“[5513

   

  

.UBRARY

te

rsity

G
u

Univ

bun;-

-wwtl‘ . .3

This is to certify that the
thesis entitled

A RE-ASSESSMENT OF RAINBOW SMELT (OSMERUS
MORDAX) PREDATION ON CISCO (COREGONUS ARTEDI)
IN LAKE SUPERIOR

presented by

JARED T. MYERS

has been accepted towards fulfillment
of the requirements for the

MS. degree in Fisheries & Wildlife

 

m lice

Major Professér’s Signature
4735/05

Date

MSU is an affinnative-action, equal-opportunity employer

 

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

DATE DUE DATE DUE DATE DUE

 

 

A RE-ASSESSMENT OF RAINBOW SMELT (OSMERUS MORDAX)
PREDATION ON CISCO (COREGONUS ARTEDI) IN LAKE SUPERIOR

By

Jared T. Myers

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF SCIENCE

Department of Fisheries and Wildlife

2008

 

ABSTRACT

A RE-ASSESSMENT OF RAINBOW SMELT (OSMERUS MORDAX)
PREDATION ON CISCO (COREGONUS ARTEDI) IN LAKE SUPERIOR

By

Jared T. Myers

The work reported in my thesis addresses two fundamental issues for
understanding larval cisco dynamics. In the first chapter I report on our study
comparing several sampling strategies for estimating the abundance of larval
cisco. Development of a sampling program that effectively captures the larval
stage is necessary to allow estimation of important population metrics and for
investigating hypotheses concerning recruitment. Our work improved our ability
to assess larval cisco populations, which will facilitate interagency coordination
and large-scale research initiatives. In the second chapter we examine the
influence of rainbow smelt predation on the survival of larval cisco in two large
Lake Superior embayments. Past research attempting to understand the role of
rainbow smelt in determining cisco recruitment has led to varying results. We
show that incidental predation by rainbow smelt can be a substantial source of
mortality for larval cisco. Our evidence also suggests that rainbow smelt may
have played an important role in the historic collapse of cisco. We recognize that
rainbow smelt predation may not be the only mechanism inhibiting the
recruitment of cisco but believe that this added mortality could be detrimental to

the occasional successful cisco year class.

ACKOWLEDGEMENTS

I am indebted to my mentors, Jason Stockwell, Mike Jones, and Dan Yule
for their guidance and support. I am also thankful for the efforts of Ken Cullis,
Casey Frantz, Allison Gamble, Jeff Black, and Jon Chicone. As always, I am
grateful to my friends and colleagues at the MSU—Quantitative Fisheries Center,
USGS-GLSC—Lake Superior Biological Station, and OMNR—Upper Great Lakes
Management Unit. Jason Stockwell, Dan Yule, Mike Jones, Jean Adams, David
"Bo" Bunnell, Tom Hrabik, Sture Hansson, and Jon Deroba have provided

constructive comments throughout the writing process.

TABLE OF CONTENTS

LIST OF TABLES

 

LIST OF FIGURES

vi

 

INTRODUCTION

 

CHAPTER I
EVALUATING SAMPLING STRATEGIES FOR LARVAL CISCO
(COREGONUS ARTEDI)

 

Introduction

 

Methods

 

Results

 

Discussion

 

CHAPTER II
AN EMPIRICAL RE-ASSESSMENT OF THE PREDATORY EFFECTS OF
RAINBOW SMELT (OSMERUS MORDAX) ON CISCO (COREGONUS

ARTEDI) IN LAKE SUPERIOR

 

Introduction

 

Methods

 

Results

 

Discussion

 

REFERENCES

 

_L_;
NOOQODU'I

21
25
41
49

54

 

LIST OF TABLES

Table 1. The number of samples collected, number of coregonid larvae

captured, and the mean density (#/1,000 m )of cisco larvae
sampled for the evaluation of sampling strategies in Thunder Bay

and Black Bay, Ontario during 2-25 May 2006. .................................

Table 2. Relative contribution of each prey item to the diet of rainbow
smelt. Percentages are based on the dry weight of stomach
contents.

 

Table 3. Model inputs for rainbow smelt energy density (J/g wet weight)
in Thunder Bay and Black Bay. Simulation day 1 is May 1.
Average length within each size class was determined using the

 

information provided in Figure 6.

Table 4. Energy density (J/g wet weight) of prey items found in the
stomachs of rainbow smelt from Thunder and Black Bays in Lake

 

Superior.

Table 5. Rainbow smelt density estimates (rainbow smelt/ha [:tSE]) using
night acoustics and midwater trawling (AC-MT) and day bottom

 

trawls (BTR) in Thunder and Black Bays.

...13

36

37

37

.45

 

LIST OF FIGURES

Figure 1. Stations used to evaluate sampling strategies for larval cisco. ......

Figure 2. (A) Log1o-transformed larval cisco density (#/1,000 m3)
estimates and (B) cumulative length frequency distributions of
larval cisco collected with side-by—side tows using a 2 x 1-m
paired neuston net and a 0.5-m (diameter) conical net. (C) Log1o-

transformed larval cisco density (#/1,000 m3) and (D) the
cumulative length frequency distributions of larval cisco collected
using sinusoidal and straight tow patterns with either a 2 x 1-m
neuston net (9) or a 0.5-m (diameter) conical net (CI). The dashed
lines (A and C) are 1:1 lines.

 

Figure 3. Average larval cisco density (#/1,000 m3) as a function of time
of day. Sampling events occurred in the early morning (EM), late
morning (LM), early afternoon (EA), and late afternoon (LA) for six
individual days at different sites. Error bars represent the range of
estimated densities for each sampling event.

 

FigUre 4. Map of larval cisco sampling locations and density estimates in
Thunder and Black Bays, Lake Superior. Two synoptic surveys of
each bay were completed, 2-25 May 2006. Interpolated surfaces

..... 9

16

were created using inverse distance weighting. ................................... 26

Figure 5. Map of acoustically-derived rainbow smelt density estimates
collected in Thunder and Black Bays (Lake Superior), 17-22 May
2006. Data from acoustic transects were binned at 1km intervals.
Interpolated surfaces were created using inverse distance
weighting.

 

Figure 6. Length frequency distributions for Thunder Bay (363
measurements) and Black Bay (400 measurements) rainbow smelt
and the four size classes (5 60 mm, 61-90 mm, 91-120 mm, >

29

120mm) used to structure the bioenergetic analyses. ............................. 35

Figure 7. Observed proportions of rainbow smelt (A) and coregonids (B)
captured in midwater trawls versus the predicted proportions from
the TS distributions developed using the cutting-edge approach to
apportion targets to species. The dashed line is the 1:1 line. A
frequency distribution of the differences is located in the lower right
corner.

 

vi

44

 

K

Figure 8. The estimated percentage of the larval cisco population
consumed by rainbow smelt in both Thunder Bay and Black Bay.
Two rates of feeding (maintenance growth: P-value z 0.2; positive
growth P-value = 0.5) were used to encompass the probable
growth of rainbow smelt through the 30-day simulations. ....................... 47

Figure 9. Approach of Selgeby et al. (1978) compared to our

bioenergetics analysis and the sensitivity of results to assumptions
of temporal overlap and bias due to sampling gear. ................................ 49

vii

 

INTRODUCTION

Historically, cisco Coregonus artedi were found throughout the Great
Lakes (Scott and Crossman 1998) and played a key role in the food web by

serving as a trophic linkage between zooplankton resources and native
piscivores (Dryer and Beil 1964, Brown et al. 1999). At the turn of the 20th

century, cisco supported major commercial fisheries throughout the Great Lakes.
However, cisco stocks collapsed by 1928 in Lake Erie (Hartman 1973), 1955 in
lakes Ontario and Huron (Christie 1973, Berst and Spangler 1973), and 1960 in
Lake Michigan (Wells and McClain 1973). Populations of cisco reached a low in
Lake Superior by 1976, but these stocks were not decimated to the same degree
as stocks in the lower lakes (Jason Stockwell, United States Geological Survey,
personal communication). Lake Superior cisco populations have recovered to
support commercial fisheries, yet highly variable recruitment (Bronte et al. 2003)
has led to the perception that the stocks are unstable.

Restoration of native coregonids in the lower lakes has recently received
increased attention (Fitzsimons and O’Gorman 2006). The Great Lakes Fishery
Commission (GLFC) recognizes the economic value and desirability of striving
towards stable, native fish communities in all the Great Lakes, as evidenced by

the GLFC Strategic Vision for Healthy Great Lakes Ecosystems:

“The commission shall encourage the rehabilitation and conservation of healthy

aquatic ecosystems in the Great Lakes that provide sustainable benefits to

 

society, contain predominantly self regulating fish communities, and support
fisheries with increasing contributions of naturally reproducing fish. Conserving
biological diversity through rehabilitation of native fish populations, species,

communities, and their habitat has a high priority. ”

It is largely understood that a forage base composed of predominately
native species is desirable in each of the Great Lakes. Fisheries managers
recognize cisco as a central species in the native fish communities of the Great
Lakes coldwater systems (DesJardine et al. 1995, Eshenroderet al. 1995,
Stewart et al. 1999, Ryan et al. 2003, Horns et al. 2003). Statements made by
each of the respective lake committees call for the re-establishment of self-
sustaining cisco populations in each of the lower lakes.

There are several reasons why the functioning of the lower lakes'
ecosystems would benefit from the recovery of cisco. Cisco are not subject to the
large-scale die-offs characteristic of alewife Alosa pseudoharengus (O’Gorman
and Schneider 1986), which will help stabilize prey fish abundances in the lower
lakes. Cisco may facilitate the restoration of native piscivores by reducing the
occurrence of Early Mortality Syndrome (EMS), a condition linked to thiamine
deficiency and believed to be impeding recovery of native piscivores (Fitzsimons
et al. 1999). Rainbow smelt Osmerus mordax and alewife are high in thiaminase
and lake trout that rely on these non-native prey produce thiamine-deficient eggs,
preventing successful reproduction in the lower lakes. Researchers have shown

that size of prey relative to predator size may be an important factor influencing

 

"

predator growth rates (Werner 1974, Mittlebach 1981 ). Restoring a larger bodied
prey fish, such as cisco, could improve both growth and fecundity of native lake
trout populations. Self-sustaining populations of cisco may also provide additional
opportunities for commercial fisheries in the Great Lakes. In sum, restoration of
cisco in the lower Great Lakes appears to be an opportunity to achieve fish
community objectives and improve ecosystem health.

Reaching the goal of self-sustaining cisco stocks will be challenging.
Decisions on numbers and/or appropriate life stages to stock will be difficult
because of limited information on early life stage mortality rates and mechanisms
that control recruitment. To make informed decisions it is important to understand
how cisco function within their environment. A priori knowledge of mechanisms
important to population dynamics will increase the probability of restoration
success. Despite decades of research and management our understanding of
cisco dynamics is still vague.

A group of researchers interested in the rehabilitation of cisco convened in
2004 to discuss the status, research, and restoration needs of cisco in each of
the respective Great Lakes (Fitzsimons and O’Gorman 2006). After a period of
thought and additional research, many members of the 2004 group met again in
the spring of 2006. The opinions from these experts were that the larval stage
was critical for understanding cisco population dynamics and impediments to
their rehabilitation. Along with stressing the importance of the larval stage,
researchers hypothesized that multiple biotic and abiotic mechanisms may

determine the success of a given year class.

 

Lake Superior cisco stocks are believed to be near historic levels of
abundance, providing an environment for examining hypothesized bottlenecks in
cisco life history. The work reported in my thesis, which was conducted in
collaboration with the United States Geological Survey — Great Lakes Science
Center and the Ontario Ministry of Natural Resources — Upper Great Lakes
Management Unit, addresses two fundamental issues for understanding larval
cisco dynamics. In the first chapter I report on our study comparing several
sampling strategies for estimating the abundance of larval cisco. Development of
a sampling program that effectively captures the larval stage is necessary to
allow estimation of important population metrics and investigating hypotheses
concerning recmitment (Sammons and Bettoli 1998, Castro and Hernandez
2000, Karjalainen et al. 2000). Our work improved our ability to assess larval
cisco populations, which will facilitate interagency coordination and large-scale
research initiatives. In the second chapter we examine the influence of rainbow
smelt predation on the survival of larval cisco in two large Lake Superior
embayments. Past research attempting to understand the role of rainbow smelt
in determining cisco recruitment has led to varying results. We show that
incidental predation by rainbow smelt can be a substantial source of mortality for
larval cisco. Our evidence also suggests that rainbow smelt may have played an
important role in the historic collapse of cisco. We recognize that rainbow smelt
predation may not be the only mechanism inhibiting the recruitment of cisco but
believe that this added mortality could be detrimental to the occasional

successful cisco year class.

