(Ulvaria subbifurcata Storer) from Conception Bay, Newfoundland
by
Kelly Victoria Young B.Sc., University of Victoria, 2003 A Thesis Submitted in Partial Fulfillment
of the Requirements for the Degree of MASTER OF SCIENCE in the Department of Biology
Kelly Victoria Young, 2008 University of Victoria
All rights reserved. This thesis may not be reproduced in whole or in part, by photocopy or other means, without the permission of the author.
Variation in prey availability and feeding success of larval Radiated Shanny (Ulvaria subbifurcata Storer) from Conception Bay, Newfoundland
by
Kelly Victoria Young B.Sc., University of Victoria, 2003
Supervisory Committee Dr. John F. Dower, Supervisor (Department of Biology)
Dr. Asit Mazumder, Departmental Member (Department of Biology)
Dr. David L. Mackas, Outside Member (School of Earth and Ocean Sciences)
Abstract
Supervisory CommitteeDr. John F. Dower, Supervisor (Department of Biology)
Dr. Asit Mazumder, Departmental Member (Department of Biology)
Dr. David L. Mackas, Outside Member (School of Earth and Ocean Sciences)
Recruitment of pelagic fish populations is believed to be regulated during the planktonic larval stage due to high rates of mortality during the early life stages.
Starvation is thought to be one of the main sources of mortality, despite the fact that there is rarely a strong correlation between the feeding success of larval fish and food
availability as measured in the field. This lack of relationship may be caused in part by (i) inadequate sampling of larval fish prey and (ii) the use of total zooplankton abundance or biomass as proxies for larval food availability. Many feeding studies rely on measures of average prey abundance which do not adequately capture the variability, or patchiness, of the prey field as experienced by larval fish. Previous studies have shown that larvae may rely on these patches to increase their feeding success.
I assess the variability in the availability of larval fish prey over a range of scales and model the small-scale distribution of prey in Conception Bay, Newfoundland. I show that the greatest variability in zooplankton abundance existed at the meter scale, and that larval fish prey were not randomly distributed within the upper mixed layer. This will impact both how well we can model the stochastic nature of larval fish cohorts, as well as how well we can study larval fish feeding from gut content analyses.
Expanding on six years of previous lab and field studies on larval Radiated
Shanny (Ulvaria subbifurcata) from Conception Bay, Newfoundland, I assess the feeding success, niche breadth (S) and weight-specific feeding rates (SPC, d-1) of the larvae to determine whether there are size-based patterns evident across the years. I found that both the amount of food in the guts and the niche breadth of larvae increased with larval
size. There was a shift from low to high SPC with increasing larval size, suggesting that foraging success increases as the larvae grow.
My results suggest that efforts should be made to estimate the variability of prey abundance at scales relevant to larval fish foraging rather than using large-scale average abundance estimates, since small-scale prey patchiness likely plays a role in larval fish feeding dynamics. In addition, the characteristics of zooplankton (density, size and behaviour) should be assessed as not all zooplankton are preyed upon equally by all sizes of larval fish. Overall, this thesis demonstrates that indices based on averages fail to account for the variability in the environment and in individual larval fish, which may be confounding the relationship between food availability and larval growth.
Table of Contents
Supervisory Committee ... ii
Abstract... iii
Table of Contents... v
List of Tables ... vii
List of Figures ... ix
Acknowledgments... xi
Chapter 1 General Introduction ... 1
1.2 Thesis Objectives and Structure... 2
1.3 Spatial variability in zooplankton abundance (Chapter 2)... 2
1.3.1 Variability in prey abundance across spatial scales... 2
1.3.2 Small-scale prey distribution ... 3
1.4 Interannual variability in larval fish feeding success (Chapter 3) ... 4
Chapter 2 Spatial variability in the prey of larval fish... 5
2.1 Introduction... 5 2.1.1 Objectives ... 6 2.2 Methods... 7 2.2.1 Site description... 7 2.2.2 Sampling protocol... 9 2.2.3 Sample processing ... 11 2.2.4 Spatial variability... 12 2.2.5 Distribution of zooplankton ... 13 2.3 Results... 15 2.3.1 Environmental conditions ... 15 2.3.2 Zooplankton abundance ... 15 2.3.3 Spatial variability... 19 2.3.4 Distribution of zooplankton ... 23 2.4 Discussion... 28 2.4.1 Zooplankton abundance ... 28 2.4.2 Spatial variability... 29 2.4.3 Distribution of zooplankton ... 33 2.4.4 Conclusions... 34
Chapter 3 Interannual variability of the feeding success of Ulvaria subbifurcata larvae in Conception Bay, Newfoundland... 36
3.1 Introduction... 36
3.1.1 Relative feeding success in the field versus the lab... 37
3.1.2 Specific consumption rates ... 39
3.1.3 Niche breadth... 39
3.1.4 Objectives ... 40
3.2.1 Sampling Methods ... 40
3.2.2 Lab Methods ... 42
3.2.3 Data Analysis... 44
3.3 Results... 45
3.3.1 Gut contents and relative feeding success ... 45
3.3.2 Specific consumption rates ... 49
3.3.3 Niche breadth... 49
3.4 Discussion... 56
3.4.1 Relative feeding success ... 56
3.4.2 Specific consumption rates ... 57
3.4.3 Niche breadth... 58
3.4.4 Conclusions... 59
Chapter 4 Conclusions and Synthesis ... 62
4.1 Thesis conclusions ... 62
4.2 Consequences of zooplankton spatial variability... 62
4.2.1 Individual based models ... 63
4.3 Larval fish feeding and growth ... 64
4.3.1 Food quality ... 65
4.4 General Conclusions ... 66
List of Tables
Table 2.1. Site description and environmental conditions during sampling. Loc = location. Start time is the time of day at the start of the first station sampled, and end time is the time of day at the start of the last station sampled in Newfoundland daylight time (NDT) for each sampling location. Temperature is the average of the upper 10m (with standard deviation) for all stations per location. Average daily wind speed and direction was estimated based on the 24 h period prior to station occupation. ... 10 Table 2.2. Comparison between net and bottle abundance estimates (L-1) of the main
groups of zooplankton collected (see text for descriptions of net and bottle sampling). “Prey” are the nauplii and copepodites (calanoid and cyclopoid) combined, which compromise 80-90% of larval fish diet; “non-prey” are adult calanoid and cyclopoid as well as harpacticoid copepods. “Total” is the total abundance of all copepodid
zooplankton in the samples... 18 Table 2.3. Mean and variance (s2) of nauplii and copepodite zooplankton counts for each station (1st center station through to 7th station sampled) at each location (P1-P9)... 20 Table 2.4. Nested ANOVA and variance component results of the abundance of
zooplankton across three scales of measurement: Location, tens of kilometres; Station, kilometres; and Error, or within station plus sampling error, meter scale... 21 Table 2.5. Mean and standard deviation of depth (m), temperature (°C) and density (σt)
during rosette sampling... 22 Table 2.6. William’s corrected G-test results comparing observed data with expected normal, negative binomial or Poisson distributions. Data were pooled for each location, and grouped so that n>5 observations are in each cell. See text for details... 25 Table 2.7. AICc weights as percent support for each model at each location; k is the number of parameters and n is the number of observations per model, for locations P1-P9. ... 26 Table 2.8. Brunt-Väisälä (BV) frequency (for 1 m depth intervals) at 10 m for each location... 31 Table 3.1. Data sets used in this study. n = number of fish stomachs examined... 38 Table 3.2. Mean (sd) weight-specific consumption rates (SPC, d-1) for size classes of
Ulvaria larvae over six years. Small, <200µgC; Medium, 200-500µgC; Large, >500µgC; Overall, all sizes... 51
