EARLY LIFE HISTORY DYNAMICS AND RECRUITMENT PROCESSES
OF RAINBOW SMELT IN LAKE HURON
By
Timothy P. O’Brien

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Fisheries and Wildlife
2010

ABSTRACT
EARLY LIFE HISTORY DYNAMICS AND RECRUITMENT PROCESSES
OF RAINBOW SMELT IN LAKE HURON
By
Timothy P. O’Brien
Rainbow smelt are an important prey species for native and introduced salmonines in the
Great Lakes, as well as being a competitor and predator of native fish species. In Lake Huron,
rainbow smelt populations have been characterized by variable recruitment and year class
strength. To understand growth and survival dynamics during early life history, larval rainbow
smelt populations were sampled during 2008 and 2009 in St. Martin Bay, Lake Huron. Growth
rates of larval rainbow smelt and relative survival of stream cohorts was higher during 2009 than
in 2008, a consequence of warmer water temperatures during June and cooler, more optimal
temperatures during July. Early hatching stream cohorts experienced high mortality during 2008
and fish surviving experienced lower growth relative to cohorts during 2009. To investigate long
term recruitment dynamics for these populations, Ricker stock-recruit models were developed for
Lake Huron rainbow smelt sampled from 1976-2009. Ricker models were developed with the
inclusion of four external variables that have been reported to affect rainbow smelt production:
precipitation, water temperature, water levels, and lake trout abundance as measured by catch per
unit effort (CPE). The best model describing rainbow smelt recruitment during 1976-2009
included lake trout CPE, Lake Huron water levels, and over basin precipitation. A separate
model for the 1994-2009 time period identified lake trout CPE as an important regulator of
rainbow smelt recruitment. These findings indicate that variability in growth and survival of
rainbow smelt during early life history stages influences year class formation, but recruitment is
strongly regulated by lake trout predation on adults or age-0 recruits.

ACKNOWLEDGMENTS

The research presented in this thesis was only possible with the help and support of many
individuals, and they share in the accomplishments presented herein. Funding for the field and
analytical portions of this research was provided by the U.S. Geological Survey, Great Lakes
Science Center in Ann Arbor, Michigan, who was my employer throughout this project.
Specifically, I thank Dr. Leon Carl and Dr. Jaci Savino for giving me the opportunity to pursue
graduate research while employed full time at the Great Lakes Science Center. I was an
inquisitive research technician who wanted to learn more about fisheries science and our Great
Lakes ecosystems, and when the opportunity came for graduate school, you trusted I would
follow through on this endeavor. Michigan State University, College of Agriculture and Natural
Resources provided fellowships for graduate coursework. Research funding was also provided
by the Saginaw Bay Walleye Club Scholarship, and I am thankful for their interest and support
in my research.
My colleagues at the Great Lakes Science Center provided helpful insight and critique as
I was learning the finer points and pitfalls of field work and analytical fisheries techniques.
Thanks to Dr. Charles Madenjian and Dr. Bo Bunnell for advice regarding modeling,
programming, and computations. Dr. Dave Warner assisted with model programming, offered
advice on technical aspects of my research, and always showed a genuine interest in this project
and my success in graduate school. Dr. Stephen Riley helped acquire data for this research and
accommodated my many requests for time series data. Dr. Jeff Schaeffer showed enthusiasm for
my research, was always willing to listen to my ideas, and understood what a balancing act
family, work, and school could be.
iii 
 

The two years of field research I conducted in St. Martin Bay would not have been
possible without the many that assisted. Special thanks go to Andrew Briggs, my steadfast field
and laboratory technician who endured many long nights wading in streams or sampling larval
fish on an ominous and unpredictable northern Lake Huron. Your perseverance and dedication
on the water and in the laboratory was instrumental to the completion of my research on early
life history of rainbow smelt. Dr. Paul Steen, Dr. Jeff Schaeffer, Steve Farha, and Dr. Ed
Roseman helped sample spawning rainbow smelt in St. Martin Bay. Thank you to Greg and
Karen Wright at the Chippewa Ottawa Resource Authority, Nunns Creek Station for providing
boat storage, sample storage, and up to date reports on rainbow smelt spawning runs. The
MAES Dunbar Forest Experiment Station provided lodging for researchers during this project.
Thank you to Clark Township, MI for providing dockage for the R/V Osmerid.
I thank my thesis committee Dr. Brian Roth, Dr. Cheryl Murphy, and Dr. Ed Roseman
for providing insight and guidance during this research. I am especially grateful to my graduate
advisor, Dr. William Taylor whose guidance kept me focused and whose support was
unwavering when I encountered challenges.
I thank my parents (Joan and Tom O’Brien) for instilling in me a respect and appreciation
for the natural world, and supporting me in the path I chose for a career in science. I thank my
brothers Bob and Dan, and my sister Kate for their support and encouragement. My daughter
Sage and son Ian were always a source of inspiration. Finally, to my wife Sara, your devotion,
support, encouragement, and unending patience while I worked long hours or was away in the
field made this all possible.

iv 
 

TABLE OF CONTENTS
LIST OF TABLES ......................................................................................................................... vi
LIST OF FIGURES ...................................................................................................................... vii
THESIS INTRODUCTION .............................................................................................................1
CHAPTER 1
ABUNDANCE, GROWTH, AND SURVIVAL OF LARVAL RAINBOW SMELT IN ST.
MARTIN BAY, LAKE HURON ...................................................................................................7
Introduction ......................................................................................................................................7
Methods ........................................................................................................................................9
Study area .................................................................................................................................9
Stream surveys .......................................................................................................................11
Ichthyoplankton surveys ........................................................................................................12
Length frequency analysis of larval cohorts ..........................................................................13
Larval growth ........................................................................................................................14
Larval mortality .....................................................................................................................15
Results .......................................................................................................................................16
Stream surveys .......................................................................................................................16
Ichthyoplankton surveys ........................................................................................................19
Length frequency analysis of larval cohorts ..........................................................................22
Larval growth ........................................................................................................................27
Larval mortality .....................................................................................................................27
Discussion ..................................................................................................................................31
CHAPTER 2
EFFECTS OF STOCK SIZE, CLIMATE, AND PREDATION ON RAINBOW SMELT
RECRUITMENT IN WESTERN LAKE HURON, 1976-2009 ....................................................39
Introduction ....................................................................................................................................39
Methods ......................................................................................................................................42
Bottom trawl collections ........................................................................................................42
Stock-recruit models ..............................................................................................................45
Integrated stock-recruit models .............................................................................................48
Model selection .....................................................................................................................49
Results ........................................................................................................................................50
Stock-recruit models ..............................................................................................................53
Effects of stock size, climate, and predation .........................................................................58
Discussion ..................................................................................................................................62
REFERENCES ..............................................................................................................................69 

v 
 

LIST OF TABLES

Table 1. Parameter estimates for finite mixture distribution models fit to larval rainbow smelt
length frequency distributions during 2009; π = mixing proportions, μ = mean cohort length, σ
= standard error. Cohort 1 corresponds to the stream spawned larvae, cohort 2 corresponds to
lake spawned larvae. Model degrees of freedom (Df) and Chi-square goodness-of-fit statistic
(X^2) are presented for each model. All models were significant at α = 0.05. Modeled length
distributions for 18-June and 8-Jul identified a third mode which was not included in this
analysis due to low (i.e., < 10 larvae) sample sizes. Thus, mixing proportions for those sample
dates do not add to 1 ......................................................................................................................26
 

Table 2. Year specific YOY length cutoffs (mm) for rainbow smelt in Lake Huron, 1976-2009.
Cutoffs were determined from length frequency analysis and applied to fishing power corrections
as determined by Adams et al. (2009) ...........................................................................................47
Table 3. Original parameter estimates, approximate standard error, and 95% confidence
intervals for stock-recruit models of rainbow smelt recruitment (1976-2009, 1976-1991, and
1994-2009) .....................................................................................................................................53
Table 4. Mean, variance, and 95% confidence intervals of bootstrap parameter estimates for
stock-recruit models of rainbow smelt recruitment (1976-2009, 1976-1991, and 1994-2009).
Statistics are based on 1000 bootstrap samples .............................................................................54
Table 5. Model selection results for rainbow smelt recruitment in Lake Huron, 1976-2009, based
on Akaike weights wi and ranked top (best model) to bottom (least plausible model). Also shown
are sums of squared errorr, AICc, Δi (AICc differences), e^ (-1/2*Δi) (model likelihoods), and
evidence ratios. All model combinations are shown with Ricker parameters (α, β) and
explanatory variables; LT = lake trout CPE, P = overbasin precipitation, W = Lake HuronMichigan water levels, and T = mean lake temperature 0-20 meters. ...........................................59
Table 6. Model selection results for rainbow smelt recruitment in Lake Huron, 1976-1991, based
on Akaike weights wi and ranked top (best model) to bottom (least plausible model). Also shown
are sum of squared errorr, AICc, Δi (AICc differences), e^ (-1/2*Δi) (model likelihoods), and
evidence ratios. All model combinations are shown with Ricker parameters (α, β) and
explanatory variables; LT = lake trout CPE, P = overbasin precipitation, W = Lake HuronMichigan water levels, and T = mean lake temperature 0-20 meters ............................................60
Table 7. Model selection results for rainbow smelt recruitment in Lake Huron, 1994-2009, based
on Akaike weights wi and ranked top (best model) to bottom (least plausible model). Also shown
are sum of squared errorr, AICc, Δi (AICc differences), e^ (-1/2*Δi) (model likelihoods), and
evidence ratios. All model combinations are shown with Ricker parameters (α, β) and
explanatory variables; LT = lake trout CPE, P = overbasin precipitation, W = Lake HuronMichigan water levels, and T = mean lake temperature 0-20 meters................................................ 61

vi 
 

LIST OF FIGURES

Figure 1. Commercial harvest (1000’s of pounds) of rainbow smelt for all five Great Lakes
during 1931-2006 , adapted from Baldwin et al. (2009)..................................................................2
Figure 2. Density (No. /ha., solid line, closed circle) and relative standard error (dotted line,
open triangle) of YOY (top panel) and YAO (lower panel) rainbow smelt estimated from USGS
bottom trawl surveys, 1976-2009, adapted from Riley et al. (2009) ...............................................5
Figure 3. St. Martin Bay, Lake Huron showing ichthyoplankton sampling transects (solid black
lines) and spawning streams surveyed (Carp River, Nunns Creek, Spring Creek, and St. martin
Creek) during 2008-09 ...................................................................................................................10
Figure 4. Stream temperatures recorded from four St. Martin Bay Spawning tributaries from
April-June during 2008 (top panel) and 2009 (lower panel); St. Martin Creek, Spring Creek,
Nunns Creek, and the Carp River. Shaded regions indicate the observed duration of spawning
by rainbow smelt ............................................................................................................................18
Figure 5. Abundance (Mean No /1000 m^3) of larval rainbow smelt in St. Martin Bay, Lake
Huron from May-August during 2008 (top panel) and 2009 (lower panel). Error bars are 95%
confidence intervals .......................................................................................................................20
Figure 6. Mean water temperature (degrees C) across sampling transects recorded during larval
fish surveys, 12 May-5-August, 2008 (open circle) and 21-May-4-August,
2009 (closed circle) ........................................................................................................................21
Figure 7. Length frequency distributions of larval rainbow smelt during 2008 in St. Martin Bay,
Lake Huron; May (top panel), June (middle panel), and July (lower panel) .................................23
Figure 8. Length frequency distributions of larval rainbow smelt during 2009 in St. Martin Bay,
Lake Huron; May (top panel), June (middle panel), and July (lower panel) .................................24
Figure 9. Finite mixture models fit to length frequency distributions of larval rainbow smelt
observed on 27 May, 11 June, 18 June, 24 June, 8 July and 22 July, 2009. Dashed lines
represent component (joint) models and solid lines represent mixed probability density functions.
Mean length is indicated by the triangle along the x-axis .............................................................25
Figure 10. Modeled exponential growth curves for larval rainbow smelt during 2008 (top panel)
and 2009 (bottom panel) showing the relationship between estimated age and length of size
classes ...........................................................................................................................................28
Figure 11. General linear model regressions (log (e) length (mm) v. estimated age) comparing
growth of larval rainbow smelt during 2008 (top panel) and 2009 (lower panel) in St. Martin
Bay, Lake Huron. .........................................................................................................................29
vii 
 

Figure 12. General linear model regressions (log (e) length (mm) v. estimated age) comparing
growth of early stream spawned (circle) and late lake spawned (triangle) larval rainbow smelt
cohorts during 2009 in St. Martin Bay, Lake Huron .....................................................................30
Figure 13. Larval rainbow smelt catch curves (log (e) density v. estimated age) comparing
instantaneous mortality rate of larval rainbow smelt during 2008 (triangle) and 2009 (circle), in
St. Martin Bay, Lake Huron ...........................................................................................................31
Figure 14. Bottom trawl sampling locations in the main basin of Lake Huron, 1976-2009 ........44
Figure 15. Stock-recruit relationships for Lake Huron rainbow smelt modeled by the Ricker
function for the years 1976-2009 (top panel), 1976-1991 (middle panel), and 1994-2009 (lower
panel)..............................................................................................................................................52
Figure 16. Frequency distributions of bootstrapped parameter estimates (log α, β) for the Ricker
stock-recruit model for the 1976-2009 time series. Approximate location of “true” parameter
estimates are indicated by arrows ..................................................................................................55
Figure 17. Frequency distributions of bootstrapped parameter estimates (log α, β) for the Ricker
stock-recruit model for the 1976-1991 time series. Approximate location of “true” parameter
estimates are indicated by arrows ..................................................................................................56
Figure 18. Frequency distributions of bootstrapped parameter estimates (log α, β) for the Ricker
stock-recruit model for the 1994-2009 time series. Approximate location of “true” parameter
estimates are indicated by arrows ..................................................................................................57

viii 
 

THESIS INTRODUCTION

The rainbow smelt Osmerus mordax is indigenous to the Atlantic coastal drainage from
Labrador to New Jersey, but has been widely introduced in North America (Scott and Crossman
1973). Rainbow smelt occur as a migratory marine species, but freshwater demes are indigenous
to eastern North America where they inhabit lake systems in Maine, New Hampshire, and
eastern and maritime provinces of Canada (Scott and Crossman 1973). Introduced as a prey
species, rainbow smelt have become established in numerous North American lakes outside their
native range, including the Laurentian Great Lakes and surrounding watersheds (Van Oosten
1937; Dymond 1944; Evans and Loftus 1987; Franzin et al. 1994).
In the Great Lakes region, rainbow smelt became established following a large planting
of eggs in Crystal Lake, Michigan in 1912 (Van Oosten 1937). Rainbow smelt populations that
developed in lakes Michigan, Huron, Superior, and Erie during 1918-1940 are believed to have
been a result of this single planting (Creaser 1926; Van Oosten 1937). In the Great Lakes,
rainbow smelt were introduced as a prey species for native and stocked salmonines and their
populations built up quickly, especially in lakes Michigan and Huron where the species became
abundant by the late 1920s (Van Oosten 1937). Following their establishment in the Great
Lakes, rainbow smelt became so abundant that commercial fisheries targeted the species for
human consumption (Schneberger 1937), first in lakes Michigan and Huron by the early 1930s,
and in lakes Erie, Superior, and Ontario by the early 1950s (Baldwin and Saalfeld 1962; Christie
1974; Baldwin et al. 2009; Figure 1). Commercial fishery yield of rainbow smelt continued to
increase through the 1980s, but since the late 1990s harvest has been greatly reduced across the
Great Lakes due to dramatic declines in abundance (Figure 1). Today, small scale commercial

1 
 

fisheries for rainbow smelt operate in Canadian waters of lakes Erie and Huron, and in the
Minnesota waters of Lake Superior (Baldwin et al. 2009).
350

5000

Huron

300

Superior

4000
250
3000

200
150

2000

Commercial Harvest (1000’s of pounds)

100
1000

Michigan

50
0

Ontario

0
300

Michigan

8000

Ontario
6000

200

4000
100

Erie

2000
0

0
1940

40000

1960

1980

2000

Year

Erie

30000
20000
10000
0
1940

1960

1980

2000

Year

Figure 1. Commercial harvest (1000s of pounds) of rainbow smelt for all five Great Lakes
during 1931-2006, adapted from Baldwin et al. 2009.

