20051457 Ver 1_General Correspondence_20060606

DO I : 10.1007/s00267-005-0169-3
Fish Assemblage Responses to Water Withdrawals
and Water Suppl y Reservoirs in Pied mont Streams
MARY C. FREEMAN'
US Geological Survey
Patuxent Wildlife Research Center
University of Georgia
Athens, Georgia 30602, USA
PAULA A. MARCINEK
Institute of Ecology
University of Georgia
Athens, Georgia 30602, USA
ABSTRACT /Understanding effects of flow alteration on
stream biota is essential to developing ecologically sus-
tainable water supply strategies. We evaluated effects of
altering flows via surface water withdrawals and instream
reservoirs on stream fish assemblages, and compared ef-
fects with other hypothesized drivers of species richness
and assemblage composition. We sampled fishes during
three years in 28 streams used for municipal water supply in
the Piedmont region of Georgia, U.S.A. Study sites had
permitted average withdrawal rates that ranged from < 0.05
to > 13 times the stream's seven-day, ten-year recurrence
low flow (7010), and were located directly downstream
either from a water supply reservoir or from a withdrawal
taken from an unimpounded stream. Ordination analysis of
catch data showed a shift in assemblage composition at
reservoir sites corresponding to dominance by habitat gen-
eralist species. Richness of fluvial specialists averaged
about 3 fewer species downstream from reservoirs, and also
declined as permitted withdrawal rate increased above
about 0.5 to one 7010-equivalent of water. Reservoir pres-
ence and withdrawal rate, along with drainage area, ac-
counted for 70% of the among-site variance in fluvial
specialist richness and were better predictor variables than
percent of the catchment in urban land use or average
streambed sediment size. Increasing withdrawal rate also
increased the odds that a site's Index of Biotic Integrity
score fell below a regulatory threshold indicating biological
impairment. Estimates of reservoir and withdrawal effects on
stream biota could be used in predictive landscape models
to support adaptive water supply planning intended to meet
societal needs while conserving biological resources.
The ecological effects of meeting the water de-
mands of an expanding human population are of
concern worldwide (Postel 2000; Jackson and others
2001). Water withdrawals and diversions used to supply
municipalities, industries, and agricultural imgation
have the potential to degrade aquatic habitats to the
point that these systems fail to support native biota or
to supply other ecosystem services (Moyle and Leidy
1992; Baron and others 2002; Naiman and others
2002). Prominent examples include conflicts between
offstream water users and instream flow needs to sus-
tain imperiled species (Collares-Pereira and others
2000; Cooperman and Markle 2003; Ward and Booker
2003), and collapse of fisheries and productivity in
flow-deprived ecosystems (Postel 1996, 2000). Even in
regions where water historically has been considered
KEY WORDS: Stream fish; Species richness; Flow depletion;
Hydrologic alteration; Water supply planning
Published online Apnl 21, 2006.
*Author to whom correspondence should be addressed; email:
m ary_freeman @ usgs.gov
an abundant resource, such as eastern North America,
rapidly growing populations are placing increasing
demands on productive freshwater systems that sup-
port unique biodiversity.
The challenges of meeting growing demands for
water supply while protecting aquatic ecosystems are
exemplified in portions of the southeastern United
States. One such area is the southern Piedmont, situ-
ated between the Appalachian Mountains and the
Coastal Plain, which has experienced some of the
highest rates of population growth in the United States
in recent decades (Conroy and others 2003; Walters
and others 2005). Population growth and urbanization
are encroaching on aquatic habitats that support high
levels of aquatic biodiversity and endemism, as well as
supporting imperiled species (Abell and others 2000;
Warren and others 2000). Threats to native biodiversity
caused by altered runoff and pollution from urbaniz-
ing areas are likely to be compounded by water supply
development, largely dependent on surface water in
the Piedmont, unless specific management actions are
taken to safeguard vulnerable streams. Regulators in
the region are attempting to define instream flow
Environmental Management Vol. 38, No. 3, pp. 435-450 ©2006 Springer Science+Business Media, Inc.

436 M. C. Freeman and P. A. Marcinek
needs to protect flowing-water ecosystems, while
accommodating societal needs for water.
Management of surface waters tapped for water
supply has focused on protecting minimum flow levels,
although ecologists have stressed the importance of
flows across the range of the natural hydrograph for
maintaining structure and function of aquatic ecosys-
tems (Poll and others 1997; Richter and others 1997;
Silk and others 2000; Bunn and Arthington 2002).
Recent efforts to improve river management have in-
volved developing ecological flow requirements in-
tended to sustain floodplain and in-channel habitats,
and recognize flow seasonality and inter-annual varia-
tion as drivers of biological communities (Postel and
Richter 2003). The holistic approach to defining needs
for lotic ecosystems has also shifted the management
question from "how much flow must be provided to
meet ecosystem needs?" to that of "how much can flow
regimes be altered without incurring undesirable eco-
system change?" (Silk and others 2000; Bunn and Ar-
thington 2002). To develop ecologically sustainable
water supply policies, regulators will need clear infor-
mation linking withdrawal levels to effects on aquatic
ecosystems.
Previous studies of flow regulation effects on fish
assemblages have indicated greater detriment to fluvial
specialists, i.e., species that require flowing-water hab-
itats for at least a portion of their life-cycle (Kinsolving
and Bain 1993; Travnichek and others 1995) or rheo-
philic species (Copp 1990), in comparison with habitat
generalist species, which are able to maintain popula-
tions in lotic and lentic systems. A recent study of fishes
in aflow-depleted river in the northeastern United
States similarly has revealed a shift in species compo-
sition toward habitat generalists and a loss of fluvial
specialists (Armstrong and others 2001). Quantifying
the responses of differing faunal groups to flow alter-
ation may provide important information to resource
managers attempting to balance water use with con-
serving biota. Additionally, regulatory agencies are of-
ten interested in the status of biological communities
relative to reference or unimpaired conditions
(Barbour and others 1999), in which case effects of
water supply development on an assessment score such
as the Index of Biotic Integrity (IBI) would be useful.
Our purpose was to improve understanding of the
biological effects of water withdrawal by quantifying
variation in fish assemblages across streams that are
differentially used for municipal water supply. Streams
used for water supply vary with respect to permitted
withdrawal rate relative to the size of the stream.
Withdrawals also vary as to whether they are made
directly from an unimpounded stream or from an
instream reservoir. Reservoirs, by trapping and storing
water during periods of higher runoff, potentially alter
downstream flows over a broader range of the flow
regime than direct withdrawals. Thus, we investigated
the effects of increasing the relative withdrawal rate
and the use of water-supply reservoirs on stream fish
assemblages. We specifically examined effects on rich-
ness of fluvial specialist (FS) and habitat generalist
(HG) fishes, and compared the influences of with-
drawal rate and reservoir presence with effects of three
site-level variables chosen to represent influences of
natural (drainage area, bed sediment size) and
anthropogenic (amount of urban land use) factors on
fish assemblages. We also asked whether sample-spe-
cific instream habitat conditions improved site-level
models for predicting species richness. To analyze
assemblages relative to reference conditions, we used
multivariate ordination to analyze assemblage similarity
between our samples and samples taken in Piedmont
reference streams, and also evaluated IBI scores for our
samples in relation to a regulatory threshold used to
indicate biological impairment. We used results to
estimate quantitative effects of increasing water alto-
cations and using instream reservoirs on stream fish
assemblages, and considered how these estimates could
be used to incorporate biodiversity conservation in
water supply planning.
Methods
Study Site Selection
To hold other landscape influences more constant,
we restricted the study to one physiographic area, the
lower portion of the Piedmont physiographic region of
Georgia (Figure 1) . We used a GIS database of 53
permitted water withdrawals, obtained from the Geor-
gia Department of Natural Resources, Environmental
Protection Division (GDNR, EPD), to identify potential
study sites within 6 river systems (Savannah, Ogeechee,
Oconee, Ocmulgee, Flint, and Chattahoochee). We
selected 27 study sites in 2000 that represented all of
the wadeable streams with apparently active withdraw-
als and drainage areas exceeding about 12 kmZ (cor-
responding to the smallest reference site sampled in
2000 by the GDNR Stream Survey Team). In addition
to sites located innon-wadeable streams, sites that were
dry (2), affected by ongoing construction or pump-
storage operations (2), or that were not accessible (2)
were not included in the study. Of these 27 sites, 13
were situated downstream from direct water withdraw-
als ("intakes") and 14 were -ocated downstream from
water supply reservoirs ("reservoirs"; Appendix 1) . We

