EPA/904-R-99-002
Savannah River Basin REMAP: A Demonstration of
the Usefulness of Probability Sampling for the
Purpose of Estimating Ecological Condition in
State Monitoring Programs
R. L. Raschke
H. S. Howard
R. J. Lewis
R. L. Quinn
B. L. Berrang
APRIL 1999
U.S. Environmental Protection Agency
Science & Ecosystem Support Division
Ecological Assessment Branch
980 College Station Road
Athens, GA 30605
-------
TABLE OF CONTENTS
SECTION
Table of Contents
Acknowledgments
Summary
1.0 Introduction
1.1 Purpose
1.2 Policy-Relevant Questions
1.3 Program Objectives
1.4 Description of the Savannah River Basin
2.0 Study Design
2.1 Resources of Interest
2.1.1 Streams - First through Third Order
2.1.2 Large Lake Embayments
2.2 Statistical Sampling Design
2.2.1 Frame Material
2.2.2 Sample Site Selection
2.3 Temporal Sampling Rationale
3.0 Indicators
3.1 Societal Values
3.2 Types and Selection of Indicators
3.2.1 Streams
-i-
Page
i
iv
vi
1.1
1.1
1.2
1.3
1.4
2.1
2.1
2.1
2.1
2.2
2.4
2.4
2.6
3.1
3.1
3.2
3.3
US EPA REGION 4 LIBRARY
AFC-TOWER 9th FLOOR
61 FORSYTH STREET SW
ATLANTA, GA. 30303
-------
3.2.2 Large Lake Embayments 3.5
4.0 Methods 4.1
4.1 Streams 4.1
4.1.1 Field Sampling 4.1
4.1.2 Analytical Methods 4.2
4.2 Large Lake Embayments 4.2
4.2.1 Field Sampling 4.2
4.2.2 Analytical Methods 4.3
4.3 Quality Assurance/Quality Control 4.3
4.3.1 Lakes 4.3
4.3.2 Streams 4.4
5.0 Findings 5.1
5.1 Basin Perspective 5.1
5.1.1 Large Lake Embayments 5.1
5.1.2 Streams 5.11
5.2 Ecoregion Perspective 5.17
5.2.1 Development of Scoring Criteria for Ecological 5.17
Health Assessment of the Lower Piedmont
5.2.2 LPE1 and Ecological Condition of Lower 5.18
Piedmont Streams
6.0 Discussion 6.1
7.0 References 7.1
Appendix A QA Data
¦ii-
-------
Appendix B Lake Data
Appendix C Fish Protocol & Stream Data
Appendix D "Guidelines for Locating and Accessing Sites on Wadeable Streams
in Watersheds of the Southeastern United States"
Appendix E "Cramer von Mises Test for Environmental Data"
Appendix F "Savannah River Basin Landscape Analysis"
Appendix G "Sampling Design Issues for Section 305(b) Water Quality Monitoring
Appendix H "The Ecological Condition of Small Streams in the Savannah
Basin: A REMAP Progress Report"
Appendix I "Savannah River Basin REMAP Interim Report: Large Lake
Embayments*
Appendix J "Peer Review Comments and the Response"
iii-
-------
ACKNOWLEDGEMENT S
Completion of this project was dependent upon EPA's
management and the talents of numerous individuals within and
outside of EPA. Special thanks is extended to EPA Region 4's
Administrator, Mr. John Hankinson, for his financial support and
faith in the Savannah REMAP team. Likewise, appreciation goes to
Russ Wright, Director of Region 4's Science and Ecosystem Support
Division (SESD), Delbert Hicks, Chief of the Ecological
Assessment Branch (EAB), Antonio Quinones, Acting Chief of the
EAB, and Meredith Anderson, EPA Region 4's Savannah River Basin
Coordinator for their support over the life-time of the project.
Trudy Stiber, Mel Parsons, Bo Noakes, and Jim Davee of the EAB
contributed significantly to the graphics of this report. Jim
Harrison, Dave Melgaard, Jim Greenfield, John Greis, John
Deatrick, Dan Ahern and Larry Burns deserve a special thanks for
their ideas and critiques of the early work. Gratitude is
extended to Don Schultz and Alana Conley of the Algal Assay
Laboratory who provided timely assistance. Thanks also to David
Millie and Larry Boihem of USDAs Southern Regional Research
Center and Lila Melendez and Arthur Burks of the Region 4
Analytical Support Branch (ASB) for their assistance in HPLC
analysis. Additionally, we would like to thank Jenny Scifres,
Anthony Carroll, Tom Sack, and Pam Betts of ASB for phosphorus
and solids analyses. Kent Thornton, Don Stevens Jr., Steve
Paulsen, Steve Weisberg, Gary Saul, and Christina Laurin of the
the Office of Research and Development's EMAP team provided
invaluable assistance during the planning and data evaluation
stages. Special acknowledgement is due Jim Maudsley and his team
who assisted in the field sampling and development of guidelines
for locating and accessing the sites. Likewise, a special debt
of gratitude is extended to Mike Van Den Avayle of the National
Biological Survey who worked with us on an interagency agreement
that provided the services of Dr. Steven Rathbun of the
University of Georgia. Dr. Rathbun was invaluable in providing
statistical guidance and significant work products during the
study. Likewise, the expertise of Dr. Bud Freeman of the
University of Georgia was most helpful during confirmation of
fish species.
The cooperation of the state's of Georgia and South Carolina
was the key to accomplishing numerous tasks from the planning
stage through the field work and review of reports. We
especially extend our thanks to Alan Halum, Mork Winn, Trish
Foster, David Kamps and Bill Kennedy of the Georgia Environmental
Protection Division and Lonnie Dorn of the Georgia Department of
Natural Resources; Butch Younginer, Kathy Stecker, Harry Gaymon,
-iv-
-------
Rick Renfrow, Jim Glover,Scott Castleberry, Chad Altman,and Dave
Chestnut of the South Carolina Department of Health and
Environmental Control for their steadfast assistance.
Finally, we owe a debt of gratitude to the many private
homeowners, lumber/paper companies, and forest mangers who
permitted access through their properties to the sites.
-v-
-------
SUMMARY
The Environmental Protection Agency's Environmental Monitoring
and Assessment Program (EMAP) is an outcome of EPA's National
Eutrophication and Acid Lake Monitoring Programs of the 1980s.
EMAP is a statistical sampling program that has adopted a uniform
approach for national and regional monitoring assessments across
ecosystem types. EMAP uses a serially alternative probability-
based sampling design that systematically allocates sampling
effort over space and time to ensure adequate coverage followed
with randomization to ensure unbiased estimates of status
throughout the life of a project. The design does not rely on
assumptions of population distribution, but describes the
underlying structure of the population of interest. The approach
is flexible and applicable to all landscape media. It has the
ability to increase or reduce sampling density down to the
ecoregion level, respond quickly to environmental problems,
maintain representative coverage of environmental resources, and
provide for sampling of fewer sites in an area but over rotating
cycles. Through this project, an interval-overlap technique is
presented that minimizes the loss of monitoring data when the
EMAP approach is incorporated into a fixed station (judgement)
monitoring program. The technique uses a back-prediction method
with a bias-corrective factor to best fit the two types of
monitoring derived data.
In cooperation with EMAP's desire to transfer this
monitoring approach to the EPA regions and states, Region 4
established the Regional Environmental Monitoring and Assessment
Program (REMAP). Region 4 teamed with scientists and managers in
EPA's Office of Research and Development and the states of
Georgia and South Carolina to conduct a demonstration of the new
monitoring approach, answer questions about probability sampling
and analysis, and address the concerns about the ecological
condition of streams and large lake tributary embayments in the
Savannah River Basin.
From a basin perspective, the tributary embayments with
regard to trophic condition are in good condition. At worst,
only about 5% of the acreage exhibited less than desirable
conditions. There appeared to be a general decline southward
with respect to stream EPT Index, dissolved oxygen, and
conductivity. Average stream temperatures increased southward.
Water quality violations were noted for dissolved oxygen and pH.
A dissolved oxygen violation was noted on an unnamed tributary to
Cliatt Creek in Columbia County, Georgia. Likewise about 8% of
the stream miles were less than both state's pH standard of 6.0
and 2% of the miles were greater than the allowable South
Carolina standard. An examination of basin-wide stream
-vi-
-------
conditions over a two-year period indicated that up to 52% of the
stream miles were in poor ecological condition.
Because of a sufficient number of reference and sampling
stations in the Lower Piedmont Ecoregion, EPA scientists focused
on that scale in assessing stream condition over a four-year
period. Consolidating information from an EPT Index, Fish Index,
and Habitat Score, scientists developed a Lower Piedmont
Ecological Index (LPEI). The LPEI showed that 69% of the
Ecoregion's stream miles are in fair to poor ecological
condition. Most of this adverse impact is attributed to habitat
degradation in the form of excessive sedimentation. One area of
the landscape along the 185 corridor showed an unusually high
number of poor stream sites and it is the conclusion of the
scientists that this area is in need of further study.
vii-
-------
1.0 INTRODUCTION
1.1 PURPOSE
Responding to increased population growth and demands for
multiple uses of natural resources, The Environmental Protection
Agency (EPA) established the Watershed Protection Approach (WPA)
in 1991 (EPA, 1991; 1996) . The WPA is a program for identifying
and preventing environmental problems, setting priorities, and
developing solutions through an open, inclusive process with the
people (stakeholders) who live in a geographical setting.
Consideration of economic prosperity and environmental well-being
is the cornerstone of WPA. The Savannah River Basin was one of
two areas selected in 1993 for the WPA in Region 4 because of its
high public use, known environmental problems, susceptibility for
further degradation, interest in participation by the users, and
the likelihood of success. Through the WPA initiative, EPA
Region 4 brought together scientists and stakeholders who
developed a strategy to provide an ecological focus for
resolving problems. This strategy gave birth to the Savannah
River Basin Watershed Project (SRBWP) (Management Committee,
1995) . The goal of the SRBWP is to develop and implement a
multi-agency environmental protection management project which
incorporates the authorities and expertise of all interested
parties in an effort to accomplish the vision of conserving,
restoring, enhancing, and protecting the Basin's ecosystems in a
way that allows the balancing of multiple uses. Further details
on objectives and issues within the basin can be found in Volume
I of the "SRBWP Initial Assessment and Prioritzation Report" by
the Management Committee (1995). Part of the SRBWP strategy
included a monitoring component, The Regional Environmental
Monitoring and Assessment Program (REMAP)(FTN §£. £l., 1994).
Environmental monitoring programs have developed in response
to specific needs, such as compliance monitoring by regulating
agencies responsible for the condition of surface waters, or
fixed-station monitoring networks that primarily address
indicators of exposure and stress. Some of the monitoring
programs are driven by mandates in the Clean Water Act (CWA).
The reports required by Sections 305(b) and 314 of the CWA are an
example. Programs that collect data on other ecosystem types
have also been established. For example, the U. S. Department of
Agriculture (USDA) National Agricultural Statistical Survey
collects data for agricultural resources; The Forest Service's
Inventory and Analysis Surveys analyze forest resources; and the
U. S. Geological Survey's National Water Quality Assessment
(NAWQA) program monitors water quality in selected basins. None
of the programs, however, have adopted a uniform approach for
1.1
-------
national and regional assessments across and among ecosystem
types. The Environmental Monitoring and Assessment Program
(EMAP) and its counterpart, REMAP, is intended to fill that gap
by providing the U. S. EPA Administrator, Congress, and the
public with statistical data summaries and periodic
interpretative reports on ecological status and trends. Because
knowledge about uncertainty is important for interpreting
quantitative environmental data, EMAP is designed to make
rigorous uncertainty estimates as well (Larsen si, , 1991).
The REMAP was developed as a partnership between EMAP, EPA's
Regional Offices, and States to promote the use of EMAP science.
The objectives of REMAP follow:
1. To evaluate and improve EMAP concepts for State and local
use.
2. To assess the applicability of EMAP indicators and the
EMAP approach at differing spatial scales.
3. To demonstrate the utility of EMAP for resolving issues
of importance to the EPA, Regions, and States.
The REMAP strategy lends itself to the benefits of a full
partnership between states and federal agencies because both
national and state monitoring needs can be met in a cost-
effective manner. The EMAP approach can provide a cost-effective
approach for assessing ecological data and reporting estimates of
status and trends in indicators of condition with known
confidence. State reporting requirements under several sections
of the Clean Water Act (CWA) can be accomplished using an EMAP
monitoring approach. Section 305(b) of the CWA requires states
to submit biennial reports that include analysis of water quality
data of all navigable waterways to estimate environmental
impacts. The Clean Lakes Section 314 requires states to submit
biennial reports that identify, classify, describe, and assess
status and trends in water quality of publicly owned lakes.
REMAP projects are being designed to provide meaningful
information to decision-makers within a 1- to 2-year period.
1.2 POLICY-RELEVANT QUESTIONS
The Science and Ecosystem Support Division (SESD) of EPA
Region 4 was asked by the Savannah River Watershed Project Policy
Committee to implement the REMAP strategy as a demonstration
project for the states of South Carolina and Georgia. These
states were interested in reducing sampling frequency and
analyses, having the ability to reduce or increase sampling
1.2
-------
density, responding quickly to emerging environmental problems,
and maintaining representative coverage of environmental
resources through a systematic-random means of sampling. Before
the monitoring study, a set of questions was posed by the states
of Georgia and South Carolina to provide direction for the
monitoring design. The following policy-relevant questions were
identified to guide the development of a plan of study and
subsequent monitoring efforts.
*¦ What is the status of condition of the water resources of
the Savannah River Basin?
* What proportion of the Savannah River Basin surface waters
are attaining designated uses?
* What are the changes of ecological condition over time?
* What factors might be associated with changes?
» Is there a tendency for distribution of condition in a
specific direction (spatial gradient) over the basin
landscape? What are the possible reasons for these
gradients?
* What resources are at risk in the Savannah River Basin?
1.3 PROGRAM OBJECTIVES
In response to the needs of the states and policy-relevant
questions posed, The Ecological Assessment Branch (EAB) of the
SESD developed the following study objectives with the
concurrence of the Policy Committee of the Savannah River
Watershed Project.
» Estimate the status and change of the condition of water
resources in the Savannah River Basin;
~ Identify water quality spatial gradients that exist withiu
the Savannah River Basin and associate current and changing
condition with factors that may be contributing to this
condition and spatial gradients;
» Demonstrate the utility of the REMAP approach for ecoregion
and river basin monitoring and its applicability for state
1.3
-------
monitoring programs;
~ Incorporate the REMAP approach in the formulation and
accomplishment of tha Stata River Basin Management Plans;
and
~ Provide baseline information required to conduct comparative
risk assessments in the Savannah River Basin.
1.4 DESCRIPTION 07 THE SAVANNAH RIVER BASIN
The Savannah River originates in the mountains of Georgia,
South Carolina, and North Carolina and flows south-southeasterly
312 miles to the Atlantic Ocean near the port city of Savannah,
Georgia {Figure 1.1). The Savannah River is formed at Hartwell
Reservoir by the Seneca and Tugaloo Rivers.
Headwater streams of the Seneca River are the Keowee River
and Twelve-Mile Creek. The Tugaloo River is formed by the
confluence of the Tallulah and Chattooga Rivers. The Savannah
River flowing in a south-southeasterly direction forms the border
between the states of Georgia and South Carolina. The river's
entire length of 312 miles is regulated by three adjoining Corps
of Engineers multipurpose reservoirs, each with appreciable
storage. The three lakes, Hartwell, Russell, and Thurmond, form
a chain along the Georgia-South Carolina border 120 miles long.
Six power developments that are part of the Georgia Power Company
hydropower network exist upstream of Hartwell Lake on the Tugaloo
River system; Yonah and Tugaloo lakes on the Tugaloo River, and
Tallulah Falls, Rabun, Seed, and Burton lakes on the Tallulah
River. Upstream of Lake Hartwell, on the Seneca River, is Duke
Power Company's Keowee-Toxaway Project. The project is composed
of three adjoining reservoirs, the most downstream of which is
Keowee Lake, and the other two, Jocassee and Bad Creek Lakes are
pump storage projects(Figure 1.2).
The Savannah River Basin has a surface area of 10,577 square
miles, of which 4,581 square miles are in South Carolina, 5,821
square miles are in Georgia, and approximately 175 square miles
are in North Carolina. Likre oth£r basins of large rivers in the
Southeast which flow into the Atlantic Ocean, the Savannah River
Basin embraces three distinct areas: the Mountain Province, the
Piedmont Province, and the Coastal Plain (Figure 1.3). The
mountains and Piedmont are part of the Appalachian area. The
division between the Mountain and Piedmont is an irregular line
extending from northeast to southwest, crossing the Tallulah •
River at Tallulah Falls. The Fall Line, or division between the
Piedmont Province and the Coastal Plain, also crosses the basin
in a generally northeast to southwest direction, near Augusta,
1.4
-------
Georgia. Elevations within the Mountain Province of the basin
vary from 1,500 feet National Geodetic Vertical Datum (NGVD) on
the Tallulah River to 5,030 feet NGVD for the highest peak,
Little Bald Mountain, in North Carolina along the watershed
divide. The Blue Ridge is characterized by mountains covered
naturally with Appalachian oak. Forests and ungrazed woodlands
are the predominant land uses with some cropland and pastures.
The Piedmont Province, due to its great width of over a hundred
miles, is truly Piedmont only in the upper parts, and gives way
to a midland area before reaching the Coastal Plain. Exclusive
of river valleys, its elevation generally varies from 500 feet
NGVD at the Fall Line to about 1,800 feet NGVD at its upper
extremity. The Piedmont is characterized by gently sloping hills
and smooth to irregular plains. This province is underlain
naturally with nutrient poor soils supporting oak/hickory/pine
and southern mixed forests. Land use is a mixture of crop lands,
pasture, and woodlands with some urban areas. Within the Coastal
Plain, elevations vary from 500 feet NGVD at the Fall Line to sea
level at the Atlantic Ocean. Flat plains dominated naturally by
oak/hickory/pine forests, pocasin (pine, holly) forests, southern
flood plain forests (oak/tupelo, bald cypress), and southern
mixed forests (beech, sweetgum, magnolia, pine and oak) are
characteristic of the Coastal Plain.
Within the three physiographic provinces there exist
distinct ecosystems based on the interrelationships between
organisms and their environment. These distinct ecosystems are
defined as ecoregions. Ecoregions are ecologically distinctive
areas that result from the mesh and interplay of the geologic
landform, soil, vegetative, climatic, wildlife, water and human
factors which may be present (from Wilken, 1986) While
physiographic provinces may prove suitable for regional or
national assessments, definition of ecoregions among broad
physiographic areas is necessary to accurately assess ecological
condition or health. Ecoregions are distinct areas grouped by
climate, soils, land forms, and vegetative cover. The Blue Ridge
physiographic province stands alone as a separate ecoregion as
does the Piedmont physiographic province. However, the Coastal
Plains physiographic province is composed of three distinct
ecoregions: the Fall Line Hills (or Sand Hills), the Southeastern
Plains and Hills, and the Coastal Plains.
Land use in the basin is agriculturally oriented. Sixty-six
percent of the basin is considered timberland and 34.1% is
nonforested. The number of acres farmed remains constant.
Between 1987 and 1992 there was little change in the total farm
acreage in the basin. However, Georgia had 330 fewer farms and
lesser acreage in 1992 than in 1987 while South Carolina had an
increase of 931 farms and an increase of 110,134 acres in farm
land. There was a shift over the same five-year period in the
1.5
-------
types of crops grown. An increase in the number of acres
cultivated have occurred in corn (18%), cotton (86%), peanuts
(12%), and tobacco (31%). These gains have been made with
corresponding decreases in primarily wheat (-30%) and soybeans
(-32%).
* The Savannah River Basin contains all or part of 43 counties
in Georgia, South Carolina, and North Carolina. Four of the
counties are in North Carolina, thirteen in South Carolina, and
twenty-six in Georgia. The population of the basin in 1990 was
about 1,500,000 and is expected to grow to 1,800,000 by the year
2030. About 53% of the population resides in Georgia, 42% in
South Carolina, and 5% in the headwaters located in North
Carolina. Four metropolitan areas contain 62% of the basin's
population. Savannah, Georgia is the largest city with 137,560
persons followed by Augusta, Georgia with a population of 44,619
(FTNfiial., 1994; SRBWP, 1995; EPA, 1991; EPA, 1996).
1.6
-------
CHATTQW. nrVER
NORTH CAROLINA
TALLLILAH
FUYE& ,
i KEDttEE
RIVER.
'TWELVE HLE
SOUTH CAROLINA
$
Figure 1.1 Savannah River Basin.
1. 7
-------
%ALLULAH FALLS
BURTON
• THURMOND
• RUSSELL
_
• HARTWELL
KEOWEE
RABUN
TUGALOO
YONAH
Figure 1.2 Location of Major Lakes in the Savannah River Basin
1.8
-------
Calhoun Falls
Augusta
PROVINCES
Q". gl Ccwtal Plaint
r | Mfinlwfn
I ] Piedmont
Figure 1.3. Physiographic Provinces and the Lower
Piedmont Ecoregion of the Savannah River Basin.
1.9
-------
2.0 STUDY DESIGN
2.1 Resources of Int«r«st
2.1.1 Streams
Within the basin's 10,579 square miles, there are 17,354
stream miles. An estimated 1,503 stream miles or 5.4% are
wadeable (first through tKird order) stream miles. The
population of wadeable streams of interest is those permanent
streams as indicated by a blue-line segment on a USGS 1:100,000-
scale topographic map series in digital format (DLGs) and the
modification of the DLGs represented by the U.S. EPA River Reach
File (RF3). Streams typically exhibit unilateral gravity flow
that under normal conditions are confined to a channel. All
permanent wadeable streams from Strahler first order to third
order (Chow, 1964) were included in the target population.
2.1.2 Large Lak« Embaymants
The statistical population of interest included all
tributary embayments >20 hectares associated with lakes >500
hectares. A tributary embayment is defined as a body of water
associated with, but offset from, the main lake that has a
permanent, blue-line stream at its headwaters. The embayment
begins at the plunge point, the stream stretch where the inflow
water density is greater than the density of the lake surface
water, and it joins the main body of the lake at the plane
created by' intersecting break points of the shoreline of the
embayment with the main body. Tributary embayments are
associated only with lakes that have ,9 shore line development
ratio >3.0 and a surface area >500 hectares (FTN e£ ai.., 1994) .
Shore line development is the ratio of the actual length of
shore line of a lake to the length of the circumference of a
circle the area of which is equal to that of a lake. If a lake
had a shoreline in the form of a circle, the shore line
development would be 1.0 (Welch, 1948).
Tributary embayments of six major lakes were studied over a
three-year period (1995 to 1997). These lakes were Burton,
Jocassee, and Keowee, located in the Mountain Province. The
other three lakes, Hartwell, Russell, a«d Thurmond, were located
in the Piedmont Province.
Lake Burton, controlled by Georgia Power Company, is located
near Clayton, Georgia. It is an old reaexvoir impounded in 1919.
The lake has a shoreline .length., of 62 miles surrounding 2,775
acres containing 1,000,080 acre-feet of water.
Hartwell Lake is 7 miles east of Hartwell, Georgia. A dam
2.1
-------
is located at river mile 305.0. When the lake level is at
elevation 660 ft. NGVD, the top of the conservation pool, the
lake extends 49 miles up the Tugaloo River in Georgia, and 45
miles up the Seneca and Keowee Rivers in South Carolina, covering
55,900 acres. The shoreline at elevation 660 NGVD is about 962
miles long, excluding island areas. The lake has a total storage
capacity of 2,550,000 acre-feet below elevation 660 NGVD.
Hartwell dam began operation in 1963.
Russell dam is at River Mile 275.2 in Elbert County, Georgia
and Abbeville County, South Carolina. The dam is 18 miles
southwest of Calhoun Falls, South Carolina, and 40 miles
northeast of Athens, Georgia. At the top of conservation pool
elevation of 475 NGVD, the lake has a useable storage capacity of
126,800 acre-feet and a shoreline of 523 miles encompassing
26,000 acres. Operation of the project began in January 1984.
Thurmond Lake is 22 miles upstream of Augusta Georgia. At
elevation 330 NGVD, at the top of the lake pool, the lake extends
40 miles up the Savannah River and about 30 miles up the Little
River in Georgia. The lake has about 1,050 miles of shoreline,
excluding island areas. At the top of the flood control pool
(elevation 335 NGVD), the lake has an area of 78,500 acres with a
total storage capacity of 2,510,000 acre-feet.
The three-project system is authorized and operated by the
U.S. Corps of Engineers for fish and wildlife, flood control,
hydro power, navigation, recreation, water quality, and water
supply.
Duke Power Company built and controls Lakes Jocassee and
Keowee. The upper lake, Jocassee, was built in 1973. It
contains an area of 7,318 acres holding 1,077 acre-feet of water
with a shoreline length of 75 miles. Lake Keowee, built in 1971,
has a shoreline length of 300 miles encompassing 18,373 acres
with a storage holding capacity of 955 acre-feet.
2.2 Statistical Sampling Design
A probabilistic sampling survey strategy was used to
characterize the wadeable streams and tributary embayments of the
Savannah River Basin. The sampling design was derived from the
approach used in EMAP (Messer sXl al., 1991; Overton al., 1990;
Stevens £t , 1992).
Probability sampling designs use randomization in the sample
site selection process. Probability sampling is the general term
applied to sampling plans in which
~ every member of the population (i.e., the total assemblage
from which individual sample units can be selected) has an
equal chance of being included in the sample;
2.2
-------
*¦ the sample is drawn by some method of random selection
consistent with these probabilities; and
* the probabilities of selection are used in making inferences
from the sample to the target population (Snedecor and
Cochran, 1967),
One advantage of probability-based surveys is their minimal
reliance on assumptions about the underlying structure of the
population (e.g., normal distribution). In fact, one of the
goals of probability-based surveys is to describe the underlying
structure of the population. Randomization is an important
aspect of probability-based surveys. Randomization ensures that
the sample represents the population. Without probability
sampling, each sample often is assumed to have equal
representation in the target population, even though selection
criteria clearly indicate this is not the case. Without the
underlying statistical design and probability samples, the
representativeness of an individual sample is unknown. Drawing
inferences from samples selected without randomization and
without incorporating inclusion probabilities can lead to
misleading conclusions.
One can study conditions of streams in two ways. The first
is by census, which entails examining every point on the streams.
This method is impracticable. A more practicable approach is to
examine some points systematically to ensure adequate coverage of
the basin, and randomly to prevent bias in selection of stream
points. For example, we would not obtain a good estimate of the
percent of all students in a region with hepatitis if we polled
i only students in small towns of less than two thousand people.
This preferential or biased sample would most likely include a
much lower proportion of students with hepatitis than the general
population of students. Similarly, in a stream study,
preferential sampling occurs if the sample includes only sites,
for example, downstream of sewage outfalls where sewage outfalls
affect only a small percentage of total stream length. This kind
of sampling program may provide useful information about
conditions downstream of sewage outfalls, but it will not produce
estimates that accurately represent conditions of the whole
basin. Preferential selection can be avoided by collecting
random samples.
Randomization can be thought of as a kind of lottery
drawing to determine which points are included in the sample.
Randomization is important. When used, it is possible to
estimate condition of streams with a known degree of confidence.
In REMAP, hexagons are used to add the systematic element to the
design. The hexagonal grid is positioned randomly over the basin
2.3
-------
map, and sampling points from within each hexagon are selected
randomly. The grid ensures spatial separation of selected
sampling points. This design's sampling requirements reduce
sampling locations to a logistically and economically feasible
number. It allows fewer sites to be sampled annually, but
provides for sampling of all randomly selected sites over a
rotating year period.
2.2.1 Frame Material
A sampling frame is an explicit representation of a
population from which a sample can be selected. The sampling
frame for wadeable streams and tributary embayments is the USGS
1:100,000-scale map series in digital format (DLGs) and the
modification of the DLGs represented by the U.S. EPA River Reach
File (RF3), which established edge matching and directionality in
the DLG files.
2.2.2 Sample Site Selection
The survey design follows the general design strategy
proposed for EMAP (Overton €& aJL-, 19.90; Messer gt al.,1991) . The
EMAP sampling design (Overton et. al.., 1990) achieves
comprehensive coverage of ecological resources through the use of
a grid structure. White et al. (1992) describe the construction
of the underlying triangular point grid and its associated
tessellation of hexagonal areas.
A two stage sampling approach was used to select the sample
units. The same general approach was used to select the Stage I
samples of wadeable streams. A 7x7x7 fold enhancement of the
random EMAP base grid was placed over the Savannah River Basin
(Fig. 2.1). Each grid point was circumscribed by a hexagonal
area 1.86 km2. These 1.86-km2 hexagons are aggregated into
groups of seven, one central hexagon surrounded by six other 1.86
km2 hexagons. These seven hexagons form a rough, crenulated
hexagon, or hexal of about 13 km2. Seven 13-km2 hexals comprise
one 90-km2 hexagon and there are seven 90-km2 in the EMAP base
grid hexagon which covers 640 km2 (Fig.2.1). This results in the
7x7x7 fold enhancement of"the Savannah River grid over the
original EMAP base grid. There are about forty-three 640 km2
hexagons (hex) located within the Savannah River Basin.
Stage I sampling selected three 13-km2 hexals at random
within each EMAP 640-km2 hexagon (Fig. 2.2). The process
constituted a probability sample and preserved the spatial
distribution of samples throughout the basin; Every stream reach
within each of the selected 13-km2 hexals was identified and
designated with a unique code. These streams constituted the
elements for the Stage II sample.-
2.4
-------
Stage I samples streams in direct proportion to their
occurrence on the landscape. There are orders of magnitude of
more small streams than there are large streams. Different
weights were assigned to the streams based on stream order. If
these sampling units are not weighted for size, random selection
will result in a preponderance of smaller streams in the
monitoring program.
The exact weighting procedure is based on the population
distribution of the streams. For streams in the Savannah River
Basin, a weight of 1.0 was assigned to first order streams (i.e.,
the smallest streams), a weight of 3.5 was assigned to second
order streams and a weight of 6.0 was assigned to third order
streams.
The selection process for streams illustrates the
randomization and spatial distribution preservation inherent in
the EMAP approach: For each stream segment located within each 13
km2 hexal, the length (km) of the segment and its classification
(e.g., first order, etc.) are transposed onto a line that
constitutes the total length (km) of streams of all stream orders
located within the hexal (Fig. 2.3). The individual stream
length segments are then multiplied by an appropriate weight.
All first order segments, all second order segments, etc. are
added to this line until the line contains all segment lengths
for the subject hexal. The total stream length contained within
a hexal is the sum of the stream reaches in the hexal (Fig.2.3 ).
The order of the segments on the line is randomized but the
location of each uniquely identified segment is preserved.
Following this same pattern, hexals within the EMAP 64 0-km2 hexes
are randomized (Fig. 2.4). The final line represents the total
length of all wadeable streams selected in the Stage I sample.
Spatial distributions are preserved through the randomization
process (all stream segment lengths randomized within a hexal,
hexals randomized within an EMAP 640-km2 hex and the 640-km2
hexes randomized). Once the sample size has been determined, the
total wadeable stream length (weighted) is divided by the
required sample number to derive a length interval for sample
selection. A random start location on the weighted line is
selected and sample sites are systematically drawn using the
derived length interval. For example if the weighted line is 200
km long and the sample size is 50 (200/50=4km), then a station is
selected every 4 km along the line beginning from the random
start point (Fig. 2.4).
In a similar manner large lake embayment stations were
selected for sampling. The hexagonal tessellation was randomonly
located over the area covered by the embayment population.
Within each hexagon, a point was randomonly selected. If the
point fell within one of the embayments, then that point became a
2.5
-------
sample point. The selection process ensured that each location
in the embayment population was equally likely to be sampled, and
that the set of sites was spatially distributed throughout all
embayments (Stevens, 1997).
2.3 Temporal Sampling Rational*
The EMAP has developed an approach that permits fewer sites
to be sampled annually, but provides for sampling all sites over
a rotating year period. Currently, this rotation period, or
interpenetrating cycle, is four years for the wadeable streams
and two years for the lake embayment sampling, but it can be two,
three, five years etc. This approach preserves the spatial
distribution of the samples throughout the Basin and randomonly
assigns similar numbers of streams or embayments in each year.
This reduces the sampling requirements in any year to a
logistically or economically feasible number while still
permitting estimates of resource condition. The design is well
adapted for detecting persistent, gradual change on dispersed
populations or sub populations and for representing patterns in
indicators of condition. The period for rotation is based on the
desired precision of estimates for any given year. For this
demonstration project, precision was set at +/-10% with a 95%
confidence Interval (CI) .
The large lake embayment study extended over a period of
three years. Two independent systematic random samples were
selected - one for each year. A total of 111 embayment sample
locations was selected such that 52 were allocated in 1995 and 59
in 1996. During the third year, we cycled back to the first set
of samples allocated for the embayments. For the three-year
period, 126 embayment stations out of 163 (77%) were sampled.
Those stations not sampled were non-targets, that is, the
location was on land, less than one meter deep, or inaccessible.
Sixty sites per year for a total of 240 sites over a four-
year period were selected for stream sampling. Only 119 sites
were sampled because of access denial, some were intermittent
streams, some were ponds or embayments, some were on dry land,
some were in wetlands, and a few did not meet our criteria of
hour to walk to the site.
2.6
-------
Hexal
(13 km2)
Savannah
River
Basin
EMAP Hex
(640 km2)
Figure 2.1. Illustration of Base Grid for the
Savannah River Basin
Figure 2.2. Illustration of Random Selection
of Hexals from 640~k»a Hexagon
2.7
-------
Hexal
Total Hexal Stream Length (km)
Figure 2.3. Weighted Hexal Stream Length
EMAP Hex (640 km1)
¦Hexmln—| Hexal u | Hex*la
EMAP Hex,
Figure 2.4. Total Weighted Stream
Length Selected in Stage I Sample
2.8
-------
3.0 INDICATORS
REMAP monitors ecological indicators to assess status,
trends, and changes in the condition and extent of the Region's
ecological resources (Bromberg, 1990, Hunsaker and Carpenter,
1990; Hunsaker at &!.., 1990). Indicators are defined as any
characteristic of the environment that estimates the condition of
ecological resources, magnitude of stress, exposure of a
biological component to stress, or the amount of change in
condition.
Ecological principles state that ecosystem responses and
condition are determined by the interaction of all the physical,
chemical, and biological components in the system. Because it is
impossible to measure all these components, REMAP's strategy
emphasizes indicators of ecological structure, composition, and
function that represent the condition of ecological resources
relative to societal values. The challenge is to determine which
ecological indicators to monitor. One approach for selecting
these indicators starts with those attributes valued by society
and determines which indicators might be associated with these
values.
3.1 Societal Values
To be effective, information from the monitoring program
must prompt action when required. This means the information
produced must be related to perceptions of aquatic health and
represent issues and values of concern and importance to the
public, aqi^atic scientists and decision makers. The selection of
these societal values drives the selection of appropriate
indicators. After extensive discussions with resource managers,
decision makers and the scientific community by members of the
EMAP - Surface Waters Resource Group (Larsen and Christie 1993),
an initial set of societal values and concerns were identified
for evaluation in EMAP. These values are:
~ Biological Integrity,
~ Trophic Condition, and
~ Fishability.
Biological integrity can be defined as the ability to
support and maintain a balanced, integrated, adaptive community
with a biological diversity, composition, and functional
organization comparable to those of natural lakes and streams of
the region (Frey 1977; Karr and Dudley 1981) and includes various
3.1
-------
levels of biological, taxonomic and ecological organization (Noss
1990). Biological integrity incorporates the idea that all is
well in the community. That is, the different groups are stable
and working well with little if any external management of the
community, whether it is a township, coral reef, or stream.
Waters in which composition, structure and function have not been
adversely impaired by human activities have biological integrity
(Karr &X,. 1986) . Karr and others (1986) also defined a system
as healthy "when its inherent potential is realized... and minimal
external support for management is needed." This value or ethic
differs considerably from values oriented toward human use or
pollution that are traditionally assessed in water quality and
fisheries programs, in which production of a particular species
of game fish is the goal (e.g., Doudoroff and Warren, 1957), and
may conflict with these definitions (Callicott 1991; Hughes and
Noss, 1992; Pister, 1987).
Fishability is defined as the catchability and edibility of
fish and shellfish by humans and wildlife (Larsen and Christie
1993). Fish represent a major human use of an aquatic ecosystem
product. Protecting fish is the goal of many water quality
agencies, and fish drive their water quality standards.
Trophic condition has been defined in EMAP as the abundance
of production of algae and macrophytes (Larsen and Christie
1993). Trophic condition involves both aesthetic (water clarity)
and fundamental ecological (production of plant biomass)
components. It is a key aspect in determining both a lake's
relative desirability to the public, its production of fish and
its ecological character or classification by limnologists
(e.g.,eutrophic or oligotrophic). Because of limited resources,
a decision was made to concentrate on trophic condition
indicators for lakes over a three-year period; and for streams,
we emphasized integrity all four years and trophic condition
(algal growth potential) only for two years.
3.2 Types and Selection of Indicators
EMAP defines two general types of ecological indicators,
condition and stressor indicators. A condition indicator is any
characteristic of the environment that estimates the condition of
ecological resources and is conceptually tied to a value. There
are two types of condition indicators: biotic and abiotic.
Condition indicators relate to EMAP's first and second
objectives: estimating the status, trends, and changes in
ecological condition; and the extent of ecological resources.
Stressor indicators are characteristics of the environment
that are suspected to elicit a change in the condition of an
ecological resource, and they include both natural and human-
induced stressors. Selected stressor indicators are monitored in
3.2
-------
EMAP only when a relationship between specific condition and
stressor indicators are known, or a testable hypotheses can be
formulated. Monitoring selected stressor and condition
indicators addresses the third EMAP objective of seeking
associations between selected indicators of stress and ecolpgical
condition. These associations can provide insight and lead to
the formulation of hypotheses regarding factors that might be
contributing to the observed condition. These associations can
provide direction for other regulatory, management, or research
programs in establishing relationships.
3.2.1 Streams
In concert with the EMAP approach, the Savannah REMAP
Project considered a suite of indicators to evaluate the
condition of ecological resources of streams in the Savannah
River basin. Selection of specific ecological indicators was
based on societal values. Upon consideration of the type of
streams (wadeable) to be investigated, a set of societal values
were first identified. They were biological integrity and
trophic condition. After identification of the values, four
indicators were selected to assess biological integrity and
trophic condition - benthic macroinvertebrates, fish, habitat,
and algal growth potential (AGP).
Benthic macroinvertebrate insects represent the first
consumer level in streams. They are important as processors of
organic matter, like leaves and sewage, that find their way into
a stream. By fragmenting or breaking down this organic matter,
stream insects prepare it for decomposition by bacteria that
attach too or colonize the organic matter. In turn, bacteria may
serve as a food source for other stream insects that seek out and
graze on the organic matter. Because of their limited mobility
and relatively long life span, stream insects provide a "window"
of cumulative impacts on ecological or resource condition. This
community is sensitive to changes; they have for many years been
used as a reliable barometer of water quality conditions. Some
groups of insects are very sensitive to stresses, like man-made
pollution, while others are tolerant. By focusing on the
presence or absence of different groups of insects, an aquatic
biologist is provided insight about the ecological health of a
stream. Sometimes pollution effects may stem from discharges of
chemicals, pesticides, or nutrients that are of a manmade origin.
Often, sediments from erosion and attributable to land clearing
or silviculture practices may adversely affect the stream
habitat. The materials that constitute a stream bottom are very
important to both fish and stream insects. For example, very
fine sediments, like silt, clay, or very fine sand, are
detrimental to the reproduction of fish and eliminate preferable
3.3
-------
habitat for stream insects (Plafkin e£. §Jl_. , 1989; Barbour g_£. &i..,
1998). Silt, especially, can interfere with a fish's or stream
insect's ability to breathe. Assessment of the insect community
was accomplished by using a standard field survey technique Known
as Rapid Bioassessment Protocol II(RBPII) (Plafkin gt
-------
outside the stream. One important food source is stream insects.
Changes in the stream insect community often result in a change
in the fish community. Like stream insect communities, fish
communities will respond to environmental change, whether it is
biological, chemical or physical. Some.^fishes are very sensitive
to environmental change while others are not. By examining all
fish groups that live in a stream, the general condition of a
stream can be assessed. For example, if there are only one or
two groups of fish in a stream who are very tolerant to
pollution, and there are no groups that are sensitive to a
pollutant, then impairment is suspected because of environmental
change that has eliminated the sensitive groups.
The Environmental Protection Agency's Rapid BioassesSment
Protocol V (RBP V) (Barbour e»t al., 1998) is an index used to
assess stream condition based on the fish community. The EPA RBP
V (Barbour et.. al.., 1998) is based primarily on the Index of
Biotic Integrity (IBI) (Karr,1981; Fausch et jil. 1984;Karr et al.
1986). The index consists of up to twelve measures scored to
assess changes in the fish oemmunity compared to a reference
stream, or a stream with least impact. For example, one of the
measures assesses the proportion of fishes in a stream considered
to be tolerant to environmental change. If the proportion of
tolerant groups are high compared to the reference stream, then
this would result in a lower score for that measure. Another
measure looks at the number of fish groups. If the number of
fish groups collected is similar to that of the reference stream
then this would result in a high score. After all twelve
measures have been given a score, the scores are totaled and the
condition of the fish community is then characterized as either
good, fair, or poor depending on how far the total score deviates
from that of a reference stream.
The primary indicator selected.to address trophic
condition in streams for the first two years was the algal growth
potential test (AGPT) (APHA, 1995). The AGPT is based on the
premise that maximum yield of plants (e.g. algae) is limited by
the amount of nutrients available to the test alga. With higher
algal growth concentrations (AGPT), there is good likelihood
that obnoxious plant growths can occur in a stream. The test was
selected as the indicator of choice to assess trophic condition
primarily because of its specific sensitivity, reliability and
the ease and economy of using it as a monitoring tool.
3.2.2 Largs Laic* Enbaymants
We focused on condition indicators related to trophic
condition because of limited resources. The original study plan
(FTN si. al-, 1994) proposed sampling for fishability indicators,
Fish Health Index and Fish Tissue Residues; biological integrity,
3.5
-------
phytoplankton and zooplankton identification and counts; and one
other trophic condition indicator, zeaxanthin, a marker pigment
for blue-green algae. Work is continuing on this pigment, but
the information was not sufficient for inclusion into this
report.
The trophic condition indicators measured during this study
were chlorophyll n., total phosphorus (TP), algal growth potential
(AGP), Secchi disc transparency, and total suspended solids
(TSS). These indicators were selected because they provide
different insights into the condition of the embayment waters.
Chlorophyll & is commonly used to estimate the degree of
phytoplankton bloom conditions that can affect aesthetics,
fishing and swimming quality, taste and odor of fishes and
drinking water, and the health of fish, waterfowl, and livestock.
Chlorophyll is a measure of instantaneous standing crop, whereas
TP and AGP indicate potential for blooms. Total phosphorus
reveals insights about nutrient input and the potential for
serious bloom conditions if we assume all of it is available.
However, much of the TP is not normally available. The AGP can
show how much of the TP is available for algal growth and the
potential, under optimum conditions, for blooms. Secchi disc
transparency is related to swimming conditions. Total suspended
solids is related to transparency, but it also can be used to
indicate effects upon fish production.
3.6
-------
4.0 METHODS
4.1 Streams
4.1.1 Fi«ld Sampling
Benthic macroinvertebrate sampling and habitat evaluation
followed basic guidelines put forth in the EPA document "Rapid
Bioassessment Protocols for Use in Streams and Rivers" (Barbour
SI Si., 1998). Multiple habitats (riffles, undercut banks, leaf
packs, woody debris, and pools) were sampled with D-frame and A-
frame biological dipnets according to the Ecological Assessment
Branch's (EAB) Standard Operating Procedures (SOP) (EPA, 1998).
In addition to the benthic macroinvertebrate sampling or
biosurvey, the RBP II also includes in-situ water quality
measurements (dissolved oxygen, pH, temperature, and
conductivity). These parameters were measured with a
multiparameter in-situ water quality device (HYDROLAB SCOUT)
prior to the habitat evaluation phase according to EAB's SOP.
Stream fish sampling followed basic guidelines set forth in
Barbour £t ^2.. (1998). A Smith-Root Type VII backpack
electrofishing unit was used to collect stream fish. A single
pass electrofishing run moving from downstream to upstream,
thoroughly sampling each habitat type (pools, runs, riffles,
eddies, undercut banks, etc.) was conducted at each stream
sampling location. Equal effort was given at each location.
Fish were identified at stream side and released. A few
individuals of each species were preserved in 10% formalin and
transported back to the lab for identification verification.
Based on the guidance provided in the EPA RBP V(Barbour et:.
all. 1998) document, nine metrics were utilized to evaluate the
data to assess the condition of stream fish assemblages. The
metrics were selected from a pool of metrics listed in the EPA
RBP document and other studies that have been conducted in
Georgia (DeVivo 1996). A list of metrics utilized and the
scoring criteria for each are presented in Appendix C.
Habitat assessment was based on a matrix of nine parameters
(EPA, 1989). These nine parameters fall into three principal
categories: primary, secondary, and tertiary parameters. Primary
parameters (bottom substrate, available cover, embeddedness, and
flow regime) characterize the stream "microscale" habitat and are
most influential to community structure. Secondary parameters
(channel alteration, bottom scouring/deposition, and sinuosity)
measure the "macroscale" habitat such as channel morphology.
Tertiary parameters (bank stability, bank vegetation, and stream
side cover) evaluate the integrity and composition of the
riparian zone.
4.1
-------
4.1.2 Analytical Methods
RBP II and V do not require analytical methods because the
organism identifications usually are made in the field. When
organisms need to be returned to the laboratory for
identification, they are sorted by specialists and identified by
an expert following protocols spelled out in the EAB's SOP
(1998). Algal growth potential tests conducted the first two
years followed the protocols of standard methods (APHA, 1995) as
modified by Schultz (1994) (EPA, 1998).
4.2 Large Lake Embayments
4.2.1 Field Sampling
Standard operating procedures (SOP) of EAB were followed as
the principle means of sample collection and measurement (EPA,
1998). All lake sampling and measurements took place the weeks
of 7/17 to 7/21, 1995, 6/21 through 7/5, 1996, and 7/7 through
7/10, 1997. One hundred and twenty-four stations were sampled
over the three-year period. This annual sampling window was
selected because it is a time of maximum recreational use, and
maximum water supply use.
Secchi disc transparency was measured according to EAB's SOP
that was adopted from EPA methodology (Klessig, 1988) using a 30
cm black and white disc lowered on the shady side of the boat.
Photic zone was determined by multiplying the Secchi measurement
by a factor of 2.1 (Raschke, 1993).
Collection of water samples consisted of using a battery
operated pump to fill a 5 gallon carboy with a composite depth
integrated sample taken from the photic zone (1% light level).
The water sample was mixed thoroughly and then the various
individual sample containers were filled, labeled and stored on
ice. Samples were collected for total phosphorus (TP), total
suspended solids (TSS), algal growth potential tests (AGPT), and
chlorophyll Field duplicates were collected at a minimum of
once in every ten samples. For the field duplicate, the carboy
was emptied, rinsed, and a second sample collected.
Chlorophyll & sampling" followed basic guidelines set forth
in Standard Methods. 19th Edition, section 10200. A 100 to 250 ml
sample was filtered through a 24 mm diameter Whatman GF/F glass
fiber filter. The filters were folded, blotted dry, enclosed in
aluminum foil, labeled and stored in a cooler containing dry ice
and returned to SESD for analyses. Samples were filtered in
triplicate.
4.2
-------
4.2.2 Analytical Mathods
Total phosphorus and total suspended solids were analyzed
using methods given in the EPA document " Methods for Chemical
Analysis of Water and Wastes" (EPA, 1983). In 1995, Cycle 1,
total phosphorus was analyzed using EPA Method 365.1. R®^u t/T°
most analyses were below the minimum detection level of 20 ug/
for this method. In 1996 and 1997, Cycles 2 and 3, a low
detection level method was used (EPA, 1992a) that allowed for
detection of phosphorus at 3 ug/L. Total suspended solids were
determined by using EPA Method 160.2.
Chlorophyll samples were measured by high performance liquid
chromatography (HPLC) following the basic guidelines given m
Standard Methods and in EPA Method 4 47.0. The chlorophyll was
extracted in a 90% acetone solution.
Algal growth potential test (AGPT) maximum standing crop
(MSC) and limiting nutrient was determined using The Selenastru^
Capricornutum Printz Algal Assay Bottle Test (Miller et. al.,19 )
as modified by Schultz e£. fil. (1994) .
4.3 Quality Assuranca/Quality Control
Standard operating procedures of the Ecological Assessment
Branch and the Analytical Support Branch of EPA's Region 4 SESD
were followed as the principal means of monitoring appropriate
quality assurance/quality control (QA/QC). Quality control checks
were included in sample collection, physical measurements
performed in- the field, chemical analyses, and data gathering and
processing. Data were subject to verification and validation.
Verification included range checks and internal consistency
checks. Validation consisted of a review of the data from a data
user's perspective for consistency based on known numerical
relationships.
4.3.1 Lakaa
Secchi disk transparency was measured at each site to
determine the photic zone for lake sampling. Prior repetitive
test measurements of Secchi depth in a variety of water bodies
showed that the coefficient of variation (CV) ranged from 5 to
15% among several investigators.
Water samples were collected as depth integrated samples
throughout the photic zone. Samples were collected for total
phosphorus (TP), total suspended solids (TSS), chlorophyll and
algal growth potential tests (AGPT). Field duplicates were
collected at a minimum of once in every ten samples. Results of
precision as coefficient of variation (CV) are given in Appendix
4.3
-------
A. In 1997, field blanks were collected along with the
duplicates. In this case, each of the sample containers was
filled with deionized water, preserved or filtered as
appropriate, and returned to the laboratory for analyses. Results
are given in Appendix A.
In 1995, (Cycle 1), TP in most of the samples was below the
minimum detection level of 20 ug/L for the method used. In 1996
and 1997 (Cycles 2 and 3), a low level phosphorus method was used
(EPA, 1992a). The CV for the field duplicates ranged from 0 to
71.2% with an average CV of 20.9% (Appendix A).
All of the field TSS duplicates in cycles 1 and 2 were below
the laboratory's detection limit of 4.0 mg/L. For Cycle 3, ASB
modified their procedure by filtering a greater volume of sample
(APHA, 1995). This modification reduced the detection limit to
1.0 mg/L. The CV ranged from 0 to 23.6% with an average CV of
18.6%. Standard Methods gives the CV as 33% at a concentration
of 15 mg/L TSS. Both laboratory and field precision were well
within the values of Standard Methods (APHA, 1995).
Chlorophyll a. and AGPT were measured to determine the
trophic status of the lakes. For chlorophyll a the CV for field
duplicates ranged from zero to 53.8% with an average CV of 16%.
The standard method (APHA,1995) does not give any precision data
for field duplicates that include a filtration step. The method
does state that for multiple injections on the HPLC, the average
CV for seven pigments is 10 percent.
The precision of the field duplicates for AGPT ranged from
1.3 to 53.1% with the average CV equal to 15.7%. The test gave
an average CV of 26.4% for the 1.0 to 2.0 Maximum Standing Crop
(MSC) level (Miller et. al., 1978) which was typical for the
Savannah lake samples.
4.3.2 Streams
Field measurements at each sampling station included
temperature, DO, pH, and conductivity. Measurements were taken
using a Hydrolab Scout. The Hydrolab was calibrated each morning
and then again at the end of each day according to EAB's SOP
(EPA, 1998).
Biological integrity "was accomplished in part by using a
standard field survey technique known as Rapid Bioassessment
Protocols II (RBPII) (Barbour ££. §2.., 1998) to assess the benthic
macroinvertebrate community. This is a screening procedure in
which the macroinvertebrates are identified in the field to the
family level. If identification is uncertain, the specimen is
brought back to the laboratory for verification. No replication
of sites were performed as this is a screening method.
The Rapid Bioassessment V Protocol (RBP V) (Barbour al..
1998) was the index used to assess- stream condition based on the
4.4
-------
fish community. To insure fish were properly identified during
the study, all fish that were captured during the first year were
preserved and sent to Dr. Byron Freeman at the Institute of
Ecology at the University of Georgia for identification. In
subsequent years, voucher specimens of s-ach species collected in
the field were preserved for identification verification at the
US EPA SESD laboratory. At the end of the four year study,
preserved fish with questionable identifications, were sent to
the Institute of Ecology for verification.
The primary indicator selected to address trophic conditions
in streams is the algal growth potential test. This test was also
used in the lake work and the QA/QC used is the same as given in
Section 4.3.1 except that limiting nutrient was not determined
for the streams.
4.5
-------
5.0 Findings
5.1 Basin Perspective
5.1.1 Largs Lake Exnbayments
The distribution of data for each variable can be characterized
by its cumulative distribution frequency (cdf). These curves
show the percent of embayment acreage in the basin equal to or
less than some specified measurement plus or minus a confidence
level. For the purpose of this study, we have set a confidence
level of 95%. This means that we are 95% sure that the acreage
estimated to be equal to or less than a given measurement is
within the bounds of our confidence lines on the graph (Fig.
5.1). There is a 1 in 20 chance (5% error) that the true or real
percent of acreage affected at a particular measurement is not
within the confidence bounds.
Chlorophyll 3. ranged from a low of 0.84 at Lake Hartwell to
11.56 ug/L at the most downstream lake, Lake Thurmond (Table
5.1) .
Table 5.1. Range of Values for the Savannah River Lakes
CHL. A
AGPT
Limit
TP
SD
TSS
Lakes
ug/L
mg/L
NUT.
ug/L
Maters
mg/L
Thurmond
0.98-11.56
0.66-11.0
N+P
3-50
1.2-4.8
0.7-27
Russell
1.10-5.47
0.39-2.01
N+P
3-60
0.7-3.4
2-32
Hartwell
0.84-6.84
0.33-2.27
N+P
3-30
1.4-10
1.0-6
Keowee
0.91-2.03
0.49-5.08
N+P
3-11
2.4-5.5
0.7-5.5
Jocassee
1.35-2.59
0.66-1.95
N+P
3-10
3.3-6.0
1.2-34
Burton
1.60
1.62
N
6
2.2
2
This range of concentrations at the times of sampling exhibit
trophic conditions related to classical lake classifications of
oligotrophic to mesotrophic (Olem and Flock, 1990) . Chlorophyll
3, was less than 12 ug/L over the entire basin's large lakes
(Figure 5.1). Based on experience (Raschke, 1994) over the past
30 years, generally, when chlorophyll 3, ranges from 0 to 10 ug/L,
there is no discoloration of the water and no problems. At a
range of 10 to 15 ug/L, waters can become discolored and algal
scums could develop. Between 20 to 30 ug/L, the water is deeply
discolored/ scums are more frequent, and matting of algae can
occur (Raschke, 1993). EPA Region 4 (Raschke, 1993) has shown
5.1
-------
that a mean photic zone growing season average of equal to or
less than 15 ug/L of chlorophyll a. should satisfactorily meet
multiple uses, including drinking water supply. .
One of the objectives of the Savannah River REMAP is
detect trends in important environmental variables over both time
and space. One means of comparison is through the testing ° e
null hypothesis that the population's distributions from two or
more annual cycles are identically distributed. This can e
accomplished through use of the Cramer von Mises test statis ic
Table 5.2. Cramer-von Mises Teats for Equality of Cumulative
Distribution Functions for the Savannah River Basin Embayments-
Equality of Cumulative Distribution Functions Between Cycles
(Years) is Tested.
Variable
W 1
Chlorophyll a
1.70* |
Agpt
CD
•
1 0
1 +
Total Phosphorus
3.16* |
Secchi Disc
0.44 J
Total Suspended Solids
2.84* 1
*Significant at alpha=.05
(W) which is founded on design-based methods of statistical
inference (Appendix E). For design-based statistical inference,
,the source of random variation is the random selection of sample
sites. This is in contrast to model-based statistical inference,
where the source of random variation is in the assessed
deviations from the statistical model (e.g., a regression model).
Thus, designed-based statistical inference has the advantage that
no model assumptions are required. The distribution of a
population can be characterized through its cumulative
distribution function (cdf). This is equivalent to testing the
null hypothesis that the cdf's are identical. A test of cdf
differences at alpha .05 (Table 5.2) using the Cramer-von Mises
test statistic (W) showed that four variables, chlorophyll 4,
AGPT, total phosphorus (TP), and total suspended solids (TSS) had
significantly different distributions from one cycle to the
other. Chlorophyll Cycles 2 and 3 are intertwined and slightly
different from Cycle 1 (W=1.70, k=3). The curve for Cycle 1
rises more gradually than that of Cycles 2 and 3 (Figure 5.2)
culminating in a high of 11.56 ug/L thus suggesting the mean is
higher for Cycle 1.
5.2
-------
Chlorophyll a represents phytoplankton standing crop or
yield at given time periods, whereas AGPT is representative of
the potential phytoplankton production, given optimum conditions
of sufficient nutrients, light, time and temperature. Algal
growth potential ranged from 0.33 mg dry weight (DW)/L at Lake
Hartwell to 11.0 mg DW/L at Lake Thurmond (Table 5.1)(Figure
5.3). Approximately 99.7% of the AGPT dry weights were equal to
or less than 5 mg/L (Fig. 5.3 ), an in-lake action level that
will reasonably assure protection from nuisance algal blooms and
fish kills in southeastern lakes (Raschke and Schultz, 1987) .
The 5 mg/L of dry weight translates to a potential chlorophyll
standing crop of approximately 57 ug/L of chlorophyll £ based on
the following equation:
Log10 chl a = 1.15 Log10(DW) + 0.95 (Raschke and
Schultz,1987).
The sampled maximum chlorophyll a of 12 ug/L is much lower than
the 57 ug/L of chlorophyll a. derived from the 5mg DW/L AGPT
action level suggesting that the present phytoplankton biomass
does not pose a threat to the integrity of the lake system.
Figure 5.4 depicts the AGPT cdf's for cycles one through three.
The curve for Cycle 2 rises more gradually than that for cycles
one and three suggesting the mean AGPT is not only higher in
Cycle 2, but also shows greater variability within this cycle.
The Cramer-von Mises test statistic confirms that the difference
between the three cycles at the alpha .05 level is statistically
significant (W=8.60; k=3).
Total phosphorus (TP), another indicator like AGPT of
potential production, ranged from 3.0 ug/L in most lakes to 60
ug/L in Lake Russell (Table 5.1). Approximately 87.0% of the
embayment acreage was equal to or less than 10 ug/L TP (Figure
5.5). If all of the phosphorus were available for algal growth,
at high values of 40 to 60 ug/L one could expect severe bloom
conditions, but this was not the case as seen by the relatively
low chlorophyll £ values. This is not surprising; besides
needing optimum conditions for maximum growth, the phytoplankton
need sufficient nutrients that are bioavailable to them.
Generally, not all of the TP in lakes is available for
phytoplankton growth. Peters (1981) estimated that bioavailable
phosphorus (BP) is 83% of TP in natural lakes and 18 to 57% in
rivers. Since our lakes are reservoirs and thus an extension of
a river system one would expect bioavailability to be much less
than that found in natural lakes. Previous work on the 18 Mile
Creek embayment of Lake Hartwell showed that the average percent
of BP to TP was 38% (Raschke §&. , 1985). Sometimes the BP
portion of TP can be as low at 3% (Raschke and Schultz, 1987).
At the alpha .05 level there was a significant difference
5.3
-------
(W=3.16; k=3) between Cycle 1 and the other two cycles, but
higher values were observed in Cycle 1 (Figure 5.6). The
significant differences between cycles for chlorophyll, AGPT, and
TP suggests that other variables are influencing differences from
one cycle to the other. We are not in a position with three
years of data to focus on particular stress indicators at this
time. Samples were collected from two to three weeks after
rainfall events in the basin. Thus rainfall or unusually high
stream flows would not seemingly cause the differences observed
between cycles with respect to these three phytoplankton growth
related indicators. Presumably the cyclic differences were
caused by internal lake influences like internal nutrient
cycling. Even these differences may be within the normal suite
of variability experienced in a natural setting.
For water supply, a mean growing season average Secchi disc
(SD) transparency of equal to or greater than 1.5 meters is
desirable (Raschke, 1993). For non-water supply embayment
situations a mean SD of greater than 1 meter is acceptable for
fishing and swimming (Raschke, 1993). Secchi disc transparency
ranged from 0.7 meters at Lake Russell to a high of 10 meters at
Lake Hartwell (Table 5.1). An examination of Figure 5.7 shows
that in about 2.6% of the embayment acreage, less than desirable
conditions exist for recreational purposes, and only 5.3% of the
acreage was less than the water supply criterion of equal to or
greater than 1.5 meters. Where SD was less than one meter,
measurements were located near shore or at the upper end of the
tributary embayments.
The National Academy of Sciences (1973) has set TSS levels
for different levels of stream protection. High protection can
be maintained if the TSS is 25 mg/L or less, moderate protection
is possible if the range is between 25 to 80 mg/L, low protection
is from 80 to 400 mg/L, and there is very little protection from
TSS at concentrations greater than 400 mg/L TSS. According to
these criteria, our embayment population is highly protected in
more than 95% of the embayment acreage and moderately protected
in the remaining acreage (Fig. 5.8). Buck (1956) divided
impoundments into 3 categories: clear with total suspended solids
(TSS) less than 25 mg/L; intermediate with TSS 25-100 mg/L; and
muddy with TSS greater than 100 mg/L. The mean harvest of game
fish was 162 lbs/acre for clear lakes, 94 lbs/acre in
intermediate lakes, and muddy lakes only yielded 30 lbs/acre.
The TSS ranged from a low of 0.7 mg/L at Lakes Keowee and
Thurmond to a high of 34 mg/L at Lake Jocassee, the uppermost
lake in the Savannah Chain of lakes (Table 5.1). Again these
high values were attributed to near shore stations receiving wind
fetch at the time of sampling. Ninety-seven percent of the
embayment acreage would fall into Buck's clean category, with
only 3% being intermediate with respect to water clarity (Fig.
5.4
-------
5.8). There were significant differences between the cycles
(W=2.84, k=3)(Figure 5.9). Presumably, cycle three was
significantly different from the other two cycles, because there
were no significant differences at alpha .05 between cycles one
and two (W=0.15; k=2).
5.5
-------
SAVANNAH RIVER BASIN REMAP
LAKE EMBAYMENT CHLOROPHYLL A
100
ui
o
<
UJ
on
o
<
80
60
UJ
5
>-
<
m 40
UJ
UJ
< 20
I m f
Hi
_
if
)ti
Measured Value
$
/
95% Confidence
4 6 8
CHLOROPHYLL A - ug/L
10
12
Figure 5.1. Cdf for Chlorophyll a.
LU
O
«t
LU
a:
c
>-
<
CD
100
80
60
40
20
0
SAVANNAH RIVER BASIN REMAP
LAKE EMBAYMENT CHLOROPHYLL A
0 2 4 6 8 10 12
CHLOROPHYLL A - ug/L
CYCLE 1 CYCLE 2 CYCLE 3
Figure 5.2. Cdf Curve Showing Differences
Between Cycles.
-------
SAVANNAH RIVER BASIN REMAP
LAKE EMBAYMENT AGPT
0 2 4 6 8 10
AGPT - mg/L
Figure 5.3. Cdf Curve for AGPT.
CYCLE 1 —— CYCLE 2 CYCLE 3
100
LU
o
s 80
o
<
IE 60
LU
5=
m 40
SAVANNAH RIVER BASIN REMAP
LAKE EMBAYMENT AGPT
2 4 6 8
AGPT - mg/L
Figure 5.4. Cdf Curve Showing Differences Between
Cycles.