2 
 

As a naturalized exotic species, rainbow smelt have integrated into the pelagic food webs
of the Great Lakes as both a prey species and predator. Accordingly, rainbow smelt are found in
the diets of sundry salmonine predators in lakes Huron (Kocik and Taylor 1987; Diana 1990;
Madenjian et al. 2006), Michigan (Stewart et al. 1981; Madenjian et al. 1998), Superior (Dryer et
al. 1965; Conner et al. 1993; Ray et al. 2007), Erie (Henderson and Nepszy 1989), and Ontario
(Brandt 1986). As a predator, rainbow smelt have been related to negative ecological
interactions with other fish species as they consume the larvae of native fish or compete for
zooplankton food resources, both of which inhibit young of the year recruitment in some species
(Selgeby et al. 1978; Loftus and Hulsman 1986; Evans and Loftus 1987; Hrabik et al. 1998;
Mercado-Silva et al. 2007; Myers et al. 2009). In Lake Huron, rainbow smelt are particularly
important as a food source for salmonines and percids as they have been the dominant prey
species following the collapse of alewife populations after 2004 (Riley et al. 2008).
Characteristic of small, pelagic fishes, rainbow smelt populations are typified by
stochastic recruitment processes that result in variable year class formation (Nellbring 1989).
Recruitment variability and population declines of rainbow smelt in the Great Lakes have been
related to a combination of abiotic and biotic factors influencing growth and survival of youngof- the-year (YOY). Sources of direct mortality such as predation (He and LaBar 1994; Gorman
2007) and cannibalism (Henderson and Nepszy 1989; Lantry and Stewart 2000; Parker Stetter et
al. 2007) have been shown to be important in structuring mean size and abundance of rainbow
smelt populations. Alternatively, Hoff (2004) and Gorman (2007) found that when other prey
species are available to lake trout (Salvelinus namaycush) and other large predators, rainbow
smelt recruitment is positively affected through reduced predation. Mortality and growth during
the egg or larval stage has also been demonstrated to impact overall year-class strength of

3 
 

rainbow smelt. Factors reported to influence early life history stages include spring climate
variables such as precipitation, mean air temperature, and solar intensity (Sluka 1991; Hoff
2004) all of which can modify the suitability of stream or nearshore habitats for egg or larval
stages of rainbow smelt. For example, McKenzie (1964) found a significant negative association
between larval production and rainfall occurring after spawning, likely due to displacement or
stranding of eggs relating to stream flashiness.
In Lake Huron, rainbow smelt abundance has been monitored annually since 1973 by the
U.S. Geological Survey (USGS) Great Lakes Science Center (GLSC), through an annual bottom
trawl survey that effectively targets all prey species (Riley et al. 2010). Between 1976 and 1991,
yearling and older (YAO) rainbow smelt were relatively abundant in this survey, but abundance
of YOY rainbow smelt was generally lower than during 1994-2009 (Figure 2). However, from
1994 to 2009, YAO rainbow smelt abundance has been declining steadily, while abundance of
YOY has oscillated dramatically with several large year classes forming during this period
(Figure 2). Current abundance estimates of YAO rainbow smelt in Lake Huron are
approximately 40% lower than during 1976-1994, a period of relatively high abundance (Riley et
al. 2008; Figure 2). This steady recent decline of rainbow smelt over the past decade in this lake
has been a concern for fishery managers tasked with matching stocking numbers of predators
with available prey biomass.

4 
 

2500

4000

2000

3000

1500

2000

1000

1000

500

0

0

3000

600

2000

400

1000

Relative Standard Error

3000

5000

Density (No./ha.)

6000

200

0
1980

1990

2000

0
2010

Year

Figure 2. Density (No /ha., solid line, closed circle) and relative standard error (dotted line, open
triangle) of YOY (top panel) and YAO (lower panel) rainbow smelt estimated from USGS
bottom trawl surveys, 1976-2009, adapted from Riley et al. (2010).

5 
 

The goal of this thesis was to evaluate the influence of selected abiotic and biotic factors
on the growth, survival and recruitment dynamics of rainbow smelt in Lake Huron. This thesis is
based on 2 years of field research in St. Martin Bay, northern Lake Huron (Chapter 1) combined
with stock-recruit analysis of a thirty year time series of rainbow smelt abundance (chapter 2).
The specific objective of research presented in Chapter 1 was to determine patterns in
abundance, growth, and survival of larval rainbow smelt populations in St. Martin Bay, Lake
Huron during 2008-2009, while the focus of Chapter 2 was on stock-recruit modeling using data
from 1976-2009. This latter chapter further evaluated the role of water temperature, water
levels, precipitation, and lake trout CPE in causing the noted variation in recruitment during this
time period.
The research presented in this thesis increases our understanding of prey fish dynamics in
Lake Huron by adding to our knowledge of environmental and biological factors influencing
recruitment processes of rainbow smelt. For example, knowledge of rainbow smelt recruitment
in relation to lake trout abundance emphasizes to fishery managers the important link between
rainbow smelt as a prey resource and the restoration of lake trout populations in the Great Lakes.
Factors such as precipitation and water temperature that are projected to increase in the Great
Lakes basin given current climate change scenarios (Loftgren 2002) are important to consider in
relation to future fish recruitment and restoration efforts. For example, increasing water
temperatures in the Great Lakes are expected to have significant impacts on growth rates of fish
populations with rainbow smelt and other coldwater species expected to have reductions in
recruitment (Casselman 2002), further reducing the availability of this key prey species to our
valuable salmonine and percid communities.

6 
 

CHAPTER ONE
ABUNDANCE, GROWTH, AND SURVIVAL OF LARVAL RAINBOW SMELT IN ST.
MARTIN BAY, LAKE HURON, 2008-2009

Introduction
Rainbow smelt (Osmerus mordax) are indigenous to northeastern North America where
they originally inhabited the Atlantic drainage from Labrador to coastal New Jersey (Scott and
Crossman 1973). Rainbow smelt occur as a migratory marine species, and are also native to
inland lake systems in Maine, New Hampshire, and the eastern and maritime provinces of
Canada (Scott and Crossman 1973). Beyond their native range, rainbow smelt have been
introduced widely as a prey species in North America, primarily in freshwater systems including
the Laurentian Great Lakes and surrounding watersheds (Van Oosten 1937; Dymond 1944;
Evans and Loftus 1987; Franzin et al. 1994).
The reproductive cycle of rainbow smelt is initiated during April, usually during the
winter-spring transition period. Rainbow smelt spawn typically near ice out or when water
temperatures reach 3-4 C (Creaser 1926; Rupp 1959). Adult rainbow smelt migrate into streams,
typically at night, where females deposit eggs over a variety of substrates including sand, gravel,
and sparse vegetation (McKenzie 1964). Smelt eggs sink rapidly and adhere readily to substrates
where they are in turn fertilized by males (Rupp 1965). Incubation times vary according to water
temperature and range from 32 days at 5 C to 9 days at 20 C (Cooper 1978; Ayer et al. 2005).
After hatching, larvae are carried by currents into receiving waters where they transition to
exogenous feeding and begin a pelagic larval phase as yolk reserves are expended.

7 
 

Researchers previously studying larval rainbow smelt have recognized that many
populations consist of early hatching cohorts originating from streams and late hatching cohorts
that come from lake spawning sources. An intensive three year study conducted in St. Martin
Bay, Lake Huron by Brown (1994) determined that larval contributions from lake spawning were
moderate and less variable than larvae originating in streams. Further, the abundance of stream
spawned larval populations was highly variable among years and as a result drove the variability
in rainbow smelt year class formation in St. Martin Bay (Brown 1994). In eastern Lake
Michigan, Tin and Jude (1983) found high densities of larvae in shallow water during May, and
noted the presence of late June and July hatching cohorts originating in deeper, colder waters of
the lake where incubation occurred more slowly. While the timing and location of lake
spawning is not well understood, this reproductive strategy appears to be common among
rainbow smelt populations (Rupp 1965; MacCallum and Regier 1970; Owens 1982), and the
source of larval cohorts hatching during late June and July. Both stream and lake sources of
larvae contribute to year class strength and form the basis of rainbow smelt populations in
northern Lake Huron.
Rainbow smelt are both a pelagic prey species and predator across the Great Lakes and
populations are characterized by highly variable recruitment and more recently, declining
abundance across the region (Bronte et al. 2003; Mills et al. 2005; Bunnell et al. 2006; Riley et
al. 2008). Reductions in the biomass of larger (>100 mm TL) rainbow smelt have been
attributed to high levels of predation by salmonines (Bronte et al. 2003; Mills et al. 2005;
Gorman 2007), leaving many stocks to rely on consistently strong year classes to sustain the
population. Recruitment variation however and specifically, factors such as temperature that
influence growth and survival during larval stages are not well understood.

8 
 

The goal of this research was to evaluate the growth and survival dynamics of larval
rainbow smelt in northern Lake Huron and to relate these findings to recruitment variability for
this species in the Great Lakes. A model system to conduct this research, based on the work of
Brown (1994), was St. Martin Bay, Lake Huron. The specific research objectives for this study
were to (1) determine the influence of tributary water temperatures on the duration of rainbow
smelt spawning during April and subsequent larval production during May, and (2) determine the
abundances, growth rates, and mortality rates of larval rainbow smelt cohorts in St. Martin Bay,
Lake Huron during May-July, 2008-2009. For objective (1) it was hypothesized that a longer
spawning period would result in higher densities of larvae originating from streams in May.
Methods
Study Area
St. Martin Bay is a large embayment in northwestern Lake Huron and borders the south
shore of Michigan’s eastern Upper Peninsula (Figure 3). The bay averages 8 m in depth and has
a maximum depth of approximately 30 m. At its widest points, St. Martin Bay is approximately
12 km east to west, and 8.5 km north to south. Approximately 16 km of sand, boulder and
cobble covered shoreline border the bay, most of which is undeveloped. St. Martin Bay is
characteristically mesotrophic (Brown et al. 1995) with emergent vegetation in productive littoral
areas near tributary mouths and clear waters offshore. Outside the littoral zone (>1 m) are
typically sand or silt substrates with isolated aquatic macrophyte zones. St. Martin Bay is fed by
three larger (3rd and 4th order) and four smaller (1st and 2nd order) tributaries that were historic
spawning areas for migratory rainbow smelt (Goodyear et al. 1982). St. Martin Bay and the Les

9 
 

Chenaux Islands area provide excellent nursery habitat for developing rainbow smelt larvae
(Goodyear et al. 1982), which made this area ideal to investigate growth and survival dynamics.

Figure 3. St. Martin Bay, Lake Huron showing ichthyoplankton sampling transects (solid black
lines) and spawning streams surveyed (Carp River, Nunns Creek, Spring Creek, and St. Martin
Creek) during 2008-09.

10 
 

Stream surveys
Rainbow smelt spawning occurs from mid-April to early May in the upper Great Lakes
(Scott and Crossman 1973) and is initiated by a spring freshet and increasing water temperatures.
To determine the influence of water temperature on spawning duration, streams were monitored
for temperatures between early April and late June during both years of the study to estimate
length of the spawning run and characterize warming patterns in spawning streams. The four
streams studied in this bay had been identified as spawning habitats in St. Martin Bay (Goodyear
et al. 1982; Brown 1994) including the Carp River, Nunns Creek, Spring Creek, and St. Martin
Creek (Figure 3). In all four study streams HOBO Water Temp pro v2 temperature loggers
(Onset Computer Corporation) were placed upstream from known spawning habitats in the lower
reaches of rivers and streams, where rainbow smelt typically spawn (Bailey 1964). Data loggers
collected water temperatures (degrees C) at 15 minute intervals before and during spawning and
throughout the larval outmigration period.
Rainbow smelt spawning in study tributaries was evaluated by conducting stream dip-net
surveys during April of 2008 and 2009. Spawning rainbow smelt were sampled in all four study
streams that were monitored with temperature loggers. Each stream was sampled twice nightly
for spawning rainbow smelt (between 2200-0100 hrs and 0100-0400 hrs). Rainbow smelt were
sampled with wire mesh dip-nets (10 mm mesh) in spawning locations in each stream that were
verified by visual observation. Each site was sampled for approximately 45 minutes, or until 100
fish were collected. Dip-net surveys were important for verifying that fish were spawning in
area streams, and when combined with temperature data allowed for estimation of the length of
the spawning run.

11 
 

Ichthyoplankton surveys
To sample St. Martin Bay for larval rainbow smelt, fixed site icthyolplankton surveys
were conducted between May and August during 2008-2009, using a study design similar to
Brown (1994). All larval fish sampling was conducted in the surface waters (0.5-1 m) of St.
Martin Bay. From May through early July 2008, larval rainbow smelt were sampled with 0.5 m
diameter, 500 micron conical plankton net with a metal frame, modified from a design by Nester
(1987), with a 22 kg brass depressor suspended below the frame. Flow meters were attached to
all larval fish gear to determine volume of water filtered. In order to increase sample sizes of
larval fish, rainbow smelt were sampled with a 1x2x4 m neuston net with 500 micron mesh for
the remainder of the 2008 sampling season and all of 2009. This mesh size has been shown to be
effective at catching larval rainbow smelt (O’Gorman 1984), and has been used in previous
rainbow smelt research in St. Martin Bay (Brown 1994). To adequately sample depth strata and
surface area within St. Martin Bay, 12 larval transects were selected over 2.5, 5, and 10 m
contours and distributed around the bay to representatively sample available nearshore and
offshore habitats (Figure 3). All ichthyoplankton sampling occurred at night (McKenzie 1964);
sampling typically occurred between 2200-0300 hrs. Tows were conducted for 5 minutes at a
speed of approximately 4 km/hr. Transects were sampled weekly during May and June, and bimonthly during July and August in both years. Survey dates were from 12 May – 19 August
during 2008 and 21 May – 24 August during 2009. In 2008, 136 larval fish samples were
collected while 105 samples were collected during 2009. Water temperatures were recorded at
the start of all larval fish tows.
Once collected, all larval fish samples were rinsed into 1 liter bottles and stored in 90%
EtOH until analyzed. In the laboratory, all larval fish were identified to species (Auer 1982) and
12 
 

enumerated. Catches of larval rainbow smelt were converted to density and expressed as
numbers per 1000 m^3 of water. Rainbow smelt larvae were measured to the nearest 0.1 mm
through a stereoscopic microscope interfaced with a digital image analysis system.