CHAPTER I:
EVALUATING SAMPLING STRATEGIES FOR

LARVAL CISCO (COREGONUS ARTEDI)

Abstract. To improve our ability to assess larval cisco Coregonus artedi
populations in Lake Superior, we conducted a study to compare several sampling
strategies. First, we compared density estimates of larval cisco concurrently
captured in surface waters with a 2 x 1-m paired neuston net and a 0.5-m
(diameter) conical net. Density estimates obtained from the two gear types were
not significantly different, suggesting that the conical net is a reasonable
alternative to the more cumbersome and costly neuston net. Next, we assessed
the effect of tow pattern (sinusoidal versus straight tows) to examine if propeller
wash affected larval density. We found no effect of propeller wash on the
catchability of larval cisco. Given the availability of global positioning systems, we
recommend sampling larval cisco using straight tows to simplify protocols and
facilitate straightforward measurements of volume filtered. Finally, we
investigated potential trends in larval cisco density estimates by sampling four
time periods during the light period of a day at individual sites. Our results
indicate no significant trends in larval density estimates during the day. We
conclude estimates of larval cisco density across space are not confounded by
time at a daily timescale. Well-designed, cost effective surveys of larval cisco
abundance will help to further our understanding of this important Great Lakes

forage species.

 

Introduction

The larval stage of cisco Coregonus artedi development is characterized
by a pelagic existence (Anderson and Smith 1971, Selgeby et al. 1978, Hatch
and Underhill 1988, Oyadomari and Auer 2004), where a host of physical and
biotic influences may act to limit recruitment (Roughgarden et al. 1988, Heath
1992). Research has shown that erratic fluctuations in adult abundance of many
fish species, like those observed for Lake Superior cisco (Yule et al. 2008a), are
often the result of extremely high mortality rates and variable survivorship during
the larval phase (Ricker 1954, Houde 1987). Many mechanisms have been
proposed to explain variations in recruitment, but the relative importance of
alternative mechanisms remains poorly understood (Cushing 1996).
Development of a sampling program that effectively captures the larval stage is a
promising avenue for estimating important population metrics and investigating
hypotheses concerning recruitment (Sammons and Bettoli 1998, Castro and
Hernandez 2000, Karjalainen et al. 2000).

Reliable assessment techniques are critical so that attempts to understand
complex ecological processes are not undermined by poor survey design. Net
avoidance is a serious consideration when attempting to measure the abundance
of agile planktonic organisms (Barkley 1972). If cisco larvae exhibit an escape
response, nets with a smaller effective sampling area would be subject to
conservative estimates of density compared to more encompassing gears. Also,

it would be expected that smaller nets would be more sensitive to the patchiness

 

of larval cisco (Oyadomari and Auer 2004) while larger framed nets would reduce
variability caused by fine-scale patchiness. Recent studies have suggested that
larval fish may react to the turbulence induced by a vessel's propeller (Claramunt
et al. 2005, Overton and Rulifson 2007). This creates a potential for differences
in estimated density due to the pattern in which the gear is towed and the time
spent directly in the boat wash. An added consideration is that samples within a
survey are confounded by the inherent limitation of sampling over both space
and time simultaneously (Levin 1992). Researchers have found that coregonid
larvae in Finnish lakes are aggregated near the surface during daytime, with
maximum catches occurring at approximately 1100-1400 hours (Viljanen et al.
1995). In Lake Superior it has been demonstrated that larval cisco density is
generally highest in the surface stratum (Selgeby et al. 1978, Hatch and Underhill
1988) during daytime (Oyadomari and Auer 2004), yet no study has investigated
the repeatability of estimatesthrough this period.

In this study we compare the performance of two gears for estimating
larval cisco abundance and make recommendations for survey tactics for this
important Great Lakes species. The present work was part of a larger study to
investigate the impact of rainbow smelt predation on larval cisco in Thunder Bay
and Black Bay, Ontario. Sampling of both bays in a timely fashion required us to
collect larval cisco using two vessels equipped with different gears. We assessed
the performance of the two gear types (a 2 x 1-m paired neuston net and a 0.5-m
[diameter] conical net) to determine whether density and mean length estimates

from the smaller conical net were comparable to those estimated from the

neuston net. We also investigated whether propeller wash influenced density and
mean length estimates by comparing samples collected off the stern of vessels
using sinusoidal and straight tow patterns. Finally, we examined potential trends
in density estimates from early morning to late afternoon to examine if time of
day could be ignored when generating mean density estimates across sites

within a day.

Methods

Data for this study were collected in Thunder Bay and Black Bay, Ontario
(Figure 1) during 2-25 May 2006. Two vessels were used to collect samples. The
first vessel was a 7.9-m Bertram with twin, 3.6 liter, stem-driven, Mercruiser
engines and was equipped with a hydraulic winch. Larval fishes were sampled
from this vessel with a 2 x 1-m paired neuston net (hereafter referred to as
“neuston net”) towed in the surface stratum. The second vessel was a 6-m
Boston Whaler with twin 150 horse power Evinrude E-TEC outboard motors.
Larval fishes were sampled with a 0.5-m (diameter) conical net (hereafter
referred to as “conical net”) towed in the upper 1 m of the water column from this
vessel. Both gears were equipped with 500-um mesh nets.

A single sample consisted of a 5-min tow in the 0-1 m stratum. Nets were
deployed behind each vessel with approximately 30 m of warp line out. During
sampling, each vessel weaved in a sinusoidal pattern to minimize towing time in

the respective vessels’ wash. The exception to this protocol was when testing for

 

 

Black Bay

   
    

 

Lake Superior

 

 

Thunder Bay

 

 

 

Wfilomtas

Figure 1. Stations used to eValuate sampling strategies for larval cisco.

the effects of propeller wash, in which one sample was towed in a sinusoidal
manner while the other paired sample (eg. from the same vessel) was collected
using a straight pull. To maintain consistency, all tows were pulled with the wind
on the stem at a speed of 3.2 to 3.9 km/hr. The volume of water filtered was
determined using General Oceanics model 2030 flowmeters. Both collections
from the paired neuston net were pooled for a single sample. All specimens were
preserved in 95% ethyl alcohol.

Larval lake whitefish Coregonus clupeaformis were distinguishable from

other larval coregonids based on size (Hinrichs 1979) and melanophore patterns

 

across the dorsum (Hinrichs 1979, Auer 1982). Differentiating larval cisco from
bloater Coregonus hoyi and kiyi Coregonus kiyi is not possible with visual
observation (Anderson and Smith 1971, Hinrichs 1979). Because bloater are
believed to spawn later than cisco (Hinrichs 1979, Auer 1982), we assumed that
emergence of cisco was separated temporally from the emergence of bloater.
Kiyi are not known to spawn in Thunder Bay and Black Bay (Ken Cullis, Ontario
Ministry of Natural Resources, personal communication). Additionally, the
maximum bathymetric depth in Thunder Bay, the deeper of the two bays, is 91 m
and the spawning habitat of kiyi is reported to be 91 to 168 m depth (Scott and
Crossman 1998). Thus, all emergent coregonids collected in May 2006 were
identified as cisco or lake whitefish.

All larval coregonids were counted for each sample. When samples
contained 5100 larval fish all specimens were identified as cisco or lake whitefish
and photographed using a Leica S6D dissecting scope equipped with a
Panasonic digital camera. When samples contained >100 individuals a random
sample of 50 specimens were identified as cisco or lake whitefish and
photographed. Photographs were used to digitally measure total length of each
fish to the nearest 0.1 mm. Density of larval cisco (#/1,000 m3 of water) in each
sample was calculated by dividing number caught by volume of water filtered.
We assumed net efficiency was 100%. All statistical tests were performed using

the statistical software SAS (SAS Institute Inc. 2003). Our significance level for

all analyses was a = 0.05.

Gear Comparison

To compare density estimates and length measurements between the two
gears, we collected samples using side-by—side tows with the two vessels. The
vessels were separated by approximately 50 m. Coordination by way of VHF

radio allowed for synchrony in gear deployment and tow duration. To satisfy the
assumption of normally distributed errors, density estimates were log1o-

transformed prior to analysis. A paired t-test was used to test the null hypothesis
that there was no difference between density estimates from the two gears. For
length comparisons we used a Kolmogorov-Smirnov test to test the null
hypothesis that there was no difference between the distributions of preserved

lengths from the two gears.

Tow Pattern

To examine the effect of turbulence caused by the propellers on density
estimates and average preserved length, paired samples were collected
successively at various sites. One sample of each pair was collected using a
sinusoidal pattern and the other sample was collected using a straight pull. A
single gear was used for sampling a specific site yet both vessels (and the
respective gears of each vessel) were used within this analysis. The sequence of
collections at a specific site was determined randomly. Both tows began at
approximately the same starting position yet the respective transects were not

overlapped. Collection of both samples took approximately 25 minutes. To satisfy

assumptions of normality we applied a log1o-transformation, with a small

11

 

constant added to the one zero observation (Johnson and Rausser 1971 ).
Mosteller and Tukey (1977) recommended adding a constant that is 1/6 the
minimum observable value, leading us to use a constant of 3 (Le. with a single

larvae captured by the conical net the minimum density estimate was 18 larval
cisco / 1,000 m3). A paired t—test was used to assess the null hypothesis of no

differences between density estimates derived from sinusoidal and straight tow
patterns. A Kolmogorov-Smirnov test was used to test for differences in the
length frequency distributions resulting from the pattern in which the net was

towed .

Daytime Trends

To investigate potential patterns in larval cisco density estimates during
light hours, we sampled a site during four time periods over the course of a day
(approximately 0600, 1000, 1400, 1800). Within each time period, three
consecutive samples were collected using the neuston net. This strategy was

replicated for a total of six days (six different sites) throughout the study period. A
log1o-transfomtation was applied to the density estimates prior to analysis to

stabilize variances and better meet assumptions of normality. A repeated
measures ANOVA was used to test the null hypothesis that there were no
temporal trends within the light period of an individual day. Because the model is
incapable of handling missing data we performed the analysis twice. In the first
analysis we used only the days in which a complete set of 12 samples were

collected (n = 3). In the second analysis we used only the first sample from each

‘

12

 

time period, thereby increasing the sample size (n = 5). On May 5, 2006 rough
seas prevented the collection of all late afternoon samples, causing us to exclude

this day from both analyses.

Results

A total of 162 icthyoplankton samples were collected to meet the
objectives of this study. We collected 114 samples using the neuston net and 48
samples using the conical net. We captured 31,246 and 1,331 cisco larvae with
the neuston and conical nets, respectively. Very few lake whitefish larvae were
identified during the study. Rainbow smelt larvae were observed in the samples
but were not counted. The number of samples collected and coregonid larvae

counted for the three respective analyses are summarized in Table 1.

Table 1. The number of samples collected, number of coregonid larvae

captured, and the mean density (#l1,000 m ) of cisco larvae sampled for the
evaluation of sampling strategies in Thunder Bay and Black Bay, Ontario during
2-25 May 2006.

 

 

 

Gear Tow Daytime
Comparison Pattern Trends
Neuston Conical Sine Straight Neuston
No. Samples 28 28 20 20 66
No. Whitefish Larvae 2 0 5 0 17
No. Cisco Larvae 8,088 1,089 5,145 242 18,013
Mean Cisco Density 474 672 299 331 453
is (98) (212) (75) (110) (69)
13

 

'K

Gear Comparison

A total of 28 side-by-side tows were made with the two gears. Density

estimates ranged between 50 and 2,317 larvae/1,000 m3 for the neuston net and

33 and 4,651 larvae/1,000 m3 for the conical net. Density estimates were not

significantly different between the two gears (t-value = -0.473, P = 0.64, DF = 27;
Figure 2A). Given the observed variance for the 28 paired collections we would
have been able to detect an effect size of 1.46 with 80% power (Lenth 2006).
Because we are ultimately interested in mechanisms that cause orders of
magnitude differences in abundance, we feel that a difference between gears of
less than 50% is not meaningful for larval cisco assessments and our sampling
effort was appropriate for the objectives of this investigation. Cisco larvae caught
using the neuston net had a mean length of 10.85 mm (SE = 0.026, n = 1,389)
and those caught by the conical net had a mean length of 10.73 mm (SE = 0.042,
n = 548). The distribution of larval cisco lengths were significantly different
between the two respective gears (K—S test: D = 0.094, P < 0.005) although the

differences in lengths captured were very small (Figure 2B).

