Content uploaded by Michael Miller
Author content
All content in this area was uploaded by Michael Miller on Sep 10, 2020
Content may be subject to copyright.
Environ Monit Assess (2009) 150:75–89
DOI 10.1007/s10661-008-0679-6
Assessment of wadeable stream resources
in the driftless area ecoregion in Western Wisconsin
using a probabilistic sampling design
Michael A. Miller ·Alison C. C. Colby ·
Paul D. Kanehl ·Karen Blocksom
Received: 17 March 2008 / Accepted: 31 March 2008 / Published online: 5 December 2008
© US Government 2008
Abstract The Wisconsin Department of Natural
Resources (WDNR), with support from the U.S.
EPA, conducted an assessment of wadeable
streams in the Driftless Area ecoregion in western
Wisconsin using a probabilistic sampling design.
This ecoregion encompasses 20% of Wisconsin’s
land area and contains 8,800 miles of perennial
streams. Randomly-selected stream sites (n=60)
equally distributed among stream orders 1–4 were
sampled. Watershed land use, riparian and in-
stream habitat, water chemistry, macroinverte-
brate, and fish assemblage data were collected
at each true random site and an associated
“modified-random” site on each stream that was
accessed via a road crossing nearest to the true
random site. Targeted least-disturbed reference
sites (n=22) were also sampled to develop ref-
M. A. Miller (B)·A. C. C. Colby
Bureau of Fisheries Management,
Wisconsin Department of Natural Resources,
101 S. Webster St., Madison, WI 53703, USA
e-mail: Michaela.miller@wisconsin.gov
P. D. Kanehl
Bureau of Integrated Science Services,
Wisconsin Department of Natural Resources,
Madison, WI, USA
K. Blocksom
Office of Research and Development,
U. S. Environmental Protection Agency,
Cincinnati, OH, USA
erence conditions for various physical, chemical,
and biological measures. Cumulative distribution
function plots of various measures collected at the
true random sites evaluated with reference con-
dition thresholds, indicate that high proportions
of the random sites (and by inference the entire
Driftless Area wadeable stream population) show
some level of degradation. Study results show
no statistically significant differences between the
true random and modified-random sample sites
for any of the nine physical habitat, 11 water
chemistry, seven macroinvertebrate, or eight fish
metrics analyzed. In Wisconsin’s Driftless Area,
79% of wadeable stream lengths were accessible
via road crossings. While further evaluation of
the statistical rigor of using a modified-random
sampling design is warranted, sampling randomly-
selected stream sites accessed via the nearest road
crossing may provide a more economical way to
apply probabilistic sampling in stream monitoring
programs.
Keywords Stream assessment ·Driftless
ecoregion ·Probability survey design ·
Biological assessment
Introduction
To date, stream monitoring conducted by the
Wisconsin Department of Natural Resources
76 Environ Monit Assess (2009) 150:75–89
(WDNR) has primarily been targeted sampling to
provide site-specific information for local resource
management. Data are often collected from either
degraded streams affected by anthropogenic ac-
tivities, or from high quality streams where game-
fish management efforts are being evaluated. This
resulting data set can be biased if used for making
inferences about overall broad-scale conditions
of stream resources. Spatial clustering of moni-
toring efforts on a relatively small proportion of
Wisconsin’s 22,000 perennial streams and a focus
on larger streams that support adult gamefish,
also limit the ability to make meaningful state-
ments about the state’s entire wadeable stream
population.
Beginning in 2003, the WDNR, with support
from the U.S. EPA’s Regional Environmental
Monitoring and Assessment Program (R-EMAP),
conducted an assessment of wadeable streams in
the Driftless Area ecoregion in western Wisconsin
using a probabilistic sampling design. The pri-
mary objectives of the study were to: (1) Use
a statistically defensible sample survey design to
assess the physical, chemical, and biological con-
ditions of the entire wadeable stream popula-
tion in the Driftless Area ecoregion; (2) Sample
least-disturbed reference sites to develop numeric
criteria with which to evaluate the condition of
individual and populations of streams; (3) Evalu-
ate whether a modified-random sample survey de-
sign (sampling randomly selected stream reaches
near road-accessible access points) characterizes
individual or populations of streams similarly to
a truly randomized survey sampling design, to
explore whether a more efficient sample survey
design could be developed for WDNR’s stream
monitoring program. Additional study objectives
and more detailed statistical analyses and find-
ings of this study are reported in a U.S. EPA
Office of Research and Development (ORD)
report (Miller et al. 2006).