Water Withdrawals, Reservoirs, and Stream Fishes
s
e
:r
.t
Ridge and Yailsy . q t~ ~ ,
~~
Blw Ridge
% ~ ~<~
~ °
r
S '` `'~`.
6 a t~,~
2 ~ ~~ ~ w~
23 ~ '
~ ~ ;*
g 6 .-
17 k_ ~ ~'
..t `~ 4
,~~. ~ ~~ ,
i, ias . r... ~ :iii ~ San4hiUs
- ~ _-..
Legend ? §`_~
~ sites ~ ~, f ~:` .. ~~ ~, _-
Pl~dmont -- .- 16
-~- eao~ion twundary f ~ ,~-~~~~
_~ ~ 3
wMershed boundary ,~ t„ ?~ ~' °~ ( d
_ _ . infer bodies ~ ', .~; s~ -.~ 6
- ~. r -
~' ~ ~ i
t~
~rf ~~ ~ ' ~ ~ `~*,
,. to
~. V 'r <
12~s,
2 111
~~ ~ 1$. ,._. cd q
fit` `~ ~ ` `~
~ ~~ /' ~
,,~-T.
qt ~ -c h ~l~'.R
~ : ~ ~' °S~t ~1°~ `~ s.
,,~ ~~~ ~ t." `` °~'' Coastal Plain
~ ~,.".^. f r ,
1 t ~ ~', ~,
o az:s zs so ~s aoo
w~o+nat.ts .,
437
Figure 1. Watershed boundaries for 28 municipal water supply withdrawals in the Piedmont ecoregion, and adjacent ecoregion
boundaries. Numbers correspond to sites listed in Appendix 1.
repeated sampling in two subsequent years, 2001 and
2003, to assess effects over a range of instream habitat
conditions. For the 2001 field season, we added one
intake site that was dry during 2000 and eliminated one
intake site that became non-wadeable. During 2003,
which had substantially higher rainfall than in 2000
and 2001, we were able to sample 20 (12 reservoir sites
and eight intakes) of the 28 sites sampled in the pre-
vious years; the remaining eight sites were not sampled
because they became non-wadeable under the higher
flow conditions.
The GDNR sampled streams chosen as Piedmont
stream reference sites (i.e., for bioassessment pur-
poses) in 2000 and 2001 and made those data available
for comparison to our study sites. Reference sites were
chosen by GDNR on the basis of appearing relatively
unimpaired and supporting relatively intact fish
assemblages, and were located within the lower Pied-
mont in five of the six river systems containing study
sites. None of the sites were directly downstream from
withdrawals or reservoirs. Seven Piedmont reference
sites were sampled in 2000, ranging from 12.3 to 690
km2 in drainage area. These seven and two additional
sites (9.7 and 20 km2) were sampled in 2001. The
GDNR did not sample Piedmont reference sites in
2003.
Sampling Procedures
We sampled withdrawal sites between June and
September in each study year; reference sites were
sampled during September and October in 2000 and
2001. The June through October timeframe
represented a period of relative assemblage stability
(Matthews 1990; Peterson and Rabeni 1995) occurring
after the spring period of spawning migrations by some
species, and was within the sampling period (April
through October) used by GDNR to assess integrity of
stream fish assemblages. We sampled fishes following
protocols developed by GDNR (GDNR 2000). Reach
length approximated 35 times mean wetted-channel
width (estimated from width measurements at five
randomly chosen locations for each 100 m length in-
cluded) to a maximum of 500 m. We sampled fishes
from the downstream to upstream boundaries of the

438 M. C. Freeman and P. A. Marcinek
sample reach using one or two backpack electrofishers,
or abarge-mounted boat electrofisher (if mean stream
width exceeded about 8 m and the stream was deep
throughout), dip nets, and seines. Captured fishes
were transferred to buckets and coolers; fish to be
measured and released were kept in aerated, fre-
quently exchanged water. Fish that were preserved
were first anesthetized with iricaine methanesulfonate
(MS-222) and subsequently transferred to 10% forma-
lin. All fish were identified to species and measured
(total length) either in the field or laboratory. We also
recorded incidences of individuals with evident dis-
ease, eroded fins, lesions, and tumors, and with "black
spot" (trematode cysts).
We measured stream discharge, water temperature,
and turbidity on each sampling date. Turbidity was
measured in nephelometric turbidity units (NTU) with
a Hach® Model 2100P turbidity meter. In 2000 and
2001, we also measured dissolved oxygen (DO) con-
centration at the upstream end of each sample reach
with a Hydrolab® multiprobe. We estimated stream
discharge near the upstream end of the site using a
Marsh-McBirney F1o-Mate® electromagnetic current
velocity meter and top-setting wading rod. Discharge
was based on depth and velocity measured every 0.5 m
along a tape measure stretched from bank to bank, or
at intervals sufficient to give at least 20 measurements
across the stream.
Habitat data were collected during low flow condi-
tions, usually immediately following fish collections.
The length of each pool, riffle, and run in the sample
reach was measured and recorded in sequence by type.
We randomly selected three to five locations for cross-
section measurements from the total length of each
habitat type. At each cross-section, we recorded depth,
velocity, and dominant bed sediment size [recorded in
phi intervals (Gordon and others 1992)] at channel
edges and at three equally spaced positions across the
channel (0.25, 0.5, and 0.75 times wetted width).
Depth, velocity, and bed sediment averages for each
site were computed as means of pool, run, and riffle
measurements, weighted by the proportion of the site
in each habitat type. In 2003, dominant bed sediment
size was recorded longitudinally along the mid-channel
at intervals equal to 0.5 times average wetted-channel
width to give a more complete profile of sediment
variation.
Analyses
Site Characteristics. We summarized average stream
flow, depth, velocity, and water quality measurements
to compare flow and habitat conditions among years
and between intake and reservoir sites. Land use in the
catchment upstream from each site was estimated
using a statewide land cover map based on 1998
Landsat Thematic Mapper imagery, with 30-m resolu-
tion (produced by the Natural Resources Spatial
Analysis Laboratory, University of Georgia, in 2001).
We examined three land use categories: urban (low-
and high-intensity), forest (including deciduous, ever-
green, and mixed forest), and agriculture (pasture and
row crop). We estimated mean bed sediment size at
sites by averaging the mean values computed for each
year the site was sampled.
To facilitate among-site comparison of potential
withdrawal rate relative to stream size, we computed a
"withdrawal index" (WI) for each site as the maximum
permitted monthly average withdrawal rate (in million
gallons per day, mgd) divided by the estimated seven-
day, ten-year recurrence low flow (7Q10) at the with-
drawal site (also expressed in mgd). The WI thus rep-
resented the fraction or multiple of the 7Q10 flow
permitted for withdrawal on a monthly average basis.
The 7Q10 flow is commonly used by regulatory agen-
cies to set wastewater discharge criteria, and has also
been used to set minimum flow requirements. We used
7Q10 flows to standardize permitted withdrawal rates
across sites because 7Q10 estimates could be obtained
from EPD files or estimated from low-flow profile data
(Carter and others 1986, 1988a,b). Other flow statistics
that could also be used to standardize withdrawal rates,
such as average annual flow, were not available because
27 of the 28 sites lacked stream gages. This also pre-
cluded quantifying actual hydrologic patterns or flow
alteration. To examine actual water withdrawal in
relation to WI, we used the maximum monthly with-
drawal rate reported for 25 sites during the 12-month
period prior to our fish samples in 2000 and 2001,
divided by 7Q10 for the site. Monthly withdrawal data
were provided by EPD or by permit holders. Because
we observed strong relations between WI and water use
(results reported below), and because water use varied
at many sites across months and water use data were
not available for all sites and months, we used WI as
our measure of potential withdrawal effects in all
analyses.
Fish Assemblage Patterns. We used the catch data
across sites and years in an ordination analysis to
examine the relative similarity of fish assemblage data
in our samples to those from the GDNR Piedmont
reference sites. We employed nonmetric multidimen-
sional scaling (NMDS), using Bray-Curtis dissimilarity,
to ordinate our study-site samples (n = 74) and Pied-
mont reference site samples (n = 16). NMDS provided
graphic representation of the relative similarity among
samples based on taxa abundances (Field and others

