Published June 23, 2014

Journal of Environmental Quality TECHNICAL REPORTS

Surface Water Quality

Linking Spatial Variations in Water Quality with Water and Land

Management using Multivariate Techniques
Yongshan Wan,* Yun Qian, Kati White Migliaccio, Yuncong Li, and Cecilia Conrad

Most studies using multivariate techniques for pollution source

evaluation are conducted in free-flowing rivers with distinct point
and nonpoint sources. This study expanded on previous research
to a managed “canal” system discharging into the Indian River
Lagoon, Florida, where water and land management is the single
S urface water quality, which is greatly influenced
by anthropogenic activities, has gained national and inter-
national interest due to its correlation with human health
and the structure and function of aquatic ecosystems (Osborne
and Wiley, 1988; Ravichandrana et al., 1996; Ferrier et al., 2001;
most important anthropogenic factor influencing water quality. Simeonov et al., 2003; Shrestha and Kazama, 2007; Samsudin
Hydrometric and land use data of four drainage basins were
uniquely integrated into the analysis of 25 yr of monthly water
et al., 2011). The assessment of water quality requires the mea-
quality data collected at seven stations to determine the impact surement of multiple parameters at varying spatial and temporal
of water and land management on the spatial variability of water scales. Such monitoring programs generally produce complex
quality. Cluster analysis (CA) classified seven monitoring stations multidimensional data that require multivariate statistical tech-
into four groups (CA groups). All water quality parameters niques as analytical tools for the characterization and evaluation
identified by discriminant analysis showed distinct spatial patterns
among the four CA groups. Two-step principal component
of water quality (e.g., Vega et al., 1998; Wunderlin et al., 2001;
analysis/factor analysis (PCA/FA) was conducted with (i) water Singh et al., 2004; Papatheodorou et al., 2006). A review of the
quality data alone and (ii) water quality data in conjunction with literature indicated that numerous studies on water quality appli-
rainfall, flow, and land use data. The results indicated that PCA/FA cations of multivariate techniques have been published. Table 1
of water quality data alone was unable to identify factors associated summarizes a subset of these studies with a focus on the water
with management activities. The addition of hydrometric and land
use data into PCA/FA revealed close associations of nutrients and
quality of surface water bodies. Most studies used data collected
color with land management and storm-water retention in pasture for several years, with biweekly or monthly sampling intervals at
and citrus lands; total suspended solids, turbidity, and NO3 + NO2 multiple sampling locations. Multivariate statistical techniques
with flow and Lake Okeechobee releases; specific conductivity with were applied for evaluation of spatial and temporal variations,
supplemental irrigation supply; and dissolved O2 with wetland data reduction, and identification and apportionment of pollu-
preservation. The practical implication emphasizes the importance
of basin-specific land and water management for ongoing pollutant
tion sources, with only a few exceptions such as identification
loading reduction and ecosystem restoration programs. of the biologic and physical processes that affect water quality
(Petersen et al., 2001) and evaluation of a monitoring network
(Ouyang, 2005).
The most commonly used multivariate statistical techniques
are cluster analysis (CA), discriminant analysis (DA), and
principal component analysis/factor analysis (PCA/FA),
although other statistical methods such as analysis of variance,
multiple linear regression, and receptor modeling have also been
used to supplement the evaluation (Table 1). Details of the
mathematics of CA, DA, and PCA/FA can be found elsewhere
(e.g., Johnson and Wichern, 1992; Fraley and Raftery, 2002).
Briefly, CA identifies groups (clusters) in the data based on the
similarities within a cluster and dissimilarities among clusters so
that objects can be grouped by variability (Fraley and Raftery,

Copyright © American Society of Agronomy, Crop Science Society of America, Y. Wan and C. Conrad, South Florida Water Management District, 3301 Gun Club
and Soil Science Society of America. 5585 Guilford Rd., Madison, WI 53711 USA. Road, West Palm Beach, FL 33406; Y. Qian, K.W. Migliaccio, and Y. Li, Soil and Water
All rights reserved. No part of this periodical may be reproduced or transmitted Science Dep. and Agricultural and Biological Engineering Dep. at Tropical Research
in any form or by any means, electronic or mechanical, including photocopying, and Education Center, IFAS, Univ. of Florida, 18905 SW 280th Street, Homestead, FL
recording, or any information storage and retrieval system, without permission in 33031. Assigned to Associate Editor Ying Ouyang.
writing from the publisher.
Abbreviations: CA, cluster analysis; CPC, common principal component; DA,
J. Environ. Qual. 43:599–610 (2014) discriminant analysis; FA, factor analysis; IRL, Indian River Lagoon; PCA, principal
doi:10.2134/jeq2013.09.0355 component analysis; SLE, St. Lucie Estuary; SFWMD, South Florida Water
Received 6 Sep. 2013. Management District; STA, storm-water treatment area; TFE, total iron; TKN, total
*Corresponding author ([email protected]). Kjeldahl nitrogen; TP, total phosphorus; TSS, total suspended solids.

599
2002; Kaufman and Rousseeuw, 1990; Wunderlin et al., 2001). Discriminant analysis is also useful for determining temporal and
Thus, the results of CA typically exhibit significant homogeneity spatial variations caused by natural and anthropogenic factors
within clusters and heterogeneity between clusters. Cluster linked to seasonality. Santos-Roman et al. (2003) related water
analysis is a convenient way to explore water quality patterns quality groups to physical characteristics of watersheds using DA
associated mostly with spatial variability (by sampling sites) and thus predicted water quality for unmonitored watersheds.
(e.g., Singh et al., 2004; Shrestha and Kazama, 2007; Zhang Both PCA and FA are exploratory methods concerned with
et al., 2009). Some researchers have also used CA for seasonal explaining the variance–covariance structure of the data. Principal
grouping (by months) to evaluation temporal variations (e.g., component analysis is generally applied for data reduction and
Zhou et al., 2007; Yang et al., 2010). Because it is not necessary interpretation of water quality data (e.g., Petersen et al., 2001;
to know the class characteristics of the data in advance, CA is Ouyang, 2005). Factor analysis, as an extension of PCA, aims
called unsupervised pattern recognition (Fraley and Raftery, 2002; to identify the underlying but unobservable quantities, called
Kowalkowski et al., 2006; Wunderlin et al., 2001). In contrast, factors, by analyzing the covariance relationships among multiple
DA is a statistical supervised pattern recognition method used to variables. This is commonly achieved through varimax rotation
discriminate a priori known groups or clusters (e.g., data from (orthogonal), which redistributes the variance of each variable
different seasons or regions) by determining the variables with to allow a high loading on a single factor and low loadings on
significant mean differences among them (Statsoft Inc., 1984; the other factors ( Johnson and Wichern, 1992). Thus, PCA
Fraley and Raftery, 2002; Insightful Corporation, 2005). is a linear combination of observable water quality variables,
Discriminant analysis has been applied to identify the most whereas FA identifies unobservable, hypothetical, latent variables
significant water quality parameters discriminating between (Vega et al., 1998). Principal component analysis and FA are
groups, and these parameters are considered to account for the commonly applied together in water quality data analysis to
spatial or temporal variations in water quality (Singh et al., 2004; identify pollution sources, i.e., naturally occurring (weathering or
Shrestha and Kazama, 2007). Thus, DA can render considerable geological processes) or anthropogenic (agricultural, industrial,
reduction of the dimensionality of the original data matrix. or domestic origins) (e.g., Wunderlin et al., 2001; Singh et al.,