Length frequency analysis of larval cohorts
Length frequency analysis was used to determine if early and late spawned cohorts were
present in the population (i.e., more than one mode or age class) and to determine mean length of
surviving larval cohorts for the computation of mean population growth rates (Ricker 1975). For
each year, composite length frequency plots were created by grouping all larvae caught by
month. Length frequency distributions showing distinct modal separation indicate multiple age
groups are present in the population being sampled (Macdonald 1987). If multiple modes were
present in composite distributions, length frequency distributions were computed by sampling
date. When clear separation of length groups was still evident during individual sampling
events, mixed probability density functions (mixture models) were fit to each distribution to
identify changes in mean length of each group through time (Ricker 1975). The mixed
probability density function in the general form is a weighted sum of k component probability
densities:

| , ,

| ,

|

,

Where μ corresponds to the means of component length distributions, σ is standard deviations of
component distributions, and π is the corresponding mixing proportions of components (Everitt
and Hand 1981). To fit mixture models to observed frequency distributions, R statistical
programming software (R Development Core Team 2009) was used with the Mixdist package
13 
 

for fitting finite mixture distribution models (Peter Macdonald with contributions from Juan Du
2008). All length frequency distributions were developed using 1 mm bins. The R package
Mixdist fits finite mixture models using maximum likelihood and Newton-type and EM
algorithms. All mixture models were fit using normal distributions and constant coefficient of
variation (CCV). For each model, a goodness of fit test based on the chi-square likelihood ratio
statistic (Rao 1965) was used to determine how well mixture distributions g(x) fit observed
length frequency histograms (Du 2008). This test statistic is expressed as:

2 ln
Where L(β) is the maximum value of the likelihood function for the full model and L(β1) is the
maximum value for the likelihood function for the reduced model (Montgomery et al. 2001). All
Chi-square tests were evaluated with α = 0.05.
Larval growth
Growth rates of rainbow smelt from late May through July were estimated using changes
in larval lengths throughout the sampling period (Hackney and Webb 1978, DeAngelis et al.
1980, Zigler and Jennings 1993). For each length group of larvae (1mm bins), a plot of density
(No./1000m^3) through time yielded an estimated age for each length group by determining the
mean day that an individual larvae attained a given size (Hackney and Webb 1978; Zigler and
Jennings 1993). This method allows growth to be estimated empirically under existing
environmental conditions and allows for variable growth without a “fixed” rate (Hackney and
Webb 1978). Length was then plotted against calculated mean day of capture and modeled using
the exponential function:

14 
 

Where L = larval length following the growth period, a = the initial larval length or the length
axis intercept, G = the instantaneous growth coefficient, and t = time in days. This derived
growth curve estimates average growth rates for larvae over the sampling period (Hatch and
Underhill 1988).
Growth between years and between 2009 early and late hatching cohorts was compared
using ANCOVA (Proc GLM, SAS Institute 1999). For general linear models, all length data
were log transformed to meet the assumption of normality. Larval growth was compared
between years by regressing natural log transformed length against estimated age. The slope of
each regression was compared with an F-test to test the assumption of parallel slopes (Neter et
al. 1996). During 2009, mean population growth rate was compared between early and late
hatching cohorts by tracking the progression of mean length of cohorts through time (Ricker
1975). Linear regression (i.e., log (e) mean length via mixture models vs. date of capture) was
then employed to determine if average growth rate was different between cohorts in the same
year. Each regression slope was tested for equality with an F-test, as in the cross-year
comparison. All statistical tests were evaluated with α = 0.05.

Larval Mortality
Mortality rates of larval rainbow smelt were estimated by plotting age against abundance
through time (No/1000 m^3) for each 1 mm length group. Mortality was then modeled using the
exponential decay function:

15 
 

Where Nt = abundance at time t, No = initial abundance, Z = the instantaneous mortality rate,
and t = time in days (DeAngelis et al. 1980). This relationship describes average mortality rates
for larval rainbow smelt cohorts over the growing season. Larval rainbow smelt densities
(No/1000 m^3) were natural log transformed (Log (e) (x+1), Brown and Guy 2007) for linearity
and regressed against age to yield a catch curve, from which instantaneous mortality could be
estimated (Ricker 1975). Differences in mortality were compared among years using ANCOVA
(Proc GLM, SAS Institute) and all statistical tests were evaluated with α = 0.05.

Results
Stream surveys
Patterns in stream temperatures and warming rates varied considerably between years
during the early spring spawning period in the St. Martin Bay region. During 2008, temperature
loggers were placed in streams at the onset of spawning on April 17th. During the first week of
documented spawning, water temperatures increased quickly on all St. Martin Bay tributaries
due to warmer than average air temperatures in the region. Nunns Creek, Spring Creek, and St.
Martin Creek were between 3-4 C when water temperature sampling began, but increased to
between 9 and 11 C the following week (Figure 4). The Carp River, the warmest and largest
tributary, was approximately 6 C when temperature sampling began and increased to near 12 C
by the following week (Figure 4). This period of rapid warming during 2008, resulted in a
spawning period between 7-10 days. Temperatures in tributaries declined following this

16 
 

warming period (Figure 4), but surveys conducted on spawning rainbow smelt thereafter
indicated that spawning had ceased following this time period in all study tributaries.
During 2009, temperature loggers were placed in streams on April 6th, nearly two weeks
before spawning was observed. Stream temperatures remained cooler during the 2009 spawning
period, and spawning was first observed in the Carp River on April 13th when stream
temperatures were approximately 3.5 C. Nunns Creek warmed more quickly than Spring or St.
Martin creeks during 2009, but remained cooler (e.g., 6 C cooler on 23-May 2009 compared to
the same date in 2008) and few spawning rainbow smelt were observed in Nunns Creek during
2009. Spring Creek and St. Martin Creek remained relatively cold and below 3 C until
approximately April 23, and no spawning by rainbow smelt was observed in these smaller
tributaries during 2009. Observations of spawning rainbow smelt and recorded water
temperatures indicated that most stream spawning in St. Martin Bay during 2009 occurred in the
Carp River. Fewer tributaries were utilized by spawning rainbow smelt during 2009 but the
estimated spawning period in the Carp River was nearly two weeks longer than 2008 (Figure 4).

17 
 

20

St. Martin Creek
Spring Creek
Nunns Creek
Carp River

15

10

2008

Temperature (C)

5

0
20

15

10

2009

5

0
4/20/09

5/4/09

5/18/09

6/1/09
Date

6/15/09

6/29/09
 

Figure 4. Stream temperatures recorded from four St. Martin Bay Spawning tributaries from
April-June during 2008 (top panel) and 2009 (lower panel); St. Martin Creek, Spring Creek,
Nunns Creek, and the Carp River. Shaded regions indicate the observed duration of spawning by
rainbow smelt.

18 
 

Ichthyoplankton surveys
During 2008 and 2009, larval rainbow smelt populations in St. Martin Bay had similar
abundance patterns with the highest larval rainbow smelt densities occurring during May. Peak
larval density during May 2008 was nearly double that observed during May of 2009 (Figure 5).
Irrespective of a shorter spawning period during 2008 (2 weeks shorter than 2009), initial
densities of larvae originating from streams were higher than those found in 2009 (Figure 5).
Mean larval rainbow smelt density peaked at approximately 60 / 1000 m^3 in 2008, and 30 /
1000 m^3 in 2009 (Figure 5) occurring on May 19 and May 21st, respectively. During 2008,
density declined rapidly through June and early July, and then increased during the 2nd week of
July concurrent with an increase in smaller, newly hatched larvae entering the population from
lake spawning populations. Following this increase, larval abundance declined further during
July and August. During 2009, larval density decreased following peak abundance in mid-May,
but then increased along with smaller, newly hatched fish during the last week of June (Figure
5). This increase in density during 2009 occurred nearly two weeks earlier than during 2008 due
to warmer water temperatures in St. Martin Bay in late June (Figure 6), which resulted in shorter
incubation periods for eggs. Abundance trends of larval rainbow smelt in St. Martin Bay
indicated that one or more pulses of larvae entered the population during late June and July of
2008 and 2009, after spawning and larval outmigration in tributaries had ceased.

19 
 

60

2008

Density (No./1000m3 )

40

20

0
50

2009

40
30
20
10
0
5/11
 

5/25

6/08

6/22

7/06
Date

7/20

8/03

8/17
 

Figure 5. Abundance (Mean No /1000 m^3) of larval rainbow smelt in St. Martin Bay, Lake
Huron from May-August during 2008 (top panel) and 2009 (lower panel). Error bars are 95%
confidence intervals.

20 
 

Spring and summer mean water temperatures and rates of warming in St. Martin Bay
differed over the two year period (Figure 6). During both years, water temperatures averaged
approximately 13 C over the period from mid May to late August. Mean water temperature
increased steadily during 2008 and reached a peak of 19.67 C on August 4th, with an overall
warming rate of approximately 0.13 degrees per day. During 2009, water temperature increased
rapidly during May and June and reached a peak mean temperature of 20.2 C on June 24, over a
month earlier than peak temperatures were reached during 2008. Following this period of rapid
warming, mean water temperatures decreased by nearly 10 C during early July and remained
notably cooler relative to 2008 (Figure 6). As a result the calculated rate of warming during
2009 was 0.06 degrees per day, approximately half the estimated warming rate during 2008.

22

Temperature (degrees C)

20

2008

18
16
14

2009

12
10
8
6
5/17

5/31

6/14

6/28
Date

7/12

7/26

8/9
 

Figure 6. Mean water temperature (degrees C) across sampling transects recorded during larval
fish surveys, 12 May-5-August, 2008 (closed circle) and 21-May-4-August, 2009 (open circle).

21 
 

Length frequency analysis of larval cohorts
Analysis of length frequency distributions of larval rainbow smelt indicated differences
within years relative to the survival of different cohorts. During both years, May larval length
frequencies were similar and depicted newly hatched larvae had entered St. Martin Bay (Figures
7, 8). During June 2008, uni-modal length frequency distributions indicated the population was
comprised of a single cohort (Figure 7) which was distributed across newly hatched larval
lengths. This pattern indicated that larvae hatched during May had experienced high mortality.
During June and July 2009, a bi-modal length distribution was evidence that the population
consisted of an older (i.e., stream spawned) cohort and a smaller (i.e., lake spawned) cohort
(Figure 8).
The occurrence of larval cohorts distinguishable by size facilitated comparison of
population growth rates for each larval cohort from late May-July 2009. Only data from 2009
were used to compare growth among cohorts, as this was the only year where early hatched
larvae remained in the population throughout the May-July sampling period. Length frequency
distributions for 2009 were constructed by individual sampling date to determine the presence of
both stream and lake cohorts at a differing temporal scale, and to determine mean length of each
cohort. Six out of nine sampling events during 2009 yielded bi-modal length distributions
indicative of multiple larval cohorts. Finite mixture models were fit to length frequency
distributions observed on 27 May (n = 117), 11 June (n = 69), 18 June (n = 34), 24 June (n =
134), 8 July (n = 47) and 22 July (n = 48), and the mean length of each component distribution
was determined (Figure 9). Chi-square tests indicated all model fits were significant and
adequately modeled the grouped length frequency data for a given sample period (Table 1).

22 
 

160

May
120

80

40

0

Frequency

June
20

10

0

July
20

10

0

0

5

10

15

20

25

30

Length (mm)
Figure 7. Length frequency distributions of larval rainbow smelt during 2008 in St. Martin Bay,
Lake Huron; May (top panel), June (middle panel), and July (lower panel).

23 
 

200

May
150

100

50

0

June
Frequency

30

20

10

0

July
30

20

10

0
0

5

10

15

20

25

30

Length (mm)
Figure 8. Length frequency distributions of larval rainbow smelt during 2009 in St. Martin Bay,
Lake Huron; May (top panel), June (middle panel), and July (lower panel).

24 
 

11 June

0.0

0.00

0.2

0.15

0.4

0.30

27 May

5

6

7

8

9

10

6

8

10

12

14

16

24 June

0.00

0.15

0.10

0.30

18 June

0.00

Probability Density

4

5

10

15

5

10

20

22 July

0.00

0.00

0.10

0.10

0.20

8 July

15

5

10

15

20

5

10

15

20

25

30

35

Length (mm)
Figure 9. Finite mixture models fit to length frequency distributions of larval rainbow smelt
observed on 27 May, 11 June, 18 June, 24 June, 8 July and 22 July, 2009. Dashed lines
represent component (joint) models and solid lines represent mixed probability density functions.
Mean length is indicated by the triangle along the x-axis.

25 
 

Table 1. Parameter estimates for finite mixture distribution models fit to larval rainbow smelt
length frequency distributions during 2009; π = mixing proportions, μ = mean cohort length, σ
= standard error. Cohort 1 corresponds to the stream spawned larvae, cohort 2 corresponds to
lake spawned larvae. Model degrees of freedom (Df) and Chi-square goodness-of-fit statistic
(X^2) are presented for each model. All models were significant at α = 0.05. Modeled length
distributions for 18-June and 8-Jul identified a third mode which was not included in this
analysis due to low (i.e., < 10 larvae) sample sizes. Thus, mixing proportions for those sample
dates do not add to 1.

2

Date

Cohort

π

μ

σ

Df

X

p-value

27-May

1

0.068

5.070

0.066

5

4.624

0.463

2

0.932

6.865

0.772

1

0.184

6.949

0.292

15

20.752

0.145

2

0.816

10.664

2.078

1

0.434

7.498

1.160

5

10.531

0.062

2

0.448

11.310

0.906

1

0.467

7.057

1.248

11

15.242

0.172

2

0.532

12.469

2.223

1

0.306

9.832

1.181

6

3.588

0.732

2

0.629

14.068

1.287

1

0.196

9.734

1.356

20

21.599

0.3627

2

0.610

16.277

1.707

11-June

18-June

24-June

8-July

22-July

26 
 

Larval Growth
Growth and mortality rates were calculated for larvae ranging from 5 to 19 mm in length,
corresponding approximately to the first 50 days of growth. Growth rates varied between years
and exponential growth models indicated that larval growth was higher during 2009; the average
-1

-1

growth rate was 0.16 mm d during 2008 and 0.23 mm d during 2009 (Figure 10). Daily
-1

growth rates ranged between 0.11 - 0.19 mm d during 2008 and between 0.15 – 0.30 mm d

-1

during 2009. General linear models confirmed that the average growth rate was significantly
higher during 2009 (F 3, 25 = 5.72, P = 0.02, Figure 11). During 2009, mean growth rates were
higher for the early hatching stream cohorts, but this difference was not significant between
stream (G=0.014) and lake cohorts (G=0.012) (F3, 8 = 1.35, P = 0.28, Figure 12). However,
length intercepts were significantly different indicating cohorts had distinct growth trajectories
(F2, 9 = 41.23, P = 0.0001).

Larval Mortality
Exponential models indicated instantaneous mortality rates during 2009 were higher (z =
0.050) than those found for 2008 cohorts (z = 0.045). However, general linear models confirmed
that mortality was not significantly different between years (F3, 24 = 0.27, P = 0.61). Log
density at peak abundance was significantly different (F2, 25 = 16.65, P = 0.0004) and higher
during 2008, indicating significantly more larvae were present during May in 2008 (Figure 13).

27 
 

20

2008

16

12

8

Length (mm)

4

0
20

2009

16

12

8

4
0
 

10

20

30
Age

40

50
 

Figure 10. Modeled exponential growth curves for larval rainbow smelt during 2008 (top panel)
and 2009 (bottom panel) showing the relationship between estimated age and length of size
classes.
 

28 
 

 

3.0

2008

2.5

Log e length (mm)

2.0

1.5

3.0

2009
2.5

2.0

1.5

0
 

10

20

30

40

50

Age

Figure 11. General linear model regressions (log (e) length (mm) v. estimated age) comparing
growth of larval rainbow smelt during 2008 (top panel) and 2009 (lower panel) in St. Martin
Bay, Lake Huron.

29 
 

3.0
Late Cohort
Early Cohort

Log e length (mm)

2.8
2.6
2.4
2.2
2.0
1.8
1.6
10

20

30

40
Age

  

50

60
 

Figure 12. General linear model regressions (log (e) length (mm) v. estimated age) comparing
growth of early stream spawned (circle) and late lake spawned (triangle) larval rainbow smelt
cohorts during 2009 in St. Martin Bay, Lake Huron.

