Water quality criteria for implementation of antidegradation water quality standards


Material Information

Water quality criteria for implementation of antidegradation water quality standards
Physical Description:
xiv, 237 leaves : ill. ; 29 cm.
Kenner, Scott James
Publication Date:


bibliography   ( marcgt )
non-fiction   ( marcgt )


Thesis (Ph. D.)--University of Florida, 1992.
Includes bibliographical references (leaves 230-236).
Statement of Responsibility:
by Scott James Kenner.
General Note:
General Note:

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
oclc - 27483449
System ID:

Full Text







I dedicate this work to the memory of my brother, Mark

James Kenner, whose passion for excitement and perseverance

for perfection provides me never ending inspiration.


I would like to express my sincerest gratitude to Dr.

Wayne C. Huber and Dr. James P. Heaney for the many learning

opportunities provided throughout my study program and the

guidance for completing my research. I would also like to

thank Dr. Warren Viessman and Dr. Joseph J. Delfino for

looking out for my best interest and providing me with several

opportunities to improve my teaching skills. I would also

like to thank Dr. R. Best and Dr. R. Ballerini for serving on

my graduate committee.

I want to thank the State of Florida and The Suwannee

River Water Management District through whom the funding for

this research was provided. Special thanks go to Janice

Fellers for her enthusiastic support and guidance throughout

the research project.

I want to thank Doris Smithson for her help with word

processing of this dissertation and her moral support over the

past four years.

Finally, I would like to thank my wife and children for

their patience and love; my parents for providing continuous

support; and my grandfather who instilled in me a love for our

environment. My ultimate thanks go to my creator and

sustainer, Jesus Christ.






. . iii

. . vii

. . x

- . xiii



Background.. .
Antidegradation Policy .
Numerical Criteria .
Water Quality Data
Data Analysis .

. . 1

Numerical Representation .
Application . .
Research Objectives and Organization


Study Area . . .
Rationale . .. .
General Description . .
Regionalization ..... .
Antidegradation Water Quality Standards
Data . .. ... .
Surface Water Quality Data .......
Flow Data . .




. 45
. 45

Literature Review . .
General . .
Two-Sample Tests .
Concentration-Flow Relationships
Seasonal Analysis .
Trend Analysis .
Serial Correlation .
Formalized Data Analysis ..



. .

. .
. .
. .

Data Analysis Protocol .
Analysis Objectives .
Data Aggregation .
Concentration flow relationships
Trends .
Seasonal Analysis. .. ...
Ambient Water Quality .
Application to Suwannee River: Results
Data Aggregation .
Flow-Concentration Relationships
Seasonality . .
Trend Analysis . .


Beneficial Use Criteria .
Water Quality Based Criteria .
Percentile Estimation .
Nonparametric Methods .
Confidence Intervals .
Criterion Event Definition .
Event Frequency .
Event Duration .

Application to the Suwannee River: Results
Deseasonalization vs. FAC .
Individual vs. Combined Data .
Ambient Water Quality Percentiles .
Deseasonalized Ambient Water Quality
Percentiles .
Percentile Confidence Intervals .
Seasonal and Annual Series .
Summary . . .


Evaluating Changes In Water Quality .
Individual Data Measurements .
Multiple data measurements .
Long-term data . .
Wasteload Allocation and Permitting .
Summary . . .


Summary . .
Conclusions . .
Future Research . .



* 197

* 197
* 200
. 202
. 208
. 209
. 212

. 214





. 230


. 140









. .
. .
. .




Table Pafe
2-1. Suwannee River Water Quality Monitoring Stations 27

2-2. Reach Definitions for the Suwannee River .. 31

2-3. Summary of Flow Gaging Stations on the Suwannee
River . . .. 41

2-4. Comparison of Period of Record for Flow and Water
Quality . . ... .. 43

3-1. Concentration-Flow Functional Relationships 52

3-2. Quarterly Seasonal Cycles . ... 75

3-3. Summary Statistics for Specific Conductance 86

3-4. Summary Statistics for Total Phosphorus ... 87

3-5. Summary Statistics for Nitrogen ... 88

3-6. Mann-Whitney for Specific Conductance .. 90

3-7. Mann-Whitney for Total Phosphorus . 91

3-8. Mann-Whitney for Total Nitrogen ... 92

3-9. Regression Analysis Summary for Specific
Conductance . . ... 96

3-10. Regression Analysis Summary for Total Phosphorus 96

3-11. Regression Analysis Summary for Total Nitrogen 97

3-12. Kruskal-Wallis Seasonality Test Results for
Specific Conductance . ... 112

3-13. Kruskal-Wallis Seasonality Test Results for
Total Phosphorus . ... 113

3-14. Kruskal-Wallis Seasonality Test Results for
Total Nitrogen . ... .114


3-15. Seasonal Characteristics For Specific
Conductance . ... 121

3-16. Seasonal Characteristics For Total Phosphorus 121

3-17. Seasonal Characteristics For Total Nitrogen 121

3-18. Summary of Median Analyses on Specific
Conductance . ... 124

3-19. Summary of Monotonic Trend Analyses on Specific
Conductance . .... 125

3-20. Summary of Median Analyses on Total Phosphorus 127

3-21. Summary of Monotonic Trend Analyses on Total
Phosphorus . ... 128

3-22. Summary of Median Analyses on Total Nitrogen 135

3-23. Summary of Monotonic Trend Analyses on Total
Nitrogen . . ... 136

4-1. Comparison of Percentiles for Deseasonalized RAW
and FAC Specific Conductance at Branford .. 163

4-2. Comparison of Percentiles for 112WRD and Combined
Total Phosphorus at Branford ... 166

4-3. Proposed Ambient Specific Conductance Percentile
Points At Selected Locations Along the Suwannee
River . . ... .. 168

4-4. Percentile Table for Ambient Specific Conductance
Water Quality . .. 169

4-5. Proposed Ambinet Total Phosphorus Percentile
Points At Selected Locations Along the Suwannee
River . . .. .. 172

4-6. Percentile Table for Ambient Total Phosphorus
Water Quality . ... .174

4-7. Proposed Ambient Total Nitrogen Percentile Points
At Selected Locations Along the Suwannee River 176

4-8. Percentile Table for Ambient Total Nitrogen Water
Quality . .... 177

4-9. Percentile Table for Deseasonalized Ambient
Specific Conductance Water Quality .. 181


4-10. Percentile Table for Deseasonalized Ambient Total
Phosphorus Water Quality ..... .. 182

4-11. Percentile Table for Deseasonalized Ambient Total
Nitrogen Water Quality .. .. 183

4-12. Confidence Intervals About Selected Percentiles
For Specific Conductance . ... 184

4-13. Confidence Intervals About Selected Percentiles
For Total Phosphorus . .. 184

4-14. Confidence Intervals About Selected Percentiles
For Total Nitrogen Data Sets ... 185

4-15. Annual-Seasonal and Annual Series Percentile
Concentrations at State Road 6 ... 188

4-16. Annual-Seasonal and Annual Series Percentile
Concentrations at Suwannee Springs ...... .189

4-17. Annual-Seasonal and Annual Series Percentile
Concentrations at Branford . 190

4-18. Ambient and Seasonal Ambient Concentrations at
State Road 6 . . 191

4-19. Ambient and Seasonal Ambient Concentrations at
Suwannee Springs . .. 192

4-20. Ambient and Seasonal Ambient Concentrations at
Branford . ..... . 193

5-1. Specific Conductance Deseasonalized Concentrations
for Selected Percentiles at Branford .. 204

5-2. Seasonal Statistics for Specific Conductance at
Branford . . 205

5-3. Deseasonalized Concentrations and Exceedences in
the Monitoring Sample for Selected Percentiles .206

5-4. Expected Exceedences in Monitoring Sample for
Selected Percentiles (n = 25) ... 207


Figure Page
2-1. Suwannee River Basin . 19

2-2. Surface Water Quality Monitoring Locations 32

2-3. Specific Conductance Sampling Frequency at
Branford . . .. .. 36

2-4. Suwannee River Flow Gaging Stations 42

3-1. Data Analysis Protocol Flow Diagram .. 65

3-2a. Specific Conductance Time Series Plot At
Branford . . 77

3-2b. Specific Conductance Time Series Plot at Suwannee
Springs . . .. .. 78

3-2c. Specific Conductance Times Series at State Road
6 . . . 79

3-3a. Total Phosphorus Time Series Plot at Branford 80

3-3b. Total Phosphorus Time Series Plot at Suwannee
Springs . . .... .. 81

3-3c. Total Phosphorus Time Series Plot at State Road
6 . . . 82

3-4a. Total Nitrogen Time Series Plot at Branford 83

3-4b. Total Nitrogen Time Series Plot at Suwannee
Springs . . .. .. 84

3-4c. Total Nitrogen Time Series Plot at State Road 6 85

3-5a. Specific Conductance vs Flow at Branford 98

3-5b. Specific Conductance vs Flow at Suwannee Spr. 98

3-5c. Specific Conductance vs Flow at State Road 6 99

3-6a. Total Phosphorus vs Flow at Branford .. 99


3-6b. Total Phosphorus vs Flow at Suwannee Springs .

3-6c. Total Phosphorus vs Flow at State Road 6 100

3-7a. Total Nitrogen vs Flow at Branford .. 101

3-7b. Total Nitrogen vs Flow at Suwannee Springs .. 101

3-7c. Total Nitrogen vs Flow at State Road 6 102

3-8a. Quarterly Box Whiskers Plot for Specific
Conductance (umhos/cm) at Branford showing the
median, upper and lower quartiles, minimum and
maximum concentrations . ... 107

3-8b. Quarterly Box Whiskers Plot for Specific
Conductance (umhos/cm) at Suwannee Springs
showing the median, upper and lower quartiles,
minimum and maximum concentrations .. 107

3-8c. Quarterly Box Whiskers Plot for Specific
Conductance (umhos/cm) at State Road 6 showing
the median, upper and lower quartiles, minimum
and maximum concentrations . 108

3-9a. Quarterly Box Whiskers Plot for Total Phosphorus
(mg/l) at Branford showing the median, upper and
lower quartiles, minimum and maximum
concentrations . 108

3-9b. Quarterly Box Whiskers Plot for Total Phosphorus
(mg/l) at Suwannee Springs showing the median,
upper and lower quartiles, minimum and maximum
concentrations . .... 109

3-9c. Quarterly Box Whiskers Plot for Total Phosphorus
(mg/l) at Suwannee Springs showing the median,
upper and lower quartiles, minimum and maximum
concentrations . ... 109

3-10a. Quarterly Box Whiskers Plot for Total Nitrogen
(mg/l) at Branford showing the median, upper and
lower quartiles, minimum and maximum
concentrations . .. 110

3-10b. Quarterly Box Whiskers Plot for Total Nitrogen
(mg/l) at Suwannee Springs showing the median,
upper and lower quartiles, minimum and maximum
concentrations . .. 110


3-10c. Quarterly Box Whiskers Plot for Total Nitrogen
(mg/1) at State Road 6 showing the median, upper
and lower quartiles, minimum and maximum
concentrations . ... 111

4-1. Deaseasonalized FAC versus Deseasonalized RAW
Specific Conductance at Branford . 163

4-2. Cumulative Percentiles for 112WRD and Combined
Total Phosphorus at Branford ... 166

4-3. Ambient Percentile Profiles For Specific
Conductance representing sequentially from top to
bottom the 99th, 95th, 80th, 65th, 50th, 35th,
20th and 5th percentiles . .... 168

4-4. Ambient Percentile Profiles For Total Phosphorus
representing sequentially from top to bottom the
99th, 95th, 80th, 65th, 50th, 35th, 20th and 5th
percentiles .. . 173

4-5. Ambient Percentile Profiles For Total Nitrogen
representing sequentially from top to bottom the
99th, 95th, 80th, 65th, 50th, 35th, 20th and 5th
percentiles . ... 176


Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy



Scott James Kenner

August 1992

Chairperson: Dr. Wayne C. Huber
Major Department: Environmental Engineering Sciences

Current water quality criteria have been established for

protection of designated beneficial uses. These criteria do

not represent ambient conditions nor incorporate natural water

quality variability. Many water bodies have ambient water

quality better than that established by the beneficial use

criteria. For these water bodies the difference between the

water quality necessary to maintain designated beneficial uses

and ambient water quality is considered assimilative capacity.

