<%BANNER%>

Fish Compensatory Responses Following Whole-Lake Experimental Density Reduction

Permanent Link: http://ufdc.ufl.edu/UFE0024914/00001

Material Information

Title: Fish Compensatory Responses Following Whole-Lake Experimental Density Reduction
Physical Description: 1 online resource (108 p.)
Language: english
Creator: Catalano, Matthew
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: Forest Resources and Conservation -- Dissertations, Academic -- UF
Genre: Fisheries and Aquatic Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: I evaluated fish compensatory responses following a whole-lake experimental density reduction of gizzard shad at Lake Dora, Florida. Gizzard shad were removed in 2005 and 2006 via a subsidized commercial fishery by the St. Johns River Water Management District. Approximately 30% of the total gizzard shad biomass and 70% of the total spawner biomass was removed. My first objective was to develop an age and length-structured population model that could be used to assess recruitment responses following density reduction. The model provided unbiased estimates of mortality, growth, gear selectivity and recruitment for long-lived species. For short-lived species there was a small upward bias in early recruitment estimates when the time series length was half of the maximum age, but all other scenarios provided unbiased estimates. My second objective was to evaluate compensatory responses of gizzard shad growth, maturity, and pre-recruit survival following density reduction. I found no evidence for density dependence in growth and maturity. Pre-recruit survival was negatively related to spawner biomass suggesting density dependence in early life survival. The strength of density dependence in pre-recruit survival was quantified by estimating the maximum lifetime reproductive rate, which was 7.3 (95% confidence interval: 1.9 ? 16.5) at Lake Dora. Thus, pre-recruit survival increases 7.3-fold at very low spawner abundance when compared to the unfished condition at Lake Dora. My third objective was to assess the relative efficacy of different gill net mesh sizes, exploitation rate, and harvest interval (number of years between removals) on total biomass reduction and spawning potential ratio of gizzard shad following removals at hypereutrophic Florida lakes using a simulation model. Specifically, I was interested in whether various combinations of these factors could achieve a 75% reduction in total population biomass and Spawning potential ratio (SPR). These targets were obtained from the literature and represent the reductions needed to achieve changes in phytoplankton biomass and to effectively reduce fish recruitment following biomanipulation. Gizzard shad biomass reduction failed to meet the target of 75% reduction for all mesh sizes and harvest intervals except for the smallest (51-mm) mesh size. Reductions in spawning potential ratio exceeded total biomass reductions but remained above a 75% reduction target except when using the smallest mesh size (51 mm). I conclude that gill net removals are unlikely to result in substantial biomass reductions at Lake Dora unless removals adopt very small mesh sizes, which may not be preferred by commercial fishers.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Matthew Catalano.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Allen, Micheal S.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024914:00001

Permanent Link: http://ufdc.ufl.edu/UFE0024914/00001

Material Information

Title: Fish Compensatory Responses Following Whole-Lake Experimental Density Reduction
Physical Description: 1 online resource (108 p.)
Language: english
Creator: Catalano, Matthew
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: Forest Resources and Conservation -- Dissertations, Academic -- UF
Genre: Fisheries and Aquatic Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: I evaluated fish compensatory responses following a whole-lake experimental density reduction of gizzard shad at Lake Dora, Florida. Gizzard shad were removed in 2005 and 2006 via a subsidized commercial fishery by the St. Johns River Water Management District. Approximately 30% of the total gizzard shad biomass and 70% of the total spawner biomass was removed. My first objective was to develop an age and length-structured population model that could be used to assess recruitment responses following density reduction. The model provided unbiased estimates of mortality, growth, gear selectivity and recruitment for long-lived species. For short-lived species there was a small upward bias in early recruitment estimates when the time series length was half of the maximum age, but all other scenarios provided unbiased estimates. My second objective was to evaluate compensatory responses of gizzard shad growth, maturity, and pre-recruit survival following density reduction. I found no evidence for density dependence in growth and maturity. Pre-recruit survival was negatively related to spawner biomass suggesting density dependence in early life survival. The strength of density dependence in pre-recruit survival was quantified by estimating the maximum lifetime reproductive rate, which was 7.3 (95% confidence interval: 1.9 ? 16.5) at Lake Dora. Thus, pre-recruit survival increases 7.3-fold at very low spawner abundance when compared to the unfished condition at Lake Dora. My third objective was to assess the relative efficacy of different gill net mesh sizes, exploitation rate, and harvest interval (number of years between removals) on total biomass reduction and spawning potential ratio of gizzard shad following removals at hypereutrophic Florida lakes using a simulation model. Specifically, I was interested in whether various combinations of these factors could achieve a 75% reduction in total population biomass and Spawning potential ratio (SPR). These targets were obtained from the literature and represent the reductions needed to achieve changes in phytoplankton biomass and to effectively reduce fish recruitment following biomanipulation. Gizzard shad biomass reduction failed to meet the target of 75% reduction for all mesh sizes and harvest intervals except for the smallest (51-mm) mesh size. Reductions in spawning potential ratio exceeded total biomass reductions but remained above a 75% reduction target except when using the smallest mesh size (51 mm). I conclude that gill net removals are unlikely to result in substantial biomass reductions at Lake Dora unless removals adopt very small mesh sizes, which may not be preferred by commercial fishers.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Matthew Catalano.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Allen, Micheal S.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024914:00001