5.7
-------
SAVANNAH RIVER BASIN REMAP
LAKE EMBAYMENT TOTAL PHOSPHORUS
100
O
<
LU
a:
o
<
80
60
>
<
m 40
uj
<
20
Measured Value
95% Confidence
10 20 30 40
TOTAL PHOSPHORUS - ug/L
50
60
Figure 5.5. Cdf Curve for TP.
LU
c
LU
C£
o
¦
in
s
UJ
LU
-------
SAVANNAH RIVER BASIN REMAP
LAKE EMBAYMENT SECCHI DISK
Measured Value
95% Confidence
6
- METERS
2 4
SECCHI DISK
Figure 5.7. Cdf Curve for Transparency.
1 —
SAVANNAH RIVER BASIN REMAP
LAKE EMBAYMENT TOTAL SUSPENDED SOLIDS
100
:NT ACREAGE
Ct) 00
o o
r~"
MBAYME
a.
o
LU
LAKE
w
o
Measured Value
0s
— —
n
1
95% Confidence
0
20 40 60
TOTAL SUSPENDED SOLIDS - mg/L
80
Figure 5.8. Cdf Curve for TSS.
5.9
-------
100
80
LU
O
<£
LU
cn
o
-
=c
CD 40 j
LU
LU
-------
5.1.2 Streams
The report by Raschke et al. (1996) (Appendix H)
demonstrated the applicability of the EMAP approach to stream
monitoring in basins. The information in this section is a
summary of four years of stream data. It is not an exhaustive
analysis of basin response. Rather, we devoted our energies to
demonstrating the applicability of the EMAP approach to an
ecoregion and the application of modified indicators and a new
index that incorporates macroinvertebrate and fish metrics
(Section 5.2).
The family level EPT Index ranged from 1-20 across all six
ecoregions (Appendix C). EPT Index scores exhibited a general
decline southward along successive ecoregion belts (Fig. 5.10).
However it should be pointed out that the small sample sizes
within each ecoregion, with the exception of the Lower Piedmont,
is inadequate to confirm this observation. The Blue Ridge
Mountain Ecoregion had the highest EPT Index scores (range = 8 -
20; n = 11). Mean EPT Index value in the Blue Ridge was 15. Only
3 sampling stations were in the Upper Piedmont; EPT Index scores
for the 3 Upper Piedmont stations were 9, 16, and 16. EPT Index
scores in the Lower Piedmont (range = 1 - 18; n = 88) were lower
than the Blue Ridge and Upper Piedmont and the mean EPT Index
value of 7 was much lower than that of the Blue Ridge (15). Five
stations were located in the Sand Hills where the EPT Index
ranged from 3 to 11. Ten stations were located in the
Southeastern Plains; the EPT Index ranged from 2 to 11 with a
mean EPT Index value of 7. Only two stream stations were located
in the Middle Atlantic Coastal Plain. An EPT Index value of 1
was recorded for the Middle Atlantic Coastal Plain stations.
Habitat evaluation scores for all sites ranged from 30 to
123 Figure 5.11). Habitat evaluation scores for each stream
station are presented in Appendix C. Unlike the EPT Index
results, habitat evaluation scores did not reveal any marked
patterns from an ecoregional perspective (Figure 5.10). The Blue
Ridge habitat evaluation scores(N = 11) ranged from 58 to 123
with a mean of 90. The Upper Piedmont (only 3 stations) had
habitat evaluation scores of 82, 102, and 112. The Lower
Piedmont's 88 stations had a wide range in habitat scores (30 to
119) with a mean score of 71. The Sand Hills ecoregion stations
(N = 5) had a range in habitat evaluation scores of 92 to 108.
Habitat evaluation scores for stations in the Southeastern
Plains(N = 10)ranged from 73 to 120. The two stations in the
Middle Atlantic Coastal Plain had habitat evaluation scores of 96
and 99.
Of the 118 sampling stations for the Savannah REMAP Project,
88 of them are in the Lower Piedmont ecoregion. Seventy-eight of
these Lower Piedmont stations had data for all three indicators
5.11
-------
(EPT, Fish IBI, and Habitat) utilized for ecological assessment.
The other ecoregions within the project area did not have a
sufficient number of sampling stations to adequately assess
ecological condition. Statistical analysis was therefore
restricted to the 78 station data set for the Lower Piedmont
ecoregion.
During the four year study period, fish were collected from
108 stream stations. Over 10,000 fish, comprising 49 different
species (Table 5.3), were collected. Appendix C list the species
and the number collected at each stream station.
Stream fish were collected from six different ecoregions in
Table 5.3 Summary of the number of fish collected over the four year
study.
Ecoregions
Stream
Stations
Number
Fish Species
Number of Fish
Identified
Blue Ridge
11
17
318
Upper Piedmont
3
8
267
Lower Piedmont
82
43
9103
Sand Hills
3
9
48
Southern Plains
8
26
329
Mid-Atlantic
Coastal Plain
1
2
9
Total
108
49*
10074
* - Number represents
the study, not the
total number
column total,
of different species
collected during
the Savannah Basin {Table 5.3). Eighty eight (over 75%) of the
stream stations were located in the Lower Piedmont ecoregion.
The Lower Piedmont is the largest ecoregion in the Savannah River
Basin. Only one stream station was located in the Mid-Atlantic
Coastal Plain.
Ranges of in-situ watex quality measurements (pH, dissolved
oxygen, conductivity, and temperature) are presented in Table
5.4. In regard to pH, no ecoregional pattern or characteristic
emerged. Although the remaining water quality parameters are
lacking in number of observations for the Uj^per Piedmont, Sand
Hills, and Middle Atlantic Coastal Plain, there appears to be a
gradient from the mountains to the coast (Figure 5.11). This
occurs as a decrease in dissolved oxygen and an increase in the
temperature regime from the Blue Ridge to the Middle Atlantic
Coastal Plain. Although not as apparent as dissolved oxygen and
5.12
-------
temperature, conductivity, with the exception of the Sand Hills,
also increased along this same ecoregional gradient. Again, more
data points are necessary to validate this pattern.
Table 5.4 In-situ Water Quality Data
Ecoregion
PH
D.O.
(mg/1)
Conductivity
CuS/ca)
Temperature
rc)
Blue Ridge
6.6 - 7.6
7.9 - 9.5
16 - 29
16.5 - 23.7
Upper Piedmont
6.3 - 7.0
8.2 - 8.5
20 - 40
22.0 - 23.2
Lower Piedmont
5.1 - 9.1
3.6 - 11.3
15 - 3260
17.5 - 28.2
Sand Hills
5.2 - 6.9
6.7 - 7.9
18 - 914
20.9 - 25.6
South Eastern
Plains
6.1 - 7.5
6.3 - 8.3
36 - 184
20.9 - 25.5
Mid-Atlantic
Coastal Plain
4.1 - 6.0
5.1 - 6.9
58 - 60
25.6 - 25.8
Water quality violations were noted for dissolved oxygen and
pH during the in-situ water quality measurements. Dissolved
oxygen at Station 98, an unnamed tributary to Cliatt Creek, in
Columbia County, Georgia was measured at 3.6 mg/L which is below
the two state's water quality standards of 4.0 mg/L. This
translates into about 2% of the stream miles being below the
minimum standard dissolved oxygen in the basin (Figure 5.12).
(Likewise, about 8% of the stream miles were below both state's pH
standard of 6.0 and approximately 2% were greater than the
allowable level for streams in Georgia (8.5) and South Carolina
(8.0) (Figure 5.13).
Algal growth potential tests were conducted for the first
two years and analyzed from a basin perspective. The results of
that effort and interpretation of the data are in a report by
Raschke, gt al. (1997) (Appendix H).
5.13
-------
EPT Indax Scores
£
8
CO
t
UJ
22
18
14
10
-2
r-Q-i
BR UP LP SH US/SP MACP
Ecoregion
ZH Non-Outlier Max
Non-Outlier Min
[ZD 75%
25%
~ Median
o Outliers
Habitat Evaluation Scoras
=§
BR
140
120
100
40
UP
LP SH
Ecoregion
US/SP MACP
_1— Non-Outlier Max
Non-Outlier Mln
CZD 75%
25%
~ Median
Figure 5.10 Box and Whisker Plots of Ecoregion EPT Index
Scores and Habitat Scores.
5.14
-------
Noo-Ou«i« Mm
MACP • e***na»
Dissolved Oxygen
USfSP MACP
X NmOiiMrMn
Non-Oultor Mta
a 75*
25%
a IMn
• Outtars
! NocvOuttor Max
NonOuferMin
~ 75%
25%
Conductivity
MO
~ 180
100
g 60
o
20
1
o
1
i
>
i
J:
1
a
—
.0.
. -CI-
X NoivOuHirMK
Non-OufltorftfUn
~ 75%
25%
BR UP LP SH US/SP MACP
Ecoceglon
• Outlfers
Figure 5.11. Box Whisker Plots of Ecoregion In-Situ Water Quality Parameters.
5.15
-------
c
l>
c
V
-I
E
ra
100
80
60
E 40
20
0 Ss
2
Dissolved Oxygen in Streams
Savannah REMAP (1994-1997)
Ua. and
SC
Standard
Measured Value
95% Confidence
6 8
Dissolved Oxygen (mg/L)
10
12
Figure 5.12 Cdf curve of Dissolved Oxygen Data.
Measured Value
95% Confidence
Standard
Outside
Standards
| Outside
Standards
pH in Streams
Savannah REMAP (1994-1997)
Figure 5.13 Cdf curve of pH data.
5 . 16
-------
5.2 Ecoregion Perspective
Because of our original emphasis on Basin ecological
condition, sampling locations were randomly selected over the
whole Savannah River Basin, not by ecoregion. This skews the
number of sampling locations in favor of the largest ecoregion,
which was the Lower Piedmont. The Lower Piedmont ecoregion is a
large geographical area that encompasses two states and many
subwatersheds. There were not enough stream stations in all of
the ecoregions to adequately develop an index for each
ecoregion. Only the Lower Piedmont region had sufficient number
of stream stations to produce enough data, in our opinion, to
develop an index that realistically assesses ecological
condition.
5.2.1 Development of Scoring Criteria for Ecological Health
Assessment of the Lower Piedmont Ecoregion
Benthic macroinvertebrate, fish, and habitat were the basis
for interpreting the ecological health of Savannah REMAP
wadeable stream sites in the Lower Piedmont Ecoregion.
Specifically, the EPT Index (macroinvertebrates), the fish IBI
(Index of Biotic Integrity), and habitat evaluation scores were
utilized to develop a scoring system for classifying Lower
Piedmont streams into three categories (good, fair, poor).
Sampling stations for the Savannah REMAP were located in six
different ecoregions, however, 88 of the 119 were in the Lower
Piedmont ecoregion which provided a sufficient database to
examine ecological health in this ecoregion.
The choice of metrics was determined by correlation
analysis. Correlation analysis is important in the choice of
metrics because it identifies redundancy. Metrics that are very
highly correlated should be interpreted with caution since they
may indicate some overlap or redundancy; metrics that are highly
correlated do not contribute new information to an assessment
(Barbour £t jjil., 1996). Habitat evaluation scores and EPT Index
results were not significantly correlated thus both of these
ecological indicators were acceptable tools for bioassessment.
Although Fish IBI and habitat evaluation scores were
significantly correlated (p<.05 = 0.42), the correlation was
more on the order of moderate rather than strong correlation
(Appendix C).
Descriptive statistics of all seven variables examined in
all 88 Lower Piedmont stations are presented in Table 5.5.
Box and whisker plots (Figure 5.15) were performed on the
results for each indicator to define the boundaries for three
categories (Good, Fair, and Poor). A scoring matrix based on
boundaries defined by box and whisker plots was completed for
the Lower Piedmont Ecoregion. The scoring matrix for the EPT
Index, Fish IBI, and Habitat is provided in Table 5.6.
5.17
-------
Table 5.5 Descriptive Statistics of the Stream Variables.
Descriptive
Statistics
I
Variables
# of
Standard
Stations
Mean
Minimum
Maximum
Deviation
Fish IBI Scores
82
26.00
13.00
43.00
6.26
Habitat Scores
84
70.99
30.00
119.00
21.68
EPT Scores
87
7.14
1.00
18.00
3.19
PH
84
6.91
5.10
9.10
0.53
Temperature (C)
83
23.02
17.5
28.2
2.05
Dissolved Oxygen (mg/1)
75
7.29
3.6
11.3
1.14
Conductivitv («S/cm)
76
80.58
15.00
243.00
43.92
Table 5.6 Scoring Matrix for Ecological Health of Lower
Piedmont Streams
Indicator
5 points
GOOD
3 points
FAIR
1 point
POOR
EPT Index
> 9
6 -
8
< 5
Fish IBI
> 31
22 -
- 30
< 21
Habitat
> 87
53 -
- 86
< 52
The next step was defining a final classification system
based on the total score obtained from all three indicators for
the 78 station Lower Piedmont data set. Again, box plots were
utilized to define the boundaries for total scores in the
"Good", "Fair", and "Poor" categories. This final
classification system is termed the Savannah Basin-Lower
Piedmont Ecological Index (SB-LPEI).
5.2.2 SB-LPEI and Ecological Condition of Lower Piedmont Streams
Final ecological health classification of Lower Piedmont
streams, based on total points derived from the three ecological
indicators (EPT Index, Fish IBI, and Habitat), was determined by
the following scheme:
5.18
-------
Classification
Total points
Good
> 11
Fair
8-10
Poor
<7
(Note: a score of 1 in either of three ecological
indicators does not warrant a "Good" ranking)
Based on this scoring scheme, 69% of the stream miles
indicated some degree of impairment ("Fair" and "Poor" rankings)
(Figure 5.14). A complete listing, by station, of the
individual ecological indicator results and the final ecological
health classification from the results of the SB-LPEI is
provided in Appendix C. Habitat degradation, primarily from
sedimentation, is apparently the leading cause affecting the
aquatic life in Lower Piedmont streams. Habitat evaluation
parameters such as bottom substrate/available cover, channel
alteration, and bottom scouring and deposition specifically
identify sedimentation concerns. Low scores in these three
sediment-related parameters of the habitat evaluation worksheet
translated into less than desirable benthic macroinvertebrate
and fish populations. Conversely, ecoregional reference sites
scored higher in these three sediment-related parameters and
supported diverse fish and macroinvertebrate communities.
5.19
-------
Habitat Evaluation
EPT Indax
» Non-Outliar Max * 11
Non-Outli«r Min * 30
' » 75* - 87.S
25% - 52
a Median - 68
22
18
10
S
H
» Hon-Outlier Mu ¦ 15
Non-Outlier Hln * 1
ZH3 75% • 9
25% - 5
Median - 7
Outliers
Fish IBI
SO
45
40
M 35
o
o
a
« 30
rfS 25
20
15
10
1 Non-Outlier Max » 43
Non-Outlier Min * 13
ZZ) 75% - 31
25% - 21
o Median - 25
LPEI
Non-Outlier Max
Hon-Outlier Min <
75% - 11
25% - 7
Median • 9
15
3
Figure 5.14. Box and Whisker Plots Used to Develop the Scoring Criteria for the Savanna
Basin-Lower Piedmont Ecological Index.
ua
-------
V)
0)
E
ra
0)
+-I
to
100
80
60
40
20
0
2 3
Ecological Index Score
Savannah REMAP (1994-1997)
Measured Value
95% Confidence
Fair
Poor
Good
8 9 10 11 12 13 14 15 16
Index Score
Figure 5.15. Cdf Curve of Savannah Basin-Lower Piedmont
Ecological Index.
5 .21
-------
6.0 Discussion of Objectivea
Estimate the status and change of the condition of water
resources in the Savannah River Basin.
Based on three years of measuring trophic condition of the
tributary embayments of large lakes in the basin, the data show
that the lakes' embayments are in good condition. Only about 5%
of the embayment acreage exhibited less than desirable conditions
with respect to recreation and water supply use (Raschke, 1993)-
Much of that could be attributed to wind fetch at the near-shore
stations. Significant changes from cycle to cycle possibly are
within the realm of natural variability or some unmeasured
stressor indicators within the lakes' environs. Sampling took
place several weeks after rainfall events, therefore, external
stream inputs were not expected to cause the observed differences
tostween cycles•
In evaluating the status of ecological health of streams in
the Savannah Basin, both biological and habitat parameters were
examined to arrive at a final estimate of the ecological
condition of wadeable streams. There appeared to be a genera
decline southward with respect to EPT Index, DO, and
conductivity. The temperature gradient decreased in a "ar
direction. Water quality violations were noted for DO and p .
DO violation of <4.0 was observed at Station 98 on an unnamed
tributary to Cliatt Creek in Columbia County, Georgia. ^Likewise,
about 8% of the stream miles were less than both states p
standard of 6.0, and 2% of the miles were greater than the
allowable South Carolina level of 8.0.
In-depth data analysis, as indicated in Section . '
restricted to streams in the Lower Piedmont Ecoregion .
there was not sufficient biological data for a t or g _
Of other ecoregions. Data analysis ^ £o *he ^"components of
22 EPT Index, an,
the RBP v habitat evaluation scores. .
This SB-LPEI was successful in establis:i™?nt Ecoregion.
"status" of wadeable streams in the Lower *Jere
Based on the SB-LPEI, sixty-nine percent of the stre,a
classified as -fair" or "poor" indicati^,e^^?!ation
Impairment at these sites pointed to habitat degrad
primarily from excessive sedimentation.* The *®su future
LPEI can be utilized to establish areas of concern
SValUSang; in ecological condition was not "tablishedjuring
this study. There was not enough data for all st y ^
confidently evaluate change over the four year study period.
6.1
-------
Identify water quality spatial gradients that exist within the
Savannah River Basin and associate current and changing condition
with factors that may be contributing to this condition and
spatial gradients.
Analysis of information by ORD, NERL-LasVegas (Appendix F)
showed that landscape indicators like percent forest cover,
forest edge, proportion of watershed area with agriculture or
urban land cover(U-Index), agriculture edge, average patch,
average forest patch, and agriculture on slopes >3% were
significantly correlated with the stream indicators AGPT, EP
Index, Fish IBI, and Habitat Score (Appendix F). NERL-LasVegas
showed that both the proportion and patterns of land use are
useful in assessing potential causative effects of stream
condition. Landscape indicators at the subbasin scale provi e
the best characterization of the basin.
In a previous Savannah REMAP report using two years of
stream data, Raschke et al. (1996) identified one area that had
an inordinate amount of bad sites clustered around Hart and
Franklin Counties, Georgia near Interstate 85. Upon review of
four years of data and taking a very conservative approach in
developing criteria for poor ecological health, the information
revealed that this area is much larger than expected. It has
expanded into South Carolina (Figure 6.1). This area includes
all or part of Hart and Elbert Counties, Georgia and Oconee,
Pickens, and Anderson Counties, South Carolina. The designation
of an area does not imply that every stream is in "poor
condition nor that the area has a certain confidence band. Our
observations are qualitative, that is, there is an unusual number
of poor areas clustered, in our professional opinion, along the
Interstate 85 corridor. We believe streams in this area are most
vulnerable to landscape perturbations and in need of further
detailed investigation.
The landscape analysis showed that approximately 64% of this
"poor" area is forest, 22.3% agriculture, 2.6% urban, and 3%
barren. Two percent of the area is in agriculture on slopes >3%,
there is approximately 21% agriculture on moderately erodible
soils, and approximately 1% on highly erodible soils, and <0.1%
agriculture on slopes >3% in highly erodible soils.
This area has been subjected to a considerable increase in
population growth because of the large impoundments in the upper
part of the Savannah River Basin. Furthermore, examination of
GIS information shows that it has a high density of chicken
production, extensive agriculture in large blocks, and the
headwaters of streams in the subbasins have a high density of
roads. In some subbasins of this "poor" area, the forest land is
highly fragmented and the land has been opened up to
industrial/urban/and agriculture development in the headwaters of
6.2
-------
some of the streams.
Damonstrata tha utility of the REMAP approach for acoragxon and
rxver basin monitoring and its applicability for stats monitoring
programs.
In the arena of state monitoring, the concept of probability
sampling is like the "new kid on the-block" - the one who dresses
differently and acts differently. And we, the regions and
mirr°rin9 real life, have been slow in warming up to this
id, and rightfully so! For he embraces a new way of thinking
at threatens stability, cultural traditions, and the past
istorical record. From the inception of this project, we were
aware of the potential disruption that probability sampling could
create among our state partners. So we diligently set a course
o testing the EMAP approach and determined how we could bes't
incorporate it into state monitoring schemes with as little
disruption as possible. We sought out and found Dr. Steve
a hbun of the University of Georgia Statistical Department. He
1S a statistician who has experience in different types of
probability sampling approaches and experience with the problems
° incorporating the "new kid on the block" into traditional
s ate monitoring programs. Rathbun addressed concerns regarding
probability-based designs posed by the "Assessment Design Focus
roup of the 305(b) Consistency Workgroup (Appendix G ). His
u ^ report in Appendix G is an important first step in the
integration of judgement and probability monitoring data without
losing most of the historical data.
States and the federal government historically have
established monitoring networks based on judgmental sampling.
at is, stations were usually located where there were pollution
pr°klems or the area was vulnerable to pollution because of man's
HiVities- Unf°rtunately, this type of site selection is biased
and it is virtually impossible to relate to a whole population of
streams/lakes, watersheds, basins, ecoregions etc. Sampling
designs based on judgement sampling are not likely to yield
representative samples.
With the need for preserving historical monitoring data and
marrying it to a probability-based design, Rathbun (Appendix G)
tested an approach using an interval overlap technique with
historical judgement sites and probability-based sites located
near judgement sites. The technique uses a back-prediction
method that determines what the historical data should have been
had a probability-based sample design been implemented from the
very beginning of the program. If the above methods shows there
is still some bias in the data, then a bias-corrective factor is
calculated to best fit the data.
6.3
-------
LOWER PIEDMONT ECOLOGICAL INDEX
Ecological Index
» Poor 5.7
Pak $-10
• Good > 11
A/PwAim
/\/ Interstate*
A/ Savannah Lakee
Figure 6.1. Area in the Lower Piedmont with an Unusual Amount
Poor Sites.
6.4
-------
Incorporate the REMAP approach in the formulation and
accomplishment of the state river basin management plans.
Most states are monitoring their basins on a cyclic schedule
rather than doing state-wide monitoring every year. This report
shows that it is possible to incorporate probabilistic sampling
(the EMAP approach) into state monitoring programs at the basin
level and even the ecoregional level. Rathbun (Appendix G)
presents a method of incorporating historical judgement station
data into a probabilistic design. This is important because the
states can better estimate stream miles impacted etc. and have
sufficient data for trend analysis. We can't predict to what
degree each state will incorporate probability sampling into
their monitoring programs. As of the distribution of this
report, we have had a workshop on integration of judgement data
with probability data. The workshop addressed state concerns and
opened the door for joint discussions. Likewise, the Office of
Water has directed the states to move toward probability sampling
for purposes of including better estimates of ecological
condition into the 305(b) reports. South Carolina is moving
toward probability sampling, Alabama has partially incorporated
it into their monitoring program and Kentucky is evaluating it
presently.
Provide baseline information required to conduct cojsparative risk
assessments in the Savannah River Basin.
REMAP is not a problem-specific program. It focuses on
monitoring the condition or system response, and changes in the
condition of the ecological resource; not specific physical
alterations, chemical species or associated problems. Biological
1u
-------
7.0 REFERENCES
APHA. 1995. Standard Methods for the Examination of Water and
Wastewater, 19th Edition. American Public Health Association,
Washington, D.C.
Buck, D. H. 1956. Effects of turbidity on fishing. Transactions
of the American Wildlife Conference 21:249-261.
Barbour, M. T., J. L. Plafkin, B. P. Bradley, C. 6. Graves, and
R. W. Wiseman. 1992. Evolution of EPA's Rapid Bioassessment
Benthic Metrics: Metric Redundancy and Variability Among
Reference Stream Sites. Environmental Toxicology and Chemistry
11:437-449.
Barbour, M. T., J. Gerritsen, B. D. Snyder, and J. B. Stribling.
1998. Draft Revision to Rapid Bioassessment Protocols for Use in
Streams and Rivers. EPA 841-D-97-002, Office of Water,
Washington, D. C.
Bromberg, S. M. 1990. Identifying ecological indicators: An
environmental monitoring and assessment program. Journal of the
Air Pollution Control Association 40:976-978.
Callicott, J. B. 1991. Conservation ethics and fishery
management. Fisheries 16:22-28.
Chow, Ven te (Editor). 1964. Handbook of Applied Hydrology: A
Compendium of Water-Resources Technology. McGraw-Hill Book Co.
New York, New York.
Doudoroff, P., and C. E. Warren. 1957. Biological indices of
water pollution with special reference to fish populations. Pages
144-163. In: Biological Problems in Water Pollution. U. S. Public
Health Service, Cincinnati, Ohio.
EPA. 1983. Methods for the Chemical Analysis of Water and Wastes.
EPA-600/4-79-020. Revised March, 1983. y.S. Environmental
Protection Agency, Cincinnati, OH.
EPA. 1991. The Watershed Protection Approach, an Overview. EPA
503/9-92/002. U. S. Environmental Protection Agency, Washington,
D. C.
7.1
-------
EPA. 1992a. Methods for the Determination of Chemical Substances
in Marine and Estuarine Environmental Samples. U.S. Environmental
Protection Agency, Cincinnati, OH.
EPA. 1992b. Framework of Ecological Risk Assessment: Risk
Assessement Forum. EPA 630/ R-92/001. EPA, Washington, D. C.
EPA. 1996. Watershed Approach Framework. EPA 840/S-96/001. Office
of Water. U. S. Environmental Protection Agency, Washington, D.
C.
EPA, 1998. EPA Region 4, Science and Ecosystem Support Division,
Ecological Assessment Branch SOP and QA Manual. Athens, 6A.
Frey, D. 6. 1977. Biological integrity of water - A historical
approach. Pages 127-140. In: R. K. Ballentine and L. J. Guarria
(Eds.). The Integrity of Water. U. S. Environmental Protection
Agency, Washington, D. C.
FTN Associates, R. L. Raschke, H. Howard. 1994. Plan of Study for
Characterizing Ecological Condition in the Savannah River Basin.
U. S. Environmental Protection Agency, Region 4, Science and
Ecosystem Support Division, Ecological Assessment Branch, Athens,
Georgia.
Hughes, R. M. , and R. F. Noss. 1992. Biological diversity and
biological integrity: Current concerns for lakes and streams.
Fisheries 17:11-19.
Hunsaker, C. T., D.E. Carpenter, and J. Messer. 1990. Ecological
indicators for regional monitoring. Bulletin of the Ecological
Society of America 71 (2):165-172.
Hunsaker, C. T. , and D. E. Carpenter. 1990. Environmental
Monitoring and Assessment Program - Ecological Indicators. EPA
600/3-90/060. U. S. Environmental Protection Agency, Research
Triangle Park, North Carolina.
Karr, J. R. , and D. R. Dudley. 1981. Ecological perspective on
water quality goals. Environmental Management 5:55-68.
Karr, J. R. , K. D. Fausch, P. L. Angermeier, P. R. Yant, and I.
J. Schlosser. 1986. Assessing Biological Integrity in Running
Waters: A Method and its Rationale. Publication No. 15. Illinois
Natural History Survey, Champaign, Illinois.
7.2
-------
Klessig, L. 1988. Problem Identification. In: The Lake and
Reservoir Restoration Guidance Manual. EPA 440/5-88-002. Criteria
and Standards Division, Washington, D. C.
Larsen, D. P., D. J. Stevens Jr., A. R. Selle, and S. G. Paulsen.
1991. Environmental Monitoring and Assessment Program, EMAP -
surface waters, a Northeast lakes pilot. Lake and Reservoir
Management 7(1): 1-11.
Larsen, D. P., and S. J. Christie (Eds.). 1993. EMAP-Surface
Waters. 1991 Pilot Report. EPA 620/R-93/003. Office of Research
and Development, U. S. Environmental Protection Agency,
Washington, D. C.
Management Committee. 1995. Savannah River Watershed Project:
Initial Assessment and Prioritization Report for the Savannah
River Basin. Volumes 1 and 2. Prepared by the Management
Committee of the Savannah River Basin Watershed Project.
Messer, J. J., R. A. Linthurst, and W. S. Overton. 1991. An EPA
program for monitoring ecological status and trends.
Environmental Monitoring and Assessment 17:67-78.
Miller, W. , J. Green, and T. Shiroyama. 1978. The Salt>n««<-riim
caoricornutum Prints Algal Assay Bottle Test. EPA-600/9-78-018.
National Academy of Sciences and National Academy of Engineering.
1973. Water Quality Criteria 1972. Superintendent of Documents,
U. S. Government Printing Office. Washington, D. C.
Noss, R. F. 1990. Indicators for monitoring biodiversity: A
hierarchical approach. Conservation Biology 4:355-364.
Olem, H. and G. Flock (Eds.). 1990. Lake and Reservoir
Restoration Guidance Manual. EPA 440/4-90/006. Second Edition
Prepared by The North American Lake Management Society for the U.
S. Environmental Protection Agency, Washington, D. C.
Overton, W. S., D. Stevens Jr., D. White 1990. Environmental
Monitoring and Assessment Program - Design Report for EMAP. EPA
600/3-91/053. U. S. Office of Research and Development,
Environmental Protection Agency, Washington, D. C.
Peters, H. 1981. Phosphorus availability in Lake Mexnphren&gog and
7.3
-------
its tributaries. Limnology and Oceanography 26:1150-?
Pister, E. P. 1987. A pilgrims progress from Group A to Group B.
Pages 221-232. In: J. B. Callicott (Ed.). Companion to Sand
County Almanac. University of Wisconsin Press, Madison,
Wisconsin.
Plafkin, J. L., M. T. Barbour, K. D. Porter, S. K. Gross, and R.
M. Hughes. 1989. Rapid Bioassessment Protocols for Use in Streams
and Rivers: Benthic Hacroinvertebrate and Fish. EPA 440/4-89-001.
Office of Water, U. S. EPA.
Raschke, R. L. 1994. Phytoplankton bloom frequencies in a
population of small southeastern impoundments. Lake and Reservoir
Management 8 (2):205-210.
Raschke, R. L. 1993. Guidelines for Assessing and Predicting
Eutrophication Status of Small Southeastern Piedmont
Impoundments. U. S. Environmental Protection Agency, Region 4,
Science and Ecosystem Support Division, Ecological Assessment
Branch, Athens, Georgia.
Raschke, R. L., and D. A. Schultz. 1987. The use of the algal
growth potential test for data assessment. Journal of the Water
Pollution Control Federation 59(4):222-227.
Raschke, R. L., B. Cosgrove, and M. Koenig. 1985. Phosphorus
loading effects upon phytoplankton standing crop in a tributary
embayment of a southern piedmont reservoir. Pages 139-148. In:
Lake and Reservoir Management: Practical Applications. North
American Lake Management Society, Madison, Wisconsin.
Raschke, R. L., H. S. Howard, J. R. Maudsley, and R. J. Lewis.
1996. The Ecological Condition of Small Streams in the Savannah
River Basin: A REMAP Progress Report. EPA Region 4, Science and
Ecosystem Support Division, Ecological Assessment Branch, Athens,
GA.
Raschke, R. L., R. Quinn, T. Olsen, and D. Stevens Jr. 1997.
Savannah River Basin REMAP Interim Report: Large Lake Embayments.
EPA Region 4, Science and Ecosystem Support Division, Ecological
Assessment Branch, Athens, GA.
Rathbun, S. C., R. Houghton, and C. Laurin. 1996. Cramer-von
Mises Tests for Environmental Monitoring Data. A report for the
7.4
-------
U. S. Environmental Protection Agency, Science and Ecosystem
Support Division, Ecosystem Assessment Branch, Athens, Georgia.
Schultz, D., R. Raschke, and R. Jones. 1994. A shortened algal
growth potential test. Environmental Monitoring and Assessment
32:201-205.
Snedecor, 6. W., and W. 6. Cochran. 1967. Statistical Methods.
Sixth Edition. The Iowa State University Press, Ames, IA.
Stevens, D. L. Jr. 1994. implementation of a national
environmental monitoring program. Journal of Environmental
Management 42:1-29.
Stevens, D. L. Jr. 1997. Variable density grid-based sampling
designs for continuous spatial populations. Environmetrics
8:165-197.
Stevens, D. L. Jr., A. R. Olsen, and D. White. 1992. Integrated
Sampling Design: Environmental Monitoring and Assessment Program.
EPA 600/xxx/xxx. U. S. Environmental Protection Agency,
Environmental Research Laboratory, Corvallis, Oregon.
Urquhart, M. S., W. S. Overton, and D. S. Birkes. 1993. Comparing
sampling designs for ecological status and trends: Impact of
temporal patterns. In: V. Barnett and K. F. Turkman (Eds.).
Statistics for the Environment. Wiley, Mew York.
Wallace, J. B., J. W. Grubaugh, and M. R. Whiles. 1996. Biotic
Indices and Stream Ecosystem Resources: Results from an
Experimental Study. Ecological Applications 6(1):140-151.
Welch, P. S. 1948. Limnological Methods. McGraw-Hill Book Co.,
Inc., Mew York.
White, D., A. J. Kimerling, and W. S. Overton. 1992. Cartographic
and geometric components of a global sampling design for
environmental monitoring. Cartography and Geographic Information
Systems 19:5-22.
7.5
-------
>
-------
Appendix A
QA Data
-------
Table QC-1.
SAVANNAH REMAP STUDY- FIELD DUPLICATES
AGPT in milligram Dry Weight per Liter
STA#
5
19
39
49
59
79
97
101
105
8
16
41
43
CYCLE
2
2
2
2
3
3
3
3
REP1
REP2
MEAN
SD
CV
1.01
0.86
0.94
0.134
14.3%
0.84
0.55
0.70
0.258
37.1%
0.66
0.67
0.67
0.009
1.3%
0.46
0.41
0.45
0.062
14.0%
4.46
5.08
4.77
0.552
11.6%
1.82
1.51
1.67
0.276
16.6%
2.96
1.60
2.28
1.210
53.1%
1.81
2.07
1.94
0.231
11.9%
1.99
1.71
1,85
0.249
13.5%
0.93
0.79
0.86
0.125
14.5%
1.41
1.35
1.38
0.053
3.9%
1.20
1.13
1.17
0.062
5.3%
1.08
1.17
1.13
0.080
7.1%
15.7%
CHLOROPHYLL a in ug/L
STA#
5
19
39
49
59
79
101
8
16
41
43
CYCLE
2
2
3
3
3
3
TOTAL PHOSPHORUS in ug/L
STA#
79
97
101
105
8
16
41
43
TSS in mg/L
STA#
8
16
41
43
CYCLE
2
2
2
2
3
3
3
3
CYCLE
3
3
3
3
REP1
REP2
MEAN
SD
CV
1.51
1.98
1.75
0.418
24.0%
2.44
2.13
2.29
0.276
12.1%
1.65
1.70
1.68
0.045
2.7%
2.98
2.83
2.91
0.134
4.6%
0.91
0.93
0.92
0.018
1.9%
0.83
1.55
1.19
0.641
53.8%
2.12
1.80
1.96
0.285
14.5%
1.50
0.90
1.20
0.534
44.5%
1.60
1.60
1.60
0.000
0.0%
7.80
8.30
8.05
0.445
5.5%
2.60
3.00
2.80
0.356
12.7%
16.0%
REP1
REP2
MEAN
SD
CV
6
7
6.5
0.890
13.7%
12
7
9.5
4.450
46.8%
14
6
10.0
7.120
71.2%
9
10
9.5
0.890
9.4%
4
4
4.0
0.000
0.0%
5
4
4.5
0.890
19.8%
29
27
28.0
1.780
6.4%
12
12
12.0
0.000
0.0%
20.9%
REP1
REP2
MEAN
SD
CV
1U
2.0
2.2
1.7
1.95
0.445
22.8%
4.7
3.6
4.15
0.979
23.6%
1.8
2.0
1.9
0.178
9.4%
18.6%
U - material was analyzed for but not detected. The number is the minimum quantitation limit.
-------
Table QC-2
SAVANNAH REMAP STUDY - FIELD BLANKS
SAMPLE#
STA#
CYCLE
TSS
TP
CHL a
(mg/L)
(ug/L)
(ug/L)
9609
41C
3
1U
3U
0.5U
9683
8C
3
1U
3U
0.5U
9696
43C
3
1U
3
0.5U
9741
16C
3
1U
3U
0.5U
U = Material was analyzed for but not detected. The number is the minimum detection limit.
-------
CO
-------
Appendix B
Lake Data
-------
SAVANNAH RIVERBASIN LAKE DATA
LAKE
STATION
JOCASSEE
1
KEOWEE
4
KEOWEE
5A
KEOWEE
5B
KEOWEE
7
KEOWEE
8
HARTWELL
10
HARTWELL
11
HARTWELL
13
HARTWELL
14
HARTWELL
15
HARTWELL
16
HARTWELL
17
HARTWELL
18
HARTWELL
19A
HARTWELL
19B
HARTWELL
20
HARTWELL
22
HARTWELL
24
HARTWELL
25
HARTWELL
26
HARTWELL
27
THURMOND
28
THURMOND
29
THURMOND
30
THURMOND
31
THURMOND
32
THURMOND
34
THURMOND
35
THURMOND
36
THURMOND
38
THURMOND
39A
THURMOND
39B
THURMOND
40
THURMOND
41
THURMOND
42
RUSSELL
43
RUSSELL
44
RUSSELL
45
RUSSELL
46
RUSSELL
48
RUSSELL
49A
RUSSELL
49B
AGPT
DATE
(MG/L)
07/20/95
0.68
07/21/95
1.12
07/21/95
1.01
07/21/95
0.86
07/21/95
0.64
07/21/95
0.79
07/20/95
1.49
07/20/95
0.91
07/20/95
1.97
07/20/95
0.81
07/20/95
0.86
07/20/95
0.88
07/20/95
0.80
07/20/95
1.26
07/20/95
0.84
07/20/95
0.55
07/20/95
1.63
07/20/95
0.62
07/19/95
0.71
07/19/95
0.65
07/19/95
0.55
07/19/95
0.70
07/19/95
0.79
07/19/95
0.75
07/19/95
0.69
07/19/95
0.67
07/19/95
0.98
07/18/95
0.92
07/18/95
0.71
07/18/95
0.66
07/18/95
0.66
07/18/95
0.67
07/18/95
0.66
07/18/95
1.01
07/18/95
1.34
07/18/95
0.74
07/17/97
0.85
07/17/95
0.95
07/17/95
0.64
07/17/95
0.75
07/17/95
0.63
07/17/95
0.48
07/17/95
0.41
LIMITING
CHLa
JUTRIENT
(UG/L)
N+P
2.54
N+P
2.03
N+P
1.51
N+P
1.98
N+P
1.77
N+P
1.66
N+P
3.55
N+P
3.89
N+P
2.35
N+P
2.37
N+P
2.39
N+P
2.02
N+P
1.22
N+P
6.84
N+P
2.44
N+P
2.13
N+P
1.55
N+P
1.74
N+P
0.84
N+P
0.98
N+P
1.77
N+P
4.16
N+P
1.87
N+P
1.31
N+P
2.35
N+P
2.18
N+P
4.50
N+P
2.16
N+P
1.95
N+P
1.60
N+P
1.94
N+P
1.65
N+P
1.70
N+P
3.40
N
11.56
N
11.17
N+P
2.92
N
3.70
N
5.47
N+P
2.59
N+P
3.39
N+P
2.98
N+P
2.83
TPHOS
TSS
(UG/L)
(MG/L)
20 U
34
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
6.0
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
30
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
20 U
6.0
20 U
4.0 U
20 U
4.0 U
20
4.0 U
50
4.0 U
30
4.0 U
20 U
4.0 U
20 U
4.0 U
60
32
20 U
4.0 U
20 U
4.0 U
20 U
4.0 U
-------
SAVANNAH RIVERBASIN LAKE DATA
AGPT
CYCLE LAKE STATION DATE (MG/L)
1 RUSSELL 50 07/17/95 0.42
1 RUSSELL 51 07/17/95 0.41
1 RUSSELL 52 07/17/95 0.39
2 JOCASSEE 53 07/05/96 1.67
2 JOCASSEE 55 07/05/96 1.84
2 JOCASSEE 56 07/05/96 1.95
2 BURTON 57 07/05/96 1.62
2 KEOWEE 59A 07/03/96 4.46
2 KEOWEE 59B 07/03/96 5.08
2 KEOWEE 60 07/03/96 NA
2 KEOWEE 62 07/03/96 2.51
2 KEOWEE 65 07/03/96 1.81
2 KEOWEE 66 07/03/96 1.11
2 KEOWEE 67 07/03/96 1.25
2 HARTWELL 70 07/02/96 1.33
2 HARTWELL 73 07/02/96 2.27
2 HARTWELL 74 07/02/96 1.62
2 HARTWELL 75 07/02/96 1.51
2 HARTWELL 77 07/02/96 1.41
2 HARTWELL 78 07/02/96 2.02
2 HARTWELL 79A 07/02/96 1.82
2 HARTWELL 793 07/02/96 1.51
2 HARTWELL 80 07/02/96 1.87
2 HARTWELL 81 07/02/96 1.51
2 HARTWELL 84 07/01/96 1.44
2 THURMOND 87 06/28/96 8.35
2 THURMOND 88 06/26/96 3.47
2 THURMOND 89 06/26/96 1.60
2 THURMOND 93 06/25/96 1.97
2 THURMOND 95 06/25/96 1.83
2 THURMOND 96 06/24/96 2.82
2 THURMOND 97A 06/25/96 2.96
2 THURMOND 97B 06/25/96 1.60
2 THURMOND 98 06/25/96 3.49
2 THURMOND 99 06/24/96 1.76
2 THURMOND 100 06/24/96 2.90
2 THURMOND 101A 06/26/96 1 81
2 THURMOND 101B 06/26/96 2 07
2 THURMOND 103 06/24/96 1 80
2 RUSSELL 105A 07/01/96 1 99
2 RUSSELL 105B 07/01/96 1.71
2 RUSSELL 109 07/01/96 2.01
2 RUSSELL 110 07/01/96 1.90
LIMITING
NUTRIENT
N+P
N+P
N+P
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N+P
N
P
P
P
N+P
P
N
P
N
N
P
N+P
N
N
N
N
CHL a
(UG/L)
2.17
2.11
1.88
2.59
1.89
1.35
1.60
0.91
0.93
0.80
1.04
0.58
0.77
0.77
1.28
2.32
1.54
1.31
1.85
1.72
0.83
1.55
0.77
1.21
1.17
0.98
6.95
3.19
1.87
1.42
2.56
0
1.62
3.90
2.20
3.39
2.12
1.80
2.47
1.2 U
1.85
3.05
2.38
TPHOS
(UG/L)
20 U
20 U
20 U
6 U
6 U
6 U
6
6 U
6
6 U
6 U
6 U
6
6
7
7
6
8
6
6
6
7
8
6
6
11
28
19
7
6
10
12
7
23
9
9
14
6
9
9
10
11
8
TSS
(MG/L)
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
72
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
4.0 U
-------
SAVANNAH RIVERBASIN LAKE DATA
AGPT
LIMITING
CHLa
TPHOS
TSS
LE LAKE
STATION
DATE
(MG/L) NUTRIENT
(UG/L)
(UG/L)
(MG/L)
3 JOCASSEE
1
07/08/97
0.66
N+P
1.8
4
1.2
3 KEOWEE
4
07/09/97
0.87
N
1.4
11
5.5
3 KEOWEE
5
07/09/97
0.72
N
1.6
4
3.8
3 KEOWEE
7
07/09/97
0.88
N+P
0.93
3 U
0.8
3 KEOWEE
8A
07/09/97
0.93
N
1.5
4
0.9
3 KEOWEE
8B
07/09/97
0.79
N
0.90
4
2.0
3 KEOWEE
8C
07/09/97
0.49
N+P
0.5 U
3 U
0.0
3 KEOWEE
9
07/09/97
1.22
N
1.3
4
0.7
3 HARTWELL
10
07/09/97
1.25
N
1.6
8
2.1
3 HARTWELL
11
07/09/97
0.33
N+P
2.1
11
1.8
3 HARTWELL
13
07/10/97
1.18
N+P
1.6
8
2.4
3 HARTWELL
14
07/10/97
1.35
N+P
1.8
4
1.6
3 HARTWELL
15
07/10/97
1.73
N
1.6
4
1.6
3 HARTWELL
16A
07/10/97
1.41
N+P
1.6
5
2.2
3 HARTWELL
16B
07/10/97
1.35
N+P
1.6
4
1.7
3 HARTWELL
16C
07/10/97
0.42
N+P
0.5 U
3 U
0.1
3 HARTWELL
17
07/10/97
1.13
N+P
1.1
4
1.6
3 HARTWELL
18
07/10/97
2.27
N+P
5.4
18
4.6
3 HARTWELL
19
07/10/97
1.40
N+P
1.6
4
1.9
3 HARTWELL
20
07/10/97
1.37
N+P
1.4
4
1.5
3 HARTWELL
22
07/10/97
1.30
N+P
1.2
4
1.0
3 HARTWELL
24
07/09/97
0.93
N+P
0.99
4
1.0
3 HARTWELL
25
07/09/97
1.16
N
1.1
4
1.2
3 HARTWELL
26
07/09/97
0.98
N+P
0.94
4
0.9
3 HARTWELL
27
07/09/97
1.12
N+P
2.3
12
2.5
3 THURMOND
28
07/07/97
1.16
N
3.7
8
2.6
3 THURMOND
29
07/07/97
0.84
N+P
2.6
5
2.3
3 THURMOND
30
07/07/97
0.91
N
3.5
7
2.8
3 THURMOND
32
07/07/97
0.84
N+P
3.2
6
2.5
3 THURMOND
34
07/07/97
1.38
N
1.4
6
1.9
3 THURMOND
35
07/07/97
1.23
P
1.6
3
1.5
3 THURMOND
36
07/07/97
1.21
N+P
1.3
8
1.2
3 THURMOND
38
07/07/97
1.09
P
1.3
5
0.7
3 THURMOND
39
07/07/97
0.95
P
1.5
8
1.9
3 THURMOND
40
07/07/97
0.99
N+P
1.9
10
2.9 U
3 THURMOND
41A
07/07/97
1.20
N
7.8
29
4.7
3 THURMOND
41B
07/07/97
1.13
N
8.3
27
3.6
3 THURMOND
41C
07/07/97
0.40
N+P
0.5 U
3 U
0.8
3 THURMOND
42
07/07/97
1.48
N
5.4
12
2.6
3 RUSSELL
43A
07/09/97
1.08
N+P
2.6
12
1.8
3 RUSSELL
43B
07/09/97
1.17
N+P
3.0
12
2.0
3 RUSSELL
43C
07/09/97
0.44
N+P
0.5 U
3
0.0
3 RUSSELL
44
07/09/97
1.29
N
3.4
10
1.7
-------
SAVANNAH RIVERBASIN LAKE DATA
AGPT
LIMITING
CHLa
TPHOS
TSS
CYCLE
LAKE
STATION
DATE
(MG/L) NUTRIENT
(UG/L)
(UG/L)
(MG/L)
3
RUSSELL
45
07/09/97
1.33
N
1.6
10
1.9
3
RUSSELL
46
07/09/97
1.32
N
1.6
9
2.1
3
RUSSELL
48
07/09/97
1.32
N+P
3.3
7
1.4
3
RUSSELL
49
07/09/97
1.19
N
2.5
12
2.1
3
RUSSELL
50
07/09/97
1.31
N+P
1.6
15
3.8
3
RUSSELL
51
07/09/97
1.23
N+P
2.1
12
22
3
RUSSELL
52
07/09/97
1.26
N
1.1
3
2.8
U = Material was analyzed for but not detected. The number is the minimum quantitation limit.
Chlorophyll a
Cycle 1 - USDA (HPLC)
Cycle 2 - USDA (HPLC)
Cycle 3 - EPA (HPLC)
-------
ft
-------
Appendix C
Fish Protocol & Stream Data
-------
Rapid Bioassessment Protocol V Metric Development
The EPA RBP V (Barbour et. al., 1998)is based primarily on
the Index of Biotic Integrity (IBI)(Karr,1981; Fausch et al.
1984;Karr et al. 1986). The IBI incorporates up to twelve
metrics which are scored to assess changes in the fish community
compared to a reference stream, or a stream with minimal impact,
of similar size and geographic area to that of the stream being
sampled. Like stream insect communities, fish communities will
respond to environmental change.
The EPA RBP V's (Barbour et. al., 1998) twelve metrics were
originally developed for Midwestern streams. They are not
intended to be used verbatim in other geographical areas. The
metrics presented in the RBP document are prototypes to be used
as guidance for developing metrics in other geographical areas.
Barbour et. al. (1998) also present modifications to the IBI that
other researches have made to make the IBI more applicable to
their regions or study area. Metric development is based on
reference fish community or a fish community with minimal impact.
After evaluating the fish data collected over the four year
study, it was determined that selecting metrics that assessed
the basic fish community structure was the most effective way to
screen, or evaluate, streams in such a large geographical area,
especially due to the nature of the study design. Ideally, when
conducting any IBI study, metric development is based on a
reference fish community in the area of study. As part of the
study design a reference fish community would be established.
However, the approach used in this study did not focus on
establishing reference fish community data. This makes it
difficult to develop metrics, with confidence, that are more
discriminating of the subtle differences within a large
watershed. The results of this IBI analysis should be used to
identify problem areas in the Basin, at which point an IBI study,
which involves the agusition of the necessary reference data, can
be implemented to address the problems areas.
Nine metrics were utilized to evaluate the data to assess
the condition of stream fish assemblages (table 1). The metrics
were selected from a pool of metrics listed in the EPA RBP
document and other studies that have been conducted in Georgia
(DeVivo 1996).
The first seven metrics assess the fish assemblage structure
and the last two assess the fish assemblage function. The
assemblage structure metrics will all decrease with increased
stream degradation. Combined, these metrics assess impacts to
the stream from physical and chemical degradation. Of the two
assemblage function metrics, proportion of omnivores will
increase with increased stream degradation, and proportion of
benthic Invertivores will decrease with increased stream
degradation.
No metrics that assess fish abundance and condition were
-------
utilized for this study. Metrics 11 and 12 listed in the RBP
document (Barbour et. al.1998), "Proportion of disease/anomalies"
and "Proportion of Hybrids" requires a certain level of training
to properly assess these metrics. The skill level among the
sampling crews varied. Therefore, consistent assessment of these
metrics was not possible. The abundance metric was not
incorporated because of too much variance in the data. It was too
difficult to determine any patterns or trends and establish
scoring criteria for the
metric.
The scoring criteria for
each of the nine metrics were
based on the data collected
from the 82 lower Piedmont
streams. Reliable reference
data was unavailable for this
study due to the nature in
which sampling locations were
selected. Therefore, an
alternative method for
developing scoring criteria
was utilized. The range of
metric results were trisected
to produce three different
ranges of results. A good
result is given a score of 5,
a medium range result is give:
a score of 3 and a lower rang
result is given a score of 1.
This is considered an
acceptable method for
developing scoring criteria
for IBI metrics (Karr, 1996). Metric scoring criteria are
presented in Table 2.
Initially, scoring criteria were developed regardless of
stream order. That is, the same criteria was applied to all
three stream orders (1st, 2nd , and 3rd) that were sampled.
Pearson Product Moment Correlation were calculated for all stream
sampling parameters, which also included stream order and IBI
score. A positive correlation was indicated for stream order and
IBI score. This indicated that the IBI score increased with
stream order designation, suggesting that the scoring criteria
favored third order streams and that the other streams were
scored unfairly. To resolve this issue the metric scores were
recalibrated based on stream order. Separate scoring criteria
were developed for each stream order, so that each stream order
had its own set of scoring criteria for each metric. After the
metrics were recalibrated to compensate for differences in
stream order, Pearson Product Moment Correlation were
recalculated and the results indicated there was no significant
Table 1 List of IBI metrics utilized for
the HBP analysis.
Index of Biotic Integrity Metrics
Fish Assemblage Structure Metrics
1 Number of Species
2 Proportion of Non-Native Species
3 Brillioun Diversity Index
4 Number of Native Suckers
5 Number of Native Sunfishes
6 Number of Minnow Species
7 Number of Darter Species
Fish Assemblage Function Metrics
8 Proportion of Omnivores
9 Benthic Inveitivores
-------
correlation between stream order and IBI score.
After all metrics were calculated and scored for a
particular stream station, the metric scores were summed to give
one final IBI score for that particular stream station. The
condition of the fish community is then usually characterized as
either "Good"/ "Fair", or "Poor", depending on how far the total
score deviates from the total possible score. These
characterizations were developed by applying box and wisker plots
to the range of final scores. Scores that were in the upper 25
percentile (>29) were classified as being in Good condition.
Scores that fell between the 25 and 75 percentile (22-29) were
classified as being in Fair condition, and scores that were in
the lower 25 percentile {<22) were classified as being Poor
Table 2 Scoring criteria for the metrics utilized for the RBP V (IBI) analysis.
Stream
Metric Score Criteria
Community Structure Metrics
Order
]
3
5
1 Number of Species
1
<5
5-7
>7
2
<7
7-13
>13
3
<9
9-14
>14
2 Proprotion of Non-Native Species
I
>27
14-26
<14
2
>16
8*16
<8
3
>20
10-20
<10
3 Brillioun Diversity Index
1
<0.303
0.303-0.523
>0.523 1
2
<0.34
0.34-0.60
>0.60
3
<0.70
0.82-0.70
>0.82
4 Number of Native Suckers
1
0
0
1 1
2
<2
2
>2 I
3
<2
2
>2 I
5 Number Native Sunfish
1
<3
3-4
>5 I
2
<3
3-4
>4 1
3
<2
2-3
>3
6 Number of Minnow Species
1
<2
3-4
>4 1
2
<3
3-4
>4
3
<4
4-5
>5
7 Number of Darter Species
1
<1
1
>1 1
2
<2
2
>2 1
3
<1
1-2
>2 1
3 >43 22-43 <22
condition. An example of how a stream station is scored and
-------
Table 3 Example of scoring IBI metric results.
Stream Station
Index of Biotic Integrity Metrics
93
122
37
Result Score
Result Score
Result Score
I Number of Species
19 5
2 1
13 3
2 Proportion of Non-Native Species
0.39 5
o
o
2.94 5
3 Brillioun Diversity Index
0.88 5
0.09 1
0.77 3
4 Number of Native Suckers
3 5
1 1
1 1
5 Number of Native Sunfishes
4 5
0 1
2 3
6 Number of Minnow Species
4 3
2 1
5 3
7 Number of Darter Species
4 S
0 1
3 5
S Proportion of Omnivores
18.99 5
100 1
31.62 3
9 Benthic Invertivores
34.11 5
0.00 1
11.76 1
Total IBI Score
43
13
27
Classification
Good
Poor
Fair
categorized is presented in Table 3.
-------
References
Barbour, M. T., J. Gerritsen, B. D. Snyder, and J. B. Stribling.
1998. Draft Revision to Rapid Bioassessment Protocols for Use in
Streams and Rivers. EPA 841-D-97-002, Office of Water,
Washington, D. C.
DeVivo, J. C., C.A Couch, B.J. Freeman. 1996. Preliminary Index
of Biotic Integrity in Urban Streams Around Atlanta, Georgia.
Proceedings of the 1997 Georgia Water Resources Conference, held
at The University of Georgia.
Fausch, K.D., J.R. Karr, and P.R. Yant. 1984. Regional
Application of an Index of Biotic Integrity based on Stream Fish
Communities. Transactions of the American Fisheries Society
113:39-55.
Karr, J.R. 1981. Assessment of Biotic Integrity Using Fish
Communities. Fisheries 6:21-27
Karr, J.R. 1986, K.D. Fausch, P.L. Angermeier, P.R. Yant, and
I.J. Schlosser. Assessment of Biological Integrity in Running
Water: a Method and its Rationale. Illinois Natural History
Survey Special Publication No. 5. Champaign Illinois. 28 pages.
Karr, J.R. 1996. Rivers as Sentinels: Using the Biology of
Rivers to Guide Landscape Management in R.J. Naiman amd R.E.
Bilby (eds), Ecology and Management of Streams and Rivers in the
Pacific Northwest Coastal Ecoregion. Springer-Verlag, New York,
New York.
-------
Summary of all Savannah REMAP Stream Data
Stream Eco
Onto
StaflonlP CYCLE
4
8
9
10
11
12
13
14
15
19
21
22
27
20
29
30
31
32
33
34
37
38
39
41
44
45
48
49
51
57
61
84
85
88
89
71
72
74
75
77
71.
