Environ. Sci. Technol. 2006, 40, 4990-4995
Variability of Indicator Bacteria at Different Time Scales in the Upper Hoosic River Watershed ELENA TRAISTER† AND SHIMON C. ANISFELD* School of Forestry and Environmental Studies, Yale University, New Haven, Connecticut 06511
Accurately evaluating whether a water body is meeting water quality criteria for indicator bacteria requires an understanding of the spatial and temporal variability in concentrations of these indicators. We have collected data on concentrations of Escherichia coli at 12 sites within the upper Hoosic River Basin, spanning a range of land uses and levels of development. Sampling was conducted with the goal of assessing the variation in E. coli levels over different time scales: seasonal, storm-related, and diurnal. General linear models were constructed to describe the factors contributing to E. coli concentrations at a given location and time. We found that bacterial levels were higher in more developed watersheds; in summer rather than winter; in storms rather than baseflow; and in the early morning rather than afternoon. Seasonal and storm sampling captured different portions of the range of E. coli concentrations, but the levels of variability at these two scales were similar. Diurnal sampling produced concentrations intermediate between seasonal and storm sampling. Compared to a pristine stream, a more urbanized stream exhibited greater diurnal variability, but less variation from baseflow to stormflow. We recommend collecting both seasonal and storm data, but not necessarily diurnal data, in assessment of stream bacterial quality.
Introduction Contamination of surface waters with fecal-derived pathogens poses a significant threat to human health and represents an important barrier to recreational and other uses of these waters. To protect human health, water bodies are required to meet water quality criteria for the levels of indicator organisms such as fecal coliform or Escherichia coli (1). Exceedance of indicator bacteria criteria has been found to be the single most common stressor in streams and rivers throughout the United States (2). A number of studies have investigated the sources of indicator bacteria to specific water bodies. Some of the more important sources identified have included the following: septic systems and waterfowl in Buzzards Bay (3); bird feces, storm drains, and river water in Lake Michigan (4); urban runoff and leaking sewer lines in Catalina Island (5); septic systems and livestock in Thomas Brook, Nova Scotia (6); and impervious areas in Rawls Creek, South Carolina (7). One of * Corresponding author e-mail:
[email protected]; phone: (203)432-5748; fax: (203)432-3929. † Present address: Massachusetts College of Liberal Arts, Biology Department, 375 Church St., North Adams, MA 01247. 4990
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 16, 2006
the best-studied sites is Huntington Beach, California and its watershed, where research has established that indicator bacteria most likely originate not from sewage (8), but from some combination of urban runoff (9, 10), contaminated groundwater (11), and tidal wetlands (12). Several studies have identified stream bed sediments (6, 13) and wrack (3) as reservoirs of indicator bacteria that can lead to continuing water contamination upon resuspension. Assessing whether a water body is meeting water quality criteria is made difficult by the high spatial and temporal variability typical of indicator bacteria concentrations. For example, Whitman and Nevers (14) found that at a Chicago beach, high variation in E. coli levels over spatial scales of ∼100 m meant that a large number of samples was required to determine water quality with reasonable precision. The assessment challenge is most pressing at beaches, where decisions must be made as to whether to close the beach (or post warning signs) in response to exceedances (15). A number of researchers have examined the temporal variability in indicator concentrations at different time scales. At the daily time scale, several studies at swimming beaches have observed a diurnal cycle caused by sunlight-induced dieoff, with higher indicator concentrations in the morning and lower concentrations in the afternoon (16, 17). It is not clear whether this diurnal variation would also be relevant in streams and rivers, which are more likely to be shaded (18). Weather-related variation is an important factor in determining bacterial concentrations. Storm events generally lead to much higher indicator concentrations, due to washoff of bacteria from the watershed and erosion or resuspension of sediment (10, 16, 19). In Charlotte Harbor, Florida, fecal coliform levels were positively correlated with 7-day antecedent rainfall and with streamflow (20). In the River Don, Scotland, the magnitude of the storm-associated increase in fecal coliform appeared to be linked not to the size of the storm, but to the amount of time for bacterial buildup in the watershed before the storm; the highest concentrations were measured during a moderate storm following a period of dry weather (21). At the seasonal time scale, several different patterns have been observed. At many sites, higher concentrations are found during wetter months due to increased runoff (3, 10, 16, 20) while others have found that frequent storm events during wetter months can lead to depletion of bacterial reservoirs in the watershed and streambed, and consequently lower concentrations (21-22). Higher water temperatures in summer can also lead to more rapid inactivation of bacteria and thus lower concentrations (20, 23-24). Only a few studies have examined the temporal variation in indicator concentrations at all the relevant time scales, and these have been focused mostly on saltwater bathing beaches, rather than on freshwater streams. We report here results from a number of streams with different levels of indicator bacteria within the Upper Hoosic River Basin, Massachusetts, including the first examples of diurnal E. coli variation in streams. We use statistical models to examine both the spatial variability in bacterial levels as a function of land use, and the temporal variability in bacterial levels at different time scales.
