Field Investigation of Vapor-Phase-Based Groundwater Monitoring

Transcription

1 Field Investigation of Vapor-Phase-Based Groundwater Monitoring by David T. Adamson, Thomas E. McHugh, Michal W. Rysz, Roberto Landazuri, and Charles J. Newell Abstract The use of in-field analysis of vapor-phase samples to provide real-time volatile organic compound (VOC) concentrations in groundwater has the potential to streamline monitoring by simplifying the sample collection and analysis process. A field validation program was completed to (1) evaluate methods for collection of vapor samples from monitoring wells and (2) evaluate the accuracy and precision of field-portable instruments for the analysis of vapor-phase samples. The field program evaluated three vapor-phase sample collection methods: (1) headspace samples from two locations within the well, (2) passive vapor diffusion (PVD) samplers placed at the screened interval of the well, and (3) field vapor headspace analysis of groundwater samples. Two types of instruments were tested: a field-portable gas chromatograph (GC) and a photoionization detector (PID). Field GC analysis of PVD samples showed no bias and good correlation to laboratory analysis of groundwater collected by low-flow sampling (slope = 0.96, R 2 = 0.85) and laboratory analysis of passive water diffusion bag samples from the well screen (slope = 3; R 2 = 0.96). Field GC analysis of well headspace samples, either from the upper portion of the well or at the water-vapor interface, resulted in higher variability and much poorer correlation (consistently biased low) relative to laboratory analysis of groundwater samples collected by low-flow sample or passive diffusion bags (PDBs) (slope = 0.69 to 0.76; R 2 = 0.60 to 0.64). These results indicate that field analysis of vapor-phase samples can be used to obtain accurate measurements of VOC concentrations in groundwater. However, vapor samples collected from the well headspace were not in equilibrium with water collected from the well screen. Instead, PVD samplers placed in the screened interval represent the most promising approach for field-based measurement of groundwater concentrations using vapor monitoring techniques and will be the focus of further field testing. Introduction Long-term monitoring (LTM) programs represent a significant cost burden for federal and commercial sites managing contaminant plumes in groundwater. Typically, these programs rely on an extensive network of monitoring wells installed throughout the site, sampled on a routine basis, to establish (or determine progress toward) concentration-based compliance goals. Sampling these wells is an intensive process, requiring field personnel to visit each well, collect a groundwater sample, label and package the sample(s) for shipment to an off-site commercial laboratory. Because of the number of steps and different parties involved, the process of obtaining a single concentration datapoint represents a significant time and cost investment. Although the per-sample charge for volatile organic compound (VOC) analysis by an accredited commercial 2011, GSI Environmental INC Ground Water Monitoring & Remediation 2011, National Ground Water Association. doi: /j x laboratory is typically close to $100 per sample, the total cost burden when including all other steps in the conventional sampling and analysis process is more likely to be in the range of $600 to $1000 per sample Multiply this cost by the sheer number of sites, as well as the number of monitoring wells present at each of these sites, and it is not surprising that estimates of annual LTM liabilities for the Department of Defense (DoD) easily exceed $100 million. Furthermore, the reliance on multistep processes can introduce significant variability into groundwater data, hindering our ability to make well-informed decisions about site management. There has been a distinct shift toward adopting less-intensive methods for groundwater monitoring in the past few decades, with the objective of reducing both the cost of obtaining data as well as the time-independent variability in the data. The most important trend has been the widespread acceptance of low-flow sampling as the primary method for groundwater monitoring. Low-flow sampling is a low stress method of purging and sampling a monitoring well that has gained widespread regulatory acceptance, with standardized practices (ASTM D 6771) and guidance documents (e.g., Puls and Barcelona 1996). NGWA.org Ground Water Monitoring & Remediation 32, no. 1/ Winter 2012/pages

2 The key procedure involves pumping small volumes of groundwater from the well screen at low-flow rates until stabilization of geochemical parameters is achieved and drawdown is stable. Because not all of the well casing volume is removed, this method is reliant on the natural (ambient) flow of groundwater through the screened interval (Robin and Gilham 1987; Powell and Puls 1993). This approach is intended to minimize (1) the disturbance of potentially stagnant water from above or below the well screen; (2) the movement of groundwater from the formation to the well that is normally induced by high volume purging; and (3) the amount of purge water that must be handled as waste. The goal of low-flow purging is to obtain a sample that is representative of the groundwater adjacent to the screen. A second trend in groundwater monitoring has been the validation of even less-intensive no-purge methods, including various passive samplers. An early comparative survey of monitoring data collected in the absence of purging demonstrated that at many sites, there was no statistically significant difference between no-purge samples and those collected using conventional purging techniques (Newell et al. 2000). At the same time, passive sampling methods such as diffusion bags were being tested and shown to be widely applicable for measuring volatile compounds in groundwater (Vroblesky 2001; ITRC 2004, 2006). These samplers are submerged below the water level and generally rely on the diffusion of volatile groundwater constituents across a semipermeable membrane. Equilibrium is obtained over the course of 1 to 3 weeks, and the concentration in the sampler is intended to represent the concentration at the depth in which the device was deployed. As such, they are well suited for determining