HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Related articles in JEQ Similar articles in this journal Similar articles in Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Web of Science (8) Citing Articles via Google Scholar Google Scholar Articles by Zeng, X. Articles by Rasmussen, T. C. Search for Related Content PubMed PubMed Citation Articles by Zeng, X. Articles by Rasmussen, T. C. Agricola Articles by Zeng, X. Articles by Rasmussen, T. C. Related Collections Surface Water Quality Biogeochemical Processes Nutrients Other Environmental Contamination Water Pollution

TECHNICAL REPORTS

Surface Water Quality

Multivariate Statistical Characterization of Water Quality in Lake Lanier, Georgia, USA

^a Warnell School of Forest Resources, Univ. of Georgia, Athens, GA 30602-2152
^b Currently, Stetson Engineers, 2171 E. Francisco Blvd., Suite K, San Rafael, CA 94901

^* Corresponding author (trasmuss{at}uga.edu)

Received for publication September 1, 2004.

	ABSTRACT

TOP ABSTRACT INTRODUCTION LAKE LANIER WATERSHED SOUTHEASTERN PIEDMONT RIVER... SOUTHEASTERN PIEDMONT LAKE WATER... MULTIVARIATE STATISTICAL... MATERIALS AND METHODS RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS REFERENCES

 
Watershed monitoring programs depend on water quality characterization data collected for many parameters, at many times and places, and with limited resources. Our objective is to present a strategy that reduces the measured parameters, locations, and frequency without compromising the quality of the monitoring program. One year of twice-monthly (growing season) and monthly (dormant season) water quality data collected from 17 lake and 10 tributary sites are used in conjunction with multivariate statistical techniques to improve the utility of collected data by identifying key parameters and monitoring locations. Factor analysis shows that tributary water quality data consists of three components—stormwater runoff, municipal and industrial discharges, and ground water—which can be distinguished using total suspended solids, total dissolved solids, and alkalinity plus soluble reactive P, respectively. Lake water quality characterization is more ambiguous than tributary water quality characterization, but factor analysis indicates that anoxia associated with lake stratification is the largest source of lake water quality variation, followed by nutrient abundance, and finally by biomass abundance. Cluster analysis suggests that tributary and lake monitoring stations can be consolidated. Reducing the number of parameters and stations frees up resources for increased monitoring elsewhere.

Abbreviations: ALK, alkalinity • Cha, chlorophyll a • DFe, dissolved iron • DMn, dissolved manganese • DOC, dissolved organic carbon • DSu, dissolved sulfide • NH₄, ammonium • NO_x, nitrite plus nitrate • SRP, soluble reactive phosphorus • SO₄, sulfate • TDS, total dissolved solids • TFe, total iron • TIC, total inorganic carbon • TKN, total Kjeldahl nitrogen • TMn, total manganese • TN, total nitrogen • TOC, total organic carbon • TP, total phosphorus • TSS, total suspended solids

	INTRODUCTION

TOP ABSTRACT INTRODUCTION LAKE LANIER WATERSHED SOUTHEASTERN PIEDMONT RIVER... SOUTHEASTERN PIEDMONT LAKE WATER... MULTIVARIATE STATISTICAL... MATERIALS AND METHODS RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS REFERENCES

 
WATER QUALITY MONITORING is the cornerstone of lake management, yet the desire to collect additional information is often frustrated by the lack of resources to support the sampling effort. A well-designed water-quality monitoring plan should preserve scarce resources by minimizing the redundancy of nearby monitoring stations and the plethora of possible variables monitored, while at the same time maximizing the information content of the collected data.

Determining the appropriate density of monitoring stations and their sampling parameters and frequency requires balancing the marginal costs of the additional monitoring infrastructure with the marginal value of the acquired data. This article examines water quality data, primarily data related to nutrients and algal biomass as well as suspended solids and the metals Fe and Mn. These data collected within Lake Lanier (a eutrophic lake in the southeastern USA) and its headwater tributaries are characterized using multivariate statistical analysis to identify a watershed-monitoring network that balances parsimony with robustness. Like other watersheds undergoing accelerated eutrophication and sedimentation, these water quality measurements are being used to assess the ecologic health of Lake Lanier.

Our intent is to evaluate the possibility that a smaller subgroup of analytes, locations, and times might provide sufficient information for evaluating watershed conditions and, if so, which set of analytes, locations, and times would be most useful. Sampling at fewer existing sites, or for a reduced set of existing analytes, or at a reduced frequency, may allow for additional sampling at other currently unsampled sites, for currently unsampled analytes, or for more frequent sampling.

	LAKE LANIER WATERSHED

TOP ABSTRACT INTRODUCTION LAKE LANIER WATERSHED SOUTHEASTERN PIEDMONT RIVER... SOUTHEASTERN PIEDMONT LAKE WATER... MULTIVARIATE STATISTICAL... MATERIALS AND METHODS RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS REFERENCES

 
Lake Lanier is a eutrophic lake in the Southeastern Piedmont Region of the USA (Hatcher, 1994, 1998). The lake has been the focus of recent research on the ecological integrity of rapidly urbanizing watersheds, under the USEPA-NSF Water and Watersheds Program (Beck et al., 2002). The need for analysis derived from the rapid growth of metropolitan Atlanta, GA, especially during the period from 1990 to 1998 when the Atlanta metropolitan area recorded a 20% increase in population—higher than any other metropolitan area in the USA (Metro Atlanta Chamber of Commerce, 2000). This ongoing development places enormous pressure on the water resources of the surrounding river basins, and especially on Lake Lanier. Thus, in anticipation of further urbanization within and around the watershed, it is of the utmost importance to develop measures that are protective of the future water quality and ecological integrity of Lake Lanier.

Because water quality degradation is a long-term concern of the communities surrounding the reservoir, Lanier, its watershed, and the future management of associated nutrient and sediment loadings are focal issues of often contentious and acrimonious debate among all the many stakeholders (Beck et al., 2002). Rapid growth within the lake's watershed and beyond is placing new burdens on the reservoir in the form of increased municipal and industrial wastewater discharges, agricultural waste inputs, and urban stormwater inflows.