Water Withdrawals, Reservoirs, and Stream Fishes 439
1982; Clarke 1993; McCune and Grace 2002). For each
sample, we applied afourth-root transformation to
abundance data in order to dampen the influence of
common taxa (Clarke 1993), retaining for analyses all
taxa that occurred in at least six samples overall and
that had ranges encompassing all sample locations. To
meet the latter criterion, we combined abundances of
some congeneric species with ranges restricted to a
subset of the sampled river systems (Appendix 2). We
used a total of 34 taxonomic entities (24 species and 10
multispecies genera) in NMDS analysis. Analyses were
accomplished with PC-ORD (McCune and Mefford
1999), using astep-down procedure (from 6- to 1-
dimensions) to find the most appropriate solution,
with 200 iterations, 10 runs with the real data, 20 runs
with randomized data, and stability criterion set to
0.0001 (McCune and Grace 2002). Because of reviewer
concerns that the PC-ORD solution might be far from
the minimum stress solution, we also conducted NMDS
ordination using the function isoMDS, package vegan,
in the R programming environment (Oksanen and
others 2005), using Bray-Curtis dissimilarity and the
same number of dimensions (3) as in the final solution
from PGORD. We used product-moment correlation
coefficients between taxa abundances and ordination
axes, and graphical representation, to examine pat-
terns of assemblage differences among sites.
Water Withdrawal and Reservoir Effects on Species
Richness. We used the limiting form of the jackknife
estimator for model My, (Burnham and Overton 1979;
Williams and others 2002) to estimate richness of FS
and HG species in each sample, given the observed
numbers of species and numbers of species in each
sample represented by 1, 2, 3, 4, or 5 individuals. We
used the program SPEGRICH, available at http://
www.mbr-pwrc.usgs.gov/software.html. The purpose of
using the jackknife estimator for richness rather than
actual sample counts of species was to reduce bias
resulting from incomplete species detection and
among-species differences in detectability.
We used an information-theoretic approach
(Burnham and Anderson 2002) to evaluate the rela-
tive effects of WI and upstream presence of a reser-
voir on fish species richness. Our approach was to
construct a set of alternative linear regression models
that, first, would allow us to compare the effects of WI
and reservoir presence with each other and with
other site-level variables hypothesized to influence
species richness. Secondly, we wished to evaluate ef-
fects of among-year differences in instream habitat on
species richness. Our data set comprised 72 observa-
tions (one to three observations at 27 sites; see Re-
sults). To avoid model over-fitting, we restricted
models to a maximum of 7 parameters (i.e., to keep
the ratio of observations to parameters to about 10 to
1; Burnham and Anderson 2002). All regression
models included a term for random variation among
sites to account for unmeasured site-specific influ-
ences on the repeated observations (Snijders and
Bosker 1999) in addition to an intercept and within-
site residual error term. Thus, we included a maxi-
mum of four explanatory variables in regression
models to limit the total number of model parame-
ters to seven.
We considered the effects of three site-level variables
in addition to WI and reservoir presence on species
richness for FS and HG fishes. First, we included
drainage area in all models because fish species rich-
ness generally increases as a function of stream size
(Matthews 1998). Given that our sites spanned two
orders of magnitude in drainage area, we did not be-
lieve that any credible explanatory model could ignore
drainage area as a predictor variable. We also hypoth-
esized that the average size of the stream bed sedi-
ments would influence species richness of FS and HG
fishes, based on observations in upper Piedmont
streams of a shift in fish assemblage structure from
dominance by fluvial specialists in steeper, rockier
streams to habitat generalists in lower gradient streams
with finer bed sediments (Walters and others 2003b).
Finally, we hypothesized that the level of urbanization
upstream from the study sites could depress species
richness, at least of FS species, as observed in other
Piedmont streams (Weaver and Garman 1994; Walters
and others 2003a; Roy and others 2005). Therefore, we
evaluated 13 models with drainage area and combina-
tions of WI (ln transformed), reservoir presence, mean
phi, and percent urban land use upstream from the site
(aresine transformed). We also evaluated whether
adding terms indicating basin identity, presence or
absence of a minimum flow requirement, and an
interaction between WI and reservoir presence im-
proved the best-fit site-level model predicting species
richness. Our purpose was to construct a small set of
preselected candidate models (Burnham and
Anderson 2002) intended specifically to compare ef-
fects of WI and reservoir presence with variables rep-
resenting natural and land use influences, while also
testing for the potential influences of differences
attributable to basins, minimum flow requirements,
and the possibility of a reservoir-withdrawal level
interaction. Basin identity was coded as Apalachicola
(i.e., Chattahoochee and Flint river systems), Altama-
ha/Ogeechee (i.e., Ocmulgee, Oconee, and Ogeechee
river systems), or Savannah following Warren and
others (2000).

440 M. C. Freeman and P. A. Marcinek
We used estimates of mean depth and mean velocity
measured for each yearly sample to evaluate the rela-
tive importance of hydraulic habitat conditions on
species richness. We compared linear regression mod-
els with mean depth and mean velocity in addition to
drainage area alone and in combination with WI and
reservoir presence. Because of missing measurements,
we used reduced data sets to compare the predictive
ability of depth and velocity with that of turbidity and
water temperature (data for 71 samples) and DO (data
for 49 samples). Velocity, turbidity, and DO data were
In transformed to reduce departures from normality.
Models were compared on the basis of differences in
Akaike's Information Criterion values, corrected for
small sample size (AIC~; Burnham and Anderson
2002). AIC provides an estimate of the expected, rela-
tive distance between a fitted model and the unknown
mechanism (s) that generated the data, so that the best-
supported model among a set of plausible models is
indicated by the lowest AIC (Burnham and Anderson
2002). The approach thus allows selection of the sim-
plest, best-fitting model(s) given the observed data,
while avoiding the problems of multiple hypothesis
testing and arbitrary probability levels associated with
stepwise variable selection. An information-theoretic
approach also allows one to make inferences (in this
case, regarding WI and reservoir effects on species
richness) based on multiple models that share rela-
tively strong support, thereby incorporating uncer-
tainty in model selection. To this end, we used Akaike
weights (w;), which vary from 0 to 1 with the best-fitting
model having the highest weight, to evaluate relative
support for each model. Summing weights across the
models that included each variable provided an esti-
mate of relative variable importance. We also used
weights to estimate model-averaged effect sizes of WI
and reservoir presence on species richness, with 90%
confidence intervals (CIs) based on unconditional
standard errors (Burnham and Anderson 2002). All
models were fit in SAS version 8.01 (SAS Institute,
Cary, NC, USA) using PROC MIXED and specifying
maximum likelihood estimation to derive AIC values,
or restricted maximum likelihood to estimate among-
site and residual variance components (Snijders and
Bosker 1999). Unconditional models (i.e., with no
explanatory variables) were fit to estimate total vari-
ances attributable to among-site and residual compo-
nents, and to provide a basis for computing how much
variance in each component was accounted for by
explanatory variables.
Water Withdrawal and Reservoir Effects on Probability of
Site Impairment. We computed IBI values for each
sample using the GDNR protocol to evaluate evidence
that WI or presence of an upstream reservoir contrib-
uted to sites qualifying as "impaired." Under the
GDNR protocol, IBI values are computed as sums of
scores of 1, 3, or 5 for 12 metrics that reflect aspects of
species richness and composition, trophic composi-
tion, and fish abundance and condition (GDNR 2000).
Sites scoring lower than 34 are assigned to classes of
"poor" or "very poor" and are classified by the State as
biologically impaired for purposes of water quality
assessment under section 305(b) of the Clean Water
Act. We thus analyzed effects of WI and reservoirs on
probability of impairment because of the regulatory
relevance of the question, and because IBI calculations
are widely used to assess biological aspects of stream
quality (Karr and Chu 2000). We note, however, that
IBI estimates maybe biased by uncontrollable variation
in capture efficiency among sites (e.g., resulting from
habitat differences; J. T. Peterson, USGS, Athens GA,
personal communication). Additionally, our methods
differed somewhat from those employed by the GDNR
survey team. Specifically, whereas the GDNR team fre-
quently used multiple electrofishers moving upstream
in parallel and closely followed by dip-netters, we more
frequently sampled by shocking downstream into a
seine set in flowing-water habitats, and we combined
electrofishing and dip-netting with seine hauls in slack
water. We have not estimated capture efficiency for
either protocol or as a function of site characteristics,
and therefore we interpret results based on IBI values
cautiously.
We used logistic regression to compare the effects of
(1) WI and reservoir presence compared to other site
variables (drainage area, mean bed sediment size, and
percent urban land use) and (2) hydraulic habitat
conditions (mean depth and velocity) on probability
that a site scored as "impaired" in a given sample. Our
model set included: all five site variables; four variables
(omitting either WI or reservoir presence); three vari-
ables (omitting both WI and reservoir presence), and
WI and reservoir presence alone and in combination
with each of the other site variables and each other (16
models total). We also tested whether the presence or
absence of a minimum flow requirement or an inter-
action between WI and reservoir presence improved
the best-fit site-level model. We did not constrain
models to include drainage area or test for an effect of
basin identity because IBI computations explicitly ac-
count for these variables. Finally, we tested whether
adding mean depth and velocity improved fit of the
highest-ranked site-level models. Logistic regression
models were fit in SAS using PROC GENMOD and
specifying repeated measurements at sites to account
for clustered data (Allison 1991). As above, models

Water Withdrawals, Reservoirs, and Stream Fishes 441
Table 1. Habitat conditions at intake and reservoir sites in each year, showing means (standard error, range)

