 

ON

,-
.)
\

 

I' LIBRARY
Michigan State
University

 

This is to certify that the
thesis entitled

MEASURING ANTHROPOGENIC DISTURBANCES IN A
HYDROGEOMORPHlC-BASED LAKE CLASSIFICATION BUILT
USING FISH ASSEMBLAGES IN 360 NORTH-TEMPERATE LAKES

presented by

BRETT MICHAEL ALGER

has been accepted towards fulﬁllment
of the requirements for the

MS degree in FISHERIES AND WILDLIFE

 

 

MM KE/mmo Ly

/Major Professor's Sjgnature

it: Aeﬂembef 2009

Date

MSU is an Afﬁrmative Action/Equal Opportunity Employer

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K1IPrqlAccaPrelelRC/DateDue.indd

MEASURING ANTHROPOGENIC DISTURBANCES IN A HYDROGEOMORPHIC-
BASED LAKE CLASSIFICATION BUILT USING FISH ASSEMBLAGES IN 360
NORTH-TEMPERATE LAKES

By

Brett Michael Alger

A THESIS
Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of
MASTER OF SCIENCE
Fisheries and Wildlife

2009

ABSTRACT

MEASURING ANTHROPOGENIC DISTURBANCES IN A HYDROGEOMORPHIC-
BASED LAKE CLASSIFICATION BUILT USING FISH ASSEMBLAGES IN 360
NORTH-TEMPERATE LAKES

By
Brett Michael Alger
Hydrogeomorphic features and anthropogenic disturbances affect ﬁsh
assemblages but the relative importance of these variables is inadequately understood.
Quantifying these linkages, across multiple spatial scales, is key to developing lake
ecosystem assessment tools and management plans. Accordingly, I classiﬁed lakes with
similar hydrogeomorphic features and within these lake groups, identiﬁed the most
important anthropogenic inﬂuences on ﬁsh species richness. I used ﬁsh assemblage,
hydrogeomorphic, and anthropogenic data from 360 lakes in Maine, New Hampshire,
Iowa, Michigan, and Wisconsin. To build lake classes, I used classiﬁcation and
regression trees; lakes having similar ﬁsh species richness were best grouped by lake
area, ecological drainage unit, and watershed area. Results indicated that in addition to
positive associations of lake and watershed area with species richness, regionalization
frameworks were useful tools in identifying unique areas of freshwater biodiversity.
Within classes of naturally similar lakes, species richness generally had a negative
association with ﬁsh stocking, dams, and urban and agricultural land cover/use. Species
richness was positively associated with human population density and road density. The
methodology of classifying lakes appears to be a tool that lake managers can utilize to
better understand ﬁsh assemblages in inland lakes. Classiﬁcations were also useful in

grouping naturally similar lakes to identify the effects of human disturbances.

TABLE OF CONTENTS

 

 

 

 

 

 

 

LIST OF TABLES _ -- .................. - -- - - ..... iv
LIST OF FIGURES - ..... - - - - - - - - -- - ---v
THESIS INTRODUCTION- - ..... -- - -- 1
CHAPTER I: HYDROGEOMORPHIC LAKE CLASSIFICATION - ““4
CHAPTER II: EFFECTS OF ANTHROPOGENIC DISTURBANCE ...................... 36
THESIS CONCLUSION 60
APPENDIX... - -- - 63
BIBLIOGRAPHY ........................ - - _- - - -_ -- 99

iii

TABLE 1:

TABLE 2:

TABLE 3:

TABLE 4:

TABLE 5:

TABLE 6:

TABLE 7:

TABLE 8:

TABLE 9:

TABLE 10

TABLE 11:

TABLE 12:

TABLE 13:

TABLE 14:

TABLE 15:

TABLE 16:

TABLE 17:

TABLE 18:

TABLE 19:

LIST OF TABLES

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

CATEGORIES OF DATA 63
HYDROGEOMORPHIC DATA - -64
ANTHROPOGENIC DATA 65
FISH ASSEMBLAGE DATA - 66
FISH RESPONSE METRICS 67
FISH STOCKING DATA 68
MULTIPLE REGRESSION MODELS 69
PEARSON CORRELATION COEFFICIENTS 70
LAKE CLASSIFICATIONS 71
: RECREATIONAL AND NON-RECREATIONAL SPECIES ............... 72
SMALL AND LARGE LAKES 73
LAKE CLASSES A, B, C, AND D 74
LAKE CLASSES E, F, G, H, I, AND J 75
ANTHROPOGENIC DATA (CLASSES A, B, C, AND D) 76
ANTHROPOGENIC DATA (CLASSES E, F, G, H, I, AND J) ............. 77
NATIVE SR CLASS REGRESSION MODELS 78
NATIVE SR LAKE SIZE REGRESSION MODELS 79
RECREATIONAL SR CLASS REGRESSION MODELS 80

 

NON-RECREATIONAL SR CLASS REGRESSION MODELS .......... 81

iv

FIGURE 1:

FIGURE 2:

FIGURE 3:

FIGURE 4:

FIGURE 5:

FIGURE 6:

FIGURE 7:

FIGURE 8:

FIGURE 9:

FIGURE 10

FIGURE 11:

FIGURE 12:

FIGURE 13:

FIGURE 14:

FIGURE 15:

FIGURE 16:

FIGURE 17:

LIST OF FIGURES

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MAP OF LAKES 82
TOTAL SR CLASSIFICATION 83
NATIVE SR CLASSIFICATION ..... -84
MAP OF TOTAL SR SMALL LAKE CLASSES 85
MAP OF TOTAL SR LARGE LAKE CLASSES ..... 86
MAP OF NATIVE SR SMALL LAKE CLASSES 87
MAP OF NATIVE SR LARGE LAKE CLASSES - 88
NATIVE SR CLASS A FISH ASSEMBLAGE 89
NATIVE SR CLASS B FISH ASSEMBLAGE 90
: NATIVE SR CLASS C FISH ASSEMBLAGE 91
NATIVE SR CLASS D FISH ASSEMBLAGE 92
NATIVE SR CLASS E FISH ASSEMBLAGE 93
NATIVE SR CLASS F FISH ASSEMBLAGE 94
NATIVE SR CLASS G FISH ASSEMBLAGE 95
NATIVE SR CLASS H FISH ASSEMBLAGE 96
NATIVE SR CLASS I FISH ASSEMBLAGE 97
NATIVE SR CLASS J FISH ASSEMBLAGE 98

Thesis Introduction

The natural landscape has shaped freshwater ecosystems across the Earth;
glaciers, climate, and other natural processes have affected the distributions and
abundances of biota in streams, inland lakes, and other aquatic ecosystems. Because
freshwater ecosystems are so complex and inﬂuenced by many features, natural and
human, at multiple temporal and spatial scales, lake managers and researchers have
attempted to partition the many effects of the landscape on in-lake biota by classifying
ecosystems. These efforts have resulted in a strong understanding of how land use/cover
types (e.g., forest, wetlands), lake morphometry (e.g., lake area, lake depth), water quality
(e.g., pH, total phosphorous), hydrology, and surface connectivity affect ﬁsh
assemblages, yet often all of these factors have not often been included comprehensively
in the same lake classiﬁcation study (Johnson et al. 1977, Dolman 1990, Schupp 1992).

Regionalization frameworks have been created to identify area of unique
biodiversity; they represent sets of natural features across a large region, with subregions
distinguished by ﬁner scale variation among features (Seelbach et. al. 2002). These
schemes can help to explain variation among ecosystems at different spatial scales
(Omemik 1987, McMahon et al. 2001, Irz et al. 2004, Cheruvelil et al. 2008).
Subsequently, I attempt to compare the ability of these various metrics of the
hydrogeomorphic landscape across scales and in combination to explain patterns of ﬁsh
assemblages using a lake classiﬁcation approach.

Classiﬁcations have been built as a way to comprehensively explain the variation
in ﬁsh species composition across the landscape, resulting in relatively similar lakes

within a class and higher variability among classes Iowa (Johnson et a1. 1977, Dolman

1990, Schupp 1992). Since ﬁsh management agencies for most states are often
ﬁnancially limited and lack the ability to gather data for every lake, classiﬁcations are
potentially an efﬁcient tool for tackling the pitfalls of managing thousands of lakes with
limited information. While each lake classiﬁcation is best developed with distinctive
goals in mind (and often unique predictor variables as a result) to explain any number of
characteristics of ﬁsh assemblages; the entire method (that of developing classiﬁcations
and using them in a management context) is an important concept to further investigate
and upon which to build.

Similar to the natural landscape, human activities also impact aquatic systems;
lake managers and researchers have attempted to use anthropogenic variables to explain
variation within and among lakes as well (Brunberg and Blomqvist 2001, Pierce and
Tomcko 2003). While some anthropogenic disturbances positively affect valued qualities
of ﬁsh assemblages through species additions (stocking), many have negative effects,
such as blockage of movement (e.g., dams), and landscape alterations (e.g., land
use/cover) (Radomski and Goeman 1995, Carpenter et al. 2007, Johnson et a1. 2008).
Because management agencies are often ﬁnancially limited, they are often resigned to
weigh trade-offs between preventing human-induced species invasions while at the same
time supplementing faltering ﬁsheries through stocking and habitat alterations, all the
while maintaining relatively stable and healthy ecosystems. Although individual lakes
are unique in many ways, there are similarities of how human disturbances affect ﬁsh
assemblages. Therefore, I hypothesized that after grouping naturally-similar lakes using

landscape-based classiﬁcation techniques, the effects of humans on ﬁsh assemblages can

be quantiﬁed in a way that is relevant to management agencies that then can implement
plans and account for humans across classes of lakes.

The research presented here examines how natural features and human
disturbances affect ﬁsh species richness in inland lakes. I build upon previous lake
classiﬁcation methods to investigate the ability of regional-scale features in addition to
previously well studied local-scale features to classify 360 lakes across Maine, New
Hampshire, Michigan, Wisconsin, and Iowa (Johnson et a1. 1977, Dolman 1990, Schupp
1992). In chapter 1, I create a hydrogeomorphic lake classiﬁcation using natural
landscape metrics from four major categories: regionalization frameworks, species
extinction/isolation, lake hydrology, and land use/cover to predict patterns of total and
native species richness (having statistically removed the effects of disturbance on the
response metric). In chapter 2, I measure the effects of anthropogenic disturbances on
native species richness within classes of lakes from chapter 1 using land use/cover, dam,
road density, ﬁsh stocking, and human population density as predictor metrics. Because
no other study has covered such a broad geographical region and large number of lakes
and examined regional-scale features in conjunction with local features, my research will
contribute to the science of landscape limnology and to the further development of
methods for lake classiﬁcation. Additionally, this research will provide managers with
helpful information and a fruitful approach for developing classes of lakes that are

naturally similar, and likewise, respond similarly to human disturbances.

Chapter l-What natural lake and landscape features best classify ﬁsh species
richness in 360 north-temperate lakes using CART analysis?

Introduction

Researchers and lake managers have long attempted to explain and understand variation
of in-lake features among lakes, often using local landscape features and lake
morphometry to predict different metrics of water quality (e.g., total phosphorous, pH)
and ﬁsh assemblages (e.g., spatial distribution, species richness). Lake classiﬁcation
systems are one way to group lakes according to the variation of a response metric (e.g.,
species richness), using the most signiﬁcant predictor variables for a set of lakes (Johnson
et a1. 1977, Dolman 1990, Schupp 1992). Such studies provide a basic approach to
classifying ﬁsh assemblages based on local landscape, lake morphometry, and in-lake
water quality features. Previous studies typically were conducted in a single state or
Canadian province, and their focus was to explain variation in ﬁsh assemblages using
local landscape and in-lake features (Rahel 1984, Dolman 1990, Schupp 1992, Jones et
al.2004). They provide a method and context for investigating how to monitor and
manage statewide ﬁsh populations through the use of predictive lake classiﬁcations.
State and federal agencies are primarily responsible for managing inland lake ﬁsh
assemblages; they are often ﬁnancially limited and must weigh trade-offs between
investing in assessment and implementing decisions from gathered information (Hansen
and Jones 2008). Management agencies in lake-rich regions therefore must implement
their programs on all lakes, while only being able to collect data, particularly of in-lake
biota, from a subset of lakes. However, based on known lake-to-landscape relationships
established using data from sampled lakes, managers can now extrapolate these results

from sampled lakes to unsampled lakes (using readily-available landscape data for

unsampled lakes) to help make decisions. To build a natural lake classiﬁcation system
based on hydro geomorhic landscape features, it is important to understand and include
those features already known to shape lake ﬁsh assemblages, while still accounting for
human disturbance. Therefore, in this chapter, I created a series of lake classiﬁcations
based on natural hydrogeomorphic features of the landscape (a combination of natural
hydrologic, geologic, and topographic factors) that range in spatial scale to explain the
variation in ﬁsh species richness present among lakes.

We can learn a lot from previous research of the relationship between the natural
landscape and ﬁsh assemblages conducted at the local scale (i.e., lakes within a single
state). Studies have extensively measured the landscape-to-ﬁsh linkages in terms of lake
surface area (Johnson et al. 1977, Tom and Magnuson 1982, Rahel 1984, Eadie et al.
1986, Jackson and Harvey 1989, Holmgren and Appleberg 2000, Riera et al. 2000, Irz et
al. 2004), lake perimeter (Olden et al. 2001 , Olden and Jackson 2002, Hershey et al.
2005, Olden et al. 2006), lake depth (Marshall and Ryan 1987, Mehner et al. 2005),
surface water connectivity as an isolation metric and as a proxy for landscape position
(Kratz et al. 1997, Magnuson et al. 1998, Riera et al. 2000, Olden et al. 2001), and
catchment area (Allan and Johnson 1997, Tonn et al. 2003, Irz et al. 2004). These studies
show positive relationships between lake surface area, depth, and/or perimeter and ﬁsh
species richness; likewise, lakes lower in the landscape with more surface connections
have higher species richness. Conversely, lakes with relatively low habitat complexity
(e.g., low values for lake area, depth or perimeter) or isolated lakes (e.g., higher in the
landscape, with no surface connections) typically have relatively low species richness.

By identifying the linkages between the landscape and species richness through past

literature, it is possible to create a lake classiﬁcation that is interpretable and applicable to
lake managers.

Past lake classiﬁcation systems were created for unique purposes in single states
or provinces using a limited number of natural landscape, in-lake water quality, and
human disturbance predictors to group inland lake ﬁsh assemblages (Johnson et al. 1977,
Tonn et al. 1983, Dolman 1990, Schupp 1992). The term “ﬁsh assemblage” has included
a variety of response metrics: presence or absence of individual or key ﬁsh species
(Johnson et al. 1977), relative species abundance (Marshall and Johnson 1987),
associations of dominant species (Dolman 1990), or species richness (SR) and
composition (Eadie et al. 1986, Kratz et al. 1997). This previous research has
incorporated many different lake morphometry characteristics to group inland lake ﬁsh
assemblages (Tonn et al. 1983), water quality data such as pH and temperature (Dolman
1990), physical substrate (Johnson et al. 1977), or local human disturbances such as
distance to nearest road (Magnuson et al. 1998) to explain among-lake differences in ﬁsh
assemblage metrics. Lake ecosystems are complex and there are many landscape factors
that shape ﬁsh assemblages at the local level; however, regional-scale predictors (e. g.,
spatial patterns of past glaciation, regional climate characteristics) may play a role as
well. While unique in their purpose and spatial scale (across a single state), many past
attempts at classifying lakes have disregarded regional-scale factors when attempting to
better understand the distribution and variation of ﬁsh assemblages across the landscape.

Past studies have laid a strong foundation for developing lake classiﬁcations and
determining what local features shape inland lake ﬁsh assemblages; however, research

suggests incorporating landscape features at a much larger spatial scale can aid in

explaining variation among lakes (Tonn 1990, Poff 1997, Cheruvelil et al. 2008).
Although individual lakes have somewhat unique physical, chemical, and biological
characteristics, there also are similarities across a state and even across broader regions,
exempliﬁed by the wide-ranging distribution of many ﬁsh species. Therefore, in
developing an approach to classify lakes across large lake-rich areas for a variety of
purpose, it is important to incorporate hydrogeomorphic landscape predictors at multiple
spatial scales. When working at such spatial scales, lakes can be viewed as patches
within a landscape that are hierarchically organized, with regional factors such as climate
and soil type constraining local and in-lake impacts on lake biota (Tom 1990, Poff 1997,
Cheruvelil et al. 2008, Soranno et al. 2009).

One way to incorporate a broader-scale landscape perspective into a classiﬁcation
framework is to include a regionalization scheme as a classifying variable. Such schemes
have been developed for both terrestrial and aquatic systems, and they can serve as strong
proxies for natural features (e. g., soil types, sub-surface geology, land use; Omemik
1987). Most regionalizations represent a set of natural features across a large region,
with subregions distinguished by ﬁner scale variation among features (Seelbach et. al.
2002). These schemes can help to explain variation among ecosystems at different
spatial scales (Omemik 1987, McMahon et al. 2001, Irz et al. 2004, Cheruvelil et al.
2008), and thus should be included as a potential classiﬁcation variable to group lake ﬁsh
assemblages.

I included in my analysis three regionalization frameworks that have been shown
to help to explain variation among aquatic systems: United States Geological Survey

(USGS) hydrologic units, (HUC, Seaber et al. 1987), ecological drainage units (EDU,

Higgins et al. 2005), and freshwater ecoregions (Abell et al. 2000). HUC’s were
developed by the USGS to approximate watersheds at multiple scales across the country
and provide a standardized method of locating, storing, retrieving, and exchanging
hydrologic data for water-resources organizations (Seaber et al. 1987). HUC’s are
hierarchical in spatial scale; six-digit HUC’s are further divided by eight-digit HU C’s
with 12-digit HUC’s being an even ﬁner spatial scale. EDU’s were developed by
agglomerating eight-digit HUC’s using landscape features, such as climate and landforrn,
to form contiguous, geographically dependent regions. These regions were developed to
classify freshwater ecosystems to represent coarSe-scale regions of freshwater
biodiversity for regional conservation planning (Higgins et al. 2005). Freshwater
ecoregions were created using native ﬁsh distributions (presence/absence data) to
delineate sub-regions across North America; they represent likely freshwater aquatic
biodiversity patterns for conservation purposes (Abell et al. 2000). All of these
regionalization schemes have been delineated at different spatial scales; across the ﬁve-
state region there are 97 six-digit HUC’s, 28 EDU’s, and ﬁve freshwater ecoregions, each
allowing for an examination of the most appropriate spatial scale for grouping lake ﬁsh
assemblages.

