Methods


Site Selection and Sampling Frequency


Based on observed patterns of pesticide occurrence from other Oregon studies (Anderson and others, 1996, 1997; Rinella and Janet, 1997; Wood, 2001; Waite and others, 2006), pesticide surveys were conducted approximately twice yearly during storm runoff in fall and spring, beginning in the fall 2002 through spring 2010. Twenty-four sites that represent one predominant upstream land-use type as determined by EWEB personnel (urban, agricultural, or forestry) were selected from among a total of 28 sites in the mainstem McKenzie River and its tributaries downstream of inflow from the Blue River (table 1). Four additional mixed land-use sites include the mainstem McKenzie River sites at the Hendricks Park boat ramp and above Hayden Bridge (the intake of raw water to the drinking water treatment facility), and sites near the mouths of major tributaries Camp Creek and Cedar Creek that flow into the mainstem of the river in the lower basin. The Hendricks Park site is downstream of EWEB’s Walterville Power Canal diversion. Land use in the vicinity includes residential homes along the river, the small community of Walterville, agricultural crops in the valley, and commercial forestlands in the surrounding hills. As such, the Hendricks Park site is representative of the influence of both forestry and agricultural land-use activities, with less impact from urban areas that are more pronounced downstream. The site on Camp Creek represents a small watershed that also is predominantly a mixture of forested and agricultural land use. Sampling sites on Cedar Creek at Springfield and the McKenzie River above Hayden Bridge incorporate a larger amount of urban area and thereby represent a mix of all land-use categories. 


No attempt was made to sample different site types equivalently; rather, sampling was designed primarily to characterize representative storm events during spring and fall runoff conditions in order to assess or confirm the perceived importance of the different site types as sources for pesticides. Sites were selected for sampling in a non-fixed design that included both explicitly stormwater/urban and forestry surveys, as well as other surveys focused on a mix of land-use types (table 2). Additionally, winter storm conditions were sampled in 2003 and 2009, and baseline or non-storm samples were collected in spring and summer 2009, and in spring 2010, to investigate the importance of non-runoff periods for pesticide occurrence.


The mixed sites formed the backbone of the sampling strategy, and were sampled most consistently throughout the study period because they represented the largest potential loading to the drinking water source. Because urbanization was perceived as an important degrading influence on drinking water quality, the first five storm surveys in 2002-05 focused primarily on urban sites in addition to the mixed sites. To evaluate the potential influence of forestry management, these urban surveys were followed by two storm surveys in the fall of 2005 and 2006 that focused primarily on forestry sites. Sampling intensity was subsequently reduced to focus only on the mixed sites in 2008, partly as a result of poor runoff conditions, and then expanded to include urban sites in fall 2009, and a comprehensive suite of agricultural, forestry, urban, and mixed sites in the spring of 2009 and 2010. Despite the irregular sampling frequency, the number of non-replicate samples collected to characterize each land use was roughly equivalent for forestry sites (n = 33) and urban sites (n = 35), with slightly more samples associated with mixed sites (n = 45); agricultural sites were relatively undersampled (n = 5). More forestry samples were collected in the fall (n = 22) than in the spring (n = 10), although samples were more evenly distributed for urban sites (n = 16 for fall, n = 11 for spring) and mixed sites (n = 18 for fall, n = 15 for spring). Agricultural sites were only sampled in the spring. Winter storm samples included forestry (n = 1), mixed (n = 5), and urban (n = 6) sites, and non-storm samples were collected at urban (n = 1) and mixed (n = 7) sites.


As a consequence of the reconnaissance nature of the sampling, and the unpredictable nature of storm sampling in general, there are large differences in the number of samples collected among the sampled sites (table 1). The EWEB intake at the McKenzie River above Hayden Bridge (site 5) was sampled the most frequently. Cedar Creek at Springfield (site 9) and Camp Creek (site 14) also were sampled more frequently because they represent the largest tributaries in the lower basin. Urban surveys show a good mix of coverage throughout the year, with each site sampled from four to eight times. Forestry sampling was more inconsistent because sites were selected based on forestry spray information, which reflected irregular and intermittent application patterns among the different tributaries. This meant that each individual forestry site generally was sampled only once or twice, after being identified as a target for spraying upstream. Spring forestry surveys were not conducted consistently because of the difficulty in capturing runoff events in the spring following application. Finally, agricultural surveys were not conducted until near the end of the reconnaissance period because site selection was complicated by logistics and lack of access to private lands. 