depth-discrete concentrations within a well. Several other grab-sample devices have been developed and tested extensively, including in situ sealed samplers (i.e., Snap Samplers) (ITRC 2007; Parker and Mulherin 2007; Britt et al. 2010). Extensive technical and regulatory guidance has been published, including the results of comparative validation studies. Regardless of the design, all samplers share similar goals, specifically to provide an easy and reliable measure of groundwater concentration while minimizing purge water generation and agitation within the well (ITRC 2007; Verreydt et al. 2010). A further opportunity for developing low-intensity monitoring methods lies in the use of equilibrium partitioning principles for vapor-phase-based groundwater monitoring (Adamson et al. 2009). This approach measures the concentration of vapor that is in equilibrium with groundwater in a monitoring well, and then correlates that vapor measurement to a groundwater concentration using the constituent-specific Henry s coefficient. By employing field-portable instruments capable of low-level detections of volatile compounds, vapor-phase-based groundwater monitoring may provide a reliable and low-cost alternative. Specifically, this approach eliminates several of the timeconsuming steps associated with conventional monitoring by forgoing purging and completing the sample analysis in the field. This approach is based on the hypotheses that (1) portable vapor-phase monitoring instruments can be used to accurately determine VOC concentrations in water under equilibrium conditions and (2) typical conditions within monitoring wells are conducive to simple vapor sampling and analysis methods for establishing a correlation with groundwater concentrations. Therefore, the overall objective of this research project was to determine if in-field vaporphase monitoring of in-well vapor samples will provide an accurate proxy of VOC concentrations within groundwater at monitoring well locations. These hypotheses and the utility of vapor-phase methods for groundwater monitoring were tested as part of research project funded by the Strategic Environmental Research and Development Program (SERDP). First, a laboratory screening study was completed that successfully validated the performance of two types of field equipment for monitoring vapor concentrations, with high degree of accuracy and precision at levels below typical groundwater maximum concentration levels (MCLs) (Adamson et al. 2009). Furthermore, analysis of headspace in equilibrium with water containing trace levels of volatile contaminants achieved accuracy objectives for 92% measurements and precision objectives for 80% measurements. This demonstrated that these instruments can be used to measure VOC concentrations in water with sufficient accuracy, precision, and sensitivity to achieve typical groundwater monitoring objectives. Finally, three methods for collecting vapor samples were examined during the laboratory study, with direct headspace sampling and diffusion-based sampling retained for further field testing. On the basis of these results, three different sampling configurations were included in the first phase of field testing: 1. Upper well headspace. This represents the simplest method for collecting and analyzing vapor samples. A sampling tube is placed through a modified well cap, allowing for direct surface sampling of vapor from the upper portion of the well headspace. Field personnel can then generate an estimate of the groundwater concentration from analysis of the vapor sample using a suitable field instrument. Factors that contribute to accuracy include the presence of a suitable impermeable well seal, a uniformly mixed air column above the water level, and in-well mixing to reduce the impact of vertical stratification within the water column (Elci et al. 2001; Britt 2005). Britt (2005) concluded that in-well mixing appears to be the rule rather than the exception in homogeneous flow fields, a condition that would greatly enhance the utility of this sampling method in determining groundwater concentrations. 2. Water-vapor interface. The objective of collecting a second headspace sampling at the interface was to determine if poor mixing and/or limited diffusion within the air column impacts accuracy, such that interface samples would provide a more representative method of determining equilibrium groundwater concentrations. The well cap configuration is identical to the upper headspace sampling method, with the exception that a sampling tube terminates as close as possible to the water-vapor interface. As such, this approach is slightly more involved because the tubing length must be designed to account 60 D.T. Adamson et al./ Ground Water Monitoring & Remediation 32, no. 1: NGWA.org

3 for the depth to water in each individual well, as well as typical water level fluctuations. 3. Passive vapor diffusion (PVD) sampler at screen. A simple passive method for collecting vapor samples was envisioned for wells where excessive stratification limited the utility of direct headspace sampling for determining groundwater concentrations. PVD samplers for estimating water concentrations were originally developed by the USGS (2002), and we successfully tested a similar design (using a rigid vial wrapped in layers of gas-permeable membrane) during the laboratory study (Adamson et al. 2009). The notable difference is that the current project aims to deploy these passive samplers within a monitoring well at the screen to determine groundwater concentrations at equilibrium, as opposed to the USGS s focus on the sediment-surface water interface. When submerged below the water level in a monitoring well, equilibrium between the well water and the vapor in the sampler is achieved within several weeks. While this requirement for extended deployment is more complicated than the direct headspace methods, installation of passive samplers can typically be piggy-backed on the previous round of monitoring, eliminating the need for an extra mobilization. As mentioned earlier, for those wells where there is evidence of vertical stratification (i.e., poor in-well mixing), a vapor sample from the well headspace (Method 1) may not be representative of groundwater concentrations. If a correlation is to be attempted in these cases, attention must be