Because I was attempting to build classes based upon natural landscape features,
it is imperative to recognize that human disturbances can also signiﬁcantly alter lake ﬁsh
assemblages at multiple spatial scales and mediate the effects of natural features. For
example, land use/cover, the prevalence of roads and dams, and ﬁsh stocking rates all
vary across the landscape and affect inland lake ﬁsh assemblages (Radomski and

Goeman 1995, Magnuson et al. 1998, Roth et al. 2007). Human disturbances such as ﬁsh

stocking or dam and reservoir creation can mask natural patterns by directly and/or
indirectly altering ﬁsh assemblage patterns through species introductions and food web
alterations (Radomski and Goeman 1995, Johnson et al. 2008). Because natural lake
classiﬁcations are based on predictable relationships between natural landscape features
and ﬁsh assemblages, human disturbances, which can mask natural patterns, must be
considered and accounted for when attempting to explain natural variation in ﬁsh
assemblages.

My study examines the variation of ﬁsh species richness across ﬁve states
(Maine, New Hampshire, Michigan, Wisconsin, Iowa), ranging from the Atlantic Ocean
to west of the Mississippi River. My work incorporates both local and regional-scale
predictor metrics; the focus is to use the natural landscape across multiple states and
spatial scales to identify which metrics explain the greatest amount of variation in lake
ﬁsh species richness. Because previous classiﬁcations are unique to their own regions,
hypotheses, and goals, my study is not necessarily intended to compare my classiﬁcation
with previous results, but rather to examine the approach of using regional-scale
predictors along with local landscape features to explain variation in ﬁsh species richness
among lakes.

In this chapter, I developed lake classiﬁcations to explain patterns of ﬁsh species
richness based on natural hydrogeomorphic features of the landscape (combination of
natural hydrologic, geologic, and topographic factors) that range in spatial scale. My
study provides a good opportunity to compare local and regional patterns of ﬁsh
assemblages, unlike past studies that only focused on a single state. Species richness, for

this study, was judged to be a more reliable response metric than others (e. g., ﬁsh

abundance) because ﬁsh assemblage data was collected from multiple agencies each with
their own methods and purposes. Each state had a unique set of species to examine, so
species counts were determined to be a more universal and comparable response metric
than assemblage type. I also advanced the methodology behind building natural lake
classiﬁcation systems by taking advantage of advances in geographical information
systems (GIS) and incorporating regional-scale predictors with previously well-studied
local landscape factors and by accounting for human disturbances. I restricted the
potential predictor variables to those that are map-based and hence available for all lakes.
Although in-lake water quality can predict ﬁsh assemblages (Bachmann et al. 1996,
Jeppesen et al. 2000), I did not include water quality metrics as predictors. This way, it is
possible to extrapolate my natural lake classiﬁcation results that are based on map and
GIS data to ‘new,’ unsampled lakes. I also focused on the ability of natural
hydrogeomorphic features (as opposed to human-mediated landscape features) to build
lake classes because management and conservation efforts often need to know what the
‘natural’ state of a lake would be in the absence of anthropogenic factors. Natural
features such as lake depth or lake area are relatively unchanging as compared to human
disturbances; therefore, the relationship between the landscape and ﬁsh species richness

may be more predictable when including only natural features.

Methods

Data Collection and Lake Selection
I collected data on hydrogeomorphic features (e.g., regionalization schemes, lake

hydrology, lake morphometry) and anthropogenic disturbances (e. g., land use, road

10

density, human population) for approximately 2300 inland lakes as part of a larger study
(e.g., Cheruvelil et al. 2008, Webster et al. 2008) to determine how the natural and human
landscapes shape lake water quality and ﬁsh assemblages (Tables 1-3). The initial data
collection focused primarily on lakes with water quality response metrics that are
typically easier to collect and more readily available than ﬁsh assemblage information.
Available ﬁsh assemblage data were then obtained for New Hampshire, Maine,
Michigan, Wisconsin, and Iowa (Figurel, Table 1). Because water quality and ﬁsh
assemblage monitoring is often conducted by different agencies using different methods
for lake selection, not every lake that existed in the original hydrogeomorphic (2,348
lakes) and landscape (2,573 lakes) databases existed in the ﬁsh assemblage database (555
lakes). Therefore, I created an overlap database that contained 360 lakes with all three
data categories (hydrogeomorphic, anthropogenic, and ﬁsh assemblage; Table 1). For
these 360 lakes, I obtained additional hydrogeomorphic (e.g., lake isolation metrics,
number of surface connections) and anthropogenic disturbance (e.g., dam density, ﬁsh
stocking metrics) data from agency sources and the use of Geographic Information

System (GIS; Table 1).

Fish Assemblage Response Metrics

I gathered ﬁsh assemblage data from state and federal agencies and universities
(Table 4); each entity’s goal was to quantify presence/absence of ﬁsh species or estimate
abundance based upon sampling effort to effectively monitor ﬁshery resources. For lakes
represented multiple times in the ﬁsh assemblage data, I used the most recent year

sampled. Most ﬁsh had been identiﬁed to the species level. However, some ﬁsh were

11

identiﬁed only to an aggregate level of genus or family, or were of unknown identity
(7%-24% of ﬁsh across ME, NH, MI, WI, and IA). For example, some states reported
ﬁsh as “unknown bullhead” or “minnow sp”. Depending on the response metric (see
below for response metric details), these aggregate terms were either included or
removed based upon lake-speciﬁc ﬁsh assemblages and by using the following decision
rules. If a species was present in a lake, higher taxonomic aggregates and unknown
ﬁshes related to that species were eliminated. However, the aggregate term was included
in the species richness count if no other species representing that aggregate term was
found in the lake. For example, if the aggregate “unknown bullhea ” was found in a lake
but no speciﬁc bullhead species (yellow, black, or brown) was recorded, I counted the
aggregate as representing one bullhead species. On the other hand, if a speciﬁc bullhead
species was captured in a lake in addition to an aggregate being recorded, the aggregate
was eliminated from the total species count.

I created two metrics to use as response metrics in our natural lake classes: Total Species
Richness (Total SR) and Native Species Richness (Native SR) (Table 5). To standardize
across all 360 lakes in ﬁve states, species richness was used as a consistent response
metric for total and native species (native by state). While metrics such as species
abundance or assemblage type (e.g., bass and bluegill co-occru'rence) were considered,
ﬁsh assemblage data came from different agencies and were collected using different
methods and amount of effort (e. g., electroshocking, fyke netting). Fish assemblages
were quite different from state to state and management agencies have different social
values to manage for in their lakes; these facts made species abundance or assemblage

type less advantageous response metrics than species richness.

12

I selected the two most prominent gears for each state (Maine and New
Hampshire (gill net and trap net); Michigan, Wisconsin, and Iowa (electroshocking and
fyke net)). I used only the species caught with these corresponding gears to create my
response metrics to include the greatest number of ﬁsh species from each lake while
minimizing gear biases across lakes and management agencies (Table 4). In other words,
I did not want to include species caught with a gear that was only used in a few lakes
and/or a single state. State-speciﬁc records (not lake-speciﬁc) were used to determine
whether a ﬁsh was native or not for all lakes in that state (Maine (David Halliwell,
personal contact), New Hampshire (Scarola 1973), Michigan (Hubbs and Lagler 2007),
Wisconsin (Becker 1983), and Iowa (Harlan and Speaker 1956)). Because it is
impossible to determine lake-by-lake historical assemblages, I considered a ﬁsh to be
native to a state if it was at least native to a single lake in the state. Since my two
response metrics were derived using the same ﬁsh assemblage data, to quantify the
association between these two variables, I ran a correlation of one (Total SR) against the

other (Native SR).

Hydrogeomorphic Features and Anthropogenic Disturbances

I included hydrogeomorphic features and anthropogenic disturbance data in
categories of regionalization, extinction/isolation, lake hydrology, and land use/cover (see
details of methods in Webster et al. 2008). I compiled maximum lake depth, mean lake
depth, and watershed cumulative catchment area from existing state databases, and
quantiﬁed lake surface area and perimeter from state geographic information systems

(GIS) data at a 1: 24,000 resolution. A lake’s runoff was calculated using mean annual

l3

runoff from 1951 to 1980 within a SOO-m buffer from a GIS coverage for a lake (Webster
et al. 2008). Land use/ cover data within each lake’s 500-m buffer were obtained from
the 1992 National Land Cover Dataset (Webster et al. 2008).

I calculated additional hydrogeomorphic features not described in Webster et al.
(2008) as summarized in Table 2. The regionalization frameworks used are as described
above: USGS six-digit hydrologic units (97 total regions) (HUC, Seaber et al. 1987),
ecological drainage units (28 total regions) (EDU, Higgins et a1. 2005), and freshwater
ecoregions (5 total regions) (Abell et al. 2000). To obtain a measure of species isolation,
I displayed maps showing the furthest glacial extent (> 15,000 years ago) in a GIS to
determine whether a lake’s current location had been covered by glaciers or not (Prest
1969)
In order to get a measure of potential immigration of new species (leading to an increase
in species richness), I included measures of lake connectivity and surﬁcial hydrology as
predictor variables in my classiﬁcations. I determined the number of surface water
connections of each lake, both upstream and downstream, using GIS and the National
Hydrography Dataset (N HD) at a 1: 100,000 resolution (http://nhd.usgs.gov, 2007).
Lakes with no surface connection were declared “seepage” while lakes with one or more
surface connections were determined to be “drainage”. I used GIS to calculate regional
and local lake density, deﬁned as the number of lakes within each EDU (range = 159-
2856) and 12-digit HUC (range = 1-89; 10,609 total regions across ﬁve-state region). An
individual lake’s regional and local lake density were measured by translating all water
bodies (>0.001 kmz) across the entire spatial range of EDU’s and 12-digit HUC’s from

polygons to points; these centroids were placed within the polygon of each water body.

14

Each lake was assigned a region (EDU and 12-digit HUC) and the total number of water
bodies within each EDU and 12-digit HUC was counted. My 360 study lakes were then
assigned to a region (EDU and 12-digit HUC) with its corresponding lake density for
each regionalization scheme. At a ﬁner spatial scale, I collected measures of lake hydrology
such as average annual precipitation and mean base-ﬂow index (BFI), an estimate of sub-
surface water inputs relative to surface water inputs (USGS) (http://water.usgs.gov)).
Average annual precipitation for years 1971-2000 was obtained from the Spatial Climate
Analysis Service at Oregon State University (www.0cs.oregonstate.edu).

I also collected anthropogenic disturbance variables at multiple spatial scales
(summarized in Table 3) to account for the effects of disturbances on my response
metrics before creating lake classes. Related to measures of potential ﬁsh species
isolation, I calculated dam density within each EDU and 12-digit HUC using dam GIS
layers from state agency databases. Just like calculating lake density at multiple spatial
scales, dams were assigned a region, and each study lake was assigned a regional and
local dam density using EDU’s (range = 48-3,634) and 12 digit-HUC’S (range = 0-59).
Other anthropogenic variables related to potential ﬁshing pressure and shoreline habitat
modiﬁcation were collected for each lake including human population and housing
density (United States 2000 Census (http://factﬁnder.census.gov)), and local road density
(it/acre) (calculated within a 500 m buffer of each lake).

Lake-speciﬁc ﬁsh stocking records were collected from each state to control for
the effects of ﬁsh stocking on ﬁsh species richness (Maine (http://www.pearl.maine.edu),
New Hampshire (Gabe Gries, personal contact), Michigan

(http://www.michigandnr.com/ﬁshstock), Wisconsin

15

(httpz/linfotrek.er.usgs.gov/doc/wdnr_biology/Public_Stocking), and Iowa
(http://limnoweb.eeob.iastate.edu/ﬁshgrowth)) (Table 6). I created three different metrics
of stocking pressure for each individual lake: total number of species stocked, total
number of stocking events (regardless of species), and total number of stocking events
divided by the number of years for which I had stocking data for each state (Table 6).
These metrics likely vary some from state to state because each state has different angler
pressures for particular species, varying ﬁnancial resources for stocking, and many
biological factors to consider when determining which species to stock, when, and how
often. The third stocking metric was created to help standardize stocking across states by
accounting for the different number of years of stocking data available for each state

(Table 6).

Developing Natural Lake Classes

To build natural lake classes based on hydrogeomorphic features, I wanted to ﬁrst
account for the effects of human disturbances on lake ﬁsh assemblages. I ﬁrst
transformed all anthropogenic and ﬁsh response metrics to meet normality assumptions
for regressions models including arcsine-square root transformation for land use/cover
variables and natural log, squared, or square root for all other variables. Because every
transformation did not necessarily result in meeting assumptions (approximately 75% of
the variables did not meet assumptions); I transformed the remaining metrics using the
best relative transformation based on the overall distribution of data and that with the
least amount of outliers. I then ran multiple linear regressions of each response metric

(Total SR and Native SR) against all human disturbance variables (Table 3), regardless of

16

correlations among variables, to obtain residual values to be response metrics when
building natural lake classes. Best model selection was used to determine which
disturbance variables to include in the ﬁnal model (SAS version 9.1). I used condition
index (Cl) and variance inﬂation factors (VIP) to determine if multicollinearity among
predictor variables was unduly inﬂuencing our models. Variables with CI values greater
than 10 and VIP values greater than 30 in the regression were removed from the model
(Hair et al. 1995). The best model was signiﬁcant using 0.05 alpha and had both the
highest adjusted R-squared and lowest Akaike's information criterion (AIC) value (Table
7).

The residuals, related to the amount of variation unexplained by these multiple
regression models of human disturbance variables and ﬁsh metrics, were tested to see if
they met normality assumptions after these initial regression models. They were
normally distributed; therefore, no subsequent transformations were needed. The
residuals were then used as the response variables in classiﬁcation and regression tree
(CART) analyses (R version 2.9.0) with multiple hydrogeomorphic variables as
predictors (Table 2). All hydrogeomorphic variables were left untransformed and
included in the CART models (this method allows for non-normal data); a correlation
matrix of highly correlated (>.6) predictors was used, however, when interpreting results.
CART models were built using the recursive partitioning algorithm “rpart” in the R
software system that initially splits a parent group into two daughter groups based on the
value of a predictor variable that maximizes the reduction in the total residual sum of

squares of the response metric (Breiman et al. 1984). Each daughter group is then split

17

using the same algorithm; I examined the proportional reduction in error (PRE) for each
split in the tree and summed them to produce an overall PRE value.

Each CART model was run using a range of complexity parameters (CP; 0.01,
0.015, 0.02, 0.025, and 0.03). Increased CP values result in models with relatively
coarse, less reﬁned tree (few splits and subsequent terminal classes); decreased CP values
result in a more reﬁned tree with more splits and terminal classes. Since my other
objective for this research was to examine human disturbance on lake classes (chapter 2),
I sought to ﬁnd a reasonable and pragmatic ﬁnal classiﬁcation. Therefore, I compared
the number of classes and the number of lakes within each class for the classiﬁcations
that were made using each CP cutoff. To be sure that each classiﬁcation resulted in
classes that were signiﬁcantly different from one another, I ran analyses of variance
(ANOVA) and post hoc Tukey tests to compare the variation between classes within a
classiﬁcation (SAS version 9.1). I did not objectively select a ﬁnal, “best” lake
classiﬁcation but rather, subjectively determined the most useful classiﬁcation based on
the total number of classes, the range of the number of lakes in the classes, and the

ANOVA results.

Results
Controlling for Human Disturbance

The two response metrics were highly correlated (r = .97); state-by—state, the total
number of species in a lake is very similar to the total number of native species (native by
state) (Table 5). Human disturbances explained 16% (Total SR) and 22% (Native SR) of

the variation in lake ﬁsh assemblages and both ﬁnal models contained similar

18

disturbances variables: number of ﬁsh species stocked, number of stocking events/years
of data, dam density (12-digit HUC), local agricultural land use, and human population
density (Table 7). Even though some predictor metrics were highly correlated, all were
included in the models so that I could explain the greatest amount of variation by human
disturbances. With the exception of agricultural land use metrics (Total SR -, Native SR
+), both models had similar parameter estimates: number of ﬁsh species stocked (-),
number of stocking events/years of data (+), dam density (12-digit HU C) (-), and human
population density (+) (Table 7). Within a general landscape category, many of the land
use/cover and ﬁsh stocking metrics exhibited high correlation with one another (Table 8).
For example, agricultural land use metrics (crop, pasture, all agriculture, and agriculture
combined with urban) had coefﬁcients greater than 0.66as were all three stocking metrics

(r > 0.85) (Table 8).

Analysis of Variance Across Classes

Before I present the results of my lake classiﬁcations (below), I discuss how I
chose a CP value, and thus a ﬁnal classiﬁcation for each response metric (Total SR and
Native SR). ANOVA analysis of all classiﬁcations for Total SR and Native SR revealed
all models to be statistically signiﬁcant (Table 9). The classiﬁcation with the fewest
number of classes (CP=0.03) had the greatest percentage of signiﬁcantly different class-
by-class comparisons; few classes were not signiﬁcantly different from another.
Conversely, the classiﬁcation with the most classes (CP=0.01) had the lowest percentage
of signiﬁcantly different classes; many classes were not signiﬁcantly different from

another. The tradeoff seen here makes intuitive sense, as the number of classes increases,

19

there is less in-class variation but not as great of a difference in species richness from one
group of lakes to the next. Because any classiﬁcation is created for different purposes
(with this tradeoff likely driving how reﬁned or unreﬁned ﬁnal classes are), lake
managers will likely determine on a case-by-case basis how many clasSes (and how
different those classes should be from one another) they would like to include in a
classiﬁcation.

My resulting classiﬁcations could all be considered and further analyzed since
they all had signiﬁcant among-class variation. The ﬁnal classiﬁcations selected for
presentation in chapter 1 (CP = 0.015) were not meant to be the statistically “best”
classiﬁcation. Rather, I chose these classiﬁcations based on the number of ﬁnal classes
and the range of the number of lakes across classes; in both classiﬁcations (Total SR and
Native SR), over 75% of the pair-wise comparisons were signiﬁcantly different (Table 9).
Since my ultimate goal was to measure the effects of human disturbances on classes of
lakes (chapter 2), I selected a classiﬁcation (Native SR) that did not have very few classes
with many lakes (e. g., CP of 0.03) or a classiﬁcation with a large number of classes, each
with very few lakes (e.g., CP of 0.01).

In the ﬁrst scenario (too many lakes in a class), some classes are not reﬁned
enough (split into smaller groups) to make ﬁnal classes management-applicable. For
example, if all 360 lakes were split into only small and large lakes (< or > 53 ha), little
would be known about lakes within the two classes (e.g., land use/cover) other than large
lakes have more species than small lakes. Therefore, it is likely more helpful to lake
managers to have smaller classes and know more about the lakes in them by reﬁning

these larger classes. In the second scenario (too few lakes in a class), there would not be

20

enough observations to run multiple regression analysis (chapter 2) given the number of
disturbance predictors in this study. For example, a class with only ﬁve lakes would
result in a non-signiﬁcant regression model, making the class (and its lakes) statistically
unhelpful. Given this issue, it was important to have a moderate ntunber of lakes in an
individual class. For chapter 1, I presented the results for classiﬁcations using a CP of
0.015, which appeared to fall in the middle of ranges for class number and class size

(e. g., number of lakes per class). Ultimately, for chapter 2, I selected a classiﬁcation that
had 10 classes ranging in size from 10 to 69 lakes; this decision led to a more pragmatic

and thorough examination of human disturbances across the landscape on my study lakes.