Historically, wasteload allocations of point source discharges

have used assimilative capacity of receiving water bodies by

allowing water quality to be lowered. The lower water quality

is acceptable as long as designated beneficial use criteria

are maintained. The focus of antidegradation policy is

maintaining and protecting ambient water quality of high

quality waters.


Implementation of antidegradation water quality standards

requires numerical definition of ambient water quality.

Inherently, representation and analysis of water quality data

require statistical techniques. Standard statistical methods

are not readily applicable to water quality data due to

several characteristics: nonnormal data, missing values,

seasonality, detection limits, and serial dependence. A

formalized approach for statistical analysis of historical

water quality data is established for representing ambient

water quality.

Ambient water quality data are then used to establish

water quality criteria based on criterion event definition.

The criterion event definition consists of magnitude, duration

and frequency of the water quality event of interest.

Statistical representation incorporates natural variability of

ambient water quality and can be effectively implemented for

water quality management.

A formalized approach to data analysis is applied to the

Suwannee River for development of antidegradation water

quality criteria. The analysis does not focus on any specific

percentile concentration and any percentile of interest can be

identified and applied as a numerical criterion.




Current water quality standards consist of three

principal components: 1) designation of beneficial uses, 2)

definition of water quality criteria necessary to maintain the

designated beneficial uses, and 3) an antidegradation policy.

The first two components, beneficial uses and water quality

criteria, have played a major role in water quality management

through control of point-source discharges. Most water

quality criteria established for a specific designated

beneficial use are constant numerical values that are not to

be exceeded. Many water bodies have ambient water quality

better than that established by the beneficial use criteria

and are often referred to as high quality waters. For these

water bodies the difference between the water quality

necessary to maintain the designated beneficial use and the

ambient water quality is considered assimilative capacity.

Historically, wasteload allocations of point source discharges

have used the assimilative capacity of receiving water bodies

by allowing water quality to be lowered. The lower water

quality is acceptable as long as the designated beneficial use


criteria are maintained. The focus of antidegradation policy

is maintaining and protecting the ambient water quality of

high quality waters.

Although a national antidegradation policy has existed

since 1968 (USEPA, 1985b), actual implementation has been

limited until recently. Public awareness for environmental

concern has grown significantly in the past two decades. This

concern has heightened the desires to maintain existing high

quality waters. Subsequently, water quality management

agencies are turning to antidegradation policy as the

regulatory mechanism to protect and maintain ambient

conditions of high quality waters.

Antideqradation Policy

Development of specific standards for unique or

outstanding waters was recommended in the 1968 report of the

National Technical Advisory Committee on Water Quality

(Federal Water Pollution Control Asso., 1968). The basis of

antidegradation policy was initially established in 1968 by

the U.S. Department of the Interior and subsequently included

in the U.S. Environmental Protection Agency's (USEPA) first

water quality standards of 1975 (USEPA, 1985). This national

antidegradation policy has remained mostly unchanged.

Apart from an initial objective to protect unique high-

quality waters, antidegradation policy is formulated with

regard to all surface water bodies. The national policy


consists of three components: 1) at a minimum the level of

water quality necessary for existing uses shall be protected

and maintained; 2) where water quality is better than that

necessary to support the existing uses, it shall be protected

and maintained unless allowing lower water quality is

necessary to provide important economic and social

development; and 3) high quality waters of outstanding

national resource shall be maintained and protected (40 CFR

Part 131, 1990). All states are required to adopt an

antidegradation policy reflecting these three components and

identifying how the policy will be implemented. A compilation

of antidegradation criteria for several states (USEPA, 1988c)

shows that antidegradation criteria are mostly narrative and

essentially emulate the national policy.

The fundamental requirement for implementation of

antidegradation policy is definition of numerical criteria

representing ambient water quality with respect to a defined

baseline time period. The numerical antidegradation criteria

define 1) the ambient condition that must be maintained for

Outstanding National Resource Waters (ONRW), 2) the ambient

water quality baseline from which the amount of "allowable"

degradation economically and socially acceptable can be

defined, and 3) water quality level to be restored for water

quality limited waters. Antidegradation water quality

criteria are based on the ambient water quality of the


specific water body and can be referred to as water quality

based criteria.

Terminology used for representing water quality can vary

depending on regulatory definitions and interpretation.

Within this document natural water quality refers to the water

quality prior to any changes or impacts due to human

influence. Ambient water quality refers to water quality

conditions which incorporate changes in water quality due to

previously permitted discharges with reference to the defined

baseline time period. The baseline time period for

establishing ambient conditions will vary from state to state

depending on specific regulations.

Numerical Criteria

Development of numerical water quality criteria to

represent ambient conditions can be presented in terms of four

primary components: 1) quantity and characteristics of

available water quality data, 2) preparation and analysis of

the data to represent ambient conditions, 3) the specific

numerical representation as criteria, and 4) application of

the numerical criteria for water quality management.

Water Quality Data

Ideally water quality data should reflect the desired

objectives of the analyses to be conducted. Objectives of the

analyses include both long-term and short-term


characteristics. The sampling frequency controls the level at

which short-term characteristics can be defined, while the

sampling duration or period of record controls how well long-

term characteristics can be defined.

Available water quality data are basically represented by

either long-term data collected by a single agency or short-

term data sets that have been collected by different agencies

for special studies. The sampling frequency can vary over the

period of record within a single data set and will vary

between different data sets. Much of the available historical

data has been collected without full consideration of the uses

or objectives of the data analysis to be conducted. Short-

term water quality monitoring efforts do not represent the

long-term variability inherent in water quality. Past long-

term monitoring efforts have lacked a clear definition of the

desired information and statistical design necessary to make

management decisions (Bell, 1989). The inadequacies of water

quality information systems have been evaluated by several

researchers (Ward et al., 1989). The two most often neglected

considerations in the establishment of a monitoring program

are a clear definition of the desired information and an

understanding of the statistical nature of the analytical

methods to be used in the analyses of the data. The ideal

situation would be to obtain water quality data that have been

designed with the objectives of the analysis taken into

consideration. Unfortunately, this is not the case and much


of the available data have deficiencies with respect to the

various analyses that must be conducted.

The lack of "quality" data often results in limited use

of existing historical data or selection of study sites with

adequate data for analysis. Water quality studies for

evaluation of trends require consistent long-term records.

Ten years of continuous monthly samples are recommended for

evaluation of trends in water quality(Hirsch et al., 1984;

Lettenmaier et al., 1991). Other common types of analyses

performed on water quality data are development of a

representative probability distribution and development of

percentiles. With short data records it is often assumed that

the data fit a lognormal distribution. The resulting

distribution parameters do not incorporate long-term trends or

cycles. Conversely, earlier data records are excluded from

the analysis due to water quality trends. Crabtree et al.

(1987) limited the period of record for development of water

quality percentiles to three years to avoid long-term trends

in the data. Representing receiving water quality as a

lognormal distribution, Male and Soucie (1989) split their

data visually to compare differences between earlier and

current data records. When the current data were

significantly different the early record was excluded from the


Many studies are conducted with fairly strict data

requirements that result in elimination of much of the


available data. The available data are not included due to

various reasons (not the same agency, variable sampling

frequency, type of analytical test, consistency etc.). The

data that do not meet the analysis requirements are excluded

before the desired results are determined. The effect of

including the poorer quality data in the actual analysis has

not been reported.

The historical nature of the needed information precludes

the development of a formalized monitoring network which takes

into consideration the above concerns. The situation that

confronts us is to gather available historical water quality

data and, through statistical analysis, obtain as much

information as possible to quantify the existing ambient water

quality. Antidegradation inherently specifies a baseline time

period around which the analysis must be conducted. This does

not afford one the flexibility to select time periods which

have the better data for analysis. Additionally, long-term

data are necessary to identify trends and natural cycles in

water quality. Representation of these characteristics of

water quality requires use of as much data as are available.

Subsequently, a primary objective of this research is to use

as much of the available data as possible.

All available data at a specific location are considered

by combining different agency data sets into one

representative data set. It is evident that the data

collected by different agencies can vary with regard to

sampling frequency, period of record, analytical techniques,

sampling location, and minimum level of detection. These

differences can result in different representations of the

true underlying population. Therefore, combining of data sets

collected by different agencies during different time periods

must be done judiciously.

Data Analysis

Analysis of water quality data is formulated based on the

desired information objectives and the characteristics of the

available data. Some of the more common objectives include

determining average conditions, long-term trends, percentiles,

concentration-flow relationships, wasteload allocation and

violations of water quality standards.

For establishing antidegradation water quality criteria

three primary objectives are defined for the data analysis: 1)

maximize the use of available data by combining multiple data

sets at a single location into one representative data set, 2)

evaluate and identify trends with respect to the baseline time

period and remove trends that do not reflect ambient

conditions of the baseline time period, and 3) numerically

characterize ambient water quality variability due to

seasonality and flow. As stated above the degree to which

this can be done is based on the available data.

These objectives require several types of data analysis

techniques. Statistical characterization of water quality has


become common practice for the management of water quality

within all types of water bodies. Various methods are well

documented in the literature for comparison of different data

sets, evaluation of both monotonic and step trends, regression

analysis of concentration and flow relationships, and defining

seasonality. The need here is to combine these various

methods together into a consistent formalized approach for

characterizing ambient water quality.

Very little information is available regarding analysis

and characterization of the data to be used to represent

ambient water quality conditions for antidegradation. More

emphasis has been placed on numerical representation as a

criterion. Ward et al. (1988) established a data analysis

protocol for evaluating short-term groundwater monitoring

data. Although the objective of their analysis is exploratory

relative to understanding the characteristics of the

groundwater quality variable, the importance of a consistent

data analysis approach is directly applicable for developing

antidegradation water quality criteria. Breidt et al. (1991)

go through a detailed data analysis prior to establishing

antidegradation water quality standards for the Delaware Water

Gap National Recreation Area. In their analysis they exclude

data that indicate any possible trends and separate seasonal

and non-seasonal data sets. This approach limits the analysis

to data sets that meet the requirements of the statistical

methods. When the only data available for a specific water


quality parameter indicate trend, no criteria are developed.

Clearly a defined approach to data analysis is needed to

provide consistent and repeatable analysis, uniformity for

application to other locations within a basin, and yield the

necessary information for water quality management.

Numerical Representation

Numerical definition of ambient water quality in terms of

water quality criteria requires representation of natural

cycles. Variability of water quality data is due to natural

cycles within nature. Natural cycles that are readily

apparent include seasonal variability of temperature and

precipitation, and daily cycles of temperature and light. The

natural variability of ambient water quality inherently leads

to a probabalistic representation. A probabalistic

representation also provides for the assessment of risks

associated with various water quality management decisions.

The current trend for representation of ambient water

quality is the use of percentiles. A percentile represents

the percent of data measurements less than or equal to a

specific concentration. Representation of ambient water

quality using percentiles was established in European

countries about 1980 with the policy of River Quality

Objectives (Bernard, 1988) and is becoming more common in the

United States. Typically a high percentile (95h) is chosen as

the criterion with the objective of protection against extreme


events. Extreme events are typically characterized by high

concentrations for short durations. This type of event is not

the only event than can result in changes in water quality.

