A multivariable approach for evaluating major impacts on water quality in Murrells and North Inlets, South Carolina.
KEY WORDS: regression, Escherichia coli, estuaries, southeast, septic system
Coastal marshes and estuaries serve as nursery and spawning grounds for numerous aquatic species. For instance, it is estimated that at least 70% of recreationally important and 96% of commercial shellfish and fish of the southeastern United States use estuaries and near-shore marine habitats at some point in their life cycle (Fulton et al. 1993). Specifically, marsh grasses in the wetlands provide shelter and nourishment for growing larvae and juvenile invertebrates and fishes (Charleston Harbor Project 1992). Also, the more than 300 estuaries in the southeastern and Gulf coasts of the United States, Puerto Rico and the Virgin Islands support $3.4 billion in recreational and over $850 million in commercial fisheries (Fulton et al. 1993). Over the last couple of decades, the United States has become increasingly dependent on imported shellfish. Specifically, United States imports of molluscan shellfish have grown from 20 million pounds in 1970 to 57 million pounds in 1990. In spite of inflation, the actual value of total United States harvests of scallops, clams and oysters has declined from $368 million in 1985 to $360 million in 1989. This pattern has held in spite of restoration efforts such as selective breeding, hatchery operations and oyster reef replenishment (Leonard 1993, Leonard et al. 1991).
In addition, coastal areas of the United States are being rapidly developed. It had been estimated that by the year 2000, 75% of the United States population would live within 50 miles of the ocean coastline and the great lakes (Vernberg et al. 1992). In the southeast, the number of people has more than doubled from 4 million in 1960 to 9.1 million in 1990 (Culliton et al. 1990). Encompassing the South Carolina (SC) study sites, Georgetown and Horry counties experienced a substantial growth in population during recent decades. Based on census data, population figures for Georgetown County in 10-year increments were 34,798 (1960), 33,500 (1970), 42,461 (1980) and 46,802 respectively (1990). For Horry County, analogous population figures were 68,247 (1960), 69,992 (1970), 101,419 (1980) and 144,053 (1990) respectively (United States Dept. of Commerce, Bureau of the Census 1960-1990). This change in population distributions is expected to have a harmful impact on coastal ecosystems. Most believe that, other things being equal, an increase in the concentrations of enteric pathogens is typically associated with increasing densities of human populations (Maiolo & Tschetter 1981). Several large monitoring studies in coastal areas have shown increased contaminant concentrations near urban areas (O'Connor 1996, Daskalakis & O'Connor 1995, Sericano et al. 1993). One impact of urbanization on coastal wetlands is the loading of human bacterial pathogens into wetland waters and the subsequent contamination of aquatic species and the lowering of water quality (Blood et al. 1992, USEPA 1988). Consequently, there have been reductions in the amount of domestic seafood that can be harvested commercially and an increase in our dependence on imported seafood. Similarly, there have been restrictions of recreational uses of these nearshore waters (e.g., recreational gathering of shellfish or swimming; Weiskel et al. 1996, Maiolo & Tschetter 1981).
Urbanization poses a particular threat to those coastal areas of the southeastern United States (i.e., parts of South Carolina and Georgia) where the lands surrounding the wetlands are still undeveloped compared with other regions. The estuaries there also tend to be shallow and the rivers are not of sufficiently large volume to flush out contaminants compared with other parts of the country (Vernberg et al. 1992, Leonard et al. 1991). Problems associated with ground absorption, (e.g., septic systems) and overburdened central treatment facilities have become critical in coastal communities, particularly in the Southeastern United States. This area has been noted as being of major importance in the coastal area management act (CAMA) in North Carolina 1974 (Maiolo & Tschetter 1981). Elevated levels of fecal coliform bacteria have been associated with urbanization and human activities. Some research has showed that fecal bacteria densities were directly related to housing density, population, development, apparent animal density and percent impervious area (Young & Thackston 1999). The coliform group consists of all the aerobic and facultative anaerobic gram negative, rod-shaped, nonspore forming bacteria, which ferment lactose with the formation of gas within a 48 h period at 35[degrees]C (Koren 1991). At the present time, the fecal coliform bacterial group is the standard for fecal pollution in surface waters and is comprised of species that reside in the intestines of warm-blooded mammals and birds (Dufour 1977). This group includes Escherichia coli (E. coli), Escherichia aurescens, Escherichia intermedia, Escherichia freundii, Aerobacter aloacae, Aerobacter aerogenes and the biochemical intermediates between Escherichia and Aerobacter (Koren 1991). Fecal coliforms are important indicators of public health. On the average, [10.sup.10) E. coli organisms are generated per human per day (Edberg et al. 1994). The threat to human health occurs when human and/or animal feces come in contact with and contaminate drinking water supplies, recreational waters and/or growing waters of filter-feeding shellfish (Gersberg et al. 1995, Zonderman & Shader 1993). Shellfish absorb food by means of a filtering process that has a low level of selectivity. Besides food, suspended sediments, toxic algae, bacteria and viruses may also be absorbed. Thus, shellfish can be contaminated with various disease-causing microorganisms residing in overlying water, especially in shallow water localities. These contaminants can become concentrated in molluscs at levels from 3 to 20 times that found in surrounding waters (Koren 1991, Kuenstner 1991, Ballentine 1985, Wood 1976). The degradation of water quality from fecal contamination often leads to the closure of shellfish harvesting waters (ISSC 1997). In the United States, point and nonpoint sources each account for about 38% of all closures of shellfish harvesting areas (Leonard et al. 1991). Pollutants of the latter category may be distinguished from the former because they seep or run off into waterways instead of entering the water through a discrete conduit or discharge (Nadakavukaren 1990). Examples of nonpoint sources of contamination include agricultural runoff, malfunctioning or poorly located septic systems, runoff from vegetated areas (e.g., wetlands) and deposition directly by waterfowl feces (Weiskel et al. 1996). Compared with point sources, nonpoint sources are more difficult to control because of problems in localizing and identifying their origins (Leonard et al. 1991). The most important nonpoint source of pollutants are septic systems, also known as ground absorption treatment, typically found in outlying rural and beach areas (Maiolo & Tschetter 1981). Both sources are associated with a seasonal population influx, an expansion in tourism and development in housing and accompanying infrastructure (e.g., roadways, solid waste and human waste disposal; Leonard et al. 1991).
