnant conditions. These deployment strategies are envisioned to include a gradient sampling approach with PS monitors placed progressive distances from the fence line. Another area for improvement is optimized duplicate deployment strategies to provide additional quality assurance information in the event of anomalous primary readings. For cases of complex sources or joint property fence-line deployments, low-cost, open-path, time-resolved monitoring will be evaluated as a way to cost-effectively augment the PS screening approach.

ACKNOWLEDGMENTS This work reflects the contributions of many individuals. In particular, the authors acknowledge the efforts of Jan Golden and Eric Kaysen with Flint Hills Resources for their collaboration; Karen Oliver, Hunter Daughtrey, Tamira Cou-sett, and Herb Jacumin with Alion Science and Technology for analytical support under EPA Office of Research and Development (ORD) contract EP-D-05-065; Mark Modrak of ARCADIS for project coordination under EPA ORD contracts EP-C-04-023 and EP-C-09-027; and many individuals with Shaw Environmental, Inc., for deployment of the PSs. The authors thank Edward Michael, Vincent Torres, and David Allen with the University of Texas and David Brymer and Chris Owen with TCEQ for their assistance in acquiring automated GC validation data. The authors appreciate the direction and support of Robin Segall, Jason DeWees, Raymond Merrill, and Connie Sue Oldham with EPA’s Office of Air Quality Planning and Standards and the quality assurance assistance of Bob Wright and Dr. Joan Bursey with EPA ORD. This article was reviewed by EPA ORD and was approved for publication. Approval does not signify that the contents necessarily reflect the views and policies of the agency nor does mention of trade names or commercial products constitute endorsement or recommendation for use.

REFERENCES (1.) Fischer, C; van Harcn, G.; Chaudhry, A.; Weber, K. Detection of Potential Leakage at a Biogas Production Plant with Open-Path KFIR Measurement Techniques–a Case Study; Gefhhrst. Reinhatt. Luft 2O06, 66, 426-430.

(2.) Mickunas, D.B.; Zarus, G.M.; Turpin, R.D.; Campagna, P.R. Remote Optical Sensing Instrument Monitoring to Demonstrate Compliance with Short-Term Exposure Action Limits during Cleanup Operations at Uncontrolled Hazardous Waste Sites; J. Hazard. Mater. 1995, 43, 55-65.

(3.) Takach, S.F.; Schulz, S.P.; Minnich, T.R.; Scotto, R.L. Results of Gas Technology Institute’s ORS Methods Development Project for Perimeter Air Monitoring during MGP Site Cleanups. In Proceedings of the 101st Annual Conference of the A& A&WMA: Pittsburgh, PA, 2008; Paper No. 756.

(4.) Lin, C.S.; Liou, N.W.; Chang, P.E.; Yang, J.C.; Sun, E. Fugitive Coke Oven Gas Emission Profile by Continuous Line Averaged Open-Path Fourier Transform Infrared Monitoring; J. Air & Waste Manage. ASSOC. 2007, 57, 472-479; doi: IQ.31SS/W47-3289.57.4A72.

(5.) Ciaparra, D.J Aries, E.; Booth, M.J.; Anderson, D.R.; Almeida, S.M.; Harrad, S. Characterization of Volatile Organic Compounds and Poly-cvclic Aromatic Hydrocarbons in the Ambient Air of Steelworks; Atmos. Environ. 2009, 43, 2070-2079.

(6.) Thoma, E.D.; Secrest, C; Hall, E.S.; Jones, D.L.; Shores, R.C.; Modrak, M.; Hashmonay, R.; Norwood, P. Measurement of Total Site Mercury Emissions from a Chlor-Alkali Plant using Ultraviolet Differential Optical Absorption Spectroscopy and Cell Room Roof-Vent Monitoring; Atmos. Environ. 2009, 43, 753-757.

(7.) Brown, R.H. Environmental Use of Diffusive Samplers: Evaluation of Reliable Diffusive Uptake Rates for Benzene, Toluene and Xylene; J. Environ. Monit. 1999, 1, 115-116.

(8.) Ballach, J.; Greuter, B.; Schultz, E.; Jaeschke, W. Variations of Uptake Rates in Benzene Diffusive Sampling as a Function of Ambient Conditions; Sci. Total Environ. 1999, 244, 203-217.

(9.) Brown, R.H. Monitoring the Ambient Environment with Diffusive Samplers: Theory and Practical Considerations; J. Environ. Monit. 2000, 2, 1-9.