Variable
Withdrawal type
2000 Year
2001
2003
Flow, discharge/7Q10 Intake 0.62 (0.20, 0 - 2.31) 3.22 (0.97, 0 - 14.0) 7.41 (2.52, 1.15 - 20.8)
Reservoir 0.62 (0.16, 0 - 1.62) 3.92 (0.86, 0.03 - 12.1) 9.30 (1.68, 0.85 - 21.7)
Depth, m Intake 0.28 (0.04, 0.10 - 0.51) 0.42 (0.04, 0.16 - 0.60) 0.36 (0.04, 0.23 - 0.55)
Reservoir 0.28 (0.03, 0.06 - 0.49) 0.34 (0.04, 0.10 - 0.52) 0.35 (0.04, 0.16 - 0.55)
Velocity, m/s Intake 0.07 (0.025, 0 - 0.29) 0.15 (0.021, 0.03 - 0.30) 0.17 (0.05, 0.07 - 0.44)
Reservoir 0.04 (0.012, 0 - 0.13) 0.09 (0.019, 0.02 - 0.24) 0.17 (0.03, 0.03 - 0.40)
Temperature, °C Intake 25.5 (0.86, 21.8 - 33) 24.7 (0.54, 21.2 - 27.3) 23.8 (1.0, 20 - 28)
Reservoir 26.6 (0.62, 22.9 - 29.3) 26.2 (0.63, 21.3 - 29.9) 27.4 (0.8, 23 - 32)
Turbidity, NTU Intake 32.0 (8.0, 6.5 - 94.6) 17.7 (2.6, 5.5 - 41.5) 19.1 (3.6, 8.2 - 39)
Reservoir 33.3 (12.8, 2.6 - 151) 13.2 (3.4, 3.7 - 54.4) 11.4 (1.6, 4.8 - 21.8)
Dissolved oxygen, mg/L Intake 5.4 (0.39, 1.8 - 6.7) 7.1 (0.26,4.7 - 8.5) Not measured
Reservoir 4.5 (0.45, 1.5 - 6.7) 6.7 (0.31, 4 - 8) Not measured
were compared on the basis of AIC~ values and Akaike
weights, which we also used to derive model-averaged
estimates of effect sizes.
Results
Site Characteristics
Permitted water withdrawal levels ranged from 1.5
to 45 million liters per day (0.4 to 12 mgd) across the
28 study sites (Appendix 1). WI values ranged from 0
(where withdrawals were discontinued) to 13.33 across
sites and were negatively correlated with drainage area
(r = -0.43, n = 27). Actual water withdrawal amounts
during the 12 months preceding 2000 and 2001 sam-
ples varied substantially among sites and among
months at a given site. The maximum monthly with-
drawal reported during a year ranged from 0.02 to 13.5
times 7Q10, and was strongly correlated with WI in
2000 (r = 0.97, n = 25) and 2001 (r = 0.90, n = 24).
Reservoir sites had a smaller average drainage area,
54 kmz, compared to 270 km2 at intake sites, and a
larger average WI (2.90 compared to 1.50 at intake
sites). Reservoir and intake sites were similar in other
characteristics. Catchment land use upstream from
intake and reservoir sites was predominantly forest
(means = 57% and 59%, respectively), followed by
agriculture (means = 21% and 20%, respectively) and
urban (means = 4.7% and 3.5%, respectively). Average
dominant bed sediment size ranged from small cobble
to sand (phi = -7 to 1) at all except for two sites that
were dominated by bedrock. The 11 sites with the
smallest average bed sediment sizes (phi = -1 to 1)
were nearly evenly divided between intake and reser-
voir sites (Appendix 1).
Stream flow, depths and velocities during sampling
periods corresponded to drought conditions during
2000 and 2001, and to elevated flows during 2003, and
were similar between intake and reservoir sites in all
years (Table 1). Temperature ranges were similar
among years, although during the wetter conditions of
2003, streams below intakes averaged 3.6°C cooler than
those below reservoirs. Dissolved oxygen levels were
lowest in 2000 (Table 1), when we recorded levels less
than 5 mg/L at seven reservoir and four intake sites.
One reservoir and one intake site had a DO level <5
mg/L in 2001. Turbidity varied widely across sites, from
<5 to >150 NTU in 2000, with lower values in
subsequent years (Table 1).
Fish Assemblage Patterns
The NMDS ordinations produced by PGORD and R
were strongly similar, with stress values of 14.43 and
14.37, respectively, for a three{limensional solution.
We have used the PGORD results to illustrate among-
site differences. In this ordination, axes 2 and 3
cumulatively represented 76% (33 and 43%, respec-
tively) of the variance in the original among-sample
distance matrix. Axis 1 represented less (8%) of the
among-sample variance; therefore, we used scores on
axes 2 and 3 to illustrate assemblage similarity among
samples. The ordination showed a shift in fish assem-
blage structure at reservoir sites away from reference
sites (Figure 2), corresponding to dominance by habi-
tat generalist taxa in reservoir site samples. Reservoir
sites mostly scored relatively high on axis 2 and low on
axis 3, and displayed almost no overlap with reference
sites, which scored oppositely. Intake samples over-
lappedwith reference and reservoir samples, but mostly
scored higher on axis 3 than did reservoir sites. Higher
sample scores on axis 2 corresponded to higher abun-
dances of primarily HG taxa (Table 2), specifically
several sunfish species (genus Lepomis), largemouth

442 M. C. Freeman and P. A. Marcinek
• Piedmont reference sites
Intake sites ~ ~~
D Reservoir sites ~ ~ ~ ~
o °p~ ~ ^~°
• fit, ^ °o ~ ~ o
• V p • LS7
• ° o ~ °~ o
° ° o ° o
00 0 A ~~ E~ ~~
o g o
~ ~
o ~
Axis 2
Figure 2. NMDS ordination of fish sample data from 2000,
2001, and 2003 at intake, reservoir, and Piedmont reference
sites, illustrating relative similarities among sites based on
assemblage composition.
bass Micropterus salmoides, and crappie (genus Pomoxis).
Conversely, higher scores on axis 3 corresponded to
higher abundances of FS taxa, especially insectivorous
cyprinids ("shiners" Cyprinella spp., spottail shiner
Notropis hudsonius), benthic invertivores (madtom cat-
fishes Noturus spp., blackbanded darter Percina nigro-
fasciata), catostomids (` jumprocks" Scartomyzon spp.,
"redhorses" Moxostoma spp.), snail bullhead (Ameiurus
brunneus), and black basses other than largemouth bass
(Micropterus spp., Table 2). One HG species, redbreast
sunfish Lepomis auritus, also had a relatively strong
positive correlation with axis 3 (Table 2).
Species Richness Variation
We collected a total of 47 FS species, all of which
were native to one or more of the sampled river sys-
tems; we collected a total of 39 HG species, of which six
were not native to any of the sampled river systems. We
based all FS and HG richness estimates on 151-order
jackknife calculations, using the number of species
observed and the number represented by a single
individual. The 151-order estimate provided the best fit
to our data in 85% of the samples, and higher-order
estimates had substantially larger standard errors and
generally were inconsistent at a site across years.
Estimated richness of FS fishes varied from 0 to 24
species across study sites and years, and was, on aver-
age, lowest at reservoir sites and highest at reference
sites (Figure 3A). Overall, 75.6% of the variance in FS
richness at reservoir and intake sites was attributable to
differences among sites (i.e., among-site vari-
ance = 20.30; residual = 6.57). Estimated FS richness
increased with drainage area (Figure 4A). Estimates of
HG richness varied from 2 to 20 species across sites,
Table 2. Correlations between transformed taxa
abundances and sample scores on NMDS axes 2 and 3
Species or taxonomic group Category Axis 2 Axis 3
Lepomis macrochirus HG 0.68 0.00
Micropterus salmoides HG 0.59 0.05
Lefiomis microlophus HG 0.56 0.18
Lepomis gulosus HG 0.55 0.00
Pomoxis spp. HG 0.51 0.29
Lepomis punctatus HG 0.41 0.24
Notropis spp. FS 0.41 0.28
Dorosoma cepedianum HG 0.27 0.28
Gambusia spp. HG 0.26 -0.28
Minytrema melanops FS 0.21 0.32
Percina nigrofasciata FS 0.14 0.80
Iclalurus rrunctatus HG 0.14 0.47
Perca flavescens HG 0.14 0.39
Cyprinus carpio HG 0.14 0.32
Lepomis cyanellus HG 0.10 0.44
Lepomis auritus HG 0.07 0.69
Cyprinella spp. FS 0.05 0.84
Moxostoma spp. FS 0.03 0.73
Noturus spp. FS 0.03 0.63
Notropis hudsonius FS -0.01 0.70
Ameiurus brunneus FS -0.16 0.79
Micropterus spp. FS -0.28 0.60
Scartomyzon spp. FS -0.31 0.70
Etheostoma spp. FS -0.35. 0.37
Hybopsis spp. FS -0.38 0.50
Nocomis leptocephalus FS -0.59 0.40
Notropis lutipinnis FS -0.65 0.17
Note: Taxa are listed in descending order by score on axis 2, and are
categorized as habitat generalists (HG) or fluvial specialists (FS). Se-
ven HG taxa for which correlation coefficients were <_ 0.2 are not
listed: Anguilla rostrata, Esox niger, E. americanus, Notemigonus
crysoleucas, Erimyzon oblongus, Ameiuras spp., and A. natalis.
were similar between intake and reservoir sites, and
were similar to estimates at reference sites in 2001
(Figure 3B). Contrary to our expectations, HG rich-
ness did not increase with drainage area (Figure 4B);
overall, 26.8% of the variance in HG richness was
attributable to differences among sites (i.e., among-site
variance = 5.00; residual = 13.68).
Water Withdrawal and Reservoir Effects on Species
Richness
The most strongly supported linear regression
models of fluvial specialist richness incorporated
drainage area, WI, and reservoir presence (Table 3).
Support for including any of the other site level vari-
ables (% urban, mean phi, basin identity, or occurrence
of a minimum flow requirement) was relatively weak;
e.g., AIC~ differences between the best site-level model
(using DA, WI, and reservoir presence) and models
with other site-level variables ranged from 2.3 to 9.6
(Table 3). Drainage area, WI, and reservoir presence,