Nonpoint source discharges are typically due to landuse

changes that take place over relatively long time periods, and

the corresponding change in water quality is a gradual change

in the average conditions. Establishment of numerical water

quality criteria must recognize both types of events: 1)

extreme events characterized by high concentrations and short

durations, and 2) long-term events characterized by gradual

changes in average conditions.

Recent work with regard to water quality based criteria

for toxics uses three components for defining a specific

criterion: 1) magnitude, 2) duration, and 3) frequency (USEPA,

1991a). The duration is the time period over which the

parameter concentration (magnitude) is averaged and it is

often referred to as the averaging period. The magnitude and

duration are established based on laboratory toxicity tests.

The magnitude is based on the concentration that produces a

specific toxicity effect and the duration relates to the

exposure time of the toxicity test. The frequency is

established somewhat arbitrarily based on a qualitative

allowable exceedence frequency for the defined event (EPA,

1991a). The concept of magnitude, duration and frequency is

readily applicable to conventional water quality parameters

based on ambient water quality.


A critical aspect of this type of representation is

assigning a frequency (return period) to the criterion or

event. The USEPA (1991a) makes reference to use of the

concentration-duration curve of daily values to establish the

frequency of the event. Theoretically, use of the

concentration-duration curve to establish return period is

incorrect (Searcy, 1959). The duration of the event must be

defined and represented within each data point used in a

subsequent frequency analysis. For example, to establish the

return period for the average monthly concentration the

frequency analysis must be conducted on data that represent

average monthly values. This type of analysis parallels the

development of intensity-duration-frequency curves for design

rainfall. Selection of a duration should take into

consideration the cause-effect relationship of the water

quality parameter being evaluated. Toxic parameters typically

have short response times, on the order of hours to days,

whereas the response to increased concentrations of nutrients

is likely to be on the order of months to years, depending on

the level of increase in concentration. The specific duration

that can be defined will also be based on the sampling

frequency of the available data. This research examines the

application of the magnitude-duration-frequency concept for

representation of ambient water quality in combination with



Application of water quality criteria for management of

water quality falls into two general categories: 1) wasteload

allocation for establishing permit discharges and 2)

evaluating and detecting future changes in water quality.

Establishment of the numerical water quality criteria for

antidegradation must take into consideration how the criteria

are actually used if they are to be effective.

Implementation of water quality standards has

historically been carried out through the National Pollution

Discharge Elimination System (NPDES) permit program

established in the 1972 Amendment to the Water Pollution

Control Act (PL 92-500). Discharge permits are established in

one of two ways: 1) treatment technology based effluent limits

or 2) water quality based effluent limits. Treatment

technology effluent limits are based on the "best practicable

technology" (BPT) with reference to current practices. Water

quality based effluent limits are established based on

maintaining a specified level of water quality in the

receiving water. When effluent limits based on BPT do not

meet the desired receiving water quality, additional treatment

measures are required. Implementation of antidegradation

water quality standards clearly falls within the category of

water quality based effluent limits. Current practice focuses

on development of total maximum daily loads (TMDL) for

specific parameters. The TMDL incorporates the waste load


allocation (WLA) for point source discharges and the load

allocation (LA) for nonpoint source discharges. Several

methods representing different levels of technical analysis

are used for development of TMDLs (USEPA, 1991a; USEPA,

1991b). Although the objective of the TMDL process is to

establish allowable parameter loadings, the receiving water

quality is represented by an allowable concentration. An

important consideration is which methods can best represent

the variability of the water quality.

Monitoring future water quality provides information to

detect changes in water quality and evaluate violations of

water quality standards. Monitoring of future water quality

is essential for evaluating the effect of implementing

advanced treatment technology for point sources and best

management practices for nonpoint sources initiated to either

improve or maintain water quality. Effects on water quality

due to point source discharges are readily evaluated by

monitoring the effluent discharge and specific locations

within the receiving water. However, effects of nonpoint

source discharges rely primarily on monitoring the quality of

the receiving water.

The ambient water quality representation must serve as a

basis for changes in future water quality. Depending on the

specific numerical representation used, various methods of

comparison are available for detecting changes in water

quality (trend analysis, changes in the mean, changes in the


standard deviation, and goodness of fit tests). Previous

research has focused on the detection of exceedences of a

single criterion by an extreme event (Hathhorn and Tung, 1988;

Loftis and Ward, 1981; Male and Soucie, 1989; Warn and

Matthews, 1984). Recognizing that the antidegradation

criteria will incorporate both short-term (extreme event) and

long-term (average event) criterion appropriate methods must

be used for assessing each type of change in water quality.

Research Objectives and Organization

Two primary objectives are established for this research:

1) formalize several data analysis techniques into a

consistent "data analysis protocol" that is generally

applicable for establishing a water quality database

representative of ambient conditions and

2) develop a numerical representation of the ambient

water quality conditions that incorporates the natural

variability of ambient water quality, and can be effectively

implemented for water quality management.

The study area used for this research is the Suwannee

River. A description of the study area, site specific

antidegradation water quality standards, and characteristics

of the available water quality data are presented in Chapter

2. Review of the appropriate literature is presented with

each subsequent chapter topic. This is done to provide better

continuity within the dissertation. Data analysis techniques


and a formalized approach to establishing an ambient water

quality database are presented in Chapter 3. Numerical

criteria based on ambient water quality are developed in

Chapter 4 followed by an evaluation of appropriate methods for

application of the established criteria in Chapter 5. Chapter

6 presents a brief summary of the research conducted,

conclusions and directions for future research.


Study Area


Under the Surface Water Improvement Act (SWIM), Sec.

373.451 FS, the Suwannee River Water Management District

(District) is charged with the responsibility to restore and

protect the quality of priority surface waters within the

District boundaries. The significance of the Suwannee River

is expressed by its designation as an Outstanding Florida

Water (OFW) according to water quality standards of the State

of Florida, Department of Environmental Regulation (FDER) (17-

302.700, F.A.C.). This designation is in addition to the

Class III usage classification for the Suwannee River (17-

302.600 F.A.C.). Beyond the Class III water quality criteria,

special protection for OFWs is provided through implementation

of antidegradation permitting requirements (17-4.242 F.A.C.).

The water quality criteria for implementing the

antidegradation standard are referenced to the "existing

ambient water quality" within the OFW. Subsequently, to

implement antidegradation water quality standards for the

Suwannee River ambient water quality must be defined.



The Suwannee River owes its popularity to its outstanding

scenic beauty and undeveloped, unspoiled nature. Compared to

other regions of Florida the Suwannee River Basin is

relatively undeveloped except for one major phosphate mining

operation. Recent pressure on new development in South

Florida has generated increased development within the

Suwannee basin. A heightened awareness of the potential

increase in development within the Suwannee basin adds to the

need for definition of ambient water quality for proper water

quality management.

General Description

The following description is taken primarily from the

Suwannee River Task Force, Report to Governor Bob Martinez


The Suwannee River originates in the Okefenokee Swamp in

Georgia and flows south for 265 miles through Florida into the

Gulf of Mexico (Figure 2-1). The entire drainage area of the

Suwannee River is approximately 9,950 square miles, with about

4,230 square miles in north central Florida.

There are three principal tributaries to the Suwannee

River. The Withlacoochee and Alapaha river basins lie mostly

in Georgia, while the Santa Fe river basin is primarily in

Florida. The river's flow is estimated to be approximately

1,550 cubic feet per second (cfs) at its entrance into Florida

and 17,000 cfs when it empties into the Gulf of Mexico.




Unlike many rivers, the Suwannee's water quality is

generally better downstream than upstream. The headwaters of

the Suwannee River are characterized by dark and muddy flow,

high in tannic and humic acids from decay of vegetation. The

major tributaries and numerous springs in the lower reaches

provide the Suwannee with higher quality water.


The Suwannee River within Florida can be divided into

five regions based on a variety of features including

floodplain morphology, geology, hydrology, water quality and

biotic habitat.

Region I begins in the Okeefenokee Swamp and continues

downstream about forty miles to White Springs. This river

segment is characterized by numerous shoals and swift

currents. Surrounding land uses are typically farmland,

forest, and low-density residential development. The river

flows over unconsolidated sands and the sand/clay/limestone of

the Hawthorne formation. Water quality is dominated by

surface runoff characterized by high color, low mineral

content and slight acidity.

Region II covers the area from White Springs to the

confluence of the Withlacoochee River, approximately 53 miles

in length. This is a transitional area where the river

crosses from the northern highland to the coastal lowland

physiographic regions. Several springs enter the river


causing wide variations in water quality along this reach.

The water quality will range from acidic to neutral pH, low to

high conductivity, and high color to relatively clear. This

reach is also influenced by phosphate mining discharges at

Swift Creek.

Region III is 62 miles in length and includes the area

from the confluence of the Withlacoochee to the Santa Fe River

confluence. This reach also has many spring inflows. The

river bed is mostly Ocala limestone. Water quality is now

dominated by groundwater inflows and is generally clear to

moderately colored, neutral pH, and higher conductivity.

Region IV is very similar to Region III and covers about

42 miles from the confluence of the Santa Fe River to just

below Fowlers Bluff. The river bed lies on Ocala limestone

and water quality continues to exhibit characteristics of

groundwater inflows. Most of the drainage basin area is

contributing to the river flow below the Santa Fe River and

the flow volume is much higher and less variable than the

upper region. The lower portion of this reach is influenced

by moderate tidal activity.

Region V represents the river estuary. The estuary

begins roughly between Fowlers Bluff and Gopher River. This

region is relatively free of springs. The river bed is a mix

of rock and sand, turning to sand and mud at the river mouth.

The river is strongly influenced by tidal action and water


quality is becoming marine influenced with conductivities

ranging from 300 to 26,000 umhos/cm.

Antidegradation Water Quality Standards

The current FDER antidegradation standard is set forth in

Rules 17-302.300 F.A.C., 17-302.700 F.A.C., and 17-4.242

F.A.C. The three primary components of the standard are: 1)

antidegradation policy (17-302.300), 2) special protection and

designation of OFWs and Outstanding National Resource Waters

(ONRW) (17-302.700), and 3) implementation of the policy

through antidegradation permitting requirements (17-4.242).

These three components are discussed in general and specific

sections that apply directly to the objectives of this

research are emphasized.

Antidegradation policy. Rule 17-302.300 (1)-(8) F.A.C.

contains general antidegradation policy statements for all

surface waters. Three key elements of the policy are

summarized below.

1) The water quality necessary to protect existing

beneficial uses (Class III) shall be maintained and discharge

of wastes into Florida waters without treatment necessary to

protect those uses is prohibited.

2) One of the most severe water quality problems is

excessive nutrients (total nitrogen and total phosphorus) and

the introduction of man-induced nutrients shall be limited.


3) If the proposed new or expanded discharge does not

lower the quality below that necessary to maintain the

beneficial use, a permit may be issued if the degradation that

does occur is necessary or desirable and is clearly in the

public interest.

This policy assures that the minimum water quality

protection for all Florida waters is the water quality

necessary to maintain the existing beneficial use

classification. The Class III narrative criteria for

nutrients (nitrogen and phosphorus), together with the

emphasis on limiting introduction of excessive nutrients, is

the basis for selection of total phosphorus and total nitrogen

as parameters for the analysis.

Special protection OFW. Rule 17-302.700 (1)-(10) F.A.C.

presents specific antidegradation policy for special

protection of OFWs and ONRWs. Key elements of this policy are

summarized below.

1) No degradation of water quality other than that

allowed under the permitting process which is implemented

pursuant to Section 17-4.242 F.A.C.

2) Designation of a Special Water (i.e. OFW) is based on

finding that environmental, social and economic benefits

provided by the ambient water quality conditions far outweigh

costs to protect and maintain them.


3) For each OFW listed under Section 17-302.700(9) the

baseline year for defining the "existing ambient" water

quality is the year ending March 1, 1979.

This establishes the permitting process as the mechanism

to implement the policy for protection of "Special Waters."

