oEPA
United States
Environmental Protection
Agency
Comparison of Random and
Systematic Site Selection for
Assessing Attainment of Aquatic
Life Uses in Segments of the Ohio
River
RESEARCH AND DEVELOPMENT
-------
-------
EPA 600/R-06/089
September 2006
www.epa.gov
Comparison of Random and Systematic
Site Selection for Assessing Attainment
of Aquatic Life Uses in Segments of the
Ohio River
Karen Blocksom1, Erich Emery2, and Jeff Thomas2
1U.S. Environmental Protection Agency
Ecological Exposure Research Division
National Exposure Research Laboratory
Cincinnati, OH 45268
20hio River Valley Water Sanitation Commission
5735 Kellogg Avenue
Cincinnati, OH 45228
Notice: Although this work was reviewed by EPA and approved for publication, it may not necessarily
reflect official Agency policy. Mention of trade names and commercial products does not
constitute endorsement or recommendation for use.
U.S. Environmental Protection Agency
Office of Research and Development
Washington, DC 20460
-------
NOTICE
The research described in this document has been funded in part by the U.S.
Environmental Protection Agency. This report has been subjected to the
Agency's peer and administrative review process and has been approved for
publication as an EPA document.
Mention of trade names or commercial products does not constitute endorsement
or recommendation for use.
-------
EXECUTIVE SUMMARY
Introduction
The Clean Water Act (CWA) section 305(b) requires that states report biennially
on the water quality standards (WQS) attainment status of all waters. The United
States Environmental Protection Agency (USEPA) has recognized that biological
assessment data are very valuable and should serve as a core indicator in
determining aquatic life use attainment status (USEPA 2002). However, great
(or large-floodplain) rivers, such as the Ohio River, are one type of aquatic
resource for which biological assessment data typically have been deficient,
primarily due to difficulty in sampling (Emery et al. 2003). The Ohio River Valley
Water Sanitation Commission (ORSANCO), a compact of eight states and the
federal government is responsible for assessment of the Ohio River.
Recognizing the abundance of stresses on this system and the lack of biological
data for the river, ORSANCO developed a bioassessment program several years
ago. In 1990, ORSANCO began a Long-Term Intensive Survey (LTIS) of the
Ohio River to provide high sample density (one sample every 3.2-6.4 km) from
selected reach segments of the river. Beginning with this effort, a fish-based
index of biotic integrity (IBI) for the Ohio River was developed (Emery et al.
2003). Although the LTIS approach provided valuable data for the development
of a fish IBI and biocriteria, this approach was very resource-intensive and time-
consuming and was not feasible for routine sampling of the entire river. To
support the development of a cost-effective program for routine monitoring, the
minimal amount of effort required to produce an adequate assessment needed to
be determined.
Methods
A probability-based sampling design, in which sites are randomly selected from
the population of possible sites, is an effective way to reduce effort and still
collect data representative of an entire resource. By simulating a probability
design using the intensive survey data already collected, the number of sites
required to adequately represent the condition of a pool was determined. Two
approaches were used to analyze the simulation data, and both assume that the
intensive survey data provide the most accurate representation of condition
available. The first approach was to determine the minimum number of sites that
provided a similar condition estimate to that of the full set of sites, indicating
some stability in the estimate. This was achieved by calculating variables for
various subsets of sites and identifying the number of sites at which distributions
of variables no longer differed from those of the full set of sites. Simultaneously,
the number of sites required to obtain an estimate of condition with a specific
level of precision was determined. A second and more commonly used
approach to determining the adequacy of sampling was based on the relative
proportion of species collected. The number of 500 m reaches required to collect
-------
80-90% of the observed taxa in the pool was calculated as a way to determine
the number of sites at which a sufficient level of effort has been expended to
estimate biological condition. By examining the results of both approaches for
this study, a probability sampling design was developed that will still be rigorous
and will provide known confidence around any estimate of condition for Ohio
River navigational pools while reducing the level of effort (i.e., number of sites
sampled) in the field.
Results and Recommendations
This research indicated that 15 sites (compared to 20-32 sites sampled in the
LTIS) may be adequate to draw conclusions about the overall condition of a
navigational pool. However, in some cases, additional sites may need to be
sampled to achieve the desired level of precision around the condition
assessment. Because there are a large number of pools in the Ohio River, an
approach that allows ORSANCO to sample and assess more pools each year
will result in a more robust assessment of the river for the 305(b) report. A
suggested approach is to sample all of the navigational pools of the Ohio River
over a 5-year period. In each pool, an initial sampling of 15 sites is carried out,
and additional sampling is completed only if required to make a definitive
assessment of the pool. This approach would limit the resources required to
assess an individual pool, such that additional effort would only be required in
those pools that are of more marginal condition. In addition, this approach will
help ORSANCO to identify those pools in which biological condition is most
impacted and to prioritize any mitigation or restoration efforts. Additional
sampling of individual sites may be required to determine causes of impairment
within pools but may be guided by the data acquired through the random
sampling design.
IV
-------
TABLE OF CONTENTS
NOTICE ii
EXECUTIVE SUMMARY iii
LIST OF TABLES vi
LIST OF FIGURES vii
APPENDIX FIGURES viii
ACKNOWLEDGMENTS x
1. INTRODUCTION 1
2. METHODS 3
2.1. Study Area 3
2.2. Sampling Protocols 4
2.2.1. Fish 4
2.2.1.1. Electrofishing 4
2.2.1.2. Fish Sample Processing 5
2.2.2. Habitat 5
2.3. Data Analysis 7
2.3.1. IBI and Metrics 7
2.3.1.1. Simulations 8
2.3.1.2. Condition Measures 9
2.3.2. Species Richness 10
3. RESULTS 11
3.1. IBI and Metrics 11
3.2. Species Richness 20
4. DISCUSSION 22
5. RECOMMENDATIONS 23
6. LITERATURE CITED 24
APPENDIX A 27
-------
LIST OF TABLES
Table 1. The five pools used for this study, along with sampling information
associated with the LTIS and selected background information 3
Table 2. Selected potential influences on water quality (from ORSANCO
1994) in study pools 4
Table 3. The metrics included in the ORFIn and their expected response to
stress 8
Table 4. Frequency of samples in each habitat class by pool 8
Table 5. Metrics used in analyses for each pool are denoted by an (X) for
that metric. Metrics were excluded if they had a maximum of less than 5 for
richness metrics and DELT anomalies and less than 10 for percentage
metrics 12
Table 6. Number of occurrences in 500 simulations for which the K-S test of
differences in distributions was significant (p < 0.20), shown for the ORFIn
scores and metrics by pool and number of sites. If a metric was not
evaluated fora particular pool (see Table 3), (NA) is listed for that cell 18
VI
-------
LIST OF FIGURES
Figure 1. Locations of intensive survey pools used in this study (highlighted
in red) along the Ohio River 6
Figure 2. Cumulative distribution functions with 95% confidence bounds for
ORFIn scores in R.C. Byrd pool based on bootstrapped subsets of sites (red
lines) and the full set of sites (black lines) 13
Figure 3. Cumulative distribution functions with 95% confidence bounds for
ORFIn scores in Hannibal pool based on bootstrapped subsets of sites (red
lines) and the full set of sites (black lines) 14
Figure 4. Cumulative distribution functions with 95% confidence bounds for
ORFIn scores in McAlpine pool based on bootstrapped subsets of sites (red
lines) and the full set of sites (black lines) 15
Figure 5. Cumulative distribution functions with 95% confidence bounds for
ORFIn scores in Newburgh pool based on bootstrapped subsets of sites (red
lines) and the full set of sites (black lines) 16
Figure 6. Cumulative distribution functions with 95% confidence bounds for
ORFIn scores in Smithland pool based on bootstrapped subsets of sites (red
lines) and the full set of sites (black lines) 17