30 
 

Log e density (No./1000m3)

5
2008
2009

4

3

2

1
0
  

10

20

30
Age

40

50
 

Figure 13. Larval rainbow smelt catch curves (log (e) density v. estimated age) comparing
instantaneous mortality rate of larval rainbow smelt during 2008 (triangle) and 2009 (circle), in
St. Martin Bay, Lake Huron.

Discussion
Rainbow smelt were the most abundant ichthyoplankter in St. Martin Bay, Lake Huron
from May-July during 2008 and 2009. These populations were characterized by contrasting
patterns in abundance, growth, and survival during both years of this study, with growth rates
differing significantly between years. During 2009, growth rates were significantly higher for
larval rainbow smelt, indicating that water temperatures and zooplankton dynamics were more
favorable for growth and survival of larvae than during 2008. Despite a relatively short (7-10
31 
 

day) spawning period in 2008 and a more protracted spawning period in 2009 (15-20 days),
initial densities of larval rainbow smelt were higher during May of 2008, with peak density
nearly double that estimated during 2009. The original hypothesis that a longer spawning period
would result in higher densities of larval rainbow smelt was thus unsubstantiated, and it appears
that the number of streams utilized for spawning is more important than length of the spawning
period. This would suggest that given optimal water temperatures and flow rates in streams,
spawning can successfully occur over a shorter time period (7-10 days) than has previously been
observed in the Great Lakes region (Owens 1982; Brown 1994).
Within years, changes in size distributions of larval rainbow smelt indicated that
populations consisted of at least two cohorts; early hatching larvae entering the bay during May
and early June and late hatching larvae entering the bay during late June and early July, a pattern
also seen by Brown (1994) in this same system. These late hatching larval cohorts were likely
from lake spawned origins, as larval outmigration from tributaries typically ends by the first
week in June (Brown 1994). During 2009, these cohorts of late hatching larvae began
approximately two weeks earlier than in 2008, likely due to warmer water temperatures in St.
Martin Bay which accelerated incubation during June of 2009. Although distinct stream and
lake cohorts developed during both sampling seasons, stream hatched larvae apparently
experienced large-scale mortality between May and early June of 2008, as indicated by a paucity
of larger (15-20 mm) larvae and a population consisting primarily of smaller (6-7 mm) larvae
during June.
The present study was in part a reassessment of the work of Brown (1994) who
conducted research on rainbow smelt populations in St. Martin Bay, Lake Huron during 1991-93.
That study determined that strong year classes developed when tributary production of larvae

32 
 

was strong and spatial overlap with their zooplankton food resources was high. More
importantly, estimates of growth and mortality appeared highly correlated with zooplankton
populations. For instance, high growth rates and relatively low mortality during 1991 (G=0.031,
Z=0.098) resulted in the highest year class recorded and an estimated 46 million larvae produced
from tributaries within St. Martin Bay (Brown 1994). During 1992, low tributary production
estimated at 13 million larvae was associated with low zooplankton densities, low spatial overlap
with prey, and subsequent low growth and high mortality of larvae within the bay (G=0.018,
Z=0.184; Brown 1994).
Significant inter-annual differences in growth of larval rainbow smelt populations were
observed in St. Martin Bay during 2008-2009, similar to variability in growth rates observed by
Brown (1994) in this same embayment. This finding, that large variation in growth rates occurs
during rainbow smelt early life history, has important implications for recruitment of this species
in the Great Lakes, as growth and mortality are linked processes during fish early life history that
are often highly correlated (Cushing 1975; Houde 1997). Thus, growth and mortality can act to
control year-class strength. For instance, reduced growth rates can result in protracted
developmental stage durations, effectively prolonging vulnerability of larvae to size dependant
mortality sources such as starvation or predation (Pepin 1991; Rice et al. 1993). Thus, in years
where larval growth is higher, probability of survival would accordingly be higher for
developing larval cohorts. High recruitment from larval to juvenile stages in fish populations is
associated with fast growth, which allows for individual larvae to better exploit optimal
temperature and food gradients (Cowan and Shaw 2002), and to evade predation, one of the
primary sources of early life stage mortality in fishes (Houde 2008). As an example, reduced
risk of size-dependant predation and an increase in recruitment potential resulting from increased

33 
 

and variable growth rates were described through model simulations for bloater (Coregonus
hoyi), a common pelagic species in the Great Lakes (Rice et al. 1993).
Many environmental factors can influence fish growth (e.g., zooplankton, salinity,
oxygen, solar intensity; Brett 1979), but temperature can impart profound changes on both the
development of eggs and growth of larvae (Blaxter 1969; Fry 1971). Because temperature can
influence the availability of zooplankton and the amount of food required by larval fish,
temperature directly influences fish growth rates (Jones 2002). While water temperature was not
directly correlated to fish growth rates for the present study, contrasting patterns in St. Martin
Bay water temperatures that were observed over the course of this study could have resulted in
the observed disparity in population growth rates between years. For instance, water
temperatures in St. Martin Bay during June 2008 slowly increased (i.e., from 10-13 C over a 14
day period) but during June 2009 bay temperatures increased dramatically (i.e., from 12-20 C
over a similar 14 day period). During July of 2008, water temperatures increased gradually and
averaged 17 C, while notably cooler temperatures persisted during July 2009 when the average
surface temperature was 13 C. Young of the year rainbow smelt prefer water temperatures in the
range of 13-14 C (Brandt et al. 1980), and temperatures during July 2009, when most larval
cohorts had entered the population, appear to have created optimal conditions for rainbow smelt
growth within St. Martin Bay. Conversely, during July 2008 average water temperatures in St.
Martin Bay exceeded the preferred range for rainbow smelt and likely contributed to the lower
growth rates observed.
The earlier emergence of lake spawned larval cohorts during 2009 may also have been a
result of temperature patterns in St. Martin Bay. Elevated temperatures increase development
rates and accelerate hatching in fish eggs (Braum 1978). In rainbow smelt, the relationship

34 
 

between temperature and time to hatch appears to be nonlinear, with cold temperatures (i.e., 2
and 4 C) resulting in hatch times of approximately 78 and 55 days, respectively, and warmer
temperatures (i.e., 8 and 10 C) resulting in hatch times of 24 and 17 days, respectively (Owen
1982). In the Miramachi River, McKenzie (1964) reported hatching of rainbow smelt eggs in 29
days at 6-7 C, but temperatures near 12 C resulted in hatching occurring after only 11 days.
Average water temperatures of 16.5 C resulted in Lake Erie rainbow smelt hatching between 7-8
days (Cooper 1978). These empirical studies indicate that rainbow smelt hatch over a wide
range of temperatures, and warmer temperatures (i.e., > 8 C) appear to substantially accelerate
egg development, likely as a result of cumulative degree days, which is the product of time to
development and temperature (Braum 1978). Although temperatures were not recorded at
deeper depths where lake spawned cohorts were likely hatching, St. Martin Bay is fairly shallow
(average depth 8 meters) and rarely stratifies during summer (Brown 1995), which would
suggest that warmer surface water temperatures during June 2009 would have in turn warmed the
deeper waters of the bay, thus causing the earlier emergence of lake spawning cohorts that was
observed during this year.
Mortality during the early life history of fishes can occur for a variety of reasons
including starvation, advection to suboptimal habitats, and predation (Houde 2002). During
2008, causes for the failure of early hatching stream cohorts were not apparent but were likely
related to lower growth rates and increased susceptibility of larvae to size specific mortality
sources, namely predation. Because growth is highly correlated to mortality in larval fish
(Houde 2002), it is reasoned that lower growth rates would complement higher mortality during
2008. Poor growth can function as a source of indirect mortality by increasing vulnerability to
predation mortality, which is considered to be the largest single cause of mortality during fish

35 
 

early life history (Paling 1971; Houde 2002). Size-dependent predation is closely associated
with growth and survival in larval fish populations because fish larvae are exposed to predators
longer and have a higher capture probability due to small size (Luecke 1990; Houde 2002). For
example, bloater larvae held in laboratory experiments under varied growth conditions showed a
50 fold increase in survival rate when rate of growth per day was increased from 0.3 to 0.5 mm
-1

d , with predation being the only form of mortality induced (Luecke et al. 1990). Rainbow
smelt are a key prey in the Great Lakes for lake trout Salvelinus namaykush (Diana 1990;
Madenjian et al. 2006) and introduced salmonines Oncorhynchus spp. (Kocik and Taylor 1987;
Diana 1990). In addition, rainbow smelt predation can also occur through cannibalism, which
has been shown to be an important source of mortality on young of the year rainbow smelt
populations (Henderson and Nepszy 1989; Lantry and Stewart 2000; Parker Stetter et al. 2007).
This researched has described the relationship between growth and survival dynamics in
larval rainbow smelt populations, and provides an understanding of the influence of these
processes on recruitment variation of rainbow smelt in Lake Huron. To fully understand and
appreciate the complexities of relating larval growth and mortality rates to recruitment, one must
acknowledge first that many processes act in concert to influence these fundamental rates
throughout fish early life history (Cowan and Shaw 2002; Houde 2008). Secondly, poor growth
and survival rates can function as a “single process”, which effectively introduces sources of size
specific mortality, such as predation (Cushing 1975). Finally, while the exact mechanisms
through which larval growth and mortality are influenced may not be evident, it has been shown
that small changes in these rates, especially growth rates, can strongly influence recruitment by
modifying survival to the end of the first year of life (Houde 1987; 1989). Processes reducing
survival during early life stages may be episodic or subtle in nature, and thus are typically
36 
 

difficult to detect, and the result of both biotic and abiotic factors affecting growth and mortality
rates through modification of larval dispersal, nutrition, and predation (Houde 2002).
Higher growth and survival rates during 2009 would suggest the formation of a stronger
year class relative to 2008, and this was substantiated through fishery independent mid-water
trawl surveys conducted by the USGS Great Lakes Science Center in the main basin of Lake
Huron. Hydroacoustic mid-water trawl surveys of Lake Huron’s main basin determined a sharp
increase in YOY rainbow smelt abundance during 2009, but during 2008 YOY rainbow smelt
abundance was low, reflecting poorer conditions for growth and survival during that year
(Schaeffer et al. 2009). Relative to 2008, this survey detected a 13-fold increase in YOY
rainbow smelt density during September 2009, with the highest densities of YOY rainbow smelt
during this survey occurring in US waters (Schaeffer et al. 2009). These high estimates of age-0
density were comparable in magnitude to densities observed during 1997, when rainbow smelt
populations were relatively abundant in Lake Huron (Warner et al. 2005).
Relating the 2009 rainbow smelt year class to predator response was accomplished
through a piscivore diet study that documented high predation on YOY rainbow smelt in the
main basin of Lake Huron. Species such as chinook salmon Oncorhynchus tshawytscha, coho
salmon Oncorhynchus kisutch, and lake trout contained high numbers of YOY rainbow smelt,
occurring in 25-35% of stomachs examined for these species (Edward Roseman, USGS Great
Lakes Science Center, personal communication).
This research has increased our understanding of growth and survival during rainbow
smelt early life history, and provides insight into factors that influence recruitment, including
variable survival of stream hatching cohorts and significant variation in growth rates during
May-July. Rainbow smelt year class formation in St. Martin Bay is in part controlled by survival

37 
 

of stream cohorts, which establishes the initial age-0 population during May and early June.
However, once these populations are established, high growth must be maintained by larvae
throughout the summer months for age-0 recruits to reach the end of their first year of life. Both
Brown (1994) and the present study have documented this important variation in survival and
growth during rainbow smelt early life history, and its relationship to year class strength. This
variation in growth and survival appears to be mediated by water temperatures which can drive
early emergence of lake spawning cohorts and sustain high growth rates throughout the larval
period. After the first year of life, subsequent recruitment of rainbow smelt to older age classes
is likely a function of predation, suggesting that recruitment of rainbow smelt in Lake Huron
may be controlled at the juvenile stage.

38 
 

CHAPTER TWO
EFFECTS OF STOCK SIZE, CLIMATE, AND PREDATION ON RAINBOW SMELT
RECRUITMENT IN WESTERN LAKE HURON, 1976-2009
Introduction
Recruitment and its relationship to adult spawning stock is central to the study and
management of fish populations (Myers 1996), but is generally the most challenging aspect in
the biological assessment of fish populations (Hilborn and Walters 1992). In studying fish
population dynamics the goal is often to determine if spawning stock, environmental factors,
biological processes, or some combination of these is responsible for observed patterns in the
relationship between the stock and its recruitment. Environmental factors such as temperature
and precipitation have been studied in relation to fishery recruitment for at least the last century
(Cushing 1982), but many apparent environment-recruitment correlations do not appear to persist
over long periods of time (Myers 1998). Furthermore, the number of apparent correlations
between environmental variables and recruitment can increase as the number of predictor
variables increases (Ricker 1975) and thus, the legitimacy of specific environmental predictors
can be difficult to ascertain (Francis 2006). Regardless, processes responsible for recruitment in
fish populations often relate to multiple biotic and abiotic factors working in concert to modify
growth and survival during their larval and juvenile stages.
Environmental variability during spring and summer has been linked to rainbow smelt
(Osmerus mordax) recruitment dynamics in the Great Lakes region. Sluka (1991) reported that
rainbow smelt abundance in Green Bay, Lake Michigan during 1966-89 was related to spawner
abundance, mean solar intensity in the previous spring, and air temperature in the previous spring
(Sluka 1991). Additionally he found that during the time period 1975-86, young-of-the-year
39 
 

(YOY) rainbow smelt abundance at stations off Alpena, Michigan (Lake Huron) was related to
spawner abundance and spring temperature, with the effect of cannibalism also presumed to be
an important source of YOY predation. Brown (1994), however, found that year class strength
and recruitment of these fish in St. Martin Bay, Lake Huron during 1991-1993 was influenced by
differences in growth rates mediated by water temperatures and zooplankton dynamics.
Furthermore, he reported that depressed stream water temperatures during spring spawning
resulted in low survival of eggs and larvae from tributaries (Brown 1991), further influencing
year class strength and recruitment to the population in Lake Huron in the fall.
Recruitment trends of rainbow smelt in Lake Huron have been monitored since the early
1970s through annual bottom trawl surveys conducted by the US Geological Survey Great Lakes
Science Center (GLSC). Between 1973 and 1991, yearling-and-older (YOA) rainbow smelt
were abundant (i.e., mean density > 700 fish/ha., Figure 2). However, from 1994 to the present,
YAO rainbow smelt abundance has been declining (i.e., mean density < 475 fish/ha., Figure 2)
and currently the population is approximately 40% reduced from the 1976-1991 time period
(Riley et al. 2008). Abundance of YOY rainbow smelt was typically low during 1976-1991 (i.e.,
mean density < 600 fish/ha., Figure 2), but during 1994-2009 YOY abundance was relatively
high (i.e., mean density > 1900 fish/ha., Figure 2), primarily as a result of several large year
classes that occurred during this period (e.g., 2005, YOY density > 5500 fish/ha.). However,
underscoring the complexity of the recruitment process, substantial recruitment to older age
groups did not occur following the formation of these large year classes in this lake.
Traditional stock-recruit methods typically involve analysis of empirical estimates of
spawning stock size and subsequent progeny produced in a given year (Hilborn and Walters
1992). One of the inherent difficulties in assessing the influence of stock size on subsequent

40 
 

production of recruits is that variation in annual production can also arise from changes in
environmental conditions, which often match or exceed the influence of changing stock size
(Ricker 1975). As such, environmental variation is often suspected to influence survival during
early life stages, and these specific variables can be integrated into traditional stock-recruit
models (Quinn and Deriso 1999). Additionally, species interactions such as predation or
competition can also be integrated into traditional stock-recruit models much in the same way
that environmental variation is incorporated (Hilborn and Walters 1992; Quinn and Deriso
1999).
The goal of this research was to examine stock-recruit relationships for Lake Huron
rainbow smelt in the context of potentially important environmental and biological variables.
Two factors that have been shown to significantly affect egg survival in streams and larval
growth rates during summer include spring precipitation (Hoff 2004) and water temperature
(Sluka 1991), and both were selected for integration into stock-recruit models. Additionally,
water levels have been linked to production and recruitment in fish populations (Jude and Pappas
1992; Rowe et al. 2002; Clark et al. 2008) and because rainbow smelt spawn and spend part of
their early life history in streams and nearshore areas, Lake Huron water levels were selected as a
third variable for stock recruit models. I hypothesized that higher water levels and warmer water
temperatures would influence recruitment positively, and that above average precipitation would
negatively influence recruitment. Finally, since lake trout predation is important in Great Lakes
rainbow smelt populations (Diana 1990; Madenjian et al. 1998; Madenjian et al. 2006; Gorman
2007), lake trout abundance (catch-per-unit-effort) was included as a species interaction in stockrecruit models that was hypothesized to negatively influence recruitment.