Determining the Hydrogeomorphic Variables that Classify Fish Species Richness
Both lake classiﬁcation trees (Total SR and Native SR) included predictors
representing multiple spatial scales and general landscape categories. Most notably, both
models included the extinction metric lake area as the main split in the model, a regional
scale predictor (EDU), and a lake hydrology metric (watershed area) (Figures 2 and 3).
Many of the morphometry variables were highly correlated with one another, (e. g., lake
area vs. lake perimeter; r = 0.92, mean depth vs. maximum depth; r = 0.88, Table 8).
Other highly correlated variables that appeared in at least one CART model include:
wetland (woody) which is correlated with wetland (all) (r = 0.97), and cumulative
watershed area which is correlated with lake perimeter (r = 0.74), lake area (r = 0.70),
and number of surface connections (r = 0.69) (Table 8). It is important to recognize how

these signiﬁcant hydrogeomorphic variables are correlated because it affects our ability

21

to interpret our classiﬁcations; it is difﬁcult to distinguish the effects of some predictors .
from each other (e. g., lake perimeter and lake area).

The ﬁnal results presented here use a CP cutoff of 0.015 in the Total SR
classiﬁcation. Five natural landscape predictor metrics partitioned the 360 lakes into 11
ﬁnal lake classes that included 7 to 105 lakes and explained 60% of the variation in Total
SR disturbance residuals (Figure 2). Lake area was the ﬁrst split that separated the entire
group of lakes; the regionalization framework EDU split the two main subgroups further.
In addition to lake area and EDU, wetland (woody), cumulative watershed area, and
mean base ﬂow partitioned the remaining subgroups of lakes (Figure 2). The smaller
lakes (< 53 ha), classes A and B, which have the lowest total species richness, lay
primarily in Maine and New Hampshire, with a few lakes in parts of Wisconsin and Iowa
(Figure 2 and 4). Lake classes C, D, and E primarily lay in Michigan, Wisconsin, and
Iowa and have relatively higher total species richness than Maine and New Hampshire
lakes (Figure 2 and 4). Larger lakes (> 53 ha), classes F, G, H, and I (151 total lakes), lay
entirely in Maine and New Hampshire and make up most of Michigan, Wisconsin, and
Iowa (Figure 5). These regions have relatively higher mean total species richness than
classes A through E, but have relatively lower total species richness than classes J and K
(Figures 2 and 5). Classes J and K lay primarily in Southeastern Michigan, Northwestern
Wisconsin, and Eastern Iowa; these lakes have the highest total species richness across all
360 lakes in the 5-state region (Figures 2 and 5).

In the Native SR classiﬁcation, again using a CP cutoff of 0.015, only four natural
landscape predictor metrics partitioned the 360 lakes into 10 ﬁnal lake classes that

included 10 to 69 lakes and explained 58% of the variation in Native SR disturbance

22

residuals (Figure 3). Lake area was the ﬁrst split that separated the entire group of lakes;
the regionalization framework EDU split the two main subgroups further. While the
response metric was different in the Native SR model, lake area and EDU separated the
initial group of 360 lakes similarly to what I saw with Total SR. In addition to lake area
and EDU, lake elevation and cumulative watershed area partitioned the remaining
subgroups of lakes (Figure 3). In the smaller lakes (< 53 ha), classes A and B had the
same distribution as classes A and B from the Total SR model; they lay primarily in
Maine and New Hampshire, with a few lakes in parts of Wisconsin and Iowa and these
lakes had the lowest relative native species richness (Figure 3 and 6). Lake classes C and
D primarily lay in Michigan, Wisconsin, and Iowa and had relatively higher total species
richness than lakes in classes A and B (Figure 3 and 6). The larger lakes (> 53 ha),
classes E, F, G, H, I, and J (227 total lakes), had a similar distribution to the classes in the
Total SR classiﬁcation. Classes E, F, and G made up all of Maine and New Hampshire
and almost all of Iowa and had relatively higher mean total species richness than classes
A through D, but have relatively lower total species richness than classes H, I, and J
(Figures 3 and 7). Classes H, I, and J lay primarily in the entire Lower Peninsula of
Michigan, the Western part of the Upper Peninsula in Michigan, and Northwestern
Wisconsin; there were also some small regions (with few lakes) in Southern Wisconsin
and Iowa (Figure 7). Classes H, I, and I had the highest total native species richness

across all 360 lakes in the 5-state region (Figure 3).

23

Comparing the Total SR and Native SR Classiﬁcations

The variation explained (PRE) in disturbance residuals across the two
classiﬁcations using a CP of 0.015 (Total SR = 0.602 and Native SR = 0.575) was
relatively similar. The two response metrics were highly correlated (r = .97) and the
hydrogeomorphic predictors and spatial scales that were included and represented in the
classiﬁcations were similar. Both classiﬁcations were driven primarily by lake area
(extinction), EDU (regionalization), and cumulative watershed area (hydrology). Lake
area, an extinction predictor and proxy for lake habitat complexity, split all 360 lakes into
two subgroups (small lakes (< 53 ha) and large lakes (> 53 ha)) and explained a majority
of the variation at the initial split of each classiﬁcation (Total SR = 29%, Native SR =
27%) (Figures 2 and 3). Lake area also explained an additional portion of the variation in
both classiﬁcations in the subgroups of lakes (Total SR = 6%, Native SR = 4%).

In addition to lake surface area, EDU’s explained a signiﬁcant amount of total
variation in each of the classiﬁcations (Total SR =l7%, Native SR = 17%). Beyond the
initial split of lake area in both classiﬁcations, EDU’s split the two main subgroups of
lakes into four subgroups, two subgroups of smaller lakes (<53 ha) and two subgroups of
larger lakes (> 53 ha) (Figures 2 and 3). In smaller lakes for both Total SR and Native
SR, EDU’s primarily classiﬁed lakes into two main groups, Maine/New Hampshire lakes
and Michigan/Wisconsin/Iowa lakes. Although there were a few exceptions, generally
speaking, small lakes in the Northeastern United States had lower species richness than
lakes in the Midwestern states (Figures 4 and 6). Similar patterns were seen in the larger
lakes for both classiﬁcations. Maine and New Hampshire had relatively lower species

richness, although most of Iowa and parts of Michigan and Wisconsin also shared similar

24

EDU patterns and corresponding ﬁsh species richness. In fact for larger lakes, the
highest species richness lake classes were in EDU’s in parts of Michigan and Wisconsin.

In addition to lake area and EDU, cumulative watershed area explained variation
in subgroups of lakes in both classiﬁcations (Total SR = 3%, Native SR = 8%), and mean
base ﬂow (hydrology) split a subgroup of lakes in the Total SR classiﬁcation.
Speciﬁcally, in the Total SR classiﬁcation, of the larger lakes located in classes I and K,
lakes > 40 km2 watershed area (class K) had a greater species richness than lakes < 40
km2 watershed area (class J) (Figure 2). Similarly in the Native SR classiﬁcation, greater
cumulative watershed area resulted in relatively greater species richness when comparing
classes F to G, H to I and J, and I to I (Figure 3).

Lake elevation, an isolation metric related to the probability of a species
introduction, split a subgroup of lakes in the Native SR classiﬁcation. Of the larger lakes,
lakes < 165.4 m in elevation (class E) had lower species richness than lakes > 165.4 m in
elevation (classes F and G), opposite of what would be predicted. For these lakes, it
appeared that other landscape predictors (e. g., lake area, EDU) signiﬁcantly explained the
variation seen in the response metric and lake elevation merely grouped the lake
according to location. Lake elevation, which explained 2% of the variation among lakes,
split lakes close to the ocean in Maine and New Hampshire (lower elevation, class B) and
those higher in elevation in Maine, New Hampshire, and Iowa (classes F and G). The
last isolation metric included in either classiﬁcation was wetland (woody), which split a
subgroup of lakes in the Total SR model (Figure 2). Lakes with greater local wetland
cover > 1.6% (measured within 500 m buffer) had greater species richness (classes H and

1) than lakes with less wetland cover (classes F and G).

25

Discussion

Hydrogeomorphic Features That Explain Variation in Lake Fish Assemblages

Lake area explained over half of the variation in the disturbance residuals in both
classiﬁcations (Total SR = 35% of 60%, Native SR = 31% of 58%); the literature reﬂects
the signiﬁcance of lake surface area in predicting species richness; (r = 0.43-0.81; Rahel
1984, Eadie et al. 1986, Jackson and Harvey 1989). These studies were conducted in a
much smaller region than my study; it appears that lake area may be even more important
when only a smaller spatial scale is being studied. My results follow the literature in that
at every branch where lake area was the primary splitting variable, in both classiﬁcations,
larger lakes had signiﬁcantly higher species richness (less likely of species extinctions)
than smaller lakes. Lake area was highly correlated with other extinction metrics (e. g.,
lake perimeter) so it could be broadly stated that greater habitat complexity (for my 360
lakes) was the primary hydrogeomorphic predictor in shaping either total or native
species richness across a 5-state region. Lakes with larger surface areas tend to have
extended littoral zones allowing for increased habitat and refuge for smaller species of
ﬁsh (e. g., protection from predation). Deeper lakes typically stratify thermally, which
can provide habitat for cool water species in the open water zone (pelagic) of a lake.
With more complex habitat available in larger, deeper lakes, unique species are able to
exist (upon introduction), which supports the potential for large populations and less
vulnerability to extinction. The relationship between greater species richness and greater
habitat complexity has been shown in many other studies; extinction metrics play a large
role in shaping inland lake ﬁsh assemblage (Tom and Magnuson 1982, Magnuson et al.

1998).

26

These studies, however, did not explicitly incorporate regional-scale metrics or
attempt to remove the effects of anthropogenic disturbances. By using a large geographic
area, by including regionalization frameworks, and by accounting for the effects of
humans, other predictor metrics (e.g., EDU) explained some of the variation seen in ﬁsh
species richness. Cumulative watershed area was highly correlated (r > .7) with habitat
complexity metrics (lake area and lake perimeter) (Table 8); it can generally be stated
that a greater cumulative watershed area (relatively higher habitat complexity) may also
result in greater species richness for all sizes of lakes. Cumulative watershed area was
also correlated with the number of surface water connections (r = .69) (Table 8), a strong
proxy for landscape position (isolating metric). Lakes that are higher in the landscape are
less likely to experience species introductions whereas lakes lower in the landscape are
less isolated (higher likelihood of an introduction). Lakes with greater cumulative
watershed area (complex habitat, lower in landscape) exhibit positive correlations with
species richness (Kratz et al. 1997, Magnuson et al. 1998, Riera et al. 2000, Olden et al.
2001); this was exhibited in the Total SR classiﬁcation (Figure 3).

Overall, local extinction and isolation metrics explained the greatest amount of
variation in both total species richness and native species richness. In every case, with
the exception of the lake elevation split in the Total SR model (Figure 3), lakes lower in
the landscape, less isolated, had greater species richness than lakes higher in the
landscape (less likely for an introduction). Lake elevation in a local region (e.g., single
state or group of counties) followed that of landscape position; lower elevations had more
species than lakes at a higher elevation. However, since my study region spanned from

sea level (Maine) to relatively higher elevations in the Midwest (Michigan and

27

Wisconsin), elevation in this study was likely not an appropriate proxy for landscape
position. For example, a lake with a relatively low landscape position in a high elevation
area may be at the same elevation as a lake with a relatively high landscape position in
another watershed.

While local metrics explained a large portion of the natural variation, EDU’s
(large spatial scale predictor) explained a signiﬁcant amount of variation as well (17% in
both classiﬁcations). Across classiﬁcations and lake sizes, EDU’s in Maine and New
Hampshire generally split from those in Michigan, Wisconsin, and Iowa. However,
exceptions existed for smaller lakes in the Total and Native SR classiﬁcations. Some
EDU’s in Michigan (Upper Peninsula), Wisconsin, and Iowa shared characteristics with
those in the Maine and New Hampshire (Figures 4 and 6). Likewise, for larger lakes,
EDU’s in parts of Michigan, Wisconsin, and most of Iowa shared some commonalities
with the Maine and New Hampshire lakes, as reﬂected in the distribution of lake classes
(Figures 5 and 7).

Regionalization frameworks of three different spatial scales were included in
analysis; however, EDU’s were the only regional-scale metric to explain variation in
either Total SR or Native SR. Although freshwater ecoregions were delineated for
freshwater systems to identify distributions of ﬁsh species across all of North America
(Abell et a1. 2000), they appear to be too large in scale (ﬁve regions across ﬁve states) to
signiﬁcantly explain variation in my study lakes. In contrast, EDU’s were delineated
primarily using drainage boundaries such as watersheds (eight-digit HUC’s) and 1
identiﬁed areas with similar biotic patterns (Higgins et al. 2005). EDU’s, generally

between 1,000 and 10,000 km2 in area, are at a scale where watershed boundaries,

28

climate, and landscape features would reveal aquatic biodiversity patterns. Their
consistent inclusion in my lake classiﬁcations, regardless of lake size, reﬂected their
ability to detect broad—scale species richness patterns. The classes of lakes here,
partitioned by a regional-scale predictor, support earlier ﬁndings that demonstrate the
importance of including relatively broad spatial-scale features (Tonn 1990, Poff 1997,
Cheruvelil et al. 2008). Past attempts at classifying lakes have not included
regionalization schemes when building classiﬁcations based on ﬁsh assemblages (e.g.,
Dolman 1990, Schupp 1992). Research on classiﬁcation schemes for ﬁsh assemblages
was conducted before many of the regionalization frameworks were created. Likewise,
these studies were uniquely developed within single states and may not have warranted
including such broad-scale predictors to examine ﬁsh assemblages. Some of the
regionalization frameworks themselves sought to identify unique biodiversity patterns
nationally and internationally. Current research and conservation efforts are focused on
multi-state, and even nationwide collaboration and management (e.g., the Environmental
Protection Agencies’ National Lakes Assessment
(http://www.epa.gov/owow/lakes/lakessurveyl)). Therefore, examining variation of ﬁsh
assemblages across state boundaries warrants the inclusion of regional-scale predictors in
new classiﬁcation schemes.

It is important to note that there are some limitations that I faced when building
these lake classiﬁcations. For example, I utilized the National Land Cover Dataset from
1992, but my ﬁsh assemblage data came from many years following that, as late as 2008.
It is possible that land use/cover changes that occurred between when the two sets of data

were collected inﬂuenced my results. In addition, the stream layer I used to categorize

29

stream connections was at a 1:100,000 resolution. It is possible that I did not capture all
relevant hydrological connections such as small perennial streams that could be sources
of immigration of new species and/or reﬁiges from hypoxia (extinction metric). In
addition, by not completely and accurately identify all hydrological connections, I was
likely not able to properly identify seepage or drainage status for every lake or determine
for all lakes which ones were natural and man-made (reservoirs). Reservoirs have
different hydrological characteristics than natural lakes, including a short water retention
time and typically larger watersheds and surface areas. Lake area and cumulative
watershed area explained a great amount of variation in my lakes; however, speciﬁc
classes had both natural lakes and reservoirs (e.g., Iowa reservoirs shared classes with
natural lakes from other states). Although different types of lakes (e. g., natural vs.
reservoir) can share ﬁsh assemblage characteristics (e. g., same number of species), a
rigorous search into the background of all 360 lakes could have identiﬁed whether it was
natural or man-made. Using continuous data I already have, a metric such as the ratio of

cumulative watershed area to lake area may identify hydrological difference among lake

types.

Differences in Species Richness Between the Northeastern and Midwestern States
The glacier predictor metric I included in CART analysis was not signiﬁcant in
any tree splits. However, the effects from glaciers and local glacial refugia appeared to
play a role in my results, and might have been captured in part by EDU. During the
furthest glacial extent thousands of years ago, ﬁsh in the Midwest (Michigan, Wisconsin,

and Iowa) and parts of Canada had the Mississippi drainage system as a freshwater retreat

30

from lakes being covered in ice. Once the glaciers retreated, ﬁsh were able to colonize or
disperse to previously glaciated areas, resulting in the upper Midwestern states and
Canada becoming relatively speciose (Rempel and Smith 1997). Fish in Maine and New
Hampshire did not have as accessible of a freshwater refuge from glaciers. In fact, many
freshwater species were likely displaced into the Atlantic Ocean (saltwater) and
extirpated due to the inability to quickly evolve with the rapid changes, resulting in the
Northeastern inland lakes having fewer ﬁsh species (Table 5). Once the glaciers revealed
freshwater ecosystems upon their retreat, there were fewer species left behind to colonize
these lake and streams.

As discussed in the introduction, I selected ﬁsh species richness (a measure of
biodiversity) as my response metric over other potential ﬁsh assemblage metrics (e.g.,
ﬁsh abundance, assemblage types). Although other studies have used species
associations in streams (Brenden et al. 2008) or species abundance in lakes (Marshall and
Ryan 1987), I felt that it was more pragmatic to use general species counts; this metric
has been successfully used in other classiﬁcation studies (Magnuson et al. 1998). There
were differences in the biodiversity patterns of the ﬁve states; as discussed above,
glaciations patterns result in Maine and New Hampshire naturally having many less
species than Michigan, Wisconsin, and Iowa. Additionally, there are differences in the
types of species found in different areas; the Northeast has multiple species that are not
present in the Midwest and vice versa. I also used general species counts rather than
species abundance because my ﬁsh assemblage data was from different management
agencies that for the most part, attempt to examine the status and trends of their own

lakes and are typically concerned mainly with recreational species. Therefore, I could

31

not be certain that agencies were not targeting recreational species when selecting lakes
to study or at least speciﬁcally targeting and capturing particular species while sampling
in a lake. Although species richness as a response metric has its drawbacks, species
abundance, in this case, potentially could have had even more biases associated with it.
My decision to use species richness was also based on my inability to control for having
data from ﬁve different states (four different agencies). I used species richness based on
considerations of what would best integrate all ﬁve states and their ﬁsh assemblages.
Species richness is useful because even though management agencies do not always
speciﬁcally concern themselves with biodiversity, they certainly take precautions to
avoid species loss across the state. Some may not be greatly concerned if they stock a
species into a lake that results in the loss of a minnow or sunﬁsh species for that speciﬁc
lake. However, state-wide, there are divisions within state management agencies that are
concerned with biodiversity, maintain a conservative approach, and care for all public
resources, not just those that have recreational value.