Figure 7. Confidence interval (Cl) lengths for estimates of proportion of river
km failing in each pool as a function of the number of sites sampled. The
horizontal dotted line represents the desired maximum Cl length of 0.25, and
the vertical solid lines represent the approximate number of sites for which
the desired Cl length is achieved 19
Figure 8. Estimates of the proportion of river km failing with 90% confidence
bounds, based on bootstrap randomizations. The dotted line represents the
threshold delineating failure of the entire pool (0.25) 20
Figure 9 . Mean bootstrap species richness as a function of sample size,
based on 1000 bootstrap randomizations of sites. The horizontal dotted lines
represent 80% and 90% of the maximum number of species and the vertical
lines represent the numbers of sites at which these are attained 21
VII
-------
APPENDIX FIGURES
Figure A-1. Cumulative distribution functions for (a) species richness, (b)
sucker species richness, (c) centrarchid species richness, and (d) % simple
lithophils in R.C. Byrd pool for random subsets and the full set of sites 28
Figure A-2. Cumulative distribution functions for (a) % detritivores, (b) %
invertivores, (c) % piscivores, and (d) CPUE in R.C. Byrd pool for random
subsets and the full set of sites 29
Figure A-3. Cumulative distribution functions for (a) species richness, (b)
sucker species richness, (c) intolerant species richness, and (d) % tolerant
individuals in Hannibal pool for random subsets and the full set of sites 30
Figure A- 4. Cumulative distribution functions for (a) % simple lithophils, (b)
% non-native individuals, (c) % detritivores, and (d) % invertivores in
Hannibal pool for random subsets and the full set of sites 31
Figure A-5. Cumulative distribution functions for (a) % piscivores, (b) DELT
anomalies, and (c) CPUE in Hannibal pool for random subsets and the full
set of sites 32
Figure A-6. Cumulative distribution functions for (a) species richness, and
(b) sucker species richness in McAlpine pool for random subsets and the full
set of sites 33
Figure A-7. Cumulative distribution functions for (a) centrarchid species
richness, and (b) intolerant species richness in McAlpine pool for random
subsets and the full set of sites 34
Figure A-8. Cumulative distribution functions for (a) % simple lithophils, and
(b) % detritivores in McAlpine pool for random subsets and the full set of
sites 35
Figure A-9. Cumulative distribution functions for (a) % invertivores, and (b)
% piscivores in McAlpine pool for random subsets and the full set of sites 36
Figure A-10. Cumulative distribution functions for (a) DELT anomalies, and
(b) CPUE in McAlpine pool for random subsets and the full set of sites 37
Figure A-11. Cumulative distribution functions for (a) species richness, (b)
sucker species richness, (c) centrarchid species richness, and (d) % simple
lithophils in Newburgh pool for random subsets and the full set of sites 38
VIM
-------
Figure A-12. Cumulative distribution functions for (a) % non-natives, (b) %
detritivores, (c) % invertivores, and (d) % piscivores in Newburgh pool for
random subsets and the full set of sites 39
Figure A-13. Cumulative distribution functions for CPUE in Newburgh pool
for random subsets and the full set of sites 40
Figure A-14. Cumulative distribution functions for (a) species richness, and
(b) centrarchid species richness in Smithland pool for random subsets and
the full set of sites 41
Figure A-15. Cumulative distribution functions for (a) intolerant species
richness, and (b) % simple lithophils in Smithland pool for random subsets
and the full set of sites 42
Figure A-16. Cumulative distribution functions for (a) % detritivores, and (b)
% invertivores in Smithland pool for random subsets and the full set of sites 43
Figure A-17. Cumulative distribution functions for (a) % piscivores, and (b)
DELT anomalies in Smithland pool for random subsets and the full set of
sites 44
Figure A-18. Cumulative distribution functions for CPUE in Smithland pool
for random subsets and the full set of sites 45
IX
-------
ACKNOWLEDGMENTS
We thank Mark Barath, U.S. EPA Region III in Philadelphia, PA, and Tony
Olsen and Dave Bolgrien, U.S. EPA Office of Research and Development in
Corvallis, OR, and Duluth, MN, respectively, for comments on an earlier
draft of this report. The data used in this research could not have been
collected without the tireless efforts and skill of ORSANCO field crews.
x
-------
1. INTRODUCTION
The Clean Water Act (CWA) section 305(b) requires that states report biennially
on the water quality standards (WQS) attainment status of all waters. Currently,
an Integrated Report is prepared to fulfill requirements related to sections 303(d),
305(b), and 314 of the Clean Water Act. The United States Environmental
Protection Agency (USEPA) has recognized that biological assessment data are
very valuable and should serve as a core indicator in determining aquatic life use
attainment status (USEPA 2002). Biological data can reflect overall ecological
condition (i.e., chemical, physical, and biological), and they provide a spatially
and temporally integrated measure of the aggregate effect of stressors and
prevailing environmental conditions. USEPA suggests that biological criteria, or
biocriteria, in WQS are most effective as numerical values, which set ranges of
indicators representing acceptable conditions for attainment of the Aquatic Life
Use designation of a water body (USEPA 2002). In practice, however, state
WQS often include only narrative biological criteria, although numeric thresholds
of biological indicators are often used to interpret the narrative criteria. In any
case, the ability to report on the condition of all waters over which a state has
jurisdiction is often limited because of inadequate assessment of certain types of
water bodies.
Great (or large-floodplain) rivers are one type of resource for which biological
assessment data typically have been deficient, primarily due to difficulty in
sampling (Emery et al. 2003). However, the data that are available suggest that,
despite reductions in chemical and organic pollution, the biological condition of
running waters of the United States, including large and great rivers, has
continued to decline since the inception of the Clean Water Act (Karr and Chu
1999). Great rivers are distinctive in that they are few in number but comprise a
large and highly visible component of lotic resources in terms of volume. They
are disproportionately degraded by many human actions that "pollute" them,
including water withdrawals, overharvesting offish, impoundments, and changes
in the landscape that can affect natural processes (Emery et al. 2003, Karr and
Chu 1999). Recognizing such stresses on the system and the lack of biological
data, a bioassessment program was developed for the Ohio River. Assessment
of the Ohio River is the responsibility of the Ohio River Valley Water Sanitation
Commission (ORSANCO), a compact of eight states and the federal government.
In 1990, ORSANCO began a Long-Term Intensive Survey (LTIS) of the Ohio
River to provide high sample density (one sample every 3.2-6.4 km) from
selected reach segments of the river. From these 741 fish assemblage samples,
efforts began in 1991 to develop numeric biocriteria for the Ohio River (Simon
and Emery 1995). A fish-based index of biotic integrity (IBI) that incorporates
river kilometer (rkm) has since been developed for the Ohio River (Emery et al.
2003). Recently, biocriteria have been refined to account for habitat type and
sampling date as well.
-------
Although the intensive survey approach provided valuable data for the
development of a fish IBI and biocriteria, this approach is very resource- and
time-consuming and is not feasible for sampling the entire river on a regular
basis. To support the development of a cost-effective program for routine
monitoring, the minimal amount of effort required to produce an adequate
assessment needed to be determined. A probability-based sampling design, in
which sites are randomly selected from the population of possible sites, is an
effective way to reduce effort and still collect data representative of an entire
resource. By simulating a probability design using the intensive survey data
already collected, the number of sites required to adequately represent the
condition of a pool can be determined. Although the LTIS samples do not
represent a complete sampling of a pool because they only consist of a 500 m
reach every few km, the systematic nature of the sampling effort allows for the
assumption that the data adequately characterize actual distributions in a given
pool. Two approaches were used to analyze the simulation data, and both
assume that the intensive survey data provide the most accurate representation
of condition available.
