STREAMFLOW VARIABILITY, FISH COMMUNITY STRUCTURE, AND
IMPLICATIONS FOR CLIMATIC CHANGE
hy
N. LeRoy Poff
Department of Zoology
University of Maryland
College Park, Maryland 20742
and
J. David Allan
School of Natural Resources & Environment
University of Michigan
Ann Arbor> Michigan 48109
CR-816540010
Project Officer
J. David Yount
Environmental Research Laboratory - Duluth
6201 Congdon Boulevard
Duluth, Minnesota 55804
This study was conducted in cooperation with the U.S.
Environmental Protection Agency
ERL-D
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
. DULUTH, MINNESOTA 55804
-------
GOO19943
ABSTRACT
Relatively undisturbed U.S. streams were classified according to
variation in 11 ecologically-relevant hydrologic characteristics. A group of
420 "best" stream sites and a group of 816 "acceptable" sites (including the
420 best sites) were evaluated with similar results. Cluster analysis
resulted in the identification of 10 distinctly different stream types, seven
perennial and three intermittent. Most of these stream types exhibit
reasonable geographic affiliation and can be interpreted in terms of regional
climatic patterns and local variation in geologic characteristics. The
classification provides a comprehensive catalog for identifying streams that,
according to ecological theory, may differ in major aspects of ecological
organization. It further offers a basis for hypothesis-generation and affords
an objective framework for matching streams for purposes of comparative
ecological investigations. Moreover, the streamflow classification can be
used to assess the potential ecological consequences of hydrologic changes,
because specific kinds of hydrologic change induced by climatic changes (e.g.,
increased intermittency, reduced flood seasonality, etc.) can be compared to
the historical hydrologic patterns summarized in this classification.
The sensitivity of measures of streamflow predictability and of high
flow disturbance regime to variation in the time scale of analysis is
investigated for the classified stream types. As the temporal scale was
changed from daily to weekly to monthly to seasonal, six of the 12 stream
types used in the analysis became more predictable, four less predictable, and
two showed no change. For analysis of the high flow disturbance regimes
across different stream types, monthly and annual data are not capable of
capturing the information available in the daily hydrograph for most stream
types. These results indicate the importance of regional climatic conditions
and local catchment characteristics in influencing the calculation of
predictability and high flow regimes at different time scales.
Stream fish assemblage data were analyzed for 34 sites in Wisconsin and
Minnesota where long-term hydrologic data exist. Assemblages were analyzed in
terms of both taxonomic and functional organization, and then related to
independent hydrologic factors using multivariate statistical techniques. The
taxonomic analysis showed strong geographic patterns among taxonomically-
similar groups that reflected an interaction of species zoogeography and
hydrologic regimes of surveyed streams. The functional analysis of the 34
assemblages revealed that two or three groups of sites could be defined in
terms of functional organization (i.e., body morphology, trophic guild,
habitat preferences, and tolerance values). These functionally-similar groups
were strongly correlated with independent hydrologic factors that differed
significantly among the 34 sites. Fish assemblages defined in functional
terms could be assigned to hydrologically variable, stable, and very stable
sites. These hydrologic-comraunity relations suggest that climatic change
which alters hydrologic regimes in this region can modify stream fish
assemblage structure. Community changes are more likely to be detected using
a functional rather than a purely taxonomic perspective.
-------
ACKNOWLEDGMENTS
We would like to thank several individuals and organizations for
contributing to this project. Nancy Roth of the School of Natural Resources,
University of Michigan, Ann Arbor, reviewed the literature and helped to
summarize the trophic and habitat categorizations used in the report. Ors.
Gerald R. Smith (Museum of Zoology, University of Michigan, Ann Arbor) and
Paul S. Seelbach (Institute for Fisheries Research, Michigan Department of
Natural Resources) offered their expertise by reviewing the trophic and
habitat categorizations. Dr. Paul Webb (School of Natural Resources,
University of Michigan, Ann Arbor) assisted us in identifying measures of body
morphology relevant to hydrologic regime. Dr. Daniel Denman (University of
Maryland, Department of Statistics) was provided statistical expertise and was
instrumental in developing the bootstrapping technique for testing stream
cluster coherence.
Several people associated with the Environmental Research Laboratory in
Duluth were also instrumental in the successful completion of this report.
Mr. Howard McCormick of ERL-D contributed to extraction of body morphological
measurements from the literature. Dan O'Brien and James Westman of the AScI
Corporation were essential in building the biological database and in
generating the critical geographic information systems maps. Finally, we
would like to extend special thanks to the Project Officer, Dr. J. David
Yount, who provided persistent and thoughtful guidance and who fostered a
cooperative working relationship between us and the ERL-D staff.
iii
-------
-------
CONTENTS
Abstract ii
Acknowledgment iii
Section 1. A Streamflow Classification of U.S. Streams and Rivers, and
its Implications for Climate Change Research
1. Introduction 1
2. Methods 3
Site Selection 3
Definition of Flow Variables 5
Statistics 9
3. Results 11
Statistical Correlations among Variables 13
Cluster Results: Statistical Relations 14
Cluster Results: Geographic Distribution 19
Cluster Stability 20
4. Discussion 21
5. Conclusions 25
Section 2. Importance of Temporal Scale in Assessing Streamflow
Predictability and Flood Regime in Streams from Different
Geographic Areas
1. Introduction 27
2. Methods 30
Colwell's Index 30
Temporal Scale and Spates 31
3. Results 33
Colwell's Index 33
Temporal Scale and Spates 36
4. Discussion 38
5. Conclusions 40
Section 3. Fish Community Structure along Hydrologic Gradients in
Wisconsin and Minnesota Streams, and some Implications
for Community Response to Climate Change
1. Introduction 42
2. Methods .46
Fish Data 46
Derivation of Functional Measures 47
Hydrologic Data . 51
Data Analysis 53
3. Results 59
Hydrologic Variables 59
Fish Species Occurrences 60
Fish Taxonomic Relations and Hydrology ... 60
Fish Functional Relations and Hydrology 70
Robustness of Multivariate Results 74
Functional Organization of Assemblages 75
4. Discussion 81
5. Conclusions and Recommendations 88
References 90
Figures 104
Appendices
A. Listing of 816 Sites for Streamflow Classification 173
B. Computer Program Code Used to Derive Streamflow Statistics . 189
C. Listing of Statistics Used in Streamflow Classification . . 265
D. Fish Species' Functional Attributes 309
-------
SECTION 1
A STREAMFLOW CLASSIFICATION OF U.S. STREAMS AND RIVERS,
AND ITS USEFULNESS FOR CLIMATE CHANGE RESEARCH
INTRODUCTION
Streamflow is arguably the most characteristic physical attribute of
stream ecosystems, and it plays a central role in stream ecology (see Hynes
1970). Because many important structural attributes in streams, such as
habitat volume, current velocity, channel geomorphology and substrate
stability, are under the direct influence of streamflow, measurement of
streamflow can represent an integration of complex environmental conditions
that are individually difficult to quantify.
Global climate change is expected to alter large-scale patterns of
precipitation, which will in turn affect lotic ecosystems by modifying
streamflow regimes (Poff 1992). To the extent that species distributions and
abundances reflect physical streamflow patterns, climate-induced changes could
potentially affect lotic ecosystems throughout the entire United States. It
is thus imperative that the existing relationships between physical
characteristics of lotic ecosystems and ecological patterns be established in
order to provide a baseline frame of reference for future interpretation of
impacts attributed to climate change. A necessary first step in this process
is the identification of currently existing hydrological characteristics that
constrain the distributions and abundances of species at regional scales.
This requires that hydrologic data be expressed explicitly in terms of
environmental selective forces for lotic biota.
In streams, flow fluctuations and extreme conditions such as floods and
low flow are primary sources of environmental variability and disturbance (cf.
Stanford and Ward 1983) . Patterns of diversity of all major lotic
assemblages, including fish (Seegrist and Gard 1972, Harrell 1978, Horwitz
1978, Minckley and Meffe 1987), invertebrates (Vannote et al. 1980, Ward and
Stanford 1983, Bournard et al. 1987), attached algae (Patrick 1975, Peterson
1987, Power and Stewart 1987) and macrophytes (Haslam 1978, Ladle and Bass
-------
1981) have been related to patterns of temporal variation in flow. Moreover,
there is a substantial body of evidence indicating that both high flow (flood)
and low flow (intermittency) disturbances play a central role in structuring
stream communities (Hynes 1970, Williams and Hynes 1976, 1977, Iversen et al.
1978, Fisher 1983, Stanford and Ward 1983, Ward and Stanford 1983, Schlosser
1987, Delucchi 1988, Minshall 1988, Power et al. 1988, Resh et al. 1988).
Geographic variation in streamflow patterns can be evaluated in many
ways, but one of the most useful is classification, where similar "types" of
hydrologic regimes are identified and associated. The scale at which
streamflow classifications have been previously constructed range from
regional (Gentilli 1952, Dewberry 1980, Alexander 1985, Hughes and James 1989,
Jowett and Duncan 1990, Richards 1990) to continental (Grimm 1968, Poff and
Ward 1989) to global (Beckinsale 1969, McMahon 1979, 1982, Haines et al.
1988) .
Different combinations of streamflow variation (e.g., range and
predictability), patterns of flooding (e.g., frequency and predictability) and
extent of intermittency presumably result in different degrees of physical
control over biotic organization {Minshall 1988, Resh et al. 1988, Poff and
Ward 1989, 1990). Most comparative geographical studies have considered only
a few measures of flow variability, e.g., mean flow conditions (Hawkes et al.
1986, Moss et al. 1987, Townsend et al. 1987), variation about the mean flow
(Horwitz 1978), short-term estimates of flood frequency (Gushing et al. 1980,
1983, Minckley and Meffe 1987, Fisher and Grimm 1988), and predictability of
monthly flow patterns (Resh et al. 1988, Bunn et al. 1986). Fewer comparative
studies have considered several hydrologic factors simultaneously (Poff and
Ward 1989, Hughes and James 1989, Jowett and Duncan 1990).
There are several reasons to classify streamflow regimes with an eye
toward the potential impacts of climate change. First, it is necessary to
establish historical (background) patterns in hydrologic regimes to establish
a baseline against which future changes must be measured. The importance of
establishing baseline hydrologic patterns prior to potential climate change
from a general water resources perspective has been recently emphasized
(Wallis et al. 1991, Dolph and Marks 1992). Second, spatial distribution of
hydrologic regimes may help us to develop testable hypotheses about
environmental constraints on regional processes and patterns in lotic systems
(cf. Resh et al. 1988, Poff and Ward 1989, Minshall 1988). Placing stream
-------
ecology into a regional context is needed to consider potential impacts of
climate change (Poff 1992, Grimm 1992, Carpenter et al. 1992). Third, a
hydrogeographic classification can help to identify specific regions that are
especially sensitive to potential alterations in amount and/or timing of
precipitation brought on by climate change. Thus, we may infer something
useful about what specific types of changes may be most important ecologically
in different regions, which differ in historical (i.e., pre-climate change)
hydrology. For example, desert streams are very sensitive to increased
aridity (Dahm and Holies 1992, Grimm and Fisher 1992) while more baseflow-
driven systems may be buffered against aridity but are potentially sensitive
to increased flow variability (Poff 1992). We need to know the geography of
such ecological sensitivities, because potentially valuable resource
management information can be revealed, as has been demonstrated when this
approach is applied to questions of water resource distribution (Gleick 1990) .
The objective of this research was to provide a large-scale framework to
allow classification of naturally-flowing streams and rivers on the basis of
ecologically-important hydrologic characteristics. An earlier analysis (Poff
and Ward 1989) used 78 streams to show that extensive hydrologic variation of
ecological interest exists for U.S. streams and rivers. This paper reports on
an extension of that earlier effort and it serves two major purposes. First,
the characterization of hydrologic similarity among all gauged, free-flowing
streams in the U.S. allows individuals/agencies interested in examining
hydrological-ecological relationships to identify sites that share
ecologically-important hydrologic regimes. Proper site-matching is a
necessary condition in any cross-community comparisons in streams (Resh et al.
1988, Fisher and Grimm 1991). Second, by examining all candidate streamflow
data available for the entire continental U.S., the paper provides a
comprehensive classification that characterizes "pre climate change"
hydrologic regimes at a continental scale.
METHODS
Site Selection
Data were acquired from a commercially available database (Earthlnfo
1990) that consists of a digital compilation of the U.S. Geological Survey
(USGS) daily and peak values files on CD-ROM. For each of the ca. 7000
-------
stations in the dataset, the "Remarks" section was read to determine if the
stream met certain specific criteria: (1) minimal or no flow regulation (e.g.,
diversion, damming, groundwater pumping), (2) minimal or no watershed
urbanization, (3) > 20 yr of continuous daily streamflow data, preferably
extending up through water year 1985, and (4) an accuracy rating of "good" or
better for the recorded flow values in the water years chosen. However, the
occurrence of a few days with "poor" or "fair" records (due to, for example,
ice blockage) was not sufficient reason alone to exclude the station. This
process generated ca. 2200 sites that potentially could be used for the
streamflow classification. Of these 2200 sites, ca. 1000 were of unknown
quality because there was no available description available for them in the
"Remarks" section of the Earth-Info database (usually because the gauge was
discontinued). This exhaustive list of potential sites was then compared with
two independently-derived datasets. First, Slack and Landwehr (1992)
identified 1659 stream gauging sites in the U.S. and its territories having at
least 20 yr of record extending through 1988 during which no "confounding
anthropogenic influences" occurred. These sites were identified based on
information collected from state water resource experts in the district
offices of the USGS, and they therefore represent a very reliable dataset
characterizing gauging station quality. However, not all sites identified by
Slack and Landwehr (1992) could be used for the streamflow classification,
because their criteria allowed inclusion of sites having only monthly flow
data and sites having significant flow impairment over the entire period of
record (FOR). For example, a regulated river with highly modified flow regime
could be included in their dataset if the extent of flow regulation had not
changed significantly over the FOR. A second dataset identifying reliable,
long-term stream gauging stations with unimpaired flow was put together by
Wallis et al. (1991). They selected sites based on readings of the "Remarks"
section in the Earth-Info database. They identified 1014 unregulated or
minimally regulated sites with daily flow records extending from 1948-1988.
Some of the sites used by Wallis et al. (1991) had gaps in the record, and
they estimated missing values with regression analysis, using data from nearby
gauges.
Comparison of the original 2200 sites with the two independent datasets
resulted in 812 sites matching the Slack and Landwehr (1992) dataset and an
additional 90 matching with the Wallis et al. (1991) dataset. Only sites
-------
having catchment area < 5000 km^ and estimated average flood durations < 15 d
were retained in the final dataset, which consisted ultimately of 733 sites
from Slack and Landwehr (1992) and 83 sites from Wallis et al. (1991) . This
final dataset (N = 816) was partitioned into two subsets: one containing 420
sites with FOR ending in 1985 and at least 36 yr in length, and a second
containing all 816 matched sites having at least 20 yr of continuous daily
flow data with the last year in the record not occurring before 1978. These
two subsets were analyzed separately to provide 1) a core classification of
the "best" gauged sites available (N = 420) in the U.S., and 2) a
classification of all "acceptable" sites (N = 816), which, while satisfying
less rigorous criteria, nonetheless represents the exhaustive classification
based on available data. The names and locations of these sites are provided
in Appendix A (data file on diskette).
Definition of Flow Variables
Several variables were defined for extraction from the long-term
hydrologic records. A computer program written in True-Basic language
(Appendix B, also included on diskette) to extract hydrologic variables. They
fell into four general categories (listed as I-IV below). The periods of
record for different streams varied in length, and this may have caused subtle
differences to arise in the values of derived statistics that were used in the
classification process. In order to account for this bias in length of
available record, many of these variables were also derived for a 20-yr period
as well as for the entire period of record. For sites exceeding 20 yr in
length of record, the period 1966-1985 was selected where possible so that all
sites shared the same short period. For sites with records terminating prior
to 1985, the most recent 20 yr period was selected, except for sites with
records terminating prior to 1978, when no short record was analyzed. The
values of the extracted variables for each site are included in Appendix C
(also on diskette).
Basin Descriptors
These are static measures indicative of stream/catchment size. These
variables are readily available from the "Remarks" section of the Earth-Info
CD-ROM database (an electronic facsimile to the gauging station summaries in
-------
the Water Supply papers published annually by the USGS) and contain no
information about hydrologic variability.
Basin Drainage Area (AREA, km^)The surface area of the catchment
topographically above the gauge elevation.
Daily Mean Discharge (QMEAN, m^ sec"1)--The average daily flow at the
site over all years in the record.
Mean Annual Runoff (MAR, mm yr"1)Ratio of QMEAN/AREA expressed as a
depth. MAR represents the difference between annual evaporation and
precipitation (Gordon et al. 1992). This index was suggested by Hughes and
Omernik (1983) as a substitute for stream order in classifying stream and
catchment size across different hydroclimatological regions.
In order to allow unbiased comparisons of hydrologic variability among
streams that vary greatly in catchment size and annual runoff, it was
necessary to "modularize" the data (Yevjevich 1972) by dividing each daily
flow value by QMEAN for the entire FOR. Thus, for each stream, the
modularized mean flow was 1.0.
Measures of Flow Variability and Predictability
These measures assess the degree of variation in the hydrologic signal
with no special treatment of extreme flow values.
Baseflow Index (BFI, %)For each year in the FOR, the ratio of the
lowest daily flow to the average daily flow indexes flow stability and
susceptibility to drying. The average of all the annual ratios was calculated
as the BFI and multiplied times 100.
Coefficient of Variation (DAYCVf %)This dimensionless index represents
the average (across all years) of the ratios between the annual mean daily
flow and the standard deviation of the daily flows, multiplied by 100 and
expressed as a percent. DAYCV describes overall flow variability without
considering the temporal sequence of flow variation.
Predictability of Flow (DAYPRED, %)DAYPRED was determined using an
index developed by Colwell (1974) which is based on information theory. When
expressed as a percent, this index ranges in value from 0 to 100 and is
composed of two independent, additive compoenents: constancy (C), a measure of
temporal invariance, and contingency (M), a measure of periodicity. The index
can be used to express the degree to which flow "states" (i.e., quantity of
discharge) are predictably distributed across specified time intervals.
-------
Predictability values are sensitive to definition of flow states, which are
ultimately arbitrary (Gordon et al. 1992). In this analysis, 11 categories
were defined with a Iog2 series with boundaries at 2~3, 2~^, 2'1, 2^, 21, 22,
2^, 24, 2^, and 2^ times modularized mean flow. Thus, the 11 flow states
ranged from < 12.5% of mean flow to >640% of mean flow. Predictability
measures are also sensitive to length of record. Gan et al. (1991) used
monthly flow data to show that short POR tend to yield overestimates of
predictability, whereas records in excess of ca. 40 yr yielded stable
estimates of predictability. Gordon et al. (1992) point out that more
comparative studies are needed to develop consistent methodologies in the
application of Colwell's index to ecologically-relevant hydrologic variability
(see Section 2 this report.)
High Flow Disturbance
This was defined as flows of magnitude exceeding the theoretically-
expected return interval of 1.67-yr based on a log-normal distribution. The
"annual peak flow series" for each station was used to determine these flood
values, because the peak series provides the maximum instantaneous (rather
than 24-hr) flow values that are needed to properly determine flood
frequencies. The peak values were assumed to represent a sample from a log-
normal distribution (see Dunne and Leopold 1978, p. 306); hence, by knowing
the mean and variance of the sample, one can calculate floods of specified
probability of occurrence (e.g., a 2-yr flood has a 50% probability of
occurring in any given year and is represented by the mean value of the annual
series on a logarithmic scale). A flow with a 1.67-yr return interval is
often recognized as "bankfull," but this may vary regionally and with climate.
The bankfull stage, according to Dunne and Leopold (1978, p. 608) corresponds
to the "discharge at which channel maintenance is most effective ... in
doing work that results in the average morphologic characteristics of the
channels." Thus, this level of flow can be considered a non-arbitrary index
of physical habitat disturbance in streams (see Poff and Ward 1989).
Flood flows exceeding the return intervals of 1.67 yr were determined to
encompass the range of potential bankfull discharge across many stream types.
(Ideally, geomorphic information is needed to determine the most appropriate
level of flow to be selected as the proper site-specific index of flood
disturbance.) After determining these threshold flow levels, the flood history
-------
for a site was determined by regressing the log of the peak flow values
against the log of the corresponding 24-hr mean flows (i.e., those occurring
on the same dates as the peak flows) so that "daily flood" values could be
determined. The long-term daily flow record was then analyzed with respect to
these "daily flood" values. Several measures were derived for the entire
period of record.
Inter-annual Variability (FLODVAR, dimensionless)The standard
deviation of the annual peak flow series on a log scale is a measure of
between-year variation in flood magnitude. This statistic was called the
Flash Flood Frequency Index by Beard (1975, see also Baker 1977, 1988), and it
can be used to assess stability in maximum annual flows.
Frequency (FLODFREQ, yr~^)The average number of discrete flood events
per year having a magnitude equalling or exceeding that associated with the
1.67 yr return-interval flood. The number of days that separate independent
flood events may vary geographically; therefore, a 10-d period separating
individual bankfull events was used as a criterion to identify separate
spates.
Duration (FLODDUR, d)The average number of days that flow remains
above the flood threshold for a site.
Seasonal Predictability of Flooding (FLDPRED, dimensionless)Maximum
proportion of all floods over the period of record that fall in any 60-d
"seasonal window." This index ranges from 0.167 ("random" flooding) to 1.0
(perfectly seasonally predictable) For this metric, the "partial duration
series" from the Earth-Info CD-ROM was used. All instantaneous flows > 1.67
yr return-interval flow in the period of record were ordered according to on
which day of the year they occurred and the temporal distribution of this
collapsed data set was analyzed for seasonal patterns. High flows occurring
within 60 days of the beginning or end of the water year were considered to
fall within the same "season".
.In addition, the day of the water year marking the beginning of the 60-d
period when FLDPRED was highest was recorded by the variable FLDTIME. This
variable was not used as a primary classification variable, but was used to
evaluate the range of timing of flood-onset within groups of hydrologically-
similar streams as identified by the cluster analysis.
Seasonal Predictability of Non-flooding (FLDFREE, dimensionless)
Maximum proportion of year (#days/365) during which no floods have ever
8
-------
occurred over period of record. Again, the partial series was used and no-
flood periods were allowed to pass through the end of one water year into the
beginning of the next.
Low Flow Disturbance
Low flows were characterized both by identifying periods of zero
discharge and by calculating site-specific lowflows of specified return
intervals. The latter was accomplished by taking the annual 1-day minimum 24-
hr low flow values for a station and assuming that they represent a sample
from a population with a Gumbel (extreme value) distribution (Linsley et al.
1982, p. 375). The parameters from this distribution were used to calculate
one-day low flows with various return intervals. The long-term daily record
was scanned to locate periods when low flows with > 5-yr recurrence intervals
occurred. Several variables were derived from this analysis.
Extent a£ Intermitteney (ZERODAY, d)Average annual number of days
having zero discharge.
Seasonal Predictability of Lowflow (LOWPRED, dimensionless)Proportion
of low flow events > 5-yr magnitude falling in a 60-d "seasonal window" (as
described above for flood predictability). Also, the variable LOWTIME was
derived to evaluate within-cluster variation in timing of lowflow-onset.
Seasonal Predictability of Non-lowflow (LOWFREE, dimensionless)Maximum
proportion of year (#days/365) during which no 5-yr+ low flows have ever
occurred over period of record.
Statistics
Basic relationships among hydrologic variables
Pearson correlation coefficients were derived for linear relations among
all 14 variable combinations. For the 11 hydrologic variables, the
correlation matrix was used as input into a principal components analysis
(PCA), a multivariate technique for examining relationships among several
quantitative variables. The goal of PCA is to derive a small number of linear
combinations of the original variables that retain the maximum possible amount
of information in the original variables (SAS 1988) . This procedure
represents a dimensional reduction of many variables to a few principal
-------
components, the interpretation of which is often facilitated by a further
transformation known as "rotation" (SAS 1988) .
Streamflow Classification
Several criteria were used to identify meaningful classifications of the
stream sites. First, three categories were defined based on a priori
ecological considerations: (1) "permanent" streams (ZERODAY < 10 days per
year); (2) "intermittent" streams {ZERODAY between 10 and 90 days, inclusive);
(3) "harsh" streams (ZERODAY > 90 days per year). These divisions, while
arbitrary, reflect the established ecological importance of flow permanence in
regulating lotic process and pattern (e.g., Grimm 1992, Ward 1992, Delucchi
1988, 1989, Valett and Stanley 1992)
The "permanent" group was a heterogeneous set of 383 data points,
representing streams with reasonably stable characteristics. To further
characterize these streams we applied a cluster analysis using the two-stage
density linkage method provided in SAS's PROC CLUSTER. This is a non-
parametric clustering, utilizing a k-th nearest neighbor criterion, which
seeks regions surrounding local maxima in the estimated probability density
function associated with a set of variables. Simulations suggest that the
method does reasonably well when the true clusters are known to be of unequal
size and variability, or when the clusters are irregularly (e.g. nonconvex)
shaped (SAS 1988) . Since we had no a priori expectations regarding the shape,
size, or dispersion of stream clusters, we felt that such a non-parametric
approach was reasonable. The two-stage density linkage method was also
applied separately to the intermittent and harsh streams.
Determining the number of clusters in an arbitrary data set is a problem
which lacks a clear statistical solution. Following the idea of Wong and
Schaack (1982), we applied the clustering algorithm with different values for
the nearest neighbor parameter. We looked for clustering solutions which gave
fairly consistent estimates of the number of modes of the distribution across
a range of parameter values, while at the same time yielding clusters with
clear interpretations. In this admittedly subjective process, we gave
precedence to scientific interpretability rather than to arbitrary criteria
based on unsupportable statistical assumptions. A distinct advantage of this
method versus other k-th nearest neighbor methods (e.g., K-means clustering
used by Poff and Ward 1989) is that individual sites are always assigned to
10
-------
the same cluster for a fixed sample size. K-means clustering produces
slightly different cluster memberships depending on the initial input order of
the sites.
Stream cluster stability
Given that the number of clusters was arbitrarily set, it is critical
that the stability of the putative clusters be examined. In order to avoid
reliance on parametric assumptions, a method based on bootstrapping, a
computer-intensive method of drawing repeated resamples from the original data
(Efron 1979), was developed by Dr. Daniel Denman (Department of Statistics,
University of Maryland). First, 383 resamples were drawn with replacement
(i.e., the same stream could be drawn more than once) from the original set of
383 points in the "permanent" stream dataset. Second, the two-stage density
linkage clustering algorithm was applied to identify 7 clusters from the "new"
collection of 383 points. Third, the set of resampled points was assessed in
terms of the number of points sharing membership both in a resample-derived
cluster as well as in an original cluster. Two indices were computed for the
resample. Mojr is the proportion of points in resample clusters which also
were together in original clusters, and it represents the extent to which
resample clusters represent the membership structure of the original clusters.
Mr|O is the proportion of points in original clusters which also were together
in resample clusters, and it represents the tendency for points originally
clustered together to stay together, regardless of to which resample cluster
they are assigned. Fourth, this process was repeated until the average
values of Mrjo and MO|r stabilized (N=200).
RESULTS
The geographical locations of the 420- and 816-site samples are shown
for the 48 conterminous states in Figure 1. Gauged stream sites were
available from all states and for most ecoregions and USGS hydrologic units.
The gauged sites exhibited a wide range of values for several important static
descriptors, including catchment area (Figure 2), mean daily flow (Figure 3),
mean annual runoff (Figure 4), gauge elevation (Figure 5), and period of
record .(Figure 6) .
11
-------
TABLE 1. MATRIX OF PEARSON CORRELATION COEFFICIENTS FOR 14 VARIABLES ON 420 STREAMS. VARIABLE DEFINITIONS ARE GIVEN IN METHODS.
SIGNIFICANT CORRELATIONS (BONFERRONI TEST) ARE: R > 0.12, P < 0.05; R > 0.13, P < 0.01; P > 0.15, P < 0.001.
AREA
QMEAN
MAR
DAYCV
DAYPRED
BFI
ZERODAY
FLDFREQ
FLDPRED
FLDFREE
FLDDOR
FLDVAR
LOWPRED
LOWFREE
AREA
1.00
0.57
-0.23
-0.03
0.10
-0.01
-0.07
0.15
0.09
0.07
0.25
-0.07
-0.19
-0.18
QMEAN
1.00
0.41
-0.27
0.44
0.07
-0.16
-0.08
0.16
0.15
0.02
-0.16
0.16
0.20
MAR
1.00
-0.34
0.42
0.09
-0.18
-0.23
0.09
0.15
-0.24
-0.19
0.38
0.39
DAYCV
1.00
-0.73
-0.56
0.60
0.39
-0.12
-0.02
-0.12
0.26
-0.23
-0.32
DAYPRED
1.00
0.60
-0.23
-0.32
0.28
0.20
0.07
-0.23
0.25
0.28
BFI
1.00
-0.19
-0.12
0.06
0.03
0.07
-0.10
-0.26
-0.28
ZERODAY
1.00
0.34
0.03
0.09
0.10
0.11
-0.20
-0.28
FLDFREQ
1.00
-0.14
-0.04
0.13
0.22
-0.31
-0.31
FLDPRED
1.00
0.81
0.59
-0.24
-0.03
-0.04
FLDFREE
1.00
0.55
-0.13
-0.01
-0.04
FLDDUR FLDVAR LOWPRED LOWFREE
1.00
0.12 1.00
-0.19 -0.07 1.00
-0.19 -0.06 0.85 1.00
TABLE 3. MATRIX OF PEARSON CORRELATION COEFFICIENTS FOR 14 VARIABLES ON 816 STREAMS. VARIABLE DEFINITIONS ARE GIVEN IN METHODS.
SIGNIFICANT CORRELATIONS (BONFERRONI TEST) ARE: R > 0.12, P < 0.05; R > 0.13, P < 0.01; P > 0.15, P < 0.001.
AREA
QMEAN
MAR
DAYCV
DAYPRED
BFI
ZERODAY
FLDFREQ
FLDPRED
FLDFREE
FLDDOR
FLDVAR
LOWPRED
LOWFREE
AREA
1.00
0.57
-0.23
-0.01
0.09
0.02
-0.02
0.09
0.07
0.05
0.26
-0.00
-0.20
-0.23
QMEAN
1.00
0.36
-0.27
0.41
0.09
-0.18
-0.08
0.12
0.09
0.02
-0.15
0.14
0.15
MAR
1.00
-0.36
0.40
0.09
-0.24
-0.22
0.08
0.14
-0.24
-0.22
0.38
0.42
DAYCV
1.00
-0.57
-0.52
0.67
0.25
-0.11
-0.04
-0.07
0.28
-0.31
-0.39
DAYPRED
1.00
0.58
-0.13
-0.29
0.31
0.26
0.13
-0.22
0.22
0.28
BFI
1.00
-0.24
-0.12
0.04
0.02
0.04
-0.10
-0.16
-0.13
ZERODAY
1.00
0.21
0.04
0.09
0.17
0.20
-0.30
-0.38
FLDFREQ
1.00
-0.15
-0.09
0.08
0.22
-0.24
-0.23
FLDPRED
1.00
0.82
0.61
0.01
0.02
0.01
FLDFREE FLDDUR
1.00
0.56 1.00
0.04 0.13
0.03 -0.17
0.02 -0.18
FLDVAR LOWPRED LOWFREE
1.00
-0.12 1.00
-0.13 0.84 1.00
-------
Statistical Correlations among Variables
The statistical relationships between the hydrological variables were
explored with correlation analysis and principal components analysis (PCA).
Table 1 shows that for the 420-site sample, many variables were significantly
correlated, despite the low correlation coefficients, because sample size was
very large. For the 11 primary classification variables, the correlation
matrix was used as input into a principal components analysis (PCA) to
determine how much of the total variance in the variable data space could be
explained by dimensional reduction. Table 2 shows that 4 principal component
axes explained 76% of the total variation among the 420 sites. The first axis
essentially represents a contrast between flow stability (high predictability
and baseflow) and variability (high negative coefficient of variation and
intermittency). The second axis emphasizes flood predictability and duration,
while the third reflects low flow predictability. The fourth axis describes
variation due to inter-annual flood intensity. Interestingly, the flood
frequency variable does not have a high loading on any factor, which suggests
that among-site variation in this variable is small relative to the other
variables.