Data Collection


Ancillary Data


Data evaluated in this analysis include not only pesticide concentrations, but also various ancillary data that were collected to provide context for understanding the patterns of pesticide occurrence. These include land-use data, precipitation data, discharge data, data describing physical and chemical characteristics of the compound, health-based screening criteria, and potential for endocrine disruption for detected pesticide compounds. Additionally, reported or estimated pesticide-use data were acquired to evaluate the detection of compounds relative to their presumed use.


All drainage basin delineations and land-use analyses were conducted by the Lane Council of Governments (LCOG) for EWEB. Land-use data were associated with target drainage basins by intersecting Geographic Information System (GIS) layers with zoning, land use, and land-cover data (David Richey, LCOG, written comm., November 18, 2010). Zoning and land-use data were from 2010 and land-cover data were from the 2001 National Land Cover Database (Homer and others, 2004). Each of these data sources were summarized by forestry, agricultural, and urban/residential categories, and mean values for selected drainage basins across the range of data sources are presented in this report.


Hourly precipitation data acquired for two precipitation gages provide complementary perspectives on precipitation patterns for the study area (fig. 1). The first gage, generally representative of precipitation patterns in the lower to middle region of the study area, is located near the mouth at Springfield City Hall (altitude 456 ft) (site 2, fig. 1) and is maintained by Lane Regional Air Protection Agency. The second is located near the upper end of the study area at Trout Creek (altitude 2,400 ft) (site 4, fig. 1), and is a Remote Automated Weather Station (RAWS) operated by the Western Regional Climate Center. This gage generally describes precipitation patterns in the middle to upper regions of the study area.


Continuous instantaneous discharge data, collected and published according to standard USGS procedures (Turnipseed and Sauer, 2010), were evaluated for Cedar Creek at Springfield (site 9, table 1). Water velocity also was measured for selected samples over the course of storm sampling by HACH^TM velocity sensors associated with automatic samplers deployed in stormwater drains. The water velocity data were converted to instantaneous discharge data using standard area-velocity equations based on channel dimensions (Turnipseed and Sauer, 2010). This information provides qualitative estimates for station hydrographs (for example, to describe the timing of peak flows) prior to and during sample collection. Although these data also were collected at other sites, evaluation of instantaneous discharge data in this report is limited to data from stormwater drains and USGS data collected at Cedar Creek at Springfield.


Estimates for selected physical-chemical characteristics for detected pesticide compounds were assembled from literature sources (Vogue and others, 1994; Mackay and others, 1997; Gilliom and others, 2006) in order to provide insight into the expected behavior and mode of transport for these compounds. Important factors related to pesticide transport in runoff include the tendency of the specific compound to partition into water or to be associated with soil particles, as measured by the sorption coefficient (Koc), as well as its persistence in its original chemical form, as measured by the soil half-life (T_1/2) (Gustafson, 1989). Koc describes the potential for sorption of the compound from the dissolved phase to the solid phase, with low Koc associated with small sorption potential. Koc is inversely related to solubility in water, so that transport of compounds with low Koc will be largely determined by the flow of water because these compounds primarily are associated with the dissolved phase. In contrast, compounds with high Koc that are strongly associated with soil/suspended sediment particles exhibit different transport behavior, tending to settle out in slow‑moving regions of the hydrologic system. Compounds with low Koc are described as hydrophilic while those with high Koc are described as hydrophobic. Although Koc provides a measure of a compound’s mobility in runoff, T_1/2 is important because it represents how quickly degradation occurs in soil prior to runoff. T_1/2 is the length of time required for one-half of the amount of compound to degrade in the soil, so that each half-life that passes continues to reduce the amount present by one-half. In general terms, pesticides may be considered to be non-persistent if their half-life is less than or equal to 30 days, moderately persistent if their half-life is 31–99 days, and persistent if their half-life is greater than 100 days (Vogue and others, 1994). Values presented for Koc and T_1/2 in this report cannot be considered as absolute, because these attributes vary depending on many factors, but are nonetheless considered useful for comparing relative differences among different compounds.