Table 3.3. Regression coefficients and significance of log-mean prey size (width, µm) to log-fish size (SL, mm) relationship for all years (shown in Fig. 3.6). Normality and homogenous error assumptions were met except where noted... 53 Table 3.4. ANCOVA results comparing the regressions from Table 3.3. ... 53 Table 3.5. Regression coefficients and significance of niche breadth to larval length (mm) relationship for all years (shown in Fig. 3.7). Normality and homogenous error assumptions were met except where noted... 55 Table 3.6. ANCOVA results comparing the regressions from Table 3.5. ... 55
List of Figures
Figure 2.1. Sampling locations in Conception Bay, Newfoundland. Each location consisted of 6 sampling stations arranged in a hexagon around a central station. The 2nd and 7th stations sampled are numbered. Inset, top left: Location of Conception Bay, on the Atlantic coast of Canada. ... 8 Figure 2.2. Temperature (°C, solid line), salinity (PSU, dotted line) and fluorescence (volts, dashed line) profiles for the upper 100m at the central station from each location. Horizontal dashed line: depth of rosette sample collections (see text)... 16 Figure 2.3. Top: Density (L-1) of zooplankton as measured from (a) the vertical net tow (70 µm); and (b) rosette fired at 10 m (shown is average from all stations sampled within a location)... 17 Figure 2.4. Correlation between the coefficient of variation (CV) of zooplankton
abundance, and CV of depth, temperature, and density over the time of the rosette
sampling (~ 4 minutes). Note expanded scale in panel a... 24 Figure 2.5. Frequency distributions of number of prey per bottle, pooled for each
location (n= 42 for each location). Black bars represent the observed data, hatched is the expected distribution based on a normal distribution, grey is Poisson and white negative binomial. ... 27 Figure 2.6. Abundance of zooplankton (total count) over the firing sequence of the rosette sampler. Firing set represents the sequence of bottle groupings during firing; bottles 1-3 were triggered first (firing set 1), followed by 4-6 (set 2), 7-9 (set 3) and 10 last (set 4; bottles 11 and 12 were disregarded due to technical problems)... 32 Figure 3.1. Map of study area on the east coast of Canada (inset), showing Bonavista, Trinity and Conception Bay... 41 Figure 3.2. Gut content of Ulvaria larvae from 6 years, grouped by larval size. Small: <200 µgC; medium 200-500 µgC; large >500 µgC. ... 46 Figure 3.3. Gut contents of Ulvaria larvae collected in 1986-2005. Thick line represents maximum gut content regression from Bochdansky et al. (in press) of lab-reared larvae feeding in saturated prey levels at 14 °C. Dotted and dashed lines shows the gut content and fish size regressions for the field data. Note that in 1997 the three bays sampled have significantly different regressions... 47
Figure 3.4. Feeding success of different sizes of field caught Ulvaria relative to
maximum feeding rates as measured in the lab by Bochdansky et al. (in press). Dashed line indicates maximum feeding as scaled to lab estimate (“lab max” regression from Figure 3.3). Each box shows the median (thick bar), interquartile range (box) and
outliers (open circles; extreme outliers, stars). Number of observations per year is shown above each panel... 48 Figure 3.5. Boxplots of SPC (d-1) across size classes (small <200 µgC; medium 200-500 µgC; large >500 µgC) and years for Ulvaria larvae. Dashed horizontal line shows maintenance consumption rate as estimated by Bochdansky et al. (in press). Solid
horizontal lines indicate the estimated maximum consumption rates for each size class. 50 Figure 3.6. Regression of log-transformed mean prey size (width, µm) and length (mm) of Ulvaria larvae across six years. Regression coefficients and significance values are given in Table 3.3. ... 52 Figure 3.7. Regression of niche breadth (standard deviation of the log-transformed prey width) against larval length (mm) for Ulvaria larvae collected in 1986-2005. Regression coefficients and significance values are given in Table 3.5... 54 Figure 3.8. Correlation between niche breadth (standard deviation of the log-
Acknowledgments
I would like to thank Pierre Pepin and John Dower for their direction, support and advice throughout this thesis. Thank you for keeping me motivated and on track, and for your advice: “If this were easy it would’ve been done already”.
I would like to thank my advisory committee members, Dave Mackas and Asit Mazumder for their helpful suggestions.
Thanks are due to Tim Shears for organising and assisting during the field sampling and all the extra help once the field work was done. I am grateful for Andrea Bartsch’s help processing the 2005 fish gut contents
I would like to thank Brad Anholt and Laura Cowen for their statistical advice on Chapter 2. Thanks are also due to Pavel Kratina and Chris Lowe for their reviews and comments. I am grateful for all the people from the Dower Lab, past and present. I will miss the tea and cookies while on watch with Akash Sastri, and his thoughtful discussions and comments in the lab. I am indebted to Rana El-Sabaawi for all of her help and advice over the years. And thank you Frances Fee for the delicious cake!
I am also grateful for everyone I’ve had a chance to meet and work with while at UVic, in particular everyone from the Mazumder, Varela, Anholt and Tunnicliffe labs.
Thank you Eleanore Blaskovich for all you do behind the scenes.
And thanks to my family and friends for all of your love and support. I thank Wendy Kane and Jerika for all of your help in getting me through the stressful times. I am forever grateful to Paul Hyland for his love and support, and his patience throughout my studies.
Chapter 1
General Introduction
Recruitment success in pelagic fish populations can be defined as the number of fish that survive to a particular stage, such as the end of their first year or to
metamorphosis (Trippel and Chambers 1997). Variability in recruitment is widely held to be linked to variability in the survival probability of the larvae due to the high
cumulative mortality during the larval stage (usually at least ninety-nine percent; Trippel and Chambers 1997); however, even small changes in the mortality rate can have
dramatic impacts on the recruitment success of the population through the production of strong year classes (Houde 1987). Despite almost a century of research on this subject (Hjort 1914) the underlying mechanisms that result in “good” versus “bad” years remain unclear, limiting our ability to predict recruitment success. Linking variations in larval fish survival to any one particular cause is often quite difficult due to the confounding influences and co-variation of factors that cannot be easily separated, such as various abiotic (temperature, turbulence) and biotic factors (predation and starvation).
Previous work has linked mortality during the larval stage to three main factors: (i) advection out of favourable areas (e.g. member-vagrant hypothesis, Sinclair and Tremblay 1984), (ii) predation (e.g. bigger-is-better hypothesis, Miller et al. 1988; stage duration hypothesis, Houde 1987) and (iii) starvation (critical period hypothesis, Hjort 1914; match-mismatch hypothesis, Cushing 1974; 1990). To date, most research has focused on predation and starvation since, as Cushing (1974) so succinctly put it,
individuals need to eat or be eaten. Although factors affecting predation and starvation in the early life of fish have been well documented in the lab (e.g. Bailey and Houde 1989; Leggett and Deblois 1994), it remains unclear whether lab results are necessarily
applicable to the field (MacKenzie et al. 1990).
Recent work has suggested that the lack of a strong, coherent relationship between food availability and larval growth under field conditions may be partly due to the way in which fish larvae and their prey are sampled (MacKenzie 2000; Pepin 2004). The timing, availability and composition of food likely play a critical role in determining larval survival because prey abundance is both variable and patchy. To date, however, most field studies have sampled prey fields at spatiotemporal scales that are orders of
magnitude greater than the scales “sampled” by individual larval fish on a given day (Pepin 2004). This mismatch in scales has made it even more difficult to unravel the link between the variability in prey availability and larval growth as measured in the field (Dower et al. 2002; Baumann et al. 2003; Pepin 2004).