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
YEAR
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
94
95
95
95
95
95
95
95
95
95
1
3
2
3
3
3
1
3
1
3
3
2
3
2
2
3
2
3
1
3
3
3
3
2
1
2
3
3
1
3
2
2
2
2
1
1
1
2
2
1
1
Region I LaWute (QMS) I Longitude (QMS)
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
LP
SH
SH
SP
SP
BR
LP
LP
LPREF
LP
LP
LP
LP
LP
LPREF
LP
34
34
34
34
34
34
34
34
34
34
34
34
34
33
33
33
34
33
33
33
33
33
33
33
33
33
33
33
33
32
34
34
34
34
34
34
34
34
34
34
34
38 53.7993
28 48.5681
28 47.0579
25 49.8077
27 22.0911
25 3.2074
28 46.6378
28 42.8550
19 42.0003
14 48.3613
14 28.5278
14 38.4142
2 9.1629
58 4.9672
57 2.2850
58 35.4090
2 42.8822
58 2.0761
56 36.8479
54 54.1530
53 51.7284
55 38.8549
55 42.2439
50 6.3653
43 56.8046
41 49.2998
15 21.9660
15 19.3664
0 5.8158
56 4.1333
55 43.5823
47 48.7828
47 3.6141
38 17.2747
37 422113
30 41.9353
30 57.1580
20 37.3851
19 58.7644
20
10
82 41
83 2
83 3
83 2
83 21
83 2
83 20
83 21
83 2
82 56
82 55
82 56
82 3
81 54
81 53
81 54
82 50
81 54
82 14
82 1
82 1
82 52
82 53
82 39
81 54
82 43
82 20
82 19
81 24
81 48
63 7
82 37
82 36
83 19
83 18
82 27
83 28
82 39
82 38
83 2
n it
37.7301
40.6523
14.25531
28.9258
41.38151
40.6590
19.02411
27.46031
1.1208
14.6220
12.76441
46.4356
55.53021
45.53021
56.61451
24.9065
58.52921
52.44631
21.7438
9.6500
9.40291
49.86S0
46.58351
2.7338
24.7820
32.5372
1.51071
59.96911
2.8706
1.80281
5010261
51.87201
22.02551
55.03561
29.64891
584868|
26.7501 [
48.50951
39.84801
3.15631
8 MM1
AGPT
Ml HAB
Ml RICH
MLEPT
FISHJBI
PH
TEMP
(c)
DO
(mg/D
COND
(uS/cm)
3406
76
11
3
6.9
21.3
7.9
45.3
28.19
64
11
3
19
6.4
17.5
0.0
*52.0
36.17
94
12
4
25
6.5
23.0
0.0
600
33.82
80
11
4
21
6.4
21.5
7.2
53.2
11.84
47
12
3
25
6.8
21.0
0.0
55.0
40.28
83
16
5
27
6.0
22.0
0.0
50.0
18.63
61
9
3
33
6.5
19.5
0.0
50.0
10.12
45
10
3
29
6.4
21.0
0.0
55.0
41.33
87
7
2
23
5.4
21.0
00
45.0
30.48
91
13
5
21
5.8
23.0
0.0
35.0
38.43
67
14
5
6.9
21.0
7.9
33.8
33.87
93
15
5
23
5.5
24.0
0.0
35.0
8.28
64
11
3
17
7.0
22.3
4.7
211.0
3.12
92
12
2
27
6.5
227
5.1
117.2
1.24
119
24
7
37
72
22.8
7.1
96.7
8.62
116
16
5
33
7 2
22.6
6.0
104.0
8.16
67
7.0
20.9
8.1
780
4.23
112
14
5
36
7.4
22.7
7.9
101.8
8.42
75
15
4
21
7.2
22.3
7.2
243.0
10.18
113
20
6
39
7.0
22.4
5.9
93.2
23.09
111
21
7
27
7.1
22.9
7.5
91.4
4.13
88
12
5
23
6.8
23.5
7.5
76.5
82
16
2
31
6.7
23.2
6.1
76.7
3.25
48
16
4
35
6.7
25.3
62
110.4
3.46
75
9
3
6.9
23.4
7.6
67.1
59.51
45
7
1
7.5
27.4
5.9
3260.0
30.36
108
8
3
8.7
24.8
6.9
914.0
28.45
92
9
4
6.8
25.6
7.5
677.0
1.54
120
11
2
7.4
22.7
7.1
36.1
8.74
111
17
3
7.5
21.2
64
184.0
1.41
99
23
14
6.9
16.6
8.9
16.0
1.71
45
15
7
19
6.8
20.0
8.1
3.33
41
11
5
17
6.5
22.0
7.4
2.46
104
30
15
31
7.2
21.4
8.3
35.0
1.40
84
23
12
19
7.4
21.2
8.4
37.0
17.96
76
19
8
19
6.4
22.5
7.6
3.79
82
27
18
21
7.3
20.4
6.4
36.0
2.06
41
20
9
35
6.8
21.1
7.8
50.6
1.23
54
23
9
31
6.9
21.4
7.7
49.6
257
105
25
10
21
6.9
20.4
7.8
35.0
1 m
91
A
74
TQ
-------
Summary of all Savannah REMAP Stream Data
Stmm Eco
Station 10
CYCLE
YEAR
(Mar
Region
UMucfefOMS)
LonoRude (DMS)
Slate
79
1
96
3
LP
34
8
27.1407
82
17
14.9103
SC
ao
1
95
3
LP
34
8
11.3256
82
17
51.3748
SC
81
1
05
3
LP
34
7
43.1358
82
18
12.2856
SC
82
1
95
3
LP
34
8
6.7529
82
28
39.8788
SC
83
1
95
1
LP
34
12
15.4883
83
25
26.0343
Ga
85
1
95
1
LP
34
8
43.1670
82
57
23.1062
Ga
88
1
95
3
LP
34
5
6.1063
82
28
36.4788
SC
87
1
95
2
LP
34
8
18.6496
82
57
24.4861
6a
88
1
95
3
LP
34
4
32.6378
82
28
27.1207
SC
89
1
95
2
LP
34
4
38.9293
82
30
18.6980
SC
93
1
95
3
LPREF
33
48
16.0738
62
7
57.4444
SC
94
1
95
3
LPREF
33
48
7.1642
82
8
13.9907
SC
95
1
95
3
LP
33
47
59.2036
82
7
23.9716
SC
98
1
95
2
LP
33
48
51.1624
82
8
34.7175
SC
97
1
95
1
LP
33
52
1.2324
83
9
52.4866
Ga
98
1
95
1
LP
33
37
11.4077
82
22
33 5640
Ga
99
1
95
2
LP
33
35
15.4617
62
12
46.2592
Ga
100
1
95
3
LP
33
35
31.1643
82
41
13.5212
Ga
101
1
95
3
LP
33
35
28.7033
82
41
51.5903
Ga
102
1
95
3
LP
33
35
7.9421
82
42
12.1633
Ga
103
1
95
1
LP
33
34
54.0984
82
40
22.8367
Ga
104
1
95
1
LP
33
32
41.1923
82
39
50.4017
Ga
113
1
95
2
SH
33
15
55.8165
81
57
21.6129
Ga
114
I
95
f
SP
33
7
22.9653
81
51
2.9919
Ga
121
2
98
2
BR
34
49
59.4444
83
35
29.5255
Ga
122
2
98
2
LP
34
32
45.2830
83
18
11.5174
Ga
123
2
98
2
LP
34
10
27.1961
83
17
2.2760
Ga
127
2
98
1
UP
34
49
25.9638
82
58
42.8228
SC
130
2
98
1
LP
34
32
36.8223
82
58
0.7518
SC
131
2
98
2
LP
34
32
45.4489
82
57
22.3896
SC
132
2
98
2
LP
34
31
28.9133
82
57
22 5618
SC
133
2
98
3
LP
34
31
46.5814
82
57
5.6190
SC
135
2
96
2
LP
33
48
25.7746
62
58
59.2704
Ga
138
2
98
1
UP
34
54
4.4831
62
46
53.2661
SC
138
2
98
2
UP
34
55
12.6118
62
45
38.6099
SC
143
2
98
3
LP
34
39
20.7134
82
38
31.9222
SC
144
2
98
3
LP
34
38
22.9730
82
38
37.6329
SC
145
2
98
3
LP
34
38
50.4543
82
38
25.1805
SC
147
2
98
3
LP
34
39
40.2584
82
37
47.9485
SC
148
2
98
1
LP
33
37
37.8644
82
47
12.3996
Ga
149
2
98
2
LP
34
28
55 3141
82
37
390RIW
SC
AGPT
Ml HAB
Ml RICH
MLEPT
FISH IBI
PH
TEMP
icj
DO CONO
(wig/I) (uSfcm)
53
16
7
23
7.6
7.9
125.0
13.12
52
22
6
21
7.4
22.0
7.9
123.0
8.26
52
23
8
25
7.3
23.0
7.6
125.0
396
107
28
8
25
7.3
25.0
7.6
101.0
0.55
66
23
11
19
6.7
20.3
7.7
440
4.80
45
23
5
33
6.8
22.3
S.4
520
4.58
72
25
9
25
7.8
26.7
85
102.7
7.03
30
19
7
21
6.7
22.2
5.9
51.0
5.97
68
27
11
27
7.3
22.8
6.5
106.0'
4.55
16
4
27
7.2
25.1
4.8
150.0
104
26
10
43
7.6
26.0
8.2
146.9
59
17
8
39
1.34
99
32
10
39
9.1
28.2
11.3
146.5
2.06
103
22
10
23
7.4
21.5
7.2
1.46
40
21
11
6.8
21.7
6.3
62.0
69
30
12
25
6.4
22.2
3.6
76.0
8.57
72
23
8
31
7.1
25.2
5.2
105.0
4.97
49
21
8
29
7.2
26.4
6.2
107.0
11.06
46
20
8
27
7.2
26.6
6.2
104.0
0.82
49
15
6
27
7.3
26.3
5.9
103.0
1.34
89
22
6
35
7.6
24.4
7.1
187.0
3.07
70
12
8
37
7.0
22.7
7.9
91.1
7.37
104
26
10
6.9
24.4
6.7
32.0
6.66
78
14
6
6.5
20.9
6.3
37.0
89
22
14
6.9
16.5
7.9
24.0
53
16
5
13
7.2
24.2
8.2
691.0
66
21
6
27
6.8
22.3
7.3
52.0
112
32
16
7.0
22.0
8.2
40.0
54
16
1
25
6.8
26.0
7.2
50.0
62
22
8
33
6.8
24.5
8.6
400
48
19
6
25
7.1
26.5
7.5
45.0
62
21
7
23
6.8
26.0
8.0
500
75
22
7
31
6.9
24.1
5.4
142.0
102
19
11
6.3
19.5
8.5
200
82
26
16
7.0
22.0
8.2
30.0
93
22
10
21
7.1
24.0
80
500
51
24
10
19
7.1
25.0
7.5
500
47
26
If
19
6.9
22.0
8.0
45.0
48
19
10
21
7.1
25.5
8.1
45.0
55
24
9
2S
7.6
23.2
7.2
42.0
«
10
ft T
*v> r
y t
4 ft
-------
Summary of all Savannah REMAP Stream Data
8>r—11 Eco
YEAR Onht Raqton 1 LaWud»pMS| I LoncHhjde(PMS) | Stele AGPT
StaHonlD CYCLE
2
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
151
154
155
155.1
182
163
164
166
167
176
177
186
107
169
191
192
193
194
195
196
197
200
205
207
210
211
213
214
216
221
222
224
231
232
236
237
236
96
96
96
96
96
96
96
96
96
96
96
97
97
97
97
97
97
97
97
97
97
97
97
97
97
97
97
97
97
97
97
97
•7
97
97
97
97
JjJr
2
2
2
3
1
2
2
2
3
2
1
2
3
3
2
2
1
2
3
1
1
2
2
1
3
3
1
1
2
2
2
3
3
3
2
2
1
34
33
A
IB
LP
LP
LP
LP
LP I 33
SPREF I 33
SPREF I 33
SP REF I 33
SP I 33
MACP I 32
MACP I 32
BR I 35
BR I 35
BR 34
BR I 34
BR I 34
BR I 34
BR I 34
BR I 34
BR I 34
LP I 34
LP I 34
LP 34
LP I 34
LP I 34
LP I 34
LP I 34
LP I 34
LP 1 34
LP I 33
LP I 33
LP I 33
SHREF I 33
SH I 33
SP I 33
SP I 33
SP I 33
35
3
4
2
3
35
31
0
0
25 50.3552
46 13.7570
46 7.6710
49 17.0757
49.7966
1.0792
3.3065
57.2257
40.2696
19.1747
57.0026
31.3210
12.3063
59 46.4191
52 3.5167
51 41.4279
51 15.5672
51 6.1637
50 36.4122
46 55.1103
22 46.7363
17 23.0683
12 17.0808
14 14.2111
47.0635
23.5813
23.9805
48.7219
42*311
0.1361
22.6985
9
9
8
7
0
42
41
40 40.14991
26 43.80631
25 57.19611
5 35.60711
4 44.96321
8 15.75791
82 36
82 29
82 29
82 29
81 55
81 54
81 54
81 54
81 S3
81 26
61 27
82 49
82 49
82 49
83 8
83 9
83 9
83 8
83 9
83 14
82 54
82 43
82 50
83 22
83 3
83 6
63 5
83 17
82 22
82 2
82 0
82 36
81 38
81 38
81 31
81 30
81 47
54.0540
420120
41.8078
26.3316
35.8942
53.5055
42.1236
21.7342
41.3849
41.7862
18.6783
14.8715
37.5836
40.7735
10.6764
58.7073
0.6364
40.5458
12.9677
10.5392
2.9064
33.0550
58.5352
43.9620
59.4713
5.3189
18.1675
19.2083
52.18741
45.80111
43.3443
7.4409
4.64891
17.6399
9.3483
452is|
38.8681
SC
Ga
Ga
Ga
SC
Ga
Ga
Ga
Ga
Ga
Ga
SC
SC
SC
SC
SC
SC
SC
SC
SC
Ga
SC
Ga
Ga
Ga
Ga
Ga
Ga
SC
SC
SC
Ga
SC
SC
SC
SC
Ga
TEMP
DO
CONO
HAB
Ml RICH
MLEPT
FISHJBI
PH
(CI
(mpfl) (uStan)
51
16
5
23
6.8
23.0
7.5
130.0
%
68
27
11
27
6.9
23.6
7.0
820
71
32
10
29
6.9
23.2
6.8
77.0
72
20
6
27
7.1
24.7
7.4
112.0
95
16
9
15
5.1
20.0
7.7
15.0
102
26
7
7.3
24.0
6.6
128.0
08
24
7
7.3
23.3
7.1
119.0
104
27
11
6.8
22.2
8.3
86.0
99
22
7
7.2
24.6
8.2
100.0
99
14
1
6.0
25.8
6.9
60.0
96
14
1
4.1
25.6
5.1
58.0
123
25
18
7.0
23.7
8.0
24.0
91
36
20
7.4
22.3
8.5
29.0
87
35
19
7.6
22.4
8.5
27.0
92
32
20
7.4
19.3
8.7
240
91
29
17
6.8
19.4
8.7
26.0
60
13
5
6.9
18.9
9.5
32.0
82
31
17
7.1
19.8
9.3
24.0
110
30
18
6.6
21.9
8.7
24.0
56
20
8
6.9
19.5
8.7
21.0
52
16
6
21
6.2
21.9
7.2
29.0
56
24
11
19
7.0
24.0
7.3
650
63
17
5
23
8.7
21.5
8.2
32.0
19
6
33
103
18
9
23
7.2
22.7
9.0
40.0
19
7
25
57
24
12
21
6.9
21.0
7.9
541.0
74
15
7
27
6.5
19.8
8.0
379.0
46
21
9
17
6.8
22.8
8.1
111 0
67
21
12
23
7.0
25.3
7.5
80.0
66
18
6
25
8.9
26.4
6.4
121.0
24
11
35
92
24
11
5.2
20.9
7.9
19.0
95
21
11
5.3
21.2
74
180
73
19
6
6.1
25.4
6.4
38.0
88
21
9
6.4
25.5
7.3
47.0
16
7
BR* BM&tL LP* lamr PladbonL UP* UBOar PladtnonL Sand HBa. 8P* SauHmaalam PUna. UUCP* MkUUartBc Caatlal Platom. BEE* Rato***
-------
STATIONS
ORGANISMS 4 8 9 10 11 12 13 14 15 19 21 22 27 28 29 30 32 33 34 37
DIPTERA
Tabanidae
X XXX
Empididae
TipuHdae
X
Tanyderidae
Rhagionklae
Cuncidae
X
Ptychopteridae
Dbridae
Ceratopogonidae
X
Chironofnidae
XXXXXXXXXXXX XX XXX
Blephariceridae
Simuliidae
X XXX XX X xxxx
TRICHOPTERA
Hydroplilklae
HelicopsychkJae
Hydropsychkfae
XX xxx x xxxxxx
RhyacophilkJae
UmnepftWdae
x
Phryganekfae
Potycentropidae
xxx XX
Psychomttdae
Philopotamidae
X X XX
Dipseuopsldae
Brachycemridae
Goeridae
LepkJostomatkJae
Calamoceratkfae
Uenoidae
Molannidae
Odontoceridae
Leptoceridae
GlossosomaMdae
PLECOPTERA
Periidae
x xxxx
Perfodidae
XX x
Peltopertidae
Capnikfae
Leuctridae
-------
ORGANISMS
STATIONS
4 8 0
10 11 12 13 14 15 19 21 22 27 28 29 30 32 33 34 37
Nemouridae
Pteronaicykfae
EPHEMEROPTERA
Baetidae
Tricorythidae
Heptagenidae
OHgoneuriidae
LeptopWetottdae
Caenidae
Neoephemeridae
Ephemeridae
EphemeroMKJae
X X X X
X X X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
OOONATA
UbeNulidae
Confuliidae
Cordulegasteridae
Gomphidae
Aeschnidae
MacromNdae
Calopterygidae
Coenagrionidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
XXX
X X
XXX
MEGALOPTERA
CorydaHdae
SiaWdae
NEUROPTERA
Steyridae
HEMIPTERA
Corixidae
VelUdae
BakMtoinalidae
Nepklae
Gewfclae
COLEOPTERA
Elmidae
Hyqropnmgpe
GyrinMae
Dytiscidae
NoterMae
DryopMaa
X
X
X
X
X
X
X X X X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
-------
STATIONS
ORGANISMS 4 8 9 10 11 12 13 14 15 19 21 22 27 28 29 30 32 33 34 37
Psephenkfae
X X
> l^« H-l
1 IOKMjNjM
X XX
HalipHdae
Eubriidae
PtUodadvHdae
CRUSTACEA
Astaddae
XXX XXXX XXXXXX XXX
Isopoda
AmpNpoda
XX X
Palaemonidae
OLIGOCHAETA
Glossosophoniidae
Naidkla«
Tubificklae
X XX
Lumbriculklae
X X
HIRUDINEA
X X
HYDRACARINA
XXX XX
MOLLUSCA
Bivalvia
Gastropoda undet. sp.
Unionidae
XX X
CorttculkJae
XX X
Lymnaekfae
Sphaeriidae
X
Physidae
Viviparidae
RanofMdaa
X X
Pleurocertdae
Ancytktaa
TAXA RICHNESS
11
11
12
11
12
16
9
10
7
13
14
15
11
12
24
16
14
15
20
21
EPT INDEX
3
3
4
4
3
5
3
3
2
5
5
5
3
2
7
5
5
4
6
7
HABITAT SCORE
78
84
94
80
47
83
61
45
87
91
67
93
64
92
119
116
112
75
113
111
-------
ORGANISMS
DIPT6RA
TabaoidM
EmpMidae
TipuHdae
Tanyderidae
Rhagkmidae
CuHddae
Ptychopterfdae
Dixidae
Ceratopogonkiae
ChironomJdae
Btepftariceridae
SfarnuWdae
38 30 41 44 45 48 49 51 57 01 64 85 68 89 71 72 74 75 77 78
X
X
X
X
X
X
X X
X
XXX
X
X
X
X X X X
X X X X X X
X X X X X X
X
X
X
X
X
TRICHOPTERA
HydroptilMae
HeNcopsychidae
Hydropsychidae
Rhyacophflidae
UmnephHidae
PhryganeMae
Polycentropidae
PsychomHdae
PhUopotamidae
Dipseuopsidae
Brachyoenlridaa
Goeridae
Lepidoslomatidae
Calamooeralidae
(Jenoidae
MolannkSae
Odontoceridae
Leptooaridae
Gtossosomaltdaa
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
PLECOPTERA
Periidae
Pertodidae
PeKopertidae
CapnNdae
Leuctridae
X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
-------
ORGANISMS
38
39
41
44
45
48
49
51
57
61
64
65
68
69
71
72
74
75
77
78
Netnouridae
Pteronarcyidae
X
X
X
X
X
EPHEMEROPTERA
Baettdae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
TricocytWdae
Heptagenidae
CXigoneurUdae
LeptophieMdae
Caenidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Neoephemeridae
Ephemerkfae
Ephemerellidae
X
X
X
X
X
X
ODONATA
Ubeituildae
X
Cordutiklae
X
Cordulegasteridae
Gomphidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Aeschnidae
Macromiidae
X
X
X
X
X
X
X
X
X
X
CaloptorygkJae
Coenagrtonklaa
X
X
X
X
X
X
X
X
X
X
X
X
X
X
MEGALOPTERA
CotydaHdae
Sialidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
NEUROPTERA
Sisyridae
HEMIPTERA
Corbddae
VelWae
BetostomaHdac
X
X
X
X
X
X
X
X
Nepidae
Gerridae
X
X
X
X
X
COLEOPTERA
Elmklae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
HydrophHidM
Gyrinidae
Oyliscidae
Noleridae
X
X
X
X
X
X
X
X
X
X
X
X
Dryopkfae
X
X
X
X
X
X
-------
ORGANISMS 38 39 41 44 45 48 49 51 57 61 64 65 68 69 71 72 74 75 77 78
Pseptonidae
Heiodidae
HaHpiidae
Eubriidae
PtHQdactvNdae
XXX X
X X
X
CRUSTACEA
Astaddaa
isopoda
AmpNpoda
Palaemonidae
XX XXX XXXXXX
X X
XX X
X X
OUGOCHAETA
Glossasophonfidae
NaJdkJae
TubHiddaa
Lumbriculttae
X X
X X
XX X
HIRUDINEA
HYDRACARINA
MOLLUSCA
Bivatvia
Gastropoda undei. sp.
Unionidae
CorMcuHdae
Lymnaeidae
Sphaeriidae
Physidae
Viviparidae
Planorbidae
Pleuiocecldae
AncyNdae
X X
X X
X X
X
X
X
X
TAXA RICHNESS
12
16
16
9
7
8
9
11
17
23
15
11
30
23
19
27
20
23
25
11
EPT INDEX
5
2
4
3
1
3
4
2
3
14
7
5
15
12
8
18
9
9
10
4
HABITAT SCORE
88
82
48
75
45
108
92
120
111
99
45
41
104
84
78
82
41
54
105
91
-------
ORGANISMS
DIPTERA
79 80 61 82 83 85 86 67 88 69 93 94 95 96 97 98 99 100 101 102
Tabanidae
Empkfidae
HpuRdae
Tanydertdae
Rhagionklae
Culicidae
Ptychopteridae
Dbddae
Ceratopogonidae
Chkonomkfae
Blephariceridae
Simuliidae
X
X X
X
X
X
X
X
X
X
X
X
XXX
X X X X
X X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X X
X X
TRICHOPTERA
Mydroptilidae
Helicopsychidae
Hydropsychkfae
Rhyacophilidae
Umnephttidae
Phryganektae
Polycentropidae
PsychomMdae
Philopotamklae
Oipseuopsidae
Brachycentridae
Goeridaa
LepidostomaHdaa
Caiamooeralidae
Uenoklae
Molannidae
Odontoceridae
Leptocerfcfae
GlossosomaHdae
X
X
XX XXXXXX
XX X XXX
X
X
X
X
XXX
XXXXXXXXX
X X
X
XXX XXX
X X
X
X
X
XXX
PLECOPTERA
Perlidae
Perlodklae
PeKopertidae
Capniidaa
Leuctridae
X XXX
X X
X
X
X
X
X
X
X
X X
X
-------
ORGANISMS 78 80 81 82 83 85 86 87 88 89 93 94 95 96 97 98 99 100 101—102
vmwni^fna
Nemouridae
Pteronarcvidae
X X X X _____
EPf^MEROPTERA
Baetkfae
TricorythkJae
Heptagenidae
ONgoneuriidae
LeptopMabiktoe
CaerNdae
Neoephemeridae
EphernerMae
EohemerefHdaa
xxxxx X xxxxxx xxxx
x xxxx x x
xxxxxxxxx xxxx xxxxx
xxxxx X XXXXXX XX
x X XXXXXX XX
X xxxxxxx x XXX
x X XX X
OOONATA
UbeUuHdae
Cordutikiaa
Cordufegasteridao
Gomphidae
AeschnkJae
MacromNdae
CalopterygMaa
Coenaarionkfa*
X x XX x
X X XX XX XX
XXXXXXXXXXXXXX XX X
xxxxxxxxx X XXXXXX X
XXX XXXXX x
xx xxx x X X xxxx
XX xxxxx X XX X
MEGALOPTERA
Corydalidae
XXXX XXX X XXX X 5 * J J J
XXXX XXXXXX X x XXX
NEUROPTERA
Sisyrtdae
X X
HEMIPTERA
Corbddae
VeNMaa
Beiostomatidae
Nepkfae
Gentdae
XXXXX XXX
X
COLEOPTERA
Elmidae
¦ L -i
riyoropniMM
Gyrinidaa
DyttsckJaa
r>OI9nON
DryopWae
XXXXXXXXX xxxxxxxxxx
XX X XX
XXX X x XX
XXXXXX
X XX X X XXXX x X
-------
ORGANISMS 79 80 81 82 83 8S 86 87 88 89 93 94 95 96 97 98 99 100 101 102
Psephenfctae
Hetodklae
HaNpKdae
EubrHdae
PtHodactvMae
X
X XX
CRUSTACEA
Astaddae
Isopoda
Amphipoda
Palaemonidae
X XXXX XX XXXX
X
X
X X
OLIGOCHAETA
GlossosophoniWae
Naididae
Tubiflckfae
Lumbriculidae
X XX
X
HIRUDINEA
HYDRACARINA
MOLLUSCA
Bivatvia
Gastropoda undat. sp.
Unkmidae
Corbiculidae
Lymnaeidae
Sphaerfklae
Physkfae
VMparidae
Planorbkfae
Pteunwerfdae
AncvHdae
X
X
X X
XX X X XXXXXX XXXX
X
X X
X XX
XXX
XXX XX
X
TAXA RICHNESS
16
22
23
28
23
23
25
19
27
18
26
17
32
22
21
30
23
21
20
15
EPT INDEX
7
8
8
8
11
5
9
7
11
4
10
8
10
10
11
12
8
6
6
6
HABITAT SCORE
53
52
52
107
66
45
72
30
68
N/A
104
59
99
103
40
69
72
49
48
49
-------
ORGANISMS
103 104 113 114 121 122 123 127 130 131 132 133 135 136 138 143 144 145 147 148
X
XXX XXXX X X
DIPTERA
TabankJae
EmpMidae
Tipulidae
Tanyderidae
Rhagionfdae
CulkAfae
Ptychopteridae
Dbddae
CeratopogonkJae
CNronomidae
Bfophariceridae
SimulikJae
X X X X X
X
X XX
XXX
X X
X
X
X
X
X
X
X
X
X
X
X
TRICHOPTERA
nyaropMKMW
HeltcopsychkJae
Hydropsychidae
RhyaoophMidae
Limnephilkfae
Plwyganeidae
PotycentropkJae
PsychomUdae
Phitopotamidae
DipsetN)psidae
BractiycenlrWae
Goeridae
LepMostomatklae
Calamoceratidae
Uenoktee
Molannidae
Odontoceridae
Leptoceridae
Glossosomatidae
X
X
X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X X X X X
X
X
X
X
X
XXXX
X X X X X X
PLECOPTERA
Perttdae
Perfodidae
CapnNdae
LeucMdae
X X
X
X
X
X
X
X
X
XXX
X
X X X X X X X
X X
X X
-------
ORGANISMS
103
104
113
114
121
122
123
127
130
131
132
133
135
136
138
143
144
145
147
148
Nemouridae
X
Pteronarcyidae
X
X
X
X
X
X
EPHEMEROPTERA
BaetMae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Tricorythktoe
X
Iteptagenidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
ONgonourMae
X
X
X
X
X
X
X
X
X
X
X
LeptophlebHdae
X
X
X
X
Caenidae
X
X
X
Neoephemeridae
X
Ephemeridae
X
X
X
X
X
X
X
X
X
X
X
Ephemerellidae
X
X
X
X
X
X
X
X
X
X
X
X
OOONATA
UbelhiKdae
X
X
X
X
Confuliidaa
X
X
X
Cordulegastaridae
X
Gomphidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Aeschnidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Macromiidae
X
X
X
X
Calopterygidae
X
X
X
X
X
X
X
X
X
X
X
X
Coetiaorionidae
X
X
X
X
X
MEGALOPTERA
Corydalklae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Sialidae
X
X
X
X
X
X
X
X
X
NEUROPTERA
Sisyridae
HEMIPTERA
X
X
Corbddae
X
X
VeliWae
X
X
X
X
X
X
BelostomaUdae
X
Nepidae
Gerridae
X
X
X
X
X
X
X
COLEOPTERA
Elmklae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
HydrophHkJae
X
X
X
X
X
X
X
Gyrinidae
X
X
X
X
X
X
X
X
X
X
Dytisddae
X
X
X
Noteridae
DryopWae
X
X
X
X
X
X
X
X
X
X
-------
ORGANISMS 103 104 113 114 121 122 123 127 130 131 132 133 135 136 138 143 144 145 147 148
Psephenkfae
X
Hekxfkfae
HaHpNdae
X X
Eubriktee
PWodactvNdae
X
CRUSTACEA
Astacidae
X X XXXX XXX XXXXXX
Isopoda
X X
AmpNpoda
X
PaiaemonMae
X
OUGOCHAETA
X XX
GtossosophonHdae
UgMMaa
Tubffiddae
LumbricuHdaa
HIRUDINEA
X
HYDRACARINA
X
MOLLUSGA
Bivalvia
X
Gastropoda undet. sp.
X
Unionidae
CoiMcuNdae
X X
Lymnaektoe
Sphaeriidac
Physidae
X
VMparidaa
Planorbidae
X
Pteurooartdae
XX x
InnHdaa
MvCMQM
TAXA RICHNESS
22
12
28
14
22
18
21
32
16
22
19
21
22
19
26
22
27
26
19
24
EPT INDEX
8
8
10
8
15
5
8
18
1
8
8
7
7
11
16
10
10
11
10
9
HABITAT SCORE
89
70
104
78
89
54
88
112
54
62
48
62
97
102
82
93
51
47
48
55
-------
ORGANISMS
DIPTERA
Tabantdae
EmpkMae
Tipulidae
Tanyderidae
RhagionkJae
Culiddae
PtychopterkJae
Dixidae
CeratopogonkJae
ChironomMae
Blephariceridae
SimuHidae
149 1S1 154 1S5 162 163 164 166 167 178 177 186 187 169 191 192 193 194 19S 196
X XXX X
XXXXXX XXXXXX XXX
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
TRICHOPTERA
HydroptiNdae
Helicopsychkfae
Hydropsycftidae
Rhyacophilktae
UmnephHkfae
Phryganeidae
Polycentiopidae
PsychomHdae
PhilopotamMae
DipseuopsWae
Bractiycentridae
Goefktae
Lepklostomatidae
Calamoceratidae
Uenoidae
Molannidae
Odontoceridae
Leptoceridae
Glossosomatidaa
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X X
X
X
X X
X
X X
X
X
X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
PLECOPTERA
Periidae
Periodidae
Peltoperlidae
Capnikfae
Leuctridae
X
X
X
X
X
X X X X XXX
X X X X XX
xxxxxxxx
-------
ORGANISMS
Nemouridae
Pteronarcyfclae
EPHEMEROPTERA
Baetidae
TricorytMdae
Heptagenidae
ONgoneurNdae
Leplophlebfldao
CaenkJae
Neoephemerfdae
Ephemeridae
Ephtunereffldae
149 151 154 155 162 163 164 166 167 176 177 186 167 169 191 192 193 194 195 196
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
OOONATA
LHwHuiidaa
CorcMHdae
Confulegasteridae
Gomphidae
Aeschnidae
Mac.omHdae
Calopterygidae
Coenagrtontdae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
MEGALOPTERA
CorydaNdae
SiaHdae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
NEUROPTERA
HEMIPTERA
X
X
X
Corbddae
VeMdae
X
X
X
X
X
Betostomattdae
Nepidae
Gentdae
I X
X
X
COLEOPTERA
ElmkJae
HydropMNdae
Gyrinidae
Dytiscidae
Noteridae
Dryopidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
-------
ORGANISMS
149
151
154
155
162
163
164
166
167
176
177
186
187
189
191
192
193
194
195
196
Psephenidae
Helodidae
HaBpHdae
Eubriidae
PtHodactyNdae
X
X
X
X
X
X
X
X
X
CRUSTACEA
Astacklae
Isopoda
Amphipoda
PaJaemonidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
OLIGOCHAETA
GlossosophonHdae
Naididae
Tubificidae
Lumbriculidae
X
X
X
X
X
X
X
X
HIRUDINEA
HYDRACARINA
MOLLUSCA
Bivalvia
Gastropoda undet. sp.
Unionidae
CorWcuNdae
Lymnaekfae
Sphaeriidae
Physidae
Viviparidae
PlanorMdae
Pleurocoridae
AncvNdae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
TAXA RICHNESS
EPT INDEX
HABITAT SCORE
15
6
52
18
5
51
28
11
68
32
11
71
16
9
95
27
7
102
22
7
96
27
11
104
23
7
99
24
6
99
14
1
96
25
16
123
36
20
91
35
19
87
32
20
92
29
17
91
13
7
60
31
17
82
30
18
110
20
8
58
-------
ORGANISMS
DIPTERA
TabanMae
EmpkMae
Tlpufldae
Tanyderidae
Rhagionidae
CuNcMae
Ptychopteridae
Dtoddae
Ceratopogonidae
CNrononMae
Btephariceridae
SfcnuHktae
197 200 205 207 210 211 213 214 216 221 222 224 231 232 236 237 238 155a
X
X
X
X
TRfCHOPTERA
HydroptiUdae
HeMoopsycMdae
Hydropsychidae
RhyacophiMae
UnmephNidae
Phryganeidae
PolyoentropfcJae
PsychomUdae
PhUopotamkJae
OiK^euopsidaa
Bractiycantridae
Goeridae
Lepktostomatidae
Catamocaratkfae
Uenoidae
Motannidae
Odontoceridae
Leptoceridae
GlossoaomaHdae
PLECOPTERA
PerMaa
Periodidaa
Peltopertkfae
CapnNdae
Leuctrfdae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
-------
ORGANISMS
1f7
200
205
207
210
211
213
214
216
221
222
224
231
232
236
237
238
155a
Nemouridae
PteronarcvkJa®
X
EPHEMEROPTERA
Baetidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Tricofythklae
X
X
X
X
X
X
X
X
X
X
Heptagenidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
OligoneurHdae
X
X
X
X
X
X
X
Leptophlebiidae
X
X
X
Caenidae
X
X
X
X
X
X
X
X
Neoephemeridae
Ephemeridae
X
X
X
X
X
Ephemerellidae
X
X
X
X
X
X
X
X
ODONATA
UbellulkJae
X
Cofduliidae
X
Coidulegasteridae
X
X
X
Gomphidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Aeschnidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Macromifdae
X
X
X
X
X
X
Calopterygidae
X
X
X
X
X
X
X
X
X
X
X
X
CoenagrionkJae
X
X
X
X
X
MEGALOPTERA
CorydaNdae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Sialkfae
X
X
X
X
NEUROPTERA
Sisyridae
HEMIPTERA
X
Corixidae
X
VelikJae
X
X
X
Belostomatidae
Nepidae
Gerridae
X
X
COLEOPTERA
Elmidae
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Hydrophilktae
X
X
X
X
Gyrinidae
X
X
X
X
X
X
X
X
X
X
Dytisddae
X
Noterklae
Dryopidae
X
X
X
X
X
X
-------
ORGANISMS
197 200 205 207 210 211 213 214 216 221 222 224 231 232 238 237 238 155a
Psephenidae
X X
HelocNdM
HalipNdae
X
EubrNdae
PtHodactvNdae
X
X
X
CRUSTACEA
Astaddae
X XX XXX xxxxxxx
Isopoda
AmpMpoda
PalaemonMae
X XX
OLIGOCHAETA
GlossosophonHdae
NakHdae
Tubificidae
LumbricuHdae
X
HIRUDINEA
X
HYDRACARINA
MOLLUSCA
Bivalvia
Gastropoda undet. sp.
X X
Union idae
X X
CorbicuHdae
X XX XXXXX
Lymnaeidae
X
Sphaeriidae
Physidae
Vtvtparidae
X X
PianortMae
Pteuroceridae
AncvH^e
TAXA RICHNESS
18
24
17
19
18
19
24
15
21
21
18
24
24
21
19
21
18
20
EPT INDEX
6
11
5
8
9
7
12
7
9
12
8
11
11
11
8
9
7
8
HABITAT SCORE
52
58
83
fflA
103
N/A
57
74
48
67
66
N/A
92
95
73
88
N/A
71
-------
Savannah River REMAP Fish Collection
Common Kent 193 196 121 191 l«6 192 194 61
American Eel
Pirate Pcrcfa
Crack Cbubiudccr
Northern bopucker
Spoaad Suck*
Striped juaprock
Silver radhene
Flier
Blue qtoOed nmfish
Redbreut Sunfisb 2
Green Sunfah
Wamouth 1
Bluefill
Pumpkinseed
Lanfeeriunfixfa
Redear lunfisb
spotted suafoh
Redeye Beit
Urgtmouth Bus j
Black Crappie
Mooted Sculpts 10 2
Whilefin Shiner
Silvery Minnow
Roeyface Chub
Blucbcad Chub 1 4 1
Golden Shiner
Highfin Shiner I
Spooail Shiner
Yellowfin Shiner 73 21 7 16 11 18
Sandbar Shiner
Creek Chub 22 1 2 6
Chain Pickerel
Redfin Pickerel
Yellow Bullhead
Brown Bullhead
Black Bullhead
Snail Bullhead
Margined Madtom
Tadpole Madum
Speckled Madum
Flal Bullhead
Savannah Darter
turquioei darter 1
Twaillited Dwter 1
Blackhanded darter
Yellow perch
p.mKow trout 7
Toul 73 44 17 » 24 16 22 7
Stream Order 1 1 2 2 2 2 2 2
Number of Specie* 13233432
Eoorefioo BR BR BR BR BR BR BR BR
Page 1
-------
Savannah River REMAP Pish Collection
Common N«n«
195
1S7
189
162
13
197
15
33
American Eel
Pinte Perch
Creek Chubeucker
Northern hopucker
Spodad Sucker
Striped junprock
Silver radiane
Flier
Radbnut Sunfish
Ona Suofixfa
WamouSh
9
21
Redeye Be*
larfcnxxith Baa
Black Qrappie
MeBled Sculpin
WhAefin Shiner
Silvery Minnow
Roeyfac* Chub
Btuehead Chub
Golden Shiner
Hi|WiD Shiner
SpoOail Shiner
Yeliawfin Shiner
IS
25
14
33
25
181
15
Creek Chub
Chain Pidcerd
Redfin Pickerel
Yellow Bullhead
Brown Bullhead
Black Bullhead
Snail Bullhead
Margined Maduxn
Tadpole Maduan
Speckled Madtom
FlatBuUheert
31
47
101
35
11
19
43
ChriitnM defter
TeeMlietad Darter
dsrtcr
Yellow perch
Rainbow trout
1
Total
Stream Order
Number cfSpeciee
Eooregion
24
3
4
BR
51
3
»
BR
31
3
10
BR
33
1
2
LP
272
I
II
LP
31
1
5
LP
196
1
6
LP
75
1
S
LP
Page 2
-------
130
10
4
S
1
1
1
1
26
1
7
LP
Savannah River REMAP Fish Collection
207 72 213 214 103 83
83
1
35
IS
23
81
2
14
24
90
8
60
26
62
39
2
70
112
17
19
10
10
IS
153
1
8
LP
101
1
3
LP
104
1
3
LP
86
1
5
LP
233
1
10
LP
62
I
II
LP
49
1
3
LP
Page 3
-------
Savannah River REMAP Fish Collection
78 69 71 104 98 T1 68
American Ed
Piralc Perch 1 2
Crack Chuteidccr 2
Northern hopudier '
SpOQtd
Striped juaprock 3
SilwiiAgm
Flier
Bhif ^Mttsd iunfiih
Radbnaft Sunfnh ' 3 3
QmSmttt 3
Wamouth
BhiegQI 9 1 1
Radear suofidi 1
H"«^ ¦¦«**
Radeyt Ban 2
tarfanoutb Bas 3
Blade Crappie
MoBlad Scutpin
WUufin Shiner
Silvery Mianow
Roeyfaee Chub 1 3 14
Bluchead Chub 35 19 8 ] 9 41 35 35
ShtMT ]
Higtxfin Shiner
Spofliil Shiner
YcUowfin Shiner 31 66 18 17 j9 72 99 22
CmkCfaub 8 19 23 125 20 20 34 1
Cham Pickerel
Redfin Pickerel
Yellow Bullhead 2
Brown Bullhead 1
Blade Bullhead
s«»«ii P"fBr*<<
Mwyhmf
Tadpole Maduxn
Spwtiod Uidun
Flat Bullhead
T—nmd Darter 1 j
Blackbaaded darter j
Yeflowprcfc j
Tottl 7< "9 49 136 69 140 171 92
*WM«Qrt«r 1 1 1 1 1 1 1 2
Number of Specie* s'36|749
lplplplplplplplp
Page 4
-------
Savannah River R*=MAP Fish Collection
Common Name ' 65 64 79 9 22 28 41 74
AiMricaaEel
PmtePcrch 4
CrMic Chubaucktr 3 2 13
Northern hopudccr 3 2 2 It
Spotted Suck*
Striped junprock j
Sitvwradhont
Fliar
Ulltf ¦wfiJ>
1 2* 3 10 38 10
3 21 13
W-»outh , 6
Bh-gOJ <1 13 3 2
7 2 4
tpoO^d wnfish
Redeye Baa 3
largMBOiilh Baa 3 2 1
Black Grappie
MoOJad Scutpin
Whitcfm Shiner
Silvery Mmnow
Jtoayfaee Cbub J j 2 12 5
Bluahaad Chub 16 27 61 1 19 6 34 34
Ooldan Sfaiaar
Highfm ShiiMr jg
1 7
12
2
Spooail Shiner
Yallowfin Shiner 9 43 46 5 ig 54 66
Sandbar Shiner
Greek Chub 25 36 3 2 IS 7 »
Cham Pickerel
Redfin Pick vel
Yellow Bullhaad 4 j j
Brown Bullhead
Black Bullhead
Snail Bullhead
Margined Maduan 1
Tadpole Madtam
SptckM
Flat Bullhaad
SavaaoahDaitar
TaeerilHeri DHar 4 1
Bladcbanded darter , 2 3
Yellow pan*
1 J 9
Fairiww tiout
Tottl 50 107 206 13 » 57 ISO 157
Stream Order 22222222
Number of Spades 3 4 II 7 12 11 16 14
Eoorepon LPUULPlPLPLPLP
Page 5
-------
Savannah River REMAP Fish Collection
Comma Name* 87 75 89 96 222 135 154 Hi
American Eel
Pirate Perth 4 1
Cmic Cfaufacucker 1 1
Northern hopucker 18 2 3 1 3
Spotted Sucker 2
Stripadjumprock *
Stiver ratxne 1
Flier
Blue ipattod wofbh
Redbraa Suofieli 12 9 1 28 7 9
OrMnSaafidi 4 8 10 4 7
WMBOUft 1 I $ 3
BiutpD « 9 21 7
spooad mfiih
Redeye Ba« 2
IstgenemhBm 112
Black Cnppie
Mooted Sculpin
Whnefin Shner 2
Silvery Miaww
Rosy&ceOaab 7 4 1
Bluehead Chub 2 64 13 88 8 18 35
Qolden Shiner
Higb&o Shiscr
SpotUil Shiner 4 4 29
YeUowfin Shiner 67 47 117 it ]6 9 137
Sandbar Shiner
Creek Chub 1 19 5 16 14 20
Cham Pickerel
Redfis Pickerel
Yellow Bullhead 1 1
Browc Bullhead
pi-<4, nmihwii
S&aflBuUbead 1
MarpnadMadum 3 4
Tadpot* kladtaiB
Spacklad MadUnii
Scvmah Defter
Tueillrtit Darter
Bbokbaoded daftar
YilMpnb
RiiuUrw liuut
Totol 23 192 102 244 4* 97 63 238
Swmb Order 22222222
Number of Special 6 10 12 7 t 12 13 •
Eoarapaa IPLPLPLPlPLPLPLP
Page 6
-------
Savannah River REMAP Fish Collection
221 135 222 122 JiL
Cooiibop N«n* 99 131 133
American Ed 2
PimePerch 1
Creek Chubeucker *
Northern bopuckcr 6 1 2
Spotted Sucker
Striped junprodc
Silver redtone
Flier
Blue ipaaed eunfieh
r«l»««H Thinfirti 17 3 6 1 1'
Otmd Suafixfa 13 2
Wamouth 1
BbMgtU 1° 35
Pwapkneeed
LoofMrfuaStb
FiiIt—*-*- 5
jotted eunfish
Redeye Ba» 2
largtmouth Bus 1 1
Black Crappit
Mottled Scutpin
Whrufic Shiner 1 1
Silvery Minnow
Rocyfaoe Chub 21 5 6
BluabeadChub 14 10 24 1
Golden Shiner 2«
Highfin Shiner
Spottail Shiner 11 3 2 42
25 72
YeUowfin Shiner 4 2 60 31 31
Sandbar Shiner
OmIc Chub : 6 22 S 13
Chain Pidcerel
Hedfin Pidterd
VdtowBuUheail 1 2 2
Brown Bullhead 2
Black Bullhead 1
Snail Bullhead
Margined Madun 2
Tadpote MaAam 1
Speckled Madun
FlatDuHheeri 1
7
turqukM darter '
Tweellled Darter 5
Backhanded darter 10
Yellow perth
19
47
140
79
107
2
2
2
2
IS
t
7
12
LP
LP
LP
LP
Page 7
-------
102
1
1
9
6
I
3
12
1
3
3
2
I
43
3
12
LP
Savannah River REMAP Fish Collection
122 216 205 «1 224 §2 86_
1 1
1 7 5 5 12
4
2
3 4 9 6
7 7 14
1
6 9
3
3 10 6
12 62 « 1« 54 SS
1
1 6 5
1 S2 53 16 66 98
1 3 1 6
4 1
2 1 12
1 3
4 3 6 M
1
13
2
2
LP
14
2
2
LP
133
2
9
IP
169
3
15
LP
S3
3
II
IP
173
3
12
LP
232
3
11
LP
PageB
-------
Savannah River REMAP Fish Collection
Common Name
101
94
143
144
88
147
79
100
American Eel
Piraic Perch
1
1
Owk nmhrvrfriT
Northern hopucker
5
23
9
3
2
3
6
SpoOad Sucker
2
ftnpidjiiBpmt
7
1
Silver redhone
2
Flier
1
Bhte^oOediunfuh
Ridbeait Sunfiih
7
IS
2
42
8
8
8
GnenSunfbh
3
3
13
12
Wmauth
1
3
1
1
Bluagill
1
1
1
11
PuopkioMsd
2
LeofMrwnfuh
RidiirNflfidi
1
gpCf&td
RadeyeBam
2
largcmouth Baa
Black Crappie
Mottled Sculpin
Whitcfin Shiner
Silvery Minnow
3
1
Rtxyfacc Chub
6
23
10
BluebaadCbub
38
33
24
12
34
33
89
11
Oolden Shiner
Hifhfin Shiner
SpoOail Shiner
9
22
4
YeUowfin Shmcr
f awtllwi ClMMr
46
31
28
10
34
26
78
19
MRODV Pnuww
Creek Chub
4
2
19
Chain Pickerel
1
Redfin Pickerel
Yellow Bullhead
2
3
Brown Bullhead
BUck Bullhead
Snail Bullhead
2
1
Margined Madura
2
4
1
1
3
1
Tadpole Madum
Speckled Madum
1
1
Fbt Butted
2
2
2
1
turquioae darter
3
2
Teaee 11 alert Darter
9
3
6
Blacfchended defter
10
30
4
3
8
4
8
Yellow perch
Rainbow trout
Total
133
232
73
32
1(3
87
236
71
Stream Order
3
3
3
3
3
3
3
3
Number of Specia
13
16
7
I
14
11
13
10
Eoongian
LP
LP
LP
LP
LP
LP
LP
LP
Page9
-------
Savannah River REMAP Fish Collection
Common Nunc 93 95 80 135A 39 38 211 210_
American Ed
Pirate Perch 3 1
Crack Owffrfficktr
Northern bopucker 23 7 9 1 1 2
Spooad Sucker
Sui^ juu4.ua 3 2 2 3
Sihwradhorae 3 1
Flier
BkM^NOedwafitb
5 16 1 16 17 11 9 3
Otmo Stmfah 1 7 3 7 4 3 1
Wamoutb 1 3 2
1 4 11 3 2 1
Radcyt But 3 2 1 12
tergsnouth Baai 4
Black Crappie
Mooted Sculpin
Whiufia Sfamcr 2
Silvery Minnow
Roayfaoe Chub 6 9 4 2 2 2
Bluebead Chub 47 11 76 7 20 3 18 42
Ookbn Shiner 1
Higbfia Shiner
Spotuil Shiner 35 6 2
YtUowfinShmer 94 13 73 4 3 3 11 9
Sandbar Shiner 3 3
Crack Chub 12 1 $
ChaanPidurtl U
Redfin Pickerel
Yellow Bullhead 1 l 7
Brawn Bullhead
Snail Bullhead
Ta^obMtAon
Speckled Madtem
16 2 4
TmmHhiI Darter 9 1 12 1
Btackfaaadtd daiur 17 3 4 20 1
Yellow parch 2 5
mtwtwi
T«Ul 231 7« 227 61 99 33 34 74
Stream Order 33333333
Number of Spebet 19 16 n 12 14 n 12 11
EcongwB LPLPLPULPLPLPLP
Page 10
-------
Savannah River REMAP Fish Collection
CoaanoB Name *
133
37
34
32
12
27
19
14
American Eel
Piralc Perch
2
1
1
9
1
1
Northern bepuefcer
1
5
7
1
7
1
2
Spotted Sucker
2
Striped jumprock
3
2
1
SBwrndbone
1
Flier
11
7
21
6
5
5
3
5
GmnSuofoh
2
4
13
2
4
21
Wamouth
1
2
2
BhMgill
7
3
1
1
1
Redeye Baa
largemnuth Bui
1
1
Black Qrappie
Mottled Salvia
Whitefin Shiner
2
Silvery Minnow
RoeyftoeChub
13
4
16
1
S
1
5
Bhiebead Chub
3
41
60
50
>8
5
IS
27
Golden Shiner
1
2
Highfin Shiner
2
20
1
SpoCtail Shiner
3
Ytllowfm Shiner
fiMAar Cketxr
43
66
66
31
13
11
23
9HHHNH OilllNi
Creek Chub
2
3
14
4
1
Chain Pickerel
Redfin Pickerel
Yellow Bullhead
3
1
1
Brawn Bullhead
1
2
1
Black Butlhaad
Snail Bullhead
1
1
1
3
Margined MadUm
1
6
2
3
3
Tulpulf
Flat Bullhead
CfariAmaedwier
7
6
3
TiiDfil Darter
1
3
9
1
Blarir handed darter
3
2
3
3
Yellow perofa
S
3
1
Kainhow trout
Total
21
136
22*
161
162
36
69
14
Stream Order
3
3
3
3
3
3
3
3
Number of Speciee
9
13
it
15
IS
7
12
16
Emegion
LP
LP
LP
LP
LP
LP
LP
LP
Page 11
-------
Savannah River REMAP Fish Collection
CammwMimt * 10 11 J lfJ 30 176 113 232
Americas Eel
PiMe Perch 3 3 1
Crack Chubaucker 2
Northern hof*udcer 11 4 3 2
SpcOad Sucker
Mpadjdnpradt
Silwmbane
Fte
Blue ^ottad wnfixfa 1
L Sunfub 2 7 9 S 10 1
13 1 M I
W«MMdk 3 11 3
Btttafitl 3 3 6
4
yOBidMBfilh
RadeyeBan
la^mouth Bas
Black Crappie
MoaMScalfiD
Wbufin Shiner 2 2
Silvery Minnow 1
Roayface Chub 1 2 3
Biuebaad Chub 72 13 47 21 23
Golden Shiner
Hijfafio Shiner 2 1
SpoOail Shiner 1 1
YtUowfin Shiner SO 17 26 6 16
OmkCfaub 11 4 2 2
Chain Pickerel 2
Radfin Pickerel
YeflowBuUhaad
Brown Bullhead 2
Black Bullhead
Snail Bullhead
Margined Madtom j
Ti^tkMaAm
tparHiil tffarttmri ]
FhtBuBhaaii
TiillmitPMKr
YaBompwcfa
Rainbow trout
2
T#ttI 165 3* 110 46 S3 9 23 11
SttwmCMar 3 3 3 3 2 2 3
Number of Spaoiai 9 12 U 10 IS 2 6 4
Eoowjian LP LP LP LP U» MACP SH 8H
Page 12
-------
Savannah River REMAP Fish Collection
Common Name ,
231
127
136
138
238
SI
163
166
Americas Eel
1
Pirate Perch
5
2
2
2
Crack Cbubeucker
4
1
Northern bopudcer
2
Spotted Sucker
Striped jumprodt
SilwndxM
Tom
1
Blue ^x*l«d ttmfixh
RadbraaatSua&fa
1
1
3
1
2
QnaSuoUi
Wamouth
4
1
BluefOl
2
1
Redeariunfufa
tpottid
2
5
11
Redeye Ba»
brgemouth Ban
1
Black Crappie
Mottled Scutpm
6
3
Whitefin Shiner
Silvwy Minnow
Roey&ceChub
Blir*****1 Chub
27
7
1
2
20
Golden Shiner
Hifhfin Shiner
Sponail Shiner
3
YcUowfin Shiner
fiaMflttf CUlfia#
11
119
13
10
2
2
38
15
MUUIWI JlMlm
Creak Chub
35
34
5
2
11
3
Chain Pickerel
2
3
Radfin Pickerel
1
Yellow Bullhead
2
Brown Bullhead
Black Bullhead
Snail Bullhead
Mergmed Madtom
1
1
Tadpole Madura
Speckled MadUan
1
2
Flat Bullhead
SavMUkh Dartar
2
2
tutquioae derter
T—lined Dtrttr
3
1
1
Blidcbttdad d#w
2
Yellow path
1
Rainbow trout
Total
14
112
S3
32
31
20
39
St
Stream Order
3
1
1
2
1
1
2
2
Number of Special
3
4
3
S
11
10
11
9
Eeongion
SH
UP
UP
UP
US/SP
US/SP
US/SP
US/SP
Page 13
-------
Savannah River REMAP Fish Collection
CoBunoaName •» 236 237 1«4 167 Totel
American E«] 2
KnttPadi 1 1 3 3 45
CM Cbubwcker 33
Northern bopucker 237
SpoBid Sttcta 10
tayadjuyrak 29
SihMrnAem 10
Fber 1
Bh»voa*d«uo&h 1 1
Redbreast Sunfah 1 W 13 533
QmSut&b 1 250
WmhuOi 2 1 44
Bhwpll 6 194
Pmuflin i 19
Inntear ¦wfiiti 3 I 11
12
^inB«il wmfiih 3 3 22
Redeye Bas 15
kit*noulhBaas 2 29
Black Gnppie 2
ltlaOlad Sculpin 42
Whiufin Shiner ig
SitwyMioBSw g
Rseyface Chub 225
Dluihwil Chub 10 2098
OoMrn Shiner 42
Hilbfin Shiner 54
SpoWiil Shiner ^47
Yeltowfin Shiner 4 6 42 2866
Smiif Shaar 6
CmkCbub 3 B16
Chain Pickerel 2 2 25
Radfin Pickerel 1 q
Yallow Bullhead 40
Brown Bullhead g
Black Bullhead ^
Snail Bullhead 15
MafgjMdMadun 1 1 75
Tadpole Madun 1
FtatBuUtaad ^
favaonahOmr w
CtataMdrtr 4Q
tanpio«drtr 43
Teeeelletiii Perler H 1 70
Btackbadadteur 4 1W
Y«Bowpinb 1 j
48
Bailihfm Bout ^
T*«l 7 14 101 39 10074
toeatn Order 2 2 2 3
Number of Special 4 6 13 9
Eoorepon US/SP US/SP US/SP US/SP
Page 14
-------
1
1
1
1
1
1
1
1
1
3
1
3
5
1
1
1
1
3
1
3
S
1
1
3
1
1
1
S
s
s
Savannah River REMAP IBI Final Results
Number of wtiw N—rfiu of minx Nwnherof ¦
Mm
Ml 1iw luiti Mckcn
Wtt
¦ies
Uuai^> - ^ - *
w MCM
Ma
Mefcic
IBI
RNVR
im
Metric
im
Metric
im
Met**
MCUK
IBI
Metric
IBI
Metric
IBI
Rcnll
Scar*
r«mH
Score
Result
Score
Remit
Score
Result
Score
Result
Scam
Result
Score
0.51
3
1.47
5
200
100
1
300
5
1100
5
1.00
3
0.51
3
0.00
5
000
3
100
1
3.00
3
600
3
0.00
1
0.43
3
1.33
5
000
3
100
1
3.00
3
500
3
0.00
1
0.41
3
0.00
5
000
3
000
1
3.00
3
300
1
0.00
1
0.29
1
000
5
000
3
1.00
1
300
3
600
3
0.00
1
0.30
3
0.00
5
0.00
3
000
1
3.00
3
300
1
0.00
1
0.43
3
0.00
5
000
3
1.00
1
300
3
4.00
1
0.00
1
0.S4
5
4.20
5
000
3
300
4.00
3
900
5
0.00
1
0.41
3
0.00
5
0.00
3
0.00
1
300
3
300
1
0.00
1
0.74
5
323
5
0.00
3
4.00
400
3
1100
5
1.00
3
0.40
3
0.00
5
0.00
3
1.00
1
3.00
5
7.00
3
0.00
1
0.58
5
0.00
5
1.00
5
1.00
1
4.00
3
1000
5
zoo
5
0.90
5
0.00
5
1.00
5
200
1
300
3
900
5
2.00
5
0.54
5
15.30
3
000
3
400
1.00
1
7.00
3
0.00
1
0.43
3
0.00
5
0.00
3
0.00
1
3.00
3
5.00
3
1.00
3
0.00
1
0.00
5
0.00
3
0.00
1
1.00
1
200
0.00
1
0.S2
5
41.10
1
0.00
3
300
2.00
1
5.00
3
000
1
0.51
3
0.00
5
100
5
300
400
3
900
5
0.00
1
0.32
3
0.00
5
0.00
3
000
1
300
3
3.00
0.00
1
0.27
1
0.00
5
1.00
5
1.00
1
3.00
3
500
3
0.00
1
0.00
1
0.00
5
100
1
1.00
1
4.00
3
700
3
0.00
1
0.74
5
23.00
1
2.00
3
200
1
3.00
5
1200
3
1.00
1
0.77
5
3.51
5
too
1
4 00
3
3.00
3
11.00
3
1.00
1
0.00
5
4.37
5
200
3
4.00
3
500
5
1900
5
2.00
3
0.03
5
2.79
5
1.00
1
3.00
5
4.00
3
1600
3.00
5
0.40
3
0.00
5
0.00
1
1.00
1
3.00
3
4.00
0.00
1
0.41
3
000
5
0.00
1
000
1
3.00
3
3.00
0.00
1
0.07
5
1.09
5
200
3
0.00
1
4.00
3
900
3
1.00
1
0.73
5
1.91
5
200
3
3.00
3
400
3
14.00
5
1.00
1
0.00
5
2.00
5
300
5
200
1
3.00
3
1000
3
1.00
1
0.46
3
000
5
000
1
3.00
3
200
1
600
1.00
1
0.72
S
7.94
5
1.00
1
3.00
3
4.00
3
12.00
3
1.00
1
0.52
3
4.10
5
1.00
1
1.00
1
3.00
3
7.00
3
200
3
0.95
S
0.00
5
100
1
100
1
400
3
13.00
3
200
3
0.09
1
0.00
5
100
1
000
1
200
1
2.00
0.00
1
0.02
s
2.14
5
1.00
1
300
3
400
3
900
3
0.00
1
0.07
5
2.13
5
1.00
1
300
5
3.00
5
1500
0.00
1
0.52
3
0.00
5
0.00
1
3.00
3
3.00
5
900
3
0.00
1
0.01
5
1.57
5
1.00
1
2.00
1
3.00
5
12.00
3
200
3
0.42
3
075
5
000
1
1.00
1
300
3
4.00
1
0.00
1
0.S3
*
0.00
S
1.00
1
1.00
1
DnnM 4
4.00
3
900
3
1.00
1
-------
ret
Son
5
5
3
3
1
3
S
1
1
3
1
3
3
1
3
5
3
3
5
1
1
3
3
1
3
S
5
5
3
3
3
Savannah River REMAP IBI Final Results
NtMitber Mtivc
5 6
Nwnhcr of native Hiw*w ofiiwuKW
NiiiAif if
MCW
IBI
Metric
IBI
Metric
IBI
Metric
IBI
Metric
IBI
ImK
Score
Renik
Score
R«mH
Score
Result
Result
Scare
11.11
3
200
3
4.00
3
4.00
3
13.00
3
4.12
5
200
3
4.00
3
4.00
3
12.00
3
0.00
5
1.00
1
2.00
1
200
1
6.00
1
005
5
1.00
1
2.00
1
4.00
3
900
3
0.00
5
0.00
1
0.00
1
200
1
200
1
1.27
5
1.00
1
100
1
300
3
7.00
3
0.00
5
000
1
1.00
1
3.00
3
800
3
1X73
3
1.00
1
2.00
3
7.00
5
11.00
3
7.88
5
1.00
1
2.00
3
300
3
900
3
1.72
5
1.00
1
3.00
3
300
3
1200
3
2.47
5
3.00
S
300
3
600
5
15.00
5
5.95
5
3.00
5
3.00
3
500
3
16.00
5
30.43
1
1.00
1
3.00
3
300
3
1200
3
0.00
5
0.00
1
100
1
300
1
700
1
1.19
5
3.00
5
2.00
3
400
3
1500
5
5.70
5
3.00
5
4.00
5
300
3
18.00
5
2.94
5
100
1
200
3
300
3
1300
3
11.43
3
000
1
300
3
3.00
3
11 00
3
7.07
5
000
1
400
5
300
3
1400
3
5.93
5
1.00
1
4.00
5
400
3
1300
3
3.08
5
too
1
300
3
400
3
11 00
3
4.14
5
1.00
1
300
3
300
3
1500
5
4.00
5
2.00
3
2.00
3
300
1
1200
3
5.58
5
1.00
1
2.00
3
300
1
11.00
3
7.98
5
2.00
3
4.00
5
300
1
1400
3
0.39
5
3.00
5
4.00
5
400
3
1900
5
2.18
5
3.00
5
3.00
3
400
3
1600
5
0.00
5
3.00
5
1.00
1
400
3
1800
5
0.00
5
100
1
100
1
400
3
1000
3
222
5
1.00
1
3.00
3
300
3
1300
3
4.85
5
1.00
1
200
3
3.00
1
1200
3
7.14
5
100
1
4.00
5
100
1
900
3
0.00
5
200
000
1
300
1
700
1
0.00
5
000
1
400
5
2.00
1
8.00
1
0.00
5
100
1
300
3
3.00
1
1000
3
000
5
1.00
1
300
3
200
1
11 00
3
4.92
5
1.00
1
400
5
4 00
1200
3
8.11
5
200
3
200
3
3.00
1
11 00
3
9.28
5
2.00
3
300
3
3.00
1
1200
3
1.20
5
200
3
400
5
3.00
3
1800
5
1.20
J5
1.88
1
4-00
»
3.88
3
1500
5
RwmH
IBI
tow
1.00
1.00
1.00
1.00
0.00
0.00
0.00
0.00
100
1.00
000
000
2.00
0.00
3.00
3.00
3.00
2.00
200
1.00
100
2.00
2.00
3.00
2.00
4.00
3.00
3.00
2.00
2.00
200
0.00
1.00
0.00
1.00
1.00
too
1.00
1.00
1.00
109
1
1
1
1
1
1
1
I
3
3
1
1
3
1
5
5
5
3
3
3
3
3
3
5
3
S
S
5
3
3
3
1
3
1
3
3
3
3
3
3
3
-------
1
1
1
3
1
3
S
1
1
1
1
3
1
3
5
1
1
3
1
1
1
S
S
5
1
1
1
1
1
1
1
I
1
Savannah River RfeMAP 1B1 Final Results
4 5 6 7
hade*
Nn mKm «um
wcfcm
mhTnIms
»• > '
MCV16
IM
»« - 1
IBI
Metric
IBI
Metric
IDI
Metric
IBI
Metric
IBI
IBI
IUmH
Scon
RcmH
Score
Retail
Sent
Result
Seate
Re*uK
Sewe
Remit
Scan
Retail
Scare
0.51
3
1.47
5
200
1 00
1
5 00
5
11.00
5
100
3
051
3
0.00
5
000
3
100
1
300
3
600
3
000
1
0.43
3
1.33
5
0.00
3
1.00
1
300
3
500
3
0.00
1
0.41
3
0.00
5
0.00
3
000
1
3.00
3
300
1
0.00
1
029
1
000
5
0.00
3
1 00
1
3.00
3
600
3
0.00
1
0.30
3
0.00
5
000
3
0.00
1
300
3
300
1
0.00
1
0.43
3
0.00
5
000
3
1 00
1
300
3
400
1
000
1
0.54
5
4.20
5
0.00
3
300
400
3
8.00
5
0.00
1
0.41
3
0.00
5
000
3
000
1
300
3
300
1
0.00
1
0.74
5
3.23
5
0.00
3
4 00
4.00
3
11 00
5
1.00
3
0.40
3
000
5
0.00
3
1.00
1
300
5
7.00
3
0.00
1
0.50
5
0.00
5
1.00
5
1 00
1
4.00
3
1000
5
2.00
5
0.50
5
0.00
5
1.00
5
200
1
300
3
800
5
2.00
5
0.54
5
15.38
3
0.00
3
4.00
100
1
700
3
0.00
1
0.43
3
0.00
5
0.00
3
000
1
3.00
3
500
3
100
3
0.08
1
0.00
5
000
3
000
1
100
1
200
1
000
1
0.52
5
41.18
1
000
3
3.00
200
1
500
3
0.00
1
0.51
3
0.00
5
100
5
300
4.00
3
600
5
0.00
1
0.32
3
0.00
5
000
3
0.00
1
3.00
3
300
1
0.00
1
027
1
0.00
5
1.00
5
1.00
1
300
3
500
3
0.00
1
0.00
1
0.00
5
1.00
1
1.00
1
400
3
700
3
0.00
1
0.74
5
23.80
1
2.00
3
200
1
300
5
1200
3
1.00
1
0.77
5
3.51
5
1.00
1
400
3
300
3
11.00
3
1.00
1
0.88
5
4.37
5
200
3
400
3
300
5
1800
5
2.00
3
0.83
5
2.78
5
1.00
1
5 00
5
400
3
1600
5
300
5
0.48
3
0.00
5
0.00
1
1.00
1
3.00
3
4.00
1
0.00
1
0.41
3
000
5
0.00
1
0.00
1
300
3
3.00
1
000
1
087
5
1.00
5
2.00
3
0.00
1
400
3
9.00
3
1.00
1
0.73
5
1.81
5
200
3
3.00
3
400
3
1400
5
1.00
1
0.88
5
2.08
5
3.00
5
2.00
1
3.00
3
10.00
3
1.00
1
0.48
3
0.00
5
0.00
1
3.00
3
2.00
1
600
1
1.00
1
0.72
S
7.84
5
1.00
1
3.00
3
400
3
12.00
3
1.00
1
0.52
3
4.10
5
1.00
1
1.00
1
3.00
3
7.00
3
2.00
3
0.85
5
0.00
5
100
1
1.00
1
400
3
1300
3
100
3
0.00
1
0.00
5
1.00
1
000
1
200
1
2.00
1
000
1
062
S
2.14
5
1.00
1
3.00
3
4.00
3
800
3
0.00
1
0.87
5
2.13
5
too
1
5 00
5
500
5
1500
5
0.00
1
0.52
3
000
5
0.00
1
300
3
300
5
900
3
000
1
081
5
157
5
1.00
1
200
1
300
5
1200
3
2.00
3
0.42
3
0.75
5
080
1
1.00
1
300
3
400
1
0.00
1
0.53
3
0.00
S
1.08
1
1.00
1
400
3
800
3
1.00
1
Paoe 1
-------
3
1
5
S
S
5
3
3
1
3
1
1
1
1
3
1
3
9
Savannah River REMAP IB! Final Results
5 ®
Number of Mine Number of Hiiuxaw*
Scwt
083
0.80
0.52
0.41
0.10
0.37
0.00
008
0.«t
0.78
0.03
0.80
0.73
082
0.71
0.97
0.77
0.70
088
0.70
0.70
0.71
0.71
0«7
079
088
0.90
0.90
0.70
0.77
0.77
0.61
0.00
0.50
0.07
eoo
081
081
0.74
QJH
MS
S
S
3
3
1
3
S
1
1
3
1
3
3
1
3
S
3
3
S
1
1
3
3
1
3
5
5
5
3
3
3
11.11
4.12
0.00
oes
0.00
1.27
0.00
12.73
7.08
1.72
2.47
S.0S
3043
0.00
1.10
5.70
2.9*
11.43
7.07
593
308
4.14
4.00
sse
7.96
030
*10
0.00
0.00
2.22
4.05
7.14
0.00
0.00
0.00
0.00
492
6.11
0.28
120
\X
IBI
Scott
3
5
5
5
5
5
5
3
5
5
5
5
1
5
S
5
S
3
5
5
S
5
5
5
5
S
5
S
5
S
S
5
5
5
5
S
S
5
5
5
S
too
200
1.00
1.00
000
1.00
000
too
1.00
1.00
3.00
300
1.00
0.00
300
300
100
000
000
1.00
100
too
200
1.00
200
3.00
300
300
1.00
1.00
100
1.00
200
000
100
too
1.00
100
200
200
lot
I
1
1
1
1
1
1
1
5
5
1
1
5
S
1
1
1
1
1
1
3
1
3
5
5
5
3
3
3
»
Metric
RcMk
400
4.00
200
2.00
000
1.00
1.00
2.00
2.00
3.00
3.00
3.00
300
1.00
2.00
4.00
2.00
100
4.00
400
3.00
300
200
200
400
400
300
100
100
3.00
2.00
4.00
000
4.00
300
3.00
400
200
3.00
400
4.66
3
3
3
3
3
3
1
3
5
3
3
5
5
3
3
3
3
5
5
3
1
1
3
3
5
1
5
3
3
5
3
3
5
5
Metric
RewH
4.00
400
200
4.00
2.00
3.00
3.00
7.00
3.00
500
COO
300
300
3.00
4.00
$00
5.00
500
500
4.00
400
5 00
300
300
3.00
400
4.00
4.00
4.00
300
300
100
300
200
300
200
400
300
3.80
3.00
56#
tm
Scow
3
3
1
3
1
3
3
5
3
3
5
3
3
1
3
3
3
3
3
3
3
3
1
I
1
3
3
3
3
3
3
V
Metric
Rratfc
1300
1200
6.00
9.00
2.00
7.00
8.00
1100
900
t2.00
15.00
1600
1200
7.00
15.00
18.00
13.00
11.00
14.00
1300
11.00
1500
12.00
11.00
14.00
19.00
16.00
1600
10.00
13.00
1200
900
700
600
1000
1100
1200
1100
12.00
16.00
1S.0B
IM
tan.