Methods
Study area
The geographic extent of the R-EMAP study area
is 7,418,000 acres (20.6% of Wisconsin’s total land
area), and encompasses the Western Coulee and
Ridges and the Southwest Savanna ecoregions in
western Wisconsin (Fig. 1). This land cover classi-
fication is based on the United States Department
of Agriculture—Forest Service’s National Hier-
archical Framework of Ecological Units (Keys
et al. 1995). These two ecological units closely
approximate Omernik’s Level III Driftless Area
Ecoregion (Omernik 1987) and are collectively
referred to as the Driftless Area in this report.
Random and modified-random
site selection
U.S. EPA National Health and Environmental
Effects Research Laboratory—Western Ecology
Division (NHEERL—WED) staff in Corvallis,
Oregon used an unequal-probability stratified
random sampling design described by Stevens
and Olsen (2004) and the WDNR’s 1:24,000-scale
Geographic Information System (GIS) statewide
stream hydrography layer, to identify 100 candi-
date sample sites. The final random sample pop-
ulation was comprised of 15 perennial wadeable
stream sites in each Strahler (1957) stream or-
der one through four (n=60). NHEERL also
identified 100 over-sample sites to be used as re-
placements if initial random sites were rejected.
The sample population for each stream order was
weighted to account for the number of stream
miles they represent in the target population.
Each random site was paired with a correspond-
ing “modified-random” sample site on the same
stream, located either upstream or downstream
at the nearest “easy” access point (typically ac-
cessed from a roadway or driveway bridge that
crossed each stream) closest to the true random
site. In this report, random sites are subsequently
referred to as “X” sites and the modified-random
sites are referred to as “B” (bridge) sites. We
tried to locate the B sites sufficient distances
(typically 10 ×the mean stream width (MSW))
away from the roadway stream crossings to avoid
hydraulic influences of bridge abutments, culverts,
or other manmade structures on stream physi-
cal characteristics, or that created artificial fish
habitat. We dropped candidate stream sites from
the sample population if there were intervening
Environ Monit Assess (2009) 150:75–89 77
Fig. 1 Random and least
disturbed reference sites
in the R-EMAP
study area
tributaries between the random and modified-
random sites that resulted in a Strahler stream
order difference of more than one order between
the random and modified-random sites. Candidate
stream sites that were deemed unsampleable for
any of a variety of reasons (e.g. too deep to wade)
were dropped and replaced with sites from the
oversample population.
Reference site selection and criteria
development
Candidate least-disturbed reference sites were
identified based on evaluation of watershed land
ownership (i.e., county, state, or federal lands
tended to have less agricultural or urban land) and
land use, using a statewide GIS land cover data-
base (WDNR 1998). The apriorimethod used
to select these candidate watersheds and least-
disturbed reference stream sites was based on
guidance developed by EPA and others (Hughes
et al. 1986; Gibson et al. 1996). Reconnaissance
of each candidate reference site was conducted
to verify there were no apparent watershed or
riparian land use factors significantly degrading
the site, and to do a cursory evaluation of in-
stream physical habitat conditions prior to desig-
nating it as a reference site. Reference conditions
were developed using the data collected from 22
least-disturbed stream sites located throughout
the study area (Fig. 1).
78 Environ Monit Assess (2009) 150:75–89
Stream site sampling protocols
In-stream and riparian physical habitat charac-
teristics were measured or visually estimated
at each X, B, and reference site following
WDNR standard operating procedures (SOPs)
(http://infotrek.er.usgs.gov/doc/wdnr_biology/
monitoring_protocols_field_forms.html). Stream
assessment reach-lengths for physical habitat and
fish assemblage sampling were 35 ×MSW (Lyons
1992a; Simonson et al. 1993). For streams less
than 2.9 m MSW, the minimum assessment-reach
was 100 m long.