The measurement of the concentration of fecal coliforms in the vicinity of a given molluscan shellfish bed is currently the major criterion for deciding when and if shellfish harvesting should be approved for human consumption (Dufour 1977). However, the current regulations could be improved by incorporating existing meteorological and land use change data into predictive models. Moreover, they do not anticipate the potential downgrading of approved harvest areas that may arise from high rates of coastal development. What are needed are predictive methods that would correlate information on land use change/development so that downgrades in water quality can be predicted before they occur and corrective development/environmental management approach can be made to prevent water quality degradation. This approach used for this study involved an historical comparison of land use change and fecal coliform bacterial densities on Murrells Inlet (MI) and North Inlet (NI), determining important benchmark development dates, developing a database and appropriate models to which to test hypotheses or investigate research questions and then performing statistical analyses on these data.
MATERIALS AND METHODS
Study and Control Sites
Murrells Inlet (33[degrees]33'N, 79[degrees]01'W) is a highly suburbanized, bar-built estuary located approximately 32 km north of NI and 90 km north of Charleston, SC and covers sections of Horry and Georgetown Counties (Fig. 1, Table 1). The northern two-thirds of the estuary consist largely of commercial, residential and tourism related land uses, whereas the remaining part is predominantly undeveloped state park (Newell 1997). MI has experienced a high rate of development in recent years. Its population of permanent residents had increased 230% (from 1,000-3,300 people) during the 1980s and had one of the highest population densities in SC as of 1990. In MI there are neither discharges of standard industrial code industries nor agricultural runoff into shellfish harvesting waters. As a result, MI was selected for inclusion in this study because there is a negligible amount of inputs from these sources. Thus, the vast majority of closures would be because of fecal coliforms from suburban housing, marinas and service industries (Fulton et al. 1993), with wildlife sources being negligible, except on the southern end of the estuary that is maintained as a state recreational area.
North Inlet (33[degrees]20'N, 79[degrees]10'W) served as a pristine reference site, relatively free from human agricultural, urban or industrial development (Fig. 1). Nevertheless, wildlife there has significantly affected the numbers of fecal coliform bacteria that are loaded into the estuary through natural forest runoff. NI is a bar-built estuary that is bounded to the northeast by Debidue Island and to the southeast by North Island (Fulton et al. 1993, Vernberg et al. 1992, Newell 1990). It is made up of stable and temporary habitats that are linked by a tidally fluctuating water height. The major functional and structural habitats of NI consist of a water column (tidal creeks and pools), oyster reefs, a subtidal benthos (the whole assemblage of plants or animals living on the sea bottom) and a vegetated marsh surface and subsurface (Vernberg et al. 1992). NI is rather unique because private institutions (Yawkey Foundation and Belle W. Baruch Foundation) own large tracts of this estuary (Vernberg et al. 1992). NI has also been designated by the national oceanic and atmospheric administration (NOAA) as a national estuaries research reserve (NERR) site. This has limited developmental activity and probably contributed to the overall pristine condition of NI. Because there is a negligible level of industrial or agricultural activity and because red tides are a very infrequent occurrence on the SC coastline, nearly all the closures of harvesting areas in this estuary could be attributed to fecal coliforms that were primarily from wildlife sources (Fulton et al. 1993).
The microbiological and water quality data used in this research covers the period of 1967-1995 and were available on a Storet CD-ROM that was compiled by the SC department of health and environmental control (DHEC) this data was also utilized in (Nelson 1998) from which much of the present work is drawn. Among the parameters obtained were the time and date of sampling, most probable number (MPN) of fecal coliform bacteria, salinity (parts per thousand) and the water temperature (degrees Celsius). Daily rainfall records for the study period have been recorded by the national weather service at Brookgreen Gardens (BG) (near MI and NI) and at Beaufort, SC. These two stations were found to have a high correlation (r = 0.763, P < 0.0001) so it is likely that the precipitation patterns at BG were a better approximation to those at MI and NI because these sites are much closer to BG than is Beaufort. The BG site is approximately 4 km from MI and 28 km from NI. Moreover, the relevant information on the history of urban development of the respective estuaries came from Mary Culberson (Georgetown Water and Sewer) and Charles Newell (DHEC Myrtle Beach). The principal statistical method for analyzing the available data was intervention analysis. Because the data is of a temporal nature, it was important to focus on large-scale events in the developmental history of a given estuary and examine how the response variable (i.e., fecal coliform (FEC) bacteria concentration) was affected by important developmental events. There are several types of intervention models that are summarized in Table 2. Figure 2 A depicts a theoretical example of the Mean-Shift intervention model. This is a sudden-impact model with no trend before the date of interest (the intervention date). The striking feature of this model is that there is no chronologic trend but there is abrupt increase in the parameter of interest (e.g., fecal coliform bacterial density, coinciding with a 1980 intervention). Its form may resemble the following equation:
[log.sub.10] (FEC) = [[alpha].sub.1] Salin + [[alpha].sub.2] Temp + [[alpha].sub.3] [[log.sub.e] (Rain) + [[alpha].sub.4] Il + [epsilon] (1)
[log.sub.10] (FEC) = transformed MPN fecal coliform concentration
Salin = salinity
Temp = water temperature
Rain = rainfall
[FIGURE 2 OMITTED]
Il = indicator variable of date (1 if the date occurs after the intervention, 0 otherwise)
The Il term of the equation captures the effect of this jump. Bacterial MPN concentration and rainfall are logarithmically transformed because they are skewed to the right and deviate substantially from normality.