Table 1. Selected surface water quality studies using multivariate techniques published since 2001.
Multivariate Auxiliary Study area Data analysis period and Sampling Water quality
Study Study objectives
techniques† analysis‡ (Country) sampling interval sites parameters
yr ——— no. ———
Wunderlin et CA, FA, DA Suquia River 2 (1998–2000, monthly) 9 22 spatial and temporal analysis, data
al. (2001) (Argentina) reduction, source identification
Petersen et al. PCA bootstrap Elbe River 5 (1994–1997, once every 14 7 identification of processes
(2001) procedure (Germany) 14 d) affecting water quality
Simeonov et CA, PCA RM River systems 3 (1997–2000, monthly) 25 27 spatial analysis, data reduction,
al. (2003) (Greece) source identification and
apportionment
Singh et al. CA, PCA/FA, Gomti River 5 (1994–1998, monthly) 8 24 spatial and temporal analysis, data
(2004) DA (India) reduction, source identification
Singh et al. CA, PCA/FA, RM Gomti River 3 (1999–2001, monthly) 8 34 spatial and temporal analysis, data
(2005) DA (India) reduction, source identification
and apportionment
Ouyang (2005) CA, FA St. Johns River 3 (1999–2001, daily or 22 42 evaluation of monitoring network,
(USA) monthly) identification of essential
parameters
Zhou et al. CA, DA rivers (Hong 5 (2000–2004, monthly) 23 23 spatial and temporal analysis, data
(2007) Kong) reduction. source identification
Shrestha and CA, PCA/FA, Fuji River (Japan) 8 (1995–2002, monthly) 13 25 spatial and temporal analysis, data
Kazama (2007) DA reduction, source identification
Pejman et al. CA, PCA/FA Haraz River (Iran) 2 (2007–2008, seasonal) 8 10 spatial and temporal analysis,
(2009) source identification
Zhang et al. CA, PCA/FA, Xiangjiang River 7 (1994– 2000) 34 12 spatial analysis, data reduction,
(2009) (China) source identification
Varol and Şen CA, PCA/FA Behrimaz Stream 1 (2003, monthly) 4 20 spatial and temporal analysis, data
(2009) (Turkey) reduction, source identification
Yang et al. CA, PCA/FA, IDW Lake Dianchi 5 (2003–2007, monthly) 8 12 spatial and temporal analysis, data
(2010) DA (China) reduction, source identification
Koklu et al. PCA/FA,DA MLR Melen River 11 (1995–2006, once 2–3 5 26 data reduction, source
(2010) (Turkey) mo) identification
Li et al. (2011) CA, PCA/FA RM, ANOVA 19 rivers (China) 2 (4 sampling trips in 2006 19 21 spatial and temporal analysis,
and 2007) source identification, source
apportionment
Mustapha and PCA/FA MLR Jakara River 0.17 (July 31 to Sept. 30, 4 15 source identification
Abdu (2012) (Nigeria) 2011, daily)
† CA, cluster analysis; PCA, principal components analysis; FA, factor analysis; DA, discriminant analysis.
‡ RM, receptor modeling; MLR, multiple linear regression; IDW, inverse distance weighting; ANOVA, analysis of variance.

600 Journal of Environmental Quality

2004; Shrestha and Kazama, 2007; Li et al., 2011). Receptor 2694 km2. The large primary canals managed by the South
modeling is also used in conjunction with PCA/FA to quantify Florida Water Management District (SFWMD) and the U.S.
source contributions associated with each factor (e.g., Singh et Army Corps of Engineers (USACE), including C-44 (completed
al., 2005; Li et al., 2011). in 1924 and enlarged to its current size in 1949), C-23, C-24,
Several observations are worth noting with respect to and C-25 (completed about 1961), were constructed to allow
applications of multivariate techniques in water quality widespread agricultural and urban development in the watershed.
evaluation. First, most studies have dealt with water quality In particular, the C-44 canal connects Lake Okeechobee, the
issues in rivers that are free flowing, with distinct nonpoint and second largest lake in the contiguous United States, with the
domestic and/or industrial pollution sources in the drainage SLE, allowing large volumes of lake water released for flood
basins (e.g., Singh et al., 2005; Zhou et al., 2007; Mustapha control into the estuary. Because there are two distinct seasons
and Abdu, 2012). For example, Zhou et al. (2007) found that (wet and dry) in South Florida, the canals act as a riverine
the highly polluted zones of the river system in Hong Kong system during the wet season ( June–October), providing flood
were associated with untreated sewage and domestic sources, protection through rapid transport of storm water into the
whereas the relatively low-pollution zones were free of major estuary. During the dry season (November–May), the canal
point sources. Li et al. (2011) indicated that the highly polluted system stores water for irrigation and maintains water table
areas in rivers along the water conveyance system of China elevations at coastal water control structures to prevent saltwater
were associated with industrial and domestic wastewaters. Few intrusion. Discharges occur only when the canal stage exceeds a
studies have examined data collected in highly managed “canal” critical level during a storm event. Due to the high water table
systems where water and land management are the single most and flat terrain in the region, runoff is delivered to the primary
important anthropogenic factor influencing water quality. canals mostly through water control structure such as culverts or
Second, when spatial variations in water quality were evaluated, pumps. Table 2 summarizes the basin characteristics of the study
source identification using PCA/FA was exercised by simply area.
contemplating the nature of water quality constituents within a The altered drainage pattern and increased hydraulic and
factor (e.g., Koklu et al., 2010; Mustapha and Abdu, 2012). It can pollutant loading to the estuary, especially from Lake Okeechobee
be difficult to relate all water quality parameters within a factor releases, have become a growing concern for the IRL/SLE system
to pollution sources, especially when several factors and multiple (Chamberlain and Hayward, 1996; Sigua et al., 2000; Wan et
parameters are involved. Few studies have validated the results al., 2006, 2012). A recent water quality assessment conducted by
of multivariate techniques with data pertaining to pollution the Florida Department of Environmental Protection (FDEP)
sources or management activities that cause the spatial variations. indicated that dissolved O2 (DO) and nutrients are common
This study attempted to expand on previous research parameters causing water quality impairment throughout the
on water quality applications using multivariate statistical watershed (Florida Department of Environmental Protection,
techniques to a managed “canal” system where land use and 2008). To address these issues, the SFWMD and the USACE
water management were uniquely integrated into the analysis developed a comprehensive restoration plan with a budget of
to determine their impacts on water quality. Twenty-five years US$1 billion to construct reservoirs and treatment wetlands
of monthly water quality data collected at seven stations in a (i.e., storm-water treatment areas, STAs) to restore historic
dense canal network discharging into the Indian River Lagoon flow patterns and to reduce nutrient loads (U.S. Army Corps
(IRL), Florida, were used. Specifically, the objective of the study of Engineers and South Florida Water Management District,
was to demonstrate the utility of the multivariate techniques in 2004). Meanwhile, the FDEP developed total maximum daily
(i) delineating the spatial variability of water quality in the canal loads for the SLE using total N and total P (TP) concentrations
network where water and land resources are highly managed for to set targets for load reduction allocations (Florida Department
agricultural production and flood control, and (ii) linking the of Environmental Protection, 2008). The Florida legislature also
spatial variations in water quality with specific water and land mandated that the SFWMD, in coordination with the FDEP and
management practices. To aid in the analysis, land use coverage the Florida Department of Agriculture and Consumer Services,
and hydrometric data (canal flow and rainfall depth) were develop and implement a watershed protection plan to reduce
integrated into a PCA/FA of the water quality data. nutrient loading to the SLE (South Florida Water Management
District, 2009). Essential to these restoration and load reduction
Materials and Methods programs is identifying water quality characteristics of surface
water delivered into the IRL/SLE system.
Study Area
The IRL, located on the southeast coast of Florida (Fig. 1), is Data
regarded as one of the most biologically diverse ecosystems in Freshwater discharges through the primary canals (C-44,
North America, with approximately 2200 identified plant and C-23, C-24, and C-25) are controlled through seven water
animal species (Swain et al., 1995). The St. Lucie Estuary (SLE) control structures: S-308, S-80, S-97, S-48, S-49, S-99, and
is the largest tributary to the IRL. Before development, the S-50 (Fig. 1). The naming convention for each station is the
IRL/SLE watershed was characterized by nearly level terrain combination of the primary canal and the structure; for example,
and poorly drained forested wetlands and wet prairie. During the station at S-308 in the C-44 canal is C44S308. Water quality
the last century, land use and drainage patterns have undergone and flow data measured at these stations were downloaded from
substantial changes after construction of a network of primary, the SFWMD database. Rainfall data collected at 11 rainfall
secondary, and tertiary canals (Fig. 1) draining a total area of stations in the region were used to derive the average rainfall
www.agronomy.org • www.crops.org • www.soils.org 601
Fig. 1. Location of water monitoring stations (red circles) and their respective drainage basins in the southern Indian River Lagoon watershed.
The drainage canal network and surface-water retention reservoirs are shown in blue. Note that the large reservoir in the C-44 basin belongs to a
power plant (not for water retention in citrus land).