In situ instantaneous measures of water tem-
perature, dissolved oxygen concentration, dis-
solved oxygen percent saturation, conductivity,
and pH were collected at all X, B, and reference
sites during baseflow conditions using electronic
meters. Water column transparency was mea-
sured at all sites with a 1.2 m-long transparency
tube. One laboratory-analyzed water chemistry
grab sample was collected during base flow con-
ditions at each site to measure total phospho-
rus, total dissolved phosphorus, total Kjeldahl
nitrogen, ammonia, and nitrate–nitrite nitrogen
concentrations.
We collected benthic macroinvertebrate sam-
ples at all X, B, and reference stream sites in
the fall of 2003 using a D-frame net with 500 μm
mesh, following WDNR SOPs. One kick sample
was collected from a single riffle (coarse gravel
or cobble substrate) located within the habitat
and fish sampling reach. In the absence of coarse
substrate, riparian vegetation overhanging into
the water, and in-stream snags primarily com-
posed of leaves, grass, or aquatic macrophytes
were sampled. We used a 125+ organism fixed-
count random sub-sampling method in the lab
to process all of the macroinvertebrate samples,
following WDNR SOPs. Guidelines developed by
Hilsenhoff (1987) were followed for determin-
ing which taxa were identified and the level of
taxonomic resolution that was applied to the taxa
in the sub-samples. Lifestages and taxa not in-
cluded in the sub-samples were: adult insects,
empty or sealed Trichoptera cases, Hemiptera,
Coleoptera (non-dryopids), Collembola, Mol-
lusca, Annelida, Decapoda, Nematoda, Nemato-
morpha, Hydracarina, and Turbellaria.
We used WDNR SOPs to sample fish assem-
blages by electrofishing each of the X, B, and
reference sites (Lyons 1992b; Lyons et al. 1996).
All stream habitats within the 35 ×MSW sam-
pling reach were thoroughly sampled, and an ef-
fort was made to capture all fish greater than
25 mm total length. The live fish were identified
to species, measured, counted, and released. Fish
that could not be identified to species in the field
were preserved and subsequently identified using
a taxonomic key (Becker 1983).
Data analytical methods
A total of nine physical habitat, 11 water qual-
ity and water chemistry, seven macroinvertebrate,
and eight fish metrics were analyzed. Prior to
analyses, variables were “weighted” using an as-
signed weight that accounted for the total num-
ber of stream miles represented by the sample
size for each stream order (n=15). Weighted
means and Horvitz–Thompson estimates of stan-
dard deviation (Diaz-Ramos et al. 1996)were
calculated for X and B sites using the PSUR-
VEY.ANALYSIS package (v. 2.7) developed
by EPA’s EMAP program (http://www.epa.gov/
nheerl/arm/analysispages/software.htm). An alpha
level of 0.05 was applied to determine the signif-
icance of all tests reported. Pearson correlation
coefficients were calculated to identify significant
correlations, and paired t-tests incorporating the
weighted means and standard deviations were
used with a Bonferroni adjustment to determine
if statistically significant differences existed be-
tween the X and B sites for any of the variables
collected.
Empirical cumulative distribution function
(CDF) plots of X and B site data were made to
estimate the percentage of the X and B stream
populations that met reference condition thresh-
olds for the various physical, chemical, and bio-
logical measures. The 25th percentile (for those
metrics where lower values indicated lower en-
vironmental quality) or the 75th percentile (for
those metrics where higher values indicated lower
environmental quality) of physical habitat, chem-
ical, and biological measures calculated from
the reference sites’ data were used to develop
the reference condition threshold scores. The
Environ Monit Assess (2009) 150:75–89 79
CDFs and percentage estimates were calculated
using the PSURVEY.ANALYSIS package in R -
language software (Ihaka and Gentleman 1996).