3
3
1
3
1
3
3
3
3
3
5
5
3
1
5
5
3
3
3
3
3
5
3
3
3
5
5
5
3
3
3
3
1
1
3
3
3
3
3
5
S
l.<
1.00
1.00
1.00
0.00
6.00
6.00
000
1.00
1.00
0.00
0.00
2.00
0.00
3.00
3.00
3.00
2.00
2.00
1.00
1.00
2.00
2.00
3.00
2.00
4.00
3.00
3.00
1.00
2.00
2.00
0.00
1.06
0.00
1.00
1.00
1.00
100
1.00
1.66
166
TBI
Scow
1
1
1
1
1
1
1
1
3
3
1
1
3
1
5
5
5
3
3
3
3
3
3
5
3
5
5
5
3
3
3
1
3
1
3
3
3
3
3
3
3
34.92
12.37
44.63
40.65
21.43
27.65
29.17
45.45
50.91
29.31
63.56
36.90
34.78
13.66
29.76
27.63
31.02
22.86
22.22
48.31
54.19
4436
30.66
37.30
35.56
18.99
26.72
15.36
16.90
3259
2.33
1071
35.62
37.50
50.00
37.93
22.95
5676
33.33
25.30
30.12
-------
1
a
i
s
2
3
2
2
3
3
1
3
3
3
3
2
2
2
2
1
1
1
2
2
1
1
3
3
3
3
1
1
3
2
3
3
3
3
2
1
2
3
3
3
1
1
2
. 2
1
2
2
3
Summary of biological indicator results and scores for the
Lower Piedmont Ecological Index
Macroinvertebrate
EPT Ftoh IBI
Station
Ordar
Habitat
Seora
Max Seora
RaauN
Seora
Max Seora
Claaaifieat
8
3
64
3
3
1
19
1
6
Poor
9
2
94
5
4
1
25
3
9
Pair
10
3
80
3
4
1
21
1
8
floor
11
3
47
1
3
1
25
3
6
Poor
12
3
83
3
5
1
27
3
1
Poor
13
1
61
3
3
1
33
8
9
Mr
14
3
46
1
3
1
29
3
8
Pfcor
15
1
87
3
2
1
23
3
7
Poor
19
3
91
6
6
1
21
1
t
Poor
22
2
93
6
8
1
23
3
•
Mr
27
3
64
3
9
1
17
1
B
Poor
28
2
92
5
2
1
27
3
1
Mr
29
2
119
6
7
37
6
*3
Oood
30
3
116
8
8'
1
33
8
*1
Om4
32
3
112
5
5
1
36
8
T1
Oood
33
1
75
3
4
1
21
1
6
34
3
113
5
6
3
39
5
13
Oood
37
3
111
5
7
27
3
11
Oood
38
3
68
5
5
1
23
3
9
Pair
39
3
82
3
2
1
31
5
8
Pair
41
2
48
1
4
1
35
5
1
Poor
64
2
45
1
7
3
19
1
8
Poor
65
2
41
1
5
1
17
1
3
Poor
68
2
104
5
15
5
31
5
18
Hood
69
1
•4
3
12
5
19
1
9
Pair
71
1
76
3
8
3
19
1
f
Poor
72
1
82
3
18
5
21
1
9
Fair
74
41
1
9
5
35
5
11
Oood
75
2
54
3
9
5
31
5
13
Bood
77
1
106
5
10
5
21
1
M
Bood
78
1
91
5
4
1
29
3
9
Pair
79
53
3
7
3
23
3
9
Pair
80
3
52
1
8
3
21
1
6
Poor
61
3
52
1
8
3
25
3
"7
Poor
82
3
107
5
8
3
25
3
11
Dood
83
1
66
3
11
5
19
1
9
Pair
85
1
45
1
5
1
33
6
1
Poor
66
3
72
3
9
5
25
3
M
Oood
67
2
30
1
7
3
21
1
1
Poor
88
3
68
3
11
6
27
3
11
Oood
S3
3
104
6
10
8
43
8
18
Oood
94
3
69
3
8
3
39
8
H
Oood
SS
3
99
6
10
6
39
8
18
Oood
96
2
103
6
10
8
23
3
19
Oood
86
1
69
3
12
8
25
3
<1
¦ood
99
2
72
3
6
3
31
8
tl
Oood
100
3
49
1
8
3
29
3
f
Poor
101
3
46
1
8
3
27
3
I
Poor
102
3
49
1
8
3
27
3
r
Poor
103
1
89
5
8
3
35
8
«
Oood
104
1
70
3
6
3
37
8
fi
Oood
122
2
53
3
8
1
13
1
8
fear
123
2
86
3
6
3
27
3
»
Mr
130
1
64
3
1
1
25
9
f
poor
131
2
82
3
8
3
33
8
11
Booa
132
2
46
1
8
3
25
3
7
Poor
133
3
82
3
7
3
23
3
9
tar
LPEI
Total Stream
-------
Station
1»~
143
144
145
147
148
146
1S1
154
186
155.1
162
197
200
206
210
213
214
216
221
222
oummaiy ot urofOQieai indicator results and scores for the
Lower Piedmont Ecological Index
Macroinvertebrate
EPT
FfchlBl
LPEI
Total Stream
0rtr
Station
Ordar
Habitat
Scot*
2
135
2
3
7
3
31
5
It
0Mtf
3
143
3
83
5
10
5
21
1
H
Hood
3
144
3
51
1
10
5
19
1
i
iter
3
145
3
47
1
11
5
18
1
1
3
147
3
46
1
10
5
21
1
I
Poor
1
148
1
56
3
8
5
25
3
H
2
148
2
52
1
6
3
18
1
«
Mar
2
151
2
51
1
5
1
23
3
»
fear
2
154
2
68
3
11
5
27
3
M
2
156
2
71
3
5
28
3
ft
•end
3
156.1
a
72
3
6
3
27
3
•
1
162
i
85
5
8
5
15
1
m
0Ottf
1
187
i
52
1
6
3
21
1
•
Poor
2
200
2
58
3
11
5
18
1
»
M
2
206
2
83
3
5
1
23
t
Pfeor
3
210
3
103
5
8
5
23
3
w
Oood
1
213
1
57
3
12
5
21
1
8
Pair
1
214
1
74
3
7
3
27
3
8
fair
Poor
2
216
2
46
1
8
S
17
1
1
2
221
2
67
3
12
S
23
3
11
Oood
2
222
2
66
3
8
3
25
3
9
Pair
-------
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
¦ wvai dirwm
Habitat
Seort
Indtx
Seort
Seort Seort
Index Seort Claudication
76
3
3
1
61
3
3
1
33 5
9 Fair
67
3
2
1
23 3
7 Peer
75
3
4
1
21 1
6 Poor
75
3
3
1
1
64
3
12
5
19 1
9 Fair
76
3
6
3
19 1
7 Poor
62
3
16
5
21 1
9 Fair
105
5
10
5
21 1
11 Fair
91
5
4
1
29 3
9 Fair
66
3
11
6
19 1
9 Fair
45
1
5
1
33 5
7 Poor
40
1
11
5
1
69
3
12
5
25 3
11 Fair
99
5
6
3
35 5
13 Good
70
3
6
3
37 5
11 Fair
64
3
1
1
25 3
7 Poor
55
3
9
5
25 3
11 Fair
95
5
9
5
15 1
11 Fair
52
1
6
3
21 1
6 Poor
6
3
33 5
57
3
12
5
21 1
9 Fair
74
3
7
3
27 3
9 Fair
94
5
4
1
25 3
9 Fair
93
5
5
1
23 3
9 Fair
92
5
2
1
27 3
9 Fair
119
5
7
3
37 5
13 Good
67
3
1
46
1
4
1
35 5
7 Poor
45
1
1
1
45
1
7
3
19 1
6 Poor
41
1
5
1
17 1
3 Poor
104
5
15
5
31 5
16 Good
41
1
9
5
35 5
11 Fair
54
3
9
5
31 5
13 Good
30
1
7
3
21 1
6 Poor
4
1
27 3
103
5
10
5
23 3
13 Good
72
3
6
3
31 5
11 Fair
53
3
5
1
13 1
6 Poor
86
3
6
3
27 3
9 Fair
62
3
6
3
33 5
11 Fair
46
1
6
3
25 3
7 Poor
75
3
7
3
31 5
11 Fair
82
1
6
3
19 1
6 Poor
51
1
5
1
23 3
f Poor
•6
3
11
5
27 3
11 Fair
71
3
10
6
29 3
11 Fair
56
3
11
5
19 1
9 Fair
63
3
s
1
2:' 3
7 Poor
46
1
9
5
17 1
7 Poor
67
3
12
5
23 3
11 Fair
66
3
6
3
25 3
9 Fair
64
3
3
1
19 1
• Poor
60
3
4
1
21 1
6 Poor
47
1
3
1
25 3
• Poor
63
3
1
27 3
7 Poor
45
1
3
1
29 3
f Poor
91
5
5
1
21 1
7 Poor
67
3
S
1
1
-------
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
EPT
IBI
Total
Straam
Habitat
Score
Index
Scora
Scora
Scora
Indax Scora
Classification
64
3
3
1
17
1
8
Poor
11S
5
5
1
33
5
11
Fair
112
5
5
1
35
5
11
Fair
113
5
6
3
39
5
13
Good
111
5
7
3
27
3
11
Fair
M
5
5
1
23
3
9
Fair
82
3
2
1
31
5
9
Fair
53
3
7
3
23
3
9
Fair
52
1
8
3
21
1
8
Poor
52
1
8
3
25
3
7
Poor
107
5
8
3
25
3
11
Fair
72
3
9
5
25
3
11
Fair
66
3
11
27
3
11
Fair
104
5
10
5
43
5
18
Good
59
3
8
3
39
5
11
Fair
M
5
10
5
39
5
18
Good
49
1
8
3
29
3
7
Poor
48
1
6
3
27
3
7
Poor
49
1
6
3
27
3
7
Poor
62
3
7
3
23
3
9
Fair
93
5
10
S
21
1
11
Fair
51
1
10
5
19
1
7
Poor
47
1
11
5
19
1
7
Poor
46
1
10
21
1
7
Poor
103
5
9
5
23
3
13
Good
7
3
25
3
11
5
35
5
72
3
8
3
27
3
9
Fair
Scons
Parcant
Good
9
11.5
Fair
38
487
Poor
31
397
Total
78
100
-------
to
-------
Appendix D
Guidelines for Locating and Accessing Sites on
Wadable Streams in Watersheds of the
Southeastern United States
By
James R. Maudsley
and
Robert J. Lewis
-------
Guidelines for Locating and Accessing EMAP Sites on Wadeable
Streams in Watersheds of the Southeastern United States
Prepared by:
James R. Maudsley Ph.D. And Robert J. Lewis
ManTech Environmental Technology Inc.
under contract to Integrated Laboratory Systems
Athens, Georgia
For:
U.S. Environmental Protection Agency
Region IV
Athens, Georgia
DCN #ESAT-4B-7004
1997
-------
Table of Contents page
Introduction 1
Acquiring Permission To Access Stream Sampling Sites.. 1
Locating EMAP Coordinates on a Map 1
Acquiring Names and Addresses of Landowners 1
Requesting Permission to access Stream Sampling Sites 3
Locating EMAP Sites in the Field - Field Reconnaissance 6
Direction Packets - For Relocating EMAP Sites in the Field at a Later Date 8
List of Figures
Fig. 1. Computer Generated Map 2
Fig. 2. Typical County Map in Tax Assessor's Office 4
Fig. 3. Parcel Map 4
Fig. 4. Sketch of Parcels of Land Surrounding Two Sampling Locations 5
Appendices
Accurately Locating EMAP Coordinates on a Computer Generated Map Appendix A
Form Letter to Landowners Requesting Permission to Access a Stream Site and
Sample Permission Slip Appendix B
Sample Direction Packet Appendix C
-------
1 of 9
Introduction
The following guidelines summarize the steps employed by EPA Region IV to locate and
access stream sampling sites selected by the Environmental Monitoring and Assessment Program
(EMAP) approach. The strength of the EMAP approach is that it is a probability-based survey
(Volstad et. al. 1995) Sampling locations are randomly selected. The net result is that fewer
stream sampling locations are needed to characterize a watershed or river basin than would be
required with a non-random selection process (ie. bridge crossings), thereby effectively reducing
the effort needed to characterize large, regional river basins to a logistically and economically
feasible level. While the EMAP approach reduces the effort required to conduct regional surveys,
it presents some unique challenges. Because stream sampling locations are randomly selected, the
stream sites may not be near identifiable physical land structures (bridge crossing, bend in the
stream) and miles from the nearest road or highway. Through trial and error EPA Region IV has
developed the following set of procedures for efficiently locating and accessing EMAP sites:
Acquiring Permission to Access Stream Sampling Sites
Step 1: Locating EMAP Coordinates on a Map
The first step in locating a randomly selected EMAP sampling location is to correctly
pinpoint the site on a map. Accuracy is essential to prevent costly and time consuming mistakes.
Each stream location generated by the EMAP approach is supplied as a pair of map coordinates, a
latitude and a longitude. The method used to locate EMAP streams sites in EPA Region IV relies
on computer software to mark the exact location of each set of coordinates on a computer
generated map (see figure 1). Region IV uses a mapping program called MapeXpert® by
Delorme but other mapping programs should work equally as well. The exact procedure for
placing EMAP coordinates on a map using MapeXpert can be found in Appendix A of this
document.
Step 2: Acquiring Names and Addresses of Landowners
Once a stream sampling location is accurately located on a map, the next step is to
determine who owns the property adjacent to the stream and ask permission to access the site.
This requires a visit to the Tax Assessor's Office in the county in which the site is located.
County tax assessor offices house records of property ownership. Most often the offices are
located in the city or town serving as the county seat. Some offices have all records as a hard
copy only, making the search for land ownership a rather slow process. However, more and more
tax offices are converting to electronically stored information, easily accessible from computer
terminals located at various stations throughout the office. Regardless of the information retrieval
system certain basic information must be obtained to properly locate the owners of a given
property. Personnel working in the tax offices are usually more than willing to guide you through
the process.
-------
^Sr S-39-jJjMeece Mill Road
.-^-Middle Fork Twehemile
\iiU i re^k
, . nagooa Branch
dSy f '
1781
vMilll f
J ll U>>
r S-39-33|M
Sta.,136
\.UddU'Fork Vwfamife C
\rs,rth Fork vmke/tiile Creak
Water W
Creek Ro
v.1 u 11111! I) hn.
Mount Bethel Road
Sr S-39-?^Rcece Mill Road
ih.i.iihW'X'^
t-fay*.i And ColtrnsJ-Mh:
XII1IWI1
7tjyrn Lrafk
K
183
t Fvrk Cmtgory Crvrk
2 of 9
''IIIIIIL
IO 1993 DcLormc Mapping \
LEGEND
a Geo Feature
US Highway
Population Center
Street, Road
—_ Major Street/Road
- » US Highway
River
Utility (powerline)
Open Water
Contours
Scale 1:37,500 (at center)
2000 Feet
1000 Meters
Mag 14.00
Thu Jul 25 15:57:52 1996
Fig. 1. Computer Generated Map
-------
3 of 9
Essentially, two pieces of information are needed to lookup a property owner, a map # and a
parcel #. The map # is obtained by looking a large map of the county prominently displayed
somewhere in the office (see fig 2). This map has a grid of large squares superimposed on it.
Each square has a number. Each nunibered square corresponds to another more detailed map of
just that portion of the county bounded by the sides of the square. With a MapeXpert map
prepared in step 1 (above) in hand, identify the numbered square on the county map that contains
the desired sampling location(s). Record the number of the square. Now (ask for help if
necessary) physically locate in the tax office the map corresponding to that number and remove it
from its file drawer or rack. The map is usually about 3 ft. square and appears either as a blue on
white blueprint of roads, streams and other land structures or as an aerial photograph (see fig 3).
Again with the MapeXpert map from step 1 in hand, locate on the numbered map the exact
location of the sampling site. The numbered map will have superimposed on it a mosaic of
polygons (see fig 3 again). Each polygon is a parcel. Each parcel will have a whole number in it.
This is the parcel #. Identify and record the number(s) of all parcels that must be crossed to
access the stream site from the nearest road or pathway (power line cut, railroad track, trail etc).
When the site is surrounded by multiple parcels, sketch a diagram of the position of each parcel in
relation to the sampling site (see fig 4). If possible, sketch the parcels on the computer generated
map made in step 1. This information will be important in Step 3.
Now, physically locate county records of land ownership. The records may be index cards in a
file drawer, bound in books, or accessed through a computer terminal. Look up property
owner(s) by map # and parcel #. Usually, file drawers and books are labeled by map number.
Upon opening the drawer or book, individual records will be filed consecutively by parcel number
starting with parcel number 1. Leaf through the records until the desired parcel # appears.
Record the name and address of the property owner that appears on the record. As a cross check,
look up each landowner by last name in the current year's tax records (another set of bound
volumes listing all landowners alphabetically by last name) to obtain a current mailing address.
For electronic retrieval, ask office personnel for assistance. Accessing records by computer
usually requires that map and parcel numbers be entered into the computer in a specific format.
This format often varies from tax office to tax office. With electronic retrieval, the names and
addresses that appear on the computer screen will be current.
Step 3: Requesting Permission to Access Stream Sampling Sites
Send each property owner an envelope that includes a letter requesting permission to cross their
property, a permission slip that grants permission for you to access their property, and an
addressed, stampled envelop in which to return permission slips. An example of the letter used
by Region IV to request permission to access a given property and an example of the permission
slip that is to be returned by the landowner to EPA are included in this document in Appendix B.
Note that the sampling location (e.g. site 175) must appear on the permission slip as well as on
the letter requesting permission to access the property. This facilitates the matching of returned
permission slips with the appropriate sample locations.
-------
ijCNERM- Mi oh way map
QGLLTHQRRE COUNTY
GEORGIA
una Pt9ir& - wnur.snaj wajcwct
' tlMORFE CuiHIY [WfWlI .
¦,Trj ¦ r m>
Fig. 2. Typical county map in tax
assessor's office.
4 of 9
Fig 3. Parcel Map
-------
5 of 9
Fig. 4. Sketch of parcels of land surrounding two sampling stations.
pJ&Au>y
aq / WJUsajuA.
"Viioutfc
r ,
f U)
-------
6 of 9
Locating EMAP Sites in the Field - Field Reconnaissance
Equipment and Supplies
Vehicle (4 wheel drive recommended) with a trip odometer
Hand held GPS unit
compass
flagging tape
machete
insect repellent
Maps accurately showing location of sampling sites.
Permission slips from property owners
Laptop computer loaded with software to link GPS signals to mapping program (optional)
Step 1: Drive as close to a stream sample site as possible
This is where a 4 wheel drive vehicle can make a difference. Once a vehicle can go no farther, the
site must be within 30 minutes walking time from the vehicle. If not, the site is designated a
"non-target" and will not be scheduled for sampling. So, the closer the better.
A factor that can significantly increase the success of finding a stream site is knowing where you
are relative to the site. Following county road maps, computer generated maps, and topo maps
work well as long as you can recognize landmarks along the route that pinpoint your location on a
map. This is often difficult to do on large tracts of undeveloped forest or timberland where many
of the roads, especially logging roads, are not marked or not shown on maps. This is where a
GPS/Mapping System is employed. The basic system consists of a notebook (lap top) computer
outfitted with a CD ROM drive and a GPS Receiver-PCMCIA card. A mapping program is run
on the computer while the computer receives information from the GPS receiver. The result is a
map on the computer screen which not only displays all previously marked stream locations but
also displays the continuously updated position of the vehicle carrying the computer. In this way
a vehicle's progress toward a predetermined location can be monitored to guide the vehicle as
close the to site as possible, even on unmarked roads. The specifications of the GPS/Mapping
system employed by EPA Region IV is as follows:
Notebook Computer
• 486-66mhz processor or greater
• CD ROM drive, 4X or greater
• Active matrix color screen (recommended for viewing screen in bright sunlight)
• 8MB memory minimum (recommend 16 MB or more)
-------
7 of 9
MapeXpert with GPS Link (called Mapkit, $495.00)
available from: Delorme Mapping
2 Delorme Drive
P.O. Box 298
Yarmouth, ME. 04096
1-800-227-1656
GPS Receiver
PCMCIA Card, SATNAV LP NMEA Output Model PM50154 $550.00
with GPSpac software for Windows Model PM50154 $99.00
available from: Centennial Technologies Inc.
180 Cherokee Rd.
Ashville, NC 28804
(704)281-0044
Step 2: Locate the site on foot
i
From the parked vehicle navigate to the stream site with the aid of a hand-held GPS and a
compass. To do this enter the coordinates of the stream site into the GPS unit as a WAYPOINT.
Then, after allowing the GPS unit time to acquire your position at the parked vehicle, ask the unit
to navigate to the WAYPOINT (refer to the GPS User's Manual for details on how to enter,
retrieve, and navigate using way points'). The GPS unit will display a bearing (in degrees) and
range (in meters or fractions of a mile). Using the compass, determine the direction of travel
specified by the bearing on the GPS screen and start walking. Take advantage of deer trails,
power line cuts, railroad beds and other pre-existing paths whenever possible. If necessary, hack
through underbrush and briers with the machete. The GPS unit will update your progress as you
go. Set new bearings with the compass as needed. Technically you have reached a site when the
range displayed on the GPS unit reads 0.1 mile (500 feet) or less, but get as close to the stream
site as physically possible. The "500 ft." rule was adopted for situations in which it is physically
impossible to reach the exact WAYPOINT (e.g. site just over property line where permission to
access was denied or stream site is impounded by a beaver dam). EMAP grid units are
approximately 2000 feet across. The 500 ft limit should keep the sampling site a valid sampling
point within the grid.
1 EPA Region IV utilizes an ENSIGN XL GPS unit manufactured by Trimble Navigation. Copyrights
prohibit the reproduction, translation, transformation or adaptation of the ENSIGN User's Manual in any form.
Therefore, refer directly to the user's manual for details on the basic operation of the GPS unit and for specific
procedures for entering (section 2-17) and navigating to (section 5-7) a waypoint.
-------
8 of 9
Once a stream site is located, mark it with flagging. As you return to the vehicle mark the trail as
needed and record walking time (remember, if walking time exceeds 30 minutes, eliminate the
sampling site from the sampling schedule). Also eliminate streams that are impounded (e.g.
beaver pond) or dry. On the trip back to the vehicle make notes (or take photographs) of any
recognizable landmarks (big oak tree, rocks, ridges, gullies etc.) that would help guide someone
back to the site should the flagging be lost or torn down. Immediately upon reaching the vehicle,
sketch a map showing the trail and position of landmarks relative to the sampling site. As you
drive away from the site continue to note landmarks on unimproved roads and use the trip
odometer to measure distances between turn ofFs, forks etc. This information will be needed later
to return to the site for the actual sampling.
Upon returning from the field, make necessary changes in maps (e.g. new roads, roads renamed or
moved), write clearly written directions to each site, and redraw detailed site maps based on the
notes and sketches made in the field.
It is best to conduct field reconnaissance after hunting season and not more than two months
before sampling is scheduled. If not, hunters tend to remove flagging and paths hacked through
underbrush and thickets tend to become overgrown again, making it difficult to follow paths and
relocate the sampling sites.
Direction Packets - for Relocating EMAP Sites in the Field at a Later Date.
Because it may take considerable time to locate EMAP sites in the field, reconnaissance and
sampling are treated as separate events. The reconnaissance is completed first. Then, at a later
date, teams return to each site for the actual sampling. Because the personnel conducting the
sampling may not be the same persons that conducted the field reconnaissance, a direction packet
is prepared by the reconnaissance team that provides samplers with all the information needed to
locate and safely access each stream site. A direction packet contains:
• Updated computer generated Map pinpointing the sampling site.
• Written directions to each site from the nearest town or main highway that contain
highway routes, descriptions (or photos) of landmarks and the detailed site maps
sketched during reconnaissance.
• Copies of the permission slips signed by property owners.
A sample direction packet is included at the end of this document as Appendix C.
-------
9 of 9
References
Volstad, J. H., S. Weisberg, D. Heimbuch, H. Wilson, and J. Seibel. 1995. Answers to Commonly
Asked Questions about REMAP Sampling Designs and Data Analyses. U. S. Environmental
Protection Agency, Research Triangle Park, NC.
-------
Appendix A
Accurately Locating EMAP Coordinates
on a Computer Generated Map
-------
Using MapeXpert™ Computerized Mapping Program to Pinpoint a Stream Sampling Site
Given a Specific Latitude and Longitude.
MapeXpert ™ version 2.0 for Windows is available from:
DELORME MAPPING
2 Delorme Drive
P.O. Box 298
Yarmouth, Maine 04096
207-227-1656
estimated cost $495
Required Computer Hardware
IBM or 100% IBM-compatible microcomputer outfitted with an Intel® 80386 or higher
or 100%-compatible processor.
Minimum of 2 MB RAM (4 MB RAM recommended).
3 MB of available hard drive space.
ISO 9660-compatible CD-ROM drive with Microsoft CD-ROM extensions.
Microsoft Windows ™-compatible mouse (recommended).
Microsoft Windows-compatible VGA card and monitor.
Microsoft Windows-compatible printer (recommended).
Microsoft Windows version 3.1
MS-DOS® 4.01 or higher.
Additional skills required to operate the MapeXpert program include a basic knowledge of
personal computer operation, use of a "mouse," and familiarity with the features and techniques
of Windows. The Windows Tutorial that accompanies Microsoft Windows Version 3.1 provides
and excellent introduction to the skills needed to use MapeXpert.
-------
To locate and label a stream sampling site given a set of map coordinates, use the Point Box
feature of the MapeXpert program. The procedure is as follows:
1.
2.
Start the MapeXpert program. You will see a map of the United States displayed at a
magnification of 3 with the Toolbox, Mag(nification) box, and Cursor box displayed at the
margins of the screen (see below).
I
DeLorme MapExpert 2.0 - [Map: 1 ]
File Edit Overlays Geography Display Window Help
j Toolbox
•*v—
AL AyiKM
Ml
^ HAWAII
Cursor box
I 1000 ml H30* 12* . W84* 6
Tools
Active map^ :c.^^^K-.
' • it— ~ h '-ir&i
- T -r
r_
Using the mouse, center the cross hairs of the cursor over the region of the state where
the study stream lies and click the left mouse button. This will place the area of the map
that contains the study stream in the center of the viewing screen.
Slide the cursor to the Mag box and double click on Mag 9. Small streams will not be
displayed at this magnification. Continue .
Move the cursor to the top of the screen . Choose Display...
Preferences... and then click on the check box next to the Point Box
option in the Preference dialog box (see right). Now click on OK at
the bottom of the Preference box. The point box will appear in the
lower right corner of the screen (see below). You will notice that the
point box covers the Mag box. At this point click and hold the left
mouse button anywhere on the words Point Box [Distance] and
drag the entire box to the left until the Mag box is fully exposed.
| Point Box (Distance)
Lat ]
N43-50-40 5779"
Lon
W070*05'58.66G3"
Dist.
1
Azim
270 1 AptfTl
Preferences
Display Options"
0 Tool box|
H Mag box
0 Cursor box
O Point Box
Cj Map Legend
Lat/Lon format
|N1 T 11111
-------
5. Choose a symbol with which to mark the stream
site. A red circle seems to stand out best. To
choose a symbol move the cursor to the Toolbox
and click on the Symbol tool J|] . The Symbols
dialogue box will appear (see right). Now click on
the Map check box. This will add color to the
symbols. Next use the scroll arrows on the Type
list box to find circle and highlight it by clicking
on it. Finally, click on Close.
6. Now return to the Point Box to enter the given latitude and longitude. To do this move
the cursor into the Lat box to the right of the letter N and click. Type the numbers
representing latitude in the order: degrees, minutes, seconds (to the nearest tenth), leaving
a space between each component. Next drop the cursor in the Lon box to the right of the
letter W, click, and enter the numbers representing longitude
7. Click on the word Apply in the bottom right corner of the Point Box. A red circle should
appear on the map displayed on the screen at the exact coordinates entered into the Point
Box. Close the Point Box by clicking in the square in the upper right hand corner of the
box. Now position the cursor (which now appears as a pencil) over the circle and click
the right mouse button once. (If the red circle is not visible, it is probably just off-screen.
Temporarily drop to Mag 8 to find the circle, then position the cursor over the circle and
click the right mouse button once).
8. Now increase magnification so that even the smallest streams are displayed. This usually
requires Mag 13 or higher. The red circle should now be located over a stream even if no
stream was visible under the circle at lower magnification.
At this point additional sites can be located and marked on the map without repeating steps 1
through 5. by simply entering new numbers for latitude and longitude in the Point Box (step 6)
and clicking Apply (step 7).
9. Finally, after all stream locations have been marked
on the map, add text to the map to identify the
stream site. To do this click on Text tool [ Tpi
The cursor should be flashing in the top box of the
Text dialogue box (see right). If not, move the
cursor into that box and type a label for the stream
site. Now move the cursor (still a pencil) to the
right, above, or below the red circle (wherever
space permits) and click the left mouse button.
The label should appear next to the circle. When
finished adding text, click on Close
Symbols
Type
Information
Golf Couise
¦
E3 Map
Size Anqle
|40 | |0
~ Tent
MB
Add Text
The Commons
Bold ~ Italic Alignjcap center ^
Sent Size ma Anglejo"
10.
To print a copy of the map click on the word File in the upper left corner of the screen
-------
and then on Print. The Print screen will appear (see below). The Preview Map window
shows the portion of the map that will be printed. To print the map click on the Print
box. Refer to the User's Manual to use any of the other features (e.g. Scale) displayed in
the print box.
Print
Panasonic KX-P4455 v51 4 on LPT2 DOS
10 in (W) x 8 in (H)
Map Title: |Poitet Landing. ME
Preview Map —~
" S cale "
1:62.500
mA
Li
s&ligi
is3
|~Piev Detail"^
{Medium
a
•> MapMakei
'.J MuralMakec
iaagga
||
\
Page Preview Mode: Use the Left Mouse Button to Pan
Any red dot(s) and text added to the map is an overlay on the original base map. The
overlay can be saved for future use by clicking on the word Overlay at the top of the
screen and then following the "Save As" feature designed for Windows.
-------
Appendix B
Form Letter to Landowners Requesting Permission to Access a Stream Site
and
Permission Slip
-------
>ttO ir4
* \
| 3 UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
REGION 4
Science and Ecosystem Support Division
980 Collsgs Station Road
Athens, Georgia 30605-2720
Robert Sims
RR2 Box 198
Comer, Georgia 30629
Dear Mr. Sims:
The Science and Ecosystem Support Division of the United States
Environmental Protection Agency, Region IV, in conjunction with the
states of South Carolina and Georgia, will be conducting biological
stream monitoring of the Savannah River Basin. The study, known as the
Savannah River - Regional Environmental Monitoring and Assessment
Program (SR-REMAP), is intended to gather data on condition of stream
ecosystems in the Savannah River Basin, identify stressors to these
systems, and provide a baseline of information for future trends. The
data is not intended for uses pertaining to litigation, tax assessment,
law enforcement, or similar purposes.
Randomly selected sites have been chosen in the Savannah River
Basin and a site or sites are located on your property. The EPA and the
state agencies assisting in this effort are seeking permission to access
the following site(s):
211
Scull Shoal Creek
Enclosed you will find a self-addressed envelope and access
permission form. Please sign the access permission form and mail by
March 21, 1997. We will visit the sampling site on two occasions, once
for fish sampling and once for macroinvertebrate (aquatic insects)
sampling. On receiving permission, we will contact you by telephone or
letter informing you of the scheduled sampling dates for the stream(s)
on or near your property.
Mr. Hoke Howard of EPA, Region IV, Science and Ecosystems Support
Division is the project coordinator for the SR-REMAP team. He can be
reached at EPA in Athens, Georgia at (706)355-8721 if you have questions
concerning this project. The entire team is grateful for your
willingness to participate in this project.
Sincerely,
Hoke S. Howard
Project Coordinator
SAVANNAH RIVER
-------
SAVANNAH RIVER
REGIONAL ENVIRONMENTAL MONITORING AND ASSESSMENT PROGRAM
ACCESS TO STREAM MONITORING SITES
I, , owner or
representative of the owner of property adjacent to
_____ , grant permission to the
staff of the U.S, EPA and state agencies assisting in the
Savannah River Regional Environmental Monitoring and Assessment
Program (SR-REMAP) access to said property for the purpose of
stream monitoring as part of the SR-REMAP.
SIGNATURE
DATE
Sites 183
-------
57^/3^
Picte*.5 do- So
/Yl.fittfc. Jlut/trt Milt. 6t.
SAVANNAH RIVER
REGIONAL ENVIRONMENTAL MONITORING AND ASSESSMENT PROGRAM
ACCESS TO STREAM MONITORING SITES
T r ( cdJ-iA uJ , owner or
representative of the owner of property adjacent to
. grant permission to the
staff of the U.S. EPA and state agencies assisting in the
Savannah River Regional Environmental Monitoring and Assessment
Program (SR-REMAP) access to said property for the purpose of
stream monitoring as part of the SR-REMAP.
0 J,, u)
SIGNATURE V di U U ' -pcytAi
DATE 2 - 7^ 5 &
-------
Appendix C
Sample Direction Packet
-------
Site 138 Pickens Co. SC
This site is located about 4 miles east of Pickens, SC offMeese Mill Rd. (not to be confused with
Reece Mill Rd.)
To reach the site leave Pickens on Reese Mill Rd. as you did for Site 136. Follow Reece Mill Rd. for
about to 3 mi. watching for the Meese Mill Rd. intersection. Turn right onto Meese Mill Rd. and
drive approx. 1.4 mi. to an old mill (red building) on the right side of the road (see photo). There is
a small parking area on the right Immediately after the mill (see inset map) Park From the Mill
walk back uphill to a telephone line cut Enter the stream here and walk upstream about 100' to the
site. The site is not flagged.
-------
LEGEND
~ Geo Feature
US Highway
Population Center
Street, Road
... i . Major Street/Road
¦¦ ¦!.> US Highway
River
Utility (powerline)
Open Water
Contours
Scale 1 37,500 (at center)
2000 Feet
1000 Meters
Mag 14.00
ThuJul 25 15:57:52 1996
-------
C*3
-------
Appendix E
Cramer Von Mises Test for Environmental Data
By-
Stephen L. Rathbun
-------
CRAMER-VON MISES TESTS FOR ENVIRONMENTAL MONITORING DATA
by
Stephen L. Rathbun, Richard Houghton
Department of Statistics
University of Georgia
Athens, Georgia
Christina Laurin
FTN Associates, LTD
3 Innwood Circle, Suite 220
Little Rock, Arkansas 72211
-------
1 Introduction
One of the objectives of both the Savannah River Initiative and the South Florida Initiative
is to detect trends in important environmental variables (e.g., algal growth potential test,
mercury) over both time and "space. Both of these environmental monitoring initiatives are
conducted under the auspices of the Regional Environmental Monitoring and Assessment
Program (REMAP), and employ the probability sampling design of the Environmental Mon-
itoring and Assessment Program (Overton, White, and Stevens 1991) on a regional scale.
Observations are collected over time and space in a serially alternating design (Urquhart,
Overton, and Birkes 1993) that is spatially interpenetrating (Overton et al. 1991). The
region of interest is partitioned into a grid of contiguous hexagonal quadrats, and quadrats
are systematically partitioned into four cycles, corresponding to annual (Savannah River
Initiative) or biannual (wet and dry seasons in the South Florida Initiative) sampling times.
The interpenetrating component of the design comes from assigning two each of the six
neighboring hexagons to the three remaining cycles. Thus, a hexagon assigned to cycle 0,
will have a pair of neighboring hexagons each assigned to cycles 1, 2, and 3. Following
the assignment of hexagons to cycles, sample points are randomly located within hexagons
belonging to each cycle according to some probability sampling design. Then, in each of the
first four sampling intervals, sites assigned to a successive cycles are sampled. This sampling
pattern is then repeated in subsequent groups of four successive sample intervals.
The following considers statistical methods for testing the null hypothesis that the data
0
-------
obtained from two or more cycles are identically distributed. Although this discussion is
couched in terms of comparing observations over time, these methods can also be used to test
the null hypothesis that observations from different subregions are identically distributed.
For example, Section 5.2 considers a test of the null hypothesis that data from different
stream orders are identically distributed.
This paper is restricted to design-based methods of statistical inference. For each cycle,
say t, the sampling units are locations, and the population is comprised of the collection of
locations s in the region At, the set of hexagons assigned to that cycle. For design-based
inference, the value of the variable of interest Zt{s) at a location s € At is assumed to
be fixed, not random. For each cycle, the data Zt(s*i),- • •, Zt(stnt) are obtained from a
probability sampling design, under which the locations stl, • - stn, are sampled with known
probabilities. The simplest example of such a design is the simple random sampling, where
Sti, • • •. s(n, are independently sampled from a uniform distribution on A. For design-based
statistical inference, the source of random variation is the random selection of sample sites.
This is in contrast to model-based statistical inference, where the source of random variation
is in the assumed statistical model (e.g., a regression model). Thus, design-based statistical
inference has the advantage that no model assumptions are required. Design-based statistical
inference for spatial sampling designs, such as employed by REMAP, is introduced by Cordy
(1993), who considers Horvitz-Thompson estimation of population parameters.
There are several approaches that may be taken to comparing 2 or more cycles. Assuming
homogeneity of variance, and that independent simple random samples are obtained from
1
-------
normally distributed populations with identical variances, a one-way analysis of variance
may be used to test the null hypothesis that the population means are identical against the
general alternative that at least one population mean is ditterent. For large sample sizes, the
central limit theorem says that sample means are approximately normally distributed even
if the original data are not; so, the normality assumption can be relaxed. If a nonparametric
procedure is desired, the Kruskal-Wallis test (Hollander and Wolfe 1973) can be used to test
for identical population means.
More generally, we may wish to avoid making any assumptions concerning the forms of
the distributions of the populations we wish to compare. Thus, we may wish to test the
null hypothesis that the populations are identically distributed. Since the distribution of a
population can be characterized through its cumulative distribution function (cdf), this is
equivalent to testing the null hypothesis that the cdf's are identical. There are two general
classes of test statistics for comparing cd/'s. Kolmogorov-Smirnov test statistics are based on
largest absolute differences between cdf's, while Cramer-von Mises test statistics integrate
squared differences between cdf's over the possible values of the variable of interest. Since
the latter looks at differences between cdf's at more than one point, and not just the point
where absolute differences are largest, Cramdr-von Mises tests should be more powerful than
Kolmogorov-Smirnov tests. The large-sample distributions of both classes of test statistics
under simple random sampling were tabulated by Kiefer (1959).
This paper shall consider the application of Cram6r-von Mises tests to the Savannah
River and South Florida data. A Quattro-Pro Template for computing Cramer-von Mises
2
-------
test statistics shall be described. The South Florida Initiative involves data collection in four
"cycles" corresponding to two different seasons (wet and dry) over a two year period. An
"Analysis of Variance" analog shall be developed for partitioning variation in cdf's between
cycles into sources of variation due to year, season, and year by season interaction. Kiefer's
large-sample results assume that simple random samples are obtained from each population,
but neither the Savannah River nor the South Florida initiatives use simple random sampling
designs. Results of a simulation study will be presented to investigate the distribution of
the test statistic under sampling designs used by the two initiatives.
2 Cumulative Distribution Function
The distribution of data obtained from cycle t can be characterized through its cumulative
distribution function (cdf). Since cycles are systematically assigned to hexagons, different
regions are sampled in each cycle; let At denote the region sampled in cycle t. For south
Florida marshes, At corresponds to that portion of the hexagons assigned to cycle t that
are in marshlands, while for South Florida canals, and Savannah River Basin streams, At
denotes the portion of the hexagons in these waterways. Let Zt{s) denote the variable of
interest at location s in region At. Then the cumulative distribution function for Zt(-) is
defined to be
(1)
3
-------
where the indicator function I{Zt(s) < z) is equal to one if Zt(s) < z and is equal to zero
if otherwise, and |^4t| is the area of region At. For rivers, streams, and canals, the integral
is over the lengths of these waterways, and |i4t| becomes the total length of waterways in
region (cycle) t. The function Ft(z) can be interpreted as the portion of the area (length)
of region At for which the variable of interest takes values less than or equal to z.
Since it is not possible with a finite budget to observe Zt(s) at all locations s € At, the
population cdf Ft(z) is unknown and must be estimated from a sample. Let zt 1, •••, ztn, denote
the values of the variable of interest at the nt sites sampled at time t, and let ^(t), - • •, 7rn,(t)
denote the corresponding inclusion probability densities. The inclusion probability ir^t) is
defined to be equal to the likelihood that the t-th site is included in the sample at time t.
Then the cumulative distribution function at time t may be estimated using the Horvitz-
Thompson estimator
£W = iTiLJ{%7/} <2>
(Cordy 1993). If sampling probabilities are equal (within cycles), as in the case of South
Florida canals, (2) reduces to
am-r £'{***>• (3)
The variance of /J (2) may be estimated by
where »ry(t) is the pairwise inclusion probability dexisity that both sites t and j are included
in the sample at time t. If sample sites are selected according to a simple random sampling
4
-------
design, the variance of Ft(z) is given by
v,(t) = *('>{'-f<*» (5)
Tit — 1
More generally, if sample sites are selected independently with inclusion probabilities *", (£),
then the variance of Ft{z) may be estimated by
The (emprical) cdj Ft{z) is most readily interpreted by plotting Ft(z) against z. Figure
1 shows a plot of the cdf for total mercury in water in cycle 0 of the south Florida canals.
Notice that the cdf is a step function (solid line) ; each step occurs at the location of a data
point so that all of the information contained in the data is retained. The curve is steepest
at low levels of total mercury indicating that about 40% of the values lie below 1.2 ng/£, and
an additoinal 45% of the values lie between 1.2 and 4.5 ng/£. The curce is very shallow for
large values of total mercury indicating that only 15% of the values lie above 4.5 ng/£. The
dashed lines give 95% confidence bands for Ft(z). Notice that these bands are widest for
intemediate values of total mercury and that the width of these bands converges to zero as
total mercury decreases towards the smallest observed value, or increases towards its largest
observed value.
Although V\ (t) is an unbiased estimator for the variance of it is unstable and can
sometimes take negative values. Moreover, Vi(t) requires values for the pairwise inclusion
probability densities ttu(£), which cannot be easily obtained for either South Florida canals
or Savannah River Basin streams. Thus, two ad hoc procedures may be considered:
-------
Procedure 1: If inclusion probability densities are identical, then we might treat the data
as if it came from a simple random sampling design, and the variance of Ft(z) might be
estimated using Vi(t). For South Florida canals, this might be justified as follows: Here
canals are partitioned and placed in random order. Then sites are placed along the ran-
domized canal segments according to a systematic design. Now consider partitioning canals
into larger and larger numbers of smaller and smaller segments. If we assume that canal
segments are placed in completely random order, then as the length of the smallest canal
segment converges to zero, the distribution of the sample sites converges to that of a simple
random sampling design. Thus, assuming that canal segments are partitioned finely enough,
tlje sample sites can be treated as if they came from a simple random sampling design. Note,
however, that canal segments are not placed in completely random order, but according to
a clustered sampling design in which clusters of canals are placed in random order, and then
locations of canals are randomized within clusters. This was done to achieve better spatial
coverage of sample sites. I; also has the consequence that Vtit) should over-estimate the
variance of Ft(z), as confirmed by results of simulations to be described later. A better
estimate of the variance of Ft(z) might be achieved using the post-stratification estimator
given by Procedure 2.
Procedure 2: In the Savannah River Basin, one might expect differences in environmental
variables between different orders of streams. Likewise, there may be differences among
and between the various water conservation areas, Big Cypress National Preserve, and the
Everglades National Park in South Florida. Similar differences may also exist between
6
-------
different canal reaches in South Florida. In response to this heterogeneity in environmental
conditions, the respective sample regions can be partitioned into strata corresponding to
orders of streams in the Savannah River Basin, water conservation areas and parks for
south Florida marshes, and canal reaches in South Florida. The sample designs used in
both environmental monitoring initiatives further lend themselves to this post-stratification,
since, within each cycle, sampling probability densities are constant within these strata.
Let ZtM denote the data from sample site i in stratum h in cycle t, let n* denote the
number of observations from stratum h, and let L denote the number of strata. Then the
cumulative distribution function in stratum h and t is estimated by
1 n,h
Fth(z) = < *}, (7)
nth :=1
and the variance of Fth{z) is approximately
Vth = (8)
The cumulative distribution function for the population in cycle t is then given by expression
(2), but an approximate expression for its variance is given by
™-E5P§(Sv*- <•'
where irth is the inclusion probability density for sites in stratum h in cyle t. This approxi-
mation is based on treating sample sites as if there generated by a simple random sampling
design within strata.
7
-------
3 Cramer-von Mises Test
Consider testing the null hypothesis that the cumulative distribution functions (cdf's) of
two or more populations are identical against the general alternative that at least one of the
cdf's is different. A Cramer-von Mises test statistic may be defined as
= /-"EL, n,\F,(z) - P(z)fdP(z)
where
/•(*) «.-!>£(*). <">
and n = nj + • • • + n*. The function F(z) can be interpreted to be the average cdf over
all cycles. The right-hand side of (10) involves integration of the squared difference between
the cdf's Ft(z) and the average cdf F(z) over all possible values of the variable z. Thus,
cdf's are compared over the whole range of the variable, not just at a single point as in the
A--sample Kolmogorov-Smirnov test, yielding improved power when compared to the latter
testing procedure.
The large-sample distribution of W was derived by Kiefer (1959), under the assumption
of simple random sampling from all populations. Table 1 gives the critical values for a-level
tests of Ha : Fi = F? = ¦¦¦ = Fk. Thus, an a * 0.05 level test for equality of cdf's over
k as 4 cycles would reject H0 ifW> 1.2373.
As indicated above, the critical values in Table 1 are obtained under the assumption
that simple random samples are obtained from each of the fc-populations. However, none
8
-------
of the data considered here are obtained from simple random samples. Results of Monte
Carlo simulations (Section 6) suggest that the critical values in Table 1 are conservative in
the sense that we may fail to reject the null hypothesis when it is false less often than we
should. Likewise, under the null hypothesis, the true a level is less than what is tabulated
in Table 1.
The South Florida Initiative involves the collection of data over two seasons (wet and
dry) over a two-year period. If we reject the null hypothesis that the cdf's over the four
cycles, then the next step would be to ask if there are significant differences between years,
between seasons, or if there is an interaction between years and seasons. Since the integrand
on the right-hand side of expression (10) takes the form of a sum of squares over all fc-cycles,
we can partition W into terms for testing differences between years, seasons, and interactions
between years and seasons. Assume that the data are balanced; that is, n\ = = • • • = n*;
let n denote the common sample size for all seasons. Suppose observations are collected over
a years and b seasons. Let zijk denote the observation from sample k in season j in year z,
let Ftj(z) denote the cdf for season j in year z, let
*•(*> = iESjM
(12)
denote the average cdf for year i, let
= jM
u
(13)
9
-------
denote the average cdf for season j, and let
ao
tsl J*=l
denote the average cdf over all seasons and years. Then, we may partition the Cramer-von
Mises test statistic as follows?
W = Wy + W, + Wyx. , (15)
where
(16)
(17)
(18)
= i e?., e;=,
tests for variation between cycles,
W'v =
= £ E?., E?«, EI,, l£.(**) - f-Ml'
tests for variation between years,
w. = n0rocE,'.,|f.J»-A.WP
-------
If a = b = 2, then we may also compute
W.», = n ;r„ - ~P-U)f - - F-wf j <&¦¦(*)
(20)
The a-level tests for the various null hypotheses may then be tested as follows:
• The null hypothesis that there are no differences between cycles is rejected if W > Watab.
• The null hypothesis that there are no differences between years is rejected if Wy > W0,a.
• The null hypothesis that there are no differences between seasons is rejected if \Vt >
Wa, 6-
4 Analysis of cdf's using Quattro Pro
The QuattroPro template CDFTEST2.WB1 can be used to calculate cumulative distribution
functions for environmental monitoring data and 95% confidence intervals assuming simple
random sampling. The template is set up to use three elements of the data; the station name,
a grouping variable (e.g., cycle, year, season, stratum), and the variable to be analyzed. The
grouping variable must be a numerical variable with values between 0 and 9 yielding a
possible 10 groups. Currently the template is limited to 500 total observations. Before
using it, a working copy of the template CDFTEST2.WB1 should be created, and a backup
copy should be kept in case the working copy is damaged. The template file is rather large,
therefore we advise that a new working copy be created for each new variable analyzed
11
-------
instead of putting multiple analyses in a single file on multiple pages. Large Quattro Pro
files become very slow to work with and are prone to causing memory errors.
A step by step procedure for using the Quattro Pro template is outlined as follows:
Step 1: Start Quattro Pro and open the Quattro Pro data file that contains the data
for the variable of interest.
Step 2: Open the working copy of the Quattro Pro template file CDFTEST2.WB1.
Step 3: Go into the data file, copy the required columns (station ID, grouping variable,
variable of interest), and paste them into page B of the template file. If sampling probabilities
are not equal, they also must be copied and pasted into page B of the template file. Grouping
variables may include, e.g., cycles, years, seasons, or stream orders, and must take numerical
values between 0 and 9. On page B of the template file, any extra data or lines containing
missing values for the variable of interest must be deleted using block delete.
Step 4: Make sure the data in the grouping variable column, variable column, and
sampling probability column are numerical data instead of alphanumeric. To convert al-
phanumeric data to numeric data, mark the columns to be changed, and perform a Search
and Replace on the label indicator (', ", or *), replacing these symbols with a plus sign (+).
Search and Replace is either under EDIT in the main menu, or Block (Quattro Pro 6.0).
Step 5; Sort the data based on the variable of interest. Use the mouse to mark all of the
data in the columns for station ID, grouping variable, variable of interest, and if necessary,
sampling probability. Do not mark the column titles. The Sort option is under DATA in
Quattro Pro 5.0 and under Block in Quattro Pro 6.0. Click in the first box in the Sort Keys
-------
section of the Sort box. Then on the notebook page, click on the top of the column with the
variable of interest to select the entire column as the sort key. Click OK and then the data
will be sorted.
Step 6: Copy the sorted data into the corresponding columns of page A using Copy and
Paste. Do not select the entire column to copy to page A, select only the block of data. On
page A, the data starts in the third row; i.e., A3. Once the data is copied, the calculations
and graphs are complete. The estimated cdf's for each group are found in columns P to Y,
and corresponding 95% confidence intervals appear in columns AK to BD. The average cdf
appears in column Z. The Cramer-von Mises statistic for testing the null hypothesis that
groups have identical cdf's can be found at AL505.
Step 7: Save the completed template file under a new name.
Step 8: Graphs of the first four cdf's and their corresponding 95% confidence bands are
created in the completed file. Before these are printed, titles need to be modified to reflect
the parameter being analyzed, and how they are grouped. Pull down Graph on the menu
and select Edit Graph. Select the graph you wish to edit and a graph box will appear on
the screen. Once the graph is open, click on the Graph menu again and select Titles. The
variable, media, and source are specified in the Main Title. The group is specified as the
Subtitle. Make sure the correct variable and units are specified in the Main Title and the
axes titles. Once the necessary corrections are complete, click OK.
Step 9: To print a graph, open it (using Edit Graph or some other method) and then
select File/Print in the main menu.
-------
Step 10: If changes to the graphs are made, be sure to save the file again after they are
complete.
Step 11: Close the file, reopen the template file and start again with the next variable
of interest.
5 Examples
The following illustrates the use of the Cramer-von Mises test on data from both the Savan-
nah River Initiative and the South Florida Initiative.
5.1 South Florida Canals
Data on a wide variety of physical variables were obtained from water and sediment samples
collected from canals in southern Florida during the wet and dry seasons over a two-year
period (fall 1993 to spring 1995). Fifty observations were available from each sampling cycle.
The following considers temporal variation in total mercury and methyl mercury in water
samples. To test for interaction between seasons and years, a grouping variable was defined
to be equal to 0 if the observation was from cyles 0 or 3 (wet season year one or dry season
year two, respectively), and equal to 1 if the observation was from cycles 1 or 2 (dry season
year one or wet season year two, respectively).
Figure 2 depicts the cdf's (solid lines) and corresponding 95% confidence bands (dashed
lines) for methyl mercury in cycles 0, 1,2, and 3. Note that the curves rise more gradually
14
-------
in cycles 0 and 3 than in cycles 1 and 2. This indicates that water samples from the former
cycles not only tend to contain higher concentrations of methyl mercury, but also that methyl
mercury readings show greater variability in these two cycles. Moreover, since cycle 0 was
carried out in the wet season of the first year, while cycle 3 was carried out in the dry season
of the second year, this pattern to temporal variation in cdf's suggests a strong interaction
between year and season. This result is confirmed by the Cramer-von Mises tests (Table 2).
First the null hypothesis that the four cdf's are identical is firmly rejected (W = 4.69; k = 4;
p < 0.0001) with W = 4.69. The partitioning of W into terms for years, seasons, and year
by season interaction indicates that while season accounts for most of the variation between
cycles (W, = 3.11; k = 2; p < 0.0001), there is a significant interaction between year and
season (Wyx, = 1.31; k = 2; p = 0.003). However, there is no evidence for a significant
difference between years (U''y = 0.28; k = 2; p = 0.494).
Figure 3 depicts the cdf's (solid lines) and corresponding 95% confidence bands (dashed
lines) for total mercury. The curves for the wet season (cycles 0 and 2) rise more gradually
than those for the dry season (cycles 1 and 3). This suggest the total mercury concentrations
are not only higher in the wet season, but they also show greater variability. The Cramfr-
von Mises test W gives strong evidence that four cdf's are not identical {W as 7.24; k ~ A\
p < 0.0001). The partitioning of W suggests that the year by season interaction accounts
for most of the variation between cycles {Wyx, = 4.93; k = 2; p < 0.0001). This interaction
appears since total mercury concentrations are higher in the wet season of the first year than
in the wet season of the second year, while the two dry seasons appear to be similar to one
-------
another. Moreover, there are significant differences between years and between seasons.
5.2 Savannah River Basin Streams
Data on a number of physical and biotic variables were obtained from the streams of the
Savannah River Basin over two sampling cycles. The following considers temporal variation
and variation between stream orders for the variables Algal Growth Potential Test (AGPT),
a habitat code (Habitat), and a condition code (Condition).
Figure 4 depicts the cdf's (solid lines) and corresponding 95% confidence bands (dashed
lines) for AGPT for the two cycles. The curve for cycle 0 rises more gradually than that
for cycle 1 suggesting that mean AGPT is not only higher in cycle 0, but also shows greater
variability within this cycle. The Cramer-von Mises test W (Table 3) confirms that the
difference between the two cycles is statistically significant (W — 1.78; k = 2;p < 0.0001). A
comparison of cdf's between cycles for Habitat and Condition reveals no significant difference
between cycles (Table 3), as can be seen from Figures 5 and 6, respectively.
Comparing cdf's for AGPT across stream orders reveals third order streams tend to
have higher and more variable values of this variable than first or second order streams
(Figure 7). The Cramer-von Mises test indicates that this difference between stream orders
is statistically significant (W — 1.31; k = 3; p = 0.012). Likewise, first order streams lend
to have lower condition codes than higher order streams (Figure 9); this difference is also
statistically significant (W = 1.71; k = 3; p = 0.002). In contrast there is no significant
difference between cdf's for Habitat for different orders of streams (W = 0.58; k = 3;
-------
p = 0.294); see Figure 8 for the cdf plots.
6 Monte Carlo Simulations
The Q-level critical values Wa* for the Cramer-von Mises test appearing in Table 1 are com-
puted under the assumption of simple random sampling from the respective populations.
However, neither the Savannah River Initiative nor the South Florida Initiative employ sim-
ple random sampling designs. Moreover, the tabulated critical values assume that we have
a large sample. This section employs Monte Carlo simulations to investigate the distribu-
tion of the Cramer-von Mises test statistic under the designs employed by these initiatives.
The general approach to carrying out the Monte Carlo simulation consists of simulating one
realization of a Gaussian random field (defined below) with a given range of spatial correla-
tion. Then 1000 independent samples of n sites per cycle are obtained from the realization
according to a remap sampling design. For each sample, the 4-population Cramer-von Mises
test statistic is computed; let W* denote the value of the Cram6r-von Mises test statistic
for the i-th sample. Finally, the VV^'s are ranked from smallest to largest. The simulated
critical value for an o-level test is then given by (1 - a) • 1000th ranked value of W*. The
proportion a of the W/'s falling above the tabulated Q-level critical value Wa was also com-
puted. The tabulated values are conservative if the simulated critical values fall below Wa,
or, equivalently if S < a.
A stationary Gaussian random field is one of the simplest models for generating random
17
-------
functions Z{s) of spatial locations s in a region A. Assume that
*(•)«#. + «(•), (21)
where y. is the population mean, and c(s) is a zero-mean normally-distributed error. The
spatial dependence between data at locations 8 and u is modeled through the covariance
function
C(lls — u||) s cov{Z(s) - Z(u)}, (22)
which we shall assume to be a function of only the distance ||s - u|| between the two sites.
Simulations are carried out under an exponential covariance function model
C(r) = a2e~v' (23)
using the spectral method described by Shinozuka (1971) and Mejia and Rodriguez-Iturbe
(1974). For the exponential covariance, the range of spatial correlation is defined to be equal
to 3/7. The results of simulations for the Cramer-von Mises test statistic do not depend on
fi and a2. So, without loss of generality, all simulations shall be carried out under n = 0 and
-------
the canals into short segments, that were placed into random order as follows: Hexagons
were systematically partitioned into cycles, corresponding to the biannual sampling times
used in south Florida, in such a manner that for each hexagon assigned to a given cycle,
two of its six neighboring hexagons are assigned to each of the three remaining cycles. To
ensure uniform spatial coverage of the study region, hexagons were also partitioned into 22
clusters. Then for each simulated realization, the clusters were randomly ordered, and canal
segments were randomly ordered within clusters. Finally, the ordered canal segments were
strung together, and a systematic random sample of 50 sites was located along the strung
segments. This was accomplished by picking a random starting point between 0 and £/50 km
and sampling every £/50 km thereafter, where £ is the total length of the strung segments.
The results show that the sample proportions of simulated Cramer-von Mises test sta-
tistics greater than the tabulated a-level critical values fall well below corresponding values
of q (Table 4). Moreover, simulated estimates of a-level critical values all fall well below
the tabulated critical values. These results indicate that the tabulated critical values are
conservative in the sense that we will reject null hypothesis of equality of cdf's less often
than we should. Moreover, there is some suggestion that the test becomes more conservative
with increasing range of spatial correlation.
6.2 South Florida Marsh
Simulation of the design used to sample the south Florida marshes was carried out on a
22 x 22 grid of contiguous hexagons whose centers are one unit apart. Hexagons were
19
-------
systematically partitioned into cyeles, as described above, yielding a total of 121 hexagons
per cycle. This is comparable to the sampling effort used in the south Florida marsh. One
random sample sites was then located in each of the hexagons.
The results show that the sample proportions of simulated Cram£r-von Mises test statis-
tics greater than the tabulated a-level critical values fall well below corresponding values of
a, and that simulated estimates of a-level critical values fall well below the tabulated critical
values (Table 5). These results indicate that the tabulated critical values are conservative in
the sense that we will fail the null hypothesis of equality of cdf's less often than we should.
Moreover, the test becomes more conservative with increasing range of spatial correlation,
as indicated by decreasing simulated a-levels with increasing correlation.
7 Conclusions
The Cramer-von Mises test is a powerful procedure for testing the null hypothesis that two or
more populations are identically distributed; that is, have identical cumulative distribution
functions. This null hypothesis can be rejected if any feature of the distributions, including
the means, variances, or even their shapes, varies significantly between the populations.
Thus, the alternative hypothesis under a Cram6r-von Mises test is very general. The Cram£r-
von Mises test is a nonparametric test; so, no distributional assumptions (e.g., normality,
homogeneous variances) are required. However, this comes at the cost of loss of power in
comparison to tests designed to detect specific alternatives. For example, the F-test in an
20
-------
analysis of variance is more powerful for detecting differences between population means,
assuming identical population variances.
The results of the Monte Carlo simulations suggest that Keifer's tabulated critical values
are conservative in the sense that this test will reject the null hypothesis of equality of
cdf's less often as it should. However, application of this test indicates that it has sufficient
power to detect trends over time in the canals of southern Florida and the streams of the
Savannah River Basin, and sufficient power to detect differences between stream orders in
the Savannah River Basin.
The partitioning of the Cramer-von Mises test statistic into terms for year effects, season
effects and year by season interaction can be extended to any factorial arrangement of factors
(e.g., strata, habitat types, etc.) provided that the data are balanced; that is, there are
identical numbers of observations in each of the factorial combinations. If the data are not
balanced, then the results of hypothesis tests can become ambiguous with regards to their
interpretations. For example, if one year has more wet season sample stations than another
year, a test based on Wv for differences between years might be rejected if their are signficant
between seasons, regardless of whether or not there are significant differences between years.
Further research is required to test for differences between years after differences between
seasons are taken into account.
REFERENCES
21
-------
Cordy, C.B. (1993). An extension of the Horvitz-Thompson theorem to point sampling
from a continuous universe. Statistics and Probability Letters, 18, 353-362.
Hollander, M., and Wolf, D.A. (1973). Nonparametric Statistical Methods. Wiley, New
York.
Kiefer, J. (1959). /f-sample analogues of the Kolmogorov-Smirnov and Cramer-V. Mises
tests. Annals of Mathematical Statistics, 30, 420-447.
Mejia, J.M., and Rodriguez-Iturbe, I. (1974). On the synthesis of random field sampling
from the spectrum: An application to the generation of hydrologic spatial processes. Water
Resources Research, 10, 705-711.
Overton. W.S., White, D., and Stevens, D.L. (1991). Design Report for EMAP, the En-
vironmental Monitoring and Assessment Program. Washington, D.C., U.S. Environmental
Protection Agency (EPA/600/3 - 91/053).