More importantly it establishes February 1978 through February

1979 as the baseline time period for defining the existing

ambient water quality.

Antideqradation permitting requirements. With regard to

OFWs, the two major components of the permitting process are

established in 17-4.242 (1) and (2) F.A.C. Section (1)

prescribes the process required to determine whether a

discharge is "necessary or desirable" and "clearly in the

public interest." Section (2) establishes specific standards

for OFWs. The initial foundation of the standard is provided

in 1714.242(2)(a) which states:

No Department permit or water quality certification shall
be issued for any proposed activity or discharge within
an Outstanding Florida Water, or which significantly
degrades, either alone or in combination with other
stationary installations, any Outstanding Florida Waters,
unless the applicant affirmatively demonstrates that:
the proposed activity or discharge is clearly in
the public interest.

The current FDER interpretation of this statement is that no

direct discharges to any OFWs can degrade the existing ambient

water quality (as specified in 17-4.242(2)(c)), and that

indirect discharges cannot "significantly degrade" the

existing ambient water quality. This implies that any direct

discharges must be of the same quality as the OFWs. However,


for indirect discharges "significant" degradation is left

undefined. Defining ambient water quality provides a

definitive tool for use in assessing the significance of the

degradation along with the environmental, social, and economic


Water quality degradation is measured as a change in

water quality from the "existing ambient" water quality. The

meaning of "existing ambient" water quality is defined in 17-

4.242(2)(c) and states:

The term existing ambient water quality shall mean (based
on the best scientific information available) the better
water quality of either (1) that which could reasonably
be expected to have existed for the baseline year of an
Outstanding Florida Water designation, or (2) that which
existed during the year prior to the date of a permit
application. It shall include daily, seasonal, and other
cyclic fluctuations, taking into consideration the
effects of allowable discharges for which Department
permits were issued or applications for such permits were
filed and complete on the effective date of designation.

The above definition for existing ambient water quality

provides the impetus for evaluating the available surface

water quality and quantity data in a probabilistic framework.


Surface Water Quality Data

Water quality data are essential to development of

ambient water quality criteria. A significant effort was made

to identify and obtain all available surface water quality

data for the Suwannee River. A primary source of water

quality information is the USEPA water quality database


STORET. In addition to obtaining water quality data from the

STORET data base, the following federal and state agencies

were contacted:

United States Geological Survey (USGS),

United States Environmental Protection Agency (EPA)

United States Army Corps of Engineers (COE),

Suwannee River Water Management District (SRWMD),

Florida Department of Environmental Regulation


Florida Department of Natural Resources (FDNR), and

Florida Game and Fresh Water Fish Commission


The established surface water quality data base for the

Suwannee River represents data gathered by seven different

agencies at twenty-four locations along the river. These

locations consist of seventy-eight significant monitoring

station data sets with periods of record ranging from two

years up to thirty-six years and from twenty-three to one

hundred fifty different parameters being measured. Eight of

these monitoring locations were selected for inclusion in the

analysis. These locations represent the temporal and spatial

variability of water quality in the Suwannee River.

Characteristics of the available water quality data at each

location are presented in Table 2-1. This table presents

information on the water quality data gathered by each agency

at each location along the river. The location of each

CD 0 0 CD C 0D NI 0 CD CD 0 0 CD CD CD CD -
CI D O C D CD iD C O I D CI r- CID 40 to Co O C D CD ID

CD NB CD CD .- -C o C D CD 0 ) .
ilf J 5 S q I Ui ^ I 5 I

*c -^ -; 0 0 0 0> 0 0 r. C' C' C' C' C' CD CD CD CD CD CD
ol to in c o R S ao il r R i 1 0 1 9t S K ^ O in r

E E. co N 4

SC a C C CD N X 6 4 4 N w CL

0L0IDD 0 cr R 0R -I 0

4J U. aL D~C
4CLU N cc
) U) ) U> 0 (>D (O> 4> > 5 5( o. Ino

ID ~00

z z z z

0 ~ ~ ~ ~ ~ CID m ju UL
41 r 0 0 0It a M 0

.2 ut fe tiL 5 LL.

(U I
C N C 5 5
1 L U. 0 w -i j l I I G I

4' wOO C>U .
Ir* C D N N C D C D N "

*'0 z Zi ZO 0 0 F Z 3 << < < <
SU- C C Z) w| m c c

D 4 Z4
00 C 0 o < CD I o C I
Sa cUc x cc c c li X g C 0

LU UJ UJ LU w .- cr LU -J
i z z z z z z z z z z z z z z Z i z z z
z zzzz z z Z z Z ZZ

'~ o-
3 : 3 3: 3: 3: 3: ?: : t0 3: 3 3 3
^1 ui~ ~ ~ m =t :) D> ni nn :) Mi ui :) ni n> Di u > m < u > v

0 d 8C C! CD
N C 4 m 0 -
u w o C4

IL 0

CIO -0i
0 LU4>
s I I I1 1 I1 I PC P: PI CI g I I (L I i I i I L

cr 1 g Sr S w 8
*d LL 0 >

N -j LL z^
3: F! z

0 Sx g 4 x
1 L s i l
U> >5EN
10 CD 3 3 U* *
I Q, V, 5,
d I Ii

c 10 2o
(08 s~
^~~~ ~ ~ ^^ ==== ^ ^ =====^ ^ ^==== ^ ^^=== = ^ ^ ^=

S0 M- (o m < o) o8 m
'S0~ *P S S5 g c -i-S >< a $8 tS< .g 2S
iico ow ^ ^ co 00 00 0 Go
0 >0
V 0 5 r r
|t 0 N 0 5 X > > z
P,. (0, 0 00 (0 C, r- D C iior
OD z D r, Co

oi z 0 4 0 U 0 )


0 w m h0 0 0m
N N C4 0 0 ( 0 C 0 0a m 0) !2 N I
E N N N tN 0 D N N N N 0 C C N o

0 C D W0 0 o o D 0 0 m m 0 0 woI -
S c (I iO ( 0l d (0 t0 oil cB o oi oi oia ai c o .-. .- T-

'5 ( N N M CM t N tM d <* l t ^ l l-

U) 0)

0 0 -z a S o
IE C 1 0 CD ( (

(a L 0 a L w (0 4 z I
0 j LU 0 N ( ( 4 C I
*I J I | J 5 ~ Z S5 5 ^^fi|J
C 0 a > 0 U4
U uj LL co LU LU aU
z 0 z -J Z ^ z 4 X a -
a Z wJ LU LU LU L 1

0 4 Z Z 2 Z Z Z Z J Z 4 Z Z 4 CL 4 Z 4 Z Z
u ( a i (i w 0i
Q) < <

g O o)Iu, 5

J |D 1 I 0 | S z
s' L s ^i -i j _j U) l l a ti i
CC lu w w LU :1 L w

0U S 3
z = w z z z I w z I I

wt K > > > ~ > > > > > ^ > ._ >

(0 (0 (0 (0 (0 (0 (0 (0 (0 (0 (0 (0 (0 (0 ( 0. C (0 (0 (0 (0
IrI m ItC "C r C r r C L i rx x C
L UU N CD w -2 u A
z z z z Z Z Z Z ZZ Z. Z Z z z z z

z Q =z z z/ Iz S
N LU 0 0 N 4 0 N N

a) I 3 |? ?? ~4 w ,Z
8 0 0
000 z (o
o z=
LU to a U) 00 (n U) 0 U

0 0
a)~~~0 ~ U~~ 0 YIw Z mo o 3I u
41 LO UO~ ~ ) LO N

x 00 C t

0 UL (0 N C 0 N (0 C- (0 U 0 L (0 N C

LU LU w o(W r O M) LL

o P g e Is n w
> Z o

02 z

- c o ic

5S 0

UL I L .


d ( 01

5 -I I I IS
s 2 N N N CD

0 0 0

u. c 0 Z Z
SZ Z X r
gz z zzz

S0 o 0

i i l i w c w w
>- > > I> >

Z Z Z Z Z z z

0 0 0 0 0

C 4


H ______


monitoring station is identified by the Group-ID. The Group-

ID refers to a general location point along the river where

several different agencies have gathered water quality data.

The original source of the data is referenced by the agency

code and the station is the specific name used by the agency

for that monitoring location. The agency codes are defined


SRWMDW Suwannee River Water Management District data

collected prior to SWIM program and obtained from the District

on floppy disk.

SRWMDT Suwannee River Water Management District data

collected under the SWIM program and obtained from the

District on floppy disk.

21FLA Florida Department of Environmental Regulation,

Jacksonville data retrieved from STORET.

DERTLH Florida Department of Environmental Regulation,

Tallahassee; Suwannee River Limnology Study 1982-1983, data

obtained from the Department on floppy disk.

112WRD US Geological Survey data retrieved from STORET.

Additional information presented in Table 2-1 for each

monitoring location includes approximate river miles measured

in an upstream direction from the mouth of the Suwannee River

at the Gulf of Mexico, number of parameters measured, number

of samples collected over the period of record, and the actual

period of record. Further discussion of this information



Spatial distribution. Each location represents several

agency monitoring stations that are located near a common

reference point. At each location the different agency

sampling points are within one mile of each other on the

average. Figure 2-2 gives a visual representation of the

eight locations selected for analysis. These locations

provide a spatial representation with reference to the five

regions that were described previously. These reaches are

again defined in Table 2-2. The Ellaville location represents

Table 2-2. Reach Definitions for the Suwannee River
Reach Description River Mile
5 Upstream from the mouth 0 to 10
to above Gopher Creek
4 Upstream from above 10 to 65
Gopher Creek to mouth of
Santa Fe River
3 Upstream from mouth of 65 to 125
Santa Fe to mouth of
Withlacoochee River
2 Upstream from mouth of 125 to 168
Withlacoochee to White
1 Upstream from White 168 to 205
SSprings to the State line

monitoring stations below the confluence of the Withlacoochee

in reach 3. It should be noted that no monitoring locations

within the estuary (reach 5) have been selected for analysis.

Period of record. The period of record reflects two

characteristics: 1) the length of the time period over which

the data were gathered, and 2) the specific time period the

0 c 5 0
c o L L 3 3 cI 1 (I
i) L ULL u ) J 4 vM

J 0 D 00 0) (
U1 ) r^ cn N


data were collected. The available period of record will

affect how well cycles and trends in water quality can be

defined. The estimated long-term averages of cycles in water

quality (i.e., seasonality) will improve as the period of

record increases. Evaluation of cycles and trends is also

affected by sampling frequency.

The evaluation of trends is necessary to determine how

water quality has changed with respect to the baseline year

ending March 1, 1979. Lettenmaier et al. (1991) recommend a

10 year period of record for analysis of trends. However,

studies have been conducted on periods of record as short as

two years (Hirsch et al. 1982). Ideally, the time frame from

which the data were gathered would include a significant time

period before and after the baseline year. The period of

record for the various data sets range from two to thirty-six


The five USGS (112WRD) stations: Wilcox (Fanning

Springs), Branford, Suwannee Springs, White Springs, and State

Road 6 (SR6) near Benton, have long-term records ranging from

twenty-four to thirty-six years covering the period from 1954

to 1990. These stations all incorporate the baseline year and

provide the longest period of record of the available data.

The Jacksonville, FDER (21FLA) stations have varying

periods of record. At Fanning Springs the period of record is

about twenty years covering 1970-1990. At Branford,

Luraville, Ellaville, and White Springs the starting date


varies from 1972 to 1979 and all end in 1986, with record

lengths ranging from seven to fourteen years. These stations

incorporate the baseline year. The Fowlers Bluff and Suwannee

Springs stations all have five years of record covering the

period 1981-1986.

The SRWMD has two data sets at each location representing

data collected prior to SWIM legislation (SRWMDW) and data

collected as part of the current SWIM program (SRWMDT).