To assess the potential threat to drinking-water quality, observed concentrations at the EWEB drinking-water intake were compared to the USEPA MCLs for compounds subject to established drinking-water standards. For unregulated compounds for which standards have not been established, concentrations were compared to HBSLs compiled by the USGS (Toccalino and others, 2003; Toccalino, 2007). HBSLs represent benchmark concentrations from various sources that may be of concern for human health where the HBSLs are exceeded. HBSL guidelines are not legally enforceable regulatory standards, although they are based on the standard USEPA cancer classifications, toxicity data, and procedures for establishing drinking-water guidelines. Accordingly, they are consistent with USEPA methods for establishing Lifetime Health Advisory and Risk-Specific Dose values. They provide important context for identifying pesticides that may represent a potential concern, although they do not provide information on specific effects on human health. 


In order to compare the relative toxicity for each detected compound, measured concentrations were normalized by the appropriate drinking water MCL or HBSL concentration to calculate a Benchmark Quotient (the ratio of the measured concentration to the benchmark concentration). Caution must be used in evaluating these metrics because they are relevant only in terms of concentrations that represent long‑term exposure to contaminants in drinking water (generally defined as drinking 2 L daily for 70 years). Concentrations measured in samples from storm runoff do not meet this definition, especially concentrations in storm runoff samples associated with tributaries that are not used for drinking water, or where insufficient data are available to determine a reliable mean concentration over the full range of storm conditions. Nonetheless, these criteria provide a way to identify contaminants, sources, or conditions that may warrant additional monitoring or source reduction efforts where storm concentrations approach the benchmark. The recommended guidelines for interpretation of the Benchmark Quotients are described in table 3 (Toccalino, 2007).


Further assessment of potential threats to human health was provided by the evaluation of detected compounds as potential endocrine disrupting compounds. Endocrine disrupting compounds interfere with the function of estrogen, androgen, and thyroid hormones in humans and animals. Because exposure to these chemicals can cause serious reproductive and developmental consequences, they may present a threat to drinking-water quality if exposure exceeds the threshold of the dose-dependent endocrine response. Designation of detected compounds as suspected endocrine disruptors was based on ratings presented by the Pesticide Action Network (PAN, 2012). This designation was made when any of several different sources of information lists the compound as suspected of endocrine disrupting activity. The various sources of information used by the PAN in this designation include the Illinois State Environmental Protection Agency, the Danish Environmental Protection Agency, the European Union Prioritization List, and three published references (Colborn and others, 1993; Benbrook, 1996; Keith, 1997).


Data to estimate pesticide use for different categories of land use were acquired to identify compounds expected to be present in the basin, as well as to focus and guide sampling times and locations for forestry sites; no attempt was made to quantify pesticide application rates or volumes. Reported pesticide-use data for urban lands were acquired from the Oregon Department of Agriculture, based on residential surveys conducted during 2007 (Sunny Jones, Oregon Department of Agriculture, written commun., May 17, 2010). These data were summarized for the three postal zip codes in Springfield, and represent the pounds of active ingredient reported to be used. Pesticide-use estimates for forestry were acquired from LCOG (David Richey, written commun., September 14, 2010), compiled from projected pesticide spray notifications provided to the Oregon Department of Forestry by commercial timber companies for specific sub-basins. These spray notifications generally include more compounds than are actually applied. Pesticide-use data for agriculture were estimated based on application data for specific crop types in the Willamette Valley from Anderson and others (1997), where those crops were characteristic of the McKenzie River basin; these estimates were verified with the Oregon State Extension agent for Lane County (Ross Penhallegon, oral commun., August 14, 2011). 