In the future, consideration could be given to using ﬁsh abundance or co-
occurrence of species as a response metric. For example, a lake with seven species in
New Hampshire could have a totally different composition than a lake with seven species
in Wisconsin. Tolman Pond in New Hampshire shared four species with Friendship Lake
in Wisconsin, both of which had seven total species and were in the same class.
However, both Pine Lake and Squaw Lake in Wisconsin, also with seven total species
and in the same class, only shared two species with Tolam Pond. By using other
response metrics such as co-occurrence or abundance of ﬁsh species, natural features

may identify different lake classes across such a broad region. However, species richness

32

warrants as much consideration by management agencies, resulted in interesting and
useful classiﬁcations in my study, and has been shown to reveal signiﬁcant features
shaping inland lake ﬁshes (Magnuson et al. 1998)

Although I attempted to look at differences of how well the natural landscape
shaped total species richness versus total native species richness, my response metrics
were likely too similar (r = 0.97) to conﬁdently discern differences between one another ,
which may indicate that my count of native species for each lake was too coarse. I
considered any species that was native to a state to be native to any lake; however, there
are cases where a species may only be native to a few watersheds in a state and non-
native in any other lakes outside those watersheds. Given the way I calculated the
number of native species in a lake across a state, I was not able to identify which species
were truly non-native to a lake. Most of my lakes were estimated as having zero or one
non-native species (~300 lakes out of 360); this is likely not what is present in most lakes
that are affected by some form of human disturbance. A more reﬁned method of
identifying native and non-native species in each lake may reveal a greater presence of
non-natives in my data than what I present here. Ultimately, my two metrics were likely
too similar; more work is required to best identify what natural features speciﬁcally affect

all or just native ﬁsh species.

Conclusions
Despite the shortcomings of my research and some classiﬁcation frameworks in
general, my predictive lake classiﬁcation using different categories of hyrdogeomorphic

and anthropogenic landscape features at multiple spatial scales explained variation in

33

different metrics of ﬁsh assemblages. Given the recent advances in technology, GIS data
can be combined with easily obtainable descriptive lake information to form landscape-
based classes with a variety of predictor variables such as regionalizations,
isolation/extinction metrics, and lake hydrology. Based on known lake-to-landscape
relationships tested using data from sampled lakes, managers can then apply these results
to unsampled lakes to help guide their management decisions. Even though individual
lakes are unique in some respects, they have many local and regional characteristics in
common that help to shape ﬁsh communities.

By examining these shared characteristics across lakes within and even among
states, and identifying the similarities and differences in the ﬁsh communities, managers
can better understand the landscape’s effects on lakes in order to conserve and protect
unique aquatic resources while providing recreational opportunities in inland lakes. More
credence should be given to species richness (i.e., biodiversity) and conserving by
management agencies. Sport ﬁsheries can be created through ﬁsh stocking in lakes with
poor biodiversity, but recreational opportunities are more stable in lakes with high
biodiversity and strong ecosystem-level interactions. Since lake managers are required to
implement ﬁshing regulations and restrictions on all public waters while explicitly having
ﬁsh assemblage data on only a small portion of them, classiﬁcations provide a tool that
can aid those charged with the task of managing inland lakes. Additionally, by dividing
lakes into classes using map‘based characteristics, lake managers may be able to
appropriate resources to other projects otherwise used for rigorous and expensive
sampling programs. Classiﬁcations can also help to indentify regional and national

biodiversity patterns by grouping lakes of relatively lower species richness and higher

34

species richness. Choosing lakes to monitor across very broad regions using a
classiﬁcation approach could also reveal the loss or gain of species due to human impacts
and allow us to better-understand the natural state of aquatic biota in the absence of
humans.

My research explored previously overlooked predictor metrics and their ability to
explain variation in ﬁsh assemblages using a predictive classiﬁcation framework. My
study was unique in that I incorporated many states, each with a wide ranging set of
natural features and human disturbances, as well as regional-scale landscape predictors
such as regionalization frameworks and the number of dams per EDU. My classiﬁcation
results highlight the importance of regional-scale predictors and suggest that they be
considered when attempting to explain the variation seen in ﬁsh assemblages, especially

when using a classiﬁcation approach and a large geographic area.

35

Chapter 2-Using a hydrogeomorphic-based lake classiﬁcation to compare
effects of anthropogenic disturbances on ﬁsh species richness in 360 north
temperate lakes.

Introduction

Lake managers and researchers seek to understand and explain the variation
among ﬁsh assemblages in inland lakes, often trying to distinguish between effects
stemming from natural features and human disturbances. In the absence of human
disturbances, hydrogeomorphic predictors shape in-lake biota. For example, lakes with
greater habitat complexity (e. g., larger lake area), increased potential for species invasion
(e. g., greater number of surface connections past and/or present), and lower hydrologic
position in the landscape (e.g., greater cumulative watershed area) generally have greater
species richness (Tom and Magnuson 1982, Rahel 1984, Kratz et al. 1997, Hershey et al.
2000, and Irz et al. 2004). However, because humans can impact inland lake ﬁsh species
through species additions (e. g., ﬁsh stocking), blockage of movement (e. g., dams), and
landscape alterations (e.g., land use/cover), among other disturbances, the relationship
between the natural landscape and ﬁsh assemblages can be disrupted (Radomski and
Goeman 1995, Carpenter et al. 2007, Johnson et al. 2008). Therefore, understanding how
the hydrogeomorphic landscape structures ﬁsh assemblages must be augmented with an
understanding of how anthropogenic disturbances affect ﬁsh assemblages in inland lakes.

With advances in technology such as Geographic Information Systems (GIS),
researchers have powerﬁrl tools to investigate the effects of land use/cover (e. g.,
agriculture, urban) and human population densities on all aquatic systems. To date, much
attention has been paid to the impacts of humans on water quality and/or ﬁsh

assemblages in lotic systems (Gergel et al. 2002, Buck et al. 2003, Jones et al. 2004),

36

whereas research on lentic systems has lagged behind. However, recent research has
demonstrated that agricultural and urban land within catchments can negatively impact
ﬁsh communities in lakes (Carpenter et al. 2007). Riparian alterations around a lake have
also caused negative impacts to ﬁsh assemblages such as the loss of forested land due to
shoreline development (i.e., increased agriculture land use, higher human populations)
(Jennings et al. 1999, Carpenter et al. 2007). Increased shoreline development has
resulted in the loss of coarse woody debris (CWD) within a lake and directly caused
negative changes to the ﬁsh community (Roth et al. 2007). GIS has been useful in
identifying these lake catchment and riparian alterations which may aid in measuring the
negative effects on ﬁsh assemblages.

In addition to land use/cover, inland lake ﬁsh assemblages can be affected by
roads. Roads are large, ﬂat, impervious surfaces that alter local hydrology and pH (Rahel
1984); additionally, lakes near roads (areas of relatively higher human populations) can
have higher ﬁshing pressure and ﬁsh stocking rates than remote lakes (Kratz et al. 1997,
and Magnuson et al. 1998). Roads have been viewed as a direct source of species
immigration (e. g., lake to nearest road distance), but also as a catchment-wide inﬂuence
from hydrological alterations that may ultimately inﬂuence a lake’s ﬁsh assemblage.

Another disturbance that can affect ﬁsh movement is dams. These structures
fragment aquatic systems and have been found to change species composition of
vegetation and ﬁsh communities in streams (Hill et al. 1998, Brunberg and Blomqvist
2001). In this sense, for lakes connected by streams, movement of species between lakes
is more likely in unfragmented systems than in those with dams blocking movement. Not

only do dams fragment stream connections between lakes, they have also been used to

37

create large impoundments where otherwise a lake would not be present. Recent research
suggests that impoundments have higher likelihood of non-native species invasions than
natural lakes (Johnson et al. 2008). These researchers found that impoundments are
likely to be more accessible to humans than natural lakes (e.g., greater number of boat
landings in impoundments). In addition, impoundments are more likely to be connected
to stream networks and have more surface connections than natural lakes. Therefore, it
appears that dams can both fragment stream systems (i.e., limit movement of native
species) and, through the creation of man-made systems (e. g.,' impoundments), increase
non-native species (Johnson et al. 2008).

Perhaps the most profound and well-established anthropogenic disturbance to
affect ﬁsh assemblages in inland lakes is ﬁsh stocking. Research on ﬁsh stocking is quite
extensive both at the individual species level and ﬁsh assemblage levels (Anderson and
Schupp 1986, Radomski and Goeman 1995, Holmlund and Hammer 2004, Irz et al.
2004). Adding a species to a lake (stocking or unintended release) can result in altered
food web dynamics, the elimination of a species from a system (through competition or
predation), indirect changes in water clarity, and changes in ﬁsh growth rates (Anderson
and Schupp 1986, Radomski and Goeman 1995, Holmlund and Hammer 2004, Irz et al.
2004). For example, Irz et al. (2004) found native ﬁsh species richness and non-native
species richness to be negatively correlated, indicating that introductions have led to
saturated communities (limited potential of ﬁsh species richness to increase). Although
ﬁsh stocking across the landscape is primarily intended to help manage faltering
ecosystems and/or create new ﬁsheries, stocking can result in unintended negative

consequences that alter ﬁsh assemblages in inland lakes.

38

Some human disturbances cause an increase to total species richness (e. g., ﬁsh
stocking, road proximity), while other disturbances result in a decrease in species
richness by fragmenting lake connectivity (e.g., dams). Since anthropogenic disturbances
vary across the landscape (e. g., dam densities, stocking rates within and across states),
and because features of the natural landscape may mediate effects of human disturbances
on ﬁsh assemblages, impacts from human disturbances on inland lake ﬁsh assemblages
should be measured more broadly than in a single lake or even single state and should
measure the effects of disturbances on groups of naturally similar lakes (i.e. ﬁsh
assemblages). However, to date, research investigating effects of human disturbances on
ﬁsh assemblages has typically looked at a particular species, only in-lake features (such
as CWD), or a small set of water bodies in a single state (Rahel 1984, Roth et al. 2007).
Therefore, I hypothesize that measuring effects of anthropogenic disturbances on ﬁsh
assemblages in classes of naturally similar lakes across a broad region will reveal unique
patterns otherwise masked by lake heterogeneity.

In order to measure the effects of anthropogenic disturbances on subgroups of
naturally similar lakes (lake classes), I ﬁrst developed a hydrogeomorphic-based lake
classiﬁcation system (that accounted for anthropogenic disturbances) using only the
native ﬁsh species within each lake (see Chapter 1 for details). I chose native species
richness (as opposed to species abundance or assemblage type) to make a more
straightforward comparison between different regions, each with different ﬁsh species
composition. In this effort, I identiﬁed the most signiﬁcant natural predictor variables
across 360 lakes, subdividing the original group of lakes into 10 classes, ranging in size

from 10 to 69 lakes. The classiﬁcation system was driven primarily by lake area,

39

cumulative catchment area, and the ecoregion (ecological drainage unit, EDU; Higgins et
al. 2005) in which each lake was located (Figure 3). The mean native species richness
across classes ranged from a low of 5.1 in Class A to a high of 20.4 in Class J (Figure 3
described in Chapter 1). In this chapter, I then sought to compare the effects of
anthropogenic disturbances on total native species richness within each lake class.
Because lake area was the primary variable in splitting all lakes into two subgroups,
small lakes (< 53 ha, Classes A, B, C, and D) and large lakes (> 53 ha, Classes E, F, G,
H, and I) (Figure 3 described in chapter 1), I also compared the effects of anthropogenic
disturbances between lakes in these broader size-based groupings.

Lastly, to look at ﬁsh assemblage responses in more detail than total native
species richness as affected by human disturbances across sizes or classes of lakes, I
determined what disturbance variables affected speciﬁc groups. To do so, I assigned ﬁsh
species to one of two groups: recreational or non-recreational. These two categories of
ﬁsh likely function somewhat differently in ecosystems, and they often are associated
with contrasting social values. Recreational species often are highly sought after by
humans, hold relatively high economic value, and are typically at higher trophic levels.
Non-recreational species have little economic value and hold lower trophic status; in
some cases, they are conserved only to maintain biodiversity in a lake. Because each ﬁsh
type has different ecosystem and social value, I hypothesized that recreational and non-
recreational species would differ in their response to a given disturbance. For example,
walleye (recreational) and emerald Shiner (non-recreational) may both be native to a
given state, but each holds unique values within that state. In this sense, a speciﬁc

anthropogenic disturbance (e.g., ﬁsh stocking) may affect recreational native species

40

differently than non-recreational native species in a set of lakes within or across states.
Disturbance metrics such as road density or human population density, which have been
attributed to ﬁshing pressure (Magnuson et al. 1998), is likely attributable to the presence
of recreational species in a lake rather than non-recreational species. Therefore, these
species disturbances may affect recreational species differently than non-recreational.
Therefore, I measured the effects of human disturbances on recreational native species
and non-recreational native species within each of my 10 hydrogeomorphic-based lake

classes.

Methods

Selecting a Lake Classiﬁcation from Chapter 1

In Chapter 1, I created two classiﬁcations of lakes based on the hydrogeomorphic
landscape using total species richness (Total SR) and native species richness (Native SR)
as response variables, while accounting for anthropogenic disturbances (Chapter 1:
Figures 2 and 3). Both classiﬁcations explained 67% of the total variation in species
richness (combined variation explained by multiple regression and CART analysis), had a
similar number of lake classes (Total SR, 11 classes; Native SR, 10 classes), and each
included very similar natural landscape predictors (e.g., lake area, EDU). Because the
Total SR and Native SR classiﬁcations were driven by similar predictors, contained lake
classes located in relatively similar areas of the 5-state region (Chapter 1: Figures 4, 5, 6,
and 7), and explained variation equally well, neither classiﬁcation was objectively
deemed “better” than the other. However, since Chapter 2 is primarily focused on

studying how humans have impacted “natural” lake ﬁsh assemblages, I used the Native

41

SR classiﬁcation in Chapter 2. The Native SR classiﬁcation of 360 lakes is made up of
10 classes with membership ranging from 10 to 69 lakes across Maine, New Hampshire,

Michigan, Wisconsin, and Iowa (Chapter 1: Figures 3, 6, and 7).

Fish Assemblages of the 10 Lake Classes

Each of the 10 ﬁnal classes from the Native SR classiﬁcation from Chapter 1
(Figure 3) was comprised of a relatively unique range and mean value of total native
species richness. Because all lakes spanned across a wide spatial range and contained
such a variable assemblage of ﬁsh species from state to state, I ﬁrst wanted to determine
if classes differed in terms of their predominant ﬁsh species. To do so, I determined the
most prevalent species in each class, and if the most prevalent species were common for
the vast majority of the lakes in each class. I conducted this analysis to determine if most
of the lakes in a class contained the same three or four most common species (indicating
some similarity in species composition in that class) or whether species composition,
even of the most prevalent species, was highly variable among lakes in a class. Lastly, I
compared the most prevalent species across classes to see if similar species were most
prevalent in several classes, or whether the identity of predominant species could be used
to distinguish among lake classes. For all species assemblage comparisons within and
across classes, I created frequency histograms and displayed the percentage of lakes in

each class for which each species was present.

Native Species Richness Response Metrics

42

By state, I classiﬁed each ﬁsh species as either recreational or non-recreational. I
used each state’s recreational ﬁshing guide (published by the state’s ﬁsh management
agency) as a standard, (Maine, Department of Inland Fisheries and Wildlife 2008; New
Hampshire, Fish and Game Department 2009; Michigan, Department of Natural
Resources 2009; Wisconsin, Department of Natural Resources 2009; and Iowa,
Department of Natural Resources 2009). If a species had a bag/size limit or
seasonal/ gear restriction associated with it, it was declared recreational; ﬁsh with no
restrictions were deemed non-recreational (Table 10).

In total, three ﬁsh species response metrics were created, total native species
richness (SR), native recreational SR, and native non-recreational SR (Table 10). Total
native SR was used as the response variable to analyze the effects of disturbances
between small and large lakes (Table 1 1) and within each of the 10 classes of lakes
created in Chapter 1. Within an individual lake class (Chapter 1: Figure 3; Classes A
through J), native recreational SR and native non-recreational SR also were used to
compare the effects of human disturbances on these different ﬁsh types (Tables 12 and

13).

Effects of Anthropogenic Disturbances on Fish Species Richness

I transformed all anthropogenic disturbance and ﬁsh response metrics to meet
normality assumptions for regression models including arcsine-square root
transformation for land use/cover variables and natural log, squared, or square root for all
other variables. For metrics that did not meet normality assumptions, even after

transformation (approximately 75% of the variables did not meet the assumptions for any

43

of the transformations), 1 selected the data transformation that resulted in the distribution
most approaching normality and with the fewest number of outliers. In short, because for
some metrics there was not a transformation that completely worked, I used the
transformation that did the “best relative job”.

In order to measure the effects of human disturbances on ﬁsh species richness, I
ﬁrst ran multiple linear regressions of total native SR against non-highly-correlated (r <
0.6) human disturbance variables that included metrics of ﬁsh stocking, dam density, land
use/cover, road density, and human population density (Tables 14 and 15). Best model
selection was used to determine which disturbance variables were signiﬁcant (p <0.05)
and to include in the ﬁnal model (SAS version 9.1). I used condition index (CI) and
variance inﬂation factors (VIP) to determine whether multicolinearity among predictor
variables was unduly inﬂuencing my models. Variables with CI values greater than 10
and VIP values greater than 30 in the regression were removed from the model (Hair et
al. 1995). The best model had both the highest adjusted R-squared and lowest Akaike's
information criterion (AIC) value.

I ran multiple linear regression models using disturbance variables with total
native SR as the response metric to determine any effects (positive or negative) of human
disturbances. Models were run for each of the Native SR lake classes from Chapter 1
(Figure 3) and for each of the two broader groups of small lakes (< 53 ha, 133 lakes) and
large lakes (> 53 ha, 227 lakes). Lastly, to explore how different species vary in their
response to disturbances, I conducted the same analysis for recreational and non-
recreational species within each of the 10 classes. To compare the relative ability of the

models to explain variation in species richness, weighted R-squared values were

44

calculated by multiplying the number of lakes in a class (or size group) by the total R-
squared value in that speciﬁc class (or size grouP), adding those values all together, and

then dividing by all lakes (360).

Results

Fish Species Assemblages in 10 Lake Classes

Across all 10 classes, sunﬁshes such as bluegill and pumpkinseeds were
“dominant” species (at least one was in the top three species by frequency in every class);
largemouth bass and black crappie were consistently “dominant” as well (Figures 8
through 17). Brown bullhead, yellow perch, golden shiner, walleye, chain pickerel and
northern pike were also some of the more prevalent species; each of these species was
one of the top three “dominant” species in at least one class (Figures 8 through 17). Eight
classes of the 10 were comprised of dominant species in a vast majority of the lakes
(Classes B, C, D, E, F, H, 1 and J), at least 83 % of the lakes (Class F) and as high as 100
% of the lakes in two other classes (Classes D and J). Meaning, for a given class, the
most dominant species were typically the same three or four species in most of the lakes
rather than high variability in assemblages lake to lake in a class. Because the most
prevalent species in these eight classes were common to so many lakes in the class, it
indicates a lack of within-class variability of assemblage type (as deﬁned by the most
prevalent species). In fact, Classes B, C, D, and F had the same three species as their
most frequent: largemouth bass, bluegill, and black crappie (Figures 9, 10, 11, and 13).
Therefore, what distinguishes these classes is not their most common species (as

determined by presence/absence), but rather the presence/absence of rarer taxa.