Figure 2B depicts an example of the Trend-Plus-Jump Model.
This is a sudden impact model with an underlying increasing trend. Again, there is an abrupt increase in the parameter of interest mean [Log.sub.10] FEC because of the 1980 intervention. The general equation for this model has the following form:
[log.sub.10] (FEC) = [[alpha].sub.1] Salin + [[alpha].sub.2] Temp + [[alpha].sub.3] [log.sub.e] (Rain) + [[alpha].sub.4] Date + [[alpha].sub.5] Il + [epsilon] (2)
Here, the additional Date term captures the trend effect, whereas the Il term is representative of the sudden increased effect.
Figure 2C depicts the Change-In-Trend model. This model allows for the assessment of a sudden impact and/or change in trend. Figure 2C shows an abrupt increase in the parameter of interest mean [log.sub.10] FEC with the 1980 intervention. At the same time, an increasing pattern of [log.sub.10] FEC ceases after 1980. A general equation for this would take a form as follows:
[log.sub.10] (FEC) = [[alpha].sub.1] + [[alpha].sub.2]Temp + [[alpha].sub.3] [log..sub.e] (Rain) + [[alpha].sub.4]Date + [[alpha].sub.5] Il + [[alpha].sub.6] Il Date + [epsilon] (3)
Here, the interaction term I1Date accounts for the change in trend. Equation 3 is a desirable model for the water quality data because it is the most general and powerful as the following cases demonstrate:
a) If [[alpha].sub.4] [not equal to] 0, [[alpha].sub.5] = 0 and [[alpha].sub.6] = 0, then there is a Constant-Trend Model (no intervention).
b) If [[alpha].sub.4] = 0, [[alpha].sub.5] [not equal to] 0 and [[alpha].sub.6] = 0, then we have a Mean-Shift Model (no trend).
c) If [[alpha].sub.4] [not equal to] 0, [[alpha].sub.5] [not equal to] 0 and [[alpha].sub.6] = 0, then we have a Trend- Plus-Jump Model.
d) If [[alpha].sub.4] [not equal to] 0, [[alpha].sub.5] [not equal to] 0 and [[alpha].sub.6] [not equal to] 0, Change in-Trend Model
After evaluating fecal coliform data compiled before the intervention date of 1980 in MI, it was concluded that the Change-in-Trend Model was the appropriate focus in the analyses because it is the only model that accounts for long-term trends in water quality and short-term changes that are attributable to an intervention. It accommodates the instantaneous and gradual changes that a given intervention would have on the concentration of fecal coliform bacteria. All models used the logarithm transformation for normalizing the right-skewed distributions of coliform bacteria and rainfall. Some of the observations were outside the detectable range of the microbiologic multiple tube dilution test (generally 2 to 2,400 MPN bacteria per 100 mL). For convenience, extreme readings outside the detectable range of the fecal coliform estimates were conservatively estimated by adding or subtracting one unit above or below, respectively, the limits of detection. Thus, a measure that was <2 MPN was coded as 1, whereas a reading that was >2400 MPN was coded as 2401.
On MI, three land use development events in the period 1967 to 1995 were identified as having the potential to affect coliform concentrations (Table 3). The first was the completion and conversion of most of the septic tanks to the central sewer line by April 1980 (Fred Earnhart, Georgetown Water and Sewer, Georgetown, SC, pers. comm.). This would be expected to decrease E. coli concentrations because septic tanks periodically leak and thus discharge fecal coliform bacteria into the waterways. Second, there was a completion of the construction of a large jetty by August 1981 that had begun in September 1977 (Dr. Dave Bushack, Belle W. Baruch Laboratory, Georgetown, SC, pers. comm.). In contrast to the sewer conversion, the jetty construction was predicted to increase coliform and decrease salinity levels by reducing the tidal amplitude of the estuary because the jetty blocks some of the intrusion of clean ocean water into the estuary. Offshore oceanic water dilutes and thus reduces the concentration of fecal coliform contaminants in the estuary. It was not clear which of these events was dominant. Because both events were so close, chronologically, and because there were few samples taken between them, the date 6/15/80 was selected to account for both events. The third key intervention event was the completion of the construction of public boat ramp in January 1986. A boat ramp increased the volume of boat traffic and would probably be an important source of pollutants because many boats may potentially release marine sanitation debris into estuarine surface water. Thus, 1/15/86 was selected as the intervention date for assessing effects of boat ramp construction.
For NI, there has been little anthropogenic activity during the time of data collection. Nonetheless, two potential sources of contamination are outlined in Table 3. These were the completion of the construction of Baruch Laboratory in June 1973. Thus, the date 6/15/73 was selected. Later, the construction of the Debidue housing development began in the middle part of the 1970s. An intervention date of 1/15/77 was selected to approximate the completion of the initial phase of residential development at Debidue Island near NI. Whereas pristine, it is important to realize that adjacent to NI, and to the south is Winyah Bay (WB). Unlike NI, WB is a partially mixed and low salinity estuary that receives fresh water from the Black, Pee Dee and Waccamaw Rivers that drain extensive agricultural areas. The relatively high flow rates of these rivers enables anthropogenic wastes to be readily transported from nearby industrialized Georgetown (Lackland et al. 1982). In the analysis, some water sampling stations on the southern end of NI near WB were excluded because they were located within the tidal node for WB.
Descriptive statistical outputs (monthly/yearly) of fecal coliform bacteria and rainfall data were coupled for NI and MI to indicate the trend in each parameter over time. In addition, regression analysis was conducted to assess the variables affecting fecal coliform bacteria densities such as rainfall and physiochemical water quality.