This item has the following downloads:


Full Text

PAGE 1

1 FISH COMPENSATORY RESPONSES FOLL OWING WHOLE-LAKE EXPERIMENTAL DENSITY REDUCTION By MATTHEW JEROME CATALANO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARITAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2009

PAGE 2

2 2009 Matthew Jerome Catalano

PAGE 3

3 ACKNOWLEDGMENTS My research was funded by the St. Johns Ri ver Water Managem ent District (SJRWMD) and I was supported by a University of Florida (UF) College of Agricultural and Life Sciences Alumni Fellowship for four years. Travel funds for professional conferences were provided by the UF Institute of Food and Agricultural Scienc es, Graduate Student Co uncil, Florida Chapter of the American Fisheries Soci ety, and the Office of Research. Other funds were provided by the Roger Rottman Scholarship. I thank my advisor Micheal Allen who provide d me with every opportunity to succeed. He gave me great advice, plenty of funds, lots of computing resources and friendship. There were many people who helped me in the lab and field: Brandon Baker, Christian Barrientos, Mo Bennett, Greg Binion, Aaron Bunch, Meredith B unch, Troy Davis, Loreto DeBrabandere, Jason Dotson, Drew Dutterer, Dan Gwinn, Porter Ha ll, Galen Kaufman, Va ughn Maceina, Patrick ORouke Vince Politano, Mark Rogers, Nick Seipker, Erika Thompson, and Allison Watts; I could not have finished without their efforts. My graduate committee Karl Havens, Franklin Percival, Bill Pine, and Carl Walters helped with development of my project and provided important advice and criticism. Finally, I thank my wife Miri am Wyman and my family for their support and love. My parents Jerry Catalano and Emily Catalano, brothe r Joe Catalano, and sister Jane Catalano gave me the best upbringing I could have asked for.

PAGE 4

4 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................3LIST OF TABLES................................................................................................................. ..........6LIST OF FIGURES.........................................................................................................................7ABSTRACT.....................................................................................................................................9 CHAP TER 1 GENERAL INTRODUCTION.............................................................................................. 11Introduction................................................................................................................... ..........11Compensation During Early Life............................................................................................ 13Evidence from Adult Life Stages........................................................................................... 14Study Objectives.....................................................................................................................162 A SIZEAND AGE-STRUCTURED MODEL TO ESTIM ATE FISH RECRUITMENT, GROWTH, MORTALITY, AND GEAR SELECTIVITY...................... 19Introduction................................................................................................................... ..........19Methods..................................................................................................................................20Results.....................................................................................................................................27Discussion...............................................................................................................................303 DOES INCREASED PRE-RECRUIT SU RVIVAL DRIVE FISH DENS ITY DEPENDENCE?: EVIDENCE FROM A WHOLE-LAKE EXPERIMENTAL DENSITY REDUCTION....................................................................................................... 46Introduction................................................................................................................... ..........46Study Site..................................................................................................................... ...........49Methods..................................................................................................................................50Results.....................................................................................................................................58Discussion...............................................................................................................................624 EXPLORING FISH REMOVAL STRATE GIES FOR BIOM ANIPULATION THAT ACCOUNT FOR UNCERTAINTY IN THE STRENGTH OF DENSITY DEPENDENCE OF TARGET SPECIES...............................................................................74Introduction................................................................................................................... ..........74Study Site..................................................................................................................... ...........75Methods..................................................................................................................................76Results.....................................................................................................................................83Discussion...............................................................................................................................84

PAGE 5

5 5 SYNTHESIS AND FUTURE RESEARCH........................................................................... 96LIST OF REFERENCES...............................................................................................................99BIOGRAPHICAL SKETCH.......................................................................................................108