Threats to lake water quality include both point and nonpoint sources of pollution (Hatcher, 1994, 1998). Point sources within the Lake Lanier watershed include 13 municipal wastewater treatment plants and 33 private and industrial facilities. Nonpoint sources are ubiquitous, including eight landfills, numerous onsite septic disposal systems, land application of dairy and poultry wastes, as well as stormwater runoff from agricultural and rapidly expanding urban sources.

	SOUTHEASTERN PIEDMONT RIVER WATER QUALITY

TOP ABSTRACT INTRODUCTION LAKE LANIER WATERSHED SOUTHEASTERN PIEDMONT RIVER... SOUTHEASTERN PIEDMONT LAKE WATER... MULTIVARIATE STATISTICAL... MATERIALS AND METHODS RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS REFERENCES

 
The Southeastern Piedmont region of the USA, which includes parts of northern Georgia, has a humid, warm-temperate climate (Breckle, 2002). The region's abundant rainfall generates runoff which—when combined with the region's deeply weathered and highly erodible soils—carries large amounts of sediments to receiving waterbodies. These sediments consist of silt and clay along with abundant Fe and nutrients (Zeng and Rasmussen, 2001; Mayhew et al., 2001).

Both observations and models of suspended sediment dynamics and their interactions with nutrients in Southeastern Piedmont rivers indicate strong positive correlations between discharges and sediment and nutrient (particularly P) concentrations (Frick et al., 1996; Holmbeck-Pelham and Rasmussen, 1997; Zeng and Beck, 2001). Also, the land application of poultry and dairy wastes containing organic forms of nutrients may reach nearby rivers and streams during storm events. Increases in discharge are caused by stormwater runoff, which is associated with nonpoint overland flow from urban and agricultural land uses (Frick et al., 1996).

A second possible cause of water quality degradation within the Lake Lanier watershed is discharge from municipal and industrial wastewater treatment facilities (Richman et al., 1998). Discharges from point sources—as well as from land-application facilities—tend to have higher dissolved solids than natural surface waters (Kent and Belitz, 2004), presumably because of the accumulation of soluble wastes. Oxidized forms of N are more commonly associated with point-source discharges and land application facilities (Kent and Belitz, 2004), presumably because organic forms of N are converted to oxidized forms as part of the waste-treatment process, and there are few natural sources of oxidized N species (Frick et al., 1996). Oxidation of organic matter in wastewater treatment facilities also causes increases in dissolved inorganic C and alkalinity (Richman et al., 1998).

Ground water discharge to streams is the primary component of baseflow in streams not dominated by wastewater discharges. Ground water in the Georgia Piedmont generally has lower suspended solids, dissolved organic C, and P concentrations (both total and soluble reactive), and high alkalinity (Peters, 1994; Richter and Markewitz, 2001). Low P concentrations in ground water are caused by sorption to Fe-rich soils by ligand exchange (Stumm and Morgan, 1996; Reynolds and Davies, 2000). Also, Mn is elevated because it is generally the first metal to be solubilized as conditions become more reducing due to organic matter decomposition (Stumm and Morgan, 1996). Alkalinity is elevated due to biodegradation, while dissolved organic matter is reduced due to clay sorption (Nelson et al., 1993).

	SOUTHEASTERN PIEDMONT LAKE WATER QUALITY

TOP ABSTRACT INTRODUCTION LAKE LANIER WATERSHED SOUTHEASTERN PIEDMONT RIVER... SOUTHEASTERN PIEDMONT LAKE WATER... MULTIVARIATE STATISTICAL... MATERIALS AND METHODS RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS REFERENCES

 
One of the important features of lake water quality is the variation in oxygen concentration with depth (Stumm and Morgan, 1996). Lake Lanier, like many other eutrophic lakes, has an oxic epilimnion during the summer and an anoxic hypolimnion throughout most of the period of summer and fall stratification (Zeng and Rasmussen, 1999; Zeng, 2001). Thus, lake surface waters generally have different geochemical characteristics than deeper waters. Within the anoxic zone, the water is dominated by reduced forms of Mn, Fe, S, and ammonium (NH₄), along with the accumulation of respiration products, principally dissolved inorganic C (Stumm and Morgan, 1996).

Phosphorus–iron cycling in lake systems usually causes Fe³⁺ oxides to sorb soluble reactive P via ligand exchange (Stumm and Morgan, 1996; Reynolds and Davies, 2000). These Fe³⁺–phosphate complexes can be transformed by reduction of Fe³⁺ ions to the more soluble Fe²⁺ form, which solubilizes both Fe²⁺ and phosphate ions (Stumm and Morgan, 1996; Wetzel, 2000). Traditional models of aquatic sediment P biogeochemistry ascribe P mobilization to the release of soluble reactive phosphorus (SRP) from Fe³⁺ oxides as these compounds are reduced and solubilized with the onset of low redox potentials in sediments, especially in conjunction with oxygen depletion from overlying waters (Mortimer, 1941, 1942; Kamp-Nielsen, 1974; Wetzel, 1983; Caraco et al., 1991). Summer hypolimnetic anoxic conditions should promote SRP release from bed sediments, which is then available for algal production in the photic zone during subsequent mixus. The scientific basis for this conventional paradigm of P cycling in lakes was developed with data primarily from north-temperate systems (Hutchinson, 1957; Carlson, 1977).

However, this and other recent cross-system comparisons in Lake Lanier and other southeastern lakes fails to show a correlation between anoxia and sediment SRP release (Reckhow, 1988; Mayhew and Mayhew, 1992; Parker, 2004). A hypothesis has recently been proposed that attributes this lack of sediment SRP release to the Fe-rich clay soils that dominate Southeastern Piedmont watersheds (Parker and Rasmussen, 2001; Parker, 2004). Iron-rich sediments are observed in stormwater runoff, causing SRP sorption, a lowering of SRP during anoxic conditions, and a reduction in SRP bioavailability following mixus (Stumm and Morgan, 1996; Parker and Rasmussen, 2001; Parker, 2004).