Tow Pattern

We collected 10 paired samples using each respective net for a total of 20
comparisons. Because we found no difference between the two gears we
combined data from both nets for one analysis examining the effect of propeller
wash. We found no significant difference in density estimates based on tow

pattern (t-value = 1.715, P = 0.104, DF = 19; Figure 2C). Cisco larvae captured

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.1: . o ,”
g A , ’ 2
5 3 ° 2:" "E‘ 75
Q /.0 0
1 0 0 IO, 2
.- o I o 0
m . A: o a
3 2 . / ° 2 so
V 5’0 '5
3 I] ’ .5
z ,r E
8 1 X/ = 25
.E [I o
O l
o ,’
9 o
0 1 2 3 4 8
Neuston Net (Log1o[Density])
4
.170 C 0 ,TI 2
5 3 J 2’ I:
% Ire g
9 ‘
d 2 x" g
I O .—
2 1 E
9 z, 8
E
w A ,
0 1 2 3 4
Sinusoidal Tow (Log1o[Density]) Length (mm)

Figure 2. (A) Log1o-transformed larval cisco density (#/1,000 m3) estimates
and (B) cumulative length frequency distributions of larval cisco collected with
side-by-side tows using a 2 x 1-m paired neuston net and a 0.5-m (diameter)

conical net. (C) Log1o-transformed larval cisco density (#/1,000 m3) and (D) the
cumulative length frequency distributions of larval cisco collected using
sinusoidal and straight tow patterns with either a 2 x 1-m neuston net (0) or a
0.5-m (diameter) conical net (CI). The dashed lines (A and C) are 1:1 lines.

using the sinusoidal tow pattern had a mean length of 10.45 mm (SE = 0.031, n =
688) and those caught using a straight tow pattern had a mean length of 10.50
mm (SE = 0.035, n = 677). The Kolmogorov-Smirnov test revealed a significant
difference between the length frequency distributions resulting from the
alternative tow patterns (D = 0.078, P < 0.05), despite the fact that the cumulative

length frequency distributions were remarkably similar (Figure 2D).

15 ‘

Daytime Trends

Despite the variability in. density estimates within a single day at some sites
(Figure 3), on average there was no pattern in density estimates during daylight.
When analyzing days with three replicate tows for all four time periods, there was
no significant trends throughout the period of an individual day (F = 2.16, P =
0.194, n = 3). Likewise, when repeating the analysis using only the first tow for
the four time periods, there was no evidence of a pattern in larval cisco density

within a day (F = 0.32, P > 0.8, n = 5).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

‘E 2000 A May 5, 2005 2000 D May 19, 2005
O 1500 1 150°
8 1000 /\ 1000
t—' 500 l 500
g 2509 2500
E 2000 B May 7, 2005 2000 E May 22, 2005
3' 15001 1500
i 1000 1000 ‘
Q, 500 m 500 \/\
> o o
"*5 2500 2500 ,
8 2000 C May 13’ 2005 2000 F ‘ May 23, 2005
Q 1500 1500 . 1
g 1000 1000 x
g 500 500

o A 0

EM LM EA LA EM LM EA LA

Sampling Period

Figure 3. Average larval cisco density (#/1,000 m3) as a function of time of day.
Sampling events occurred in the early morning (EM), late morning (LM), early
afternoon (EA), and late afternoon (LA) for six individual days at different sites.
Error bars represent the range of estimated densities for each sampling event.

 

Discussion

Several studies have examined the dynamics of cisco early life history by
attempting to estimate abundance of swim-up larvae. Evidence from many of
these studies has shown emergent cisco to be distributed throughout the water
column but concentrated primarily in the surface stratum (Anderson and Smith
1971, Selgeby et al. 1978, Hatch and Underhill 1988, Oyadomari and Auer
2004), leading us to use only surface horizontal tows to index density. Selgeby et
al. (1978) evaluated the performance of several gear types (Clarke-Bumpus
sampler, 0.5—m tow net, 1 x 1-m net) and concluded, similar to the results of our
study, that the effective sampling area of a gear does not appear to affect
estimated density of larval cisco. Hatch and Underhill (1988) also conducted a
gear comparison (0.5-m Bongo sampler and 1-m Tucker trawl) and found no
difference between density estimates from the two gears. However, neither study
addressed sampling design issues in light of their results. Because Selgeby et al.
(1978) conducted a limited number of paired comparisons (n = 5) and Hatch and
Underhill (1988) did not report detailed diagnostics for their analysis, we felt that
our further investigation of appropriate sampling devices was warranted.

Our results suggest that samples from the conical net are comparable to
those from the much larger neuston net. Because conical nets are affordable,
can be deployed from any small vessel without special requirements (e.g.,
hydraulic winch), and result in smaller catches (equating to less laboratory

processing time), we recommend the conical net for future surveys of larval

17

 

cisco. We note that smaller sample volumes and the patchiness of local
distributions may cause surveys using conical nets to be vulnerable to greater
variation in density estimates. To some extent this can be offset by increasing the
number of tows conducted. Depending on the objectives of future studies this
may be an important consideration when choosing a gear to deploy. The length
frequency distributions of larvae captured by the two gears were significantly
different, with the neuston net capturing slightly larger individuals (Figure 2B). We
consider this difference, while significant according to the K-S test, is small
enough to be unimportant from a practical viewpoint. Both nets captured a limited
range of larval cisco sizes, possibly due to the emigration of larger age-0 cisco
from the surface waters or gear avoidance by larger larvae. We feel alternate
sampling strategies and gears would be needed to assess age-0 cisco once they
reach larger sizes, a goal beyond the scope of the present study.

A recent study showed bow-mounted push nets had an increased
efficiency for sampling of larval sunfish Lepomis spp. but not for gizzard shad
Dorosoma cepedianum, suggesting that vessel interference may influence larval
fish catchability for certain species (Claramunt et al. 2005). Because of the sea
conditions encountered in Lake Superior and the rigid mounting apparatus used
to deploy this gear type, its use was not considered as it would often rise above
the waters surface with the rise and fall of the vessels bow. Instead, we tested for
effects of vessel interference within our sinusoidal tow versus straight tow
analysis. Despite the concerns, we did not observe an effect on estimated

density caused by tow pattern. The difference in length frequency distributions for

the two tow patterns, while statistically significant, were very small (Figure 2D)
and are very unlikely to represent differences that would influence the biological
interpretation of survey results. Given the availability of global positioning
systems, we recommend sampling larval cisco using straight tows to simplify
protocols and facilitate straightforward measurements of volume filtered (distance
x net mouth area). We do not however suggest abandoning the use of
flowmeters, as their use may serve as an added measure of distance.

Finally, our investigation showed no trends in density over the light period
of a day, which indicates density estimates across space within a given day are
comparable and can be used to estimate mean density for the sampled region.
Recent efforts to measure spawner and recruit abundance are based on
statistical grids (10’ latitude by 10' longitude) because this scale is of interest to
managers within Lake Superior. Assessments of larval cisco will be of more utility
when they are executed to the same scale. Collectively, our results suggest that
multiple vessels/agencies can be used to conduct larval cisco surveys of major
geographic regions within Lake Superior, facilitating large-scale research
initiatives.

When designing sampling programs, it is imperative to understand the
biases associated with different gears and strategies. Studies targeting the larval
phase of cisco should employ gears known to efficiently capture the
demographics of this life stage. However, it is also important to make effective
use of time and resources. Developing protocols that balance these conditions

will ensure that surveys achieve a rigorous estimate of abundance across the

 

widest possible spatial range. Using a coordinated effort to advance the level of
understanding of cisco early life history in Lake Superior will facilitate research

initiatives and ultimately provide a greater understanding of cisco population

dynamics.

20

 

CHAPTER II:
AN EMPIRICAL RE-ASSESSMENT OF THE PREDATORY EFFECTS OF
RAINBOW SMELT (OSMERUS MORDAX) ON CISCO (COREGONUS ARTEDI)

IN LAKE SUPERIOR

Abstract. Evidence from small lakes suggests that predation on larval cisco
Coregonus artedi by non-native rainbow smelt Osmerus mordax can lead to
cisco suppression or extirpation. However, research on larger lakes has led to
equivocal conclusions. In this study, we examine the potential predation effects
of rainbow smelt in two adjacent but contrasting embayments on Lake Superior.
We sampled icthyoplankton, pelagic fish communities, and diet composition of
rainbow smelt and used a bioenergetics model to estimate the proportion of cisco
larvae consumed by rainbow smelt in each bay. Our results suggest predation by
rainbow smelt can account for up to 100% mortality of larval cisco. We
summarized density estimates of predators and prey at different spatial scales
and found predation impact decreased as spatial resolution increased. Finally,
we examined the sensitivity of past conclusions to estimates of rainbow smelt
abundance derived from bottom trawl samples and assumptions of temporal
overlap between rainbow smelt and larval cisco. After adjusting these parameters
to reflect current understanding, we found previous predation estimates may
have been conservative. We conclude rainbow smelt may have been an

important contributor to the demise and slow recovery of cisco in Lake Superior.

21 k

Introduction

Exotic species introductions threaten ecosystem integrity worldwide (Mack
et al. 2000; Mooney and Hobbs 2000). Freshwater systems are especially
vulnerable to biological invasions (Sala et al. 2000; Lodge 2001 ). The loss of
diversity, as a consequence of species invasions, appears to be greater in
freshwater systems compared to terrestrial systems (Ricciardi and Rasmussen
1999; Mooney and Cleland 2001). A successful invasion event often features a
rapid increase in the invading species followed by declines of previously
abundant native species (Capelli 1982; Worthington and Lowe-McConnell 1994;
Ricciardi et al 1998). Despite the practical importance of understanding the
relationship between exotic and native species, the mechanisms by which exotic
species impact native species are often poorly understood (Mercado-Silva 2007).

After intentional introduction to Crystal Lake, Michigan, in 1912, rainbow
smelt Osmerus mordax escaped into Lake Michigan by 1923 and subsequently
spread throughout the Great Lakes (Van Oosten 1937). Within small lake
systems, the effect of exotic rainbow smelt on native cisco Coregonus artedi
populations has been well documented. Evans and Loftus (1987) reported the
negative impact of rainbow smelt invasions in Ontario Lakes, noting the
substantial change in food web interactions and displacement of native species
such as cisco. Loftus and Hulsman (1986) combined empirical evidence with
modeling to show that rainbow smelt predation on coregonid larvae was the

primary cause for recruitment failure in Twelve Mile Lake, Ontario. Similarly,

” A

 

Hrabik et al. (1998) showed that predation by rainbow smelt led to the extirpation
of cisco in Sparkling Lake, Wisconsin, only eight years after rainbow smelt were
first detected. Further, restoration of walleye Sander vitreus populations in
northern Wisconsin lakes has increased predation pressure on rainbow smelt,
leading to cisco recovery (Krueger and Hrabik 2005). However, inferences drawn
from these small-lake studies may not be applicable to the much larger Great
Lakes because ecosystem size could lead to differences in spatial and temporal
overlap of predator and prey (Peterson et al. 1999).

Biological invasions have had many effects on native fish communities
within the Great Lakes (Mills et al. 1993; Ricciardi 2001; Madenjian et al. 2008).
However, investigators have reached different conclusions about the affect of
rainbow smelt on cisco in Lake Superior. Anderson and Smith (1971) and
Christie (1974) found negative correlations between rainbow smelt and cisco and
concluded that rainbow smelt were responsible for the collapse of cisco stocks.
Anderson and Smith (1971) suggested competition was the mechanism that
caused cisco to decline. Swanson (1978) suggested that turbidity contributed
indirectly to the decline of cisco by concentrating larval cisco and predatory
rainbow smelt in surface waters. Further, Cox and Kitchell (2004) developed an
ecosystem simulation model to evaluate hypotheses concerning recruitment
failure of Lake Superior cisco and found that their “strong rainbow smelt
predation” hypothesis best fit the time series data (1929-1998).