PAGE 6

6 LIST OF TABLES Table page 2-1 Point estimates and 95% confidence interv als for model param eters estimated from length-age data for gizzard shad at Lake Dora, Florida, USA........................................... 36 3-1 Parameter estimates (95% confidence intervals) from the length age model for lakes Dora, Eustis and Harris. ..................................................................................................... 67 3-2 Delta Akaik es Incormation Criterion (AIC ) values for competing models describing associations between growth increments a nd age, lake, and population density (i.e., total population biomass)................................................................................................... 67 3-3 Delta AIC values for competing models describing associati ons between gizzard shad m aturity and lake, population density ( Bt), cohort size, year, and cohort................. 68 3-4 Delta AIC values for competing models describing associati ons between gizzard shad pre-recruit survival and sp awner biom ass (SB) and year.......................................... 68 4-1 Gear selectivity parameter estimates (95% confidence interval) for each gill net mesh size from the from th e length age model........................................................................... 90 4-2 Probability that total population biomass is less th an 25% of equilibrium unharvested value for a one and two year harvest in terval, a range of exploitation rates ( ), and five gill net mesh sizes ranging from 51 to 102 mm.......................................................... 91 4-3 Probability that transitional spawning poten tial r atio (SPR) is less than 25% of for a one and two year harvest interval a range of exploitation rates ( ), and five gill net mesh sizes ranging from 51 to 102 mm.............................................................................91

PAGE 7

7 LIST OF FIGURES Figure page 2-1 Proportional bias of model parameters at constant low fishing mortality (F = 0.2 yr1). .......................................................................................................................................37 2-2 Proportional bias of model parameters with fishing m ortality in creasing annually to a low level (F = 0.2 yr-1)...................................................................................................... 38 2-3 Proportional bias of model parameters at constant high fishing m ortality (F = 0.7 yr1)........................................................................................................................................39 2-4 Proportional bias of model parameters with fishing m ortality in creasing annually to a high level (F = 0.7 yr-1)..................................................................................................... 40 2-5 Observed (points) and model-predicted (lines) length-age survey catch proportions for gizzard shad at Lake Dora, Florid a from January 20005 to January 2009................... 41 2-6 Observed (points) and model-predicted (l ines) g izzard shad length distributions from the 2005 (upper) and 2006 (lower) fishery at Lake Dora, Florida..................................... 42 2-7 Gizzard shad recruitment estimates (milli ons of age-1 recruits ) from 1999 to 2009. Error bars represent 95 % confidence intervals.................................................................. 43 2-8 Model-estimated gear sel ectiv ity curves for the fisher y-independent gill net survey (solid line), 2005 fishery (dashed line), and 2006 fishery (fine dashed line).................... 44 2-9 Observed (points) and model-predicted (s olid line) gizzard shad m ean length-at-age from a fishery-independent gill net surv ey from 2005-2009 at Lake Dora, Florida.......... 45 3-1 Annual age-1 recruitment estimates (+/95% confidence interval) for lakes Dora (a), Eustis (b) and Harris (c) fr om the length-age model......................................................... 69 3-2 Cohort-specific maturity ogives for gizzard shad at lakes Dora (a), Eustis (b) and Harris (c) with respect to fish length.................................................................................. 70 3-3 Time series of predicted spawner biom ass for 2003 2008 at lakes Dora (a), Eustis (b), and Harris (c) from the length-age model................................................................... 71 3-4 Loge pre-recruit survival as a function of spawner biom a ss at lakes Dora (circles), Eustis (triangles) and Harris (plus).................................................................................... 72 3-5 Kernel density of maximum lifetime reproduc tive rate estim ates for lakes Dora (solid line), Eustis (dashed line) and Harris (fine dashed line).................................................... 73 4-1 Estimated gear selectivity curves for 51, 64, 76, 89, and 102-mm gill net m esh sizes..... 92

PAGE 8

8 4-2 Observed (points) and predicted (lines) leng th distributions of catches for each gill net mesh size.................................................................................................................. ....93 4-3 Total population biomass as a function of exploitation rate and gill net mesh size for an annual harvest interval.................................................................................................. 94 4-4 Spawning potential ratio (SPR) as a func tion of exploitation rate and gill net mesh size for an annual harvest interval..................................................................................... 95