TABLE 2. MATRIX OF ROTATED (VARIMAX) PCA FACTOR LOADINGS
THE 11 VARIABLES USED IN THE ANALYSIS OF 420 STREAMS. BOLDFACE
NUMBERS INDICATE VARIABLES WITH HIGHEST FACTOR LOADINGS
DAYCV
DAYPRED
BFI
ZERODAY
FLDFRQ
FLDPRED
FLDFREE
FLDDUR
FLDVAR
LOWPRED
LOWFREE
Variance explained
Cumulative proportion
FACTOR1
-0.902
0.775
0.794
-0.615
-0.363
0.079
-0.018
0.053
-0.109
0.066
0.145
2.60
0.24
FACTOR2
-0.062
0.236
0.024
0.157
-0.041
0.923
0.910
0.771
0.075
-0.038
-0.065
2.37
0.44
FACTORS
-0.197
0.151
-0.430
-0.294
-0.407
0.032
0.029
-0.185
0.007
0.923
0.924
2.24
0.65
FACTOR4
-0.132
0.223
0.044
0.059
-0.402
0.073
0.032
-0.190
-0.931
0.047
0.012
1.153
0.76
13
-------
TABLE 4. MATRIX OF ROTATED (VARIMAX) PCA FACTOR LOADINGS
THE 11 VARIABLES USED IN THE ANALYSIS OF 816 STREAMS. BOLDFACE
NUMBERS INDICATE VARIABLES WITH HIGHEST FACTOR LOADINGS
DAYCV
DAYPRED
BFI
ZERODAY
FLDFRQ
FLDPRED
FLDFREE
FLDDUR
FLDVAR
LOWPRED
LOWFREE
Variance explained
Cumulative proportion
FACTOR1
-0.540
0.838
0.867
-0.084
-0.281
0.089
0.053
0.040
0.017
-0.010
0.036
1.85
0.17
FACTOR2
-0.073
0.226
-0.035
0.092
-0.043
0.923
0.901
0 . 805
0.050
-0.025
-0.032
2.34
0.38
FACTORS
-0.237
0.244
-0.260
-0.228
-0.299
0.053
0.072
-0.194
0.064
0.926
0.918
2.07
0.57
FACTOR4
-0.134
0.196
0.002
-0.104
-0.647
0.097
0.048
-0.171
-0. 842
0.079
0.070
1.25
0.68
FACTORS
-0.719
0.037
0.249
-0.884
0.110
-0.015
-0.077
0.026
-0.283
0.138
0.227
1.53
0.82
Similar patterns emerged for the 816-site sample. Significant
correlations among hydrologic variables (Table 3) were frequent, despite low
correlation coefficients. The correlation structure was very similar to that
for the 420-site sample, indicating that variable periods of record did not
seriously influence the underlying relationships among hydrologic variables.
Five factors were required to explain 82% of the total among-site variation.
The PCA for the 816-site sample was similar to the 420-site sample, except
that the importance of flow variability (coefficient of variation and
intermittency) was transferred from the 1st to the 5th axis, and flood
frequency had a relatively high weight on the 4th axis (Table 4).
Cluster Results: Statistical Relationships
For the 420- and 816-site samples, 10 clusters were formed into groups
of "permanent" and "intermittent" streams, which were given 2-letter or 3-
character alpha-numeric abbreviations (Table 5). Among the intermittent
streams, sites having > 90 days of zero flow per year were characterized as
"Harsh Intermittent" (HI). Streams with less continuous intermittent
conditions (10 < ZERODAY < 90) fell into two clusters: "Intermittent Flashy"
14
-------
TABLE 5. LISTING OF ABBREVIATIONS FOR 10 CLUSTERS
FORMED IN THE 420- AND 816-SITE SAMPLES.
Category
Abbreviation
Description
Perennial
Intermittent
* Clusters
** Clusters
PR
GW
SS
SR
SRI*
SR2*
SN
SN1**
SN2**
PF
IR
IF
HI
formed only for N = 420
formed only for N = 816
Perennial Runoff
Stable Groundwater
Supterstable Groundwater
Snow + Rain
Snow + Rain, type
Snow + Rain, type
Snowmelt
Snowmelt, type 1
Snowmelt, type 2
Perennial Flashy
Intermittent Runoff
Intermittent Flashy
Harsh Intermittent
sites
sites
1
2
(IF) and "Intermittent Runoff" (IR) streams. The permanent groups of streams
had less than 10 days of zero flow per year on average. For this group,
solutions with 4 to 9 clusters were considered and compared in terms of
interpretability and repeatability. A 7-cluster solution was accepted as
optimal for the permanent streams. For the 420- and 816-site samples, 6
clusters were shared: "Perennial Runoff" (PR), "Stable Groundwater" (GW),
"Superstable Groundwater" (SS), "Snow+Rain" (SR), "Snowmelt" (SN), and
"Perennial Flashy" (PF) streams. For the 420-site sample, the 7th cluster was
a variant of "Snow+Rain" (SR2), while for the 816-site sample, the 7th
cluster was a variant of "Snowmelt" (SN2).
The statistical properties of these 10 clusters (7 permanent + 3
intermittent) are given for the 420-site sample in Table 6, which indicates
sources of significant variation among clusters in terms of the ecologically-
important hydrologic variables. For the intermittent streams, the Harsh
Intermittent (HI) group averages 190 days per year without flow. The
Intermittent Flashy (IF) and Intermittent Runoff (IR) groups show many fewer
15
-------
TABLE 6. NUMERICAL MEANS AND STANDARD DEVIATIONS (IN PARENTHESES) OF 11 VARIABLES FOR 10
CLUSTERS FOR 420 GAUGED STREAM SITES. VARIABLES INDICATED BY "*" WERE NOT INCLUDED IN
THE TWO-STAGE CLUSTER ROUTINE AT ANY TIME, BUT ARE INCLUDED FOR COMPARATIVE PURPOSES.
FOR INTERMITTENT STREAMS, EXCLUDED VARIABLES ARE INDICATED BY ""
Perennial Mesic Super
Runoff Groundwater Stable
(PR) (GW) (SS)
N
DAYCV
DAYPRED
BFI
ZERODAY
FLDFREQ
FLDVAR
FLDPRED
FLDFREE
FLDDUR
LOWPRED
LOWFREE
AREA*
QMEAN* .
MAR*
POR*
209
173.5
(46.9)
56.0
(11.6)
7.3
(5.6)
0.4
(1.3)
0.69
(0.13)
0.32
(0.21)
0.47
(0.12)
0.23
(0.12)
3.0
(2.5)
0.74
(0.16)
0.64
(0.17)
1128
(1068)
13.0
(12.8)
0.44
(0.26)
49.7
(7.0)
55
114.3
(36.9)
72.7
(3.9)
27.8
(8.1)
0
(0)
0.80
(0.16)
0.31
(0.10)
0.43
(0.09)
0.19
(0.11)
2.6
(1.6)
0.61
(0.14)
0.51
(0.17)
1324
(1155)
17.5
(14.7)
0.49
(0.21)
50.4
(6.5)
17
86.4
(49.1)
73.2
(5.5)
47.5
(16.8)
0
(0)
0.66
(0.09)
0.27
(0.14)
0.62
(0.14)
0.37
(0.14)
4.3
(2.4)
0.43
(0.07)
0.26
(0.09)
1144
(1100)
11.5
(8.8)
0.44
(0.40)
50.4
(5.8)
Snow
+ Rain
(SRI)
27
167.4
(40.5)
64.9
(9.3)
5.4
(3.9)
0.4
(1.7)
0.75
(0.16)
0.37
(0.23)
0.68
(0.11)
0.57
(0.09)
2.9
(1.6)
0.87
(0.10)
0.76
(0.10)
546
(518)
20.2
(24.3)
1.28
(0.94)
46.1
(7.8)
Snow
+ Rain
(SR2)
29
111.6
(28.8)
76.0
(7.0)
18.2
(8.3)
0.02
(0.12)
0.68
(0.08)
0.26
(0.11)
0.68
(0.14)
0.46
(0.10)
3.7
(2.5)
0.78
(0.16)
0.66
(0.18)
1099
(983)
27.6
(27.8)
1.11
(0.96)
49.2
(7.2)
Snowmelt
(SN)
22
134.1
(27.3)
78.8
(10.5)
15.8
(8.5)
0.05
(0.02)
0.65
(0.10)
0.25
(0.18)
0.97
(0.06)
0.79
(0.11)
9.6
(2.7)
0.70
(0.17)
0.62
(0.16)
1417
(1322)
23.7
(30.5)
0.43
(0.20)
49.1
(7.0)
Peren. Intermit.
Flashy Runoff
(PF) (IR)
24
270.4
(42.1)
34.0
(9.8)
4.4
(6.4)
2.4
(2.9)
0.97
(0.13)
0.37
(0.17)
0.48
(0.11)
0.31
(0.13)
2.8
(0.9)
0.48
(0.11)
0.31
(0.13)
1310
(890)
8.4
(5.9)
, 0.22
(0.12)
47.0
(7.3)
20
289.2
(91.0)
29.8
(7.2)
21.7
(8.5)
0.79
(0.16)
0.43
(0.30)
0.55
(0.15)
0.39
(0.20)
0.61
(0.13)
0.53
(0.20)
1133
(986)
5.2
(5.0)
0.22
(0.18)
43.9
(7.1)
Intermit.
Flashy
(IF)
10
474.5
(85.3)
20.4
(4.7)
43.7
(17.4)
0.94
(0.19)
0.40
(0.13)
0.46
(0.13)
0.19
(0.12)
0.46
(0.05)
0.16
(0.05)
845
(573)
3.0
(2.4)
0.12
(0.02)
43.8
(5.7)
Harsh
Intermit:
(HI)
7
481.0
(123.0)
47.1
(13.6)
189.7
(58.7)
1.04
(0.30)
0.44
(0.32)
0.58
(0.14)
0.46
(0.16)
0.49
(0.16)
0.25
(0.17)
644
(524)
0.4
(0.2)
0.03
(0.03)
43.0
(6.9)
16
-------
days of no discharge while differing from one another both in terms of average
intermittency and average flood frequency. Among the permanent streams, the
Snowmelt (SN) group sites have very high seasonality of flooding, the
Snow+Rain streams (SR) have intermediate seasonality of flooding coupled with
very high seasonality of lowflow and either a stable (SR2) or variable (SRI)
daily flow. The Perennial Runoff streams (PR) are characterized by low flood
seasonality coupled with high seasonality of lowflow and variable daily flow.
The Stable Groundwater (GW) group has low variability in daily flow coupled
with aseasonal flooding, while the Superstable (SS) streams express extremely
stable daily flow. The Perennial Flashy (PF) group exhibits high flow
variability and high flood frequency with low seasonality for both floods and
lowflow events . --
The clusters for the 816-site sample are segregated in very similar
fashion (Table 7) except that two snowmelt groups (rather than two snow+rain
groups) were formed. The main statistical distinction between these two
groups in terms of the hydrologic variables was that one group had greater
daily flow stability and higher seasonal predictability of lowflow conditions.
Graphical comparison of clusters in terms of hydrologic variables
reveals striking or subtle differences for DAYCV (Figure 7), DAYPRED (Figure
8), FLDFRQ (Figure 9), FLDPRED (Figure 10), FLDFREE (Figure 11), FLDVAR
(Figure 12), FLDDUR (Figure 13), BFI (Figure 14), ZERODAY (Figure 15), LOWPRED
(Figure 16), and LOWFREE (Figure 17). Less discrimination among clusters
occurs when static basin descriptors are examined such as area (Figure 18),
elevation (Figure 19), mean daily flow (Figure 20), and mean annual runoff
(Figure 21).
The distinction between SN1 and SN2 clusters in the 816-site sample can
be appreciated by examining the differences among clusters in terms of average
timing of flood onset and elevation (Figs.19 and 22). SN1 sites tend to be at
higher elevations, which probably contributes to greater seasonality through a
more enduring snowpack, as indicated by a later seasonal onset of snowmelt
(Figure 22). In practical terms, SN1 and SN2 could be lumped together to form
a unitary SN group for the 816-site sample, since the ecological distinction
among the two groups is subtle at best.
The distinction between SRI and SR2 clusters in the 420-site sample
cannot be easily discerned from the hydrologic variables or basin descriptors
used in the study. However, SRI streams tend to be lower elevation (Figure
17
-------
TABLE 7. NUMERICAL MEANS AND STANDARD DEVIATIONS (IN PARENTHESES) OF 11 VARIABLES FOR 10
CLUSTERS FOR 816 GAUGED STREAM SITES. VARIABLES INDICATED BY "*" WERE NOT INCLUDED IN
THE TWO-STAGE CLUSTER ROUTINE AT ANY TIME, BUT ARE INCLUDED FOR COMPARATIVE PURPOSES.
FOR INTERMITTENT STREAMS, EXCLUDED VARIABLES ARE INDICATED BY "--"
Perennial Mesic Super
Runoff Groundwater Stable
(PR) (GW)
-------
19) and to have an earlier onset of flooding (Figure 22) than SR2 sites. This
suggests that the SRI group is very similar to Poff and Ward's (1989) "Winter
Rain" streams, while the SR2 cluster more closely represents their "Snow+Rain"
streams. Poff and Ward (1989) separated these two types of streams using a
"flood timing" variable; however, their variable was not used here because
they failed to take into account the underlying circular distribution of the
seasonal variable. In short, a flood that occurs on day 10 of the water year
is more similar seasonally to a flood occurring on day 364 than it is to a
flood occurring on day 270, but standard statistical techniques will not
detect this because they measure distance as a function of absolute difference
(see Figure 22). Unfortunately, no single numerical representation can be
used to indicate this circular relationship. However, inspection of Figure 22
shows that SRI streams have median FLODTIME at day 81 of the water year, which
is December 19, a time of heavy winter precipitation generated by Pacific
storms. High elevation and inland SR streams (see below) have seasonal
flooding more heavily influenced by snowmelt (median day = February 21).
Cluster Results: Geographic Distribution
The spatial distribution of the identified clusters shows reasonably
good geographic affiliation. For the 420-site sample, PR streams occur
primarily in the Eastern Deciduou^Biome, with some representation in the West
(Figure 23a) . Stable GW streams also occur primarily in the East (Figure
23b). SS groundwater streams are largely restricted to the upper Midwest,
although a few of them occur in the Northwest (Figure 23c). The SR streams
are primarily in the Pacific Northwest (Figure 23d), although a significant
portion of the SR2 streams are scattered across the northern tier of states.
Some likely misclassifications occur in this group, as indicated by the
identification of SR streams in the Southeast. The Snowmelt streams show high
regional fidelity, being almost restricted to high elevation Rocky Mountain
states (Figure 23e) . The PF group also shows good regional affiliation for
the forest-prairie transition of the Midwest (Figure 23f). A small group of
PF streams also occur along the humid western Gulf Coast. The intermittent
streams (HI, IR, IF) tend to occur in the arid Southwest and Great Plains,
though some occurrences in the East are seen (Figure 23g).
For the 816-site sample, similar, geographic patterns were observed. PR
sites (Figure 24a) were mostly in the East, as were GW streams (Figure 24b).
19
-------
SS sites remained predominantly upper midwestern in distribution (Figure 24c),
and SR streams displayed a northwestern and northern-tier geographic pattern
(Figure 24d). Snowmelt streams were mostly restricted to the Rocky Mountain
region, but a noticeable proportion (7/12) of SN2 streams occurred in the
upper Midwest (Figure 24e). Perennial Flashy streams retained a strong
signature for the forest-prairie transition zone (Figure 24f). Among the
intermittent stream types, HI streams were most strongly associated with the
northern prairie, the southern prairie and the far southwest (Figure 24g). IF
streams appeared to occur in the forest-prairie transition zone of the
southern plains, while IR streams were more widespread, showing a more
eastern, northern and far western distribution (Figure 24g).
Cluster Stability
It can be noted that when sample size was shifted from 420 to 816, some
stream sites changed cluster affiliation (e.g., cf. PR streams in CA, NV, OR,
and WA in Figs.Sa and 9a). These "misclassifications" illustrate an important
limitation to the use of cluster analysis or any other classification scheme:
results are always relative to total variation expressed in the input matrix.
Thus, because the total sample variability changes when the analysis is
expanded from 420 to 816 streams, some sites will shift cluster membership
because the among-site distances in multivariate space have been slightly
altered. To assess the "robustness" of the clusters identified in the
analysis of the 383 "permanent" streams, we employed a bootstrapping technique
(see Methods). Based on 200 replicates, we found:
Mean 5%tile 95%tile
Mr|0 .80. .69 .87
M0|r .54 .42 .68
This suggests that streams which originally clustered together tend to stay
together (mean Mr|O = .80), but that points which are together in the
resample-based clusters do not necessarily share common original clusters
(mean Moir = .54) . In other words, the resample clusters tended to join
together several of the original clusters. Inspection of the resample
clusters showed that original clusters PR, SS, SN, and PF (1, 5, 6 & 7) were
20
-------
reproducible, but that clusters GW, SRI, and SR2 (2, 3, & 4) often merged
together into the PR group. These results indicate that the statistical
separation between the PR, GW, and SR clusters is weak, and, on purely
statistical grounds, these groups cannot be said to constitute discreet
groupings. However, on ecological grounds, there is some justification for
retaining these as separate groups. Given the importance of seasonality of
flood regime, it is reasonable to accord FLDPRED a high weight in
discriminating among clusters. Thus, the argument can be made that the
average differences in FLDPRED and DAYPRED/DAYCV between the PR streams and
the SR streams (particularly the SR2 streams) are of ecological interest, if
not strict statistical significance. The geographic separation of the PR and
SR sites further argues for their segregation (cf. Figure 23a and 23d). The
differences between the PR and GW sites are less obvious. Clearly, the
overlap for these clusters is great with the exception of DAYCV, PREDDAY, and
BFI (see Table 6). These variables, however, are often used as primary
discriminating variables in exploring hydrological-ecological relationships
(see Section 3, this report), so it is not unreasonable to argue that there
are ecological reasons for retaining the distinction between PR and GW
streams. It should be emphasized here that the SS group is an extreme
representation of the GW group, and the SS group did show statistical
stability during the bootstrapping simulations.
DISCUSSION
Climate change that modifies local precipitation regime and that
increases air temperature will modify the physical habitat template of a
stream via an integrated catchment response in which vegetational,
hydrological, thermal, and geomorphological components interactively adjust to
the substantially altered conditions (Poff 1992a). As mentioned previously,
the magnitude, frequency, duration and predictability of environmental
extremes (flow, temperature) are viewed as agents of disturbance and hence
important regulators of ecological processes and patterns in lotic systems
(see Resh et al. 1988, Poff and Ward 1989, 1990 for extended discussions).
Local community assembly and species persistence (including riparian
vegetation) have presumably been influenced by the characteristic disturbance
regime, such that freo^iently "disturbed" communities are relatively resilient
21
-------
(see Poff and Ward 1990, Reice et al. 1990, Schlosser 1990, Wallace 1990) .
Given this argument, climate change that substantially alters the magnitudes
and temporal distribution of extreme events may be expected to have the
greatest relative impact on resident ecosystem structure and function. Thus,
historical variability regime becomes a relevant variable in projecting the
possible direct hydrologic consequences of specified climate change (see
Figures 9, 10). Over longer periods of time, if temperatures increase,
resident fauna may simply be removed as thermal tolerances are exceeded.
Presently, the geographical distribution of climate change cannot be
predicted, thus it is impossible to know where physical habitat templates may
be most severely modified. However, it seems likely that at biome boundaries,
where relatively rapid, coupled transitions in vegetation and precipitation
patterns occur, geomorphic systems may be near thresholds of change and
ecological systems may be most vulnerable.
Several types of stream responses to altered hydrologic regimes are
worthy of discussion. First, changes in the absolute amount of annual
precipitation will influence stream permanence. Several studies have shown
that, in arid-land streams particularly, large relative decreases in
streamflow are correlated with small absolute declines in precipitation
(Langbein 1949, Revelle and Waggoner 1983, Karl and Riebsame 1989).
Simulation studies have also emphasized this sensitivity (e.g., Flaschka et
al. 1987, Schaake 1990), and the ecological consequences of any systematic
reduction in flow in arid-land streams has been addressed by ecologists (Dahm
and Molles 1992, Grimm and Fisher 1992) . Streams in cool and/or humid regions
do not exhibit substantial reductions in annual runoff with temperature
increases of ca. 1°C (Karl and Riebsame 1989), but severe reductions in
regional precipitation could clearly diminish average flow and push some
perennial streams into greater intermittency by increasing the number of days
with zero flow. For example, a fairly clear geographic break in the
distribution of perennial and intermittent streams corresponds to the forest-
prairie transition in the Midwest (cf. PR and PF streams to IR and IF streams
in Fig. 24). Climatic change that causes the Midwest to become drier and
encourages eastward progression of prairie would likely endanger many
permanent midwestern streams. However, a mitigating factor in this potential
response would be underlying geologic control on stream baseflow. In
particular, mesic groundwater and superstable streams would appear to be
22
-------
buffered against immediate negative responses to reduced precipitation inputs
because of aquifer storage. These streams, which also are likely to have
cooler summer thermal regimes due to groundwater inputs, might be viewed as
providing significant regional refuges for biota threatened by increasing
aridity and elevated air temperatures (cf. Meisner et al. 1988, Meisner 1990).
At present, predictions for regional climate responses to global wanning
cannot be given with confidence. Regional precipitation in climatically-
distinct regions may change by ± 20% and runoff may change by ± 50% (Schneider
et al. 1990). Regions that are now semi-arid may experience increased annual
precipitation, thus allowing presently intermittent streams to become
permanent. Such a change would represent a significant modification of the
present hydrologic template, and resident lotic communities would presumably
adjust in response to new habitat conditions.
A second major type of hydrologic change that could influence lotic
communities is an altered regime of hydrologic extremes, the frequency and
intensity of which are likely by-products of climatic change (Rind et al.
1989). For example, in stable groundwater-fed streams, more frequent flooding
due to a more energetic atmosphere would increase overall flow variability.
Since biotic interactions are presumably strong in streams having very stable
hydrographs (see Meffe 1984, Schlosser 1987, Poff and Ward 1989), increased
hydrologic variability would be expected (in theory) to modify community
structure by favoring generalists and/or inferior competitors. In other types
of streams, flashier hydrographs could result in major geomorphic changes that
would directly modify habitat conditions for resident biota. There are
several good paleohydrological and direct observational studies that clearly
illustrate the important interaction between mean annual precipitation and
flood regime in influencing channel morphology and geometry (see Schumm and
Lichty 1963, Schumm 1968, Burkham 1972, Know 1972, Williams 1978). In
general, mean annual precipitation controls the extent of vegetative cover and
thus influences runoff and sediment supply rate, and catastrophic channel
responses tend to result from frequent, intense flooding during periods of low
annual precipitation and reduced vegetation. These rapid channel responses
occur over a time scale of years under "normal" climatic fluctuations and are
thus likely to be observed as global climate change progresses (see Poff 1992
for further discussion). Most of these changes have been documented in semi-
arid areas; however, even in humid regions with their heavy vegetation cover
23
-------
and relatively cohesive soils, intense storms may cause localized episodic
channel widening (Costa 1974) or erosion (Kochel 1988) .
A third important ecological consequence of climatic change on stream
hydrology will result from altered seasonality of extreme events. The most
likely mechanism producing this response is increased temperature that reduces
the amount of time that frozen precipitation remains in storage as snowpack.
Several simulation studies have shown how moderate increases in annual
temperatures can cause early onset of snowmelt and lead to winter flooding
(e.g., Gleick 1987, Lettenmaier and Can 1990, Schaake 1990). High altitude
and latitude streams may shift from highly seasonal, primarily snowmelt-driven
systems to more variable snow+rain streams with more predominant winter
floods. Similarly, snow+rain streams might shift to entirely rainfall-
dominated systems with no seasonal storage of precipitation. Flood
predictability and timing may be important environmental correlates of
spawning and recruitment success (e.g., Seegrist and Gard 1972, Nesler et al.
1988), so changes in seasonality of high flows can be expected to produce
strong ecological responses. An additional consequence of a potential shift
from snowmelt to rainfall-dominated systems is that many snowmelt streams
occur in semi-arid regions where late summer baseflow is provided by the slow
melting of the lingering snowpack. Early melting of snowpacks, if not
accompanied by an increase in summerprecipitation, will result in a tendency
for late summer drying and a loss of habitat for present communities. Another
consequence of warmer winter temperatures might be loss of extensive seasonal
ice cover, which often serves to limit diversity and abundance of fish species
in small, northern streams (Schlosser 1987) .
A problem with any speculative projection about how climate change may
affect hydrologic regimes and hence lotic communities is that systematic
streamflow data extend only on the order of decades into the past. The
natural variability in climate makes even detection of trends in hydrologic
means very difficult (Karl 1988, Matalas 1990) and ensures that the prospect
of statistically documenting regional climate change will remain uncertain for
several decades (Schneider et al. 1990). Further, the assumption of temporal
invariance in hydrologic events is likely to be invalid under a changing
climate, such that prediction of spatial and temporal distribution of
hydrologic means and extremes based on empirical records will be dubious (Moss
24
-------
and Lins 1989) . All these considerations mitigate against unconditional
"predications" about regional hydrologic responses to climate change.
Beyond the concern for prediction of changes in the physical environment
is the great uncertainty associated with establishing robust relationships
between hydrologic regimes and ecological patterns and processes. Much
present theory in stream ecology emphasizes the regulatory role that hydrology
and disturbance regime play in organizing community structure and constraining
species distributions and abundances (see overviews in Resh et al. 1988, Poff
and Ward 1989, 1990, Minshall 1988, Fisher and Grimm 1992). However,
compelling empirical data linking ecological organization to hydrologic regime
are largely lacking because the appropriate large-scale "controlled"
ecological comparisons have not been done (but see Section 3). Thus, any
"predictions" about the ecological consequences of climatic change must
explicitly acknowledge a high degree of uncertainty in the knowledge base.
Despite the uncertainties of both hydrological and ecological
prediction, it is important for stream ecologists to have a "phenomenology" of
streamflow types in order to better appreciate the existing range of
hydrologic variability in North American streams and rivers. The typology
presented in this paper is based on ecologically-relevant hydrologic
attributes; therefore, it can be expected to provide a meaningful basis for
identifying streams that differ in major aspects of ecological organization.
An objective framework that allows comparable sites to be matched and
establishes reasonable a priori expectation is required to enhance potential
success of research and management of climatic change.
CONCLUSIONS
Relatively undisturbed U.S. streams were classified according to
variation in 11 ecologically-relevant hydrologic characteristics. A group of
420 "best" stream sites and a group of 816 "acceptable" sites (including the
420 best sites) were evaluated with similar results. Cluster analysis
resulted in the identification of 10 distinctly different stream types. Most
of these stream types exhibit reasonable geographic affiliation and can be
interpreted in terms of regional climatic patterns and local variation in
geologic characteristics. Four of the seven perennial stream groups (PR, SS,
SN, and PF) were found to be robustly defined when subjected to a
25
-------
bootstrapping technique. The other three groups (GW, SRI, and SR2) typically
merged together into the PR group and are thus not statistically distinct.
However, these groups can probably be retained on ecological grounds, because
of their inherent differences in flood predictability and baseflow
characteristics.
The derived classification for U.S. streams based on ecologically-
relevant hydrologic characteristics provides a comprehensive catalog that
identifies streams that, according to ecological theory, may differ in major
aspects of ecological organization. This classification thus provides a basis
for hypothesis-generation and affords an objective framework for matching
streams for purposes of comparative ecological investigations. Moreover, this
hydrologic classification can be used to assess the potential ecological
consequences of hydrologic changes, because specific kinds of hydrologic
change induced by climatic changes (e.g., increased intermittency, reduced
flood seasonality, etc.) can be compared to the historical hydrologic patterns
summarized in this classification.
26
-------
SECTION 2
IMPORTANCE OF TEMPORAL SCALE IN ASSESSING STREAMFLOW PREDICTABILITY
AND FLOOD REGIME IN STREAMS OF DIFFERENT GEOGRAPHIC AREAS
INTRODUCTION
A growing need in stream ecology is the establishment of criteria with
which study sites can be selected for comparative work (e.g., Fisher and Grimm
1991). Stream ecologists have relied on criteria such as stream order and
trophic base (e.g., Vannote et al. 1980) or ecoregions (Omernik 1987).
Increasingly, similarity in hydrologic regime is seen as an important
criterion for site matching because hydrologic events play important
structuring role in stream ecosystems (Resh et al. 1988). Indeed, there are
now several examples of comparative hydrologic analyses that ecologists may
use to guide site selection for biological studies (e.g., Resh et al. 1988,
Poff and Ward 1989, Hughes and James 1989, Jowett and Duncan 1991) .
Several papers over the past few years have shown the great variation in
hydrologic regimes that exists among streams at regional and continental
scales at a variety of locations throughout the world (see Section 1 of this
report for a summary). Classification of streams having similar hydrology can
certainly assist in proper site-matching; however, the criteria by which
different researchers classify streams are variable and often based on non-
formal or ad hoc statistical descriptors of hydrologic variation. Implicit in
any hydrographic analysis is the treatment of temporal scale. The length of
the hydrologic time series is an obvious factor that determines what kinds of
statistical summaries can be extracted from the data. Less obviously, the
temporal resolution with which the available data are viewed can influence the
interpretation of the derived statistical summaries. Hydrographs can be
analyzed with numerous degrees of temporal resolution, and the scale chosen
typically reflects specific interest in hydrologic phenomena having different
periodicities. For example, an annual time scale may be appropriate if the
question involves assessing inter-annual variation in spring runoff (e.g.,
Molles and Dahm 1990). Intra-annual seasonal patterns of hydrologic variation
might be assessed with monthly data (Resh et al., 1988, Bunn et al. 1986) .
27
-------
Transient events, like spates, may require high resolution, daily data (Poff
and Ward 1989).
Approaches to comparing the hydrologic similarity of lotic ecosystems
are relatively new, and no "standardized" methods have been accepted with
which to proceed. A necessary first step toward this end is determining
whether available methodologies can be unambiguously applied under a range of
circumstances that are likely to be encountered in applying the methodology to
actual problems. An important aspect of this effort is to examine how robust
the methodologies are to variation in the temporal resolution with which the
data are viewed. If standard methodologies provide different answers
depending on the scale of observation and analysis, then those limitations
need to be acknowledged and taken into account when applying the
methodologies. Thus, the issue of scale-dependency in pattern analysis is
very important for ecologists interested in using hydrologic records to
establish "comparable" sites for biological studies.
There are two objectives in this section. First, we wanted to examine
how sensitive a widely-used, formal measure of predictability (Colwell's
index) is to variation in the size of the temporal window with which the raw
data are initially viewed. Second, we wanted to focus on the question of what
degree of temporal resolution in a hydrologic dataset is required to
adequately describe the occurrence of extreme high flows. A further aspect
for both of these objectives was our interest in examining the geographic
variation in sensitivities to temporal scale of analysis. Therefore, a large
number of streams representing a wide range of climatic regions were selected
from across the continental United States.
Colwell (1974) introduced a formal measure of environmental
predictability that has received widespread use in many stream ecological
studies over the last several years (Resh et al. 1988, Poff and Ward 1989, see
Gordon et al. 1992) . This method essentially reconfigures a data time series
into a two-dimensional matrix consisting of n states by p time intervals.
The states are merely categories of discharge level. The time intervals are
divisions of some natural period, typically an annual cycle. For example, a
long time series of annual hydrologic data could be broken into time intervals
of months, weeks, days, etc. Colwell's index is quantitative, yet it has
several drawbacks. First, there are no objective criteria for establishing
the number of discharge levels used to calculate the index, and the calculated
28
-------
value is sensitive to variation in this parameter (Can et al. 1991). Second,
the index value is sensitive to length of hydrologic data record, requiring
about 40 years of continuous record to produce a stable output (Can et al.
1991). Third, the partitioning of the time scale is arbitrary. Individual
researchers have calculated Colwell's index using monthly (Resh et al. 1988,
Can et al. 1991, Bunn et al. 1986) and daily (Poff and Ward 1989) time scales.
The consequences of arbitrary temporal resolution in the definition of time
scale using Colwell's index have not been explored.
Hydrologic extremes are recognized for the control they exert on stream
communities, and a host of important ecological questions arise from
considering the temporal distribution (regime) of such extremes (see Resh et
al. 1988, Poff 1992). Yet there is no consensus as to what .degree of temporal
resolution in hydrologic data is required to adequately characterize a
stream's spate regime. If monthly data are as useful as daily data in
allowing the accurate description of the frequency and predictability of
extreme events, then the task of site matching is made much easier, because
monthly data are generally more accessible and can be more easily extrapolated
from readily available precipitation data. However, situations may arise
where average monthly flow does not correlate with flood events because large,
transient floods leave little signature on the monthly average. Similarly,
annual flow data can be misleading in terms of inferring spate regimes because
dry years may be dry on average, but have significant spates (e.g., Grimm
1992) . The adequacy of using coarse grain or aggregated hydrologic data
(e.g., monthly, annual) to make inferences about transient hydrologic
phenomena such as floods may vary among streams depending on total
precipitation or local catchment geology. For example, intermittent streams
may be less sensitive than perennial streams to changes in time scale used to
characterize flood regimes because a large flood in an intermittent stream may
leave a permanent signature on the monthly average. Similarly, highly
seasonal streamflow regimes, such as those dominated by snowmelt, may not
require fine grain data to characterize the distribution of extremely high
flows accurately. The temporal scale of hydrologic data needed to
characterize floods and the geographic variability of this characterization
has not received previous attention in the stream ecology literature.
29
-------
METHODS
Data were taken from long-term U.S.Geological Survey flow records using
the methods described in Section 1 of this report. All sites represented flow
regimes relatively unimpaired by anthropogenic modifications. Additionally,
each site was evaluated for a 36 yr period (1950-1985), a length of time
sufficient to allow a stable estimate for Colwell's index of predictability
(see Can et al. 1991). In an earlier analysis, a total of 480 sites were
selected and analyzed using non-hierarchical cluster analysis for a streamflow
classification (see Section 1). All sites in this analysis met the 36-yr
criterion. Each of the 10 groups resulting from the classification was
"subsampled" to provide data for the present analysis. No more than 12 sites
per classification group were selected for a total sample size of 118 gauged
streams. Sites were not randomly subsampled from the initial classification
groups. Rather, for each of the 10 groups, the sites closest to the
multivariate centroid of the group were selected by inspection to insure that
the sample would represent the sites most characteristic of each group. The
10 groups themselves essentially span the range of hydrologic variability
available in the best available hydrologic data set for the entire U.S.
Selecting the sites in this manner allowed us to examine the entire range of
hydrologic variability and to test for significant differences among
previously-defined hydrological types. An additional two groups were defined
because two of the original 10 groups had members that occurred both in the
eastern and the western U.S., two broadly-defined regions which differ in
climatic seasonality. These two groups were partitioned geographically (for a
total of 12 groups) so we could evaluate the extent to which the
classification results were robust against geographic location.