depth for each basin using the Thiessen Polygon method. The through secondary and tertiary canals, land use of the “buffer
period of record for the analysis was from 1981 through 2004. zone” along the primary canal was considered an insignificant
The 12 water quality parameters selected for analysis in this factor influencing spatial variation in water quality (Carey et al.,
study were: DO (mg/L), specific conductivity (mS/cm), pH, 2011), and thus the “buffer zone” concept was not examined in
turbidity (nephelometric turbidity units, NTU), color (Pt–Co this study.
units, PCU), total suspended solids (TSS, mg/L), NO3–N +
NO2–N (NOX–N, mg/L), NH4–N (mg/L), total Kjeldahl N Data Treatment and Statistical Methods
(TKN, mg/L), PO4–P (mg/L), TP (mg/L), and total Fe (TFe, Before performing statistical analyses, the data were
mg/L). Analytical methods are summarized in Table 3. logarithmically transformed to normalize the distribution of
The SFWMD developed land use/land cover GIS data layers each water quality parameter and minimize the effects of outliers.
for 1988, 1995, 1999, and 2004 by photointerpretation of aerial All statistical analyses were performed using S-Plus (Version 7.0,
photography and digital orthophotographic quarter quadrangles. Insightful Corporation, 2005).
Each layer was processed to derive the percentage of major Cluster Analysis
land use types for each basin. Land use types were aggregated
into seven categories: pasture, citrus, other agriculture, urban For each station, the mean values of the transformed data
(including transportation), wetland, forest, and water (Table 2). were normalized to minimize the effects of the scale of the units
These categories are reflective of basin-specific land and water on the clustering, so that each variable was considered equally
management practices and are consistent with the land use important (Kaufman and Rousseeuw, 1990):
types used for hydrologic simulations of the basins (Wan et al., x if - m f [1]
zif =
2006). Because flows are routed into the primary canals mainly sf

602 Journal of Environmental Quality

where mf and sf are the mean and the Table 2. Characteristics of the C-23, C-24, C-25, and C-44 basins.
mean absolute deviation, respectively, of Characteristic C-23 C-24 C-25 C-44
the logarithmically transformed variable Basin area, km2 454.9 355.1 437.2 471.8
f across all stations; and xif and zif are the Land use, %†
original and normalized mean values, Pasture 36.3(2.1) 46.0(3.1) 23.3(5.9) 20.3(1.9)
respectively, of the transformed variable f Citrus 28.5(1.8) 18.7(3.1) 47.4(9.2) 32.5(1.1)
at the ith station. Other agriculture 6.5(4.9) 6.3(4.4) 5.5(0.5) 8.6(2.6)
The agglomerative hierarchical Urban 3.9(1.0) 5.1(1.2) 3.1(0.8) 7.3(2.6)
clustering method, which merges each Wetland 16.4(4.8) 14.9(1.9) 10.2(2.0) 14.4(5.5)
group until all observations are in a single Forest 6.7(3.8) 7.4(6.6) 6.6(6.4) 13.0(4.1)
group, was adopted to cluster the seven Water 1.6(0.9) 1.7(0.7) 4.0(3.6) 3.9(2.3)
monitoring stations. In this method, each Rainfall (depth)‡
object is initially considered as a separate Annual mean, mm/yr 1285 1303 1348 1341
group (cluster); the two groups with the Dry-season monthly mean, mm/mo 66 67 69 70
smallest dissimilarity are then merged. Wet-season monthly mean, mm/mo 165 167 173 170
The dissimilarity (distance) between Flow (depth)‡
two objects, d(i,j), was computed using Annual mean, mm/yr 427 482 390 993
Euclidean distance: Dry-season monthly mean, mm/mo 23 19 15 74
2
Wet-season monthly mean, mm/mo 52 68 56 92
å f =1( zif - z jf )
p [2]
d (i , j ) = † The mean percentages of different land use types were calculated from the South Florida Water
Management District land use GIS layers for 1988, 1995, 1999, and 2004. Standard deviations are
included in the parentheses.
where i and j represent different stations
and p is the number of variables (Kaufman ‡ Based on long-term hydrologic data (1965–2005) collected by the South Florida Water Management
District. The C-44 flow was measured at C44S80 and included releases from Lake Okeechobee.
and Rousseeuw, 1990). Distances between
clusters were determined by linkages. seasonality. The second data set included annual means of water
Ward’s and group-average linkage algorithms were used for cross- quality constituents plus land use coverage and hydrometric data
validation of the results. (canal flow and rainfall depth) for each basin. The percentage of
Discriminant Analysis each land use category for each survey year was assigned for the
years in between (four or five preceding survey years) assuming
Discriminant analysis was applied to identify individual
there was no significant land use change during that period. The
variables that discriminate between groups identified by CA. Several
number of principal components in PCA with eigenvalues >1
models have been developed for discriminating between k groups
was selected as the number of factors to use in FA to extract the
based on the relationships among the groups’ covariance matrices,
loading scores (e.g., Singh et al., 2004; Zeng and Rasmussen,
ranging from homoscedastic to heteroscedastic models (Insightful
2005; Shrestha and Kazama, 2007). This two-step PCA/FA
Corporation, 2005). The common principal component (CPC)
exercise allowed better detection of links between water quality
model was selected to identify the discriminating constituents, i.e.,
and basin-specific water and land management practices by
station means calculated from logarithmically transformed data,
comparing the loading values in rotated components. To verify
with CA groups as the dependent variable. The CPC is a robust
nonlinear model developed from PCA. In the CPC model, the Table 3. Analytical methods for selected water-quality parameters.
covariance matrices of different k groups are assumed to have Water-quality Analytical Detection
some common features (i.e., the k groups’ covariance structures are parameter method† limit‡
simultaneously diagonalizable) (Flury, 1984). To verify the results Dissolved O2 EPA 360.1 0.1 mg/L
from the CPC model, DA was also performed using a canonical Specific conductivity EPA 120.1 1 mS/cm
discriminant function based on the homoscedastic covariance pH EPA 150.1 –
structure, as well as a classical heteroscedastic model. Discriminant Turbidity SM 2130B 0.1 NTU§
analysis was performed using all data (year around) and seasonal Color SM 2120B 1.0 PCU¶
data representing the dry (November–May) or wet (June– Total suspended solids EPA 160.2 1 mg/L
October) seasons (Qian et al., 2007). The Kruskal–Wallis test NOX–N SM 4500NO3F 0.004 mg/L
(significance level a = 0.01) was performed to test the significance NH4–N SM 4500-NH3H 0.008 mg/L
of differences among the CA groups. Total Kjeldahl N EPA 351.2 0.05 mg/L
Principal Component Analysis/Factor Analysis PO4–P SM 4500PF 0.002 mg/L
Total P SM 4500PF 0.002 mg/L
To further elucidate the relationship between spatial variations
Total Fe SM 3120 B 1 mg/L
in water quality and basin management characteristics, PCA/FA
with varimax rotation (orthogonal solution) was conducted with † EPA methods are from the USEPA; SM methods are from the American
Public Health Association.
two sets of data. The first data set was the annual means of water
‡ Detection limits could vary due to the improvement of laboratory
quality data for each basin. Station C44S308 was excluded from techniques during the study period (1981–2004).
this analysis because basin runoff was primarily discharged via § Nephelometric turbidity units.
C44S80. Annual means were used to remove the influence of
¶ Platinum–cobalt units.

www.agronomy.org • www.crops.org • www.soils.org 603

the PCA/FA result, Pearson correlation analysis of the annual
means of water quality constituents against land use percentage,
rainfall, and flow rate was conducted. These data were also ranked
for Spearman’s rank correlation analysis for cross-validation. The
significance criterion of the correlation analyses was p ≤ 0.01.

