Pesticides in Streams of the United States--Initial Results from the National Water-Quality Assessment Program

U.S. GEOLOGICAL SURVEY
Water-Resources Investigations Report 98-4222
Sacramento, California, 1999

METHODS FOR SAMPLE COLLECTION AND ANALYSIS

Water samples were collected at 58 sites in accordance with the NAWQA national sampling strategy (Gilliom and others, 1995). Samples were collected for 1 to 3 years at each site using a combination of fixed-interval and extreme-flow sampling. For the fixed-interval sampling, 4 to 8 samples generally were collected each month during critical periods of high pesticide use and runoff and 1 to 2 samples were collected each month during other periods. Additional samples were collected during periods of extreme high and low flows. Samples were collected more frequently for some sites where short-term fluctuations were a concern. The sampling frequencies for the 58 sites included in this report are shown in figure 4. The total number of samples collected at each site from 1992 through May of 1995 is shown in table 1. The focus of this report is primarily on results from samples collected during 1993, 1994, and the first few months of 1995.

All samples were depth- and width-integrated using standard USGS methods (Shelton, 1994). All equipment used for collecting and processing water samples was constructed of Teflon, glass, aluminum, or stainless steel and was cleaned and rinsed with residue-grade methanol. All samples were filtered using pre-combusted glass-fiber filters with a nominal 0.7-mm pore diameter to remove suspended particulate matter. Complete collection and processing methods are described by Shelton (1994).

The target compounds were extracted from water samples using C-18 solid-phase extraction (SPE) columns and identified and quantified using capillary-column GC/MS. Complete details of the analytical procedure are given in Zaugg and others (1995) and will be described only briefly here.

One-liter, filtered water samples were drawn through C-18 SPE columns under vacuum. The target compounds sorb to the C-18 phase, effectively removing them from the water. The SPE columns were then dried using a gentle stream of carbon dioxide to remove residual water. The target compounds were removed from the SPE columns by elution with hexane:isopropanol (9:1). The volume of the eluate was reduced using a gentle stream of nitrogen. The sample extract was then injected onto a capillary-column gas chromatograph for separation of the target compounds. Pesticides and pesticide metabolites were identified and quantified using a mass selective detector operating in the selected-ion monitoring (SIM) mode.

Method detection limits (MDLs) were determined using standard U.S. Environmental Protection Agency (USEPA) procedures (U.S. Environmental Protection Agency, 1994; Zaugg and others, 1995; U.S. Geological Survey, 1998). MDLs for the 46 target compounds ranged from 0.001 µg/L to 0.018  (table 2). Target compounds were quantified at concentrations less than the reported MDL if compound identification criteria were met (Zaugg and others, 1995). Reported concentrations lower than the MDL were used in some of the calculations of summary statistics in this report.

Analytical recoveries for the target compounds are shown in table 2. The values of the analytical recoveries are based on analysis of laboratory-spiked reagent-grade water using pesticide concentrations ranging from 0.01 to 0.3 µg/L (Zaugg and others, 1995; U.S. Geological Survey, 1998). Recoveries ranged from 13 to 156 percent, with recoveries for 80 percent of the compounds between 39 and 113 percent. This analytical method was developed to enable detection of very low concentrations of a maximum number of compounds; analysis conditions, however, were not optimal for all of the target compounds. Performance of this method was relatively poor or inconsistent for six compounds--azinphos-methyl, carbaryl, carbofuran, DEA, linuron, and terbacil. The analytical recovery for these compounds was low, and all reported concentrations are regarded as estimates. When these six compounds are excluded, recoveries ranged from 42 to 156 percent, with recoveries for 80 percent of the compounds between 50 and 113 percent. Concentrations were not corrected for analytical recovery in the analyses for this report; therefore, it is important to keep in mind the differences in recovery among the target compounds when evaluating results.

METHODS FOR DATA ANALYSIS

Analysis of data on pesticides in streams presents many unique problems stemming from the characteristics of trace-level organic contaminants, strong seasonal variations in concentrations, variable data-collection strategies, and other factors. Although there are many possible ways to address these problems in data analysis, only the methods used for this report are described below.