41 
 

This chapter has two primary objectives; 1) Develop stock-recruit models for rainbow
smelt populations in Lake Huron during 1976-2009 based on fishery independent bottom trawl
collections, and 2) Integrate water temperature, water level, precipitation, and lake trout
abundance time series data into stock-recruit models as explanatory variables. Specifically, I
wanted to test the hypothesis that the abundance of young-of-the-year rainbow smelt in Lake
Huron was influenced by water temperature, water levels, precipitation, rainbow smelt stock
size, and lake trout CPE, or some combination of the these variables.

Methods
Bottom trawl collections
Rainbow smelt populations were sampled in Lake Huron from 1976-2009 during bottom
trawl surveys conducted by the USGS GLSC. From 1976-1991, a 12 m headrope bottom trawl
with 4.76 mm square cod end mesh was used. From 1992-2009, a 21 m headrope bottom trawl
was used with 4.76 mm cod end mesh (Riley et al. 2009). Bottom trawls were towed on contour
at fixed transects for 10 minutes at approximately 2.2-2.4 m/h during daylight hours. These
surveys sample the same transects annually at depths from 9-110 m spaced at 9 m intervals
throughout the western main basin of Lake Huron (Figure 13). Surveys were typically carried
out between early October and mid November; however surveys during 1992-93 were conducted
during September and thus were excluded from analysis due to potential seasonal differences in
fish distributions (Riley et al. 2009). Additionally, no data were available from 2000 and 2008
as surveys were not completed during these years due to inclement weather. To account for a
change in trawling gear that occurred during the 1992 field season, fishing power corrections
42 
 

(FPC) were developed to standardize estimates of CPE among gears (Adams et al. 2009). FPC
were developed for yearling-and-older (YAO) rainbow smelt, but high variability in catches of
young-of-the-year (YOY) rainbow smelt resulted in no estimable FPC for this life stage (Adams
et al. 2009). To adequately evaluate rainbow smelt densities given the change in sampling gear,
stock recruit models were constructed for 1976-2009 with corrected YAO catch data (1976-1991
data corrected to reflect 21 m trawl), and for the 1976-1991 and 1994-2009 time periods
separately with no corrected data.

43 
 

Figure 14. Bottom trawl sampling locations in the main basin of Lake Huron, 1976-2009.

44 
 

Stock-recruit models
Analysis of stock-recruit models used length to partition bottom trawl catch data into
YOY (recruits) and YAO (stock) components for the time series. While age-length keys have
not been developed for Lake Huron rainbow smelt, previous work has indicated spawners are
typically 100 mm in length or greater (Brown 1994). This was further substantiated from a
survey of spawning rainbow smelt in St. Martin Bay, Lake Huron during 2008-09 (see Chapter
1) where length data indicated that spawners averaged between 135-140 mm, and no spawners
were less than 108 mm. For the 1976-2009 time series, catches of rainbow smelt were
partitioned into YOY and YAO components based on length frequency analysis, which was year
specific and resulted in YOY length cutoffs ranging from 75-105 mm (Riley and Adams 2010,
Table 2). Based on this information, the spawner component of stock-recruit models was
determined as numeric density (No./ha.) of rainbow smelt ≥ the estimated YOY length cutoff for
a given year, while recruits were considered to be the numerical density of fish ≤ year specific
cutoffs.
Stock-recruit models were initially constructed for all time periods (1976-2009, 19761991, and 1994-2009) using the models of Ricker (1975) and Beverton and Holt (1957). In the
general form the Ricker model is represented by the equation:

Where R = number of recruits, S = stock size, α is a model parameter describing productivity, β
is a parameter describing density dependence in the population, and

is a lognormal error. This

relationship assumes that increasing spawning stock will hinder recruitment at some stock size
through density dependant processes such as cannibalism (Ricker 1975; Quinn and Deriso 1999).
45 
 

This model was linearized and parameters were estimated through linear regression using the
equation:

log

log ε

log

The Beverton-Holt model in the general form is represented by the equation:

Where R = number of recruits, S = stock size, a is the productivity parameter, b is a parameter
controlling the level of density dependence, and e^ (σ2/2) is the bias adjustment factor. This
relationship increases with increasing spawner abundance and approaches an asymptote at some
level of stock size (Quinn and Deriso 1999). This model was transformed and parameters
estimated through non-linear regression using the form:

log

log ε

log

After fitting stock recruit models, second order Akaike’s Information Criterion (AICc, Burnham
and Anderson 2002) was used to compare models and determine which stock recruit relationship
was the most accurate reflection of rainbow smelt dynamics in Lake Huron. Results indicated
that both models were similar based on reduced SSE, however, the Ricker model was selected
because it accounts for compensation at high stock sizes from sources such as cannibalism
(Ricker 1954; 1975), and because initial stock recruit models appeared to show densitydependence at higher stock sizes; a phenomenon noted by earlier investigators for Lake Huron

46 
 

rainbow smelt (Sluka 1992). Additionally, this aspect of the model is particularly relevant to
rainbow smelt, which are known to exhibit cannibalism (Henderson and Nepszy 1989, Parker
Stetter et al. 2007). Final model fitting for basic Ricker stock-recruit functions was
accomplished through non-linear regression (PROC NLIN, SAS Institute), and model parameter
estimates were then corroborated with initial estimates determined through linear regression.
Measures of uncertainty for Ricker stock-recruit parameters were estimated through a
bootstrap procedure (Hillborn and Walters 1992; Efron 1993), which removes the assumption of
normality and allows the computation of parameter confidence intervals. For all three stockrecruit models (1976-2009, 1976-1991, and 1994-2009), linear regression residuals were resampled with replacement and added to predicted values to estimate new parameters. This
procedure was repeated 1000 times for each model and bootstrapped parameters were saved after
each step (Neter et al. 1996; Hilborn and Walters 1992). Saved parameter estimates from the
bootstrap procedures were then used to construct frequency distributions, estimate the variance,
and determine 95% confidence intervals of simulated parameters. All bootstrap procedures were
programmed using Visual Basic for Applications (VBA), implemented through Microsoft Excel.

Table 2. Year specific YOY length cutoffs (mm) for rainbow smelt in Lake Huron, 1976-2009.
Cutoffs were determined from length frequency analysis and applied to fishing power corrections
as determined by Adams et al. (2009).
YOY
length
cutoff
Years
applied

75

85

85, 89, 92, 81, 83, 8693
87, 90-91,
95-96, 99,
01-02, 04

90

95

100

105

05-06

73-76, 7880, 82, 84,
88, 94, 97,
03

07-09

77

47 
 

Integrated stock-recruit models
In order to evaluate predator impact on rainbow smelt population dynamics, lake trout
CPE (No./1000 ft of gill net) for the Michigan waters of Lake Huron’s main basin was
determined from spring gill net surveys conducted by the Michigan Dept. of Natural Resources
and Environment and the Chippewa Ottawa Resource Authority. Lake trout captured in this
survey were larger than 250 mm (Ji He, Michigan Department of Natural Resources and
Environment, personal communication), which is consistent with lake trout sizes known to prey
on rainbow smelt in the Great lakes (Madenjian et al. 1998). For inclusion in stock recruit
models, lake trout CPE was expressed as the deviation from the mean CPE for the whole time
period.
Lakes Huron-Michigan water levels (meters) were included as an additional variable in
stock recruit models that could potentially influence survival of rainbow smelt during early life
stages. Water level data (meters) for Lakes Michigan-Huron were obtained from the United
States Army Corps of Engineers, Detroit District (USACE). Water level parameters in stock
recruit models were calculated as the deviation of March-May mean values from annual mean
values, as suggested by Hilborn and Walters (1992).
Precipitation data included in stock-recruit models were obtained from NOAA’s Great
Lakes Environmental Research Laboratory’s hydrologic database (Croley and Hunter 1994) of
monthly records of overbasin precipitation in m^3/sec. (not including Georgian Bay) during
1976-2007. Precipitation data for 2009 were estimated by using the mean precipitation
(m^3/sec.) for the whole time period. Precipitation parameters in stock recruit models were
calculated as the deviation of March-May mean values from annual mean values (Hilborn and
Walters 1992).
48 
 

Water temperatures (degrees C, 0-20 meter average) were determined from ice
thermodynamic models developed by researchers at the Great Lakes Environmental Research
Laboratory (Croley and Assesl 1994). This model generates daily water temperature values
throughout the water column by incorporating air temperature, wind speed, cloud cover, and
humidity recorded at several locations across the lake (Croley and Assel 1994). Mean water
temperatures from the top 20 meters were used, as this corresponds to epilimnetic waters where
age-0 rainbow smelt are typically found (Parker Stetter et al. 2008). Similar to lake trout CPE,
water temperatures were expressed as the deviation from the grand mean for the time series.
Water temperatures, water levels, over basin precipitation, and lake trout CPE were first
incorporated into Ricker stock-recruit models using the form:

Where x1 to xn are n explanatory variables and c1 to cn are the estimated coefficients (Maceina
and Pereira 2007). For each time series, 16 model combinations were considered and parameters
were estimated through linear regression (Proc REG, SAS Institute) with the model form:

/

49 
 

Model selection
For each model considered, AICc was calculated (sum of squared error) using the
equation:

2 log

2

2

1
1

Where θ is the SSE objective function, K is the number of parameters, and n is the number of
observations in the data set. AICc differences were then calculated using the equation:

Where Δi is the AICc deviation of each model from the best model (Burhnam and Anderson
2002). The relative likelihood of each model (expressed as e^-1/2 Δi) was then used to compute
Akaike weights wi , which are expressed as:

exp
∑

exp

1
2

1
2

Where r is the number of candidate models (Burhnam and Anderson 2002). These normalized
relative likelihoods provide a weight of evidence for model plausibility and were used in
selecting the best model(s) for the 1976-2009, 1976-1991, and 1994-2009 time series. Evidence
ratios (w1/wj) were then computed as a further quantitative measure of distance between two
models (Burhnam and Anderson 2002).

50 
 

Results
Rainbow smelt recruitment in Lake Huron during 1976-2009 varied considerably and
was characterized by a period of high stock and low numbers of recruits (1976-1991) and a
period of low stock and higher production of age-0 fish (1994-2009). Mean density of YAO
rainbow smelt was approximately 1,975 fish/ha during 1976-1991, but decreased to 474 fish/ha.
during 1994-2009. The mean density of YOY rainbow smelt was highly variable between time
periods, ranging from 598 fish/ha. during 1976-1994 to over 1,950 fish/ha. during 1994-2009.
There tended to be distinct time periods (almost decadal) where successive years would cluster
along the stock-recruitment curve. For instance, most observations for 1980-89 occur over the
period of time where Lake Huron populations had the highest recorded stock sizes and during a
period of low recruitment (Figure 14). During 1999-2009, however, the highest indices of
recruitment were recorded, and observations occurred over the ascending portion of the Ricker
curve where stock size is lowest. The recruitment model for 1976-1991 is nearly asymptotic in
appearance with little variation in recruitment occurring over a broad range of stock sizes (Figure
14). Furthermore, during 1976-1991, the production of age-0 fish was limited despite relatively
abundant spawning stocks. This relatively flat stock-recruitment curve for 1976-1991 was
contrasted by a more classical Ricker curve for 1994-2009, when compensatory affects were
more evident at higher stock sizes (Figure 14). During 1994-2009, density of age-0 recruits in
the fall was high relative to 1976-1991 when abundances were lower (< 400 fish/ha.); however,
overcompensation due to cannibalism or predation (Ricker 1954) is evident at higher stock sizes
(> 500 fish/ha).

51 
 

0

2000

4000

6000

1976-2009

0

1000

2000

3000

4000

1500
500
0

Recruits

1976-1991

0

500

1000

1500

0

2000

4000

6000

1994-2009

0

200

400

600

800

1000

1200

1400

Stock Size

Figure 15. Stock-recruit relationships for Lake Huron rainbow smelt modeled by the Ricker
function for the years 1976-2009 (top panel), 1976-1991 (middle panel), and 1994-2009 (lower
panel).
52 
 

Stock-recruit models
Stock recruit models converged quickly with one or two iterations, and multiple starting
values produced repeatability in parameter estimates (Table 3). For the three time series
analyzed, stock-recruit relationships were modeled best by the following equations;

1976

2009

5.42

.

1976

1991

1.54

.

1994

2009

13.79

.

Table 3. Original parameter estimates, approximate standard error, and 95% confidence
intervals for stock-recruit models of rainbow smelt recruitment (1976-2009, 1976-1991, and
1994-2009).
Model

Parameters

Estimate

Std. Error

95% CI

1976-2009

Ln(α)

1.690

0.294

[1.086, 2.2938]

β

-0.0013

0.00018

[-0.0016, -0.0009]

Ln(α)

0.0431

0.326

[-0.269, 1.131]

β

-0.0009

0.00043

[-0.0018, -3.29E-6]

Ln(α)

2.624

0.3942

[1.756, 3.4912]

β

-0.0024

0.00069

[-0.0039, -0.0009]

1976-1991

1994-2009

Precision estimates of the α and β parameters were evaluated through a bootstrap procedure and
these results indicate that parameters were well estimated and approximated results attained
analytically. Frequency distributions for bootstrap parameter estimates were normally
distributed with means close to original estimates (Figures 16-18). Variance ranged from 0.076
53 
 

– 0.129 for the alpha parameter and 2.95E-08-3.86E-07 for beta, and the 1994-2009 time series
had the largest amount of variability around bootstrap parameter estimates (Table 4).
Confidence intervals around bootstrap parameter estimates were quite similar to those
determined through analytical procedures (Table 3, 4). Bootstrap re-sampling indicated
parameters were reasonably well estimated, with the 1994-2009 time series having the most
uncertainty regarding accuracy in parameter estimates.