Yearly Trends in Fecal Coliform Bacteria
It is of interest to visually describe the yearly distributions of mean coliforms for each inlet with the distributions for rainfall at BG. Because BG is close to both study sites, one can see how the distribution of bacteria compares to that of rainfall depicted in Figure 3 and Figure 4. To avoid confusion, it should be noted that the zero bacteria concentration readings for MI in 1968 and for NI in 1969 and 1971 are caused by a lack of measurements taken during those years, rather than pristine water quality. In MI, note that fecal coliform densities were generally higher after the intervention than before. However, levels were not statistically different. In NI, fecal coliform densities were similar before and after the intervention dates. For these graphs, it is a bit difficult to visually assess the intervention. For instance, in MI mean annual bacteria levels were somewhat higher in the period following the intervention. However, the standard deviation for this variable is sufficiently large that one cannot say for certain how bacteria levels were affected by the intervention (i.e., there was no statistically significant response). Because the analytic approach accounts for the potential influence of other variables and underlying trends in the data, it should be given more emphasis than the graphical approach depicted in Figures 3 and 4. These data show inconsistencies regarding the effect of the interventions. For NI, the limited number of sampling years before and after the intervention, respectively, makes it more difficult to draw conclusions about trends in the data. As mentioned, large rainfall events have been observed to elevate bacterial concentrations. The correlation coefficients (r) between loge (rainfall) and [log.sub.10] (coliform) concentrations generally support this observation, although some of these are weak but significant. For the MI station Main Creek at Atlantic Ave Bridge, r was found to be 0.363 (P = 0.0004) and for MI station Main Creek at Mickey Spillane's Home, an r = 0.436 (P < 0.0001) was observed. Similarly, for NI station North Inlet, an r = 0.251 (P = 0.0408) was observed and for NI station Town Creek at Debidue Creek, an r = 0.255 (P = 0.0388) was found. It should be noted that the correlations between rainfall and fecal coliform counts were considerably stronger and more significant for MI than for NI. This is suggestive of a land use effect. In more developed areas rainfall/runoff would have more of an effect than a largely wooded area like North Inlet. Developed areas have more impervious surface coverage and less opportunity for rainfall to percolate into the ground; thus, greater pollution impacts.
[FIGURES 3-4 OMITTED]
Monthly Trends in Fecal Coliform Bacteria
For the monthly trends in fecal coliform bacterial density graphs (Figs. 5, 6), it can be seen that for MI, the before intervention (BI) fecal coliform bacteria densities peak in January, April and July and the after intervention (AI) densities peak in June and October. Rainfall appears to peak in January and the summer, a time when there is peak tourism. In comparison, for NI, the BI bacteria concentrations peak in January, April and September, whereas AI bacteria concentrations peak in March, April and November. The January and April coliform peaks may be attributed in part to winter waterfowl, a high water table condition, high wildlife activity, low wildlife feces degradation rates and a low rate of plant transpiration. Paired t-tests were run to compare the monthly mean coliform concentrations before and after the intervention. For MI, significant differences in coliform levels were found for January (P = 0.0188, BI > AI), March (P = 0.0051, BI > AI), May (P < 0.0001, AI > BI), June (P < 0.0001, AI > BI), October (P < 0.0001, AI > BI) and November (P = 0.0053, AI > BI). For NI, the mean bacterial densities also differed significantly for January (P = 0.0026, BI > AI), March (P = 0.0056, AI > BI), April (P = 0.0058, BI > AI) and September (P < 0.0001, BI > AI). The temporal similarity of January and March differences before and after the intervention indicated that similar variables (e.g., seasonal rainfall, high water table conditions and low feces degradation rates for wildlife) may affect fecal coliform densities in each estuary. Although the months where significant differences in the means varied in each inlet, the interventions seemed to influence the seasonal distributions of coliforms for MI and NI. As with the yearly plots, there were apparently 0 bacterial concentration readings for May and June during the period before the intervention, because no samples were collected for those months.
[FIGURES 5-6 OMITTED]
The intervention Change-in-Trend model was applied to MI where the individual station data were pooled. Several runs were done for the reduced model and models with different combinations of interactions. The 1986 intervention (boat ramp) proved not to be significant. The criterion that was used for determining a good model fit was a comparatively high adjusted [R.sup.2] (>0.25) and a low P value (<0.05) for the intervention parameter. After running regressions with the reduced model and models involving several interaction terms, the best-fitting model included rain-salinity interaction and found a significant change in trend resulting from the 1980 intervention. Table 4 gives a table of these analytic results. Here, date was significant (P = 0.0001), indicating an increasing trend in bacteria concentration over the period of data collection. I1 Date was also significant (P = 0.0001), suggesting a decrease in the increasing trend of coliform bacteria. I1 was marginally significant (P = 0.0573) suggesting the possibility for an instantaneous decline in bacteria concurrent with the intervention. When the earlier mentioned model was applied to the individual sampling stations, the following sites were found to have significantly changed (decreased fecal coliforms) after the intervention: Main Creek at Marlin Quay Marina (P = 0.0323), Main Creek at Inlet Range Marker (P = 0.0337), Allston Creek at Weston Flat (P = 0.0001), Allston Creek at Pogs Head Landing (P = 0.0017) and Parsonnage Creek at Nance's Dock (P = 0.0008). These stations can be seen spatially in Figure 7. Note that, spatially, most of the improved water quality was observed at stations nearest the mouth of the estuary, closest to clean, higher salinity oceanic seawater. When stratified by locality (inner [farthest], middle and outer [closest]), salinity was observed to increase approaching the open ocean: the mean salinity levels (parts per thousand) were 30.9 (inner), 34.3 (middle) and 35.0 (outer). Although regression analyses point out anticipated changes in water quality in response to a given intervention, in reality one would expect estuarine systems to respond nonlinearly to a given intervention. Figure 8 shows a possible interpretation for how water quality may have changed at Murrells Inlet. Here, a cubic function was used to fit the yearly mean coliform concentrations:
MLFEC = -904078.2 + 1364.51 x Year - 0.686469 x [Year.sup.2] + 0.000115 x [Year.sup.3] (4)
[FIGURES 7-8 OMITTED]
This shows a gradual rather than sudden decrease in bacteria concentrations after the intervention and this seems reasonable given what is known about the geophysical nature of water quality in coastal South Carolina. However, extrapolation outside the range of available data is uncertain because the wide 95% confidence limits for the mean predicted values in the plot illustrate. Figure 9 shows the plot of what MLFEC concentrations would have been assuming that no intervention had taken place. This takes on the following linear function:
MLFEC = -90.96471 + 0.046591 x Year (5)
[FIGURE 9 OMITTED]
Compared with the previous figure, it is clear that bacteria concentrations would have been higher in the absence of the septic tank/sewer convergence compared with that of the intervention. In particular, the expected MLFEC for 1995 without the intervention would be 1.98, whereas the corresponding value for the cubic model would be 1.08 with the intervention (a 45% reduction).