PAGE 9

9 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy FISH COMPENSATORY RESPONSES FOLL OWING WHOLE-LAKE EXPERIMENTAL DENSITY REDUCTION By Matthew Jerome Catalano August 2009 Chair: Micheal S. Allen Major: Fisheries a nd Aquatic Sciences I evaluated fish compensatory responses fo llowing a whole-lake experimental density reduction of gizzard shad at Lake Dora, Florida. Gizzard shad were removed in 2005 and 2006 via a subsidized commercial fishery by the St. Johns River Water Management District. Approximately 30% of the total gizzard shad biomass and 70% of the total spawner biomass was removed. My first objective was to develop an age and length-structured population model that could be used to assess recruitment responses following density reduction. The model provided unbiased estimates of mortality, growth, gear sel ectivity and recruitment for long-lived species. For short-lived species there was a small upward bias in early recruitment estimates when the time series length was half of the maximum ag e, but all other scen arios provided unbiased estimates. My second objective was to evaluate compensatory responses of gizzard shad growth, maturity, and pre-recruit survival following density reduction. I found no evidence for density dependence in growth and matur ity. Pre-recruit survival was negatively related to spawner biomass suggesting density dependence in earl y life survival. The strength of density dependence in pre-recruit survival was qua ntified by estimating the maximum lifetime reproductive rate, which was 7.3 (95% confidence in terval: 1.9 16.5) at La ke Dora. Thus, prerecruit survival increases 7.3 -f old at very low spawner abun dance when compared to the

PAGE 10

10 unfished condition at Lake Dora. My third ob jective was to assess the relative efficacy of different gill net mesh sizes, exploitation rate, and harvest interval (number of years between removals) on total biomass reduc tion and spawning potential ratio of gizzard shad following removals at hypereutrophic Florida lakes us ing a simulation model. Specifically, I was interested in whether various co mbinations of these factors co uld achieve a 75% reduction in total population biomass and Spawni ng potential ratio (SPR). Thes e targets were obtained from the literature and represent th e reductions needed to achieve changes in phytoplankton biomass and to effectively reduce fish recruitment fo llowing biomanipulation. Gizzard shad biomass reduction failed to meet the target of 75% reduction for all mesh size s and harvest intervals except for the smallest (51-mm) mesh size. Reductions in spawning pot ential ratio exceeded total biomass reductions but remained above a 75% reduction target except when using the smallest mesh size (51 mm). I conclude that gill net removals are unlikely to result in substantial biomass reductions at Lake Dora unless removals adopt very small mesh sizes, which may not be preferred by commercial fishers.

PAGE 11

11 CHAPTER 1 GENERAL INTRODUCTION Introduction Density dependent population regulation is a pe rvasive theme in ecol ogy. Populations of living organism s are density dependent if their birth and death rates ar e functions of some measure of population density (Murray 1994; Gote lli 1995). Ecologists typically refer to two types of density dependence: depensatory and compensatory. Depensatory density dependence is a positive feedback on population size wher eby population growth in creases as density increases (Gotelli 1995; Rose et al 2001). Depensation can result in reduced reproduction at low population densities, which is known as the All ee effect (Allee et al. 1949). Compensatory density dependence is the opposite; population growth decreases as density increases (Gotelli 1995; Rose et al. 2001). Compensation results in high per capita reproductiv e rates in fishes at low spawner abundance and relatively low reproduc tive rates at high abun dance (Myers et al. 1999). There is considerable debate about th e relative importance of stochastic versus equilibrium (density dependent regulation) dynamics in animal populations, but densitydependence likely plays an important role in regulating populations (Murdoch 1994; Brooks and Bradshaw 2006). Understanding the mechanisms and specific life histor y stages leading to density dependence can provide insight into how populations might resp ond to perturbations such as harvest (Fogarty et al. 1992) and changes in habitat quality and quantity. Density dependence is a particularly importa nt concept in the life history of fish populations. There is strong support for the exis tence of compensatory density dependence in fish populations (Goodyear 1980; Myer s et al. 1999; Rose et al. 2001). Sustainable harvest of fish populations is predicated on the assumption that fish populati ons can compensate for harvest of individuals via increases in survival and reproduction. Without compensation, only

PAGE 12

12 populations with positive intrinsi c rates of population growth c ould be harvested sustainably (Rose et al. 2001). However, such populations would increase unbounded in the absence of harvest, which is unrealistic and typically not observed in nature (Murdoch 1994). Evidence for compensation in fish populations led to the early development surplus production models (Schaefer 1954), which predict ha rvest rates that fully utilize this surplus fish biomass. Depensatory density dependence, although impo rtant for many animal populations, does not have similar broad support in the fish eries literature (Myers et al. 1995). Compensatory population regulati on processes must affect the number of individuals in the population in order to act as a stabilizing for ce on population growth. These processes can act directly, via changes in survival and reproducti on, or indirectly via changes in growth or behavioral processes that in turn affect survival and reproducti on (Rose et al. 2001). Compensation in fish populations is thought to occur through reproduction and early life dynamics and can result from changes in vital rate s such as larval/juvenile survival and growth, which leads to changes in age-specific fecundity and survival. Such changes may result from shifts in availability of food a nd space due to relaxation of intraspecific competition. Larval and juvenile life stages are particularly importa nt regulators of fish populations (Hjort 1914; reviewed by Heath 1992), and even small changes in these rates can cause substantial change in subsequent adult abundance (Houde 1989). Density dependence in adult life stages such as changes in maturation schedules, length-specific fecundity, egg size and quality, and growth can also regulate population grow th (Trippel 1995; Rochet 1998 ; Rochet 2000), but these mechanisms are thought to be subordinate to pr ocesses operating during early life. However, few studies have evaluated the re lative importance of various co mpensatory mechanisms at the population level.