Table 4. Mean, variance, and 95% confidence intervals of bootstrap parameter estimates for
stock-recruit models of rainbow smelt recruitment (1976-2009, 1976-1991, and 1994-2009).
Statistics are based on 1000 bootstrap samples.
Model

Parameters

Mean

Variance

95% CI

1976-2009

Ln(α)

1.681

0.076

[1.127, 2.220]

β

-0.0013

2.95E-08

[-0.0016, -0.0009]

Ln(α)

0.438

0.086

[-0.129, 1.014]

β

-0.0009

1.50E-07

[-0.0017, -0.0001]

Ln(α)

2.609

0.129

[1.880, 3.283]

β

-0.0024

3.86E-07

[-0.0036, -0.0012]

1976-1991

1994-2009

54 
 

140
120

ln α

100
80
60

Frequency

40
20
0
140

1.0

1.5

2.0

2.5

120

β

100
80
60
40
20
0

-0.0018 -0.0016 -0.0014 -0.0012 -0.0010 -0.0008 -0.0006

Parameter estimate
Figure 16. Frequency distributions of bootstrapped parameter estimates (log α, β) for the Ricker
stock-recruit model for the 1976-2009 time series. Approximate location of “true” parameter
estimates are indicated by arrows.
55 
 

160
140
120

ln α

100
80
60

Frequency

40
20
0
160

-0.5

0.0

0.5

1.0

1.5

140

β

120
100
80
60
40
20
0

-0.0020 -0.0015 -0.0010 -0.0005 0.0000

0.0005

Parameter estimate
Figure 17. Frequency distributions of bootstrapped parameter estimates (log α, β) for the Ricker
stock-recruit model for the 1976-1991 time series. Approximate location of “true” parameter
estimates are indicated by arrows.
56 
 

140
120

ln α

100
80
60

Frequency

40
20
0
140

1.0

1.5

2.0

2.5

120

β

100
80
60
40
20
0

-0.0018 -0.0016 -0.0014 -0.0012 -0.0010 -0.0008 -0.0006

Parameter estimate
Figure 18. Frequency distributions of bootstrapped parameter estimates (log α, β) for the Ricker
stock-recruit model for the 1994-2009 time series. Approximate location of “true” parameter
estimates are indicated by arrows.
57 
 

Effects of stock size, climate, and predation
Rainbow smelt recruitment in Lake Huron during 1976-2009 was influenced by stock
size, lake trout abundance, Lake Huron water levels, and over basin precipitation. Moreover,
based on Akaike weights (AIC weights) and interpretation of evidence ratios, recruitment models
considered most plausible for each time period had a strong predatory component as a unifying
factor. For the three time periods modeled, a total of 48 models were compared, 16 for each time
period. For the 1976-2009 stock-recruit analysis, the most plausible model (≈ 31% of the weight
of evidence) included rainbow smelt stock size, lake trout abundance, water levels, and over
basin precipitation as factors influencing recruitment (Table 5). AIC weight for the model
incorporating stock size, lake trout abundance, and precipitation accounted for 13% of the
evidence and was ranked as the next most plausible model (Table 5). All subsequent models
weighted between 3-10% included stock size, lake trout abundance, precipitation, and water
levels, with temperature not factoring in as an important environmental variable (Table 5).
For the time period from 1976-1991, the best model included only lake trout abundance,
and accounted for 24% of the weight of evidence among models (Table 6). The second best
model included only stock size and accounted for 20% of the evidence among models. The
model combining water levels and lake trout abundance was weighted ≈ 13% and AIC weights
were distributed among subsequent models containing lake trout abundance, precipitation, or
temperature (Table 6). The most important factors influencing recruitment during 1976-1991
were stock size, lake trout abundance, and water levels, with much of the evidence suggesting
that the combined influence of lake trout abundance and stock size have the strongest influence
on age-0 recruitment to the population.

58 
 

During 1994-2009, the model containing rainbow smelt stock size and lake trout
abundance accounted for ≈ 62% of the weight of evidence among models (Table 7). The next
most plausible model, accounting for only 12% of the weight included precipitation in addition
to the previous model variables. The evidence ratio between these two models indicates that the
best model is five times more likely and better supported by the data than the second best model.
Table 5. Model selection results for rainbow smelt recruitment in Lake Huron, 1976-2009, based
on Akaike weights wi and ranked top (best model) to bottom (least plausible model). Also
shown are sums of squared errorr, AICc, Δi (AICc differences), e^ (-1/2*Δi) (model likelihoods),
and evidence ratios. All model combinations are shown with Ricker parameters (α, β) and
explanatory variables; LT = lake trout CPE, P = overbasin precipitation, W = Lake HuronMichigan water levels, and T = mean lake temperature 0-20 meters.
Model

SSE

AICc

Δi

e(-1/2*Δi)
1.000

Akaike
weight
0.309

Evidence
ratio
1.00

α-βS-LT-W+P

18.65

-0.193

0.000

α-βS-LT+P

21.87

1.483

1.676

0.433

0.134

2.31

α-βS-LT

24.71

2.322

2.516

0.284

0.088

3.52

α-βS+P

24.74

2.356

2.549

0.280

0.086

3.58

α-βS-LT-W+PT
α-βS-W+P

18.37

2.576

2.769

0.250

0.077

3.99

22.98

2.924

3.117

0.210

0.065

4.75

α-βS

27.82

3.257

3.450

0.178

0.055

5.61

α-βS-LT-W

23.83

3.969

4.162

0.125

0.039

8.01

α-βS-LT+P-T

21.87

4.425

4.618

0.099

0.031

10.06

α-βS+P+T

24.42

4.684

4.877

0.087

0.027

11.46

α-βS-LT-T

24.71

5.028

5.221

0.073

0.023

13.61

α-βS+T

27.50

5.424

5.618

0.060

0.019

16.59

α-βS-W

27.58

5.499

5.692

0.058

0.018

17.22

α-βS -W+P+T

22.87

5.718

5.911

0.052

0.016

19.21

α-βS-LT-W-T

23.74

6.807

7.000

0.030

0.009

33.12

α-βS-W+T

27.34

7.953

8.146

0.017

0.005

58.74

 
59 
 

Table 6. Model selection results for rainbow smelt recruitment in Lake Huron, 1976-1991, based
on Akaike weights wi and ranked top (best model) to bottom (least plausible model). Also
shown are sum of squared errorr, AICc, Δi (AICc differences), e^ (-1/2*Δi) (model likelihoods),
and evidence ratios. All model combinations are shown with Ricker parameters (α, β) and
explanatory variables; LT = lake trout CPE, P = overbasin precipitation, W = Lake HuronMichigan water levels, and T = mean lake temperature 0-20 meters.
Model

SSE

AICc

Δi

e(-1/2*Δi)
1.000

Akaike
weight
0.242

Evidence
ratio
1.00

α-βS-LT

1.95

-25.689

0.000

α-βS

2.42

-25.316

0.374

0.830

0.200

1.21

α-βS-LT-W

1.67

-24.509

1.180

0.554

0.134

1.80

α-βS-W

2.19

-23.849

1.840

0.399

0.096

2.51

α-βS-P

2.21

-23.703

1.986

0.370

0.089

2.70

α-βS-LT-P

1.90

-22.489

3.200

0.202

0.049

4.95

α-βS-T

2.39

-22.393

3.297

0.192

0.046

5.20

α-βS-LT-T

1.93

-22.211

3.479

0.176

0.042

5.69

α-βS-W-P

2.07

-21.090

4.599

0.100

0.024

9.97

α-βS-LT-W-T

1.64

-20.425

5.265

0.072

0.017

13.91

α-βS-P+T

2.16

-20.382

5.308

0.070

0.017

14.21

α-βS-W+T

2.17

-20.354

5.335

0.069

0.017

14.40

α-βS-LT-W-P

1.67

-20.186

5.504

0.064

0.015

15.67

α-βS-LT-P-T

1.89

-18.168

7.521

0.023

0.006

42.97

α-βS-W-P+T

2.04

-16.992

8.697

0.013

0.003

77.38

α-βS-LT-W-P-T

1.64

-15.092

10.598

0.005

0.001

200.10

 
 

60 
 

Table 7. Model selection results for rainbow smelt recruitment in Lake Huron, 1994-2009, based
on Akaike weights wi and ranked top (best model) to bottom (least plausible model). Also
shown are sum of squared errorr, AICc, Δi (AICc differences), e^ (-1/2*Δi) (model likelihoods),
and evidence ratios. All model combinations are shown with Ricker parameters (α, β) and
explanatory variables; LT = lake trout CPE, P = overbasin precipitation, W = Lake HuronMichigan water levels, and T = mean lake temperature 0-20 meters.
Model

SSE

AICc

Δi

e(-1/2*Δi)
1.000

Akaike
weight
0.619

Evidence
ratio
1.0

α-βS-LT

5.18

-3.299

0.000

α-βS-LT-P

4.76

-0.071

3.228

0.199

0.123

5.0

α-βS-LT-W

4.94

0.414

3.713

0.156

0.097

6.4

α-βS-LT+T

5.17

1.025

4.324

0.115

0.071

8.7

α-βS

10.35

2.237

5.536

0.063

0.039

15.9

α-βS+T

9.79

4.979

8.277

0.016

0.010

62.7

α-βS-LT-W-P

4.72

5.403

8.702

0.013

0.008

77.5

α-βS-LT-P-T

4.75

5.488

8.787

0.012

0.008

80.9

α-βS+P

10.21

5.528

8.827

0.012

0.008

82.6

α-βS+W

10.23

5.549

8.848

0.012

0.007

83.4

α-βS-LT-W-T

4.91

5.927

9.225

0.010

0.006

100.8

α-βS+W+T

9.48

8.896

12.195

0.002

0.001

444.7

α-βS+P+T

9.63

9.093

12.392

0.002

0.001

490.8

α-βS+W+P

10.18

9.827

13.126

0.001

0.001

708.3

α-βS-LT-W-P-T

4.70

12.773

16.072

0.000

0.000

3089.6

α-βS+W+P+T

9.47

14.455

17.754

0.000

0.000

7163.8

61 
 

Discussion
Variation in density of age-0 rainbow smelt recruits in the fall over the entire time series
(1976-2009) was best described by a Ricker stock recruit model that incorporated values of the
stock abundances of rainbow smelt and lake trout CPE as biological variables, and water levels
and over basin precipitation as physical factors contributing to recruitment variability during the
first year of life. These results were taken to indicate precipitation had a strong positive
influence on stream conditions for egg and larval survival through modification of stream
environments, while water levels change had a negative effect on recruitment. Substantial
empirical support (i.e., AIC Δi < 2, Burnham and Anderson 2002) was also found for the Ricker
model with lake trout CPE and over basin precipitation for the entire time series, with Lake
Huron water levels not factoring in as an important influence on recruitment. Additionally,
water temperature was not an important environmental variable that influenced age-0 recruitment
to the population during these years. When this time series was split into two periods (19761991 and 1994-2009) to eliminate any possible sampling biases based on the change in trawling
gears that were used, similar results were obtained. During 1976-1991, the Ricker stock recruit
model including stock size and lake trout CPE best explained the variation in recruitment.
Substantial support (i.e., AIC Δi < 2) also existed for one model including lake trout CPE and
Lake Huron water levels, and for models including only stock size, Lake Huron water levels, or
precipitation alone. The Ricker model that best described variation in recruitment from 19942009 included only lake trout CPE as an explanatory variable. Evidence in support of this model
was confirmed by a relatively large AIC weight (AIC w = 0.62) and no other model with
substantial empirical support (i.e., no model with AIC Δi < 2). These results suggest that
rainbow smelt stock size and lake trout CPE are key factors regulating recruitment of rainbow
62 
 

smelt in Lake Huron, as these factors were common to the most plausible models identified for
the three time periods. These results were interpreted to mean lake trout predation had a strong
influence on the abundance of rainbow smelt spawning stock, thereby limiting reproductive
capacity of the population through removal of these spawning stocks. Lake trout predation was
also considered to be an important regulator of age-0 or juvenile abundance during this time
period. Rainbow smelt stock size was probably important for two reasons. First, increased
density of reproductive age fish increases the egg and larval densities produced by the population
(Ricker 1975). Secondly, at higher stock sizes, cannibalism became an important regulator of
age-0 density. This phenomenon can be seen for most observation during the 1980s in the 19762009 model.
No similar stock-recruit analysis has been published for Lake Huron rainbow smelt
populations, but previous studies focusing on rainbow smelt populations in Lakes Superior and
Erie have considered abiotic and biotic influences on its recruitment dynamics. For instance,
lake trout populations during 1974-2005 were closely associated with population dynamics of
rainbow smelt in Lake Superior (Gorman 2007). Successful plantings built up the lake trout
population during the 1970s and large predatory demand by these populations resulted in large
declines in rainbow smelt abundance during the early 1980s (Gorman 2007). Gorman (2007)
further suggested that cisco populations served as a predatory buffer between lake trout and
rainbow smelt, and that poor cisco production was linked with further declines in rainbow smelt.
Further support for the recruitment model developed for rainbow smelt in Lake Huron
comes from Hoff (2004) who found May rainfall, rainbow smelt stock size, and bloater
(Coregonus hoyi) biomass to be the most important factors influencing rainbow smelt
recruitment in Lake Superior. The interaction between bloater biomass and lake trout was found

63 
 

to positively influence rainbow smelt recruitment in Lake Superior, with bloater presumably
serving as a buffer to lake trout predation on rainbow smelt (Hoff 2004). While this relationship
to lake trout was indirect, this represents the important predatory influence lake trout may have
on rainbow smelt populations in other Great Lakes. In Lake Erie, recruitment of YOY rainbow
smelt was not related to water temperature or rainbow smelt stock size but appeared inversely
related to age-1+ rainbow smelt abundance, indicating cannibalism was a significant controlling
factor at times in these populations (Henderson and Nepszy 1989). While salmonid predation
was not found to be correlated with recruitment, the authors posit that rainbow smelt year classes
were largest in Lake Erie during years when lake trout and pacific salmon plantings were low
(Henderson and Nepszy 1989).
The approach of modeling rainbow smelt recruitment for three separate time blocks was
to account for a change in bottom trawling gear that occurred in the USGS GLSC bottom trawl
survey after 1991. Because fishing power corrections developed by Adams et al. (2009) could
not resolve differences in trawling gear catchability for age-0 rainbow smelt (i.e., no correction
factor to recast age-0 catches during 1976-1991), splitting the 1976-2009 time series into two
periods (1976-1991 and 1994-2009) allowed for the separate analysis of stock recruit
relationships for each trawl and served as an alternative approach that would acknowledge
potential differences in age-0 catchability between trawling gears. This approach resulted in the
model for 1976-1991 having a limited number of observations at low stock sizes, often where the
stock-recruit relationship is best defined (Hilborn and Walters 1992). In contrast, models for
1976-2009 and 1994-2009 have a broad range of stocks sizes which help define depensatory and
compensatory processes important to understanding stock-recruit relationships (Hilborn and
Walters 1992). Thus, the attempt to deal with variable catchability by treating these data as two

64 
 

time series yielded two models that described the stock-recruit relationship well and one model
that had low power to detect the stock recruit relationship at low stock sizes (i.e., 1976-1991).
These recruitment models are useful for understanding factors associated with rainbow
smelt recruitment in Lake Huron, the Great Lakes, and large systems where this species occurs.
Based on the history of lake trout stocking in Lake Huron, recent evidence of natural
reproduction of this species in Lake Huron (Riley et al. 2007), and models presented here, results
indicate that predation by lake trout was a strong regulator of recruitment during 1994-2009.
During 1994-2002, lake trout stocked in US waters of Lake Huron averaged 4 million fish, and
during subsequent years an average 1.5 million lake trout have been stocked annually
(FWS/GLFC). Evidence of natural reproduction by lake trout was first detected in 2004,
indicating that wild lake trout were in the population (Riley et al. 2007). Furthermore, lake trout
predation on rainbow smelt has been well documented in Lake Huron (Diana et al. 1991,
Madenjian et al. 2006) and Lake Michigan (Madenjian et al. 1998).
The best model selected for recruitment during 1994-2009, which included only lake
trout abundance, is also supported based on the recent changes in size structure of the rainbow
smelt population in Lake Huron. During 1997, hydroacoustic surveys of Lake Huron’s main
basin identified a population comprised of multiple age classes, with approximately 40% of the
population age-0 fish (Warner et al. 2004). However, in 2004 this survey found age-0 fish to
make up approximately 60% of the total rainbow smelt density, indicating substantial declines
had occurred in the abundance of adults. Bottom trawl surveys during 2004-2009 corroborated
this general trend of the rainbow smelt population dominated by small (<90mm) rainbow smelt
with greatly reduced numbers of older age classes (Roseman et al 2008; Riley et al. 2009). Low
stock sizes during 2005 produced the largest index of age-0 abundance during the entire 1976-