In Georgia, economic activities and land-use changes affect the behavior of impoundment systems primarily through the accumulation of silt- and clay-dominated sediments deriving from land-surface disturbances and the accumulation of nutrients (Frick et al., 1996). The observed record of Lake Lanier shows that concentrations of soluble species of N and P are significantly lower than those expected from currently estimated inputs of these materials to the lake (Zeng, 2001). Hatcher (1994) estimated that 80 to 90% of the P load entering Lake Lanier is probably sequestered by sediment deposits.

Lake Lanier appears to behave as a most effective trap for sequestering nutrients and thereby lessening their potential impacts on biological cycles, presumably due to their sorption onto mineral sediments, with subsequent deposition, and then burial. This hypothesis, regarding the strong association between suspended Piedmont sediments and nutrients (especially P), has indeed been substantiated through fertilization manipulations of an aquaculture pond, also located in the Georgia Piedmont some 100 km from Lake Lanier (Parker, 2004).

	MULTIVARIATE STATISTICAL APPROACHES

TOP ABSTRACT INTRODUCTION LAKE LANIER WATERSHED SOUTHEASTERN PIEDMONT RIVER... SOUTHEASTERN PIEDMONT LAKE WATER... MULTIVARIATE STATISTICAL... MATERIALS AND METHODS RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS REFERENCES

 
Factor analysis and cluster analysis are multivariate statistical techniques that can be used to characterize watershed water quality and assist in water quality monitoring planning. Factor analysis is used to understand the correlation structure of collected data and identify the most important factors contributing to the data structure (Schoer, 1985; Buckley and Winters, 1992; Padro et al., 1993). In factor analysis, the relationship among a number of observed quantitative variables are represented in terms of a few underlying, independent variables called factors, which may not be directly measured or even measurable.

Although not commonly used in water quality analysis, several studies have used factor analysis to identify primary sources of contamination. Borovec (1996) used factor analysis to extract dominant types of heavy metals in river sediments. The study was able to identify three parameter groups, which were then used to specify contamination sources. Factor analysis is also used to find associations between parameters so that the number of measured parameters can be reduced. Known associations are then used to predict unmeasured water quality parameters.

Cluster analysis is another data reduction method that is used to classify entities with similar properties. The method divides a large number of objects into a smaller number of homogeneous groups on the basis of their correlation structure (Hartigan, 1975). The objective of cluster analysis is to identify the complex nature of multivariate relationships (by searching for natural groupings or types) among the data under investigation, so as to foster further hypothesis development about the phenomena being studied. Cluster analysis imposes a characteristic structure on the data analysis for exploratory purposes.

Colby (1993) used cluster analysis to objectively analyze a large number of physical and chemical ground water variables to identify zones with similar physical and chemical hydrogeologic characteristics. Suk and Lee (1999) also applied cluster analysis to identify distinct hydrochemical regimes and zones. Within the lake water quality literature, Salmaso (1996) applied cluster analysis in a phytoplankton community study to identify similar algal groups. Yet water quality studies have not employed this approach to assist in the characterization of lake water quality and the design of water quality monitoring programs.

Momen et al. (1999) used cluster analysis to segregate lake nitrate concentration data in lakes in the Adirondack region of the northeastern USA. Momen and Zehr (1998) and Momen et al. (1999) also employed canonical discriminant analysis to segregate a wide range of lakes in the same region. The purpose of these studies was to evaluate the effects of acidic deposition and atmospheric nutrient (N) addition on lake ecosystems. These studies are particularly important because of possible temporal changes in watershed disturbances (Momen et al., 1997; Lawrence et al., 2004).

In this article, both factor analysis and cluster analysis are used to characterize the spatial and temporal variability of Lake Lanier tributary and lake water quality. Because of over-sampling in space and time, the Lake Lanier dataset may contain redundancy in water quality information. The goal of this analysis is to identify a method that reduces the monitoring and assessment requirements of this lake system, yet still accurately represents the system. The ability to reduce the sampling effort—yet to also obtain better information from this reduced effort—has important ramifications for both water quality monitoring and modeling.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION LAKE LANIER WATERSHED SOUTHEASTERN PIEDMONT RIVER... SOUTHEASTERN PIEDMONT LAKE WATER... MULTIVARIATE STATISTICAL... MATERIALS AND METHODS RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS REFERENCES

 
Site Description
Lake Lanier is situated in the headwaters of the Apalachicola–Chattahoochee–Flint river basin (Hydrologic Unit Code 03130001), which is shared by the three contiguous states of Alabama, Georgia, and Florida. The lake is a key resource in the proposed tri-state water compact designed to apportion basin water use among the three states (Kundell, 1999). Lake Lanier is a major impoundment on the Chattahoochee River about 80 km northeast of Atlanta, GA (Fig. 1) . The reservoir, when full, covers approximately 156 km² and has a total storage volume of approximately 2.4 x 10⁹ m³. The reservoir is used for drinking water supply, hydroelectric power generation, flood control, navigation, and recreation. Long-term average discharge from the 2689-km² watershed is approximately 59 m³ s^–1. Of this flow, approximately 70% is contributed from two large tributaries, the Chattahoochee and Chestatee Rivers. It is important to note that smaller, tributary watersheds nearer the lake shoreline are more intensively developed, and often have a disproportionate impact on lake water quality due to their proximity to the lake.

View larger version (57K): [in this window] [in a new window]   Fig. 1. Lake Lanier location and water quality sampling stations.

Tributary monitoring sites were located upstream of the mouths of 10 significant tributaries to Lake Lanier and at 17 points within the reservoir. Monitored parameters are provided in Table 1 and the locations of the sampling stations are shown in Fig. 1. Tributary water quality monitoring stations include two major stations on the Chattahoochee and Chestatee Rivers plus eight minor stations on lesser tributaries. The two major stations are located immediately downstream of U.S. Geological Survey (USGS) stream flow gauges on the Chattahoochee River at Cornenia (02331600) and the Chestatee River near Dahlonega (02333500).