Shinozuka, M. (1971). Simulation of multivariate and multidimensional random processes.
Journal of the Acoustical Society of America, 49, 357-367.
Urquhart, N.S., Overton, W.S., and Birkes, D.S. (1993). Comparing sampling designs for
ecological status and trends: impact of temporal patterns, in V. Barnett and K.F. Turkman
(eds.), Statistics for the Environment, Wiley, New York.
22
-------
Table 1. Critical values of a-4evel Cramer—von Mises tests of H0: F| = F2 = • • • Fk
against the general alternative that at least one cdf is different. This table is extracted
from Kiefer (1959).
k
a
2
3
4
5
0.75
0.18545
0.31472
0.45103
0.59161
0.50
0.27757
0.44138
0.60668
0.77253
0.25
0.42098
0.62227
0.81775
1.00947
0.20
0.46640
0.67691
0.87980
1.07785
0.15
0.52481
0.74592
0.95734
1.16268
0.10
0.60704
0.84116
1.06311
1.27748
0.05
0.74752
1.00018
1.23730
1.46466
0.02
0.93320
1.20561
1.45913
1.70028
0.01
1.07366
1.35861
1.62263
1.87215
0.005
1.21412
1.51010
1.78345
2.03935
0.001
1.54027
1.85773
2.14949
2.40774
0.0001
2.00691
2.34950
2.66130
2.82500
-------
Table 2. Cramer von Mises tests for equality of cumulative distribution functions for South
Florida canals.
Source
Total Mercury in Water
W
Years
Seasons
Years x Seasons
1.01
1.31
4.93
0.014
0.003
<0.0001
Total
7.24
<0.0001
Source
Methyl Mercury in Water
W
Years
Seasons
Years x Seasons
0.28
3.11
1.31
0.494
<0.0001
0.003
Total
4.69
<0.0001
-------
Table 3. Cramer-von Mises tests for equality of cumulative distribution functions for
Savannah River Basin streams. Equality of cumulative distribution functions between
cycles and between stream orders is tested.
Cycles Stream Orders
Variable W P W P
AGPT 1.78 <0.0001 1.31 0.012
Habitat 0.27 0.518 0.58 0.294
Condition 0.12 0.929 1.71 0.002
-------
Table 4. Simulated critical values W* for 4-sample a-level Cramer-von Mises tests under
different ranges of spatial correlation in the south Florida canals, and corresponding
proportion a of simulated Cramer—von Mises test statistics greater than a-ievel critical
yalues. Tabulated critical values from Kiefer (1959) are given at the bottom.
a
0.1 0.05 0.01
Range (km)
A
a
W£
A
a
w*
a
0
0.8341
0.048
0.9903
0.017
1.3393
0.001
10
0.6887
0.013
0.8085
0.006
1.1559
0.001
20
0.8037
0.032
0.9467
0.011
1.2953
0.001
30
0.5913
0.002
0.6731
0.001
0.8573
0.000
60
0.6911
0.009
0.7773
0.002
1.0023
0.000
120
0.7959
0.019
0.9117
0.005
1.1885
0.000
tabulated
1.0631
1.2373
1.6226
-------
Table 5. Simulated critical W* values for 4—sample ot-level Cramer—von Mises tests under
different ranges of spatial correlation in the south Florida marsh, and corresponding
proportion a of simulated Cramer-von Mises test statistics greater than a-level critical
values. Tabulated critical values from Kiefer (1959) are given at the bottom.
a
0.1 0.05 0.01
Range (km)
W£
A
a
*
a
W£
*
a
0.5
0.8094
0.029
0.9564
0.014
1.2906
0.002
1.0
0.8279
0.040
0.9885
0.013
1.3666
0.002
2.0
0.5819
0.002
0.6981
0.002
0.9355
0.000
4.0
0.4557
0.000
0.5171
0.000
0.6233
0.000
8.0
0.3096
0.000
0.3489
0.000
0.5604
0.000
tabulated
1.0631
1.2373
1.6226
-------
FIGURE LEGENDS
Figure 1. Cumulative distribution function (solid line) for total mercury concen-
tration in water samples collected from south Florida canals in cycle 0. The
corresponding 95% confidence bands are given by the dashed lines.
Figure 2. Cumulative distribution functions (solid lines) for methyl mercury
concentration in water samples collected from south Florida canals in cycles
0, 1,2, and 3. The corresponding 95% confidence bands are given by the
dashed lines.
Figure 3. Cumulative distribution functions (solid lines) for total mercury con-
centration in water samples collected from south Florida canals in cycles
0, 1,2, and 3. The corresponding 95% confidence bands are given by the
dashed lines.
Figure 4. Cumulative distribution functions (solid lines) for Algal Growth Poten-
tial Test in samples collected from Savannah River Basin streams in cycles
0 and 1. The corresponding 95% confidence bands are given by the dashed
lines.
Figure 5. Cumulative distribution functions (solid lines) for Habitat Score in
samples collected from Savannah River Basin streams in cycles 0 and 1.
The corresponding 95% confidence bands are given by the dashed lines.
Figure 6. Cumulative distribution functions (solid lines) for Condition in sam-
ples collected from Savannah River Basin streams in cycles 0 and 1. The
corresponding 95% confidence bands are given by the dashed lines.
-------
Figure 7. Cumulative distribution functions (solid lines) for Algal Growth Poten-
tial Test in samples collected from Savannah River Basin streams of orders
1,2, and 3 (top to bottom). The corresponding 95% confidence bands are
given by the dashed lines.
Figure 8. Cumulative distribution functions (solid lines) for Habitat Score in
samples collected from Savannah River Basin streams of orders 1, 2, and 3
(top to bottom). The corresponding 95% confidence bands are given by the
dashed lines.
Figure 9. Cumulative distribution functions (solid lines) for Condition in samples
collected from Savannah River Basin streams of orders 1, 2, and 3 (top to
bottom). The corresponding 95% confidence bands are given by the dashed
lines.
2
-------
c d f
1 . 0
n
M
0
u
7
Y\
D
A
i
3
?
0 . 1
0.0
I I I I I I II I 1 I 1 I I
0 2
4 6
T 0 t Q
T—I—I I I i I—I—I—I—r
8 10
Me r c u r
"I I I |
Figure 1
-------
cycle 0
cycle 2
cdf
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1-1
O.Ov
0.0 0.2 0.4 0.6 0.8 1.0
Methyl Mercury
1.2 1.4
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
Methyl Mercury
cycle 1
cycle 3
cdf
l.Oi
0.9i
.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
0.4 0.6 0.8 1.0 1.2 1.4
Methyl Mercury
Methyl Mercury
Figure 2
-------
cycle 0
cycle 2
4 6 8 10 12 14 16
Total Mercury
cdf
1.0
0.9
0.8
0.4
0.3
0.2
0.1
0.0
4 6 8 10 12 14 16
Total Mercury
cycle 1
cycle 3
8 10 12 14 16
4 6
Total Mercury
cdf
1.0
0.
0,
0,
0,
0.5
0.4
0.3
0.2
0.1
0.0
4 6 8 10 12 14 16
Total Mercury
Figure 3
-------
cycle 0
20 30 40
AGPT
50 60
cycle 1
cdf
1.01 j __
0.9'
o.8- !>f.J
0.7- rT
0.6 jfj
0.5 :/•
o.4- /:
0.3 /
0.2 •'
o.i /
ftnV ¦ '¦
o 10 20 30 40 50 60
AGPT
Figure 4
-------
cycle 0
cdf
0.2
T
0 20 40 60 80 100 120
Habitat
cycle 1
cdf
20 40 60 80 100 120
Habitat
Figure 5
-------
cycle 0
cdf
0.7
T
T
i - • * ¦ r 1 ' 1 ¦ • ¦ ¦ mm * » ^
50 100 150 200 250
Condition
cycle 1
cdf
0.4
T
T
1 1 ,
50 100 150 200 250
Condition
Figure 6
-------
0.4"1
Order 1
0.0
Order 2
Order 3
AGPT
Figure 7
-------
cdf
1.0-
0.
0,
0.
0,
0
9-
e
7
6
5
0.4
0.3
0.2
0.1;
o.o-
I
cdf
1.
0,
0,
0,
0
0
0
9
8
7
6
5
0.4
0.3
0,
0,
0
2
1
o-
cdf
1.
0.
0,
0
0
0
o-
9"
8"
7-
6
5
0.4
0.3
0.2
0.1
0.0
20
40
60
' ' I 1 1
80
100 120
20 40 60 80 100 120
20 40 60 80 100 120
Habitat
Order 1
Order 2
Order 3
Figure 8
-------
cdf
Order 1
50 100 150 200
250
---r,
Order 2
50
100 150
DO
250
Order 3
100 150 200
Condition
250
Figure 9
-------
-------
Appendix F
Savannah River Basin Landscape Analysis
By
Chaloud et al.
-------
Savannah River Basin
Landscape Analysis
by
Deborah J. Chaloud, Curtis M. Edmonds, and Daniel T. heggem
U.S. Environmental Protection Agency
Office of Research and Development
National Exposure Research Laboratory
Environmental Sciences Division
Landscape Ecology Branch
Las Vegas, Nevada
-------
Savannah River Basin
Landscape Analysis
Deborah J. Chaloud, Curtis M. Edmonds, and Daniel T. Heggem
U.S. Environmental Protection Agency,
Office of Research and Development, National Exposure
Research Laboratory, Environmental Sciences Division,
Landscape Ecology Branch, Las Vegas, Nevada
INTRODUCTION
Scientists from the U.S. Environmental Protection Agency (EPA),
Region 4, Science and Ecosystem Support Division, enlisted the
assistance of the landscape ecology group of U.S. EPA, Office of
Research and Development (ORD), National Exposure Research
Laboratory, Environmental Sciences Division (ESD), in conducting
a landscape assessment of the Savannah River Basin (Figure 1) as
part of their ongoing Regional Environmental Monitoring and
Assessment Program (REMAP) demonstration project. In the Scope
of Work provided by Region 4, the goal was stated as "provide
technical/scientific assistance ...to EPA Region 4 in assessing
current wadeable stream conditions in the Savannah River Basin
with landscape factors that may be contributing to these
conditions or gradients." Three specific objectives were
presented in the form of questions. These were:
Are both the proportions of land uses and the spatial
pattern of land uses important for characterizing and
modeling stream condition in watersheds/ecoregions of
different areas?
Can land uses near the streams better account for the
variability in ecological condition than land use for
the entire watershed/ecoregion?
Does the size of the watershed/ecoregion influence
statistical relationships between landscape
characteristics and ecological condition?
In addition, an assessment of landscape change was to be
conducted as part of continuing ESD research in application of
change detection techniques.
1
-------
South Carolina
Georgia
North Carolina The data analysis plan developed
to address the objectives given
above called for calculation of
a specific suite of landscape
indicators for all nine United
States Geological Survey (USGS)
8-digit hydrological unit codes
(HUC; USGS, 1982), a selected
subset of the 94 Georgia and
South Carolina subbasins, and
the riparian corridors in the
HUCs and selected subbasins.
The subbasins are generally
equivalent in area to USGS 11-
digit HUCs. The riparian
corridor was defined as 100
meters on either side of stream
arcs; this size was selected
from a review of state laws and
literature available on the
Internet (e.g., Santa Cruz
County, 1998; U.S. EPA, 1998;
South Carolina Department of
Natural Resources, 1998). The
suite of indicators included:
landcover types, u-index,
agriculture on slopes greater
than 3 percent, agriculture on
highly erodible soils,
agriculture on moderately
erodible soils, agriculture on highly erodible soils with slopes
greater than 3 percent, number of occurrences of roads crossing
streams, and number of impoundments. Landscape indicator
statistics were also computed for the drainage areas and
associated riparian corridors of a selected set of sites sampled
by Region 4 using REMAP protocols. Region 4 provided an ARC/INFO
coverage of the sampling locations and Quattro Pro spreadsheets
of the water quality and biotic measurements.
Figure 1. Savannah River Basin
METHODS
The selected landscape indicators are identical to, or based on,
indicators used in the mid-Atlantic atlas (Jones et al, 1997).
In the atlas, the indicators were calculated only for 8-digit
HUCs; in this study, indicators are additionally calculated for
smaller spatial units. The basic methodology is the same,
however. In general, calculation of the landscape indicators
involves ARC/INFO techniques of extracting or "cookie cutting"
2
-------
the desired area from a spatial data set. The data are formatted
in an ARC/INFO grid of uniform cell size. In this study, a 30-m
cell size is used for all grids. For indicators which are
produced from more than one data set (e.g., roads crossing
streams), ARC/INFO overlay and intersection techniques are used.
A few indicators, used only on the drainage areas of the
individual sampling sites, are produced from an in-house custom
statistics program. These are indicators of fragmentation, i.e.,
the degree to which landcover types are present in patches rather
than in continuous, homogenous blocks. The landscape change
indicator is produced from comparison of satellite imagery from
two dates. This is the only indicator which does not use
ARC/INFO as the primary data analysis software. Landscape change
assessment employs ENVI, an image processing software package
available for PC or Unix systems.
Data Sets Used
The spatial data sets used are obtained from a variety of
sources. The primary data sets used in this landscape assessment
include: Multi Resolution Land Characteristics (MRLC) Interagency
Consortium land cover/land use (Bara, 1994), State Soil
Geographic data base(STATSGO)soils (Natural Resources
Conservation Service, 1996), RF3 streams (U.S. EPA, 1997), USGS
8-digit HUCs, Georgia and South Carolina subbasins, Region 4
sampling site locational and sampling data, 30-m and 100-m
digital elevation models (DEM; USGS, 1990), digital line graph
(DLG) roads (USGS, 1989), and National Inventory of Dams
impoundments (U.S. Army Corps of Engineers, 1997). Landscape
change assessment used North American Land Characterization
(NALC) imagery from the 1970s and 1990s (U.S. EPA, 1993). Data
sets were subset to the area of interest using the basin boundary
coverage.
Sampling Site Ranking, Selection, and Drainage Area Creation
A simple, unweighted scoring system was used to rank the sampling
sites, shown in Figure 2, by their results. Water quality
variables (pH, dissolved oxygen, conductivity) and biota [algal
growth potential test (AGPT); Ephemeroptera, Plecoptera, and
Trichoptera index (EPT); fish index of biological integrity
(fish_ibi), macroinvertebrate habitat, and macroinvertebrate
richness] were scored separately. The frequency distributions
for each variable was examined. Most indicated a bimodal
distribution, with reduced frequencies near the lower and upper
ends of the variable's range. Measurement values corresponding
to the inflection points of the curve were selected to divide the
range into three classes. A score value was ascribed to the
3
-------
jCy&x '; '/-*> • \ S189
aSHfc .i£S> -ST '&Utifiiia
&%0- -Ac-^'w
k^/Jwww/.
»£*•?
i? MOB-
b«*{;--
sa® S2ta #«"
sa+
•¦V- isias
?*y ;?
Figure 2. Sampling Site Locations
The figures were useful in charaCte
across the basin and making prelimi
further investigation.
rizi
nary
ng
measurement value, 1
for bad, 2 for fair, 3
for good, and 0 for
missing data. Although
these are labeled as
good, fair, and bad,
these terms apply to
the measurement value
compared to the range
of measurement values,
not to any applicable
water quality standards
or other measurement
system. The scores
were summed and
recorded. The number
of measurements used in
the summation was also
recorded; this was
necessary because of
the large number of
sites missing results
for one or more
variables. The
measurement data and
scoring data were then
associated with the
site location coverage.
Map compositions were
prepared for each HUC,
presented here as
figures 3 through 11.
relative conditions
decisions about areas for
The sampling locations had been selected by the Region using the
EMAP site selection protocol. Several discussions and
correspondences were conducted with a lead EMAP Statistician, Dr.
Tony Olson, about the spatial area represented by the sampling
sites. It was determined that it would be necessary to develop
the specific drainage area of each sampling location and to treat
the water quality and biota information as point data. Accurate
drainage area computation requires DEMs of 30-m intervals or
better; at the time of analysis, these were available for only
portions of the Savannah River Basin, primarily the north end and
part of the central area.
4
-------
y.tK
Figure 3. HUC 3060101 Sampling Site Results
5
-------
r
Bioteey Score
Water Quality
15 ffifhwjy
Figure 4. HUC 3060102 Sampling Site Results
-------
Biology Son
WttrQufcy
ISHtfcw
HUC 3060103 Sampling Site Results
-------
BioJwy Score
WacrQurfty
US Hi?h*ay
HUC 3060104 Sampling Site Results
8
-------
Figure 7. HUC 3060105 Sampling Site Results
Biotoiy Scat
ttata Quality
Gaul
Fair
BaJ
total
Dam
Soon
ImerM*
USHigtawv
Rxkad
s
9
-------
Biology Scot
Wjffl-Quatas
LS Hi?hwa>
Rbiiroad
Figure 8. HUC 3060106 Sampling Site Results
10
-------
Rnksy Scat
WurQuiy
I'Sttjhwy
HUC 3060107 Sampling Site Results
-------
Biolofy Score
W ad Qualm
IS Hi gfr*a>
Railroad
Figure 10. HUC 3060108 Sampling Site Results
-------
Figure 11. HUC 3060109 Sampling Site Results
4
Biology Score
1
Wtfer Quality
•
Goal
Far
•
B*1
•
So Dan
A
Dtm
EZ!
Sffran
s
(nwMaic
$ IS Hisduii
2./'. Rjilmad
The process used to delineate the drainage areas employs
hydrological analyses tools contained in the Grid module of
ARC/INFO. First sinks in the DEMs are identified and filled.
Flow direction is computed as the direction from each 30-m cell
towards its steepest downslope neighbor. From the flow direction
grid, a flow accumulation grid is created by calculation of the
number of cells which flow into each downslope cell; this grid
resembles the existing stream network. The sampling station
locations are input as pour points. In some cases, the sampling
point coordinates did not fall directly on a flow accumulation
path; in these instances, the pour point was placed on the flow
13
-------
accumulation in the cell nearest to the given station
coordinates.
In the selection of the subset of sites for landscape indicator
assessment, efforts were made to select sites that met the
following criteria: 1. Full suite of measurement variables, 2.
Located in the areas indicated to be of greatest interest to the
Region, 3. 30-m DEM data available to use in drainage area
determination, 4. Representation of the full range of
measurement values, and 5. Representation of first through third
stream order classes. Using these criteria, sixteen sites were
selected.
3060101
3060103
3060107
3060108
3060106
The selection of
subbasins for
presentation of
landscape
indicators was
made after
selection of the
sampling sites.
The selected
subbasins are
all in HUC
3060103 and each
includes one or
more of the
sampling site
subset. This
provides the
nested hierarchy
of spatial units
in the
assessment. An
arbitrary number
was assigned to
each subbasin
after merging
the separate
Georgia and
South Carolina
3060109 coverages. The
subbasins are
shown in Figure
12.
3060102
3060104
3060105
Figure 12.
HUCs and Subbasins
14
-------
LANDSCAPE ASSESSMENT
HUC Indicators
The Savannah River Basin
is arrowhead-shaped,
trending generally
northwest to southeast.
The basin is comprised of
nine USGS 8-digit HUCs
(numbered 3060101 through
3060109/ hereafter
referred simply by the
last digit), spanning
three ecoregions: Blue
Ridge, Piedmont and
Coastal Plains. As
shown in Figure 13, HUCs
1 and 2 are primarily in
the Blue Ridge ecoregion,
HUCs 3, 4, 5, and 7 lie
in the Piedmont, and the
majority of HUCs 6, 8,
and 9 are in the Coastal
Plain. As shown in Table
1, the size of the HUCs
varies from 200,987.55 ha
(HUC 7) to 488,842.20 ha
(HUC 6). Associated
riparian areas vary from
31,324.14 ha (HUC 7) to
88,651.85 ha (HUC 3),
based on a 100-m
corridor on either side
of all RF3 stream arcs.
Landcover types are
derived from MRLC data,
nominal base year 1992.
Differences among the
three ecoregions are
evident in the forest
landcover statistics
for the HUCs, Table 2.
Deciduous and evergreen
forests predominant in
HUCs 1 through 5 and 7,
Tabl* 1. Physic*! Dj_ra»n«j-On Statistic* for 8-digit HUCs
HUC
Total Area
(ha)
Riparian
Corridor
(ha)
Stream
Length
(km)
Stream
Density
(m/ha)
3060101
272,812.23
55,585.95
3066.68
11.24
3060102
258,218.91
54,114.18
2994.61
11.60
3060103
483,189.03
88,651.85
4803.99
9. 94
3060104
398,298.06
65,842.94
3463.28
8.70
3060105
204,446.97
32,453.26
1636.33
8.00
3060106
488,842.20
83,668.92
1765.63
3.61
3060107
200,987.55
31,324.14
4771.76
23.74
| 3060108
220,108.41
37,124.59
2044.32
9.29
|| 3060109
248,158.71
47,316.49
2679.38
10.80
15
3060103
3060104
3060107
3060108
3060101
Coastal Plain
Figure 13.
Atlantic Coastal Plain
Ecoregions of the Savannah
River Basin
Blue Ridge Mountains
Piedmont
-------
T«bl« 2. For«»t Com Typ«», P«ro«nt Covr by BUC
0 HUC
Evergreen
Mixed
Deciduous
Woody
Wetlands
| 3060101
23.36
10.45
37.92
0.62
I 3060102
25.66
12.15
45.68
0.27
| 3060103
28.35
11.07
25.28
0.55
| 3060104
23.72
9.65
38.66
0.36
H 3060105
39.95
8.85
28.57
0.69
| 3060106
33.39
7.22
12.74
11.54
J 3060107
50.21
9.72
18.69
0.74
H 3060108
24.17
7.50
15.38
10.86
I 3060109
25.24
4.63
7.33
31.46
the HUCs comprising the
Blue Ridge and Piedmont
ecoregions; all forest
types account for 64.7 to
83.4 9% of the total land
cover. Forest landcover
accounts for 37.20 to
53.35% of the landcover in
the Coastal Plain HUCs,
with evergreen forests the
predominant forest type.
Wetland landcover types
are found primarily in the
Coastal Plain HUCs,
accounting for 11.10 to
35.93% of the total
landcover, most of it in
woody wetlands. Wetlands comprise less than one percent of the
landcover in the HUCs outside the Coastal Plains.
Agricultural landcover types, Table 3, comprise 9.91 to 32.47% of
the total landcover in each HUC. Pasture/hay is the dominant
agricultural land use in the upper part of the basin, while row
crops are the largest agricultural land use in the lower basin.
Urban landcover types, Table 4, account for between 0.85 to 5.33%
of the total land use in all HUCs. There is no ecoregion-related
pattern to the distribution of urban landcover. Barren landcover
types, Table 5, comprise less than one percent of the total
landcover in HUCs 1 and 2, and approximately 2 to 10 percent of
the landcover of the Piedmont and Coastal Plains HUCs.
tarecnt Com
it by IU
C
HUC
Low
Intensity
Residential
High
Intensity
Residential
High
Intensity
Camnercial/
Industrial
3060101
Pasture
/Hay
Row
Crops
Other
Grasses
10.18
5.05
0.76
3060101
3.35
0.29
1.16
3060102
6.76
3.46
0.28
3060102
1.03
0.06
0.42
3060103
13.21
9.08
0.48
3060103
1.74
0.25
0.60
3060104
15.51
7.59
0.35
3060104
0.88
0.06
0.49
3060105
4.01
6.90
0.12
3060105
0.60
0.08
0.32
3060106
1.60
14.60
0.46
3060106
2.72
1.01
1.60
3060107
3.32
6.59
0.09
3060107
0.71
0.10
0.30
3060108
2.76
29.71
0.06
3060108
0.49
0.10
0.26
3060109
1.78
14.15
0.38
3060109
1.03
0.64
1.18
16
-------
1
1 HUC
Water
Emergent
Wetlands
Barren:
Quarries/
Strip Mines
Barren:
Bare
Rock/Sand
Barren:
Transitional
3060101
6.41
0.04
0.19
<0.01
0.22
3060102
3.61
0.07
0.04
<0.01
0.49
3060103
6.60
0.03
0.14
<0.01
2.63
3060104
0.57
0.02
0.14
<0.01
1.98
| 3060105
4.87
0.03
0.19
<0.01
4.81
3060106
1.46
0.48
0.60
0.01
10.57
3060107
0.39
0.03
0.07
<0.01
9.04
3060108
0.47
0.24
0.54
<0.01
7.45
3060109
3.10
4.47
0.14
0.05
4.40
The patterns
of
landcover/1and
use within the
riparian
corridors,
Table 6, are
not
substantially
different than
those for the
HUCs overall,
with the
exception that
water is an
appreciable
percentage of
the landcover
within
riparian
corridors in
most HUCs.
Agricultural
land use
within the
riparian
corridor
ranges from
4.63% to
12.78% and
urban land use
ranges from
0.33% to 3.51%. Barren landcover ranges from less than 1% to a
little more than 6%. The predominant landcover types in the HUC
riparian corridors are forest and wetlands in the Coastal Plains
and forest in the other ecoregions.
HUC
Forest
Agriculture
Urban
Wetland
Barren
Water
3060101
71.96
8.16
3.19
1.92
0.44
14.51
3060102
79.06
7.03
1.31
1.09
0.33
11.20
3060103
70.63
10.10
1.38
1.77
1.26
14.86
3060104
83.20
12.16
0.76
1.21
0.64
2.04
3060105
76.83
4.84
0.57
2.23
4.17
11.33
3060106
48.79
6.54
3.51
28.90
6.23
6.06
3060107
86.00
4.63
0.40
1.99
5.51
1.50
3060108
48.53
12.78
0.33
32.43
4.20
1.77
3060109
24.13
6.38
1.34
57.23
2.48
8.44
While there is some variation in landcover types among the three
ecoregions, overall the HUCs are relatively homogeneous in
landcover/land use pattern. In all HUCs, natural landcover types
comprise greater than 50% of the total landcover. Urban land
uses account for only a small percent of the total landcover and
agricultural uses account for 1/10 to approximately 1/3 of the
total land cover/land use. These results contrast greatly with
the results obtained for 8-digit HUCs in the mid-Atlantic region
(Jones et al, 1997), where large differences were evident at this
scale. The broad-scale patterns evident in the mid-Atlantic
(e.g., intensive urbanization of the Coastal Plains, concentrated
17
-------
agricultural land uses in valleys, and isolation of forests to
highland areas) are not in evidence in the Savannah River Basin.
Agriculture on slopes greater than 3% grade has been developed as
a landscape indicator because the potential for erosion increases
significantly at this grade. Similarly, agriculture practiced on
highly or moderately erodible soils has a higher potential for
erosion. These indicators are developed from overlays of DEMs,
MRLC land
cover/land
use, and
erodibility
factors
contained
in the
STATS GO
soils data
base.
Results for
all of
these
indicators
are
generally
low, as
shown in
Table 7.
Only HUCs 3
and 4 showed
greater than
20% total
land area
for any of
the
agriculture-
soil-slope
indicators,
that being
agriculture
on
moderately
erodible
soils, most
of it in
pasture/hay. Results for these indicators within the riparian
corridors are lower, ranging from nonexistent to less than 12%
agriculture on moderately erodible soil in HUC 4, as shown in
Table 8.
HUC
Pasture/
Hay on
Slopes >
3%
Row
Crops
on
Slopes
>3%
Pasture/
Hay on
Moderately
Erodible
Soils
Row Crops
on
Moderately
Erodible
Soils
Pasture/
Hay on
Highly
Erodible
Soils
Row
Crops on
Highly
Erodible
Soils
3060101
2.24
0.98
10.03
4.96
—
—
3060102
1.33
0.67
6.54
3.36
—
—
3060103
1.25
0.75
12.51
8.41
0.61
0.58
3060104
2.25
0.95
15.24
7.45
0.26
0.14
3060105
0.22
0.42
2.22
3.84
1.76
2.71
3060106
0.13
0.72
0.22
1.67
<0.01
0.02
3060107
0.10
0.26
4.20
3.06
—
—
3060108
0.03
0.47
0.48
3.23
~
—
3060109
<0.01
0.02
0.41
2.51
—
Sabl* 8.
Agriculture
-Itelatad I
ndicatora in I
lip&rian Corrii
iora, Pttou
t im
HUC
Pasture/
Hay on
Slopes >
3%
ROW
Crops
on
Slopes
>3%
Pasture/
Hay on
Moderately
Erodible
Soils
Row Crops
on
Moderately
Erodible
Soils
Pasture/
Hay on
Highly
Erodible
Soils
Row
Crops on
Highly
Erodible
Soils
3060101
1.04
0.55
4.66
2.50
~
—
3060102
0.95
0.48
4.08
1.94
--
—
3060103
0.51
0.33
5.68
3.53
0.20
0.21
3060104
1.00
0.47
7.98
3.89
0.08
0.07
3060105
0.11
0.16
0.73
1.46
0.98
1.56 |
3060106
0.05
0.23
0.21
1.17
<0.01
0.01 A
3060107
0.03
0.12
1.88
1.78
--
" 1
3060108
0.01
0.10
0.25
1.35
~
1
3060109
<0.01
<0.01
0.21
1.30
—
-- 1
18
-------
T«bl« 9. Rotdi Cro»mini
HUC
3060101
3060102
3060103
3060104
3060105
3060106
3060107
3060108
3060109
Road
^Crossings
1235
964
1487
1227
362
1914
637
642
723
fltg—a and Im>p\inAi>»nt»
No. Crossings/
Stream km
0.40
0.32
0.31
0.35
0.22
1.08
0.13
0.41
0.27
Dams
117
58
98
102
35
191
60
52
31
No. Dams/
Stream km
0.038
0.019
0.020
0.029
0.021
0.108
0.013
0.025
0.012
Roads frequently cause
increased runoff to
streams and contribute
pollutants washed off
the road surfaces.
This phenomenon is
represented by the
roads-crossing-streams
indicator, computed
from intersecting
digital line graph
roads with RF3 stream
arcs. As shown in
Table 9, values for
this indicator range
from 362 in HUC 5 to
1,914 in HUC 6. Normalizing these values to the number of road
crossings per stream kilometer, also shown in Table 9, shows the
greatest frequency of roads crossings per stream kilometer is in
HUC 6 with more than one road crossing per kilometer of stream
length. The lowest frequency is in HUC 7 with approximately one
road crossing for every 8 kilometers of stream length. The
remaining HUCs have frequencies in the range of one road crossing
for every 2.5 to 5 kilometers of stream length.
Information for dams was obtained from the National Inventory of
Dams which tracks all dams greater than 6 feet in height for
inspection purposes. As shown in Table 9, the fewest number of
dams in any HUC is 31 in HUC 9 while the greatest number is 191
in HUC 6. Normalizing by the total stream length within each HUC
shows the greatest frequency of dams is also in HUC 6, with one
dam for every 9 kilometers of stream length. The lowest
frequencies of dams are in HUC -9 and-HUC 7, with roughly one dam
for every 80 kilometers of stream .length. The locations of dams
are depicted in Figure 1.
Subbasin Indicators
As discussed above, the landscape indicators at the HUC level
show some variation among HUCs attributable to natural landcover
variation at the ecoregion level. However, the patterns of land
use are generally consistent across ecoregions and among HUCs.
This section focuses on the next scale, the subbasin. Landscape
indicators are presented for several subbasins of HUC 3. These
particular subbasins were selected because they each contain one
or more of the sampling sites selected for analysis. The
landscape indicators produced for the subbasins are the same as
those produced for the HUCs.
19
-------
Table 10. Physical Dimension Statistics tor 8el*ot*d
I Subbasin
Total
Area (ha)
Riparian
Corridor
(ha)
Stream
Length
(Jon)
Stream I
Density 1
(m/ha) 1
1 20
55,797.39
9,089.19
486.79
8.72
26
53,225.73
10,089.00
528.63
9.93
32
17,195.76
4,311.00
255.27
14.84
36
61, 462.62
9,704.07
499.79
8.13
53
68,295.33
12,800.34
695.35
10.18
Physical dimensions of
the selected subbasins
are shown in Table 10.
The total land area in
each subbasin ranges from
17,195.76 ha in #32 to
68,295.33 ha in #53. The
associated riparian
corridors range from
4,311.00 to 12,800.34 ha.
ii. I^nd Cow typw for »«Uctsd
The landcover
statistics for HUC 3
overall are 64.70%
forest (approximately
28% evergreen, 25%
deciduous and 11%
mixed forest), 22.29%
agriculture
(approximately 13%
pasture/hay and 9%
row crops), 2.59%
urban, 6.60% water,
approximately 3%
barren, and less than
one percent wetlands.
Among the subbasins,
the forest landcover
classes vary from
40.26% in #32 to
73.66% in #53. As
shown in Table 11,
evergreen forests are
the largest forest
class in #26, #36,
and #53; deciduous is
the largest class in
#20 and #32. Agricultural land use in #26 is about the same as
in the HUC overall (23.65% of which approximately 16% is in
pasture/hay). Greater agricultural land use is evident in #20
(34 07% with about 20% in pasture/hay) and #32 (32.45% of which
almost 18% is pasture/hay). Less landcover is in agricultural
land uses in #36 (15.16%, with more than 8% pasture/hay) and #53
(11 58%, with row crops slightly exceeding pasture/hay). Urban
land use is lowest in #53 at less than one percent and highest in
#20 at 8.05%. The remaining three subbasins have urban land use
land Cover Type
Subbasin
20
26
32
36
53
Water
1.30
8.11
20.59
0.57
9.36
Low Intensity
Residential
5.60
2.87
3.34
2.10
0.45
High Intensity
Residential
0.83
0.57
0.43
0.28
0.03
High Intensity
Cotemeccial/Industrial
1.62
1.23
1.04
0.67
0.15
Pasture/ Hay
20.23
16.33
17.54
8.43
5.58
Row Crops
13.84
7.32
14.91
6.73
6.00
Other Grasses
1.53
0.95
0.75
0.36
0.14
Evergreen Forest
16.09
25.32
10.53
37.08
36.70
Mixed Forest
9.55
11.00
5.96
13.12
10.98
Deciduous Forest
28.30
21.12
23.77
24.61
25.98
Woody Wetlands
0.57
0.75
0.82
0.30
0.63
Emergent Wetlands
0.03
0.02
0.11
0.02
0.03
Barren: Quarries/
Strip Mines
0.26
0.14
0.21
0.07
0.09
Barren: Transitional
0.24
4.27
<0.01
5.66
3.88
20
-------
in slightly higher percentages than for the HUC overall, ranging
3.05% in #36 to 4.81% in #32.
Table 12. Land Cover Type* for
P*se«nt ATM
Selected Subbaaln Riparian . Corridors,
1
Subbasin |
1 L&iul Covci Typti
20
26
32
36
53 1
Water
5.89
22.38
25.31
2.25
18.58 J
Low Intensity Residential
3.97
1.68
1.82
1.36
0.31
High Intensity Residential
0.26
0.19
0.07
0.08
<0.01
High Intensity
Commercial/Industrial
0.76
0.47
0.39
0.20
0.08
Pasture/ Hay
8.84
8.68
7.60
3.84
1.94
Row Crops
5.54
3.40
8.32
2.84
2.47
Other Grasses
0.36
0.24
0.21
0.05
0.02 |
Evergreen Forest
15.78
18.59
14.30
25.59
29.40
Mixed Forest
12.76
12.30
8.49
15.37
10.54
Deciduous Forest
43.15
28.94
30.23
44.56
33.20
Woody Wetlands
2.11
1.69
2.73
0.79
2.27
Emergent Wetlands
0.09
0.05
0.37
0.04
0.10
Barren: Quarries/ Strip
Mines
0.34
0.06
0.15
0.03
0.02 |
Barren: Transitional
0.16
1.32
<0.01
2.99
1.07 1
In the riparian
corridors, forest
comprises 53.02 to
85.52% of the
total cover, with
deciduous the most
dominant forest
cover type, as
shown in Table 12.
Agricultural land
use within the
riparian corridor
ranges from 4.41%
in #53 to 15.92%
in #32 and urban
land use comprises
from 0.39% to
4.99% of the total
riparian land
cover. Wetlands
account for
approximately 3%
or less of the
riparian land
cover types.
Table 13. Jt0cleultuN-ftalat«d Indicator* tor S*l*etad Subfcaains
Subbasin
20 |
26 |
32
36
53
Pwxeaat of
lubbasin Total Area
Agriculture on Slopes >3%
3.54
2.71
4.29
1.17
0.60
Agriculture on Moderately
Erodible Soils
34.03
23.53
31.64
15.15
4.39
Agriculture on Highly
Erodible Soils
—
—
7.11
Ncowt of tubbaain lipuiia Corridor |
Agriculture on Slopes >3%
1.31
1.34
1.24
0.39
0.27
Agriculture on Moderately
Erodible Soils
14.13
11.99
13.02
6.67
1.43
Agriculture on Highly
Erodible Soils
—
—
2.61
The agriculture-
soil-slope
indicator results
for HUC 3 are 2%
agriculture on
slopes greater
than 3%,
approximately 21%
agriculture on
moderately
erodible soils,
approximately 1%
agriculture on
highly erodible
soils, and less
than 0.1%
agriculture on
slopes greater
than 3% in highly
21
-------
erodible soils. Among the subbasins, #20, #26, and #32 have more
agriculture on slopes greater than 3% and more agriculture on
moderately erodible soil than for the HUC overall; the remaining
two subbasins are substantially lower than the HUC overall for
both these indicators, as shown in Table 13. Only #53 has any
agriculture on highly erodible soil (about 7%) and agriculture on
slopes greater than 3% and highly erodible soils (0.35%).
Results for these indicators are lower for the riparian
corridors, with only subbasins #20, #26, and #32 having more than
10% riparian land cover in agriculture on moderately erodible
soils.
Tabl* 14. Roads Crossing Stxaams and Hnwrrnvtaants foe
H Subbasin
Road
Crossings
No. Crossings/
Stream km
Dams
No. Dams/
Stream km
20
299
0.61
19
0.039
26
227
0.43
15
0.028
32
56
0.22
0
--
36
170
0.24
13
0.026
a
82
0.11
4
0.006
Table 14 provides
results for the number
and frequency of roads
crossing streams -and
dams; these indicators
are depicted in Figure
14. Roads crossing
streams ranges from 56
in #32 to 299 in #20.
There are no dams in
#32, but 19 dams in #20.
The frequency of roads
crossing, streams is highest in #20 with approximately one road
crossing for every 1.6 kilometers of stream length; the lowest
frequency is in #53 with one road crossing per approximately 9
kilometers of stream length. The frequency of roads crossing
streams for the HUC overall is approximately one crossing per 3
kilometers of stream length. The frequency of impoundments for
the HUC overall is approximately one dam for every 50 stream
kilometers. The frequency of dams is lower than for the HUC
overall in #32 with no dams and in #53 with approximately one dam
for every 167 kilometers of stream length. The greatest
frequency of dams among the subbasins is in #20 with one dam per
approximately 25 stream kilometers.
At this scale, patterns which may impact water quality begin to
be evident. In Figure 5, the sampling stations in #53 are
indicated as fair to good (as compared to the overall data
ranqe) This subbasin has the highest proportion of landcover in
forest among the subbasins, the lowest proportion of agriculture
and urban land uses, and a low proportion of agriculture on
slopes greater than 3%. Among the selected subbasins, it has the
lowest frequency of roads crossing streams. Although is the
largest of the subbasins in total area, this subbasin has only
dams. However, #53 is the only subbasin among those examined
22
-------
Figure 14. Roads crossing Streams and Dams in Selected Subbasins
-------
with agriculture on highly erodible soils and agriculture on
slopes greater than 3% and highly erodible soils.
In contrast, the sampling sites in #32 and #20 rank as fair to
bad compared to the overall data ranges. These two subbasins
contain the greatest proportion of agriculture among the
subbasins, 28 to 33% agriculture on moderately erodible soils,
and 3 to 4% agriculture on slopes greater than 3%. In addition,
#20 has the highest proportion of urban land use, the highest
normalized roads crossing streams value, and the greatest
frequency of dams among the selected subbasins.
Sampling Site Drainage Landscape Indicators
As described above, landscape analysis at the subbasin scale may
be adequate to provide a generalized characterization of the
Savannah River Basin. One of the objectives of this project,
however, is to try to establish relationships among landscape
indicators and water quality/aquatic biota indicators. The water
quality data were collected at specific sampling sites. To
investigate relationships with landscape indicators, it is
necessary to delineate the
drainage area to the 15"
individual sampling site.
This was done for a subset of
16 sampling sites. The
selection process was
described earlier, as was the
methodology for delineating
the drainage areas.
Phy*ic*i Dimension StafcLatiea for
The drainage areas for the
sampling sites range from
122.58 to 10,665.18 ha, as
shown in Table 15. In
delineating the drainage
areas, the locations for the
sampling sites frequently did
not lie on a stream arc,
necessitating a best guess,
based on the indicated stream
order and proximity to stream
arc, as to the point on the
arc to use as the pour point.
In addition to the landscape
indicators calculated for the
HUCs and subbasins, indicators
of fragmentation were
Site
Total
Area (ha)
Riparian
Corridor
(ha)
Stream
Length
(to)
Stream
Density
(m/ha)
S22
973.98
149.85
7.34
7.54
S27
4,950.90
939.78
47.57
9.61
S68
468.09
88.65
4.34
9.28
S80
6,499.71
884.16
44.90
6.91
S8X
6,612.21
908.01
45.85
6.93
S95
10,665.18
1,727.73
89.30
8.38
S103
572.76
69.30
3.32
5.80
S113
747.00
83.79
4.51
6.03
S130
1,169.73
163.80
8.68
7.42
S149
776.52
139.50
6.87
8.84
S151
1,076.22
191.88
9.53
8.85
S155
2,556.72
361.06
18.98
7.42
S195
4,279.41
660.94
46.05
10.76
S197
122.58
43.56
2.06
16.78
S200
1,798.47
377.37
19.11
10.62
S216
551.16
116.19
5.69
10.33
24
-------
Table 16.
Aggregated Land Cover Types for Sa
Percent Area.
pllng Site Drainages,
generated using
a custom, in-
house software
program. For
the
fragmentation
indicators, the
15
landcover/land
use classes of
the MRLC data
were aggregated
to six classes:
water, urban,
forest,
agriculture,
wetlands, and
barren, as
shown in Table
16 for the
overall
drainage area
and in Table 17
for the
riparian
corridors. In these aggregated land cover types, other grasses
are included in agriculture and woody wetlands are included in
the wetlands cover type.
Results for agriculture-related indicators over the entire
drainage area and the riparian corridor are presented in Table
18. The number of road crossing streams and dams are shown in
Table 19. Ten of the 16 sampling site drainages contain no dams;
however, where dams are present, they are generally greater in
frequency than in the HUC or subbasins overall. The frequency of
roads crossing streams ranges from approximately one road
crossing per 5.5 kilometers of stream length to a maximum of one
road crossing for every stream kilometer.
Results for each indicator were encoded into ARC/INFO Grids. A
Grid stack was generated and used to develop a correlation
matrix. A separate Grid stack was generated for the riparian
corridors contained in the drainage areas for the sixteen
sampling sites. With an n of 16, the correlation coefficients
are significant at values greater than 0.666 for a = 0.005, at
values greater than 0.601 for a = 0.01, at values greater than
0.507 for a = 0.025, and at values greater than 0.425 for a =
Site
Water |
Urban |
Agriculture |
Forests |
Wetlands
Barren
S22
0.11
<0.01
59.40
40.21
0.24
0.04
S27
0.34
0.34
18.76
80.06
0.37
0.11
S68
<0.01
<0.01
0.27
96.01
<0.01
3.73
S80
0.56
7.75
18.28
72.71
0.45
0.26
S81
0.55
7.62
17.97
73.17
0.44
0.25
S95
0.62
2.84
8.40
81.52
0.87
5.75
S103
<0.01
<0.01
0.04
89.92
0.08
9.96
S113
0.59
2.80
23.70
61.78
4.56
6.55
S130
1.02
<0.01
62.27
36.28
0.40
0.03
S149
0.03
16.02
44.33
39.38
0.10
0.13
S151
0.09
13.01
40.59
46.00
0.19
0.10
S1S5
0.11
3.21
17.77
73.21
0.56
5.14
S195
0.85
0.44
3.45
94.49
0.04
0.73 j
S197
<0.01
5.50
42.59
51.62
0.29
<0.01 |
1 S200
0.19
6.77
47.64
42.37
0.23
0.11 |
bsis-
0.10
0.07
4.90
87.62
0.13
7.18 |
25
-------
Tabl* 17. Aggragratad Land Covmx Typma fox Stapling Sit* Riparian
0.05. Using these
values, a number
of significant
correlations
between water
quality/aquatic
biology indicators
and landscape
indicators were
indicated, as
shown in Table 20.
In general,
correlations were
the same or less
for the riparian
corridor than for
landscape
indicators over
the whole drainage
area. The primary
exception is
dissolved oxygen,
which exhibited
significant
correlation only
with total
anthropogenic cover (U-index, comprised of an aggregation of
urban and agriculture land cover types) in the riparian corridor.
It should be noted that this analysis is preliminary and is based
only on the nonrandomly selected subset of sixteen sampling
locations. The data set size was insufficient to perform a
cluster analysis. The strongest correlations were between
landscape indicators; this is not surprising as several of the
landscape indicators contain similar information. The redundancy
is needed at this point in the research until the strongest and
most sensitive relationships with aquatic indicators can be
established.
Figures 15 through 20 depict six of the sampling station drainage
areas. Sites S68, S113, and S195 are ranked as good data sites,
based on the relative rankings of the data measurements. Site
S68 is a small forested drainage located in HUC 2. Site 113 is
also relatively small and is located in HUC 6; although
agriculture and urban areas are evident within the drainage, they
are fragmented as compared to the forest landcover; much of the
riparian corridor is wetlands. Site S195 is a larger drainage
and higher order stream located in HUC 2. All of the landcover
Site
S22
Water
Urban
Agriculture
Forests
Wetlands
Barren
<0.01
<0.01
29.91
69.37
0.72
<0.01
327
0.48
0.05
8.19
90.41
0.82
0.06
S68
<0.01
<0.01
0.41
99.19
<0.01
0.41
seo
2.28
6.17
11.81
78.70
0.92
0.11
S81
2.21
6.02
11.63
79.14
0.88
0.11
395
2.97
2.06
2.83
88.22
1.79
2.14
3103
<0.01
<0.01
<0.01
95.84
0.13
4.03
S113
4.19
<0.01
4.72
60.36
29.53
1.18
S130
6.21
<0.01
25.60
66.60
1.59
<0.01
S149
<0.01
11.35
20.70
67.93
<0.01
<0.01
S151
<0.01
9.52
17.82
72.47
0.19
<0.01
3155
0.07
1.49
6.55
89.14
0.64
2.13
S195
4.21
0.08
7.07
87.39
0.09
1.15
S197
<0.01
9.09
10.75
80.17
<0.01
<0.01
S200
0.45
6.13
26.18
66.71
0.45
0.07
S216
0.0 8
0.15
2.47
92.56
0.23
4.49
26
-------
Savannah River
Sampling Site S68 Drainage
Barren
Row Crops
Evergreen Forest
Mixed Foresi
Deciduous Forest
/y Roads
/\J River Reach
• Dam Locations
Figure 15. Sampling Site S68 Drainage.
27
-------
Savannah River
Sampling Site S113
jk'+t-c! i?sjB
" » . .
r»V A
/
, /x;r / r L
-u
-------
Savannah River
Sampling Site SI95 Drainage
Water
Other Grasses
Barren
Low Residential
Commercial Urban
Evergreen Forest /%'' 2?ad® .
y /\y River Reach
Mixed Forest
Darns
Hay
Row Crops
Deciduous Forest
Wetlands
Figure 17. Sampling Site S195 Drainage
29
-------
Table 18. Jkgrieultura-Aalatad Indicators for Sapling Sit* Drainages Ripaxian
Corridors, Earoant
Total Araa
Riparian Corridor 1
Site
Agriculture
on Slopes >
3%
Agriculture
on
Moderately
Erodible
Soil
Agriculture
on Highly
Erodible
Soils
Agriculture
on Slopes
>3%
Agriculture
on
Moderately
Erodible
Soils
Agriculture |
on Highly |
Erodible I
Soils I
S22
3.91
59.40
—
2.22
29.91
1
S27
0.26
IB.69
—
0. 03
8.19
I
S68
0.19
0.27
—
0.30
0.41
1
seo
1.57
17.72
—
0.66
11.64
1
S81
1.54
17.42
—
0.65
11.43
1
S95
0.28
5.87
—
0.02
2.30
I
S103
—
0.04
0.04
—
—
1
S113
2.14
—
—
2.04
--
B
S130
9.59
62.27
—
1.70
25.60
1
S149
1.85
41.40
0.32
18.83
1
S151
2.35
38.36
~
0.28
16.46
H
SX55
1.07
17.33
17.33
0.02
6.38
6.38 J
S195
0.64
3.37
—
0.96
7.07
—
S197
9.92
42.59
—
0.21
10.75
—
3200
4.25
45.51
—
2.43
25.42
—
S216
0.09
4.90
—
0.15
2.47
—
types are present, as are a number of roads and a few dams. The
predominant landcover, however, is unfragmented forest.
The remaining three figures are indicative of sites with fair to
bad relative rankings. Site S22, located in HUC 3, subbasin #39
has extensive agriculture, much of it in large blocks while the
forest landcover types are fragmented. Site S80 is a large
drainage area located in HUC 3 subbasin #36; the sampling site is
located in an area of unfragmented forest, but the upper reaches
of the drainage, including the headwaters of most of the streams
are dominated by urban and agricultural landcovers and extensive
road networks. Site S149 is a fairly small drainage located in
HUC 3, subbasin #20. There is extensive agriculture and urban
land use; the forest landcover is highly fragmented. The
headwaters of one of the two streams in thfe drainage is found in
an area of high intensity commercial/industrial land use.
30
-------
Savannah River
Sampling Site S22 Drainage
0.7
0.7
1.4 Miles
Water
Deciduous Forest
Hay
Row Crops
Wetlands
Barren
Evergreen Forest /\ / Roads
'/\/ River Reach
• Dams
Mixed Forest
Figure 18. Sampling Site S22 Drainage
31
-------
Savannah River
Sampling Site S80 Drainage
2 Miles
Water
Low Residential
Row Crops
Other Grasses
High Residential Evergreen Forest
Commercial Urban
Hay
Mixed Forest
Deciduous Forest
Figure 19. Sampling Site S80 Drainage
Wetlands
Barren
/S/ Roads
/\/ River Reach
• Dams
32
-------
Savannah River
Sampling Site S149 Drainage
0.7
A
0.7 Miles
Water
Low Residential
High Residential
Commercial Urban
Hay
Figure 20. Sampling Site S149 Drainage
Row Crops
Other Grasses
Evergreen Forest
Wetlands
Barren
\/ Roads
Mixed Forest /\/ Rlvcr Rcach
Deciduous Forest _ Dams
33
-------
labia 19. Roads Ciouiag Strums and Tmirmnrtwnts for IiSfldSCftp6 Ch&n^B
Two mosaics were developed
from the NALC data base
for the 1970s (Figure 21)
and the 1990s (Figure 22)
Savannah River Basin study
area. The mosaics were
matched to provide
analysis across similar
areas of the two mosaics.
Both mosaics were
processed into normalized
difference vegetation
index (NDVI) images and
the values in the 70s
mosaic was subtracted from
the 90s. Positive
numbers indicate gains in
vegetation and negative
numbers equate to losses
in vegetation; in Figure
23 vegetation gains are
shown in green while
vegetation losses are
shown in red. A standard
deviation was calculated
using n-1, for the entire change NDVI image. The Arc/Info grid
coverages depicting the various areas of interest were then
converted to image files (hereafter referred to as masks), and
the UTM coordinates for each were recorded. The resolution for
each mask was converted to 60 meters to match the resolution of
the change NDVI image. The change NDVI image was repeatedly sub-
sampled to select the matching areas of each mask. Each sub-
sampled change NDVI image and its corresponding mask were then
used as inputs to a custom in-house software program which
calculates the amount of cells (pixels) that are inside the mask
and groups them into 4 categories. They are: cells which are
greater than or equal to 4 standard deviations of loss in
vegetation, those cells which are greater than or equal to 2
standard deviations of loss in vegetation, and the corresponding
numbers of cells for gains in vegetation. In the following
tables the losses and gains have been grouped together and shown
as either a negative number for percent of loss or a positive
number for percent of gain.
An additional column is used to represent the cells removed from
the study area, which contain negative NDVI indices in either the
Xttpllttg iiti Prtiaigii
Site
Road
Crossings
No. Crossings/
Stream km
Dams
No. Dams/
Stream km
322
3
0.41
0
--
S27
21
0.44
1
0.021
368
2
0.46
0
—
S80
30
0.67
6
0.134
381
30
0.65
6
0.131
S95
37
0.41
4
0.045
S103
1
0.30
0
—
S113
1
0.22
0
—
S130
4
0.46
1
0.115
S149
5
0.73
0
—
S151
8
0.84
0
S155
3195
4
0.21
0
—
27
0.59
4
0.087
S197
1
0.49
0
—
S200
19
0.99
0
—
3216
1
0.18
0
34
-------
1970s Savannah River Basin
Figure 21. Mosaic of circa 1970 NALC images
35
-------
Mosaic of circa 1992 NALC images
-------
r
.. •' •¦*nwrttfliWhiiZ.J'.
r;
¦ ¦ - •< -".v.' \ .
' ppjfe-gm \ {'¦ -*' § '
: 'A . •-v^y J-; .. ,
if.' .»»> ' - «-'•' >.
;V; >- ' •«',
:yi *- 4-
. V ^ s
t. > * .
, ***. V
* "1* - •
. ' -'X-' - "
-•' -Mv
' * *' r-
• .• i ( .
vr*-;* -e; .
v •- ¦ :•*,'?•' ;..*>? -
- C'V :.,.v -V • ¦
- /; »• -v. i
•?r>- > :V-
,K •- i
* *
' ' i'-t
Savannah River 90s - 70s NDVI Change
2.0 Std. Dev.
-V •
" ' * •
< n- '' !
I
r y*
w
Figure 23. Gains (green) and losses (red) in vegetation, 1970s to 1990s
37
-------
Table 20. Correlation of Aquatio and Landscape Indicators for
I Aquatic
| Indicator
Landscape
Indicator
Correlation
S ignif icanoe
Level (a » )
I AGPT
%Forest Landcover
forest edge
U-index
ag__edge
avg ag patch
avg foreat patch
negative
negative
positive
positive
positive
negative
0.005/0.01R
0.005/0.025R
0.005*
0.025/0.05R
0.025
0.025/0.01R
EPT
avg forest patch
%forest cover
U-index
avg ag patch
forest edge
ag edge
ag on slopes > 3%
positive
positive
negative
negative
positive
negative
negative
0.005/0.025R
0.01
0.01
0.01
0.025
0.025
0.05
Richness
%forest cover
forest edge
U-index
avg ag patch
avg forest patch
ag edge
ag on slopes > 3%'
positive
positive
negative
negative
positive
negative
negative
0.025
0.025
0.025*
0.05
0.05
0.05
0.05
1 Fish_ibi
avg forest patch
forest edge
%forest cover
U-index
positive
positive
positive
negative
0.025*
0.05*
0.05
0.05
|pH
roads/streams
%forest cover
forest edge
U-index
ag on slopes >3%
positive
positive
positive
negative
negative
0.025
0.05
0.05
0.05
0.05
Dissolved
Oxygen
U-index
negative
—/0.05R
Habitat
avg forest patch
ag edge
%forest cover
forest edge
U-index
positive
negative
positive
positive
negative
0.025*
0.025
0.05
0.05
0.05/0.01R
[conductivity
%forest cover
ag on slopes >3%
negative
positive
0.005
0.05
r m riparian corridor
* - significant at the same level for both the full drainage and
riparian corridor correlations.
image. Subbasin #26 is the white-shaded
area shown in Figure 25.
Subbasin #32, differences in the water
surface (solar glare) produced a positive
change in vegetation which offset the loss
in that area. When the water areas
70s or 90s NDVI
image. Negative
NDVI indices are
generated by
clouds, water and
other non-
vegetation. This
also helps to
remove erratic NDVI
values caused by
differences in
solar illumination.
However, sometimes
these values are
meaningful, as in
the ease where an
impoundment may
have been installed
after the 70s image
and before the 90s.
An example of this
is shown in Figure
24.
Table 21 depicts
change in selected
subbasins of HUC 3.
Subbasin #26
reflects greater
than 3% negative
change, because an
impoundment was
installed between
the 70s and the 90s
Table 21. Landscape Change for
Selected Subbaaina
Subbasin
Percent
NDVI
Change
Percent
NDVI
Change,
Negative
numbers
removed
20
-9.803
-9.670
26
-9.993
-6.838
32
0.661
-6.384
36
-2.695
-2.826
52
-0.366
-1.873
38
-------
"
Savannah River Basin
Flgure 24. Lake Russell was created between the 1970s and 1990s.
(negative NDVI numbers) were removed the overall sub-watershed
had a loss of greater than 6%.
Table 22 shows the NDVI change in the drainage areas
selected sixteen sampling sites.
1970s
1990s
summary
Ths t-
. -hree questions posed as objectives by the Region can now be
Pressed:
^anciD°tl1 the Pr0P°rti°ns of land uses and the spatial pattern of
Co ,Uses important for characterizing and modeling stream
1^ion in watersheds/ecoregions of different areas?
39
-------
K
. -1 V
-' --V •
''As shown in this landscape
* assessment, both the proportion and
I the patterns of land use are
. important in assessing impacts on
streams. In the correlation analysis
conducted on the sampling site
. drainages, both total landcover types
^ (%forest, U-index) and pattern
indicators (fragmentation indicators
.-.4 including average patch size, forest
.and agriculture edges) were found to
p correlate with aquatic indicators. A
third important element is the scale
at which
,... *? analysis is
' aW M done. As
demonstrated
IBHBBhere, landscape
Figure 25. Land Surface Loss to indicators at
Lake Russell in Subbasin #26. the HUC level
were too coarse
to provide any
indications of water quality. In the
analysis of selected subbasins, patterns of
land use began to emerge; this scale may be
sufficient to provide a generalized
characterization of the basin.
Can land uses near the streams better
account for the variability in ecological
condition than land use for the entire
watershed/ecoregion?
In this particular assessment, landscape
indicators for the riparian corridors did
not provide stronger correlation with
aquatic indicators, with the exception of
dissolved oxygen. It should be remembered,
though, that this is one analysis of a small
spatial area in one region with a particular
suite of indicators. In other situations
the riparian corridor may be of greater
importance than the overall watershed. Even
in this region, the southern portion of the
basin has riparian corridors dominated by
wetlands. Only one site from this area was
used in the analysis and the entire sampling
data set contains only a few sites in this
Table 22. Landscape Change
for Sanplimg Site
Drainages
| Site
Percent
NDVX
Change
Percent
NDVI
Change,
Negative
numbers
removed
Class "Good"
S155
-0.873
-0.859
S 68
-0.613
-0.613
S195
-0.976
-1.245
S113
-4.193
-4.000
Class "Bad"
S80
-2.975
-2.748
S197
-8.235
-8.235
S149
-8.994
-7.325
S22
-18.404
-17.591
Class "Fair"
S81
-2.930
-2.707
S216
-1.235
-1.235
S103
-0.626
-1.627
327
-4.717
-4.674
Clas3 "Other"
S151
-6. 992
-5.721
S200
-4.528
-4.328
S13Q
-13.372
-11.149
S95
-2.932
-2.871
40
-------
ecoregion. A separate analysis of wetlands-dominated systems is
probably worthwhile.
Does the size of the watershed/ecoregion influence statistical
relationships between landscape characteristics and ecological
condition?
There was no indication in this analysis of any relationship with
the spatial extent of the drainage areas. This includes the
landscape indicators developed for the HUCs and subbasins. In
the sampling site analysis, one of the selection criteria was to
include streams of varying order; by doing so, both small and
large drainage areas were included. Drainage area was included
in the correlation analysis; no correlation was shown with any of
the aquatic indicators.
REFERENCES
Bara, T.J., ed. 1994. Multi-Resolution Land Characteristics
Consortium-Documentation Notebook. U.S. Environmental
Protection Agency, Environmental Monitoring and Assessment
Program, Research Triangle Park, NC.
Jones, K.B., K.H. Riitters, J.D. Wickham, R.D. Tankersley, Jr.,
R.V. O'Neill, D.J. Chaloud, E.R. Smith, and A.C. Neale.
1997. An Ecological Assessment of the United States Mid-
Atlantic Region: A Landscape Atlas. U.S. Environmental
Protection Agency, Office of Research and Development,
Washington, D.C.
Natural Resource Conservation Service. 1996. State Soil Survey
Geographic Data Base (STATSGO) Metadata. U.S. Department
of Agriculture, Natural Resource Conservation Service,
Washington, D.C.
Santa Cruz County. 1998. Riparian Corridors
http://gate.cruzio.com/~countysc/pln/riparian.htm
South Carolina Department of Natural Resources. 1998. South
Carolina Rivers Program: Recommended Best Management
Practices for River-bordering Lands
http://www.dnr.state.sc.us/water/envaff/river/bmps.html
U.S. Army Corps of Engineers. 1997. The National Inventory of
Dams http://corpsgeol.usace.army.mil/
41
-------
U.S. Environmental Protection Agency. 1998. Ecological
Restoration
http://www.epa.gov/owowwtrl/NPS/Ecology/chapl.html
U.S. Environmental Protection Agency. 1997. History of the U.S.
EPA's River Reach File. U.S. Environmental Protection
Agency, Office of Water, Washington, D.C.
U.S. Environmental Protection Agency. 1993. North American
Landscape Characterization (NALC) Research Plan. U.S.
Environmental Protection Agency, Office of Research and
Development, Washington, D.C.
U.S. Geological Survey. 1990. Digital Elevation Models,
National Mapping Program Technical Instructions, Data Users
Guide 5. U.S. Geological Survey, Second Printing (Revised),
Reston, Virginia, 1990.
U.S. Geological Survey. 1989. Digital Line Graphs from
1:100,000-Scale Maps-Data Users Guide 2. U.S. Geological
Survey, Reston, Virginia.
U.S. Geological Survey. 1982. Codes for identification of
hydrologic units in the United States and the Caribbean
outlying areas. U.S. Geological Survey, Circular 878-A,
Reston, VA.
Acknowledgments
Data processing and analysis support was provided by Lockheed-
Martin under Contract 68-C5-0065 to the U.S. Environmental
Protection Agency. In particular, the contributions of Lee Bice,
Karen Lee, and Dick Dulaney of Lockheed-Martin were essential in
the preparation of analyses used in this report.
42
-------
0
-------
Appendix G
Sampling Design Issues for Section 305(B) Water
Quality Monitoring
By
Stephen L. Rathbun
-------
Sampling Design Issues for Section 305(b) Water Quality
Monitoring
by
Stephen L. Rathbun
-------
Sampling Design Issues for Section 305(b) Water Quality
Monitoring
by
Stephen L. Rathbun
Abstract
State 305(b) water quality monitoring programs typically employ judgment sampling
designs, in which sample sites are selected according to a number of often vaguely
defined criteria. The resulting data are likely to yield biased estimates of parameters
such as the percent of the water resources that are satisfactory for their designated uses
(e.g., swimming, drinking, fishing, etc.). Moreover, there is no statistically justifiable
method for combining such data across states as mandated by Section 305(b) of
the Clean Water Act. This paper describes how probability-based sampling designs
can be implemented to sample water resources. A diverse variety of probability-
based sampling designs are available, the scientific judgment of the investigator can
be taken into account during the selection of strata, and multiple-stage designs can
be used to reduce sampling costs. Data resulting from probability-based sampling
1
-------
designs can be used to obtain unbiased estimates of such quantities as the percent
of water resources meeting environmental criteria for designated uses, and the total
mass of a chemical contaminant is a state's water resources. Moreover, data from the
various states can be easily combined even if different states use different probability-
based sampling designs. Despite these advantages, managers of state water quality
monitoring programs are reluctant to implement probability-based sampling designs.