SRWMDW stations Fanning Springs, Branford, Suwannee Springs,

White Springs and State Road 6 have six years of record

covering the period 1977-1983; Fowlers Bluff has two years of

record covering the period 1982-1983; and Luraville and

Ellaville have ten years of record covering the period 1977-

1988. The data obtained from the on going SWIM program

(SRWMDT) cover the period 1989-1991 representing about two

years of record.

The Tallahassee, FDER (DERTLH) stations were all

established for a study of the limnology of the Suwannee River

(FDER, 1985). These stations have two years of record

covering the period 1982-1983.

The data gathered by the five agencies discussed above

(112WRD, 21FLA, SRWMDW, SRWMDT, and DERTLH) represent the most

significant data available for the project.

Sampling frequency. The sampling frequency represents

the number of samples and time interval between the samples

collected over the period of record. The time sampling


interval will dictate the time frame at which cycles can be

determined. Typical cycles in water quality include short-

term cycles reflecting daily variability, seasonal cycles

(monthly, bi-monthly, quarterly) and long-term annual cycles.

To characterize these cycles requires data collected within

the time frame of the particular cycle being evaluated. For

example daily cycles require sampling on an hourly basis and

monthly cycles require sampling on a daily basis, whereas

long-term seasonal cycles are based upon long-term averages of

data collected within each season to be considered (monthly,

bi-monthly or quarterly).

The average sampling interval for any specific agency and

sampling location can be estimated by dividing the number of

samples collected (Table 2-1) by the years of record. This

average sampling frequency for each of the agencies can be

misleading and should only be used as a rough estimate. For

example, the average sampling frequency for 112WRD 02320500

(Branford) is approximately eleven samples per year indicating

a monthly sampling frequency. However, the true sampling

frequency is determined from an annual sampling frequency

histogram. The sampling frequency histogram of specific

conductance measurements at 112WRD 02320500 is shown in Figure

2-3 and indicates significant variability in sampling

frequency over the period of record. From 1954 through 1957

two to three samples per month were collected, whereas over

the most recent monitoring period (1984 to 1990) only



--- r-- --o

SD -

I z

r- I .. 1 .m .: .. .; ; ; 1


I_ I ,,, I-

0 0 0 0 0 0
H 1)


-___ ____ ___ ____ ____ ___ ____ ___-f 1


quarterly samples have been collected. Estimated long term

averages or trends will tend to be biased toward the time

period with a higher sampling frequency. Thus, the analyses

must consider closely the variability of the sampling

frequency over the period of record for each location and


Analytical methods. Specific analytical methods used to

determine concentration of various water quality parameters

can vary over the period of record and among different


Changes in analytical methods often lead to differences

in accuracy and levels of detection between different time

periods of the water quality data. The desired level of

detection for a specific method can also vary based on the

dilution or concentration used during preparation of the

sample. Thus, methods used by different laboratories or at

different times might be the same, but the level of detection

can vary depending on the sample preparation. The level of

detection can also change with improved methods and analytical


The difficulty of the method and experience of laboratory

personnel can be a potential source of differences between

data reported by different agencies. For example, specific

conductance is a fairly straight forward analysis that has not

varied significantly over time, whereas total phosphorus

analysis can be done using various solvents for extraction of


the SRP fraction of total phosphorus. Extraction requires

good laboratory procedure and experience of the lab personnel

can influence the results.

In addition to laboratory procedures, collection of

samples in the field can also be a source of variability. The

type of sample collected (grab versus composite) and consis-

tency of the specific location within the water body must be

considered. With changes in personnel, water level, and

accessibility the specific sample location can vary. This can

affect results especially for monitoring locations just below

tributaries or discharges as previously discussed.

From the discussion above potential for variability in

the data due to differences and variability of analytical

methods is quite evident. A detailed evaluation of differ-

ences between data analysis methods used by various agencies

and for each water quality parameter would require a signifi-

cant effort beyond the scope of this project. Even after a

detailed evaluation of available information on the various

analytical methods and laboratory procedures, often many

questions still remain due to a lack of documentation. A

major portion of water quality data were obtained from the

STORET database. Within STORET each parameter code represents

a specific analytical method. Parameter codes for STORET were

used as a reference to verify analytical methods between

agencies. Available information on analytical methods and

sampling protocol of various agencies was obtained to gain an


understanding that will aid in knowledgeable interpretation of

the data. In addition to gathering available information,

personal communication with some of the agencies and

laboratories was made.

Selection of parameters. Three parameters have been

selected for use in developing the methodology to numerically

define ambient water quality. The three parameters selected

are specific conductance, total phosphorus, and total

nitrogen. However, it should be noted that the methodology

is developed for future application to other water quality


Specific conductance (SC) is a conservative parameter and

provides consistent tracking for comparison between locations.

Furthermore, it is monitored at most locations along the

stream and has significant periods of record.

Total phosphorus (TP) and total nitrogen (TN) are used

because problems associated with excessive amounts of these

nutrients are well documented for various water bodies.

Importance and concern for these two nutrients are the basis

of one of the statements in the antidegradation policy (17-

302.300 (3) F.A.C.). Additionally, the current Class III

criteria for nutrients is narrative and numerical classifica-

tion will provide a definitive basis for water quality

management. FDER (1985) identified ortho-phosphorus as a

better tracking parameter than total phosphorus. However,

their study also indicated a significant positive correlation


exists between total phosphorus and ortho-phosphorus for the

Suwannee River data they evaluated. Thus, analysis of TP will

provide a general representation reflecting the character-

istics of the sub-components of TP. TN values are not

measured by all the agencies. However, TN values can be

calculated by summing up total kjeldahl-N (TKN) and total

nitrite-nitrate-N. Where the components of total nitrogen

were available TN was calculated. In general TP and TN

provide an overall assessment of the phosphorus and nitrogen

components in the Suwannee River.

Flow Data

Flow data play an important role in the assessment of

water quality. Flow data are necessary to evaluate possible

relationships with concentration and for comparison of the

long-term flow variability with the flow variability over the

water quality period of record.

Available flow data for the project are taken from the

USGS flow gaging stations within the basin. The main stem

Suwannee River and tributary stations are summarized in Table

2-3. Figure 2-4 shows the location of the flow gaging

stations within the basin.

Table 2-4 compares the period of record for flow data and

water quality data. The water quality periods of record given

in Table 2-4 represent the maximum combined period of record

for all agencies at the referenced location.


-4) -4- 4J -4)
E4 l a c 0 a c >0 0 0a
Idn01 4 cda ) id Id 0 0 0
i i I I I I I I I I
41C +1 4 4JC 4 4J VCrN 4Jr-
o U f U (ol U UU U C U 3
W 0 f 0 0 04-IN 0U

m V (VI nine ( OOn ( V iimm
H I O'00' 01 00 01 I 00 0) 00
, I ,-i- ,- ,- -i- ,- I I -r4U) M -E,-4

% O O 0 %0 0 (4 q %0 o00 O
OU in (N k.o I- '.0 %- 'l W

00 0
o o T T

O l
0 I I (0,
I r-. Mr

M 4 N 0 c 0 0 G O
at 0 II oI I I-
S-4 (4 r4 rn in

0 0M N N

m h a aa

i (0 O N0 U) O0 O0 U0 0
( H H 0 H 0
0 I i LA 0 0 0 H

W N 0 0 0 0 0 0 00 0
H U) 0 01 H N 0
O 00 00 ,oo n 0 O 0o

E4 4 1r4 1f1

Bao C o oo o o a Go 8 I o G o
S-m14 4

0 .) o i o ) >x> > I m >

) x N a) N WIN -4 -0 4, (1 4 0
SI 0 I0 I- IW- 0 0 r ) .n o r. o -

S 0 0 0* r. 0 r- 0 r. *r *r
0 0 00 00 0 0 04 0

H (N 0 N 14N U)Me 4 ( 4- H ).0r a H-
id N (N (N N (N 14 ** (N *E
F0l WO l 0 3 04V0 qV 10 V (40 N) 0)
to Z I UZI aU I cdc aH d a(iUu) HI a_ I QZ _

(0 1^%'^~~~c Oa ^ C0i' e t iIBd r-l rl1flM 11 r-l l ( N 2 -



i m

v to
~r- Dv-r

- ll
110 *
L 00
















J -.

Table 2-4. Comparison of Period of Record for Flow and Water
Location Flow Water Quality
Period of Record Period of Record
Wilcox 1942 1990 1960 1991
(Fanning Springs)
Branford 1931 1990 1954 1991
Luraville 1927 1938 1966 1991
Ellaville 1927 1990 1966 1991
Suwannee Springs 1975 1990 1956 1991
White Springs 1927 1990 1966 1991
Benton 1976 1990 1956 1991
(State Road 6)__

Wilcox, Branford, Ellaville, and White Springs all have

long-term flow records that overlap the quality monitoring

period of record. The available flow record at Bell (Rock

Bluff) does not overlap with the available water quality record.

At Luraville the continuous flow record does not overlap the

available water quality record. The USGS water quality period

of record at Luraville covers the period 1966-1978 and

instantaneous flow measurements are also recorded in the water

quality database for this time period. The available water

quality records for Suwannee Springs and Benton are longer than

the available continuous flow records. For the time periods

without continuous flow records, some instantaneous flow

measurements are available at these locations from the USGS

water quality data and reflect only the flow during the water

quality monitoring period.


Flow data are recorded as daily averages for the continuous

monitoring period of record. Thus, sampling frequency for flow

data is daily and uniform throughout the continuous period of

record except for missing periods. Instantaneous flow data

recorded in the water quality database have a sampling frequency

similar to the water quality data collected by the USGS (112WRD)

for that specific location.


Literature Review


Sound water quality management decisions depend on

competent characterization and interpretation of water quality

data. The recognition of both natural water quality

variability and sampling variability has led to the use of

statistical techniques for characterization of water quality

data. Two general objectives underlie all statistical

analysis: 1) descriptive characterization in terms of specific

parameters, and 2) use of sample parameters to infer or deduct

something about the population (Gibbons, 1985). Recent

research efforts have focused on the appropriate statistical

methods for analysis of water quality data and have led to the

emphasis of nonparametric methods (Hirsch et al., 1982; Hirsch

and Slack, 1984; Lettenmaier, 1976; Montgomery et al., 1986;

van Belle and Hughes, 1984; Ward et al. 1988). Parametric

procedures make inference and tests of hypothesis that are

based on parameter values of the underlying distribution.

Traditional statistical inference is based on the assumption

of normality. The Student's t-test and analysis of variance



(ANOVA) are examples of parametric techniques that assume the

underlying population is normal. Nonparametric statistics

make no assumptions as to the distribution of the sample

population. Nonparametric implies a test for a hypothesis

which is not a statement about the parameters) of the sample

distribution. Nonparametric tests are formulated as a

function of the order statistics or rank-order statistics of

a random sample.

The collective order statistics of a random sample from

a population with a continuous distribution are represented by

the unique ordered arrangement such that X(, is the smallest,

X. is the second smallest, etc.; and X( is the largest of the

random sample X,, X2, ..., X- X., for 1 s r s n is called the

ri order statistic. Rank-order statistics for a random sample

are any set of constants which indicate the order of

observations. Commonly the set of numbers used to establish

the ranks are the first n positive integers. The rank of the

r& order statistic, X., is simply r and r(X,) is called a rank

statistic. The magnitude of any observation is only used to

establish the rank. Thus, any statistical procedure based on

rank-order statistics depends only on the relative magnitudes

of the observations not on parameters of the distribution.

Additionally, the existence of outliers does not affect the

test results.

Another class of methods associated with nonparametrics,

and often mistakenly referred to as nonparametric tests, are


distribution-free tests. Distribution-free inference, whether

for testing or estimation, uses functions of the sample

observations whose corresponding random variable has a

distribution which does not depend on the specific

distribution function of the population from which it was

drawn. Order statistics have the property such that the

transformation U(. = Fx(X()) (where Fx is the cumulative

distribution function) produces a random variable which is the

r order statistic from the uniform population on the interval

(0,1) regardless of the parent distribution function Fx, and

therefore U( is distribution-free (Gibbons, 1985).