In contrast, neither predation (Selgeby et al. 1978) nor competition

(Selgeby et al. 1994) by rainbow smelt were identified as important factors

23

contributing to the decline of Lake Superior cisco in the Apostle Island region,
Wisconsin, and Black Bay, Ontario. Selgeby (1982) examined commercial catch
and effort statistics for the Wisconsin cisco fishery spanning 1936 to 1978 and
concluded overfishing was the likely cause of stock collapse. Bronte et al. (2003)
added that if rainbow smelt were depressing cisco abundance prior to the late
1970s, then the 90% reduction in rainbow smelt biomass during 1978-1981
should have resulted in immediate strong recruitment of cisco, which did not
occur until the 1984 year class (MacCallum and Selgeby 1987).

In this study, we performed a comparative assessment of the influence of
rainbow smelt predation on larval cisco populations in Thunder and Black Bays,
Ontario. Despite their close proximity, these bays exhibit contrasting physical and
biological characteristics. Thunder Bay (74,000 ha) is oligotrophic with moderate
densities of rainbow smelt while Black Bay (51,000 ha) is more productive and
currently supports higher densities of rainbow smelt. Historically, both bays
supported large commercial fisheries for cisco (Jacobson et al. 1987), but these
fisheries declined throughout the 19803. The Thunder Bay cisco stock is
presently composed of several year classes and the stock appears strong (Yule
et al. 2008a), but the Black Bay cisco stock is likely well below historic levels. We
used a bioenergetics model to estimate consumption of larval cisco by rainbow
smelt and compared this to estimates of total larval cisco abundance for each
bay. We also examined the effect of spatial averaging of density estimates on our
conclusions. Specifically, we compared estimates when we assumed each bay

had homogonous densities of rainbow smelt and larval cisco with estimates

24

obtained using the observed spatial variation in predator and prey densities.
Finally, we re-estimated predation rates from the data of Selgeby et al. (1978),
using the WI bioenergetics model and more recent information on several key
assumptions, including duration of predator prey overlap and availability of

rainbow smelt to bottom trawl gear.

Methods

Larval cisco survey

We conducted two synoptic surveys of both Thunder and Black Bays in
2006. Thunder Bay sampling occurred between 2-7 May and 21-25 May, while
sampling of Black Bay occurred on 13 May and 19-20 May. A systematic
sampling design (Figure 4) was employed after pilot data suggested 1 site/1,000
ha would result in standard errors that were 20-30% of the mean density (US.
Geological Survey, unpublished data). Two vessels were used during the survey.
A 2 x 1-m paired neuston net (“neuston net”) was deployed from one vessel and
a 0.5-m (diameter) conical net (“conical net”) was deployed from the second
vessel. Both nets had 500-um mesh. A single sample consisted of a 5-min tow in
the 0-1 m depth stratum. Nets were towed 30 m behind each vessel. To maintain
consistency, all tows were pulled with the wind on the stern with vessel speed
ranging from 3.2 to 3.9 km/hr. The volume of water filtered was determined using
General Oceanics (Miami, Florida) model 2030 flowmeters. The two net samples

from the paired neuston net were pooled into a single sample for each tow. Cisco

25

larvae were present in the surface stratum throughout the study period and all
collected specimens were preserved in 95% ethyl alcohol. Previous analyses
demonstrated the two gears provide comparable larval cisco density estimates
and repeated sampling at several sites showed estimates did not vary over
daylight hours (Myers et al. In Press).

Previous studies have demonstrated that larval cisco are generally
concentrated near the surface (Selgeby et al. 1978, Hatch and Underhill 1988,
Oyadomari and Auer 2004). We estimated abundance of larval cisco in each bay
during each synoptic survey by multiplying mean densities by the volume of

water in the uppermost 1 m of the water column of each bay. An analysis of

 

Larval Cisco Density
# Fish l Hectare

: ‘ 0-500 3 0
ET”? 501 _ 1,000
E91: 1001. 5,000
- 5.001 - 10,000
- 10,001 - 20.000

 

 

 

 

  

. g, '5.“
u 7%.; O

   

Survey 1 Survey 2 x

“a“ I - Afi‘

_ _ lGlometers
0 5 10 20 30 40

 

 

 

 

 

Figure 4. Map of larval cisco sampling locations and density estimates in
Thunder and Black Bays, Lake Superior. Two synoptic surveys of each bay were
completed between 2-25 May 2006. Interpolated surfaces were created using
inverse distance weighting.

26

larvae length-frequency distributions showed that fish collected during the
second synoptic survey of both bays were likely new emergers, so we opted to
estimate total larval abundance by summing the two estimates for each bay.

Visual identification of larval coregonids is inexact. Lake whitefish
Coregonus clupeaformis larvae are distinguishable from other larval coregonids
based on larger size (Hinrichs 1979) and melanophore patterns across the
dorsum (Hinrichs 1979, Auer1982). Differentiating larval cisco from other
coregonids like bloater Coregonus hoyi and kiyi Coregonus kiyi is difficult
(Anderson and Smith 1971, Hinrichs 1979). Because bloater are believed to
spawn later than cisco (Hinrichs 1979, Auer 1982), we assumed that emergence
of cisco was separated temporally from the emergence of bloater. Kiyi are not
known to spawn in Thunder Bay and Black Bay. We therefore identified larval
coregonids as cisco or lake whitefish.

All larval fish in each sample were counted. When'samples contained
5100 larval fish all specimens were identified and photographed using a Leica
86D dissecting scope equipped with a Panasonic digital camera. When samples
contained >100 individuals a random sample of 50 specimens were selected to
estimate proportions of cisco and lake whitefish. We measured total length (TL)

of each photographed fish to the nearest 0.1 mm using SigmaScan Pro 5
software (SPSS, Inc., Chicago, IL). Density (larval cisco/1,000 m3 of water) was

calculated for each sample. Because the northernmost extent of the
hydroacoustic survey in Black Bay was approximately 48.58° N latitude, average

larval density values were calculated using only sites south of this latitude (Figure

27

4). It is unlikely that cisco spawn in the northern reaches of Black Bay as
commercial fisherman have not been successful there (Ken Cullis, Ontario

Ministry of Natural Resources, personal communication).

Hydroacoustics and trawling survey

Multiple gears were used to estimate densities of age-1 and older fish in
Thunder and Black Bays between 13-22 May 2006. All samples were collected
using the US. Geological Survey (USGS) research vessel Kiyi. Between 13-15
May, we collected four daytime bottom trawl samples in each bay at standardized
sites sampled annually each spring (Stockwell et al. 20063). We also collected a
daytime midwater trawl sample (between 20-39 m depth) at a Black Bay site on
17 May where high densities of larval cisco had been collected on 16 May. We
systematically surveyed each bay during the nighttime hours of 16-22 May using
a series of parallel acoustic transects spaced at roughly 3 km intervals (Figure 5).
Nighttime work commenced 30 minutes after the start of nautical twilight and
ended 1-2 hours before nautical twilight ended. We collected night midwater trawl
samples periodically along the acoustic survey path, with a total of 7 and 14
samples collected from Black and Thunder Bays, respectively. Bottom trawl
samples were not collected at night because rainbow smelt (the species of main
interest in this study) perform diel vertical migration in Lake Superior (Heist and
Swenson 1983; Yule et al. 2007).

The depth distribution, acoustic size, and densities of fish were estimated

using a DT-X digital echosounder (BioSonics, Inc., Seattle, Washington)

28

 

        
     
 

Black Bay

 

 

 

Thunder Bay

 

Rainbow Smelt Density
# Fish l Hectare

I 1001—5000
I 5001-10000
I 10.001-25.000

 

 

 

 

cc—Z— Kilometers
0 2.5 5 1 0 1 5 20

Figure 5. Map of acoustically-derived rainbow smelt density estimates collected
in Thunder and Black Bays (Lake Superior), 17-22 May 2006. Acoustic densities
were estimated for each1 km of boat travel. Interpolated surfaces were created
using inverse distance weighting.

equipped with a 120 kHz circular split-beam transducer with a half-power beam

width of 67°. The transducer was mounted on a 1.2-m-long tow body and was

deployed 1 m below the surface. The transducer emitted 3 pings/s with the pulse

29

‘

duration set at 0.4 ms. Vessel position was measured with an Ashtech model
BRG2 differentially corrected global positioning system (accurate to 1 m) and
positional information was embedded in the acoustic data files. Standard target

calibration was performed on 11 April 2006 using a 33-mm tungsten carbide

sphere (theoretical target strength [TS] of -41 .5 decibels [dB] at 7° C). Mean TS

of the sphere (—42.5 dB 1 1.0 dB SD, N = 619 echoes) was within 1 dB of the
expected TS so we did not apply corrective offsets to the acoustic data before
calculating fish densities.

The midwater trawl (Gourock Trawls, Femdale, Washington) had 15.2-m
headrope and footrope lines and 13.7 m breast lines. The nylon mesh graduated
from 300 mm stretch measure at the mouth to 12 mm at the cod end. The bottom
trawl used (3/4 Yankee trawl number 35) had an 11.9 m headrope, 15.5 m
footrope and 2.2-m-high wing lines with 89-mm stretch measure mesh at the
mouth, 64-mm mesh at the trammel and 12-mm mesh at the cod end. We placed
miniature depth and temperature loggers (VEMCO, Shad Bay, Nova Scotia) on
the midwater trawl headrope and footrope to measure the headrope depth and
trawl mouth height when fishing.

Trawl catches were sorted by species and weighed in aggregate to the
nearest gram. For small catches (< 50 individuals per species) all fish were
measured to the nearest mm total length. For larger catches at least 50
individuals per species were selected randomly and measured and the remaining

fish were counted.

30

Fish density estimates and apportioning acoustic targets to species.

Acoustic data were processed using Echoview software version
3.25.55.06 (SonarData Ltd., Tasmania, Australia). Fish density estimates were
developed using echo integration methods. We used a bottom tracking algorithm

to establish a line 0.2 m above the bottom (i.e., “a bottom exclusion line”). Before

measuring mean volume backscattering strength (i.e., Sv) we double-checked all

echograms to ensure bottom echoes were properly excluded. A second line was
added 3-5 m below the transducer of all echograms (“surface line”) to exclude
echoes in the transducer near field and any air bubbles resulting from high seas.
We used the embedded vessel position data to define 1,000 m intervals on
echograms and each interval was further divided into 10-m-high cells from the
surface line to the bottom exclusion line. We measured the area backscattering

coefficient (ABC) in each cell. MacLennan and Fernandez (2000) defined ABC as

10(SV/1O) x T, where T is the mean thickness of the cell (m) being integrated.

Before measuring ABC we applied a -65 dB 8,, threshold to minimize the

inclusion of backscattering from the macroinvertebrate Mysis relicta. We

estimated the mean backscattering cross section (Obs) of the average size fish in

each cell from mean target strength (obs = 10TS/1O). We used identical criteria for

single-target acceptance defined by Yule et al. (2006) except the minimum TS in
the present study was set at -60 dB. This lower threshold was chosen to ensure
that age-1 rainbow smelt were included in the calculation of mean TS. Parker

Stetter et al. (2006) showed that yearling and older rainbow smelt consistently

31

had 3 TS larger than -60 dB in Lake Champlain (New York - Vermont, United

States, and Quebec, Canada). Fish density (#lha) by cell was calculated by
dividing ABC by 005 and multiplying the resultant by 10,000 (i.e., 10,000 m2 = 1

ha). All 10—m-high cells in each 1000-m-long interval were summed to estimate
total density in each interval.

To apportion acoustic targets to species we developed an approach for
separating rainbow smelt from cisco based on differences in their acoustic sizes.
Using a parallelogram drawing tool available in the Echoview software, we drew
midwater trawl paths on echograms (all data were time tagged). We then
exported single targets from each trawl path so that TS distribution data could be
compared to length distributions in each trawl sample. We used TS distributions
from nearly-pure catches of cisco and rainbow smelt to determine the relative
proportions of these species at locations where trawl samples were not collected.
Of the 21 night midwater trawl samples, two were nearly-pure cisco (> 97% by
number) and two were nearly pure rainbow smelt (98%). The single targets from
these trawl paths were used in a recursive partitioning platform available in JMP
5.1 (SAS Institute, Inc, Cary, NC) to determine the TS “cutting values” which
most significantly separated the mean TS of these species. The recursive
partitioning technique was applied to the four possible pair-wise comparisons to
determine four cutting values that we then averaged.