Water-Quality Sample Collection 


Pesticide samples were collected as a mix of composite and grab samples, depending on the location and flow rates. The emphasis on capturing extreme flow conditions during storm sampling in urban drains and small tributaries meant that water levels tended to rise and decline at most sites during a short period of time. Because of limited personnel to collect storm samples, the use of automatic samplers was determined to be appropriate. Approximate flow-weighted samples for stormwater drains and most tributary locations were composited manually from aliquots collected into 1-L glass bottles by Sigma^TM automatic samplers; aliquots were composited based on stage data collected over the course of the storm (fig. 4). After evaluation of the distribution of aliquot collection over the sample hydrograph, each sample was identified with the region of the hydrograph where it was collected. Where stage data were not available, time-weighted composite samples were collected manually into 1-L amber glass bottles over a 30-minute period as equal-volume aliquots collected every 3 minutes by peristaltic pump. Grab samples were collected manually into 1-L amber glass bottles by peristaltic pump or via a sampling pole; samples from the EWEB intake were collected directly from a spigot. All flow- and time-weighted composite samples were composited and split using a Teflon^TM churn splitter, and were processed in accordance with the USGS “National Field Manual for the Collection of Water-Quality Data”(U.S. Geological Survey, variously dated), and EWEB’s “Lower McKenzie Watershed, Stormwater and Urban Runoff Monitoring Plan”(EWEB, written commun, 2004). To the greatest extent possible, the USGS parts-per-billion protocol was used in sample processing.


Several sets of special-topic ambient samples were collected early in the program from selected stormwater drains to evaluate specific questions related to timing and quality control of sample collection (table 4). Three sets of paired pre-storm and storm samples were collected to compare pesticide occurrence prior to the onset of storm conditions with occurrence during the peak of storm runoff. These were collected in fall 2002, and fall and spring 2004. Three sets of paired samples also were collected to examine the influence of sampling over different regions of the storm hydrograph on pesticide occurrence. These samples were collected both early and late in the storms sampled during fall 2002 and spring 2005. All these samples were composited and processed similar to ambient samples.


All samples were filtered within 8 hours of collection through 0.7-mm baked glass-fiber (GF/F) filters into 1-L baked amber glass bottles, and shipped overnight on ice to the USGS National Water-Quality Laboratory (NWQL) in Denver, Colorado. Once samples reached the NWQL, they were analyzed using carbon-based and resin-based solid‑phase extraction, capillary-column gas chromatography/mass spectrometry (GC/MS—USGS schedule 2010/2033), and high-performance liquid chromatography/mass spectrometry (HPLC—USGS schedules 2060 and 2080). Additional filtered samples (125 mL amber glass bottles) were shipped to the USGS Organic Geochemistry Research Laboratory in Lawrence, Kansas, for high performance liquid chromatography/mass spectrometry (HPLC/MS) analysis of glyphosate compounds not included in the NWQL schedules (USGS analysis code LCGY). Analytical methods are documented in Werner and others (1996), Sandstrom and others (2001), Furlong and others (2001,2008), and Meyer and others (2009). The suite of compounds analyzed are listed in appendix A. Four compounds were measured by multiple USGS schedules under different parameter codes: carbaryl, carbofuran, linuron, and terbacil. Using guidance from the NWQL that ranks the quality control from mass spectrometry and gas chromatography methods higher than liquid chromatography, data for these compounds were included in the analysis for this report preferentially from schedule 2010/2033. 


Two levels of reporting concentrations are used for all data: the long-term detection limit (LT-MDL) and the laboratory reporting level (LRL) (Childress and others, 1999). The LT-MDL is statistically defined as the smallest concentration that can be measured and reported with 99 percent confidence, and is calculated over an extended period of time (generally 6–12 months). At the LT-MDL concentration, the risk of a false positive detection is less than or equal to 1 percent; however, the risk of a false negative occurrence can be as much as 50 percent. To reduce this unacceptably high risk of reporting an analyte as not present when it actually is present, the LRL is defined by the laboratory as twice the LT-MDL. 


All analytical results greater than the LRL in use at the date of laboratory analysis are reported as unqualified. Analytical results less than the relevant LRL but greater than the LT-MDL are reported only when the identity of the analyte is confirmed, indicating that the compound is present although there is uncertainty in the absolute value of the reported concentration. These results are reported with a qualifying E-code. If the analytical result is less than the LT-MDL, the concentration is reported as less than the LRL. Occasionally, concentrations are censored at values greater than the LRL because interference from the sample matrix introduces greater uncertainty. The range of LRLs for each compound during the period of this study are shown in appendix A. 