Colwell's Index
The input for calculation of Colwell's index was the daily hydrograph.
All hydrographs were organized according to the "water year", which commences
on October 1 and ends on September 30 of the following calendar year. Data
were standardized by dividing each entry in the matrix by the overall mean
value of the matrix. This allows streams of different sizes to be directly
compared. Nine class intervals (states) were used for discharge, and they
were bounded at the following increments of logio units: -1.5, -1.0, -0.5, 0,
30
-------
0.5, 1.0, 1.5, 2.0. These categories span the range from <0.03 to >100 times
the average flow (on an arithmetic scale). The number of flow states selected
can influence the calculation of predictability (Can et al. 1991), yet there
is no objective criterion for selection of the number of states or their
scaling (see Gordon et al. 1992). We chose these nine flow states to span a
wide enough range to capture "extreme" events that occurred for the entire
range of data under consideration.
Separate indices of predictability at each site were determined for four
time steps that comprise segments of the annual cycle: daily, weekly, monthly
and seasonal. Our hypothesis was that calculated predictability values would
not vary within the 12 streamflow types as a function of temporal scale used
to calculate the value. These four intervals span the range of temporal
frames with which stream ecologists generally work. For each site, a daily
matrix consisted of 365 days x 36 yr of data. Day 366 of leap years was
omitted. The weekly matrix was derived by taking the average weekly flow for
52 7-d periods for each of the 36 yr. Days 365 and 366 (in leap years) were
omitted. The monthly matrix was derived by determining the average flow for
each calendar month of the year. The seasonal matrix consisted of four, 3-
month "seasons", starting with October.
Importance of temporal scale in assessing; spates
To determine how effectively hydrographic data of coarse temporal
resolution can assess transient events (spates) in streams, we compared
monthly and annual flow statistics to daily flow statistics. If coarse grain
data (e.g., monthly to annual) are capable of allowing detection of transient
events, then, at a minimum, the signatures of the maximum daily
(instantaneous) flow ought to be present in the time series of the coarse
grain data. For the monthly time scale, we determined for each of the 36
years the month having the highest average flow. Using the daily flow matrix,
we then recorded for each year the month in which the highest daily flow
occurred. We calculated the proportion of the 36 observations where the high
monthly flow and the month of highest daily flow matched. For example, a
stream with the highest daily flow each year in the same month having the
highest average monthly flow would score a maximum proportion of 1.0.
Differences among the 12 groups were tested with respect to proportion of
31
-------
TABLE 8. ABBREVIATIONS AND SAMPLE SIZES FOR 12 DIFFERENT CLASSIFICATION
GROUPS USED IN THE ANALYSIS
Descriptive Title
Perennial Runoff
Stable Groundwater
Superstable Groundwater
Snow + Rain
Snowmelt
Perennial Flashy
Intermittent Runoff
Intermittent Flashy
Harsh Intermittent
Abbreviation
PR1 (East)
PR2 (West)
GW1 (East)
GW2 (West)
SS
SRI (Type 1)
SR2 (Type 2)
SN
PF
IR
IF
HI
sample size
11
11
11
4
12
9
12
11
12
10
9
6
matched flow events after arcsine transformation with one-way ANOVA and
Student-Newman-Keuls multiple comparison tests (SAS 1988).
The utility of annual flow data in assessing peak daily flows was
assessed using rank correlation. For each year in the 36-yr record, the
series for both the maximum daily flows and for the annual average flows were
ranked (allowing for ties). Spearman's rank correlation coefficient (rs) was
determined for the ranks of these two series. The null hypothesis of no
positive correlation between the daily and annual series for each individual
stream site was tested using a one-tailed Spearman's p (Conover 1971, p. 248).
Differences among the 12 groups with respect to average rank correlation were
examined after arcsine transformation with one-way ANOVA and Student-Newman-
Keuls multiple comparison tests (SAS 1988).
32
-------
RESULTS
The 12 groups used for the analysis are defined in Table 8. The 118
individual stream sites used in the analysis, and their scores for the various
derived variables, are provided in Table 9.
Colwell's Indey
Differences among Groups for a Fixed Time Step
Clear differences among the 12 identified groups were found in terms of
flow predictability at all temporal scales. For all four time steps,
significant among-group differences existed at p < 0.0001. Figure 25
summarizes the differences among the 12 groups for each of the time steps used
to calculate streamflow predictability. At the shortest (daily) time step,
superstable (SS) and snowmelt (SN) streams had highest predictability values,
while perennial flashy (PF), intermittent runoff (IR), and intermittent flashy
(IF) streams had the lowest predictability values (Figure 25). For the
longest (seasonal) time step, SS and SN streams retained the highest
predictability values, but eastern groundwater (GW1) and high elevation
snow+rain (SR2) streams were a close second. The lowest predictability scores
were recorded for the IF, IR, and PF streams, in addition to the harsh
intermittent (HI) streams.
The null hypothesis that all group means were equal was tested with
oneway ANOVA for each of the four predictability time steps. In each case the
null hypothesis was rejected at p < 0.0001. Significant pairwise differences
among groups were examined for each predictability measure using the SNK
multiple comparisons test (SAS 1988). These results are given in Table 10 and
they correspond directly to Figure 25. Several interesting observations can be
made. For example, the following pairs of groups were always statistically
similar for all four predictability time steps: SN and SS, SRI and GW2, GW2
and PR2, GW1 and GW2. The three groups in the triplet of PF, IR and IF were
also statistically indistinguishable for all time steps. PR1 and PR2 streams
were the same only for the monthly and seasonal time steps, while SRI and SR2
streams were always different. Interestingly, the eastern perennial runoff
(PR1) and stable groundwater (GW1) streams were always different from one
another, whereas the western PR2 and GW2 streams were always the same (see
below).
33
-------
TABLE 9. SUMMARY OF 6 CATEGORIES OF STATISTICAL DATA DERIVED FOR 118 SITEi
OVER A COMMON 36-YR PERIOD, BY STATE ABBREVIATION,
GAUGED STREAM NUMBER, AND FLOW GROUP AFFILIATION.
State
WV
PA
VA
NY
ME
IL
MN
IN
IA
LA
FL
CA
CA
NM
CA
CA
NM
CA
NM
NV
OR
WA
MO
GA
IN
AL
FL
VA
MD
IL
NC
PA
NJ
CA
NM
CA
TX
IL
WI
WI
MI
OR
WI
MN
OR
SD
CA
Gauge*
03198500
03102500
02017500
01350000
01055000
03345500
04014500
03334500
06808500
07352000
02231000
11315000
11264500
07203000
11282000
11266500
07218000
11230500
08380500
10329500
14020000
12186000
07067000
02217500
04094000
02374500
02376500
03170000
01583500
05438500
02111000
01555000
01396500
11381500
09430500
11383500
08172000
05435500
05379500
05434500
04033000
14328000
04071000
05286000
14010000
06409000
11367500
Group
PR1
PR1
PR1
PR1
PR1
PR1
PR1
PR1
PR1
PR1
PR1
PR2
PR2
PR2
PR2
PR2
PR2
PR2
PR2
PR2
PR2
PR2
GW1
GW1
GW1
GW1
GW1
GW1
GW1
GW1
GW1
GW1
GW1
GW2
GW2
GW2
GW2
SS
SS
SS
SS
SS
SS
SS
SS
SS
SS
Day-Month
Matches
55.56
55.56
63.89
61.11
50.00
83.33
69.44
58.33
69.44
66.67
80.56
36.11
63.89
66.67
55.56
47.22
44.44
69.44
58.33
38.89
41.67
30.56
77.78
75.00
55.56
58.33
77.78
63.89
61.11
91.67
44.44
41.67
58.33
61.11
72.22
58.33
69.44
69.44
75.00
75.00
72.22
47.22
69.44
83.33
41.67
66.67
50.00
Day-Year
r-value*
0.677
0.509
0.424
0.785
0.481
0.786
0.388
0.492
0.502
0.694
0.625
0.769
0.887
0.456
0.922
0.910
0.810
0.837
0.662
0.879
0.713
0.569
0.614
0.642
0.772
0.661
0.577
0.523
0.546
0.889
0.319
0.675
0.721
0.855
0.805
0.818
0.736
0.655
0.398
0.623
0.627
0.597
0.590
0.806
0.621
0.532
0.816
Predictabilitv of Streamflow
Day
43.75
48.30
51.76
45.38
50.53
33.83
47.16
50.57
42.02
41.74
35.23
54.16
48.08
39.57
45.23
49.63
41.48
55.83
45.75
68.18
69.29
63.62
65.81
60.86
69.53
60.77
68.64
63.71
60.94
50.60
60.95
54.76
58.68
64.40
61.96
59.63
51.45
59.39
70.55
59.23
78.37
78.64
76.65
55.47
82.77
72.89
77.18
Week
45.55
49.47
51.92
44.94
51.34
33.81
48.13
49.74
42.95
42.79
35.80
54.12
48.48
40.83
45.19
49.60
42.17
56.43
46.20
67.80
69.79
66.28
64.89
61.50
68.82
61.95
68.18
65.65
61.94
50.76
61.18
55.60
60.30
64.26
61.66
59.66
50.78
60.06
69.83
59.88
78.66
78.71
76.90
55.58
83.42
73.83
77.11
Month
52.54
56.34
57.71
51.05
56.79
36.92
53.16
52.10
46.26
45.22
37.24
53.63
51.73
45.72
46.92
51.39
43.96
58.41
47.63
66.50
73.29
73.02
66.19
66.30
74.22
67.72
69.88
72.24
63.94
53.02
65.04
60.12
64.41
66.49
60.07
60.84
49.76
63.80
70.42
63.41
81.25
79.64
78.30
54.76
84.38
74.19
77.09
Season
62.52
67.53
69.18
60.43
70.52
43.89
58.35
63.31
48.52
53.40
42.70
49.46
55.40
49.67
45.64
54.25
46.85
65.35
49.78
69.72
81.15
78.21
70.60
74.75
77.91
71.59
73.55
75.65
68.09
55.87
67.12
68.48
68.87
73.13
59.07
63.35
52.83
64.58
70.94
65.16
83.69
79.91
84.91
57.18
84.91
75.16
76.84
34
-------
FL
CA
NV
OR
CA
OR
CA
WA
OR
CA
OR
WA
WI
OR
ME
MN
CA
WA
NY
OR
WA
MI
ID
ID
MT
ID
MT
CO
ID
CO
MT
ME
MT
CO
MO
IA
MO
IA
IL
TX
TX
SD
KS
NE
LA
CA
AR
CA
SD
KY
ND
ND
KY
MN
CA
KS
KS
KS
02359500
11355500
10316500
14325000
11402000
14042500
11532500
12020000
14308000
11382000
14193000
12488500
05405000
10396000
01022500
05275000
11522500
12451000
04256000
14179000
12035000
04105000
12413000
13120000
12332000
13336500
12358500
09112500
13120500
09124500
12355500
01013500
06207500
07083000
05498000
05486490
05500000
06898000
03346000
08070000
08068520
06481500
06892000
06811500
08010000
11063500
07261500
11098000
06356500
03298000
06354500
05066500
03320500
05300000
11111500
06917000
07167500
06911500
SS
SS
SRI
SRI
SRI
SRI
SRI
SRI
SRI
SRI
SRI
SR2
SR2
SR2
SR2
SR2
SR2
SR2
SR2
SR2
SR2
SR2
SR2
SN
SN
SN
SN
SN
SN
SN
SN
SN
SN
SN
PF
PF
PF
PF
PF
PF
PF
PF
PF
PF
PF
PF
IR
IR
IR
IR
IR
IR
IR
IR
IR
IR
IF
IF
61.11
47.22
83.33
63.89
58.33
55.56
63.89
61.11
58.33
66.67
72.22
72.22
77.78
50.00
52.78
77.78
52.78
77.78
75.00
55.56
72.22
63.89
72.22
80.56
75.00
86.11
83.33
83.33
83.33
86.11
86.11
80.56
88.89
80.56
72.22
61.11
55.56
66.67
75.00
72.22
83.33
77.78
66.67
77.78
66.67
83.33
72.22
80.56
72.22
66.67
83.33
88.89
66.67
83.33
75.00
66.67
69.44
66.67
0.651
0.766
0.846
0.739
0.845
0.737
0.609
0.681
0.702
0.862
0.475
0.517
0.449
0.810
0.583
0.790
0.714
0.701
0.474
0.533
0.496
0.772
0.678
0.886
0.708
0.665
0.717
0.914
0.860
0.840
0.718
0.692
0.721
0.869
0.744
0.686
0.832
0.771
0.743
0.835
0.881
0.904
0.828
0.794
0.662
0.716
0.712
0.923
0.872
0.592
0.921
0.949
0.659
0.878
0.928
0.792
0.804
0.802
70.14
74.05
63.31
52.99
54.64
53.71
59.81
58.18
58.81
49.42
62.74
63.70
65.36
62.25
54.92
58.60
57.24
68.26
59.12
64.51
67.32
62.30
64.76
74.92
72.26
68.22
68.43
74.95
71.45
75.00
71.15
55.58
70.16
72.84
29.96
28.73
29.04
28.07
31.77
39.92
34.24
32.80
28.13
44.18
27.45
33.93
37.72
36.95
46.36
36.34
40.32
40.27
37.29
35.56
42.80
32.52
31.91
33.67
69.81
73.86
63.91
54.30
54.95
54.81
60.03
59.85
60.09
49.45
63.80
63.43
66.90
62.88
56.63
58.76
57.87
69.36
60.20
65.51
69.25
62.85
65.26
75.67
73.73
68.97
69.52
75.07
72.34
75.71
71.95
55.57
70.95
73.38
27.48
28.27
27.26
27.47
30.07
37.75
32.12
32.26
26.47
41.66
27.30
33.36
35.93
35.91
46.16
34.02
39.99
39.92
36.46
35.40
41.64
29.87
29.18
30.59
69.46
73.90
66.26
60.94
56.08
58.48
65.08
66.24
64.98
50.47
68.17
65.69
69.09
65.75
60.45
60.11
61.90
72.03
64.98
70.48
76.78
65.95
67.74
76.85
76.27
70.47
71.74
76.59
74.60
77.66
72.68
56.02
74.33
76.28
27.15
29.92
28.56
29.03
31.49
36.39
31.66
31.39
27.84
43.13
39.62
33.09
38.19
36.50
48.95
36.68
40.38
38.77
42.14
34.59
38.25
27.69
29.04
28.80
69.26
72.27
65.13
79.03
57.65
64.41
72.76
85.06
65.12
52.52
77.83
72.44
70.85
68.69
68.01
62.82
66.88
75.78
77.15
70.13
90.17
72.43
70.19
78.01
76.30
71.09
76.12
76.14
76.18
80.03
77.60
64.78
83.14
80.74
34.44
38.38
36.19
36.63
40.07
38.92
34.48
29.24
32.09
46.75
55.54
33.17
44.88
36.65
51.94
49.57
33.93
33.28
49.06
32.59
36.91
30.10
29.19
30.92
35
-------
OK
KS
AZ
KS
KS
KS
MM
SD
CA
ND
MM
CA
CA
07311500
07180500
09480000
06914000
07172000
06889500
08408500
06425500
11274500
05060500
07222500
11124500
10258500
IF
IF
IF
IF
IF
IF
IF
HI
HI
HI
HI
HI
HI
83.33
72.22
69.44
80.56
75.00
72.22
83.33
88.89
69.44
91.67
80.56
83.33
75.00
0.793
0.684
0.895
0.809
0.764
0.814
0.897
0.935
0.927
0.919
0.786
0.944
0.841
43.39
31.84
34.67
32.98
30.44
28.71
45.81
51.05
67.29
58.41
52.68
45.80
53.82
37.54
29.13
33.22
29.45
28.67
27.33
42.27
50.17
65.80
57.46
49.46
44.46
50.98
27.83
28.02
32.37
27.94
25.95
28.61
39.28
44.87
61.66
53.63
43.39
41.80
43.72
27.73
30.96
34.37
31.25
28.29
32.47
42.05
36.40
49.63
36.55
39.41
39.79
32.55
rs > 0.55 is significant at a = 0.001; rs > 0.30 is significant at a = 0.05.
Differences within Groups for Variable Time Steps
The 12 groups fell into three categories. First, some groups showed an
increase in the estimate of predictability as time scale increased from daily
to seasonal. Included in this group were SN, SRI, SR2, GW1, PR1, and PF
streams (Figure 26). A second category of four stream types showed no
discernible change in predictability with increasing time scale. Streams
included in this category included PR2, GW2, SS and IR streams (Figure 27).
The third category consisted of two groups of streams that were characterized
by declining predictability values as temporal scale increased (Figure 27,
bottom panels). IF streams predictability values reached a minimum at the
monthly time step before increasing again at the seasonal time step. HI
streams consistently declined and were thus the only stream type that had the
lowest predictability at the longest time scale.
Importance of temporal scale in assessing- spates
Daily vs. Monthly Data
The correspondence between months with highest average flow and months
with highest daily flow ranged from 30-90% across the 12 groups (Figure 28a).
The groups with the greatest percentage of concurrence were SN and HI streams,
both of which had median scores > 80%. All other groups had medians > 60%,
with the exception of PR2 streams, which had a median score < 50%. Many of
36
-------
TABLE 10. SUMMARY OF COMPARISONS OF SIGNIFICANT DIFFERENCES AMONG CLUSTER
MEANS FOR ONEWAY ANALYSES OF VARIANCE FOR FOUR MEASURES OF OVERALL
PREDICTABILITY OF STREAMFLOW (DAILY, WEEKLY, MONTHLY, AND SEASONALLY),
FOR NUMBER OF MATCHES BETWEEN MAXIMUM MONTHLY AVERAGE AND MONTH WITH
MAXIMUM DAILY FLOW, AND FOR RANK CORRELATION BETWEEN ANNUAL PEAK DAILY
FLOWS AND ANNUAL AVERAGE FLOW. FOR EACH TEST, THE 12 CLUSTERS ARE
ARRANGED FROM LEFT TO RIGHT IN DESCENDING ORDER OF CLUSTER MEAN SCORE.
CLUSTER MEANS THAT ARE NOT SIGNIFICANTLY DIFFERENT FROM ONE ANOTHER
(SNK TEST, O = 0.05) HAVE A COMMON UNDERLINE. SIGNIFICANTLY
DIFFERENT CLUSTERS HAVE DIFFERENT UNDERLINES. CLUSTER
ABBREVIATIONS AND SAMPLE SIZES ARE GIVEN IN TABLE 8
Daily Predictability
SS SN SR2 GW1 GW2 SRI HI PR2 PR1 IR IF PF
Weekly Predictability
SS SN SR2 GW1 GW2 SRI PR2 HI PR1 IR IF PF
Monthly Predictability
SN SS SR2 GW1 SRI GW2 PR2 PR1 HI IR PF IF
Seasonal Predictability
SN SS SR2 GW1 SRI GW2 PR2 PR1 IR HI PF IF
Day-Month Matches
SN HI IR IF PF SR2 GW1 PR1 GW2 SRI SS PR2
Rank-Co rrelat ion
HI IR IF GW2 SN PF PR2 SRI SS GW1 SR2 PR1
37
-------
these groups were significantly different at a = 0.05 using the SNK multiple
comparisons test (Table 10).
Daily vs. Annual Data
When the rank correlations between the series of annual maximum daily
flow and annual mean flows were determined, values ranging from 0.32 to 0.95
were observed (Table 9, Figure 28b). For a sample size of 36, a rs value >
0.30 was significant at a = 0.05; therefore, all individual stream sites had
significant correlations between annual maximum and annual average flows. For
all 12 stream groups, medians exceeded 0.6, with the exception of PR1 streams.
HI streams were the only group with a median exceeding 0.9. Although there
was ample overlap among groups, some statistically significant differences
were found (Table 10). One interesting pattern was that the western perennial
runoff and stable groundwater streams (PR2 and GW2) had significantly greater
correlations between maximum annual average and maximum annual daily flow
(Figure 28b). For the perennial runoff streams, at least, this pattern was
the opposite of that observed for the daily-monthly matches (Figure 28b),
where eastern PR streams were less sensitive to a change in time scale than
their western "counterparts".
DISCUSSION
This analysis shows that the temporal scale with which one analyzes a
hydrologic time series has significant implications for the inferences drawn
about streamflow predictability and disturbance (spate) regime. The magnitude
of the "impact" of using different scales varies with geographic location, so
any comparative hydrologic analysis at a regional or larger scale should take
these relationships into consideration.
The results from the repeated calculation of Colwell's index of
predictability applied to streamflow records for the 12 groups of streams
identified here show two things. First, there are generally large among-group
differences regardless of what temporal scale is used. This fact allows us to
identify "types" of streams that are in a sense independent of the scale that
is chosen. For example, snowmelt streams have very high predictability
relative to most other stream types, regardless of whether predictability is
calculated with daily, weekly, monthly, or seasonal data. Therefore, as long
38
-------
as temporal grain is consistent, it is possible to discern major differences
in predictability for a wide range of hydrologically diverse streams.
Second, within certain groups, significant changes in the estimates of
predictability occur, depending on the time scale chosen. For some groups,
predictability increases, for others it stays the same, and for others,
declines occur. These substantial changes can even exceed the among-group
differences (cf. HI in Figure 25). These observations suggest that, when
comparing measures of predictability among hydrologically and/or
geographically diverse streams, one should give due consideration to the
rationale for choosing a particular temporal window for analysis.
Third, there appears to be a geographic signature on the calculation of
predictability that to some extent overrides the importance of streamflow
classification type. The eastern PR1 and GW1 streams were always different in
terms of mean predictability at different time scales (see Figure 25, Table
10); however, the western PR2 and GW2 streams were never statistically
distinguishable. This finding suggests that only in the eastern U.S. are
these two stream types easily separable (where they are most abundant).
Runoff streams appear to respond rapidly to precipitation events whereas
stable groundwater streams are presumably less responsive due to substantial
aquifer storage. In a climate where significant precipitation can occur at
any time of the year (i.e., eastern U.S.), PR streams would exhibit frequent
discharge fluctuations in response to precipitation, but under a strongly
seasonal precipitation regime (as in the West), streamflow variation in PR
streams would be more seasonal and hence less sensitive to temporal scale
chosen to estimate predictability.
For estimation of the flood regime, our results show that coarse-grain
data can be used if one is willing to accept a high error rate. For certain
streams (SN), monthly data can be used to identify the month of the year with
the maximum single daily flow with up to 80% accuracy. However, for others
(e.g., PR2) the correspondence is less than 50%. These results caution
against using monthly flow data to accurately characterize the spate regimes
for most streams. Many streams experience high flows of ecological
significance more than once per year, and as the number of high flows of
interest increases, it is increasingly unlikely that monthly data can
adequately capture the temporal distribution of those transient but important
events. By contrast, monthly data are probably adequate for analysis of
39
-------
lowflow events (which we did not analyze here), because lowflow events are
generally of greater duration than high flow events.
The two groups with the highest day-month match were SN and HI streams.
Given the high seasonality of high flows in snowmelt streams, it is not
surprising that these streams would be relatively insensitive to the temporal.
scales used in this study. The high correspondence across temporal scales for
harsh intermittent streams can probably be explained by considering that these
streams are defined as having zero flow for at least half of each year, on
average. Thus, any large flow occurring in these streams, however transient,
will probably make a significant contribution to flow values averaged over
longer time scales.
The rank correlations between the series of annual average flows and
annual maximum daily flows were always statistically significant (p < 0.05)
for individual streams (rs with 35 d.f. > 0.30). This suggests that annual
flow data might be useful in reconstructing hydrologic extremes for some types
of streams for which only annual flow data exist or for which only
precipitation data exist. However, the low absolute value of the rank
correlations for most streams indicates that the relationship between high
annual flows and high flow years contains substantial scatter. Further, the
correlations provided do not take into account the possibility that, in
particular years, several high flows may occur that exceed the magnitude of
even the highest annual flow of some other years. Thus, the information
contained in the rank correlation structure cannot be used reliably to
describe the frequency of flooding of a particular intensity for many stream
types. However, for highly seasonal streams (e.g., snowmelt), where maximum
annual flows almost always occur at a particular time of year (during
snowmelt), average annual flow statistics may provide very useful information
in determining magnitudes of major flows on a yearly basis (Dahm and Molles
1990, see Figure 28).
CONCLUSIONS
Colwell's index of predictability is sensitive to the time scale of data
analysis for some streams but not for others. The sensitivity varies
according to previously-defined stream types (i.e., those derived in Section 1
of this report). As the time scale of analysis increases (from daily to
40
-------
weekly to monthly to seasonal), some stream types become more predictable (SN,
SRI, SR2, GW1, PR1, and PF), others less predictable (IF and HI) and the
remainder do not change (PR2, GW2, SS, and IR). These patterns indicate the
importance of regional climatic conditions and local catchment characteristics
in influencing the calculation of predictability at different time scales.
Thus, the use of Colwell's index to assess streamflow predictability requires
some justification for the selection of a particular time scale.
For analysis of the high flow disturbance regimes across different
stream types, monthly data are not capable of capturing the information
available in the daily hydrograph for most streams. Only snowmelt and harsh
intermittent streams are consistently above 80% in the correspondence between
the timing of high monthly flows and months having the annual maximum flow.
The rank correlation between the annual mean flows and the annual peak
daily flows is statistically significant for all streams types. Intermittent
streams generally express higher rank correlations than perennial streams.
The low absolute value of the correlation for several stream types (especially
abundant eastern PR and GW streams) indicates that annual flow data are
limited in their ability to confidently extract information on high flow
regimes for streams where daily data are lacking.
41
-------
SECTION 3
FISH COMMUNITY STRUCTURE ALONG HYDROLOGIC GRADIENTS IN
WISCONSIN AND MINNESOTA STREAMS, AND SOME IMPLICATIONS
FOR COMMUNITY RESPONSE TO CLIMATE CHANGE
INTRODUCTION
How ecological communities will respond to rapid climate change is a
difficult question for ecologists to answer for several reasons. First, we
generally lack sufficient ecological data that can serve as a baseline against
any inferred change. Second, high natural variability in both environmental
signals and in community composition makes responses to anthropogenic climate
change difficult to document. This is particularly the case when data records
are short. Third, limited resources typically constrain our ability to gather
the types of information required to rectify the first two problems.
Moreover, suffused into each of these concerns is the inescapable issue of
"scale": At what levels of spatial, temporal, and ecological resolution must
we conduct work in order to establish bona fide patterns? Indeed, the problem
of scale has emerged in the last few years as arguably the major general
intellectual hurdle for ecologists to overcome in producing a unified and
predictive science (see O'Neill et al. 1986, Wiens 1989, Allen and Hoekstra
1992, Levin 1992).
Recent volumes on ecological responses to climate change (Regier et al.
1990, Firth and Fisher 1992, Kingsolver et al. 1992) have drawn attention to
the necessity of ecologists' "scaling-up" from their traditional local focus
to a broader regional research context, because climate change will induce
alterations at the regional scale. At this broad level of spatial resolution,
experimental studies are generally not feasible. Instead, comparative studies
that incorporate multivariate analyses are often appropriate tests of large-
scale hypotheses (Diamond 1986, Ricklefs 1987, Brown and Maurer 1989),
assuming sites for comparison are well-matched and historical processes are
comparable across sites (Orians 1987, Tonn et al. 1990). The application of
well-designed comparative studies is likely to provide timely information on
potential ecological responses to climate change and to suggest critical
42
-------
ecological experiments (Pace 1993, cf. Cole et al. 1991) . This comparative
approach has been successfully applied to fish assemblages (Mahon 1984, Tonn
et al. 1983, 1990, Jackson and Harvey 1989).
In community ecology, an emerging view is that the dynamic interplay
among species in a local community is constrained by larger-scale
environmental factors and available species pool (Ricklefs 1987, Roughgarden
1989, Menge and Olson 1990). This implies that when ecological questions are
asked at regional scales, information on abiotic constraining factors and
historical processes that determine the regional species pool will be crucial
to understanding the regional pattern. Local processes or habitat constraints
may be need to be invoked to explain residual variation in the regional
pattern (Tonn 1990, Duarte 1991). For questions of variation in community
structure across geographic scales, consideration both of accompanying
variation in regional environmental factors and of constraints imposed by
regional species pool are clearly needed. To the extent that local-regional
relations for lotic communities can be established and related to climate, a
potentially predictive basis can be developed for assessing ecological
responses to future climate change.
In stream systems specifically, local community structure is known to be
influenced by regional climatic factors. For example, temperature directly
constrains performance of aquatic species and latitudinal/altitudinal
gradients in thermal regimes define geographic distributions for species, such
that the implications of regional warming for certain fish species can be
evaluated (e.g., see Meisner 1990a) . Also, energy inputs into stream
ecosystems directly influence local trophic characteristics (Vannote et al.
1980), and differences among biomes in vegetation biomass and growth form
result in significant variation in community structure (Corkum and Ciborowski
1988, Corkum 1989). Anthropogenic activities that alter natural thermal
regimes (e.g., by impounding free-flowing rivers, Ward and Stanford 1979) or
modify the energy input base (e.g., through land-use changes, Karr et al.
1986) can radically alter lotic communities, and are indicative of the kinds
of changes that natural climate change likely will engender.
One climatically-influenced factor that can significantly constrain
lotic community structure is the hydrologic regime. Extremes of flow (spates,
droughts) and variable flow can directly influence community patterns, as has
been demonstrated by a number of studies for both fish (Horwitz 1978, Coon
43
-------
1987, Fausch and Bramblett 1991, Bain et al. 1988, Jowett and Duncan 1990,
Meffe 1984) and invertebrates (reviewed in Resh et al. 1988, Poff and Ward
1989, Fisher and Grimm 1991) . For individual fish species, spates may serve
as sources of direct mortality for both juvenile (Seegrist and Gard 1972,
Hanson and Waters 1974, Schlosser 1985, Harvey 1987) and adult (Toth et al.
1982, Schlosser and Toth 1984) fishes, and the timing of high flows may serve
as environmental cues for spawning (John 1963, 1964, Nesler et al. 1988) .
Although ecologists have long intuited the importance of hydrologic regime in
structuring lotic community structure, the theoretical basis for this idea was
not considered until the 1980's, when stream ecologists explored the relevance
of marine ecology's harsh-benign hypothesis (Peckarsky 1983) and intermediate
disturbance hypothesis (Ward and Stanford 1983) to lotic communities. As the
importance of physical disturbance to streams becomes increasingly recognized
(Resh et al. 1988), interest has grown in testing the hypothesis that
significant variation in community structure among streams is explicable in
terms of hydrologic patterns, which can vary substantially over even short
geographic distances (Poff and Ward 1989, Biggs et al. 1990). In the past few
years, much attention has been paid to the application of Southwood's (1977,
1988) "habitat template" idea to lotic communities (Frissell et al. 1986,
Minshall 1988, Poff and Ward 1990, Schlosser 1987) . If habitat (and resource)
availability and duration vary among locations, then differences in species
assemblages across these habitats should reflect, in some measurable way, the
differential abilities of species to persist and succeed under local
environmental conditions. In other words, species should possess the
attributes that enhance fitness given a particular template. . If the template
changes, due to climate change (or other factors) then one expects shifts in
species composition or functional organization to occur (Carpenter et al.
1992, Grimm 1992, Poff 1992a).
Of course, hydrologic regime alone is not expected to fully explain
patterns in community structure, because other important habitat features,
known to have local influence, are to differing extents independent of
discharge (e.g., channel morphology, substrate, gradient). The. importance of
these other physical habitat factors has been amply documented in the
literature (e.g., Angermeier 1987, Bozek and Hubert 1992). Additionally,
local heterogeneity in habitat features can create flow refugia (Sedell et al.
1990, Lancaster and Hildrew 1993) that surely disguise the signal a hydrologic
44
-------
regime imposes on community structure. Nonetheless, at regional (among-
stream) scales, there is reason to hypothesize that a substantial and
ecologically interesting portion of variation in lotic community structure may
reflect broad-scale hydrologic constraints (Poff 1992a), because stream
discharge can serve as an integrator of catchment-scale processes (Resh et al.
1988) .
If some relationships between fish community structure and hydrologic
regimes can be established based on present patterns, then it should be
possible to infer something about likely alterations in community structure
under scenarios of climate change that modify existing hydrologic regimes in
specific ways. Demonstration of such relationships would provide a basis for
identifying communities (or community types) "at risk" given projected changes
in climate. A similar point of view has been expressed for freshwater fish
assemblages by Tonn (1990), who notes that, even in lakes, hydrologic factors
can indirectly maintain local assemblage types by influencing dispersal
success, extinction rates and colonization rates of species in the regional
species pool.
In attempting to find fish-flow patterns, it is critical to insure that
there is agreement between the scale at which the community and the
environmental factors are viewed. Margalef (1968) pointed this out when he
argued that large-scale patterns are best detected with coarse-grain data.
Hydrologic data can be described by any number of arbitrary time scales, from
instantaneous to multi-annual. The most appropriate approach is to scale the
hydrologic variables to the ecological data. For ecological factors, the
units of measurement can profoundly alter the perception of pattern (e.g.,
Allen and Starr 1982, Rahel 1990, Allen and Hoekstra 1992). Relative
abundance data provide fine-grain information because they emphasize local
peaks in species performance, while species presence/absence data emphasize a
coarser grain of environmental tolerance (Allen and Skagen 1973, Allen and
Starr 1982).
The method of aggregation of the ecological entities also influences
perceived pattern. Two traditional approaches to viewing communities are in
taxonomic terms (species identities) or in functional terms (aggregations of
species into guilds of species possessing similar ecological roles and needs).