1.2 Thesis Objectives and Structure
The goal of this thesis is to estimate the variability of larval fish prey availability and larval fish feeding at scales relevant to individuals in order to address the mismatch in scales of previous work. The two main objectives are to (i) quantify patterns of variability in zooplankton abundance across a range of spatial scales (Chapter 2), and (ii) to describe the interannual variability in feeding success of Ulvaria subbifurcata larvae as a function of larval size (Chapter 3). This research was carried out in partnership with Dr. Pierre Pepin from the Department of Fisheries and Oceans (St. John’s) on Conception Bay, Newfoundland. Chapters 2 and 3 are formatted as manuscripts to be submitted for publication, co-authored with Dr. John F. Dower and Dr. Pierre Pepin. As such, they are not meant to be an exhaustive literature review and there may be some replication
between chapters.
1.3 Spatial variability in zooplankton abundance (Chapter 2)
In this chapter, I quantify the small-scale spatial distribution of zooplankton to investigate (i) the variability in prey abundance across a range of spatial scales; and (ii) the distribution of prey at scales relevant to larval fish foraging.
1.3.1 Variability in prey abundance across spatial scales
Zooplankton abundance is often estimated from samples collected from large volumes of water and at large spatial scales compared to gut content analysis carried out on individual larval fish (Dower et al. 2002). This assumes that large-scale zooplankton
samples are representative of the variability in zooplankton abundance experienced by larvae at much smaller scales (Pepin 2004). It has been suggested that zooplankton biomass is more variable at smaller scales and, as such, finer-scale sampling is needed to properly characterize the variability in zooplankton biomass (Mackas 1984; Pepin 2004). To estimate the variability in zooplankton abundance across a range of scales, I used a hierarchal sampling program that was designed to collect zooplankton at spatial scales spanning meters to kilometres to assess which scales were most important in describing the variability in zooplankton abundance.
1.3.2 Small-scale prey distribution
It has long been known that zooplankton are patchily distributed (Cassie 1963), and it has long been supposed that larval fish can utilize prey patches to enhance their feeding (Hunter and Thomas 1974). As such, large-scale zooplankton sampling that integrates over volumes much larger than those “sampled” by a larval fish during its daily foraging routine may confound the relationship between prey availability and larval fish feeding (Pepin 2004). Sampling zooplankton at scales relevant to larval fish is also quite difficult due to the time and effort that it takes to collect the necessary samples (Dower et al. 2002). For this thesis, over 700 zooplankton samples were collected using a rosette sampler that collected replicate, small volume samples (i.e. comparable to the volume searched daily by a foraging larva) over a two week period. The goal was not merely to demonstrate that patchiness exists, but to try and quantify the form of the patchiness. I compared the observed distribution of zooplankton at each location to normal, negative binomial and Poisson distributions using the log-likelihood G statistic. I also applied a generalized linear model framework to assess the model of best fit using Akaike’s Information Criterion (AIC). It is important to assess the degree and form of patchiness, as previous studies have shown that prey patchiness may be a significant factor
1.4 Interannual variability in larval fish feeding success (Chapter 3)
Initially, the goal of this chapter was to assess the in-situ feeding success (as measured by gut contents) of larvae collected simultaneously with the zooplankton described in Chapter 2 in order to examine the relationship with variability in
zooplankton abundance. However, the range of environmental conditions encountered, as well as the number of larvae collected, limited my ability to construct a functional response. Consequently this chapter ended up being more of an exploratory analysis of the variability in feeding rates of different size classes of radiated shanny (Ulvaria
subbifurcata) larvae from coastal Newfoundland over six years. I expand on previous lab
and field studies that have investigated larval feeding and growth by quantifying (i) the variability in feeding habits displayed by different size classes of larvae, (ii) size-based differences in niche breadth and (iii) weight-specific feeding rates, to determine whether feeding patterns differ with larval size and whether the same size-based patterns are consistent across years.
It is well known that all zooplankton are not preyed upon equally by larval fish and that larval diets shift ontogenetically (Miller et al. 1992; Pepin and Penney 1997; Scharf et al. 2000; Voss et al. 2003). Despite this, total zooplankton abundance (or total biomass) is often still used as a proxy for the availability of prey for larvae (Meekan et al. 2003). Given that size-based differences in larval feeding behaviour are know to occur, it seems likely that a prey field that constitutes good feeding conditions for small larvae may not necessarily be optimal for larger larvae and vice versa. By assessing the
variability in feeding habits across different larval size-classes, I show that differences in feeding success depends both on the size of the larva (through ontogeny) and the
availability of suitable prey. As such, measures of feeding success of larval fish will change from year to year depending on the size structure of both the larvae and their prey.
The thesis concludes with a synthesis of the main results and a brief discussion of ideas for future work in Chapter 4.
Chapter 2 Spatial variability in the prey of larval fish
2.1 IntroductionZooplankton are patchily distributed across a range of spatial scales. Although we have a reasonable understanding of the nature of this patchiness at mesoscales, few field-based studies have quantified zooplankton distributions at small scales in detail since the pioneering work of Cassie (1963) and Wiebe and Holland (1968). In general, small-scale patchiness is still largely ignored by zooplankton ecologists. This is mainly the result of (i) the logistical constraints of accurately quantifying small-scale variability and (ii) the fact that field studies are more often concerned with larger scale phenomena.
It has been hypothesized that larval fish may rely on finding dense patches of zooplankton prey because the average densities observed in the field appear insufficient to support positive larval fish growth (Houde and Schekter 1980). Feeding by larval fish depends not only on the overall abundance of prey in the environment but on factors that affect the encounter rate between an individual and its prey, such as turbulence and the distribution of prey at small scales (MacKenzie et al. 1990; Letcher and Rice 1997). As such, the spatial scales that must be sampled to quantify the effect of prey-predator interactions of larval fish should include micro- (cm-m) and fine-scale (m-100’s m) variability in addition to broader scale patterns (Jenkins 1988).
There is still no clear consensus regarding the effect of food availability on larval survival (Leggett & Deblois 1994). This is partly because zooplankton samples typically integrate across much larger volumes of water (e.g. 10-100 m3) than those searched by an individual larval fish (e.g. litres per day; Dower et al. 2002; Pepin 2004). When
integrated over large spatial scales, estimates of prey abundance are typically biased because of the smoothing of extremes, thereby further confounding any underlying relationship between prey abundance and larval feeding (MacKenzie et al. 1990; Zenitani et al. 2007). Studies investigating the relationship between food abundance and larval fish growth have also been hampered by improper sampling methods, such as using nets with a mesh size too large to adequately collect micro- and mesozooplankton (nauplii and copepodite stages of copepods), the typical prey item of larval fish (Frank 1988).
Although improvements in sampling gear may have enhanced our ability to accurately measure the abundance of larval fish prey, the continued reliance on abundance estimates averaged over scales larger than the typical larval foraging ambit may not be suitable to understand larval fish feeding (Vlymen 1977; Houde and Schekter 1980).