Lake water quality monitoring design generally mirrored that employed during the Clean Lakes Study (Hatcher, 1994). Lake monitoring stations were distributed in the lake embayments and main body for the purpose of providing a more robust evaluation of spatial variation in water quality. Lake sampling was conducted twice monthly during the algal growing season (April–October) and monthly during the nongrowing season. Nongrowing season samples were collected just below the lake surface and 1 m above the lake bottom and analyzed for the entire suite of lake water quality parameters. Growing-season samples were collected at approximately five depths at each station. Samples at the middle of the epilimnion and hypolimnion were analyzed for all lake parameters, while the remaining three samples were analyzed for nutrients and Chlorophyll a, shown in Table 1.

Quality assurance and control for this monitoring program was documented in a field sampling manual, which consisted of elements designed to ensure consistency of sampling methods, quantify the reliability of the analytical system, and ensure the integrity of the resulting database (Limnotech, 1998). All methods conformed to USEPA and state protocols. Field duplicate samples and field blanks were used to quantify measurement precision and to detect significant contamination. Both field duplicate and blank samples comprised from 5 to 10% of the total number of samples collected. Laboratory splits, blanks, spikes, control checks, and periodic reviews of performance indices were performed (Limnotech, 1998).

Data Preprocessing
All monitored parameters listed in Table 1 are used in the multivariate statistical analysis for the purpose of examining whether the sampling parameters can be reduced. Thus, sampling dates with missing observations were excluded; only observations with all data records present were used. Observations with values below the detection limit were replaced with a value equal to one-half of the detection limit. Data were obtained at different times and locations, as well as different depths in the case of lake station data. For factor analysis, the entire pool of data was used. For cluster analysis, tributary water quality data were organized by sampling station and month, and lake water quality data were arranged by station, month, and epilimnion and hypolimnion. For spatial data analysis, the data were aggregated over time. For temporal data analysis, the data were aggregated over space. A base-10 logarithmic transform was used to remove the heteroscedastic effect of a variation in measurement error with the magnitude of the observation and the transformed multi-dimensional data were used in this multivariate analysis. The multivariate analysis was conducted using STATISTICA software (Statsoft, 1995).

Factor Analysis
The initial step was the determination of the parameter correlation matrix, which is used to account for the degree of mutually shared variability between individual pairs of water quality parameters. The second step was the estimation of the eigenvalues and factor loadings for the correlation matrix. Each eigenvalue corresponded to an eigenfactor that identifies the groups of variables that were most highly correlated among themselves. The first eigenfactor accounted for the greatest variation among the observed variables, while each subsequent eigenfactor was orthogonal to all preceding factors, and provided incrementally smaller contributions to the overall descriptive ability of the model.

Because lower eigenvalues may contribute little to the explanatory capability of the data, only the first few factors were needed to account for much of the parameter variability. In this study, the factor extraction was performed using the method of principal components. Guidelines have been developed for determining how many factors to use and how many to ignore (Browne, 1968; Linn, 1968; Tucker et al., 1969; Hakstian et al., 1982). The most widely used method is the Kaiser criterion (Kaiser, 1960), which retains only those factors with eigenvalues >1. This means that each retained factor provides as much explanatory capability as one original variable. The Scree test (Cattell, 1966) identifies the useful factors as those whose eigenvalues are substantially larger than subsequent eigenvalues.

Once the correlation matrix and eigenvalues were obtained, factor loadings were used to measure the correlation between variables and factors—the squared factor loading being the variance explained by that factor. To obtain the proportion of variance in all variables accounted for by each factor, the squared factor loadings were summed for that factor and then divided by the number of variables, which is equivalent to dividing the factor's eigenvalue by the number of variables.

Factor rotation was used to facilitate interpretation by providing a simpler factor structure. The factors were rotated so that the observed axes were aligned with a dominant set of variables, which assisted in understanding how factors were related to the observed variables. While our study used the varimax rotation, which is a standard rotation method (Kaiser, 1958), other rotations have also been proposed, including quartimax, biquartimax, and equamax (Harman, 1967).

Cluster Analysis
Cluster analysis was first conducted to group sampling stations using Ward's method with Euclidean distance measure. A representative group of homogeneous stations was then selected to analyze temporal variability of water quality within the group for the purpose of evaluating whether the sampling frequency can be reduced. We used cluster analysis to link variables in the configuration of a tree with different branches—branches that have linkages closer to each other indicate a stronger relationship among variables or clusters of variables. The dendrogram generated from tree clustering provides a useful graphical tool for determining the number of clusters that describe underlying processes that lead to spatial variation.

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION LAKE LANIER WATERSHED SOUTHEASTERN PIEDMONT RIVER... SOUTHEASTERN PIEDMONT LAKE WATER... MULTIVARIATE STATISTICAL... MATERIALS AND METHODS RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS REFERENCES

 
Table 2 provides a statistical summary of the water quality data for the parameters listed in Table 1 for both tributary and within-lake data. Note that lake stations, for every parameter, have lower mean and median concentrations than tributary stations. While a definitive conclusion of fate and transport of lake inputs is not possible because samples were not taken at the same frequency or times, it does suggest that a general attenuation of water quality inputs is provided by the lake system. This behavior is consistent with the conclusion reached by Hatcher (1994)( 1998), which showed that >90% of influent sediment and nutrients are sequestered by the reservoir.

View this table: [in this window] [in a new window]   Table 2. Statistical summary of Lake Lanier water quality data. All units are mg/L except Chlorophyll a (µg/L).

View larger version (21K): [in this window] [in a new window]   Fig. 2. Monthly rainfall and inflows to Lake Lanier during the water quality sampling period.

These correlations were used to identify groups of highly correlated water quality variables. Table 3 provides the eigenvalues and the explanatory capability for the Lake Lanier tributary water quality data. The first factor accounts for almost half the total variability, whereas subsequent factors assist in describing the water quality information, but with a rapid diminishment in magnitude.

Factor loadings reflect the correlations between the variables and the extracted factors. Factor loadings for the three retained eigenvalues are shown in Table 4. Factor loadings are shown without rotation because many of the factors are aligned with one or the other axes. Factor loadings >0.5 are considered significant in this study.

View larger version (15K): [in this window] [in a new window]   Fig. 3. Correlation between total suspended solids (TSS) concentration and discharge (Flow) in the Chattahoochee and Chestatee Rivers.