We compared the efficacy of our mean “cutting value” approach for
separating rainbow smelt and cisco acoustic targets by comparing predicted

proportions of cisco and rainbow smelt (separated with the cutting value) to

32

observed proportions in each of the 16 night midwater trawl samples that were
not used to develop the cutting values. We excluded one deepwater midwater
trawl sample because the catches of both rainbow smelt and cisco were minimal.
We used a t-test for independent samples to test the null hypothesis that

hydroacoustics and midwater trawling achieve similar proportions of rainbow

smelt and coregonids. We set a = 0.05 for all statistical tests.

Rainbow smelt food habits

Rainbow smelt stomachs were collected to assess food habits. Upon
capture, rainbow smelt were quickly sorted into 10-mm-wide length bins (i.e., <
50 mm, 50-59 mm,...,>150 mm) targeting up to 10 fish per bin per trawl sample
and placed on ice to slow digestion. We then excised stomachs and combined
them into one sample vial per length bin per trawl sample. Stomachs were
preserved in alcohol, and vials were labeled with the trawl serial number and
length bin so we could later determine where and when the stomachs were

collected.

Consumption and predation mortality estimates

We used the Wisconsin bioenergetics model (Wisconsin SeaGrant, UW-
Madison, Center for Limnology, Bioenergetics v 3.0 software) to estimate
rainbow smelt predation on larval cisco populations. Because larval cisco and
rainbow smelt were caught during the duration of our studyn(2-25 May 2006) we

opted to model consumption of rainbow smelt for 30 days (i.e., 1-30 May). We

33

separated rainbow smelt into four size classes (560 mm, 61-90 mm, 91-120 mm,
and >120 mm; Figure 6) to structure the bioenergetics analysis with respect to
rainbow smelt size.

The bioenergetics model required several inputs, including average
temperature experienced by rainbow smelt, rainbow smelt diet composition, and

the caloric content of both rainbow smelt and their prey. Because Thunder Bay

was isothermic at 4.7°C, we used this temperature in the Thunder Bay

simulations. In Black Bay, water temperature near the surface was slightly

warmer (58°C) than near the lakebed (49°C) so we used a value of 5.4°C. We

examined rainbow smelt stomach contents under a dissecting microscope (Nikon
SMZ-ZB). Contents were separated into the following categories: Mysis relicta,
Diporeia spp., zooplankton, larval coregonids, or other items (terrestrial insects,
benthic organisms, yearling and older fish, etc.). Because the larvae in the
icthyoplankton survey were identified predominately as cisco (>99%), coregonid

larvae in rainbow smelt were assumed to be cisco. Each respective diet item was

placed in a drying oven for 24 hours at 60°C and weighed to the nearest 0.0001

gram using an analytical balance (Sartorius ED2245). We determined the
average diet composition (as determined by dry weight) in each bay by pooling
all stomachs in each rainbow smelt size class (Table 2). We used the estimates
of rainbow smelt energy density reported by Lantry and Stewart (1993). A

relationship provided by Elliot et al. (1996) was used to convert standard

34

 

I 4
>120mm

  

Thunder Bay

 

 

W w 80'
61-90 mm 91-120 mm

 
 

 

 

 

W
560 mm

 

.//////
/////////
////////

//
//////////
////////////////
////////////////////////////

%/

       
 
  
 

r//

A

     
 
 

 

150

5
2
1
9.5

05
52

m m
1

305030 u.o LonEzz

O

 

Black Bay

 

 

 

 
 

 

 

  

//////
////////////////
/////////////////////////////////////
//////////////////////o

/////////////6

fl

//

%7/
4O

 

    
 
 
 

   

 

150

5
2
1

c0

05
52

m m
1

$2030 Co 23:52

0

120 140 160

00

1

m

Length (mm)

Figure 6. Length frequency distributions for Thunder Bay (363 measurements)

and Black Bay (400 measurements) rainbow smelt and the four size classes (5

60 mm, 61-90 mm, 91-120 mm, > 120mm) used to structure the bioenergetic

analyses.

35

Table 2. Relative contribution of each prey item to the diet of rainbow smelt.
Percentages are based on the dry weight of stomach contents.

 

 

 

 

Length Stomachs Stomachs DiGt Item
(mm) Examined wl Food Cisco

Mysis Diporeia Plankton Larvae Other
3 560 171 44 18.9% 0.0% 78.8% 0.0% 2.3%
3', 61 -90 268 151 43.3% 0.7% 54.1% 0.0% 1.9%
2 91-120 165 101 58.2% 8.5% 30.0% 2.0% 1.3%
E >120 134 71 60.1% 15.9% 7.6% 0.1% 16.4%
g‘ 560 107 53 0.5% 0.0% 99.5% 0.0% 0.0%
3 61-90 342 306 5.8% 1.0% 91.5% 0.1% 1.8%
3 91-120 266 240 3.4% 4.6% 88.1% 0.2% 3.6%
6 >120 96 57 12.5% 8.5% 43.7% 0.1% 35.3%

 

Iength(SL) reported by Lantry and Stewart (1993) to an estimate of total length

(TL). We calculated the average length of rainbow smelt in each size class in

each bay and used a length-weight relationship for rainbow smelt in Lake

Superior (Yule et al. 2007) to estimate the average individual weight in each size

class (Table 3).

We used literature values to estimate energy densities of prey consumed

by rainbow smelt (Table 4). Johnson et al. (1998) reported a linear relationship

between energy density and weight of cisco up to 500 9. Because the weight of

individual cisco larvae is negligible, we used the intercept (4,550 J/g) of the

Johnson et al. (1998) relationship to estimate energy density of larval cisco

(Table 4). This estimate of energy density falls within the range of values

36

Table 3. Model inputs for rainbow smelt energy density (J/g wet weight) in
Thunder Bay and Black Bay. Simulation day 1 is May 1. Average length within
each size class was determined using the information provided in Figure 6.

 

 

 

 

 

 

 

 

 

 

Simulation Rainbow Smelt Total Length
Day 560 61-90 91-120 >120
Average Length (mm) within Bin
3‘ 54 72 108 132
m
3 Average Weight (g)
.5 0.5 1.4 5.7 11.2
c
2 Energy Density (J/g)
I- 1 3268 3526 4997 5205
30 3372 351 1 4425 4674
Average Length (mm) within Bin
55 75 103 123
>5
(U
m Average Weight (g)
x 0.6 1.6 4.8 8.8
o
E
m Energy Density (J/g)
1 3268 3784 4954 5127
30 3372 3650 4373 4581

 

Table 4. Energy density (J/g wet weight) of prey items found in the stomachs of
rainbow smelt from Thunder and Black Bays in Lake Superior.

 

 

Energy
Density
Prey (Jig) Source
Mysis relicta 3,537 Johnson et al. (1993)
Diporeia spp. 4,386 Johnson et al. (1993)
Zooplankton 3,016 Johnson et al. (1993)
Larval Cisco 4,550 Johnson et al. (1993)

Other (Insects and Fish) 3,786 Lantry and Stewart (1993)

37

reported for other Great Lakes larval fishes (Hartman and Margraf 1992). Energy
density was held constant throughout our 30-day model simulations.

Growth of rainbow smelt during our 30 day simulations was too small to
estimate accurately from our data. Therefore, we used a range of P-values
(proportion of maximum consumption) to estimate consumption. We modeled two
scenarios: 1) the P-value needed for maintenance (i.e., start wt = end wt) for
each size rainbow smelt size class (P-value = 0.18-0.22), and 2) a P-value of 0.5.
which corresponds with the observed growth reported by Baily (1964). These P-
values provided a range of realistic growth trajectories and thus predation
intensities.

The bioenergetics model integrated all variables and predicted the
biomass of each prey item consumed, given the rate of feeding. We multiplied
the biomass consumed by our estimates of rainbow smelt abundance to estimate
the total biomass of larval cisco consumed. The biomass of larval cisco
consumed was divided by their average mass to estimate the total number
consumed. Average mass of cisco larvae (0.003 g) was estimated by weighing
89 individuals (ranging from 9512 mm) to the nearest 0.001 grams. After
completing a model simulation for each size class we summed across size
classes to estimate total numbers of cisco larvae consumed in each bay. We
estimated the mortality attributable to rainbow smelt predation by dividing the
number of cisco larvae consumed by the total abundance of cisco larvae in each
bay. Predation mortality (PM) was calculated in following manner:

PM=(D*C*R)/L

38

where D is the duration of feeding (days in contact), C is the number of cisco
larvae consumed per day (dependent on the daily ration), R is rainbow smelt
abundance, and L is larval cisco abundance. We did not allow predation mortality

to exceed 100%.

Effect of spatial scale on predation mortality

To examine the influence of spatial scale on our estimates of predation
mortality, we analyzed the data at three different scales. In the first analysis we
calculated the arithmetic mean densities of rainbow smelt in each bay using the
1,000 m acoustic intervals as sample units. We calculated the arithmetic mean
density of larval cisco for each synoptic survey using all 5-minute larval tows.
Abundance of rainbow smelt and cisco larvae was calculated by multiplying
density estimates by the area of each respective bay. Total abundance of cisco
larvae was calculated by adding estimates of abundance from the two synoptic
surveys of each bay. In the second analysis, Thunder Bay was divided into
quadrants, Black Bay was halved, and mean regional densities were calculated.
We determined which samples were collected in each region and calculated
mean regional densities. Abundance of each region was calculated by multiplying
average density (#lha) for the region by its area (ha). Total abundance of cisco
larvae within a region was calculated by adding results from the two synoptic
surveys. Total abundance of available cisco larvae was calculated by summing
regional abundances for each bay. In the final analysis we divided both bays into

grids measuring 51,000 ha (i.e., a grid was less than 1,000 ha if the shoreline did

39

not permit for a complete grid). We used the Geostatistical Analyst Extension of
ArcMap version 9.0 (ESRI, Redlands, CA) to create interpolated surfaces using
inverse distance weighting. The number of pixels (1 ha) in each grid were
counted to determine the area of each grid. We used the average of all pixels
within a grid to estimate densities of rainbow smelt and cisco larvae during the
first and second synoptic survey. We calculated abundance by multiplying the
density estimates by the area of each respective grid. Total abundance of cisco
larvae in each grid was determined by adding the abundance estimates from
each synoptic survey. Total abundance of cisco larvae was calculated by

summing the abundances from each respective grid.

Re-assessment of eartier findings

Selgeby et al. (1978) and Selgeby et al. (1994) examined the influence of
rainbow smelt on the survival of larval cisco in Black Bay and the Apostle Islands
in 1974. Because of similarities in the objectives and sampling location (Black
Bay) of Selgeby et al. (1978) and our current study, we compared and contrasted
assumptions of the two approaches. Selgeby et al. (1978) estimated rainbow
smelt densities with day bottom trawls but MaSon et al. (2005) later showed day
bottom trawl samples provided lower estimates of rainbow smelt densities when
compared to acoustic and midwater trawl (AC-MT) methods. Also, Selgeby et al.
(1978) assumed that rainbow smelt consumed larval cisco during a 5-10 day
period. However, results of our study suggest that both larval cisco and rainbow

smelt overlap for a much longer period (35 30 days).

40

To determine the average number of cisco larvae consumed by rainbow
smelt within a given day in May 1974, Selgeby et al. (1978) scaled the mean
stomach content mass and the average number of cisco larvae present in the
guts to project daily rations of 1.5% and 2.5% of their body weight. Using
information provided by Selgeby et al. (1978) and Selgey et al. (1994) we
developed a bioenergetics model to re-estimate the average number of cisco
larvae consumed by rainbow smelt within a given day in May 1974. The
proportion of diet items found in the stomachs of rainbow smelt in the spring of
1974 was reported in Selgeby et al. (1994). The value of rainbow smelt energy
density, energy density of their prey, and temperature in Black Bay was assumed
to be the same as our current study. We used daily rations of 1.5% and 2.5% (in
lieu of P-values) to estimate the average number of cisco larvae consumed by
rainbow smelt in each of the four rainbow smelt size classes (560 mm, 61-90
mm, 91-120 mm, and >120 mm). Predation mortality was calculated using the
equation described previously. Abundance of rainbow smelt and larval cisco in

May 1974 was reported in Selgeby et al. (1978).