Water Withdrawals, Reservoirs, and Stream Fishes 443
22
20
18
16
14
12
10
8
6
N 4
~U
0) 2
Q
rq 0
....
~ 22
L
~ 20
18
Z 16
14
t2
10
8
6
4
2
0
Intake sites Reservoir sites Reference sites
2000 ~ 2001 ~ 2003
Figure 3. Mean estimated richness for fluvial specialist (A)
and habitat generalist (B) fishes at intake, reservoir, and
Piedmont reference sites in 2000, 2001, and 2003. Error bars
indicate one standard error.
together, accounted for 70.3% of the among-site vari-
ation in FS richness estimates (i.e., among-site variation
declined from 20.30 to 6.02). FS richness declined with
increasing WI and with reservoir presence (Figure 5A).
Summed weights for the 14 models with reservoir
presence was 0.81 compared to 0.56 for the 14 models
that included WI, indicating greater support for the
influence of reservoir presence on FS richness. Incor-
porating average water velocity and, to a lesser degree,
average depth improved model fit for FS richness
(Table 3), and accounted for 18.8 and 9.1% of residual
variance, respectively. FS richness increased with
increasing velocity and depth. Tests with reduced data
sets showed that velocity was substantially more
predictive of FS richness than turbidity or water tem-
perature (n = 71, AIC~ differences from velocity mod-
el = 8.1 and 11.2, respectively), or DO (n = 49, AIC~
difference from velocity model = 4.8).
None of the tested models provided a good fit to
HG richness estimates, which displayed no relation to
WI or reservoir presence (Figure 5B). Including six
site-level variables (drainage area, WI, reservoir pres-
30 1
25 -
20 -
15-
10-
y 5
N
U
N 0-
a
N
O 25 -
N
~ 20 -
7
Z
15-
10-
51
A Fluvial specialists
^
^
~ ^ 1
(~ O ^
^ ~ ~ ^^ ^ ~
^ ^ Q ~' ^
b~b
~^ ~~ ~~~
D1 ~
B Habitat generalists
~ ~
m Q ~ d
D b ~ ^ ^~
p' ~ Dm ^~ Q ^~ ^
i ~^ i ^r i
3 4 5 6 7 8
Drainage area (In transform)
~ Intake sites Q Reservoir sites
0a
2
Figure 4. Richness estimates for fluvial specialist (A) and
habitat generalist (B) fishes plotted in relation to drainage
area at intake and reservoir sites, data for all years.
ence, % urban, mean phi, basin identity) accounted
for only 18.9% of the total variance in HG richness
estimates. The best-supported model included depth
alone and accounted for 8.4% of total variance.
Model-averaged effect sizes show a decrease in FS
richness of about 1.7 species for a unit increase in ln-
transformed WI values (mean = -1.69, 90% CI = -0.34
to -3.04). This equates to a loss of about one species as
permitted withdrawal increases from 0 to 0.8 x 7Q10,
and loss of an additional three species as WI increases
to 12 x 7Q10. Sites below reservoirs are predicted to
have an average of about three fewer FS species
(mean = -3.26, 90% CI = -1.09 to -5.42) compared
with sites below intakes. The large confidence intervals
for these estimates reflect variation in the data and
uncertainty associated with model selection.
Water Withdrawal and Reservoir Effects on Site
Impairment
Twelve (86%) of the reservoir sites and seven (50%)
of the intake sites scored as impaired in at least one
year, although fewer sites (four reservoir and two

444 M. C. Freeman and P. A. Marcinek
Table 3. Alternative linear regression models of fluvial specialist richness listed in order from best- to
least-supported
Site variables Habitat variables AAIC, w;
DA, WI, Reservoir Velocity 0 0.39
DA, Reservoir Velocity 1.2 0.21
DA, Reservoir Depth, Velocity 1.8 0.16
DA, WI Velocity 3.0 0.09
DA,WI Depth, Velocity 4.3 0.05
DA Velocity 4.9 0.03
DA, WI, Reservoir Depth 5.7 0.02
DA Depth, Velocity 6.2 0.02
DA, Reservoir Depth 6.9 0.01
DA, WI Depth 9.1 0.00
DA, WI, Reservoir 9.5 0.00
DA, WI, Reservoir, WI*Reservoir 11 0.00
DA Depth 11.2 0.00
DA, Reservoir 11.4 0.00
DA, WI, Reservoir, Minimum Q 11.8 0.00
DA, WI, Reservoir, Mean phi 11.9 0.00
DA, WI 12.0 0.00
DA, WI, Reservoir, % Urban 12.0 0.00
DA, WI, Reservoir, Basin 12.4 0.00
DA, Reservoir, % Urban 12.8 0.00
DA, Reservoir, Mean phi 13.8 0.00
DA, WI, % Urban 14.1 0.00
DA, WI, Mean phi 14.4 0.00
DA 14.9 0.00
DA, % Urban 16.7 0.00
DA, Mean phi 17 0.00
DA, Mean phi, % Urban 19.1 0.00
Note: Results show variables included, differences in AIC~ from the best-fit model, and model weights (u ,). Results shown in bold are for models
used to derive model-averaged effect sizes. DA = in-transformed drainage area; WI = in-transformed withdrawal index; Reservoir =reservoir
presence; % Urban = aresine-transformed percent urban land use; Velocity = in-transformed average velocity.
intake) scored as impaired in all years sampled. The
best-supported logistic regression models for proba-
bility of a sample at a site scoring as impaired incor-
porated WI, drainage area, average velocity, and, with
lesser support, average depth (Table 4). In contrast to
the models for FS richness, adding reservoir presence
to the model was not as strongly supported (e.g., 2.2
difference in AIC~ values for site-level models with WI
and reservoir presence versus WI alone, Table 4). The
odds of scoring as impaired increased by about 300%
with a unit increase in In-transformed WI value (model-
averaged odds ratio = 4.07; 90% CI = 0.95 to 17.43), or
as WI increased from 0 to about 2 x 7Q10. However,
the large 90% confidence interval for the odds ratio
encompassed no effect and indicated substantial
uncertainty in the predictions.
Discussion
The results of this study have allowed us to estimate
the effects of increasing the allowable rate of water
withdrawal from streams, and of using instream reser-
voirs, on composition of downstream fish assemblages.
In the wadeable, lower Piedmont streams included in
our study, increasing the potential for water withdrawal
and use of an instream reservoir were associated with a
loss of native fish species that are dependent on flow-
ing-water habitats. Vulnerable species included a vari-
ety of minnows (Cyprinidae), redhorse and jumprock
suckers (Catostomidae), darters (Percidae), and
catfishes (Ictaluridae), as well as stream-dwelling basses
(Centrarchidae). Models using drainage area, WI, and
reservoir presence to predict fluvial specialist richness
were better supported by our data than models using
drainage area alone or in combination with any of the
other site-level variables tested. Models were improved
by including sample-specific average water velocity,
indicating the influence of velocity on site habitability
by fluvial specialist fishes.
Our results are consistent with the hypothesis that
altering flow regimes will affect stream biota in relation
to the degree of alteration (Poff and others 1997; Bunn
and Arthington 2002). Raising the permitted water
withdrawal rate potentially increases the proportion
of flow removed across the hydrograph, reducing
seasonal and interannual variability in baseflow

Water Withdrawals, Reservoirs, and Stream Fishes
30
A Fluvial specialists
25 ^
20 ^
w! ^
15 ^ZS
~ ~^ m ~
~ o

~
o ~ o ~
a o ~ ~

O
` 25
B Habitat generalists
a~
~
20
0
Z 'g ° g o
15 ~ ^~ ^ p D ^
10 i ~ ~ p p ~
^~ ^ ~
5 ^ ~~ ~ ^
^ ^ ~ ~ ^
^ Q

o +-
o.oo
0.50 1.00 1.50 2.00 2.50 3.00
Permitted monthly average withdrawaV7Q10 (In transform)
~ Intake sites Q Reservoir sites
Figure 5. Richness estimates for fluvial specialist (A) and
habitat generalist (B) fishes plotted in relation to withdrawal
index at intake and reservoir sites, data for all years.
conditions. Creating water storage with instream res-
ervoirs further increases the capacity for altering flows.
Reservoirs use runoff to refill depleted storage, thereby
diminishing high flows (e.g., during wetter seasons or
years). The effect on downstream fishes may be ex-
pected to vary as a function of reservoir volume relative
to inflow and rate of water withdrawal. However,
despite likely variation in reservoir operations, we ob-
served ageneral effect on downstream richness of flu-
vial specialist species. In contrast, habitat generalist
species displayed no association with either the rate of
permitted water withdrawal or upstream reservoir
presence.
Reservoirs can also influence downstream water
quality, depending on temperature and dissolved oxy-
gen conditions within the reservoir and from what
reservoir stratum water is released downstream. The
water supply reservoirs in our study primarily release
surface water. Surface releases can send warm water
downstream during summer, whereas hypolimnetic
releases may be cooler and low in dissolved oxygen
(Baxter 1977; Collier and others 1996). Our data show
445
higher average water temperatures downstream from
reservoirs than from intakes, whereas we have observed
instances of low dissolved oxygen below intakes as well
as reservoirs. Elevated water temperatures would be
expected to be more detrimental to fluvial specialists
adapted to forested streams than to pond- and lake-
adapted habitat generalists (Scott and Helfman 2001).
Thus, increasing downstream water temperatures may
be a mechanism by which water-supply reservoirs can
cause a shift in fish assemblages. Reservoirs may also
trap sediments, resulting in lower downstream turbid-
ities (Collier and others 1996), although our turbidity
measurements below intakes and reservoirs are variable
with broadly overlapping ranges. We also have not
observed large average differences between intake and
reservoir sites in mean depth, mean water velocity, or
relative flows. These comparisons are based on mea-
surements made on a few days during low-flow periods;
continuous data would provide better information on
the occurrence of periodically stressful instream con-
ditions. However, the similarities in observed condi-
tions during low flows below intakes and reservoirs
supports the hypothesis that the apparent reduction in
habitat suitability for fluvial specialists downstream
from reservoirs is not solely a function of altered water
quality or low-flow habitat, but also results from alter-
ation in flow regimes.
Our estimates of effect sizes of withdrawals and
reservoirs on fluvial specialist fishes reflect consider-
able uncertainty, resulting in part from the broad
array of factors that may influence stream fish
assemblages. Present and past land use, geomor-
phology and instream habitat structure, pollutants,
species interactions, and stochastic variation may all
influence the composition of local fish assemblages.
Moreover, each stream reach is part of a landscape in
which phenomena operating at larger temporal and
spatial scales, notably geomorphic processes and ani-
mal movements, likely influence local populations
(Fausch and others 2002). Given the large number of
factors potentially affecting local stream assemblages,
it is not surprising that only a portion of the variance
in assemblage characteristics can be accounted for by
models based on drainage area and permitted water
supply rates or use of an upstream reservoir for sup-
ply. For example, isolation by reservoirs (upstream
and downstream) as well as close proximity to
downstream urban areas and point-source discharges
are likely to diminish local species assemblages,
whereas connections with nearby tributary systems
having intact faunal communities are likely to aug-
ment local species richness, independently of flow
alteration effects. The observation that water supply