To circumvent the logistic constraints associated with quantifying small-scale prey distributions in the field, researchers have modelled the encounter process using individual based models (IBMs; Letcher and Rice 1997). In most instances prey are assumed be randomly distributed (i.e. a Poisson distribution) despite evidence to the contrary (Owen 1989; Power and Moser 1999; Lough and Broughton 2007). Briefly, a Poisson distribution is commonly used when animals are assumed to be randomly distributed and the population variance is equal to its mean, making the Poisson
distribution convenient to model (Cassie 1962; Fasham 1978). Given that most studies have found zooplankton to be overdispersed (i.e. variance greater than the mean), a more appropriate distribution may be the negative binomial distribution, which explicitly includes overdispersion (Cassie 1962; Pepin 1989; Lough and Broughton 2007). Pepin (1989) found that IBMs were sensitive to the number of successful encounters and the underlying pattern of prey distribution; therefore, the choice of prey field distribution is likely to affect model results (Letcher et al. 1996). The use of an unrealistic prey distribution can lead to the underestimation of encounter rates in the field, or the overestimation of the minimum prey densities required by larval fish. It is therefore important to understand how well models can predict the distributions of animals in order to determine whether the models are indeed realistic.
2.1.1 Objectives
Zooplankton are known to be patchy, however there is very little empirical information about the extent of patchiness especially across a range of scales. My objective is to quantify the small-scale distribution of larval fish prey, not merely to demonstrate patchiness, but to describe the level of patchiness across three scales of sampling in order to answer two questions. First, how is variability in prey abundance partitioned across spatial scales? Second, which statistical model best fits the observed
pattern of prey distribution at scales relevant to larval foraging? The first question is of interest when considering how best to quantify food availability for larval fish in the field, while the second is primarily of interest in trying to improve our ability to model larval fish - zooplankton interactions.
To address the first question, I used a hierarchal sampling program to collect zooplankton at scales from meters to kilometres. Patchiness can be caused by a variety of environmental factors, all of which operate over a range of scales (Mackas et al. 1985). Therefore, it is important to understand the contribution of the various scales to the observed variance, as choice of scale can influence the results (Lindegarth et al. 1995; Mason and Brandt 1996). For the second question I first compared the observed
distribution of zooplankton at each location to an expected normal, negative binomial or Poisson distribution using a log-likelihood G statistic (Sokal and Rohlf 1981; White and Bennetts 1996). Next, I applied a generalized linear model framework to investigate the underlying distribution of our zooplankton data. Generalized linear models (GLM) allow one to specify the error structure explicitly to represent variance within data (Dick 2004). Akaike’s information criterion (AIC) was used to assess which distribution best fit our data (Burnham and Anderson 2002). AIC provides a quantitative method of selecting the best model without requiring observations to be binned (as in goodness of fit tests), and is less subjective than the graphical methods commonly used. It also provides a relative measure of support for each model under consideration through the use of AIC weights (Dick 2004).
2.2 Methods
2.2.1 Site description
Sampling was conducted on Conception Bay, Newfoundland, Canada from 17 - 31 July, 2005 aboard the CCGS Shamook during daylight hours (08:00-17:00 NDT; Fig. 2.1). Conception Bay is a long (~100 km), narrow (~25 km) and deep (~300 m)
53.15
53.0
47.45
47.3
P1 P2 P3 P6 P5 P7 P8 P9 °W °N Conception Bay Atlantic Canada53.15
53.0
47.45
47.3
P1 P2 P3 P6 P5 P7 P8 P9 °W °N Conception Bay Atlantic Canada http://www.aquarius.geomar.de/omc/make_map.htmlFigure 2.1. Sampling locations in Conception Bay, Newfoundland. Each location consisted of 6 sampling stations arranged in a hexagon around a central station. The 2nd and 7th stations sampled are numbered. Inset, top left: Location of Conception Bay, on the Atlantic coast of Canada.
Currents are variable with no dominant pattern (de Young and Sanderson 1995). The closest weather station is located at St. John’s airport, 15 km to the east of Conception Bay. The zooplankton community is characterized as sub-arctic, dominated in the upper waters by the nauplii and copepodites of calanoid and cyclopoid copepods such as
Calanus finmarchicus, Pseudocalanus sp., Temora longicornis, and Oithona similis
(Davis 1982; Pepin and Penney 1997). Conception Bay supports a wide number of fish species throughout the year, with Pleuronectes ferrugineus, Ulvaria subbifurcata,
Hippoglossoides platessoides and Mallotus villosus being the four dominant species in
late July (Laprise and Pepin 1995).
2.2.2 Sampling protocol
Zooplankton and ichthyoplankton samples were collected at 9 locations (denoted P1-P9) throughout Conception Bay (Fig. 2.1). Spacing between adjacent locations was 10km. Each location consisted of six stations separated by 1km, organized hexagonally around the 7th centre station. At each location, sampling started at the centre station and proceeded in a counter-clockwise fashion around the hexagon (Fig. 2.1). A Seabird SBE25 conductivity-temperature-depth sensor (CTD) was used to collect hydrographic data throughout the full water column at the centre of each hexagon. CTD casts at the surrounding six stations only covered the top 10 m (i.e. the depth from which
zooplankton samples were collected, see below). The sequence of sites and environmental conditions on the day of sampling are detailed in Table 2.1.
Zooplankton samples were collected at each centre station using a 30 cm diameter ring net (70 µm mesh) towed vertically from 0-40 m at 1 m s-1. These zooplankton
samples (hereafter as “net zooplankton”) were preserved in 2% buffered formalin. To quantify their small-scale distribution zooplankton were also collected using a 12-bottle rosette sampler at all seven stations within each location. Each bottle collected 2.5 L of water, with the total volume sampled by the rosette comparable to that searched by a larval fish per day (Hunter 1981; Pepin 2004). Samples were collected at a depth of 10 m, the mid-range of larval fish distribution in Conception Bay (Pepin unpubl.).
Table 2.1. Site description and environmental conditions during sampling. Loc = location. Start time is the time of day at the start of the first station sampled, and end time is the time of day at the start of the last station sampled in Newfoundland daylight time (NDT) for each sampling location. Temperature is the average of the upper 10m (with standard deviation) for all stations per location. Average daily wind speed and direction was estimated based on the 24 h period prior to station occupation.
Date Loc Station Depth (m)
Start time (NDT) End time (NDT) Average temp (sd) (°C) Wind direction (deg) 24 hr wind (m s-1) 17-Jul-05 P6 17-23 95 14:09 17:50 10.81 (0.48) 261.7 3.9 21-Jul-05 P7 34-40 170 13:45 15:48 12.61 (0.30) 156.7 2.4 24-Jul-05 P9 60-66 197 9:30 11:37 13.36 (0.37) 167.5 4.0 24-Jul-05 P1 74-80 189 17:10 18:51 13.02 (0.36) 167.5 4.0 27-Jul-05 P8 90-96 193 9:19 11:03 13.27 (0.17) 242.5 5.6 27-Jul-05 P5 101-107 242 14:36 16:13 10.03 (0.22) 242.5 5.6 28-Jul-05 P2 112-118 66 8:25 10:09 12.26 (0.21) 248.3 7.0 28-Jul-05 P3 119-125 270 12:15 14:00 10.91 (0.38) 248.3 7.0
Bottles were fired as close to simultaneously as possible (~ 4 minutes to collect all 12 samples). Zooplankton from each bottle (hereafter referred to as “bottle zooplankton”) were filtered onto 70 µm mesh and preserved in 2% buffered formalin.
In all, 756 zooplankton samples were collected using the rosette sampler. Three bottles consistently misfired (bottles 5, 11 and 12) and were excluded from further analysis, as were samples from Location P4 because of sampling difficulties caused by the rosette misfiring. This resulted in a total of 560 useable samples. After processing all of the samples from two locations a power analysis revealed that processing only 6 bottles per rosette provided sufficient power to detect differences within and among locations. Therefore, only 6 randomly chosen bottles per rosette were analysed from each location (for a total of 336 samples).