(10.) Buzica, D.; Gerboles, M.; Plaisance, H. The Equivalence of Diffusive Samplers to Reference Methods for Monitoring O3, Benzene and N[O.sub.2] in Ambient Air; J. Environ. Monit. 2008, 10, 1052-1059.

(11.) Woolfenden, E. Sorbent-Based Sampling Methods for Volatile and Semi-Volatile Organic Compounds in Air. Part 2. Sorbent Selection and Other Aspects of Optimizing Air Monitoring Methods; J. Chro-matogr. A 2010, 1217, 2685-2694.

(12.) Pfeffer, H.U.; Breuer, L. BTX Measurements with Diffusive Samplers in rile Vicinity of a Cokery: Comparison between ORSA-Type Samplers and Pumped Sampling; J. Environ. Monit. 2000, 2, 483-486.

(13.) Ambient Air Quality Standard Method for Measurement of Benzene Concentrations. Part 4: Diffusive Sampling followed by Thermal Desorption and Gas Chromatography; Technical Committee CEN/TC 264; BS EN 14662-4:2005; British Standards Institution: London, United Kingdom, 2005.

(14.) Strandberg, B.; Sunesson, A.L.; Olsson, K.; Levin, J.O.; Ljungqvist, G.; Sundgren, M.; Sallsten, G.; Barregard, L. Evaluation of Two Types of Diffusive Samplers and Adsorbents for Measuring 1, 3-Butadiene and Benzene in Air; Atmos. Environ. 2005, 39, 4101-4110.

(15.) Martin, N.A.; Marlow, D.J.; Henderson, M.H.; Goody, B.A.; Quincey, P.G. Studies Using the Sorbent Carbopack X for Measuring Environmental Benzene with Perkin-Elmer-Type Pumped and Diffusive Samplers; Atmos. Environ. 2003, 37, 871-879.

(16.) McClenny, W.A.; Oliver, K.D.; Jacumin, H.H.; Daughtrey, K.H.; Whi-taker, D.A. 24 h Diffusive Sampling of Toxic VOCs in Air onto Carbopack X Solid Adsorbent followed by Thermal Desorption/GC/MS Analysis–Laboratory Studies; J. Environ. Monit. 2005, 7, 248-256.

(17.) McClenny, W.A.; Jacumin, H.H., Jr.; Oliver, K.D.; Daughtrey, E.H., Jr.; Whitaker, D.A. Comparison of 24 h Averaged VOC Monitoring Results for Residential Indoor and Outdoor Air Using Carbopack X-Filled Diffusive Samplers and Active Sampling–a Pilot Study; J. Environ. Monit. 2006, 8, 263-269.

(18.) Mukerjee, S.; Oliver, K.D.; Seila, R.L; Jacumin, H.H., Jr.; Croghan, C; Daughtrey, E.H., Jr.; Neas, L.M.; Smith, LA. Field Comparison of Passive Air Samplers with Reference Monitors for Ambient Volatile Organic Compounds and Nitrogen Dioxide under Week-Long Integrals; J. Environ. Monit. 2009, 11, 220-227.

(19.) Mukerjee, S.; Smith, L.A.; Norris, G.A.; Morandi, M.T.; Gonzales, M.; Noble, C.A.; Neas, L.M.; Ozkaynak, A.H. Field Method Comparison between Passive Air Samplers and Continuous Monitors for VOCs and N[O.sub.2] in El Paso, Texas; J. Air& Waste Manage. Assoc 2004, 54, 307-319.

RELATED ARTICLE: IMPLICATIONS Improved knowledge of air pollution concentrations at industrial facility fence lines is a topic of increasing environmental importance. Fence-line and process monitoring can yield many benefits, ranging from enhanced risk management to cost savings through improved process control. Efforts are underway within EPA to develop and test various cost-effective fence-line monitoring strategies for potential use in a range of research and regulatory applications. Among these, passive diffusive sampling is emerging as a promising technique for time-integrated fence-line monitoring applications.

RELATED ARTICLE: About the Authors Eben D. Thoma U.S. Environmental Protection Agency, Office of Research and Development, National Risk Management Research Laboratory, Research Triangle Park, NC Michael C. Miller and Kuenja C. Chung U.S. Environmental Protection Agency, Region 6, Dallas, TX Nicholas L. Parsons and Brenda C. Shine U.S. Environmental Protection Agency, Office of Air Quality Planning and Standards, Research Triangle Park, NC Thoma, Eben D.; Miller, Michael C.; Chung, Kuenja C.; Parsons, Nicholas L.; Shine, Brenda C.