The methodology outlined earlier was also applied to NI where all the individual station data of estuaries were combined. Initially, a model was run containing the Date, I1, and I1 Date terms. However, this model had high variance inflation for some of the terms in the model. This indicated a probable multicollinearity among the independent parameters and the results would have to be interpreted with caution. At this point, the sudden impact term I1 was dropped, whereas I1 Date was retained because estuaries would be expected to respond gradually instead of instantaneously to a given intervention. The 1977 intervention proved not to be significant (P = 0.9191). The best-fitting model included rain-salinity interaction and found a marginally insignificant change in trend resulting from the 1973 intervention (P = 0.1189). Hence the 1973 intervention was used, whereas the 1977 intervention was not. The results from the regression model for the 1973 intervention may be found in Table 5. When the earlier mentioned model was applied to the individual sampling stations, only Town Creek at Sixty Bass Creek was statistically significant (P = 0.0285) (Fig. 10). This site is close to the tidal node from Winyah Bay and may indicate an increase in the rate of decrease in fecal coliform densities since 1973.
For MI, the overall inlet results are generally what one might expect. For an urbanized estuary, it was easier to detect improvements in water quality than for a pristine inlet. These results were consistent with the current water quality classification scheme in that rainfall was found to be an important variable in determining coliform concentrations and large rainfall events may contribute to closure of shellfish harvesting waters immediately after an event. When there was such a high level of development, anthropogenic inputs to the system may dominate over natural sources. Vernberg et al. (1997, 1999) found that fecal coliform bacteria in MI bad a much higher percentage of E. coli comprising the coliform group (83%) and only a small percentage of samples were free of coliforms (<5%) when compared with NI (53% E. coli and 23% coliform free respectively). This implies that MI is a greater potential source of human bacterial pollution than NI. Chestnut et al. (2000) have also shown that other urban areas of South Carolina also have high levels of E. coli comprising the fecal coliform group. Thus, the model was able to pick up an improvement in water quality resulting from the septic tank to sewer conversion. Although the jetty construction would be expected to act in the opposite direction (e.g., increase coliform counts caused by poorer flushing conditions), our data suggested that the septic tank to sewer conversion would likely have had a more pronounced impact on water quality than the jetty construction, because fecal coliform counts decreased based on our nonlinear model. Additionally the improvements in water quality in MI were most pronounced at the stations near the mouth of the estuary, near cleaner oceanic water.
Recent research by Kelsey et al. (2003) used multiple antibiotic resistance (MAR) at MI to help identify fecal coliform by source. This research suggests that most of the fecal coliform pollution detected in MI estuary may be from nonhuman sources, including fecal coliforms found in areas adjacent to high densities of active septic tanks. For this study, human and pet wasteloads were ascertained for MI and it is estimated that the human to pet wasteload is 0.14. The process by which this figure was found is highlighted in Table 6. It has also been estimated that about 90% of residences in MI are currently connected to sewer lines. The product of these two coefficients is about 0.12 that is strikingly similar to the sudden decrease in bacteria concentrations highlighted in Table 4 (-0.12). This helps to validate our model and shows that manmade changes in land use for MI can affect water quality in spite of the fact that domestic animals and wildlife account for most of the bacteria concentration in this system.
Similarly, the findings at NI were not too surprising. Here, the construction of Baruch Laboratory and Debidue development only accounted for a small number of septic tanks and human inputs into the ecosystem. Thus, the background levels of bacteria contamination from natural wildlife sources would clearly outweigh the magnitude of contamination from any limited watershed development. In analyzing fecal coliform data for NI and MI over the entire 1967 to 1995 period, the fecal coliform density of NI should reflect primarily bacterial inputs from wildlife sources. Based on NI levels of fecal coliform representing wildlife sources (mean [log.sub.10] 0.83), mean fecal coliform densities in urbanized MI (mean [log.sub.10] 1.052) were only elevated by 27%. This implies the anthropogenic sources of fecal coliform bacteria may only account for 27% of the total fecal coliform burden in MI and NI. Also, natural fluctuations resulting from seasonal patterns in rainfall, salinity and temperature would potentially confound the effects from the intervention itself. Evidence of this was seen in the fact that the same monthly differences were seen before and after the interventions in NI and MI (January, March, April and September). Thus, rainfall and water table level(s) should be considered to be major factors affecting fecal coliform densities in each estuary.