2.2.3 Sample processing
Zooplankton from the vertical net tows were subsampled three times using a calibrated large-bore pipette (usually 1/500th or 1/250th) and enumerated using a Bogorov counting tray and dissecting microscope; at least 100 (usually more than 300) individuals were counted per subsample. The average of the three subsamples was then calculated. Zooplankton were identified as either calanoid, cyclopoid or harpacticoid, and staged as nauplii, copepodite or adult (nauplii were not identified to order). Differences in the total abundance among the replicates were less than 10%.
For the rosette samples, all of the zooplankton from each bottle were counted and identified in the same manner as the net zooplankton. One bottle was chosen at random from each station and re-counted; the average difference in total abundance between replicate counts was less than 5%.
A t-test was used to compare zooplankton abundance between the net and bottle samples. This was done, not to compare gear efficiencies, but to illustrate that nets can underestimate abundance due to the simple fact that they average over too large of a volume. This integrates vertical patchiness of zooplankton as compared to small, discrete samples taken at the depth that a larval fish would be found.
All further analyses focus on nauplii and copepodites stages, the main prey of larval fish in Conception Bay (Pepin and Penney 1997). Because the firing of the 12 rosette bottles was not truly simultaneous, we also tested for a ‘bottle effect’ to determine whether the firing sequence affected the amount of zooplankton collected. Pooling data across locations, a one-way ANOVA (with prey zooplankton count as dependent and bottle firing sequence as the independent factor) revealed no effect of firing sequence (p = 0.751), therefore data were pooled for each station.
2.2.4 Spatial variability
My sampling scheme was designed to quantify variability in prey availability at three spatial scales: among locations (10s km), within a location (kms), and within a rosette (<1 m). I used a nested ANOVA to investigate the contribution of each scale to the overall variance in zooplankton abundance (Lindegarth et al. 1995):
ijk i j i
ijk location station location
X =µ+ + ( ) () +ε (Eq. 2.1)
where Xijk is the abundance of zooplankton in sample k from station j at location i (total count per bottle); µ is the overall mean, locationi is the effect of the ith location,
station(location) j(i) the effect of the jth station from the ith location, plus an error term ε, which reflects the variability among bottles as well as sampling error. The three spatial scales were considered random factors. F-tests were used to investigate the extent to which each spatial scale contributes to total variation (Lindegarth et al. 1995). Data were analyzed using SPSS v. 12.0.
Environmental variability was estimated as the difference in temperature at 10 m between stations within a location. Temperature was chosen to assess environmental variability as it was more variable than salinity. Using only the downcast from the CTD, data were binned into one meter depth intervals. The variance in temperature was then calculated as the difference between the 10 m temperature values at each station within a location. To investigate whether the variance in plankton abundance within a location was related to the variability in environment, the variance in abundance was regressed with the variance in temperature at 10 m. Data were analyzed using Sigmaplot v. 8.0.
To investigate the amount of vertical movement in the rosette during sampling, the average and standard deviation of the depth, temperature, and density were calculated for the centre station of each location for the time that the rosette was sampling at 10m. To investigate if the movement in the rosette affected the variability in zooplankton abundance as estimated with the rosette, the coefficient of variation (CV) of zooplankton abundance at each location was regressed against the CV of depth, temperature and density.
2.2.5 Distribution of zooplankton
To properly model and investigate the distribution of zooplankton, abundances are reported as the number of prey items counted per bottle, not as densities (Cassie 1963). Tests of departure from randomness, or comparing the variance to the mean are only valid when counts, not densities, are used. This is due to the differences in units between the variance and the mean densities (Bez 2000).
Zooplankton counts were compared to an expected distribution (either the normal, negative binomial or Poisson) using the likelihood ratio G-test (Sokal and Rohlf 1981). Observations were pooled for each location to ensure adequate sample size and were grouped into classes. The size range of each group was chosen to ensure the number of observations per class was greater than 5, as the G-test is not accurate for small
frequencies (Sokal and Rohlf 1981). This resulted in multiple groups at each tail of the distribution being pooled to ensure n>5, decreasing the overall degrees of freedom. The expected frequencies were calculated using the probabilities of counts in each class, with parameters estimated from the data using the ‘fitdistr’ function from the MASS package (version 7.2-30) in the R statistical package (version 2.4.1; Venables and Ripley 2002; R Core Development Team 2006). The normal and negative binomial distribution had two parameters estimated from the data and the Poisson distribution had one parameter. The G-test was calculated as:
=
∑
i i a i f f f G 2 ln ˆ (Eq. 2.2)where fi is the observed and fˆi expected for a number of classes (Sokal and Rohlf 1981).
The value of G was corrected using Williams correction (q = 1+ (a2-1)/6nv, where v is the degrees of freedom) to better approximate the chi-square distribution (Sokal and Rohlf 1981). The degrees of freedom was calculated by subtracting the number of intrinsic parameters plus 1 from the number of classes (Sokal and Rohlf 1981).
Histograms were then created to compare the observed and expected frequencies based on the different distributions (Sileshi 2006).
The distribution of counts was modelled in R using the ‘glm’ function for normal and Poisson distribution (choosing the appropriate link function for each distribution), and the ‘glm.nb’ function from the MASS package to model the negative binomial distribution. Each location was modelled separately, with data from all bottles at a given station pooled in order to have enough observations for analysis (42 observations per model). The number of zooplankton per bottle was modelled as the dependent factor, with station number as the independent factor and the error assumption chosen based on the underlying error distribution (i.e. normal, negative binomial or Poisson). To choose which distribution best fit the data, we used Akaike’s Information Criterion corrected for small sample size (AICc):
1) -( 1) ( 2 2 -AICc K n K K ll − + + = (Eq. 2.3)
where ll is the complete log-likelihood of the model, extracted using the ‘logLik’ function in R, K is the number of parameters in the model, and n the number of observations (Burnham and Anderson 2002). AIC can be used to compare models with different error distributions, provided that the likelihoods used in the calculations are complete (i.e. the constant terms are not dropped; Burnham and Anderson 2002). AIC weights were then calculated as:
∑
= ∆ ∆ = R 1 r 2 1 - exp 2 1 - exp r i wi (Eq. 2.4)where ∆i = AICci – min AICc for ith model and min AICc is the smallest AICc value for
of all models. The model with the highest weight is considered the ‘best’ model of the set of candidate models (Burnham and Anderson 2002). It should be noted that the model selected may not be the true underlying distribution, merely the best of all the models under consideration. In addition, AIC performs well at discriminating between error distributions with a sufficient sample size (Dick 2004; Sileshi 2006).
2.3 Results
2.3.1 Environmental conditions
Winds were generally weak from the southeast (<5 m s-1) until 27th July at which point they shifted to the southwest and became stronger (Table 2.1). Increased wind speeds corresponded with a deepening of the thermocline in the upper 10 m for locations P5 and P8 (Fig. 2.2). Average SST was ~15 °C at the start of the sampling program, decreasing to ~10 °C by the end of the survey (Fig. 2.2). The temperature of the upper 10m increased from 10-12 °C to 13 °C in the first half of the survey, after which the average temperature in the upper 10 m decreased once more to 10-11 °C corresponding to the strengthening southwest winds. Surface salinity was generally ~31.5 PSU and
increased with depth at all stations (32-33 PSU; Fig. 2.2).