Experimental Section Study Sites. The Upper Hoosic River, a tributary of the Hudson River, drains approximately 500 km2 of northwestern Massachusetts, along with small portions of Vermont and New York, before flowing north into Vermont (Figure 1). This 10.1021/es0601437 CCC: $33.50
2006 American Chemical Society Published on Web 07/21/2006
FIGURE 1. Sampling sites (numbered) and land use in the upper Hoosic River watershed. watershed contains a mix of land covers, including industrial and urbanized areas as well as farms and forested lands. The most recent water quality assessment of this watershed by the Massachusetts Department of Environmental Protection (MA DEP, 25) identified several stream reaches where results from past bacterial sampling have proven inconclusive and where further study was recommended. For this study, twelve sampling sites were selected which defined a set of partly nested watersheds. Land use data were compiled for these watersheds using data from Massachusetts (1999 data, www.mass.gov/mgis), Vermont (2002 data, www.vcgi.org), and New York (1994-1999 data, www.nygis.ny.us). The watersheds ranged in size from 16 to >500 km2, and ranged in land use from 71% to 91% forested (Table 1). Streamflow data were available for one site (site 5) from the USGS (water.usgs.gov, station 01332500). At
several of the other sites, relative water level data were obtained from staff gauges installed at the commencement of this project. Methods. Three different types of sampling were carried out. First, samples were collected from all twelve sites approximately every fourteen days from March 28, 2004 to December 5, 2004 (“seasonal sampling”); these samples were generally collected between 9 a.m. and 1 p.m.. In addition, samples were collected over the course of storm eventss every 2 h on the rising limb of the hydrograph; every 4 h on the falling limbsat 2 of the sites (sites 3 and 5) for 5 different storms during the summer of 2004 (“storm sampling”). Finally, samples were collected every 2 h over the course of a 24 h period of baseflow at sites 3 and 5 on 5 different occasions during the summer of 2004 (“diurnal sampling”). The 2 sites selected for the intensive storm and diurnal VOL. 40, NO. 16, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4991
TABLE 1. Size and Land Use Statistics for Sampled Watersheds land use (%) area water/ open/ site (km2) forest agriculture urban wetland park residential 1 2 3 4 5 6 7 8 9 10 11 12
33.8 110.4 38.3 36.1 330.8 106.4 16.5 87.0 193.1 135.9 51.2 528.1
83.5 80.1 81.4 91.3 74.3 78.6 83.3 78.4 70.7 74.4 75.8 75.7
8.0 12.3 12.9 5.4 10.1 8.6 5.7 9.1 12.6 13.5 9.3 10.4
0.5 0.2 0.0 0.3 2.4 2.4 2.9 2.3 2.5 0.7 1.5 1.8
0.1 0.3 0.4 0.6 3.1 5.3 4.8 5.5 2.3 3.0 5.4 2.4
2.4 3.8 2.5 1.0 3.1 0.5 0.7 0.3 4.6 2.7 3.0 3.3
5.6 3.2 2.8 1.5 7.0 4.6 2.7 4.4 7.5 5.6 5.0 6.4
sampling represent a small, forested/agricultural watershed (site 3) and a larger watershed with more urban and residential land use (site 5). Samples were analyzed for E. coli concentrations using the Colilert system from Idexx Laboratories, which uses defined substrate technology to identify E. coli based on the ability to cleave 4-methyl-umbelliferyl-β-D-glucuronide (MUG). This EPA-approved method provides a most probable number (MPN) of E. coli by incubating multiple wells and determining how many contain E. coli. Duplicate samples were collected with a 14% frequency (n ) 66), and had an average coefficient of variation of 19%. Field blanks were collected with a 7% frequency (n ) 33) and