65 
 

2009 time period, but subsequent years failed to indicate substantial recruitment to older age
classes. In fact, the 1994-2009 time series had the five highest recruitment events for the entire
29 year period, all of which failed to result in increased abundance of adults in subsequent years.
The paucity of older age classes in recent years suggests that predation occurring on age-1+
rainbow smelt was substantial in Lake Huron, similar to predator induced changes that occurred
in Lake Superior rainbow smelt populations during 1974-2005 (Gorman 2007).
This research has identified a strong trophic link between lake trout and rainbow smelt
that has implications for future rehabilitation efforts of lake trout in Lake Huron and the Great
Lakes. First, rainbow smelt are the dominant prey species in Lake Huron following the collapse
of alewife (Alosa psuedoharengus) populations after 2003 (Riley et al. 2010). Secondly,
rainbow smelt are important in the diet of lake trout during young-of-the-year (Swedberg and
Peck 1984), juvenile (Eck and Wells 1983), and adult (Madenjian 2006) stages, thereby
indicating their importance to lake trout populations over a wide range of age classes. Finally,
lake trout will need to compete with stocked salmonines, percids and other predators for low
prey resources in Lake Huron. Managers should be concerned that consumption by predators in
Lake Huron could approach the level of production by prey species (Bence 2005), and result in
poor growth or condition of economically valuable sport fishes.
Because the prey fish community available to salmonine predators in Lake Huron is
currently dominated by one species, the rainbow smelt, and because Lake Huron is a
multijurisdictional resource with competing socioeconomic and ecological needs (i.e., sustenance
of hatchery dependant sport fisheries, lake trout restoration, and commercial fisheries), allocation
of this important prey resource among fishery components will remain a challenge for fishery
managers. As such, the primary tools for allocating prey resources are either to control the

66 
 

number of predators via stocking or limit the harvest of prey or predator species (Brown et al.
1999). Furthermore, while state or provincial boundaries exist for management of Lake Huron’s
fisheries, the fish communities, particularly the predator-prey assemblages, are part of larger
ecosystems, and thus must be managed through cooperative means across jurisdictions (Brown et
al. 2009). This research further supports the assertion that continued rehabilitation efforts of lake
trout and maintenance of current stocks of introduced salmonines in Lake Huron will be
dependent on the availability of rainbow smelt populations as a principal forage base, with other
species such as bloater (Coregonus hoyi) or cisco (Coregonus artedii) further diversifying the
prey assemblage.

67 
 

REFERENCES

68 
 

REFERENCES

Adams, J. V., S. C. Riley, and S. A. Adlerstein. 2009. Development of fishing power
corrections for 12-m Yankee and 21-m wing trawls used in the USGS Lake Huron fall bottom
trawl survey. Great Lakes Fishery Commission Technical Report 68.
Auer, N.A. (ed). 1982. Identification of larval fishes of the Greta Lakes basin, with emphasis on
the Lake Michigan drainage. Great Lakes Fishery Commission Special Publication 823:744p.
Ayer, M.H., C. Benton, W. King V, J. Kneebone, S. Elzey, M. Toran, K. Grange, and D.L.
Berlinskey. 2005. Development of practical culture methods for rainbow smelt larvae. North
American Journal of Aquaculture. 67:202-209.
Bailey, M.M. 1964. Age, growth, maturity, and sex condition of American smelt, Osmerus
mordax (Mitchill), of western Lake Superior. Transactions of the American Fisheries Society
9:382-395.
Baldwin, N. S., and R.W. Saalfeld. 1962. Commercial fish production in the Great Lakes, 18671960. Great Lakes Fishery Commision Technical Report No. 3: 166p.
Baldwin, N. A., R. W. Saalfeld, M. R. Dochoda, H. J. Buettner, and R.L. Eshenroder. 2009.
Commercial Fish Production in the Great Lakes 1867-2006. Great Lakes Fishery
Commission. Available: http://www.glfc.org/databases/commercial/commerc.php (Accessed
August 23, 2010)
Bence, J.R. 2005. A critique of fish-community objectives. Pages 119-123 in M.P. Ebener,
editor. The state of Lake Huron 1999. Great Lakes Fishery Commission Special Publication
05-02.
Beverton, R. J. H., and S. J. Holt. 1957. On the dynamics of exploited fish populations.
Fisheries Investigations Series II, XIX, London: HMSO.
Blaxter, J. H. S. 1969. Development: eggs and larvae. Pages 177-252 in W.S. Hoar and D.J.
Randall, editors. Fish Physiology Vol. 3. Academic Press, New York.
Brandt, S.B., J.J. Magnuson, and L.B. Crowder. 1980. Thermal habitat partitioning by fishes in
Lake Michigan. Canadian Journal of Fisheries and Aquatic Sciences 37:1557-1564.
Brandt, S. B., and S. P. Madon. 1986. Rainbow smelt (Osmerus mordax) predation on slimy
sculpin (Cottus cognatus) in Lake Ontario. Journal of Great Lakes Research 12(4):322-325.

69 
 

Braum, E. 1978. Ecological aspects of the survival of fish eggs, embryos, and larvae. Pages
102-131 in S.D. Gerking, editor. Ecology of freshwater fish production. Halstead Press, New
York.
Brett, J.R. 1979. Environmental factors and growth. Pages 599-667 in: W.S. Horar, D.J.
Randall, and J.R. Brett, editors. Fish Physiology, vol. 8. Academic Press, New York.
Bronte, C. R., M. P. Ebener, D. R. Schreiner, D. S. DeVault, M. M. Petzold, D. A. Jensen, C.
Richards, and S. J. Lozano. 2003. Fish community change in Lake Superior, 1970–2000.
Canadian Journal of Fisheries and Aquatic Sciences 60:1552–1574.
Brown, E.H., T.R. Busiahn, M.L. Jones, and R.L. Argyle. 1999. Allocating Great Lakes forage
bases in response to multiple demand. Pages 355-394 in W.W. Taylor and C.P. Ferreri,
editors. Great Lakes fisheries policy and management: A binational perspective. Michigan
State University Press, East Lansing.
Brown, M.L., and Guy, C.S. 2007. Science and statistics in fisheries research. Pages 1-29 in .S.
Guy and M.L. Brown, editors. Analysis and interpretation of freshwater fisheries data.
American Fisheries Society, Bethesda, Maryland.
Brown, R. W. 1994. Reproduction, early life history, and recruitment of rainbow smelt in St.
Martin Bay, Lake Huron. Ph.D. Dissertation. Michigan State University, East Lansing.
Brown, R.W., M.P Ebener, T.J. Sledge, and W.W. Taylor. 1995. Forage fish assemblage
structure in the littoral and nearshore area of St. Martin Bay, Lake Huron. Pages 207-222 in:
M. Munawar, T. Edsall, and J. Leach, editors. The Lake Huron ecosystem: ecology, fisheries,
and management. Academic Publishing, Amsterdam.
Brown, R.W., and W.W. Taylor. 1995. Effects of a recreational dip-net fishery on rainbow
smelt egg deposition. North American Journal of Fisheries Management 15:165-169.
Bunnell, B. B., C. P. Madenjian, and R. M. Claramunt. 2006. Long-term changes of the Lake
Michigan fish community following the reduction of exotic alewife (Alosa pseudoharengus).
Canadian Journal of Fisheries and Aquatic Sciences 63:2434-2446.
Burnham, K. P., and D. R. Anderson. 2002. Model Selection and Multimodel Inference: A
Practical Information-Theoretic Approach. Springer-Verlag, New York.
Cada, G. F., and G. L. Hergenrader. 1980. Natural mortality rates of freshwater drum larvae in
the Missouri River. Transactions of the American Fisheries Society 109:479-483.
Casselman, J.M. 2002. Effects of temperature, global extremes, and climate change on yearclass production of warmwater, coolwater, and coldwater fishes in the Great Lakes basin.
Pages 39-59 in N.A. McGinn, editor. Fisheries in a changing climate. American Fisheries
Society, Symposium 32, Bethesda, Maryland.
70 
 

Christie, W. J. 1974. Changes in the fish species composition of the Great Lakes. Journal of the
Fisheries Research Board of Canada. 31:827-854.
Clark, M.E., K. A. Rose, J.A. Chandler, T.J. Richter, D.J. Orth, and W. Van Winkle. 2008.
Water-level fluctuation effects on Centrarchid reproductive success in reservoirs: A modeling
analysis. North American Journal of Fisheries Management 28:1138-1156.
Conner, D. J., C. R. Bronte, J. H. Selgeby, and H. L. Collins. 1993. Food of salmonine
predators in Lake Superior, 1981-1987. Great Lakes Fishery Commission Technical Report
No. 59
Cooper, J.E. 1978. Identification of eggs, larvae, and juveniles of the rainbow smelt, Osmerus
mordax, with comparisons to larval alewife, Alosa psuedoharengus, and gizzard shad,
Dorosoma cepedianum. Transactions of the American Fisheries Society 107:56-62.
Cowan, J.H., and R.F. Shaw. 2002. Recruitment. Pages 88-111 in L.A. Fuiman and R.G.
Werner, editors. Fishery science: the unique contributions of early life stages. Blackwell
Publishing, Oxford.
Creaser, C. W. 1926. The establishment of the American smelt in the upper waters of the Great
Lakes. Papers of the Michigan Academy of Science, Arts and Letters 5:405–423.
Croley, T. E., II, and R. A. Assel. 1994. A one-dimensional ice thermodynamics
model for the Laurentian Great Lakes. Water Resources Research 30(3):625-639.
Croley, T. E., II; Hunter, T. S. 1994. Great Lakes Monthly Hydrologic Data; NOAA Technical
Memorandum ERL GLERL-83, Great Lakes Environmental Research Laboratory: AnnArbor,
Michigan.
Cushing, D. H. 1975. Marine ecology and fisheries. Cambridge University Press, Cambridge.
Cushing, D. H. 1982. Climate and fisheries. Academic Press, New York.
DeAngelis, D. L., P. A. Hackney, and J. C. Webb. 1980. A partial differential equation model
of changing sizes and numbers in a cohort of juvenile fish. Environmental Biology of Fishes
5(3)261-266.
Diana, J. S. 1990. Food habits of angler caught salmonines in western Lake Huron. Journal of
Great Lakes Research 16(2):271-278.
Dryer, W. R., L. F. Erkkila, and C. L. Tezloff. 1965. Food of the lake trout of Lake Superior.
Transactions of the American Fisheries Society 94:169-176.
Du, J. 2002. Combined algorithms for fitting finite mixture distributions. Master’s thesis,
McMaster University, Hamilton, Ontario.
71 
 

Dymond, J. R. 1944. Spread of the smelt (Osmerus mardax) in the Canadian waters of the Great
Lakes. Canadian Field Naturalist 58:12-14.
Eck, G. W., and L. Wells. 1983. Biology, population structure, and estimated forage
requirements of lake trout in Lake Michigan. U.S. Fish and Wildlife Service Technical
Paper 111.
Efron, B., and R.J. Tibshirani. 1993. An introduction to the bootstrap. Chapman and Hall, New
York.
Evans, D. O., and D. H. Loftus. 1987. Colonization of inland lakes in the Great Lakes region by
rainbow smelt, Osmerus mordax: their freshwater niche and effects on indigenous fishes.
Canadian Journal of Fisheries and Aquatic Sciences 44:249-266.
Everitt, B.S. and D.J. Hand. 1981. Finite mixture distributions. Chapman and Hall, London. 143
pp.
Francis, R. I. C. C. 2006. Measuring the strength of environment-recruitment relationships: the
importance of including predictor screening within cross-validations. ICES Journal of Marine
Science 63: 594-599
Franzin, W. G., B. A. Barton, and R. A. Remnant. 1994. Range extension, present and potential
distribution, and possible effects of rainbow smelt in Hudson Bay drainage waters of
northwestern Ontario, Manitoba, and Minnesota. North American Journal of Fisheries
Management 14:65-76.
Fry, F. E. J. 1971. The effect of environmental factors on the physiology of fish. Pages 1-89 in
W.S. Horar and D.J. Randall, editors. Fish Physiology, vol 6. Academic Press, New York.
FWS/GLFC. 2010. Great Lakes Fish Stocking database. U.S. Fish and Wildlife Service, Region
3 Fisheries Program, and Great Lakes Fishery Commission. Available:
www.glfc.org/fishstocking/rangesearch.htm (June 2010).
Gorman, O. T. 2007. Changes in a population of exotic rainbow smelt in Lake Superior: boom
to bust, 1974–2005. Journal of Great Lakes Research 33(1):75-90.
Goodyear, C. S., T. A. Edsall, D. M. Ormsby Dempsey, G. D. Moss, and P. E. Polanski. 1982.
Atlas of the spawning and nursery areas of Great Lakes fishes. Volumes I-VIII. U.S. Fish and
Wildlife Service, Washington, DC FWS/OBS-82/52.
Hackney, P. A., and J. C. Webb. 1978. A method for determining growth and mortality rates of
ichthyoplankton. In: Fourth National Workshop on Entrainment and Impingement (eds. L.D.
Jensen), pp. 115-124. Ecological Analysts Communications, Melville, New York.