The first approach was to determine the minimum number of sites that provided a
similar condition estimate to that of the full set of sites, indicating some stability in
the estimate. This was achieved by calculating variables for various subsets of
sites and identifying the number of sites at which distributions of variables no
longer differed from those of the full set of sites. Simultaneously, the number of
sites required to obtain an estimate of condition with a specific level of precision
was determined.
A second and more commonly used approach to determining the adequacy of
sampling was based on the relative proportion of species collected. Previous
studies have examined the level of effort necessary to produce a representative
sample based on the proportion of the available species captured by that method
in wadeable streams (Dauwalter and Pert 2003, Reynolds et al. 2003, Lyons
1992) and nonwadeable rivers (Meador 2005, Hughes et al. 2002, Lyons et al.
2001). These studies have been focused on determination of the appropriate
continuous sampling distance at a single location, which may represent an entire
stream or river. However, ORSANCO already had set the electrofishing distance
at 500 m, based on work by Simon and Sanders (1999), who determined that
500 m was a sufficient distance to characterize biological integrity. For this
study, the number of 500 m reaches required to collect 80-90% of the observed
taxa in the pool was calculated as a way to determine the number of sites at
which a sufficient level of effort has been expended to estimate biological
condition. By examining the results of both approaches for this study, a
probability sampling design was developed that will still be rigorous and will
provide known confidence around any estimate of condition for Ohio River
navigational pools while reducing the level of effort (i.e., number of sites
sampled) exerted in the field.
-------
2. METHODS
2.1. Study Area
The Ohio River begins in Pittsburgh, Pennsylvania, at the confluence of the
Monongahela and Allegheny rivers (rkm 0) and flows southwesterly for
approximately 1579 km through six states to the confluence with the Mississippi
River (Figure 1). Currently, there are 18 high-lift and 2 low-head dams on the
Ohio River, each providing a minimum of 2.75 m depth for commercial
navigation. These dams define major pools on the river and significantly limit fish
populations to a specific pool. For this reason, and because any watershed
management actions would likely be carried out at this scale, the pool was
viewed as the appropriate level of assessment in the Ohio River for this study. In
addition, probability designs are not intended for assessment of individual sites
but for larger areas. Thus, the pool is the most well-defined unit of assessment
using a probability sampling design in the Ohio River.
From the LTIS dataset, data were selected from five pools, each sampled in one
of four different years. Table 1 provides information on the pools and the
samples collected in them, and Table 2 characterizes factors that may influence
water quality in these pools. These five pools were selected because they
represent the most intensive sampling, with a sample frequency of more than
one site every 4 km (2.5 mi). Variations in condition were expected both across
pools and among years, although the goal of this study was to identify patterns
with sample size and not to specifically evaluate the condition of individual pools.
Table 1. The five pools used for this study, along with sampling information
associated with the LTIS and selected background information.
Pool
R.C. Byrd
Hannibal
McAlpine
Newburgh
Smithland
Year of
sampling
2002
1996
1997
1994
1996
Dam location
(river km)
449.3
203.4
976.5
1249.0
1478.2
Length
(km)
67.1
58.3
118.3
89.2
116.7
Average Number of sites
width (m) sampled
352
345
622
755
1255
24
20
32
25
30
-------
Table 2. Selected potential influences on water quality (from ORSANCO 1994) in
study pools.
Pool
R.C. Byrd
Hannibal
Tributaries
(drainage >200 mi2
(518km2))
2
2
No.
permitted
discharges
30
52
General influences
Intersected by heavily industrialized
Kanawha River
Many small-moderate sized towns,
McAlpine
Newburgh
Smithland
2
2
industrial sites and associated barge
traffic
44 Includes city of Louisville, Kentucky
53 Smaller industry, small towns, one
moderate-sized city, minimal barge
traffic or heavy industry
10 Intersected by agriculture-influenced
Wabash River, few towns, no heavy
industry
2.2. Sampling Protocols
2.2.1. Fish
2.2.1.1. Electrofishing
All sampling was conducted during the low-flow, stable conditions of July through
October and during water conditions meeting sampling criteria (i.e., minimum
secchi depths of 38 cm; water levels within 61 cm of normal-flat-pool).
Procedures for electrofishing followed that described by Emery et al. (2003). At
each site, a 500 m reach was electrofished with a 5.5 m jon boat outfitted with an
onboard generator. Electrofishing was conducted at night, as this is the
established protocol used by ORSANCO and has been documented to provide
for a more representative sample of the resident fauna in deeper rivers when
compared to day electrofishing (Sanders 1992). The onboard generator supplied
AC power to 150-W floodlights on the bow of the boat, and to a Smith-Root Type
VI-A alternator-pulsator used to convert the AC generator output to DC and then
regulate the output for electrofishing. A single stainless steel ball suspended
from a bow-mounted retractable aluminum boom served as the anode, with the
aluminum boat hull serving as the cathode.
Each site was electrofished proceeding downstream along the shoreline at a
speed equal to, or slightly greater than the prevailing current velocity. The
electrofishing time at each site generally ranged from 1800 to 5000 seconds
depending on the current velocity, available cover, and the number offish
encountered. Efforts were made to capture every fish sighted by the crew.
-------
2.2.1.2. Fish Sample Processing
Upon capture, fish were placed in an aerated, recirculating on-board live well for
processing. Particular care was used in handling species of special concern.
The majority of captured fish were identified to species, examined for external
anomalies, weighed, measured for total length, and released in the field. Those
requiring laboratory identification were preserved in buffered 10% formalin and
later identified using regional ichthyological keys (e.g., Fishes of Ohio (Trautman
1981), Fishes of Missouri (Pflieger 1997), and Fishes of Tennessee (Etnierand
Starnes 1993)). Fish measuring less than 20 mm in length (e.g., larval fish) were
not recorded as they are difficult to identify accurately and offer data of
questionable value to an assemblage assessment (Angermeier and Karr 1986).
The occurrence of external DELT (deformities, eroded fins and body parts,
lesions, and tumors) anomalies was recorded following procedures outlined by
Ohio EPA (1989) and refined by Sanders et al. (1999). The frequency of DELT
anomalies has been shown to be a good indication of stress caused by chronic
agents, intermittent stresses, and chemically contaminated sediments. As a
result, it is a commonly used metric for assessment of rivers throughout the
United States (Emery et al. 2003).
2.2.2. Habitat
Substrate information used in data analysis was collected in 2000 for sites
sampled prior to that year, with the assumption that the basic characteristics of
the habitat at a site have not changed over time. For sites sampled in 2000 and
beyond, habitat was sampled within 4 months of fish sampling (i.e., during the
fish sampling index period). At each site, the shoreline of each 500 m sampling
zone was divided into five 100 m segments, creating 6 points of reference for the
zone along the shoreline (i.e., Om, 100 m, 200 m, 300 m, 400 m, and 500 m). At
each interval, a 6 m copper pole was used to characterize the substrate at 11
points. The first measurement was taken at the shoreline with subsequent
measurements at 3 m intervals towards mid-channel (total distance = 30 m).
This resulted in a total of 66 point measurements within each 500 m fishing zone.
Substrate was recorded as boulder, cobble, gravel, sand, fines, hardpan, or as a
combination of these substrate types and used to estimate the percentage of
each sediment-type within the 500 m sample area. Habitat data for some sites
was unavailable.
-------
Figure 1. Locations of intensive survey pools used in this study (highlighted in red) along the Ohio River.