446 M. C. Freeman and P. A. Marcinek
Table 4. Alternative logistic regression models for probability of a stream sample scoring as impaired, listed in
order from best- to least-supported
Site variables Habitat variables ~AIC~ w;
WI Velocity 0 0.51
DA, WI Velocity 1.6 0.23
WI Depth, Velocity 2.1 0.18
DA, WI Depth, Velocity 3.9 0.07
WI 12.0 0.00
DA, WI 12.4 0.00
WI Depth 13.3 0.00
WI, WI*Reservoir 13.7 0.00
WI, Minimum Q 13.8 0.00
WI, Mean phi 14.2 0.00
WI, % Urban 14.2 0.00
WI, Reservoir 14.2 0.00
DA, WI Depth 14.4 0.00
DA, WI, Reservoir 14.5 0.00
WI, Reservoir, Mean phi 16.4 0.00
WI, Reservoir, % Urban 16.4 0.00
DA, WI, Mean phi, % Urban 16.8 0.00
DA, Mean phi, % Urban 19.0 0.00
DA, WI, Reservoir, Mean phi, % Urban 19.1 0.00
DA, Reservoir 20.1 0.00
DA, Reservoir, Mean phi, % Urban 21.2 0.00
Reservoir 22.7 0.00
Reservoir, % Urban 22.8 0.00
Reservoir, Mean phi 24.5 0.00
Note: Results show the variables included, differences in AIC~ from the best-fit model, and model weights tw;~. Results shown in bold are for models
used to derive model-averaged effect sizes. DA = in-transformed drainage area; WI = in-transformed withdrawal index; Reservoir =reservoir
presence; % Urban = aresine transformed percent urban land use; Velocity = in-transformed average velocity.
variables do improve predictive models for richness of
fluvial-dependent species (or probability that a site
scores as impaired) implies that decisions concerning
how to supply water for offstream uses will have
measurable consequences for biotic integrity, even
though other landscape factors may add to or modify
those effects.
Model selection uncertainty and sampling error
also contributed to uncertainty in estimated effects of
water supply variables on stream fishes. For example,
models predicting FS richness as functions of drain-
age area and reservoir presence were weighted almost
as highly as the best-supported model, which also in-
cluded relative rate of permitted water withdrawal. We
could not entirely separate the effects of reservoirs
from withdrawals in this study because higher vvith-
drawal rates tended to be associated with reservoirs
(although not absolutely, as one of the highest rela-
tive withdrawals was at a stream intake). Model aver-
aging increased estimated standard errors, but
allowed us to estimate effects of withdrawals and res-
ervoirs while incorporating alternative hypotheses
(e.g., that either or both factors were responsible for
decreased FS richness).
Management Implications
Understanding potential effects of alternatives for
landscape development on native biological commu-
nities is fundamental to making informed manage-
ment decisions. This study has provided information
on relations between increasing the allocation of
water for offstream use, and of supplying water
through creation of instream reservoirs, on species
richness of native fishes in Piedmont streams. Across
our study streams, water supply variables appear more
predictive of species richness than catchment urban-
ization or average bed sediment size, although the
effects are not precisely known. How could decision
makers best use this information? Specifically, ques-
tions are posed such as: What minimum flow levels
will protect aquatic ecosystems? What levels of with-
drawal are compatible with maintaining integrity of
aquatic ecosystems? And, given options such as
building large, regional supply reservoirs versus
using multiple smaller reservoirs or withdrawals from
unimpounded streams, which provide the best bal-
ance of maximizing water supply and conserving
ecosystem integrity?

Water Withdrawals, Reservoirs, and Stream Fishes 447
The question of protective minimum flow levels
remains unanswered, except to note that there is no
evidence that providing for a minimum flow of 7Q10
protects stream fish assemblages, either from our data
or more generally (Stalnaker and others 1995). Higher
minimum flow provisions may mitigate some effects of
withdrawals and reservoirs, but only if periodic low-flow
depletion is the primary pathway by which hydrologic
alteration influences stream biota. If biotic integrity is
diminished by flow reduction during periods of nor-
mally higher base flows, then requiring a protected
minimum flow level will be insufficient to protect
stream ecosystem integrity (Poff and others 1997;
Richter and others 1997).
The results of this study support two hypotheses that
could be applied, tested, and refined through the
process of developing water supply in the rapidly
growing regions of the eastern United States. Our re-
sults indicate that (1) increasing permitted water
withdrawal levels is likely to result in local loss of stream
fish species, specifically fluvial-dependent species, and
(2) construction of instream water supply reservoirs is
similarly likely to result in reduced richness of fluvial-
dependent species. Based on our data, streams in the
lower Piedmont may begin to experience species losses
if permitted withdrawal exceeds about 0.5 to one 7Q10-
equivalent of water. Additional research to broaden the
geographic scope and size of the data set could im-
prove our ability to predict effects of water withdrawal
and use of reservoirs on stream fishes. However, given
the ecological complexity of stream systems, i.e.,
structural uncertainty (Williams and others 1996), and
the difficulties in precisely quantifying the richness and
abundance of many stream species (i.e., partial
observability), there likely always will be considerable
uncertainty when predicting the effects of any given
withdrawal or reservoir on stream biota.
Taking an adaptive management approach (Walters
1986) to future water supply development could allow
communities to meet their water needs while working
with managers and regulators to conserve the biologi-
cal diversity native to a region's streams. Conroy and
others (2003) argue that, given uncertainty and the
high ecological stakes of current development trajec-
tories in areas such as the southern Piedmont, man-
agement should be based on adaptive decision making
utilizing predictive models that relate policy decisions
to integrity of stream ecosystems at varying scales.
Estimates such as generated in this study could be
useful in beginning this process. For example, in the
Piedmont region of Georgia, managers could use our
results to hypothesize that withdrawals exceeding a gi-
ven rate are likely to result in species losses, and that
supplying water by way of multiple, dispersed with-
drawals capped below that level will have fewer effects
than concentrating supply at large withdrawals or in-
stream reservoirs. Decision makers could identify
streams that appear "over-allocated" with respect to
supporting native fishes, and test and refine this
hypothesis, as well as identifying areas within basins
where further allocation is likely to lead to faunal de-
cline. Decision makers could also evaluate alternative
supply scenarios with respect to predicted biological
effects given this hypothesis, preferably in the context
of a predictive model incorporating other influences
such as changing land use. Decisions regarding indi-
vidual projects will be influenced by multiple factors,
including the presence of rare or imperiled stream
biota (e.g., species protected under the Endangered
Species Act or Georgia's Endangered Wildlife Act) and
economic considerations, but whatever decisions are
made, one could predict effects on biological integrity
in the affected stream systems. Importantly, monitor-
ing stream biota before and after implementation of
new withdrawals could then test those predictions, with
the results used to improve our understanding of
relations between withdrawals, water supply reservoirs,
land use change, and stream biota. Applied at a re-
gional scale, water supply development could be plan-
ned to avoid excessive depletion and fragmentation in
stream systems critical for supporting unique faunal
assemblages. This approach would differ from that
currently taken by shifting the emphasis from mini-
mum flow policy and provisions, to the adaptive
development of water supply strategies that conserve
biological resources.
Acknowledgments
We appreciate field assistance provided by John
Seginak, USGS, and the following students and
technicians from the University of Georgia: Richard
Weyers, Shane Hawthorne, Peter Esselman, Jesslyn
Shields, Patricia Rodriguez, Sandra Helms, Leif
Stephens, Judith Barkstedt, Clay Brady, Erica Curry,
Megan Hagler, Sarah McClurg, Casey Storey, Dean
Sedgewick, John Knight, and Allison Roy. Bud Free-
man (UGA) and John Biagi and Patti Lanford (GDNR)
provided logistic support and data for reference sites.
This research was funded through the USGS State
Partnership Program and by additional funding
supplied by the GA Department of Natural Resources
to the University of Georgia. The manuscript was sub-
stantially improved by comments from Jonathan
Kennen, Mike Harris, Brett Albanese, Lance Williams,
and three anonymous reviewers.