Selection of Critical Period

The sampling frequency varied considerably among the 58 sites (fig. 4), and thus some comparisons between sites would not be appropriate if data from an entire year were used. For example, a comparison of mean concentrations for 1994 for Duck Creek (wmic-duck) and for Cherry Creek (splt-cherry) would be biased because the samples were collected only from April through October at Duck Creek but throughout the year at Cherry Creek. For this reason, some of the comparisons made in this report are based on data from samples collected during a designated critical period.

A 5-month critical period was selected for each site during which pesticide concentrations were highest and sampling was relatively frequent. Samples collected during the critical period are shown in red in figure 4. The choice of a critical period for each site was based on the temporal distribution of the total (summed) pesticide concentrations measured during the 1- to 3-year sampling period and on the sampling frequency at the site. When sufficient samples were collected at a site during 2 different years, a critical period was chosen for each year. Comparisons that are based on concentrations measured during the critical periods are influenced less by differences in sampling frequency and timing of pesticide application than are comparisons that are based on an entire year of data. For most sites, the chosen critical period was from April through August. For some sites in the South, the Northwest, and California, the critical period began in autumn or winter.

Calculation of Detection Frequencies

In general, more samples were collected during periods when elevated concentrations of pesticides were expected and stream discharge was high. At most sites, this occurs during May, June, and July. For the calculation of summary statistics, such as detection frequencies and mean or median concentrations, this targeted sampling must be accounted for or the results will be biased.

Annual mean detection frequencies for each compound at each site were derived by first calculating the detection frequency for each month. The mean of the 12 monthly detection frequencies was then used as an estimate of the detection frequency for a 1-year period. The 1-year period with the most intense sampling at each site was used, which for most sites was from spring 1993 to spring 1994. For months when no samples were collected at a site, the mean of the detection frequencies for the 2 adjacent months was used. Detection frequencies calculated in this way are estimates of the detection frequencies that would be obtained if samples were collected at even intervals throughout the year. At eight sites, samples had not been collected for several months of the 1-year period and thus unbiased detection frequencies were not calculated for these sites.

Differences in the detectability of the target compounds also must be accounted for if detection frequencies of the compounds are to be compared. For example, a comparison between the detection frequencies for atrazine, with an MDL of 0.001/, and tebuthiuron, with an MDL of 0.01 µg/L, would not necessarily reflect the true difference in occurrence of these compounds. Any difference could be due to the tenfold difference in analytical sensitivity. Although detections less than the MDLs were reported, the MDL values give an indication of the relative detectability of the target compounds. To account for these differences, a minimum concentration, or common reporting level of 0.01 µg/L, was used for comparisons of the detection frequencies of the 46 compounds. Thus, detection frequencies that are based on the common reporting level represent the proportion of samples in which the concentration of a specific compound equaled or exceeded 0.01 µg/L rather than the proportion of samples in which the compound was detected. Four compounds--disulfoton, prometon, propargite, and terbufos--have MDLs higher than 0.01 µg/L, but the reporting of detections less than the MDL moderate this potential bias. Propargite and terbufos have MDLs of 0.013 µg/L, which is only slightly higher than the common reporting level of 0.01 µg/L. Disulfoton, which has an MDL of 0.017 µg/L, was detected in only 6 samples, all of which had concentrations less than 0.01 µg/L. Prometon which has an MDL of 0.018 µg/L, was detected in 1,179 samples (U.S. Geological Survey, 1999). However, one-third of the reported concentrations of prometon were less than the MDL; the most commonly reported concentration was 0.01 µg/L. Thus, the detection frequencies reported for the compounds with MDL's greater than 0.01 µg/L are reasonably accurate and comparable with detection frequencies for the other compounds.