45

Classes A and G were more variable in species composition (Figures 8 and 14).
The most prevalent (frequently observed) species (Class A, brown bullhead, golden
shiner, pumpkinseed, yellow perch; Class G, bluegill, largemouth bass, yellow perch)
were in fewer than half the lakes for Class A and approximately 63 % of the lakes in
Class G (Figures 8 and 14). This result suggests that compared to the other eight classes,
these two classes, A and G, have more variability in species composition across their
lakes. Because these classes contain lakes from all ﬁve states; there are many
assemblage types within the class with the dominant species being highly variable. The
other eight classes have lakes from only four states (Classes C and F), three states
(Classes B, D, H, 1, and J) and two states (Class E), indicating less variability in

assemblage types and therefore more consistent dominant species for a given class.

Comparing Disturbance Effects Among Lake Classes

My exploration of how anthropogenic disturbances were associated with total
native species richness across lake classes included ten class-speciﬁc models. Of these
models, eight had one to three signiﬁcant predictor variables resulting in adjusted R-
squared values ranging from 0.17 to 0.38 across the classes (Table 16). Two class-
speciﬁc models (B and J) had no signiﬁcant predictor variables, four classes (A, D, E,
and F) had a single signiﬁcant predictor, two classes (H and I) had two signiﬁcant
predictors, and two classes (C and G) had three signiﬁcant predictors (Table 16). The
total weighted adjusted R-squared across all 360 lakes (10 classes) was 0.20; 20% of the
variation of total native SR in the natural lake classes was explained by anthropogenic

disturbances (Table 16). Human population density was positively associated with six

46

different lake classes (small and large lakes) and road density was the only disturbance
metric to have no signiﬁcant association with any lake classes (Table 16). The number of
species stocked (Class G), dam density per EDU (Class E), dam density per 12-digit
HUC (Class C), agricultural land use (Class H), urban land use (Class I) and agricultural
and urban combined land use (Class E, F, and G) all had a negative association with total

native SR in some lake classes (Table 16).

Comparing Disturbance Effects Between Small and Large Lakes

My exploration of how anthropogenic disturbances were associated with total
native species richness across a large range of lake sizes included two main regression
models based on lake size. The small lake (< 53 ha) regression model had three
signiﬁcant predictors; number of dams per 12-digit HUC (negative), road density
(positive), and human population density (positive) together explained 21 % of the
variation in total native species richness across 133 lakes (Table 17). The large lake (>
53 ha) regression model had ﬁve signiﬁcant predictors. The number of dams per 12-digit
HUC (negative association), road density (positive association), and human population
density (positive association) had the same impacts on large lakes as small lakes and the
number of species stocked (negative association) and agricultural land use (negative
association) additionally affected large lakes (Table 17). However, even though there
were two additional human disturbance metrics in the large lakes model, the ﬁve
signiﬁcant metrics collectively explained only 12 % of the variation across 227 lakes

(Table 17). The total weighted adjusted R-squared across the two lake size groups is

47

0.15; 15% of the variation of total native SR is explained by anthropogenic disturbances

(Table 17).

Comparing Disturbance Effects on Recreational and Non-Recreational Species
Richness

My exploration of how anthropogenic disturbances have affected native
recreational and non-recreational species richness included ten class-speciﬁc models. Of
the native recreational species richness models, eight of them had one to three signiﬁcant
predictor variables resulting in adjusted R-squared values ranging from 0.08 to 0.71
across the classes (Table 18). Two class-speciﬁc models (I and J) had no signiﬁcant
predictor variables, two classes (F and H) had only a single signiﬁcant predictor, three
classes (A, B, and C) had two signiﬁcant predictors, and three classes (D, E, and G) had
three signiﬁcant predictors (Table 18). The total weighted adjusted R-squared across all
360 lakes (10 classes) was 0.25; 25 % of the variation of native recreational SR in the
natural lake classes was explained by anthropogenic disturbances.

In the native non-recreational species richness models, eight of them had one to
three signiﬁcant predictor variables resulting in adjusted R-squared values ranging from
0.14 to 0.81 across the classes (Table 19). Two class-speciﬁc models (B and J) had no
signiﬁcant predictor variables, one class (C) had only a single signiﬁcant predictor, ﬁve
classes (A, D, F, H, and I) had two signiﬁcant predictors, and two classes (E and G) had
three signiﬁcant predictors (Table 19). The total weighted adjusted R-squared across all
360 lakes (10 classes) was 0.33; 33 % of the variation of native non-recreational SR in

the natural lake classes was explained by anthropogenic disturbances.

48

In smaller lakes (Classes A, B, C, and D), human disturbances affected native
recreational species differently than native non-recreational species; these differences
were seen in the signiﬁcant predictors in the ﬁnal regression models. Native recreational
SR was inﬂuenced primarily by the negative associations with dam density (exception
Class D) and positive associations with urban land, agriculture and urban land combined,
road density, and human population density (Table 18). Native non-recreational SR was
also positively associated with road density and human population density; however, land
use/cover metrics (e.g., agriculture, urban, and agriculture and urban combined) had
negative associations (Table 19).

In large lakes (Classes E, F, G, H, I, and J), native recreational SR was affected by
negative associations with ﬁsh stocking and the combined land cover/use of agriculture
and urban (Table 18). Native recreational SR was positively associated with dam
density, agriculture, road density, and human population density. Native non-recreational
SR was negatively associated with darn density for some classes (F and G) but positively
associated with Class H (Table 19). Combined agriculture and urban land use/cover also
had differing associations depending on lake class for native non-recreational SR,
positive in Class E, but negative in Classes F and H (Table 19). Agriculture and urban
land cover/use as individual disturbances were negatively associated with native non-
recreational species; human population density was a positive association. Overall,
within a given class, recreational species and non-recreational species were affected
differently by anthropogenic disturbances. Because these two types of ﬁsh hold such

different social values and ecosystem functions, humans appear to have varying

49

associations with different native species rather than consistent impacts across all ﬁsh in

inland lakes.

Discussion

Measuring Anthropogenic Disturbance Effects on Native Species in Lake Classes

Almost all of the anthropogenic disturbances included in my models (all variables
but road density and human population density) were negatively related to total native SR
(for the eight classes with signiﬁcant models). An increase in the number of ﬁsh species
stocked was associated with a decrease in native ﬁshes (Class G); literature has shown
that an introduction of non-native species into lakes has caused declines in native species
abundance (Irz et al. 2004). My metric of stocking (number of species) included all
native and non-native species, therefore, it can not be concluded that the negative impact
on native ﬁshes was due solely to stocking non-native species. However, because
stocking can result in altered food web dynamics, and the loss of a species from a system
(through competition or predation), among many other changes, ﬁsh stocking can
certainly have the effect of lowering species richness in lakes (Anderson and Schupp
1986, Radomski and Goeman 1995, Holmlund and Hammer 2004).

Dam densities had similar associations as ﬁsh stocking (negative) in Classes C
(per 12-digit HUC) and E (per EDU) (Table 16). Johnson et a1. (2008) found that non-
indigenous species were more likely to invade impoundments (dam created) than natural
lakes; this certainly has repercussions on native species. Darn density, either scaled in

12-digit HUC’s or EDU’s, had a negative association with total native SR. Dams

50

fragment systems that would otherwise be connected in their absence, causing changes to
the species composition (Hill et al. 1998, Brunberg and Blomqvist 2001).

Individual land use/cover metrics of agriculture (Class H) and urban (Class I), as
well as their combined use/cover (Classes C, F, and G) all impacted total native SR
negatively (Table 16). Agriculture was highly correlated with agriculture and urban
combined (r = 0.96). Further, forest land cover was highly negatively correlated with
agriculture (r = -0.85) and agriculture and urban combined (r = -0.89). Literature has
shown that loss of forested land in the riparian of a lake (i.e., loss of CWD) can result in
changes to the ﬁsh assemblage in a lake (Roth et al. 2007). In Classes C, F, and G,
higher agriculture and urban combined land use (i.e., lower forested land cover) had
negative associations with total native SR. Lakes with the highest relative total native SR
(by class) were mostly found in Michigan and Wisconsin (i.e., higher forest, lower
agriculture) whereas less species-rich lakes were mostly in Iowa (i.e., lower forest, high
agriculture) (Chapter 1: Figures 6 and 7). However, it can be generally stated that Iowa
had lower species richness overall than Michigan and Wisconsin, regardless of lake class,
so agricultural and urban land were likely not the only factors, natural or human, causing
differences between the two broad regions.

Human population density appeared positively associated with total native SR
across the entire ﬁve state region; higher human population densities in six classes (small
lakes, A, C, D; large lakes, G, H, and 1) correspond to higher native SR. While road
density did not signiﬁcantly affect any class, lakes near roads (i.e., areas of relatively
higher human populations) can have impacts on ﬁsh assemblages through higher ﬁshing

pressure (boat movement and bait buckets) and intentional ﬁsh stocking rates as

51

compared to remote lakes (Rahel 1984, Kratz et al. 1997, Magnuson et al. 1998). While
species from these introductions were often non-native to the speciﬁc lake, it was likely
the species was native to the state. Given that we determined native/non-native status for
each species by state, an increase in total native SR in areas of higher human population
densities could actually be a species native to the state (non-native to the lake) that has
been moved from one lake to another, increasing total native SR in lakes near human
population centers. In addition, humans may choose to live near lakes with higher
species richness; therefore, it is not necessarily that higher SR is an effect from humans

but an association with higher human populations.

Measuring Anthropogenic Disturbance Effects on Native Species in Small and
Large Lakes

The number of species stocked was negatively associated with total native SR in
large lakes while dam density (per 12-digit HUC) was negatively associated with total
native SR in both sizes of lakes. These results corroborate the literature in that stocking
can have negative impacts on a lake’s ﬁsh community while dams fragment systems and
often are associated with impoundments (i.e., higher human access) (Irz et al. 2004,
Johnson et al. 2008) (Table 17). Agricultural land use had a negative association with
total native SR in larger lakes, potentially due to the inverse relationship agriculture had
with forested land cover (r = -0.85) (data not shown). Many of the lakes with relatively
lower total native SR were located in Iowa (i.e., high agriculture, low forest) while the
- lakes with the highest relative native SR were in Michigan and Wisconsin, where there is
less agricultural land (Chapter 1: Figure 6). These results reﬂect the literature; areas of

lower forested cover (i.e., higher agriculture, increased shoreline development) have led

52

to changes in ﬁsh community structure (Jennings et al. 1999, Carpenter et a1. 2007). This
does not imply that Iowa was once vastly covered by forest and now is completely altered
by humans to be covered with agriculture, but broad-scale patterns of species richness
(e.g., low versus high native species richness) may reﬂect the broad land cover types that
currently exist in our study region.

Road density and human population density were positively associated with total
native SR in small and large lakes (Table 17). A lake’s proximity to roads has previously
been shown to negatively impact species richness (Rahel 1984, Magnuson et al. 1998).
However, since road density was moderately correlated with human population density in
our study (r = 0.52, data not shown), it is entirely possible that road density is exhibiting
the same positive effects on total native SR as human population density (Tables 16 and
17). Additionally, our metric for roads was calculated as the road density around the lake
(within a 500-m buffer), not distance to the nearest road (Rahel 1984, Magnuson et al.
1998). Our measure of road density around a lake may actually better-reﬂect the
likelihood of human access to lakes because as road density increases it is likely that
access is easier, potentially leading to a greater likelihood of species richness to increase
through introductions. Similarly to what we found with road density, as human
populations increased around small and large lakes, total native SR increased (Table 17).
Regardless of lake size, native species richness was higher in areas with increased road
and human population density, increased accessibility of humans to lakes was positively

associated with greater biodiversity.

53

Examining Recreational and Non-Recreational Species Within Lake Classes

The effects of anthropogenic disturbances explained the variation in native
recreational species (R2 = 0.25) and native non-recreational species (R2 = 0.33) across all
classes of lakes (Tables 18 and 19), the variation explained in total native SR was 20 %
(Table 16). Human disturbances explained over 50% of the variation in recreational
native SR for Classes D, E, and G and over 50% of the variation in non-recreational
native SR for Classes A, E, F, and G. Conversely, the most variation human disturbances
were able to explain in total native SR was only 38% (Class F). More variation was
explained by human disturbances in the Native SR lake classes when the speciﬁc type of
species was identiﬁed for all native ﬁsh (recreational or non-recreational), rather than
total native SR.

Speciﬁcally comparing recreational and non-recreational species within classes,
the effects of human disturbances explain recreational native SR better in classes B (R2 =
0.43) and D (R2 = 0.71) than non-recreational native SR (Class B, R2 = Non signiﬁcant
model; Class D, R2 = 0.34) (Tables 18 and 19). Conversely, the effects of human
disturbances explained a greater amount of variation in non-recreational native SR (Class
E, R2 = 0.81; Class F, R2 = 0.59) than recreational native SR (Class E, R2 = 0.52; Class F,
R2 = 0.08). The same trend could be seen in Class A where native non—recreational SR
(R2 = 0.55) was explained much better by disturbances than was native recreational (R2 =
0.22). Because non-recreational species were more prevalent in lake classes than
recreational species (Figures 8 through 17), variability was higher within a class for non-

recreational species. In this sense, the greater within-class variability of non-recreational

54

species likely made them a better response metric that total native SR or native
recreational SR for measuring the effects of human disturbances.

Recreational species dominated individual lakes in almost all classes by
frequency; for the most part, recreational species comprised the top three or four most
frequent species in a lake. However, non-recreational species far outnumbered
recreational species in every state (Table 10), therefore, it was reasonable to expect that
more variation would be explained using non-recreational species as a response metric.
Non-recreational species were more frequent for an entire class, however, unique taxa of
families such as darters, minnows, madtoms, sculpins and shiners were only present in a
few lakes; whereas, recreational species were typically the dominant species in almost all
lakes, (Figures 8 through 17). In fact, many non-recreational species occurred in only a
few lakes within a class, many of which were likely intolerant of human disturbances.
Other researchers have shown that human disturbances are likely to affect species
intolerant of human-induced stress; these species are therefore strong biotic indicators of
anthropogenic disturbance (Drake and Pereira 2002). The presence of these unique
species appeared to identify lakes with high biodiversity, a value many lake managers
strongly consider when making decisions. Non-recreational species appear to be a key
group of ﬁsh that managers and researchers could examine within speciﬁc lake classes
for management in the face of anthropogenic disturbances. Ultimately, they may help to
indicate what lakes human-induced stressors are affecting intolerant species, therefore

warranting conservation efforts to maintain biodiversity.

55

Conclusions

The effects of human disturbances on my study lakes were generally consistent
with the literature, although they were exhibited at a much broader scale than previous
studies. The disturbances that had generally positive associations with ﬁsh species
richness were road density and human population density. With greater accessibility to
lakes (i.e., more roads and people), ﬁshing pressure and stocking rates have been shown
to increase, therefore, likely more accidental ﬁsh introductions through bait buckets (i.e.,
ﬁshing) (Kratz et al. 1997, and Magnuson et al. 1998). Dams (local scale and regional
scale), agricultural and urban land use/cover, and ﬁsh stocking were negatively
associated with ﬁsh species richness in my study lakes. Dams fragment connectivity of
freshwater systems, restricting the movement of ﬁsh to naturally colonize a new lake, and
cuting off access to stream refuge for ﬁsh during winter (hypoxic) conditions. Shoreline
development (i.e., increased agriculture and urban land and corresponding loss of forest)
results in the loss of habitat (CWD) for ﬁsh and likely causes negative impacts through
increased nutrient inputs and alterations to the near shore zones (e.g., removal of
macrophytes).Stocking of non-native species (either to a lake or even entire state (e.g.,
brown trout)) may initially increases species richness, although long-term food web
dynamics and loss of niche space can end in the loss of other species (i.e., decrease in
species richness). The metrics I created using the best, readily-available stocking data do
not account for all purposeful species introductions into lakes. For example, I acquired
state agency stocking information, and for Michigan and Wisconsin, these data were
post-1970. However, stocking is known to have been conducted long before then, and in

most states, private stocking programs likely result in introductions for which agencies

56

 

have no records. Besides intentional stocking, unintentional introductions (e.g., bait
buckets, boat movement) also affect the distribution of species and surely inﬂuence the
overall distribution of ﬁsh species across my ﬁve states. Altogether, my estimates of ﬁsh
stocking were likely conservative and did not capture all forms of human-induced
introductions of ﬁsh species (native or non-native) into inland lakes.

Many different ﬁsh response metrics were considered for analysis, ultimately, I
decided on using species richness. Because of the large spatial scale of this study and
how much ﬁsh distributions vary across all ﬁve states, there was high variability in the
ﬁsh assemblages among all lakes and classes. Because of this, it seemed most practical
to use species richness rather than abundance of each species or assemblage type.
Species richness ultimately drives conservation approaches to management (e. g.,
maintaining biodiversity) and helps to account for the distributional differences of ﬁsh
species that is inherent when researching lakes across 5 states. Non-recreational species
appear to vary across and within classes making them more useful for identifying
disturbances than recreational species. Species abundance in lakes where recreational
ﬁshing is prevalent is often driven by management decisions to maintain opportunities to
catch game species. This fact likely compromises the use of recreational species richness
as a response metric because each agency has a different ﬁsh assemblage to manage and
constituents for which to manage. Different recreational values such as bass ﬁshing
versus trout ﬁshing makes comparing lake-to-lake species abundances more difﬁcult than
a general species count. Managing with biodiversity (species richness) in mind crosses

political boundaries and somewhat mitigates differences in species distributions and

57

abundances across such a large region like in this study, therefore, species richness was
selected as the sole response metric.

Future research of relationships between anthropogenic disturbances and ﬁsh
species richness might also beneﬁt from analyses such as general additive models (GAM)
to identify such non-linear relationships. For example, I would expect that non-linear
relationships might be revealed in ﬁsh stocking because adding a new species into a lake
has a positive effect on species richness. However, continued addition of new species
can alter ecosystem dynamics leading to an overall species richness decrease. Because
human disturbances can likely have non-linear relationships with ﬁsh assemblages, future
research should consider additional analysis to linear regression.

In conclusion, my research highlighted the ability to identify speciﬁc effects of
humans on inland lake ﬁsh assemblages; species richness was positively and negatively
associated with a wide range of anthropogenic disturbances. Lake managers could use
this information to help mitigate the loss of species and conserve biodiversity;
additionally, managers may able to create and improve species richness using other
management tools. For the most part, despite the many varying effects (positive and
negative) on ﬁsh within a class, lakes were dominated primarily by the same three or four
species. Therefore, future research and management may be able to identify these
dominant species and make inferences as to how speciose a lake might be given the
presence or absence of a few key species. This may help to reduce the amount of in-lake
sampling is needed to make decisions across a large number of lakes. It was more

effective and informative to measure human disturbances in classes of lakes rather than

58

across all lakes; this would also help to make decisions for a large number of lakes while

only have resources to collect in-lake data on a subset.