-------
2.3. Data Analysis
There were three main objectives of the analysis. The first was to use a
multimetric index offish assemblage condition and the raw values of its
component metrics to compare the intensive survey sampling design with
subsets of those sites. The second was to compare pool-level condition
estimates and precision using all of the sampling sites and subsets of sites. For
this portion of the analysis, the condition estimate represented the proportion of
sites in a pool for which the multimetric index score failed to meet a minimum
threshold value. The final objective was to determine the number of sites in each
pool at which 80% and 90% offish species were captured, as an additional way
to gauge representativeness of sampling in the pool. For all three objectives,
only the sample from the first visit to each site was used in analyses.
2.3.1. IBI and Metrics
Analyses for this study involved the use of an existing multimetric index
specifically developed for the Ohio River fish assemblage. The Ohio River Fish
Index (ORFIn) was originally developed using LTIS data and consists of 13
metrics describing various characteristics of the fish assemblage (Emery et al.
2003) (Table 3). As part of this process, least disturbed (reference) sites were
identified and defined as being at least 1 km upstream or downstream from
navigational dams, at least 1.61 km downstream from any point source
discharge, and at least 500 m from the mouth of any tributary (Emery et al.
2003).
Subsequent to the ORFIn's development, three habitat-based classes were
developed that rely on substrate composition (ORSANCO, unpublished data):
(A) cobble >= 14%; (B) cobble < 14% and sand < 70%; and (C) cobble <14% and
sand >= 70%. Within each of these habitat classes, the 25th percentile ORFIn
score among least disturbed sites was set as the minimum score required to be
considered passing (unimpaired). This threshold for determining impairment was
selected following the logic of Yoder and Rankin (1995), because the reference
condition represents only least disturbed and not truly pristine conditions. The
three habitat-specific thresholds for ORFIn scores were quite different from one
another. For habitat A, this threshold was 39, and for habitat B, it was 33 (of a
possible score of 65). For habitat C, considered to be sand flats, the condition
assessment was determined to be dependent on sampling date, with higher
ORFIn scores obtained later in the sampling index period. Thus, the threshold
was adjusted for Julian day of the sample collection based on a 25th percentile
regression using a quantile regression method (Koenker and Bassett 1978). This
adjustment resulted in the following formula for the minimum adjusted score
required to "Pass": (0.12*Julian Day - 2.4) (unpublished data).
-------
For this study, each site was evaluated with regard to impairment thresholds
using the ORFIn. First, ORFIn metrics and index scores were calculated for
each sample. Then, each site was classified into habitat A, B, or C using the
criteria above. Each site was assessed as passing or failing (impaired) based on
the ORFIn score, habitat classification, and Julian day (when relevant). For sites
lacking habitat data, ORFIn scores were compared to thresholds for all 3 habitat
types. If the score exceeded the minimum required to pass for all habitat types,
the site condition was designated as 'PASS'. If the score fell below the threshold
for all 3 habitats, its condition was designated as 'FAIL'. Those sites with mixed
results were assessed as 'UNKNOWN' condition.
Table 3. The metrics included in the ORFIn and their expected response to stress.
.... Expected response
Metnc to stress
Number of species Decrease
Number of sucker species Decrease
Number of centrarchid species Decrease
Number of great-river species Decrease
Number of intolerant species Decrease
% Tolerant individuals Increase
% Simple lithophilic individuals Decrease
% Non-native individuals Increase
% Detritivore individuals Increase
% Invertivore individuals Decrease
% Piscivore individuals Decrease
Number of DELT anomalies Increase
Catch per unit effort (CPUE) Decrease
Table 4. Frequency of samples in each habitat class by pool.
Pool
Byrd
Hannibal
McAlpine
Newburgh
Smithland
A (cobble)
4
3
8
0
3
B (mixed)
19
7
13
10
12
C (sand)
1
2
5
13
6
Unavailable
0
8
6
2
9
Total
24
20
32
25
30
2.3.1.1. Simulations
Bootstrap methods were used to simulate the random selection of sites within
pools. A bootstrap approach randomly draws sites with replacement from the
original dataset and assumes that the set of sites in this original dataset
adequately reflects the distribution of conditions in the pool. Sampling with
replacement means that during each random draw, each site has an equal
probability of being selected. For example, when a subset of five sites is
selected with replacement, it is possible for any particular site to be drawn all five
times. This methodology is appropriate for these data because, although pools
were sampled intensively for LTIS, they were not sampled in their entirety. Thus,
8
-------
although the systematic nature of the sampling likely resulted in a set of samples
representative of the distribution of conditions in the pool, all of the possible
sample locations in the pool were not specifically included in the dataset. By
creating a large number of sample sets and subsets (of sites) using
bootstrapping, almost any population parameter and its variance can be
estimated robustly from the original dataset (Chernick 1999).
In this analysis, the general process was as follows for each pool. A number of
sites equal to that in the full set of LTIS sites (e.g., 24 sites in R.C. Byrd pool)
were selected with replacement from the set of LTIS sites (original set) to create
a bootstrap set of sites (full set). The first five of those were combined and used
to estimate certain characteristics related to condition. Then the next five were
added to the first five and the characteristics re-estimated for that set of ten. This
process was repeated with five additional sites added each time until the full set
of bootstrap sites was included. The full set of bootstrap sites was treated during
each simulation as the true measure of condition in the pool, and the
characteristics for each subset of sites was compared to this full set. The entire
process was repeated 500 times to obtain 95% confidence intervals for each
statistic at each sample size.
2.3.1.2. Condition Measures
The cumulative distribution function (CDF) of a variable indicates the probability
that any particular observation of that variable will be at or below a specified
value (Sokal and Rohlf 1995). By comparing CDFs between subsets of sites and
the full set of bootstrap sites, the minimum sample size at which the distributions
strongly overlap and provide very similar information can be determined. For
each pool, empirical cumulative distribution functions (CDFs) were calculated for
each metric and the ORFIn based on various subsets of sites and for the full set
of bootstrap sites. For example, for R.C. Byrd pool, there were 24 sites in the full
set of data, and CDFs were generated using 5, 10, 15, 20, and 24 sites for each
of the 500 simulation runs. Calculation of the empirical CDF consisted of
determining the following percentiles for each run on each set of sites: 0, 1, 5,
10, 25, 50, 75, 90, 95, 99, and 100. Then, the 2.5th, 50th, and 97.5th quantiles
were determined for each percentile across all 500 runs. These values provide
the median CDF and its 95% confidence intervals (Cl) for plotting. For each
metric or the ORFIN scores and each subset size, the median CDF and its 95%
Cl was plotted, along with the CDF for the full set of bootstrap sites for
comparison. Visual comparisons of CDFs and confidence intervals were used to
evaluate the degree of similarity between the full set of sites and subsets of
different sizes.
In addition to line plots of CDFs, distributions of metrics or the ORFIn were
directly compared between subsets and the full set of sites using a
nonparametric Kolmogorov-Smirnov (K-S) two-sample test. The test was
performed using PROC NPAR1WAY in SAS with Monte Carlo estimation of
-------
exact p-values (v. 9.1, SAS Institute, Gary, NC). The K-S test assumes that the
two samples being compared are independent of one another, but this
assumption was not met because one sample was simply a subset of the other.
This would tend to lead to an actual Type I error rate much smaller than the
nominal value. To improve the ability to detect differences between distributions
(power) and to achieve an actual Type I error rate closer to 0.05, a p-value of
0.20 or less was used to identify significant differences between distributions,
rather than the preferred significance level of 0.05. The number of runs out of
500 in which the distributions differed significantly (p < 0.20) was recorded for
each metric and sample size.