The second factor incorporates those water quality variables that are characteristic of wastewater discharges, including total dissolved solids (TDS), ALK, NO_x, and TN. Total N is correlated with the second factor due to the large contribution of NO_x to the calculated value of TN. In addition, several other variables load negatively on the second factor, including TSS and TFe, which is consistent with discharges from these facilities (Richman et al., 1998).

It is noteworthy that TDS is a primary indicator of the second factor. Total dissolved solids are easily and reliably measured in the field using specific conductance. Thus, elevated TDS measurements indicate the possibility of anomalous N concentrations. Clearly, sampling for elevated N species would be more effective when it is targeted to those samples where elevated TDS is observed.

The third factor has lower factor loadings, the largest positive values (0.41 and 0.40) being for TMn and ALK. More extreme negative values (–0.60 and –0.50) are found for SRP and TP, respectively. The third factor is consistent with a ground water signature because of the high P retention of Southeastern Piedmont soils along with increased soil metabolism generating carbonate and noncarbonate alkalinity (Richter and Markewitz, 2001). Probably the most likely diagonstic indicator would be a combination of elevated alkalinity in association with low SRP concentrations. Additional field inspection of these relationships should be pursued to evaluate the reliability of this approach.

It appears that each of the three factors has a physical basis for observation. The primary contributors to stream-flow water quality variation are: (i) nonpoint source, or stormwater runoff (46.4%); (ii) point source, or wastewater treatment facility, discharges (20%); and (iii) ground water inflows (8.6%). These three factors account for three-quarters of the observed water quality variability. The remaining variability may result from localized instream processes such as upstream reservoirs or unique watershed contributions.

Factor Analysis of Lake Data
Lake water samples included all parameters collected during tributary sampling along with six additional water quality variables, including Chlorophyll a (Cha), total inorganic C (TIC), dissolved sulfide (DSu), sulfate (SO₄), dissolved Fe (DFe), and dissolved Mn (DMn). Like the tributary data, many variables are positively correlated with each other, including strong correlations between TMn and DMn, r = 0.84; TMn and TFe, r = 0.73; TFe and DFe, r = 0.71; along with good correlations between TMn and TIC, r = 0.66; TFe and NH₄, r = 0.63; NO_x and TN, r = 0.62. Only a few of the variables are negatively correlated, the largest being between DMn with Cha, r = –0.34; and SO₄ with TIC, r = –0.33.

Table 5 provides the eigenvalues and the explanatory capability for the lake water quality data. The first factor only accounts for 27.2% of the total variability, while subsequent factors again provide a diminishing ability to predict water quality variation. The first six factors each have an eigenvalue 1, yet the greatest improvement in predictive ability ends after the first three factors. Using the Scree test result, only the first three factors are used in the subsequent analysis.

View larger version (27K): [in this window] [in a new window]   Fig. 4. Spatial and temporal variations of soluble reactive phosphorus (SRP) and total iron (TFe) concentrations in Lake Lanier.

The third factor is strongly related to Cha, TP, TKN, TSS, and TOC. This appears to be a biologic productivity measure with perhaps organic forms of nutrients, and is consistent with the model developed by Mayhew and Mayhew (1992). It is interesting to note that the eigenvalues for the second and third factors are approximately equal, 2.35 and 2.26, respectively, indicating that their relative effects are similar.

Discriminating between lake water quality types is more difficult than for tributary flows. This is likely due to incomplete knowledge of lake biogeochemical processes (Mayhew and Mayhew, 1992; Hatcher, 1994; Mayhew et al., 2001; Parker and Rasmussen, 2001). Improvements in lake water quality monitoring programs will require a better understanding of the complex relationships between algal blooms and nutrient availability. Clearly, additional biogeochemical research, monitoring, and modeling of nutrient cycling in Southeastern Piedmont impoundments are needed.

Cluster Analysis of Tributary Data
Factor analysis was used to combine water quality parameters into homogeneous groups; it is also possible to evaluate whether water quality samples at various locations can be combined into homogeneous regions so that the number of sampling sites can be reduced. Figure 5 presents the results of cluster analysis for water quality monitoring data from the stations tributary to Lake Lanier. Two associations are evident. The association between East Fork Little River and Flat Creek North is most significant, with bonds to Wahoo Creek, and, to a lesser degree, Balus Creek, then to West Fork Little River. These are small tributaries with large nonpoint-source pollution inputs (Smith and Sellers, 1994; Hatcher, 1994, 1998).

View larger version (13K): [in this window] [in a new window]   Fig. 5. Similarity dendrogram among tributary sampling stations from cluster analysis. Five lake tributary groupings shown: (i) Chestatee and Chattahoochee Rivers; (ii) Wahoo Creek, East Fork Little River, Flat Creek North, Balus Creek, and West Fork Little River; (iii) Limestone Creek; (iv) Six Mile Creek; and (v) Flat Creek South.

The identified group including sampling stations at Chattahoochee and Chestatee Rivers was selected to analyze temporal variations and to evaluate whether the sampling frequency can be reduced. Figure 6 shows the cluster analysis result on sampling times. One association is evident, which includes May through November 1996 except September. This was a relatively dry period (Fig. 2) and base flow could dominate tributary flows. The association among the relatively wet months is not significant. March 1996 was the wettest month and has weak association with other months due to significant stormwater inflows.

View larger version (16K): [in this window] [in a new window]   Fig. 6. Similarity dendrogram among tributary sampling times in the group of the Chattahoochee and Chestatee Rivers sampling stations. Evident association includes May through November 1996 except September.

View larger version (11K): [in this window] [in a new window]   Fig. 7. Similarity dendrogram among lake sampling stations from cluster analysis. Four lake station groupings shown: (i) The lowest part of the lake near the dam (Stations 1, 2, 3, 4, 5); (ii) the central part of the lake (Stations 6, 7, 8, 9, 10); (iii) the upper part of the lake (Stations 12, 13, 14, 15, 16, 17); and (iv) the embayment near Flat Creek South (Station 11).