Quality Assurance


A total of 39 quality-assurance (QA) samples, including equipment blanks, replicates, and laboratory spikes, were collected and represent about 24 percent of the total number of samples collected. The QA program was particularly intensive in the first years of sampling in order to ensure that methods used would provide reliable data at low detection levels without contamination. These samples were collected to identify the potential for bias as well as to assess the level of precision in the ambient data. Analyses of surrogate compounds added by the NWQL provide additional information on the accuracy and potential for bias of the analytical process. Finally, a special QA study was conducted to evaluate the potential for bias from sample collection using plastic tubing in the automatic samplers. This section summarizes all the QA results; figures are presented in appendix B.


Field equipment blanks were used to document the potential for contamination occurring during equipment cleaning, sample processing, shipment, and analysis at the laboratory. A total of nine blank samples were collected from 2002 to 2010 using certified organic-free water, and were submitted to the entire process of sample collection, compositing, filtering, shipment, and analysis. Only one detection was observed—caffeine (0.010 µg/L) in the first blank sample from the 42nd Street stormwater culvert (site 7) in September 2002. This sample was associated with an exceptionally high caffeine concentration (11.4 µg/L) that was subsequently traced to a coffee kiosk disposing gray water directly into the storm drain upstream of the sampling site. This blank concentration was considered to be relatively insignificant because it was approximately equal to the LRL (0.0096 µg/L) and represented a very small fraction (< 0.1 percent) of the ambient concentration. No other detections, either unqualified or E-coded, were observed in any blank samples. These results indicate that contamination is not a concern for the ambient data.


A total of 18 pairs of replicate samples were collected to describe variability inherent in the sampling and analytical process. These samples were collected in several ways, encompassing all the sampling methods used for ambient data—11 pairs were collected with the same method (either grab or composite) and 7 pairs were collected with different methods (either split concurrent, collected over a different region of the hydrograph, or composite versus grab) (table 5). Data from the replicate samples were evaluated in two ways: first, the concentrations in the pair of replicate samples were compared to determine the relative percent difference (RPD) as the difference between the two concentrations divided by the mean. RPD results were further examined to determine if there was a consistent effect of sample collection method. RPD values greater than 10 percent between the replicate pairs were considered to represent significant variability. Second, ratios of standard deviations from replicate samples were compared to those from routine samples to determine the relative importance of sampling variability to overall variability in the measurement process.


Most compound pairs (1,954, or 98 percent, of the 2,003 pairs of analyses) were associated with non-detections in both replicate samples. Of the remaining 49 pairs of analyses,18 were associated with one value coded as less than LRL and either one E-coded or unqualified value; all E-coded or unqualified values were either less than (n = 10) or close to the threshold defined by the LRL (n = 8, median difference = 0.02 µg/L). These replicate pairs were not evaluated further because the values less than LRL were not quantified. For 16 of the other 31 replicate pairs, both replicate values were E-coded. Eight of these pairs were associated with an RPD greater than 10 percent, primarily associated with replicates collected over different regions of the hydrograph (appendix B, fig. B1-A); these compounds included carbaryl (n = 4), caffeine (n = 1), diuron (n = 1), malathion (n = 1), and prometon (n = 1). Fifteen remaining compound pairs were associated with at least one unqualified value, and 12 of these showed a high level of variability between the two samples (RPD > 10 percent); affected compounds include 2,4-D (n = 4), 2,4-DB (n = 2), caffeine (n = 2), diazinon (n = 1), diuron (n = 1), imidacloprid (n = 1), and triclopyr (n = 1) (appendix B, fig. B1-B and B1-C). Eight of the replicate pairs with RPD greater than 10 percent were associated with different sampling methods (six comparing composites collected over different regions of the hydrograph, and two comparing composite samples with grab samples); four pairs were collected as split concurrent replicates (appendix B, fig. B1-B and B1-C). These results demonstrate that sampling/analytical variability can be important for a small number of compounds, especially where concentrations are E-coded, and that sampling method is implicated as a source of variability. Although the small number of samples with significant variability precludes determination of a clear pattern of bias in general, results indicate that sampling over different ranges of the hydrograph may be the most important source of variability. 