The habitat-specific ORFIn thresholds described previously were used to
evaluate each sample as passing or failing, and the pool-level condition was
determined as the proportion of samples (river km) failing. For each run, an
estimate of the proportion of river km failing (Pfei|) in the pool was calculated, and
over the 500 runs, 95% confidence intervals were estimated for that proportion.
This was done for each of the subsets of sites, and the full set of sites from the
intensive survey was viewed as providing the 'true' assessment of the pool. For
these estimates of condition, the desired precision was within 12.5 percentage
points, or a total confidence interval length of 25 percentage points. A proportion
of 0.25 of river km failing has been set tentatively as the threshold for assessing
an entire pool as failing (for placement on the 303(d) list of impaired waters).
The selection of this criterion value stemmed from the decision to set the sample-
level threshold for failing at the 25th percentile of the reference distribution.
Based on this threshold, it would be possible for up to 25% of sites within a pool
to be considered failing, even within a pool made up entirely of reference sites
(as defined by ORSANCO). Whether this variation among reference sites is due
to natural factors or actual differences in disturbance level, this approach would
tend to limit only the Type I error rate (i.e., incorrectly assessing a pool as failing)
when assessing a pool and not the Type II error rate (incorrectly assessing a
pool as passing). As a way to limit both Type I and Type II errors, a definitive
assessment of a pool is attained only if the confidence interval for the proportion
failing does not include the threshold. This means that proportions either above
or below the threshold by a small amount may not provide a sufficient
assessment of the pool as a whole.
2.3.2. Species Richness
To estimate the number of samples required to obtain approximately 80% or 90%
of the total species collected within a pool, Estimates software was used to
model the species richness as a function of the number of samples (Colwell
2005). For each simulation run in Estimates, a sample was selected from the full
set of samples with replacement. The bias-adjusted bootstrap estimate of
species richness (Sboot) was calculated based on Smith and van Belle (1984) as:
10
-------
Sobs
+ Z (1 ~ Pk )
where S0/5S is the number of species observed in the pooled samples, pk is the
proportion of samples containing species k, and m is the total number of
samples. From this calculation, it can be shown that the value of Sboot is
maximized when each species occurs in only one sample, and Sboot equals S0bs
when each species occurs in all samples from a pool. After calculating Sboot,
another sample was selected with replacement and combined with the first
sample, and the species richness was again estimated. This process continued
until a number of samples equivalent to the original set had been selected and
combined, with species richness estimated each time using the bias-adjusted
bootstrap approach. The entire procedure was repeated 1000 times, and the
average value and standard deviation across runs was determined. From this
information, the minimum sample size (i.e., number of sites) required to collect
an average of 80% or 90% of observed taxa within a pool was determined.
3. RESULTS
3.1. IBI and Metrics
Those component metrics with very small ranges tend to reflect more rare
components of the fish population and contribute very little variation to the overall
ORFIn score. Thus, metrics with limited ranges of values were excluded from
plotting and comparing distributions (Table 5). Richness metrics with a maximum
of 5 or greater and percentage metrics with a maximum of 10 or greater were
included in analyses. The DELT anomalies metric was used only if the maximum
value was 5 or greater. As a result of these criteria, one or more metrics was
excluded from each analysis, and the metric for great-river species richness was
excluded from all analyses. The ORFIN score was used in analyses for all pools.
The CDFs of ORFIn scores for subsets of pools were usually very similar to
those for the full set of bootstrap sites (Figures 2-6). The confidence bounds for
the CDFs narrowed predictably with increasing numbers of sites relative to the
maximum number of sites sampled within a pool. Within a sample size of 15
sites for Byrd, Hannibal, and Newburgh pools and 20 sites for McAlpine and
Smithland pools, the confidence bounds were closely aligned between the subset
and the full set of sites. Plots of CDFs of metrics can be found in Appendix A.
11
-------
Table 5. Metrics used in analyses for each pool are denoted by an (X) for that
metric. Metrics were excluded if they had a maximum of less than 5 for richness
metrics and PELT anomalies and less than 10 for percentage metrics.
Metric
Number of species
Number of sucker
species
Number of centrarchid
species
Number of great-river
species
Number of intolerant
species
% Tolerant individuals
% Simple lithophilic
individuals
% Non-native individuals
% Detritivore individuals
% Invertivore individuals
% Piscivore individuals
Number of DELT
anomalies
Catch per unit effort
(CPUE)
R.C. Byrd
Pool
X
X
X
X
X
X
X
X
Hannibal
Pool
X
X
X
X
X
X
X
X
X
X
McAlpine
Pool
X
X
X
X
X
X
X
X
X
Newburgh
Pool
X
X
X
X
X
X
X
X
X
Smithland
Pool
X
X
X
X
X
X
X
X
12
-------
§
!_-
CD
Q_
O
100"
80-
60-
40-
20-
80-
60-
40-
20-
5 sites
15 sites
10 sites
Subset 95% Cl
~~ Subset median
Full set 95% Cl
Full set mediarr
20 sites
10 20 30 40
50 0 10 20 30 40
ORFIn Score
50 60
Figure 2. Cumulative distribution functions with 95% confidence bounds for
ORFIn scores in R.C. Byrd pool based on bootstrapped subsets of sites (red lines)
and the full set of sites (black lines, N=24).
13
-------
c
CD
^
CD
Q_
CD
E
D
o
100
80-
60-
40-
20-
10 sites
- - Subset 95% Cl
~~ Subset median
- ~ Full set 95% Cl
Full set median
10 20 30 40 50
10 20 30 40 50 60
ORFIn Score
Figure 3. Cumulative distribution functions with 95% confidence bounds for
ORFIn scores in Hannibal pool based on bootstrapped subsets of sites (red lines)
and the full set of sites (black lines, N=20).
14
-------
CD
Q_
CD
E
D
o
100
80-
60-
40-
20-
80-
60-
40-
20-
80-
60-
40-
20-
5 sites
15 sites
25 sites
t
10 sites
20 sites
30 sites
~ Subset 95% Cl
— Subset median
Full set 95% Cl
Full set median
0 10 20 30 40 50 0 10 20 30 40 50 60
ORFIn Score
Figure 4. Cumulative distribution functions with 95% confidence bounds for
ORFIn scores in McAlpine pool based on bootstrapped subsets of sites (red lines)
and the full set of sites (black lines, N=32).
15
-------
CD
O
CD
Q_
CD
E
D
O
100'
80-
60-
40-
20-
80-
60-
40-
20-
5 sites
15 sites
10 sites
Subset 95% Cl y
~~ Subset median /, /./
-- Full set 95% Cl / ' /'/
Full set median, / //,
20 sites
10 20 30 40 50
10 20 30 40 50 60
ORFIn Score
Figure 5. Cumulative distribution functions with 95% confidence bounds for
ORFIn scores in Newburgh pool based on bootstrapped subsets of sites (red
lines) and the full set of sites (black lines, N=25).
16
-------
100
I
CD
Q_
CD
E
D
o
80-
60-
40-
20-
25 sites
10 20 30 40 50 60
- Subset 95% Cl
~~~ Subset median
- - Full set 95% Cl
Full set median
10 20 30 40
50 60
ORFIn Score
Figure 6. Cumulative distribution functions with 95% confidence bounds for
ORFIn scores in Smithland pool based on bootstrapped subsets of sites (red
lines) and the full set of sites (black lines, N=30).
17
-------
Table 6. Number of occurrences in 500 simulations for which the K-S test of differences in distributions was significant (p <
0.20), shown for the ORFIn scores and metrics by pool and number of sites. If a metric was not evaluated for a particular
pool (see Table 3), (NA) is listed for that cell.
Pool
"E
m
d
of
"co
.a
'c
CO
I
CD
|5.