59

Thesis Conclusions

In Chapter 1, I successfully developed lake classiﬁcations for total and native
species richness using hyrdogeomorphic features of the landscape at multiple spatial
scales. In the future, managers can use classiﬁcations such as these to assign lakes for
which only hydrogeomorphic data are available to classes. Class-speciﬁc management
options can then be pursued by managers, allowing them to incorporate the natural
variation among lakes even when it is difﬁcult to obtain in-lake data. Even though
individual lakes are unique in some respects, they share similar characteristics due to
having similar local and even regional characteristics with other lakes which collectively
shape ﬁsh communities. My research and lake classiﬁcations were not necessarily
intended to compare results with other studies (indeed the variety of approaches and
variables used across studies challenges statistical comparisons of classiﬁcation systems),
but rather to explore previously overlooked predictor metrics and their ability to explain
variation in ﬁsh assemblages in a classiﬁcation.

My study was unique in that I incorporated multiple states, each with a wide
ranging set of natural features and human disturbances. Regional-scale predictors, which
have been incorporated into classiﬁcation efforts relatively recently and lack thorough
evaluation, were useful in explaining variation in species richness, and valuable for
creating disturbance metrics such as the number of dams per EDU. Even though the
primary splits in all classiﬁcations were local features, regional-scale predictors also were
included in several of the classiﬁcations, highlighting the importance of regional-scale

predictors.

60

To investigate the effects of human disturbances on inland lake ﬁsh assemblages,
it was more useful to quantify these relations within classes than across all lakes. 1 found
that the effects of human disturbances such as dams (local scale and regional scale),
agricultural land use and urban land use, and stocking all are associated with lower
species richness. In contrast, road density and human population density generally were
associated with higher species richness. Species richness was a useful response metric
for studying lakes across ﬁve states and native non-recreational species richness, in
particular, appeared to be the best indicator of human disturbances on lakes of those I
considered. Non-recreational species were the most prevalent by frequency in all classes;
presence and absence of unique taxa therefore helped to identify patterns of biodiversity.
Species richness should be a concern of management agencies for conservation goals,
even while working to ensure the presence of recreational species for sport ﬁsheries. In
conclusion, lake classiﬁcations should be considered by state agencies as an approach to
managing a large number of lakes while having information on a subset. By dividing
lakes into classes using map-based characteristics, lake managers may be able to: (a)
appropriate resources otherwise used for rigorous and expensive sampling programs to
other projects, and (b) make somewhat informed decisions about the management of
individual lakes, even when in—lake data are lacking.

My research highlights the importance of within and among management agency
collaboration, or lack there of. I was able to collect natural landscape and human
disturbance data on more than 2,300 lakes. In contrast, ﬁsh assemblage data were only
available for approximately 550 lakes across all ﬁve states, reﬂecting the fact that ﬁsh

assemblage data are more difﬁcult to gather than map-based features. Even within a

61

single state, agencies had non-overlapping monitoring programs, meaning that agencies
collecting water quality data did not coordinate with agencies collecting ﬁsh data in order
to sample the same lakes. As a result, nearly 200 lakes of my 550 lakes with ﬁsh
assemblage data lacked the other data that I needed for my analysis, ultimately because
they had not been sampled by the water quality management agency. From the
standpoint of developing classiﬁcation models such as those described here, coordinated
monitoring efforts across agencies would be extremely beneﬁcial. Efforts such as the
Environmental Protection Agencies’ National Lakes Assessment help align multi-state
and even nationwide collaboration and management, which I view as a step in the right

direction.

62

Appendix

Tables

Table l. The number of lakes in each state for which data were obtained according to
variable category. ‘Overlap’ indicates the lakes used in analysis because complete data
were available.

 

 

 

Variable Category ME NH MI WI IA TOTAL
Hydro geomorphic 569 740 486 426 127 2348
Anthropogenic 658 725 637 426 127 2573
Fish Assemblage 65 31 125 219 115 555
Overlap 28 27 105 90 110 360

 

63

Table 2. List of hydrogeomorphic variables (regionalization, extinction/isolation,
hydrology, morphometry, land use/cover) and their measurement units arranged
according to category. NA indicates a non-continuous variable and therefore does not
have a median, minimum, or maximum. Land use/cover coverage’s were calculated
within a 500-m buffer of the lake.

 

 

Category/Name Median Minimum Maximum
Regionalization
6-Digit Hydrologic Unit Code (HUC) 97 Total Regions
Ecological Drainage Unit (EDU) 28 Total Regions
Freshwater Ecoregion 5 Total Regions
Isolation
Lake Glaciated or Not Glaciated NA NA NA
Lake Seepage or Drainage NA NA NA
# of Lakes within an EDU 1,549 159 2,856
# of Lakes within a 12-digit HUC 9 1 89
# of Surface Water Connections 2 0 37
Lake Elevation (m) 303 17 570
Extinction
Lake Area (ha) 84.4 3.3 6,039.1
Lake Maximum Depth (m) 8.5 1.2 40.8
Lake Mean Depth (m) 3.1 0.8 12.2
Lake Perimeter (m) 6,444 722 185,962
Hydrology
Mean Base Flow (%) 57.0 14.0 89.0
Mean Precipitation (cm/yr) 84.7 65.4 124.3
Mean Runoff (cm/yr) 27.9 5.1 76.2
Watershed Cumulative Catchment Area (kmz) 11 0.1 31,080
Land Use/Cover (%)
Water 2.4 0 35.2
Forest 49.8 .6 99.7
Grassland/Field 1.0 0.0 36.4
Wetland Woody 3.9 0.0 71.8
Wetland Herbaceous 1.4 0.0 16.3
Wetland Combined 5.7 0.0 74.9
All Other 0.0 0.0 11.0

 

64

Table 3. List of anthropogenic disturbance variables and their measurement units (land
use/cover, dams, roads, ﬁsh stocking, human populations) arranged according to
category. Land use/cover coverage’s were calculated within a 500-m buffer of the lake.

 

Categogy/Name Median Minimum Maximum
Land Use/Cover (%)

Urban 1.2 0.0 66.1

Agriculture Pasture 6.0 0.0 72.9

Agriculture Row Crop 10.2 0.0 89.8

Agriculture Other 0.0 0.0 33.5

Agriculture Combined 20.1 0.0 94.7

Agriculture and Urban 25.6 0.0 95.9
Dams

# of Dams within an EDU 558 48 3,634

# of Dams within a 12-digit HUC 2 0 59
Roads

Road Area in a 500 m Buffer (ft/acre) 38.7 0.0 141.7
Fish Stocking

# of Fish Species Stocked per Lake 1 0 22

# of Fish Stocking Events per Lake 14.5 0 393

# of Fish Stocking Events per Lake/Y ears of Data 0.3 O 5.1
Human Population

Human Population Density in 2000 1,083 18 68,636

Housing Densitv in 2000 ~ 816 34 29.749

 

65

 

 

 

3m mmm Sam
88:88 - wet—0058805 5683::
o: m: 83 023-086 watSEoE «com Son use £02 83m 88m «33 <—
EEwE maze =8 2: emu - memo—0059585 o a
om 2m a . H. E ._ m meow Sow can 902 85 zZQ H Hm 55
game wczozco 2: own - was—cosmoboﬂm 38 —2
m3 mg m . . E ._ m ooom meow can 302 85 ”“75 m
88:88 Egg—coo mBZ
mm 3” 6:26: mo 3:05 use mama @3752 a m<2m :2
o w w EH new :5
5 £888 98 58:52
38:82 Rama—coo 302
mm 3 “Emcee .«o €55th 355 @3732 92,—. can :5 m<2m H2
5 Emmommm new E8382
woe—«A moo—a.— =e=oo=e0 3a: 23> com: magma. 83m
Ema—.22.. .35. .3..— ue .9..an 1:: omen—ash wEEEam 9:36

 

.bﬁaozaaD 28m «32 98 .929 mooSomom .8382 .8 3:085qu
538$? 28 $3522 2: .3321”: 88on “5838?, 98 wcmhozcoE Bomwo—oom m.<mm o5 ova—05 momocowxx .809
£3.28 5 3.0.: on 8 moo—ﬂ me 838:: was A33 8% own—£883 Emu E moo—a. me 528:: 32 £28215 8% 5mm Maocowm

mo 3093a £83 =« $82» 3383 $3» mo awash 53w 838:8 Sesame ,3 8% own—988mm 5..“ me $885 .v 933—.

66

Table 5. List of ﬁsh assemblage response metrics used in building natural lake classes
(Total SR and Native SR). Values of each metric are for individual lakes but displayed
for the 5-state region and each individual state.

 

Response Metric Median Minimum Maximum
5-State

Total SR 11 1 26

Native SR 11 1 25
Maine

Total SR 9 l 16

Native SR 9 1 16
New Hampshire

Total SR 8 3 13

Native SR 6 2 10
Michigan

Total SR 15 3 26

Native SR 14 2 24
Wisconsin

Total SR 14 5 25

Native SR 14 5 25
Iowa

Total SR 8 4 21

Native SR 7 3 19

 

67

Table 6. Summary of ﬁsh stocking by state agency (Maine-Department of Inland
Fisheries and Wildlife (DIF W), New Hampshire-Fish and Game Department (FGD),
Michigan-Department of Natural Resources (DNR), Wisconsin-Department of Natural
Resources (DNR), and Iowa-Department of Natural Resources (DNR)), range of years
across all lakes, statistics (median, minimum, and maximum) of the number of species
stocked per lake, and the total number of species stocked across all lakes in a state.

N o. Sp. Stocked per Lake Total No. Sp.
State Ageng Years in Data Median Min. Max. Stocked in All Lakes

ME DIF W 1937-1996 2 0 9 10
NH FGD 1926-1996 2 0 8 13
MI DNR 1 979-2006 1 0 7 20
W1 DNR 1972-2005 2 0 5 16
IA DNR 1928_;2004 l 0 22 27

 

 

68

Table 7. Final multiple regression model results for each response metric versus human
disturbance variables (n = 360 lakes for all analyses). All variables were signiﬁcant at p
< .05.

 

 

Response Adjusted Parameter Standard
Variable R-Sguared Estimate Error
Total SR 0.16
Number of Fish Species Stocked -0.19 0.10
Number of Fish Stocking Events/Years in Data 0.21 0.10
Number of Dams per 12-Digit HUC -0.10 0.03
Agricultural Land Use -0.29 0.10
Human Population Density 0.40 0.06
Native SR 0.22
Number of Fish Species Stocked -0.18 0.10
Number of Stocking Events/Years in Data 0.22 0.10
Number of Dams per 12-Digit HUC -0.14 0.03
Agricultural Land Use (Pasture) 1.14 0.30
Agricultural Land Use (Crop) 0.58 0.30
All Agricultural and Urban Land Use Combined -l .30 0.28
Hum_an Population Densitv 0.52 0.07

 

69

 

Table 8. Pearson correlation coefﬁcients for those predictor variables with an r > 0.60.
Correlations are grouped by major category; anthropogenic disturbances only,
hydrogeomorphic features only, and anthropogenic disturbances and hydrogeomorphic
features combined.

Variable r

Anthropogenic Disturbances

Agriculture (All) Land Use vs Agriculture and Urban Land Use 0.96
Number of Stocking Events vs Number of Stocking Events/Years 0.95
Agriculture (Crop) Land Use vs Agriculture (All) Land Use 0.92
Number of Species Stocked vs Number of Stocking Events 0.90
Agriculture (Pasture) Land Use vs Agriculture (All) Land Use 0.89
Human Population Density 2000 vs Housing Density 2000 0.88
Agriculture (Crop) Land Use vs Agriculture and Urban Land Use 0.87
Number of Species Stocked vs Number of Stocking Events/Years 0.85
Agriculture (Pasture) Land Use vs Agriculture and Urban Land Use 0.82
Agriculture (Pasture) Land Use vs Agriculture (Crop) Land Use 0.67
Urban Land Use vs Agriculture (Other) Land Use 0.63
Hydrogeomorphic Features
Wetland (Woody) Cover vs Wetland (All) Cover 0.97
Lake Area vs Lake Perimeter 0.92
Lake Max Depth vs Lake Mean Depth 0.88
Forest Land Cover vs Mean Runoff 0.79
Lake Perimeter vs Cumulative Watershed Area 0.74
Lake Area vs Cumulative Watershed Area 0.70
Number of Surface Connections vs. Cumulative Watershed Area 0.69

Anthropogenic Disturbances and Hydrogeomorphic Features

Forest Land Cover vs Agriculture and Urban Land Use 089
Forest Land Cover vs Agriculture (All) Land Use -0.85
Mean Runoff vs Agriculture (All) Land Use -0.79
Mean Runoff vs Agriculture and Urban Land Use -0.78
Forest Land Cover vs Agriculture (Crop) Land Use -0.76
Mean Runoff vs Agriculture (Pasture) Land Use -0.73
Forest Land Cover vs Agriculture (Pasture) Land Use -0.72
Mean Runoff vs Agg’culture (Crop) Land Use -0.68

 

 

70

Table 9. Summary of Total SR and Native SR lake classiﬁcations developed using a
range of CP values; listed is the number of classes, the range of lakes per class, total PRE
value, ANOVA F value, and ANOVA P-value. The last column indicates the percentage
of signiﬁcant class-by-class pair-wise comparisons.

 

 

Total SR
Pair-wise
CP # of Classes # of Lakes PRE AN OVA F ANOVA P Comparisons 1%)
0.030 6 20-151 0.50 72.4 <.0001 93.3
0.025 7 20-1 17 0.53 64.8 <.0001 90.4
0.020 9 12-105 0.57 57.1 <.0001 75.0
0.015 11 7-105 0.60 51.3 <.0001 74.5
0.010 16 7-66 0.67 43.2 <.0001 62.5
Native SR
Pair-wise
CP # of Classes # of Lakes PRE ANOVA F ANOVA P Comparisons (%)
0.030 6 27-98 0.49 86.6 <.0001 86.6
0.025 6 27-98 0.49 86.6 <.0001 86.6
0.020 8 10-82 0.54 70.6 <.0001 85.7
0.015 10 10-69 0.58 64.8 <.0001 77.7
0.010 14 10-46 0.63 48.7 <.0001 67.0

 

71

Table 10. Summary of the total number of native ﬁsh species by state with all native
separated into species that are native recreational and native non-recreational.

 

ﬁfe Total Native Native Recreational Native Non-Recreational
Maine 34 6 28
New Hampshire 19 7 12
Michigan 78 18 60
Wisconsin 86 l 7 69
Iowa 32 17 15

 

72

Table 11. List of ﬁsh assemblage response metrics used in measuring anthropogenic
disturbance across natural lake classes (Native SR). Values of each metric are for
individual lakes but displayed across all lakes, small lakes only (< 53 ha), and large lakes
only (> 53 ha).

 

Response Metric Median Minimum Maximum
All Lakes (360)

Total Native SR 11 l 26
Small Lakes (133)

Total Native SR 7 l 21
Large Lakes (227)

Total Native SR 13 3 25

 

73

Table 12. List of ﬁsh assemblage response metrics used in measuring anthropogenic
disturbance across natural lake classes (Native SR). Values of each metric are for
individual lakes but displayed across classes (A, B, C, and D) of small lakes only (< 53
ha).

 

Response Metric Median Minimum Maximum
Class A Lakes (31)
Total Native SR 5 1 10
Native Recreational SR 2 1 8
Native Non-Recreational SR 2 0 6
Class B Lakes (27)
Total Native SR 5 1 1
Native Recreational SR 5 1 8
Native Non-Recreational SR 0 0 6
Class C Lakes (57)
Total Native SR 8 16
Native Recreational SR 6 l 10
Native Non-Recreational SR 2 9
Class D Lakes (18)
Total Native SR 13 7 21
Native Recreational SR 7 5 12
Native Non-Recreational SR 5 0 14

 

 

74

Table 13. List of ﬁsh assemblage response metrics used in measuring anthropogenic
disturbance across natural lake classes (Native SR). Values of each metric are for
individual lakes but displayed across classes (E, F, G, H, I, and J) of large lakes only (>
53 ha).

 

Response Metric Median Minimum Maximum
Class E Lakes (16)

Total Native SR 8 3 12

Native Recreational SR 5 1 5

Native Non-Recreational SR 2 2 9
Class F Lakes (36)

Total Native SR 10 6 15

Native Recreational SR 7 5 9

Native Non-Recreational SR 3 0 8
Class G Lakes (46)

Total Native SR 1 1 6 20

Native Recreational SR 7 l 13

Native Non-Recreational SR 6 0 13
Class H Lakes (50)

Total Native SR 13 5 20

Native Recreational SR 7 3 11

Native Non-Recreational SR 6 0 12
Class I Lakes (69)

Total Native SR 17 7 25

Native Recreational SR 8 4 12

Native Non-Recreational SR 8 2 16
Class J Lakes (10)

Total Native SR 21 16 24

Native Recreational SR 8 7 12

Native Non-Recreational SR 1 1 8 16

 

75

Table 14. List of anthropogenic disturbance variables and their measurement units (land
use/cover, dams, roads, ﬁsh stocking, human populations) used in multiple regression
analysis, data displayed here are for small lakes only (<53 ha), Native SR classes A, B, C,
and D. Land use/cover coverage’s were calculated within a 500-m buffer of the lake.

 

Categog/Name Median Minimum Maximum
Land Use/Cover (%)

Urban 1.3 0.0 66.1

All Agriculture 33.8 0.0 94.7

Agriculture and Urban 47.1 0.0 95.9
Dams

# of Dams within an EDU 615 48 3,634

# of Dams within a 12-digit HUC 2 0 59
Roads

Road Area in a 500 m Buffer (ft/acre) 35.8 0.0 141.7
Fish Stocking

# of Fish Species Stocked per Lake 1 0 6
Human Population

Human Population Densitv in 2000 1L075 25 68.636

 

76

 

Table 15. List of anthropogenic disturbance variables and their measurement units (land
use/cover, dams, roads, ﬁsh stocking, human populations) used in multiple regression
analysis, data displayed here are for large lakes only (>53 ha), Native SR classes E, F, G,
H, I, and J. Land use/cover coverage’s were calculated within a 500-m buffer of the lake.