Appeal of nonparametric methods is due to their

flexibility to the characteristics and variability of water

quality data. Parametric methods require strict assumptions

that most often include independence, normally distributed

population and equal variance. These assumptions are commonly

violated by water quality data characteristics that include

nonnormal data, missing values, seasonality, values below

detection limits, and serial dependence (Hirsch and Slack,

1984; Lettenmaier, 1976). Only two requirements are essential

to nonparametric methods: 1) the population is continuous, and

2) random sampling is required (Gibbons, 1985; Runyon 1977).

The first assumption is intrinsic to water quality. The

second requirement essentially deals with independence of the

data. Often water quality data are collected at a relatively

constant frequency over time and serial correlation must be


considered. However, the ability to evaluate serial

correlation is dependent upon the consistency and duration of


An important consideration with regard to application of

nonparametric procedures is power. Power of a test is the

probability that the test statistic will lead to rejection of

the null hypothesis (Ho) when it is not true. With regard to

trend analysis power refers to the probability of rejecting

the null hypothesis (no trend exists) when trend does not

actually exist. Parametric tests are more powerful than

nonparametric tests when the basic assumptions of the

parametric test are met. However, when these assumptions are

not met nonparametric tests have been shown to be more

powerful than their parametric counterparts (Harris, 1988;

Hirsch and Slack, 1984).

The data analysis necessary for developing the ambient

water quality representation must address four primary

objectives: 1) comparison of different agency data sets at

the same location, 2) identifying concentration-flow

relationships, 3) detecting and quantifying seasonality, and

4) detection and quantification of trends.

Two-Sample Tests

The analysis presents two problems that require a two-

sample test. Different data sets at the same location need to

be compared to justify combining them into one data set and


analysis of step trends involves comparison of two different

periods within the same data set. Several statistical tests

can be used to evaluate the hypothesis that two random samples

(data sets) come from the same population (Conover, 1980;

Haan, 1977; Gibbons, 1985; Kite, 1978). The Mann-Whitney test

(also called the Wilcoxon test) is considered the most

powerful non-parametric tests for differences in location

(Gibbons, 1985; Kite, 1978). Harris et al. (1987) and Ward et

al. (1988) both recommend application of the Mann-Whitney test

for analysis of groundwater quality data especially when

parametric assumptions are not meet or unknown. Berryman et

al. (1988) review several techniques for trend detection and

conclude that the Mann-Whitney test is the best choice for

step trend analysis.

The null hypothesis, Ho, is that the compared data sets

are from the same population. For two random samples X,,

X2,*...X and YI, Y2,...Yn, the Mann-Whitney criterion is based

on the sum of the rank statistics of the Y's (or X's) in the

combined ordered sequence N = m + n. The Mann-Whitney U

statistic represents the number of times a Y precedes an X in

the combined ordered arrangement. Large values of U indicate

the Y values are generally larger than the X values and

results in the rejection of the null hypothesis. When the

sample sizes are large an alternative test statistic can be

calculated whose distribution is approximately standard

normal. The Mann-Whitney test procedure is presented in


several standard texts on statistics (Gibbons, 1985; Kite,

1978; Runyun, 1977) and is a common procedure in statistical

computer analysis programs. The Mann-Whitney procedure is

presented in the Appendix.

Concentration-Flow Relationships

Correlations between river flow and concentration of

water quality parameters have been studied by several

researchers (Delfino, 1977; Hull et al., 1981; Noss and

Gladstone, 1987; Smith et al., 1982). The actual correlation

will vary from one river system to the next and between

different water quality parameters. The existence of a

correlation between concentration and flow on any given river

system usually follows one of two characteristic responses, a

dilution effect or a wash-off (erosion) effect (Hirsch et al.,

1991; Lettenmaier et al., 1991; Smith et al., 1982). The

dilution effect is characterized by a decrease in concen-

tration with increasing flow (negative correlation) and is

normally associated with dissolved constituents. The dilution

effect is typically associated with water quality parameters

where the main source is relatively constant, such as a point

source discharge or a constant base flow from an underground


The wash-off/erosion effect is characterized by increases

in concentration with increasing flow (positive correlation)

and is normally associated with sediment and parameters that


are attached to sediment. The primary sources for increasing

concentration with flow are overland flow and stream bank

erosion. Certain conditions can lead to both responses

(dilution and wash-off) for a specific parameter (Smith et

al., 1982), where the initial increase in discharge has a

dilution effect but as the discharge continues to increase the

concentration increases with increased wash-off and erosion.

Variability of the parameter concentration due to flow is

accounted for by developing a flow adjusted concentration

(FAC). The FAC is calculated by first developing a

relationship between concentration and flow. This

relationship is then used to calculate a predicted

concentration based on the measured flow. The FAC is then

calculated by subtracting the predicted concentration from the

actual measured concentration. Several relationships between

concentration and flow have been used to develop FACs for

analysis of trends in water quality (Lettenmaier et al., 1991;

Hirsch et al., 1991; Spooner, 1990; Smith et al., 1982). All

of the methods identified except one use linear regression to

estimate the coefficients of the relationship. Linear regres-

sion is conducted using least squares analysis which is not a

nonparametric procedure. However, it does provide an accepted

procedure to account for variability due to flow. The other

method described by Lettenmaier, et al. (1991) is a locally

weighted regression technique that uses a nonlinear smoothing

algorithm developed by Cleveland (1979). The data analysis


procedure does not incorporate the smoothing technique. The

functional relationships used in the data analysis are given

in Table 3-1.

Table 3-1. Concentration-Flow Functional Relationships
Functional Relationship Equation
linear C = a + bQ
log-linear C = a + b(lnQ)
hyperbolic C = a + b(1/1+fQ)
inverse C = a + b(l/Q)
quadratic C = a + biQ + b2Q2
exponential InC = a + bQ
log-log InC = a + b(lnQ)
log-quadratic-log InC = a + b,(lnQ) + b2(lnQ)2

The parameters for each of the relationships are

determined using linear or multiple linear regression, except

for beta (f) in the hyperbolic function. Eight values of f

are used resulting in eight possible hyperbolic equations. 3

is initially estimated as a function of the integer portion of


S = 10-2.5(NT(logQ))

where logQ is base 10. 3 is then incremented by a factor of

100.5 for each subsequent regression.

Significance of the various regressions is determined by

evaluating the null hypothesis, Ho: b = 0, or the assumption

that the slope of the regression line equal zero. The

regression is considered significant for p < 0.05. Selection


of the appropriate regression is based on the following


1. whether concentration increases or decreases with


2. do the data exhibit heterogeneity of variance (varying

levels of variance over the range of discharges),

3. the r2 and p values for the functional forms


4. homogeneity of the FAC residuals, and

5. the function must predict realistic concentrations

over the range of flows (e.g., no zero concentrations).

Seasonal Analysis

Seasonality within the data is characterized by an annual

cycle where the mean and or variance of one season (monthly,

bimonthly, or quarterly) is different from the mean or

variance of the next season. These seasonal variations can be

a consequence of both natural activity (biological) and

managed activity (agriculture or process management).

Additionally, changes in river flow and the source of river

flow are seasonal and impart seasonal characteristics to the

water quality data. Although, removal of variations due to

flow by calculating FACs does remove some seasonality, Hirsch

et al. (1991) show that in some cases seasonality does remain

in the flow adjusted data.


The analysis required for developing ambient water

quality is both detection and quantification of seasonality.

In traditional time series analysis, specific detection of

periodicities (seasonal cycles) within the data is

accomplished by spectral analysis (Brockwell and Davis, 1987;

Chatfield, 1984). Spectral analysis assumes equally spaced

data which is not a common characteristic for water quality

data. With respect to nonparametric statistical analysis,

detection of seasonality becomes a two-sample or multiple-

sample comparison. If prior knowledge indicates that one

season is different from the others, a two-sample test such as

the Mann-Whitney test can be applied. When seasonal charac-

teristics are not known, multiple seasons are usually

anticipated. The Kruskal-Wallis test is used to detect

differences in the medians between several samples (multiple

seasons). This procedure is simply an extension of the Mann-

Whitney test to more than two data sets and is described in

several texts (Kite, 1978; Conover, 1980; Gibbons, 1985). If

one or more of the seasonal medians is significantly different

from the others, the test statistic indicates significant

seasonality. The Kruskal-Wallis test has the same power

characteristics as the Mann-Whitney test (Harris et al.,

1987). Harris (1988) and Phillips et al. (1989) both

recommend the Kruskal-Wallis test for detection of

seasonality. The Kruskal-Wallis test procedure is presented

in the Appendix.

The Kruskal-Wallis test does not provide a quantitative

measure of seasonality. In time series analysis seasonality

is quantified and removed using various methods including: 1)

moving average filter, 2) differencing, and 3) sine or cosine

functions. These methods rely on a continuous time series

with constant time intervals between data points (Chatfield,

1975; Brockwell and Davis, 1987; Montgomery and Reckhow, 1984)

and are not generally applicable to water quality data. The

simplest approach for estimating the seasonal component is to

calculate the overall mean and variance for each seasonal

period. The simple calculation of seasonal means is readily

applicable to time series that exhibit little or no trend

(Brockwell and Davis, 1987). Where significant trend exists

it must be removed prior to estimation of the seasonal

components. In addition to seasonal means it is suspected

that variance of the data is also seasonal. Subsequently the

seasonal component includes a seasonal mean and variance.

A common graphical procedure used to evaluate seasonality

is the development of seasonal box-whiskers plots (Ward et

al., 1988). These plots have three components: 1) the median

represented by the center line in the box, 2) the

interquartiles or middle 50% of the data represented by the

box, and 3) the maximum and minimum values represented by the

whiskers. The plots are interpreted by comparing the location

and size of the boxes on the plots. Significant seasonality

in the calculated medians is indicated when the boxes do not


overlap, while differences in the length of the boxes

indicates possible seasonality in the variance (Phillips et

al., 1989). Box-whiskers plots provide a graphical means of

interpreting seasonality.

Trend Analysis

The specific method selected for the analysis depends on

the type of trend present. Water quality data are commonly

evaluated for two types of trend; monotonic trend (uniformly

increasing or decreasing) and step trend. Methods for trend

analysis of water quality data and their application have been

well documented over the past ten years (Hirsch et al., 1982;

Hirsch and Slack, 1984; Hirsch, 1988; Hirsch et al., 1991;

Lettenmaier et al., 1991; Reckhow and Stow, 1990; Smith et

al., 1982; van Belle and Hughes, 1984). Berryman et al.

(1988), provide a review of nonparametric methods for trend

analysis and conclude that Mann-Whitney, Spearman, and Kendall

tests are the best choice for trend detection in water quality

time series.

Several methods for evaluating step trends are presented

by Hirsch (1988) and Phillips et al. (1989). Two common

nonparametric procedures are the Wilcoxon Signed Rank test and

the Mann-Whitney test. The Wilcoxon Signed Rank test is

designed for comparing two data sets with paired values. This

requires each data set to have the same number of data points.

The Mann-Whitney test is a similar test but allows for the


analysis of unpaired data sets. The Mann-Whitney test

provides the flexibility needed to compare different time

periods with different sampling frequencies.

The method most often applied for evaluation of monotonic

trend in water quality data is the seasonal Kendall test de-

scribed by Hirsch et al.(1982) and Hirsch and Slack (1984).

This test is distribution free and not affected by seasonality

which makes it highly applicable to analyzing water quality

data. The Kendall test compares all possible pairs of data

values and assigns a rank of plus one if the later value in

time is higher and a negative one if it is lower. If no trend

is present, the sum of the pluses and minuses should equal

zero or be relatively close to zero. Seasonality is taken

into consideration by conducting the rankings and summing over

each season individually. The seasonal Kendall test statistic

is then calculated as the sum of the individual seasonal rank

sums and converted to a standard normal test statistic using

the mean and variance. A detailed description of the test can

be found in Hirsch et al. (1982). The seasonal Kendall test

procedure is presented in the Appendix.