PAGE 13

13 Understanding recruitment in fish populations is critical to evaluating effects of compensatory population regulation. Recruitmen t is the conversion of eggs, through density dependent and density-independent pr ocesses, to the fish that re produce in the next generation (Myers 2002). Definition of the age and size at which recruitment occurs lacks standardization but is generally considered the age or size at wh ich fish become vulnerable to a fishery. Average annual recruitment is relatively constant over a wide range of spaw ner abundance for many species, but exhibits high lognormal variati on around expected values due largely to environmental variation (Walters and Martell 200 4). This has led some to conclude that recruitment is independent of spawner abundance. However, density-dependence must be strong in order for average recruitmen t to remain stable over a wide range of spawner abundances. Recent meta-analyses of spawner-recruit data have confirmed that fish populations are subject to strong compensatory density-dependence th rough reproduction (Myers and Barrowman 1996; Myers et al. 1999; Myers 2001; Myers 2002). Compensation During Early Life Density-dependent survival during early lif e stag es can result in large changes in subsequent recruitment. Early researchers surm ised that survival during the larval stage was most important and that starvation during some critical period had th e greatest influence on recruitment (Hjort 1914). More recently, results from indi vidual-based-model simulations indicate that density dependent feedbacks on recru itment are more likely during the late larval to early juvenile phase (Cowan et al. 2000) becau se total cohort consumption rates are highest during this period. Cushing (1990) proposed the match/mismatch hypothesis as a model explaining recruitment variability in fish populations. The mode l states that the degree of temporal overlap between peak larval abundan ce and maximum prey abundance determines the magnitude of recruitment (Cushing 1990), and ha s been supported by empirical data. Density-

PAGE 14

14 dependent change in juvenile survival is thought to be the primary mechanism for compensation in fish populations (Rose et al. 2001). There are several proposed mechanisms for density dependent survival of juvenile fishes. Walters and Juanes (1993) proposed that reduced survival at high juvenile densities results from increased risk taking at small spatial and tempor al scales by individuals attempting to procure scarce resources in a competitive environment. For example, juveniles may be forced to leave food-poor habitat refugia and spe nd more time in predator-dense feeding zones in order to maintain adequate growth rates when density of conspecifics is high. Conversely, juveniles can procure adequate prey without s ubstantial risk taking when density is low. This hypothesis has not been fully tested empiricall y, but Gilliam and Fraser (1987) showed that predation risk and food availability inte ract to affect individual behavior leading to increas ed predation risk when food availability is low. Density dependent growth rates of juveniles can also affect recruitment to adulthood. Growth rates of age-0 fish are often density-d ependent (Zijlstra and Witte 1985; van der Veer 1986; Peterman and Bradford 1987). Numerous studies have shown increased predation risk for slower growing individuals with in a cohort (reviewed by Sogard 1997). The bigger is better hypothesis (Shepherd and Cushing 1980) proposes that larger age-0 individuals have lower rates of mortality, because faster grow th decreases the duration of exposure to stages where mortality is high (Houde 1987; Miller et al. 1988; Hovenkamp 1992; Sogard 1997). In addition, smaller individuals may have higher rates of starvationinduced mortality during winter (Post and Evans 1989). Evidence from Adult Life Stages Density-dep endence in adult stages can al so result in compensation although these mechanisms are generally consider ed subordinate to j uvenile survival. Fecundity can change