Results

Larval cisco survey

Density (:l: SE) of cisco larvae in Thunder Bay averaged 2,796 i 69.7

fish/ha (n = 63) during the first synoptic survey (2-7 May) and 3,343 i 54.5

fish/ha (n = 74) during the second survey (21-25 May). In Black Bay, density of

41

cisco larvae declined from an average of 1,850 :1: 208 fish/ha (n = 16) during the
first survey (13 May) to 471 i 53 fish/ha (n = 19) during the second survey.
Differences in larval distributions were apparent during the two surveys of both
bays (Figure 4). With exception of the extreme northeast and southeast portions
of the Thunder Bay, relatively high densities of larval cisco were found
throughout the bay during both surveys. In Black Bay there appeared to be a
north-to-south gradient in larval cisco density. Four larval samples were collected
in the most northern waters of Black Bay (Figure 4) on 19 May 2006 but no cisco

larvae were captured.

Hydroacoustics and trawling survey

During 17-22 May 2006 we collected 220 and 96 km of acoustic data in
Thunder Bay and Black Bay, respectively (Figure 5). Cisco were predominant in
Thunder Bay, representing 57.3% of 2,156 fish caught by night midwater trawl
samples, followed by rainbow smelt (30.9%), kiyi (5.0%), and bloater (4.9%).
Rainbow smelt were the principal species in Black Bay, representing 95.4% of
the fish caught in night midwater trawl samples (4,701 total fish). The mean
cutting-edge TS value for separating cisco and rainbow smelt was -44.59 dB. We
calculated expected proportions of rainbow smelt and cisco in each of the 16
midwater trawl samples not used to develop this mean cutoff value. Expected
proportions of rainbow smelt and coregonids did not vary significantly from the

actual proportions captured in the 16 midwater trawl samples (rainbow smelt: t-

42

value = -0.335, P > 0.7, DF = 30, Figure 7A; coregonids: t-value = -0.934, P >
0.3, DF = 30, Figure 7B).

We concluded that the approach used to apportion acoustic targets to
species was reasonable. Thus, when calculating species densities we assumed
all targets smaller than -44.59 db were rainbow smelt, regardless of depth. This
empirically derived cutting edge value (-44.59) is consistent with recent literature
reporting acoustic versus real size relationships for rainbow smelt and
coregonids. For example, an equation by Rudstam et al. (2003) predicts the TS
of a 150 mm rainbow smelt to be -44.40 dB at 120 kHz, which is within 0.2 dB of
our cutting value. Meanwhile, an equation by Mehner (2006) (developed for
vendace Coregonus albula) predicts that a 150 mm coregonid has a TS of -40.91
dB, which is considerably larger than an equivalent size rainbow smelt. We were
confident in our identification of rainbow smelt using acoustic TS because most
captured rainbow smelt (98.9%) were less than 150 mm total length (TL) and
most captured cisco exceeded 225 mm TL.

Acoustically derived estimates of rainbow smelt density were generally
high in Black Bay (3,435 i 460 fish/ha; Table 5), with the greatest values along
the southwest shore and the central east portions of the bay (Figure 5). The
estimate of rainbow smelt density using acoustics at night was approximately
three times higher than the value obtained from four day bottom trawl samples
collected in Black Bay (1,181 :I: 460 fish/ha; Table 5). In Thunder Bay, there were
small areas of moderate rainbow smelt abundance but densities were generally

low across most of the bay (Figure 5). The estimated mean density of rainbow

43

..x
O

 

A) Rainbow Smelt

.0 .0 .o
A O) CO
‘\
0
\\
O
\
O
0

Predicted Proportion
(Proportion Targets < -44 59 db)
.0
N

 

 

 

 

4.0
'00

.0 .0 .0
-P- O) 00
0
\‘x
N.
\

Predicted Proportion
(Proportion Targets > -44 59 db)
3
0° 0
\ \ &
O
0

 

 

.0
o

 

0.0 02 0.4 06 0:8 1.0

Proportion in Midwater Trawl

Figure 7. Observed proportions of rainbow smelt (A) and coregonids (B)
captured in midwater trawls versus the predicted proportions from the TS
distributions developed using the cutting-edge approach to apportion targets to
species. The dashed line is the 1:1 line. A frequency distribution of the
differences is located in the lower right corner.

44

Table 5. Rainbow smelt density estimates (rainbow smelt/ha [tSE]) using night
acoustics and midwater trawling (AC-MT) and day bottom trawls (BTR) in
Thunder and Black Bays.

 

AC-MT BTR

Thunder Bay 530 (34) 932 (486)
Black Bay 3,435 (460) 1,181 (460)

smelt in Thunder Bay was 530 i 34 fish/ha using night acoustics. Based on four
day bottom trawls, the mean density of rainbow in Thunder Bay was 932 1 486
fish/ha. Use of day bottom trawls to characterize the fish communities would lead
to the inference that rainbow smelt densities were similar between Thunder and
Black Bays (Table 5). However, evidence from acoustic coverage of both bays
suggests that rainbow smelt densities were 6-7 times greater in Black Bay
compared to Thunder Bay. The day midwater trawl collected in Black Bay caught
XXXX rainbow smelt. Because of the concerns associated with using bottom
trawls to target pelagic species, we only used acoustically-derived estimates of

rainbow smelt density in our bioenergetic analyses.

Rainbow smelt size-structure, consumption, predation mortality, and the effect of
spatial scale

A total of 386 and 400 rainbow smelt were randomly selected and
measured for total length in Thunder and Black Bays, respectively. There were
two prominent size classes in Thunder Bay while there was proportionally more
yearling (fish < 100 mm) rainbow smelt in Black Bay (Figure 6). A total of 1,549

rainbow smelt stomachs were examined (Table 2). Fifty percent of the stomachs

45

from Thunder Bay contained food compared to 81% of the stomachs from Black
Bay. As rainbow smelt increased in size their diet shifted from predominately
zooplankton to larger prey (Mysis relicta, terrestrial insects, fish). The incidence
of larval cisco in rainbow smelt guts was low in both Thunder and Black Bays
(Table 2). 1

Using arithmetic mean density estimates of rainbow smelt and larval cisco
for Thunder Bay, we estimated that rainbow smelt consumed 107-261 million
cisco larvae (depending on the p-value), equivalent to a predation mortality of 24-
58% (Figure 8). Similarly, dividing Thunder Bay into four regions resulted in a
projected 104—254 million larvae consumed, which was 23-55% of the larval cisco
population (Figure 8). When densities of predators and prey were estimated for
each 1,000 ha grid cell in Thunder Bay, we projected that rainbow smelt
consumed only 80-178 million larvae, or 17-38% of the estimated larval cisco
population (Figure 8).

Using arithmetic mean density estimates of rainbow smelt and larval cisco
for Black bay we projected that 23-48 million cisco larvae were consumed during
May 2006, or 48-100% of the larval cisco population (Figure 8). Dividing Black
Bay into a north and south region resulted in a projected 23-52 million larvae
consumed (43-100% of the larval cisco population). When densities of predators
and prey were estimated for each 1,000 ha grid cell we projected that rainbow
smelt consumed only 22-31 million larvae, 38-53% of the larval cisco population

(Figure 8).

46

 

 

 

 

 

 

 

 

 

 

 

  
   

N
0|

45:5
.55:

o
0'

C
.0

4:"
3?

'0
C

":3-

'~:
$0

"
O
a
0'0 .
:'
U
.9-

5; .

Predation Mortality
S

N
U!
. . W,
933*

-. 3.,
i=2.

...g’fi‘

 

 

ff???

5:;

Bayw'de Regional 1, 00 ha

 

Thunder Bay E:

v. L-J

Maintenance Growth Black Bay Maintenance Growth

Positive Growth ; Positive Growth

   

Figure 8. The estimated percentage of the larval cisco population consumed by
rainbow smelt in both Thunder Bay and Black Bay. Two rates of feeding
(maintenance growth: P-value = 0.2; positive growth P-value = 0.5) were used to
encompass the probable growth of rainbow smelt through the 30-day
simulations.

47

Re-assessment of earlier findings

The earlier study (Selgeby et al. 1978) and our study yielded similar
estimates of the daily consumption of cisco larvae by rainbow smelt. Selgeby et
al. (1978) calculated average consumption rates of 4.6 (1 .5% daily ration) and
7.7 (2.5% daily ration) cisco larvae per day per rainbow smelt. The bioenergetics
model gave consumption estimates of 6.0 (1 .5% daily ration) and 10.9 (2.5%
daily ration) cisco larvae per day in May 1974. Using the 1974 estimates of
rainbow smelt and larval cisco densities (368/ha and 258,000/ha, respectively)
reported by Selgeby et al. (1978), rainbow smelt were expected to consume 3.3-
55% (Selgeby et al 1978) or 5.294% (bioenergetics model) of the larval cisco .
population during a five-day period (Figure 9). Increasing the period of overlap to
30 days resulted in a proportional increase in the predation mortality (Figure 9).
Selgeby et al. (1978) acknowledged that their use of bottom trawls could have
caused conservative estimates of rainbow smelt density. The gear comparison of
this study showed that bottom trawl estimates of rainbow smelt density were
conservative compared to AC-MT estimates for Black Bay but not Thunder Bay.
For the two bays, AC-MT estimates of rainbow smelt density were on average
1.1.72 times greater than day bottom trawl estimates. Correcting rainbow smelt
density estimates by a factor of 1.72, along with a 30 day period of overlap,

resulted in predation mortality estimates approaching 100% (Figure 9).

48

5 Days

30 Days

 

 

30 Days 8.
BTR Bias

Correction

 

 

0 25 50 75 100
Predation Mortality

7“": 1 5% Body Weight Present Study 195% 3095’ Weight

Selgeby et al. _. ~
:3 2. 5% Body Weight - 25% Body Wejgm

Figure 9. Approach of Selgeby et al. (1978) compared to our bioenergetics
analysis and the sensitivity of results to assumptions of temporal overlap and

bias due to sampling gear.

Discussion

We found that predation mortality by rainbow smelt could account for 16.9-
57.8% and 37.8-100% of the mortality imposed on larval cisco populations in
Thunder and Black Bays, respectively. Our consideration of different feeding
rates and spatial scales led to the range in our estimates of predation mortality.
Fisheries ecologists are recognizing more and more that fish distributions are

patchy (Kolasa and Pickett 1991) and that the scale of an investigation influences

49

how we perceive ecological processes (Wiens 1989, Mason and Brandt 1999).
The heterogeneity associated with spatial distributions can have fundamental
effects on biological processes (Mason and Brandt 1996, Stockwell and
Johnson), interactions amongst species (Tilman and Karevia 1997, Ault et al.
1999), and how natural resources are managed (Loehle 1991 ). Despite the
concern about scaling effects, it is often still ignored in investigations of fish
growth and production, predator-prey interactions, and sustainability of fisheries
(Mason and Brandt 1999). Our study provides support for the claim that spatial
scale of an investigation may affect the outcome of the research. If we had
assumed that rainbow smelt and larval cisco were evenly distributed across the
bays, we would have concluded that predation by rainbow smelt had a
substantial effect on cisco. However, if we account for the spatial distributions of
predator and prey by dividing the bays into 1,000 ha grids, the apparent impact of
rainbow smelt decreases. It is difficult to recommend the appropriate spatial
resolution for investigating these questions because we do not fully understand
the movement of predators or prey through time. Multiple estimates of predator
abundance throughout our simulation period would have improved our
understanding of the degree of movement within a bay and the importance of
spatiotemporal distributions.