paid to the vertical location where the corresponding groundwater sample is collected. For a low-flow groundwater sample that is typically collected near the center of the well screen, a vapor sample from the same vertical location would likely yield more comparable results. In these cases, passive methods for collecting vapor samples may be a better option for determining the groundwater concentration. This article presents results from the first phase of a field program that is investigating the use of vapor-phase methods as a more sustainable, cost-effective, and faster approach for LTM programs. The objectives of this portion of the study were to test the performance of the fieldportable instruments and several proposed sample collection methods under field conditions, and in the process, identify conditions that affect equilibrium partitioning and contribute to sampling variability. For this task, data collected using vapor-phase-based methods were evaluated in terms of accuracy relative to more conventional groundwater sampling and analyses methods. The field program consisted of the collection of a series of vapor and groundwater samples from a set of existing monitoring wells, followed by analysis of data to establish relevant correlations between each of the sampling and analysis methods. Materials and Methods Analytical Equipment Two types of field instruments were selected for this project based on functionality and cost: (1) gas chromatograph (GC): Voyager (Photovac Inc., Waltham, Massachusetts) equipped with a 10.6 ev photoionization lamp detector and an electron capture detector (ECD) and (2) photoionization detector (PID): ppbrae 3000 (RAE Systems Inc., San Jose, California) equipped with a 10.6 ev photoionization lamp detector. The PID is relatively inexpensive, and its widespread use in environmental monitoring means that it is familiar to most field personnel. While PIDs are generally sensitive at low vapor concentrations, they are capable of measuring only the total concentration of VOCs and have limited applicability for sites where multiple contaminants are present. The field-portable GC represents a higher cost investment (both in terms of equipment and training), but has the distinct advantage of quantifying individual compounds with a mixed vapor stream. Because the GC used during this project was equipped with multiple detectors, it was ideal for monitoring several different contaminant classes (e.g., chlorinated ethenes, chlorinated ethanes, chlorinated methanes, aromatics). The results of the laboratory evaluation of instrument performance (Adamson et al. 2009) showed that the GC and PID achieved the project criteria for accuracy, precision, and sensitivity. The ppbrae 3000 PID achieved the accuracy and precision criteria for 100% measurements. The instrument method detection limit (MDL) corresponds to a water-phase concentration of 1.3 mg/l benzene, less than the MCL of 5 mg/l. The Voyager GCs achieved the accuracy criteria for 94% measurements and the precision criteria for 100% measurements. For each of the three VOCs tested (1,1- dichloroethene, benzene, and vinyl chloride), the instrument MDL corresponded to a water-phase concentration of less than 0.5 μg/l which was less than the MCL of 5 to 7 μg/l. Sampling Methods Six different sampling methods were included as part of the field program. These methods are shown in Figure 1 and described in the following paragraph. To collect water and vapor samples, special well caps were designed and installed in advance of the first monitoring event. These were constructed by project personnel using modified compression caps and tested prior to use to ensure the seal integrity. 1. Upper well headspace. A vapor sample was collected from a port that terminated immediately below the sealed well cap. The port consisted of a short piece of rigid tubing (1/8-inch diameter Nylaflow, Professional Plastics, Inc., Fullerton, California) that was installed through a small hole drilled in the cap and then held in place using silicon caulking. The port exited the top of the cap and was connected to a four-way valve (polycarbonate stopcocks with Luer connections) via flexible tubing. Vapor samples were collected with a disposable plastic syringe (60 ml), transferred to a Tedlar bag and allowed to equilibrate several minutes prior to analysis. Replicate 100-μL vapor samples were analyzed with the portable-field GC. 2. Water-vapor interface. A second vapor sample was collected from a pre-installed port that extended to a depth 1 foot above the water-vapor interface. Sample collection and analysis was conducted in the same fashion as for the upper well headspace samples. NGWA.org D.T. Adamson et al./ Ground Water Monitoring & Remediation 32, no. 1:

4 Water / vapor interface vapor sampling port Groundwater level Passive diffusion bag (PDB) Passive vapor diffusion sampler (PVD) Valve Upper headspace vapor sampling port String 3 5 Vapor-Phase Collection 1 Upper Well Headspace 2 Headspace at Water-Vapor Interface 3 Passive Vapor Diffusion (PVD) sampler at screen Groundwater-Phase Collection Passive Diffusion Bag (PDB) at Water- Vapor Interface Passive Diffusion Bag (PDB) at screen Low-flow at screen (basis for comparison) Well screen Weight at bottom Figure 1. Sampling methods tested during preliminary field program. Schematic displays six sampling methods. 3. PVD sampler at screen. The final vapor sampling method involved placing a 40-mL VOA vial (sealed in two layers of gas-permeable low-density polyethylene (LDPE) membrane) at the center of the screened interval by attaching it to a support string extending from the well cap. Following submersion below the water surface, this configuration prevented water from entering the vial but allowed passive diffusion of vapor-phase contaminants across the membrane. Extended deployment within the wells allowed for equilibration to occur; after 3 weeks the samplers were retrieved and capped with polytetrafluoroethylene (PTFE)/Silicone septa caps. 100-μL vapor samples were removed from the capped PVDs with a gastight syringe (Hamilton, Co., Reno, Nevada) and analyzed for organic constituents with the portable-field GC. PVD samplers used during the field studies were constructed according to previous guidance documents (USGS 2002). 