When investigating environment-community patterns across zoogeographic scales,
where species compositions naturally change, a functional perspective may be
45
-------
necessary to provide a basis for comparison of taxonomically dissimilar
communities (see Schoener 1986, 1987). Further, if environmental change
adversely affects species in a community, it is likely to result in a common
response among species of similar functional attributes. Such functional
analyses are the basis for much of comparative ecology of both stream fish
(Karr et al. 1986) and invertebrates (Vannote et al. 1980).
The primary goal of this research was to test the hypothesis that fish
assemblage structure varies among unregulated streams as a function of
ecologically important components of the hydrologic regime, as speculatively
proposed by Poff and Ward (1989). In order to do this, we identified
locations where both fish assemblage data and hydrologic data were available
so that quantitative relations could be established. We wanted to compare the
usefulness of taxonomic vs. functional approaches to assessing the habitat-
assemblage relationships; therefore, we defined several functional attributes
that we expected to be sensitive to hydrological variation and thus provide a
basis for identifying hydrological-biological associations. These attributes
included life history variables, trophic guild characteristics, body
morphology, habitat preferences (stream size, microhabitat, and substrate),
and environmental tolerance (see Methods for full description).
METHODS
Fish Data
Fish data from sites in three states (AR, MN, WI) were collected from
various sources and examined for proximity to existing USGS gauging stations
in those states. Fish sampling data from all three states were downloaded
from the EPA's national repository of fish and water quality storage and
retrieval system (STORET). Additionally, private source data were acquired
from each state: the Master Fish and Waterbody file from the Wisconsin DNR
(Fago 1992), a stream survey from the University of Minnesota Bell Museum of
Natural History, and a collection of samples taken during a study by the
Arkansas Department of Pollution Control (Giese 1987) . All data are currently
housed in the ERL-D Global Climate Change Information Management System.
Data collected in this fashion pose problems, including variable
motivations for collection, dissimilar collection techniques and efficiency,
and differential taxonomic resolution and accuracy across sites and states.
46
-------
These problems are nearly impossible to rectify because the collections often
date back years to decades, thus invalidating reconstruction of original field
survey conditions, taxonomic accuracy, etc. Despite these limitations,
archived (historical) fish collection data can be used to examine specific
ecological hypotheses (e.g., see Horwitz 1978), though the strength of the
interpretations must be qualified by the uncertain quality of the available
data.
Locational information from the fish sample data was converted from its
native state in a personal computer database to a geographic information
system (CIS) coverage so it could be examined along with the USGS gauging
stations. All fish collections on the gauged stream that were within a 15 km
radius of the stream gauge were included as candidate sites in the dataset.
For each gauged site, a map was generated that showed the local stream network
and the locations of the USGS gauge and the candidate fish sampling sites.
This list of potential sites was then closely examined to make sure that
candidate sites were on the gauged stream and not on small tributaries or in
adjacent lentic habitats. Several individual collections may have occurred
within the 15 km radius of the gauge over a period of several years. Such
spatial replicates were considered to be individual collections occurring at a
unique location on the stream. Repeated samplings at any unique location were
considered as temporal replicates for that location. For each site, the
number of spatial and temporal replicates was tallied over the period for
which fish data were available.
Derivation of functional measures
Life History Variables
Life history traits are considered to be products of natural selection
(Stearns 1976), and thus provide ideal functional descriptors for assessing
variation in community structure along environmental selection gradients.
Stream ecologists have used this approach to hypothesize that community
structure should reflect variation in hydrologic selective pressures and thus
show some predictable within- and among-catchment differences (e.g., Schlosser
1987, Minshall 1988, Resh et al. 1988, Poff and Ward 1989, 1990, Biggs et al.
1990, Poff 1992a). Despite the theoretical appeal of using life history
variables to describe the functional composition of stream fish communities,
the available compilations of appropriate life history information (Winemiller
47
-------
and Rose 1992, Detenbeck et al. 1992) are inadequate for the present analysis
because too. many species are not included in these databases.
Body morphology measurements
Fish ecologists have long recognized variation in body morphology of
fishes from different lotic biotypes, such as riffles, runs and pools (see
Nikolskii 1963, p. 75). Based on limited work, it appears that dominant
morphological types in lotic fish assemblages can be influenced by hydrologic
variability (Bain et al. 1988). Discussions with Dr. Paul Webb (University of
Michigan) resulted in selection of two morphological ratios (Webb and Weihs
1986) that are likely to describe morphological relationships to. the
hydrologic environment: 1) A body shape factor defined by the ratio of total
body length (TBL) to maximum body depth (MBD) describes the hydrodynamic
profile of the fish (e.g., fusiform vs. bluff-body profile), which influences
energetic costs of position maintenance. 2) The caudal peduncle depth factor
is the ratio of minimum depth of the caudal peduncle (MDCP) and the maximum
caudal fin depth (MCFD). This factor is a rough correlate for swimming
ability of the fish, in that fish having a large ratio are relatively strong
swimmers (e.g., thunniform fishes).
Morphological measurements for fishes were taken mainly from the
Peterson Field Guide to the Freshwater Fishes (Page and Burr 1991) and the
Atlas of North American Freshwater Fishes (Lee et al. 1980) . Page and Burr
present illustrations taken from live fishes or photographs of live or freshly
killed fishes. Lee's atlas includes photographs of fishes. Measurements for
three species not available from these two sources were taken from Scott and
Grossman (1974) and Trautman (1981).
Four measurements were required to produce the two ratios. Total body
length (TBL) was the overall length of the pictured fish, including all of the
caudal fin and all of the mouth. This was the maximum length of the picture
on the page. Maximum depth of body (MDB) was the maximum height of the body
only. Fins, stickles, and other appendages were not included in this
measurement. Minimum depth of the caudal peduncle (MDCP) was the smallest
point of the caudal peduncle, not including any fins. Maximum Caudal Fin
Depth (MCFD) was the maximum height of the caudal fin, regardless of shape.
All measurements were taken using a dial caliper graded to 0.254 mm (=
0.01 in). In some cases, an estimate was made to the nearest 0.127 mm.
48
-------
Because the ratios are dimensionless, the units of measurement for the four
variables needed only to be consistent for a particular fish. The photographs
from Lee's atlas may have introduced some inaccuracy in the cases where the
caudal fin was slightly compressed from top to bottom, where photographs were
too light in the caudal fin section to allow accurate measurement, or where
photographs were trimmed too much for inclusion in the atlas. (Some photos
had noticeable trim lines, with extra portions of the background showing, and
some had no background showing, which could mean a portion of the fish was
trimmed from the photo.)
Trophic Guild
A number of authors have proposed trophic or feeding categories for
stream fishes (Horwitz 1978, Moyle and Li 1979, Grossman et al. 1982,
Schlosser 1982) . K.R. Allen (1969) offered a simple classification for North
American stream fishes: most are invertivores, some become piscivorous later
in their life cycle, and a few are herbivorous. While broadly correct, this
scheme fails to capture significant additional variation in feeding habits,
primarily in terms of where and how fishes capture their food (Table 11).
Some invertivores feed primarily from the benthos, others from the water
column and surface, and still others appear to be generalists. There also is
considerable variation in the amount of plant matter consumed. Fishes whose
diet included substantial plant matter were classified as herbivore-
detritivores, while those reported to ingest only occasional amounts of plant
matter were classified as omnivores. Planktivore and parasitic categories are
also represented in our data set. Fish were assigned to trophic categories
based on descriptions in Lee et al. (1980) and Scott and Grossman (1974) .
Habitat Classification
Habitat preferences were established from descriptions of fish habitat
included in references describing stream fishes from North America of the
midwestern region (Becker 1983, Lee et al. 1980, Scott and Grossman 1974,
Trautman 1981) . Although there is a large amount of literature on fish
habitat utilization, to our knowledge there are no generally accepted habitat
categories parallel to trophic categories. We first examined habitat
descriptions from standard references, and then by trial and error developed
49
-------
categories that appeared to allow useful separation of species into major
habitat categories in terms of current, substrate, and stream size. For each
TABLE 11. FUNCTIONAL MEASURES OF FISH SPECIES. SCORES FOR EACH
SPECIES GIVEN IN APPENDIX D
Body Morphology (continuous)
1. Swimming Factor
2. Shape Factor
Trophic Guild (categorical)
1. Herbivore-detritivore
2. Qmnivore
3. General Invertivore
4. Surface/Water column Invertivore
5. Benthic Invertivore
6. Piscivore3
7. Planktivore
8. Parasite
Habitat Classification
Flow Habitat (categorical)
1. Fast
2. Moderate
3. Slow-none
4. Generalist
Substrate Preference (categorical)
1. Rubble (rocky, gravel)
2. Sand
3. Silt
4. Generalist
Stream Size (categorical)
1. Smallb
2. Medium-Large0
3. Small-Large4
4. Lake
Silt Tolerance (categorical)
1. Intolerant
2. Moderately Tolerant
3. Tolerant
a includes fishes feeding on large invertebrates such as crayfish
b fishes of small streams and headwaters, and of both small and medium
streams
c found in medium-sized streams and large rivers
<* reported in small, medium, and large streams and rivers
50
-------
habitat category, subcategories were established (Table 11). All efforts to
place species into habitat categories were hampered to varying degrees either
by lack of specific information in published descriptions, or by the species'
apparent breadth of habitat use. In a number of instances it was necessary to
use a "generalist" category.
Environmental Tolerance
Tolerance refers to tolerance of silt and is based on expert opinion.
Species in our database that were not classified into tolerance categories by
the Ohio EPA (1979) were given tolerance scores based on the other references
cited above and on expert opinion (see below). Three tolerance categories
were established (Table 11).
Confidence
Our confidence in the accuracy with which fish species are placed into
trophic, habitat, and tolerance categories is variable. Many species
inhabited a range of categories, and information varied in its completeness,
sometimes appearing inconsistent. After developing the categories of Table 11
and assigning each species a code, the results were submitted to expert fish
biologists for evaluation (G.R. Smith, Museum of Zoology, University of
Michigan, Paul Seelbach, Institute for Fisheries Research, Michigan Department
of Natural Resources). However, it is not yet possible to designate all
species unambiguously into appropriate categories. When sufficient
information was available and all sources agreed, a code of 3 (high
confidence) was given. When little information was available, when sources
conflicted, or when the biological information did not allow a species to be
well described by a single category, a confidence code of 1 or 2 was assigned.
Species' scores for the attributes listed in Table 11 are given in Appendix D
(and also provided on diskette).
Hydrologrie Data
Variables for each gauged site were derived from the long-term
hydrologic dataset and named in Section 1. The following variables were used:
DAYCV, DAYPRED, FLDFREQ, FLDPRED, FLDFREE, BFI, ZERODAY, LOWPRED, LOWFREE.
These variables represent long-term averages over variable periods of record
51
-------
TABLE 12. SAMPLING CHARACTERISTICS OF 34 WIMN SITES. FOR EACH SITE, THE NUMBER OF
UNIQUE SPECIES COMBINED ACROSS ALL COLLECTIONS SINCE 1960 ARE GIVEN, AS ARE THE
TOTAL NUMBER OF COLLECTIONS AND THE PERIOD OF AVAILABLE HYDROLOGIC DATA
Gauge #
4063700
4078500
4080000
4081000
4085200
4086500
5069000
5293000
5300-000
5311400
5313500
5315000
5316500
5317000
5332500
5333500
5367500
5368000
5374000
5379500
5381000
5383000
5394500
5397500
5406500
5413500
5414000
5415000
5423000
5423500
5424000
5432500
5433000
5543830
state
WI
WI
WI
WI
WI
WI
MN
MN
MN
MN
MN
MN
MN
MN
WI
WI
WI
WI
MN
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
number
species
22
41
44
51
25
35
21
25
34
25
27
21
32
29
31
24
36
34
23
54
24
51
24
21
32
35
45
26
31
22
34
39
36
34
number
samples
4
12
6
17
5
32
21
3
9
5
3
4
5
2
6
2
10
5
3
40
5
21
4
4
11
7
16
6
16
14
19
9
9
8
streamflow
record
1964-1986
1929-1985
1929-1970
1929-1963
1967-1986
1931-1970
1948-1983
1940-1985
1934-1985
1961-1981
1940-1985
1941-1985
1936-1985
1939-1985
1929-1970
1929-1981
1929-1961
1951-1986
1931-1980
1935-1986
1929-1986
1929-1970
1940-1986
1940-1986
1955-1986
1935-1986
1935-1986
1940-1986
1950-1970
1949-1969
1950-1970
1940-1986
1940-1986
1964-1986
at gauged locations (see Table 12). (The derivation of these variables is
provided in detail in Section 1 of this report). All gauged stream sites in
the database within within 15 km of a fish sampling station were screened for
unacceptable sources of hydrologic disturbance (impoundment, hydroelectric
facilities, irrigation withdrawal, etc.). The list of candidate sites was
then cross-checked against the two independently-derived datasets listing
stream gauges that have been stable over the entire period of record (Wallis
52
-------
et al. 1991, Slack and Landwehr 1992). Some sites did not appear on these
lists. Telephone conversations with U.S. Geological Survey personnel
responsible for original screening of sites for inclusion in the Slack and
Landwehr (1992) list confirmed that the unmatched sites could have been
included in the original listings, but typically were not because the gauges
had been discontinued several years ago. Thus, all sites used in the analysis
have acceptably unmodified hydrologic regimes.
Data Analysis
Fish Data Matrices
Several decisions had to be made to determine which data were suitable
for inclusion. First, we excluded all fish samples collected prior to 1960,
in order to keep the fish data relatively contemporary. A total of 24, 33 and
43 potential sites with fish data were identified for AR, MN, and WI,
respectively. The intensity of collections varied markedly among sites, as
did the number of species collected per site. The relationships between
species number and sampling intensity are shown for all the AR, MN, and WI
sites in Figure 29. In order to diminish the likelihood that any observed
pattern in assemblage structure simply reflected sampling intensity, a
decision was made to include only sites with >. 20 species present across all
spatial-temporal replicates, regardless of how many collections were made
(Table 12). We considered this an acceptable tradeoff between minimizing
artifacts caused by sampling intensity and retaining enough sites in the
database to perform a valid analysis.
All data from AR were excluded from further analysis because too few
sites had sufficient species richness. Fish diversity in AR is very high
(Robison and Buchanan 1988), and the paucity of species in the EPA database
indicates a problem with the original data source in terms of sampling
efficiency and/or archival techniques. In WI and MN, 25 and 9 sites,
respectively, contained £ 20 species and were acceptable hydrologically. To
increase sample size, sites from these two states were combined into a
"regional" dataset (hereafter referred to as WIMN) containing 34 locations
(Figure 30). This is reasonable since the two states are adjacent and share
common drainages.
A linear relationship between species richness and sampling intensity
persisted when the 34 WIMN sites were considered together (Figure 31), though
53
-------
there was much scatter about the line of central tendency. This relationship
indicates that any observed hydrological-ecological pattern must be carefully
examined for possible confounding with sampling intensity. Another
potentially confounding factor is stream size, which can influence taxonomic
composition and species richness. Neither the entire WI and MN datasets
(Figure 32) nor the 34 WIMN sites (Figure 33) showed any systematic
relationship between species richness and catchment area (a strong correlate
of stream size).
Data from the WIMN sites varied in terms of its information content.
Almost half of the retained sites contained only presence/absence (+/-) data.
Thus, it was necessary to collapse all abundance data from sites into binary
data. Given the aforementioned problems with the original data, an analysis
based solely on +/- data is prudent.
Community Structure: Taxonomic Organization
The data for this analysis consisted of a 34 site by 106 species matrix.
The 106 species represented the sum of unique species present across all sites
at all times (Table 13). In the original dataset, 12 "species" were
identified as unknowns and eliminated from the analysis. Each entry in the
matrix was either a "1" (species present at site) or an "0" (species absent).
There were two objectives of this taxonomic analysis. First, we wished to
determine similarity of sites based on species composition and use
multivariate techniques to explore any patterns that related to among-site
hydrologic variables. To do this we used a variety of classification and
ordination techniques (described below). Second, we wished to test the
hypothesis that taxonomic composition across sites was explainable in terms of
hydrologic variables. To do this we used discriminant analysis.
Ordination and ClassificationA variety of multivariate techniques can
be used to explore patterns in ecological communities (see Gauch 1982, Ludwig
and Reynolds 1988). Two frequently used approaches are ordination and
classification. Ordination summarizes community data by reducing a complex
multi-dimensional dataset to a low-dimensional (typically 1-3) space in which
similar samples (sites) fall closely together along species' gradients and
dissimilar samples are well separated (Gauch 1982) . The positions of sites in
54
-------
species space can then be compared with environmental information for the
sites to see if differences in community composition correlate with the
TABLE 13. LIST OF UNIQUE SPECIES RECORDED ACROSS ALL 34 WIMN SITES (UPPER PANEL)
AND NUMBER OF SITES AT WHICH EACH SPECIES OCCURRED (LOWER PANEL). NAMES
FOR SPECIES ABBREVIATIONS ARE GIVEN IN TABLE 16
BMS
GLR
LKC
IOD
BNS
RBT
RVD
WDS
LCS
LND
ROS
OSS
WHC
WHB
COS
RSD
ABL
GPK
SDS
CCF
BLC
QBS
SNG
CNM
3RD
MOE
TRP
NOP
BLG
FTD
LNG
BKT
BTM
MOD
CSH
SAR
TPM
RBD
RCS
SIL
RES
HFS
JND
WAE
YEP
BDD
SAB
BUB
GRR
SLC
WTS
FWD
BST
NHS
SPC
LKS
SRS
BOF
SPS
SHR
SLR
SMB
SKM
LSR
GLD
AME
CRC
SRL
CAP
RKB
LGP
PMK
CHL
MSM
FHM
BLM
SPO
BHM
NRD
LMB
RRH
GOE
BSD
HHC
SHD
RVS
MTS
MMS
MUE
OZM
8KB
END
GSF
BNT
PRO
WSD
BKS
BLT
SCT
EMS
BIB
CIS
FND
YEB
FCF
BRB
20
19
5
3
6
9
1
1
1
16
14
9
5
6
8
1
3
1
21
10
12
10
2
12
9
3
2
24
17
17
3
4
1
1
34
8
8
9
5
4
2
3
30
18
14
13
2
8
3
1
33
6
11
16
1
1
1
2
23
21
18
20
8
10
2
1
33
8
24
15
10
11
5
1
24
30
7
4
5
11
1
1
23
26
11
6
6
5
1
1
18
18
18
13
4
2
4
1
22
14
5
4
1
9
3
1
environmental data. Of the many ordination techniques available, detrended
correspondence analysis (DCA, Hill 1979a) has many favorable characteristics
which combine to make it a powerful technique for examining community
relations (Gauch 1982). The principal strength of DCA is that it imposes the
stringent condition that second and higher order axes have no systematic
relations of any kind to lower axes, unlike all other ordination techniques
(e.g., principal components analysis, polar ordination, canonical correlation
analysis), which simply constrain higher axes to be linearly uncorrelated
(orthogonal) with lower axes (Gauch 1982) . DCA has been used extensively in
the ecological literature, including in the analysis of lotic community
structure for both abundance and binary data (Ormerod 1987, Ormerod and
Edwards 1987, Moss et al. 1987, Poff et al. 1990, Rahel and Hubert 1991,
Boulton et al. 1992).
55
-------
Classification is a method of placing ecological entities (species,
sites) into classes or groups. In this paper, sites having similar taxonomic
composition would group together while dissimilar sites would not, based on a
measures of multivariate distance in overall species space. Hierarchical
clustering routines produce dendrograms that can be viewed at various levels
of resolution to identify various numbers of classes or groups. Hierarchical
classification is best suited to small datasets (n < 100) whose hierarchical
structure can be visually comprehended (Gauch 1982). Of the many techniques
available for classification of ecological community data, two-way indicator
species analysis (TWINSPAN, Hill 1979b) has the unique property of
"deliberately arrang[ing] the two clusters at each node in the way that
results in placing the most similar samples together in the dendrogram1s
sample sequence," and thus facilitating interpretation (Gauch 1982, p. 201).
TWINSPAN has been widely used as a classification tool for ecological
communities, including those in streams (Wright et al. 1983, Ormerod 1987,
Ormerod and Edwards 1987, Moss et al. 1987, Poff et al. 1990) . It is also
appropriate for binary data and is often used as a complement to DCA.
Discriminant AnalysesCanonical discriminant analysis (CDA) is a
multivariate technique that derives canonical variables (linear combinations
of quantitative variables) that have the highest possible multiple correlation
with previously-defined classes (SAS 1988). In this paper, the classes are
groups of ecologically-similar sites and the quantitative variables are the 9
hydrologic variables for each of the 34 WIMN sites. The maximum number of
canonical variables that can be derived equals the number of previously-
defined groups, minus one. CDA is a three step process. First, classes
(groups of sites) are defined in ecological terms that are independent of the
hydrologic variables. Second, linear combinations of the hydrologic variables
are derived that best describe the separation of the classes. Third, the
hypothesis that class means are the same in terms of the hydrologic variables
is tested, as is the hypothesis that the correlation between the canonical
variables and the ecological groups is equal to zero. Rejection of these
hypotheses indicates that the hydrologic variables can discriminate among the
ecologically-defined groups. Interpretation is based on examining the
correlations between the original hydrologic variables and the canonical
variables to see which hydrologic variables are most important in explaining
56
-------
the discrimination among the classes. CDA requires the assumption of
multivariate normality of the (hydrologic) variables to be examined. A good
rule of thumb is that if all variables show univariate normality, multivariate
normality is a reasonable assumption. Visual inspection of the 9 hydrologic
variables indicated that all but ZERODAY (which has many zero values) were
approximately normally distributed. CDA is essentially a dimension reduction
technique useful in exploring ecological-environmental relations, and it has
been used extensively in the ecological literature, including in studies of
fish ecology (e.g., Hawkes et al. 1986, Bozek and Hubert 1992, Nelson et al.
1992) .
Discriminant function analysis (DFA) is another, related multivariate
technique that classifies observations on the basis of one or more
quantitative variables into two or more known (previously defined) groups
(Johnson and Wichern 1982, SAS 1988). DFA also produces a quantitative
function relating group affiliation (based on ecological similarity) to
environmental variables, and it has two major strengths. First, it can be
used to determine the classification error rate for the original observations.
If the quantitative function derived from the hydrologic variables can
completely discriminate among the previously-defined ecological groups, then
the DFA is 100% effective in predicting ecological class from environmental
data. Second, the quantitative function can be used to predict the membership
of new sites into ecological groups based only on environmental data (e.g.,
see Marchant et al. 1984, Bozek and Hubert 1992, Norris and Georges 1993).
DFA also requires the assumption of multivariate normality of the (hydrologic)
variables to be examined. Two analyses were made with DFA: a non-parametric
test including ZERODAY and a parametric test excluding ZERODAY from the
analysis (see SAS 1988, p. 360 ff.).
Three steps were required to test the hypothesis that taxonomic
composition of fish assemblages varied with hydrologic factors for the 34 WIMN
sites. First, assemblage similarity among sites was calculated using the
Jaccard coefficient, a measure specifically designed for binary data (Dyer
1978) . Second, sites were classified into 2-3 groups based on similarity
using a hierarchical clustering routine (Ward's minimum variance method) on
SAS. A separate analysis was performed where groups were defined by the
TWINSPAN classification. Third, hydrologic factors for each site were used as
57
-------
quantitative variables to try to discriminate (using CDA and DFA) among the
previously-defined ecological groups.
Community Structure: Functional Organization
The data for this analysis consisted initially of a 34 site x 25
functional attribute matrix. The structure of this matrix requires some
detailed explanation. As described above, for each unique species in the
overall dataset, functional information was available for 25 individual
categories and subcategories. The two morphology categories contained
continuous data, whereas the 5 trophic, habitat, and tolerance categories
(total of 23 subcategories) contained only categorical data (see Table 11).
For each site (row in the matrix), the following procedure was applied: for
the continuous (morphological) variables, the average value for all species
present was calculated and entered, whereas for the categorical attributes,
the proportion of all species falling into subcategories within a major
category was determined. For example, if 40 species were present at Site Z,
the value entered in the matrix for.the first morphology attribute was
calculated by averaging the 40 species' values for that attribute. As an
example of the categorical calculation, the trophic category contained 8
subcategories. If 10 of the hypothetical 40 species present at Site Z were
"omnivores", and 4 species were "herbivores", then 0.25 and 0.10 would be
entered as the omnivore and herbivore attribute scores, respectively, for Site
Z. The remaining six trophic subcategory proportions were similarly
determined, so that within the trophic attribute category, all subcategory
scores would sum to 1.00 (or slightly less in some cases where a species with
an undefined attribute occurred). This step was repeated for each of the five
major categorical attributes. After the matrix was constructed, the variables
were standardized to mean of 0 and standard deviation of 1 and a correlation
matrix was derived that described the similarity among sites in the functional
attribute space (i.e., sites with functionally similar fish assemblages would
be highly correlated). This correlation matrix (dimension 34 x 34) was then
used as input into a hierarchical cluster analysis (Ward's method) using SAS
and 2-3 ecologically-similar groups were specified and defined. The groups
were then used in DFA and CDA to test the null hypothesis that ecologically-
similar assemblages (defined in functional terms) cannot be discriminated by
the hydrologic variables.
58
-------
Ward's method defines distance between two clusters as the "ANOVA sum of
squares added up over all the [functional] variables" (SAS 1988, p. 296). The
procedure assumes a multivariate normality and is sensitive to outliers.
Visual inspection of normal probability plots of the 25 functional attribute
variables showed that almost all were approximately normally distributed.
Exceptions included only those variables that were not represented at all
sites (e.g., parasitic trophic category).
For the major functional attributes (Table 11), means and standard
errors for the groups identified by cluster analysis were plotted. Oneway
analysis of variance (SAS 1988) was used for each functional category to test
the hypothesis of no statistically significant difference among groups. To
meet parametric assumptions, morphological data were log-transformed and
other, proportional functional data were arcsin-transformed prior to
performing ANOVA (Steel and Torrie 1980). Differences among groups were
tested with the Student-Newman-Keuls multiple comparison test, with a = 0.05.
As a check against possible violations of the parametric assumption of
normality, non-parametric one-way analysis of variance was also performed
using the Wilcoxon (2-sample) and Kruskal-Wallis (3-sample) tests (Steel and
Torrie 1980, SAS 1988).
RESULTS
Hydrologic Variables
The long-term averages for the various hydrologic variables for each of
the 34 WIMN suites are given in Table 14. Of the 34 sites, 21 were cross-
referenced by Slack and Landwehr (1992) and Wallis et al. (1991) and were
classified according to the methods presented in Section 1'of this report.
Six of the 10 classification types present in the entire U.S. were represented
by the 21 streams in WI and MN. Hydrologic characteristics for the 13
unclassified sites were visually inspected and an assignment was made to one
of the 10 groups defined in Section 1 (in parentheses in Table 14). Almost
all streams are perennial, with the greatest proportions represented by
superstable (SS, n - 12), mesic groundwater (GW, n = 6), and perennial runoff
(PR, n = 12) streams. Intermittent runoff (IR) streams occur twice, while
snow+rain (SR) and snowmelt (SN) streams were represented by one site each.
59
-------
The correlations among the hydrologic variables across the 34 WIMN sites are
given in Table 15.
Fish Species Occurrences
A total of 106 unique fish species were represented across the 34 WIMN
sites (Table 16). Of these 106 species, only one (common shiner) occurred at
all 34 sites, although 4 others were recorded at >30 sites (see Table 22,
"total sites" column). A total of 15 species occurred at >20 sites, while 66
species were found at 10 or fewer sites. Twenty species were recorded at only
one of the 34 WIMN sites (see Table 22).
Fish Taxonomic Relationships to Hydrologrie Factors
The TWINSPAN analysis resulted in the hierarchical dendrogram shown in
Figure 34. The number of groups of sites specified in the dendrogram depends
on the level at which the user defines a division. At 1 level of division, 2
groups resulted; at 2 levels of division, 4 groups of sites resulted; and , at
3 levels of division, 8 groups of sites were generated. Individual
"indicator" species that have high relative weighting in splitting a large
group of sites into two smaller groups are shown at each branch in the
dendrogram (Figure 34). For example, at the first division, largescale
stoneroller (LSR) is an "indicator" species for groups 1 and 2, while the
sand shiner (SDS), big-mouth shiner (BMS), common stoneroller (SRL), and plains
carpsucker (= quillback) (QBS) are "indicators" for groups 3 and 4. Indicator
species can occur at more than one branch point, as evidenced by SRL, which
occurs twice in the 2-level division. At the third division, groups 1 and 2
are distinguished largely on the basis of northern hogsucker (NHS), while
groups 3 and 4 are divided according to the proportional presence of SRL and
brown trout (BRT).
The geographic distribution of the sites comprising the four TWINSPAN
groups is shown in Figure 35. There is clear regional clustering of the sites
when clustered according to taxonomic affiliation, and these patterns probably
reflect zoogeography. TWIN 1 sites tend to occur along the Lake Michigan
shore of Wisconsin; TWIN 2 sites are located in northern, interior Wisconsin;
and, TWIN 4 sites are restricted to southwestern Wisconsin. TWIN 3 sites are
almost entirely restricted to Minnesota, with the exception of one SW
60
-------
TABLE 14. SUMMARY OF HYDROLOGIC VARIABLES FOR 34 WIMN SITES. CLASS REFERS TO STREAM CLASSIFICATION (FROM
SECTION 1, THIS REPORT). INFERRED CLASSIFICATIONS FOR PREVIOUSLY UNCLASSIFIED SITES ARE ENCLOSED IN
PARENTHESES. GROUP REFERS TO ECOLOGICAL GROUP (AS IDENTIFIED WITH CANONOICAL DISCRIMINANT
ANALYSIS SEE TEXT), WHERE 1 = "VARIABLE", 2 = "MODERATELY STABLE", AND 3 = "VERY STABLE".
TWIN REFERS TO THE 2-LEVEL TWINSPAN CLASSIFICATION FOR EACH SITE (REFER TO FIG. 34).
Gauge
4063700
4078500
4080000
4081000
4085200
4086500
5069000
5293000
5300000
5311400
5313500
5315000
5316500
5317000
5332500
5333500
5367500
5368000
5374000
5379500
5381000
5383000
5394500
5397500
5406500
5413500
5414000
5415000
5423000
5423500
5424000
5432500
5433000
5543830
State
WI
WI
WI
WI
WI
WI
MN
MN
MN
MN
MN
MN
MN
MN
WI
WI
WI
WI
MN
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
WI
Claaa
SR
GW
(GW)
(SS)
PR
(PR)
SN
PR
IR
(IR)
PR
(PR)
PR
PR
(SS)
SS
(SS)
GW
PR
SS
PR
(SS)
SS
SS
(SS)
GW
GW
SS
(PR)
(PR)
(PR)
SS
SS
(GW)
Group
1
3
2
1
2
1
2
1
1
1
2
1
1
1
3
3
3
3
2 .
1
3
1
3
3
2
2
2
1
1
1
1
2
2
1
Twin
1
2
2
2
1
1
3
3
3
3
3
3
3
3
2
2
2
2
3
4
2
4
2
2
4
4
4
4
1
1
1
3
4
1
DAYCV
103.55
102.71
95.82
44.45
211.14
186.68
212.20
257.11
236.15
321.88
231.90
204.08
201.53
192.94
39.91
54.18
84.89
103.88
178.29
93.53
230.84
60.40
97.12
146.29
68.09
174.55
161.39
191.44
196.57
205.16
167.96
155.34
121.09
112.63
DAY-
PRED
73.47
73.71
75.98
74.71
58.52
45.73
55.94
28.66
23.99
24.99
33.99
28.44
32.49
45.33
78.59
78.47
74.27
73.96
70.65
74.05
53.12
74.12
75.93
71.69
69.82
68.43
67.48
62.37
26.90
27.19
46.78
67.73
69.83
55.23
FLD-
FREO
0.43
0.70
0.62
0.57
0.50
0.88
0.58
0.76
0.87
0.67
0.70
0.78
1.06
0.89
0.74
0.60
0.73
0.58
0.74
0.60
0.76
0.60
0.68
0.60
0.66
0.85
0.77
0.66
0.95
0.86
0.81
0.60
0.70
0.57
FLD-
PRED
0.75
0.68
0.72
0.79
0.73
0.64
0.81
0.61
0.79
0.62
0.67
0.75
0.56
0.63
0.58
0.61
0.73
0.58
0.53
0.71
0.52
0.48
0.60
0.71
0.54
0.55
0.40
0.46
0.65
0.53
0.55
0.64
0.60
0.64
FLD-
FREE
0.45
0.35
0.24
0.23
0.29
0.37
0.69
0.50
0.59
0.54
0.59
0.61
0.28
0.32
0.41
0.54
0.44
0.22
0.42
0.24
0.38
0.31
0.42
0.44
0.28
0.14
0.17
0.29
0.35
0.23
0.19
0.42
0.30
0.37
BFI
0.24
0.25
0.34
0.55
0.14
0.08
0.10
0.02
0.01
0.00
0.02
0.02
0.02
0.05
0.51
0.53
0.37
0.48
0.17
0.46
0.04
0.48
0.38
0.22
0.60
0.41
0.37
0.34
0.04
0.02
0.05
'0.34
0.47
0.13
ZERO
-DAY
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.01
0.08
0.2
0.0
0.02
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.02
0.01
0.0
0.0
0.0
0.0
LOW-
EBEQ
0.63
0.56
0.48
0.34
0.57
0.39
0.50
0.60
0.48
0.48
0.50
0.54
0.39
0.29
0.47
0.35
0.53
0.57
0.58
0.43
0.55
0.47
0.38
0.40
0.36
0.41
0.62
0.30
0.50
0.50
0.80
0.42
0.54
0.71
LOW-
FREE
0.37
0.44
0.32
0.27
0.53
0.29
0.45
0.35
0.30
0.21
0.41
0.39
0.38
0.16
0.35
0.21
0.33
0.39
0.51
0.39
0.35
0.18
0.35
0.27
0.28
0.23
0.34
0.15
0.51
0.46
0.57
0.23
0.19
0.59
-------
TABLE 15. PEARSON CORRELATION MATRIX FOR HYDROLOGIC VARIABLES ACROSS 34 WIMN SITES. THE R-VALOES THAT
ARE SIGNIFICANT AT P < 0.05 USING A BONFERRONI TEST ARE INDICATED BY BOLDFACE AND "*"
AREA
AREA
DAYAVE
DAYCV
DAYPRED
FLDFREQ
FLDPRED
FLDFREE
BFI
ZERODAY
LOHPRED
LOWFREE
1
0
-0
0
0
0
0
0
-0
-0
-0
.000
.784*
.087
.098
.149
.120
.305
.005
.069
.224
.187
DAYAVE
1.000
-0.452
0.490
-0.165
-0.015
0.107
0.395
-0.199
-0.176
-0.201
DAYCV
1
-0
0
-0
0
-0
0
0
0
.000
.822*
.412
.013
.328
.843*
.521
.138
.120
DAYPRED
1.