Results
Groups of Monitoring Stations
The classification patterns obtained by both group-average
linkage and Ward’s method using Euclidean distance were
similar, and only the dendrogram obtained from the group-
average method is presented (Fig. 2). The seven monitoring
Fig. 2. Dendrogram of the cluster analysis of seven water quality
stations were clustered into four groups. Group 1 (G1) included monitoring stations on the C-23, C-24, C-25, and C-44 canals.
C23S48, C23S97, and C24S49; Group 2 (G2) included
C25S50 and C25S99; Group 3 (G3) consisted of C44S80; and explaining 75% of the total variance of the water quality data set
Group 4 (G4) consisted of C44S308. This grouping is consistent (Table 6). A high loading of a variable on a factor shows a strong
with the basin delineation and water/land management. Group relationship between the factor and the respective variable. Liu
4 (C44S308) had the greatest separation from the other stations. et al. (2003) classified the significant loadings as strong (absolute
Group 3 (C44S80) also had distinctly different features than the loading value >0.75), moderate (0.50–0.75), and weak (0.30–
other groups. These two stations control the discharge from Lake 0.50). This classification was commonly adopted in later studies
Okeechobee and C-44 basin runoff to the SLE (Fig. 1). Group (e.g., Ouyang, 2005; Singh et al., 2005). In the present study,
1 stations receive flows from the C-23 and C-24 basins, which factor loadings >0.5 were considered significant. As shown in
have similar land use and development history, with pasture as Table 6, the first factor, explaining 26.88% of the total variance,
the predominant land use (36 and 43%, respectively) (Table 2). had significant positive loadings on TP, PO4–P, NH4–N, TKN,
In contrast, G2 stations receive storm water from the C-25 basin, and color. The second factor, explaining 18.47% of the total
with citrus being by far the dominant land use (47%). variance, had positive loadings on TSS, turbidity, and NOx–N.
The third factor, explaining 16.48% of the total variance, had
Spatial Variation of Discriminating Constituents positive loadings on pH and DO and a negative loading on color.
The DA results obtained from the CPC model, the canonical The fourth factor, explaining 13.44% of the total variance, had a
discriminant function, and the classical heteroscedastic model negative loading on specific conductivity and a positive loading
provided the same result (Table 4). Dissolved O2, specific on TFe.
conductivity, pH, color, TSS, and turbidity were selected as The PCA/FA of the annual means of the water quality data as
discriminating constituents using all data or seasonal data. well as rainfall depth, flow, and land use percentage identified six
Among the nutrients, PO4–P, TP, and NH4–N were also selected factors with eigenvalues >1, explaining 76% of the total variance
as discriminating constituents using all data or seasonal subsets; of the data set (Table 7). The first factor explained 23.73% of the
NOX–N was a discriminating constituent only for the dry-season total variance, with significant positive loadings on TP, PO4–P,
data, while TKN and TFe were discriminating constituents NH4–N, TKN, color, and pasture and negative loadings on citrus
using all data and dry-season data. During the dry season, all CA and water. The second factor, explaining 13.97% of the total
groups exhibited greater spatial heterogeneity than in the wet variance, had positive loadings on TSS, turbidity, NOx–N, and
season, reflecting the influence of seasons on water quality (Qian flow. The third factor, explaining 10.94% of the total variance,
et al., 2007). had significant positive loadings on rainfall, flow, and TFe and a
All water quality constituents showed distinct spatial patterns
(Table 5). The p values derived from the Kruskal–Wallis test for Table 4. Water quality discriminating constituents identified by
water quality constituents were all <0.01, indicating significant discriminant analysis.
spatial variation in water quality among CA groups. For example, Water-quality Dry-season Wet-season
All data
G1 had the highest color, TP and PO4–P; G2 had the lowest parameters data data
DO, turbidity, and TP and the highest specific conductivity Dissolved O2 yes yes yes
and NH4–N; while G4 had the highest DO, turbidity, TSS, and Specific conductivity yes yes yes
TKN and the lowest NH4–N, specific conductivity, and color. pH yes yes yes
Both G3 and G4 had greater DO, turbidity, and TSS and lower Turbidity yes yes yes
specific conductivity and color than G1 and G2. The median Color yes yes yes
values of NOX–N were similar between G3 and G4, as well as Total suspended solids yes yes yes
between G1 and G2, although G3 and G4 exhibited higher NOX–N no yes no
medians than G1 and G2. NH4–N yes yes yes
Total Kjeldahl N yes yes no
Two-Step Principal Component Analysis and Factor Analysis PO4–P yes yes yes
The PCA/FA of the annual mean concentrations of the water Total P yes yes yes
quality constituents identified four factors with eigenvalues >1, Total Fe yes yes no

604 Journal of Environmental Quality

Table 5. Descriptive statistics of water quality data on dissolved O2 (DO), specific conductivity, pH, turbidity (in nephelometric turbidity units, NTU),
color (in platinum–cobalt units, PCU), total suspended solids (TSS), NOx–N, NH4–N, total Kjeldahl N (TKN), PO4–P, total P (TP), and total Fe (TFe) for
the four groups identified by cluster analysis.
Specific
Statistic DO pH Turbidity Color TSS NOX–N NH4–N TKN PO4–P TP TFe
conductivity
mg/L mS/cm NTU PCU ———————————— mg/L ———————————— mg/L
Group 1 (C24S49, C23S48, and C23S97)
Min. 0.1 249 5.13 0.7 10 –† – – – – 0.025 –
Mean 5.0 945 7.23 3.9 129 3 0.11 0.10 1.33 0.183 0.290 428
Median 4.9 a‡ 857 c 7.20 a 3.0 b 110 d 2a 0.07 a 0.06 c 1.30 c 0.161 c 0.262 c 342 a
Max. 12.9 2140 8.45 52.0 400 92 1.64 0.87 10.48 0.879 1.400 2300
Skewness 1.91 0.76 0.01 6.10 0.70 5.39 3.46 1.77 3.72 1.29 1.25 1.78
Group 2 (C25S50 and C25S99)
Min. 0.3 286 5.94 0.4 20 – – – – – – –
Mean 4.8 1072 7.23 3.3 90 4 0.12 0.08 1.24 0.085 0.162 325
Median 4.6 a 1029 d 7.19 a 2.2 a 80 c 2a 0.09 a 0.07 d 1.22 b 0.062 b 0.127 a 277 a
Max. 12.8 2920 8.63 170.0 272 356 1.29 1.42 6.19 0.710 1.046 2100
Skewness 0.46 1.06 0.39 17.02 0.99 12.89 3.39 5.42 0.88 2.30 2.17 2.54
Group 3 (C44S80)
Min. 0.8 125 5.45 1.1 5 – – – – 0.005 0.045 40
Mean 6.1 666 7.51 9.4 70 8 0.23 0.05 1.21 0.085 0.156 381
Median 6.3 b 619 b 7.51 b 4.5 c 55 b 5b 0.22 b 0.04 b 1.13 a 0.064 b 0.134 a 286 a
Max. 12.4 1937 8.91 90.3 221 110 0.89 0.44 5.87 0.368 0.446 2420
Skewness −0.01 1.48 −0.34 3.32 1.28 3.28 0.96 2.21 0.10 1.65 1.16 3.54
Group 4 (C44S308)
Min. 0.3 260 6.30 1.5 2 – – – – – 0.050 28
Mean 6.9 545 7.79 37.1 57 37 0.25 0.05 1.64 0.065 0.175 939
Median 7.2 c 531 a 7.82 c 24.1 d 43 a 20 c 0.22 b 0.02 a 1.49 d 0.04 7a 0.146 b 595 b
Max. 12.6 3500 9.09 321.0 416 541 1.72 0.70 7.96 0.977 1.088 4673
Skewness −0.39 8.24 −0.30 2.79 3.10 4.97 1.71 2.64 3.41 5.47 3.68 1.95
† Value below detection limits.
‡ Medians followed by the same letter are not significantly different at a = 0.01 with pairwise least significance difference contrast of the Kruskal–Wallis test.