4. PDB at water-vapor interface. PDBs consisting of 24-inch long and 1.25-inch diameter sealed LDPE bags filled with deionized water were installed at the watervapor interface using a string that connected the top of the bag with a loop on the underside of the well cap. Using static water level data collected prior to deployment, string lengths for individual wells were chosen that ensured complete water submersion of bag samplers. Similar to the PVD samplers, diffusion of organic contaminants across the gas-permeable membrane permits equilibration of the concentration in the monitoring well with the concentration of the water inside the sampler. Following retrieval of the PDB samplers from monitoring wells, water was transferred to 40-mL VOA vials using a screwcap located at the top of the bags. Vials were then shipped to a commercial laboratory for analysis. PDB samples from the interface were collected in an attempt to demonstrate potential impacts of vertical stratification on correlations between headspace and groundwater samples. 5. PDB at screen. PDB bags were also installed within the screened interval using the same stringing technique with a weight attached to the bottom to minimize movement. Samples were collected in a manner identical to that described previously, and then shipped for off-site analysis at a commercial laboratory. Because these samples were collected from the middle of the screened interval, they were intended to provide a direct basis for comparison to lowflow groundwater samples and correlated PVD samples. 6. Low-flow at screen. Low-flow groundwater samples were collected at all wells to provide a baseline groundwater concentration for comparison to those obtained using vapor-phase-based methods and passive groundwater sampling. This conventional sampling method used a peristaltic pump to draw small volumes of groundwater through tubing that terminated at the center of the screened interval. Pumping was continued until field measurements of geochemical parameters (i.e., electrical conductivity, ph, temperature) stabilized. Water samples were then collected into 40-mL VOA vials for off-site laboratory analysis. In addition to the three vapor sampling methods and three water sampling methods described above, a supplemental vapor analysis method was employed that involved transferring 20-mL water samples (collected using either low-flow or PDBs) to a 40-mL VOA vial. Capped vials were mixed vigorously by hand, allowed to equilibrate for approximately 1 h, and then the headspace was analyzed using the field GC. This method was designed to eliminate potential variability introduced by collecting a vapor sample from the well, while still rapidly generating a groundwater concentration through a combination of field vapor analyses and equilibrium calculations. Sampling and Analysis Program The field monitoring program was conducted in 10 monitoring wells in the Houston area over the course of approximately 6 weeks (Table 1). Well construction details are provided in Table S1 (Supporting Information). At each site, well materials (caps, passive sampling devices) were installed in a single mobilization in December 2009 at the onset of the study. A monitoring event was then completed in January 2010, approximately 3 weeks after the installation event, a time interval that is appropriate for passive diffusion samplers based on technology guidance documents. At the conclusion of the first monitoring event, well materials (i.e., passive samplers) were 62 D.T. Adamson et al./ Ground Water Monitoring & Remediation 32, no. 1: NGWA.org

5 Table 1 Sampling Events for Preliminary Field Program Parameter Number of sites 3 Number of wells 10 Number of sampling events 2 (January 2010, February 2010) Frequency of sampling events Every 3 weeks replaced and allowed to equilibrate for a period of 3 weeks, after which the second monitoring event was conducted (February 2010). To minimize the effect of sample collection on the sample results, the samples were collected from each well in sequential order with the samples most likely to be affected by short-term mixing/disturbance collected first. The sample collection order was as follows: 1. Vapor analysis of well headspace samples from top of well. 2. Vapor analysis of well headspace samples from close to water table. 3. Removal of passive diffusion samplers from monitoring well: î transfer of water from PDBs to 40-mL VOA vials for laboratory analysis. î field analysis of PVD vapor samples. 4. Collection of low-flow water samples for laboratory analysis. 5. Headspace analysis of vapors in equilibrium with water samples in 40-mL VOA vials. A total of 180 samples were collected and 364 analyses were completed as part of the field program, including field duplicates and replicate analyses (Table 2). For the six primary methods, the objective was to collect 20 individual samples per method. Several factors prevented this objective from being achieved, including: (1) one well was inadvertently opened prior to the start of the first sampling event, disturbing equilibrium conditions; (2) one or more pieces of equipment were compromised (e.g., obstructed tubing, pump failure) in several wells during the first sampling event; and (3) the water level rose above the depth where the tubing for collecting a vapor sample terminated in three wells. Data Analysis Vapor-phase concentrations measured in the field were converted to equivalent groundwater concentrations using Henry s law and the groundwater temperature measured at the well screen at the time of sampling (Staudinger and Roberts 2001). Equilibrium partitioning methods were deemed appropriate for estimating groundwater concentration based on a laboratory-based study successfully validating these methods and showing that equilibrium (Adamson et al. 2009). However, it is recognized that dynamic conditions within a well likely preclude true equilibrium. Note that the Henry s law constant is temperature-dependent, and Table 2 Sampling and Analysis Plan for Preliminary Field Program Sample Type (Location) 1. Well headspace (upper) 2. Well headspace (interface) 3. Passive vapor diffusion (screen) 4. Passive diffusion bag (interface) 5. Passive diffusion bag (screen) 6. Lowflow water (screen) Matrix Sampled/Matrix Analyzed Field or Laboratory Analysis Number of Samples Collected 2 Vapor/vapor Field 18 (GC); (PID) 16 