A CDF summarizes the overall distribution of
some variable (x-axis) measured at many random
sites. The probability that any variable within the
target population will be less than or greater than
some specified value can be estimated from a CDF
curve (Sokal and Rohlf 1981). If X and B CDFs
provide equivalent data, then their population
estimates should be similar and their confidence
intervals should overlap. Kolmogorov–Smirnov
two-sample tests were used to determine whether
the X and B probability distributions differed
statistically (Sokal and Rohlf 1981).
Results
Sample size summary
Of the 100 random and 100 over-sample sites
provided by EPA, we rejected a total of 71 can-
didate sites during the initial field reconnaissance.
The final sample population consisted of 58 true
random and their associated ‘modified-random’
bridge sites, and 22 targeted least-disturbed refer-
ence sites. Reasons for rejecting X sites included:
dry stream channel (38 sites), Strahler stream
order changes greater than one between X and
B site pairs (ten sites), sample site was located
in a wetland with no definable stream channel
Fig. 2 Bivariate scatterplots and Pearson correlation coefficients (r) comparing physical habitat measures collected at the
random ‘X’ and associated modified-random ‘ B’sites
80 Environ Monit Assess (2009) 150:75–89
Fig. 3 Bivariate scatterplots and Pearson correlation coefficients (r) comparing water chemistry and quality measures
collected at the random ‘X’ and associated modified-random ‘ B’sites
(six sites), site not wadeable due to beaver dams
or artificial impoundments (five sites), site actu-
ally a Mississippi River backwater channel (five
sites), site access denied by landowner(s) (four
sites), no stream channel located at the sample
site coordinates (two sites), and a 3 m tall barrier
fence prevented access to the stream channel (one
site).
Environ Monit Assess (2009) 150:75–89 81
Random and modified-random sites
comparisons
Pearson correlation coefficients and correspond-
ing p-values comparing the X and B site data
were calculated for nine physical habitat, 11 wa-
ter chemistry, seven macroinvertebrate, and eight
fish assemblage measures (Figs. 2,3,4and 5).
Large correlation coefficients (r) and significant
p-values were observed for all water chemistry,
macroinvertebrate, and fish assemblage measures,
and eight of the nine physical habitat measures. Of
the 35 variables, only the mean pool habitat length
did not show a significant correlation between the
X and B sites. Weighted paired t-tests comparing
the X and B site data did not produce statistically
significant p-values for any of the 35 physical,
chemical, and biological variables evaluated.
The above comparisons do not account for
streams where no nearby road crossing could be
found meeting the Strahler order change criteria.
Based on the ten sites where no road crossing met
the criteria, it is estimated that 21.3% (11.6–31.0
95% CI) or 2,457 miles (1,245–3,669 miles 95%
Fig. 4 Bivariate scatterplots and Pearson correlation coefficients (r) comparing macroinvertebrate assemblage measures
collected at the random ‘X’ and associated modified-random ‘ B’sites
82 Environ Monit Assess (2009) 150:75–89
Fig. 5 Bivariate scatterplots and Pearson correlation coefficients (r) comparing fish assemblage measures collected at the
random ‘X’ and associated modified-random ‘ B’sites
CI) of the target population stream length would
be excluded. All excluded sites occurred on first
order streams.
Probability estimates and evaluation
of population distributions
Empirical CDFs were plotted to further evalu-
ate differences between the X and B sites data,
and to estimate the percentage of the X and B
sample populations that met reference condition
threshold values. The upper 75th percentiles cal-
culated from least disturbed reference sites data
were used as the thresholds for the percentage of
sand, silt, and clay substrates; the mean depth of
fine sediments; the width/depth ratio; the mean
distance between bends; mean streambank ero-
sion width; Kjeldahl-N; NH3;NO
3–NO2; total dis-
solved P; total P; Hilsenhoff Biotic Index (HBI)
scores; and the percentage of ‘tolerant’ fish indi-
viduals. The 25th percentile was used as the refer-
ence condition threshold for mean riparian buffer
width; dissolved oxygen percent saturation; dis-
solved oxygen concentration; water transparency;
percent ephemeroptera, plecoptera, trichoptera
(% EPT) individuals in macroinvertebrate sam-
ples; the percentages of stenothermal and top car-
nivore fish individuals present; and the fish index
of biotic integrity (IBI) score.