"5
5
.c
E
.a
g
CD
Z
•a
_co
.c
W
No.
sites
5
10
15
20
5
10
15
5
10
15
20
25
30
5
10
15
20
5
10
15
20
25
ORFIn
2Q
7
0
0
8
3
0
30
16
5
0
0
0
26
11
1
0
44
9
0
0
0
Species
30
3
0
0
13
0
0
47
24
0
0
0
0
28
4
0
0
36
3
2
0
0
Suckers
31
5
1
0
18
0
0
39
21
3
0
0
0
25
8
0
0
NA
NA
NA
NA
NA
Centra rch.
27
7
0
0
NA
NA
NA
44
27
6
0
0
0
20
9
0
0
38
7
1
0
0
Grt
River
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
Intol.
NA
NA
NA
NA
13
1
0
45
16
6
0
0
0
NA
NA
NA
NA
28
12
4
0
0
Toler.
NA
NA
NA
NA
13
0
0
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
Simple
Lith
24
2
0
0
8
0
0
37
17
5
0
0
0
25
5
0
0
33
11
0
0
0
% Non-
native
NA
NA
NA
NA
15
1
0
31
20
9
1
0
0
21
10
0
0
NA
NA
NA
NA
NA
Detr.
30
5
0
0
13
2
0
32
16
1
0
0
0
29
4
0
0
49
12
1
0
0
Invert.
26
4
0
0
9
1
0
50
13
2
0
0
0
25
3
0
0
28
7
1
0
0
Pise.
29
3
0
0
11
0
0
46
14
4
0
0
0
27
4
0
0
27
13
3
0
0
DELT
NA
NA
NA
NA
0
0
0
0
0
0
0
0
0
NA
NA
NA
NA
0
0
0
0
0
CPUE
32
3
0
0
14
0
0
48
10
4
0
0
0
23
5
0
0
43
12
0
0
0
18
-------
The K-S tests of differences between distributions of subsets of sites and the full
set of sites for each pool were rarely significant, typically only for subsets of 5 or
10 sites (TableS).
Confidence intervals around Pfei| varied with the number of sites in the subset, but
for 15 or more sites, the estimate itself was usually very close to that for the full
set (Figures 7 and 8). The confidence interval length around Pfei| tended to
decrease with increasing sample size, but this pattern was not entirely
consistent. Approximately 25 sites were required for Smithland and Newburgh
pools and 20 sites for Hannibal pool to reach the desired 90% Cl length of 25
points (0.25). This target Cl length was never reached for the other two pools,
even when the Cl was based on bootstrapping with the full number of sites
(Figure 7). Only in McAlpine pool did the confidence bounds not include the 0.25
threshold for at least some sample sizes.
R.C. Byrd
Hannibal
McAlpine
Newburgh
Smithland
10 20 30
Number of sites
40
Figure 7. Confidence interval (Cl) lengths for estimates of proportion of river km
failing in each pool as a function of the number of sites sampled. The horizontal
dotted line represents the desired maximum Cl length of 0.25, and the vertical
solid lines represent the approximate number of sites for which the desired Cl
length is achieved.
19
-------
o
L
o
Q.
O
1.0:
0.8:
0.6:
0.4:
0.2:
0.0
R.C. Byrd pool
0.8
0.6:
0.4:
0.2
0.0
0.8
0.6:
0.4
0.2
0.0
10 15 20
McAlpine pool
25
5 10 15 20 25 30 35
Smithland pool
1.0:
0.8:
0.6:
0.4:
0.2:
0.0
Hannibal pool
4 6 8 10 12 14 16 18 20 22
Newburgh pool
0.8
0.6:
0.4:
0.2
0.0
10 15 20 25 30
~*~~ 90% confidence bounds
~^~ Median
Condition threshold
5 10 15 20 25 30 35
Number of sites
Figure 8. Estimates with 90% confidence bounds of the proportion of river km
failing, based on bootstrap randomizations. The dotted line represents the
threshold delineating failure of the entire pool (0.25).
3.2. Species Richness
On average, 80% of the total observed species richness in a pool were collected
within 10 sites and 90% of species within approximately 15 sites (Figure 9). The
two longest pools, McAlpine and Smithland, tended to require more samples than
shorter pools to reach these benchmarks. Variability around species richness
estimates only decreased slightly with increasing sample size.
20
-------
35r
30
25
20
15
10
OT
o
CD
'o
CD
w
Q.
50 r
30
20
R.C. Byrd
, i , , , , i , , , , i
Hannibal
50r
40
30
20
10 15 20 25 30 0
50r
40
30
20
10 15 20 25
McAlpine
, i .... i .... i ....... .1
10
Newburgh
OT
o 10
o 0 5 10 15 20 25 30 35 05 10 15 20 25 30
60r
0 5 10 15 20 25 30 35
Number of sites
Figure 9 . Mean bootstrap species richness as a function of sample size, based
on 1000 bootstrap randomizations of sites. The horizontal dotted lines represent
80% and 90% of the maximum number of species and the vertical lines represent
the numbers of sites at which these are attained.
21
-------
4. DISCUSSION
The use of random sampling is clearly advantageous for assessing the Ohio
River. By using random sampling, many fewer samples per pool would be
required for an adequate assessment, compared to the number of samples
collected using the non-random intensive survey design. This reduced effort per
pool would reduce the resources required to adequately assess a pool and could
allow for more pools to be sampled each year. In addition, a random design
brings with it certain statistical properties that result in known confidence levels
around estimates of condition (Diaz-Ramos et al. 1996). The results of this study
show that a random subset of samples may be able to represent condition in the
pool adequately. The CDFs of both ORFIn scores and component metrics were
very similar for all but the smallest subset sizes of 5 and 10 sites. The
confidence intervals around the CDFs did vary with sample size, as expected,
but generally were not excessively large even for sample sizes of only 10. The
analysis of species richness also showed that 90% of species observed pool-
wide could be captured within 10-15 samples. In addition, estimates of Pfei|
generally were very close to the estimate based on the full set for 15 sites or
more.
In contrast, the confidence intervals around the estimates of Pfaii for a given pool
were generally larger than desired, sometimes even for the full set of sites.
Under ideal circumstances for drawing sound conclusions about a pool's
condition, the 90% confidence interval of the condition estimate would not include
the threshold of 0.25. In such a situation, the size of the Cl would be irrelevant,
and one could assess the pool definitively as impaired or unimpaired with some
known level of confidence. Based on the Pfaii estimated from the original set of
data, which can be viewed as the best representation of the true condition in the
pools, only two of the five pools (i.e., Byrd and McAlpine) would be considered
failing. However, the closer the actual Pfei| is to the threshold, the more likely it is
that the Cl will include the threshold. For example, in R.C. Byrd pool, with a Pfaii
of 0.20, even the Cl based on the full set of 24 sites included the threshold of
0.25. Thus, in cases with an estimate close to the threshold, more samples
would be required to make definitive statements about the condition of the pool.
On the other hand, in cases where Pfei| was either very large or very small, a
relatively small number of sites was sufficient to avoid inclusion of the 0.25
threshold in the Cl (e.g., McAlpine pool, 10 sites).
Although some of the analyses may lead to different conclusions, the patterns
that emerge from the data generally are consistent across pools, regardless of
the level of variability within a pool. The five pools included in this study varied in
potential impacts to water quality and in habitat diversity, but patterns in species
richness and in variability associated with the ORFIn were similar across or
seemingly unrelated to these differences. Habitat classified as a mixture of sand
and cobble (habitat B) was common in all five pools, but the presence of cobble
and sand habitats varied across pools (Table 4). For example, Newburgh pool
22
-------
had no cobble (habitat A) and mostly sandy (habitat C) sites, whereas Byrd and
McAlpine pools tended to have more cobble than sand sites. Smithland had
more sandy than cobble sites, and Hannibal was approximately evenly divided
between cobble and sandy sites. The influences in each pool that might affect
water quality (e.g., contribution by tributaries, discharges) also differed across
pools (Table 2), with varying levels of industry and sizes of towns along the
banks of each pool. Still, these differences do not seem to directly affect the
patterns seen.