Vapor/vapor Laboratory 10 Vapor/vapor Field 16 (GC); (PID) 16 Vapor/vapor Laboratory 7 Vapor/vapor Field 18 Water/water Water/vapor 1 Water/water Water/vapor 1 Water/water Water/vapor 1 Laboratory Field Laboratory Field Laboratory Field Water sample was collected using designated sampling method (either low-flow, passive diffusion bag at well screen, or passive diffusion bag at water/vapor interface). Field analysis of sample headspace following rapid induction of equilibrium partitioning in partially water-filled vial. 2 Total does not include replicate samples analyzed in field or laboratory. For vapor samples collected for field analysis, duplicate or triplicate analyses were completed for all samples. For vapor or groundwater samples collected for laboratory analysis, field duplicate samples were collected at a minimum frequency of one duplicate sample for every 10 field samples. there is a potential for the temperature at the water-vapor interface to be slightly different than the temperature at the well screen. Temperature at the water-vapor interface was not collected at all wells, but at two wells where it was measured, it differed by 1.3 to 3.7 C from the temperature at the well screen during this sampling period (data not shown). While this introduced a slight amount of low bias to the groundwater concentrations estimated from well headspace samples, no temperature correction was attempted because temperature data were not available for all wells. Data sets consisted of groundwater concentrations measured or estimated using one of the sampling and analysis methods from the entire population of monitoring wells. A test for normality (Anderson-Darling) was performed on all data sets. In all cases, concentration data spanned several orders of magnitude, and the results of these tests confirmed expectations that they did not represent normal distributions To improve the normality of this data sets and thus improve the power of the statistical methods used to evaluate the data (USEPA 2009) two different types of transformations were attempted (log-transformation and normalization to low-flow groundwater concentrations), and the Anderson-Darling test was re-run on the transformed NGWA.org D.T. Adamson et al./ Ground Water Monitoring & Remediation 32, no. 1:

6 Log Groundwater Concentration Calculated Using PVD (µg/l) y = 0.98x R 2 = Log Groundwater Concentration Calculated Using Headspace-Interface (µg/l) y = 0.76x R 2 = Log Groundwater Concentration Measured Log Groundwater Concentration Measured (a) Using Low-Flow (µg/l) (b) Using Low-Flow (µg/l) (c) y = 0.69x R 2 = 0.60 Figure 2. (a) PVD samplers at well screen vs. low-flow groundwater samples from well screen. (b) Headspace samples from watervapor interface vs. low-flow groundwater samples at well screen. (c) Headspace samples from upper portion of well vs. low-flow groundwater samples from well screen. All field analyses were completed with field-portable GC (PID data not shown). Log Groundwater Concentration Calculated Using Headspace-Upper (µg/l) 6.0 Log Groundwater Concentration Measured Using Low-Flow (µg/l) data (α = 5). On the basis of the results of these tests, log-transformed data were used in further evaluation of the data. Comparisons were performed between groundwater concentration data calculated using field measurements of equilibrium vapor samples and groundwater concentration data measured using laboratory analysis of (1) low-flow groundwater samples and (2) groundwater samples collected using PDBs. Results of these comparisons are presented for the following evaluations: 1. Linear regression analysis was the primary means for examining relationships between the between logtransformed concentration data sets generated from the same population (i.e., the same set of wells). The R 2 value (or the squared correlation coefficient) serves as an indicator of the goodness-of-fit, and thus was used to demonstrate variability. The slope of the regression line was used to evaluate a predictive relationship between data collected from the same set of wells but using different methods. A slope near 1 indicated that data sets are similar, while a slope of less than 1 indicated underprediction or low bias relative to the baseline case (e.g., groundwater concentration from a low-flow sample), and a slope greater than 1 suggested overprediction or high bias. 2. Two-sample tests (parametric and nonparametric) to determine if there is a statistically significant difference between the means of the low-flow groundwater data and the groundwater data calculated using the vapor-phasebased methods. The nonparameteric (Wilcoxon ranksum test) and parametric (paired t test) methods both used log-transformed data with α = 5 (i.e., significance level of 95%). 3. RPD is the relative percent difference between individual data pairs (e.g., low-flow vs. vapor-phase-based concentration). For paired measurements, the RPD is the ratio of the difference between the two values to the average of the two values. It can serve as an indicator of the accuracy of a measured value by comparing it to a value which is expected to be equivalent. Results and Discussion PVD Samplers PVD samplers were installed at the screened interval for the monitoring well, at approximately the same depth interval where low-flow groundwater samples were collected. All vapor measurements were completed using the field GC because insufficient sample volume was available for PID analysis. Collectively, the PVD data (Figure 1, location 3) correlate well with the low-flow groundwater samples analyzed with the field GC (Figure 1, location 6), with no bias (slope = 0.98) and only moderate variability (R 2 = 0.85) (Figure 2a). Because the water and vapor samples were collected from the same vertical location in each well, there was unlikely to be any influence from in-well factors such as stratification. PDBs installed at the well screen (Figure 1, location 5) were also used to collect groundwater samples that were sent for off-site laboratory analysis, and a similar comparison was made to measured concentrations from low-flow groundwater samples (Figure 1, location 6). This comparison yielded similar results (slope = 0.96; R 2 = 0.85) to those obtained when PVD concentrations were compared to low-flow data. In fact, a simple linear regression between the PVD and