2.3.2 Zooplankton abundance
Patterns of total zooplankton abundance differed between the ring net and Niskin bottles. The lowest and highest abundances from the ring net samples were 10.5 L-1 and 33 L-1 at Locations P2 and P3, whereas the lowest and highest values from the Niskin bottles were 69 L-1 and 96 L-1 at Locations P6 and P9, respectively (Fig. 2.3). Nauplii were the dominant stage in all samples, comprising almost 80% of the Niskin bottle samples and 60% of the ring net samples. Cyclopoid copepodites were the next most abundant, accounting for 10 and 25 % of the catch for bottles and nets, respectively. The amount of calanoid copepods (juveniles and adults) in the ring net was not significantly different than that found in the Niskin bottles (net mean (sd) = 0.14 (0.06) L-1; bottle 0.34 (0.34) L-1; p = 0.118; Table 2.2).
P7-Jul21 0.0 0.5 1.0 1.5 2.0 2.5 P1-Jul24 0.0 0.5 1.0 1.5 2.0 2.5 P6-Jul17 Depth (m) 0 20 40 60 80 100 Fluorescence 0.0 0.5 1.0 1.5 2.0 2.5 P9-Jul24 0.0 0.5 1.0 1.5 2.0 2.5 P8-Jul27 Temperature (deg C) 0 5 10 15 20 0 20 40 60 80 100 Salinity (PPT) 30 31 32 33 34 P3-Jul28 0 5 10 15 20 30 31 32 33 34 P2-Jul28 0 5 10 15 20 30 31 32 33 34 temperature salinity florescence P5-Jul27 0 5 10 15 20 30 31 32 33 34
Figure 2.2. Temperature (°C, solid line), salinity (PSU, dotted line) and fluorescence (volts, dashed line) profiles for the upper 100m at the central station from each location. Horizontal dashed line: depth of rosette sample collections (see text).
Location P1 P2 P3 P5 P6 P7 P8 P9 A bun dance (L -1) 0 20 40 60 80 100 120 nauplii harp.copepodite harpacticoid cy-copepodite cyclopoid cal-copepodite calanoid Location P1 P2 P3 P5 P6 P7 P8 P9 a b
Figure 2.3. Top: Density (L-1) of zooplankton as measured from (a) the vertical net tow (70 µm); and (b) rosette fired at 10 m (shown is average from all stations sampled within a location).
Table 2.2. Comparison between net and bottle abundance estimates (L-1) of the main groups of zooplankton collected (see text for descriptions of net and bottle sampling). “Prey” are the nauplii and copepodites (calanoid and cyclopoid) combined, which compromise 80-90% of larval fish diet; “non-prey” are adult calanoid and cyclopoid as well as harpacticoid copepods. “Total” is the total abundance of all copepodid zooplankton in the samples.
Net Bottle t p-value
Calanoid 0.14 (0.06) 0.34 (0.34) 1.67 0.118 Cal. Copepodite 1.18 (0.31) 1.71 (1.18) 1.24 0.235 Cyclopoid 2.18 (1.14) 6.25(3.18) 3.41 0.004 Cy. Copepodite 5.64 (2.69) 10.89 (4.44) 2.86 0.013 Nauplii 14.45 (6.65) 58.88 (12.53) 8.86 <0.001 Prey 21.25 (9.39) 71.48 (14.02) 8.42 <0.001 Non-prey 2.97 (1.64) 10.78 (5.60) 3.78 0.002 Total 24.20 (10.80) 82.30 (16.83) 8.21 <0.001
Overall, the bottles collected significantly more zooplankton than the nets; in particular the bottles collected almost four times more prey zooplankton (nauplii and copepodites) than the nets (net = 21.25 (9.39) L-1; bottle = 71.48 (14.02) L-1; Table 2.2).
2.3.3 Spatial variability
The abundance of zooplankton was highly variable within stations as well between stations and locations (Table 2.3). Five of the 56 stations had a variance
estimate less than the mean (e.g. P8, station 6, mean =199.8, s2 = 57.0). At the majority of stations, the variance exceeded the mean, in some cases by up to an order of magnitude difference (Table 2.3).
The nested ANOVA showed significant variability at all three spatial scales (Table 2.4). There were significant differences in the number of prey zooplankton (i) within a station, as well as (ii) between stations within a location and (iii) between locations throughout the bay (p < 0.001; Table 2.4). Variance component analysis indicated that differences among locations explained 30% of the variability, among stations 19%, and differences among samples within a station explained the most
variability with 51%. Unfortunately it is not possible to separate sampling error from the variability among samples; however, replicate error was typically less than 5% and so likely has a small influence on the variability between bottles. Levene’s test for homogeneity of variance and Shapiro-Wilk test of normality confirms that the
assumptions of the nested ANOVA were met (Levene’s test p = 0.078; SW p = 0.303). There was no correlation between the variability (or CV) in zooplankton abundance and the variability in temperature at 10 m (r2 = 0.005, p = 0.957).
There was slight movement in the rosette over the time it took to fire all of the bottles, as shown by the deviation in depth during that period (Table 2.5). Location P7 showed the most movement (mean (sd) depth – 9.65 (0.27) m), and P5 had the least (9.98 (0.07) m; Table 2.5). Overall, the amount of movement was minimal, as most locations had deviations in depth of less than 0.10 m.
Table 2.3. Mean and variance (s2) of nauplii and copepodite zooplankton counts for each station (1st center station through to 7th station sampled) at each location (P1-P9).
P1 P2 P3 P5
station mean s2 mean s2 mean s2 mean s2
1 134.8 206.6 180.0 1987.6 200.3 151.9 165.8 1981.4 2 195.5 175.5 217.8 1265.4 182.2 207.0 166.3 1022.3 3 177.5 332.3 211.5 769.1 195.0 512.0 189.2 266.2 4 173.3 301.5 193.0 407.6 165.2 945.0 215.5 961.9 5 211.3 1383.5 205.2 642.6 176.8 335.0 180.0 177.2 6 168.5 1465.1 186.2 1579.0 199.5 879.1 154.5 205.1 7 153.0 307.2 181.3 245.9 177.7 231.1 194.8 461.4 mean 173.4 196.4 185.2 180.9 s2 1075.3 1042.5 553.8 998.9 P6 P7 P8 P9
station mean s2 mean s2 mean s2 mean s2
1 162.7 1776.3 141.0 304.4 206.0 382.4 238.8 553.8 2 153.0 1310.0 197.5 798.7 223.7 521.5 194.7 2745.5 3 162.8 336.6 159.7 1669.1 220.0 1536.4 252.7 2188.7 4 170.0 1328.4 170.8 657.8 206.0 749.2 245.0 726.4 5 155.2 2272.2 158.8 1555.4 168.0 1662.0 266.2 707.0 6 153.8 749.4 213.3 1584.7 199.8 57.0 272.2 449.0 7 171.3 869.9 225.0 140.4 201.3 357.1 209.8 643.8 mean 161.3 180.9 203.5 239.9 s2 1104.1 1682.3 930.4 1690.2
Table 2.4. Nested ANOVA and variance component results of the abundance of zooplankton across three scales of measurement: Location, tens of kilometres; Station, kilometres; and Error, or within station plus sampling error, meter scale
Source df MS F p-value component Variance Variance % Location 7 24027.13 505.87 <0.001 506.601 30.2
Station 48 2749.87 8.74 <0.001 315.343 18.8 Error 280 857.81 3.21 <0.001 857.812 51.1 Levene’s test homogeneous var p = 0.078 SW test for normality p = 0.303
Table 2.5. Mean and standard deviation of depth (m), temperature (°C) and density (σt)
during rosette sampling.