The X and B site cumulative distribution func-
tions and their confidence intervals overlap for all
of the physical, chemical, and biological measures
Environ Monit Assess (2009) 150:75–89 83
Fig. 6 Cumulative
distribution function
curves of physical habitat
measures collected from
the X and B sites. The
stippled lines represent
95% confidence intervals
around the distribution
plots. The vertical stippled
lines represent the
reference condition
threshold values
analyzed, indicating the population estimates of
these measures are not significantly different be-
tween the random and modified-random sites
(Figs. 6,7,8and 9). The results of Kolmogorov–
Smirnov tests (Dmax values) also indicate no sig-
nificant differences (all Dmax p-values >0.05).
Effects of geographical distances between
random and modified-random sites
Excluding three stream sites where the X and
B sampling reaches overlapped, the average dis-
tance from the X sites to the B sites was 701 m,
with a minimum distance of 106 m and a maximum
distance of 2,283 m. We investigated the influ-
ence of the linear distances between the X to
B sampling site pairs on the following measures:
width/depth ratio, percentage of fine substrate,
percent EPT genera present, HBI score, number
of fish captured, fish IBI score, fish species rich-
ness, total dissolved phosphorus, and NH3;since
based on differences between the X and reference
stream sites data, these measures are shown to be
strong predictors of biotic potential or sensitive
measures of stream quality. Spearman correlation
coefficients (rs) values indicate no significant re-
lationships between the distance between the X
and associated B assessment sites and the strength
of the X and B site correlations for any of the
84 Environ Monit Assess (2009) 150:75–89
Fig. 7 Cumulative distribution function curves of water
chemistry and quality measures collected from the X and B
sites. The stippled lines represent 95% confidence intervals
around the distribution plots. The vertical stippled lines
represent the reference condition threshold values
physical, chemical, or biological measures re-
ported (Fig. 10).
Discussion
Characterization of stream resources
in the driftless area
A primary objective of this study was to charac-
terize the physical, chemical, and biological con-
ditions of stream resources in the Driftless Area
ecoregion using the EMAP probabilistic sampling
design. Since previous sampling efforts in the
Driftless Area and state have primarily targeted
either highly degraded or high quality streams
to address stream or site-specific management
needs, these data cannot be used to character-
ize the overall quality of stream populations in
an objective, statistically-defensible fashion. This
R-EMAP study is the first broad-scale assess-
ment of stream resources by the WDNR that has
Environ Monit Assess (2009) 150:75–89 85
Fig. 8 Cumulative
distribution function
curves of
macroinvertebrate
assemblage measures
collected from the X and
Bsites.Thestippled lines
represent 95% confidence
intervals around the
distribution plots. The
vertical stippled lines
represent the reference
condition threshold
values
produced data of known statistical quality and
that applied objective numeric criteria to assess
the physical, chemical, and biological conditions
of individual or populations of streams. This char-
acterization of stream resource conditions will
be used to improve the rigor of federal water
Fig. 9 Cumulative
distribution function
curves of fish assemblage
measures collected from
the X and B sites. The
stippled lines represent
95% confidence intervals
around the distribution
plots. The vertical stippled
lines represent the
reference condition
threshold values
86 Environ Monit Assess (2009) 150:75–89
Fig. 10 Scatterplots showing relationships between the
distance between the X and B sites and the absolute value
of the differences in physical, chemical, and biological mea-
sures collected at the X and B sites. Spearman correlation
coefficients are provided
resource quality (e.g. EPA 305b) reporting, and
provide information to WDNR administration
to direct and evaluate resource management
activities.
Study results indicate that between 77 and 100
percentage of the sample population (and by in-
ference the entire Driftless Area stream popula-
tion) show some degree of degradation depending
upon the specific physical, chemical, or biological
measure evaluated. The primary factor impact-
ing stream quality in the Driftless Area is land
use. Significant amounts of topsoil, manure, and
chemical fertilizers enter streams via surface run-
off; and livestock pastured in the valley floors also
result in trampled stream banks, and additional
manure and soil entering streams.