-0.
-0.
-0.
0.
-0.
-0.
-0.
000
5»6*
074
327
853*
478
133
220
FLDFREQ
1
-0
-0
-0
0
-0
0
.000
.295
.118
.461
.078
.117
.037
FLDPRED FLDFREE BFI ZERODAY LOWPRED LOWFREE
1
0
-0
0
-0
0
.000
.496 1.000
.168 -0.386 1.000
.116 0.346 -0.331 1.000
.014 -0.007 -0.316 0.001 1.000
.169 0.001 -0.414 -0.180 0.700* 1.000
-------
TABLE 16. LISTING OF 106 TOTAL UNIQUE SPECIES
COLLECTED ACROSS ALL 34 WIMN SITES SINCE 1960
Abbrev .
ABL
AME
BCS
BDD
BHM
BIB
BKB
BKF
BKS
BKT
BLC
BLG
BLM
BLT
BMS
BND
BNS
BNT
EOF
BRB
BSD
BST
BTM
.BUB
CAP
CCF
CHL
CNM
CRC
CSH
EMS
FCF
FHM
FND
FTD
FWD
CIS
GLD
GLR
GOS
GOE
GPK
GRR
GSF
HFS
HHC
IOD
JND
LCS
LGP
LKC
LKS
Common Name
AMERICAN BROOKLAMPREY
AMERICAN EEL
BLACKCHIN SHINER
BANDED DARTER
BULLHEAD MINNOW
BIGMOUTH BUFFALO
BLACK BULLHEAD
BANDED KILLIFISH
BROOK SILVERS IDE
BROOK TROUT
BLACK CRAPPIE
BLUEGILL
BLUNTNOSE MINNOW
BLOATER
BIGMOUTH SHINER
BLACKNOSE DACE
BLACKNOSE SHINER
BROWN TROUT
BOWFIN
BROWN BULLHEAD
BLACKS IDE DARTER
BROOK STICKLEBACK
BLACKSTRIPE TOPMINNOW
BURBOT (LING)
CARP
CHANNEL CATFISH
CHESTNUT LAMPREY
CENTRAL MUDMINNOW
CREEK CHUB
COMMON SHINER
EMERALD SHINER
FLATHEAD CATFISH (YELLOW)
FATHEAD MINNOW
FINESCALE DACE
FANTAIL DARTER
FRESHWATER DRUM
GIZZARD SHAD
GILT DARTER
GOLDEN REDHORSE
GOLDEN SHINER
GOLDEYE
GRASS PICKEREL
GREATER REDHORSE
GREEN SUNFISH
HIGHFIN CARP SUCKER
HORNYHEAD CHUB
IOWA DARTER
JOHNNY DARTER
LAKE CHUBSUCKER
LOGPERCH
LAKE CHUB
LAKE STURGEON
Genus
Lamptera
Anguilla
Notropis
Etheostoma
Pimephales
Ictiobus
Ictalurus
Fundulus
Labidesthes
Salvelinus
Pomoxis
Lepomis
Pimephales
Coregonus
Notropis
Rhinichthys
Notropis
Salmo
Amia
Ictalurus
Percina
Culaea
Fundulus
Lota
Cyprinus
Ictalurus
Icthyomyzon
Umbra
Semotilus
Notropis
Notropis
Pylodictis
Pimephales
Phoxinus
Etheostoma
Aplodinotus
Dorosoma
Percina
Moxostoma
Notemigonus
Hiodon
Esox
Moxostoma
Lepomis
Carpiodes
Nocomis
Etheostoma
Etheostoma
Erimyzon
Percina
Couesius
Acipenser
species
lamottei
rostrata
heterodon
zonale
vigilax
cyprinellus
melas
diaphanus
sicculus
fontinalis
nigromaculatus
macrochirus
notatus
hoyi
dorsalis
atratulus
heterolepis
trutta
calva
nebulosus
maculata
inconstans
notatus
lota
carpio
punctatus
castaneus
limi
atromaculatus
cornutus
atherinoides
olivaris
promelas
neogaeus
flabellare
grunniens
cepedianum
evides
erythrurum
crysoleucas
alosoides
americanus
valenciennesi
cyanellus
velifer
biguttatus
exile
nigrum
sucetta
caprodes
plumbeus
fulvescens
63
-------
LMB
LND
LNG
LSR
HDD
MMS
MOE
MSM
MTS
MUE
. NHS
NOP
NRD
OSS
OZM
PMK
PRO
QBS
RBT
RED
RCS
RES
RKB
ROS
RRH
RSD
RVS
RVD
SAB
SAR
SCT
SDS
SHD
SIL
SHR
SKM
SLC
SLR
SMB
SNG
SPO
SPS
SPC '
3RD
SRL
SRS
TPM
TRP
WAE
WDS
WHB
WHC
WSD
WTS
YEB
YEP
LARGEMOUTH BASS
LONGNOSE DACE
LONGNOSE GAR
LARGESCALE STONEROLLER
MUD DARTER
MIMIC SHINER
MOONEYE
MISSISSIPPI SILVERY MINNOW
MOTTLED SCULPIN
MUSKELLUNGE
NORTHERN HOGSUCKER
NORTHERN PIKE
NORTHERN REDBELLY DACE
ORANGE SPOTTED SUNFISH
OZARK MINNOW
PUMPKINSEED
PEARL DACE
PLAINS CARPCUCKER (QUILLBACK)
RAINBOW TROUT
RAINBOW DARTER
RIVER CARPSUCKER
REDFIN SHINER
ROCK BASS
ROSYFACE SHINER
RIVER REDHORSE
REDSIDE DACE
RIVER SHINER
RIVER DARTER
SMALLMOUTH BUFFALO
SAUCER
STONECAT
SAND SHINER
SLENDERHEAD DARTER
SILVER LAMPREY
SHORTHEAD REDHORSE
SUCKERMOUTH MINNOW
SILVER CHUB
SILVER REDHORSE
SMALLMOUTH BASS
SHORTNOSE GAR
SPOTTAIL SHINER
SPOTFIN SHINER
SPECKLED CHUB
SOUTHERN REDBELLY DACE
COMMON (CENTRAL) STONEROLLER
STRIPED SHINER
TADPOLE MADTOM
TROUT PERCH
WALLEYE
WEED SHINER
WHITE BASS
WHITE CRAPPIE
WESTERN SAND DARTER
WHITE SUCKER
YELLOW BULLHEAD
YELLOW PERCH
Micropterus
Rhinichthys
Lepisosteus
Campostoma
Etheostoma
Notropis
Hiodon
Hybognathus
Cottus
Esox
Hypentelium
Esox
Phoxinus
Lepomis
Dionda
Lepomis
Semotilus
Carpiodes
Oncorhynchus
Etheostoma
Carpiodes
Notropis
Ambloplites
Notropis
Moxostoma
Clinostoma
Notropis
Percina
Ictiobus
Stizostedion
Noturus
Notropis
Percina
Ichthyomyzon
Moxostoma
Phenacobius
Hybopsis
Moxostoma
Micropterus
Lepisosteus
Notropis
Notropis
Hybopsis
Phoxinus
Campostoma
Notropis
Noturus
Percopsis
Stizostedion
Notropis
Morone
Pomoxis
Ammocrypta
Catostomus
Ictalurus
Perca
salmoides
cataractae
osseus
oligolepis
aspirgene
volucellus
tergisus
nuchalis
bairdi
masquinongy
nigricans
lucius
eos
humilis
nubila
gibbosus
margarita
cyprinus
mykiss
caeruleum
carpio
umbratilis
rupestris
rubellus
carinatum
elongatus
blennius
shumardi
bubalus
canadense
flavus
stramineus
phoxocephala
unicuspis
macrolepidotum
mirabilis
storeriana
anisurum
dolomieui
platostomus
hudsonius
spilopterus
aestivalis
erythrogaster
anomalum
chrysocephalus
gyrinus
oiaiscomaycus
vitreum
texanus
chrysops
annularis
Clara
commersoni
natalis
flavescens
64
-------
Wisconsin stream. Lumping together TWIN 1+2 and TWIN 3+4 into only 2 large
TWINSPAN groups (see Figure 34) results in the WIMN 34 sites falling into a
cluster of eastern and northern-interior Wisconsin streams on the one hand
(all circles in Figure 34), and a cluster of western Wisconsin and Minnesota
streams on the other (all squares in Figure 34).
The DCA of the 34 sites based on the binary taxonomic dataset produced
four ordination axes, the first two of which explained 66% of the total
reported variance in species space (Table 17). The position of the 34 sites
with respect to the first three DCA axes is shown in Figure 36, which shows
that there is wide separation in species space among the groups of
taxonomically-similar sites (as classified by TWINSPAN) on the first three DCA
axes. DCA1 separates TWINSPAN groups 1 and 2 from groups 3 and 4, while DCA2
further separates groups 1 and 2 (Figure 36a). Groups 3 and 4 are clearly
separated along DCA3 (Figure 36b). The ordination of the individual
species on the first 2 DCA axes can be used to illustrate species
gradients across the sites (Figure 37). DCA1 indicates a gradient from
TABLE 17. CORRELATIONS BETWEEN DCA AXES AND 2 STATIC BASIN
DESCRIPTOR AND 9 HYDROLOGIC VARIABLES FOR 34 SITES BASED ON
BINARY DATA. (+ = P £ 0.10, * = P < 0.05, ** = P < 0.01)
AREA
DAYAVE
DAYCV
DAYPRED
FLDFREQ
FLOP RED
FLDFREE
BFI
ZERODAY
LOWPRED
LOWFREE
Eigenvalue
Cumulative
variance
explained
DCA1
-0.25
0.17
-0.35*
0.17
-0.25
0.16
-0.06
0.09
-0.19
0.24
0.33+
0.234
0.37
DCA2
0.56**
0.70**
-0.40*
0.51**
-0.27
0.04
0.35*
0.49**
-0.04
-0.30+
-0.36*
0.181
0.66
DCA3
-0.25
0.14
-0.66**
0.62**
-0.28
-0.34*
-0.64**
0.65**
-0.50**
-0.03
-0.05
0.130
0.87
DCA4
-0.26
-0.05
0.08
0.15
-0.15
-0.35*
-0.09
0.13
-0.08
-0.03
-0.06
0.082
1.00
65
-------
smallmouth buffalo (SAB) and shortnose gar (SNG) on one end to finescale dace
(FND), striped shiner (SRS), blackstripe topminnow (BTM), river redhorse (RRH)
and muskellunge (HUE) on the other. Separation along DCA2 is not as dramatic,
though SRS and BTM are positioned at one extreme, and RRH and MUE are at the
other. The "indicator" species from the 3-level TWINSPAN analysis (see Figure
34) are also indicated in Figure 37. Comparing Figures 37 and 36a reveals
that the positions of the species that separate groups 1 and 2 (LSR vs. SDS,
BMS, SRL, and QBS) correspond with the positions of the TWINSPAN groups 1 and
2.
Table 17 gives the correlation between the DCA axis scores and the
original hydrologic variables. DCA1 was negatively correlated with
coefficient of variation and positively correlated with measures of
predictable low flows. This suggests that sites with high DCA1 scores (viz.,
TWINSPAN groups 1 and 2, see Figure 36a) have predictable low flow (LOWFREE)
and low variability (DAYCV), whereas sites with low DCA1 scores (groups 3 and
4) are more variable hydrologically. For DCA2, several hydrologic variables
showed strong correlations. Low scores on DCA2 correlated with hydrologic
variability and unpredictable low flows, and high scores reflected the
importance of flow predictability (DAYPRED) and stable baseflow (BFI). This
axis essentially separated TWINSPAN groups 1 (low DCA2 = variable) and 2 (high
DCA2 = stable) (see Figure 36b). Thus, TWIN 1 and 2 are both relatively
stable hydrologically, but TWIN 2 is the more stable of the two.
Interestingly, DCA2 also correlated strongly with two static measures of
catchment size (Table 17), suggesting that larger streams are more stable.
DCA3, which explained about 20% of the overall variation in the site by
species analysis, also showed some significant correlations with hydrologic
variables. This axis represented a gradient from high variability and low
flood predictability to high baseflow stability and overall flow
predictability, and it primarily separated TWIN 3 (more variable) from TWIN 4
(higher relative baseflow). In summary, these results indicate that there are
major differences among the four TWIN groups in terms of correlations with
several hydrologic variables. However, the exact relationship between
taxonomic composition and independent hydrologic variables is difficult to
sort out, because particular hydrologic variables (DAYCV, DAYPRED, BFI,
FLDFREE, and LOWFREE) are significantly correlated with at least two of the
first three DCA axes. A multivariate technique that takes into account
66
-------
multiple correlations among TWINSPAN groups is a superior approach for
exploring the taxonomic-hydrologic associations.
The TWINSPAN group identifier for each of the 34 WIMN sites was used in
a canonical discriminant analysis (CDA) and discriminant function analysis
(DFA). This technique was highly successful in discriminating among the four
previously-defined TWIN groups in terms of the independent hydrologic
variables. The CDA derived 3 linear combinations of the 9 hydrologic factors
that discriminated among the 4 TWINSPAN-defined groups. The first canonical
variable was highly significantly different from zero (squared multiple
correlation = 0.77, p < 0.0001), and both the second and third canonical
variables had p-values < 0.10 (Table 18). The individual univariate F-tests
for among-group differences on the hydrologic variables showed that BFI,
DAYPRED, and DAYCV were highly significant; FLDFREE, FLDPRED and LOWFREE were
significant; and, LOWPRED was heavily weighted but not statistically
significant (Table 18). The correlations between the canonical variables and
the original hydrologic variables (under CAN1, CAN2 and CANS in Table 18)
showed that the first canonical variable represents a contrast between flow
stability (high positive correlation with BFI and DAYPRED) and variability
plus predictability of extreme flows (high negative correlation with DAYCV,
LOWFREE and FLDFREE). The higher correlations with BFI and DAYPRED indicate
that these factors are the most important discriminators among the two groups
(also cf. the univariate F-tests). The second canonical variable appears to
represent a contrast between sites having high predictability of periods
without floods (FLDFREE) and those having low predictability of periods
without low flows (LOWFREE). The third canonical variable has high positive
correlations with DAYPRED and FLDPRED and high negative correlations with
DAYCV, indicating this variable contrasts predictability with variability.
Examination of the mean canonical scores for the 4 TWINSPAN groups of
sites shows how the groups differ on each of these synthetic variables. When
several canonical variables (and associated original hydrologic variables) are
used to discriminate among several classes of sites, it is useful to refer to
the univariate F-tests to assist in interpretation of among-group differences.
The most important discriminating hydrologic variable is BFI (F = 20.9, CAN1 r
= 0.92), and it can be seen that TWIN 1 and 3 have low mean scores on this
variable, while TWIN 2 and 4 load heavily on BFI (Table 18). Similarly, TWIN
1 and 3 have high DAYCV, in contrast to TWIN 2 and 4, which have low mean
67
-------
TABLE 18. SUMMARY OF CANONICAL DISCRIMINANT ANALYSIS (CDA) FOR 4 TWINSPAN-DEFINED GROUPS. GROUP MEANS
AND STANDARD DEVIATIONS (IN PARENTHESES) ARE GIVEN FOR EACH HYDROLOGIC VARIABLE, AS ARE RESULTS FOR
UNIVARIATE F-TESTS, WHICH INDICATE IF THE FOUR GROUPS DIFFER WITH RESPECT TO HYDROLOGIC VARIABLES.
THE CANONICAL VARIABLES REPRESENT WEIGHTED LINEAR RECOMBINATIONS OF HYDROLOGIC VARIABLES THAT
MAXIMIZE DISCRIMINATION AMONG ECOLOGICALLY-SIMILAR GROUPS. COEFFICIENTS FOR THE 3 CANONICAL
VARIABLES INDICATE CORRELATION BETWEEN THE CANONICAL VARIABLE AND INDIVIDUAL HYDROLOGIC
VARIABLES.THE MEAN SCORE FOR EACH ECOLOGICAL GROUP IS ALSO GIVEN FOR THE 3 CANONICAL VARIATES
TWIN1
N
DAYCV
DAYPRED
FLDFREQ
FLDPRED
FLDFREE
BFI
ZERODAY
LOWPRED
LOWFREE
7
169.
(44.
47.
(16.
0.
(0.
0.
(0.
0.
(0.
0.
(0.
0.
(0.
0.
(0.
0.
(0.
7
0)
7
8)
71
21)
64
08)
32
11)
10
07)
004
007)
58
11)
47
11)
TWIN2
10
100.
(56.
73.
(7.
0.
(0.
0.
(0.
0.
(0.
0.
(0.
0
(0)
0.
(0.
0.
(0.
0
1)
0
3)
66
07)
65
09)
37
13)
37
16)
46
07)
33
07)
TWIN3
10
219
(46
41
(17
0
(0
0
(0
0
(0
0
(0
0
(0
0
(0
0
(0
.1
.5)
.2
.7)
.76
.15)
.66
.10)
.50
.07)
.08
.11)
.03
.06)
.48
.11)
.34
.11)
TWIN4
7
124.4
(52.6)
69.4
(4.0)
0.69
(0.09)
0.53
(0.10)
0.25
0.07
0.45
(0.09)
0
(0)
0.45
(0.09)
0.25
(0.09)
F3,3Q Pr
10.4 0.
13.2 0.
1.1 0.
3.2 0.
8.4 0.
20.9 0.
1.7 0.
2.6 0.
6.7 0.
> F
0001
0001
37
039
0003
0001
19
074
001
CAN1* CAN2* CAN3**
-0.64
0.76
-0.25
-0.43
-0.52
0.92
-0.28
-0.36
-0.57
0.46 -0.49
-0.28 0.48
0.23 -0.24
0.07 0.51
0.62 0.36
-0.15 0.22
0.40 -0.09
-0.42 -0.16
-0.53 0.02
Group Mean
CAN1
CAN2
CAN3
-1.
-1.
-0.
77
56
41
0.
-0.
1.
92
23
04
-1
1
-0
.49
.26
.12
2.59
0.09
-0.91
Squared canonical correlation
(- R2)
t _ _ r\X t
between
CAN1 and hydrologic variables
/>« **n . j i i * ^ j _ t _ i_ i ^ _
- 0.77 (F27,65 - 3.6, p -
e\ t- *\ »»__ ,_ f\ A _-.
p - 0.01)
A A A O
Squared canonical correlation (- R*) between CAN2 and hydrologic variables - 0.37 (F7,24 " 2-°» P " 0.09)
-------
scores for DAYCV. Thus, TWIN 1 and 3 are essentially hydrologically
"variable" sites, while TWIN 2 and 4 are hydrologically "stable". TWIN 1
sites are separated from TWIN 3 on the secbnd canonical variable (for which
these 2 groups have scores of similar magnitude but different sign). TWIN 3
sites are relatively more variable, have longer flood free periods, and
shorter lowflow free periods than TWIN 1 sites. The relatively stable TWIN 2
and 4 sites are distinguished in terms of hydrologic variables comprising CANS
(see Table 18). The univariate F-tests suggest that FLDPRED and overall
hydrologic variability (DAYCV and DAYPRED) are primary discriminants for these
groups. TWIN 2 sites have significantly higher flood predictability and are
less variable generally than the otherwise similar TWIN.4 sites. This
indicates that TWIN 2 sites are the most hydrologically stable of the four
.TWINSPAN groups.
The nonparametric DFA was able to properly classify the 34 sites into
the 4 TWINSPAN groups with 100% accuracy. The parametric DFA (which excluded
ZERODAY) was only slightly less successful, having a 5% classification error
rate.
In addition to the discriminant analysis for the TWINSPAN groups, we
used Ward's clustering method to define either 2 or 3 groups of sites from the
Jaccard similarity matrix, which essentially expresses the correlation among
the 34 sites based on shared species presence (see Methods). In neither the
2-group or the 3-group case was a discriminant function produced from the
hydrologic variables that could distinguish among the pre-defined,
taxonomically-similar groups. The multivariate test of significance (Wilk's
X) had p > 0.10, while the misclassification rate for discriminant function
approached 50% (i.e., only 1/2 of the taxonomically-similar groups could be
properly discriminated using the hydrologic variables).
Fish Functional Relationships to Hydrologrie Factors
When ecological similarity among sites was defined in terms of functional
attributes using Ward's method, the CDA and DFA were able to make significant
discriminations for both the 2-group and 3-group cases. The hierarchical
dendrogram produced by Ward's method is shown in Figure 38. The results of
the DFA and CDA are given for the 2-group and 3-group cases in Table 19 and
Table 20, respectively. The geographic locations of the 34 WIMN sites coded
according to the 3-group case are shown in Figure 30. For the 2-group case,
69
-------
TABLE 19. SUMMARY OF NONPARAMETRIC CANONICAL DISCRIMINANT ANALYSIS (CDA)
FOR 2-CLUSTER CASE. GROUP MEANS AND STANDARD DEVIATIONS (IN
PARENTHESES) ARE GIVEN FOR EACH HYDROLOGIC VARIABLE, AS ARE
RESULTS FOR UNIVARIATE F-TESTS, WHICH INDICATE IF THE 2 GROUPS
DIFFER WITH RESPECT TO HYDROLOGIC VARIABLES. THE CANONICAL VARIABLE
REPRESENTS A WEIGHTED LINEAR RECOMBINATION OF HYDROLOGIC VARIABLES
THAT MAXIMIZE DISCRIMINATION AMONG ECOLOGICALLY-SIMILAR GROUPS.
COEFFICIENTS FOR CAN1 INDICATE CORRELATION BETWEEN THE CANONICAL
VARIABLE AND INDIVIDUAL HYDROLOGIC VARIABLES. THE MEAN SCORE
FOR EACH ECOLOGICAL GROUP IS ALSO GIVEN FOR THE CANONICAL VARIABLE
Group 1
(Variable)
N
DAYCV
DAYPRED
FLDFREQ
FLDPRED
FLDFREE
BFI
ZERODAY
LOWPRED
LOWFREE
Group Mean
CAN1
16
173
(73
46
(19
0
(0
0
(0
0
(0
0
(0
0
(0
0
(0
0
(0
-1
.5
.7)
.5
.9)
.75
.17)
.63
.10)
.37
.13)
.16
.19)
.02
.05)
.49
.14)
.35
.14)
.391
Group 2
(Stable)
18
137
(60
67
(11
0
(0
0
(0
0
(0
0
(0
0
(0
0
(0
0
(0
1
.2
.7)
.7
.1)
.67
.09)
.62
.10)
.37
.14)
.32
.17)
.0001
.0003)
.49
.08)
.34
.10)
.237
Fl,32
2
15
2
0
0
6
2
0
0
.48
.08
.68
.15
.01
.78
.91
.01
.03
Pr > F
0
0
0
0
0
0
0
0
0
.13
.0005
.11
.70
.91
.014
.10
.94
.87
CAN1*
-0
0
-0
-0
0
0
-0
-0
-0
.333
.704
.346
.087
.026
.520
.360
.018
.037
Squared canonical correlation (= R^) between CAN1 and hydrologic variables
0.646 (Fg,24 = 4.87, p = 0.0009)
the 34 sites were separated into groups of 16 and 18 (cf. all circles to
triangles in Figure 30; see also Figure 38). The CDA derived a linear
combination of the 9 hydrologic factors that discriminated among the groups,
and this canonical variable was highly significantly different from zero
(squared multiple correlation different from zero (squared multiple
correlation = 0.648, p = 0.0009). The individual univariate F-tests for
among-group differences on the hydrologic variables showed that DAYPRED and
BFI were highly significant, while ZERODAY, DAYCV and FLDFREQ were heavily
70
-------
TABLE 20. SUMMARY OF CANONICAL DISCRIMINANT ANALYSIS (CDA) FOR 3-CLUSTER CASE. GROUP MEANS
AND STANDARD DEVIATIONS (IN PARENTHESES) ARE GIVEN FOR EACH HYDROLOGIC VARIABLE, AS ARE RESULTS
FOR UNIVARIATE F-TESTS, WHICH INDICATE IF THE THREE GROUPS DIFFER WITH RESPECT TO HYDROLOGIC
VARIBLES. THE CANONICAL VARIABLES REPRESENT WEIGHTED LINEAR RECOMBINATIONS OF HYDROLOGIC
VARIABLES THAT MAXIMIZE DISCRIMINATION AMONG ECOLOGICALLY-SIMILAR GROUPS. COEFFICIENTS FOR
CAN1 AND CAN2 INDICATE CORRELATION BETWEEN THE CANONICAL VARIABLE AND INDIVIDUAL HYDROLOGIC
VARIABLES. THE MEAN SCORE FOR EACH ECOLOGICAL GROUP IS ALSO GIVEN FOR THE TWO CANONICAL VARIATES
N
DAYCV
DAYPRED
FLDFREQ
FLDPRED
FLDFREE
BFI
ZERODAY
LOWPRED
LOWFREE
Group Mean
CAN1
CAN2
Group 1
(Unstable)
16
173.4
(73.7)
46.5
(19.9)
0.75
(0.17)
0.63
(0.10)
0.37
(0.13)
0.16
(0.19)
0.02
(0.05)
0.49
(0.14)
0.35
(0.14)
-1.37
-0.001
Group 2
(Moderately
Stable)
10
161.0
(52.9)
63.8
(12.0)
0.67
(0.10)
0.62
(0.12)
0.35
(0.18)
0.30
(0.18)
0.0001
(0.0004)
0.50
(0.08)
0.35
(0.12)
1.22
0.80
Group 3
(Very
Stable) ^2,31
8
107.5 2.82
(59.5)
72.5 8.28
(8.2)
0.67 1.30
(0.07)
0.62 0.08
(0.07)
0.40 0.25
(0.09)
0.35 3.52
(0.17)
0.0 1.41
(0.0)
0.48 0.08
(0.09)
0.33 0.05
(0.07)
1.22
-1.00
Pr > F CAN1* CAN2*
0.075 -0.334 0.506
0.001 0.704 -0.295
0.29 -0.346 -0.015
0.92 -0.086 -0.031
0.78 0.026 -0.219
0.042 0.520 -0.179
0.26 -0.359 -0.002
0.92 -0.018 0.125
0.96 -0.037 0.073
Squared canonical correlation (= R2) between CAN1 and hydrologic variables
Squared canonical correlation (= R2) between CAN2 and hydrologic variables
0.646 (Fie,46 = 2.65, p = 0.004)
0.319 (F8,24 = 1-40, p = 0.246)
-------
ted but not statistically significant (Table 19). The correlations between
the canonical variable and the original hydrologic variables (under CAN1 in
Table 19) show that the canonical variable represents a contrast between flow
stability (high positive correlation with DAYPRED and BFI) and extreme flows
and variability (high negative correlation with DAYCV, ZERODAY and FLDFREQ).
The higher correlations with DAYPRED and BFI indicate that these factors are
the most important discriminators among the two groups (also cf. the
univariate F-tests). The mean scores for the two groups of sites on the
canonical variable (CAN1) are of similar magnitude but opposite sign. The
negative coefficient for group 1 (-1.39) indicates that these 16 sites are
inversely correlated with DAYPRED and BFI and positively correlated with
DAYCV, ZERODAY and FLDFREQ. Group 1 sites can thus be characterized as
hydrologically "variable" sites. By contrast, the group two mean score
(+1.24) on CAN1 indicates that these 18 sites can be characterized as
hydrologically "stable". The nonparametric DFA was able to properly classify
the 34 sites into the two groups with 100% accuracy. The parametric DFA
(which excluded ZERODAY) was less successful, having a 12% classification
error rate. However, the parametric DFA also identified DAYPRED and BFI as
the two most important environmental factors in discriminating among the two
ecological groups. As in the non-parametric analysis, DAYCV and FLDFREQ were
less heavily weighted but substantially negatively correlated with CAN1. The
squared multiple correlation between CAN1 and the hydrological variables was
0.633 (p = 0.0005).
The location of individual sites (coded by ecological group) as a
function of individual scores on the canonical variable is shown in Figure 39.
There is some overlap in this 1-dimensional representation of a 9-dimensional
hydrologic space. The sites can also be viewed in the bivariate space of the
two major hydrologic factors (DAYPRED and BFI) that contribute to CAN1 (Figure
40). The group centroids are clearly separated, although there is some
overlap in these two dimensions among the individual sites. Additions of
additional axes (i.e., variables with high weight on CAN1) would further
separate the clusters from one another.
For the 3-group case, sites were separated into clusters of 16, 10 and
8, with the last two groups representing a splitting of the 18 member "stable"
group from the 2-cluster case (see Figures 30, 38). The CDA derived linear
combinations of the 9 hydrologic factors that discriminated among the three
72
-------
groups (Table 20). The first canonical variable was highly significantly
different from zero (squared multiple correlation = 0.646, p = 0.004), while
the second was not (correlation = 0.319, p = 0.246). This result indicates
that CAN2 explained little additional separation among the groups not
described by CAN1, probably because the same hydrologic variables were most
heavily weighted on both canonical variables. Nonetheless, the individual
sites in the three groups can be reasonably well distinguished when plotted in
the space generated by CAN1 and CAN2 (Figure 41). The individual univariate
F-tests for among-group differences on the hydrologic variables showed that
DAYPRED and BFI were significant (p < 0.05) and DAYCV approached significance
(p = 0.075) (Table 20). The correlations between the canonical variables and
the original hydrologic variables on CAN1 were identical (by definition) to
those in the 2-group case. For the second canonical variable (CAN2), there
was a high positive correlation with DAYCV and a relatively large negative
correlation with DAYPRED. CAN1 thus represents a contrast between flow
stability (high positive DAYPRED and BFI) and variability and flow extremes
(negative DAYCV, ZERODAY and FLDFREQ). CAN2 represents a contrast between
flow variability (high positive DAYCV) and flow predictability (negative
DAYPRED) (see Figure 41). Group 1 sites are identical to those in the 2-case
solution, having a high negative mean score on CAN1 (see Table 20). They are
characterized as hydrolbgically "variable". Group 2 and Group 3 sites both
have a high positive mean score on CAN1, indicative of hydrologic stability,
but they are discriminated on CAN2. Group 2 sites score positively on CAN2
(indicating high relative variability), but group 3 sites score negatively on
CAN2 (indicating low variability). Thus, the 10 sites in group 2 can be
characterized as hydrologically "moderately stable", whereas the 8 sites in
group 3 are hydrologically "very stable".
The nonparametric discriminant function analysis (DFA) was able to use
the hydrologic data to properly classify the 34 sites into their respective
ecological groups with 100% accuracy. The parametric DFA (which excluded
ZERODAY) was only slightly less successful, having a 2% classification error
rate. The parametric DFA also identified DAYPRED, BFI, and DAYCV as the most
important hydrological factors in discriminating among the 3 ecological
groups. The squared multiple correlation for CAN1 was 0.635 (p = 0.0037), but
the value for CAN2 (0.248) was not significant (p = 0.35).
73
-------
Possible Confounding Factors Influencing Results
The potential confounding of the hydrological-ecological patterns by
sampling intensity (Figure 31) and catchment area (Figure 33) was assessed by
examining whether the defined groups of sites differed with respect to these
factors. Figure 42 shows that both groups of ecologically-similar sites
express a strong but very similar relationship between number of species
collected and sampling intensity at each site. We can conclude that the
observed differences among groups with respect to functional organization is
not an artifact of differential sampling intensity among the two types of
streams. Systematic relationships between cluster affiliation and the
measures of stream size can be viewed in several ways. Figure 43a shows that
there are no consistent differences between groups in terms of the species
collected and upstream catchment area. However, when area is defined in terms
of average discharge (Figure 43b), the variable streams show a species-area
relation, while the stable group does not, due mostly to larger streams having
fewer species collected. But when stream size is defined as suggested by
Hughes and Omernik (1983) as mean annual runoff (mm/yr), both groups of sites
show similar species-area relations (Figure 43c). These alternative
descriptions support the conclusion that the ecologically-defined groups are
not "predictable" in terms of knowledge of catchment area alone. For the 3-
group case, neither the moderately stable nor the very stable sites showed a
strong relationship between species richness and either catchment area (Figure
44a) or average discharge (Figure 44b). However, there was a tendency for
sampling intensity of streams to vary as a function of size (Figure 45). With
the exception of one point, small streams typically were sampled more
frequently than large streams, leading to a higher apparent richness in the
small streams than might otherwise be expected.