Calculation of Total Concentrations

A total concentration of pesticide compounds was used for many of the analyses done for this report rather than the concentrations of each individual compound. Using the total pesticide concentration allowed basic comparisons to be made among basins with different crop types or land uses and where different pesticides may be used. The total concentration of herbicides, insecticides, or all pesticides in a sample was determined by summing the concentrations of individual compounds. Thus, the total herbicide concentration is defined as the sum of the concentrations of all 27 herbicide and herbicide transformation products included in the target compounds. Similarly, the total insecticide concentration is defined as the sum of the concentrations of all 19 insecticides and insecticide transformation products, and the total pesticide concentration is defined as the sum of the concentrations of all 46 target compounds. Individual compounds that were reported as not detected were assigned concentrations of zero for these sums.

Calculation of Monthly and Annual Median Concentrations

Uneven sampling frequency (fig. 4) affects the calculation of median concentrations of pesticides. Median concentrations calculated using all samples would be biased high because at most sites more samples were collected during periods of elevated pesticide concentrations. To minimize this bias, monthly median concentrations were calculated for each compound at each site. For months in which no samples were collected at a site, the mean of the median concentrations for the 2 adjacent months was used. For months when only one sample was collected, the concentration in that sample was used as the monthly median concentration. Concentrations that were reported as not detected were given a value of zero. In some cases, this resulted in a monthly median value less than the MDL for a particular compound. Monthly median concentrations also were determined for total herbicides and insecticides and for total pesticides.

For a few of the comparisons made in this report, an annual median concentration was used. Median concentrations for a 1-year period were calculated as the median of the 12 monthly median concentrations. This procedure gives equal weight to samples collected during each month so that the annual median concentration is not biased by the variable sampling frequency used during the year. In addition, using the median of the monthly values, rather than the mean, minimizes the influence of extreme values in the distribution of monthly median values. This method therefore, is a somewhat conservative way of calculating an annual median concentration. Samples were not collected at eight sites during several months of the 1-year period; annual median concentrations, therefore, were not calculated for these sites.

Calculation of Time-Weighted Mean Concentrations

Similar to median concentrations of a compound, the mean concentration calculated for a specified time interval can be affected by uneven sampling frequency. To minimize this bias, time-weighted mean concentrations were determined in which the concentration reported for a sample is assigned to a time interval that is based on the number of days between that sample and the adjacent samples. The time interval associated with each sample extends halfway to the date of the preceding sample and halfway to the date of the succeeding sample. For example, if samples were collected on May 1, 8, and 19, the concentrations reported for the May 8 sample would be assigned to all days from May 5 through May 13. The sum of concentrations assigned to all days during a specified time interval is then divided by the total number of days in the interval to obtain a time-weighted mean concentration. Because of the relatively low sampling frequency during some months at most sites, time-weighted mean concentrations were calculated only for the 5-month critical period when samples were collected frequently. By using concentrations for the critical period only, potential errors resulting from concentrations in a sample being assigned to long periods of time are avoided.

The time-weighted mean concentration of a compound can be strongly influenced by a single sample with a high concentration. For example, the highest calculated 5-month time-weighted mean concentration--9.6 µg/L for cyanazine in Kessinger Ditch in Indiana (whit-kess)--is strongly influenced by one sample collected in May 1993, which contained 160 µg/L of cyanazine. If this sample was not used in the calculation, the time-weighted mean concentration for cyanazine would be 2.1 µg/L for the 5-month period. However, because each sample is weighted according to the amount of time the sample represents and because sampling usually was more frequent when pesticide concentrations were highest, the time-weighted mean is not as sensitive to extreme values as a simple mean would be, for which all samples would be weighted equally. The time-weighted mean is a useful measure of concentration for assessing sustained exposure of ecosystems and water users to pesticides.

Calculation of Load and Yield

The load of a compound is the mass of that compound transported in a stream during a specified period. The load can be estimated as the product of the concentration of the compound and the discharge volume of the stream measured at the same location. Daily discharge values were available for each day during the sampling period for nearly all sampling sites. Daily concentration values for the target compounds were obtained by linear interpolation between measured values. The load estimates for most compounds probably are biased low because concentrations that were reported as not detected were given a value of zero in the load calculations and because concentrations were not adjusted for analytical recoveries (table 2). In addition, load estimates for small streams generally are less precise than load estimates for larger streams. This is because of the higher variability in the concentrations of pesticides in most small streams and the generally low probability of sampling at the time of peak pesticide concentrations. In most cases, peak concentrations in small streams probably were not sampled and thus the loads calculated at these sites are biased low. Pesticide loads were calculated for 1-year periods at sites with sufficient sampling and for 5-month critical periods at all sites. Load estimates also were calculated for total pesticides, total herbicides, and total insecticides by summing the concentrations of individual compounds prior to multiplication by the discharge values.