Improved data collection and analysis may improve validity and applicability of the models in other ways. For example, localized variability in water currents may have affected how the coliforms responded to an intervention and other parameters, and this may also account for some of the unexplained variation of the models. Utilizing spatial procedures such as kriging could overcome some of these difficulties. Also, excluded variables for which there were no data, could account for much of the unexplained variation. For example, pH can affect the survival of bacteria. Generally, coliforms die more readily under alkaline than acidic conditions. The combination of salinity and the 8.0 pH of seawater contribute to the rapid death of coliforms that enter the oceans via sewage outfalls and land drainage (Carlucci & Pramer 1960). The bactericidal efficiency of seawater is influenced by season, being highest in the summer and lowest in the winter (Vaccaro et al. 1950). Second, deficiencies of the inorganic nutrients of phosphorous and nitrogen are known to limit matter decomposition and bacteria development in sea water. In particular, a medium made of high levels of the inorganic nutrients of [(N[H.sub.4]).sub.2]HP[O.sub.4] and [(N[H.sub.4]).sub.2]S[O.sub.4] enable E. coli to survive (Carlucci & Pramer 1960). It is also well known that coliforms are facultative anaerobic organisms that do not directly depend on oxygen for their survival. Nonetheless, dissolved oxygen may be an important excluded variable because it has a high negative correlation with water temperature and a high negative correlation with time of year (Gross & Folts 1973).
Clearly, the earlier mentioned findings are most applicable to estuaries and intercoastal waterway along the South Carolina coastline. It would be most useful for those regulators in the region who may be concerned that a high level of housing development, over a relatively small area, may lead to a larger proportion of shellfish harvesting water closures. Results from this study have clearly indicated the utility of intervention analysis for assessing long-term water quality trends of fecal coliform pollution in shellfish harvesting areas, in both pristine and urbanized estuaries. The usefulness of this method is quite obvious and the potential to forecast future trends in bacterial water quality may be possible, especially if linkages between fecal coliform density trends and urban census tract data can be established with statistical accuracy. Future studies should attempt to better line (via GIS) fecal coliform trends with other land use data such as percentage impervious surface and percentage vegetative cover data. To be able to apply this methodology to other areas, other information may need to be incorporated into the model such as freshwater input from rivers and streams, toxic chemicals discharged from non-point sources, housing census tract data and other land use data.
Ballentine, C. 1985. For oyster and clam lovers, the water must be clean. Department of Health and Human Services--Food and Drug Administration. HHS Publication No. (FDA) 85-2200. Rockville, MD.
Blood, E., T. Chandler, A. Fortner, M. Fulton, H. McKellar, W. Michener, G. Scott, and T. Siewicki. 1992. Urbanization and southeastern systems (USES): research proposal and progress report. The Belle W. Baruch Institute for Marine Biology and Coastal Research University of South Carolina and National Marine Fisheries, NOAA Southeastern Fisheries Science Center.
Carlucci, A. F. & D. Pramer. 1960. An evaluation of factors affecting the survival of Escherichia coli in sea water. Appl. Microbiol. 7:247-250. Charleston Harbor Project. 1992. A day in the life of the harbor. Charleston, SC.
Chestnut, D. R., C. R. Sansbury, R. F. Van Dolah & G. I. Scott. 2000. An ecotoxicological characterization of the Broad Creek and Okatee River watersheds in Beaufort County, South Carolina. SC Dept of Health and Environmental Control, Final Report (In press).
Culliton, T. J., M. A. Warren, T. R. Goodspeed, D. G. Remer, C. M. Blackwell & J. J. McDonough, III. 1990. Fifty-years of population change along the nation's coasts 1960 to 2010. The second report of a coastal trends series. National oceanic and atmospheric administration. Rockville, MD.
Daskalakis, K. D. & T. P. O'Connor. 1995. Normalization and elemental sediment contamination in the United States. Environ. Sci. Technol. 29:470-477.
Dufour, A.P. 1977. Escherichia coli: The fecal coliform. In: A.W. Hoadley & B. J. Dutka, editor. Bacterial indicators/health hazards associated with water. ASTM STP635. American society for testing and materials.
Edberg, S. C., J. E. Patterson & D. B. Smith. 1994. Differentiation of distribution systems, source water, and clinical coliforms by DNA analysis. J. Clin. Microbiol. 32(1):139-142.
Fulton, M. H., G. I. Scott, A. Fortner, T. F. Bidleman & B. Myabe. 1993. The effects of urbanization on small high salinity estuaries of the Southeastern United States. Arch. Environ. Toxicol. 25:476-484.
Gersberg, R., M. Matkovits, D. Dodge, T. McPherson & J. Boland. 1995. Experimental opening of a coastal California lagoon: effect on bacteriological quality of ocean waters. J. Environ. Health 58(2):24.
Gross, A. J. & J. R. Folts. 1973. Analysis of selected, existing water quality data on the Connecticut River. Completion report FY-73-2. Water resources research center. University of Massachusetts. Amherst, MA.
ISSC. 1997. Interstate shellfish sanitation conference guide for the control of molluscan shellfish. United States Department of Health and Human Services, Public Health Service, Food and Drug Administration. 1997 Revision.
Kelsey, R. H., G. L. Scott, D. E. Porter, B. Thompson & L. Webster. 2003. Using multiple antibiotic resistance and land use characteristics to determine sources of fecal coliform bacterial pollution. Environ. Monit. Assess. 81:337-348.
Koren, H. 1991. Handbook of environmental health and safety principles and practices vol. 2. Chelsea, MI: Lewis Publishers, Inc.
Kuenstner, S. 1991. Seafood and health: risks and prevention of seafood-home Illness. New England Fisheries Development Association, Inc. Boston, MA.
Lackland, D.T., E.H. Liu, D.E. Koehler & T. H. Chrzanowski. 1982. Cellulase activity as a measure of environmental perturbation in salt marsh ecosystems. Botanica Marina 25:143-148.