 

Categopy/Name Median Minimum Maximum
Land Use/Cover (%)

Urban 1.1 0.0 60.0

All Agriculture 15.0 0.0 88.8

Agriculture and Urban 17.5 0.0 89.7
Dams

# of Dams within an EDU 558 51 3,634

# of Dams within a 12-digit HUC 2 0 46
Roads

Road Area in a 500 m Buffer (ft/acre) 41.3 0.3 131.9
Fish Stocking

# of Fish Species Stocked per Lake 1 0 22
Human Population

Hum_an Population Density in 2000 Q04 18 64.804

 

77

 

 

 

XXX :6 mm": hm... mm... mm... 3.: «N6 XXX 3.: «~—

 

83 22: A23 $2: 8:: $2:
wed .146 $6 $3 mad ovd .37.qu acre—=9:— 5:5:
3.2.. .53—
§.e 5.8 2:.8
Gd- Gd- hmd- 60:5an 53.5 4% 9.533344
3mg
mo..- 53.5
22.8
946- 95:323.. =<
A38
:6- on: 33-2 .2. 255
62:
was- :5— .5.— 2.3:
33:
3Q- eoxoeum 3325 he :55: Z

 

 

an aw— cm: 30 EM...— Bm an EU Sm 34..
$259;— mﬂooam 0232 .33.
.omd H NM 3:363 :38 .3. V a E Eeéﬁwﬁ 803 833:; :< $605539 5 Echo vacuum can 88550 $8883

05 mm A5533 $20 :80 £53, @32ng A8335 =« 8m moxﬂ own H 5 C “895va momma? 83 Mm 0332 End?
moist? 3535me 535: was.» Bonnet 360% 033: .89 .8,“ £sz .258 coﬁmewou oat—=8 35m .3 03a...

78

 

 

 

2... an... "a
83: 48.8
36 wmd has: noun—snea— 525.:
A38 398
Ed :5 8.2 23:
.55an0 55.5 Q 95:323..
53.5
$2:
mg. 23.328. 5..
A88 A89
3o- N3- 0:: .83— ..2. 255
29H .3.— 855
63:
Id- 13.35 839mm .3 3:8: Z

RN 8:30 .3 5.88

 

 

 

gau— =aEm

$05.31 33on 9:32 .8oh

.m _ .o N Na BEE?» 89 .3. v a 3 282.3:me 203 838:; =< 2355868 E 8.5 38%—8m 23 88.58 88883
05 E 36203 380 :03 5.23 3.3380 .Amombmsm =8“ .8.“ m8?— oom n 5 Am: mm A 3&8 098 can 3: mm vv max“. :85
£53» 338:; 85933:. :85:— msmho> 30:5: 36on 038: 89 .5.“ £32 .088 =23“:on 0322: 3:5 .5 03:.

79

 

 

 

 

 

xxx xxx 2.: 8... 2:. mm... 5. a; Q... a... "a
838 S8 :28
VS 23 $5 $25 5.32....— 53...:
2:8 :58
83 Ed 82 23.
858 §8
9A. 83 3:285 53.5 a 23.323
5.8
Ed 53.5
:28 88
So 85 25.355?
5.8 58 §8 :68
owe :8 38 8.8 usmzusézamaa:
3.8 8.8
£3 w x .8 new .2. was:
2:8
mmo... tau—25m @63on ha hon—=52

 

 

3.. $— cml 36 emu 3m—

 

55 :89 :~.E__AJ._M <

amen—Em 3.9on 3.533.83— «>352

.mmd H mm 603303 :39

.3. v a a “=85:me 203 333:; =< $855.8.“ 5 8.5 @8983 can 28:33 “30:8qu 05 mm €82.08 mmﬂo
none 5.23 taxa—35 .Amombmca =m 8m 8%. com H 5 C SE28 8330 83 Mm 3sz £53, 873%; 353556
SEE: mamas $233 86on E83388 033: .89 ho.“ 33mm: 338 £832on 23:38 :35 .3 935.

80

 

xxx 2... on... L8... am... a...

1'

 

 

3.... mm... xxx mm... "m
6.8 .28 8.8
mm... as one £25 8.33%.. 5:58
308
.m. «0.2V 33.
.88 .38 .88 .88
38 8... E. R... 3:355 :25 a. 23.38?
.38 .88 :08 .88
mm. .- a..- 8. _- 38 555
.88 608
S. _- we. _- 9.5.3:? =<
.88 .88
.3 88 up: .382 .2. was:
.88
£8 :8. .2. 2:5

 

 

3.. ac— cm: 30 cm... 3m

500.005 00.00am .3 80:857.

 

3:9 :30 “Sum $94..

00.08.09. 00.00am 880.80.504.57. 0387.

.20 n N8 830.03 .99

.3. v a. :0 880....cw6 0.03 00.9.0.8? ..< .m.m0..80.an. 8 .880 88.880 .05 038.60 80.0888. 0... m. 883.00. 35.0
:80 8.5.3 80.8.35 Amarwcm ..m 8.. 000.“. com u 5 A. 8089.0. 00me? 00.2 Mm 0337. 8.5.3 00.889 0083866
8082. msmh0> 3.08.0.8 00.00% 382808880: 03.08 .89 8. 3.200.. .0808 80.3080“ 0.8238 .08.“. .3 039—.

81

00380.0 0290800wo€xn 3.58: wan: 00308 0w0380mmm :0: @6020
2 00m: 2: 3 033 n5 .83 :_m:00m_3 Ame: 53:32 A53 002380: 302 as 0502 E 0003 com 05 .«0 Q02 .— 02$:—

— . d . _ . ...
3.38” 8. as o . ......
z ...

 

 

c:
c:
ﬂ
-—1

 

 

8..sz

82

.womd mo 2:? my; a E @338 .86 m0 mu a 9%: av $320 8:: 33:2:
0:: nonmaw u>onw moxom .33: mm moxﬂ mo 838:: was 2:? 028%“: :38 2t Amoxos 20$ 35 mmﬁo some 5 Ea
com: 08 039 MM; 3522?: was .o:_m> $93 633%; wEEEm 2: Amoxos 28-55 gooﬁng some E .806 mo
03? mam a E maﬁa“: 6m Ed mo no a $5: $883 Boa momma? Ham 2 53, not cosmoummmm? Mm 3on .N 2:»:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

8x3 NV 8x3 m9. 8x3 9 8x3 3 8x3 mm 8me av
5m: 92. 09 09 ad
_ 820 I 3&0 m_u v.85 u_ 330 m 820 0 $90
_ m _
_
8&3 cm $me on Fmod «No. o $0.0 $9.3 mm m 8x3 vm $3.3 5
Now of 2 82v .x. ~me 3
x $90 a 830 $3. 9.3 26E $3 392 30m 0 330 m m 330 < $.20
:66 mmod ovod mFod
NEv. ovv ﬁx. o. v m: Ev
$3 8583; 04 >uoo>>6cm=m>> m2< 9.3 30m
god mood
Dow Dom

 

 

 

 

 

 

 

«and
m... mmv
mm! 9:3

 

 

 

83

.wmmd mo 33> man. a E mEzsmo: .86 no :0 a wEmz AC momma? 35.: 986E

0:: 8:8: 03% moxom 6on: E 3%. :0 898:: :5. 02:; 8:38: :85 on. $98: 29: on». 933 :03 E :5
com: 98 9:? mam :32an :5 .039 $050 .2:ng mEE—mm 05 A898: 28:05 95:33 :03 E .mnmd mo
33> mm: a E szsmo: .m 86 90 mu m ME? G98: 22: momma—o REM OH .23 out :oumoEmmaE mm 9th .m 95w:—

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mox3 or 99.3 mo m9.3 0v m9.3 on m9.3 we 99.3 nm
You mm: 0.: for m.~.. Om
w wmm.o _ mmﬂo O mmm_0 n. wmm.0 D 3&0 o mmm.o
.
.
omod m9.3 om Cod m9.3 or 30.0 8x3 mm
N:2 meEv N? «Ex 5va E E
m9< 85.9w; I 8&0 m9< nonwhmﬁ; M $30 now m meO
95.0 vmod ovod mox3 5
“Ex 0.2V Evhmwv 9. er ﬁm
mo._< 329053 caugwm. 9.3 m9< 9.3 < $30
omod mood
30m 30m
onwd
m: mmv
mm? 9.3

 

 

 

84

.m:o:E:omow

$30 83 gm m03~> :88 Mm .90? :8 N onE 8m .5 :5 d “U momma? .323: w 3 haw :5 Am :5 < 3330
R323: 3 SEE 2.: 3 :986E 98:38 max“. :33 on: :0 Sam: Eam hangoum on: 5.3 :32? $9.2 mm-m
£33 wvd £93 de $9.3 vmd £93 5&13 mm VV momma? 93 :95 Mm ESP mo :ESDEE: .3:an .v charm

 

. . _ _ . _ _ .
Ev. 08.: com omN
m x m
D . m
0 0
z
\ m i
, .21 ,
H! i... ._ < O
_._. it: A.
1...»,,..\.., mama—U 3:3

, .... =«Em

 

...m.‘.,:.,,,._,.,__ amoEEE 862$
_\
f.“ :38

85

.m:o_Etomou awn—o 83

:5 82? :38 mm 88% :8 N 2:wE com AM :5 a momma—o 5:23: D 3% 9.: C :5 .E .0 .m momma—o .32”: o:
823 on: 3 @3365 usewnsm moxﬂ owuﬂ 2: mo 5va 5% DEEOBm on: :23 :39? $83 om-v_ :5 £33 om;
£83 NT~ “8x:— moTI £83 2-0 £83 w T”: An: mmAv momma—o 8:: emu: Mm ﬁner—L ho :osanmE 3:QO .m charm

_ _ _ _ _ _ _ . _

 

 

Ex 83 com omm o
x . w
... a m 3
_ o
2
I ...
,++++(_f,_ o i
+ M
g o MM, n. o
.. u momma—U 8:3
n owns:—
,.\<_ mug—Em 833m

3.:

86

£83988: mud—o 83 98 £03? :38 Mm 25:2 8.: m 833m 8m .5 «E: .0 .m momma—o £:o_w8 ma

hum :5 2 830 £538 $ 833 on: 3 3:865 @5233 m8?— =aEm 8: mo 5va E% CNN—9:88 2: 5:5 :32?
383 m TD £83 3-0 £83 nmé £83 85$ Am: mmvv 3820 83 :25 Mm 0352.8 :ossnEﬂw :3QO 6 2sz
_ ﬂ 8 _ _ d 4 _ _

 

 

Ex 83 com Sm o
O 4 m
0 o m 2,
7..
m a f
z
a < o
......”
1.. r momma—U 8:3
1, x. . =«Em
__ m 80:55 8325
.52

87

.mcoumtomow

mmﬁu 83 was mos—g 588 Mm 3th “oh m oSwE 3m .9. can A .: momma—o £:o_m8 2v 3% 93 A0 23 d .m 8330

£:o_m8 m C 853 8: 3 @88me asewnzm 83 womb: 05 mo 5va 5% 33:88 05 £15 550% 8:. £83 ow;

£83 omi £83 3-0 £83 on £83 3-5 8: mmAV 8330 83 093 Mm 3st mo 839533 33mm .5 25w:—
_ q 3 d 4 _ _ _ _ z :

    

 

 

Ev. 83 28 SN . z a
3,1“...
_. o w ,,,/,v or a.
_ u m $3 +th...,MmM
I . ., .. 1 \
z
0 +
(on. » .... a
_ o m o
m. . m: momma—U 83H
_, a. out“:
/,u,\,a_ 38:33 833m

9532

88

 

 

 

\O

..l
< ‘
[—5 t
E _
<_C<r
fl!
Q)
.2
Cd...
43m
to—
O
s—m
£2
E
I
22: 1
Eu
2
< —
m
(I)
2
U E
I
l
l
lll l l l ' l
VNOOONDV'NO

tuasaid sagaads 1mm saileruo #

FALLFISH

BLACK Bl'llHEAD

BANDED KILLHFISH

CREEK CHl‘B

COMMON SHINER

CHANNEL C.-\I'F|Sll

WALLEYE

NORTHERN PIKE

BLACK CRAPPIE

SNHLLNIOLTH BASS

LARGEMOLTH BASS

CHAIN PICKEREL

BLL‘EGILL

BROOK TROLT

WHlTE SL'CKER

YELLOW PERCH

PL'UPKINSEED

GOLDEN SHINER

BROWN BL'LLHEAD

l
NORTHERN REDBELLY
D;\CE

Figure 8. Number of species present in each class across all lakes and states displayed to show the predominate

ﬁsh assemblages in a class. Four species were present in two lakes and 10 species were present in one lake.

 

.33 28 E Eomoa 23» 86on _ _ can 8%: 9,3 E 6035 053 360% 0:0 .350 a E mommEEommm
BuEEonoE 2: 30am 8 vozaammn 38.5 use 3%: =m mmohum mmﬁo :03 E E305 863:0 53:52 .a 959m

\ELLOW BI'LLIIEAD

mm
4<FOP

 

GOLDEN SHINER

mm
<—

ELLOW PE RCH

“ALLEYE

\

dd:

0 _ _
S, :2 :z
”“8393 (.3252

CREEN Sl‘NFISH
\lHITE CRAPPIE
BLACK BLLHIEAD

4

3
m2 m 320

CHANNEL CATFISH

:—

BL.»\CK CRAPPIE

\RGEXIOIITH BASS I
I

I

BLUEGILL

=

 

S

2

cm

mm

cm

tuasaid saroads 1mm sonar] JO #

 

90

 

 

 

NORTHERN REDBELLY DACE

MUDMINNOW

TOTAL
57

IOWA DARTER

IA

3|

FLATHEAD CATFISH
CREEK CHL7B

WI
8

GIZZARI) SHAD

SMALLMOUTH B.—\SS

MI
l5

COMMON SHINER
m COMMON WHITE SUCKER

Number of Lakes

NH

BROWN BL'LLHE.-\D

ROCK BASS

M E
0

WHITE SUCKER
GOLDEN SHINER
BLUNTNOSE MINNOW

WALLEYE

Class C

NORTHERN PIKE
GREEN SL’NFISH
YELLOW BL'I.LHE.‘\D
WHITE CRAPPIE
PUMPKINSEED
BLACK BCLLHEAD
CHANNEL CATFISH
YELLOW PERCH
BLACK CRAPPIE
LARGEMOLITH BASS

BLUEGILL
T T r r 1 ﬂ

O O O O O O O
\D W V M N '—

iuasaid sagoads 1|th sewer] JO #

 

 

 

91

Figure 10. Number of species present in each class across all lakes and states displayed to show the predominate

assemblages in a class. 10 species were present in two lakes and 20 species were present in one lake.

 

 

 

Number of Lakes
NH MI WI IA TOTAL

M E

Class D

 

I8

(‘I

 

 

WHITE SLFCKER

HYBRID SIINFISH

BROOK SILVERSIDE

GRASS PICKEREL

BLACKNOSE SHINER

WHITE CRAPPIE

ROCKBASS

GREEN SL'NFISH

GOLDEN SHINER

CHANNEL CATFISH

BLUNTNOSE MINNOW

W A RMOL'TH

WALLEYE

COM-MON WHITE SI’CKER

BOWFIN

YELLOW BULLHEAI)

BLACK BLTLLHEAD

BROWN BULLIIEAD

NORTHERN PIKE

PUMPKINSEED

YELLOW PERCH

LARGEMOUTH BASS

BLUEGILL

BLACK CRAPPIE

 

 

Figure 11. Number of species present in each class across all lakes and states displayed to show the predominate

assemblages in a class. Nine species were present in two lakes and 13 species were present in one lake.

WI
0

IIIIII . . .

Figure 12. Number of species present in each class across all lakes and states displayed to show the predomi

Ml
assemblages in a class. Two species were present in two lakes and 10 species were present in one lake.

Number of Lakes
6

NH

Class E

tuasaid sayoadg llllM SQ)|B'[ Jo #

 

SMALLMOL I‘H BASS
ATLANTIC SALMON
ALEWII’I-I ‘
FALLFISII
RAINBOW SMELT
WHITE PERCH
REDBRIi \ST Sl‘NFISH‘
I
BANI)F.I) KII.L|I-‘ISII
AMERICAN EEL
WHITE SL'CKER
GOLDEN SIIINISR
CIIAIN I’lCKIiREL
BROWN BI'LLIIEAD
YELLOW PERCH

PI'MPKINSEED

,i L, A,, 7,,, ¥_, _ ,,_I

te

I
I
I I

 

ln
m

36

II IIIIIIIIUIIIImu

TOTAL

Number ofLakes
MI IA
6

ME
0

Class F

LOGPERCH

IOWA DARTER

WIIITE BASS

CENTRAL ML‘DMININOW
BIOMOL TH BL'FFALO
JOHNNY [)ARTER
GREEI\ SI'NFISH
COMMON SHINER
BROWN BI'I.LIIE.‘\I)

ROCKBASS

WHITE CRAPPIE
YELLOW Bl'LLHEAD
WHITE SL'CKER
GOLDEN SIIINER
BLL'NTNOSE MINNOW
SMALLMOCTII BASS
CHANNEL CATFISH
BLACK BULLHEAD
I’l'MPKINSEED
NORTHERN PIKE
WALL E YE

YELLOW PERCH

= BLACK CRAPPIE
l..-\R(}IS.\lOl?TH BASS

I ' I I I I
OWOWOV‘.
CONN—"—

iuasaid sanadS [[JIM SMIB'] J0 #

 

94

BI.l'l€(i|LL
j

0

 

COMMON WHITE SI7CKER‘

Figure 13. Number of species present in each class across all lakes and states displayed to show the predominate

assemblages in a class. Six species were present in two lakes and 17 species were present in one lake.

 

 

 

SLIMY SCULPIN
3 MUSKELLLFNGE

LAKE TROL’T
JOHNNY DARTER
CHAIN PICKEREL
BLACKNOSE DACE
BANDED KILLIFISH
QL‘ILLBACK CARPSUCKER
NORTHERN REDBELLY DACE?
SPOTTAIL SHINER
MOTTLED SCULPIN
LONGNOSE SL'CKER
LOGPERCH
BL’RBOT
YELLOW BASS
SAUGER X WALLEYE
RIVER CARPSUCKER
PEARL DACE
GIZZARD SHAD
ATLANTIC SALMON
ROCKBASS
RAINBOW SMELT
FLATHEAD CATFISH
COMMON WHITE STICKER
WHITE BASS
LAKE CHl'B
BLUNTNOSE IVIINNOW
BIGMOUTH BUFFALO
FALLFISH
BROOK TROUT
FRESHWATER DRL'M
GREEN SUNFISH
CREEK CHL'B
BROWN BULLHEAD
COMMON SHINER
YELLOW BL’LLHEAD
BLACK BULLHEAD
WHITE CRAPPIE
SMALLMOUTH BASS
CHANNEL CATFISH
PL’MPKINSEED
WHITE SL'CKER
NORTHERN PIKE
GOLDEN SHINER
BLACK CRAPPIE
WALLEYE
YELLOW PERCH
LARGEMOUTH BASS
BLUEGILL
T I I T I I I

WOWOWOWO
MMNNF-‘ﬁ

iuasaid saioads (mm $3)|B"[ JO #

IA TOTAL
7

Number ofLakes
NH Ml WI
8 5

M E

Class G

 

 

 

 

95

Figure 14. Number of species present in each class across all lakes and states displayed to show the predominate

assemblages in a class. 11 species were present in two lakes and 18 species were present in one lake.