448
M. C. Freeman and P. A. Marcinek
-d c~cnc~cncncnc; chc'~ cn c~chonchchcn cnc~c~
v o 0 0 0 0 o c o 0 o c^ o 0 0 0 0 0 0
0 0 0 0 0 o c o 0 0 0 0 o c o c c o o
LL ct cr cv cv cv ct c~ ~~ ct ti ~t ct cu ct ct ct ct cv cv
v o 0 0 0 0 0 0 o c c o 0 0 0 o c c c o c o 0 o c o 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 o c o 0 0 0 0 0 o C o 0 0
cl V 1 CV CV GV 'V GV GV GV V 1 GV GV ',~V GV GV GV .^I ^1 GV GV ,^i GV CV ~l GY GV CV
~ O C O O O C O O C O O O C O O O C C C O C O O O C O O .-~
~ O O O O O C O O C C O O C O O O C C C O C O O O C O C O
O^ O O O C O O C O O O C C O O C O C O C O O O O O C O
~" CV C4 GV GV cl c t GV GV GV ti CI CV GV GV GV cl GV .^I ti Pl G4 LY CV G`1 cl CV GV GV
O
C. y 00/ O
C v ~ ¢~ 6) L W 6t 51 ~ C 7J 7J O 61 O C O C O C~ C C CJ ~ ¢~ p 0 W
',. ~ C G C C C C C C C .-+ C C .. -y ~r C C C C
~ co 0 0 0 0 0 0 0 Or o o pr c pr o pr pr pr pr -• p1 Or o 0 o Or pr o c
-:zzzzZZZi`zzt`2i`zi`rni`cic~i`ZZ2i`i`z ~
r
~ ~
p .,~oocv.-,w.-~r„-.o~oco.. ..oi`ctcta~cu.,~cnooc~-, ,-. .•
a ....
s yb
b~
0
;~ E ~
a, ccd~ocvcncn~nctccoc~aococ~chc~ ct~~nd~orc~cci`.nct ~.
~ x .n .~, ct oo ,-, ct rn .n cn o x t` ~ o a; .-, a~ ~ ., c o cn c a! r ~
ro oc~i~-:o.:o~ oor'aoociocaid~ov~cioo ci E
~ i i i i i i i i i i i i i i i i i i i i i i i i i i c~
a
~ u
;; x
p ~Ca
x d~ctchcrocucuctacat d~cu..o~rcncc~n°p~c;~r ~i`chcr u
b b ~ $
~ ~ ..d~c~ooa:aio~d~cn~!o..cno~nt~cnC.j~~r;~cC'~ cnchcht~ ~,ti
oc~io~o-.-.c o'•.:ooooo~i-~-.':v~ci ~ cioc<io ~„ ~
3 ~ ~~
~b
-o ~ ro
~ W
,~ E
~ ocooo~noocooooooocoooccoo.nooo~° ~
p ~ ooowcti`~n~ncq.c~~rc~oo-.~c:~cocooooi`-,~.n A b
c 3 -:ci-:o~..o,-;.;woci.:..,-;~':..oociaiooocioc~i..o ~c
b ~
~ ~ ~
v v
o c
c .~
v~ ~
~~o^ooooooccoocoooooooccoococo
~ ~i`i`rcr,o:-+roo~n~ogoo-~oc~.r,cio~rco~n~noo~noi`r 3 ~
~ 'x ao ~ ~ r .; ci ai r of .; od ~ of o0 06 cd cD .; co co ad cc ai r3 ,ti
ch .., ,-, d+ o ct ct co c oo .~ oo rn cn ct oo cC w cn a t` .n .-, .n d~ ~ ...
Q ~ ~
4-
3 'o 'o 'o 'c '"o 'o 'o 'o 'o 'o 'o 'o 'o 'o o° o E
L° " .x .x'~.x ~'~,Y,x~,s'~ Z~x ~.xx ~ ~~x ~ ~ s;'s,x ~~ Co°o 'a E
.'OrC ~ at at at v y H ~ ~ at at at v v a~ o a w 7
~ ~ L ~ 67 U FCi .C-i ~ .[-i ~ L ~ r[r ~ r~i r~r ~ ~ ~ ~ 5.7 C7 7J 6) rti C ~ ~ .C+ U E
CG CG OG GG C4 P4 CG d. a; cG a! cs' a ~ ., c .5 4 'J .
3 0~ E E E
'~.
d w ::
~. b
U
~ L
a~ ~ .-. ~1 c~ d~ Lry cD t~ a0 O~ O .• GY a; d~ u~ c0 1~ oC U: O GV ch mot' u~ c0 I~ 00 `~ be y 'O a~
G r-I r-1 ~--1 r-1 r-1 r--I r-1 r-1 r~l ~-~I ~t ~t ~t ~t ~t ~t ~t ~t ~t L ~ 0. a_
C ~ '° p, a
w oo .~ ,,
y~~L~
~ ~' `~ [ i a
~ ~ ~ ~a a
u y`t ~ x .~C w w C ~ y
4., ~ 0.~ C ~ .x a~ y x x u° ~
~a -~ ~~ a~ x v x ~~ bb~yo
y ~ ~ v U U c c °
,ro L sC, L sue. .r' tC v ~. F. ~ Ci ~ y L 'O v
p Cx~ Cx~ OX L ~U > ~ ~.~ Jx.x~w,C s 3 3 ~ °~ ~
v ~ ~ v X ~ ~ ~ v ~ " yv, ~ v 3 E ~ C C7i ~ a~ „~, ~ v c ~ ''' '~ ~ ~ ,i°, b ~ 7
E C4 ~ ` o L ~ ` ~ a' U U ~ o `~ ~' ~ cc v; v ~ U U s. C ~ v ~ b ~ ~ ~ a
U W C~ C.7 -O Abdo ~ ~ U U ~ w ~ ,J e ;;
v ~ b 'C s. C~ 3 v .a ~~ ro~ 'y ~ 7 L 3 ~[~~ ti~ V C~ .~ v u x v
~¢~c~zv~a~H~n°.3::»co~xt°w~=t `3~.°,wxa°~c~~w° ~'~~a
N
Q
cv
c`o
T
C
(tl
rn
U
U
c`
crs
L
U
.~
(~
3
N
L

X
-0
Q
a
Q

Water Withdrawals, Reservoirs, and Stream Fishes
449
Appendix 2. Taxa comprising multiple species for use in NMDS analysis
Taxon name Species included
Cyprinella spp. Cyprinella callisema, C. callitaenia, C. nivea, C. venusta, C. xaenura
Hybopsis spp. Hybopsis rubrifrons, H. sp. cf. H. winchelli
Notropis spp. Notrofiis cummingsae, N. hypsilefris, N. longirostris, N. pelersoni, N. texanus
Scartomyzon spp. Scartomyzon brasseus, S. lachneri, S. rupiscartes
Moxostoma spp. Moxostoma collapsum, M. robusturn, M. sp. cf. M. fioecilurum
Noturus spp. Noturus gyrinus, N. insignis, N. lefilacanthus
Ameiurus' spp. Ameiurus talus, A. melas, A. nebulosus
Gambusia spp. Gambusia affinis, G. holbrooki
Pomoxis spp. P. annularis, P. nigromaculalus
Micropterus spp. Microfiterus cataractae, M. coosae, M punctulatus
F.theostoma spp. Elheostoma inscriptum, E. fusiforme, E. hopkinsi, E. olmstedi, E. swaini
Literature Cited
Abe11, R. A., D. M. Olson, E. Dinerstein, P. T. Hurley, J. T.
Diggs, W. Eichbaum, S. Walters, W. Wettengel, T. Allnutt,
C. Loucks, and P. Hedao. 2000. Freshwater ecoregions of
North America: a conservation assessment. World Wildlife
Fund, Washington, DC, 319 pp.
Allison, P. D. 1991. Logistic regression using the SAS system:
theory and application. SAS Institute Inc., Cary, North
Carolina, 288 pp.
Armstrong, D. S., T. A. Richards, and G. W. Parker. 2001.
Assessment of habitat, fish communities and streamflow
requirements for habitat protection, Ipswich River, Massa-
chusetts, 1998-99. US Geological Survey Water Resources
Investigations Report 01-4161.
Barbour, M. T., J. Gerritsen, B. D. Snyder, and J. B. Stribling.
1999. Rapid bioassessment protocols for use in streams
and wadeable rivers: periphyton, benthic macroinverte-
brates and fish, 2nd ed. EPA 841-B-99-002. Office of Wa-
ter, U. S. Environmental Protection Agency, Washington,
DC.
Baron, J. S., N. L. Poff, P. L. Angermeier, C. N. Dahm, P. H.
Gleick, N. G. Hairston Jr., R. B. Jackson, C. A. Johnston, B.
D. Richter, and A. D. Steinman. 2002. Meeting ecological
and societal needs for freshwater. Ecological Applications
12:1247-1260.
Baxter, R. M. 1977. Environmental effects of dams and
impoundments. Annual Review of Ecology and Systematics
8:255-283.
Bunn, S. E., and A. H. Arthington. 2002. Basic principles
and ecological consequences of altered flow regimes
for aquatic biodiversity. Environmental Management 30:492-
507.
Burnham, K. P., and D. R. Anderson. 2002. Model selection
and mttltimodel inference: a practical information-theo-
relic approach. Springer-Verlag, New York, 488 pp.
Burnham, K. P., and W. S. Overton. 1979. Robust estimation
of population size when capture probabilities vary among
animals. Ecology 60:927-936.
Carter R. F., E. H. Hopkins, H. A. Perlman. 1986. Low-flow
profiles of the upper Ocmulgee and Flint rivers in Georgia.
US Geological Survey Water Resources Investigations Re-
port 86-4176.
Carter, R. F., E. H. Hopkins, and H. A. Perlman. 1988a. Low-
flow profiles of the upper Savannah and Ogeechee rivers
and tributaries in Georgia. US Geological Survey Water
Resources Investigations Report 88-4047.
Carter, R. F., E. H. Hopkins, and H. A. Perlman. 19886. Low-
flow profiles of the upper Oconee River and tributaries in
Georgia. US Geological Survey Water Resources Investiga-
tions Report 88-4048.
Clarke, K. R. 1993. Non-parametric multivariate analysis of
changes in community structure. Australian Journal of Ecol-
ogy 18:117-143.
Collares-Pereira, M.J., I. G. Cowx, F. Ribeiro, J. A. Rodrigues,
and L. Rodago. 2000. Threats imposed by water resource
development schemes on the conservation of endangered
fish species in the Guadiana River Basin in Portugal. Fish-
eries Management and Ecology 7:167-178.
Collier, M., R. H. Webb, and J. C. Schmidt. 1996. Dams and
rivers: primer on the downstream effects of dams. USGS
Circular 1126. US Geological Survey, Tuscon, Arizona.
Conroy, M. J., C. R. Allen, J. T. Peterson, L.J. Pritchard, and
C. T. Moore. 2003. Landscape change in the southern
Piedmont: challenges, solutions, and uncertainty across
scales. Conservation Ecology 8:3. [online]URL:http//
www. consecol. org/vo18/ iss2/art3.
Cooperman, M. S., and D. F. Markle. 2003. The Endangered
Species Act and the National Research Council's interim
judgment in Klamath Basin. Fisheries 28:10-19.
Copp, G. H. 1990. Effect of regulation on 0+ fish recruitment
in the Great Ouse, a lowland river. Regulated Rivers: Research
and Management 5:251-263.
Fausch, K. D., C. E. Torgersen, C. V. Baxter, and H. W. Li.
2002. Landscapes to riverscapes: bridging the gap between
research and conservation of stream fishes. Bioscience
52:483-498.
Field, J. G., K. R. Clarke, and R. M. Warwick. 1982. A practical
strategy for analyzing multispecies distribution patterns.
Marine F,cology Progress Series 8:37-52.
GDNR (Georgia Department of Natural Resources). 2000.
Standard operating procedures for conducting biomoni-
toring on Fish communities in the Piedmont ecoregion of
Georgia. Georgia Department of Natural Resources, Wild-
life Resources Division, Fisheries Section, Social Circle,
Georgia.