PDB data demonstrate the strong correlation between these two data sets (slope = 3; R 2 = 0.96) (Figure 3a). The results of this comparison confirm that the two passive methods for collecting groundwater concentration (Figure 1, locations 3 and 5) generate statistically similar data sets (Table 5), regardless of which medium is sampled and what analysis method (field vs. laboratory) is employed. Consequently, the variability encountered when trying to use these alternate methods to match low-flow groundwater concentrations is at least partly attributable to differences between passive and low-flow methods for collecting groundwater, as opposed to problems with collecting consistent vapor samples or accurate field measurements. It is important to note that a single outlying data point in one sample at one well contributed significantly to the observed variability between PVD and low-flow 64 D.T. Adamson et al./ Ground Water Monitoring & Remediation 32, no. 1: NGWA.org

7 Log Groundwater Concentration Calculated Using PVD (µg/l) (a) 6.0 y = 3x R 2 = Log Groundwater Concentration Measured Using PDB-Screen (µg/l) Log Groundwater Concentration Calculated Using Headspace-Upper (µg/l) (b) 6.0 y = 0.95x R 2 = Log Groundwater Concentration Calculated Using Headspace-Interface (µg/l) Figure 3. (a) PVD samplers vs. passive diffusion bags at screen. (b) Headspace samples from water-vapor interface vs. passive diffusion bags at water-vapor interface. All field analyses were completed with field-portable GC. groundwater data (Figure 2a). For example, with this point omitted from the PVD data, the R 2 value for the regression with low-flow data improved to 0.96 without significantly affecting the slope (1). The concentration obtained from this well was biased low (relative to the low-flow sample) for all vapor-phase and passive methods, indicating that the sampling methods (rather than the analysis methods) were responsible for the large difference. Since an assessment of differences related to sampling methods was part of this study, the decision was made to include this data point in all evaluations and not omit it as an outlier. Headspace Sample Near the Water-Vapor Interface in Well Headspace samples were collected from the water-vapor interface (Figure 1, location 2) to determine the extent to which equilibrium vapor samples from this location could be correlated to low-flow groundwater samples collected from the screen. Vapor measurements were completed using the field GC and the PID. Groundwater concentration data calculated using the headspace sampling method and analysis with the field GC correlated relatively poorly with the low-flow data (Figure 1, location 6), showing a strong low bias (slope = 0.76) and high variability (R 2 = 0.64) (Figure 2b). Two-sample tests indicated that the means were significantly different, such that the populations could not be considered equivalent. The data obtained using PID measurements were considerably worse than those obtained using the field GC (slope = 0.59, R 2 = 0.40). In general, the water column was extended above the screened interval in this set of monitoring wells. As such, a portion of the bias may have been an artifact of lower temperatures (and thus lower Henry s constants) at the interface relative to the screened interval. The low bias and high variability also suggest that this water column may have been stagnant in many of the wells and at a lower concentration than the water at the screened interval. Consequently, vapor samples collected from the well headspace were more closely in equilibrium with water that was collected from the water-vapor interface, and data showed that this water was not particularly representative of the water collected from the well screen. PDBs installed at the water-vapor interface (Figure 1, location 4) were also used to collect groundwater samples that were analyzed at an off-site laboratory, and a similar comparison was made to measured concentrations from low-flow groundwater samples (Figure 1, location 6). This comparison yielded slightly better results than those obtained when headspace-based concentrations were used, but the correlation remained relatively poor (slope = 0.82, R 2 = 0.60) with two-sample tests demonstrating that the two methods yielded statistically different data (Table 5). Considerable variability was observed in both data sets, suggesting an apparent difficulty in obtaining a consistent sample from the water-vapor interface in a monitoring well. This is further illustrated by a direct comparison between concentrations from the PDB at the interface (Figure 1, location 4) and concentrations calculated using the headspace sampled at the interface (Figure 3b). There was less bias between the two data sets (slope = 0.87), as would be expected from their similar sampling location near the interface. The two-sample tests indicated that there was no statistically significant difference between the datsets (Table 5). However, the R 2 value remained low (0.60), confirming that sampling near the interface introduced a considerable amount of variability into the monitoring data. Upper Well Headspace Sample Headspace samples were collected from the upper portion of the well to determine the extent to which equilibrium vapor samples in this location could be correlated to lowflow groundwater samples collected from the screen. Vapor measurements were completed using the field GC and the PID. As with the headspace-interface data set, the groundwater concentration data calculated using the upper well headspace (Figure 1, location 1) sampling method correlated relatively poorly with the low-flow data (Figure 1, location 6). Using the field GC, the vapor-phase measurements were consistently biased low (slope = 0.69) with high variability (R 2 = 0.60) (Figure 2c), and two-sample tests indicated that the data set means were significantly different (Table 5). The data obtained using PID measurements (slope = 0.57, R 2 = 0.34) were again worse than those NGWA.org D.T. Adamson et al./ Ground Water Monitoring & Remediation 32, no. 1:

8 Log Groundwater Concentration Calculated Using Headspace-Upper (µg/l) 6.0 y = 0.95x R 2 = Log Groundwater Concentration Calculated Using Headspace-Interface (µg/l) Figure 4. Headspace samples from upper portion of well vs. headspace samples from water-vapor interface (GC analysis). Log Groundwater Concentration Calculated Using Field Vapor Analysis of Equilibrium Groundwater (µg/l) 6.0 y = 0.94x R 2 = Log Groundwater Concentration Measured Using Lab Groundwater Analysis (µg/l) Figure 5. Field GC analysis of vapor in equilibrium with groundwater samples vs. laboratory analysis of groundwater samples. obtained using the field GC. A portion of the bias may be attributable to slightly lower temperatures at the interface relative to the screened interval, which would result in underpredictions of groundwater concentration using equilibrium partitioning. The results indicate that the vapor in the upper well headspace was not in equilibrium with water collected for low-flow sampling. However, there was a clear consistency between the vapor data collected from the upper headspace (Figure 1, location 1) and the interface (Figure 1, location 2), as demonstrated by the strong correlation between these two data sets following a linear regression (slope = 0.95, R 2 = 0.97) (Figure 4). This suggests that mixing and air-phase diffusion within the headspace results in relatively uniform conditions within the headspace, at least within the time frame of this sampling program. Consequently, the location where the vapor sample is collected in the air-filled column above the water table does not appear to be an important contributor to variability. Field Analysis of Groundwater Samples An alternate method for determining groundwater concentrations was investigated at select locations by placing a water sample from the well in a sealed vial containing a headspace and agitating the sample for a sufficient period of time to achieve equilibrium partitioning. The field GC was used to analyze the vapor in the headspace of the vial, with the result then converted to a VOC concentration in the water sample. This method was employed for all three of the water sample collection methods: (1) low-flow groundwater samples (Figure 1, location 6); (2) PDBs installed at the screen (Figure 1, location 5); and (3) PDBs installed at the watervapor interface (Figure 1, location 4). The groundwater concentrations calculated using the vapor-phase field measurements were then compared to the concentrations measured when the corresponding groundwater samples were analyzed off-site at a commercial laboratory. The results were consistent for all three sample collection methods and confirmed a strong correlation between field and laboratory analyses, even though a different medium was being analyzed in each case (Figure 5). The slope of the regression line was approximately 0.94, indicating that the field analyses of vapor slightly underpredicted the groundwater concentration. This slight bias may be attributable to an insufficient time for equilibration following transfer of the groundwater samples to the containers. The effect of extending the equilibration time beyond 60 min was not tested. A different trend was observed at low concentrations, where the field vapor measurements slightly overpredicted relative to the laboratory groundwater measurements. Regardless, this appears to be a relatively accurate method for obtaining depth-discrete groundwater data, especially at higher concentrations. It is easy and rapid alternative to low-flow groundwater sampling as it eliminates the wait for laboratory results. Furthermore, the results emphasize that factors related to field analyses are not the sole contributors, or even the major contributors, to the variability observed when trying to match vapor-phasebased groundwater concentrations with low-flow groundwater concentrations. Evaluation of Precision and Accuracy for Field vs. Laboratory Analyses In addition to the p revious evaluations, several other methods were employed to investigate the precision and accuracy of the various sampling and analyses methods. Laboratory and Field Analyses of Replicate Samples Both groundwater and vapor duplicate samples were collected for analysis at (separate) commercial laboratories. A small set of duplicate vapor samples were also collected for field analysis. For each set of duplicates, the relative standard deviation (RSD) was calculated as a metric for assessing precision (Table 3). Although the sample set was relatively small, the level of precision for field analyses of duplicates was equal to or better than that for laboratory analyses. Note that the RSD values in Table 3 reflect variability associated with the sampling steps as well as the analysis steps. 66 D.T. Adamson et al./ Ground Water Monitoring & Remediation 32, no. 1: NGWA.org

9 Analysis Type Groundwater (laboratory) Vapor (laboratory) Table 3 Precision of Laboratory vs. Field Analyses of Duplicate Samples No. of Duplicate Sample Sets RSD (%) Range Mean Vapor (Field) Does not include two additional duplicate sample sets where concentration was below laboratory reporting limit. Table 4 Precision of Replicate Field Analyses of All Samples Analysis No. of Replicate RSD (%) Type Analysis Sets Range Mean Median Vapor (field) Replicate Field Analyses of Vapor Samples Replicate analyses of all vapor samples were completed in the field to provide a more focused assessment of the precision of the equipment under field conditions. The data in Table 4 represent the RSD values calculated from duplicate or triplicate analyses using the field GC (note that insufficient sample volume was available to complete replicate analyses with the PID). Because of the large sample size, the RSD values for this set of measurements are likely to be a more representative indicator of the precision of the field instruments. Greater than 90% of the RSD values were less than 30%, and the median RSD value for the instrument under field conditions (7.9%) was only slightly higher than the median RSD value for the same instrument during the laboratory validation study (%). There was no evidence that any particular sample type (e.g., upper headspace sample, PVD) contributed to higher RSD values following replicate analyses. Field GC vs. PID Analyses of Vapor Samples For those wells where vapor samples were analyzed by both the field GC and the PID, the resulting data were used to determine potential bias in either of the field instruments. For each instrument, all data were lumped together, regardless of sample location. Using this bulk comparison, it