The identified group—including lake sampling stations at the lowest part near the dam—was selected to analyze temporal variations of water quality within the group because this group of stations has direct effects on downstream water quality. Figure 8 shows the cluster analysis results on sampling times for both epilimnion and hypolimnion water quality observations. The results indicate different associations among sampling times for the epilimnion and hypolimnion. The only similar association for the epilimnion and hypolimnion is June, July, and August 1996. Within the epilimnion or hypolimnion, the associations among sampling times are not evident. These results suggest that the lake water quality had significant temporal variations.

View larger version (19K): [in this window] [in a new window]   Fig. 8. Similarity dendrogram among lake sampling times in the group of lake sampling stations nearest the dam (Stations 1, 2, 3, 4, 5). (a) Epilimnion; (b) Hypolimnion.

A monitoring program could use a smaller set of variables to identify times for intensive sampling; TSS as an indicator of stormwater runoff, TDS as an indicator of wastewater and land application discharges, and a combination of ALK and TP or SRP as an indicator of ground water inflows. This reduced set of parameters could be monitored over larger areas within the watershed to provide more detailed spatial information about sources and processes. Additional, concomitant analysis for a broader suite of parameters should be performed at location and times when elevated indicators of possible contamination (e.g., TDS or TSS) are observed (Richman et al., 1998). Stage-dependent sampling is recommended because the bulk of water quality variation can be attributed to stormwater (Holmbeck-Pelham and Rasmussen, 1997; Zeng and Rasmussen, 2001). Sampling should be concomitant with discharge measurements over a wide range of discharges. Additional, low-flow sampling for nitrates should focus on those tributaries where wastewater treatment and land application facilities are located (Richman et al., 1998).

Factor analysis of lake water quality observations yielded a more complex suite of processes. The first factor, which accounts for 27.2% of the total lake water quality variability, appears to describe the anoxia associated with lake stratification. Unlike other lake systems, SRP does not accumulate within Lake Lanier's anoxic zone (Zeng and Rasmussen, 1999). This has a profound impact on P biogeochemical cycling in that algal blooms fail to occur following the fall mixus. A second lake factor, which accounts for 12.3% of the total variability, appears to be related to nutrient abundance—NO_x, TN, and SRP (indicators of lake eutrophication) load positively. A third lake factor, which accounts for 11.9% of the variation, is positively related to algal biomass (Cha and TOC) as well as TP, TKN, and TSS. To better understand the complex biogeochemical cycling of this lake system, additional intensive lake monitoring without reducing sampling parameters is suggested.

Cluster analysis was used to characterize the spatial and temporal variability of water quality data. The 10 tributaries to Lake Lanier that were monitored in this study can be divided into five homogeneous groups. Our results suggest that one monitoring station should be situated on the largest tributary in the first group (Wahoo Creek) to track the behavior of those streams. Notwithstanding the fact that the Chestatee and Chattahoochee Rivers have much in common, they should both be monitored because they are the largest tributaries to Lake Lanier. Although the remaining three tributaries (Limestone Creek, Six Mile Creek, and Flat Creek South) are small, they behave sufficiently different from the other tributaries and require monitoring. In all, water quality monitoring in six tributaries to Lake Lanier is recommended. Cluster analysis of temporal variability within the two largest tributaries suggests that samples taken during low flow months can be grouped together, while months dominated by stormwater inflows have less in common.

Cluster analysis of water quality data from 17 lake stations was used to determine whether various parts of Lake Lanier could be grouped into homogeneous zones. Based on the analysis here, monitoring in just four zones is recommended: (i) the lowest part of the lake near the dam (Stations 1–5), (ii) the central part of the lake (Stations 6–10), (iii) the upper part of the lake (Stations 12–17) nearest the influent tributaries, and (iv) the embayment near Flat Creek South (Station 11). In all, water quality monitoring at four lake stations is recommended. These observations should be collected at 1- to 2-m intervals from the water surface to the maximum depth of the lake at the station on a monthly basis. Reducing lake water quality sampling frequency is not recommended due to significant temporal variations.

	SUMMARY AND CONCLUSIONS

TOP ABSTRACT INTRODUCTION LAKE LANIER WATERSHED SOUTHEASTERN PIEDMONT RIVER... SOUTHEASTERN PIEDMONT LAKE WATER... MULTIVARIATE STATISTICAL... MATERIALS AND METHODS RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS REFERENCES

 
Lake Lanier is the key water supply reservoir for the rapidly growing Atlanta metropolitan area. Rapid development in the upstream headwaters presents water quality management challenges for the area. Water quality data was collected between March 1996 and March 1997 by a local association of governments to provide a synopsis of eutrophication conditions within the lake and influent tributaries. Knowing that the scope of the original study exceeds local resources for continuation, the expectation is that a smaller set of key parameters and locations can be identified to provide a comparable level of lake and watershed characterization. This study used multivariate statistical techniques to identify the key parameters and locations for consideration and sampling frequency as well.

Factor analysis shows that tributary water quality data consists of three components (i.e., stormwater runoff, municipal and industrial discharges, and ground water), which can be distinguished using TSS, TDS, and ALK + SRP, respectively. Lake water quality characterization is more ambiguous than tributary water quality characterization, but factor analysis indicates that anoxia associated with lake stratification is the largest source of lake water quality variation, followed by nutrient abundance, and finally by biomass abundance. Cluster analysis of spatial variations shows that the existing number of monitoring stations can be reduced from five tributary stations to one station in one case, and from six lake stations to one lake station in another. Cluster analysis of temporal variability within the two largest tributaries suggests that the sampling frequency taken during low flow months can be reduced.

	ACKNOWLEDGMENTS

 
This work was funded by grants from the USEPA–National Science Foundation, Water and Watersheds Program (grant R825758). The authors thank the associate editor, Ole Wendroth, and three anonymous reviewers for their helpful comments.