While setting reference condition thresholds
(management goals) at the 25th or 75th percentile
of the reference condition data is a scientifi-
cally defensible approach, setting resource man-
agement goals should perhaps be tempered with
natural resource management agency goals and
societal decisions. The study findings and the
methods used to develop reference conditions
Environ Monit Assess (2009) 150:75–89 87
will stimulate further discussion within WDNR
on which individual or combinations of measures
should be used to assess stream quality, and what
numeric criteria thresholds for these various para-
meters should be applied.
Selection of reference sites and application
of reference condition data
While macroinvertebrate and fish assemblage in-
dices are increasingly used to objectively assess
stream resources in Wisconsin, development of
least disturbed reference conditions for stream
physical habitat and water chemistry measures
provides additional objective numeric criteria
with which to evaluate the condition of individ-
ual or populations of streams. In addition, the
macroinvertebrate and fish assemblage reference
condition data allowed calibration of these bi-
ological indices for the Driftless Area streams,
allowing increased accuracy of the individual
and stream population assessments in this study.
These findings are influencing the WDNR’s col-
lective understanding of key response and ex-
planatory variables that can be used to either
evaluate or provide insights into understanding
what factors most strongly influence the quality of
Wisconsin’s stream resources.
Differences between random
and modified-random sample sites
Another primary objective of this study was
to evaluate whether sampling randomly selected
stream segments at sites accessible from road
crossings would provide results similar to a truly
randomized sampling design. While study find-
ings indicate that none of the 35 physical, chem-
ical, or biological parameters evaluated were
significantly different between the random and
modified-random sampling sites, lack of differ-
ences for several parameters may be artifacts of
field sampling methods or limitations of labora-
tory analytical methods or field equipment. For
example, following WDNR SOPs, riparian buffer
widths are measured along transect lines laterally
from the stream within the first 10 m from the
water’s edge. Sites with buffer widths greater than
10 m are reported as 10 meters, which limited the
true buffer width differences between some of the
X and B site-pairs. Similarly, water transparency
is measured using a 1.2 m-long transparency tube,
but transparency exceeded 1.2 m at some sites,
thereby reducing the true differences between
some X and B site-pairs.
Because mixing in lotic systems results in rel-
ative homogeneity of water chemistry, few dif-
ferences would be expected to exist for these
measures between X and B sites. However, it is
possible that tributaries, pollutant point sources,
or groundwater inputs intervening between the X
and B sampling sites can change the concentration
of chemical parameters and dissolved oxygen, or
water temperature.
None of the seven macroinvertebrate metrics
analyzed were significantly different between and
X and B sites. Most macroinvertebrate taxa are
relatively sessile organisms and have been shown
to be more strongly influenced by local habitat or
reach-scale environmental factors than fish which
are thought to respond to both localized and
watershed-scale influences (Barbour et al. 1999;
Lammert and Allan 1999). Given that macroin-
vertebrates are thought to respond more strongly
to site-specific or reach-scale influences, the lack
of X and B site differences for macroinvertebrate
measures may be a more sensitive test of whether
spatial bias is being induced by the modified-
random sampling design relative to fish assem-
blage data.
Probabilistic sampling design issues
Fundamental principles of probabilistic sampling
are that every population element in the target
population has a known (and non-zero) proba-
bility of being sampled, and that it is critically
important to rigorously define both the target
population and the elements of which it is com-
prised (Cochran 1977). A major objective of the
Wisconsin R-EMAP study (Miller et al. 2006)was
to estimate the number of stream miles in the
study area that were meeting physical, chemical,
or biological, reference condition criteria. There-
fore, a continuous sampling design was applied,
where assessment measurements were taken at or
in the vicinity of randomly selected points (Larsen
88 Environ Monit Assess (2009) 150:75–89
1997). For stream population elements that
are spatially-static (e.g. stream physical habitat),
sampling at the modified-random sites changes
the target population from the total length of
all wadeable streams to the total length of all
streams that are within some specified distance
from road crossings. For spatially-dynamic popu-
lation elements (water chemistry parameters, fish,
or macroinvertebrates that are strongly influenced
by upstream land use and the ambient conditions
of flowing water) it is less clear what the spatial
boundaries of these population elements are.