Much of this reluctance stems from the fear that information from the historical
data base will be lost. A procedure for combining data from probability-based and
judgment sampling designs is demonstrated. This procedure exploits spatio-temporal
correlation among the observations from both data bases to back predict what data
would have been obtained had a probability-based sampling design been implemented
from the very beginning of the monitoring program.
2
-------
Contents
1 Introduction 4
2 Available Data 8
2.1 Savannah River Initiative (REMAP) 8
2.2 Clean Lakes Program 9
3 General Design Issues 10
3.1 Statistical Inference 12
3.2 Examples of Probability-Based Designs 14
3.3 Sampling over Space and Time 24
4 Current Status of Section 305(b) Water Resource Monitoring 38
4.1 Response to 305(b) Consistency Workgroup 42
4.2 Combining Data Across States 52
5 Design Alternatives for Section 305(b) Water Resource Monitoring 58
5.1 Sampling Lakes 58
5.1.1 Sampling Large Lakes 59
5.1.2 Sampling Small Lakes 62
5.2 Sampling Rivers and Streams 76
5.3 Sampling at Access Points 84
3
-------
6 Retaining Information from Historical Data 90
6.1 Space-Time Model 91
6.2 Spatio-Temporal Prediction 94
6.3 Simulation Model 96
6.4 Effect of Sampling Bias 100
6.5 Bias Reduction 106
6.6 Conclusions and Recommendations HO
1 Introduction
Section 305(b) of the 1972 Federal Water Pollution Control Act (usually known as the
Clean Water Act) mandates that each state submit a surface water quality assessment
report to thp Environmental Protection Agency (EPA) every two years, and that
the EPA submit a comprehensive assessment of the condition of the nation's waters
to Congress every two years. The latter requires the combining of data obtained
by the former as well as various native American tribes. However, current state
monitoring efforts present a number of obstacles to combining data at a national level
in a statistically defensible manner. Many of the obstacles arise from differences in
the objectives among the states and between the states and the EPA. While the EPA
is required to report on the condition of the totality of all of the nation's aquatic
resources, states and tribes often select monitoring stations based on local purposes
4
-------
(305(b) Consistency Workgroup, 1996).
The combination of data across states and tribes would be straightforward pro-
vided all states and tribes used probability-based sampling designs, provided that
there is some consistency in what variables are measured and how they are measured,
and provided that there is consistency in the definitions of target populations and
sample units. It is not necessary that all states employ the same probability-based
design, and so, states are free to implement designs tailored to their local require-
ments. However, few states or tribes employ probability-based sampling designs,
and for most states and tribes, the sample population covers less than 100% of their
water resources. Consequently, the representativeness of the current monitoring sta-
tions must be questioned. Statistical inference is limited to statements, for example,
regarding the percentage of sample sites showing impaired conditions, and not the
percentage the state's water resources that show impaired conditions. Efforts to com-
bine data across states and tribes are also impaired by variation among states and
tribes in site selection criteria, definition of target populations and strata, what vari-
ables are measured, sampling protocols, and analytical laboratory procedures. Some
limitations are biological: there is considerable natural variation among states in the
composition of their biota (i.e., what species are present) irrespective of anthropogenic
effects. Moreover, different states have different types of water resources (e.g., estuar-
ies in coastal states, mountain streams in states having mountains, etc.), and different
5
-------
resource types are likely to respond differently to environmental insults.
Recent years have seen new efforts to improve the quality of Section 305(b) water
resource monitoring. In 1992, the Intergovernmental Task Force on Water Quality
(ITFWQ) was established in response to Office of Management and Budget Memoran-
dum 92-01. Cochaired by the EPA and the United States Geological Service (USGS),
the ITFWQ is charged with the review and evaluation of national water quality mon-
itoring efforts, and to recommend improvements. They have recommended that Sec-
tion 305(b) change from the current 2-year reporting cycle to a 5-year reporting cycle.
This would help states achieve better coverage of their water resources through the
implementation of rotating panel and similar designs (see Section 3.3) under which
1/5 of the sample sites are monitored each year. The EPA has also established a
305(b) Consistency Workgroup, which as its name implies is tasked with improving
the consistency of Section 305(b) water quality monitoring among the states and
tribes. The 305(b) Consistency Workgroup is also exploring the implementation of
probability-based designs.
This paper considers issues regarding the replacement of current judgment sam-
pling designs used by most state water resource monitoring programs with probability-
based sampling designs. This together with efforts to improve consistency among
state and tribal programs in their sampling protocols, analytical laboratory proce-
dures, definitions of target populations, etc., would facilitate future efforts to combine
6
-------
data across states and tribes. Of particular concern is how we might replace current
sampling designs with minimal loss of historical monitoring information. Methods
are developed for combining historical data with new probabilistic data to obtain
predictions of what data would have been obtained had a probability-based design
been implemented in the very beginning of the monitoring program. Although it is
intended that sampling at judgment sample sites be discontinued at some point in
the future, sampling at a subset of such sites could continued to address site specific
questions and for purposes of model building. This paper does not consider methods
for combining judgment sample data with probability sample data collected during
the same time interval to improve estimates at that time interval. For a discussion of
such methods, see Overton, Young, and Overton (1993) and Cox Pieogorsch (1996).
After describing the available data in Section 2, Section 3 provides a general dis-
cussion of survey designs including those for sampling over space and time. The
current status of 305(b) water quality monitoring efforts is discussed in Section 4;
this includes a response to the concerns raised by the 305(b) Consistency Workgroup
regarding the replacement of current judgment sampling designs by probability based
pompling designs, and a discussion of how data may be combined across state un-
der probability-based sampling. Section 5 gives some specific design alternatives for
^tripling lakes and streams, including designs that involve sampling at access points.
Methods for combining historical judgment data with new probabilistic data are con-
7
-------
sidered in Section 6.
2 Available Data
The Regional Environmental Monitoring and Assessment Program (REMAP), and
the Clean T Program provide data on Secchi depth from lakes in the Savannah
River Basin. Secchi depth is a measure of water clarity. It is obtained by dropping a
Secchi disk over the side of a boat and measuring the depth at which the disk is no
longer visible.
2.1 Savannah River Initiative (REMAP)
The Savannah River Initiative of the Regional Environmental Monitoring and As-
sessment Program is sponsored by the Environmental Protection Agency. Data on
chlorophyll A and Secchi depth was collected in July 17-21, 1995 and June 24 to July
5 1996. Each year, 37-40 sites were sampled from the embayments of large lakes in
the Savannah River Basin, including Russell, Thurmond, Hartwell, Keowee, Jocassee,
Burton. Sample sites were selected according to the two-tiered sampling design.
A 7 x 7 x 7 fold enhancement of the EMAP base grid was placed over the Savannah
River Basin. Each grid point is circumscribed by a 1.86 km2 hexagon; 7 of these
hexagons form a 13 km2 hexal, and 7 hexals form a 635 km2 EMAP hexagon. The
tier 1 sample is comprised of 3 randomly selected 13 km hexals from each of the
8
-------
635 km2 EMAP hexagons covering the Savannah River Basin. All embayments were
enumerated within each of the selected hexals. The tier 2 sample of embayments to
be sampled each year was then selected using the procedure of Larsen and Christie
(1993).
2.2 Clean Lakes Program
The Clean Lakes Program is sponsored by the South Carolina Department of Health
and Environmental Control (SC-DHEC) and the Environmental Protection Agency.
This program involves the collection of data used to evaluate the quality of lake water
in South Carolina. Secchi depth was observed at 17 sites located in five large lakes in
South Carolina. These sites were selected according to a judgment sampling design
favoring the main channels of each lake. At each site, 0-2 monthly observations
were collected between April and October of each year. The length of the data
records depends on the sample site. This study was initiated at 10 sample sites
scattered throughout lakes Russell, Hartwell, Keowee, and Jocassee in April 1991.
Three additional sample sites were added in May 1992, one in Lake Russell and two
in Lake Keowee. Three sites in Broadway Lake were sampled only in 1994, and one
site in Lake Hartwell was sampled in 1993. In addition to Secchi depth, chlorophyll
A was measured occasionally, but records of this variable were too sparse to warrant
further analysis.
9
-------
April May June July August September October
Month
Figure 1: Mean secchi depth by month.
The seasonal pattern of variation in Secchi depth is illustrated in Figure 1. Monthly
means were adjusted to take into account variation among years in what sites were
included the sample. Mean Secchi depth was lowest in April at 2.82 m, increased to
I
a maximum of 3.83 m in June, then decreased to an asymptote of approximately 3.5
m thereafter.
3 General Design Issues
Environmental monitoring programs should be designed within the context of their
objectives in such a way as to optimize the amount of information they yield about
the resource of interest. The objectives may call for the selection of specific sites of
interest, for example sites near point sources of environmental contamination. For
10
-------
the latter, pairs of sites are often employed, one immediately downstream and the
second upstream of the point source. In such cases, inferences are restricted to the
environmental conditions that occur at those specific sites, and may include compar-
isons between upstream and downstream sites. When interest is restricted to specific
sites, sufficient monitoring resources should be made available to sample all of these
sites. If, however, the objectives call for inferences regarding the status of the envi-
ronment on a regional scale, sufficient monitoring resources are not available to census
all of the waters in that region. In such cases, a sample of sites must be selected. To
guarantee unbiased estimates of status, this sample of sites must be selected using a
probability-based sampling design. Probability-based designs involve some method of
random selection of sample sites, but are not reistricted to simple random sampling.
Probability-based sampling designs may be used to estimate the mean value of an
environmental parameter in the lakes of a region of interest, the percent of stream
miles that have impaired environmental conditions, the total mass of a contaminant in
the estuaries in a study region, or the percent of the area of lakes showing improving
environmental conditions. Probability-based sampling designs are most appropriate
for investigating nonpoint sources of environmental contamination and can also be
used to select reference sites for the investigation of the impact of point sources of
environmental contamination.
Nonprobability-based sampling designs must rely on the judgment of the investi-
11
-------
gator. Such judgment sampling designs are not likely to yield a representative sample,
and hence, can lead to biased estimates of population parameters. Unbiased estima-
tion of environmental parameters under judgement sampling requires the assumption
that the population or region of interest is homogeneous, an assumption that a*ms
unlikely to be tenable in nature.
3.1 Statistical Inference
Two types of statistical inference can be distinguished, design-based and model-based.
Design-based inference requires that data be obtained under a probability-based sam-
pling design. Under design-based inference, the values of the variable of interest in the
population or region of interest are assumed to be fixed and nonrandom. Here, the
source of random variation comes from the random selection of sample sites. Since
the sampling design is specified by the investigator, and hence is known, no model
assumptions are required. Design-based inferences are made on the actual population
or region from which the sample was drawn, and not on the parameters of some as-
sumed model. Such inferences may include unbiased estimates of the mean value of
an environmental parameter in the lakes of a region of interest, the percent of stream
*r>n«Hs that have impaired environmental conditions, the total mass of a contaminant
in the estuaries of a region of interest, or the percent of the area of lakes showing
improving environmental conditions. Standard errors and confidence intervals are
12
-------
available for all of these population parameters. Design-based hypothesis testing pro-
cedures test how likely that a sample with the observed data could have been drawn
from a population with the null parameter value(s). Since inferences are restricted to
the population from which the sample was drawn, design-based inference cannot be
used to predict future observations or data at unsampled sites.
Under model-based inference, it is assumed that the data are realized from some
random model. In multiple regression, for example, the variable of interest is «.ssnnwi
to be a linear function of some explanatory variables plus a random error. Further
assumptions may include the homoskedasticity of the errors, and that the data are un-
correlated and normally distributed. However, we may wish to assume that data are
spatially and temporally correlated, in which case, assumptions axe required regarding
the specific correlation structure. Instead of making inferences about the region from
which the data were obtained, model-based inferences are made on model parame-
ters. Such inferences may include estimates of the model parameters, together with
their corresponding standard errors, as well as predictions of future observations and
data at unsampled sites. Model-based hypothesis testing procedures test whether or
not the data are compatible with a null model. Although model-based inferences are
available for both probability-based and nonprobability-based sampling designs, pa-
rameter estimates can be biased under the latter. Typically, model-based inferences
ignore variability due to random selection of sample sites.
13
-------
3.2 Examples of Probability-Based Designs
A wide variety of probability-based sampling designs are available. The simple ran-
dom sampling design is the most basic method for selecting sample sites from a region.
For rectangular study regions, a simple random sample is obtained by random and
independent selection of X, Y coordinates from. For irregularly shaped regions, lo-
cations are sampled from the smallest rectangular region until a sufficient number of
sites are located in the study region (Figure 2); only those sites falling in the study
region are retained in the sample. Subregions will tend to be sampled in proportion
to their areas; for example, if 40% of the region is in loamy soils, then we expect 40%
of the sample sites to fall on loamy soils. Aside from the selection of the study region,
the selection of sample sites does not involve any scientific judgment.
The selection of a probability-based design need not, and should not ignore the
scientific judgment of the investigator. Under a stratified random sampling design,
the region is partitioned into strata, often corresponding the different habitats of in-
terest. For example, streams may be partitioned into first-, second-, and third-order
streams, while lakes may be partitioned by trophic level, ecoregion, size, access (public
or private), or whether they are natural or man-made. The wetlands surrounding the
Carolina Bay in Figure 3 are partitioned into five "undisturbed" habitat types. Sam-
ple units are then selected from each stratum according to a some probability-based
sampling design; a simple random sampling deign is used in Figure 3. Here, scientific
14
-------
700
(DO
400
300
200
too
100 200 300 400 500 BOO 700 BDO >00
0
X
Figure 2: Simple random sample of 100 sites in Ebenezer Aquifer (closed circles).
Sites falling outside the study region (open circles) are excluded from the sample.
15
-------
judgment is required for optimal selection of strata. Strata should be selected in such
a way that differences between strata are as large as possible, while units within strata
are as uniform as possible. By controlling for differences among strata, the stratified
random sampling design reduces the sampling variance and hence improves the pre-
cision of population parameter estimates. Therefore, a stratified random sampling
design can achieve the same precision at a smaller sample size than a simple random
sampling design, and hence reduce costs.
The optimum allocation of sampling effort among strata requires the within stra-
tum variances of the variables of interest, information that is not likely to be available
at the beginning of a new monitoring program. However, allocation proportional to
stratum size works well, and sample allocation may adjusted as data are obtained. It
js almost certain that different variables will yield different optimal allocation schemes,
and so, some compromise allocation scheme Costs may be reduced by decreasing sam-
pling effort in expensive strata, and increasing sampling effort in cheap strata. By
adjusting the allocation of sampling effort to the different strata, we may increase the
sampling effort in ecologically important strata, and ensure that an adequate sample
is obtained from rare habitats.
Another way in which the cost of sampling efforts can be reduced is to employ a
double sampling design. Double sampling can be used when a inexpensive ancillary
variable is available as a surrogate for the variable of interest. For example, Secchi
16
-------
Figure 3: Stratified random sampling design in the wetlands surrounding a Carolina
Bay. Circles are in grasslands, squares axe in briars and shrubs, triangles are in vines
and small trees, stars are in hardwoods and pines, and crosses are in pines.
17
-------
depth, which is inexpensive to obtain, may be an ancillary variable for total suspended
solids or Chlorophyll A, which require more expensive equipment and laboratory
procedures. Under a double sampling design, primary sample sites are first selected
according to any sampling design, then secondary sample sites are obtained by taking
a simple random sample of the primary sites (Figure 4). Both the ancillary variable
and the variable of interest are measured at the secondary sample sites, while only the
ancillary variable is measured at the primary sample sites. Under double sampling,
parameter estimation relies on the correlation between the variable of interest and
the ancillary variable. The ratio of secondary over primary sample sites depends on
the cost of obtaining the variable interest relative to the cost of the ancillary variable,
and on the magnitude of the correlation between the two variables. As the cost of
the variable of interest increases and the correlation increases, the optimal ratio of
secondary over primary sample sites decreases.
The above sampling designs require maps depicting all of the state's water re-
sources, from which a listing of all lakes, stream reaches, and estuaries may be ob-
tained. Such information might be obtained from River Reach File Version 3 (RF3)
(Horn and Grayman 1993). This file is not perfect; information on new man-made
reservoirs, small lakes, and higher order streams may be missing, and it also includes
some lower order ephemeral streams that may not be present if sought on the ground.
In any case, the information contained in RF3 should be verified on the ground, and
18
-------
'I I I 111 I ¦ 11 111 ¦ 11' 111 11 ¦ I 11 ¦¦¦ I 11 »'¦ 1' 1111 111 11
0 I DO 2DD 300 400 500 600 700 BOO ODD
X
Figure 4: Double sampling design in Ebenezer Aquifer. Primary sample sites are
designated by open circles, and secondary sample sites axe designated by closed circles.
19
-------
an effort should be made to fill in any missing information. If point sources of conta-
mination are of concern, then a list of all point sources is required, information that
is not available from RF3.
It may not be practical to be obtain such a list frame of all water resources within a
state. Multiple-stage sampling designs do not require list frames of all water resources,
and hence, may be more practical for water resource monitoring. Under a two-stage
sampling design (a special case of a multiple-stage sampling design), the population is
first partitioned into primary sample units, then a simple random sample of primary
units is selected, and finally, a simple random sample is obtained from each of the
selected primary units. Thus, water resources only need to be enumerated within
each of the selected primary units. Primary units should be small enough so that
all water resources within each of them can be easily enumerated. The flexibility of
multiple-stage sampling designs is illustrated by the following examples:
• To investigate the trophic levels of all small lakes in a region, the state may
first be partitioned into hexagons. Then a simple random sample of hexagons is
selected. The small lakes are enumerated within each of the selected hexagons,
and then a simple random sample of lakes is obtained from each of the selected
hexagons. Finally water samples are collected from each of the selected lakes.
• To investigate point sources of environmental contamination, the state may first
be partitioned into Natural Resource Conservation Service (NRCS) watershed
20
-------
units. Then a simple random sample of watershed units is selected. The point
sources are enumerated within each of the selected watershed units, and then a
simple random sample of the point sources is obtained. Finally, water samples
may be obtained upstream and downstream of the selected point sources.
• To investigate the soils of Ebenezer Aquifer, n parallel line transects may be
randomly located within the aquifer, and then m soil samples may be randomly
selected along the length of each transect (Figure ). Here, the transects are
treated as the primary sample units.
From the third example above, observe that the transect sampling design familiar
to ecologists is a special case of a two-stage sampling design. Two-stage sampling
designs can be extended into multiple stage designs by further partitioning each of
the sampled primary units into secondary units, partitioning sampled secondary units
into tertiary units, and so on. At each stage, a simple random sample of the units
defined that stage is obtained. Multiple-stage sampling designs may be modified to
allow stratified random sampling during any stage of the design.
Under the above conventional sampling designs, the sample selection procedure
does not depend on the observations obtained during the course of the survey. Under
adaptive sampling designs, however, the selection of future sample sites depends on
the observations that have been obtained up to the present time. Adaptive cluster
sampling designs are particularly suitable for the investigation of highly localized
21
-------
Figure 5: Line transect design in Ebenezer Aquifer.
phenomena such as clusters of a rare species or hot spots of highly contaminated
environmental resources (Thompson 1990,1992). Under an adaptive cluster sampling
design, a simple random sample of locations is first selected (Figure 6). If a given
sample site satisfies a given condition (i.e., presence of a rare species, or high levels
of contamination), addition sample sites are clustered around that site. This process
is repeated with the new sample locations until no new sites are added which satisfy
the criterion.
The above examples illustrate just a fraction of the diversity of available probability-
based sampling designs. Probability-based sampling designs can be tailored for almost
any scientific situation and can be constructed in response to many budgetary and
scientific constraints.
22
-------
Figure 6: Adaptive cluster sampling desiLn. First 10 sample units are selected at
random (dark shaded squares). Then adjacent unit are added to the sample whenever
one or more points are observed in the selected unit (light shaded squares).
23
-------
3.3 Sampling over Space and Time
So far, we have only considered probability-based designs for selecting sample sites at
a given point in time. Here, we shall consider the allocation of sampling effort over
space and time. There are at least four approaches to sampling over space and time:
• Permanent Stations: A sample of n permanent sample stations are selected
from some probability-based design; data are collected from each sample station
during every sample interval.
• Serially Alternating Design: Sample stations are selected from some probability-
based design and are partitioned into m sets of equal size n. Set i is then
sampled during intervals i,i + m, i + 2m, •••, as shown in Table 1 (Rao and
Graham 1964). This design was proposed for the Environmental Monitoring
and Assessment Program (EMAP) (Messer et. al 1991); here EMAP hexagons
are partitioned into m sets of size n, and hexagons are sampled as described
above.
• Rotating Panel Design: Sample stations are initially selected from some probability-
based design and are partitioned into m sets of equal size n. During each sample
interval, one set of sites is dropped from the sample, and is replaced by an ad-
ditional set of n sites selected from the probability-based design as shown in
Table 2 (Skalski 1990).
24
-------
Sampling Interval (ie., year, month, season)
Set 1 2 3 4 5 6 7 8 9 10 11 12
1 X
X
X
2
X
X
X
3
X
X
X
4
X
X
X
Table 1: Serially alternating design.
• Ever-Changing Stations. Under this sample design, a new and independent
probability sample is obtained during each sample interval.
The latter three sample designs can be augmented by selection of additional per-
manent sample stations which are to be sampled during each interval (Urquhart,
Overton, and Birkes 1993).
The various alternatives to spatio-temporal sampling offer a number of advantages
ynrl disadvantages with respect to design-based inference and spatio-temporal model-
ing and prediction. Correlation matrices are of block-Toeplitz form under permanent
station and serially alternating designs, and so, computationally more efficient algo-
rithms may be used during spatio-temporal modeling. If temporal trends are expected
to depend on location, permanent station and serially alternating designs are most
25
-------
Set
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Sampling Interval (ie., year, month, season)
1 23456789 10 11 12
X
X X
XXX
X X X X
X X X X
X X X X
X X X X
X X X X
X X X X
X X X X
X X X X
X X X X
XXX
X X
X
Table 2: Rotating panel design.
26
-------
suitable for estimating such quantities as the proportion of stream miles showing im-
proving or degrading environmental conditions. Permanent station designs yield the
smallest level of spatial coverage, while the greatest level of spatial coverage is ob-
tained under ever-changing station designs. Repeated sampling at pennant stations
may have an impact on the local environments at those stations, or example, through
the trampling of sensitive vegetation by observers, or through modification of the
behavior of people knowing the locations of those stations.
The optima] allocation of sampling effort over space and time depends on the
relative magnitude of spatial and temporal autocorrelation. This spatio-temporal
autocorrelation comes from the observation that data close together in space or time
are likely to be more similar than data collected far apart over space or time. Under
strong temporal autocorrelation, repeated observations at a given site will contain
a large amount of redundant information, and so the optimal design will sample a
large number of sites at infrequent times. In contrast, when spatial autocorrelation is
strong, data collected at different locations at a given point in time will contain are
large amount of redundant information, and so, the optimal design will consist of a
few sights that are sampled frequently. To quantify the optimal allocation of sampling
effort over space and time, we require estimates of the relative magnitudes of spatial
and temporal autocorrelation. The following considers Secchi depth data from two
environmental monitoring programs involving the lakes of the Savannah River Basin,
27
-------
the Clean Lakes Program of SC-DHEC, and the Regional Environmental Monitoring
and Assessment Program sponsored by EPA.
Data from the 17 sites of the Clean Lakes Program were used to estimate mag-
nitude of temporal correlation in Secchi depth. Observations were not collected at a
sufficient number of sites to effectively model spatial correlation using this data. The
Secchi depth Z(s<, tjk) at site s, and time tjk (month j in year k) was fit to the linear
model
Z{st,tjk) = H + Qi + Tj + E{&i, tjk)
where /i is the overall mean, a, is the effect of site i, Tj is the effect of month j, and
e(st,tjk) is the model error. The year of the observation did not enter significantly
into the model. Temporal dependence in between the data at times t and t! at site s
may be modeled through the temporal variogram
27i(l< " <'!) = var{Z(s,f) - Z(s,t')};
assume that the variogram depends only on the difference 11 - t'\ between the two
points in time. In general, there will be little variability (high autocorrelation) be-
tween data at times that are close together, and hence the temporal variogram will
be small for short time lags. Conversely, there will be high variability (low autocor-
relation) between data at times that are far apart, and hence the variogram will tend
to be an increasing function of time lag. If temporal trends are adequately modeled,
then the variogram will tend to approach an asymptote as the time lag increases; the
-------
time it takes to approach that asymptote is the range of temporal correlation. Pairs
of observations further apart than the range of temporal correlation are negligibly
correlated.
A nonparametric estimate of the variogram ran be obtained from the residuals
e(si,tjk) = Z(si, tjk) -Si- Tjt
where ai and fj are the ordinary least squares estimates of the parameters a* and Vj,
respectively. Then the temporal variogram at site s< may be estimated by
= aFTTT S l?(s«' bk) - tjk + r)|2,
j,k
where Ni(r) is the number of pairs of observations lag r apart in time at site s,. A
pooled estimate of the temporal variogram over all n sites may then be obtained from
,, M = L, jVi(r)7,(r)
7,< ' E?.i ATf(r) '
Figure 7 gives the nonparametric estimate of the temporal variogram for the Clean
Lakes program aata (closed circles). The curved line gives the least squares fit of the
Gaussian variogram model
27 t(r) = Co + Cg( 1 — e-or2). (1)
Estimates of the variogram parameters are cq — 0.2815, Cg — 0.207, and a = 0.144.
The large nugget effect ofco = 0.2815 suggests that there is a large amount of measure-
ment error, or short-term variability in Secchi depth. The estimate of a corresponds to
29
-------
u • w 'l I I I I I 1 1 1 I { I I I I I I 1 I I I » I I I I I » I I \ I I I I I I t I I I
D ID 2D 3D 4 D
Log Class Volue (in LACD1ST— units)
Figure 7: Temporal variogram for Clean Lakes Program data. The closed circles give
the nonparametric estimates, while the curved line gives the fitted variogram model.
a range of temporal autocorrelation of yJZ/ot = 4.6 months; observations more than
4.6 months apart are negligibly correlated (correlations are less than e~3 ^ 0.05).
Estimated monthly (Table 3) means show the same pattern as in Figure 1; Secchi
depth is lowest in April, increases to a maximum in June, and thai decreases to an
asymptote.
30
-------
Month Mean (m) Standard Error
April
2.831
0.092
May
3.247
0.090
June
3.839
0.093
July
3.684
0.093
August
3.499
0.096
September
3.454
0.091
October
3.583
0.094
Table 3: Mean secchi depth by month for Clean Lakes Program data
The REMAP data was used to model spatial correlation in Secchi depth. REMAF
observations were not collected at a sufficient number of times to effectively model
temporal autocorrelation. Moreover, the above results of analysis of the Clean Lakes
Program data suggest that the range of temporal autocorrelation is only 4.6 months,
which is shorter than the one-year time interval separating the REMAP observations.
The Secchi depth Z(sutj) at location Sj and year tj was fit to the linear model
Zfa, tj) = n + Tj + e(Si, tj),
where fj, is the overall mean, r; is the effect of year j, and is the model
error. The spatial dependence between data at locations s and u at a given time t is
31
-------
modeled through the spatial variogram
27,(||s - u||) = var{Z(s,t) - Z(u,t)};
assume that 27, depends only on the distance ||s - u|| between the two locations. In
general, there will be little variability (high spatial autocorrelation) between data at
close locations, and hence the temporal variogram will be small for short distance
lags. Conversely, there will be high variability (low spatial autocorrelation) between
data at far apart locations, and hence the variogram will tend to be an increasing
function of distance lag. If spatial trends are adequately modeled, then the variogram
will tend to approach an asymptote as the time lag increases; the distance at which
it approaches that asymptote is the range of spatial correlation. Pairs of observations
further apart than the range of spatial autocorrelation are negligibly correlated.
A nonparametric estimate of the spatial variogram at lag distance d, and at time
tj may be obtained from
where the sum is over all pairs of sites approximately d apart, and Nj(d) is the number
of such pairs of sites. A pooled estimate of the spatial variogram over all sampling
intervals may then be obtained from
27-(d)—zUw '
32
-------
Figure 8 give the nonparametric estimate of the spatial variograni for the REMAP
data (closed circles). The curved line gives the weighted least squares fit of the
exponential variograni model
27,(d) = Co + Ce(l - e~ad)- (2)
Restricted maximum likelihood estimates of the variogram parameters are cq = 0.726,
Ce = 1.2937, and 5 = 0.0797. The large nugget effect of Co = 0.726 suggests that there
is a large amount of measurement error, or microscale spatial variability in Secchi
depth. The estimate of a corresponds to a range of temporal correlation of 3/5 = 37.7
km; observations more than 37.7 km apart are negligibly correlated (correlations are
less than e~3 = 0.05).
The results described above show that Secchi depth exhibits both strong spatial
and temporal correlation in lakes of the Savannah River basin. This correlation
suggests that there is some redundancy in the data. The level of redundancy may be
quantified by computing the effective sample size, which is defined to be the number of
independent samples required to achieve the same precision of parameter estimate as a
sample of correlated observations of a given sample size. Consider, for example, model
based estimation of the mean. The variance of the sample mean of n uncorrelated
observations is equal to
33
-------
D
1 D
2D
3D
5D
4 D
6D
7D
BO
di stonei
Figure 8: Spatial Variogram for REMAP data. The closed circles give the nonpara-
metric estimates, while the curved line gives the fitted variogram model.
34
-------
while the variance of the sample of n correlated observations is equal to
2 n n
V2 = "J ,
71 «=1
where a2 is the population variance and pit is the correlation between observations i
and j. Then the effective sample size is equal to
Vx n2
nXV2~^iEU Pa
Table 4 gives the effective sample size for different sampling frequencies under the
fitted temporal Gaussian variogram model (1). When sampling up to three times per
year, the effective sample size is very close to the number of samples taken. How-
ever, as the sampling frequency increases, the redundancy in the data also increases,
resulting in effective sample sizes that are a fraction of the total number of samples
taken.
Table 5 shows the effective sample sizes of the two REMAP samples under the
fitted exponential variogram model (2). Notice that there is considerable redundancy
in the REMAP data; the effective sample size is less than a third of the number of
samples taken in each of the two years.
35
-------
Sample frequency Total Samples Effective Sample Size
Twice per Month
240
53.5
Once per Month
120
47.4
Six times per Year
60
38.6
Four times per Year
40
32.5
Three times per Year
30
27.7
Twice per Year
20
19.9
Once per Year
10
10.0
Table 4: Effective sample size as a function of sample frequency for a 10 year study.
Year Total Samples Effective Sample Size
1995 42 11.3
1996 35 10.9
Table 5: Effective sample size for the two REMAP sample years
36
-------
The optimal allocation of sampling effort over space and time was investigated un-
der varying ranges of spatial and temporal correlation. Serially alternating sampling
designs with varying sampling frequencies, number of sample stations per sampling
interval, and numbers of cycles were investigated. Each design has an equal total
sampling effort of n = 256 samples in a 16 x 16 unit region over an 8 year period.
Sampling frequencies of 0.5, 1, 2, 4, and 8 times per year were considered. The num-
ber of cycles ranged from 1,2,4, • • •, 8/, where / is the sampling frequency. Note that
when the number of cycles is equal to 1, we have a permanent station design, and
when the number of cycles is equal to 8/, we have an ever-changing station design.
Under a /c-cycle design with a sampling frequency of /, the total number of loca-
tions sampled is m = 32kff. These stations were randomly located in the 16 x 16
unit region under the constraint that no two stations be located within 8/y/m of one
another.
Table 6 gives the optimum number of cycles to estimate linear temporal trend for
a serially alternating under different ranges of spatial and temporal autocorrelation.
Here, the data Z(s, t) at the location s at time t are modeled as
Z(s,t) = 0O +fat+ e(s,t),
where the errors have exponential spatio-temporal correlation function
p(h,r) = corr{Z(s, t), Z(s + h,< + r)}
37
-------
= exp{—3 ||h|| /a, - 3r/at},
a, is the range of spatial autocorrelation, and at is the range of temporal autocorrela-
tion. The optimum design is defined to be the design under which the variance of the
general least squares estimator of f3l is minimized, and hence yields the greatest power
for detecting linear temporal trends in the data. Among the designs considered, the
optimal sampling frequency was 8 times per year. From Table 6, the optimal design
under a range of temporal autocorrelation of 1/2 year and spatial autocorrelation
of 8 units, the optimal design is an 8 cycle design. The optimal number of cycles
depends on the relative ranges of spatial and temporal autocorrelation. As the range
of temporal autocorrelation increases, the optimal number of cycles also increases,
but as the range of spatial autocorrelation increases, the optimal number of cycles
decreases.
4 Current Status of Section 305(b) Water Resource Monitoring
Although Section 305(b) of the Clean Water Act mandates that each state submit a
surface water quality assessment report to the Environment Protection Agency (EPA)
every two years, little guidance is given as to what specific data should be collected.
Consequently, states tend to design their water quality monitoring programs to meet
local priorities governing the allocation of their water resources, and in response to
local sources of environmental degradation.. Most states do not monitor all of their
38
-------
Range of Spatial Autocorrelation
1
2
4
8
16
0.0625
2
2
1
1
1
0.125
4
2
2
2
1
Range of
0.25
8
4
4
2
2
Temporal
0.5
8
8
8
4
4
Autocorrelation
1.0
16
16
8
8
8
2.0
32
16
16
16
8
4.0
32
32
32
16
16
Tabk 6: Optimum number of cycles for serially alternating designs under different
levels of spatial and temporal correlation.
39
-------
waterbodies every two years, and do not employ probability-based sampling designs
when selecting locations for sample sites. Instead sample sites are selected according
to a number of criteria, that differ among states and are not always well defined. For
example, the South Carolina Water Quality Monitoring Program selects 265 primary
stations that are influent or effluent to sub-basins, at major streams at state lines,
at the confluence of major streams, above and below major industrial and municipal
areas, in major lakes, and at the mouth of major tributaries. In Maryland, the
Basic Water Monitoring Program established a network of 68 sites in locations where
known water quality programs exist, and in rivers or major tributaries just above the
confluence with a river, but excludes areas with no serious water quality problems.
In either case, the representativeness of the sample sites cannot be readily quantified,
and hence estimates of the overall quality of the states' water resources are likely
be biased, especially in states which avoid areas thought to contain no serious water
quality problems.
In defense of state efforts, it should be pointed out that federal water resource mon-
itoring designs have not provided leadership by employing probability-based designs
themselves. The National Stream Quality Accounting Network (NAWQAN), the Na-
tional Water-Quality Assessment Program (NAWQA), and the National Status and
Trends Program (NS&T) all employ judgment sampling designs. It is interesting
to note that, while the Biomonitoring Environmental Status and Trends Program
40
-------
(BEST) uses a probability-based design to monitor presticides in starlings, it uses a
judgment sampling design to monitor pesticides in fish. The Environmental Monitor-
ing and Assessment Program (EMAP) is the only large federal program that employs a
probability-based sampling design to monitor aquatic resources, but this program was
only recently established and has a questionable future. In contrast, most programs
that monitor terrestrial resources, including EMAP, use probability-based sampling
designs (Olsen et al. 1998).
Recent years have seen attempts to improve water quality monitoring efforts. The
Intergovernmental Task Force on Monitoring Water Quality (ITFM) was established
in 1992 to review and evaluate national water quality monitoring efforts and to make
recommendations for improvements. The ITFM has recommended that states change
from a 2 year reporting cycle to a 5 or 6 year reporting cycle. By doing so, states
may increase spatial coverage of their water resources through the implementation of
serially alternating sampling designs.
In 1990, the EPA established the National 305(b) Consistency Workgroup to ad-
dress variation in sampling protocols and reporting methods among states. In re-
sponse to efforts of this workgroup, several states are exploring methods for obtaining
more representative samples of their water resources. For example, South Carolina is
establishing Watershed Water Quality Management (WWQM) Stations at the down-
stream access of every Natural Resource Conservation Service (NRCS) watershed unit.
41
-------
Thus, a census of all watershed units is obtained. However, the representativeness of
the resulting data depends on how watershed units are partitioned. Nevertheless, the
WWQM stations provide good spatial coverage of the South Carolina's watersheds.
Some states have implemented probability-based sampling designs. The Delaware
Department of Natural Resources and Environmental Conservation selected a sample
of 96 sites, randomly selected from a list frame of 3200 roadway crossings of nontidal
streams in the northern two counties of the state. The Maryland Department of
Natural Resources randomly selected a sample of about 350 sites from a list frame of
all first, second, and third order stream reaches.
4.1 Response to 305(b) Consistency Workgroup
The failure of states to adapt probability-based sampling designs in their water quality
monitoring efforts may in part be due to misperceptions regarding their limitations.
Many of these misperceptions can be found in the draft report of the Monitoring
and Assessment Design Focus Group of the 305(b) Consistency Workgroup (1996),
which lists a number of disadvantages and concerns with probability-based sampling
designs. The following shall address each these by suggesting how a probability-based
design that can be used to address each of these concerns. Note that these proposed
designs may require some modification for specific applications.
Concern 1. Probability-based designs will not identify new problem sites unless
42
-------
they happen to be selected randomly. A similar statement could be made about judg-
ment sampling designs: Judgment sampling designs will not identify new problem
sites unless they can be identified by the investigator. Thus, under a judgment sam-
pling design, the ability to identify new problem sites is limited by the judgment of
the investigator. The probability of identifying new problem sites can be increased by
increasing the spatial coverage of a sampling design either through implementation
of serially alternating or rotating panel designs, or through sampling a new set of
sites during each sampling interval. A more efficient approach would require assump-
tions regarding causal mechanisms, and then information on the causal variables,
preferrably over the entire population. For example, an investigator might attempt
to identify all potential point sources of environmental contamination (for example,
from a listing of all sewage treatment plants, or all paper mills in the state). How-
ever, sufficient resources may not be available to sample all of the potential point
sources. Farther information regarding the characteristics of the identified potential
point sources might be used to select which ones are most likely to pose environmental
hazards, but the cost of compiling such information may be prohibitive. Moreover,
some potential point sources which appear to pose to no environmental hazard, and
hence are not included in the sample, may in truth pose a significant environmental
hazard.
A probability-based sampling design can be used to identify which potential point
43
-------
sources pose a significant environmental hazard. This can be accomplished by se-
lecting a simple random sample of potential point sources in the first year of the
investigation. Selected sites that show significant environmental damage may then be
sampled in each of the next years, perhaps until they meet or exceed regulatory stan-
dards. In the second year, a simple random sample of the remaining sites is selected,
and again, those sites showing significant environmental damage are retained. This
process is repeated in subsequent years until all potential point sources are sampled
at least once.
In states where it is prohibitively expensive to identify all potential point sources
of environmental contamination, a two-stage sampling design might be used to assist
in the identification of point sources as follows: The state's water resources are par-
titioned into the NRCS watershed units. In the first year, a simple random sample
of the watershed units is selected. Then the potential point sources of environmental
contamination are identified within each of the selected watershed units. A simple
random sample of the identified point sources may then be selected. In each of the
subsequent years, a simple random sample of the heretofore unsampled watershed
units is sampled until all watershed units have been sampled. After that time, the
process may be repeated. Thus, in each year, only those potential point sources within
the selected watershed units need by enumerated, from which a simple random sample
can be selected for field sampling.
44
-------
Adaptive sampling designs are particularly well suited to the identification of new
problem sites under nonpoint sources of environmental contamination. Start with a
simple random sample of sites. Then cluster new sample sites around each site show-
ing a level of environmental degradation about some threshold. A response-surface
model (Myers 1976) may be fit to the data, to identify locations where additional
sampling is required to obtain an estimate of the location of the local mayimnm
level of environmental degradation. Occasionally, additional sample points should be
randomly selected to ensure the identification of new problem sites.
Concern 2. Probability-based designs will not determine temporal trends at priority
sites. There are a number of very legitimate reasons why specific priority sites may
be of interest. For example, we may wish to investigate the efficacy of environmental
remediation at locations of sewage or industrial discharge, or hot spots known to show
especially high levels environmental damage. To assess the efficacy of such restoration
efforts, however, it may be necessary to compare temporal trends at these priority
sites to temporal trends at reference sites, selected to represent conditions existing
prior to environmental degradation at the priority sites. If interest lies in the levels of
contaminants in the waters of a river or stream, then it may suffice to locate reference
sites upstream of priority sites and a probability-based sampling design need not be
considered. If, however, interest lies in the restoration of the ecological community at
priority sites, then upstream sites are not guaranteed to be representative of conditions
45
-------
that had existed prior to environmental degradation at priority sites, and hence, a
probability-based sampling design should be used to select reference sites. To further
ensure the representativeness of reference sample sites, a stratified random sampling
design might be used, where the allocation of sampling effort to strata is proportional
to the number of priority sites found in each stratum. Alternatively, reference sites
may be located some random distance and direction from each of the priority sites,
or if sufficient resources are available, two or more reference sites may be clustered
around each priority site.
Concern 3. Probability-based designs are not designed to assess improvements
in specific waterbodies or watersheds due to controls, enforcement, or restoration.
When assessing improvements at specific waterbodies or watersheds is of interest,
then each of the specified waterbodies or watersheds must be sampled. However,
the question remains as to what specific locations should be sampled within those
waterbodies or watersheds. If the water quality of a stream or river is of interest,
it may suffice to sample at the effluent end of that stream or river. If, on the other
hand, the status of the ecological community, or the quality of bottom sediments are
of interest, a probability-based design is required to ensure that the sample sites are
representative of the waterbody or watershed of interest. Here, individual waterbodies
or watersheds can be treated as strata for a stratified random sampling design. The
use of a judgment sampling design to select what specific sites are to be sampled
46
-------
within each waterbody or watershed can result in biased estimates of status and
temporal within that waterbody or watershed.
Concern 4¦ Probability-based designs respond poorly to political priorities. With-
out more specifics regarding what political priorities are to be considered, it is not
possible to make specific recommendations as to how a probability-based sampling
design may accommodate them. However, the sampling intensity can adjusted to
ensure that a higher density of sample sites is obtained in high priority regions at the
cost of a lower density of sample sites in low priority regions.
Concern 5. If all 305(b) assessments were based on changing probabilistic sites,
States would no longer track specific waterbodies and mapping a spatial analysis would
be curtailed. The use of changing probabilistic sites does not preclude temporal and
spatial analysis of the data. Statistical methods for such analyses shall be discussed in
Section 5.2 below. Regardless of whether a probability-based or judgment sampling
design is used, the power of analysis for temporal trends within specific waterbodies
will depend on how many observations are available within those waterbodies. How-
ever, if permanent sample sites are selected according to a judgment sampling design,
then the only statistically justifiable inferences are with respect to those specific sites.
Under a probability-based design, statistically justifiable inferences regarding tempo-
ral trends can be made regarding the waterbodies as a whole. Moreover, statistical
tests for trend are also likely to be more powerful under changing probabilistic sample
47
-------
sites than under a fixed station design (See Section 4.2).
Concern 6. Probability-based designs require significant up-front effort for proper
design and long-term, adherence to the study plan. The ability to make statistically
justifiable inferences regarding the water resources as a whole should justify the added
up-front effort required to obtain an appropriate probability-based design. The costs
of long-term adherence to the study plan can be reduced by using a serially alternating
design (see Section 4.2) instead of selecting a new set of probabilistic sample points
for each sample interval.
Concern 7. Under a probability-base design, states would lose the benefits of exist-
ing sites with many years of data. In Section 6.0, a method for combining historical
data from a judgment sample design with new data from a probability-based is de-
veloped. The proposed method calls for a period of overlap in which observations
are collected from both designs. Then the spatio-temporal autocorrelation among the
observations from both data bases is exploited to back predict what data would have
been obtained had a probability-based design been used from the very beginning of
the monitoring program. The resulting predictor relies heavily on the historical data
base, especially for predictions many years in the past.
Concern 8. Determining sources of impairment may be beyond the capability of
probability-based designs. Results of observational studies can not provide definitive
evidence that a given factor or combination of factors are responsible for environ-
48
-------
mental impairment. Correlations between levels of environmental impairment and
alleged sources of impairment may be spurious. Moreover, the highest contaminant
concentrations are not necessarily located near their sources, but may be located
downstream where local site characteristics may promote adsorbtion of contaminants
in the sediment or their entry into the food chain. Definitive evidence for causal
relationships can only be obtained through randomized experimental manipulations
of the environment. However, such manipulations may not only be impractical, but
also unethical. Nevertheless, it may be possible to gain some insight through a care-
fully planned observational study. Sites should be selected in a factorial arrangement
in which all combinations of high and low levels of each of the alleged causal fac-
tors are equally replicated. However, the information required for such a design may
not be readily available. A more cost-effective approach would be to implement a
probability-based design in which the alleged causal factors are measured along with
the measures of impairment. Supplemental sites may then be added to provide infor-
mation from factor combinations missed by the probability-based design, improving
the power to separate out causal contributions.
Concern 9. If the design does not allow sampling at access points like bridges,
sampling elsewhere will be difficult and expensive. The savings incurred by sampling
at arrepg points may allow larger sample sizes under tight budgetary constraints,
«nH hence potentially more precise estimates of environmental parameters and more
49
-------
statistical power for detecting trends. Probability-based sampling methods can be
used to select what access points are to be included in the sample. However, to
statistically justify inference to the water resource as a whole, evidence is required
that the access points are representative of that water resource, or alternatively, an
estimate of the bias introduced by sampling at access points. There are a number of
reasons why the representativeness of access points may be questioned:
• The density of access points such as bridges will tend to be higher in regions of
high human population density, and lower where human populations are sparse.
• The level of environmental impairment may vary with the suitability of locations
for bridge construction. Do we really want bridge engineers to determine where
we sample?
• The bridges themselves may adversely affect their local environments.
Section 5.3 discusses how each of these concerns may be addressed using probability-
based designs.
Concern 10. Concern over the number of years required to determine spatial or
temporal trends in a basin or state. Probability-based designs require no more years to
determine spatial or temporal trends than judgment sampling designs. The power to
detect such trends is a function of the sample size, and the degree of spatio-temporal
correlation in the data. If probability-based designs show less spatio-temporal correla-
50
-------
tion, as would often be the case, then they should be more powerful than a judgment
sample of the same size. It should be kept in mind that spatial and temporal trends
should be interpreted with caution. Ecological systems are inherently dynamic; so,
in order to investigate the impact of management on environmental impairment, we
must distinguish between trends resulting from management practices and natural
environmental fluctuations. This requires an understanding of the natural fluctua-
tions that may occur in a waterbody that might only be obtained from collecting data
over a number of years.
Concern 11. Concerns over the expense of sampling sufficient sites for statistical
rigor and also availability of technical support for States. Given the high cost of
environmental monitoring, it is essential that the sampling design yield the strongest
possible statistical inference with respect to the states' water resources. Regardless of
sample size, statistically justifiable inferences can be made regarding the status of the
water resources as a a whole under a probability-based sampling design. Since the only
statistically justifiable inferences that can be made under a judgment sampling design
are with respect to status and trends at the sample themselves, judgment sampling
designs make very inefficient use of funds allocated to environmental monitoring.
The EPA should be responsible for providing technical support to the states for
implementing probability-based sampling designs, and analyses of the resulting data.
51
-------
4.2 Combining Data Across States
In addition to the biennial water quality assessment reports that must submitted by
the states, Section 305(b) of the Clean Water Act also mandates that the EPA submit
a comprehensive assessment of the quality of the nation's water resources to Congress
every two years. The latter requires the combining of data submitted in the states'
reports. Given that most states employ judgment sampling designs, valid statistical
inference is limited to statements regarding what percentage of sample stations sup-
port their designated uses (e.g., drinking water supply, fish consumption, recreation,
etc.), and what percentage of stations show improving or degrading water quality.
Statements regarding what percentage of water resources support their designated
uses, or show improving or degrading water quality cannot be statistically justified.
The combining of data across states would be straightforward if all states were
to employ probability-based sampling designs and provided that they use the same
defintion for the target population, and consistent measurement protocols. Then the
different states can be treated as strata, and the mean level of an environmental
indicator across the 50 states can be estimated by
£ = ' £*' (3)
where is the estimated mean level of the environmental indicator in state t, Jj4*|
is the quantity of the water resource (e.g., stream miles, total surface area of lakes
52
-------
or estuaries, etc.) in state i, and \A\ is the total quantitative of that resource in the
nation (i.e., \A\ = \Ai\). The precision of this estimate can be estimated through
its variance
1 50
*»(£) = pja £ • var(^). (4)
In a similar manner, the proportion of the nation's water resources showing a given
condition (e.g., degraded, supporting designated uses, showing improving conditions,
etc.) can be estimated by
j 50
(5)
where pt is the estimated proportion of the water resources of state i that show that
condition. The corresponding variance estimate is
1 50
= MP £ '^l2' varC^»)- (6)
The above estimates do not require that the same sampling design be employed
by all states; they only require that each state employ a probability-based sampling
design. Estimates of state means fiit proportions Pi, and their corresponding variances
depend on the particular sampling designs employed by each state. However, unbiased
estimation across the 50 states requires some consistency among states with respect
to what data are collected and how the data are obtained.
Differences among states in definitions of target populations (e.g., what orders of
streams or sizes of lakes are sampled) can lead to biased estimates of the status of
53
-------
the nation's water resources. For example, if some states do not sample lower order
stream reaches, and such stream reaches tend to have better (lower) water quality
than higher order reaches, then the overall proportion of stream miles meeting a water
quality standard will be underestimated (overestimated). To avoid this source of bias,
the EPA (with input from the state agencies) should provide the states a clear and
concrete definition of the target population of water resources that should be sampled.
Depending on their needs, individual states may elect to sample sites not included in
this target population, but data from those sites should be reported separately.
Differences among states in sampling protocols (e.g., at what depth a water sample
is obtained, when samples are taken, how samples are handled and stored following
collection), and laboratory procedures for assaying samples may also lead to biased
estimates. This bias may be reduced by having states adopt consistent sampling pro-
tocols, and laboratory procedures for assaying samples (ITFM 1995). Nevertheless, it
is likely that there will remain some variation among state field crews and laboratories
with respect to how sampling protocols and laboratory procedures are applied. To
reduce the resulting biases, groups of states should engage in joint sampling efforts,
in which field crews from the various states sample the same sites using their own
sampling protocols, and their own laboratories for assaying resulting samples. The
analysis of variance model
Vij = n + 0i + 7j + £ij (7)
54
-------
can then be fit to the resulting data at site j using the field crew from state
i. Here, // is the overall mean, ft is the bias attributed to the methods for state
i, "Yj is the effect of site j, and £ij is the model error. The bias terms are not
individually estimable unless further assumptions are made; for example we wsnmp
that ori alternatively, that one of the individual states uses unbiased
methods (i.e., = 0 for some i). The parameters of (7) can be estimated using the
generalized linear model procedure (PROC GLM) of the Statistical Analysis System
(SAS Institute 199?). Given estimates of the bias terms, a bias corrected estimate of
the overall mean can then be obtained from
¦i 50
P-pfEW-CA-ft)-
Note that if the analysis of variance shows that there are no significant differences
among the states, then no bias correction is necessary.
Note that the above does not require that all 50 states sample each site. Instead,
it suffices that the data from all of the states be connected (sensu Searle 1971, pp.
319-324). To determine if all states are connected, create a table showing which state
crews sampled which sites. For example, see Figure 9 in which six sites are sampled
by six states; here state 'B' sampled sites 2 and 5, and site 2 was sampled by both
states 'B' and 'F'. lb find the connected subsets, draw horizontal and vertical line
segments connecting any pair of observations on the same row or column; observations
that can be connected by such line segments form a connected subset; in Figure 9,
55
-------
4£
CO
Figure 9: Connected subsets of states.
for example, states 'B' and 'F' form one connected subset, states 'A' and 'D' form a
second connected subset, and states 'C' and 'E' form a third connected subset. Since
there are more than one connected subsets, the data are disconnected, and hence we
would not be able to estimate the relative biases of the states1 data. The states would
be connected if, for example, state 'B' were to sample the additional sites 3 and 4.
The above analyses also assume that there is no interaction between states and
6ites, so that the bias in a given state's methods does not depend o» site. Tukey's
procedure (Snedecor and Cochran 1980, pp. 283-285) may be used to test for this
interaction. If a significant interaction is found, then the analysis variance model may
be fit to log transformed data:
lnVij = /i + & + 7j + e
-------
Then, the overall mean can be estimated by
Regardless of efforts to improve consistency among state water resource moni-
toring programs, it is likely that states will continue to differ with respect to what
variables are measured. Moreover, it is not necessarily appropriate for states with
widely different types of water resources to measure the same variables. This is espe-
cially true for biotic measurements, since there is considerable geographic variation
in the composition of aquatic communities over the United States. Obviously, esti-
mation of the overall mean level of an environmental variable across the 50 states
requires that the same variable be measured in each state. On the other hand, esti-
mation of the proportion of water resources showing a given condition (i.e., degraded,
supporting a designated use, showing improving conditions), do not require that the
same variables be measured across the states. However, the quality of the estimates
could be improved by some general agreement with respect to definitions of what is
meant by a degraded condition, or when a waterbody supports a designated use or
shows improving conditions. Without such an agreement, expression (5) would only
estimate what proportion of the nation's water resources were designated as showing
a given condition, and not necessarily in any clearly defined way what proportion
actually shows that condition.
57
-------
5 Design Alternatives for Section 305(b) Water Resource Monitoring
Statistically defensible methods for combining data across the 50 states require that
the states replace their current judgement sampling designs with probability-based
sampling designs. The specific probability-based design to be implemented by a
given state depends on the resources available from that state to support monitoring
efforts, the logistical constraints under which monitoring is to be carried out, and
the characteristics of that state's water resources. Therefore, detailed descriptions
of specific monitoring designs are beyond the scope of this report. The following
broadly outlines some alternative probability-based designs that may be implemented
for water resource monitoring. For each sampling design, methods for estimating the
population mean, population proportion, and the total mass of an environmental
contaminant are considered.
5.1 Sampling Lakes
The recommended approach to sampling lakes depends on the monitoring objectives,
the distribution of sizes and types of lakes within a state, and the information available
on the population of lakes to be sampled. The objectives may call for sampling all of
the larger lakes in the state, but resources are unlikely to be available for sampling
all of the smaller lakes each year. For the latter, we may require a random sample.
58
-------
5.1.1 Sampling Large Lakes
A stratified random sampling design can be used to sample the large lakes within
a state, where each lake is treated as a stratum. Under such a design, ti, sample
sites are randomly located within each of m large lakes according to a simple random
sampling design (Figure 2); i = 1, • • •, m. Suppose that sufficient funds are available
to sample n sites during each sample interval. Then the recommended allocation of
pampling effort calls for selecting
sites from lake i, where |^4t| and |K|, respectively, are the surface area and volume
of lake z. Thus, lakes are sampled proportional to their sizes. The allocation scheme
is optimal (minimizes sampling variance) under the assumption that the within lake
variances are homogeneous (i.e., they are identical among the large lakes). If the
within-lake variances are heterogeneous, then an optimal allocation scheme would
call for increased allocation of sampling effort within lakes showing high variability,
and decreased allocation within lakes showing low variability. Different environmen-
tal variables are likely to show different patterns of within-lake variability, so that
an allocation scheme is optimal for one variable is not likely to be optimal for the
remaining variables. Moreover, the within-lake variances are not likely to be known
a priori, and hence, allocation proportional to lake size is recommended.
59
-------
Under the stratified random sampling design described above, the mean level of
an environmental variable across the surface area of lake i can be estimated by the
sample mean y, of the n* observations in that lake. The precision of this estimate can
be estimated by
varfo) = J-
ni
where s? is the sample variance of the n* observations in lake i. The overall mean
across the surface of all m large lakes can be estimated by
with corresponding variance estimate
The proportion of the surface area of lake i showing a given condition (i.e., de-
graded, supporting designated uses, etc.) can be estimated by pt, the proportion of
sample sites showing that condition. The corresponding variance estimate is given by
The proportion of the surface area of all m large lakes showing that condition can
then be estimated by
fc-iifiw-ft
60
-------
with corresponding variance estimate
Suppose that the concentration of an environmental contaminant in a given water
sample is expressed in terms of mass per unit volume. Then the total mass of that
contaminant in lake i can be estimated by
c\Ai\ ¦*
Ti =
» j-1
where dtJ and yt3 are the water depth and concentration at site j in lalfp i, and
the constant c is defined to achieve the appropriate units of measurement. The
corresponding variance estimate is
cW
var(fj) =
n< 1 / \
2
.
n(n — 1)
Then t|he total mass of the contaminant across all m large lakes can be estimated by
m
with corresponding variance estimate
vaf(rBt) =£var(ri).
»=i
Instead of locating sample sites according to a simple random sampling design
within each of the large lakes in the population, sample sites can be located accord-
ing to a randomized-tessellation stratified design (Stevens 1997). Under such a design,
61
«=i
m
-------
a grid of contiguous polygons is randomly placed over the study region, as is shown
for example in Figure 10, where a hexagonal tessellation is randomly located over
Lake Jocassee. Then a single site is randomly located within each of the polygons.
Only sites falling in the region of interest are included in the sample. The sampling
variance under the randomized-tessellation stratified design is smaller than that un-
der the simple random sampling design, especially if the data shows strong spatial
correlation. The Yates-Grundy estimator for its variance is reasonably stable under
strong spatial correlation. If there is a large measurement error, or if there is large mi-
croscale variation in the data, however, the Yates-Grundy estimator for the variance
can be unstable; in such cases, the randomized-tessellation stratified design cannot
be recommended.
5.1.2 Sampling Small Lakes
The recommended approach to sampling small lakes depends the quality of informa-
tion that is available regarding what lakes are present in a state. Ideally a listing of all
small lakes in the state would be available, perhaps from USGS maps, aerial photos,
or satellite images. Then a simple random sample or stratified random sample could
be selected from the list frame of lakes. However, the cost of obtaining a list frame of
all 1»1ms within a state may be prohibitive. In this case, a two-stage sampling design
may be required.
62
-------
Figure 10: Randomized-tesseUation stratified design for Lake Jocassee.
63
-------
Simple Random Sample. Suppose that a list frame of all N lakes in a state is
available. Then a simple random sample n lakes can be obtained from randomly
drawing numbers between 1 and N until a sample of n unique lakes is drawn. Then
one sample site is located within each of the sampled lakes. The decision as to
what actual location is selected within each of the sampled lakes depends on the
variable that is to be measured and the monitoring objectives. If it is desired to make
inferences about the total mass of contaminants in the lakes of a given state, then sites
should be selected randomly. A random sample would also be required to estimate
the proportion of the volume of lake waters or surface area of lakes of a state that
are impaired. If, on the other hand, it is desired to make inferences about the mean
level of an environmental variable accross the population of lakes, or the proportion
of lakes showing impaired conditions, random selection of sites within lakes may not
be necessary. In such cases, water samples may be taken from the deepest part of the
lake, or biota may be sampled in the multiple habitats around the lake in which they
are found.
Under a simple random sampling design, the mean level of an environmental
variable across the lakes in a state can be estimated by the sample mean p, with
corresponding variance estimate
. /AT — n\ s2
var(S) = ("TrJ
where s2 is the sample variance. The proportion of lakes showing a given condition
-------
(i.e., degraded, supporting designated use, etc.) can be estimated by p, the proportion
of sample sites showing that condition. The corresponding variance estimate is
'N — n\ pil-p)
— fN ~n\ P(1 -P)
™(p)={—) irrr-
Instead of estimating the mean level of an environmental variable across the lakes,
we may wish to estimate the mean level of that variable across the surface area of
those lakes, or over the volume of the lakes. In such cases, sample sites should be
randomly located within each of the selected lakes. The ratio estimators
- ELi M • in ...
TX.AM ' ( )
and
fl (9)
EtilKI ' 1 J
can then be used to estimate the mean level of the variable across the surface area
and volume of lakes in the population, respectively. Here, \At\ and |V*| respectively
are the are? and volume of lake i, and y, is the value of the variable of interest in
lake i. Thus, the data are weighted by the sizes of the lakes that were sampled. The
variance estimates are
N(N-n) E^aAI-Ift-AIAI)2
—m* (10)
and
N(N-n) ET„ (|K] y,-AJKI)2 .
= ^ ' '
65
-------
where |A| and |V| respectively are the total surface area and volume of the N lakes
in the population. If \A\ and |V| are unknown, we may replace these quantities in
the expressions above by their estimates
ni?i n £T
To estimate the proportion of the total surface area or volume of lakes that shows a
given condition, replace y* in the expressions above with a binary variable that takes
the value 1 if sample site i shows that condition, and the value 0 if otherwise.
The total mass of an environmental contaminant can be estimated by
r = \V\- ft,, (12)
with cprresponding variance estimate
var(f) = |V|2var(ft). (13)
If the total volume is unknown, total mass may be estimated by
T = ^£|A|-di-V. (14)
71 tel
where e£» is the water depth at sample site i. The corresponding variance estimate is
1=^9 {ir } (i5)
To sample lakes over time, a serially alternating design with k cycles may be
implemented by randomly partitioning the small lakes into k sets of size n £ N/k.
66
-------
This may be accomplished by taking a simple random sample of size n from the N
lakes in the list frame to form the first set of lakes. The second set of lakes is obtained
by taking a simple random sample of size n from the remaining N — n lakes. This
process is repeated until all lakes have been assigned to sets. Lakes in set i are then
sampled at time intervals i, i + k, i 4- 2k, • • •, as shown in Thble 1 for a k = 4 cycle
design. Observations from each time interval can be treated is though they were
obtained from a simple random sample from the original population of N lakes, and
so, population parameters may be estimated as described above. The proportion of
lakes showing improving (deteriorating) conditions can be obtained by dividing the
number of lakes showing improving (deteriorating) conditions by N. Since the entire
population of lakes is sampled, this estimate has no sampling variance.
Stratified Random Sample. Suppose that in addition to a simple listing of lakes,
further information is available about each lake in the list frame. For example, we
may know which lakes are man made and which lakes are natural, we may have a
list of oligotrophy and eutrophic lakes, or a description of the geological formation
on which each lake lies. If the variable of interest depends on such characteristics,
then a stratified random sampling design can be used to reduce sampling variation,
«nH honrp improve the precision of population parameter estimates. Strata may also
correspond to their designated uses (i.e., drinking water, fishing, etc.). Stratified
67
-------
random sampling designs can also guarantee that rare types of lakes are included
in the sample, and to allocate more sampling effort to lakes that are deemed to be
ecologically, economically, or sociologically important. Populations of lakes will tend
to contain a very small number of larger lakes, and a very large number of small lakes,
and so, a simple random sample may not pick up any of the important large lakes in
the population. By stratifying by lake size, we can ensure that an adequate sample
of large lakes is selected.
Under a stratified random sampling design, the list frame of lakes is first parti-
tioned into K strata; let Nh denote the number of lakes in stratum h. Then a simple
random sample of rih lakes is obtained from stratum ft, h = 1,2, • • •, K. Finally, one
sample site is randomly located within each of the sampled lakes. The number of
lakes sampled from each stratum may be proportional to the total number of lakes in
each stratum
proportional to the total surface area of lakes in each stratum
or proportional to the total volume of lakes, in each stratum
Using one of these sample allocation schemes as a starting point, sampling effort can
68
-------
be increased in strata deemed to be more important, and reduced in strata deemed
to be less important.