 

 

 

WHITE CRAPPIE
SPOTFIN SHINER
MlFDMlNNOW
LOGPERCH

LAKE CHLTBSUCKER
SAND SHINER
PUMPKINSEED x BLUEGILL
__ LONGNOSE GAR

3 5?. GRAss PICKEREL
BLACKCHIN SHINER
BROOK SILVERSIDE
TADPOLE MADTOM
CENTRAL MUDMINNOW
SPOTTAIL SHINER
MOTTLED SCULPIN
MIMIC SHINER
HYBRID SUNFISH
GREEN SUNFISH
CHANNELCATFBH
BLACKNOSE SHINER
IOWA DARTER
WARMOCTH

BOWFIN
MUSKELLUNGE
COMMON SHINER
JOHNNY DARTER
COMMON WHITE SUCKER
ROCKBAss

BLACK BLILLHEAD
BROWN BULLHEAD
WHITE SUCKER
SMALLMOUTH BAss
GOLDEN SHINER
BLUNTNOSE MINNOW
WALLEYE

YELLOW BULLHEAD
NORTHERN PIKE
BLACK CRAPPIE
PUMPKINSEED
BLUEGILL

YELLOW PERCH

LARGEMOL’TH BASS
I a I I I

f,
CD CD CD CD CD CD CD
“3 V1 <r CG CV ——

mosaic] saioads 1mm sexier] to #

TOTAL
50

IA
4

I9

Number of Lakes
MI

NH
0

ME
0

Class H

 

 

 

96

Figure 15. Number of species present in each class across all lakes and states displayed to show the predominate

assemblages in a class. Six species were present in two lakes and 12 species were present in one lake.

 

 

 

TOTA L
69

WI IA

36

MI

 

Number of Lakes

NH

ME

Class 1

 

 

 

TADPOLE MADTOM
SPOTTED GAR
RAINBOW DARTER
NORTHERN REDBELLY
LONGEAR SUNFISH
FATHEAD MINN(‘)W
WHITE CRAPPIE
TROL'TPERCH
CREEK CHUB
BIGMOUTH BUFFALO
SHORTHEAD REDHORSE
SAND SHINER
MOTTLED SCULPI-N
LAKE CHL’BSLTCKER
EMERALD SHINER
BLACKCHIN SHINER
MI'DMINNOW
HORNYHEAD CHL‘B
BANDED KILLIFISH
PUMPKINSEED X
CHANNEL CATFISH
CENTRAL

MIMIC SHINER
ML'SKELLUNGE
GRASS PICKEREL
BLACKNOSE SHINER
SPOTTAIL SHINER
IOWA DARTER
HYBRID SUNFISH
GREEN SUNFISH
COMMON SHINER
WARMOLTTH
LONGNOSE GAR
BROOK SILVERSIDE
LOGPERCH

BLACK BULLHEAD
WHITE STICKER
COMMON WHITE
BOWFIN

JOHNNY DARTER
ROCKBASS

BROWN BULLHEAD
GOLDEN SHINER
BLLINTNOSE MINNOW
YELLOW BCLLI-IEAD
SMALLMOUTH BASS
WALLEYE
NORTHERN PIKE
BLACK CRAPPIE
PUMPKINSEED
LARGEMOUTH BASS
BLUEGILL

YELLOW PERCH

 

 

Figure 16. Number of species present in each class across all lakes and states displayed to show the predominate

assemblages in a class. Five species were present in two lakes and 22 species were present in one lake.

 

 

 

 

 

 

IA TOTAL

Number of Lakes
WI

NH Ml

ME

Class I

IO

 

N
—-a

I

O

F—‘

iuasaid saioadg th SMIB’] Jo #

I
00

I
No

I

V

I
N

WHITE CRAPPIE
ROCKBASS
GREATER REDHORSE

COMMON WHITE SL'CKER'

COMMON SHINER
BOWFIN
BLUNTNOSE MINNOW
WHITE BASS
MUSKELLL‘NGE
MIMIC SHINER
LOGPERCH

GOLDEN REDHORSE
TROL'TPERCH
JOHNNY DARTER
CHANNEL CATFISH
SILVER REDHORSE
SHORTHEAD REDHORSE
WHITE SUCKER
LARGEMOUTH BASS
GOLDEN SHINER
BLACK BULLHEAD
YELLOW BULLHEAD
YELLOW PERCH
BLUEGILL
WALLEYE
SMALLMOLITH BASS
PUMPKINSEED
NORTHERN PIKE
BLACK CRAPI’IE

 

 

 

 

98

Figure 17. Number of species present in each class across all lakes and states displayed to show the predominate

assemblages in a class. Seven species were present in two lakes and 17 species were present in one lake.

Bibliography

Abell, R.A., D.M. Olson, E. Dinerstein, P.T. Hurley, J .T. Diggs, W. Eichbaum, S.
Walters, W. Wettengel, T. Allnutt, C.J. Loucks, and P. Hedao. Freshwater
Ecoregions of North America. Island Press. Washington, DC. 2000.

Allan, J .D. and LB. Johnson. 1997. Catchment-scale analysis of aquatic ecosystems.
Freshwater Biology. 37: 107-11 1.

Anderson, D.W. and DH Schupp. 1986. Fish community responses to northern pike
stocking in Horseshoe Lake, Minnesota. Minnesota Department of Natural
Resources Report. No. 387

Bachmann, R.W., B.L. Jones, D.D. Fox, M. Hoyer, L.A. Bull, and DE. Canﬁeld Jr.
1996. Relations between trophic state indicators and ﬁsh in Florida (U.S.A.)
lakes. Canadian Journal of Fisheries and Aquatic Sciences. 53: 842-855.

Becker, G.C. Fishes of Wisconsin. University of Wisconsin Press. Madison, Wisconsin.
1983.

Breiman, L., J .H. Friedman, R. Olshen, and C. Stone. Classiﬁcation and Regression
Trees. Wadsworth, Belmont. 1984.

Brenden, T.O., L. Wang, and PW. Seelbach. 2008. A river valley segment classiﬁcation
of Michigan streams based on ﬁsh and physical attributes. Transactions of the
American Fisheries Society. 137: 1621-1636.

Brunberg, AK. and P. Blomqvist. 2001. Quantiﬁcation of anthropogenic threats to lakes
in a lowland county of Central Sweden. Ambio. 30(3): 127-134.

Buck, 0., D.K. Niyogi, and CR. Townsend. 2003. Scale-dependence of land use effects
on water quality of streams in agricultural catchments. Environmental
Pollution. 130: 287- 299.

Carpenter, S.R., B.J. Benson, R. Biggs, J .W. Chipman, J .A. Foley, S.A. Golding, R.B.
Hammer, P.C. Hanson, P.T.J. Johnson, A.M. Kamarainen, T.K. Kratz, R.C.
Lathrop, K.D. McMahon, B. Provencher, J .A. Rusak, C.T. Solomon, E.H.
Stanley, M.G. Turner, M.J. Vander Zanden, C. Wu, and H. Yuan. 2007.
Understanding regional change: A comparison of two lake districts. BioScience.
57(4): 323-335.

Cheruvelil, K.S., P.A. Soranno, M.T. Bremigan, T. Wagner, and S.L. Martin. 2008.

Grouping lakes for water quality assessment and monitoring: the roles of
regionalization and spatial scale. Environmental Management. 41: 425-440.

99

Dolman, W.B. 1990. Classiﬁcation of Texas reservoirs in relation to limnology and ﬁsh
community associations. Transactions of the American Fisheries Society. 19:
511-520.

Drake, M.T. and D.L. Pereira. 2002. Development of a ﬁsh-based index of biotic
integrity for small inland lakes in Central Minnesota. North American Journal of
Fisheries Management. 22: 1105-1123.

Eadie, J. McA., T.A. Hurley, R.D. Montgomerie and KL. Teather. 1986. Lakes and
rivers as islands: species-area relationships in the ﬁsh faunas of Ontario.
Environmental Biology of Fishes. 15(2): 81-89.

Gergel, S.E., M.G. Turner, J .R. Miller, J .M. Melack and EH. Stanley. 2002. Landscape
indicators of human impacts to riverine systems. Aquatic Sciences. 64: 118-128.

Hair J .F . Jr., R.E. Anderson, R.L. Tatham, and WC. Black. Multivariate Data Analysis,
3rd edn. New York, Macmillan. 1995.

Hansen, GA. and ML. Jones. 2008. The value of information in ﬁshery management.
Fisheries 33: 340-348.

Harlan, J .R. and EB. Speaker. Iowa Fish and Fishing. State of Iowa. 1956.

Hershey, A.E., S. Beaty, K. Fortino, M. Keyse, P.P. Mou, W.J. O’Brien, A.J. Ulseth,
G.A. Gettel, P.W. Lienesch, C. Luecke, M.E. McDonald, C.H. Mayer, M.C.
Miller, C. Richards, J .A. Schuldt and SC. Whalen. 2005. Effect of landscape
factors on ﬁsh distribution in arctic Alaskan lakes. Freshwater Biology. 51: 39—
55.

Hill, N.M., P.A. Keddy, and LC. Wisheu. 1998. A hydrological model for predicting the
effects of dams on the shoreline vegetation of lakes and reservoirs.
Environmental Management. 22(5): 723-736.

Higgins, J .V., M.T. Bryer, M.L. Khoury, and T.W. Fitzhugh. 2005. A freshwater
classiﬁcation approach for biodiversity conservation planning. Conservation
Biology. 19(2): 432-445.

Holmgren, K. and M. Appelberg. 2000. Size structure of benthic freshwater ﬁsh
communities in relation to environmental gradients. Journal of Fish Biology. 57:
1312-1330.

Holmlund, CM. and M. Hammer. 2004. Effects of ﬁsh stocking on ecosystem services:
An overview and case study using the Stockholm Archipelago. Environmental
Management. 33(6): 799—820.

100

Hubbs, CL. and K.F. Lagler. Fishes of the Great Lakes Region. University of Michigan
Press. Ann Arbor, MI. 2007

Iowa Department of Natural Resources. 2009. Iowa Fishing Guide.

Irz, P., C. Argillier and J .P. Proteau. 2004. Contribution of native and non-native species

to ﬁsh communities in French reservoirs. Fisheries Management and Ecology.
1 1: 165—172.

Jackson, DA. and H.H. Harvey. 1989. Biogeographic associations in ﬁsh assemblages:
local vs. regional processes. Ecology. 70(5): 1472-1484.

Jennings, M.J., M.A. Bozek, G.R. Hatzenbeler, E.E. Emmons, and MD. Staggs. 1999.
Cumulative effects of incremental shoreline habitat modiﬁcation of ﬁsh
assemblages in north temperate lakes. North American Journal of Fisheries
Management. 19: 18-27.

J eppesen, E.J., J .P. Jensen, M. Sondergaard, T. Lauridsen, and F. Landkildehus. 2000.
Freshwater Biology. 45: 201-218.

Johnson, M.G., J .H. Leach, C.K. Minns, and CH. Olver. 1977. Limnological
characteristics of Ontario lakes in relation to associations of walleye (Stizostedion
vitreum vitreum), northern pike (Esox lucius), lake trout (Salvelinus namaycush),

and smallmouth bass (Micropterus dolomieui). Journal of the Fisheries
Research Board of Canada. 34: 1592-1601.

Johnson, P.TJ., J .D. Olden, M.J. Vander Zanden. 2008. Dam invaders: impoundments
facilitate biological invasions into freshwaters. Frontiers in Ecology and the
Environment. 6(7): 357-363.

Jones, J .R., M.F. Knowlton, D.V. Obrecht, and EA. Cook. 2004. Importance of
landscape variables and morphology on nutrients in Missouri reservoirs.

Canadian Journal of Fisheries and Aquatic Sciences. 61: 1503-1512.

Kratz, T.K., K.E. Webster, C.J. Bowser, J .J . Magnuson, and B.J. Benson. 1997. The
inﬂuence of landscape position on lakes in northern Wisconsin. Freshwater
Biology. 37: 209-217.

Magnuson, J .J ., W.M. Tonn, A. Banetjee, J. Toivonen, O. Sanchez, and M. Rask. 1998.
Isolation vs. extinction in the assembly of ﬁshes in small northern lakes. Ecology.

79(8): 2941-2956.

Maine Department of Inland Fisheries and Wildlife. 2008. Open Water Fishing
Regulations.

101

Marshall, TR. and PA. Ryan. 1987. Abundance patterns and community attributes of
ﬁshes relative to environmental gradients. Canadian Journal of Fisheries and
Aquatic Science. 44(2): 198-215.

McMahon G., S.M. Gregonis, S.W. Waltman, J .M. Omemik, T.D. Thorson, J .A. F reeouf,
A.H. Rorick, and J .E. Keys. 2001. Developing a spatial framework of common

ecological regions for the conterminous United States. Environmental
Management. 28(3): 293—3 16.

Mehner, T., M. Diekmann, U. Bramick, and R. Lemcke. 2005. Composition of ﬁsh
communities in German lakes as related to lake morphology, trophic state, and
shore structure and human-use intensity. Freshwater Biology. 50: 70-85.

Michigan Department of Natural Resources. 2009. Michigan Fishing Guide.
New Hampshire Fish and Game Department. 2009. Freshwater Fishing Digest.

Olden, J .D. and DA. Jackson. 2002. A comparison of statistical approaches for modeling
ﬁsh species distributions. Freshwater Biology. 47: 1976—1995.

Olden, J .D., D.A. Jackson, and PR. Peres-Neto. 2001. Spatial isolation and ﬁsh
communities in drainage lakes. 0ecologia. 127:572—5 85.

Olden, J .D., O.P. Jensen and M.J. Vander Zanden. 2006. Implications of long-term
dynamics of ﬁsh and zooplankton communities for among-lake comparisons.
Canadian Journal of Fisheries and Aquatic Sciences. 63: 1812-1821.

Omemik, J. M. 1987. Ecoregions of the conterminous United States. Annals of the
Association of American Geographers. 77(1): 188-125.

Pierce, RB. and CM. Tomcko. 2003. Effects of walleye stocking on the ﬁsh
community of Lake Thirteen. Minnesota Department of Natural Resources
Investigational Report. No. 505.

Poff, N.L. 1997. Landscape ﬁlters and species traits: towards mechanistic understanding
and prediction in stream ecology. Journal of the North American Benthological
Society. 16(2): 391-409.

Prest, V.K. 1969. Retreat of Wisconsin and Recent ice in North America. Geological
Survey of Canada Map, 1257A, scale 125,000,000.

R Development Core Team (2008). R: A language and environment for statistical
computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN
3900051-07-0, URL http://www.R-project.org.

Radomski, R]. and T.J. Goeman. 1995. The homogenizing of Minnesota lake ﬁsh
assemblages. Fisheries. 20(7): 20-23.

102

Rahel, F.J. 1984. Factors structuring ﬁsh assemblages along a bog lake successional
gradient. Ecology. 65(4): 1276-1289.

Rempel, LL. and D.G. Smith. 1998. Postglacial ﬁsh dispersal from the Mississippi

refuge to the Mackenzie River basin. Canadian Journal of Fisheries and Aquatic
Sciences. 55: 893-898.

Riera, J .L., J .J . Magnuson, T.K. Kratz, and KB. Webster. 2000. A geomorphic template
for the analysis of lake districts applied to the Northern Highland Lake District,
Wisconsin, USA. Freshwater Biology. 43: 301-318.

Roth, B.M., I.C. Kaplan, G.G. Sass, P.T. Johnson, A.E. Marburg, A.C. Yannarell, T.D.
Havlicek, T.V. Willis, M.G. Turner, and SR. Carpenter. 2007. Linking terrestrial
and aquatic ecosystems: The role of woody habitat in lake food webs. Ecological
Modeling. 203: 439—452.

The data analysis for this paper was generated using SAS software, Version 9.1 of the
SAS System for Windows. Copyright © 2004 SAS Institute Inc. SAS and all other
SAS Institute Inc. product or service names are registered trademarks or
trademarks of SAS Institute Inc., Cary, NC, USA.

Scarola, J .F . Freshwater Fishes of New Hampshire. New Hampshire Fish and Game
Department. Division of Inland and Marine Fisheries. 1973.

Schupp, DH. 1992. An ecological classiﬁcation of Minnesota lakes with associated ﬁsh
communities. Minnesota Department of Natural Resources Report. No. 417.

Seaber, P.R., F.P Kapinos, and G.L. Knapp, 1987. Hydrologic Unit Maps. US.
Geological Survey Water-Supply Paper 2294, United States Department of the
Interior, Geological Survey, Denver, Colorado.

Seelbach, P.W., M.J. Wiley, P.A. Soranno, and M.T. Bremigan. 2002. Aquatic
conservation planning: Using landscape to maps to predict ecological reference
conditions for speciﬁc waters. Pages 454-478 in K. Gutzwiller, editor. Concepts
and Applications of Landscape Ecology in Biological Conservation. Springer-
Verlag.

Soranno, P.A., K.E. Webster, K.S. Cheruvelil and M.T. Bremigan. 2009. The lake
landscape-context framework: linking aquatic connections, terrestrial features and

human effects at multiple spatial scales. Verh. Internat. Verein. Limnol. 30(5):
695-700.

Tonn, W.M. 1990. Climate change and ﬁsh communities: A conceptual framework.
Transactions of the American Fisheries Society. 119: 337-352.

103

 

Tonn, W.M. and J.J. Magnuson. 1982. Patterns in the species composition and richness
of ﬁsh assemblages in northern Wisconsin lakes. Ecology. 63(4): 1149-1166.

Tonn, W.M., J .J . Magnuson, and A.M. Forbes. 1983. Community analysis in ﬁshery
management: an application with northern Wisconsin lakes. Transactions of the
American Fisheries Society. 112: 368-377.

Tonn, W.M., C.A. Paszkowski, G.J. Scrimgeour, P.K.M. Aku, M. Lange, E.E. Prepas and
K. Westcott. 2003. Effects of forest harvesting and ﬁre on ﬁsh assemblages in
boreal plains lakes: A reference condition approach. Transactions of the American
Fisheries Society. 132: 514-523.

Wisconsin Department of Natural Resources. Bureau of Fisheries Management. 2009.
Guide to Wisconsin Hook and Line Fishing Regulations.

Webster, K.E., P.A. Soranno, K.S. Cheruvelil, M.T. Bremigan, J .A. Downing, P.D. Vaux,
T.R. Asplund, L.C. Bacon, and J. Connor. 2008. An empirical evaluation of the

nutrient-color paradigm for lakes. Limnology and Oceanography. 53(3): 1137-
1148.

104

 

      

IIIIII

ullllliil lililliﬁillliﬂijlll“