The mean distance between the random and
modified-random assessment sites in this study
was approximately 700 m (0.44 miles). Although
no significant differences were observed between
random and modified-random assessment reaches
in this study, there may be a greater potential
for differences in other regions in Wisconsin or
elsewhere with significantly lower road density,
greater land cover or land use heterogeneity,
higher potential for intervening point sources of
pollution or groundwater discharge, or greater
topographic relief, than the R-EMAP study area.
For example, in a significant proportion of north-
ern Wisconsin road density is about 30% less than
that of the R-EMAP study area. This increases
the potential distance between random and road-
accessible sites, and may result in greater observed
differences between these sample sites. However
in this case, the northern portion of Wisconsin
is also characterized by more homogeneous land
cover and land use, and lower human and live-
stock population densities than in the R-EMAP
study area, which may result in fewer observed
differences between X and B sites.
It is hoped that the findings of this study will
generate further research and discussion on the
validity of using road accessible stream sampling
sites to characterize stream target populations.
Sample population site selection
The finding that 71 of the randomly selected
stream sites were rejected during the field recon-
naissance effort to obtain a sample population of
60 streams (nearly 120% of the original sample
population) is significant. Random sample survey
designs must include sufficient over-sample pop-
ulations to maintain target sample sizes, given
that significant numbers of random sites are likely
to be rejected. Project planning must also incor-
porate sufficient time for map work to identify
site locations, for contacting land owners, and for
field reconnaissance of assessment sites and sub-
sequent replacement of rejected sites. Only 5% of
the random sites were dropped due to landowner
access denial, which is significantly less than what
the investigators had expected prior to the start of
the study, and was a rationale for proposing sam-
pling at road crossings since all navigable streams
can be legally accessed from public roadways in
Wisconsin.
Implications for WDNR’s wadeable stream
monitoring program
Study results provide valuable insights for im-
proving the efficiency and rigor of the WDNR’s
wadeable stream monitoring program. This R-
EMAP study is the first time the WDNR has con-
ducted a probabilistic stream survey on a broad
geographic scale. The preliminary findings that a
modified-random sampling design appears to in-
duce little bias in the assessment of stream quality
may allow more efficient and cost-effective stream
sampling effort in Wisconsin, but additional study
is needed to more rigorously evaluate the utility
of applying a road-accessible sampling design in
other ecoregions within Wisconsin. Of particular
concern is the exclusion of streams with no nearby
road crossing. Finally, the process of using nu-
meric reference condition criteria to objectively
determine whether individual or populations of
streams are meeting their potential will be applied
statewide which will help reduce WDNR resource
biologists’ and managers’ reliance on subjective,
qualitative, resource evaluations.
Acknowledgements A number of WDNR staff con-
tributed to this study. We thank Sue Acre, Bob Busch,
Bill Ceelen, and Walt Jaeger. In addition, a number of
EPA staff were key contributors to this project and the
authors are very thankful for their support: Dr. Florence
Fulk-National Exposure Research Laboratory, Cincinnati,
OH, and Sarah Lehmann and Edward Hammer of Region
5 helped secure funding and initiate the project. Dr. Tony
Olsen, NHEERL-WED, provided the sample population-
draw and statistical guidance; Dr. Robert Hughes, OR
Environ Monit Assess (2009) 150:75–89 89
State Univ., and Drs. Anthony Olsen, Steve Paulsen, and
Paul Ringold of NHEERL-WED, provided reviews of ear-
lier drafts of the EPA - ORD report (Miller et al. 2006)on
which this report is based. We also thank two anonymous
reviewers of the draft manuscript.
References
Barbour, M. T., Gerritsen, J., Snyder, B. D., & Stribling,
J. B. (1999). Rapid bioassessment protocols for use
in streams and wadeable rivers: Periphyton, benthic
macroinvertebrates, and fish, second edition. EPA 841-
B-99-002. Washington: U. S. Environmental Protec-
tion Agency, Office of Water.