PAGE 15

15 with adult density through changes in condition, food availability, and gr owth rate (Baccante and Reid 1988; Henderson et al. 1996). Individuals of higher cond ition may have greater lengthspecific fecundity (Henderson et al. 1996; Marshall and Frank 1999; Oskarsson and Taggart 2006). Density reduction via exploitation has ca used increased length-specific fecundity for many species (e.g., Koslow et al. 1995), although with changes in the age structure after fishing the total population fecundity will typically decline. Many fish species exhibit phenot ypic plasticity in ma turation schedules. Age at maturity generally decreases with increased exploitation (Trippel 1995). Populations with size-dependent maturation schedules may also undergo changes in ag e at maturity via increases in growth rate (Trippel 1995). Rochet (1998) repor ted that plasticity in maturati on has led to decreased age at maturity and increased size at maturity across 77 commercially exploited fish stocks. Other studies have documented decreased age and size at maturity following exploitation (Beacham 1983). Shifts in maturation schedules may be controlled by feeding conditions during nutritionally sensitive periods in gametogenesis (B urton 1994). Juvenile gr owth rates have also been linked to density dependent changes in maturation schedules (Brophy and Danilowicz 2003; Scheuerell 2005). Somatic growth is often density dependent in fish populations. Growth typically increases when population density decreases due to decreased intraspecific competition. Many studies have documented increases in growth related to exploitation of fish stocks (Millner and Whiting 1996; Rijnsdorp and van Leeuwen 1996; Helser and Almeida 1997). Healey (1980) reported increases in growth that were proportional to the de gree of exploitation in experimentally manipulated lake whitefish Coregonus clupeiformis populations. Lorenzen (2002) concluded

PAGE 16

16 that density dependent growth alone could explain compensatory population regulation in several species. Understanding compensation in fish populati ons has been hindered by a lack of experimental manipulations. Large-scale, controlled, population-level experimental manipulations are needed but are rare in the li terature. Few studies have manipulated fish populations experimentally at the whole-lake scale and tested mechanisms for compensation. Fish biomanipulation projects are ideal situations for testing comp ensatory density dependence. In these cases, target species are often unharveste d prior to biomanipulati on and total harvest and exploitation rate can usually be estimated. Nearby unharvested lakes can be used as control systems against which to evaluate population responses at harvested lakes. Study Objectives I evalu ated fish compensatory responses following whole-lake size-selective density reduction. My model species was gizzard shad Dorosoma cepedianum Gizzard shad are an important component of aquatic food webs in Nort h American rivers, lakes, and reservoirs, and are native to Florida la kes. They are omnivores, feeding on organic detritus and zooplankton. This flexible feeding strategy allows gizzard shad to simulta neously influence lake primary productivity and piscivore biomass through middl e-out processes (DeV ries and Stein 1992). Detritus feeding by gizzard shad may increase lake primary productivity via excretion of sediment derived nutrients into the water co lumn (Schaus et al. 1997). This mechanism represents a source of new nutrients to the phytoplankton and may contribute substantially to total lake phosphorus loading in some systems (Schaus et al. 1997; Vanni et al. 2006). Gizzard shad larvae and juveniles may reduce survival of economically important piscivores via grazing effects on zooplankton during early life (Stein et al. 1995). More over, gizzard shad are dominant organisms in freshwater ecosystems and can constitute over 90 % of fish biomass in eutrophic

PAGE 17

17 and hypereutrophic lakes (Allen et al. 2000; Vanni et al. 2006). Management of gizzard shad populations has important implications for lake ma nagement and consequently has been a target species for attempts at biomani pulation to alter lake primary pr oductivity and food web structure. The St. Johns River Water Management Dist rict in Florida has sought to manipulate gizzard shad populations in hypereutrophic Florida lakes to re duce lake primary productivity. Such biomanipulations have previously been carried out at Lakes A popka, Denham, and Griffin by subsidizing commercial fishers to harvest gizzard shad using gillnets. However, these manipulations did not include a ri gorous evaluation of responses in nutrient cycling, lake primary productivity, and shad population dynamics. A new study was initiated in 2005 to reduce the population density of gizzard shad at Lake Dora, using two other lakes as reference systems. This provided a unique opportunity to evaluate the response of a previously unexploited fish population to density reduction to assess density dependence and the causal mechanisms. Gizzard shad density reduction was carried out at one treatment lake (Lake Dora) and two unharvested reference lakes (Lakes Eustis and Harri s) at the Harris Chain of Lakes, Florida. Removals occurred at Lake Dora in Marc h-April 2005 and January-March 2006. Gizzard shad were removed via an experimental gillnet fishery by hired commercial fishers. Removal was highly size-selective owing to a minimum mesh size restric tion of 102 mm. Therefore, gillnetting reduced the density of gizzard shad approximately > 300 mm. The primary mechanism for the predicted improvements in wa ter quality was the re duction in phosphorus loading via removal of large detritivorous gizz ard shad that excrete previously sediment-bound nutrients into the water column via sediment feeding. It was thought th at removal of gizzard shad would reduce this bent hic-pelagic nutrient loading.