Since a simple random sample of lakes is obtained from each of the strata, the
stratum means and proportions, the total mass of contaminant within a stratum, and
their corresponding variances can be estimated using the same methods as described
above for the simple random sampling design. The mean level of an environmental
variable across the N lakes in the population can be estimated by
I k
= T7 £ NhVh>
fc=i
where
1
^ TIl
»=1
is the sample mean of the observations • • ¦, yhnh from stratum h. The correspond-
ing variance estimate is
K 2
~ nh
where
? n=i vl - nhVh
Sh nh- 1
is the sample variance of the observations from stratum h. Similarly, the proportion
of showing a given condition can be estimated by
P* = TF £
iV fc-i
69
-------
where is the proportion of observations from stratum h showing that condition.
The corresponding variance estimate is
var(pgt) = ^5 H Nh{Nh - •
h=l nfc
The mean level of an environmental variable across the surface area or volume of
the lakes may be estimated by
1 k
Msts "|a| ^ '
and
1 x
Mstv 777T ^ 1 |Vh| '
\v I h=i
respectively, where \Ah\ and |V^| are the total surface area and volume of lakes in
stratum h, and |A| and |V| are the total surface area and volume of all lakes. Here,
p,hs and p.hv are computed from observations in stratum h using expressions (8) and
(9), respectively. The corresponding variance estimators are
I k
varfoj = pjp £ l^fvar^jJ,
and
w(Atv) = £ |V)ipvaf(/2hJ,
respectively, where varO^,) and vax(p,hv) are computed from the observations from
stratum h using expressions (10) and (11), respectively.
70
-------
The total mass of an environmental contaminant over all lakes can be estimated
by summing the estimated mass of that contaminant over the strata. That is, take
K
h=l
where r h is computed from the observations in stratum h using either expressions
(12) or (14). The variance of f8t can then be estimated by
K
va?(7v) = ]Tvar(rh).
To sample lakes over time, a serially alternating design may be implemented in
each of the strata as described above for the simple random sampling design. If a k
cycle design is implemented in each stratum, then at each time the sample allocation
is proportional to the number of lakes in each stratum. Note, however, that there is
no requirement that the number of cycles k be identical among strata. By using a
smaller number of cycles, more sampling effort can be made in more important strata,
while larger number of cycles can be used in less important strata.
Two-Stage Sample. The implementation of the above sampling designs requires a
list frame of all lakes in the target population. The cost of obtaining such a list frame
can be prohibitive. These costs can be reduced by implementing a two-stage sampling
design. Under a two-stage sampling design, the state is first partitioned into primary
sample units, which may correspond to counties, watershed units, or a contiguous
71
-------
grid of hexagonal or square quadrats. The first stage of the design is comprised of the
random selection of n primary sample units from the population of N primary units.
Then the lakes are enumerated within each of the selected primary sample units. The
second stage of the design is comprised of the random selection of lakes within each
of the selected primary units. Typically, allocation of sampling effort among primary
units is proportional to the number of lakes in each of the selected primary units.
Thus, if a total of m lakes are to be sampled, select
from primary unit i, where M, is the number of lakes in the t-th selected primary unit.
Note that for variance estimation, we require m, > 2 (unless a particular primary unit
only contains one or two lakes).
To sample lakes over time, a serially alternating design with k cycles may be
implemented by randomly partitioning the N primary units into sets of size n £ N/k.
Primary units in set i are then sampled at time intervals t, t + k, i + 2k, ¦ • •, as shown
in Table 1 for a A: = 4 cycle design. In each time interval, the lakes are enumerated
within each member of the appropriate set of primary units, from each of which, a
simple random sample of lakes is drawn. Thus, after k time intervals, all of the lakes
within the state will have been enumerated.
Within a given time interval, the mean level of an environmental variable across
72
-------
the population of lakes in the state can be estimated by
= (16)
where M is the total number of lakes in the state, and is the sample mean value of
the environmental variable among the selected lakes in primary unit i. The variance
of /}// may be estimated by
®(5"> " (I) (^) n + 55*~ m')^'
where sj is the sample variance of lakes selected from primary unit i, and
, Z?., - i (Z?„ M.V.f
S"~ n -1
Although the total number of lakes M in the state will be known after the first k time
intervals of the serially alternating design described above, this quantity may not be
known beforehand, or if this serially alternating design is not implemented. The total
number of lakes in the state may however be estimated by
_ \r "
n £
Substituting M into expression (16), we obtain the ratio estimator for the population
mean:
ST-i
"* £S.i M '
whose variance may be estimated by
/N\2/N — n\sl N
™r(^) = (l) (—) n + { ~ m<)^'
73
-------
where
S-. M?(R ~ g«)2
s" inn •
Similarly, the proportion of lakes showing a given condition (i.e., degraded, sup-
porting designated use, etc.) may be estimated by
if the total number of lakes M is known, or by the ratio estimator
~ ET-1 Mfr
Pr a, m,
if the total number of lakes is unknown. Here, p. is the proportion of lakes sampled
in stratum i that satisfy that condition. The corresponding variances are
where
ET.1 Mm - i (E?.i mf
4= — «.
p n — 1
and
where
rs _
—\ fN\2 fN-n\ N A.,,., vPt(l-P«
8* n- 1
74
-------
The mean level of an environmental variable across the surface area of the lakes
may be estimated by
where and \Aij\ are the observation from and surface area of lake j in primary unit
i. The variance of Jit may then be estimated by
var(/2(
1 f^(AT-„)^i#2
•> = jl}?] nfa-n
+
n(n-l) fr{
N " Mi(Mi - rm)
•j mj ~ mj
~ ' V*i " 5Z My I
"** J=1 7"» J=1
^ -rrii)^ flA , 1 ^, , \ ^ f, A , 1
n £ m,(mi - 1) g 1""j" "• ^
where
is the estimated total surface area of the lakes in the population. The proportion of
the surface area satisfying a given condition can be estimated by replacing ytJ in the
expressions above with a binary variable that takes the value 1 if that condition is
satisfied in lake j of primary unit t, and takes the value 0 if otherwise. The mean
level of the environmental variable across the volume of the lakes may be estimated
by replacing the lake areas |A0| by the corresponding volumes |V^|.
The total mass of an environmental contaminant may be estimated by
T = — J2 ZT IJ I A/I • • yij
where dy is the water depth at the sample site in lake j in primary unit t. The
75
-------
corresponding variance estimate is
l^Mk^lA , J \2
n(n ~ 1) ^ \m» j=i n£[mk£{ /
, J 1 ^1J1 , J V
5.2 Sampling Rivers and Streams
Rivers and streams axe unique among natural resources in that, except for regions
under tidal influence, the waters flowing past a given point originate from upstream
of that point. Thus, observations of the water quality at the effluent end of a water-
shed are in some sense representative of the waters flowing through that watershed.
This observation has led many water quality monitoring programs to target sampling
at the effluent ends of watersheds. For example, South Carolina's Watershed Water
Quality Management (WWQM) program targets sites at the downstream access of
every National Resource Conservation Service (NRCS) watershed units. Note that
not all NRCS watersheds units are watersheds unto themselves, but are subwater-
sheds. A subwatershed is a subset of a watershed obtained by subtracting out those
regions covered by other watershed units in the collection. A mass balance model can
be constructed from WWQM sample stations provided sufficient information is avail-
able. The total mass of a contaminant passing by a sample station can be computed
by the product of the concentration of that contaminant in a water sample times the
volume of water flowing past that station per unit time. Then the contribution of the
76
-------
watershed unit to that mass can be obtained by subtracting the mass of contaminants
input into that watershed unit by upstream watershed units from the mass of con-
taminants effluent from the watershed unit. However, such computations require the
assumption that no contaminants are lost due to adsorption onto bottom substrates,
uptake in organisms, or evaporation.
Unless a mass balance model or other mechanistic modeling effort is planned,
there is very little reason to target sampling at confluences of waterways. Moreover,
since representativeness of such sample site is not known, such targeted efforts are
not appropriate for sampling the bottom substrate, or biotic communities. Only a
probability sampling design can be used to obtain unbiased estimates of the mean
level of an environmental contaminant across the length of rivers and streams, the
proportion of stream and river miles that support designated uses, or the total mass
of an environmental contaminant in the streams and rivers of a state.
The following considers three broad design alternatives for sampling rivers and
streams within a state. The choice of design depends on what information is available
on the population of streams and rivers, and the resources available for planning
sampling efforts.
Simple Random Sampling. A simple random sampling design requires a digitized
map of all rivers (and streams) within the state. Such a design can be constructed by
77
-------
first partitioning the rivers into river segments, defined to be any length of river con-
taining no branches. The river segments are then laid out end to end in any arbitrary
order. Finally, n sample points are obtained by random selection of locations between
0 and L, the total length of the river segments. Since there tend to be more miles
of first-order streams, than higher-order streams, a simple random sample will tend
to be dominated by first-order stream sites. Therefore, it is generally recommended
that streams be stratified by stream order (see below).
Parameter estimation under the simple random sampling design is straightforward:
The mean level of an environmental variable across the length of the river system can
be unbiasedly estimated by the sample mean p, with corresponding variance estimate
vaf(p) =
n
where s2 is the sample variance. The proportion of river miles showing a given
condition (i.e., degraded, supporting designated uses, etc.) can be estimated by p,
the proportion of sample sites showing that condition. The corresponding variance
estimate is
Finally, the total mass of an environmental contaminant in the rivers of the state can
be estimated by
an
n i*l
78
-------
where # is the concentration of the contaminant in a water sample collected at site
i, and |^4j| is the cross-sectional area of the river at that site. The variance of r may
then be estimated by
Stratified Random Sample. A stratified random sampling design may be im-
plemented to improve the precision of parameter estimates, to facilitate comparisons
among strata, and ensure adequate sampling effort in rare strata. Here, strata may
correspond to stream orders, or designated uses (i.e., swimming, drinking water, fish-
ing, etc.). Under a stratified random sampling design, the list of river segments is first
partitioned into K strata. Then a simple random sample of n/, sites is selected from
each stratum h\ h = 1,2, • • ¦, K, as described above. The number of sites sampled
from each stratum may be proportional to the total length Lof river segments in
each stratum:
Using this sample allocation scheme as a starting point, additional sampling effort
can be designated in strata deemed to be more important, while reduced sampling
effort can be designated in strata deemed to be less important.
Since a simple random sample design is obtained from each stratum, the stratum
ynaant and proportions, the total mass of a contaminant within each stratum, and
n n — 1
(18)
79
-------
their corresponding variances can be obtained using the same methods as described
above for simple random sampling. Then mean level of an environmental contaminant
across the lengths of all rivers in the population can be estimated by
K
E
h=\
where
1 K
Vat ~ T ^ ^ Lh •
*•> t_i
J
Vh -
nhlZl
is the sample mean of the observations j/ai , • • •, ytmh from stratum h, and L is the
total river miles in the population of rivers. The corresponding variance estimate is
L & n*
where
2 yl-nhfh
nh - 1
is the sample variance of the observations from stratum h.
Similarly, the proportion of rivers miles showing a given condition can be estimated
by
Pit =
L h*i
where % is the proportion of sample stations from stratum h showing that condition.
The corresponding variance estimate is
var(p«t) = 72 Li-- r~ •
L m n" ~ 1
80
-------
The total mass of an environmental contaminant across the lengths of all rivers in
the population can be estimated by summing the estimated mass of that mnt.imiina.nt.
over the strata. That is, take
K
** =
where f* is computed from observations in stratum h using expression (17). The
variance of rh can then be estimated from
K
var(r„.) = £va?(ffc).
h=l
Two Stage Sample. The implementation of the above sampling designs requires a
digitized map of all rivers and streams in the target population. The cost of obtaining
such a map can be prohibitive. These costs may be reduced by implementing a
two-stage sampling design. Under this design, the state is first partitioned into N
primary sample units, which may correspond to counties, NRCS watershed units,
1
or a contiguous gird of hexagonal or square quadrats. The first stage of the design
is comprised of the simple random selection of n primary sample units from the
population of N primary units. Then the rivers and streams are digitized within
each of the selected primary units; there is no need to digitize waterways within the
remaining primary units. The second stage of the design is comprised of taking a
simple random sample of sites along the lengths of the waterways within each of the
selected primary units. Typically, allocation among the primary units is proportional
81
-------
to the number of river miles in each of the selected primary units. Thus, if a total of
m sites are to be sampled, select
"tro)'
from primary unit i, where Li is the total river miles in primary unit t. Note that for
variance estimation, we require that m, > 2.
The mean level of an environmental variable along the lengths of the rivers and
streams in the population may be estimated by the ratio estimator
- ST.1 UK
* ET.t '
where y, is the sample mean of the variable among the observations from primary
unit i. The corresponding variance estimate is
var(/ifl) = I y) — + TiL,1*—'
\ L / n nL2
where
g„ lm - Ur?
n — 1
s* is the sample variance of observations from primary unit i, and
L = ^±Lt
" tsl
is the estimated total length of waterways in the target population.
82
-------
Similarly, the proportion of river miles showing a given condition can be estimated
by
s E."=i U • p.
a, u '
where pt is the proportion of sites from stratum i showing that condition. The corre-
sponding variance estimate is
where
rr-i £?(P,-fe)
s' = —~i—
2
The total mass of an environmental contaminant across the volume of the popu-
lation of waterways can be estimated by
AT n r m,
« rrii
where and \Aij\ are the contaminant concentration and the cross-sectional area of
the waterway at sample site j in primary unit i. The corresponding variance estimate
is
_ #2
var(fn) =
n(n
^ £ (£ £ I'M • w - £ £ £ £ I'M • *>)
-1) ,tr n *«im* >~i y
L' (Wi•«*-=-£i-4"!•*»)
83
-------
5.3 Sampling at Access Points
The savings incurred by sampling at access points allows larger sample sizes under
tight budgetary constraints, and hence potentially more precise parameter estimates
and greater statistical power for detecting spatial and temporal trends. The collection
of access points can be treated as the sample population from which a probability
sample can be obtained. However, to statistically justify inference to the water re-
source as a whole, we require evidence that the access points are representative of
that water resource, or alternatively, we require an estimate of the bias introduced
by sampling at the access points.
There are a number of reasons why the representativeness of access points may
be questioned: First, the density of bridges will tend to be higher in regions of high
human population density, and lower where human populations are sparse. Thus,
by taking a simple random sample of bridges, the level of environmental impairment
may be over estimated. This source of bias may be reduced by weighting the data
proportional to the length of the river segment comprised of all points closer to the
selected bridge than any other bridge (Figure 11a). Thus, the population mean level
of an environmental variable is estimated by
K = Z^WiVi
H tml
where y, is the data collected at the bridge », the weight Wi = U/L, is the length
84
-------
of the river segment comprised of all points closer to bridge i than any other bridge,
and L is the total river miles of the target population. The precision of flw can be
estimated by its sampling variance:
®GU = (^) £ (20)
where N is the total number of bridges in the population of bridges, n is the number
of bridges sampled, and
2 = Si wWi - ml
w _ i
n — 1
The estimated mean (19) assumes that the bridge is representative of the river seg-
ment containing that bridge, and the corresponding variance (20) makes the further
assumption that the variable is constant over the length of that river segment (Figure
lib). So the sampling variance is likely to be underestimated.
If the lengths of the river segments vaiy considerably, then the sampling variance
of p.w can be quite large. This sampling variance can be reduced by using an unequal
probability sample of bridges: Randomly locate points along the lengths of the rivers
and streams, and then select the bridge that lies closest to each of the selected points.
Bridges are sampled with replacement; that is, if a given bridge is selected more
fhan once, data collected by that bridge should be counted as many times as that
bridge is selected. Again, we shall assume that each bridge is representative of all
points along the length of the river closer to that bridge than any other bridge. Then
the population mean can be estimated by the sample mean p, with corresponding
85
-------
a
)( Bridge
SI
w ¦
S2
S3
S4 « S5
i
t
i
S6
S7
S8
99
i
i
i
i
ft 1
ft
ft 1
A
b
Z
Rher Miles
Figure 11: Sampling Bridges, (a) The locations of nine bridges along the length of
a river. The river is partitioned into segments SI to S9 as shown. A bridge will be
sampled if a random point falls in that bridge's segment, (b) Assumed relationship
between the variable of interest and location along the length of the river.
variance estimate var($f) = s2/n, where s2 is the sample variance. This estimate of
the population mean assumes that the bridge is representative of the river segment
containing that bridge, and the corresponding variance estimate does not take into
account variation along the length of that river segment.
The level of environmental impairment may vary with the suitability of locations
86
-------
for bridge construction, also resulting in biased estimates of the mean level of an envi-
ronmental variable. This source of bias may be reduced by using a stratified random
sampling design: The river segments associated with the bridges are partitioned into
m strata defined by their suitability for bridge construction. Thus each stratum will
consist of river segments that are roughly equally suitable for bridge construction.
River segments within each stratum are then laid out aid to end, and n* points are
randomly selected along the total length of stratum t; i = 1, • • • m. Finally, select
the bridge closest to each of the selected points. Then the population mean can be
estimated by
1 m
Art = T ^iVi i
»=1
where ft is the sample mean of selected bridges in stratum t, L, is the total length
of river segments in stratum i, and L is the total river miles of the system. The
corresponding variance estimate is
var(^t) = 72 Z. — •
,=1
where s? is the sample variance of selected bridges in stratum i. Again, the estimator
assumes that the bridge site is representative of the river segment containing
that bridge, and the corresponding variance estimator does not account for variation
within river segments.
Some portion of the lengths of rivers may be completely unsuitable for bridge
construction. This portion cannot be sampled at access points, and so, should be
87
-------
treated as a separate stratum to be sampled using one of the methods described in
Section 5.2.
The bridges themselves may have adverse effects their local environment, resulting
in overestimates of environmental impairment. This source of bias might be reduced
by sampling some random distance upstream from each bridge, instead of immediately
below or adjacent to them.
Regardless of what design is used to select the access points to be sampled, evi-
dence is required to demonstrate that the resulting sample yields unbiased estimates
of environmental parameters. This requires data collected from a probability-based
design, in which sites are selected from the water resource as a whole (e.g., using
methods such as described in Section ). Let p,b denote the estimated population
mean obtained from sampling at bridges, let /20 denote the estimated population
mean obtained from sampling along the water resource as a whole, and let var(£fc)
and var(£0) denote the corresponding variances. Then the null hypothesis that sam-
pling at bridges yields an unbiased estimate of the population mean can be tested
using the test statistic
Under the null hypothesis, t is approximately t-distributed with nt + n„ — 2 degrees
of freedom, where na and nt are the number of observations from the two respective
samples. If estimates from access points are not significantly different from estimates
88
-------
obtained from the probability-based design over the resource as a whole, then sampling
at access points suffices. If not, then the bias can be estimated by
Assuming that the two samples are independent, then the variance of the estimated
bias can be estimated by
var(3) = va?(£a) + var(/i6)
This bias correction can then be applied to future data collected exclusively from
access points; that is, if is an uncorrected estimate of the population mean obtained
from access point data, then a bias corrected estimate of the population mean is given
by
with corresponding variance estimate
_ ^
vaf(fic) = var(£j + var(/?).
Note that this presumes that the same sampling design was employed, and assumes
tluvt. the bias does not change over time. It is recommended that the latter assumption
be checked periodically using data from a probability based design including non-
access points. A better approach would be to include both access and non-access
points in the design during each sampling interval. The allocation of sampling effort
89
-------
between access and non-access points can be determined so as to obtain the most
precise estimates at minimum cost. Improved performance may also be achieved by
applying different bias corrections to different strata.
6 Retaining Information from Historical Data
Despite the advantages outlined above, managers of state water quality monitoring
programs are reluctant to implement probability-based sampling designs. Much of
this reluctance stems from the fear that information from the historical data base will
be lost. Therefore, probability-based sampling designs are not likely to be widely im-
plemented unless statistical approaches to combining data from judgment and prob-
ability sampling designs are available. Unfortunately, methods for combining such
data have received very little attention in the statistical literature. Overton, Young,
and Overton (1993) use sampling frame attributes to assign judgment sites to clusters
of similar probability sites. Judgment sites assigned to a given cluster are assumed to
be representative of that cluster, and are treated as though they were obtained from a
probability-based sampling design. However, the representativeness of the judgment
sites with respect to their assigned clusters is difficult to diagnose, and if false, the
combined data may yield biased estimates (Cox and Piegorsch 1996).
The following proposes an alternative approach to combining data from historical
judgment sample sites with data from new probability-based sample sites. This ap-
90
-------
proach requires an interval of overlap in which both historical judgment sites and new
probability-based sites are sampled. Then the spatio-temporal correlation between
the two sampling designs is exploited to predict what data would have been obtained
had a probability-based sampling design been implemented from the very beginning
of the monitoring program.
6.1 Space-Time Model
The following assumes that the data are a partial realization of a spatio-temporal
random process. In particular, assume that the data Z(s, t) at site s =(x, y) and time
t are realized from the model
Z( s, t)=0o + /?i*i (s.*) + •" • + PpXp(s> t) + *(s, t), (21)
where 0o,0u---,0p are model parameters, and £(s, t) is a zero-mean error term. The
explanatory variables xi (s,t),•••, xp(s, t) may be functions of the spatial coordinates,
time, distances to known geographic features (e.g., the mouth of the river system),
or environmental variables such as water temperature, current, or turbidity.
Pairs of observations that are close together in space and time are likely to be
more gimilftr to one another than pairs of observations that are far apart. This
spatio-temporal dependence can be modeled through the spatio-temporal correlation
function
p(l|«i -82|U*i -fc|) *= conlZisutrifZfo^)},
91
-------
which depends only on the distance ||s! — S2H between the pair of sample sites Si
and s2, and the difference in sample times t\ and t2. The correlation function takes
values between -1 and 1; positive values indicating positive spatio-temporal depen-
dence, while negative values indicate negative spatio-temporal dependence. Typi-
cally, the correlation function will be a decreasing function of both |(si — 821| and
|ti — taj, asymptotically approaching zero as the spatial and temporal distances be-
tween the observations increase. The rate at which the correlation function approaches
zero determines the range of spatio-temporal correlation; correlation functions that
rapidly approach zero characterize processes where interactions occur only between
sites that are very close together, while correlations that slow approach zero charac-
terize processes where distant sites interact. Observations have a perfect correlation
of 1 with themselves so that p(0,0) = 1. However, there is often a discontinuity at
zero when the correlation function is plotted against distance in space or time. This
discontinuity is the so-called nugget effect, and is typically the result of measurement
error or small-scale sampling variation.
Alternative measures of spatio-temporal dependence in the data include the co-
variance function
C{h,r) = cPpihyT)
and the vaxiogram
1l{Kr) — var{Z(s,t) - Z(s + h,t + r)}
92
-------
= 0. ~ ?(8*' * + r)l2' (22)
93
-------
where the sum is over all pairs of observations collected at sites approximately distance
h apart and at sample times r apart, and N^r is the number of such pairs of sites.
The values e(si,t) are residuals from a multiple regression of the data against the
explanatory variables Xi(s,t),---,xp(s,t). The weighted least squares estimator of
the parameter 0 is then obtained by finding 0 that minimizes
; * var{7(/i3,r*)}
where the sum is over all spatial and temporal lags at which 2i/(h,T) is computed,
and
vax{7{hj,rk)} S 2{2'y{h,r,0)}i/Nhr.
6.2 Spatio-Temporal Prediction
Suppose that fixed sites Sj, • • • ,s„ are selected according to an arbitrary judgment
sampling design, and that the variable of interest is observed at those sites at time t =
1, • • • ,T. Thus, the judgment sample data are {Z(siyt) : i = 1,• • • ,n; t = 1,• • • ,T}.
At time t = M < T, a probability-based sampling design is implemented, selecting
sites Ui, • • •,um. Data are then collected at times t = Af, M + 1, • • •, so that the
probability sample data are (Z(ui,t) : i = 1, • • • ,m; t = M, • • • ,T}. More generally,
new probability sample sites may be selected in each sample interval, or a serially
alternating design may be implemented. For ease of notation, however, we shall use
a permanent station sample design here (see Section 3.3).
94
-------
Our objective is to back predict what data would have been obtained had a
probability-based sampling design been implemented from the very beginning of the
monitoring program; that is, predict the unobserved values of {Z(u*, t) : i = 1, • • •, m;
t = 1, • • •, M - 1}. Kriging is perhaps the most popular method of spatial predic-
tion (Cressie 1989), and can be easily extended to spatio-temporal prediction. This
popularity owes much to its stability with respect to violations of model assump-
tions (e.g., Cressie and Zimmerman 1992). In particular, kriging is not sensitive to
whether or not a spatial trend is included in the model (Journel and Rossi 1989), or
to misspecification of the variogram model (Stein and Handcock 1989).
If the complete data base were to be used, spatio-temporal prediction would re-
quire the solution of nT+m(T-M+l)+p+l linear equations for the same number
of unknowns. This may not be practical for a reasonably large data set. Therefore,
the following spatio-temporal predictor shall only use data from the judgment sample
at time t, and data from the probability design at time M to predict the unobserved
values of the data from the probability sample at time t. Then the universal kriging
predictor is
Z(uk,t) = X>
-------
system ofn + m + p+1 equations
(II* - SjII, 0) + Y, *2.7 (||Ui - 8,-11, M -1) + Co + t) = 7 Ota - u*||, 0);
i=l 1=1 »=1
j =
£ Ax<7 (||St - Uj||, 0) + £ A2<7 (||Ui - UjH , A/ -1) + Co + <) = 7 (K - u*ll. Af
»=i »=i «*i
j =
Ali + 52^21 = 1,
»=1 t=l
n m
^ AiiXj(sj,t) + y A2iXj(ui, Af) = ij(ufc,t), j = 1, • • • , p,
»=i »=i
for the n + m + p+ 1 unknowns Aii,--,Ain,A2i,---,A2m,£o»£i>*-'.£p- This system
of equations is called the kriging equations. The precision of the resulting kriging
predictor is described by the kriging variance
-------
The function /(<) models the background temporal trend (Figure 12). The spatial
random field ^(s) has unit variance and spatial correlation function
p,(h) = exp{—3/i/k,}
with a long range of spatial dependence of k, = 200 km; for the current application,
it can be considered to model the spatial trend in the data (Figure 13). Likewise, the
temporal process a(t) has unit variance and temporal correlation function
pt(r) = exp{-3 r/Kt}
with a long random of temporal correlation of «t = 3000 years. The spatio-temporal
random process 0(s,t) allows the temporal trend to depend on location; it has unit
variance and spatio-temporal correlation function
r) = exp{-3h/n, - 3r/«J
with relatively short ranges of spatial and temporal correlation set at ks = 20 km
and Kt = 10 years. All three of the above processes were simulated using the spectral
method (Shinozuka 1971; Mejia and Rodriguez-Iturbe 1974). The error e(s, t) is
Gaussian white noise with unit variance, and models the effects of measurement error.
It was simulated using the polar method (Ripley 1987, p. 62).
The relative influence of the four component processes on the resulting «fota can
be fixed by varying the levels of the coefficients a„at,a,t, and at. If we set ast = 0,
97
-------
f (t)
2.00
75
50
25
00
0.75
0.50
25
00
50
0
30
40
20
10
t
Figure 12: Temporal TVend.
then the spatial and temporal effects are additive, and the sampling bias attributed
to the judgment sampling design can be simply removed by subtraction. It seems
more likely that temporal trends occurring in the data may depend on location, and
so, sampling bias cannot be simply removed by subtraction.
Two samples of data are generated. A total of 100 probability sites Ui, • • ¦ .u^,
are obtained a simple random sample over a 100 x 100 km region. For the judgment
sample, an additional 100 sites Si, • • •, sioo are independently selected from the density
proportional to
n(«\ - expiPo + PM*)}
l+exp{/3o + 0i m(s)}'
which depends on the realization of the first component of our simulation model (24).
98
-------
o 0
Figure 13: Spatial TVend.
99
-------
Note that if /3j = 0, then we obtain another simple random sample. For > 0, the
judgment sample is biased in favor of high data values, while for /31 < 0, the judgment
sample is biased in favor of low data values. Data for both designs is generated for
years t = 1,• • • ,50, but it is assumed that the probability-based points are only
observed for years t = 41, • • •, 50.
6.4 Effect of Sampling Bias
The geostatistical methods described in Sections 6.1 and 6.2 are carried out condi-
tional on what sites are actually included in the sample, and thus, ignore the effects
of sampling variation on variogram estimates and spatio-temporal predictions. In
particular, the potential effects of sampling bias in the judgment sampling design are
not considered. These effects shall be thoroughly explored under the following values
of the model parameters: ao = 5, as = 10, at = 3, att = 3, ae = 0.5, /?0 = -1, and
= 4. Taking > 0 yields a judgment sampling design biased in favor of large val-
ues. In Figure 14, the sample means for both designs are plotted against time. Data
from the judgment sampling design show an increasing trend over time (triangles),
with a large jump in mean level occurring in year 31. The probability sites were only
sampled after year 41, but, as expected given that > 0, have lower means than
the judgment sites (circles). Our objective is to predict the unobserved values for the
probability-based design from years 1 to 40.
100
-------
mean
18
14
16
101 *
12
6
0
10
20
30
40
50
time
Figure 14: Annual means for judgment (triangles) and probability (circles) sample
sites.
The data from the two designs were fitted separately to the planar trend model
Z(x, y, t) = q0 + a\x + a2y + e(x, y, t),
where Z(x, y, t) denotes the data collected at coordinates (x, y) at time t, and e(x y t)
is the model error. Ordinary least squares estimates yield the fitted models
Z(x, y, t) = 15.7 + 0.0701* - 0.0155y
for the judgment design, and
Z(x, y, t) = 18.2 + 0.0867 - 0.0269y
for the probability-based design. Notice that the estimated partial slopes are of lower
magnitude under the judgment design than under the probability-based design.
101
-------
The method of moments estimator 27t(h) for the spatial variogram 27t(h) =
27(h, 0) (expression (22)) was computed separately for each of the two designs. The
results suggest that the biased judgment sampling design also yields a biased estimate
of the variogram. For both designs, 27, (/i) increases rapidly to an asymptote with
increasing h (Figure 15). However, the asymptote under the judgment sampling
design appears to be larger than that under the probability-based design. Weighted
least squares estimation was used to fit the exponential variogram model
27*(h) = 2
-------
100
Distance
Figure 15: Fitted spatial variogram models for probability-based (solid line fit to the
circles), and judgment (dashed line fit to the triangles) sample sites.
to 2%(r), where a1 is the variance of the data, and Kt is the range of temporal
correlation. The estimated range of temporal correlation was kt = 12.2 years
The universal kriging predictor (23) was computed for the unobserved data at the
probability sample sites between years 1 and 40. Then, within each of these years,
the mean of the predicted values was computed using
1 m ~
k = (25)
where Z(u*, t) is given by expression (23). Figure 17 compares these mean predicted
values (x's) with the unobserved mean values (open circles) of the probability sites
in years 1 to 40. Note that the means of the predicted values form a smoother curve
103
-------
25'
20
E
Z 15
O
o 10
>
5
0
0 2 4 6 B 10 12
Time Log
Figure 16: Fitted temporal variogram model.
than either the observed means of the judgment sample sites, or the unobserved
means of the probability sample sites. This is not unexpected given that kriging is
a smoothing algorithm. The means of the predicted values do tend to fall below
the observed means from the judgment sample sites (triangles) indicating that the
proposed procedure does reduce the bias attributed to the judgment sampling desigfr.
Moreover, the means of the predicted values successfully pick up the discontinuity in
the data at year 31. However, the means of the predicted values also tend to fall above
the unobserved means of the probability sample sites (open circles), which they were
intended to predict. Thus, the proposed procedure still yields biased predictions.
104
-------
mean
18
16
14
12
10'
8-
*4 *• A
1 ••••
,»l „ O •"
¦ °o° o •
e • •
A* *
* ~ »».
* »** ee0
A o A "C
° f*oO0e0 ° °co °«
g"pO pOp o "
10
20 30
time
40
50
Figure 17: Comparison of predicted (x's) and unobserved (open circles) sample means
for probability sites in years 1 to 40. In addition, observed annual means for judgment
(triangles) and probability (open circles) sample sites are given.
105
-------
6.5 Bias Reduction
The observed positive bias is not unexpected given that the judgment sample sites
are biased in favor of high data values, and given the role that the judgment sample
sites play in predicting unobserved past values at the probability sample sites. The
magnitude of this bias can be estimated using data from those years in which obser-
vations from both designs are available. This can be accomplished by predicting the
observed data from the probability based design using
n to
Zj{uk,t) = £AuZ(Si,t) + + j)\ t = Af, ¦ • • ,T- 1,
i= 1 t=l
where the coefficients An,• • •, Ai„, A21, • • •, A2m are selected to minimize the mean
squared prediction error subject to the constraint that the resulting predictor be
unbiased for the true value of the data. This predictor uses data from the judgment
sampling at time t, and data from the probability-based sampling design at time t+j
to predict the data for the probability-based design at time t. Then the prediction
bias of Zj(uk,t) is given by
bj{Uk,t) = Zj(uk,t) - Zj(uk,t).
The subscript j is included in Zj(uk,t) and 6j(u*,t) to take into account that the
prediction bias may depend on the number of time lags j in the past we are attempting
to back predict the probability-based data. The mean prediction bias in year t under
106
-------
a predictor using probability based data j years in the future is then given by
h = ^ II *)•
Table 7 gives the mean prediction bias in year t under predictors using probability
based data j time lags in the future. Notice that the prediction bias depends strongly
on what year's data we are attempting to predict. However, this is of little use for
estimating the mean bias in years 1 to 40. Within each year, the bias appears to
increase somewhat with increasing time lag. This suggests that the magnitude of bias
in the proposed predictor will increase as we attempt to back predict the probability
data further into the past.
To quantify the relationship between mean bias and time lag, the general linear
model
bjt = /* + ocj + (3t + Ejt
was fit to the observations in table 7, where n is the over mean aj is the effect of timp
lag j, said 0t is the effect of year t. Then the mean bias for time lag j was adjusted
to take into account variation among years using the general linear models procedure
of SAS (SAS Institute 1985). The adjusted mean bias is then plotted against time
lag as shown in Figure 18. Notice that the adjusted mean bias appears to increase
linearly with increasing time lag, further indicating the bias in the proposed predictor
increases as we attempt to back predict the probability data further into the past.
107
-------
lagj
Year 123456789
41 1.428 1.488 1.515 1.515 1.517 1.542 1.567 1.566 1.579
42 1.118 1.149 1.144 1.145 1.176 1.203 1.201 1.215
43 0.509 0.487 0.481 0.514 0.542 0.536 0.551
44 0.518 0.517 0.561 0.596 0.589 0.606
45 1.015 1.059 1.093 1.077 1.093
46 0.927 0.972 0.950 0.969
47 0.717 0.675 0.693
48 0.691 0.725
49 1.097
Table 7: Mean bias as a function of year in which probability data are predicted and
time lag.
108
-------
BIAS
1 .02
0.90
0.92
0.94
0.96
0 . 96
1 . DO
0 .88
23456789
LAG
Figure 18: Adjusted mean bias plotted against t.imo lag.
Fitting a linear model to the data in Figure 18, we obtain the following estimate for
the bias at time lag j:
bj = 0.87989 + 0.014164j.
Using the expression above, a bias corrected predictor for the mean of the proba-
bility sample sites in year t is given by
Figure 19 compares bias corrected predicted mean values (x's) with the unobserved
mean values (open circles) of the probability sites in years 1 to 40. Comparing the
results in Figure 19 with those previously obtained in Figure 17, notice that the bias
A— t^t ~ bM—t
= fa- 0.87989 - 0.014164 x (41 - t).
109
-------
correction was successful in reducing the bias in predicted values. However, there
is a suggestion of a small overcorrection, with biased corrected predictions falling
slightly below the unobserved means that they are attempting to predict. That the
predicted values fall well below the unobserved means in the first nine years can be
attributed to the observation that the judgment sites show very little sampling bias in
those years. This points to one of the shortcomings of the proposed approach to back
prediction; it assumes that the sampling bias shows no temporal trends. Nevertheless,
it is interesting to note that the predicted values track the trend function in Figure
12 very well.
6.6 Conclusions and Recommendations
The above approach exploits the spatio-temporal correlation with historical data from
the judgment sampling design to back predict the unobserved means at probability
sample sites. To compensate for the sampling bias of the judgment sample, a bias
correction is required. This approach requires the careful modeling of any spatial
trends that may occur over the study region, the spatio-temporal correlation structure
in the data, and the bias resulting from the judgment sample design. To ensure
that model assumptions are satisfied, appropriate diagnostic procedures should be
implemented.
Bias correction requires a period of overlap in which observations are collected
110
-------
mean
18
16
14
12
10
8
*
*4 »* •
* * '»»»,»
°o« ?»' • *
A*'
A*
* A*
t
«®Ofl
A o 4
o°o
10
20 30
t i me
40
50
Figure 19: Comparison of bias corrected predicted (x's) and unobserved (open circles)
sample means for probability sites in years 1 to 40. In addition, observed annual means
for judgment (triangles) and probability (closed circles) sample sites are given.
Ill
-------
from both sampling designs. Further research is required to determine how long that
period of overlap should be. The bias correction also assumes that the sampling
bias of the judgment design shows no temporal trends. In practice, it is not possible
to determine that validity of this assumption. Improved predictions could poten-
tially be obtained if sites from both the probability-based and judgment sampling
designs are partitioned into strata selected to minimize sampling bias of judgment
sites within strata. Such strata might be selected using the methods of Overton et
al. (1993). Stratum identification can then be used as explanatory variables in the
spatio-temporal model (21), not only improving the precision of predictions, but also
reducing the effects of sampling bias.
References
Cox, L.H., and Piegorsch, W.W. 1996. Combining environmental information I:
environmental monitoring, measurement, and assessment. Envrionmetrics 7,
299-308.
Cressie, N. 1989. The many faces of spatial prediction. In Geostatistics, Vol. 1, M.
Armstrong, ed., Kluwer, Dordrecht, 163-176.
Cressie, N. 1991. Statistics for Spatial Data. Wiley, New York.
Cressie, N., and Zimmerman, D.L. 1992. On the stability of the geostatistical
method. Mathematical Geology 24, 45-59.
112
-------
Horn, C.R., and Grayman, W.M. 1993. Water-quality monitoring with EPA reach
file system. Journal of Water Resources Planning and Management 119, 262-
274.
Journel, A.G., and Rossi, M.E. 1989. When do we need a trend model in kriging?
Mathematical Geology 21, 715-739.
Larsen, D.P., and Christie, S.J. 1993. EMAP-Surface Waters 1991 Pilot Report.
EPA/620/R-93/003. U.S. Environmental Protection Agency, Office of Research
and Development, Washington, D.C.
Mejia, J.M., and Rodriguez-Iturbe, I. 1974. On the synthesis of random field sam-
pling from the spectrum: An application to the generation of hydrologic spatial
processes. Water Resources Research 10, 705-711.
Messer, J. J., Linthurst, R.A., and Overton, W.S. 1991. An EPA program for moni-
toring ecological status and trends. Environmental Monitoring and Assessment
17, 67-78.
Myers, R.H. 1976. Response Surface Methodology.
Olsen, A.R., Sedransk, J., Edwards, D., Gotway, C.A., Liggett, W., Rathbun, S.,
Reckhow, K.H., and Young, L.J. 1998. Statistical issues for monitoring ecologi-
cal and natural resources in the United States. Environmental Monitoring and
113
-------
Assessment (in press).
Ripley, B.D. 1987. Stochastic Simulation. Wiley, New York.
Overton, J., Young, T., and Overton, W. 1993. Using found data to augment a
probability sample: procedure and a case study. Environmental Monitoring
and Assessment 26, 65-83.
Rao, J.N.K., and Graham, J.E. 1964. Rotation designs for sampling on repeated
occasions. Journal of the American Statistical Association 59, 492-509.
Searle, S.R. 1971. Linear Models. Wiley, New York.
Shinozuka, M. 1971. Simulation of multivariate and multidimensional random processes.
Journal of the Acoustical Society of America 49, 357-367.
Skalski, J.R. 1990. A design for long-term status and trends monitoring. Journal of
Environmental Management 30, 139-144.
Snedecor, G.W., and Cochron, W.G. 1980. Statistical Methods, Seventh Edition.
Iowa State University Press, Ames, Iowa.
Stein, M.L., and Handcock, M.S. Some asymptotic properties of kriging when the
covariance function is misspecified. Mathematical Geology 21, 171-190.
Stevens, D.L. (1997). Variable density grid-based sampling designs for continuous
spatial populations. Environmetrics 8, 167-195.
114
-------
Thompson, S.K. 1990. Adaptive cluster sampling. Journal of the American Statis-
tical Association 85, 1050-1059.
Thompson, S.K. 1992. Sampling. Wiley, New York.
Urquhart, N.S., Overton, W.S., and Birkes, D.S. 1993. Comparing sampling designs
for monitoring ecological status and trends: impact of temporal patterns. In
Statistics for the Environment, V. Barnett, and K.F. Turkman, eds., 71-85.
115
-------
-------
Appendix H
The Ecological Condition of Small Streams in the
Savannah Basin: A REMAP Progress Report
By
Raschke et al.
-------
(GICAL CONDITION OF
lMS IN THE SAVANNAH
IN: A REMAP
EPORT
k
2K5Cfi:
.
1
L
R. L.
H. S. HOWAI
J. R. MAUDSLEY
R. J. LEWIS
-------
THE ECOLOGICAL CONDITION OF SMALL STREAMS
IN THE SA VANNAH RIVER BASIN: A REMAP
PROGRESS REPORT
BY
'R. L. RASCHKE
'H. S. HOWARD
2 J. R. MAUDSLEY
JR. J. LEWIS
1. U. S. Environmental Protection Agency Region 4, Environmental Services
Division, Ecological Support Branch, 625 Bailey St., Athens, GA 30605
2. MANTECH Environmental Technology, Inc., Research Road, Athens, GA
30605.
-------
SUMMARY
In response to the needs of the states of Georgia and South Carolina
and their policy-relevant questions, the Ecological Support Branch of EPA Region
4 provided a monitoring strategy to help them effectively and efficiently sample
Savannah River Basin waters. During the first two summer seasons (1994 &
1995) of a four-year cycle, 64 sites on wadeable streams were monitored in a
systematic random manner to evaluate the status of ecological condition in the
basin- By sampling fish, insects and algae, and evaluating the habitat,
investigators found that water quality of most stream miles were in good condition
with respect to nutrient content. However, 38% of the stream miles were affected
by poor habitat, and 33% to 52% of the insect and fish communities respectively
were in poor ecological condition.
Although the Branch has just begun to explore the potential use of the
Geographical Information System (GIS), two areas encompassing several counties
in Georgia and South Carolina seemed to have clusters of sites of poor ecological
condition. Besides poor habitat, two other potential causes of poor conditions,
wastewater treatment plants and animal feeding operations, may be negatively
affecting the condition of insect and fish communities. Further refinement of the
data analysis and eventual rechecks of sites in this area will be necessary before
any permanent conclusions can be drawn. The cluster of poor sites in South
Carolina at this time, are attributed only to habitat effects like, sediment erosion,
deposition of sediments, and stream bank failure.
-------
TABLE OF CONTENTS
PAGE
INTRODUCTION 1
SAVANNAH RIVER BASIN 3
MONITORING DESIGN 5
INDICATORS 6
MONITORING ASSESSMENT 11
BIBLIOGRAPHY 18
ii
-------
introduction
The southeast's population growth has brought pressure on natural resources.
Responding to the challenge of adapting to more people and balancing multiple uses of natural
resources, the Environmental Protection Agency (EPA) instituted the Watershed Protection
Approach (WPA). WPA is a program for identifying and preventing environmental problems,
setting priorities, and developing solutions through an open, inclusive process with the people
(stakeholders) who live in a geographical setting. Consideration of economic prosperity and
environmental well-being is the corner stone of WPA.
The Savannah River Basin was selected for environmental protection because of high
population growth, known environmental problems, its susceptibility for further degradation, and
the likelihood of successfully enhancing quality of life in the basin. Through the WPA program,
EPA - Region 4 brought together stakeholders of varying interests who developed a
comprehensive strategy known as the Savannah River Basin Watershed Project. Part of that
Strategy included a monitoring component, the Regional Environmental Monitoring and
Assessment Program (REMAP).
REMAP represents a fundamental change in environmental-appraisal. It produces
representative measurements of overall status and trends of environmental condition. Its goal is
measure cumulative effects with a known degree of certainty, provide decision makers with
sound ecological data, and measure the effectiveness of environmental protection efforts.
The Environmental Services Division (ESD) of EPA Region 4 was asked by the
Savannah River Watershed Project Policy Committee to implement the REMAP strategy as a
1
-------
demonstration project for the states of South Carolina and Georgia. These states were interested
in reducing sampling and analyses, having the ability to reduce or increase sampling density,
responding quickly to emerging environmental problems, and maintaining representative
coverage of environmental resources through systematic yet random means of sampling.
Before the monitoring study, a set of questions was posed by the states of Georgia and
South Carolina to provide direction for the monitoring design. The following policy-relevant
questions were identified to guide the development of a plan of study and subsequent monitoring
efforts.
»• What is the status of condition of the water resources of the Savannah River Basin?
~ What proportion of the Savannah River Basin surface waters are attaining designated
uses?
~ What are the changes of ecological condition over time?
~ What factors might be associated with changes?
~ Is there a tendency for distribution of condition in a specific direction (spatial gradient)
over the basin landscape? What are the possible reasons for these gradients?
~ What resources are at risk in the Savannah River Basin?
In response to the needs of the states and the policy-relevant questions posed, the Ecological
Support Branch developed the following study objectives with the concurrence of the Policy
Committee of the Savannah River Watershed Project.
~ Estimate the status and change of the condition of water resources in the Savannah River
Basin;
»• Identify water quality spatial gradients that exist within the Savannah River Basin and
2
-------
associate current and changing condition with factors that may be contributing to this
condition and spatial gradients;
~ Demonstrate the utility of the REMAP approach for watershed and river basin monitoring
and its applicability for state monitoring programs;
~ Incorporate the REMAP approach in the formulation and accomplishment of River Basin
Management Plans; and
~ Provide baseline information required to conduct comparative risk assessments in the
Savannah River Basin.
Beneficiaries of the study are basin resource managers in state and federal agencies and
local governments. This information will help identify stressed areas and suggest strategies for
addressing high priority problems. In addition, the study will provide baseline information for
comparing trends in condition throughout the Savannah River Basin and assessing the
effectiveness of cumulative management efforts on protecting and managing these ecological
resources.
SA VANNAH RIVER BASIN
The Seneca and Tugaloo Rivers begin on the slopes of the Blue Ridge Mountains in
North Carolina. These two rivers join forming the Savannah River near Hartwell, Georgia and
Anderson, South Carolina. The river flows in a southerly direction forming the boundary
between South Carolina and Georgia. Eventually, the Savannah River empties into the Atlantic
Ocean at the port city of Savannah (Figure 1).
Within the basin's 10,579 square miles, there arel7,354 stream miles. One thousand five
3
-------
hundred and three of those
stream miles or 5.4% are
wadeable (first through third
order) stream miles. The basin
consists of three different land
forms or physiographic
provinces: the Blue Ridge,
Piedmont, and Coastal Plain.
The Blue Ridge is characterized
by mountains covered naturally
with Appalachian oak. Forests
and ungrazed woodlands are the
predominant land uses with
some cropland and pastures.
The Piedmont is characterized
SAVANNAH RIVER BASIN, STUDY AREA MAP
Figure 1. Savannah River Basin with physiographic
provinces.
by gently sloping hills and smooth to irregular plains. This province is underlain naturally with
nutrient poor soils supporting oak/hickory/pine and southern mixed forests. Land use is a
mixture of croplands and pasture, woodlands with some urban areas. Flat plains dominated
naturally by oak/hickory/pine forests, pocasin (pine, holly) forests, southern floodplain forests
(oak/tupelo, bald cypress), and southern mixed forests (beech, sweetgum, magnolia, pine and
oak) are characteristic of the Coastal Plain.
Within the three physiographic provinces there exists distinct ecosystems based on the
4
-------
interrelationships between organisms and their environment. These distinct ecosystems are
defined as ecoregions. While physiographic provinces may prove suitable for regional or
national assessments, definition of ecoregions among broad physiographic areas is necessary to
accurately assess ecological condition or health. Ecoregions are distinct areas grouped by
climate, soils, land forms, and vegetative cover. The Blue Ridge physiographic province stands
alone as a separate ecoregion as does the Piedmont physiographic province. However, the
Coastal Plains physiographic province is composed of three distinct ecoregions: the Fall Line
Hills (or Sand Hills), the Southeastern Plains and Hills, and the Coastal Plains.
monitoring design
Objectives of the monitoring design provide information about the ecological situation and
eventually trends in condition of the natural resources. One resource of interest is all perennial
wadeable streams. The design strategy selects wadeable stream sampling points that provide valid
estimates of general basin-wide stream condition.
One can study conditions of streams in two ways. The first is by consensus, which entails
examining every point on the streams. This method is impracticable. A more practicable approach
is to examine some points systematically to ensure adequate coverage of the basin, and randomly
to prevent bias in selection of stream points. For example, we would not obtain a good estimate of
the percent of all students in a region with hepatitis if we polled only students in small towns of less
thpn two thousand people. This preferential or biased sample would most likely include a much
lower proportion of students with hepatitis than the general population of students. Similarly in a
stream study preferential sampling occurs if the sample includes only sites, for example, downstream
5
-------
of sewage outfalls where sewage outfalls affect only a small percentage of total stream length. This
kind of sampling program may provide useful information about conditions downstream of sewage
outfalls, but it will not produce estimates that accurately represent conditions of the whole basin.
Preferential selection can be avoided by collecting random samples.
Randomization can be thought of as a kind of lottery drawing to determine which points are
included in the sample. Randomization is important. When used, it is possible to estimate condition
of streams with a known degree of confidence. In REMAP, hexagons are used to add the systematic
element to the design. The hexagonal grid is positioned randomly over the basin map, and sampling
points from within each hexagon are selected randomly. The grid ensures spatial separation of
selected sampling points.
This design's sampling requirements reduce sampling locations to a logistically and
economically feasible number. It allows fewer sites to be sampled annually, but provides for
sampling of all randomly selected sites over a rotating year period. Currently, this rotation period
for the Savannah REMAP project is four years. However, the period can range from two to five
years. This report is an assessment of wadeable stream condition based on a sampling of 64 sites
after the second year of a four-year cycle. It is expected that an additional 72 sites will be sampled
for a total of 136 sites by the end of the four-year cycle.
INDICATORS
REMAP monitors ecological indicators to assess condition and trends. Indicators are defined
primarily as any characteristic that estimates the condition of natural resources. The challenge is to
decide which indicators to monitor. One approach for selecting indicators starts with those attributes
6
-------
valued by society and then decide what indicators might be associated with these values. Upon
consideration of the type of streams (wadeable) to be investigated and after extensive discussions,
an initial set of societal values and concerns were identified. They are: Biological Integrity and
Trophic Condition (Table 1).
Table 1. Values, Indicators, and Measures Used to Evaluate Wadeable
Stream Conditions in the Savannah River Basin.
VALUE
INDICATOR
MEASURES
Biological Integrity
Stream Insect RBP1 II
Fish RBP2 V
Habitat Score
EPT4 Index
Fish IBI5
Trophic Condition
AGPT3
Average Maximum Yield in
Dry Weight of Algal Cells
per Liter
1. Stream insect Rapid Bioassessment Protocol based on identification of insects to family level.
2. Fish Rapid Bioassessment Protocol based on identification of different kinds of fish.
3. Algal Growth Potential Test; a measure of nutrient enrichment.
4. An index based on the identification of pollution sensitive insects known as stoneflies, mayflies,
and caddisflies.
5. An Index of Biological Integrity based on twelve different characteristics offish.
Biological integrity incorporates the idea that all is well in the community. That is, the
different groups are stable and working well with little if any external management of the
community, whether it is a township, coral reef, or stream.
Trophic condition is a measure of water quality condition based on different levels of
7
-------
available nutrients. When nutrients are in excess, overabundance of algae and larger green plants
results in nuisance conditions. Millions of dollars are spent annually to control the growth of algae
and other plants. Overabundant growth of plants can affect biological integrity but also human uses
like fishing, boating, swimming etc.
The challenge is to identify ecological indicators that can be related directly to those societal
values held by the public. There are two general types of ecological indicators; condition and
stressor. A condition indicator is any characteristic of the environment that estimates the condition
of natural resources and is conceptually tied to a societal value. Stressor indicators are suspected to
elicit a change in the condition of the natural resource. The indicators selected to address biological
integrity are stream insects and fish assemblages.
Insects represent the first consumer level in streams. They are important as processors of
organic matter, like leaves and sewage, that find their way into a stream. By fragmenting or
breaking down this organic matter, stream insects prepare it for decomposition by bacteria that attach
or colonize the organic matter. In turn, bacteria may serve as a food source for other stream insects
that seek out and graze on the organic matter. Because of their limited mobility and relatively long
life span, stream insects provide a "window" of cumulative impacts on ecological or resource
condition. This community is sensitive to changes; they have for many years been used as a reliable
barometer of water quality conditions. Some groups of insects are very sensitive to stresses, like
manmade pollution while others are tolerant. By focusing on the presence or absence of different
groups of insects, an aquatic biologist is provided insight about the ecological health of a stream.
Sometimes pollution effects may stem from discharges of chemicals, pesticides, or nutrients that are
of a manmade origin. Often, sediments from erosion and attributable to land clearing or silvaculture
8
-------
practices may adversely affect the stream habitat. The materials that constitute a stream bottom are
very important to both fish and stream insects. For example, very fine sediments, like silt, clay, or
very fine sand, are detrimental to the reproduction of fish and eliminate preferable habitat for stream
insects. Silt, especially, can interfere with a fish's or stream insect's ability to breathe. Assessment
of the insect community was accomplished by using a standard field survey technique known as
Rapid Bioassessment Protocol II(RBPII). With the RBP II protocol, most sites can be surveyed
with relative limited time and effort in the field and laboratory.
Habitat is an important consideration when evaluating aquatic systems. To examine the
quality of the habitat, habitat evaluations are conducted at each stream station. These evaluations
focus on parameters such as substrate (bottom sediments) characteristics, flow regimes, impacts to
the stream channel (eg., channelization, deposition), impacts to streamside vegetation, stability of
the stream banks, and available cover. Ecoregional reference sites provide a basis for the best
attainable conditions for all streams with similar physical dimensions for a given ecoregion.
Presently, there are two reference sites per ecoregion except for the coastal plain ecoregion. The
process of reference site identification is still ongoing.
Fish were chosen primarily for their societal value and role as a top consumer in streams.
Fish are relatively easy to identify and with minimal training most fish can be collected, sorted, and
identified at the field site and then released unharmed. They are an important part of the food web.
Fish are found in the smallest of streams and some are even found in heavily polluted streams.
They occupy positions throughout the food web and include groups that represent a variety of
feeding types. Their diet often consists of food derived from both inside the stream and outside the
stream. Fish serve as one of the major predators of stream insects. Changes in the stream insect
9
-------
community often result in a change in the fish community. Like stream insect communities, fish
communities will respond to environmental change, whether it is chemical or physical. Some fishes
are very sensitive to environmental change while others are not. By examining all fish groups that
live in a stream, the general condition of a stream can be assessed. For example, if there are only
one or two groups of fish in a stream who are very tolerant to pollution, and there are no groups that
are sensitive to environmental change, then impairment is suspected because of environmental
change that has eliminated the sensitive groups.
The Environmental Protection Agency's Rapid Bioassessment Protocol V (RBP V) is an
index used to assess stream condition based on the fish community. The index consists of twelve
measures scored to assess changes in the fish community compared to a reference stream, or a stream
with least impact. For example, one of the measures assesses the proportion of fishes in a stream
considered to be tolerant to environmental change. If the proportion of tolerant groups are high
compared to the reference stream, then this would result in a lower score for that measure. Another
measure looks at the number of fish groups. If the number of fish groups collected is similar to that
of the reference stream then this would result in a high score. After ail twelve measures have been
given a score, the scores are totaled and the condition of the fish community is then characterized
as either good, fair, or poor depending on how far the total score deviates from that of a reference
stream.
The primary indicator selected to address trophic condition in streams is the algal growth
potential test (AGPT). The AGPT is based on the premise that maximum yield of plants (e.g. algae)
is limited by the amount of nutrients available to the test alga. With higher algal growth
concentrations (AGPT), there is good likelihood that obnoxious plant growths can occur in a stream.
-------
The test was selected as the indicator of choice to assess trophic condition primarily because of its
specific sensitivity, reliability and the ease and economy of using it as a monitoring tool.
MONITORING ASSESSMENT
One of the objectives of this study is to estimate ecological condition in the Savannah River
Basin. The task is to establish action levels (index score or concentration). These levels are then
used to decide if a stream segment is in good, fair, or poor ecological condition related to a particular
societal value or issue of concern. Approaches used to establish action levels included assessment
of reference sites, data analysis, and field experience. Development of action levels for indicators
used in this study provides the opportunity to estimate miles of wadeable streams in poor, fair, or
good condition. Conforming to the adage "that a picture is worth a thousand words," estimates of
the percent of wadeable stream miles in a certain condition are made easier to understand by the
cumulative distribution curve (CDF). These curves show the percent of wadeable stream miles equal
to or less than some specified concentration or index number plus or minus a confidence level. For
the purposes of this study, we have set a confidence level of 95%. This means that we are 95% sure
that the present stream miles estimated to be equal or less than a given index score or concentration
is within the bounds of our confidence lines on the graph (Figure 2). There is only 1 in 20 chances
(5% error) that the true or real percent of stream miles affected at a particular concentration or a
score is not within the confidence bounds.
11
-------
Habitat evaluation
identifies degradation at a given
site that may be of a physical
nature (i.e., stream erosion) rather
than water quality. A good
portion of the wadeable stream
miles (70%) had degraded habitat
ranging from fair to poor (Figure
SAVANNAH RIVER BASIN REMAP
HABITAT
100
U)
111
90
—I
80 ¦
70 :
2
fin
<
LU
50 -
CXL
40
O)
30
4—
o
20
10
o
o +-¦
40 60 80 100
HABITAT SCORE - % of REFERENCE
95 % Confidence Limits
]
Figure 2. Cumulative distribution of habitat score.
2). Based on on-site habitat
evaluations, most of the degradation could be attributed to non-point source sediment erosion,
deposition of sediments, and stream bank failure due to loss of bankside vegetation. This finding
agrees with other reports from state resource agencies in the southeastern United States that have
identified non-point source pollution, and especially sedimentation as a major concern.
The stream insect EPT Index,
through the course of the Savannah
River REMAP, has emerged as a
valuable indicator for interpreting
stream biological integrity. This
indicator is simply a summation of
the total number of pollution-
sensitive stream insects. These
pollution-sensitive stream insects are
SAVANNAH RIVER BASIN REMAP
EPT INDEX
100
Sj 90
=d 80
^ 70
2 60
lu 50
£ 40
CO 30
•5 20
* 10
0
}y/
Q.
POOR;]
GOOD
FAIR
—I 1—
r
1 f t 1 1 1" 1 1 liil
0
10 15
EPT INDEX
20
95 % Confidence limits
Figure 3. Cumulative distribution of EP I index.
12
-------
the mayflies (Ephemeroptera), stoneflies (Plecoptera), and caddisflies (Trichoptera) and their
presence or absence suggests good water quality. Stream miles in the "good" category were
characterized by an EPT Index score of seven or greater. About 40% of the stream miles exhibited
EPT index scores in the "good" category. The remaining stream miles were less than seven (Figure
3). That is, biological integrity ranged from fair to poor. Not all of this impact is attributable to
habitat quality. Over half of the sites (53%) identified as impacted by the EPT Index score were
limited by fair to poor habitat quality, the remaining 47% were affected by other unknown stresses.
Of the twelve measures used for the fish analysis, four were very informative about the
condition of the fish community. The four measures were; (1) the proportion of omnivores (fish
that eats both plants and animals), (2) the total number of darters (small fish that inhabit the bottom
and primarily feed on stream insects), (3) the total number of suckers ( a group of small to relatively
large fish that inhabit the bottom and feed on stream insects), and (4) the total number of different
kinds offish collected. The proportion of omnivores reflect change in the food web of the stream.
If the food web is disrupted in some way, fish adapted to eat anything will dominate over those who
are more specialized feeders. Darters and suckers are sensitive to changes in the habitat and the
stream insect community. If the habitat has been disrupted by increased siltation or if the stream
insect community is damaged, then these kinds of fish would become reduced or absent in the
stream. The total number of different fish found reflects all environmental changes to the stream.
The results of analyzing these four measures alone would be very similar to the results of analyzing
all twelve measures. However, all twelve measures were used to determine the fish IBI.
The RBP V fish analysis with all twelve measures have a total possible score of 60. Streams
with greater scores than 43 are considered in "good" condition. Streams whose scores are between
13
-------
100
CO
LU
3 60
S 50
£ 40
co 30
"5 20 f
>p 10
10
30 and 43 are considered in "fair"
condition, and streams whose
scores were below 30 were
considered in "poor" condition.
Based on the 1994-1995 data, only
7% of the fish communities in the
basin were in good condition, 93%
were impacted. Poor sites
represent 52% of the impacted
basin streams with the remaining 41% in fair condition (Figure 4).
The Ecological Support
Branch (ESB) has conducted many
AGPT's related to studies in the
southeast. Based on experience,
literature review, and data analysis
ESB set an action level of 5mg/L
test algal (dry weight) that would
reasonably assure protection from
SAVANNAH RIVER BASIN REMAP
FISH
:
'/V
tr
/
FAIR |
jaooDj
yS
—f >¦
20 30 40
FISH IBI SCORE
50
- 95 % Confidence Lirrrts
SAVANNAH RIVER BASIN REMAP
STREAM AGPT
mo
CO
lil
90
_l
80
5
70
5
in
¦ 60
50
or
40
CO
30
y*—
O
20
10
o J*
0
10
20 30 40
AGPT mg/L
50
95 % Confidence Limits
zl
60
rjf ¦
-
7 :
I/J GOOD
¦f -- ^ ¦ ' - - ----- -
FAm
m
1 1 1
—i—f—i—i—.—
60
nuisance plant growths and fish Fi§ure 5" Cumulative distribution curve for AGPT.
kills in southeastern lakes. Their experience of applying this test to stream waters is limited, but they
have suggested AGPT levels that are conservative yet would be protective of basin streams. AGPT
dry weights of equal to or less than 20mg/L converts to 0.036mg/L of phosphorus. They believe that
14
-------
streams containing phosphorus concentrations equal to or less than this level could be considered in
good condition. Dry weight yields greater than 30 mg/L amount to 0.072 mg/L of biologically
available phosphorus. Streams at this level are considered in poor trophic condition and in need of
further investigation. An examination of Figure 5 suggests that about 15% of the wadeable stream
miles in the basin are in poor trophic condition. That is, excessive nutrient inputs may be causing
unsightly algal growths. The good news is that about 85% of the stream miles are in fair to good
condition. Thus, nutrient levels are probably not causing nuisance growths, low dissolved oxygen
problems, or fish kills in most of the wadeable streams.
Figure 6 presents a
summary of stream classification
(good, fair, poor) based on all
ecological indicators. Based on the
first two years of sampling, most of
the wadeable streams in the basin were
in good condition with respect to
the present evaluation of
enrichment as measured by the
growth response of the algae.
Stream miles classified as "poor" were more prevalent among the ecological indicators that provide
a measure of stream ecological condition over time (fish, insects, and habitat). For example, miles
classified as poor were as follows: fish = 52%, insects = 33%, and habitat = 38%. Only 15% of the
sampling sites were classified as "poor" based on the algae (AGPT). This ecological indicator is a
15
Ecological Assessment Summary
Classification of Stream indicators
Savannah River Basin REMAP
Q. Fi#i H»MW lrw»c*s Algae
Indicator
HI Good Q Fjir | Poor
Figure 6. Summary of CDF curve classification of basin
river miles.