Overall, the ability of ORSANCO to report on the biological condition of the Ohio
River should improve with the use of a random sampling design. Although the
typical recommended minimum sample size for this type of design is usually
around 50 sites (U.S. EPA Aquatic Resources Monitoring web site,
http://www.epa.gov/nheerl/arm/surdesignfaqs.htm), a much smaller number of
samples has the potential to provide enough information for assessment in some
pools of the river. This study indicated that 15-20 sites may be adequate to draw
conclusions about the overall condition of a navigational pool. However, the
sufficiency of this smaller number of sites depends on the variation in condition of
the fish assemblage within a given pool. The more consistent water and habitat
quality are throughout a pool, the fewer samples that are needed to characterize
the biological condition of that pool. Because there are a large number of pools
in the Ohio River, an approach that allows ORSANCO to sample and assess
more pools each year will result in a more robust assessment of the river for the
Integrated Report.
5. RECOMMENDATIONS
In order to be consistent with surrounding states and maximize the data available
for the Integrated Report, a five-year rotational sampling approach is highly
desirable, as data up to 5 years old may be used in the report. An ideal
approach would allow ORSANCO to sample all navigational pools over a 5-year
period and provide an adequate assessment for each pool. Tentatively,
biocriteria require less than 25% of river km in a pool be considered failing in
order to consider the pool in attainment of its Aquatic Life Use designation. A
requirement designed to ensure that an adequate assessment has been
performed is that the confidence interval around the estimate of PfaN does not
include the threshold of 0.25. By sampling only 15 randomly selected sites per
pool in a single year, there is potential for an adequate and conclusive
assessment of a pool. If, after 15 sites, a definitive assessment of the pool
cannot be made, an additional 15 sites from the same design would be sampled
the following year and combined with the first 15 sites. This may shift the Pfaii but
will also reduce the length of the Cl so that it may no longer include the
threshold. If the Cl still includes the threshold, an additional 15 sites should be
sampled the third year and combined with the first 30 sites. If the Cl still includes
the threshold at this point, the estimate of condition based on 45 samples could
23
-------
be used to assess the pool as above or below the 0.25 threshold, regardless of
the Cl. The result would be a sample size close to the recommended sample
size of 50 and should lead to desirable Cl lengths. This approach would limit the
resources required to assess an individual pool, such that additional effort would
be necessary only in those pools that are of more marginal condition. In addition,
this approach would help ORSANCO to rapidly identify those pools in which
biological condition is most impacted and to prioritize any mitigation or restoration
efforts. To ensure that an adequate number of sites would be available to meet
potential sampling needs in a given pool, an initial random draw of 60-80 sites
per pool is appropriate. This approach should provide enough oversample to
account for non-target, inaccessible, or otherwise unsampleable sites and still
obtain up to 45 samples for assessment of condition. Finally, additional sampling
of individual sites may be required to determine causes of impairment within
pools but may be guided by the data acquired through the random sampling
design.
6. LITERATURE CITED
Angermier, P.L. and J.R. Karr. 1986. Applying an index of biotic integrity based
on stream-fish communities: considerations in sampling and interpretation.
North American Journal of Fisheries Management 6: 418-427.
Chernick, M.R. 1999. Bootstrap Methods: A Practitioner's Guide. John Wiley
and Sons, Inc., New York, New York. 264 pp.
Colwell, R.K. 2005. Estimates: Statistical estimation of species richness and
shared species from samples. Version 7.5. User's Guide and
applications published at http://purl.oclc.org/estimates.
Dauwalter, D.C. and E.J. Pert. 2003. Electrofishing Effort and Fish Species
Richness and Relative Abundance in Ozark Highland Streams of
Arkansas. North American Journal of Fisheries Management. 23: 1152-
166.
Diaz-Ramos, S., D.L. Stevens, Jr., and A.R. Olsen. 1996. EMAP Statistical
Methods Manual. EPA/620/R-96/002. U.S. Environmental Protection
Agency, Office of Research and Development, National Health and
Environmental Effects Research Laboratory, Corvallis, Oregon.
Emery, E.B., T.P. Simon, F.H. McCormick, P.L. Angermeier, J.E. DeShon, C.O.
Yoder, R.E. Sanders, W.D. Pearson, G.D. Hickman, R.J. Reash, and J.A.
Thomas. 2003. Development of a Multimetric Index for Assessing the
Biological Condition of the Ohio River. Transactions of the American
Fisheries Society. 132: 791-808.
24
-------
Etnier, D.A. and W.C. Starnes. 1993. The Fishes of Tennessee. The University
of Tennessee Press. Knoxville. 681 pp.
Hughes, R.M., P.R. Kaufmann, AT. Herlihy, S.S. Intelmann, S.C. Corbett, M.C.
Arbogast, and R.C. Hjort. 2002. Electrofishing distance needed to
estimate fish species richness in raftable Oregon rivers. North American
Journal of Fisheries Management 22:1229-1240.
Karr, J.R., and E.W. Chu. 1999. Restoring Life in Running Waters: Better
Biological Monitoring. Island Press, Covelo, California. 206 pp.
Koenker, R. and G. Bassett. 1978. Regression Quantiles. Econometrica.
46(1): 33-50.
Lyons, J. 1992. The length of stream to sample with a towed electrofishing unit
when fish species richness is estimated. North American Journal of
Fisheries Management 12:198-203.
Lyons, J., R.R. Piette, and K.W. Niermeyer. 2001. Development, validation, and
application of a fish-based index of biotic integrity for Wisconsin's large
rivers. Transactions of the American Fisheries Society 130:1077-1094.
Meador, M.R. 2005. Single-pass versus two-pass boat electrofishing for
characterizing river fish assemblages: species richness estimates and
sampling distance. Transactions of the American Fisheries Society
134:59-67.
Ohio EPA. 1989. Addendum to: Biological Criteria for the Protection of Aquatic
Life: Volume II: Users Manual for Biological Field Assessment of Ohio
Surface Waters. Ohio EPA. Columbus, OH.
ORSANCO. 1994. Ohio River Water Quality Fact Book. Ohio River Valley
Water Sanitation Commission, Cincinnati, Ohio.
Pflieger, W.L. 1997. The Fishes of Missouri. Missouri Department of
Conservation. Jefferson City, MO.
Reynolds L., AT. Herlihy, P.R. Kaufmann, S.V. Gregory, and R.M. Hughes.
2003. Electrofishing Effort Requirements for Assessing Species Richness
and Biotic Integrity in Western Oregon Streams. North American Journal
of Fisheries Management. 23: 450-461.
Sanders, R.E. 1992. Day Versus Night Electrofishing Catches from Near-Shore
Waters of the Ohio and Muskingum Rivers. Ohio Journal of Science.
92(3): 51-59.
Sanders, R.E., R.J. Miltner, C.O. Yoder, and ET. Rankin. 1999. The Use of
External Deformities, Erosions, Lesions, and Tumors (DELT Anomalies) in
25
-------
Fish Assemblages for Characterizing Aquatic Resources: A Case Study of
Seven Ohio Streams. Pages 225-248 in T.P. Simon, editor. Assessing
the Sustainability and Biological Integrity of Water Resources using Fish
Communities. CRC Press, Boca Raton, Florida.