The seasonal Kendall test only indicates whether or not

a trend exists and does not determine the magnitude of the

trend. To complement the seasonal Kendall test for trend,

Hirsch et al. (1982) present the seasonal Kendall slope

estimator which is a modification of the slope estimate

developed by Sen (1968) to account for seasonality. The slope


estimator is the median of the differences of the ordered

pairs evaluated for the Seasonal Kendall trend test and

quantifies the magnitude of the trend.

Serial Correlation

The most difficult consideration for the analysis of

water quality data is that of serial correlation. The

statistical methods described make the assumption of

independent data. Water quality data often exhibit

persistence over time where the measurement at one time period

will have a tendency to be high if the concentration observed

at the previous time period was high. This persistence is

commonly referred to as serial correlation or autocorrelation.

Several investigators have evaluated the effect of serial

correlation on the power of statistical tests (Berryman et

al., 1988; Harris, 1988; Hirsch and Slack, 1984; Lettenmaier,

1976). Lettenmaier (1976) presented a power function for the

Mann-Whitney test based on a parameterized statistic called

the dimensionless trend number, NT. NT is parameterized based

on the assumption of normality. The power function is

developed for daily sampling frequency and lag 1 Markov

dependence and provides an estimate of the number of samples

necessary to detect specific levels of trend. Montgomery and

Reckhow (1984) use the power function developed by Lettenmeir

(1976) to develop a series of empirical correction factors

that are used to adjust the critical levels of the statistical


tests. Use of the methods developed by Lettenmeir (1976) and

Mongomery and Reckhow (1984) require the ability to estimate

the lag 1 correlation coefficient for daily water quality

data. Hirsch and Slack (1984) propose a modification to the

original Seasonal Kendal test (Hirsch et al., 1982) to account

for serial correlation in calculation of the test statistic

(See Appendix). Application of the correction for serial

correlation does not require estimation of the correlation

structure. However, Hirsch and Slack (1984) show that

application of the correction to data sets with monthly

sampling frequency and less than ten years of record greatly

reduces the power to detect trends and recommend that the

correction only be used for data sets 2 ten years of record.

The original test is recommended when less than ten years of

record are available.

Hirsch and Slack (1984) point out the difficulty of

differentiating between trend and persistence for any

technique when the degree of serial correlation is high. More

importantly, Loftis et al. (1991) identify the need to

understand the effect of serial correlation with regard to the

time scale of the estimations and statistical analysis. The

distinction between serial correlation and trend is much more

difficult over short time periods than over long time periods.

The existence of serial correlation basically requires larger

sample sizes to achieve a desired level of precision

(Lettenmaier, 1976; Loftis et al., 1991). Berryman et al.


(1988) present a review of nonparametric tests for trend and

make several informative conclusions: 1) tests for series with

dependence are less power than tests for series without

persistence, 2) the test significance is smaller than its

nominal level (conservative) for application of the classic

tests under conditions of dependence, and 3) the more classic

tests can be applied in the presence of modest amounts of

dependence. Thus, in the presence of modest amounts of

dependence, the loss in power for application of tests for

dependence is likely greater than the gain in exactness of the

test significance.

The ability to evaluate serial correlation within water

quality data is often limited. The primary objective of the

research on serial correlation and statistical detection

methods is to improve and optimize water quality monitoring

programs. Again, the development of ambient water quality is

limited to available historical data that is characterized by

infrequent and variable sampling that makes it impractical to

estimate the correlation structure, especially on a daily

basis. An understanding of the effect of serial correlation

is essential in the interpretation of the data analysis


Formalized Data Analysis

Formalized approaches to water quality data analysis are

based on the objectives of the analysis. Reckhow and Stow


(1990) present a formalized approach for monitoring design

with an objective to estimate the number of samples necessary

(sampling frequency) to detect specific levels of trend. This

approach can be summarized in four general steps: 1) data

analysis for detection and numerical definition of seasonality

and concentration-flow relationships, 2) removal of

deterministic components (seasonality and flow variability)

from the data, examination of residuals for stationarity and

transformation as necessary to obtain stationarity, and 4) use

variance of residuals to estimate sampling frequency required

to detect various magnitudes of trend.

Several researchers have discussed the methods and

approach to trend analysis (Berryman et al., 1988; Hirsh et

al., 1982; Hirsch and Slack, 1984; Lettenmaier, 1976;

Lettenmaier et al., 1991; Montgomery and Reckhow, 1984; van

Belle and Hughes, 1984). Hirsch et al. (1991) provide a

general summary of the approach to trend analysis that

consists of four steps: 1) trend type to be analyzed step

versus monotonic, 2) selection of statistical methods, 3) kind

of data to analyze concentration versus flux data, and 4) data

manipulations for removal of natural sources of variability

(seasonality and flow) and considerations for censored data

(below detection limit).

Ward et al. (1988) present a data analysis protocol for

assessment of ground water quality data and identify the

objectives and benefit from a formalized approach to data


analysis: 1) to make informed water quality management

decisions, 2) provide a scientifically defensible description

of the water quality behavior, 3) ensure a consistent approach

to analysis when different analysts are involved, and 4)

provide a means of tracking the analysis for verification of

results and conclusions.

Literature with regard to data analysis to represent

ambient water quality for development of antidegradation

criteria is limited. A few states have developed

antidegradation criteria, but focus on the numerical

representation of the criteria versus analysis of the ambient

data to be used for developing the criteria. The Nevada

Division of Environmental Protection (NDEP, 1987) used all

available data from several sources to develop "Requirements

to Maintain Existing Higher Quality"(RMQH). RMQH's were

developed based on the complete data record and the most

recent three year period in an effort to recognize recent

possible improvements in water quality. No other preparatory

analysis of the data was reported (NDEP, 1987). Data analysis

for development of water quality criteria for Hawaii and

American Samosa consisted of selecting "healthy" sites and

separation of the data into three seasonal time periods: wet,

seasonally wet, and dry (Hawaii Department of Health, 1977;

Krock and Sullivan, 1986).

Breidt et al. (1991) present a good description of the

data analysis used to develop a distribution free represen-


station of antidegradation water quality criteria. The data

analysis included evaluation of time series plots, data

density, seasonality and trend. However, data sets that

exhibited trend, seasonality or regime shifts were eliminated

from analysis for development of the water quality criteria

(Breidt et al, 1991). Two deficiencies in this approach are

1) eliminating data with trend, seasonality or regime shifts

neglects the deterministic components that are part of the

natural characteristics of water quality and are subsequently

not represented in the criteria, and 2) when the only data

available for a specific parameter or location exhibit trend

or seasonality no criteria would be developed. In fact,

Breidt et al. (1991) reduced their data set for analysis from

thirty-six to twelve parameters.

The approach presented in this research has a primary

objective of using, justifiably, as much of the available data

as possible and representing deterministic components within

the water quality criteria.

Data Analysis Protocol

Analysis Obiectives

The objective of the analysis is to develop a data set

that represents ambient water quality conditions with respect

to a defined baseline time period. The numerical definition

of ambient water quality must incorporate natural variability,

while variability due to anthropogenic impacts is quantified


and removed from the data. Natural variability is defined by

analyzing seasonal cycles and concentration/flow relation-

ships. Anthropogenic impacts are defined by evaluating

changes in water quality over time (trend analysis) with

respect to a baseline time period. Typically the baseline

time period will be the current time period or some earlier

time period that is defined by regulations. When the baseline

time period represents an earlier time, it is possible that

water quality may have improved since that time. This

approach analyzes the compiled data to define ambient water

quality as the better water quality of either 1) the baseline

time period, or 2) that which has existed since the baseline

time period.

To provide the benefit of a formalized "data analysis

protocol" (Ward et al., 1988) for the evaluation of additional

parameters and monitoring locations, a four step approach to

the data preparation and analysis has been established. The

four major steps of the DAP are:

a. data set aggregation at each monitoring location,

b. concentration versus flow regression analysis,

c. evaluation of seasonality, and

d. water quality trend analysis.

A general flow diagram outlining the sequence of procedures

for the DAP is given in Figure 3-1. At each step the data are

evaluated using qualitative and quantitative methods to

prepare the data for subsequent stages of analysis. Each step

Figure 3-1. Data Analysis Protocol Flow Diagram


requires a decision regarding representation of the data and

no individual analytical or statistical method provides a

complete definitive answer. Consequently, the information

from all the methods used is combined to make an informed

judgement regarding the characteristics of the data.

Data Aggregation

The objective of data aggregation is to compare and

combine where justified, the data collected by various

agencies forming a single data set at a specific monitoring

location. The data sets at each location are evaluated and

compared using a four step procedure. The first two steps are

qualitative by nature while the other two are quantitative.

Geographical orientation and location. The purpose of

this step is to identify sampling location relationships to

tributaries and point source discharges. Tributaries and

point discharges into the river can have a significant effect

on the receiving water quality. A monitoring station located

at the mouth of a discharge or tributary will likely charac-

terize the mixing of the two water bodies versus the true

ambient nature of the receiving water. Monitoring stations

that more closely reflect the tributary (or discharge) should

not be included for further analysis. Also, some monitoring

stations are located upstream while others might be located

downstream of a discharge or tributary. In order to identify

and represent their impacts on water quality, the data


collected by the agencies at stations above and below tribu-

taries and point sources should be separated for comparison.

Upstream and downstream data sets should be compared and where

no difference is identified the data sets can be combined into

one data set.

This step requires obtaining as much specific information

as possible about the sampling location and plotting the

locations on a detailed map of the area. The map should

clearly indicate tributaries and point source discharges.

Time series plots. A fundamental step to any analysis of

data is a graphical representation. Time series plots of the

data provide a means of visual comparison and interpretation

of the data. For each parameter at each location, a single

time series plot is made with the data for each agency


The time series plots are used to visually compare the

sampling period of record, apparent trends, outliers and

general differences in measured concentrations over the period

of record. As previously discussed, the time series plots

also show that sampling frequency varies over the period of

record. Initially, significant differences in the measured

concentrations for one data set compared to the other agencies

at a given location identifies the possibility of significant

differences that require further investigation of that

particular data set.


Standard statistics. As the title suggests, this step

comprises computation of the standard statistics: minimum,

mean, maximum, standard deviation, coefficient of variation

and sample size. The statistics for each data set at a

specific location provide for a numerical characterization of

the data. Visual differences apparent in the time series

plots are quantified by these statistics. The skew

coefficient provides a quantitative measure of whether or not

the data might be normal, noting that the skew of a normal

distribution is zero. The mean provides a measure of central

tendency and the standard deviation represents the spread of

the data. These standard statistics can be highly influenced

by a single high measurement indicating differences in the

data. Subsequently, the nonparametric Mann-Whitney procedure,

which is not influenced by one or two high measurements, is

used for comparing the data sets.

Mann-Whitney test. The Mann-Whitney test is used to

provide a statistical comparison of the different agency data

sets at a given location. The purpose of the comparison is to

identify possible differences due to sampling and analytical

procedures. The null hypothesis for the test is that the two

data sets compared are not different (e.g. come from the

sample population). The rejection region for H. is when the

absolute value of the calculated test statistic is greater

than Z,, based on a two-tailed probability (p) of a

(confidence level of 1-a). If the two-tailed probability is


greater than a, then the hypothesis that the data sets come

from the same population is not rejected. The significance

level, a, is chosen by the analyst and a value of 0.05 is

recommended. As the calculated value for p increases the

significance of the comparison increases.

Differences in sampling and analytical procedures are not

the only possible source of variability that can cause the

Mann-Whitney test to reject the comparison of two data sets.