In comparing the variability in replicate samples with variability in ambient data, the median standard deviation for replicate data (excluding pairs where at least one value was less than LRL or not quantified) was 0.007 overall compared to the median for ambient data (similarly excluding values not quantified) of 0.028. Analysis of variance among the different sources of variability in replicate samples showed a significant difference among the standard deviations associated with each replicate type (n=31, p<0.0001). Parsing out the different sources of sampling variability, the largest standard deviation was associated with samples collected over different regions of the hydrograph (median=0.019) compared to the variability due to sampling method (median=0.01), and the median standard deviation for split-concurrent samples (which essentially represents laboratory variability) was zero. These results indicate that variability generally can be ranked in the following order—laboratory variability < variability due to different sampling methods < variability due to sampling over changing field conditions < variability inherent in sampling different storm conditions. The small number of samples with significant variability (greater than 10 percent RPD between replicate samples) prohibits more rigorous analysis, although results indicate that field variability of pesticide concentrations over the storm hydrograph is likely to be more important than variability introduced by the combined sampling and analytical methods for dissolved pesticides in these primarily small streams.


Fourteen replicate laboratory spike samples were collected to measure the analytical recoveries for specific compounds, providing information on accuracy, precision, and potential bias in the analysis from effects of the sample matrix. Spike samples were environmental samples that were fortified with certified known concentrations of a group of analytes and paired with unspiked samples. Concentrations in the spike mixture represented an increase of about 0.1 µg/L. Percent recoveries for the spiked sample were calculated as the ratio of the difference between the spiked and unspiked concentrations to the concentration of the added spike. Because some compounds were not detected in the ambient samples even though they were reported or estimated to be used in the basin, recoveries for these compounds were evaluated by laboratory reagent spike data. These reagent spike data cannot be considered equivalent to matrix spike data because they provide a measure of analytical recovery that is independent of the sample matrix.


Because of the small number of detections in the ambient data to compare to spike results, it generally was only possible to determine a small number of matrix spike recoveries where concentrations were quantified in the ambient samples. Of these, the GC/MS method provided higher recoveries (median = 77 percent, n = 13) than the HPLC method (median = 40 percent, n = 4) (appendix B, fig. B2). Recoveries for laboratory reagent spikes generally were higher than recoveries for matrix spikes, and again the GC/MS method (median = 97 percent; n = 510) performed better than the HPLC method (median = 89 percent, n = 556). Two of the 45 compounds that were detected in the ambient data (cis- and trans-propiconazole) were associated with median recoveries higher than 140 percent, and 5 compounds (2,4-D, triclopyr, 1-naphthol, dinoseb, and metsulfuron-methyl) were associated with median recoveries less than 60 percent. All these compounds except for 1-napthol were analyzed by the HPLC method. Of compounds reported to be used in the basin but never detected, three were associated with median recoveries less than 60 percent in laboratory reagent spike samples. These included cis-permethrin, bromoxynil, and oxamyl—the latter two were analyzed by HPLC method. Because reports of pesticide use are estimated and are not reliable, these results do not reflect any certainty that these compounds should be expected to be detected, and were therefore not detected because of analytical bias. Nonetheless, the results clearly demonstrate that some compounds analyzed by HPLC method were more likely to be associated with low recoveries, which implies that measured concentrations for these compounds may be biased low. 


Surrogate compounds, or deuterated versions of compounds included in the laboratory schedules that are never detected in environmental samples, were routinely added by the laboratory in known amounts to every sample. Because surrogates are similar in structure to selected target compound groups of interest, surrogates were analyzed to further monitor the potential for matrix effects that may affect compound recoveries. Surrogate recoveries were calculated in the same way as spike recoveries. Similar to the spike results, surrogate recoveries were most consistent using the GC/MS method, largely between 80 and 120 percent (appendix B, fig. B3). Surrogate recoveries using the HPLC method were more variable and tended to be lower, generally between 60 and 100 percent. More recoveries were less than 60 percent and greater than 140 percent for the HPLC method. These results are consistent with the spike results, showing that recoveries for the GC/MS method generally are less variable than recoveries for the HPLC method. Surrogate issues with the HPLC method have been documented by others and these results are consistent with the relatively poor performance of surrogate compounds with this method (Werner and others, 1996; Munday and Domagalski, 2003). 