450 M. C. Freeman and P. A. Marcinek
Gordon, N. D., T. F. McMahon, and B. L. Finlayson. 1992.
Stream hydrology: an introduction for ecologists. John
Wiley & Sons, Chichester, 526 pp.
Jackson, R. B., S. R. Carpenter, C. N. Dahm, D. M. McKnight,
R. J. Naiman, S. L. Postel, and S. W. Running. 2001. Water
in a changing world. Ecological Afiplications 11:1027-1045.
Karr, J. R., and E. W. Chu. 2000. Sustaining living rivers.
Hydrobiologia 422/423:1-14.
Kinsolving, A. D., and M. B. Bain. 1993. Fish assemblage
recovery along a riverine disturbance gradient. Ecological
Apfilications 3:531-544.
Matthews, W. J. 1990. Spatial and temporal variation in fishes
of riffle habitats: a comparison of analytical approaches for
the Roanoke River. American Midland Naturalist 124:31-45.
Matthews, W. J. 1998. Patterns in freshwater &sh ecology.
Chapman and Hall, New York, 756 pp.
McCune, B., and M. J. Mefford. 1999. PGORD. Multivariate
analysis of ecological data, Version 4.25. MjM Software
Design, Gleneden Beach, Oregon.
McCune, B., and J. B. Grace. 2002. Analysis of ecological
communities. MjM Software Design, Gleneden Beach,
Oregon, 300 pp.
Moyle, P. B., and R. A. Leidy. 1992. Loss of biodiversity in
aquatic ecosystems: evidence from fish faunas. Pages 127-
169 in P. L. Fielder and S. K. Jain (eds.), Conservation
biology: The theory and practice of nature conservation,
preservation, and management. Chapman and Hall, New
York.
Naiman, R. J., S. E. Bunn, G. Nilsson, G. E. Petts, G. Pinay,
and L. C. Thompson. 2002. Legitimizing fluvial ecosystems
as users of water: an overview. Environmental Management
30:455167.
Oksanen, J., R. Kindt, and R. B. O'Hara. 2005. Vegan: Com-
munity Ecology Package. Available from http://cc.oulu.fi/
jarioksa/.
Peterson, J. T., and C. F. Rabeni. 1995. Optimizing sampling
effort for sampling warmwater stream fish for communities.
North America Journal of Fisheries Management 15:528-541.
Poff, N. L., J. D. Allan, M. B. Bain, J. R. Karr, K. L.
Prestegaard, B. D. Richter, R. E. Sparks, and J. C. Strom-
berg. 1997. The natural flow regime. Bioscience 47:769-784.
Postel, S. 1996. Dividing the waters: food security, ecosystem
health, and the new politics of scarcity. Worldwatch Paper
132. Worldwatch Institute, Washington, DC, 76 pp.
Postel, S., and B. Richter. 2003. Rivers for life: managing
water for people and nature. Island Press, Washington, DC,
253 pp.
Postel, S. L. 2000. Entering an era of water scarcity: the
challenges ahead. Ecological Applications 10:941-948.
Richter, B. D., J. V. Baumgartner, R. Wigington, and D. P.
Braun. 1997. How much water does a river need? Freshwater
Biology 37:231-249.
Roy, A. H., M. C. Freeman, B.J. Freeman, S.J. Wenger, W. E.
Ensign, and J. L. Meyer. 2005. Investigating hydrologic
alteration as a mechanism of fish assemblage shifts in
urbanizing streams. Journal of the North American Benthologi-
cal Society 24:656-678.
Scott, M. C., and G. S. Helfman. 2001. Native invasions,
homogenization, and the mismeasure of integrity of fish
assemblages. Fisheries 26:6-15.
Silk, N., J. McDonald, and R. Wigington. 2000. Turning in-
stream flow water rights upside down. Rivers 7:298-313.
Snijders, T. A. B., and R. J. Booker. 1999. Multilevel analysis:
an introduction to basic and advanced multilevel model-
ing. Sage Publications, London, 266 pp.
Stalnaker, C., B. L. Lamb, J. Henrickson, K. Bovee, and J.
Bartholow. 1995. The Instream Flow Incremental Meth-
odology, aprimer for IFIM. Biological Report 29. US Na-
tional Biological Service, Washington, DC.
Travnichek, V. H., M. B. Bain, and M. J. Maceina. 1995.
Recovery of a warmwater fish assemblage after initiation of a
minimum-flow release downstream from a hydroelectric
dam. Transactions of the American Fisheries Society 124:836-844.
Walters, C. J. 1986. Adaptive management of renewable re-
sources. Macmillan, New York, 374 pp.
Walters, D. M., D. S. Leigh, and A. B. Bearden. 2003a.
Urbanization, sedimentation, and homogenization
of fishes in the Etowah River basin, USA. Hydrobiologia
494:5-10.
Walters, D. M., D. S. Leigh, M. C. Freeman, B. J. Freeman, and
C. M. Pringle. 2003b. Geomorphology and fish assemblages
in a Piedmont river basin, U.S.A. Freshwater Biology 48:1950-
1970.
Walters, D. M., M. C. Freeman, D. S. Leigh, B.J. Freeman, and
C. M. Pringle. 2005. Urbanization effects on fishes and
habitat quality in a southern Piedmont river basin. Pages
69-85 in L. R. Brown, R. H. Gray, R. M. Hughes and M. R.
Meador (eds.), Effects of urbanization on stream ecosys-
tems. American Fisheries Society Symposium 47, Bethesda,
Maryland.
Ward, F. A., and J. F. Booker. 2003. Economic costs and
benefits of instream flow protection for endangered species
in an international basin. Journal of the American Water Re-
sources Association 39:427-440.
Warren, M. L., Jr., B. M. Burr, S.J. Walsh, H. L. Bart Jr., R. C.
Cashner, D. A. Etnier, B. J. Freeman, B. R. Kuhajda, R. L.
Mayden, H. W. Robison, S. T. Ross, and W. C. Starnes.
2000. Diversity, distribution, and conservation status of the
native freshwater fishes of the southern United States.
Fisheries 25 (10) :7-29.
Weaver, L. A., and G. C. Garman. 1994. Urbanization of a wa-
tershed and historical changes in a stream fish assemblage.
Transactions of the American Fisheries Society 123:162-172.
Williams, B. K., F. A. Johnson, and K. Wilkins. 1996. Uncer-
tainty and the adaptive management of waterfowl harvests.
Journal of Wildlife Management 60:223-232.
Williams, B. K., J. D. Nichols, and M.J. Conroy. 2002. Analysis
and management of animal populations. Academic Press,
San Diego, 817 pp.