PAGE 18

18 My primary objective was to evaluate compen satory responses of gizzard shad growth, maturation, and pre-recruit surviv al following density reduction (Cha pter 3). To estimate prerecruit survival, I developed a novel age and length structured population model that estimated mortality, growth, gear selectivity, and recru itment parameters (Chapter 2). Finally, I incorporated estimates of the strength compensa tory density dependence of gizzard shad to evaluate the efficacy of removal methods (exploita tion rate, gill net mesh size, harvest interval) to gizzard shad biomanipulation (Chapter 4).

PAGE 19

19 CHAPTER 2 A SIZEAND AGE-STRUCTURED MODEL TO ESTIM ATE FISH RECRUITMENT, GROWTH, MORTALITY, AND GEAR SELECTIVITY Introduction Rates of m ortality and reproduction in fish popul ations are often a function of fish size rather than age (Sauer and Slade 1987). However, stock assessments have widely adopted methods that rely solely on fish age to estimate mortality, gear selectivit y, and recruitment rates (e.g., statistical catch-at-age, vi rtual population analysis). There is substantial variation in growth rates within cohorts resulting in a distri bution of lengths around each age. Size-selective fishing practices therefore result in differential fishing mortality rates among fish of the same age (Hansen and Chouinard 1992), which is not accounted for in most age-based population models. Stock assessments and management policies could be improved if data on length as well as age could be incorporated into estimation procedures. Length and age-structured models have an additional advantage because they can estimate growth parameters that account for gear selectivity and the cumu lative effects of size-selective harvest (Taylor et al. 2005). G ear selectivity affects age and length samples due to the selective properties of a survey or fishery gear. Fishing gears typically have greater capture efficiencies for larger individuals. Thus, mean length-at-age of age classes recruiti ng to the gear may be overestimated due to higher capture probabilities for the largest (fastest-g rowing) individuals of the cohort (Taylor et al. 2005). Cumulative size-sele ctive harvest effects refer to the decay of the largest individuals in the population over time via fishing mortality. These removal effects would be strongest on fully recru ited age classes by removing the la rgest individuals in a cohort, and therefore the remaining fish that are collected in length-age samples may represent primarily slow-growing animals (Kristiansen and Svsand 1998; Sinclair et al. 2002). These effects can

PAGE 20

20 lead to underestimation of the asymptotic length parameter ( L)and overestimation of the metabolic parameter ( K ) of the von Bertalanffy growth model (Taylor et al. 2005). Taylor et al. (2005) proposed a method for estimating mortality, gear selectivity, and growth parameters simultaneously from a length-ag e catch matrix (collected in a single year) using a multinomial maximum likelihood framework. Growth parameters obtained using Taylor et al.s (2005) approach were unbi ased with respect to gear sele ctivity and the cumulative effects of size-selective ha rvest. This method has the potential to allow estimation of critical population parameters with realistic data re quirements (i.e., one year of data ) and may be par ticularly useful for species for which large amounts of catch data are lacking due to minimal exploitation. The model is unique in that it accounts for variati on in length-at-age by carrying out survival calculations for each length-age bin explicitly. Here I developed a new model that uses the Taylor et al. (2005) formulation but extends the model to estimate historical recruitment (i.e ., the number of age-1 fish in the population each year) for situations where sequential years of su rvey length-age catch matrices are available. The model estimates recruitment, growth, mortality, and gear selectivity parameters from a time series of survey catches of length and age, the le ngth distribution of the ha rvest, and total annual harvest in biomass. The objectives of this chap ter were to (1) introduce the model structure, (2) evaluate model performance using a series of simulation-estimation procedures, and (3) demonstrate the model using data on gizzard shad from a whole-lake biomanipulation experiment in Florida. Methods Model The m odel estimates a recruitment time series ( Rt), instantaneous natural mortality rate ( M ), von Bertalanffy growth parameters (asymptotic length, L; metabolic coefficient, K ; time at