Location Depth (m) Temperature (°C) Density (σt) P1 10.31 (0.09) 8.42 (0.06) 25.03 (1.14) P2 10.47 (0.08) 9.20 (0.56) 24.75 (0.73) P3 10.21 (0.10) 9.59 (0.23) 24.53 (0.25) P5 9.98 (0.07) 9.32 (0.07) 24.62 (0.29) P6 9.30 (0.09) 9.15 (0.43) 24.57 (0.07) P7 9.65 (0.27) 9.82 (0.15) 24.54 (0.40) P8 10.10 (0.10) 13.09 (0.03) 23.71 (0.06) P9 10.23 (0.15) 11.26 (0.08) 24.12 (0.10)
The standard deviation in temperature ranged from 0.03 °C at P8 to 0.56 °C at P2 and the range in deviation of density was 0.06 at P8 to 1.14 at P1 (Table 2.5). There was no significant relationship between the CV of zooplankton abundance and depth (linear regression r2– 0.344, p = 0.166; Fig. 2.4). There was also no relationship between the CV of zooplankton abundance and temperature or density (Fig 2.4).
2.3.4 Distribution of zooplankton
The observed counts for all locations were not significantly different from the normal or negative binomial distributions, but were significantly different than the
Poisson (Table 2.6). When comparing the data to the Poisson distribution, it was obvious that the Poisson distribution was a poor fit. As a result of the narrow distribution of the expected Poisson values, the tails of the observed data often had to be pooled quite a bit resulting in low degrees of freedom. The Poisson model also provided a poor fit to the data at all locations based on AICc weights (Table 2.7). In general, the Poisson
distribution underestimated the variability in the data (Fig. 2.5). In contrast, the negative binomial and the normal distribution provided better fits to the data. Overall the normal distribution received over 90% of the support among the three models (Fig 2.5, Table 2.7) except at Location P1 where the negative binomial model received 55% of the support.
CV of depth 0.005 0.010 0.015 0.020 0.025 0.030 CV of zoo p la nk to n a bund an ce 0.12 0.14 0.16 0.18 0.20 0.22 0.24 CV of temperature 0.000 0.015 0.030 0.045 0.060 CV of density 0.000 0.015 0.030 0.045 0.060 a b c
Figure 2.4. Correlation between the coefficient of variation (CV) of zooplankton abundance, and CV of depth, temperature, and density over the time of the rosette sampling (~ 4 minutes). Note expanded scale in panel a.
Table 2.6. William’s corrected G-test results comparing observed data with expected normal, negative binomial or Poisson distributions. Data were pooled for each location, and grouped so that n>5 observations are in each cell. See text for details.
Normal Negative binomial Poisson
Location G-stat p-value G-stat p-value G-stat p-value
P1 1.838 0.607 1.790 0.617 39.320 <0.001 P2 2.704 0.440 3.725 0.293 14.730 0.001 P3 2.512 0.473 3.066 0.382 10.329 0.006 P5 2.050 0.727 2.893 0.576 23.507 <0.001 P6 4.622 0.202 4.646 0.200 13.730 0.001 P7 2.498 0.476 1.744 0.627 19.062 <0.001 P8 4.898 0.179 5.214 0.157 9.099 0.011 P9 2.663 0.447 4.021 0.259 26.545 <0.001
Table 2.7. AICc weights as percent support for each model at each location; k is the number of parameters and n is the number of observations per model, for locations P1-P9. k n P1 P2 P3 P5 P6 P7 P8 P9 Normal 8 42 45.3 98.9 92.9 96.5 83.2 92.4 95.3 98.7 Poisson 7 42 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Negative Binomial 8 42 54.7 1.1 7.1 3.5 16.8 7.6 4.7 1.3
P1 50 75 100 125 150 175 200 225 250 275 0 5 10 15 20 25 30 35 observed normal poisson negative binomial P2 50 75 100 125 150 175 200 225 250 275 P3 50 75 100 125 150 175 200 225 250 275 frequenc y 0 5 10 15 20 25 30 35 P5 50 75 100 125 150 175 200 225 250 275 P7 50 75 100 125 150 175 200 225 250 275 P6 50 75 100 125 150 175 200 225 250 275 0 5 10 15 20 25 30 35 P9 count 50 75 100 125 150 175 200 225 250 275 300 325 P8 count 50 75 100 125 150 175 200 225 250 275 0 5 10 15 20 25 30 35 frequen cy
Figure 2.5. Frequency distributions of number of prey per bottle, pooled for each location (n= 42 for each location). Black bars represent the observed data, hatched is the expected distribution based on a normal distribution, grey is Poisson and white negative binomial.
2.4 Discussion
2.4.1 Zooplankton abundance
Although convenient to calculate, average abundance estimates that fail to consider the spatial structure of the zooplankton community may be of little value in individual-scale analyses such as larval fish feeding dynamics (Vlymen 1977). If plankton are randomly distributed, the accuracy of any given estimate of a population depends only on the total sample count regardless of the number or location of samples; however, this is not true if the population is overdispersed (Cassie 1963). For a given number of organisms counted from a patchy distribution, the variance of the total
population will increase with greater sample volume since the likelihood of encountering a patch increases. Thus, a large number of smaller samples provide a more accurate estimate of mean abundance than a small number of large samples (Cassie 1963). Averaging over large scales can artificially create areas of high abundance depending on whether a patch happens to be sampled, leading to the erroneous conclusion of a good foraging area (Mason and Brandt 1996). Alternatively, larger-scale sampling can smooth out extremes and underestimate the actual abundance experienced by a larval fish
(Vlymen 1977). We found that even a 70 µm mesh net, commonly used for sampling nauplii as an estimate of larval fish prey availability, underestimated the overall
abundance of zooplankton, and under-sampled the smaller zooplankton when compared to the results from our Niskin bottle samples. Most larval fish prey are concentrated within the upper ten metres in Conception Bay (60-80%; Davis 1982; Young et al.
unpubl.) and averaging over the depth range of a vertical net cast leads to an
underestimate of the abundance of prey in the upper mixed layer. Stratified net samples that sample the appropriate depth and volume where zooplankton are concentrated would provide a better estimate than larger-scale integrated vertical net tows. The scale of observations is important as well, as patchiness is also scale-dependent; the larger the difference of scales between levels, the greater the possible range of scales of patchiness that will be interposed (Kotliar & Wiens 1990; Morrisey et al. 1992; McGeoch & Gaston 2002). If larvae are indeed utilizing prey patches in the field, they may be encountering
more prey than that estimated from sampling methods which integrate over scales much larger than those experienced by a larval fish.
2.4.2 Spatial variability
Zooplankton are commonly assumed to be randomly distributed at scales below 1m3 (Gerritsen and Strickler 1977), despite previous work that has shown patchiness at even smaller scales (Owen 1989; Currie et al. 1998; McManus et al. 2003; Donaghay 2004). The abundance of zooplankton in Conception Bay shows high spatial variability across a range of scales with the most variance (over fifty percent) occurring between replicates, indicating that considerable patchiness exists at scales smaller than a meter (Lindegarth et al. 1995). The system is more uniform at the larger measured scales, for example thirty percent of the variability in abundance was measured between locations throughout the bay. This emphasizes the need to scale down how larval fish prey are sampled, as large-scale sampling programs are likely missing the majority of variability in prey abundance experienced by feeding larvae. One might assume that the variability is the result of differences in environmental conditions between stations within a location; however, there was no relationship between the variance observed in plankton abundance and the variance in temperature between stations within a location. Additionally, any vertical variability in zooplankton may influence these results, especially if the gear moves substantially during sampling. However, the variability in the CTD measurements were not large and there was no significant relationship between the variability in
zooplankton abundance and variability in temperature during sampling.