Data from different time periods can vary due to trends in

water quality or due to differences in flow regime (e.g.,

drought conditions). Comparison of data collected from the

same time period will reduce the effect of variability due to

different periods of record and provide a better represen-

tation of possible differences between agencies. Based on an

understanding of the sampling frequency and period of record

the Mann-Whitney test should be done for complete records and

for similar periods of record between two data sets.

Data adjustments. The raw data are adjusted for

outliers, multiple daily samples (laboratory replicates),

values reported as below detection limits (censored data), and

where justified different agency data sets combined into one

data set. If outliers are identified in the data they can be

removed. Identifying and removing outliers should be done

with careful consideration of whether it represents a mistake

or an actual value. Multiple daily samples are averaged to

represent one daily value.


The nonparametric procedures used in the analysis are not

affected by censored data (Hirsch et al., 1982; Ward et al.,

1988). Adjustment of censored data is basically a preference

choice by the analyst. Two common adjustments are: 1) to set

all censored data equal to one half the detection limit, or 2)

set all censored data equal to a value just below the

detection limit. When combining different data sets the

minimum detection level for a given parameter might not be the

same or the detection level can change over the period of

record. When combining the data sets the controlling minimum

detection level is the highest detection level. All censored

data over the period of record and from each data set combined

should be adjusted to the highest minimum detection level.

Based on the information obtained from this analysis step

the data sets are combined into one representative data set

for each specific parameter and location.

Concentration Flow Relationships

The concentration-flow relationship analysis provides

information for characterizing the type of source associated

with the water quality parameter and to develop FACs for the

trend analysis.

This step is fairly straight forward, however it is only

applicable where flow data are available. Daily flow data are

paired with water quality measurements for the same dates.

The functional relationships given in Table 3-1 are applied


and evaluated. When a significant correlation between

concentration and flow exists the best functional relationship

is chosen and FACs are calculated according to equation 3. The

best fit functional relationship might not necessarily be the

best regression fit. The functional relationship must be

physically justifiable in that it should not predict negative

concentrations for flow value within the possible range of

values. The FACs are used for the trend analysis.

Additionally, the functional relationship with flow can be

used in the development of the criteria.


Trend analysis is necessary to evaluate changes in water

quality that have occurred over time due to various impacts

within the basin (point and nonpoint source discharges) with

respect to the baseline year. These types of changes in water

quality do not reflect ambient conditions for the baseline

year. If trends in the data are identified, they are removed

from the data to represent ambient conditions for the baseline

year. For the trend analysis a baseline time period must be

selected. The baseline year is used as a reference point, but

does not imply any physical change occurred in the basin at

that time. The baseline time period will most likely be based

on specific regulations and typically will be the current time

period or reference some earlier time period.


The method used for the trend analysis, seasonal Kendall

or Mann-Whitney, will depend on the type of trend to be

evaluated. The Mann-Whitney test is used to evaluate step

trend and the seasonal Kendall test is used to evaluate

monotonic trend. Analysis for step trend applies to two

general cases: 1) when a specific event in time is known such

as a new point source is added or significant changes in a

treatment process occur, and 2) when there is a long gap in

the data. A monotonic trend is characterized by gradual

increasing or decreasing change over time. The decision to

evaluate for step trend or monotonic trend is based primarily

on visual characteristics of the time series plots for each


The Mann-Whitney test is applied as described in the data

aggregation step. The seasonal Kendall test evaluates the

significance of an apparent trend (increase or decrease) in

concentration over time. The null hypothesis, H,, is that no

change has occurred. Because both increasing and decreasing

trend are possible a two sided probability is used for the

test. The significance level, a, is selected by the analyst.

The recommended significance level is 0.05, with a

corresponding confidence level of 95%.

To understand the identified water quality changes and

obtain as much information as possible three general time

periods are evaluated: 1) the entire period of record, 2) the

record from the earliest date up to the baseline year, and 3)


the record starting at the baseline year to the most recent

available data. However, based on the actual period of

record, type of trends, pollution sources, and basin charac-

teristics the specific periods evaluated will vary. As

necessary, other time periods should be evaluated. Results

from the trend analysis of these three time periods will

assist in making judgements regarding detrending of data to

establish the baseline ambient water quality.

In addition to detecting and identifying the significance

of the trend, the magnitude of the trend is calculated. The

magnitude of the step trend between two time periods is

estimated by the difference in medians. The magnitude of a

monotonic trend is determined by the seasonal Kendall slope

estimator or Sen slope estimator (Hirsch et al., 1982).

Data adjustment. Where the identified trends do not

reflect ambient water quality of the baseline year the data

are detrended. Recall the objective of the analysis is to

represent the ambient water quality as the better water

quality of either 1) the baseline time period, or 2) that

which has existed since the baseline time period. When the

baseline year represents the best water quality the data are

detrended using the baseline year as the reference. For

example, if the water quality prior to the baseline year is

better it is adjusted upward to represent the baseline time

period. Conversely, if water quality has improved since the


baseline year, the data are adjusted to represent the better

water quality.

Seasonal Analysis

The analysis of seasonality consists of two primary

steps: 1) selection and determination of the most significant

seasonal period, and 2) numerical calculation of the seasonal

components (mean and standard deviation). Representation of

each parameter, at each location, with its corresponding

significant quarterly cycle would require development of

numerous seasonal relationships and subsequent criteria. The

seasonal analysis presented here identifies the most

consistent quarterly seasonal cycle within the data for use at

all locations for all parameters. The seasonal analysis uses

box-whiskers plots and the Kruskal-Wallis test. The null

hypothesis, H,, for the Kruskal-Wallis test is that the data

are all independent and identically distributed (no

seasonality exists). The null hypothesis is rejected if any

two seasons are significantly different. Similar to the Mann-

Whitney test a two tailed probability is used with

significance level of a.

Selection of the appropriate seasonal cycle is related to

the variability and sampling frequency of the water quality

data. To represent monthly seasonality the sampling frequency

must be equal to or greater than monthly. When the sampling

frequency varies over the period of record, the minimum common


sampling frequency should be the basis for the seasonal cycle

to be used.

If the selected seasonality is greater than twelve

months, it is appropriate to conduct an analysis to determine

the most significant seasonal cycle. For example, if

quarterly seasons are selected there are three possible

quarterly cycles which are shown in Table 3-2. The most

significant seasonality is determined by the largest

significance of the Kruskal-Wallis test on each cycle. For

the selected seasonal cycle the mean and standard deviation

for each season are calculated from the detrended data.

Table 3-2. Quarterly Seasonal Cycles
Quarter Cycle 1 Cycle 2 Cycle 3
1 Jan 1 to Mar 31 Feb 1 to Apr 30 Mar 1 to May 31
2 Apr 1 to Jun 30 May 1 to Jul 31 Jun 1 to Aug 30
3 Jul 1 to Sep 30 Aug 1 to Oct 31 Sep 1 to Nov 30
4 Oct 1 to Dec 31 Nov 1 to Jan 31 Dec 1 to Feb 29

Ambient Water Quality

The result of the data preparation and analysis presented

above provides the information necessary to represent ambient

water quality and natural variability. The ambient water

quality representation consists of three components: 1)

detrended ambient water quality data set, 2) concentration-

flow functional relationship, and 3) seasonal components

represented by the mean and standard deviation.

Application To Suwannee River: Results

The data analysis protocol is used to evaluate historical

water quality data for the Suwannee River. This approach is

readily adapted to the analysis of surface water quality data

gathered for the Suwannee. The objective for this project is

to use "the best scientific information available." The data

gathered and reviewed represent the best available information

for describing the ambient water quality of the Suwannee

River. As much of the available data as possible is

aggregated to form one composite data set at each location.

Data Agqregation

From Figure 2-1 and Table 2-1 we observe that the data

sets need to be split at Ellaville and White Springs. The

time series plots for specific conductance, total phosphorus

and total nitrogen are presented in Figures 3-2, 3-3 and 3-4

for each location. Comparison of the different periods of

record for each agency at each location identifies the 1982-

1983 period as the period of record when data were

consistently collected by most agencies. The complete data

set 21FLA 21020006 at White Springs was excluded from further

analysis for all three parameters. The statistics for each

agency at each location are tabulated in Tables 3-3, 3-4 and

3-5 for specific conductance, total phosphorus, and total

nitrogen respectively.







dU/so wn L
r L 0

q i oE:



o uJ U

m 0

U o --I
If) C1 C\J




O -

y-.- Cu U 1I
E) 000

U c t

0 0'


Iz- r .


DD 0 0 0 r-
0 0 0 0 0
-n oo 1






r i a
0 0 0
zo~w n_ LnU

0 co LU 4

(D -)) ca



w C

o 0 o 4





co U

0 E 3

Sco \c
I == I Lf


CO 0
0 0

a o C i

C. o I


Cd I lbw) N0
Q )

r z4




< )

n 0 N


D 04

d I /6u777 :NO

n r-

mL ri

to 0 04

co N1 IT m CdJ





U cn
0 0


n C

:3 D g

a u4

zn LL 0
m M

-m 4


L 0)

CM t- 0




CN I /6w) DNO0




LL }

I I I i
4 m c' c J~0















Table 3-3. Summary Statistics for Specific Conductance


FOWLERS 21FLA 21020022 82 230.0 365 74.5 0.32 28

SRWMDT SUW240C1 170 301.8 362 65.0 0.22 12

SRWMDW SUW241C1 57 206.0 338 83.5 0.41 26

FANNING 112WRD 02323500 54 233.0 369 91.0 0.39 114

SRWMDT SUW160C1 164 313.0 360 60.5 0.19 18

SRWMDW SUW160G1 57 213.3 341 76.0 0.36 62

21FLA 21020031 28 215.0 500 97.7 0.45 121

BRANFORD SRWMDW SUW140G1 45 203.6 370 93.0 0.46 62

112WRD 02320500 37 233.2 382 87.2 0.37 360

21FLA 21020035 42 201.0 410 89.7 0.45 70

SRWMDT SUW140C1 142 292.0 356 67.0 0.23 17

DERTLH TALDER09 25 165.0 320 85.0 0.52 23

LURAVILLE SRWMDT SUW130C1 127 274.0 348 68.5 0.25 18

112WRD 02320000 40 207.0 317 80.0 0.39 11

SRWMDW SUW130C1 43 185.3 330 85.2 0.46 94

DERTLH TALDER08 30 154.0 310 81.4 0.53 24

21FLA 21020034 40 182.7 360 84.0 0.46 34

ELLAVILLE 21FLA 21020014 42 163.0 270 66.5 0.41 28

SRWMDT SUW100C1 107 262.0 339 71.7 0.27 18

DERTLH TALDEROO6 33 145.0 282 76.0 0.52 25

SRWMDW SUW100G1 37 179.4 338 85.0 0.47 94

112WRD 02319500 68 180.0 291 81.0 0.45 8

21FLA 21010026 34 156.0 280 77.4 0.50 24

DERTLH TALDER005 33 145.0 282 77.5 0.53 25

SRWMDT SUW090C1 86 259.7 350 81.0 0.31 18

SUWANNEE 21FLA 21020015 41 128.0 350 73.0 0.57 25

SRWMDT SUW070C1 71 218.0 353 96.0 0.44 18

112WRD 02315550 39 120.0 390 78.5 0.65 250

SRWMDW SUW070G1 47 122.2 350 81.8 0.67 62

WHITE SRWMDW SUWO45G1 40 61.0 75 9.8 0.16 9

SRWMDW SUWO40G1 29 61.5 147 21.0 0.34 51

21FLA 21020042 35 63.0 145 22.6 0.36 40

112WRD 02315500 31 58.0 135 19.3 0.33 108

SRWMDT SUWO40C1 53 91.0 157 28.3 0.31 17

SRWMDT SUWO30C1 53 91.4 174 28.8 0.32 18