significant negative loading on specific conductivity. The fourth with specific conductivity (negative), TSS (positive), turbidity
factor, explaining 10.48% of the total variance, had significant (positive), NOx–N (positive), and TFe (positive).
positive loadings on pH, DO, urban, and other agriculture land.
The fifth factor, explaining 8.86% of the total variance, had a Discussion
significant negative loading on forest and a significant positive While spatial variations in water quality were logically linked
loading on TKN. The sixth factor, explaining 7.83% of the total to contributing sources associated with domestic or industrial
variance, had significant negative loadings on DO and wetland. origins in the literature (e.g., Simeonov et al., 2003; Zhou et
The Pearson and Spearman’s rank correlation procedures
resulted in similar correlations between water quality Table 6. Rotated loading matrix (varimax) using the annual mean
concentrations of water quality constituents. Significant factor
constituents and land use percentage, rainfall, and flow, with loadings are in bold type.
Pearson giving a stronger diagnosis in some cases, and thus only
Water-quality
the Pearson correlation matrix is reported (Table 8). Pasture and parameter
Factor 1 Factor 2 Factor 3 Factor 4
citrus are perhaps the most dominant land use types influencing Total P 0.92 −0.14 −0.05 0.14
water quality. For example, TP, PO4–P, TKN, and NH4–N PO4–P 0.80 −0.26 −0.20 0.25
were all significantly correlated with pastures (positive) and NH4–N 0.79 0.13 −0.16 −0.40
citrus (negative). Associated with this correlation pattern was Total Kjeldahl N 0.69 0.00 0.12 0.00
significant correlation (negative) between these nutrients and Color 0.64 −0.29 -0.59 0.16
the water land use type. Significant correlations (negative) Total suspended solids 0.00 0.86 0.04 0.02
between forest and TP, TKN, and NH4–N were detected. Urban Turbidity −0.02 0.78 0.26 0.39
land also showed significant correlations with some of these NOx–N −0.23 0.73 0.07 0.18
constituents. Color was significantly correlated with pasture pH 0.13 0.11 0.88 0.12
(positive), citrus (negative), and water (negative), while DO was Dissolved O2 −0.18 0.08 0.80 0.00
significantly correlated with pasture (negative), other agriculture Specific conductivity 0.08 −0.28 −0.26 −0.80
(positive), urban (positive), and wetland (positive). Rainfall was Total Fe 0.37 0.25 −0.34 0.72
significantly correlated with DO (negative), specific conductivity Eigenvalues 3.23 2.22 1.98 1.61
(negative), and TFe (positive). Flow was significantly correlated Variance explained, % 26.88 18.47 16.48 13.44

www.agronomy.org • www.crops.org • www.soils.org 605

Table 7. Rotated loading matrix (varimax) using the annual mean concentrations of water quality constituents and percentage of land uses, rainfall
and flow data. Significant factor loadings are in bold type.
Water-quality parameter Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6
Pasture 0.87 −0.19 −0.21 −0.26 0.08 −0.13
PO4–P 0.84 −0.24 0.20 0.07 0.05 0.06
Total P 0.83 −0.14 0.11 0.17 0.26 0.08
Citrus −0.81 −0.05 0.16 0.04 0.32 0.35
Color 0.78 −0.24 0.27 −0.36 0.09 0.11
Water −0.69 −0.09 0.31 −0.21 −0.14 0.19
NH4–N 0.61 0.11 −0.26 −0.12 0.50 0.15
Turbidity −0.05 0.83 0.23 0.33 −0.12 −0.06
Total suspended solids −0.11 0.78 −0.03 0.07 0.10 0.17
NOx–N −0.19 0.71 0.08 0.07 −0.22 −0.25
Flow −0.05 0.68 0.51 0.10 −0.11 −0.05
Rainfall −0.13 0.01 0.76 −0.22 0.04 0.26
Total Fe 0.41 0.32 0.70 −0.02 0.04 0.14
Specific conductivity 0.17 −0.28 -0.70 −0.36 −0.01 0.24
Other agriculture 0.09 0.09 0.07 0.88 0.14 0.20
pH −0.14 0.10 −0.07 0.86 0.25 −0.13
Urban 0.17 0.40 0.04 0.74 −0.30 0.08
Forest −0.05 0.32 −0.03 −0.15 −0.85 0.00
Total Kjeldahl N 0.50 0.09 0.09 0.09 0.60 −0.14
Dissolved O2 −0.30 0.08 −0.20 0.53 −0.09 −0.54
Wetland 0.27 0.02 −0.08 −0.17 0.09 −0.90
Eigenvalues 4.98 2.93 2.30 2.20 1.81 1.65
Variance explained, % 23.73 13.97 10.94 10.48 8.60 7.83

al., 2007; Li et al., 2011), source identification in this study was transport mechanisms such as leaching, dilution, and wash-off
not readily apparent with PCF/FA of the water quality data phenomena (Helsel and Hirsch, 1992; Ravichandrana et al.,
alone (Table 6). While the first factor in Table 6 (TP, PO4–P, 1996; Swanson et al., 2000). Our two-step PCA/FA suggested
NH4–N, TKN, and color) can be considered nutrient related, it that there existed a latent factor for TSS, turbidity, and NOx–N
is difficult to identify specific pollution sources or anthropogenic (Factor 2 in Table 6), and the factor was related to flow (Factor
activities in association with these factors. The PCA/FA of both 2 in Table 7). The most significant water resources management
water quality data and land use and hydrometric data (Table 7) practice affecting flow is flood control releases of freshwater from
revealed clear associations of specific water quality parameters Lake Okeechobee to the SLE via C-44 (Fig. 1). The separation of
with rainfall depth, flow, and land use, allowing identification of C44S308 and C44S80 from the remaining stations in the cluster
basin-specific water and land management practices that cause analysis (Fig. 2) reflects the influence of lake releases. A further
spatial variations in water quality. distinction between C44S308 and C44S80 shown in Fig. 2
was probably related to the dry-season water supply releases
Water Resources Management from Lake Okeechobee via C44S308, which were not readily
Spatial patterns of water quality have been linked to hydrologic captured by water quality samples collected at C44S80. High
processes and water resources management through pollutant flows at C44S80 and C44S308 associated with lake releases were