appeared that the PID provided a reasonable correlation with field GC measurements at low vapor concentrations, but that the correlation became poorer at higher concentrations. This is a function of the relatively low upper detection limit for the PID, such that high vapor concentrations are difficult to measure with this device. As a result of this limitation, considerable variability was observed (R 2 = 0.30) between the data collected using the two analytical instruments (Table 5). These results are generally consistent with those described previously when field GC and PID measurements were compared to low-flow groundwater concentrations. Specifically, the PID is less capable of generating an unbiased estimate of the groundwater concentration (especially at high concentrations), and the variability introduced limits its utility. However, most of these devices have a relatively low purchase price and are extremely easy to use, such that they may have value in screeninglevel applications where a less precise measurement is required or where high-concentration dilution techniques are employed. Field vs. Laboratory Analyses of Vapor Samples For a selected set of vapor samples, duplicate samples were sent to a commercial laboratory to assess consistency between laboratory and field analyses. Vapor samples were collected from both the water-vapor interface and the upper portion of the well. Data from both well locations were combined, and the laboratory measurements were compared with those obtained from analyses using the field GC and the PID. Regardless of the instrument used, considerable variability was observed between laboratory and field analyses (Table 5). This was a more pronounced problem for field analyses with the PID. The slopes of the regression lines indicated that field analyses slightly overpredict concentrations reported by the commercial laboratory (Figure 6). The magnitude of this bias was such that a statistically significant difference between the data sets was established using both the parametric and nonparametric two-sample tests (Table 5). The variability could be attributable to a variety of factors, and is most likely a combination of variability in precision of the field instruments and precision of the laboratory analyses. However, based on the data presented for field duplicates/replicate analyses, the contributions from these two factors would not be expected to cause the magnitude of variability observed in the data displayed in Figure 6. One factor that did not appear to contribute was the location where the vapor sample was collected; laboratory analyses of headspace samples from the same well yielded very similar results regardless of the depth where the sample was collected (slope = 6; R 2 = 0.99) (Table 5). The pattern is consistent for both laboratory and field analyses (Figure 4) of vapor samples from the two well headspace locations. Factors Contributing to Bias and Variability The data from the field program were sufficient to investigate a number of factors that may have contributed to bias and variability between data sets. A summary of relevant correlation analyses are presented in Figure 7. Vertical stratification within individual monitoring wells was a clear contributor to variability. Concentrations within the well water column did not appear to be uniform with depth within the majority of locations included in this program, as evidenced by the differences between samples collected from the well screen and the water-vapor interface NGWA.org D.T. Adamson et al./ Ground Water Monitoring & Remediation 32, no. 1:

10 Sample Set 1,2 Phase Sampled Table 5 Summary of Data Evaluation for All Sampling Methods Phase Analyzed Sample Set Compared to: 3 Linear Regression Relative Percent Difference (%) Median (Nondirectional) Median (Directional) Statistically Different? (p value) 4 Nonparametric (Wilcoxon Rank-Sum Test) 5 Parametric (Paired t Test) 6 Comparison to low-flow samples PVD Vapor Vapor Low-flow No (p = 0.47) No (p = 0.33) Headspace-interface (GC) Vapor Vapor Low-flow Yes (p = 16) Yes (p = 16) Headspace-interface (PID) Vapor Vapor Low-flow Yes (p = 14) Yes (p = 09) Headspace-upper (GC) Vapor Vapor Low-flow Yes (p = 02) Yes (p = 06) Headspace-upper (PID) Vapor Vapor Low-flow Yes (p = 098) Yes (p = 098) PDB at screen Water Water Low-flow No (p = 5) No (p = 0.12) PDB at interface Water Water Low-flow Yes (p = 19) Yes (p = 49) Field equilibration of low-flow and PDB water Water Vapor Low-flow Yes (p = 36) No (p = 0.36) Comparison between other sampling methods PVD Vapor Vapor PDB at screen No (p = 0.15) No (p = 0.11) PDB at interface Water Water PDB at screen Yes (p = 1) No (p = 0.11) Headspace-interface (GC) Vapor Vapor PDB at interface No (p = 0.62) No (p = 0.93) Headspace-interface (PID) Vapor Vapor PDB at interface No (p = 0.52) No (p = 0.60) Headspace-upper (GC) Vapor Vapor Headspace-interface (GC) Headspace-upper (PID) Vapor Vapor Headspace-interface (PID) No (p = 0.23) No (p = 0.10) No (p = 0.54) No (p = 0.31) Comparison between analytical methods All headspace (PID) Vapor Vapor All headspace (GC) No (p = 0.56) No (p = 0.93) All headspace (Lab Analysis) Vapor Vapor All headspace (GC) Yes (p = 03) Yes (p = 37) All headspace (laboratory analysis) Vapor Vapor All headspace (PID) Yes (p = 05) Yes (p = 09) Headspace-upper (laboratory analysis) Vapor Vapor Headspace-interface (laboratory analysis) No (p = 63) No (p = 0.13) 1 All data represent measured or calculated groundwater concentrations from a field program conducted in January 2010 and February Groundwater concentrations were either groundwater samples sent for analysis at a commercial laboratory or vapor samples analyzed in the field (using a field GC or PID) and converted to groundwater concentrations (in mg/l). 3 Comparisons include only for the primary constituent (either trichloroethene or vinyl chloride) in each monitoring well. 4 Statistical comparisons included data on wells where selected sampling methods were employed and all of the selected analyses yielded a non-detect value. 5 Nonparametric test: Wilcoxon rank-sum test using log-normalized data from specified methods (a = 5). 6 Parametric test: paired t test on mean of log-normalized data from specified methods (a = 5). 68 D.T. Adamson et al./ Ground Water Monitoring & Remediation 32, no. 1: NGWA.org