The yield of a target compound is defined as the load of that compound at the sampling site, divided by the area of the drainage basin. For the yield calculations in the report, we assumed that the pesticides entering the streams as a result of use on nonagricultural or nonurban land did not significantly affect the stream load. Yields at agricultural indicator sites were calculated by using the area of agricultural land (excluding pasture) in the drainage basin. Yields at urban indicator sites were calculated by using the area of urban land in the drainage basin. Yields at integrator sites were calculated by using the sum of the areas of agricultural land (excluding pasture) and urban land in the drainage basin. Thus, the yields are estimates of the amount of a pesticide transported in a stream per unit area of agricultural or urban land, or both, in the basin. Estimates of yield are subject to the same uncertainties as are estimates of load.

Graphical Representation of Results

Boxplots are used in many of the figures in this report to represent the distribution of data for different sites or different target compounds (fig. 5A). In all boxplots in this report, the box part of the figure encompasses data points between the 25th and 75th percentiles of the data, which represents the middle 50 percent of the data. The median of the data (the 50th percentile) is shown as a horizontal line through the box. Vertical lines (whiskers) extend from the box down to the 10th percentile and up to the 90th percentile so that the box and whiskers together represent the middle 80 percent of the data. In some of the boxplots, points below the 10th percentile and above the 90th percentile (extreme values) are indicated by circles. Extreme values are not shown in the figures if the main point is best expressed by showing how the bulk of the data are distributed; but for cases where the extreme values do show an important aspect of the data, the extreme values are included in the plot. Boxplots are useful for visually displaying the form of the distribution of data. For example, in the hypothetical boxplots shown in figure 5A, it can be seen that the data represented by plot A are skewed toward the high end, with more than 50 percent of the data less than about 5 and the remainder of the data extending to 30. Plot B represents data in which most observations are tightly distributed between 3 and 10, except for a few much higher values. Plot C represents data that are uniformly distributed across a relatively wide range.

Concentration distribution plots also are used in this report to represent the distribution of concentrations within a data set (fig. 5B). In this type of plot, a line is used to represent the percentage of values in the data exceeding particular concentrations. The examples in figure 5B illustrate how to interpret the frequency distribution plots. For a given concentration on the x-axis, the percentage of monthly median concentrations exceeding that concentration can be read from the y-axis. For example, the line representing compound A shows that 68 percent of the monthly median concentrations of compound A exceeded 0.01 µg/L and 6 percent exceeded 1 µg/L. For compound B, concentrations generally were lower, with only about 33 percent of monthly median concentrations exceeding 0.01 µg/L and none exceeding 1 µg/L. The information shown in the concentration distribution plots is similar to that shown in the boxplots, except that the entire distribution is shown rather than selected percentiles. The frequency distribution plots are useful for comparing measured concentrations in samples with a concentration of particular interest, such as a water-quality criterion value.

Land-Use and Pesticide-Use Estimates

The distribution of land use within the 58 drainage basins included in this study is shown in figure 3. The land-use percentages were derived from U.S. Geological Survey Land Use and Land Cover (LULC) data (U.S. Geological Survey, 1990) stored in the Geographic Information Retrieval and Analysis System, or GIRAS (Mitchell and others, 1977), and from the 1992 Census of Agriculture (U.S. Department of Commerce, 1995). Land-use classes are based on the Anderson Level I classification system (Anderson and others, 1976), except for the class for agricultural land where the Level II classification "cropland and pasture" has been divided into "cropland" and "pasture" on the basis of the proportion of cropland to pasture reported in the 1992 Census of Agriculture. The LULC data are based on aerial photographs from the mid-1970s. These photographs were used to delineate polygons of land use mainly on 1:250,000-scale maps. Both the age of the LULC data and the relatively low resolution are potential sources of error in the land-use estimates used in this report, especially for the small basins. The LULC data, however, are the highest resolution, nationally consistent classification of land use and land cover currently available for the United States. The LULC data for urban areas have been updated by incorporating1990 U.S. Census Bureau population data (Hitt, 1994). The combination of two data sources (LULC and Census of Agriculture), which differ in both time and resolution, is another source of potential error in the estimates of general land use and in the estimates of crop acreages and pesticide use described below.