Given our spatial coverage of each bay using hydroacoustics, we believe
that density estimates based on acoustics and midwater trawls (AC-MT)
represent the rainbow smelt populations of Thunder and Black Bays more

accurately than our limited day bottom trawl sampling. Bottom trawls generally

50

provide conservative estimates of abundance, especially when the target species
are above the trawl path or evade the approaching gear (Stockwell et al 2007,
Yule et al. 2008b). Selgeby et al. (1978) only sampled rainbow smelt from Black
Bay with bottom trawl samples because 41 day midwater trawl samples in the
Apostle Islands caught zero rainbow smelt. During this study we captured
rainbow smelt in the surface waters of Black Bay during daylight hours using a
midwater trawl. This provided further evidence that day bottom trawl sampling
provides conservative estimates of rainbow smelt density. In fact, the density
estimate of rainbow smelt in Black Bay using AC-MT was six to seven times the
estimate from bottom trawls (Table 5). However, bottom trawl sampling in
Thunder Bay achieved a greater estimate of density compared to AC-MT. We
believe that, by chance, the bottom trawl samples in Thunder Bay were collected
in areas of high rainbow smelt density. The correction factor of 1.72, calculated
using data from both bays, was an objective approach for recasting the bottom
trawl derived rainbow smelt density estimates reported by Selgeby et al. (1978).
Selgeby et al. (1978) concluded that rainbow smelt predation was not
affecting cisco populations because cisco stocks sustained a commercial
operation at above average (historical) catch levels during the 19703. However,
Black Bay cisco stocks declined precipitously in the 1980s (Dextrase et al. 1986).
Over a twenty-year period, Dextrase et al. (1986) observed increases in the
length, age and growth of cisco in Black Bay, which was similar to what Selgeby
(1982) reported for cisco populations that collapsed in US. waters of Lake

Superior. Selgeby’s (1982) analysis of the Wisconsin cisco fishery led him to

51

conclude that overfishing caused cisco stocks to collapse and later Dextrase et
al. (1986) reached the same conclusions for the decline of cisco stocks in Black
Bay. We believe the increases in length, age, and growth reported by Dextrase et
al. (1986) were not the consequence of overfishing, but instead were linked to
failed recruitment. Yule et al. (2008) provided evidence suggesting that cisco can
live 20+ years and that variable recruitment is likely a natural characteristic of the
species. Thus, we propose that Black Bay cisco stocks were adversely affected
by rainbow smelt predation and that cisco stocks would have declined throughout
the 19803 even in the absence of commercial fishing, albeit likely at a slower
rate.

In support of Selgeby et al. (1978), Bronte et al. (2003) suggested that
rainbow smelt were not impeding the recruitment of cisco because declines in
rainbow smelt abundance in the late 19703 should have released cisco from this
bottleneck. However, this inference relies on the assumption that large (>150
mm) rainbow smelt were the major source of cisco predation, because the
decline in smelt densities during 1978-1981 was limited to large rainbow smelt
(Gorman 2007). Despite an absence of large-bodied rainbow smelt in our
samples, our modeling showed rainbow smelt can still have a significant effect on
larval cisco. We believe that consumption of larval cisco by rainbow smelt, as
reported by Selgeby et al (1994), may have played a pivotal role in the decline of
cisco in Black Bay.

We recognize our estimates of consumption are sensitive to the suite of

assumptions we made. For example, conservative estimates of larval cisco

52

abundance would inflate estimates of rainbow smelt predatory impact. We
assumed larvae were concentrated in the surface stratum and that two synoptic
surveys was adequate for estimating abundance. This assumption seems to be
consistent with most studies (Anderson and Smith 1971, Selgeby et al. 1978,
Hatch and Underhill 1988, Oyadomari and Auer 2004) yet it is important to
mention that vertical movement patterns and ontogeny of age-0 cisco is poorly
understood (Selgeby et al. 1978). It will be important that future larval cisco
surveys be designed using an approach for estimating abundance throughout the
water column.

Finally, we recognize that predation by rainbow smelt may not be the only
mechanism contributing to low cisco recruitment events. Bronte et al. (2003)
reported year class strength in different regions of Lake Superior was
synchronized and Yule et al. (2008) suggested sporadic recruitment may be a
natural characteristic of these populations. Together, these studies suggest
large-scale physical processes are what determine the magnitude of cisco
recruitment. Our findings suggest the added pressure of rainbow smelt predation
is sufficient to dampen the magnitude of the occasional successful year class or

eliminate annual hatches altogether.

53

REFERENCES

Anderson, E. D., and L. L. Smith Jr. 1971. Factors affecting abundance of lake
herring (Coregonus artedii Lesueur) in western Lake Superior.
Transactions of the American Fisheries Society 100:691-707.

Auer, N. A. (ed.). 1982. Identification of larval fishes of the Great Lakes basin
with emphasis on the Lake Michigan drainage. Great Lakes Fishery
Commision Special Publication 82-3. Ann Arbor, Michigan.

Ault, J. S., J. Luo, S. G. Smith, J. E. Serafy, J. D. Wang, R. Humston, and G. A.
Diaz. 1999. A spatial dynamic multistock production model. Canadian
Journal of Fisheries and Aquatic Sciences 56 (Suppl. 1):4-25.

Barkley, R. A. 1972. Selectivity of towed-net samplers. Fisheries Bulletin. 70:799-
820.

Berst, A. H., and G. R. Spangler. 1973. Lake Huron: the ecology of the fish
community and man’s effects on it. Great Lakes Fishery Commission
Technical Report 21, Ann Arbor, Michigan.

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 changes in
Lake Superior, 1970-2000. Canadian Journal of Fisheries and Aquatic
Sciences 60:1 552-1574.

Brown, E. H. Jr., T. R. Busiahn, M. L. Jones, and R. L. Argyle. 1999. Allocating
Great Lakes forage bases in response to multiple demand. In Great Lakes
Fisheries Policy and Management, a Binational Perspective, eds. W. W.
Taylor, and C. P. Ferreri, pp. 259-354. East Lansing, MI: Michigan State
University Press.

Capelli, G. M. 1982. Displacement of northern Wisconsin crayfish by Orconectes
rusticus (Girard). Limnology and Oceanography 27:741-745

Castro, L. R., and E. H. Hernandez. 2000. Early life survival of the anchoveta
Engraulis ringens off central Chile during the 1995 and 1996 winter
spawning seasons. Transactions of the American Fisheries Society
129:1107-1117

Christie, W. J. 1974. Changes in the fish species composition of the Great Lakes.
Journal of Fisheries Research Board Canada 31:827-854.

54

Claramunt, R. M., D. E. Shoup, and D. H. Wahl. 2005. Comparisons of push nets
and tow nets for sampling larval fish with implications for assessing littoral

habitat utilization. North American Journal of Fisheries Management
25:86-92.

Cox, S. P., and J. F. Kitchell. 2004. Lake Superior ecosystem, 1929-1998:
simulating alternative hypotheses for recruitment failure of lake herring
(Coregonus artedi). Bulletin of Marine Science 74:671-683.

Cushing, D. H. 1996. Towards A Science of Recruitment in Fish Populations.
Oldendorf/Luhe, Germany: Ecology Institute.

Crowder, L. B. 1980. Alewife, rainbow smelt and native fishes in Lake Michigan:
competition or predation? Environmental Biology of Fishes 5: 225-233.

DesJardine, R. L., T. K. Gorenflo, R. N. Payne, and J. D. Schrouder. 1995. Fish-
community objectives for Lake Huron. Great Lakes Fishery Commission
Special Publication 95-1, 38 pp.

Dryer, W. R., and J. Beil. 1964. Life history of lake herring in Lake Superior.
Fishery Bulletin 63:493-530.

Elliott, R. F ., and eight coauthors. 1996. Conducting diet studies of Lake
Michigan piscivores — a protocol. US. Fish and Wildlife Service, Fishery
Resources Office, Report 96-2, Green Bay, Wisconsin.

Eshenroder, R. L., M. E. Holey, T. K. Gorenflo, and R. D. Clark Jr. 1995. Fish-
community objectives for Lake Michigan. Great Lakes Fishery
Commission Special Publication 95-3, 56 pp.

Evans, D. 0., 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 (Suppl. 2):249-266.

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
(Suppl. 1):75-90.

Fitzsimons, J. D., and R. O’Gorman. 2006. Status and assessment, research and
restoration needs for lake herring in the Great Lakes. Can. Tech. Rep.
Fish. Aquat. Sci. 2638.

Fitzsimons, J.D., S. B. Brown, D. C. Honeyfield, and J. G. Hnath. 1999. A review

of early mortality syndrome in Great Lakes salmonids and its relationship
with thiamine. Ambio 2819-15.

55

Foltz, J. W., and C. R. Norden. 1977. Seasonal changes in food consumption
and energy content of smelt (Osmerus mordax) in Lake Michigan).
Transactions of the American Fisheries Society 106(3):230-234.

Hartman, K. J., and F. J. Margraf. 1992. Effects of prey and predator abundances
on prey consumption and growth of walleyes in western Lake Erie.
Transactions of the American Fisheries Society 121:245-260.

Hartman, W. L. 1973. Effects of exploitation, environmental changes, and new
species on the fish habitats and resources of Lake Erie. Great Lakes
Fishery Commission Technical Report 22, Ann Arbor, Michigan.

Hatch, J. T., and J. C. Underhill. 1988. Abundance, distribution, growth, and
mortality of larval lake herring in western Lake Superior. American
Fisheries Society Symposium. 5:96-103.

Heath, M. R. 1992. Field investigations of the early life stages of marine fish.
Advances in Marine Biology. 28:1-133.

Heist, B. G., and W. A. Swenson. 1983. Distribution and abundance of rainbow
smelt in western Lake Superior as determined from acoustic sampling.
Journal of Great Lakes Research 9(3):343-353.

Hinrichs, M. A. 1979. A description and key to the eggs and larvae of five species
of fish in the subfamily Coregoninae. M. S. thesis, University of Wisconsin,
Stevens Point, Stevens Point, WI.

Horns, W., C. R. Bronte, T. R. Busiahn, M. P. Ebener, R. L. Eshenroder, T.
Gorenfio, N. Kmieck, W. Mattes, J. W. Peck, M. Petrzold, and D. R.
Schreiner. 2003. Fish-community objectives for Lake Superior. Great
Lakes Fishery Commission Special Publication 03-01, 78 pp.

Houde, E. D. 1987. Fish early life dynamics and recruitment variability. American
Fisheries Society Symposium. 2:17-29.

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.

Jacobson, L. D., W. R. MacCallum, and G. R. Spangler. 1987. Biomass
dynamics of Lake Superior lake herring (Coregonus artedi): application of
Schnute’s difference model. Canadian Journal of Fisheries and Aquatic
Sciences 44 (Suppl. 2):275-288.

56

Johnson, S. R., and G. R. Rausser. 1971. Effects of misspecifications of log-
linear functions when sample values are zero or negative. American
Journal of Agricultural Economics 53(1):120-124.

Johnson, T. B., D. M. Mason, C. R. Bronte, and J. F. Kitchell. 1998. Estimation of
invertebrate production from patterns of fish predation in western Lake
Superior. Transactions of the American Fisheries Society 127(3):496-506.

Karjalainen, J., H. Auvinen, H. Helminen, T. J. Marjomaki, T. Niva, J. Sarvala,
and M. Viljanen. 2000. Unpredictability of fish recruitment: interannual
variation in young-of-the-year abundance. Journal of Fish Biology 56:837-
857.

Kolasa, J., and S.T.A. Pickett (Editors). 1991. Ecological heterogeneity. Springer-
Verlag, New York.

Lantry, B. F., and D. J. Stewart. 1993. Ecological energetics of rainbow smelt in
the Laurentian Great Lakes: an interlake comparison. Transactions of the
American Fisheries Society 122:951-976.

Lenth, R. V. 2006. Java Applets for Power and Sample Size [Computer
software]. Retrieved 1/05/07 from
http://www.stat.uiowa.edu/~rlenth/Power.

Levin, S. A. 1992. The problem of pattern and scale in ecology. Ecology
73(6):1943-1967.

Lodge, D. M. 2001. Lakes. Pages 277-314 in: Chapin, F.S. Ill, Sala, OE, and
Huber-Sanwald, E., editors. Global Biodiversity in a Changing
Environment: Scenarios for the 21St Century. Springer-Verlag, New York.

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. '

Loehle, C. 1991. Managing and monitoring ecosystems in the face of
heterogeneity. In Ecological heterogeneity. Edited by J. Kolasa and S.T.A
Pickett. Springer-Verlag, New York.

MacCallum, W. R., and Selgeby, J. H. 1987. Lake Superior revisited 1984.
Canadian Journal of Fisheries and Aquatic Sciences 44 (Suppl. 2):23-36.