To evaluate the potential effect of plastic tubing in the autosamplers on pesticide concentrations, a special QA study was conducted in collaboration with a research chemist from the NWQL. This study involved collecting seven pairs of replicate grab samples from stormwater drains during 2002–04. The first sample in each replicate pair was collected through the autosampler tubing into clean glass bottles seated in the sampler, and then filtered into baked glass bottles without compositing. The second sample was grabbed directly into baked glass bottles. The results were evaluated by determination of RPD between the two sample collection types as well as comparison of standard deviations with those from other replicate samples and ambient data. Of 697 replicate pairs of analyses, both replicate concentrations were reported as less than the LRL for 634 pairs or about 91 percent. Of the other 63 pairs, 17 pairs were associated with one concentration qualified as less than the LRL and were not evaluated further because values less than the LRL were not quantified. Finally, 19 of the remaining 48 pairs of analyses were associated with an RPD greater than 10. Most of these pairs with significant variability were from pairs either with both (10 of a total of 20) or one (4 of a total of 6) E-coded detections, with 5 pairs (from a total of 20) where both detections were unqualified. Of these, concentrations were relatively higher in 13 pairs in the samples collected directly into glass, while concentrations were relatively higher in 6 pairs in the samples collected through the plastic tubing (appendix B, fig. B4). Compounds with smaller concentrations in the tubing samples include 2,4-D (n = 2), desulfinylfipronil (n = 1), diazinon (n = 2), diuron (n = 3), metsulfuron-methyl (n = 1), and sulfometuron-methyl (n = 2).Compounds with higher concentrations in the tubing samples include bentazon (n = 1), carbaryl (n = 1), prometon (n = 1), and triclopyr (n = 1); caffeine showed no clear pattern with two pairs having smaller and two having higher concentrations. Although the number of detections was too low for statistical conclusions, these results indicate that concentrations for some compounds in samples collected by automatic sampler may be biased low due to interaction with the plastic tubing, especially for E-coded values that are subject to high analytical uncertainty.


In order to evaluate the variability due to the use of plastic tubing within the context of other sources of sampling variability, standard deviations of replicate pairs (including only those pairs where both concentrations were greater than the LRL) from these special QA samples (median = 0.018) were compared with those from the routine replicate samples describing variability from sampling method (median = 0.01) and samples collected over different regions of the hydrograph (median = 0.019). These results indicate that the use of plastic tubing in the automatic samplers introduces variability to the sampling process that is comparable to the effect of sampling over different regions of the hydrograph.


This variability is most pronounced for E-coded concentrations, which by definition are low for most compounds. Although not insignificant, this variability was considered acceptable within the context of the reconnaissance or exploratory objectives of the project, which are focused on capturing storm runoff in order to identify important source areas for pesticides. It was decided that the logistics of event-based sampling of multiple stormwater drains and small streams, where flows tend to rise and decline relatively quickly over the course of a storm, meant that using automatic samplers provided a better means to collect a representative sample with the limited personnel resources that were available.


Data Analysis


Because of the non-fixed design of this reconnaissance study and the corresponding inconsistency in sampling timing and frequency, it was not possible to use standard statistical procedures to evaluate the data. The analysis presented in this report is essentially descriptive in nature, focusing on the distribution of detected compounds across seasons and land-use categories. The evaluation of storm conditions focuses on antecedent precipitation and sampled storm characteristics. Simple correlations are examined to evaluate relations between precipitation and/or discharge with occurrence of pesticides, as measured by number of detected compounds and the sum of all quantified or total concentrations in a single sample. Results from the various sampling surveys also are considered within the context of the physical-chemical characteristics of detected compounds, their potential for endocrine disrupting activity, and the various health-based criteria. Finally, results are synthesized with those from other studies to generate a conceptual model describing pesticide transport in the basin.


First posted May 30, 2012