Table 8. Pearson correlation matrix between water quality constituents and rainfall, flow, and land use.
Other
Water-quality parameter Rainfall Flow Pasture Citrus Urban Wetland Forest Water
agriculture
Dissolved O2 −0.31** 0.07 −0.32** 0.06 0.26 0.30** 0.29** 0.05 −0.05
Specific conductivity −0.35** −0.48** 0.44** −0.19 −0.34** −0.32** −0.04 −0.07 −0.14
pH −0.19 0.11 −0.30** 0.17 0.69** 0.57** −0.03 −0.26 −0.20
Turbidity 0.04 0.78 −0.32** 0.01 0.30** 0.57** −0.05 0.30** −0.04
Color 0.22 −0.16 0.76** −0.54** −0.21 −0.22 0.17 −0.17 −0.34**
Total suspended solids 0.06 0.37** −0.24 0.11 0.16 0.30** −0.11 0.10 0.07
NOx −0.02 0.46** −0.37** −0.01 0.11 0.36** 0.15 0.38** 0.13
NH4–N −0.13 −0.18 0.54** −0.30** 0.03 −0.06 0.12 −0.33** −0.49**
Total Kjeldahl N 0.02 0.09 0.43** −0.27** 0.18 0.07 0.23 −0.39** −0.49**
PO4–P −0.01 −0.15 0.67** −0.55** 0.10 0.05 0.16 −0.21 −0.40**
Total P −0.05 −0.13 0.63** −0.48** 0.22 0.09 0.14 −0.32** −0.48**
Total Fe 0.41** 0.50** 0.13 −0.14 0.14 0.16 −0.08 0.00 −0.13
** Significant at a = 0.01.
606 Journal of Environmental Quality
probably the most significant factor contributing to their higher correlation of specific conductivity with rainfall (Table 8) also
DO, turbidity, TSS, NOx–N, and lower specific conductivity, suggests that lower specific conductivity in high-rainfall years
color, and NH4–N than at G1 and G2 stations (Table 5). is associated with less supplemental irrigation supplies from the
Specifically, the higher DO, turbidity, and TSS were probably Floridan Aquifer.
related to enhanced aeration of water and erosion of sediment
under high-flow conditions. Figure 3 shows that increases in Land Use and Storm-Water Retention
TSS and turbidity with flows at C44S80 can be well defined by Land use and development have long been shown to
linear relationships at monthly or annual time scales. The lower influence surface water quality, with Osborne and Wiley (1988)
color, lower NH4–N, and higher NOx–N were possibly because linking urbanization to water quality changes in the St. Fork
Lake Okeechobee water has a longer residence time than storm- River in Illinois, Long and Plummer (2004) attributing high
water runoff originating from the local basins, thereby allowing levels of specific conductance to dense residential development
photolysis and oxidation of these constituents in lake water in the Wachusett Reservoir watershed in Massachusetts, and
(Doering and Chamberlain, 1999). Yin et al. (2005) relating urban development patterns to shifts
Another latent factor identified by PCA/FA was specific in pollution sources in Shanghai, China. Water quality has also
conductivity and TFe (Factor 4 in Table 6), and this factor been evaluated with respect to the amount of arable land and
was linked to rainfall and flow (Factor 3 in Table 7). Total Fe is grassland (Ferrier et al., 2001), urban land and upland agriculture
mostly in particulate form, and positive correlation with rainfall (Santos-Roman et al., 2003; Zampella et al., 2007), and land
and flow is probably transport related. Correlation of specific development intensity and impervious percentage (Carey et al.,
conductivity with flow and rainfall depth is, however, due to the 2011). The two-step PCA/FA indicated that the nutrient-related
irrigation practice on citrus lands. Since large-scale expansion factor (TP, PO4–P, NH4–N, TKN, and color) was associated
of citrus in the 1960s, storage capacity in the drainage network with land management of pasture and citrus and the area of open
has been inadequate to meet irrigation demands. In the C-23, water in a basin (Factor 1 in Tables 6 and 7). The C-23 and C-24
C-24, and C-25 basins, a common practice is to use artesian basins had the greatest percentage of pasture (Table 2). Mean
well water from the Floridan Aquifer, a confined aquifer with concentrations of TP, PO4–P, NH4–N, and color in G1 were
high mineral content, as an irrigation supplement when surface the highest (Table 5). Graves et al. (2004) reported that color in
water becomes limited. In contrast, supplemental irrigation runoff from pasture was significantly higher than from citrus and
water is supplied by Lake Okeechobee in the C-44 basin (South urban land, and they attributed the high color to the leaching
Florida Water Management District, 2004). The different of organic materials (humic and tannic acids from vegetative
supplemental irrigation sources lead to the spatial variation in decay) into surface water. Cattle wastes in pastures are subject
specific conductivity across these basins. The higher value at G2 to storm-water runoff, and that is probably the reason why TP,
than at G1 corresponds with the greater citrus area in the C-25 PO4–P, TKN, and NH4–N were positively related to pasture.
basin than in the C-23 or C-24 basins (Table 2). The positive In addition, the C-23 and C-24 basins receive applications of

Fig. 3. Linear regressions of total suspended solids (TSS) and turbidity (in nephelometric turbidity units, NTU) with flow at C44S80 on (A,B) a
monthly time scale and (C,D) an annual time scale. All regressions are significant at a = 0.01.