Simon, T.P. and E.B. Emery. 1995. Modification and Assessment of an Index of
Biotic Integrity to Quantify Water Resource Quality in Great Rivers.
Regulated Rivers: Research and Management. 11: 283-298.
Simon, T.P. and R.E. Sanders. 1999. Applying an Index of Biotic Integrity
Based on Great River Fish Communities: Considerations in Sampling and
Interpretation. Pages 475-506 in T.P. Simon, editor. Assessing the
Sustainability and Biological Integrity of Water Resources using Fish
Communities. CRC Press, Boca Raton, Florida.
Smith, E.P., and G. van Belle. 1984. Nonparametric estimation of species
richness. Biometrics 40:119-129.
Sokal, R.R., and F.J. Rohlf. 1995. Biometry. W.H. Freeman and Company,
New York. 887pp.
Trautman, M.B. 1981. The Fishes of Ohio. Ohio State University Press. 782
pp.
U.S. Environmental Protection Agency (USEPA). 2002. Consolidated
Assessment and Listing Methodology: Toward a Compendium of Best
Practices, 1st edition. U.S. Environmental Protection Agency, Office of
Water, Washington, D.C. http://www.epa.gov/owow/monitoring/calm.html.
Yoder, C.O. and E.T. Rankin. 1995. Biological criteria program development
and implementation in Ohio. Pages 109-144 in W.S. Davis and T.P.
Simon, editors. Biological Assessment and Criteria. Lewis Publishers,
Boca Raton, Florida.
26
-------
APPENDIX A
Plots of CDFs and 95% Cl of ORFIn metrics for subsets of sites (red lines),
with the CDF and 95% Cl for the full set of sites (black lines).
27
-------
E
O
-------
-------
o
o
100
80
60
40
20
0-
80-
60-
40-
20-
0
5 sites
15 sites
10 sites
5 10 15 20 25
10 15 20 25
Species richness
6 8 10
Intolerant species
IUU
80
60
40-
20
o-
80-
60
40-
20-
n
5 sites /Jyl
/\/l\ J
ff >>
f
1 k
/ 1
I 1
1
1 \l
(J A
Y /
\ i
15 sites
>
/
//
^
rf[
' i
'»
f
f
1 0 sites
/
/ I
II
\J I
YJ //
Y \
> J
//
1
1
24681
- - Subset 95% Cl
~ Subset median
- ~ Full set 95% Cl
Full set median
4 6 8 10
Sucker species richness
I/ I
5 sites
15 sites
10 sites
10 15 20 25
10 15 20 25
% Tolerant individuals
Figure A-3. Cumulative distribution functions for (a) species richness, (b) sucker species richness, (c) intolerant species richness, and
(d) % tolerant individuals in Hannibal pool for random subsets and the full set of sites.
30
-------
JO
IS
O
O
5 sites
15 sites
10 sites
10 15 20 25
- Subset 95% Cl
"" Subset median
- ~ Full set 95% Cl
Full set median
20 40 60 80 100
% Simple lithophils
10 15 20 25
% Non-native
10 15 20 25
% Detritivores
20 40 60 80 100
% Invertivores
Figure A- 4. Cumulative distribution functions for (a) % simple lithophils, (b) % non-native individuals, (c) % detritivores, and (d) %
invertivores in Hannibal pool for random subsets and the full set of sites.
31
-------
10 20 30 40 50
% Piscivores
-------
100-
80
60-
40-
20
o-
§ 80-
| 60^
|j 40
E
5 20
o
80
60
40
20
0
25 sites
20 sites
30 sites
5 10 15 20 0 5 10 15 20 25
Species richness
100
80
60
40
20
0-
80-
60-
40-
20-
0
80
60-I
40
20-I
0
15 sitegr
25 sites
/-TV
/
10 site^
20 sites
30 sites
/
- Subset 95% Cl
^~ Subset median
Full set 95% Cl
Full set median
01 234561 2345
Sucker species richness
Figure A-6. Cumulative distribution functions for (a) species richness, and (b) sucker species richness in McAlpine pool for random
subsets and the full set of sites.
33
-------
is
|
o
100
80
60
40
20
0
80
60
40
20J
0
80
60
40
20J
0
Ssit^e
25 sites
20 sites
30 sites
0
! 4 6 0 2 4 6 8
Centrarchid species richness
Figure A-7. Cumulative distribution functions for (a) centrarchid
for random subsets and the full set of sites.
- - Subset 95% Cl
~~ Subset median
- - Full set 95% Cl
Full set median
02460246
Intolerant species
species richness, and (b) intolerant species richness in McAlpine pool
34
-------
100'
60 0 20 40
6 Simple lithophils
60 0 20
f> Detritivores
40 60
Figure A-8. Cumulative distribution functions for (a) % simple lithophils, and (b) % detritivores in McAlpine pool for random subsets
and the full set of sites.
35
-------
-------
100
80
60
40
20
0
80
60
40
20
0
80
60
40
20
0
5 sites
15 sites
25 sites
10 sites
z
20 sites
30 sites
0 2 4 6 8 10 12 0 2 4 6 8 10 12 14
DELT anomalies
- Subset 95% Cl
Subset median
Full set 95% Cl
Full set median
100 200 300 0 100 200 300 400
CPUE
Figure A-10. Cumulative distribution functions for (a) DELT anomalies, and (b) CPUE in McAlpine pool for random subsets and the full
set of sites.
37
-------
J5
O
-------
JO
IS
O
-------
'*=
JO
3
E
^
O
Subset 95% Cl
Subset median
- ~ Full set 95% Cl
Full set median
CPUE
Figure A-13. Cumulative distribution functions for CPUE in Newburgh pool for random subsets and the full set of sites.
40
-------
10 15 20 25
Species richness
4 6 8 10
Centrarchid species richness
- Subset 95% Cl
Subset median
Full set 95% Cl
Full set median
Figure A-14. Cumulative distribution functions for (a) species richness, and (b) centrarchid species richness in Smithland pool for
random subsets and the full set of sites.
41
-------
-------
10
15 20
% Detritivores
100
80
60
40'
20
0-
80-
60-
40-
20-
0
80-
60-
40-
20-
0
5 sites
15 sites
25 sites
10 sites
20 sites
- Subset 95% Cl
"" Subset median
- - Full set 95% Cl
Full set median
20 40 60 80
20 40
60 80
% Invertivores
Figure A-16. Cumulative distribution functions for (a) % detritivores, and (b) % invertivores in Smithland pool for random subsets and
the full set of sites.
43
-------
-------
E
O
100
80'
60'
40'
20-
0
80'
60'
40'
20"
0"
80'
60'
40'
20-
0
5 sites
15 sites
25 sites
10 sites
20 sites
- Subset 95% Cl
Subset median
Full set 95% Cl
Full set median
100 200 300 400
0 100 200 300 400
CPUE
Figure A-18. Cumulative distribution functions for CPUE in Smithland pool for random subsets and the full set of sites.
45
-------
vvEPA
United States
Environmental Protection
Agency
Office of Research
and Development (8101R)
Washington, DC 20460
Official Business
Penalty for Private Use
$300
EPA 600/R-06/089
September 2006
www.epa.gov
Please make all necessary changes on the below label,
detach or copy, and return to the address in the upper
left-hand corner.
If you do not wish to receive these reports CHECK HERE
D; detach, or copy this cover, and return to the address in
the upper left-hand corner.
PRESORTED STANDARD
POSTAGE & FEES PAID
EPA
PERMIT No. G-35
Recycled/Recyclable
Printed with vegetable-based ink on
paper that contains a minimum of
50% post-consumer fiber content
processed chlorine free
------- |