Leonard, D. L., E. A. Slaughter, P. V. Genovese, S. L. Adamany & C. G. Clement. 1991. The 1990 National Shellfish Register of Classified Estuarine Waters. Report for United States department of commerce, NOAA, NOS.
Leonard, D.L. 1993. Turning the tide on water quality and declining shellfish resources. World Aquaculture 24(4):56-64.
Maiolo, J. P. & P. Tschetter. 1981. Relating population growth to shellfish bed closures: a case study from North Carolina. Coastal Zone Management Journal 9(1): 1-18.
Nadakavukaren, A. 1990. Man and environment: a health perspective. Prospect Heights, IL: Waveland Press, Inc. 530 pp.
Nelson, K. A. 1998. A multifactorial approach for describing the relationship between the classification of shellfish harvesting waters and adjacent land use. Dissertation. UMI dissertation services. Ann Arbor, MI. 121 pp.
Newell, C. L. 1990. Sanitary survey: area V: September 8, 1985 to May 9, 1990. South Carolina Department of Health and Environmental Control. Columbia, SC.
Newell, C. L. 1997. Shellfish management area IV; 1997 annual update. South Carolina Department of Health and Environmental Control. Columbia, SC.
O'Connor, T. P. 1996. Trends in chemical concentrations in mussels and oysters collected along the US coast from 1986 to 1993. Mar. Environ. Res. 41:183-200.
Sericano, J. L., T. L. Wade, J. M. Brooks, E. L. Atlas, R. R. Fay & D. L. Wilkinson. 1993. National status and trends mussel watch program: Chlordane-related compounds in Gulf of Mexico oysters, 1986, 1990. Environ. Pollut. 82:23-32.
United States Dept of Commerce. Economics and statistics administration, bureau of the census 1960. Census of population. General population characteristics. Washington, DC.
United States Dept of Commerce. Economics and statistics administration, bureau of the census 1970. census of population, general population characteristics. Washington, DC.
United States Dept of Commerce. Economics and statistics administration, Bureau of the census 1980. Census of population, general population characteristics. Washington, DC.
United States Dept of Commerce. Economics and statistics administration, bureau of the census 1990. Census of population. General population characteristics. Washington, DC.
United States Environmental Protection Agency. 1988. Environmental progress and challenges: EPA's update. Office of policy planning and evaluation. Washington, DC.
Vaccaro, R. F., M. P. Briggs, C. L. Carey & B. H. Ketchum. 1950. Viability of Escherichia coli in sea water. Am. J. Public Health 40:1257-1266.
Vernberg, F.J., W.B. Vernberg, E. Blood, A. Fortner, M. Fulton, H. McKellar, W. Michener, G. Scott, T. Siewicki and K. EL Figi. 1992. Impact of urbanization of high-salinity estuaries in the Southeastern United States. Neth. J. of Sea Res. 30:239-248.
Vernberg, W. B., G. I. Scott, S. H. Strozier, J. Bemiss & J. W. Daugomah. 1997. The effects of urbanization on human and ecosystem health.. In: F. J. Vernberg, W. B. Vernberg & T. Siewicki, editor. Sustainable development in the Southeastern Coastal Zone, vol. 20. Belle W. Baruch Library in Marine Science. pp. 221-240.
Vernberg, F.J., W.B. Vernberg, D.E. Porter, G.T. Chandler, H.N. McKellar, D. Tufford, T. Wiewicki, M. Fulton, G. I. Scott, D. Bushek & M. Wahl. 1999. Impact of coastal development on land-coastal waters. In: E. Ozhan, editor. Land ocean interactions: Managing coastal ecosystems. Ankara, Turkey: MEDCOAST, Middle East Technical University, pp. 612-622.
Weiskel, P. K., B. L. Howes & G. R. Heufelder. 1996. Coliform contamination of a coastal embayment: Sources and transport pathways. Environ. Sci. Technol. 30:1872-1881.
Wood, P. C. 1976. Guide to shellfish hygiene. World Health Organization, Geneva, Switzerland.
Young, K. D. & E. Thackston. 1999. Housing density and bacterial loading in urban streams. J. Environ. Eng. 125:1177-1180.
Zonderman, J. & L. Shader. 1993. Environmental diseases. New York: Henry Holt and Company, Inc.