Functional Organization of Ecologically-similar Assemblages
Several interesting and consistent patterns emerged when the functional
composition of fish assemblages were contrasted with corresponding hydrologic
characteristics for the 34 WIMN sites. These differences are summarized in a
general fashion in Figure 46. To investigate the relative contributions of
the various functional attributes to defining the ecologically-similar groups,
we compared means and standard errors of the individual functional attributes
for each group using oneway analysis of variance. The parametric and non-
74
-------
parametric anovas produced very similar results; therefore, only the
parametric results are reported here. This univariate approach shows major
differences among groups and provides the basis for the generalizations
illustrated in Figure 46. However, many of the functional variables were
highly cross-correlated (Table 21), a fact that supports the use of the
multivariate canonical discriminant analysis described previously. The
taxonomic composition of the assemblages from the 3 hydrological stream types
also varied (Table 22, Figure 46). Differences in taxonomic composition help
explain the observed functional patterns, since the functional attributes were
generated from species-specific information initially.
Morphological Variables
When two groups of functionally-similar assemblages were created,
several interesting observations were made. First, no substantial among-group
differences were apparent for the first morphological variable (swimming
factor), but fishes from variable habitats had significantly lower values (p =
0.03) on the second morphological variable, the shape factor (Figure 47a).
High scores on the shape factor reflect a relatively elongated body shape that
presumably enhances hydrodynamic profile and reduces energetic costs of
position-maintenance in fast-flowing water (Webb and Weihs 1986) .
For the 3-group case, again no differences on the swimming factor were
found, while some separation on the shape factor was observed (Figure 47b).
Highly significant differences (p < 0.0001) in the shape factor were observed
among groups, with assemblages from the very stable streams showing greatest
body elongation compared to assemblages from the stable and variable streams,
which did not differ from one another (Figure 47b).
Trophic Variables
For the 2-cluster case, several among-group differences were observed
for trophic categories (Figure 48a). Assemblages in hydrologically variable
streams had a significantly greater abundance of omnivores (p = 0.0005) and a
significantly lower abundance of benthic invertivores (p = 0.0001) than the
stable streams. Generalist invertivores were also more abundant (p = 0.07) in
variable streams, while parasitic species were less abundant (p = 0.06).
Planktivores, which were recorded only from the variable sites, were
significantly more abundant there than in the stable sites (p = 0.014). (This
75
-------
TABLE 21. PEARSON CORRELATION MATRIX FOR 25 FUNCTIONAL ATTRIBUTE VARIABLES FOR 106 SPECIES GROUPED
ACCORDING TO 5 MAJOR CATEGORIES (SEE TABLE 11). R-VALUES WITH P < 0.05 USING A BONFERRONI TEST AND
ARE INDICATED BY BOLDFACE AND "*". FOR P < 0.10, "a" IS USED
SWIM
MORPHOLOGY
SWIM 1.000
TROPHIC
HERB 0.109
OMNIVOR -0.229
GENERAL 0.014
SURFACE 0.733*
BENTHIC -0.079
PISCIV
PLANK
-0.095
0.123
SHAPE
0.045
-0.357
-0.569
-0.062
0.454
0.316
-0.176
HERB
1.000
-0.357
0.000
0.006
0.187
-0.334
-0.071
OMNIVOR
1.000
0.466
-0.412
-0.612*
-0.410
0.006
GENERAL
1.000
-0.194
-0.593a
-0.523
-0.040
SURFACE
1.000
0.048
0.072
-0.080
BENTHIC
1.000
-0.029
-0.258
PISCIV
1.000
0.243
PLANK
1.000
PARASIT
WATER VELOCITY
FAST
MODERATE
SLOW
GENERAL
SUBSTRATE
RUBBLE
SAND
SILT
0.090
-0.051
0.103
-0.383
-0.074
0.368
-0.014
0.269
0.599a
-0.571
0.290
0.570
-0.107
-0.439
0.221
0.304
-0.285
0.002
0.305
0.379
-0.290
-0.310
-0.423
0.392
0.013
-0.422
0.134
0.553
-0.118
-0.568
0.418
-0.058
-0.411
0.454
0.393
0.112
0.014
0.061
-0.410
-0.057
0.229
-0.109
0.576
0.587a
-0.679*
0.221
0.686*
-0.319
-0.631
-0.298
0.119
0.051
0.023
-0.067
-0.454
-0.059
-0.270
-0.187
0.236
-0.008
-0.202
-0.140
0.198
-0
0
-0
0
0
-0
-0
.033
.418
.248
.020
.319
.318
.309
STREAM SIZE
SMALL
LARGE
GENERAL
TOLERANCE
TOLERANT
MODERATE
INTOL
-0.380
0.235
0.196
0.002
0.532
-0.286
-0.048
0.168
-0.210
-0.604*
-0.139
0.354
-0.361
0.130
0.332
-0.272
0.239
-0.081
0.815*
-0.691*
-0.082
0.832*
-0.007
-0.506
0.304
-0.369
0.204
0.592a
0.169
-0.450
-0.376
0.317
0.079
-0.216
0.497
-0.150
-0.321
0.659
-0.455
-0.734*
-0.370
0.714*
-0.399
0.135
0.184
-0.337
-0.156
0.302
-0.292
-0.077
0.384
0.123
0.215
-0.191
-0
0
-0
-0
-0
0
.224
.257
.171
.560
.133
.362
continued
-------
TABLE 21 (continued)
FAST MODERATE SLOW GENERAL RUBBLE SAND SILT GENERAL
WATER VELOCITY
FAST 1 . 000
MODERATE 0.357
SLOW -0.726*
SUBSTRATE
RUBBLE
SAND
SILT
0.789*
-0.027
-0.641*
1.000
-0.846*
0.754*
-0.088
-0.741*
1.000
-0.941*
0.128
0.839*
0
-0
-0
.446
.212
.348
1.
-0.
-0.
000
173
861*
1.00.0
0.189
1
.000
STREAM SIZE
SMALL
LARGE
GENERAL
TOLERANCE
TOLERANT
MODERATE
INTOL
-0.093
0.298
-0.177
-0.437
-0.290
0.499
SMALL
-0.254
0.479
-0.317
-0.560
-0.298
0.451
LARGE
0.167
-0.432
0.316
0.585a
0.448
-0.627
GENERAL
0
0
-0
-0
-0
0
.104
.027
.151
.170
.467
.447
LAKE
-0.
0.
-0.
-0.
-0.
0.
204
441
276
614*
411
660
TOLERANT
0.071
-0.299
0.384
0.236
0.414
-0.525
MODERATE
0
-0
0
0
0
-0
.448
.607
.199
.665*
.399
.670*
-0.190
0.149
-0.011
0.187
-0.039
-0.036
INTOL
STREAM SIZE
SMALL
LARGE
GENERAL
TOLERANCE
TOLERANT
MODERATE
INTOL
1.000
-0.664*
-0.349
O.S90a
-0.134
-0.307
1.000
-0.448
-0.605*
-0.171
0.564
1.000
0.095
0.412
-0.385
0
-0
0
.025
.101
.006
1.
0.
-0.
000
129
675*
1.000
-0.765*
1
.000
I-
r-
-------
TABLE 22. RELATIVE PROPORTIONS OF 41 DOMINANT FISH SPECIES FOR 34 WIMN SITES FOR E
OF THREE HYDROLOGICAL STREAM TYPES. SPECIES ARE RANKED ACCORDING TO THE TOTAL
NUMBER OF SITES AT WHICH EACH SPECIES OCCURS. ONLY SPECIES HAVING A RELATIVE
PROPORTION >0.500 IN AT LEAST ONE STREAM TYPE ARE INCLUDED. SPECIES SHOWN
IN FIGURE 37 ARE INDICATED BY A "*"
Abbrev.
BDD
BKB
BLC
BLG
BLM
BNT
BMS
BND
BSD
CAP
CCF
CHL
CRC
CSH
EMS
FHM
FTD
GLR
GOS
GSF
HHC
JND
LGP
LND
LSR
NHS
NOP
PMK
RKB
ROS
SKM
SLR
SPS
SCT
SHR
SDS
SMB
WAE
WTS
YEB
YEP
Common Name
BANDED DARTER
BLACK BULLHEAD
BLACK CRAPPIE
BLUEGILL
BLUNTNOSE MINNOW
BROWN TROUT
BIGMOUTH SHINER
BLACKNOSE DACE
BLACKS IDE DARTER
CARP
CHANNEL CATFISH
CHESTNUT LAMPREY
CREEK CHUB
COMMON SHINER
EMERALD SHINER
FATHEAD MINNOW
FANTAIL DARTER
GOLDEN REDHORSE
GOLDEN SHINER
GREEN SUNFISH
HORNYHEAD CHUB
JOHNNY DARTER
LOGPERCH
LONGNOSE DACE
LARGESCALE STONEROLLER
NORTHERN HOGSUCKER
NORTHERN PIKE
PUMPKINSEED
ROCK BASS
ROSYFACE SHINER
SUCKERMOUTH MINNOW
SILVER REDHORSE
SPOTFIN SHINER
STONECAT
SHORTHEAD REDHORSE
SAND SHINER
SMALLMOUTH BASS
WALLEYE
WHITE SUCKER
YELLOW BULLHEAD
YELLOW PERCH
Total
Sites
13
18
12
17
30
13
20
18
23
24
10
5
33
34
14
24
17
19
8
18
26
30
10
16
10
16
24
11
15
14
8
18
23
22
21
21
20
18
33
9
14
Variable
.250
.750
.375
.562
.938
.375
.562
.625
.625
.812
.250
.062
1.000
1.000
.500
.938
.438
.438
.500
.625
.688
.938
.188
.250
.125
.188
.812
.500
.375
.188
.125
.438
.688
-562
.312
.688
.312
.438
.938
.500
.625
Moderately
Stable
.600
.400
.200
.500
.900
.400
.700
.400
.500
.700
.500
.000
1.000
1.000
.500
.700
.600
.600
.000
.700
.700
.900
.100
.600
.100
.500
.500
.100
.500
.700
.600
.500
.900
.900
.800
.800
.800
.500
1.000
.100
.100
Very
Stable
.375
.250
.500
.375
.750
.375
.500
.500
1.000
.500
.125
.500
.875
1.000
.125
.250
.500
.750
.000
.125
1.000
.750
.750
.750
.875
1.000
.750
.250
.500
.500
.000
.750
.375
.500
1.000
.250
.875
.750
1.000
.000
.375
*
*
*
*
*
*
*
*
*
*
78
-------
result held when a non-parametric test was employed.) These are consistent
patterns that suggest that generalist trophic strategies are associated with
hydrologic variability in the study streams.
In the 3-cluster case, further separation of the stable and the very
stable sites occurred on these trophic attributes (Figure 48b). The very
stable sites had more benthic invertivores (p < 0.0001), fewer generalist
invertivores (p = 0.0002) and more parasitic fish (p = 0.0004) than the stable
or variable sites. Very stable sites also had the fewest omnivores, though
this difference was not significantly different from omnivore proportion at
moderately stable sites.
Substrate Preference
For the 2-group case, assemblages in variable streams had relatively few
species associated with rubble (p < 0.0001) and more species typically
associated with silt (p < 0.0001). There was a tendency for substrate
generalists to be more highly represented in variable streams, but this
difference was not significant (p = 0.09) (Figure 49a).
When the stable streams were further divided in the 3-group case (Figure
49b), we observed that significant differences in species associated with
rubble and silt still occurred among the three stream types (p < 0.0001 for
both), but the very stable sites were not significantly different from the
moderately stable sites (p > 0.05). However, the proportion of sand-
associated species at very stable sites was significantly lower (p = 0.004)
than at the moderately stable or variable sites, which did not differ from one
another (Figure 49b).
Stream Size Association
For this macro-habitat attribute, the major contrast between the groups
in the 2-cluster case was that variable sites had significantly more small
stream fishes (p - 0.05) and lake fishes (p = 0.05), while stable streams had
more fishes found in medium-large systems (p = 0.0006) (Figure 50a) . Variable
streams also tended to have more generalist, small-large system fishes, though
this difference was not statistically significant (p = 0.16).
For the 3-cluster case, the very stable streams showed a significantly
greater proportion of fishes typical of medium-large systems (p = 0.0002) and
a lower proportion of generalist fishes typical of small-large systems (p =
79
-------
0.004) (Figure 50b). Lake species were absent from moderately stable streams,
which had significantly fewer (p = 0.017) of these species than very stable or
variable streams.
Biotope (Water Movement) Association
The dominant pattern observed for the 2-cluster case for this functional
variable was the high representation of slow water species (p = 0.0001) at
variable sites (Figure 51a). Stable sites had more fish species associated
with fast (p = 0.0001) and intermediate (p = 0.004) velocity macrohabitats.
There was a tendency (p = 0.08) for stable habitats to have more generalist
species.
For the 3-cluster case, stream assemblages from very stable streams had
proportionally more medium-velocity species (p = 0.0004) than did either
moderately stable or variable streams (Figure 51b). Very stable streams also
had significantly fewer slow-velocity fishes than moderately stable streams,
which themselves were significantly different from the variable streams. No
significant differences among moderately stable and very stable streams were
observed for the fast or general categories (Figure 51b).
Environmental Tolerance
Fish assemblages in variable systems had much higher proportions of
tolerant (p < 0.0001) and moderately tolerant (p = 0.02) species relative to
stable systems, which had a high proportion of intolerant species (p < 0.0001)
(Figure 52a).
In the 3-cluster case, fish assemblages from the three groups fell along
a tolerant-intolerant gradient, with very stable sites having significantly
fewer tolerant species (p < 0.0001) and more intolerant species (p = 0.0001)
than variable sites. Moderately stable sites were intermediate and
significantly distinct from the variable and the very stable sites (Figure
52b) .
80
-------
DISCUSSION
The results of this analysis clearly show that differences in fish
assemblage structure are strongly associated with specific hydrologic factors
that vary among catchments across geographic scales. We interpret this
pattern to mean that hydrologic regime constrains assemblage structure,
defined in either functional or taxonomic terms. Thus, given knowledge of
critical hydrologic factors for a stream in the geographic area covered by
this study, certain general features of the fish community should be
predictable with some confidence.
Several previous studies have related fish community structure to
habitat variables (e.g., Gorman and Karr 1978, Schlosser 1985, 1987,
Angermeier 1987, Hawkes et al. 1986, Bisson et al. 1988, Rahel and Hubert
1991, Bozek and Hubert 1992, Nelson et al. 1992, Pearsons et al. 1992), and
the influence of hydrologic variation (particularly floods) on fish community
structure has been amply documented (e.g., John 1963, Harrell 1978, Meffe
1984, Matthews 1986, Coon 1987, Schlosser 1987, Bain et al. 1988, Jowett and
Duncan 1990, Fausch and Bramblett 1991). This study represents one of the
only attempts to identify cross-stream patterns in fish assemblage structure
based solely in terms of long-term hydrologic averages (also see Horwitz
1978), and it is the only research of which we are aware that uses multiple
measures of hydrologic variability. Further, while others have developed
functional descriptions for fish species to assess assemblage response to
catchment degradation (Karr et al. 1986), this is the first study we are aware
of that describes fish assemblage structure in functional terms designed to be
sensitive to hydrologic variation.
A hierarchical classification (using TWINSPAN) of the fish assemblage
data defined in taxonomic terms identified two major divisions of sites (TWIN
1+2 and TWIN 3 + 4) . These two divisions could be further separated into
four groups with relatively distinct geographic separation (Figure 35).
Canonical discriminant analysis (CDA) showed that these four groups had
significant hydrologic correlates, but the interesting result was that neither
the two most hydrologically "stable" groups (TWIN 2 and 4) nor the two most
hydrologically "variable" groups (TWIN 1 and 3) occurred in the same major
division (cf. Table 18 and Figure 34). This result suggests the possibility
that, within a taxonomically-similar major division (e.g., TWIN 1 + 2 or TWIN
81
-------
3 + 4), taxa are further segregated among available sites depending on
catchment-scale hydrologic conditions. This possibility assumes that there is
no confounding of potential range limits of fish species and the geographic
distribution of sites having specified hydrologic characteristics. In other
words, if taxonomically-similar fish assemblages in a major division (e.g.,
TWIN 1 + 2 sites) have ranges that potentially include all 17 of the sites in
that division, then the null expectation would be a random distribution of
these species among all 17 of the sites. To assess the range of fish species
distributions in the region covered by this study, one of us (NLP) identified
19 species that were present in >50% of the sites in at least one of the four
TWINSPAN groups. The selected species included all the "indicator" species
shown in Figure 34. Without prior knowledge of the TWINSPAN or hydrologic
affiliation, the other of us (JDA) examined geographic range maps for these
species (using Lee et al. 1980) and attempted to assign fish species to the
four TWINSPAN groups. For only 7/19 species could geographic ranges be
reasonably related to the four TWINSPAN groups (BMS, COS, SRL, LKC, QBS, LSR,
and YEP). The best that could be said for each of these seven species was
that it was unlikely to occur in one particular TWINSPAN group. Of these
seven, only three (BMS, QBS, and LSR) were correctly identified as being
absent from one particular TWINSPAN group. The fact that these three species
are "indicator" species separating TWIN 1+2 from TWIN 3+4 (Figure 34)
suggests some zoogeographic constraint among the first two major TWINSPAN
divisions. However, within each major division (i.e., four TWINSPAN groups),
it appears that all species can potentially occupy all 17 sites. That they do
not suggests that the recorded distribution reflects some process of site
selection (or site exclusion) that is correlated with site hydrologic factors.
When viewed in functional terms, the relationship between fish
assemblage structure and hydrologic regime across the 34 WIMN sites is
remarkably clear (Figure 30), and assemblages can be described as coming from
hydrologically "variable", "moderately stable", and "very stable" sites. The
geographic distribution of the functionally-defined assemblages complements
the geographic distribution of the taxonomically-defined assemblages, thus
providing further evidence for hydrologic constraints on assemblage structure.
This can be seen by comparing, on the one hand, the distribution of the 16
hydrologically "variable" sites in the functional analysis (triangles in
Figure 30) to the 17 "variable" sites in the taxonomic analysis (TWIN 1 + 3 in
82
-------
TABLE 23. CROSS-CLASSIFICAION OF HYDROLOGIC DESIGNATIONS
FOR THE 34 WIMN SITES DEFINED IN TAXONOMIC VS.
FUNCTIONAL TERMS
Functional Desicmation
Taxonomic
Designation
"Variable" sites
TWIN 1 (n=7)
TWIN 3 (n=10)
"Stable" sites
TWIN 2 (n=10)
TWIN 4 (n=7)
Moderately
Variable Stable
(n=16) (n=10)
6 1
6 4
1 1
3 4
Very
Stable
(n=8)
0
0
8
0
Figure 35), and, on the other hand, the distribution of the 18 "stable"
functional sites (combined moderately stable and very stable sites in Figure
30) to the 17 "stable" taxonomic sites (TWIN 2 + 4 in Figure 35). The results
of this comparison are summarized in Table 23, which shows that, while the
hydrologic designation for sites defined in functional terms do not precisely
correspond with the hydrologic designation for the taxonomically-defined
sites, the 2-group functional analysis largely identifies the four
taxonomically-defined assemblages coming from "variable" and "stable" streams.
For the variable functional sites, 75% (12/16) of the taxonomic sites were
similarly classified, and for the stable (moderately stable + very stable)
functional sites, 72% (13/18) were similarly classified. TWIN 1 and TWIN 2
groups were the most similar in hydrologic designation to the corresponding
functionally-defined sites (Table 23). This functional approach thus suggests
a powerful technique that can be used to generalize across zoogeographic
boundaries.
The advantage of characterizing fish assemblages in terms of functional
composition is that this approach provides ecological insight that are
difficult to acquire by taxonomic approaches alone. The functional approach,
83
-------
combined with species presence/absence data, allows coarse-grain patterns to
be detected when the corresponding hydrologic data is also coarse-grained
(e.g., long-term averages of hydrologic variables). More detailed
hydrological information would probably be needed to detect fine-grained
ecological patterns such as inter-annual variation in population abundances or
size structure.
The contrasts in functional characteristics for fish assemblages in
hydrologically variable vs. stable streams (Figures 17-22) were largely in
keeping with theoretical expectations. In many instances, fish assemblages
showed differences for functional attributes that varied consistently with
average hydrologic environment. This held for both the 2-cluster case
(variable vs. stable contrast) and the 3-cluster case (variable to moderately
stable to very stable gradient). The most striking pattern was that
assemblages from hydrologically variable streams had generalized feeding
strategies, were associated with silt and general substrates, were
characterized by slow-velocity species with headwater affinities, and were
broadly tolerant. By contrast, stable and very stable streams had less
tolerant assemblages, more specialized trophic guilds and species associated
with fast-flowing and/or permanent streams (see Figure 46). Many of the
differences among the stream groups for individual functional attributes might
be expected given the correlation structure of the functional data (see Table
21). However, the ecological significance of these results is that the
patterns, taken as a whole, are consistent with the general theoretical
expectation that variable systems are dominated more by generalists (trophic,
habitat) and tolerant species than stable systems. These broad differences in
functional organization reflect differences in species composition across the
sites (see Figure 46, Table 22). Variable streams are dominated by generally
tolerant species of ictalurids (brown and yellow bullhead), percids (yellow
perch), cyprinids (golden shiner), and centrarchids (pumpkinseed) . By
contrast, stable sites are characterized by more elongate, less tolerant
catostomids (northern hogsucker, shorthead redhorse), cyprinids (rosyface
shiner, longnose dace) and a centrarchid (smallmouth bass).
The finding that assemblages from hydrologically variable streams
contain more small-stream and wide-ranging species (Figure 50) offers an
intriguing possible interpretation. To the extent that flow variability and
low baseflow (see scores on the first canonical variables in Tables V and VI)
84
-------
typify the variable streams, one might expect these sites to be comprised of
the "colonizing" headwater species that Schlosser (1987) identified as adapted
to variable headwater environments in temperate warmwater streams of the
central United States. Interestingly, these functional "headwater" sites are
not comprised only of the smaller streams in the WIMN dataset (e.g., compare
multiple measures of area in variable vs. stable sites in Figure 43),
suggesting that stream size may not be the sole correlate of species' adaptive
strategies employed by species in hydrologically-variable sites. Additional
life history information would be desirable to explore this pattern.
Some interesting possible environmental correlates of variable vs.
stable streams can be inferred. In an extensive survey of New Zealand
streams, Jowett and Duncan (1990) found that several physical characteristics
were related to flow variability. Compared to stable streams, systems with
high flow variability had much lower velocities (under average discharge
conditions), which probably resulted in higher average water temperatures.
Variable-flow streams were more complex morphologically, having well-developed
pool-riffle structures where pools were deeper and riffles shallower than the
stable stream systems. If these patterns hold for the WIMN streams, then the
high relative proportion of generalist and tolerant species in the variable
streams would be explicable. Coon (1987) studied fish communities at two
locations on the South Branch Root River in Minnesota, and found that an
upstream site was flashier than the downstream site, owing to greater relative
impermeability of the basin in the upstream reaches. The more variable site
experienced greater flow fluctuations including low summer baseflow
conditions, more rapid response to storm runoff, and much lower winter
temperatures that allowed extensive ice cover to develop. These observations
suggest a correlation between flow variability and seasonal disturbance
intensity that may hold for the WIMN sites considered in this paper.
Specifically, the very stable hydrologic sites in this study (Table 16) may be
characterized by high groundwater inflow and relatively sparse ice cover in
winter and minimal oxygen stress in summer. By contrast, the variable
hydrologic sites may represent relatively harsh sites in terms of winter ice,
low summer water velocities, and associated seasonal thermal/oxygen stress.
The absence of a strong species-area relationship suggests a potential
limitation of this study. Species-area relationships have been documented in
many systems, but the strength of the relationship may depend on the units
85
-------
used to define area. For example, Angermeier and Schlosser (1989) found
habitat volume to be a better predictor of number of species than habitat
area, while Watters (1992) found a tight species-area relationship for fishes
in 37 Ohio River drainages based on catchment area alone. A contributing
factor to the absence of a species-area relation in the present study appears
to be the differential sampling intensity of streams varying in size (Figure
45). Generally, small streams were sampled more frequently than large
streams, leading to a tendency toward higher apparent richness in the small
streams (cf. Figure 44). Given the strong relationship between sampling
intensity, species collected, and catchment area in this study, we suspect
that a species-area relation would exist for the WIMN sites had all sites been
sampled with equal effort. However, the fact that the ecologically-similar
groups of assemblages identified in this study showed similar species-area and
species-sampling intensity relationships provides support for the argument
that the hydrological-ecological patterns documented here are not simply an
artifact of the sampling limitations. A further point worth emphasizing is
that species-area relationships may themselves reflect hydrologic variability.
For example, Angermeier and Schlosser (1989) documented that species-area
relationships were stronger in Panamanian than in northern temperate zone
streams (in Minnesota and Illinois). They noted that the temperate streams
were characterized by frequent shifts between physically harsh and benign
environmental conditions, and they suggested that the community organization
in these streams was predominantly influenced by immigration/extinction
dynamics, which tend to mask species-area relations as mobile opportunists
continually move into and out of available habitats. These dynamic processes
could contribute, in part, to the absence of clear species-area relations in
the WIMN streams. However, if environmental variability masks the species-
area relationship, then we would expect to see a stronger species-area
relationship expressed in the stable streams, yet this was not observed (see
Figure 44). The question of the relationship between species-area curves and
hydrologic variability deserves closer attention in future research.
One of the most interesting findings of this analysis is that functional
species traits provide theoretically interpretable insight into assemblage
structure in differentially-variable environments that goes beyond strictly
taxonomic information. This result provides some corroboration for
Southwood's (1977) "habitat templet" hypothesis, which has been recently
86
-------
proposed as particularly relevant for lotic ecosystems (Minshall 1988, Poff
and Ward 1989, 1990, Schlosser 1990, Poff 1992a, Townsend and Hildrew 1993).
Thus, a speculative application of our results would suggest that an
alteration in hydrologic regime may lead to adjustments in assemblage
structure. The use of functional species traits may be a more powerful tool
for projecting the potential assemblage repsonse to such broad scale change
than would traditional taxonomic approaches, because generalization across
zoogegraphic domains is possible.
These results are of particular interest in the context of climatic
change, which, at regional scales, is expected to alter precipitation-runoff
regimes, thereby modifying stream hydrology and community structure (see Grimm
1992, Poff 1992b). For example, Tonn (1990) argued that changes in local fish
fauna in response to climate change are less likely if assemblages are drawn
primarily from a regional species pool comprised of generalists, which are
more broadly tolerant across a range of environmental parameters. Generalist
fish species typically do better than specialists in invading new habitats
(Holdgate 1986), and specialist fish species that invade temporally variable
habitats may neither persist (Meffe 1984) nor modify community structure of
the invaded habitat (Zaret 1982). These considerations suggest that climate
change which increases environmental harshness or variability should favor
generalist species. Several scenarios can be put forth that would increase
environmental variability and/or harshness in WIMN streams (see Section 1 of
this report) . Decreased precipitation, coupled with higher regional
temperatures, would reduce habitat volume and presumably increase
physiological stress for stream fishes. Increases in frequency or duration of
hydrologic extremes is one predicted consequence of climate change (Rind
1989), and such changes would presumably favor generalist species. However,
for any particular region, great uncertainty exists in terms of predicting how
hydrologic regimes are likely to change. For example, precipitation in
climatically-distinct regions may change by ± 20% and runoff may change by ±
50% (Schneider et al. 1990). Thus, were the Wisconsin-Minnesota area to
become wetter over the coming decades, generalist species might not gain any
advantage.
The results of this study on 34 WIMN streams are encouraging for laying
the groundwork for more detailed and definitive studies that explore the
relationship between stream fish community structure and hydrologic
87
-------
variability. Given some of the problems associated with the dataset used in
this study, we caution against concluding that this study demonstrably shows
that a specified change in climate will have a specified ecological
consequence. Additional studies of this sort are needed before such a
conclusion is justified, even for the upper Midwest region. Future work
should focus on satisfying several criteria: collect fish data with
equivalent sampling intensity and techniques; record numerical abundances of
species (not just presence/absence); estimate population size structure;
collect samples over the same time periods (years); record local habitat
characteristics (e.g., depth, width, velocity, temperature) where samples are
collected; and ensure that hydrologic data of similar duration are available.
When applied to multiple sites across a broad geographic scale, studies that
satisfy these criteria have the potential to reveal important community-
environment patterns that can assist resource scientists in mitigating the
adverse impacts associated with rapid climate change.
CONCLUSIONS AND RECOMMENDATIONS
Stream fish assemblage data (presence/absence) were collected from 34
sites in Wisconsin and Minnesota where long-term hydrologic data also existed.
Fish assemblages were analyzed both in terms of taxonomic and functional
organization, and ecologically-similar groups were identified for both and
then related to independent hydrologic. factors using multivariate statistical
techniques. The taxonomic analysis (using TWINSPAN) showed strong geographic
patterns among taxonomically-similar groups. Zoogeographic patterns (inferred
from species range maps) played some role in dividing the 34 assemblages into
two major groups, but further division into four taxonomically-defined groups
of assemblages reflected hydrologic factors, not zoogeographic constraints.
The functional analysis of the 34 fish assemblages revealed that two or three
groups of sites could be defined in terms of functional organization (i.e.,
body morphology, trophic guild, habitat preferences, and tolerance values).
These functionally-similar groups were strongly correlated with independent
hydrologic factors that differed significantly among the 34 sites. Fish
assemblages defined in functional terms could thus be assigned to
hydrologically variable sites (high flow variability, tendency to become
seasonally intermittent) and hydrologically stable sites (low flow
88
-------
variability, high perennial baseflow). The stable sites could be further
broken down into moderately stable and very stable sites. The functional
organization of fish species in these two or three groups of sites followed
theoretical expectations and provided strong support for the view that
hydrologic factors are significant environmental variables that influence fish
community structure. Several species of fish were identified as indicative of
the variable-stable hydrologic gradient among stream sites.
The strong hydrologic-community relations found in the 34 WIMN sites
suggest that hydrologic alterations induced by climatic change (or other
anthropogenic disturbances) will modify stream fish assemblage structure in
this region. These changes may be detected by evaluating the communities
either in functional terms (e.g., generalists, tolerance) or in taxonomic
terms. Shortcomings in the dataset (uneven sampling intensity across sites,
variable periods of hydrologic record, etc.) suggest that our findings
represent robust, general patterns of broad ecological interest. However,
these same shortcomings caution against making firm predictions about specific
biotic responses to climatic change until more reliable datasets have been
evaluated.
89
-------
REFERENCES
Alexander, W.J.R. 1985. Hydrology of low latitude Southern Hemisphere land
masses. Hydrobiologia 125:75-83.
Allen, K.R. 1969. Distinctive aspects of the ecology of stream fishes: a
review. J. Fish. Res. Bd. Can. 26:1429-1438.
Allen, T.F.H., and T.W. Hoekstra. 1992. Towards a unified ecology. Columbia
Univ. Press, NY. 384 pp.
Allen, T.F.H., and S. Skagen. 1973. Multivariate geometry as an approach to
algal community analysis. Brit. Phycol. J. 8:267-287.
Allen, T.F.H., and T.B. Starr. 1982. Hierarchy: perspectives for ecological
complexity. Univ. Chicago Press, IL. 310 pp.
Angermeier, P.L. 1987. Spatiotemporal variation in habitat selection by fishes
in small Illinois streams. In: H.J. Matthews and D.C. Heins (eds.).
Community and evolutionary ecology of North American stream fishes.
Universiy of Oklahoma Press, Norman, pp. 52-60.
Angermeier, P.L., and I.J. Schlosser. 1989. Species-area relationships for
stream fishes. Ecology 70:1450-1462.
Bain, M.B., J.T. Finn, and H.E. Booke. 1988. Streamflow regulation and fish
community structure. Ecology 69:382-392.
Baker, V.R. 1977. Stream-channel response to floods, with examples from
central Texas. Geol. Soc. Am. Bull. 88:1057-1071.
Baker, V.R. 1988. Flood erosion. In: V.R. Baker, R.C. Kochel, and P.C. Patton
(eds.). Flood Geomorphology. John Wiley & Sons, New York, pp 81-95.
Beard, L.R. 1975. Generalized evaluation of flash-flood potential. Tech. Rep.-
-Univ. Texas Austin, Cent. Res. Water Resour. CRWR-124, 1-27.
Beckinsale, R.P. 1969. River regimes. In: R.J. Chorley (ed.). Water, earth,
and man. Methuen, London, pp. 455-471.
Biggs, B.J.F., M.J. Duncan, I.G. Jowett, J.M. Quinn, C.W. Hickey, R.J. Davies-
Colley, and M.E. Close. 1990. Ecological characterisation, classification,
and modelling of New Zealand rivers: an introduction and synthesis. N.Z.J.
Mar. Freshwat. Res. 24: 277-304
Boulton, A.J., C.G. Peterson, N.B. Grimm, and S.G. Fisher. 1992. Stability of
an aquatic macroinvertebrate community in a multiyear hydrologic
disturbance regime. Ecology 73:2192-2207.
90
-------
Bournard, M., H. Tachet, and A.L. Roux. 1987. The effects of seasonal and
hydrological influences on the macroinvertebrates of the Rhone River,
France. 2. ecological aspects. Arch. Hydrobiol.(Suppl.) 76:25-51.
Bozek, M.A., and W.A. Hubert. 1992. Segregation of resident trout in streams
as predicted by three habitat dimensions. Can. J. Zool. 70:886-890
Brown, J.H., and B.A. Maurer. 1989. Macroecology: the division of food and
space among species on continents. Science 243:1145-1150.