Another factor that must be considered is the possibility that our results may be biased by vertical movements induced by internal wave activity. If strong vertical gradients in zooplankton abundance exist, vertical movement by internal waves at timescales comparable to our sampling procedure (i.e. 4 minutes) could increase the apparent small-scale variability in zooplankton abundance (Rinke et al. 2007; McManus et al. 2005). To address this, I compared the Brunt-Väisälä (BV) frequency from CTD casts at each location with the time taken to fire all of the rosette bottles. The bottles were fired in groups of three with one minute between each successive set (four minutes
in total). Thus, in cases where the BV frequency is less than four minutes there is the potential for internal waves to bias our results (Mackas and Owen 1982). At all
locations, the BV frequency was less than four minutes (Table 2.8). However, examining the pattern of differences in abundance across the groups of bottles in the firing sequence, no consistently clear effect is evident (Fig. 2.6). For example, there was a significant difference in zooplankton abundance between the bottle groups at P8, which had one of the longest BV frequencies (135 seconds), and yet there was no significant difference between the bottle groups at P2 and P7 even though those locations had the highest BV frequency (Fig. 2.2; Table 2.8). Overall, there were no significant differences in
abundance between the bottle groups at 5 of the 8 locations (Fig. 2.6; ANOVA with total count as dependent and firing group as factor, p > 0.05). Overall, given that the BV frequency was always less than the time taken to fire all of the rosette bottles it is still likely that internal waves may have played a role. However, the effect is likely small as we could not detect a consistently significant difference in abundance between
successively-fired bottles.
In addition to potential effects of internal wave activity, there is also the
possibility that the zooplankton may have reacted to the presence of the rosette itself and the vibrations caused by the triggering of the bottles. At most locations the last rosette bottles fired captured less zooplankton than the rest of the bottles (Fig. 2.6). Four minutes may have been sufficiently long for some of the zooplankton within the last bottles to escape. However, whereas the first three bottle groups had three bottles each, the final bottle group had only one usable replicate as bottles 11 and 12 were discarded due to technical problems. If bottle 10 is disregarded and only the first nine bottles in the rosette are considered, there is no significant difference in abundance due to firing
sequence at most locations. This pattern doesn’t hold for P9, where the bottles fired last have more zooplankton than the bottles fired first (Fig. 2.6).
Table 2.8. Brunt-Väisälä (BV) frequency (for 1 m depth intervals) at 10 m for each location. P1 P2 P3 P5 P6 P7 P8 P9 BV (s-1) 0.024 0.025 0.018 0.007 0.041 0.032 0.007 0.021 Time (seconds) 42 39 55 145 24 31 135 47
P1 P2 P3 P5 P6 P7 P8 P9 Location 0 50 100 150 200 250 300 350 Total Count Firing Set 1.00 2.00 3.00 4.00
Figure 2.6. Abundance of zooplankton (total count) over the firing sequence of the rosette sampler. Firing set represents the sequence of bottle groupings during firing; bottles 1-3 were triggered first (firing set 1), followed by 4-6 (set 2), 7-9 (set 3) and 10 last (set 4; bottles 11 and 12 were disregarded due to technical problems).
2.4.3 Distribution of zooplankton
According to the goodness of fit G-test, the data were not significantly different from either the normal or negative binomial distributions. The major drawback to goodness of fit tests is that although the data may not be significantly different from a number of distributions, it does not indicate which is best (however, it will show which distribution is not a good fit). AIC is useful in that it can rank the distributions under consideration and indicate which model was best, given the data in hand (Dick 2004). According to the AIC results, the normal distribution best described the distribution of zooplankton in the upper mixed layer of Conception Bay, and the Poisson distribution received no support at any location. The negative binomial distribution received some support, indicating evidence of patchiness, although not to a high degree. Variances were often much higher than the means (an order of magnitude higher in some cases);
however, the overall mean abundance was also quite high. The normal, Poisson and negative binomial are all descriptions of distributions along a continuum and as such, the negative binomial distribution approaches normality at high mean values (McGeoch and Gaston 2002; Lough and Broughton 2007).
Because of the heterogeneous nature of zooplankton populations, the Poisson assumption used in most foraging models may be invalid since (i) it does not accurately model the variable nature of the environment and (ii) it can lead to an underestimation of standard errors (and potentially incorrect conclusions) if the variance is ignored (Sileshi 2006). Using a stochastic simulation model of larval fish feeding, Pepin (1989) found that switching to a negative binomial distribution reduced the sensitivity of larval feeding success to variations in food abundance since, as patchiness increased, the range of prey concentrations over which the probability that a larva will starve also increased. This causes an apparent decoupling between the estimated average abundance of zooplankton and the feeding success of the larval fish. For example, at low patch intensity for a given concentration of food, the prey field experienced by a foraging larva is similar to the average abundance of food, resulting in a positive relationship between the measured average prey density and feeding success. However, as patchiness increases, the “average” concentration becomes a poor predictor of larval feeding success since only
those individuals that happen upon a patch can feed, confounding any relationship between average food availability and overall feeding success. Thus, even if the total amount of food remains constant, its statistical distribution will affect how well larval fish feed and grow (Pepin 1989).
2.4.4 Conclusions
A developing concern is whether we are sampling the environment in a way that accurately describes the environment as experienced by individual larval fish (Pepin 2004). For instance, it is known that the way in which we sample the environment (e.g., the size of the sampling unit, the number of samples taken, the total area sampled) affects how well our observations will estimate the underlying distributional characteristics of the population (Wiebe and Holland 1968; McGeoch and Gaston 2002). However, conventional sampling techniques often collect samples with volumes several orders of magnitude too large to properly characterize the variability of the small-scale
environment, leading to prey-field estimates that are not representative of what a larval fish encounters when feeding (Pepin 2004). Regional scale surveys may well be the most effective design for population-level studies but almost certainly miss important variation at small scales, which may be key to understanding individual predator-prey interactions (Pepin 2004; Lough and Broughton 2007). As the large-scale dynamics of larval fish populations become better understood, the focus is shifting to individual scale dynamics; even so, most sampling strategies have not scaled appropriately to sample at the
individual, rather than the population level (Dower et al. 2002; Pepin 2004). This will impact both how well we can model the stochastic nature of larval fish cohorts, as well as how well we can study larval fish feeding from gut content analyses.
Despite almost 90 years of research, there is still no clear consensus regarding the role of food availability on larval fish feeding and growth. Models can only go so far in determining the important factors governing larval fish growth and survival without being validated by field studies. Unfortunately the feeding dynamics of larval fish are difficult to investigate in the field because of the complexity of the behavioural and physical factors that may influence predator-prey interactions. Because of this, measures
of average prey abundance may be independent of the feeding and growth of individuals in the population. Since patchiness arises from various environmental and behavioural factors it is difficult to apply a ‘one size fits all’ sampling approach, as it is not possible to sample every species at all spatial scales with any one sampling protocol (Mackas et al. 1985; McGeoch and Gaston 2002). This study has shown that the way in which we sample the environment, from the type of gear that is used to the scale at which we sample, will affect how accurately we can quantify larval fish feeding environments. Small-scale prey patchiness likely plays a key role in larval fish feeding dynamics by creating variability among individuals (Letcher and Rice 1997). Further studies that investigate patchiness at the appropriate scale will be needed before the debate over the role of food availability on larval growth and survival is settled.