www.agronomy.org • www.crops.org • www.soils.org 607

sewage sludge and biosolids to supplement fertilization, leaving vegetation and periphyton contributed to high DO in Florida’s
a large amount of legacy P in the soil. This practice is construed pristine wetlands.
to be partly responsible for the high color and P concentrations Frequent low DO concentrations during runoff events from
observed at G1. the watershed are a major stressor to the SLE (Chamberlain
Negative correlations between TP, PO4–P, NH4–N, TKN, and Hayward, 1996; Florida Department of Environmental
color, and the percentages of both citrus and water were because Protection, 2008). At all stations, DO values were periodically
most open water, representing storm-water retention reservoirs, below the Florida State Class III water quality standard (5
are embedded in citrus land. Storm-water management permits mg/L). The median DO was 4.9 mg/L for G1, 4.6 mg/L for
in large areas of pasture in the C-23 and C-24 basins allow direct G2, 6.3 mg/L for G3, and 7.2 mg/L for G4 (Table 5). Low
discharge of storm water into the C-23 and C-24 canals without DO occurred mostly in the wet season when the temperature
water detention. In contrast, the C-25 basin has a basin rule and nutrient concentrations in runoff were high. Figure 4
requiring on-site water detention for storm-water management indicated that DO was negatively correlated with temperature
in citrus land. This is evidenced by the highest areal percentage (only C24S49 data are shown because the pattern was similar
of open water as a land use type for the C-25 basin (Table 2, Fig. at other stations), supporting the argument by Graves et al.
1). Although these reservoirs are shallow (about 2 m in depth), (2004) in that low DO in South Florida is partly caused by high
they allow a large portion of storm water to be retained on site temperature in the summer (lower DO saturation point) and
before discharging into the C-25 canal. The water quality benefit inhibited re-aeration in canal environments. Interestingly, DO
is obvious, as shown in Table 5, for its lower color, turbidity, TSS, also correlated significantly with PO4–P, NH4–N, and TP in
TP, and PO4–P than for G1. The practical implication for this a similar fashion as with temperature (negative), but not with
spatial pattern is that the existing swale, block, and canal systems NOx–N (Fig. 4). While low DO can be due to inputs of nutrient-
for citrus in the C-44, C-23, and C-24 basins could be modified rich detritus in secondary and tertiary canals from agricultural
to increase retention while still being effective for drainage lands, which increased O2 demand, TP, PO4–P, NH4–N, DO,
during heavy storms. and temperature could simply be covariants associated with
seasonality (Qian et al., 2007). The complexity of enhanced
Wetland Preservation and Dissolved Oxygen primary productivity under elevated nutrient concentrations
Although wetlands, the third largest land use type in the and DO dynamics needs to be studied further to elucidate how
watershed, are known for their nutrient retention capability reducing nutrient loading can effectively improve DO conditions
(McCormick and Laing, 2003; Graves et al., 2004), their in South Florida.
association with nutrients was not significant (Tables 7 and 8).
Water quality measured at these stations represents a blend of Management Implications
runoff originating from all types of land use in the basins. The Issues associated with the quantity, quality, timing, and
net effect of wetlands on nutrient concentrations was probably distribution of freshwater discharged into the IRL/SLE system
overshadowed by that of pasture and citrus. The
only water quality parameter that was significantly
associated with wetlands was DO. While the
higher DO in C-44 was partly attributable to Lake
Okeechobee releases, the percentage of wetland was
found to be significantly correlated with DO across
the watershed (Tables 7 and 8). This is consistent
with the observation by Graves et al. (2004) about
higher DO in storm water from wetland than from
other major land uses in the watershed. Santos-
Roman et al. (2003) also reported that watersheds
with protected forest areas had higher DO in
Puerto Rico.
Dissolved O2 dynamics in wetland is complex
due to the influence of aquatic vegetation on the
rate of O2 exchange and metabolic O2 production
and consumption. Although low DO has been
observed in temperate wetlands (e.g., Rose and
Crumpton, 1996), relatively high DO is common
in wetlands in South Florida’s subtropical setting
(McCormick and Laing, 2003; Graves et al., 2004).
Higher DO in wetlands can be attributable to lower
temperature due to the shading effect of forest and/
or enhanced aeration in an open-water environment
in contrast to narrow canals in agricultural lands.
Fig. 4. Correlation of dissolved O (DO) with (A) temperature, (B) PO4–P, (C) NH4–N, and
McCormick and Laing (2003) also indicated (D) NOX–N measured at C24S49 (n2 = 310). All correlations are significant at a = 0.01
that the metabolic activities of submerged aquatic except for NOX–N.
608 Journal of Environmental Quality
have instigated federal and state restoration efforts at both and water resources management in a dense drainage/irrigation
regional and local scales (U.S. Army Corps of Engineers and canal network where storm-water runoff serves as the single
South Florida Water Management District, 2004; Florida most important nonpoint source of water pollution. While CA
Department of Environmental Protection, 2008; South Florida and DA helped to delineate the spatial patterns of water quality,
Water Management District, 2009). These restoration programs two-step PCA/FA of water quality data in conjunction with
address both water quantity management through regulation hydrometric and land use data allowed detection of associations
of flow discharge patterns and water quality treatment to between selected water quality parameters and specific land and
reduce nutrient concentrations in runoff. The water and water management practices. Such analyses can be invaluable on
material loads associated with controlled freshwater releases many occasions when intentions are to identify anthropogenic
from Lake Okeechobee represent an artificial stressor, causing activities that influence water quality, to explore possible
adverse impacts on water quality and ecological conditions in interactions between external drivers and water quality processes
the IRL/SLE (Chamberlain and Hayward, 1996; Wan et al., that may take place in the field, or to develop technically sound
2012). With the restoration of the Greater Everglades under pollution abatement programs.
the Comprehensive Everglades Restoration Plan (CERP)
at the regional level, it is anticipated that releases from Lake References
Okeechobee into the SLE will be substantially reduced from Carey, R.O., K.W. Migliaccio, Y. Li, B. Schaffer, G.A. Kiker, and M.T. Brown.
present levels (U.S. Army Corps of Engineers and South Florida 2011. Land use disturbance indicators and water quality variability
in the Biscayne Bay Watershed, Florida. Ecol. Indic. 11:1093–1104.
Water Management District, 2004). Accompanying this regional doi:10.1016/j.ecolind.2010.12.009
restoration effort will be significant decreases in TSS, turbidity, Chamberlain, R.H., and D. Hayward. 1996. Evaluation of water quality and
and N concentrations in the C-44 canal. These parameters were monitoring in the St. Lucie Estuary, Florida. Water Resour. Bull. 32:681–
among the major discriminating constituents associated with 696. doi:10.1111/j.1752-1688.1996.tb03466.x
Doering, P.H., and R.H. Chamberlain. 1999. Water quality and source of
Lake Okeechobee releases. freshwater discharge to the Caloosahatchee Estuary, Florida. J. Am. Water
Restoration efforts at the local scale under CERP aim to Resour. Assoc. 35:793–806. doi:10.1111/j.1752-1688.1999.tb04175.x
restore the inflow pattern to the hydrologic characteristics of the Ferrier, R.C., A.C. Edwards, D. Hirst, I.G. Littlewood, C.D. Watts, and R.
Morris. 2001. Water quality of Scottish rivers: Spatial and temporal trends.
predrainage or natural condition through the construction of Sci. Total Environ. 265:327–342. doi:10.1016/S0048-9697(00)00674-4
large storage reservoirs and STAs in the C-44, C-23, C-24, and Florida Department of Environmental Protection. 2008. TMDL Report:
C-25 basins (U.S. Army Corps of Engineers and South Florida Nutrient and dissolved oxygen TMDL for the St. Lucie Basin. FDEP,
Water Management District, 2004; Wan et al., 2006). Our Tallahassee, FL.
Flury, B.N. 1984. Common principal components in k groups. J. Am. Stat. Assoc.
analyses of spatial variations in watershed water quality suggest 79:892–898.
that emphasis should be placed on the C-23 and C-24 basins, Fraley, C., and A.E. Raftery. 2002. Model-based clustering, discriminant
where nutrient concentrations and color were the highest. This analysis, and density estimation. J. Am. Stat. Assoc. 97:611–631.
doi:10.1198/016214502760047131
supports the current design of the project, which includes the
Graves, G.A., Y. Wan, and D.L. Fike. 2004. Water quality characteristics of storm
C-23/C24 reservoirs and STAs as the largest facilities in the water from major land uses in South Florida. J. Am. Water Resour. Assoc.
watershed. Our observation of better water quality (P, color, and 40:1405–1419. doi:10.1111/j.1752-1688.2004.tb01595.x
TSS) at G2 than at G1 also suggests that the proposed storage Helsel, D.R., and R.M. Hirsch. 1992. Statistical methods in water resources.
Elsevier, Amsterdam.
reservoirs will have water quality benefits in addition to their Insightful Corporation. 2005. S-PLUS 7.0 guide to statistics. Vol. 2. Insightful
capacity to regulate flow patterns. Furthermore, these storage Corp., Seattle, WA.
reservoirs will provide a supplemental irrigation supply during Johnson, R.A., and D.W. Wichern. 1992. Applied multivariate statistical analysis.
the dry season (Wan at el., 2006); thus, specific conductivity in 3rd ed. Prentice–Hall, Englewood Cliffs, NJ.
Kaufman, L., and P.J. Rousseeuw. 1990. Finding groups in data: An introduction
G1 and G2 will become lower when reliance on the Floridan to cluster analysis. John Wiley & Sons, New York.
Aquifer is reduced. Koklu, R., B. Sengorur, and B. Topal. 2010. Water quality assessment
In conjunction with constructing CERP reservoirs and STAs using multivariate statistical methods—A case study: Melen river
system (Turkey). Water Resour. Manage. 24:959–978. doi:10.1007/
is an effort to rehydrate about 36,422 ha of drained wetlands s11269-009-9481-7
throughout the watershed and an ongoing regulatory source Kowalkowski, T., R. Zbytniewski, J. Szpejna, and B. Buszewski. 2006. Application
control program to implement best management practices of chemometrics in river water classification. Water Res. 40:744–752.
(BMPs) (South Florida Water Management District, 2009). doi:10.1016/j.watres.2005.11.042
Li, S., J. Li, and Q. Zhang. 2011. Water quality assessment in the rivers along
Storm-water retention and filtering capabilities of forests and the water conveyance system of the Middle Route of the South to North
wetlands have been shown to improve water quality (Santos- Water Transfer Project (China) using multivariate statistical techniques
Roman et al., 2003; Graves et al., 2004). Ongoing BMPs specific and receptor modeling. J. Hazard. Mater. 195:306–317. doi:10.1016/j.
jhazmat.2011.08.043
to citrus and pastures will be especially effective due to their Liu, C.-W., K.-H. Lin, and Y.-M. Kuo. 2003. Application of factor analysis in the
substantial spatial extent and correlation with water quality assessment of groundwater quality in a blackfoot disease area in Taiwan.
constituents (N, P, and color) in the watershed. Sci. Total Environ. 313:77–89. doi:10.1016/S0048-9697(02)00683-6
Long, S.C., and J.D. Plummer. 2004. Assessing land use impacts on water quality
Conclusions using microbial source tracking. J. Am. Water Resour. Assoc. 40:1433–
1448. doi:10.1111/j.1752-1688.2004.tb01597.x
Multivariate statistical techniques including CA, DA, and McCormick, P.V., and J.A. Laing. 2003. Effects of increased phosphorus loading
PCA/FA are standard tools for analyzing multidimensional on dissolved oxygen in a subtropical wetland, the Florida Everglades.
Wetlands Ecol. Manage. 11:199–216. doi:10.1023/A:1024259912402
water quality data (e.g., Simeonov et al., 2003; Shrestha and Mustapha, A., and A. Abdu. 2012. Application of principal component analysis
Kazama, 2007). This study demonstrated the use of multivariate and multiple regression models in surface water quality assessment. J.
techniques to link spatial variations in water quality with land Environ. Earth Sci. 2:16–23.