were always negative, except for one occasion early in the project when samples were collected in autoclaved, previously used bottles and a low level of E. coli (5.2 MPN/100 mL) was detected (subsequent samples were collected in disposable bottles). Six samples exceeded the upper range of the method (i.e., >2419.6 MPN per 100 mL) and are reported as 2419.6 MPN per 100 mL. In addition to E. coli concentration, stream temperature (from a thermometer) and water level (from the staff gauges) were recorded for each sample. Weather records (daily precipitation) for Albany, NY were obtained from NOAA’s National Climate Data Center. All statistical analyses were carried out in MiniTab. We carried out best subsets regression to select the variables that were most important in controlling E. coli concentrations, and then used these variables to construct general linear models (GLMs) for each type of sampling. A GLM is an extension of a multiple regression model that can include both continuous and categorical predictors. Because raw data were not normally distributed (Anderson-Darling test), all models used log-transformed E. coli concentrations (which were normally distributed). Because of data limitations, the models were not validated with data independent of the data used for training.
FIGURE 2. Box plot of data from seasonal sampling (n ) 19 for each site; 18 for sites 2, 11, and 12). Levels indicated for each site are 10th, 25th, 50th, 75th, and 90th percentiles, as well as outliers (asterisks) and geometric means (crossed circles). WQS-1 corresponds to EPA’s single-sample water quality standard (WQS) for moderate, full body contact recreation (298 colonies per 100 mL), while WQS-2 corresponds to EPA’s geometric mean WQS (126 colonies per 100 mL).
Results Seasonal Sampling. Figure 2 summarizes E. coli levels from biweekly sampling over 10 months at each of the 12 sampling sites. Concentrations were generally higher in more developed sub-basins, specifically sites 5, 9, 10, 11, and 12. In addition, there was an apparent seasonal pattern to E. coli concentrations, as illustrated for sites 3 and 5 in Figure 3 (the same general pattern was observed at the other sites as well). E. coli density for baseflow samples generally follows the same seasonal pattern as temperature, increasing from spring to summer, and then decreasing again in the fall. Wet-weather samples appear as spikes in concentration relative to baseflow. 4992
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 16, 2006
FIGURE 3. (top) Daily streamflow (USGS, site 5) and precipitation (NOAA, Albany, NY) over the period of sampling. (bottom) Seasonal sampling data for sites 3 and 5, along with water temperature at site 5; wet weather samples (defined as rainfall within the 24 h prior to sampling) are shown as open symbols. A regression model was able to predict seasonal-sampling E. coli levels with an adjusted r 2 of 56% (Table 2). The patterns mentioned above were borne out by the model, in the form of positive relationships between E. coli concentrations and residential land cover, the occurrence of a recent rain event, and temperature, and a negative relationship between E. coli and forest cover. Note that using antecedent dry weather
TABLE 2. Model Variables, Coefficients, and p Values for a Regression Model of the Seasonal Sampling Data (Adjusted r 2 ) 56%) variable
coefficient
p
residential cover (%) forest cover (%) occurrence of rain event during the previous 24 h (Boolean) temperature (°C)
11.46 -4.22 0.618
0.002 0.001