Becker, G. C. (1983). Fishes of Wisconsin. Madison:
University of Wisconsin Press.
Cochran, W. G. (1977). Sampling techniques (3rd ed.). New
York: Wiley.
Diaz-Ramos, S., Stevens, D. L. Jr., & Olsen, A. R.
(1996). EMAP statistical methods manual. EPA/620/R-
96/002. U.S. Environmental Protection Agency, Office
of Research and Development, National Health
Effects and Environmental Research Laboratory,
Western Ecology Division, Corvallis, Oregon.
Gibson, G. R. Jr., Barbour, M. T., Stribling, J. B.,
Gerritsen, J. & Karr, J. R. (1996). Biological crite-
ria: Technical guidance for streams and small rivers.
EPA/822-B-96-001. Washington: U.S. Environmental
Protection Agency, Office of Science and Technology.
Hilsenhoff, W. L. (1987). An improved biotic index for
organic stream pollution. Great Lakes Entomologist,
20, 31–39.
Hughes, R. M., Larsen, D. P., & Omernik, J. M. (1986).
Regional reference sites: A method for assessing
stream potentials. Environmental Management, 10,
629–635.
Ihaka, R., & Gentleman, R. (1996). R: A language for data
analysis and graphics. Journal of Computational and
Graphical Statistics, 5, 239–314.
Keys, J. E. Jr., Carpenter, C. A., Hooks, S. L., Koeneg,
F. G., McNab, W. H.,Russell, W. E., et al. (1995). Eco-
logical units of the Eastern United States—first approx-
imation. Technical publication R8-TP 21. Map (scale
1:3,500,000). Atlanta: U.S. Department of Agriculture,
Forest Service.
Lammert, M., & Allan, J. D. (1999). Assessing biotic
integrity of streams: Effects of scale in measuring the
influence of land use/cover and habitat structure on
fish and macroinvertebrates. Environmental Manage-
ment, 23, 257–270.
Larsen, D. P. (1997). Sample survey design issues for
bioassessment of inland aquatic ecosystems. Human
and Ecological Risk Assessment, 3, 979–991.
Lyons, J. (1992a). The length of stream to sample with
a towed electrofishing unit when fish species rich-
ness is estimated. North American Journal of Fisheries
Management, 12, 198–203.
Lyons, J. (1992b). Using the index of biotic integrity (IBI) to
measure environmental quality in warmwater streams
of Wisconsin. St. Paul: General Technical Report
NC-49. North Central Forest Experiment Station, U.S.
Department of Agriculture.
Lyons, J., Wang, L., & Simonson, T. D. (1996). Devel-
opment and validation of an index of biotic integrity
for coldwater streams in Wisconsin. North American
Journal of Fisheries Management, 16, 241–256.
Miller, M. A., Colby, C. C., & Kanehl, P. D. (2006). Re-
port on the regional environmental monitoring and
assessment program study of wadeable streams in the
driftless area ecoregion in Western Wisconsin. EPA
600/R-06/165.
Omernik, J. M. (1987). Ecoregions of the conterminous
United States. Annals of the Association of American
Geographers, 77, 118–125.
Simonson, T. D., Lyons, J., & Kanehl, P. D. (1993). Guide-
lines for evaluating fish habitat in Wisconsin streams
(36 p.). Gen. Tech. Rep. NC-164. St. Paul, MN: U.S.
Dept. of Ag. Forest Service. North Central Forest
Exp. Station.
Sokal, R. R., & Rohlf, F. J. (1981). Biometry (2nd ed.). New
York: Freeman.
Stevens, D. L. Jr., & Olsen, A. R. (2004). Spatially-
balanced sampling design of natural resources. Journal
of the American Statistical Association, 99, 262–278.
Strahler, A. N. (1957). Quantitative analysis of watershed
geomorphology. Transactions of the American Geo-
physical Union, 38, 913–920.
Wisconsin Department of Natural Resources (1998).
Wisconsin initiative for statewide cooperation on land-
scape analysis and data. Madison, WI 53703.