-------
measure of ennchment potent,al during the aetual sampling event andi. •
event and ,s sensrtiVe to inputs from non-
point source pollution generated by storm runoff. Conseuuen.lv ,L
„ „h . u. the percentage of m.les classified
as "poor by algae would increase during wet periods.
We have just begun to
explore stress indicators like
landscape changes and their
association with condition
indicators like the EPT index by
using information in the Branch's
Geographical Information System
(GIS). Landscape records
ranging from non-point source
features like coverages of
industrial waste sites were
examined for association with
poor ecological areas. On-going
data analysis continues in further
SAVANNAH RIVER BASIN
Feeding Operations
WWTP Sites
Figure 7. Location of poor ecological
areas.
refining GIS coverage and
possible cause-effect relationships via association analyses. However, it was observed that many
sites in "poor" ecological condition seemed to cluster in two areas of the basin, one in Georgia and
one in South Carolina (Figure 7).
Degradation of habitat (Figure 6) and two landscape stressors may partially explain some
16
-------
causes of poor ecological condition in the Georgia area that covers six counties and ten watersheds.
The stressors are wastewater treatment plants (WWTP) and animal feeding operations. Wastewater
treatment plants were upstream of several sites in the Georgia area. They may be the chief cause of
enrichment in Hart County as evidenced by the AGPT results. Animal feeding operations are
plentiful in the northwestern Georgia piedmont. Because of the intensive feeding of animals in
confined spaces, heavy organic pollution from water running off farmland will influence stream
trophic condition and community biological integrity. Some thing(s) are affecting the stream insect
and fish community integrity in the South Carolina area that covers five counties and eight
watersheds. Besides noted habitat degradation (Figure 6) from sediment erosion, deposition, of
sediments, and stream bank failure, none of the landscape intelligence available showed obvious
clustering of landscape features associated with poor ecological condition.
17
-------
BIBLIOGRAPHY
Anonymous. 1995. Savannah River Basin Watershed Project: Initial Assessment and Prioritization
Report for the Savannah River Basin. Volume I. Prepared by the Management Committee of the
Savannah River Basin Watershed Project.
FTN Associates, Ltd., R. L. Raschke, H. S. Howard. 1994. Plan of Study for Characterizing
Ecological Condition in the Savannah River Basin. U. S. Environmental Protection Agency Region
4, Environmental Services Division, Ecological Support Branch, 625 Bailey St. Athens, GA 30605.
Klemm, D. J., Q. J. Stober, and J. M. Lazorchak. 1993. Fish Field and Laboratory Methods for
Evaluating the Biological Integrity of Surface Waters. EPA/600/R-92/111.
Larsen, D.P., D. L. Stevens, A. R. Selle, and S. G. Paulsen. 1991. Environmental Monitoring and
Assessment Program: EMAP - Surface Waters, a Northeast Lakes Pilot. Lake and Reservoir
Management 7(1): 1-11.
Hunsaker, C. T., and D. E. Carpenter, eds. 1990. Ecological Indicators for the Environmental
Monitoring and Assessment Program. EPA 600/3-90/060.
Plafkin, J. L., M. T. Barbour, K. D. Porter, S. K. Gross, and R. M. Hughes. 1989. Rapid
Bioassesment Protocols for Use in Streams and Rivers: Benthic Macroinvertebrate and Fish.
EPA/444/4-89-001.
Raschke, R. L., and D. A. Schultz. 1987. The Use of the Algal Growth Potential Test for Data
Assessment. Journal Water Pollution Control Federation 59(4):222-227.
Volstad, J. H., S. Weisberg, D. Heimbuch, H. Wilson, and J. Seibel. 1995. Answers to Commonly
Asked Questions about REMAP Sampling Designs and Data Analyses. U. S. Environmental
Protection Agency, Research Triangle Park, NC.
18
-------
-------
Appendix I
Savannah River Basin REMAP Interim Report: Large
Lake Embayments
By
Ron Raschke et al.
-------
SAVANNAH RIVER BASIN REMAP
AN INTERIM REPORT: LARGE LAKE EMBAYMENTS
• •: : : ' 'A
4 . :..." *•
&¦ \
' * v
; :
n& .. "jB
m,*
by
&
^5.
i4 >
VI.#: ( : ,
• rr"/
-5:^ if
RON RASCHKE1
f
./.v. >~>$»*» ¦ f ;;r ¦'
BOB QUINN
TONY OLSEN1
DON STEVENS JR.*
.
x in
\:hy
*'»• i w»i ' ¦¦'¦'¦ ;
r teS
' tyy'
'¦y-A; .
pi
II
-
- ¦
y ; . •,
1U- S. Environmental Protection Agency, Science and Ecosystem Support Division,
Ecological Assessment Branch, 980 College Station Rd., Athens, GA 30605
*U. S. Environmental Protection Agency, NERL-Corvallis, 200 S. W. 35th St, Corvallls,
OR 97333
"Dynamac Corporation, 200 S. W. 35th St, Corvallls, OR 97333
-------
CONVERSION FACTOR TABLE
Multiply By To Obtain
Meters 3.281 Feet
Hectares 2.469 Acres
Meters Square 2.47 x 10" Acres
-------
TABLE OF CONTENTS
CONTENTS PAGE
Conversion Factor Table i
Table of Contents ii
Summary iv
1.0 Introduction 1
1.1 Purpose 1
1.2 Monitoring 1
1.3 Policy Relevant Questions 3
1.4 Objectives 3
1.5 Description of the Savannah River Basin 4
1.6 Reservoirs (Lakes) 7
1.7 Study Design 8
1.7.1 Resources of Interest 8
1.7.2 Statistical Sampling Design 9
1.7.3 Frame Material 10
ii
-------
1.7.4 Sample Site Selection
1.8 Indicators
1.8.1 Societal Values
1.8.2 Types and Selection of Indicators
2.0 Methods/QA
3.0 Findings
4.0 References
Appendix A REMAP Sampling Points
Appendix B Lake Embayment Stations Sampled
iii
-------
SUMMARY
As part of a proposed new monitoring strategy (REMAP) demonstration for the states of
Georgia and South Carolina, a systematic-random sampling of embayments of major reservoirs
(lakes) in the Savannah River Basin was initiated in 1995. A focus was placed on tributary
embayments because they are the first portion of a reservoir to exhibit adverse nutrient impacts
from man-induced changes. This report covers an assessment of trophic condition over a two-
year period (1995-1996).
Trophic indicators showed that generally embayment acreage was in good condition, and
it was not immediately threatened with unsightly nuisance phytoplankton blooms. Chlorophyll a
was less than 12 ug/L, a concentration that meets guidelines for multiple uses, including drinking
water supply. Approximately 99% of the algal growth potential standing crop was equal to or
less than 5 mg/L, an action level that assures protection from fish kills and nuisance blooms.
Total phosphorus ranged from 2 to 60 ug/L. Eighty-seven percent of the embayment acreage
contained less than or equal to 10 ug/L of total phosphorus. With respect to water clarity, Secchi
disc depth showed that only 2.6% of the acreage was less than desirable. Based on total
suspended solids concentration, 0% of the waters were classified as muddy and only 3% of the
acreage was classified intermediate. The low percent of acreage under less than desirable
conditions were associated with near-shore stations receiving wind fetches at the time of
sampling.
iv
-------
1.0 INTRODUCTION
1.1 PURPOSE
Responding to increased growth and demands for multiple uses
of natural resources, The Environmental Protection Agency (EPA)
established the Watershed Protection Approach (WPA) in 1991 (EPA,
1991; 1996). The Savannah River Basin was one of two areas
selected in 1993 for the WPA in Region 4 because of its high
public use, known environmental problems, susceptibility for
further degradation, interest in participation by the users, and
the likelihood of success. Through the WPA initiative, EPA
Region 4 brought together scientists and stakeholders who
developed a stategy to provide an ecological focus for resolving
problems. This strategy gave birth to the Savannah River Basin
Watershed Project (SRBWP) (Management Committee, 1995). The goal
of the SRBWP is to develop and implement a multiagency
environmental protection management project which incorportates
the authorities and expertise of all interested parties in an
effort to accomplish the vision of conserving, restoring,
enhancing, and protecting the basins ecosystems in a way that
allows the balancing of multiple uses. Further details on
objectives and issues within the basin can be found in Volume I
of the "SRBWP Initial Assessment and Priorization Report" by the
Management Committee (1995). Part of the SRBWP strategy included
a monitoring component, The Regional Environmental Monitoring and
Assessment Program (REMAP) (FTN at. al., 1994) .
1.2 MONITORING
Environmental monitoring programs have developed in response
to specific needs, such as compliance monitoring by regulating
agencies responsible for the condition of surface waters, or
fixed-station monitoring networks that primarily address
indicators of exposure and stress. Some of the monitoring
programs are driven by mandates in the Clean Water Act (CWA).
The reports required by Sections 305(b) and 314 of the CWA are an
example. Programs that collect data on other ecosystem types
have also been established. For example, the u. S. Department of
Agriculture (USDA) National Agricultural Statistical Survey
collects data for agricultural resources; The Forest Service's
inventory and analysis surveys of forest resources; and the U. S.
1
-------
Geological Survey's National Water Quality Assessment (NAWQA)
program monitors water quality in selected basins. None of the
programs, however, have adopted a uniform approach for national
and regional assessments across and among ecosystem types. The
Environmental Monitoring and Assessment Program (EMAP) and its
counterpart, REMAP is intended to fill that gap by providing the
U. S. EPA Administrator, Congress, and the public with
statistical data summaries and periodic interpretative reports on
ecological status and trends. Because knowledge about
uncertainity is important for interpreting quantitative
environmental data, EMAP is designed to. make rigorous
uncertainity estimates as well (Larsen et al., 1991) .
The REMAP was developed as a partnership between EMAP, EPA's
Regional Offices, and States to promote the use of EMAP science.
The objectives of REMAP follow:
1. To evaluate and improve EMAP concepts for State and local
use.
2. To assess the applicability of EMAP indicators and the
EMAP approach at differing spatial scales, and
3. To demonstrate the utility of EMAP for resolving issues
of importance to EPA, Regions, and States.
The REMAP strategy lends itself to the benefits of a full
partnership between states and federal agencies because both
national and state monitoring needs can be met in a cost-
effective manner. The EMAP approach can provide a cost effective
approach for assessing ecological data and reporting estimates of
status and trends in indicators of condition with known
confidence. State reporting requirements under several sections
of the Clean Water Act (CWA) can be accomplished using an EMAP
monitoring approach. Section 305(b) of the CWA requires states
to submit biennial reports that include analysis of water quality
data of all navigable waterways to estimate environmental
impacts. The Clean Lakes section 314 requires states to submit
biennial reports that identify, classify, describe and assess
status and trends in water quality of publicly owned lakes.
REMAP projects are being designed to provide meaningful
information to decision-makers within a 1- to 2-year period.
2
-------
1.3 POLICY RELEVANT QUESTIONS
The Science and Ecosystem Support Division (SESD) of EPA
Region 4 was asked by the Savannah River Watershed Project Policy
Committee to implement the REMAP strategy as a demonstration
project for the states of South Carolina and Georgia. These
states were interested in reducing sampling frequency and
analyses, having the ability to reduce or increase sampling
density, responding quickly to emerging environmental problems,
and maintaining representative coverage of environmental
resources through a systematic yet random means of sampling.
Before the monitoring study, a set of questions was posed by the
states of Georgia and South Carolina to provide direction for the
monitoring design. The following policy-relevant questions were
identified to guide the development of a plan of study and
subsequent monitoring efforts.
» What is the status of condition of the water resources of
the Savannah River Basin?
~ What proportion of the Savannah River Basin surface waters
are attaining designated uses?
*¦ What are the changes of ecological condition over time?
~ What factors might be associated with changes?
»• Is there a tendency for distribution of condition in a
specific direction (spatial gradient) over the basin
landscape? What are the possible reasons for these
gradients?
*¦ What resources are at risk in the Savannah River Basin?
1.4 OBJECTIVES
In response to the needs of the states and policy-relevant
questions posed, The Ecological Assessment Branch (EAB) of the
SESD developed the following study objectives with the
concurrence of the Policy Committee of the Savannah River
Watershed Project.
~ Estimate the status and change of the condition of water
3
-------
resources in the Savannah River Basin;
~ Identify water quality spatial gradients that exist within
the Savannah River Basin and associate current and changing
condition with factors that may be contributing to this
condition and spatial gradients;
* Demonstrate the utility of the REMAP approach for watershed
and river basin monitoring and its applicability for state
monitoring programs;
~ Incorporate the REMAP approach in the formulation and
accomplishment of the
State River Basin
Management Plans; and
*• Provide baseline
information required
to conduct comparative
risk assessments in
the Savannah River
Basin.
1.5 DESCRIPTION OF THE
SAVANNAH RIVER BASIN
The Savannah River
originates in the mountains
of Georgia, South Carolina,
and North Carolina and
flows south-southeasterly
312 miles to the Atlantic
Ocean near the port city of
Savannah Georgia (Figure
1). The Savannah River is
formed at Hartwell
Reservoir by the Seneca and
Tugaloo Rivers. Above the
confluence of the Seneca
and Tugaloo Rivers, the
headwater streams of the
Seneca River are the Keowee
River and Twelve Mile
Creek. The Tugaloo River
1. Savannah River Basin.
GEORGIA
NORTH CAROLINA
SOUTH CAROLINA
4
-------
is formed by the confluence of the Taliulah and Chattooga Rivers.
The Savannah River flowing in a south-southeasterly direction
forms the border between
the states of Georgia and
South Carolina. The
rivers' entire length of
312 miles is regulated by
three adjoining Corps of
Engineers multipurpose
reservoirs, each with
appreciable storage. The
three lakes, Hartwell,
Russell, and Thurmond,
form a chain along the
Georgia-South Carolina
border 120 miles long.
Six power developments
that are part of the
Georgia Power Company
hydropower network exist
upstream of Hartwell Lake
on the Tugaloo River
system,- Yonah and Tugaloo
lakes on the Tugaloo
River, and Taliulah
Falls, Rabun, Seed, and
Burton lakes on the
Taliulah River. Upstream
of Lake Hartwell, on the Figure 2. Location of Major Lakes in the
Seneca River, is Duke Savannah River Basin.
Power Company's Keowee-
Toxaway Project. The project is composed of two adjoining
reservoirs, the most downstream of which is Keowee Lake, and the
other two, Jocassee and Bad Creek Lakes are pump storage
proj ects.
The Savannah River Basin has a surface area of 10,577 square
miles, of which 4,581 square miles are in South Carolina, 5,821
square miles are in Georgia, and approximately 175 square miles
are in North Carolina. Like other basins of large rivers in the
Southeast which flow into the Atlantic Ocean, the Savannah River
Basin embraces three distinct areas: the Mountain Province, the
Piedmont Province, and the Coastal Plain. The mountains and
"^TALLULAH ^ALLS
kAUGUSTA
• THURMOND • BURTON
• RUSSELL SEED
• HARTWELL RABUN
KEOWEE TUGALOO
JOCASSEE YONAH
BAD CREEK
5
-------
Piedmont are part of the
Appalachian area. The
division between the
Mountain and Piedmont is an
irregular line extending
from northeast to southwest,
crossing the Tallulah River
at Tallulah Falls. The Fall
Line, or
division between the
Piedmont Province and the
Coastal Plain, also crosses
the basin in a generally
northeast to southwest
direction, near Augusta,
Georgia. Elevations within
the Mountain Province of the
basin vary from 1,500 feet
National Geodetic Vertical
Datum (NGVD) on the Tallulah
River to 5,030 feet NGVD for
the highest peak, Little
Bald Mountain, in North
Carolina along the watershed
divide. The Piedmont
Province, due to its great
width of over a hundred Figure 3. Physiographic Provinces of the
miles, is truly piedmont Savannah River Basin,
only in the upper parts, and
gives way to a midland area before reaching the Coastal Plain.
Exclusive of river valleys, its elevation generally varies from
500 feet NGVD at the Fall Line to about 1,800 feet NGVD at its
upper extremity. Elevations within the Coastal Plain vary from
500 feet NGVD at the Fall Line to sea level at the Atlantic
Ocean.
ORASSTOWN BALD MOUNTAIN
f
tjf I
^ -
/ ^
Ja
4+ clq^son
3r \
\
1
'ij • calAqun falls
^ N
ATHENS •
S
\
TVf r'
AUGUSTA* 1
\ 1
V \
"s|
PALL LINE
—J
\ \
\ \
-\ ^
I
i
v
\ 5
( \
SAVAMW^^
~ MOUNTAIN ~ PEDMOMT ~ OGASTAL PLAIN
Land use in the basin is agriculturally oriented. Sixty-six
percent of the basin is considered timberland and 34.1% is
nonforested. The number of acres farmed remains constant.
Between 1987 and 1992 there was little change in the total farm
acreage in the basin. However, Georgia had 330 fewer farms and
lesser acreage in 1992 than in 1987 while South Carolina had an
increase of 931 farms and an increase of 110,134 acres in farm
6
-------
land. There was a shift over the same five-year period in the
types of crops grown. An increase in the number of acres
cultivated have occured in corn (18%), cotton (86%), peanuts
(12%), and tobacco (31%). These gains have been made with
corresponding decreases in primarily wheat (-30%) and soybeans (-
32%) .
The Savannah River Basin contains all or part of 43 counties
in Georgia, South Carolina, and North Carolina. Four of the
counties are in North Carolina, thirteen in South Carolina, anH
twenty-six in Georgia. The population of the basin in 1990 was
about 1,500,000 and is expected to grow to 1,800,000 by the year
2030. About 53% of the population resides in Georgia, 42% in
South Carolina, and 5% in the headwaters located in North
Carolina. Four metropolitan areas contain 62% of the basin's
population. Savannah, Georgia is the largest city with 137,560
persons followed by Augusta, Georgia with a population of 44,619.
1.6 RESERVOIRS (LAKES)
Reservoirs herein called lakes are defined as bodies of
water that have a surface area of at least 4 ha, with a depth of
at least 1 meter and at least 1,000 m2 of the surface area in
open water.
Tributary embayments of six major lakes were studied over a
two-year period (1995 & 1996). These lakes were Burton,
Jocassee, and Keowee located in the Mountain Province. The other
three lakes, Hartwell, Russell, and Thurmond were located in the
Piedmont Province.
Lake Burton, controlled by Georgia Power Company, is located
near Clayton Georgia. It is an old reservoir impounded in 1919.
The lake has a shoreline length of 62 miles surrounding 2,775
acres containing 1,000,080 acre-feet of water.
Hartwell Lake is 7 miles east of Hartwell, Georgia. A dam
is located at river mile 305.0. When the lake level is at
elevation 660 ft. NGVD, the top of the conservation pool, the
lake extends 49 miles up the Tugaloo River, Georgia, and 45 miles
up the Seneca and Keowee Rivers, South Carolina covering 55,900
acres. The shoreline at elevation 660 is about 962 miles long,
7
-------
excluding island areas. The lake has a total storage capacity of
2,550,000 acre-feet below elevation 660. Hartwell dam was closed
in 1963.
Russell dam is at River Mile 275.2 in Elbert County, Georgia
and Abbeville County, South Carolina. The dam is 18 miles
southwest of Calhoun Falls, South Carolina, and 40 miles
northeast of Athens, Georgia. At the top of conservation pool
elevation of 475 NGVD, the lake has a useable storage capacity of
126,800 acre-feet and a shoreline of 523 miles encompassing
26,000 acres. Operation of the project began in Jaunuary, 1984.
Thurmond Lake is 22 miles upstream of Augusta Georgia. At
elevation 330 NGVD, at the top of the lake pool, the lake extends
40 miles up the Savannah River and about 30 miles up the Little
River in Georgia. The lake has about 1,050 miles of shoreline,
excluding island areas. At the top of the flood control pool
(elevation 335 NGVD), the lake has an area of 78,500 acres with a
total storage capacity of 2,510,000 acre-feet.
The three-project system is authorized and operated for fish
and wildlife, flood control, hydropower, navigation, recreation,
water quality, and water supply.
Duke Power Company built and controls Lakes Jocassee and
Keowee. The uppermost lake, Jocassee, was built in 1973. It
contains an area of 7,318 acres holding 1,077 acre-feet of water
with a shoreline length of 75 miles. Lake Keowee built in 1971
has a shoreline length of 300 miles encompassing 18,373 acres
with a storage holding capacity of 955 acre-feet.
1.7 STUDY DESIGN
1.7.1 Resources of Interest
The statistical population of interest included all
tributary embayments >20 hectares associated with lakes >500
hectares. A tributary embayment is defined as a body of water
associated with, but offset from, the main lake that has a
permanent, blue-line stream at its headwaters. The embayment
begins at the plunge point, the stream stretch where the inflow
water density is greater than the density of the lake surface
water, and it joins the main body of the lake at the plane
8
-------
created by intersecting break points of the shoreline of the
embayment with the main body. Tributary embayments are
associated only with lakes that have a shore lins development
ratio >3.0 and a surface area >500 hectares (FTN bL al., 1994).
Shore line development is the ratio of the actual length of shore
line of a lake to the length of the circumference of a circle the
area of which is equal to that of a lake, if a lake had a
shoreline in the form of a circle, the shore line development
would be 1.0 (Welch, 1948).
1.7.2 Statistical Sampling Design
A probablistic sampling survey strategy was used to
characterize the lake embayments of the Savannah River Basin.
The sampling design was derived from the approach used in the
EMAP (Messer et al., 1991; Overton et al., 1990; Stevens et al.,
1992).
Probability sampling designs use randomization in the sample
selection process. Probability sampling is the general term
applied to sampling plans in which
»• every member of the population (i.e., the total assemblage
from which individual sample wits can be selected) a
known probability (>0) of being included in the sasg>le;
*¦ the sample is drawn by some method of random selection
consistent with these probabilities; and
»• the probabilities of selection are used in inferences
from the sample to the target population (Snedecor wid
Cochran, 1967),
One advantage of probability-based surveys is their minimal
reliance on assumptions about the underlying structure of the
population (e.g., normal distribution). In fact, one of the
goals of probability-based surveys is to describe the underlying
structure of the population. Randomization is an important
aspect of probability-based surveys. Randomization ensures that
the sample represents the population. Without probability
sampling, each sample often is assumed to have equal
representation in the target population, even though selection
criteria clearly indicate this is not the case. Without the
underlying statistical design and probability samples, the
9
-------
representativeness of an individual sample is unknown. Drawing
inferences from samples selected without randomization and
without incorporating inclusion probabilities can yield
misleading conclusions. To provide policy-relevant information,
not only is the ecological condition of the target population
important, but also the proportion of the resource that is in a
particular state of condition. Very different policy and
management alternatives might be evaluated if 50% rather than
<10% of target embayments are hypereutrophic.
1.7.3 Frame Material
A sampling frame is an explicit representation of a
population from which a sample can be selected. The sampling
frame for the lake embayments is the USGS 1:100,000-scale map
series in digital format (DLGs) and the modification of the DLGs
represented by the U.S. EPA River Reach File (RF3), which
established edge matching and directionality in the DLG files.
From this, we used all lake areas identified as tributary
embayments.
1.7.4 Sanple Site Selection
The survey design follows the general design strategy
proposed for EMAP (Overton fit al.,1990; Messer at al., The EMAP
sampling design (Overton at. al., 1990) achieves comprehensive
coverage of ecological resources through the use of a grid
structure. White e£ al. (1992) describe the construction of the
underlying triangular point grid and its associated tessellation
of hexagonal areas. The EMAP base grid has a point density of
one point per 635 Jcm2. The base grid is intensified through a 7 x
7 fold enhancement, a 49-fold increase in grid density (White et
al., 1992) which results in one point per 13 km2 hexagonal area.
The hexagonal tessellation was randomonly located over the area
covered by the embayment population. Within each hexagon, a
point was randomonly selected. If the point fell within one of
the embayments, then that point became a sample point. The
selection process ensures that each location in the embayment
population is equally likely to be sampled, and that the set of
sites are spatially distributed throughout all embayments.
Stevens (1997) defines this sampling process as a random
tessellation stratified design. Since the study extended over
two-years, two independent samples were selected - one for each
10
-------
year. A total of 111 sample locations were selected such that 52
were allocated in 1995 and 59 in 1996.
1.8 INDICATORS
REMAP monitors ecological indicators to assess status,
trends, and changes in the condition and extent of the Region's
ecological resources (Bromberg, 1990, Hunsaker and Carpenter,
1990; Hunsaker et al., 1990). Indicators are defined as any-
characteristic of the environment that estimates the condition of
ecological resources, magnitude of stress, exposure of a
biological component to stress, or the amount of change in
condition.
Ecological principles state that ecosystem responses and
condition are determined by the interaction of all the physical,
chemical, and biological components in the system. Because it is
impossible to measure all these components, REMAP's strategy
emphasizes indicators of ecological structure, composition, and
function that represent the condition of ecological resources
relative to societal values. The challenge is to determine which
ecological indicators to monitor. One approach for selecting
these indicators starts with those attributes valued by society
and determines which indicators might be associated with these
values.
1.8.1 Societal Values
To be effective, information from the monitoring program
must prompt action when required. This means the information
produced must be related to perceptions of aquatic health and
represent issues and values of concern and importance to the
public, aquatic scientists and decision makers. The selection of
these societal values drives the selection of appropriate
indicators After extensive discussions with aquatic resources
managers, decision makers and the scientific community by members
of the EMAP - Surface Waters Resource Group (Larsen and Christie
1993), an initial set of societal values and concerns were
identified for evaluation in EMAP. These values are:
~ Biological Integrity,
~ Trophic Condition, and
11
-------
~ Fishability
Biological integrity can be defined as the ability to
support and maintain a balanced, integrated, adaptive community
with a biological diversity, composition, and functional
organization comparable to those of natural lakes and streams of
the region (Frey 1977; Karr and Dudley 1981) and includes various
levels of biological, taxonomic and ecological organization (Noss
1990). Waters in which composition, structure and function have
not been adversely impaired by human activities have biological
integrity (Karr et al. 1986). Karr and others (1986) also
defined a system as healthy "when its inherent potential is
realized...and minimal external support for management is
needed." This value or ethic differs considerably from values
oriented toward human use or pollution that are traditionally
assessed in water quality and fisheries programs, in which
production of a particular species of game fish is the goal
(e.g., Doudoroff and Warren, 1957), and may conflict with these
definitions (Callicott 1991; Hughes and Noss 1992; Pister 1987).
Fishability is defined as the catchability and edibility of
fish and shellfish by humans and wildlife (Larsen and Christie
1993) . Fish represent a major human use of an aquatic ecosystem
product. Protecting fish is the goal of many water quality
agencies, and fish drive their water quality standards.
Trophic condition has been defined in EMAP as the abundance
of production of algae and macrophytes (Larsen and Christie
1993) . Trophic condition involves both aesthetic (water clarity)
and fundamental ecological (production of plant biomass)
components. It is a key aspect in determining both a lake's
relative desirability to the public, its production of fish and
its ecological character or classification by limnologists (e.g.,
eutrophic or oligotrophic). Because of limited resources, a
decision was made to concentrate on trophic condition indicators
for lakes.
1.8.2 Types and Selection of Indicators
EMAP defines two general types of ecological indicators,
condition and stressor indicators. A condition indicator is any
characteristic of the environment that estimates the condition of
12
-------
ecological resources and is conceptually tied to a value. There
are two types of condition indicators: biotic anr) abiotic.
Condition indicators relate to EMAP's first and second
objectives: estimating the status, trends, and changes in
ecological condition; and the extent of ecological resources.
Stressor indicators are characteristics of the environment
that are suspected to elicit a change in the condition of an
ecological resource, and they include both natural and human-
induced stressors. Selected stressor indicators are monitored in
EMAP only when a relationship between specific condition and
strssor indicators are known, or a testable hypothisis can be
formulated. Monitoring selected stressor and condition
indicators addresses the third EMAP objective of seeking
associations between selected indicators of stress and ecological
condition. These associations can provide insight and lead to
the formulation of hypotheses regarding factors that might be
contributing to the observed condition. These associations can
provide direction for other regulatory, management, or research
programs in establishing relationships.
We focused on condition indicators related to trophic
condition because of limited resources. The original study plan
(FTN al., 1994) proposed sampling for fishability indicators,
Fish Health Index and Fish Tissue Residues; biological integrity,
phytoplankton and zooplankton identification and counts; and one
other trophic condition indicator, zeaxanthin, a marker pigment
for blue-green algae. Work is continuing on this pigment, but
the information was not sufficient for inclusion into this
report.
The trophic condition indicators measured during this study
were corrected chlorophyll a, total phosphorus (TP), algal growth
potential (AGP), secchi disc transparency, and total suspended
solids (TSS). These indicators were selected because they
provide different insights into the condition of the embayment
waters.
Corrected chlorophyll a is commonly used to estimate the
degree of phytoplankton bloom conditions that can affect
aesthetics, fishing and swimming qua 1 y, wat-erfowi and
fishes and drinking water, and the health of fish, waterfowl, and
livestock. Chlorophyll is a measure of instantaneous standing
13
-------
crop, whereas TP and AGP indicate potential for blooms. Total
phosphorus reveals insights about nutrient input and the
potential for serious bloom conditions if we assume all of it is
available. However, much of the TP is not normally available.
The AGP can show how much of the TP is available for algal growth
and the potential, under optimum conditions, for blooms. Secchi
disc transparency is related to swimming conditions. Total
suspended solids is related to transparency, but it also can be
used to indicate effects upon fish production.
2.0 METHODS/QA
Standard operating procedures (SOP) of the Ecological
Assessment Branch and Analytical Support Branch of EPA's Region 4
SESD were followed as the principle means of monitoring
appropriate QA and QC checks on sample collections, physical
measurements, chemical analyses, data gathering and processing.
Data were subject to verification and validation. Verification
included range checks and internal consistency checks.
Validation consisted of a review of the data from a data users
perspective for consistency based on known numerical
relationships.
All lake sampling and measurements took place the weeks of
7/17 to 7/21, 1995 and 6/21 through 7/5, 1996. Eighty-two
stations were sampled over the two-year period. This annual
sampling window was selected because it is a time of maximum
recreational use, and maximum water supply use.
Secchi disc transparency was measured by R.L. Raschke.
Prior repetitive test measurements of Secchi depth in a variety
of water bodies showed that the coefficient of variation (CV) for
Secchi depth ranged from 5% to 15% among several samplers,
including Raschke. In waters >1.0 meter Secchi depth
transparency, the CV's were from 5 to 10%, but in waters of <1.0
meter, variability in measurements sometimes increased up to 15%.
Water samples were collected as depth integrated samples
throughout the photic zone. Samples were collected for total
phosphorus (TP), total suspended .suspended solids (TSS), algal
14
-------
growth potential tests (AGPT), and chlorophyll a. Field
duplicates were taken at least once in every ten samples.
For chlorophyll a, 100 to 250 mL of sample was filtered
through a 24 mm diameter Whatman GF/F glass fiber filter.
Samples were filtered in triplicate. The filter was folded,
blotted dry, enclosed in aluminum foil, labeled and stored in a
cooler containing dry ice and returned to SESD for analyses.
Chlorophyll samples were extracted in 90% acetone and measured by
visible spectrophotometry and by high performance liquid
chromatography (HPLC). The results given in Table l are from
HPLC analyses as many of the chlorophyll levels were generally
too low to be determined spectrophotometrically. For the field
duplicates, the coefficient of variation
-------
Schultz £±. al. (1994) . The CV ranged from 1.3 to 53.1% with the
average at 19.3%. This variability is similar to values listed
for the method (Miller et al., 1978) .
3.0 FINDINGS
The distribution of data for each variable can be
characterized by its cumulative distribution frequency (cdf).
These curves show the percent of embayment acreage in the basin
equal to or less than some specified measurement plus or minus a
confidence level. For the purpose of this study, we have set a
confidence level of 95%. This means that we are 95% sure that
the sampled acreage estimated to be equal to or less than a given
measurement is within the bounds of our confidence lines on the
graph (Fig. 4). There is a 1 in 20 chance (5% error) that the
true or real percent of acreage affected at a particular
measurement is not within the confidence bounds.
Chlorophyll a ranged from a low of 0.84 ug/L at Lake
Hartwell to 11.56 ug/L at the most downstream lake, Lake Thurmond
(Table 1).
16
-------
Table 1. Range of Values for the Savannah River Lakes, 1995 &
1996.
LAKES
CHL. A
AGPT
LIMITS
TP
SD
TSS
UG/L
MG/L
UG/L
METERS
MG/L
THURMOND
0 . 98-11.56
0.66-11.0
NP-N
2-50
1.2-4.8
2-27
RUSSELL
1. 88-5.47
0 .39-2.01
NP-N
10-60
0.7-2.6
2-32
HARTWELL
0.84-6.84
0.55-2.27
NP-N
3-30
1.7-10
2-6
KEOWEE
0.91-2.03
1.11-5.08
N-N
3-6
2.4-4 . 6
2
JOCASSEE
1.35-2.59
0.68-1.95
NP-N
3-10
3.3-6.0
2-34
BURTON
1.60
1.62
N
6
2.2
2
LU
<3
<
HI
s
o
<
i—
z
LU
<
QQ
5
LU
LU
O
cc
LU
CL
0
This range of concentrations at the times of sampling exhibit
trophic conditions
related to classical
lake classifications
of oligotrophic to
mesotrophic(Olem and
Flock, 1990) .
Chlorophyll a was
less than 12 ug/L
over the entire
basin's large lakes
(Fig. 4 ). Based on
experience (Raschke,
1994) over the past
3 0 years, generally,
when chlorophyll a
ranges from 0 to 10
ug/L, there is no
discoloration of the
water and no
problems. At a range
of 10 to 15 ug/L,
waters can become
discolored and algal scums could develop
the water is deeply discolored
100
80
60
40
20
....
r- —
i / w—
" __T/
"h i" " r- - 1
¦—f-J-J]'
j
-j -
! / III!
I
.
jp Jl
I -
1
0
4 6 8
CHLOROPHYLL A in ug/L
10
12
• CYCLE 1
CYCLE 2
Figure 4. Cdf for Chlorophyll a.
Between 20 to 30 ug/L,
scums are more frequent, and
17
-------
matting of algae can occur (Raschke, 1993). EPA Region 4
(Raschke, 1993) has shown that a mean photic zone growing season
average of equal to or less than 15 ug/L of chlorophyll a should
satisfactorily meet multiple uses, including drinking water
supply.
One of the objectives of the Savannah River REMAP is to
detect trends in important environmental variables over both time
and space. One means of comparison is through the testing of the
null hypothesis that the population's distributions from two or
more annual cycles are identically distributed. This can be
accomplished through use of the Cramer von Mises test statistic
(W) which is founded on design-based methods of statistical
inference. For design-based statistical inference, the source of
random variation is the random selection of sample sites. This
is in contrast to model-based statistical inference, where the
source of random variation is in the assessed statistical model
(e.g., a regression model). Thus, designed-based statistical
inference has the advantage that no model assumptions are
required. The distribution of a population can be characterized
through its cumulative distribution function (cdf). This is
equivalent to testing the null hypothesis that the cdf's are
identical. A test of cdf differences at alpha .05 (Table 2)
using the Cramer-von Mises test statistic (W) showed that three
variables, chlorophyll a, AGPT, and total phosphorus had
significantly different distributions from one cycle to the
other. Chlorophyll cycle 1 and cycle 2 were slightly different
(Wsl.26, k=2). The curve for cycle 1 rises more gradually than
that of cycle 2 (Figure 4) culminating in a high of 11.56 ug/L
thus suggesting the mean is higher for cycle 1.
18
-------
Table 2. Cramer-von Mises Tests for Equality of Cumulative
Distribution Functions for the Savannah River Basin Embayments
Equality of Cumulative Distribution Functions Between Cycles
(Years) is Tested.
VARIABLE
w
CHLOROPHYLL A
1.26*
AGPT
5 . 84*
TOTAL PHOSPHORUS
1. 86*
SECCHI DISC
0.25
TOTAL SUSPENDED SOLIDS
0 .15
~Significant at alpha=.05
100
80
60
40
20
1XJ
o
<
LL)
en
o
<
Chlorophyll a t;
represents 2
phytoplankton <
standing crop or
yield at given time
periods, whereas AGPT
is representative of
Li-
the potential q
phytoplankton 0 2 4 6 8
production, given AGPT in mg/L
optimum conditions of
sufficient nutrients,
1 . • 3 CYCLE 1 CYCLE 2
light, time and
temperature. Algal
growth potential Figure 5. Cdf for AGPT.
ranged from 0.39 mg
dry weight (DW)/L at Lake Russell to 11.0 mg DW/L at Lake
19
-------
Thurmond (Table 1). Approximately 99.7% of the AGPT dry weights
were equal to or less than 5 mg/L (Fig. 5 ), an in-lake action
level that will reasonably assure protection from nuisance algal
blooms and fish kills in southeastern lakes (Raschke and Schultz,
1987) The 5 mg/L of dry weight translates to a potential
chlorophyll standing crop of approximately 57 ug/L of chlorophyll
a based on the following equation:
Loq chl a = 1.15 Log10(DW) +0.95 (Raschke and Schultz,
1987) .
The sampled maximum chlorophyll a of 12 ug/L is much lower than
Che 57 ug/L of chlorophyll a derived from the 5mg DW/L AGPT
action level. Figure 5 depicts the AGPT cdf's for cycles one
and two. The curve
for cycle two rises
more gradually then
that for cycle one
suggesting the mean
AGPT is not only
higher in cycle one,
but also shows
greater variability
within this cycle.
The Cramer-von Mises
test statistic
confirms that the
difference between
the two cycles at the
alpha .05 level is
statistically
significant (W=.5.84;
k=2) .
Fiaure 6. Cdf for Total Phosphorus.
Total phosphorus Fl?u
(TP), another ™^ntial production, ranged from 2 ug/L
indicator like AGPT of ^tent^lpr ^ ^
;n Lake H^weg' 5° Of the embayment acreage was equal to or less
Approximately 87.0% of ^ ^ ^ phosphorus were
than 10 ug/L TP (Fig. 6). ^ values of 40 to 60 ug/L one
available for algal grow , but this was not the case
could expect ^ chlropiyli a values. This is not
as seen by the relative y
20 30 40
TOTAL PHOSPHORUS in ug/L
60
. rvri (= 4
, c 1
20
-------
100
surprising; besides needing optimum conditions for maximum
growth, the phytoplankton need sufficient nutrients that are
bioavailable to them. Generally, not all of the TP in lakes is
available for phytoplankton growth. Peters (1981) estimated that
bioavailable phosphorus (BP) is 83% of TP in natural lakes and 18
to 57% in rivers. Since our lakes are reservoirs and thus an
extension of a river system one would expect bioavailability to
be much less than that found in natural lakes. Previous work on
the 18 Mile Creek embayment of Lake Hartwell showed that the
average percent of BP to TP was 38% (Raschke at al., 1985) .
Sometimes the BP portion of TP can be as low at 3% (Raschke and
Schultz, 1987). At the alpha .05 level there was a significant
difference (W=1.86; k=2) between cycles 1 and 2, but higher
values were observed in cycle 1 (Fig. 6). The significant
differences between
cycles for
chlorophyll, AGPT,
and TP suggests that
other than normal
variables are
influencing
differences from one
cycle to the other.
We are not in a
position with two
years of data to
focus on particular
stress indicators at
this time. Samples
were collected from
two to three weeks
after rainfall
events in the basin.
Thus rainfall or
unusually high
stream flows would
not seemingly cause the differences observed between cycles with
respect to these three phytoplankton growth related indicators
The assumption of non-rainfall effects was confirmed by the non-
significant differences between cycles for Secchi Dept (
(W=o.25; k=2) and total dissolved solids (TSS) (W-0.15; k 2),
indicators of sedimentation effects from rainfall events
Presumably the cyclic differences were caused by internal la e
UJ
o
<
LU
C£
O
<
I-
z
LU
2
>-
<
CO
LU
LU
o
a.
LU
CL
4 6
SD TRANSPARENCY in METERS
• CYCLE 1
CYCLE 2
Figure 7. Cdf for Secchi Disc.
21
-------
influences like internal nutrient cycling. Even these
differences may be within the normal suite of variability
experienced in a natural setting.
For water supply, a mean growing season average Secchi disc
(SD) transparency of equal to or greater than 1.5 meters is
desirable (Raschke, 1993). For non-water supply embayment
situations a mean SD of greater than l meter is acceptable for
fishing and swimming (Raschke, 1993). Secchi disc transparency
ranged from 0.7 meters at Lake Russell to a high of 10 meters at
Lake Hartwell (Table 1). An examination of Figure 7 shows that
about 2.6% of the time less than desirable conditions existed in
embayment waters for recreation purposes and only 5.3% of the
time were they
less than the
water supply
criterium of
equal to or
greater than 1.5
meters. Where SD
was less than one
meter,
measurements were
located near
shore or at the
upper end of the
tributary
embayments.
Buck (1956) TSS in mg/L
divided
impoundments into
3 categories:
clear with total
suspended solids /T
(TSS) less than 25 mg/L; intermediate with TSS 25-100 mg/L; and
muddy with TSS greater than 100 mg/L. The mean harvest o game
fish was 162 lbs/acre for clear lakes, 94 lbs/acre in
intermediate lakes, and muddy lakes only yielded 30 lbs/acre.
The TSS ranged from a low of 2 mg/L at all lake embay^ents to a
high of 34 mg/L at Lake Jocassee, the uppermost lake m t e
a ui my/u a a (-rain these high values were
Savannah Chain of lakes (Table 1). Again tneae y
attributed to near shore stations receiving wind fetoh at the
• CYCLE 1
CYCLE 2
Figure 8. Cdf for Total Suspended Solids
22
-------
time of sampling. Ninety-seven percent of the embayment acreage
would fall into Buck's clean category, with only 3% being
intermediate with respect to water clarity (Fig. 8).
23
-------
4.0 REFERENCES
APHA. 1995. Standard Methods for the Examination of Water and
Wastewater, 19th Edition. American Public Health Association,
Washington, D.C.
Buck, D. H. 1956. Effects of turbidity on fishing. Transactions
of the American Wildlife Conference 21:249-261.
Bromberg, S. M. 1990. Identifying ecological indicators: An
environmental monitoring and assessment program. Journal of the
Air Pollution Control Association 40:976-978.
Callicott, J. B. 1991. Conservation ethics and fishery
management. Fisheries 16:22-28.
Doudoroff, P., and C. E. Warren. 1957. Biological indices of
water pollution with special reference to fish populations. Pages
144-163. In: Biological Problems in Water Pollution. U. S. Public
Health Service, Cincinnati, Ohio.
EPA. 1983. Methods for the Chemical Analysis of Water and Wastes.
EPA-600/4-79-020. Revised March, 1983. U.S. Environmental
Protection Agency, Cincinnati, OH.
EPA. 1991. The Watershed Protection Approach, an Overview. EPA
503/9-92/002. U. S. Environmental Protection Agency, Washington,
D. C.
EPA. 1992. Methods for the Determination of Chemical Substances
in Marine and Estuarine Environmental Samples. U.S. Environmental
Protection Agency, Cincinnati, OH.
EPA. 1996. Watershed Approach Framework. EPA 840/S-96/001. Office
of Water. U. S. Environmental Protection Agency, Washington, D.
C.
Frey, D. G. 1977. Biological integrity of water - A historical
approach. Pages 127-140. In: R. K. Ballentine and L. J. Guarria
(Eds.). The Integrity of Water. U. S. Environmental Protection
24
-------
Agency, Washington, D. c.
FTN Associates, R. L. Raschke, H. Howard. 1994. Plan of Study for
Characterizing Ecological Condition in the Savannah River Basin.
U. S. Environmental Protection Agency, Region 4, Science and
Ecosystem Support Division, Ecological Assessment Branch, Athens,
flonrrri =
Hughes, R. M., and R. F. Noss. 1992. Biological diversity and
biological integrity: Current concerns for lakes and streams.
Fisheries 17:11-19.
Hunsaker, C. T., D.E. Carpenter, and J. Messer. 1990. Ecological
indicators for regional monitoring. Bulletin of the Ecological
Society of America 71<2):165-172.
Hunsaker, C. T., and D. E. Carpenter. 1990. Environmental
Monitoring and Assessment Program - Ecological Indicators. EPA
600/3-90/060. U. S. Environmental Protection Agency, Research
Triangle Park, North Carolina.
Karr, J. R., and D. R. Dudley. 1981. Ecological perspective on
water quality goals. Environmental Management 5:55-68.
Karr, J. R., K. D. Fausch, P. L. Angermeier, P. R. Yant, and I.
J. Schlosser. 1986. Assessing Biological Integrity in Running
Waters: A Method and its Rationale. Publication No. 15. Illinois
Natural History Survey, Champaign, Illinois.
Larsen, D. P., D. J. Stevens Jr., A. R. Selle, and S. G. Paulsen.
1991. Environmental Monitoring and Assessment Program, EMAP
surface waters, a Northeast lakes pilot. Lake and Reservoir
Management 7(l):l-ll.
Larsen D P , and S. J. Christie (Eds.). 1993. EMAP-Surface
' •-i o ~«vt- epa 620/R-93/003 . Office of Research
Waters. 1991 Pilot Report. EPA 6^u/k
and Development, U. S. Environmental Protection Agency,
Washington, D. C.
Management Committee. 1995. Savannah River Watershed Project:
Initial Assessment and L
River Basin. Volumes 1 and 2. Preparea oy i. =
Committee of the Savannah River Basin watershed Project.
25
-------
Messer, J. J., R. A. Linthurst, and W. S. Overton. 1991. An EPA
program for monitoring ecological status and trends.
Environmental Monitoring and Assessment 17:67-78.
Miller, W., J. Green, and T. Shiroyama. 1978. The SfiI en a fit rum
raprir.nrnutum Printz Algal Assay Bottle Test. EPA-600/9-78-018.
Noss, R. F. 1990. Indicators for monitoring biodiversity: A
hierarchical approach. Conservation Biology 4:355-364.
Olem, H. and G. Flock (Eds.). 1990. Lake and Reservoir
Restoration Guidance Manual. EPA 440/4-90/006. Second Edition
Prepared by The North American Lake Management Society for the U.
S. Environmental Protection Agency, Washington, D. C.
Overton, W. S., D. Stevens Jr., D. White 1990. Environmental
Monitoring and Assessment Program - Design Report for EMAP. EPA
600/3-91/053. U. S. Office of Research and Development,
Environmental Protection Agency, Washington, D. C.
Peters, H. 1981. Phosphorus availability in Lake Memphrenagog and
its tributaries. Limnology and Oceanography 26:1150-?
Pister, E. P. 1987. A pilgrims progress from Group A to Group B.
Pages 221-232. In: J. B. Callicott (Ed.). Companion to Sand
County Almanac. University of Wisconsin Press, Madison,
Wisconsin.
Raschke, R. L. 1994. Phytoplankton bloom frequencies in a
population of small southeastern impoundments. Lake and Reservoir
Management 8 (2):205-210.
Raschke, R. L. 1993. Guidelines for Assessing and Predicting
Eutrophication Status of Small Southeastern Piedmont
Impoundments. U. S. Environmental Protection Agency, Region 4,
Science and Ecosystem Support Division, Ecological Assessment
Branch, Athens, Georgia.
Raschke, R. L., and D. A. Schultz. 1987. The use of the algal
growth potential test for data assessment. Journal of the Water
Pollution Control Federation 59(4):222-227.
26
-------
Raschke, R. L., B. Cosgrove, and M. Koenig. 1985. Phosphorus
loading effects upon phytoplankton standing crop in a tributary
embayment of a southern piedmont reservoir. Pages 139-148. In:
Lake and Reservoir Management: Practical Applications. North
American Lake Management Society, Madison, Wisconsin.
Rathbun, S. C., R. Houghton, and C. Laurin. 1996. Cramer-von
Mises Tests for Environmental Monitoring Data. A report for the
U. S. Environmental Protection Agency, Science and Ecosystem
Support Division, Ecosystem Assessment Branch, Athens, Georgia.
Schultz, D., R. Raschke, and R. Jones. 1994. A shortened algal
growth potential test. Environmental Monitoring and Assessment
32:201-205.
Snedecor, G. W., and W. G. Cochran. 1967. Statistical Methods.
Sixth Edition. The Iowa State University Press, Ames, IA.
Stevens, D. L. Jr. 1994. Implementation of a national
environmental monitoring program. Journal of Environmental
Management 42:1-29.
Stevens, D. L. Jr. 1997. Variable density grid-based sampling
designs for continuous spatial populations. Environmetries
8:165-197.
Stevens, D. L. Jr., A. R. Olsen, and.D. White. 1992. Integrated
Sampling Design: Environmental Monitoring and Assessment Program.
EPA 600/xxx/xxx. U. S. Environmental Protection Agency,
Environmental Research Laboratory, Corvallis, Oregon.
Urquhart, M. S., W. S. Overton, and D. S. Birkes. 1993. Comparing
sampling designs for ecological status and trends: Impact of
temporal patterns; In: v. Barnett and K. F. Turkman (Eds.).
Statistics for the Environment. Wiley, New York.
Welch, P. S. 1948. Limnological Methods. McGraw-Hill Book Co.
Inc., New York.
White, D., A. J. Kimerling, and W. S. Overton. 1992. Cartographic
and geometric components of a global sampling design for
environmental monitoring. Cartography and Geographic Information
Systems 19:5-22.
27
-------
APPENDIX A
REMAP LAKE EMBAYMENT SAMPLING POINTS FOR CYCLES 1 & 2
(1995 & 1996)
A-l
-------
H CYCLE 1 1 CYCLE 2
A-2
-------
LAKE RUSSELL SAMPLE POINTS
H CYCLE 1 ffl CYCLE 2
A-3
-------
LAKE HARTWELL SAMPLE POINTS
®JsiXMOi f=e Cftr&
1 CYCLE 1 a CYCLE 2
-------
LAKEKEOWEE SAMPLE POINTS
¦ CYCLE 1 ~ CYCLE 2
A-5
i
-------
LAKE JOCASSEE SAMPLE POINTS
1 CYCLE 1 ~ CYCLE 2
A-6
A
-------
LAKE BURTON SAMPLE POINTS
%
CYCLE 1 ~ CYCLE 2
A-7
-------
APPENDIX B
SAVANNAH REMAP LAKE EMBAYMENTS
LAKE EMBAYMENT STATIONS SAMPLED (1995 & 1996)
B-l
-------
LAKE EMBAYMENT STATIONS SAMPLED, 1995 -1996
CYCLE
STATIONJD
LONGDMS
LATDMS
LAKE
1
1
-82 5424.530929
35 2 9.546320
Jocassee
1
4
-82 51 7.603567
34 53.45.993569
KEOWEE
1
5
-82 51 55.611809
34 52 8.208817
KEOWEE
1
7
-82 54 49.665560
34 47 11.788171
KEOWEE
1
8
-82 5512.267418
34 45 25.877105
KEOWEE
1
&
-82 57 14.545488
34 43 25.328172
KEOWEE
1
10
-82 50 36.042482
34 41 50.982771
HARTWELL
1
11
-82 51 48.938645
34 40 1.427178
HARTWELL
13
-83 8 21.058132
34 34 24.386896
HARTWELL
1
14
-824938.283149
34 33 34.244494
HARTWELL
1
15
-83 620.511658
34 32 28.952918
HARTWELL
1
16
-824919.275926
34 31 56.433897
HARTWELL
1
17
-82 48 59.200669
34 3013.216865
HARTWELL
1
18
-83 6 7.106787
3429 38.128221
HARTWELL
1
19
-83 4 39.463439
34 29 23.852941
HARTWELL
1
20
-82 5828.372960
34 28 47.713879
HARTWELL
1
22
-82 51 49.270112
34 27 9.136690
HARTWELL
1
24
-82 50 19.866007
34 23 54.623175
HARTWELL
1
25
-82 52 31.838827
3423 51.677778
HARTWELL
1
26
-82 50 1.218515
34 2329.129219
HARTWELL
1
27
-82 57 0.566973
3422 0.973040
HARTWELL
1
28
-82 30 32.463918
33 55 29.861465
THURMOND
1
29
-82 2711.522718
3353 34.753482
THURMOND
1
30
-8225 28.144621
335311.871449
THURMOND
1
31
-82 21 18.751953
33 5t 34.332004
THURMOND
1
32
-82 23 4.224512
33 49 30.599430
THURMOND
1
34
-82 1526.752841
33 43 49.010464
THURMOND
1
35
-82 18 7.370846
3343 3.904412
THURMOND
1
36
-82 19 16.008013
3341 41.487729
THURMOND
1
38
-82 15 45.262833
33 40 6.970435
THURMOND
1
39
-82 25 10.454159
33 39 42.915041
THURMOND
1
40
-$2 27 54.152732
3338 57.791282
THURMOND
1
41
-62 32 46.441481
33 38 3.485909
THURMOND
1
42
-82 2944.619895
33 37 50.333626
THURMOND
1
43
-82 45 17.522595
3411 21.827095
RUSSELL
1
44
-82 42 52.970639
34 8 53.666517
RUSSELL
1
45
-82 37 41.619648
34 8 57.582352
RUSSELL
1
46
-82 37 25.496575
34 8 18.125663
RUSSELL
1
48
-82 40 7.213973
34 646.747787
RUSSELL
1
49
-82 40 12.298408
34 5 59.541942
RUSSELL
1
50
-82 38 19.522020
34 5 49.958644
RUSSELL
1
51
-82 36 35.699957
34 5 6.921357
RUSSELL
1
52
-82 35 13.386742
34 2 45.605296
RUSSELL
2
53
-82 5511.608256
35 1 53.581238
Jocassee
2
55
-82 56 45.543830
34 5945.144400
Inmrnrtii
2
56
-82 58 0.827130
34 58 43.138440
Jocassee
2
57
-83 32 25.841760
34 52 58.890653
BURTON
2
59
-82 5411.676408
344955.612154
KEOWEE
2
60
-82 55 9.878470
34 48 7.103444
KEOWEE
2
61
-82 53 3.040507
344? 13.030299
KEOWEE
2
62
-82 56 1.924517
3446 25.690855
KEOWEE
2
65
-82 58 11.317878
344326.490824
KEOWEE
2
66
-82 56 19.640120
3443 2.932961
KEOWEE
-------
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
LAKE EMBAYMENT STATIONS SAMPLED, 1995 -1996
67
-82 58 3.955804
34 42 30.456779
KEOWEE
68
-83 14 49.616801
34 37 41.653244
HARTWELL
70
-83 8 8.974833
34 34 17.717526
HARTWELL
73
-82 47 36.995458
34 31 23.215728
HARTWELL
74
-82 50 54.886218
34 29 57.003787
HARTWELL
75
-82 48 12.315729
34 29 37.361079
HARTWELL
77
-82 57 0.235490
34 28 56.887496
HARTWELL
78
-82 52 4.677041
34 28 34.434387
HARTWELL
79
-82 49 52.508573
342728.982305
HARTWELL
80
-82 50 3.004343
34 25 23.234650
HARTWELL
81
-82 52 37.963618
34 24 19.921871
HARTWELL
84
-82 51 50.090858
3422 45.024973
HARTWELL
87
-82 3526.567783
33 5921.234816
THURMOND
88
-82 23 10.567304
33 55 20.967239
THURMOND
89
-82 22 36.206081
33 53 1.813908
THURMOND
91
-82 22 41.289075
334942.901412
THURMOND
92
-82 25 29.984397
33 48 55.466103
THURMOND
93
-82 16 11.134728
33 47 8.846842
THURMOND
95
-82 13 41.447226
33 45 55.135352
THURMOND
96
-82 20 2.757306
33 43 30.367269
THURMOND
97
-82 14 52.721702
33 4311.368723
THURMOND
98
-82 13 35.323466
33 43 0.845442
THURMOND
99
-82 17 52.510994
3342 31.398682
THURMOND
100
-82 23 49.731993
334218.312925
THURMOND
101
-82 1540.402407
3341 43.027050
THURMOND
103
-8223 4.194887
33404&64797S
THURMOND
105
-824438.684495
341232.272620
RUSSELL
109
-82 37 53.290017
34 7 53.604337
RUSSELL
110
-62 38 2.135389
34 5 15.291914
RUSSELL
111
-82 3620.648669
34 1 41.621509
RUSSELL
-------
6-1
-------
APPENDIX J
Peer Review Comments and the Response
-------
R*vi«w«r 1
COTWSPt; You need to highlight the findings in a way that is
easy to read and understand.
T we disagree. We reviewed our outline and the logical
sequence of findings etc. Other reviewers found the
report easy to follow and understand.
Comment; The descriptions through section 3 were excessive and
sounded like project justification.
Response; We believe it is important to detail the approach and
sampling strategy for the purpose of "fully" informing
our primary client, the state monitoring and 305b
coordinators, who generally are unfamiliar with the
probability-based sampling design approach.
Comment; There are too many wholesale citations of SOPs which
make the document difficult to stand on its own merits.
ResponseWe disagree. We only referred to SOPs in conjunction
with field and laboratory efforts. The document is
more sound when SOPs are used because the methods are
based on a wealth of experience and quality control
checks.
Comment; Lake Embayments - It would be helpful to estimate what
you found and follow that with a discussion and
interpretation section. It is hard to follow as
written, but it might be a start on a scientific
publication.
Response: We have no intent of publishing the lake study results
in a scientific publication. We don't think the
results will add anything new to science.
Comment; a lot of the figures were hard to read or missing or
needed to be redrawn.
Responset Good Point. The figures were enlarged and put at the
end of their respective sections.
Comment You need somewhere near the front to say what all the
appendices deal with so that there is some
understanding of the bulk.
Response: The titles in the table of contents and at the
beginning of each appendix sufficiently describes the
contents.
-------
Comment: Important figures and tables in the appendices need to
be pulled up into the main body of the report so the
reader can get the message much more concisely and see
what is being presented.
Response: We agree. Some tables were summarized and brought
forward.
Comment: I would like to see if elevation or stream order plots
of the data show the same trends as ecoregions. I am
not convinced due to the disproportionate sample site
distribution in the Lower Piedmont that this is the
best way to parse the data. Other analytical
approaches don't appear to have been explored.
Response: Our original intent was to examine the Basin as a whole
(see appendices H and I). The "trend spatially" in the
report is very subjective and based on few data points
in some ecoregions. We noticed that there seems to be
a "trend," but a new sampling design and strategy would
have to be used to confirm our observations.
Ecoregions provide a necessary spatial framework for
monitoring ecological resources. Ecoregions represent
areas of relative homogeneity. The 1991 Science
Advisory Board's evaluation of the ecoregion concept
said, " that the ecoregions not only provide a valuable
framework for monitoring and assessment, but also
provide a geographic context for defining biological
criteria. Stream order and/or elevations could
encompass several ecoregions.
Rtvimrer 2
Comment: The only substantive comment relates to recommendations
for future studies. Add some more data for some of the
ecoregions.
Response: We agree and there will be an opportunity in the summer
of 1991 when SESD initiates the Regional REMAP study.
Comment: Add major streams to figure 1.2.
Response: We disagree. It would clutter up the figure which is
intended to show the lakes that potentially could be
sampled under our large lake criteria. The description
in the text is sufficient.
Comment: In one place of section 5.1.1, the authors say 15 ug/L
of chlorophyll A is satisfactory, but they imply that
57 ug/L of chlorophyll A when it is derived from 5 mg
-------
dryweight/L of AGP.
Pgsppnsg: The 15 ug/L is a growing season average based on
intensive sampling of small lakes in Georgia, South
Carolina, and North Carolina. The 57 ug/L is
instantaneous and based on standing crop potential
under optimum conditions. Since it is potential
growth, a higher number derived in a laboratory setting
is appropriate to initiate further investigation into a
potential problem.
R«vi«w«r 3
Comment: Overall, I think you have done an outstanding job
summarizing the methods and results. The LPEI looks
like a reasonable way to holistically portray the
ecological information. I also like the way you
answered the initial questions/objectives at the end.
Response: None.
Comment: When possible, future statistical studies should be
designed to incorporate sufficient sites in each
ecoregion to allow inferences to be drawn for each of
he ecoregions of interest.
Response: The EMAP is designed to address ecoregion sampling. We
focused at the basin scale because ecoregional sampling
would have required more sampling and time.
Additionally, ecoregions in the basin were not well
defined at the beginning of sampling. The states of
South Carolina and Georgia are in the midst of defining
ecoregion boundaries and determining reference sampling
sites. We and the states are in agreement with respect
to the Lower Piedmont Ecoregion Boundaries.
Comment: Identification of reference areas may include
subjectively selected sites if least impacted areas are
under represented by the statistical sample in an
ecoregion.
Response; We agree.
Comment: Further investigation of landscape/instream
relationships is encouraged to build on the
correlations documented here. Development of such
relationships has considerable potential as a screening
tool to identify potentially impaired sites.
-------
Response: We agree. We plan to look at these relationships in an
upcoming regional REMAP survey of wadeable streams.
Rtvimtz 4
Comment: Related to clarification and better sentence structure.
Response: Agreed with comments and expanded some sections to
better explain findings.
Comment: I have concerns about the development of the LPEI and
its use of the LPEI on the same data set used to
develop it. Usually an index or criterion is developed
on a reference set of data collected across he entire
range of the target population and then applied to
independent data. This data set only represents a part
of the Lower Piedmont Ecoregion, and it may not capture
the total range of any of the component metrics. It is
truly only a Savannah Basin Lower Piedmont Index.
Response: We agree. We had not looked at the entire range
(across the Lower Piedmont Ecoregion) for the
individual metrics used. We only focused on the
Savannah Basin. We corrected the LPEI in the text to
SB-LPEI (Savannah Basin-Lower Piedmont Ecological
Index). We will have an opportunity to test the
index's power across many ecoregions within the
Regional REMAP study beginning in the summer of 1999.
Comment: I think the appendix about locating probability sites
on maps and in the field, and obtaining access
permission will be very useful to us. That is exactly
the stage we are at in establishing our probability
network.
Response: We agree and think it is state of the art.
Comment: We have had a workshop on integration of judgement data
with probability data and adequately answered state
concerns. At that workshop, we were presented with
some theoretical approaches for integrating data, but
weren't given any procedures to use. The workshop
addressed state concerns, but it didn't provide us with
tools to accomplish integration. It did help
illustrate the beneficial uses of probability-based
designs in answering 305b and other resource-wide
condition questions, and demonstrated the limitations
of judgement-based designs in addressing those same
questions. I think you have overstated the
accomplishments of that workshop.
-------
» The statement concerning the workshop was changed to
reflect the reviewer's viewpoint. The follow up report
in Appendix G addresses the question of merging
judgement and probability data more fully.
Comment; The three-project lake system is authorized and
operated...by the Corps of Engineers... for fish...etc.
You mentioned who operates the other lakes, but failed
to mention the COE on these major lakes.
Response: Correction noted and made by authors.
Raviaw 6
Comment: We recognize the potential usefulness of probability
sampling in our river basin sampling rotation and
statewide monitoring..
Response: None
Comment: We are concerned that the results of the present report
will prove difficult to fit into our 305b/303d listing
process. That is, the "good," "fair," and "poor"
evaluations may not provide a good fit with the 305b
categories of support, partially support, and not
support., For example, will fair mean partially
support? The real concern is that we will probably
have to take these results and fit them into 305b even
though that has not been the primary purpose of the
study.
Response: We agree that the primary purpose of the study was to
demonstrate the feasibility of using the EMAP
probability sampling approach for monitoring purposes.
We believe the information gathered is amenable for
inclusion into a 305b report and will work with the
state on this concern.
------- |