72 
 

Hatch, J.T., and J.C. Underhill. 1988. Abundance, distribution, growth, and mortality of larval
lake herring in western Lake Superior. Pages 96-103 in R.D. Hoyt, editors. 11th annual larval
fish conference. American Fisheries Society, Symposium 5, Bethesda, Maryland.
He, X., and G. W. LaBar. 1994. Interactive effects of cannibalism, recruitment, and predation
on rainbow smelt in Lake Champlain: a modeling synthesis. Journal of Great Lakes Research
20(1):289-298.
Henderson, B. A., and S. J. Nepszy. 1989. Factors affecting recruitment and mortality rates of
rainbow smelt (Osmerus mordax) in Lake Erie, 1963-85. Journal of Great Lakes Research 15
(2):357-366.
Hillborn, R., and J.C. Walters. 1992. Quantitative fisheries stock assessment. Kluwer
Academic Publishers, Boston, MA.
Hoff, M. H. 2004. Biotic and abiotic factors related to rainbow smelt recruitment in the
Wisconsin waters of Lake Superior, 1978-1997. Journal of Great Lakes Research 30(1):414422.
Houde, E. D. 1987. Fish early life dynamics and recruitment variability. Pages 17-29 in R.D.
Hoyt, editors. 10th annual larval fish conference. American Fisheries Society, Symposium 2,
Bethesda, Maryland.
Houde, E.D. 1989. Subtleties and episodes in the early life history of fishes. Journal of Fish
Biology 35(Supp A):29-38.
Houde, E. D. 1997. Patterns and trends in larval-stage growth and mortality of teleost fish.
Journal of Fish Biology 51:52-83.
Houde, E.D. 2002. Mortality. Pages 64-87 in: L.A. Fuiman and R.G. Werner, editors. Fishery
science: the unique contributions of early life stages. Blackwell Publishing, Oxford.
Houde, E. D. 2008. Emerging from Hjort’s shadow. Journal of Northwest Atlantic Fishery
Science 41:53-70.
Hrabik, T. R., J. J. Magnuson, and A. S. McLain. 1998. Predicting the effects of rainbow smelt
on native fishes in small lakes: evidence from long-term research on two lakes. Canadian
Journal of Fisheries and Aquatic Sciences 55:1364-1371.
Hulbert, P. J. 1974. Factors affecting spawning site selection and hatching success in
anadromous rainbow smelt (Osmerus mordax, Mitchill). Master’s thesis. University of
Maine, Orono, Maine.
Jones, C.M. 2002. Age and growth. Pages 33-63 in: L.A. Fuiman and R.G. Werner, editors.
Fishery science: the unique contributions of early life stages. Blackwell Publishing, Oxford.
73 
 

Jude, D.J., and J. Pappas. 1992. Fish utilization of Great Lakes Coastal wetlands. Journal of
Great Lakes Research 18(4):651-672.
Kocik, J. F., and W. W. Taylor. 1987. Diet and movements of age-1+ pink salmon in western
Lake Huron. Transactions of the American Fisheries Society 116: 628-633.
Lantry, B. F., and D. J. Stewart. 2000. Population dynamics of rainbow smelt (Osmerus
mordax) in Lakes Ontario and Erie: a modeling analysis of cannibalism effects. Canadian
Journal of Fisheries and Aquatic Sciences 57:1594-1606.
Loftgren, B.M. 2002. Global warming influences on water levels, ice, and chemical and
biological cycles in lakes: some examples. Pages 15-21 in N.A. McGinn, editor. Fisheries in
a changing climate. American Fisheries Society, Symposium 32, Bethesda, Maryland.
Loftus, D. H. and P. F. Hulsman. 1986. Predation on larval lake whitefish (Coregonus
clupeaformis) and lake herring (C. artedi) by adult rainbow smelt (Osmerus mordax).
Canadian Journal of Fisheries and Aquatic Sciences 43:812-818.
Luecke, C., J.A. Rice, L.B. Crowder, S.E. Yeo, and F.P. Binkowski. 1990. Recruitment
mechanisms of bloater in Lake Michigan: an analysis of the predatory gauntlet. Canadian.
Journal of Fisheries and Aquatic Sciences 47:524-532.
MacCallum, W. R., and H. A. Regier. 1970. Distribution of smelt, Osmerus mordax, and the
smelt fishery in Lake Erie in the 1960s. Journal of the Fisheries Research Board of Canada
27:1823-1846.
Macdonald, P. D. M. 1987. Analysis of length-frequency distributions. Pages 371-384 in R.C.
Summerfelt and G.E. Hall, editors. Age and Growth of Fish. Iowa State University Press,
Ames, Iowa.
Macdonald, P. with contributions from J. Du. 2008. Mixdist: Finite Mixture Distribution
Models. R package version 0.5-2. PUBLISHER. http://www.r-project.org,
http://www.math.mcmaster.ca/peter/mix/mix.html.
Maceina, M. J., and D. L. Pereira. 2007. Recruitment. Pages 121-185 in C.S. Guy and M.L.
Brown, editors. Analysis and interpretation of freshwater fisheries data. American Fisheries
Society, Bethesda, Maryland.
Madenjian, C. P., T .J. Desorcie, and R. M. Stedman. 1998. Ontogenetic and spatial patterns in
diet and growth of lake trout in Lake Michigan. Transactions of the American Fisheries
Society 127:236-252.
Madenjian, C. P., J. D. Holuszko, and T. J. Desorcie. 2006. Spring-Summer diet of lake trout on
Six Fathom Bank and Yankee Reef in Lake Huron. Journal of Great Lakes Research 32:200208.
74 
 

McKenzie, R.A. 1964. Smelt life history and fishery of the Miramichi River, New Brunswick.
Fisheries Research Board of Canada 144:1-76.
Mercado-Silva, N., G. G. Sass, B. M. Roth, S. Gilbert, and M. J. Vander Zanden. 2007. Impact
of rainbow smelt (Osmerus mordax) invasion on walleye (Sader vitreus) recruitment in
Wisconsin lakes. Canadian Journal of Fisheries and Aquatic Sciences 64:1543-1550.
Miller, T. J., L. B. Crowder, J. A. Rice, and E. A. Marschall. 1988. Larval size and recruitment
mechanisms in fishes: Toward a conceptual framework. Canadian Journal of Fisheries and
Aquatic Sciences 45:1657-1668.
Mills, E.L., Casselman, J.M., Dermott, R., Fitzsimons, J.D., Gal, G.,Holeck, K.T., Hoyle, J.A.,
Johannsson, O.E., Lantry, B.F., Makarewicz, J.C., Millard, E.S., Munawar, I.F., Munawar,
M., O’Gorman, R., Owens, R.W., Rudstam, L.G., Schaner, T., and Stewart, T.J. 2005. A
synthesis of ecological and fish community changes in Lake Ontario, 1970-2000. Great
Lakes Fishery Commission Technical Report 67.
Montgomery, D.C., E. A. Peck, and G.G. Vining. 2001. Introduction to linear regression
analysis. John Wiley and Sons, New York.
Myers, J., M. L. Jones, J. D. Stockwell, and D. L. Yule. 2009. Reassessment of the predatory
effects of rainbow smelt on Ciscoes in Lake Superior. Transactions of the American Fisheries
Society 138:1352-1368.
Myers, R.A. 2001. Recruitment: understanding density dependence in fish populations. Pages
123-148 in P. Hart, and J. D. Reynolds, editors. Fish and Fisheries Handbook. Blackwell
Scientific, London.
Myers, R.A., and N. J. Barrowman. 1996. Is fish recruitment related to spawner abundance?
Fishery Bulletin 94(4):707-724.
Myers, R. A. 1998. When do environment-recruit correlations work? Reviews in Fish Biology
and Fisheries 8: 285–305.
Nellbring, S. 1989. The ecology of smelts (genus Osmerus): a literature review.
Nordic Journal of Freshwater Research 65:116-145.
Nepszy, S. J., J. Budd, and A. O. Dechtiar. 1978. Mortality of young-of-the-year rainbow smelt
(Osmerus mordax) in Lake Erie associated with the occurrence
of Glugea hertwigi. Journal of Wildlife Disease 14:233-239.
Nester, R.T. 1987. Horizontal ichthyoplankton tow-net system with unobstructed net opening.
North American Journal of Fisheries Management 7:148-150.

75 
 

Neter, J., M.H. Kutner, C.J. Nachtsheim, and W. Wasserman. 1996. Applied linear statistical
models. Irwin Publishers, Chicago.
Nsembukya-Katuramu, S., E. K. Balon, and R. Mahon. 1981. A comparison of spawning,
harvested, and die-off rainbow smelt, Osmerus mordax, in Eastern Lake Erie. Journal of
Great Lakes Research 7(2):144-154.
O’Gorman, R. 1974. Predation by rainbow smelt (Osmerus mordax) on young-of-the-year
alewives (Alosa pseudoharengus) in the Great Lakes. The Progressive Fish-Culturist 36:223224.
O’Gorman, R. 1983. Distribution and abundance of larval fish in the nearshore waters of
Western Lake Huron. Journal of Great Lakes Research 9(1):14-22.
O’Gorman, R. 1984. Catches of larval rainbow smelt (Osmerus mordax) and alewife (Alosa
pseudoharengus) in plankton nets of different mesh sizes. Journal of Great Lakes Research
10(1):73-77.
O’Gorman, R., B. F. Lantry, and C. P. Schneider. 2004. Effect of stock size, climate, predation,
and trophic status on recruitment of alewives in Lake Ontario, 1978-2000. Transactions of
the American Fisheries Society 133:855-867.
Owens, R.W. 1982. The ecology of spawning smelt in southeastern Lake Michigan. Master’s
thesis. Eastern Michigan University, Ypsilanti, Michigan
Oyadomari, J. K., and N. A. Auer. 2004. Inshore-offshore distribution of larval fishes in Lake
Superior off the western coast of the Keweenaw Peninsula, Michigan. Journal of Great
Lakes Research 30(1):369-384.
Paling, J.E. 1971. Causes of mortality. Pages 249-258 in W. E. Ricker, editor. Methods for
assessment of fish production in fresh waters. Blackwell Scientific, Oxford.
Parker Stetter, S. J., J. L. Stritzel-Thomson, L. G. Rudstam, D. L. Parish, and P. J. Sullivan.
2007. Importance and predictability of cannibalism in rainbow smelt. Transactions of the
American Fisheries Society 136:227-237.
Pepin, P. 1991. Effect of temperature and size on development, mortality, and survival rates of
the pelagic early life history stages of marine fish. Canadian Journal of Fisheries and Aquatic
Sciences 48:503-518.
Quinn, T.J., and R.B. Deriso. 1999. Quantitative fish dynamics. Oxford University Press, New
York.

76 
 

R Development Core Team. 2009. R: A language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL
http://www.R-project.org.
Rao, C.R. 1965. Linear statistical inference and its applications. Wiley, New York.
Ray, B. A., T. R. Hrabik, M. P. Ebener, O. T. Gorman, D. R. Schreiner, S. T. Schram, S. T.
Sitar, W. P. Mattes, and C. R. Bronte. 2007. Diet and prey selection by Lake Superior lake
trout during spring, 1986–2001. Journal of Great Lakes Research 33:104–113.
Rowe, D.K., U. Shankar, M. James, and B. Waugh. 2002. Use of GIS to predict effects of water
level on the spawning area for smelt, Retropinna retropinna, in Lake Taupo, New Zealand.
Fisheries Management and Ecology 9:205-216.
Rice, J. A., T.J. Miller, K.A. Rose, L.B. Crowder, E.A. Marschall, A.S. Trebitz, D.L. DeAngelis.
1993. Growth rate variation and larval survival: inferences from an individual-based sizedependent predation model. Canadian Journal of Fisheries and Aquatic Sciences 50:133-142.
Ricker, W.E. 1954. Stock and recruitment. Journal of the Fisheries Research Board of Canada
11:559-623.
Ricker, W.E. 1975. Computation and interpretation of biological statistics of fish populations.
Fisheries Research Board of Canada. Bulletin 191.
Riley, S. C., J. X. He, J. E. Johnson, T. P. O’Brien, and J. S. Schaeffer. 2007. Evidence of
widespread natural reproduction by lake trout in the Michigan waters of Lake Huron.
Journal of Great Lakes Research 33: 917-921.
Riley, S.C., E. F. Roseman, S. J. Nichols, T. P. O’Brien, C. S. Kiley, and J. S. Schaeffer. 2008.
Deepwater demersal fish community collapse in Lake Huron. Transactions of the American
Fisheries Society 137: 1879-1890.
Riley, S. C., E. F. Roseman, J. V. Adams, T. P. O’Brien, and S. Farha. 2010. Status and trends
of the Lake Huron offshore demersal fish community, 1976-2009. U.S.G.S Annual report to
the Great Lakes Fishery Commission. U.S.G.S. Great Lakes Science Center, Ann Arbor, MI.
Riley, S. C., and J. V. Adams. 2010. Long-term trends in habitat use of offshore demersal fishes
in Western Lake Huron suggest large-scale ecosystem change. Transactions of the American
Fisheries Society 139:1322-1334.
Roseman, E. F., T. P. O’Brien, S. C. Riley, S. Farha, and French, J. R. P. III. 2009. Status and
trends of the Lake Huron deepwater demersal fish community, 2008. U.S.G.S. Annual report
to the Great Lakes Fishery Commission. U.S.G.S. Great Lakes Science Center, Ann Arbor,
MI.
77 
 

Roseman, E. F., J. S. Schaeffer, T. P. O’Brien, J. R. P. French III, C. S. Kiley, and S. C. Riley.
2007. Status and trends of the Lake Huron deepwater demersal fish community, 2006. Great
Lakes Science Center Annual Lake Committee Report. Ann Arbor, MI: United States
Geological Survey.
Rupp, R. S. 1959. Variation in the life history of the American smelt in inland waters of Maine.
Transactions of the American Fisheries Society 88(4):241–252.
Rupp, R. S. 1965. Shore-spawning and survival of eggs of the American smelt. Transactions of
the American Fisheries Society 94(2):160–168.
SAS Institute. 1999. SAS/STAT user’s guide, version 8. SAS Institute, Inc., Cary, North
Carolina.
Schaeffer, J.S., D.M Warner, T.P. O’Brien, and S. Farha. 2010. Status and Trends of Pelagic
Prey Fishes in Lake Huron, 2009. U.S.G.S Annual report to the Great Lakes Fishery
Commission. U.S.G.S. Great Lakes Science Center, Ann Arbor, MI.
Scott, W. B., and E. J. Crossman. 1973. Freshwater Fishes of Canada. Fisheries Research
Board of Canada. Bulletin 184.
Selgeby, J. H., W. R. MacCallum, and D. V. Swedberg. 1978. Predation by rainbow smelt
(Osmerus mordax) on lake herring (Coregonus artedii) in western Lake Superior. Journal of
the Fisheries Research Board of Canada 35:1457-1463.
Sluka, R.D. 1991. The effects of spring climate, spawner abundance, and cannibalism on the
abundance of rainbow smelt (Osmerus mordax) at two sites in the upper Great Lakes.
Master’s Thesis. Michigan State University, East Lansing.
Schneberger, E. 1937. The biological and economic importance of the smelt in Green Bay.
Transactions of the American Fisheries Society 66:139-142.
Stewart, D. J., J. F. Kitchell, and L. B. Crowder. 1981. Forage fishes and their salmonid
predators in Lake Michigan. Transactions of the American Fisheries Society 110:751-763.
Stedman, R. M. and R. L. Argyle. 1985. Rainbow smelt (Osmerus mordax) as predators on
young bloaters (Coregonus hoyi) in Lake Michigan. Journal of Great Lakes Research
11(1):40-42.
Swedberg, D.V., and J.W. Peck. 1984. Food of young-of-the-year lake trout (Salvelinus
namaycush) in Presque Isle Harbor, Lake Superior. Journal of Great Lakes Research 10:280285.
Tin, H. T., and D. J. Jude. 1983. Distribution and growth of larval rainbow smelt in Eastern
Lake Michigan, 1978-1981. Transactions of the American Fisheries Society 112:517-524.
78 
 

U.S. Army Corps of Engineers, Detroit District (USACE). 2009. Historic Great Lakes water
level data 1918-2009. Available:
www.lre.usace.army.mil/greatlakes/hh/greatlakeswaterlevels/historicdata/ (December 2009)
Van Oosten, J. 1937. The dispersal of smelt, Osmerus mordax (Mitchill), in the Great Lakes
region. Transactions of the American Fisheries Society 66: 160–161.
Warner, D.M, T.P. O’Brien, C.S. Faul, and R.G. Stickel. 2005. Status of Pelagic Prey fish in
Lake Huron in 1997 and 2004. U.S.G.S. Annual report to the Great Lakes Fishery
Commission. U.S.G.S. Great Lakes Science Center, Ann Arbor, MI.
Warner, D. M., J. S. Schaeffer, and T. P. O’Brien. 2009. The Lake Huron pelagic fish
community: persistent spatial pattern along biomass and species composition gradients.
Canadian Journal of Fisheries and Aquatic Sciences 66:1199-1215.
Webb, P.W. 2008. The impact of changes in water level and human development on forage fish
assemblages in great lakes coastal marshes. Journal of Great Lakes Research 34(4): 615-630.
Wells, L. 1973. Distribution of fish fry in nearshore waters of southeastern Lake Michigan and
east central Lake Michigan, May-August 1972. Administrative Report Great Lakes Fisheries
Laboratory, US Fish and Wildlife Service, Ann Arbor, Michigan.
Zigler, S. J., and C. A. Jennings. 1993. Growth and mortality of larval sunfish in backwaters of
the upper Mississippi River. Transactions of the American Fisheries Society 122:1080-1087.

79