Bunn, S.E., D.H. Edward, and N.R. Loneragan. 1986. Spatial and temporal
variation in the macroinvertebrate fauna of streams of the northern Jarrah
Forest, Western Australia: community structure. Freshwat. Biol. 16:67-91.
Burkham, D.E. 1972. Channel changes of the Gila River in Safford Valley, Arizona,
1846-1970. U.S. Geol. Survey Prof. Paper 655G.
Carpenter, S.R., S.G. Fisher, N.B. Grimm, and J.F. Kitchell. 1992. Global
change and freshwater ecosystems. Annu. Rev. Ecol. Syst. 23:119-139.
Cole J., G. Lovett, and S. Findlay S, eds. 1991. Comparative analyses of
ecosystems: patterns, mechanisms, and theories. Springer-Verlag, New York,
375 pp
Colwell, R.K. 1974. Predictability, constancy, and contingency of periodic
phenomena. Ecology 55:1148-1153.
Coon, T.G. 1987. Responses of benthic riffle fishes to variation in stream
discharge and temperature. In: W.J. Matthews and D.C. Heins (eds.).
Community and evolutionary ecology of North American stream fishes.
Universiy of Oklahoma Press, Norman, pp. 77-92.
Conover, W.J. 1971. Practical nonparametric statistics. John Wiley & Sons, NY.
462 pp.
Corkum, L.D. 1989. Patterns of benthic invertebrate assemblages in rivers of
northwestern North America. Freshwat. Biol. 21:191-203.
Corkum, L.D., and J.J.H. Ciborowski. 1988. Use of alternative classifications
in studying broad-scale distributional patterns of lotic invertebrates.
J.N. Am. Benthol. Soc. 7:167-179.
Costa, J.E. 1974. Response and recovery of a Piedmont watershed from tropical storm
Agnes, June 1972. Water Res. Res. 10:106-112.
Gushing, C.E., C.D. Mclntire, J.R. Sedell, K.W. Cummins, G.W. Minshall, R.C.
Petersen, and R.L. Vannote. 1980. Comparative study of physical-chemical
variables of streams using multivariate analyses. Arch. Hydrobiol. 89:343-
352.
91
-------
Gushing, C.E., C.D. Mclntire, K.W. Cummins, G.W. Minshall, R.C. Petersen, J.R.
Sedell, and R.L. Vannote. 1983. Relationships among chemical, physical, and
biological indices along river continua based on multivariate analyses.
Arch. Hydrobiol. 98:317-326.
Dahm, C.N., and M.C. Holies, Jr. 1992. Streams in semiarid regions as
sensitive indicators of global climate change. In: P.L. Firth and S.G.
Fisher (eds.). Global climate change and freshwater ecosystems. Springer-
Verlag, NY. pp.250-260.
Delucchi, C.M. 1988. Comparison of community structure among streams with
different temporal flow regimes. Can. J. Zool. 66:579-586.
Delucchi, C.M. 1989. Movement patterns of invertebrates in temporary and
permanent streams. Oecologia 78:199-207.
Detenbeck, N.E., P.W. DeVore, G.J. Niemi, and A. Lima. 1992. Recovery of
temperate-stream fish communities from disturbance: a review of case
studies and synthesis of theory. Environ. Manage. 16:33-53.
Dewberry, T.C. 1980. A stream classification system for midwestern North
America. Unpublished manuscript.
Diamond, J. 1986. Overview: laboratory experiments, field experiments, and
natural experiments. In J. Diamond and T.J. Case (eds.). Community ecology.
Harper & Row, NY. pp. 3-22.
Duarte, C.M. 1991. Variance and the description of nature. In: J. Cole, G.
Lovett, and S. Findlay (eds.). Comparative analyses of ecosystems:
patterns, mechanisms, and theories, Springer-Verlag, New York, NY. pp.301-
318.
Dunne, T., and L.B. Leopold. 1978. Water in environmental planning. W.H.
Freeman and Co., San Francisco. 818 pp.
Dolph, J., and D. Marks. 1992. Characterizing the distribution of observed
precipitation and runoff over the continental United States. Climatic
Change 22:99-119.
Dyer, D.P. 1978. An analysis of species dissimilarity using multiple
environmental variables. Ecology 59:117-125.
Earthlnfo, Inc. 1990. Hydrodata users manual: USGS daily and peak flows. U.S.
West Optical Publishing, Denver, CO. 187 pp.
Efron, B. 1979. Bootstrap methods: another look at the jacknife. Annals Stat.
7:1-26.
92
-------
Fago, D. 1992. Distribution and relative abundance of fishes in Wisconsin
Vlil. Summary report. Technical Bulletin No. 175, Wisconsin Department of
Natural Resources.
Fausch, K.D., and R.G. Bramblett. 1991. Disturbance and fish communities in
intermittent tributaries of a western Great Plains river. Copeia 1991:659-
674.
Firth, P.L., and S.G. Fisher (eds.). 1992. Global climate change and
freshwater ecosystems. Springer-Verlag, New York.
Fisher, S.G. 1983. Succession in streams. In: J.R. Barnes and G.W. Minshall
(ed.). Stream ecology: application and testing of general ecological
theory. Plenum Press, New York. pp. 7-27.
Fisher, S.G., and N.B. Grimm. 1988. Disturbance as a determinant of structure
in a Sonoran Desert stream ecosystem. Verh. Internat. Verein. Limnol.
23:1193-1189.
Fisher S.G., and N.B. Grimm. 1991. Streams and disturbance: Are cross-
ecosystem comparisons useful? In: J. Cole, G. Lovett, and S. Findlay
(eds.). Comparative analyses of ecosystems: patterns, mechanisms, and
theories, Springer-Verlag, New York, NY. pp. 196-221.
Flaschka, I.M., C.W. Stockton, and W.R. Boggess. 1987. Climatic variation and
surface water resources in the Great Basin Region. Water Res. Bull. 23:45-57.
Frissell, C.A., W.J. Liss, C.E. Warren, and M.D. Hurley. 1986. A hierarchical
framework for stream habitat classification: viewing streams in a watershed
context. Environmental Management 10:199-214.
Gan, K.C., T.A. McMahon, and B.L. Findlayson. 1991. Analysis of periodicity in
streamflow and rainfall data by Colwell's indices. J. Hydrol. 123:105-118.
Gauch, H.G., Jr. 1982. Multivariate analysis in community ecology. Cambridge
University Press, Cambridge. 298pp.
Gentilli, J. 1952. Seasonal river regimes in Australia. Proc. 8th General
Assembly and 17th International Congress, International Geographical Union.
Washington, D.C. pp. 416-421.
Giese, J. 1987. Physical, chemical, and biological characteristics of least-
disturbed reference streams in Arkansas' ecoregions. Arkansas Department of
Pollution Control.
Gleick, P.H. 1987. The development and testing of a water-balance model for climate
impact assessment: modeling the Sacramento Basin. Water Res. Res. 23:1049-1061.
93
-------
Gleick, P.H. 1990. Vulnerability of water systems. In: P.E. Waggoner (ed.).
Climate change and U.S. water resources. John Wiley & Sons, NY. pp.223-240,
Gordon, N.D., T.A. McMahon, and B.L. Findlayson. 1992. Stream hydrology: an
introduction for ecologists. John Wiley and Sons, New York. 526 pp.
Gorman, O.T., and J.R. Karr. 1978. Habitat structure and stream fish
communities. Ecology 59:512-515.
Grimm, F. 1968. Das Abflufiverhalten in Europa Typen und regionale
Gliederung. Wissenschaftliche Veroffentlichungen des Deutschen Instituts
fur Landerkunde, Neue Folge 25/26: 18-180.
Grimm, N.B. 1992. Implications of climate change for stream communities. In:
J.G. Kingsolver, P.M. Kareiva, and R.B. Huey (eds.). Biotic interactions
and global change. Sinauer Assoc. Inc., Sunderland, MA. pp. 293-314.
Grimm, N.B., and S.G. Fisher. 1992. Response of arid-land streams to changing
climate. In: P.L. Firth and S.G. Fisher (eds.). Global climate change and
freshwater ecosystems. Springer-Verlag, NY. pp.211-233.
Grossman, G.D., P.B. Moyle, and J.O. Whitaker, Jr. 1982. Stochasticity in
structural and functional characteristics of an Indiana stream fish
assemblage: a test of community theory. Amer. Nat. 120:423-454.
Haines, A.T., B.L. Finlayson, and T.A. McMahon. 1988. A global classification
of river regimes. Applied Geography 8:255-272.
Hanson, D.L., and T.F. Waters. 1974. Recovery of standing crop and production
rates of a brook trout population in a flood damaged stream. Trans. Am.
Fish. Soc. 103:431-439.
Harrell, H.L. 1978. Response of the Devil's River (Texas) fish community to
flooding. Copeia 1978:60-68.
Harvey, B.C. 1987. Susceptibility of young-of-the-year fishes to downstream
displacement by flooding. Trans. Am. Fish. Soc. 116:851-855.
Haslam, S.M. 1978. River plants. Cambridge University Press. Cambridge,
England.
Hawkes, C.L., D.L. Miller, and W.G. Layher. 1986. Fish ecoregions of Kansas:
stream fish assemblage patterns and associated environmental correlates.
Environ. Biol. Fishes. 17:267-279.
Hildrew, A.G., and P.S. Giller. 1993. Patchiness, species interactions and
disturbance in the stream benthos. In: P.S. Giller, A.G. Hildrew, and D.
Raffaelli (eds.). Aquatic ecology: scale, pattern and process. Blackwell
Scientific Publ., Oxford. (In Press)
94
-------
Hill, M.O. 1979a. DECORANA - a FORTRAN program for detrended correspondence
analysis and reciprocal averaging. Ithaca, NY: Cornell University.
Hill, M.O. 1979b. TWINSPAN - a FORTRAN program for arranging multivariate data
in an ordered two-way table by classification of the individuals and
attributes. Ithaca, NY: Cornell University.
Holdgate, M.W. 1986. Summary and conclusions: characteristics and consequences
of biological invasions. Phil. Trans. Royal Soc. London B, Biol. Sci.
314:733-742.
Horwitz, R.J. 1978. Temporal variability patterns and the distributional
patterns of stream fishes. Ecol. Monogr. 48:307-321.
Hughes, J.M., and B. James. 1989. A hydrological regionalization of streams in
Victoria, Australia, with implications for stream ecology. Australian
Journal of Marine and Freshwater Research 40:303-326.
Hughes, R.M., and J.M. Omernik. 1983. An alternative for characterizing stream
size. In: T.D. Fontaine, III, and S.M. Bartell (ed.). Dynamics of lotic
ecosystems. Ann Arbor Science, Ann Arbor, MI. pp. 87-102
Hynes, H.B.N. 1970. The ecology of running waters. University of Toronto
Press, Toronto.
Iverson, M., P. Wiberg-Larsen, S.B. Hansen, and F.S. Hansen. 1978. The effect
of partial and total drought on the macroinvertebrate communities of three
small Danish streams. Hydrobiol. 60:235-242.
Jackson, D.A., and H.H. Harvey. 1989. Biogeographic associations in fish
assemblages: local vs. regional processes. Ecology 70:1472-1484.
John, K.R. 1963. The effect of torrential rains on the reproductive cycle of
Rhinichthys osculus(Girard) in the Chiricahua Mountains, Arizona. Copeia
1963:286-291.
John, K.R. 1964. Survival of fish in intermittent streams of the Chiracahua
Mountains, Arizona. Ecology 45:112-119.
Johnson, R.A., and D.W. Wichern. 1982. Applied multivariate statistical
analysis. Prentice-Hall, Englewood Cliffs, NJ. 594 pp.
Jowett, I.G, and M.J. Duncan. 1990. Flow variability in New Zealand rivers and
its relationship to in-stream habitat and biota. N.Z.J. Mar. Freshwat. Res.
24: 305-317
Karr, J.R., K.D. Fausch, P.L. Angermeier, P.R. Yant, and I.J. Schlosser. 1986.
Assessing biological integrity in running waters: a method and its
rationale. Illinois Natural History Survey Special Publication 5, 28 pp.
95
-------
Kingsolver, J.G., P.M. Kareiva, and R.B. Huey (eds.). 1992. Biotic
interactions and global change. Sinauer Associates Inc., Sunderland, Mass.
Knox, J.C. 1972. Valley alluviation in south-western Wisconsin. Ann. Assoc. Amer.
Geogr. 62:401-410.
Karl, T.R. 1988. Multi-year fluctuations of temperature and precipitation: the gray
area of climate change. Climatic Change 12:179-197.
Karl, T.R., and W.E. Riebsame. 1989. The impact of decadal fluctuations in mean
precipitation and temperature on runoff: a sensitivity study over the United
States. Climatic Change 15:423-447.
Kochel, R.C. 1988. Geomorphic impact of large floods: review and new perspectives on
magnitude and frequency. In: V.R. Baker, R.C, Kochel, and P.C. Patton (eds.).
Flood geomorphology, John Wiley & Sons, NY. pp 169-187.
Ladle, M., and J.A.B. Bass. 1981. The ecology of a small chalk stream and its
responses to drying during drought conditions. Arch. Hydrobiol. 90:448-466.
Lancaster, J., and A.G. Hildrew. 1993. Characterizing in-stream flow refugia.
Can. J. Fish. Aquat. Sci. (In Press).
Langbein, W.B. 1949. Annual runoff in the United States. U.S. Geological Survey
Circular 5.
Lee, D.S., C.R. Gilbert, C.H. Hocutt, R.E. Jenkins, D.E. McAllister, and J.R.
Stauffer, Jr. 1980. Atlas of North American Freshwater Fishes. N.C. St.
Mus. Nat. Hist., Raleigh, NC.
Lettenmaier, D.P., and T.Y. Gan. 1990. Hydrologic sensitivities of the Sacramento-
San Joaquin River Basin, California, to global warming. Water Res. Res. 26:69-86.
Levin, S.A. 1992. The problem of pattern and scale in ecology. Ecology
73:1943-1967.
Linsley, R.K., Jr., M.A. Kohler, and J.L.H. Paulhus. 1982. Hydrology for
engineers, 3rd ed. McGraw-Hill, New York. 508 pp.
Ludwig, J.A., and J.F. Reynolds. 1988. Statistical ecology. John Wiley & Sons,
NY. 337 pp.
Mahon, R. 1984. Divergent structure in fish taxocenes of north temperate
streams. Can. J. Fish. Aquat. Sci. 41:330-350.
Marchant, R., P. Mitchell, and R. Norris. 1984. Distribution of benthic
invertebrates along a disturbed section of the La Trobe River, Victoia: an
analysis based on numerical classification. Australian Journal of Marine
and Freshwater Research 35:355-374.
Margalef, R. 1968. Perspectives in ecological theory. Univ. Chicago Press, IL.
96
-------
Matalas, N.C. 1990. What statistics can tell us. In: P.E. Waggoner (ed.). Climate
change and U.S. water resources, John Wiley & Sons, NY. pp 139-149.
Matthews, W.J. 1986. Fish faunal structure in an Ozark stream: stability,
persistence, and a catastrophic flood. Copeia 1986: 388-397.
McMahon, T.A. 1982. Hydrological characteristics of selected rivers of the
world. Technical Documents in Hydrology, Unesco, Paris. (Serial # SC-
82/WS/51).
McMahon, T.A. 1979. Hydrological characteristics of arid zones. Proc. of the
Canberra Symposium - The hydrology of area of low precipitation. IAHS-AISH
Publ. No. 128, pp. 105-123.
Meffe, G.K. 1984. Effects of abiotic disturbance on coexistence of predator
and prey fish species. Ecology 65:1525-1534.
Meisner, J.D. 1990a. Effect of climatic warming on the .southern margins of the
native range of brook trout, Salvenius fontinalis. Can. J. Fish. Aq. Sci.
47:1065-1070
Meisner, J.D. 1990b. Potential loss of thermal habitat for Brook Trout, due to
climatic warming, in two southern Ontario streams. Trans. Am. Fish. Soc.
119:282-291.
Meisner, J.D., J.S. Rosenfeld, and H.A. Regier. 1988. The role of groundwater
in the impact of climate warming on stream salmonines. Fisheries 13:2-8.
Menge, B.A., and A.M.Olson. 1990. Role of scale and environmental factors in
regulation of community structure. Trends Ecol. Evol. 5:52-57.
Minckley, W.L., and G.K. Meffe. 1987. Differential selection by flooding in
stream fish communities of the arid American Southwest. In: W.J. Matthews
and D.C. Heins (eds.). Community and evolutionary ecology of North American
stream fishes. University of Oklahoma Press, Norman, pp. 93-104.
Minshall, G.W. 1988. Stream ecosystem theory: a global perspective. J. N. Am.
Benthol. Soc. 7:263-288.
Molles, M.C., Jr, and C.N. Dahm. 1990. A perspective on El Nino and La Nina:
global implications for stream ecology. J. No. Amer. Benthol. Soc. 9:68-76.
Moss, D., M.T. Furse, J.F. Wright, and P.O. Armitage. 1987. The prediction of
the macro-invertebrate fauna of unpolluted running-water sites in Great
Britain using environmental data. Freshwat. Biol. 17:41-52.
Moss, M.E., and H.F. Lins. 1989. Water resources in the Twenty-First Century a
study of the implications of climate uncertainty. U. S. Geological Survey
Circular 1030.
97
-------
Moyle, P.B., and H.W. Li. 1979. Community ecology and predator-prey relations
in warmwater streams. In: H. Clepper (ed.). Predator-prey systems in
fisheries management. Sports Fishing Institute, Washington, D.C. pp. 171-
180.
Nelson, R.L., W.S. Platts, D.P. Larsen, and S.E. Jensen. 1992. Trout
distribution and habitat in relation to geology and geomorphology in the
North Fork Humboldt River drainage, northeastern Nevada. Trans. Am. Fish.
Soc. 121:405-426.
Nesler, T.P., R.T. Muth, and A.F. Wasowicz. 1988. Evidence for baseline flow
spikes as spawning cues for Colorado squawfish in the Yampa River,
Colorado. Amer. Fish. Soc. 5:68-79.
Nikolskii, G.V. 1963. The ecology of fishes. Academic Press, NY. 352 pp.
Norris, R.H., and A. Georges. 1993. Analysis and interpretation of benthic
macroinvertebrate surveys. In: D.M. Rosenberg, and V.H. Resh (eds.).
Freshwater biomonitoring and benthic macroinvetebrates. Chapman & Hall, NY.
NY. pp. 234-286
Ohio Environmental Protection Agency. 1989. Biological criteria for the
protection of ao^iatic life: Vol. III. Standardized biological field
sampling and laboratory methods for assessing fish and macroinvertebrate
communities. Division of Water Quality Monitoring and Assessment, Columbus,
OH.
Omernik, J.M. 1987. Ecoregions of the conterminous United States. Annals
Assoc. Am. Geogr. 77:118-125.
O'Neill, R.V., D.L. DeAngelis, J.B. Waide, and T.F.H. Allen. 1986. A
hierarchical concept of ecosystems. Princeton Univ. Press, NJ. 253 pp.
Orians, G.H. 1987. Ecological comparisons. Science 235:226-227.
Ormerod, S.J. 1987. The influences of habitat and seasonal sampling regimes on
the ordination and classification of macroinvertebrate assemblages in the
catchment of the River Wye, Wales. Hydrobiol. 150:143-151.
Ormerod, S.J., and R.W. Edwards. 1987. The ordination and classification of
macroinvertebrate assemblages in the catchment of the River Wye in relation
to environmental factors. Freshwat. Biol. 17:533-546.
Pace, M.L. 1993. Forecasting ecological responses to global change: the need
for large-scale comparative studies. In: P.M. Kareiva, J.G. Kingsolver, and
R.B. Huey. (eds.). Biotic interactions and global change. Sinauer
Associates Inc., Sunderland, MA. pp. 356-363.
98
-------
Page, L.M, and B.M. Burr. 1991. A field guide to freshwater fishes: North
America north of Mexico. Houghton Mifflin Co., Boston. 432 pp.
Patrick, R. 1975. Stream communities. In: M.L. Cody and J.M. Diamond (ed.).
Ecology and evolution of communities. Belknap Press. Cambridge, MA. pp.
445-459.
Pearsons, T.N., H.W. Li, and G.A. Lamberti. 1992. Influence of habitat
complexity on resistance to flooding and resilience of stream fish
assemblages. Trans. Am. Fish. Soc. 121:427-436.
Peckarsky, B.L. 1983. Biotic interactions or abiotic limitations? A model of
lotic community structure. In: T.D. Fontaine, III, and S.M. Bartell (eds.).
Dynamics of lotic ecosystems. Ann Arbor Science, Ann Arbor, MI. pp.303-324.
Peterson, C.G. 1987. Influences of flow regime on development and desiccation
response of lotic diatom communities. Ecology 68:946-954.
Poff, N.L. 1992a. A perspective on the definition of disturbance: why
disturbances can be predictable. J. N. Am. Benthol. Soc. 11:86-92.
Poff, N.L. 1992b. Regional hydrologic response to climate change: an
ecological perspective. In: P.L. Firth and S.G. Fisher (eds.). Global
climate change and freshwater ecosystems. Springer-Verlag, NY. pp.88-115.
Poff, N.L., and J.V. Ward. 1989. Implications of streamflow variability and
predictability for lotic community structure: a regional analysis of
streamflow patterns. Canadian Journal of Fisheries and Ao^iatic Sciences.
46:1805-1818.
Poff, N.L., and J.V. Ward. 1990. The physical habitat template of lotic
systems: recovery in the context of historical pattern of spatio-temporal
heterogeneity. Environmental Management 14:629-646.
Poff, N.L., N.J. Voelz, J.V. Ward, and R.E. Lee. 1990. Algal colonization
under four experimentally-controlled current regimes in a high mountain
stream. Journal of the North American Benthological Society 9:303-318.
Power, M.E., and A.J. Stewart. 1987. Disturbance and recovery of an algal
assemblage following flooding in an Oklahoma stream. Am. Midi. Nat.
117:333-345.
Power, M.E., R.J. Stout, C.E. Gushing, P.P. Harper, F.R. Hauer, W.J. Matthews,
P.B. Moyle, B. Statzner, and I.R. Wais De Badgen. 1988. Biotic and abiotic
controls in river and stream communities. J. N. Am. Benthol. Soc. 7:456-
479.
99
-------
Rahel, F.J. 1990. The hierarchical nature of community persistence: a problem
of scale. Am. Nat. 136:328-344.
Rahel, F.J., and W.A. Hubert. 1991. Fish assemblages and habitat gradients in
a Rocky Mountain-Great Plains stream: biotic zonation and additive patterns
of community change. Trans. Am. Fish. Soc. 120:319-332.
Regier, H.A., J.A. Magnuson, and C.C. Coutant. 1990. Introduction to
proceedings: symposium on effects of climate change on fish. Trans. Am.
Fish. Soc. 119:173-175.
Reice, S.R., R.C. Wissmar, and R.J. Naiman. 1990. Disturbance regimes, resilience,
and recovery of animal communities and habitats in lotic ecosystems. Environm.
Manage. 14:647-660.
Resh, V.H., A.V. Brown, A.P. Covich, M.E. Gurtz, H.W. Li, G.W. Minshall, S.R.
Reice, A.L. Sheldon, J.B. Wallace, and R. Wissmar. 1988. The role of
disturbance in stream ecology. J. N. Am. Benthol. Soc. 7:433-455.
Revelle, R., and P. Waggoner. 1983. Effects of a carbon dioxide-induced climatic
change on water supplies in the Western United States. In: Changing climate,
National academy of sciences, National Academy Press, Washington, D.C. pp. 419-
432.
Richards, R.P. 1990. Measures of flow variability and a new flow-based
classification of Great Lakes tributaries. J. Great Lakes Res. 16:53-70.
Ricklefs, R.E. 1987. Community diversity: relative roles of local and regional
processes. Science 235:167-171.
Rind D, Goldberg R, Ruedy R (1989) Change in climate variability in the 21st
Century. Climatic Change 14:5-37.
Robison, H.W., and T.M. Buchanan. 1988. Fishes of Arkansas. University of
Arkansas Press, Fayetteville. 536 pp.
Roughgarden, J. 1989. The structure and assembly of communities. In J.
Roughgarden, R.M. May, and S.A. Levin (eds.). Perspectives in ecological
theory. Princeton University Press, pp. 203-226.
SAS Institute Inc. 1988. SAS/STAT user's guide, release 6.03 edition. Gary,
NC.
Schaake, J.C. 1990. From climate to flow In: P.E. Waggoner (ed.). Climate change and
U.S. water resources, John Wiley fi Sons, NY. pp 177-206.
Schoener, T.W. 1986. Overview: kinds of ecological communities - ecology
becomes pluralistic. In: J. Diamond, and T.J. Case (eds.). Community
ecology, Harper & Row Publ., NY. pp. 467-479.
100
-------
Schoener, T.W. 1987. Axes of controversy in community ecology. In: W.J.
Matthews and D.C. Heins (eds.). Community and evolutionary ecology of North
American stream fishes. Universiy of Oklahoma Press, Norman, pp. 8-16.
Schlosser, I.J..1982. Fish communtiy structure and function along two habitat
gradients in a headwater stream. Ecol. Monogr. 52:395-414.
Schlosser, I.J. 1985. Flow regime, juvenile abundance, and the assemblage
structure of stream fishes. Ecology 66:1484-1490.
Schlosser, I.J. 1987. A conceptual framework for fish communities in small
warmwater streams. In: W.J. Matthews and D.C. Heins (eds.). Community and
evolutionary ecology of North American stream fishes. University of
Oklahoma Press, Norman, pp. 17-24
Schlosser, I.J. 1990. Environmental variation, life history attributes, and
community structure in stream fishes: implications for environmental management
assessment. Environ. Manage. 14:621-628.
Schlosser, I.J., and L.A. Toth. 1984. Niche relationships and population
ecology of rainbow (Etheostoma caeruleum) and fantail (E. f label!are)
darters in a temporally variable environment. Oikos 42:229-238.
Schneider, S.H., P.H. Gleick, and L.O. Mearns. 1990. Prospects for climate change.
In: P.E. Waggoner (ed.). Climate change and U.S. water resources, John Wiley &
Sons, NY. pp. 41-73.
Schumm, S.A. 1968. River adjustment to altered hydrologic regimen Murrumbidgee
River and Paleochannels, Australia. U. S. Geological Survey Professional Paper
598.
Schumm, S.A., and R.W. Lichty. 1963. Channel widening and flood-plain construction
along Cimarron River in southwestern Kansas. U. S. Geological Survey Professional
Paper 352-D.
Scott, W.B., and E.J. Grossman. 1973. Freshwater fishes of Canada. Bull. Fish.
Res. Bd. Can. 184.
Sedell, J.R., G.H. Reeves, F.R. Hauer, J.A. Stanford, and C.P. Hawkins. 1990.
Role of refugia in recovery from disturbances: modern fragmented and
disconnected river systems. Environ. Manage. 14:711:724.
Seegrist, D.W., and R. Card. 1972. Effects of floods on trout in Sagehen
Creek, California. Trans. Amer. Fish. Soc. 101:478-482.
Slack, J.R., and J.M. Landwehr. 1992. Hydro-climatic data network (HCDN): a
U.S. Geological Survey streamflow data set for the United States for the
101
-------
study of climate variations, 1874-1988. U.S. Geological Survey Open-File
Report 92-129, Reston, VA. 193 pp.
Southwood, T.R.E. 1977. Habitat, the templet for ecological strategies?
Journal of Animal Ecology 46:337-365.
Southwood, T.R.E. 1988. Tactics, strategies and templets. Oikos 52:3-18.
Stanford, J.A., and J.V. Ward. 1983. Insect species diversity as a function of
environmental variability and disturbance in streams. In: J.R. Barnes, and
G.W. Minshall (eds.). Stream ecology: application and testing of general
ecological theory. Plenum Press, NY. pp. 265-278.
Stearns, S.C. 1976. Life-history tactics: a review of the ideas. Quart. Rev.
Biol. 51:3-47.
Steel, R.G.D., and J.H. Torrie. 1980. Principles and procedures of statistics:
a biometrical approach, 2nd. ed. McGraw-Hill, NY. 633 pp.
Tonn, W.M. 1990. Climate change and fish communities: a conceptual framework.
Trans. Am. Fish. Soc. 119: 337-352
Tonn, W.M., J.J. Magnuson, and A.M. Forbes. 1983. Community analysis in
fishery management: an application with northern Wisconsin lakes. Trans.
Am. Fish. Soc. 112:368-377.
Tonn, W.M., J.J. Magnuson, M. Rask, and J. Toivonen. 1990. Intercontinental
comparison of small-lake fish assemblages: the balance between local and
regional processes. Amer. Natur. 136:345-375.
Toth, L.A., D.R. Dudley, J.R. Karr, and O.T. Gorman. 1982. Natural and man-
induced variability in a silverjaw minnow (Ericymba Jbuccata) population.
Amer. Midi. Nat. 107:284-293.
Townsend, C.R., and A.G. Hildrew. 1993. Species traits in relation to a
habitat templet for river systems. Freshwat. Biol. (In Press)
Townsend, C.R., A.G. Hildrew,and K. Schofield. 1987. Persistence of stream
invertebrate communities in relation to environmental variability. J. Anim.
Ecol. 56:597-614.
Trautman, M.B. 1981. The fishes of Ohio. Ohio State University Press,
Columbus, OH.
Vannote, R.L., G.W. Minshall, K.W. Cummins, J.R. Sedell, and C.E. Gushing.
1980. The river continuum concept. Can. J. Fish. Aquat. Sci. 37:130-137.
Wallace, J.B. 1990. Recovery of lotic macroinvertebrate communities from
disturbance. Environ. Manage. 14:605-620.
102
-------
Wallis, J.R., D.P. Lettenmaier, and E.F. Wood. 1991. A daily
hydroclimatological data set for the continental United States. Water
Resources Research 27:1657-1663.
Ward, J.V. 1991. Aquatic insect ecology. 1. Biology and habitat. John Wiley &
Sons, NY. 438 pp.
Ward, J.V., and J.A. Stanford, eds. 1979. The ecology of regulated rivers.
Plenum, NY. 398 pp.
Ward, J.V., and J.A. Stanford. 1983. The intermediate disturbance hypothesis:
an explanation for biotic diversity patterns in lotic ecosystems. In: T.D.
Fontaine, III and S.M. Bartell (eds.) Dynamics of lotic ecosystems. Ann
Arbor Press, Ann Arbor, MI. pp. 347-356.
Watters, G.T. 1992. Unionids, fishes, and the species-area curve. J. Biogeogr.
19:481-490.
Webb, P.W., and D. Weihs. 1986. Functional locomotor morphology of early life
history stages of fishes. Trans. Am. Fish. Soc. 115:115-127.
wiens, J.A. 1989. Spatial scaling in ecology. Functional Ecology 3:385-397.
Williams, D.D., and H.B.N. Hynes. 1976. The ecology of temporary streams. I.
The faunas of two Canadian streams. Int. Revue ges. Hydrobiol. 61:761-787.
Williams, D.D., and H.B.N. Hynes. 1977. The ecology of temporary streams. II.
General remarks on temporary streams. Int. Revue ges. Hydrobiol. 62:53-61.
Williams, G.P. 1978. The case of the shrinking channels the North Platte and
Platte Rivers, Nebraska. U.S. Geological Survey Circular 781.
Winemiller, K.O., and K.A. Rose. 1992. Patterns of life-history
diversification in North American fishes: implications for population
regulation. Can. J. Fish. Aquat. Sci. 49:2196-2218.
Wong, M.A., and C. Schaack. 1982. Using the kth nearest neighbor clustering
procedure to determine the number of subpopulations. Proc. Stat. Computing
Section , Amer. Stat. Assoc. 1982:40-48.
Wright, J.F., D. Moss, P.O. Armitage, and M.T. Furse. 1983. A preliminary
classification of running-water sites in Great Britain based on
macroinvertebrate species and the prediction of community type using
environmental data. Freshwat. Biol. 14:221-256.
Yevjevich, V. 1972. Probability and statistics in hydrology. Water Resources
Publications, Littleton, CO. 302 p.
Zaret, T.M. 1982. The stability/diversity controversy: a test of hypotheses.
Ecology 63:721-731.
103
-------
Figure 1. Map of all 816 sites used in the analysis showing distribution for a) states, b)
hydrologic units, and c) ecoregions. Sites coded by "A" represent the 420
"best" sites, and sites coded by "+" indicate additional sites included in the
complete 816-site analsysis (see Methods).
104
-------
STATES / CLUS420 & CLUS816
IT)
O
MUM
1000
A CLUS420 POINTS
+ CLUS816 AND NOT "BEST
-------
HYDROLOGIC UNITS / CLUS420 & CLUS816
j Si \.-/^&> *T *5~^T"T^" T ^&VT-VV.,,
/«$ «^t^(V*/'^/^
M-"v^ ^'tsyw - w If
fe44 T 7? >A?Afl4l'
lfaLfyr^f*5-^^m^
'"&-)' V a yYv-' / 7 »^i^)m5 v'V'
^^CrvfesiSlI^/-^
^t^^^fife
CLUS420 POINTS
CLUS816 AND N(
ID
o
BEsr
-------
ECOREGIONS / CLUS420 & CLUS816
1000
CLUS420 POINTS
CLUS816 AND NOT 'BEST'
-------
A) N=420
0)
£
CO
o
A
3
160
140
120
100
80
60
40
20
VA
t t 4
VA
VA.
k X X
^ t *
XXX
VV
x'xV
x'xV
XXX
s'x'x'
XXX
XXX
VA
XXX
XXX
s " -N
VV
X X
X X
X X
'//,
X X
x -N '
,\\
Vx'
X X
s-s-
'x'x^
/ ' f
w
X X '
'x x '
.