57

Mason, D. M. and S. B. Brandt. 1996. Effects of spatial scale, capture efficiency,
and the spatial distribution of predators on the predictions made by
spatially-explicit models of fish growth rate. Environmental Biology of
Fishes 45:283-298.

Mason, D. M. and S. B. Brandt. 1999. Space, time, and scale: new perspectives
in fish ecology and management. Canadian Journal of Fisheries and
Aquatic Sciences 56 (Suppl. 1):1-3.

Mack, R. N., D. Simberioff, W. M. Lonsdale, H. Evans, M. Clout, and F. A.
Bazzaz. 2000. Biotic invasions: causes, epidemiology, global
consequences, and control. Ecological Applications 10(3):689-710.

Madenjian, C. P., R. O’Gorman, D. B. Bunnell, R. L. Argyle, E. F. Roseman, D.
M. Warner, J. D. Stockwell, and M. A. Stapanian. 2008. Adverse effects of
alewives on Laurentian Great Lakes fish communities. North American
Journal of Fisheries Management 28(1):263-282.

Mehner, T. 2006. Prediction of hydroacoustic target strength of vendace
(Coregonus albula) from concurrent trawl catches. Fisheries Research
79:162-169.

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
(Sander vitreus) recruitment in Wisconsin lakes. Canadian Journal of
Fisheries and Aquatic Sciences 64:1543-1550.

Mills, E. L., J. H. Leach, J. T. Carlton, and C. L. Secor. 1993. Exotic species in
the Great Lakes: 3 history of biotic crises and anthropogenic introductions.
Journal of Great Lakes Research 19(1):1-54.

Mittlebach, G. G. 1981. Foraging efficiency and size: a study of optimal diet and
habitat use by bluegills. Ecology 62:1370-1386.

Mooney, H. A., and R. J. Hobbs. 2000. invasive Species in a Changing World.
Island Press, Washington, DC.

Mooney, H. A., and E. E. Cleland. 2001. The evolutionary impact of invasive
species. Proceedings of the National Academy of Sciences 98(10):5446-
5451.

Mosteller, F., and J. W. Tukey. 1977. Data Analysis and Regression: A Second
Course in Statistics. Addison-Wesley, Reading, MA.

58

Myers, J. T., J. D. Stockwell, D. L. Yule, and J. A. Black. In Press. Evaluating
sampling strategies for larval cisco (Coregonus artedi). Journal of Great
Lakes Research 34(2).

Nyberg, P., E. Bergstrand, E. Degerman, and O. Enderlein. 2001. Recruitment of
pelagic fish in an unstable climate: studies in Sweden’s four largest lakes.
Ambio 30(8):559-564.

O’Gorman, R. and C. P. Schneider. 1986. Dynamics of alewives in Lake Ontario
following a mass mortality. Transactions of the American Fisheries Society
1 15:1-14.

Overton, S. A., and R. A. Rulifson. 2007. Evaluation of plankton surface pushnets
and oblique tows for comparing the catch of diadromous larval fish.
Fisheries Research 86299-104.

Oyadomari, J. K., and N. N. 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 (Supple. 1):369-384.

Oyadomari, J. K., and N. N. Auer. 2007. Influence of rearing temperature and
feeding regime on otolith increment deposition in larval ciscoes.
Transactions of the American Fisheries Society 136:766-777.

Parker Stetter, S. L., L. G. Rudstam, J. L. Stritzel, J. L., and D. L. Parish. 2006.
Hydroacoustic separation of rainbow smelt (Osmems mordax) age groups
in Lake Champlain. Fisheries Research 82:176-185.

Peterson, J. E., J. C. Cornwell, and W. M. Kemp. 1999. Implicit scaling in the
design of experimental aquatic ecosystems. Oikos 85:3-18.

Ricciardi, A. 2001. Facilitative interactions among aquatic invaders: Is an
“invasional meltdown” occurring in the Great Lakes? Canadian Journal of
Fisheries and Aquatic Sciences 58:2513-2525.

Ricciardi, A., R. J. Neves, and J. B. Rasmussen. 1998. Impending extinctions of
North American freshwater mussels (Unionoida) following the zebra
mussel (Dreissena polymorpha) invasion. Journal of Animal Ecology
67(4):613-619.

Ricciardi, A., and J. B. Rasmussen. 1999. Extinction rates of North American
freshwater fauna. Conservation Biology 13(5):1220-1222.

Ricker, W. E. 1954. Stock and recruitment. Journal of the Fisheries Research
Board of Canada 11:559-623.

59

Roth, B. M., J. C. Tetzlaff, M. L. Alexander, and J. F. Kitchell. 2007. Reciprocal
relationships between exotic rusty crawfish, macrophytes, and Lepomis
species in northern Wisconsin lakes. Ecosystems 10:74-85.

Roughgarden, J., S. Gaines, and H. Possingham. 1988. Recruitment dynamics in
complex life cycles. Science. 241:1460-1466.

Rudstam, L. G., S. L. Parker, D. W. Einhouse, L. D. Witzel, D. M. Warner, J.
L. Stritzel, D. L. Parrish, P. J. Sullivan. 2003. Application of in situ target-
strength estimations in lakes: example from rainbow-smelt surveys in
Lakes Erie and Champlain. ICES Journal of Marine Science. 60(3):500-
507.

Ryan, P. A., R. Knight, R. Macgregor, G. Towns, R. Hoopes, and W. Culligan.
2003. Fish-community objectives for Lake Erie. Great Lakes Fishery
Commission Special Publication 03-02, 56 pp.

Sala, O. E., F. S. Chapin, J. J. Arrnesto, E. Berlow, J. Bloomfield, R. Dirzo, E.
Huber-Sanwald, L. F. Huenneke, R. B. Jackson, A. Kinzig, R. Leemans,
D. M. Lodge, H. A. Mooney, M. Oesterheld, N. L. Poff, M. T. Sykes, B. H.
Walker, M. Walker, and D. H. Wall. 2000. Global biodiversity scenarios for
the year 2100. Science 287:1770—1774

Sammons, S. M., and P. W. Bettoli. 1998. Larval sampling as a fisheries
management tool: early detection of year-class strength. North American
Journal of Fisheries Management. 18:137-143.

SAS Institute Inc. 2003. Statistical analysis system. Version 9.1. SAS Institute
Inc., Cary, NC.

Sawada, K., M. Furusawa, and N. J. Williamson. 1993. Conditions for the precise
measurement of fish target strength in situ. Journal of the Marine
Acoustical Society of Japan 20:73-79.

Scott, W. B., and E. J. Crossman. 1998. Freshwater fishes of Canada. Galt
House Publications Ltd. Oakville, Ont. Canada. 966pp.

Selgeby, J. H. 1982. Decline of lake herring (Coregonus artedii) in Lake Superior:
an analysis of the Wisconsin herring fishery, 1936-78. Canadian Journal of
Fisheries and Aquatic Sciences 39:554-563.

Selgeby, J. H., W. R. MacCallum, and D. V. Swedberg. 1978. Predation by
rainbow smelt (Osmerus mordax) on lake herring (Coregonus artedi) in

western Lake Superior. Journal of Fisheries Research Board Canada
35:1457-1463.

60

Selgeby, J. H., W. R. MacCallum, and M. H. Hoff. 1994. Rainbow smelt— larval
lake herring interactions: competitors or casual acquaintances? National
Biological Survey, Biological Report 25, Washington, DC.

Smith, S. H. 1968. Species succession and fishery exploitation in the Great
Lakes. Journal of Fisheries Research Board Canada 25(4):667-693.

Stewart, T. J., R. E. Lange, S. D. Orsatti, C. P. Schneider, A. Mathers, and M. E.
Daniels. 1999. Fish-community objectives for Lake Ontario. Great Lakes
Fishery Commission Special Publication 99-1, 56 pp.

Stockwell, J. D. and B. M. Johnson. 1999. Field evaluation of a bioenergetics-
based foraging model for kokanee (Oncorhynchus nerka). Canadian
Journal of Fisheries and Aquatic Sciences 56 (Suppl. 1): 140-151.

Stockwell, J. D., D. L. Yule, L. M. Evrard, G. A. Cholwek, and O. T. Gorman.
2006a. Status and trends of prey fish populations in Lake Superior, 2006a.
Report to the Lake Superior Committee, Great Lakes Fish Comm., Ann
Arbor, MI. Available:
http://www.glsc.usgs.gov/_files/reports/2006LakeSuperiorPreyfish.pdf
(April 2008).

Stockwell, J. D., D. L. Yule, O. T. Gorman, E. J. Isaac, and S. A. Moore. 2006b.
Evaluation of bottom trawls as compared to acoustics to assess adult lake

herring (Coregonus artedi) abundance in Lake Superior. Journal of Great
Lakes Research 32:280-292.

Stockwell, J. D., D. L. Yule, T. R. Hrabik, J. V. Adams, 0. T. Gorman, and B. V.
Holbrook. 2007. Vertical distribution of fish biomass in Lake Superior:
Implications for day bottom trawl surveys. North American Journal of
Fisheries Management 27(3):735-749.

Swenson, W. A. 1978. Influence of turbidity on fish abundance in western Lake
Superior. US. Environmental Protection Agency Ecological Research
Report - 600/3-78-067. Springfield, Virginia.

Tilman, D., and Kareiva, P. (Editors). 1997. Spatial ecology: the role of space in
population dynamics and interspecific interactions. Princeton University
Press, Princeton, NJ.

Van Oosten, J. 1937. The dispersal of rainbow smelt, Osmerus mordax (Mitchill),

in the upper Great Lakes Region. Transactions of the American Fisheries
Society 66:160-171.

61

Viljanen. M., J. Karjalainen, H. Helminen, J. Sarvala, and A. Sydanoja. 1995.
Night-day catch ratios of coregonid larvae in three large lakes in Finland.
Archiv fiir Hydrobiologie, Special Issues, Advances in Limnology 46:195-
201.

Vondracek, B., B. D. Giese, and M. G. Henry. 1996. Energy density of three
fishes from Minnesota waters of Lake Superior. Journal of Great Lakes
Research 22(3):757-764.

Wells, L. 1966. Seasonal and depth distribution of larval bloaters (Coregonus
hoyi) in southeastern Lake Michigan. Transactions of the American
Fisheries Society 95:389-396.

Wells, L., and A. L. McClain. 1973. Lake Michigan: man’s effects on native fish
stocks and other biota. Great Lakes Fishery Commission Technical Report
20, Ann Arbor, Michigan.

Werner, E. E. 1974. The fish size, prey size, handling time relation in several
sunfishes and some implications. Journal of the Fisheries Research Board
of Canada 31:1531-1536.

Wiens, J. A. 1989. Spatial scaling in ecology. Functional Ecology 3:385-397.

Worthington, E. B., and R. Lowe-McConnell. 1994. African lakes reviewed:
creation and destruction of biodiversity. Environmental Conservation
21:199-213.

Yule, D. L., J. D. Stockwell, G. A. Cholwek, L. M. Evrard, S. T. Schram, M.
Seider, and M. Symbal. 2006. Evaluation of methods to estimate lake

herring spawner abundance in Lake Superior. Transactions of the
American Fisheries Society 135(3):680-694.

Yule, D. L., J. V. Adams, J. D. Stockwell, and O. T. Gorman. 2007. Using multiple
gears to assess acoustic detectability and biomass of fish species in Lake

Superior. North American Journal of Fisheries Management 27(1):106-
126.

Yule, D. L., J. D. Stockwell, J. A. Black, K. I. Cullis, G. A. Cholwek, and J. T.
Myers. 2008a. How systematic age underestimation can impede
understanding of fish population dynamics: lessons learned from a Lake

Superior cisco stock. Transactions of the American Fisheries Society 137:
481-495.

62

Yule, D. L., J. V. Adams, J. D. Stockwell, and O. T. Gorman. 2008b. Factors
affecting bottom trawl catches: Implications for monitoring the fishes of
Lake Superior. North American Journal of Fisheries Management
28(1):109-122.

63

 

 

 

 

 

 

 

SI IY LIBRARIES

GAPI STATE UNIV H

ICHI

 

0
1
4|
5
6
5
9
2
0
3
9
2
..I

3

 

 

 

 

 

 

 

 

 

 

 

 

I 1111'!