	REFERENCES

TOP ABSTRACT INTRODUCTION LAKE LANIER WATERSHED SOUTHEASTERN PIEDMONT RIVER... SOUTHEASTERN PIEDMONT LAKE WATER... MULTIVARIATE STATISTICAL... MATERIALS AND METHODS RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS REFERENCES

 

• Beck, M.B., B.D. Fath, A.K. Parker, O.O. Osidele, G.M. Cowie, T.C. Rasmussen, B.C. Patten, B.G. Norton, A. Steinemann, S.R. Borrett, D. Cox, M.C. Mayhew, W. Zeng, and X. Zeng. 2002. Developing a concept of adaptive community learning: Case study of a rapidly urbanizing watershed. Int. Assess. 3:299–307.[CrossRef]
• Borovec, Z. 1996. Evaluation of the concentrations of trace elements in stream sediments by factor and cluster analysis and the sequential extraction procedure. Sci. Total Environ. 177:237–250.[CrossRef]
• Breckle, S.W. 2002. Walter's vegetation of the earth. 4th ed. Springer, New York.
• Browne, M.W. 1968. A comparison of factor analytic techniques. Psychometrika 33:267–334.[CrossRef][Web of Science][Medline]
• Buckley, D.E., and G.V. Winters. 1992. Geochemical characteristics of contaminated surficial sediments in Halifax Harbor: Impact of waste discharge. Can. J. Earth Sci. 29:2617–2639.
• Caraco, N., J.J. Cole, and G.E. Likens. 1991. A cross-system study of phosphorus release from lake sediments. p. 241–258. In J.J. Cole et al. (ed.) Comparative analysis of ecosystems: Patterns, mechanisms, and theories. Springer-Verlag, New York.
• Carlson, R.E. 1977. A trophic state index for lakes. Limnol. Oceanogr. 22:361–368.
• Cattell, R.B. 1966. The Scree test for the number of factors. Multiv. Behav. Res. 1:245–276.[CrossRef]
• Colby, N.D. 1993. The use of 2-way cluster analysis as a tool for delineating trends in hydrogeologic units and development of a conceptual model. p. 91–100. In Proc. 1993 Ground Water Modeling Conf., Golden CO. 9–12 June 1993. Int. Ground Water Modeling Center, Golden, CO.
• Frick, E.A., G.R. Buell, and E.H. Hopkins. 1996. Nutrient sources and analysis of nutrient water-quality data, Apalachicola–Chattahoochee–Flint River basin, Georgia, Alabama, and Florida, 1972–90. Water-Resour. Invest. Rep. 96-4101. USGS, Reston, VA.
• Hakstian, A.P., W.D. Rogers, and R.B. Cattel. 1982. The behavior of numbers of factors rules with simulated data. Multiv. Behav. Res. 17:193–219.[CrossRef]
• Harman, H.H. 1967. Modern factor analysis. Univ. of Chicago Press, Chicago, IL.
• Hartigan, J.A. 1975. Clustering algorithms. John Wiley & Sons, New York.
• Hatcher, K.J. 1994. Diagnostic/feasibility study of Lake Sidney Lanier, Georgia. Project completion report. Univ. of Georgia, Athens, GA.
• Hatcher, K.J. 1998. Diagnostic/feasibility study of Lake Sidney Lanier, Georgia. Project completion report. Univ. of Georgia, Athens, GA.
• Holmbeck-Pelham, S.A., and T.C. Rasmussen. 1997. Characterization of temporal and spatial variability of turbidity in the upper Chattahoochee River. p. 144–147. In K.J. Hatcher (ed.) Proc. 1997 Georgia Water Resour. Conf., Athens, GA. 20–22 Mar. 1997. Univ. of Georgia, Athens, GA.
• Hutchinson, G.E. 1957. A treatise on limnology. Vol. 1. John Wiley & Sons, New York.
• Kaiser, H.F. 1958. The varimax criterion for analytic rotation in factor analysis. Pyrometrical 23:187–200.
• Kaiser, H.F. 1960. The application of electronic computers to factor analysis. Educ. Psychol. Meas. 20:141–151.[CrossRef][Web of Science]
• Kamp-Nielsen, L. 1974. Mud–water exchange of phosphate and other ions in undisturbed sediment cores and factors affecting the exchange rates. Arch. Hydrobiol. 73:218–237.
• Kent, R.H., and K. Belitz. 2004. Concentrations of dissolved solids and nutrients in water sources and selected streams of the Santa Ana Basin, California, October 1998–September 2001. Water-Resour. Invest. Rep. 03-4326. USGS, Reston, VA.
• Kundell, J.E. 1999. Water-related legislation of the 1997–1998 term of the Georgia general assembly. p. 1–5. In K.J. Hatcher (ed.) Proc. 1999 Georgia Water Resour. Conf., Athens, GA. 30–31 Mar. 1999. Univ. of Georgia, Athens, GA.
• Lawrence, G.B., B. Momen, and K.M. Roy. 2004. Use of stream chemistry for monitoring acidic deposition effects in the Adirondack region of New York. J. Environ. Qual. 33:1002–1009.[Abstract/Free Full Text]
• Limno-Tech. 1998. Development of linked watershed and water quality models for Lake Lanier. Limno-Tech, Ann Arbor, MI.
• Linn, R.L. 1968. A Monte Carlo approach to the number of factors problem. Psychometrika 33:37–71.[CrossRef][Web of Science][Medline]
• Mayhew, E.A., and M.C. Mayhew. 1992. The effects of the interaction of clay, phosphorus, and organic particulates on the trophic status of Lake Lanier. Tech. Completion Rep. ERC01-92. Environ. Res. Center, Atlanta, GA.
• Mayhew, M., T.C. Rasmussen, and A.K. Parker. 2001. Phosphorus binding by iron-rich soil in the Southeastern Piedmont: Implications for point and non-point sources of phosphorus. p. 453–456. In K.J. Hatcher (ed.) Proc. 2001 Georgia Water Resour. Conf., Athens, GA. 26–27 Mar. 2001. Univ. of Georgia, Athens, GA.
• Metro Atlanta Chamber of Commerce. 2000. Final report of the clean water initiative. Tech. Rep. Metro Atlanta Chamber of Commerce, Atlanta, GA.
• Momen, B., L.W. Eichler, C.W. Boylen, and J.P. Zehr. 1997. Are recent watershed disturbances associated with temporal and spatial changes in water quality of Lake George, New York, USA? Environ. Manage. 21:725–732.[Medline]