KEVIN A. NELSON, (1) * GEOFF I. SCOTT (2) AND PHILIP F. RUST (3)
(1) Michigan Department of Community Health, Bureau of Epidemiology Administration, Lansing, Michigan 48909; (2) National Ocean Service, Charleston, South Carolina 29412; (3) Department of Biostatistics, Bioinformatics and Epidemiology, Medical University of South Carolina, Charleston, SC 29425
* Corresponding author. E-mail: firstname.lastname@example.org
TABLE 1. Description of sampling stations at NI and MI. DHEC Location Station# NI 1 Jones Creek at Nancy Creek 2 Noble Slough 3 North Inlet 4 Town Creek at Debidue Creek 5 Oyster Bay near Cut-off Creek 6 No Man's Friend Creek and Mud Bay 7 Jones Creek and Mud Bay 8 Town Creek at Sixty Bass Creek 9 Town Creek at Southern Entrance 10 Jones Creek at Duck Creek 11 Town Creek at Bread and Butter Creek MI 01 Main Creek at Atlantic Ave Bridge OIA Main Creek at Stanley Drive 2 Main Creek at Mickey Spillane's Home 3 Main Creek at Captain Dicks' Marina 4 Main Creek at Marlin Quay Marina 5 Murrells Inlet-Range Marker 6 Allston Creek at Weston Flat 7 Allston Creek at Pog-Hughes Landing 8 Parsonnage Creek at Nance's Dock 16 Parsonnage Creek at Chicken Farm Ditch DHEC Coordinates Station# NI 1 33[degrees]17'34.0", 79[degrees]10'51.0" 2 33[degrees]18'00.0", 79[degrees]11'05.0" 3 33[degrees]19'30.0", 79[degrees]09'50.0" 4 33[degrees]20'02.0", 79[degrees]10'00.0" 5 33[degrees]18'32.0", 79[degrees]12'09.0" 6 33[degrees]18'06.0", 79[degrees]12'12.0" 7 33[degrees]16'15.0", 79[degrees]12'04.0" 8 33[degrees]19'21.0", 79[degrees]11'35.0" 9 33[degrees]19'32.0",79[degrees]11'48.0" 10 33[degrees]18'35.0",79[degrees]10'50.0" 11 33[degrees]19'53.0",79[degrees]11'17.0" MI 01 33[degrees]34'46.0", 79[degrees]00'14.0" OIA 33[degrees]34'30.0", 79[degrees]00'31.0" 2 33[degrees]24'02.0", 79[degrees]01'17.0" 3 33[degrees]33'17.0", 79[degrees]01'49.0" 4 33[degrees]33'10.0", 79[degrees]01'19.0" 5 33[degrees]31'50.0", 79[degrees]02'12.0" 6 33[degrees]32'16.0", 79[degrees]02'54.0" 7 33[degrees]31'48.0", 79[degrees]03'01.0" 8 33[degrees]34'20.0", 79[degrees]03'01.0" 16 33[degrees]32'28.0", 79[degrees]02'53.0" TABLE 2. Types of intervention models. Model Type Equation Mean-Shift [Log.sub.10]FEC = [[alpha].sub.1],Salin + [[alpha].sub.2]Temp + [[alpha].sub.3] [Log.sub.e]Rain + [[alpha].sub.4]I1 + [epsilon] Trend-Plus-Jump [Log.sub.10]FEC = [[alpha].sub.1],Salin + [[alpha].sub.2]Temp + [[alpha].sub.3] [Log.sub.e]Rain + [[alpha].sub.4]Date + [[alpha].sub.5]I1 + [epsilon] Change-in-Trend [Log.sub.10]FEC = [[alpha].sub.1],Salin + [[alpha].sub.2]Temp + [[alpha].sub.3] [Log.sub.e]Rain + [[alpha].sub.4]Date + [[alpha].sub.5]I1 + [[alpha].sub.6]Il Date + [epsilon] Constant-Trend [Log.sub.10]FEC = [[alpha].sub.1],Salin + (no intervention [[alpha].sub.2]Temp + [[alpha].sub.3] effect) [Log.sub.e]Rain + [[alpha].sub.4]Date + [epsilon] TABLE 3. Major anthropogenic interventions that had the potential to modify water quality for Murrells Inlet and North Inlet during the period 1967-95. Estuary Intervention Date Murrells Conversion of Septic Tanks to Sewer Line 6/15/80 Inlet Construction of Jetty 6/15/80 Construction of Boat Ramp 1/15/86 North Construction of Baruch Laboratory 6/15/73 Construction of Debidue Housing Development 1/15/77 TABLE 4. Analytical results Murrells Inlet, 1980 intervention, rain-salinity interaction. Source DF SS MS F P-value Model 7 392.72801 56.1040 116.93 < 0.0001 Error 2018 968.24982 0.47981 Total 2025 968.24982 Adjusted [R.sub.2] = 0.2861 Variable DF Param Est SE T P-value Intercept 1 2.496136 0.16 15.751 0.0001 Date 1 0.000126 0.00 7.348 0.0001 II 1 -0.117380 0.00 -1.902 0.0573 11 Date 1 -0.000130 0.00 -6.27 0.0001 Salin 1 -0.070463 0.00 -19.022 0.0001 Lrain 1 0.364628 0.04 9.537 0.0001 Temp 1 0.011213 0.00 4.703 0.0001 Lrain x Sal 1 -0.008687 0.00 -7.209 0.0001 Variance Variable DF Inflation Intercept 1 0.00 Date 1 11.83 I1 1 3.77 I1 Date 1 7.61 Salin 1 1.62 Lrain 1 30.11 Temp 1 1.05 Lrain x Salin 1 32.18 TABLE 5. Analytical results North Inlet, 1973 intervention, rain-temperature interaction. Source DF SS MS F P-value Model 6 370.25754 61.7096 229.066 -0.0001 Error 1650 444.50507 0.26940 Total 1656 814.76261 Adjusted [R.sub.2] = 0.4525 Variable DF Param Est SE T P-value Intercept 1 2.194261 0.13 16.914 0.0001 Date 1 0.000043 0.00 1.673 0.0946 11 Date 1 -0.000046 0.00 -1.579 0.1146 Salin 1 -0.041880 0.00 -33.782 0.0001 Lrain 1 0.169005 0.02 8.805 0.0001 Temp 1 -0.015172 0.00 -5.077 0.0001 Lrain x Sal 1 -0.004506 0.00 -4.81 0.0001 Variance Variable DF Inflation Intercept 1 0.00 Date 1 39.78 11 Date 1 40.27 Salin 1 1.02 Lrain 1 9.95 Temp 1 2.52 Lrain x Salin 1 10.97 TABLE 6. Wasteload calculations Murrells Inlet (humans and domestic animals) Human All (19,819) = 0.43 x [10.sup.14] MPNs/day Septic tank (1,585) = 0.03 x [10.sup.14] MPNs/day Domestic animals Dogs (3,993) = 1.33 x [10.sup.14] MPNs/day Cats (4,472) = 2.40 x [10.sup.14] MPNs/day Total human and pet wasteload = 2.56 x [10.sup.14] to 2.99 x [10.sup.14] MPNs/day Human/pet wasteload = 0.43/2.99 = 14%