Estimates of the amount of land in each basin planted in specific crops (table 4) also were derived from 1992 Census of Agriculture and LULC data. County-level estimates of the harvested acreage of specific crops were available from the Census of Agriculture. Using the LULC data, the amount of cropland and pasture, as defined by the Anderson Level II classification (Anderson and others, 1976), was determined for the land in each county wholly or partly included in a drainage basin. The total county acreage for a specific crop was multiplied by the ratio of cropland and pasture in the part of the county included in the drainage basin to the total cropland and pasture in the county. This provided an estimate of the acreage of a specific crop in the part of the county included in the drainage basin. The values for each county wholly or partly included in the basin were then summed to obtain an estimate of the total amount of land in the drainage basin planted in each crop. The use of the LULC data improved the estimates for crop acreage compared with the use of only the Census of Agriculture county data. The assumption was made that the part of a county included in a drainage basin is representative of the county as a whole in terms of the mix of crops grown. Deviations from this assumption result in an overestimation or underestimation of the acreage for specific crops. The estimates are probably more accurate for large basins than for the small basins because the large basins generally contain larger portions of counties or whole counties.

The estimates of the amount of land planted in specific crops are based on harvested acreage; thus not all the land identified as cropland in figure 3 is accounted for in the crop acreage estimates. Fallow land, land planted in cover crops or soil improvement crops, land on which crops failed, and land that is part of the Conservation Reserve Program (CRP) are examples of croplands that are not included in the estimates of crop acreage in the basins. Because nearly all agricultural pesticide use occurs on land where crops are harvested, the estimates of crop acreage provide a useful basis for comparison among the basins in this report.

Estimates of pesticide use in the drainage basins were derived from data compiled by the National Center for Food and Agricultural Policy (NCFAP) (Gianessi and Anderson, 1996) and from estimates of crop acreage described above. For each state, NCFAP estimated the percentage of acres treated with specific pesticides for 87 agricultural crops. NCFAP combined these values with estimates of the mass of each pesticide applied per acre for each of the crops to obtain an average use coefficient, by state, for each pesticide/crop combination. NCFAP applied these state-based use coefficients to county-level estimates of crop acreages from the 1992 Census of Agriculture to obtain estimates of the amounts of specific pesticides used within each county in the conterminous United States. To obtain estimates for the drainage basins discussed in this report, the amount of cropland and pasture in each whole or partial county included in a drainage basin was multiplied by the appropriate county-level use estimate for each pesticide. The values for each county were then summed.

The pesticide-use coefficients used by NCFAP were based on data obtained from a variety of sources during 1990-93 and 1995. These data can reasonably be applied to 1994 as well and thus the pesticide-use estimates derived from the NCFAP data are applicable to the entire sampling period discussed in this report, but with some limitations.

Several potential sources of error in the pesticide-use estimates should be noted. First, the estimates are for pesticide use on agricultural cropland only. Pesticide applications to lawns, gardens, nursery stock, forests, water bodies, rights-of-way, federally owned grazing and pasture land, and other noncropland areas are not included. Seed treatments and postharvest applications of pesticides also are not included. Second, the use coefficients developed by NCFAP are based on statewide estimates of application and treatment rates, and therefore, local variability in cropping and management practices may not be reflected in the use coefficients. Third, the crop acreages are based on 1992 Census of Agriculture data and thus may not represent acreages during the actual sampling period. Fourth, some crop acreage may not have been included in the Census of Agriculture data because of Census nondisclosure rules. Finally, there may have been changes in the pesticides used and in crop acreage or application and treatment rates in some basins during the 3-year sampling period.