• Momen, B., and J.P. Zehr. 1998. Watershed classification by discriminant analyses of lakewater-chemistry and terrestrial characteristics. Ecol. Appl. 8:497–507.
• Momen, B., J.P. Zehr, W.W. Boylen, and J.P. Sutherland. 1999. Determinants of summer nitrate concentration in a set of Adirondack Lakes, New York. Water Air Soil Pollut. 111:19–28.[CrossRef]
• Mortimer, C.H. 1941. The exchange of dissolved substances between mud and water in lakes. J. Ecol. 29:280–329.[CrossRef][Web of Science]
• Mortimer, C.H. 1942. The exchange of dissolved substances between mud and water in lakes. J. Ecol. 30:147–201.[CrossRef]
• Nelson, P.N., J.A. Baldcock, and J.M. Oades. 1993. Concentration and composition of dissolved organic carbon in streams in relation to catchment soil properties. Biogeochemistry 19:27–50.
• Padro, R., E. Barrado, Y. Castrillejo, M.A. Valasco, and M. Vaga. 1993. Study of the contents and speciation of heavy metals in river sediments by factor analysis. Anal. Lett. 26:1719–1739.
• Parker, A.K. 2004. The role of iron-rich sediment in the biogeochemical cycling of phosphorus in Georgia Piedmont impoundments. Ph.D. diss. Univ. of Georgia, Athens, GA.
• Parker, A.K., and T.C. Rasmussen. 2001. Phosphorus cycling in Southeastern Piedmont lakes: An alternative pathway. p. 457–460. In K.J. Hatcher (ed.) Proc. 2001 Georgia Water Resour. Conf., Athens, GA. 26–27 Mar. 2001. Univ. of Georgia, Athens, GA.
• Peters, N.E. 1994. Water-quality variations in a forested Piedmont catchment, Georgia, USA. J. Hydrol. (Amsterdam) 156:73–90.
• Reckhow, K.M. 1988. Empirical models for trophic state in Southeastern U.S. lakes and reservoirs. Water Resour. Bull. 24:723–734.
• Reynolds, C.S., and P.S. Davies. 2000. Sources and bioavailability of phosphorus fractions in freshwaters: A British perspective. Biol. Rev. 76:27–64.[CrossRef]
• Richman, M.T., B. Brouckaert, and A. Amirtharajah. 1998. Impact of discrete pollutant sources into Lake Lanier. p. 5.1–5.59. In K.J. Hatcher (ed.) Diagnostic/feasibility study of Lake Sidney Lanier, Georgia. Project Completion Rep. Univ. of Georgia, Athens, GA.
• Richter, D.D., Jr., and D. Markewitz. 2001. Understanding soil change. Cambridge Univ. Press, Cambridge, UK.
• Salmaso, N. 1996. Seasonal variation in the composition and rate of change of the phytoplankton community in a deep subalpine lake (Lake Garda, Northern Italy): An application of nonmetric multidimensional scaling and cluster analysis. Hydrobiologia 337:49–68.[CrossRef]
• Schoer, J. 1985. Iron-oxo-hydroxides and their significance to the behavior of heavy metals in estuaries. Environ. Technol. Lett. 6:189–202.
• Smith, M.C., and J. Sellers. 1994. Water quality impacts of poultry litter management. Technical Completion Rep. ERC04-94. Environ. Res. Center, Atlanta, GA.
• StatSoft. 1995. STATISTICA for Windows. Statsoft, Tulsa, OK.
• Stumm, W., and J.J. Morgan. 1996. Aquatic chemistry: Chemical equilibria and rates in natural waters. John Wiley & Sons, New York.
• Suk, H., and K.K. Lee. 1999. Characterization of a ground water hydrochemical system through multivariate analysis: Clustering into ground water zones. Ground Water 37:358–366.
• Tucker, L.R., R.F. Koopman, and R.L. Linn. 1969. Evaluation of factor analytic research procedures by means of simulated correlation matrices. Psychometrika 34:421–459.
• Wetzel, R.G. 1983. Limnology. Saunders College Publ., Philadelphia, PA.
• Wetzel, R.G. 2000. Limnology: Lake and river ecosystems. Academic Press, San Diego, CA.
• Zeng, W., and M.B. Beck. 2001. Development and evaluation of a mathematical model for the study of sediment-related water quality issues. Water Sci. Technol. 43:47–54.[Web of Science][Medline]
• Zeng, X. 2001. Water quality modeling in Southeastern lake management: A case study of Lake Lanier, Georgia. Ph.D. diss. Univ. of Georgia, Athens, GA.
• Zeng, X., and T.C. Rasmussen. 1999. Water quality dynamics in Lake Lanier. p. 350–353. In K.J. Hatcher (ed.) Proc. 1999 Georgia Water Resour. Conf., Athens, GA. 30–31 Mar. 1999. Univ. of Georgia, Athens, GA.
• Zeng, X., and T.C. Rasmussen. 2001. Short-term and long-term sediment and phosphorus inputs to Lake Lanier. p. 714–717. In K.J. Hatcher (ed.) Proc. 2001 Georgia Water Resour. Conf., Athens, GA. 26–27 Mar. 2001. Univ. of Georgia, Athens, GA.

Related articles in JEQ:

This Issue in Journal of Environmental Quality

JEQ 2005 34: v. [Full Text]  

This article has been cited by other articles:

				D. Kaown, Y. Hyun, G.-O. Bae, and K.-K. Lee Factors Affecting the Spatial Pattern of Nitrate Contamination in Shallow Groundwater J. Environ. Qual., August 31, 2007; 36(5): 1479 - 1487. [Abstract] [Full Text] [PDF]

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Related articles in JEQ Similar articles in this journal Similar articles in Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Web of Science (8) Citing Articles via Google Scholar Google Scholar Articles by Zeng, X. Articles by Rasmussen, T. C. Search for Related Content PubMed PubMed Citation Articles by Zeng, X. Articles by Rasmussen, T. C. Agricola Articles by Zeng, X. Articles by Rasmussen, T. C. Related Collections Surface Water Quality Biogeochemical Processes Nutrients Other Environmental Contamination Water Pollution