www.agronomy.org • www.crops.org • www.soils.org 609

Osborne, L.L., and M.J. Wiley. 1988. Empirical relationships between land use/ Statsoft Inc. 1984. Discriminant function analysis. Statsoft Inc.. Tulsa, OK. www.
cover and stream water quality in an agricultural watershed. J. Environ. statsoft.com/textbook/stdiscan.html.
Manage. 26:9–27. Swain, H.M., D.R. Breininger, D.S. Busby, K.B. Clark, S.B. Cook, R.A. Day, et
Ouyang, Y. 2005. Evaluation of river water quality monitoring stations by al. 1995. Introduction to the Indian River Biodiversity Conference. Bull.
principal component analysis. Water Res. 39:2621–2635. doi:10.1016/j. Mar. Sci. 57:1–7.
watres.2005.04.024 Swanson, F.J., F.N. Scetena, G.E. Dissmeyer, M.E. Fenn, E.S. Verry, and J.A.
Papatheodorou, G., G. Demopoulou, and N. Lambrakis. 2006. A long-term Lynch. 2000. Watershed processes: Fluxes of water, dissolved constituents,
study of temporal hydrochemical data in a shallow lake using multivariate and sediment. In: G.E. Dissmeyer, editor, Drinking water from forests and
statistical techniques. Ecol. Modell. 193:759–776. doi:10.1016/j. grasslands: A synthesis of the scientific literature. Rep. SRS-39. U.S For.
ecolmodel.2005.09.004 Serv., South. Res. Stn., Asheville, NC.
Pejman, A.H., G.R. Nabi Bidhendi, A.R. Karbassi, N. Mehrdadi, and M. Esmaeili U.S. Army Corps of Engineers and South Florida Water Management District.
Bidhendi. 2009. Evaluation of spatial and seasonal variations in surface 2004. Central and Southern Florida Project: Indian River Lagoon—South.
water quality using multivariate statistical techniques. Int. J. Environ. Sci. Final Integrated Project Implementation Report and Environmental
Technol. 6:467–476. doi:10.1007/BF03326086 Impact Statement. USACE and SFWMD, Jacksonville, FL.
Petersen, W., L. Bertino, U. Callies, and E. Zorita. 2001. Process identification Varol, M., and B. Şen. 2009. Assessment of surface water quality using multivariate
by principal component analysis of river water quality data. Ecol. Modell. statistical techniques: A case study of Behrimaz Stream, Turkey. Environ.
138:193–213. doi:10.1016/S0304-3800(00)00402-6 Monit. Assess. 159:543–553. doi:10.1007/s10661-008-0650-6
Qian, Y., K.W. Migliaccio, Y. Wan, Y.C. Li, and D. Chin. 2007. Seasonality of Vega, M., R. Pardo, E. Barrado, and L. Deban. 1998. Assessment of seasonal and
selected surface water constituents in the Indian River Lagoon, Florida. J. polluting effects on the quality of river water by exploratory data analysis.
Environ. Qual. 36:416–425. doi:10.2134/jeq2006.0185 Water Res. 32:3581–3592. doi:10.1016/S0043-1354(98)00138-9
Ravichandrana, S., R. Ramanibai, and N.V. Pundarikanthan. 1996. Ecoregions Wan, Y., Z.-G. Ji, J. Shen, G. Hu, and D. Sun. 2012. Three dimensional water
for describing water quality patterns in Tamiraparani basin, South India. J. quality modeling of a shallow subtropical estuary. Mar. Environ. Res.
Hydrol. 178:257–276. doi:10.1016/0022-1694(95)02801-3 82:76–86. doi:10.1016/j.marenvres.2012.09.007
Rose, C., and W.G. Crumpton. 1996. Effects of emergent macrophytes on Wan, Y., J.W. Labadie, K.D. Konyha, and T. Conboy. 2006. Optimization
dissolved oxygen dynamics in a prairie pothole wetland. Wetlands 16:495– of frequency distribution of freshwater inflows for coastal ecosystem
502. doi:10.1007/BF03161339 restoration. J. Water Resour. Plan. Manage. 132:320–329. doi:10.1061/
Samsudin, M.S., H. Juahir, S.M. Zain, and N.H. Adnan. 2011. Surface river water (ASCE)0733-9496(2006)132:5(320)
quality interpretation using environmetric techniques: Case study at Perlis Wunderlin, D.A., M.P. Díaz, M.V. Amé, S.F. Pesce, A.C. Hued, and M.Á.
River Basin, Malaysia. Int. J. Environ. Protect. 1(5):1–8. Bistoni. 2001. Pattern recognition techniques for the evaluation of spatial
Santos-Roman, D.M., G.S. Warner, and F. Scatena. 2003. Multivariate and temporal variations in water quality. A case study: Suquia River
analysis of water quality and physical characteristics of the selected Basin (Cordoba-Argentina). Water Res. 35:2881–2894. doi:10.1016/
watersheds in Puerto Rico. J. Am. Water Resour. Assoc. 39:829–839. S0043-1354(00)00592-3
doi:10.1111/j.1752-1688.2003.tb04408.x Yin, Z.-Y., S. Walcott, B. Kaplan, J. Cao, W. Lin, M. Chen, et al. 2005. An analysis
Shrestha, S., and F. Kazama. 2007. Assessment of surface water quality of the relationship between spatial patterns of water quality and urban
using multivariate statistical techniques: A case study of the Fuji River development in Shanghai, China. Comput. Environ. Urban Syst. 29:197–
Basin, Japan. Environ. Modell. Softw. 22:464–475. doi:10.1016/j. 221. doi:10.1016/j.compenvurbsys.2003.10.001
envsoft.2006.02.001 Yang, Y.-H., F. Zhou, H.-C. Guo, H. Sheng, H. Liu, X. Dao, and C.-J. He. 2010.
Sigua, G.C., J.S. Steward, and W.A. Tweedale. 2000. Water-quality monitoring Analysis of spatial and temporal water pollution patterns in Lake Dianchi
and biological integrity assessment in the Indian River Lagoon, Florida: using multivariate statistical methods. Environ. Monit. Assess. 170:407–
Status, trends, and loadings. Environ. Manage. 25:199–209. doi:10.1007/ 416. doi:10.1007/s10661-009-1242-9
s002679910016 Zampella, R.A., N.A. Procopio, R.G. Lathrop, and C.L. Dow. 2007.
Simeonov, V., J.A. Stratis, C. Samara, G. Zachariadis, D. Voutsa, A. Anthemidis, Relationship of land-use/land-cover patterns and surface-water quality
et al. 2003. Assessment of the surface water quality in northern Greece. in the Mullica River Basin. J. Am. Water Resour. Assoc. 43:594–604.
Water Res. 37:4119–4124. doi:10.1016/S0043-1354(03)00398-1 doi:10.1111/j.1752-1688.2007.00045.x
Singh, K.P., A. Malik, D. Mohan, and S. Sinha. 2004. Multivariate statistical Zeng, X., and T.C. Rasmussen. 2005. Multivariate statistical characterization of
techniques for the evaluation of spatial and temporal variations in water water quality in Lake Lanier, Georgia, USA. J. Environ. Qual. 34:1980–
quality of Gomti River (India): A case study. Water Res. 38:3980–3992. 1991. doi:10.2134/jeq2004.0337
doi:10.1016/j.watres.2004.06.011 Zhang, Q., Z. Li, G. Zeng, J. Li, Y. Fang, Q. Yuan, et al. 2009. Assessment of
Singh, K.P., A. Malik, and S. Sinha. 2005. Water quality assessment and surface water quality using multivariate statistical techniques in red soil
apportionment of pollution sources of Gomti River (India) using hilly region: A case study of Xiangjiang watershed, China. Environ. Monit.
multivariate statistical techniques: A case study. Anal. Chim. Acta Assess. 152:123–131. doi:10.1007/s10661-008-0301-y
538:355–374. doi:10.1016/j.aca.2005.02.006 Zhou, F., Y. Liu, and H. Gou. 2007. Application of multivariate statistical
South Florida Water Management District. 2004. Upper East Coast water supply methods to water quality assessment of the water courses in northwestern
plan. SFWMD, West Palm Beach, FL. New Territories, Hong Kong. Environ. Monit. Assess. 132:1–13.
South Florida Water Management District. 2009. St. Lucie River watershed doi:10.1007/s10661-006-9497-x
protection plan. SFWMD, West Palm Beach, FL.

610 Journal of Environmental Quality