\\Jx
-i-i-i
-V/N
1 'x IN
1
1
XXX
x'x'x'
x'x'x'
k'x'x'
1
-
*
,
1000
2000
3000
4000
5000
Area (km*)
o
55
A
E
B) N=816
400
' Y/////P7-A J ; )\JJ A J
1000
2000 3000
Area (km2)
4000 5000
Figure 2. Histograms for catchment area for a) "best" sites and b) "all" sites.
108
-------
A) N=420
26 ft
9 W
«9 n n
4 M U
0
1 c ft_
O
b.
O 100-
E
3
Z en
50-
0 -
XXX
ft*
v,\
^ X X
* * *
XXX
XXX
XXX
* * f
XXX
x'x'x'
XXX
* * t
.'.'.'
>->->-
X**XN
X X
X X
X \ X
* + *
[=
sz:
-H
HH
-
.
0 20 40 60 80 100
Mean Daily Flow (m3 sec*1)
B) N=816
0 20 40 60 80 100
Mean Daily Flow (m3 sec*1)
Figure 3. Histograms for mean daily flow (DAYAVE) for a) "best" sites and b)
"all" sites.
109
-------
A) N=420
300
(0 200
° 150
O
g 100
3
z
50
X / X X X
^'X'X
\ s / x XN
X / X X X
x x x x x
x'XvX
\N\\\S
N'S'%'V'N'>
T^\ \ \ \ N
\
\
\
\
\
V
N ^
\
\
\
V
\ \
\ \
\ \
\ \
'""
_
-
.
.
0.5 1 1.5 2 2.5
Mean Annual Runoff (m yr"1)
B) N=816
0.5 1 1.5 2 2.5
Mean Annual Runoff (m y r*1)
Figure 4. Histograms for mean annual runoff (MAR) for a) "best" sites and b)
"all" sites.
no
-------
A) N=420
50
2f\ A _
0 U
1
* 150-
1 9 U
"5
5 1 r,r,
* 100-
5 A
U
o -
\ \ \
xv:
£<
X **
* * *
NxNyN
xx xx
/ XN
\Vx
x-:
w
1 / S
»%xx/
'X/ *
* * *
\VN
W
XxV/>
\ \ '
LJ_I.
^-.
"^
500 1000 1500 2000 2500 3000
Elevation (m)
B) N=816
9 V V-
_ An n -
9 4 U U
O
55
300-
o
fe
o on n-
^ £U U ~
|
Z 1 n o-
1 w V
n .
^
1
l
vx
1
1
ov
i
i
\ \
7%
\ \
77<
\ \
\ \
EZZ3
\
tZ2=
\ \
EZZ
1 1
772
1 1
I
_
1
_
500 1000 1500 2000 2500 3000
Elevation (m)
Figure 5. Histograms for stream gauge elevation for a) "best" sites and b) "all"
sites. (Note: not all sites present due to some missing values.)
-------
A) N=420
80
70
S 60
9
>
55
O
A
50
40
30
20
1 0
0
"
'
\'\""
V >'
L*'S\
\ s
* \.
X N'
\ s
V
. X X
*
x" V
#
/ X
\ \ '
-xj
:*i
'ft
x \
f * 4
X \
'xV
^-^
kX X
>X /
*^ ^
^
/ /
*S ,
* * f
^ \
* / ^
:^
/
\ \
X
\ \
',',
\\
v
^i
\ \
'Ss
\ \
\ X
:-:-:
%
S ,
\ \
\ \
,'v
S<
\ \
* .
*/*
V
\ \
x'x
/ /^
f *
/ f^
y x
x /(
\ \
ft*
\ \
\v
:-x
;«
:«
1 X X
V,
W
'«
>>;
>xV
V/1
\ \ '
:%
%s:
\ \
X \
^
\ \
\ \
V'S
N'^
3
X .
\ \
X i
\ \
X .
sV
»>:
X X
x'V
v'v
x'x*
X ,
\ V
X i
X X
34
38 42 46 50 54 58
Period of Record (yr)
0)
a
CO
o
B) N=816
10 20 30 40 50
Period of Record (yr)
60
Figure 6. Histograms for period of record (POR) for a) "best" .sites and b) "all"
sites.
112
-------
A) N=420
800
60°
CB
o
e
400
£ 200
0)
O
o
x
B) N=816
e
o
£
75
o
o
O
'Figure 7. Range of values for coefficient of variation for daily flow (DAYCV) for 10
clusters for a) "best" sites and b) "all" sites. Each box encloses 50% of observed
values (median = horizontal line). Observed range is contained.within upper and
lower bars except for extreme outliers (circles). Stream type abbreviations are
provided in Table 5.
113
-------
A) N=420
B) N=816
cc 5 w oc
a. o en (o
f- CM
z z
(0 (0
5: E t 2
Figure 8. Range of values for Colwell's predictability of daily flow (DAYPRED)
for 10 clusters for a) "best" sites and b) "all" sites. Symbols and
abbreviations same as in Figure 7.
114
-------
N=420
1.5
tfl ** N 9
8 S g §
B) N&816
u
i
O
O
Figure 9. Range of values for flood frequency (FLODFREQ) for 10 clusters for a)
"best" sites and b) "all" sites. Symbols and abbreviations same as in
Figure 7.
115
-------
JO
CO
CD
£
a
o
o
A) N=420
W^
o
T
f
j.
i
*- «M
B) N=816
«
5
o
a
TJ
O
O
z
Figure 10. Range of values for flood predictability (FLODPRED) for 10 clusters
for a) "best" sites and b) "all" sites. Symbols and abbreviations
same as in Figure 7.
116
-------
A) N=420
a o co
co co
B) N=816
ct £ i
Figure 11. Range of values for flood-free period (FLODFREE) for 10 clusters
for a) "best" sites and b) "all" sites. Symbols and abbreviations
same as in Figure 7.
117
-------
A) N=420
4
£1 1 R -
_ I 9
« s
3 (B
e -
e to ,
2> 1"
«
£ 0.5-
U.
n
15-
_ 1.3
e ^ <
2
T}
e o
"5 0 5-
_ 0.5
o
' 0
o
h o
o
_ o
4
DC
Q.
B)
O
+
I 1
8
*
8
5 co
O co
Ns816
i
8
f
O
o
T
_!_'
i
£
CO
M
OC
CO
:
:
g
1
o
CO
I
j
:
:
3
b
o o
o o
pL
i
\
i J
n
oc $ co tr 5
^s
r
o
T
T
U.
Q.
O
O
T
***** '* J"
1
til
OC IL
O
n
I
o
1 1 i
i Q i !
i O i i
i i i
! 0 i o i
! i
i ° o T
3 -T. ' O f\
\ .
^
i
:
:
:
:
!^
W
^
ll
!9iu
!e|T
iS^
1^
i i i -1-
: ^» :
>;$5i
fe^
| JL
-
i &E t x
Figure 12. Range of values for interannual flood variability (FLODVAR) for 10
clusters for a) "best" sites and b) "all" sites. Symbols and
abbreviations same as in Figure 7
118
-------
A) N=420
DC $ co
Q. O CO
B) Ns816
co co
*
Figure 13. Range of values for average flood duration (FLODDUR) for 10
clusters for a) "best" sites and b) "all" sites. Symbols and
abbreviations same as in Figure 7.
119
-------
A) N=420
X
(0
o
o
"o
o»
9
CD
B) N=816
X
O
o
3
o
"5
0)
eg
CD
Figure 14. Range of values for baseflow index (BFI) for 10 clusters for a)
"best" sites and b) "all" sites. Symbols and abbreviations same as in
Figure 7.
120
-------
A) N=420
0.8
«. O 0.6
O 2
5 0.4
B) N=816
II
o *:
S 5
tr 5 co oc
o. o eo co
CM
to
00 _
Oft -
OA
.*
0« _
«
~
.
'
'
fl
,
@|
'
vJ
^r
Q.
r7*T*n
^ jL1'
-
]
:
.
:
I
X
-
.
sw
Si
s«s«*
s$
%
-------
A) N=420
B) N=816
.0
«
t>
5
$
O
Figure 16. Range of values for lowflow predictability (LOWPRED) for 10
clusters for a) "best" sites and b) "all" sites. Symbols and
abbreviations same as in Figure 7.
122
-------
A) N=420
0
Q.
o
0
o
o
a. o co
W (0
B) N=816
O
Q.
O
"- c u.
Figure 17. Range of values for lowflow-free period (LOWFREE) for 10 clusters
for a) "best" sites and b) "all" sites. Symbols and abbreviations
same as in Figure 7.
123
-------
A) N=420
5000
B) N=816
£ = I
Figure 18. Range of values for catchment area (AREA) for 10 clusters for a)
"best" sites and b) "all" sites. Symbols and abbreviations same as in
Figure 7.
124
-------
A) N=420
3000-
2 3 0 0
O 2000 ~
^ i e n n -
8> i 50 O
UJ
A 1000-
0)
3
5 e A A
(3 500
o -
c
f
c
(.
t
r
5
5
5
)
3
j
H
C
C
i,
k>a>
>
?
.>
J=:>»J
O
o
T
T'
I
c
c
c
L
I
)
)
N
-I
c
^v
5 !
rl
s\
S^ '-
J .
\
<>t
£
^
^
T
i
|
:
!
|c
m
P
5
ft
L
.
SI
I
b
o
£-,
I
-
P*73
r»^;
₯ ^ co S
o-otogggfeSt
N=816
Figure 19. Range of values for gauge elevation (ELEV) for 10 clusters for a)
"best" sites and b) "all" sites. Symbols and abbreviations same as in
Figure 7.
125
-------
A) N=420
120-
«i n n
_ 100
O
5
» 'o
5 2 60-
(0 00-
o «_
« E 40-
w
o
20
o -j
8
1
T
T
B
E
a.
{^
IS
J&
$
*
v*>
^
T
(0
(0
c
(
£
»A
D
)
C
s>
sv
\s
^
_J
CM
B
/n
5
<>
0
ft
//s
^/
//.
(A
o
T-
i««a
U.
Q.
,
i r*
ff
IL
-
-
-
X
B) N=816
1 ZO-
04 n A .
TOO
O)
CB
£ "" o ft
g - 80
« 0
0 2 so-
(0 O 0~
o «
2* E 40-
(B «^ 4 U
>
20
0 ~\
'i
\
_
B
^
3
j
3
B
e
f
fS
j-gr
C
1
3
j
w
^
Z3
i
(0
(0
1
^^
C
r
c
E
CO
1
3
>
>
J
i
c
f
^
\
T-
5
=)
N
x^
v^»
O
T
r/7\
\//A
CM
CO
1
0
n
^
rfffffi
U.
Q.
1 1
9
JL
B
1
I^i] LI"|
-
-
-
-
.
O
X
Figure 20. Range of values for average daily flow (DAYAVE) for 10 clusters for
a) "best" sites and b) "all" sites. Symbols and abbreviations same as
in Figure 7.
126
-------
A) N=420
9 -
*S A -
c 4
cc ^
t* A _
1 '*
c >
S E 2^
< o
e
CO
-J
H
i 1
o
^
[__, i
1
in
o
I
^-H
±
^-
DC
(0
C
p
\
S^V
«S
N
DC
(0
)
^
y
\
S
^.i
i
:
:
:
:
:
;
i
i ]
;rx/j
* .
J>*
w
0
~r^
u.
flL
-^
L^Jj
DC
r ^ ^
U.
-
z
B) N=816
o
3
CC
<
s
Figure 21, Range of values for mean annual runoff (MAR) for 10 clusters for a)
"best" sites and b) "all" sites. Symbols and abbreviations same as in
Figure 7.
127
-------
A) N=420
- _ en
Q. O W
Z U. rr u.
W Q. ± =
B) N=816
364
og
*. "O
« o
± °
u- E
273--E
182
(0 (0
Figure 22. Range of values for first day of the flood period (FLODTIME) for 10
clusters for a) "best" sites, and b) "all" sites. Symbols and
abbreviations same as in Figure 7.
128
-------
Figure 23. Maps showing geographical distribution of 420 "best" sites for a) Perennial
Runoff (PR), b) GW (Stable Groundwater), c) SS (Superstate
Groundwater), d) Snow+Rain (SR1 and SR2), e) Snowmelt (SN), f)
Perennial Flashy (PF), and g) Intermittent Runoff (IR), Intermittent
Flashy (IF), and Harsh Intermittent (HI) streams.
129
-------
STATES / CLUS420 / 'PR'
CLUS420 : CLUSNAME - 'PR' -> 209 of 420
MUec
1000
-------
STATES / CLUS420 / 'GW
CLUS420 : CLUSNAME - 'GW -> 55 of 420
MUot
1000
-------
STATES / CLUS420 / 'SS'
CLUS420 : CLUSNAME « 'SS' > 17 of 420
Mllas
1000
-------
STATES / CLUS420 / /SR1YSR2*
ro
ro
Mllat
1000
CLUS420 : CUUSNAME - 'SRV > 27 of 420
CLUS420 : CLUSNAME = 'SR2' -> 29 of 420
-------
STATES / CLUS420 / 'SIM'
CLUS420 : CLUSNAME ° 'SN' -> 22 of 420
M1I0B
1000
-------
STATES / CLUS420 / 'PP
CLUS420 : CLUSNAME » 'PF' -> 24 of 420
1000
-------
STATES / CLUS420 / ,'IRVIFVHI'
Miles
/ l ~\..
) ''*;»
'- 20 of 420
1000 A CLUS420 : CLUSNAME = 'IF' -> 10 of 420
0 CLUS420 : CLUSNAME = 'HI' -> 7 of 420
-------
Figure 24. Maps showing geographical distribution of "all" 816 sites for a) Perennial
Runoff (PR), b) GW (Stable Groundwater), c) SS (Superstate
Groundwater), d) Snow+Rain , e) Snowmelt (SN1 and SN2), f) Perennial
Flashy (PF), and g) Intermittent Runoff (IR), Intermittent Flashy (IF), and
Harsh Intermittent (HI) streams.
137
-------
STATES / CLUS816 / 'PR'
r
CLUS816 : CLUSNAME = 'PR' -> 384 of 816
MllB>
1000
-------
STATES / CLUS816 / 'GW
-
-
CLUS816 : CLUSNAME » 'GW -> 84 of 816
MIlM
1000
-------
STATES / CLUS816 / 'SS'
r/
"»-
' 'V
r- - -\ "- -,-..
'_ i /<-^,
(V f!"1'
r
; ; , L vl
I V 7 L J
it \ /
( \ [
\ \ i - -..../.
i \ /-' / - - ---
-N V /
^ /..
"" -"-J- '\ '
\
i }
J
\
<^
vfr'^ ; '! .W
* v-rAf jr
- 23 of 816
1000
-------
STATES / CLUS816 / 'SRf
&&> :&£ *--
/r.-A T
--.. i A .T0> ^
i?)
CLUS816 : CLUSNAME = 'SR' -> 101 of 816
Mile*
1000
-------
STATES / CLUS816 / /SNT/SN2'
i /" ,>;;:: 3,
,;.'-- ,-; *
Miles
1000
+ CLUS816 : CLUSNAME
A CLUS816 : CLUSNAME
'SN1' -> 48 of 816
'SN2' -> 12 of 816
-------
STATES / CLUS816 / 'PF'
V \
\,
'- '(
\J
Mile*
1000
CLUS816 : CLUSNAME ° ''PF' -> 54 of 816
-------
STATES / CLUS816 / .'IRYIFYHI
Miles
1000
CLUS816 : CLUSNAME = 'IR' -> 49 of 816
CLUS816 : CLUSNAME > 'IF' -> 21 of 816
; CLUSNAME = 'HI' -> 40 of 816
-------
Monthly Predictability
Daily Predictability
oeeooeoo
PRi
PR2'
GW1*
GW2*
SS '
SR1"
SR2'
SN '
PF "
IR '
IF '
HI *
ofo c
j3S2k
Bai
i o
-1I
4-4
i^Ff.
Seasonal Predictability
Weekly Predictability
Figure 25. Range of values for Colwell's index of predictability calculated for
118 sites in 12 clusters for each of four different time steps (daily,
weekly, monthly, seasonally) over a common 36-yr period. Each
box encloses 50% of observed values (median = horizontal line).
Observed range is contained within upper and lower bars except for
extreme outliers (circles). Stream type abbreviations are provided
in Table 8.
145
-------
>.
~ 75-
3
2 70-
o 70
1 65-
a
SN
o
-^
** J- *"
1 Q 1
^
, X X X X
i X X X X
* t t t
. V X V V
o
J
iim
1-
.
1
90
fl°
70
60
50:
40
SR1
T
Day Weak Month Season
Day Wa«k Month Season
«Q-
SR2
T
. i
T
J.
/ f * *
\ \ X \ '
\ ^ X \ '
* / * X
U , 1
T*
mm.
L
75-
>>
Tfl.
A
-.
z 8$-
^ ea
b
Q.
GW
1 I II
!
1 T _
111
T ,V,V " %%%!
J fv,v -t
I 0
0
± \ a. L , _
Day Waak Month Saaaon
Day Waak Month Saaaon
PR1
Day Waak Month Saaaon
60
55
r so
PF
| 40
I 35
30
±
f
Day Waak Month Saaaon
Figure 26. Range of values for Colwetl's index of predictability calculated for
four different time steps (daily, weekly, monthly, seasonally) for
six stream groups. Interpretation as in Figure 25. Stream type
abbreviations are given in Table 8.
146
-------
PR2
75
GW2
A
O
Waak Month Saaaon
45
Week Month Saaaon
SO
W«ak Month Season
55
SO-
>.
I «S
| 40.
! 35
a.
30
25
IR
Day Weak Month Saason
.IF
£
0
9O-
o
0
1
i
1
o
T
1
~\%;
:
I 1
;
:
O
o
0
o
T
'///S//tA
YMWfo
HI
Day Waak Month Saaaon
= 55-
so
Q
5 45-
30-
T
'
i
': 1
1
f
}
nr
i 1
T I
'
I \ \ \ *
X X X X
J.
w%%&-
i '
Day
Waak Month Saaaon
Figure 27. Range of values for Colwell's index of predictability calculated for
four different time steps (daily, weekly, monthly, seasonally) for
six stream groups. Interpretation as in Figure 25. Stream type
abbreviations are given in Table 8.
147
-------
«- CM i- CM '* 'CM
Figure 28. Range of values for 12 stream types over a common 36-yr record for
A) proportion of matches for maximum daily flow and maximum
monthly average, and B) rank correlation coefficient between
maximum daily flows and annual maximum flows. Interpretation as
in Figure 25.
148
-------
0)
£
"3
o
a
(A
6
z
20
Arkansas
RA2 > 0.591
No. Samples
Minnesota
1 o
(ft
£
o
§.
CO
6
CO
2
o
o
n.
CO
o
^m
4Q
30-
20-
1
M\
60
50-
40-
'
30-
20-
10-
i
R*2 0.259
9 *
* 9
1
No. Samples
Wisconsin
R*2 - 0.564
^
m
0 ^9 9
*
* * * *
i «*
No. Samples
100
Figure 29. Association between number of samples and log-in of number of
species for AR, MN, Wl sites.
149
-------
REGIONAL STREAM FLOW I FISH ASSEMBLAGE MAP
e
Variable (16 stations)
Stable (10 stations)
5415000"
Very Stable (8 stations) 0
Km
400
Figure 30. Geographical location of 34 WIMN sites, coded by stream gauge
number. Symbols describe similar sites based on functional
description of fish communities (see text for details).
150
-------
WIMN (34 Sites)
60
50-
a
a>
a
o>
A
40-
30-
20
RA2 » 0.385
1 0
100
Number of Samples
Figure 31. Relationship between number of samples and number of species
collected for 34 WIMN sites.
151
-------
«
£
o
*
40
30
20-
Minnesota
10
R*2 » 0.001
100
1000
Area (kmA2)
10000
60
Wisconsin
so
40 H
20
10 H
o
RA2 0.079
10 100 1000 10000 100000
Area (kmA2)
Figure 32. Relationship between number of species and catchment area for all
22 MN and 44 Wl candidate sites.
152
-------
2
"o
a
CO
e
WIMN
60
50-
40-
30-
20
RA2 » 0.004
100
1000
Area (km*2)
10000
Figure 33. Relationship between number of species and catchment area for 34
WIMN sites.
153
-------
LSR
SOS
BUS
SRL
QBS
9.
NHS
SRL
BNT
4 clusters
(2 levels of division)
406
408
542
3700 408C
5200 5423
3500 542*
5543
WAE
SPS
500 407
ooo 408(
1000 408
830 538
PMK
J500 5332
)000 5333
000 5367
1000 536B
5394
5397
500 5069
500 5293
500 5300
000 53 1 1
500 5315
500 5316
5432
NOP
000 53 i:
000 5317
000 5374
400
000
500
500
500 5401
000 54 1!
000 54 lc.
5431
>500 5379!
'500 5383(
>000 54|4(
JOOO
BLC
>00
)00
)00
8 clusters
(3 levels or division)
Figure 34. TWINSPAN dendrogram generated for 34 WIMN sites from the binary taxonomic data. "Indicator"
species that discriminate among-group differences at each branch point are shown with 3-letter
abbreviations (see Table 16 for species names). Two, 4, or 8 clusters can be identified for 1, 2,
and 3 levels of division, respectively. Numbers in circles indicate four TWINSPAN groups for 2-
level division.
-------
REGIONAL STREAM FLOW I FISH ASSEMBLAGE MAP
TWIN 1
Q TWIN 2
[] TWIN 3
TWIN 4
5415000
Km
500
830
400
Figure 35. Geographical location of 34 WIMN sites grouped according to
affiliation defined in four-group TWINSPAN classification.
155
-------
M
*X
O
Q
300
200-
100 Q
+ TWIN1
Q TWIN2
O TWIN3
A TWIN4
100 200
DCA Axis 1
300
200
CO
SL
°5
o
o
100
* D»oTa*
O a
OO
8-r-
+ TWIN1
a TWIN2
O TWIN3
A TWIN4
100 200
DCA Axis 1
300
Figure 36. Positions of individual WIMN sites plotted according to scores on DCA
axes generated from binary taxonomic data. Sites are coded by four
TWINSPAN groups indicated in Figure 34.
156
-------
CM
CO
'5
O
O
600
400
200
NHS
(RRH)
(MUE)
X K 3
(SAB, K x" "BMSK **» « $ °*_ <*ND)
(SNG) *« **x aK^UBNT x
1 ' xx »x« xOxx xx x
pp»x,SDS "
-200-
-400
-300
x x
X
*~ ««
K*
" ^""
X M
w (SRS)
(BTM)
-100
100
300
500
DCA Axis 1
Figure 37. Positions of 106 species plotted according to scores on first two DCA axes generated from binary
taxonomic data. Species abbreviations not in parentheses are indicator species (see TWINSPAN
dendogram, Fig. 34). Species abbreviations are explained in Table 16.
-------
1-
4
6
3
7
b
2
3
0
S
2
3
5
b
1
5
0
b
1
b
0
b
9
3
0
5
0
0
0
s
1
1
4
5
i
7
C
*
8
J
0
4
8
6
5
2
4
C
4
3
8
b
1
6
5
5
7
9
c
b
e
3
0
4 5
7 3
8 2
5 5
b
3
2
5
b
6
7
5
5
9
4
5
S
6
B
0
b
8
1
0
5
9
7
5
4
8
0
0
b
3
3
0
4
B
5
2
b
1
3
5
KUU
5
3
2
5
f J
b
6
9
0
5
7
4
0
5
0
6
5
5
1
3
5
-1
5
1
4
0
OOOOOOOOOC003COOOOOOOOOOOOOOOOOOOO
OOOOOOCOOOCOCCCOOQOOOOOOOOOOOOOOOO
0.45 +
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
ixxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
0.4 +XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
ixxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
0.35 +XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
S IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
e IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
m Ixxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
1 0.3 +XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
t IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
a Ixxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
r IXXXXXXXXXXXXXXXXXXXXXXXXXXXX.XXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
t 0.25 +XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
1 IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
a IXXXXXXXXXXXXXXXXXXXXXXXX.YXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
1 IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
ixxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
R 0.2 +XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
S IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
q IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
u ixxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
a 0.15 +XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
r IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
e Ixxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
d IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXX
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXX
o.i +xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxx
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXX
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXX
IXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXXXXXXXX
Ixxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxx
0.05 +xxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxx xxxxxxxxxxxxx
IXXXXXXXXXXXXX XXXXXXXXXX XXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXXXXXXXX XXXXXXXXXXXXX XXXXXXXXXXXXX
Ixxxxxxxxxxxxx xxxxxxxxxx xxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxx xxxxxxxxxxxxx
ixxxxxxx xxxx xxxxxxxxxx xxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxx xxxxxxxxxxxxx
IXXXXXXX . . XXXXXXXXXX XXXXXXXXXX XXXXXXX XXXXXXXXXXXXX XXXXXXX XXXX XXXXXXX XXXX XXXXXXX
o +. xxxx . . xxxx . . . xxxxxxx . . . xxxx xxxxxxx xxxx . xxxx xxxx .
Figure 38. Hierarchical dendrogram generated by Ward's method using
the 36 site x 25 functional attribute matrix. Group 1 is
well-separated from Groups 2 + 3 (semi-partial r-squared
distance measure = ca. 0.45), while Group 2 separates from
Group 3 at a much shorter distance measure of ca. 0.12.
158
-------
1 0
CO
0)
55
o
DID Variable (N.16)
0 Stable (N=18)
DAYCV+ . 4
ZERODAY+
FLDFREQ+
3 - 2 - 1 0 1 2
Canonical Variable 1
DAYPRED-
BFI +
Figure 39. Location of 34 individual WIMN sites with respect to scores on the
first canonical variable. Sites are coded according to membership in
either ol two groups, Variable or Stable sites. Hydrologic variables
that have significant correlations with the canonical variable are
indicated at either end of the horizontal axis..
159
-------
X
0
o
£
5
2.
"5
0)
a
00
0.6 :
.
0.5-
(
0.4-
0.3-
*
0^-
0.1 -
n
A
B
^k
R ^
A O O
n
-jr
^J^^
n
^« A ^
20 30 40 50 60 70 8
A Variable Sites
D Stable Sites
Predictability of
Daily Flow
Figure 40. Locations of 34 individual WIMN sites in bivariate space defined by
the hydrologic variables DAYPRED and BFI. Sites are coded by
affiliation in one of two groups.
160
-------
i/niuv T
To
.2
c .0
O CO
C 'Z
CO CO
0 >
DAYPRED +
a
2-
1
0-
-1
-2-
-3-
D
A D
@rj
_ LJ
A - A 1 1 O
A A
n
A ^ A ' 0 D
A 0%
Q
O
"1 ' 1 " 1 ^
- 4 - 2 0 2 4
DAYCV* Canonical DAYPRED +
ZERODAY+ Canonical BF| +
FLDFRQ+ Variable 1
A Variable
D Mod. Stable
O Very Stable
Figure 41. Locations of 34 individual WIMN sites based on their scores on the first two canonical variables in
the 3-cluster case. Original hydrologic variables that are correlated with the canonical variables
(see Table 20) are indicated by their abbreviations.
-------
w
o
0
g 50-
"5
o
01
i «
o
a
(0
| 30-
3
z
9*1 -
D
a a
A A
A.
A
A*
A4xi n D
1 A a
DA
fA
D A
D Variable
A Stable
10 20 30 40
Number of Samples
50
Figure 42. Relationship between number of species and number of samples
collected for all 34 WIMN sites, coded by cluster affiliation (2-
cluster case). For Variable sites, r2 = 0.43, and for Stable sites,
r2= 0.29.
162
-------
a
CO
E
3
o
a
CD
E
3
50-
*1
A
A A
P D
DD
n n
T r"?^
A Variable
U Stable
100 1000 10000
Area (kmA2)
so-
40-
30-
n
CD
A A
Br>l A
*° n A
M^A *' A
S Variable
Stable
0 10 20 30 40
Mean Daily Flow (mA3/sec)
0)
£
U
o
a
OT
a>
A
50-
40-
30-
A
A A
0
DD
A AA CPjDA
A A U
^
A* n ^D en
.. r~k. . .A _n , ^
a
n
A Variable
Stable
0.0 0.1 0.2 0.3 0.4
Mean Annual Runoff (m/yr)
Figure 43. Relationship between number of species and three measures of
catchment area collected for all 34 WIMN sites, coded by cluster
affiliation (2-chjster case).
163
-------
V
0)
a
CO
^
k
Numbe
60-
so-
1
40-
'
A
A A
D a
o
a
p «'*:°*
A A (j-|
A Variable
D Mod. Stable
O Very Stable
100
1000
Area (kmA2)
10000
WJ-
1 50-
o
9
a.
«- 40
0 W
».
0>
JO
1 30-
Z
>n .
A
AA
^^
a a
n°
A Pi O
3* °
*->o r\o o
T~r* O
A Variable
D Mod. Stable
O Very Stable
10 20 30
Mean Daily Flow
(mA3/sec)
40
Figure 44. Relationship between number of species and two measures of
catchment area collected for all 34 WIMN sites, coded by cluster
affiliation (3-cluster case).
164
-------
0)
c
o
o
o
o s
h.
d>
A
E
3
z
50
40-
30-
20-
A
A Variable
D Stable
100
1000
Area (kmA2)
10000
Figure 45. Relationship between sampling intensity and catchment area for 34
WIMN sites, coded by cluster affiliation (2-jcluster case).
165
-------
Variable Sites
Stable Sites
Hydrological Characteristics
Fluctuating Baseflow
Low Predictability
High Variability
Occasional Zeroflow
Frequent Flooding
Stable Baseflow
High Predictability
Low Variability
No Zeroriow
Less Frequent Flooding
Ecological Characteristics
Increasing Trophic Generalization
increasing Small Stream
ana wtae-ranging Fishes
Increasing Slow Velocity Species |-
Decreasing Body Elongation
increasing Fine Sediment Association
j increasing Silt Tolerance I
Black Bullhead
Yellow Bullhead
Yellow Perch
Golden Shiner
Pumpkinseed
Northern Hogsucker
Rosyface Shiner
Longnose Dace
Shorthead Redhorse
Smallmouth Bass
Figure 46. Conceptual summary graph of relationships between hydrological and
ecological characteristics for the 34 WIMN sites. Species that occur
almost exclusively at one end of the hydrological-ecological gradient
are indicated at the bottom of the figure.
166
-------
u
CB
o -
~
3_
2_
( -r '
L , . ,_
-r
-
-
Unstable Stable Unstable Stable
Swim Factor Shape Factor
o
CO
u.
o -
5-
2_
1
1 1
1
'
1
r^n
TT
~r
-IVvVM,
^> *j/> v
E^3
UN MS VS
Swim Factor
UN MS VS
Shape Factor
Figure 47. Mean group scores (+ 2se) for two body morphology attributes for
the 2-cluster case (upper panel) and 3-cluster case (lower panel).
For each cluster, box encloses 50% of observed values, with median
value indicated by horizontal line within box. Upper and lower bars
enclose upper and lower 25% of observations, respectively.
167
-------
c
o
u
Q.
O
0.4
0.3-
0.2-
0.1 -
0.0
Variable
Stable
HD OM Gl SI Bl PI PL PA
Trophic Guild
c
o
o
Q.
o
0.4
0.3
0.2-
0.1 -
0.0
Variable
StflhU
Very Stable
oo
VD
HD OM Gl SI Bl PI
Trophic Guild
PL
PA
Figure 48. Mean group proportions (+ 2se) for 8 trophic categories attributes for the 2-cluster case (upper
panel) and 3-cluster case (lower panel). HD = herbivore-detritivore, OM = omnivore, Gl =
generalist invertivore, SI = surface-feeding invertivore, Bl = benthic invertivore, PI =
piscivore-invertivore, PL = planktivore, PA = parasite.
-------
c
o
o
Q.
O
0.6
0.5-
0.4-
0.3-
0.2
0.1
0.0
rubble sand silt general
Substrate
D Variable
B Stable
0.6
e
o
o
a
o
0.5
0.4
0.3
0.2 H
0.1
0.0
rubble
sand silt
Substrate
general
D Variable
0 Stable
Very Stable
Figure 49. Mean group proportions (+ 2se) for 4 substrate preference
categories for the 2-cluster case (upper panel) and 3-cluster case
(lower panel).
169
-------
e
o
o
a
o
0.5
0.4-
0.3
0.2-
0.1 -
0.0
I
T
D Variable
B Stable
Small Med-Lg Small-Lg Lake
Stream Size
o
a.
o
0.5
0.4
0.3
0.2-
0.1-
0.0
Variable
Stable
Very Stable
Small Med-Lg Small-Lg Lake
Stream Size
Figure 50. Mean group proportions (+ 2se) for 4 stream size preference
categories for the 2-cluster case (upper panel) and 3-cluster case
(lower panel).
170
-------
e
o
o
a
o
a
o
a
o
0.6
0.5
0.4
04
0.2
0.1
0.0
G Variable
Stable
Fast Medium Slow General
Flow Habitat
D Variable
Q Mod. Stable
B Very Stable
Fast Medium Slow General
Flow Habitat
Figure 51. Mean group proportions (+ 2se) for 4 flow habitat categories for the
2-cluster case (upper panel) and 3-cluster case (lower panel).
171
-------
0.5
0.4 J
c
2, 0.3-
o
a
o
o
a
o
i_
&
0.2-
0.1 -
0.0
T
tolerant moderate intolerant
Tolerance
tolerant moderate intolerant
Tolerance
D Variable
Stable
Variable
MotLStabto
Vary Stabl*
Figure 52. Mean group proportions (+ 2se) for 3 tolerance categories for the
2-cluster case (upper panel) and 3-cluster case (lower panel).
172
------- |