PAGE 21

21 zero length, t0), two parameters defining the stan dard deviation in length-at-age (1, 2), and three parameters of a flexible gear selectivity f unction for a fishery-inde pendent survey (shape, s; steepness, s; length at 50% selectivity, Ls 50) and the fishery ( v, v, Lv50). The model is conditioned on total annual harvest (biomass) and fit to a time series of survey (e.g., experimental gill net) le ngth-age catch matrices ( nl,a,t) and fishery length composition data ( fl,t) using a multinomial maximum likelihood function. The survey length-age data are arranged in an array of dimensions length age year. Th e survey length-age component calculates the likelihood of the observed catch of agea fish in length bin l at time t given a model-generated set of predicted proportions at age, length, and time (Taylor et al. 2005). The survey length-age log likelihood was: lat taltalpn n )ln( ln,,,,L, (2-1) where nl,a,t is the observed catch of agea fish in discrete length interval l at time t and pl,a,t is the model-predicted catch proportion of agea fish in length interval l at time t Predicted catch proportions pl, a,t are estimated as: lat ltal ltal talalPsN alPsN p,, ,, ,,, (2-2) where Nl,a,t is the predicted abundance of agea fish in length interval l at time t sl is the lengthbased survey gear selectivity, and P ( l | a ) is the probability of being in length interval l given age a. The Nl,a,t term incorporates fishing and natural mortality (described below). The likelihood term for the fishery length distribution data was calculated similarly except that the Nl a t terms are summed across ages to result in predicted length distributions and the sl term is replaced by the length-based fishery gear selectivity ( vl). Survey and fishery likelihood terms were summed to calculate the total likelihood.

PAGE 22

22 Survey gear selectivity, sl, (and fishery gear selectivity, vl ) was estimated using the function (Thompson 1994): )( )(50 501 1 1 1lL lL s s s lss sss se e s (2-3) where s is the shape parameter that determines the shape, describes the steepness, and Ls 50 is the length at 50% selectivity. This is a flexib le selectivity function th at produces either a dome shaped or sigmoidal curve, depending on parameter values. Values of s are bounded between 0 and 1. The functional form becomes sigm oidal (i.e., knife edge selectivity) as s approaches 0 and increasingly dome-shaped as s approaches 1. The P ( l | a) term is calculated from a normal pr obability density function with mean la and standard deviation sda. Mean length-at-age, la, is assumed to follow the von Bertalanffy (1938) growth model: )(01taK aeLl (2-4) where L is the asymptotic length, K is the metabolic coefficient, and t0 is the time at zero length. The standard deviation in length-at-age is estimated using (Fournier et al. 1991): 1 1 21 1 1 1A aesda (2-5) where 1 defines the magnitude of the standard deviations, 2 controls the trend in sda over ages, and is the Brody growth coefficient ( = e-K). The Nl.a,t terms are estimated as th e recruitment that occurred a -1 years prior ( Rt-a+ 1) times the survivorship to age a and length l over the time interval t-a +1 to time t : talZ attaleRN,,1 ,, (2-6)

PAGE 23

23 where Rt-a +1 is the recruitment that gave rise to the agea cohort and Zl, a t is the cumulative lifetime instantaneous total mortality for age a fish that are in length bin l at time t The model assumes fish recruit to the population at age 1, thus one is added to the time-specific recruitment subscript. Cumulative instantaneous mortality re presents the total lifetime mortality experienced by a fish of a given length-age-time bi n as they grew from age 1 to age a along a growth trajectory with an asymptotic length L ( l,t ) = l /(1-exp(K *( a t0))) (Taylor et al. 2005). The model assumes that K is time (years), length, and age invari ant, thus a unique asymptotic length L ( l,t ) (i.e., growth trajectory) is calculated for each length-age bin. The cumulative instantaneous mortality is calculated separately for each leng th-age bin and year as (Taylor et al. 2005): )'()'( ,,)1(a atal talFvaMZ (2-7) where a is a vector of ages from age 1 up to age a -1, and Ft( a ) and vl( a ) are vectors of annual instantaneous fishing mortality rates and length-speci fic fishery gear selectivities, respectively. These terms represent the fishing mortalities and fishery gear selectivities that would have been experienced in the past by fish in a given lengt h-age-time bin. The product of the elements of vectors Ft( a ) and vl( a ) were summed over the age interval a to calculate the cumulative instantaneous fishing mortality experienced by fish of a given length-age-time bin over their lifetime prior to time t The fishery gear selectivity ( vl( a ) terms were calculated by first determining the lengths that fish of a given length-age bin would have been in past years (i.e., at ages a ). These are a function of the length-age bin specific asymptotic length L ( l,t ) using la = L ( l,t )(1-exp(K *( a t0))). The length-specific fishery gear selectivity is then calculated for each of these ages using Equation 2-3.

PAGE 24

24 The Ft( a ) values are subset from a vector of a nnual instantaneous fi shing mortality rates Ft. The model is conditioned on aggregate annual catch (i.e., biomass). Thus, the annual instantaneous fishing mortality rate was calculated recursively as: t t tB C F 1ln (2-8) where Ct are the annual observed catches and Bt is the model-predicted vulnerable biomass. Biomass is calculated using an assumed length-weight relationship of the form, wt = ala b, which was estimated outside the model. The model requires Ft values for each year during the time span of the surveys and also for the A -1 years before the surveys began. This is because the initial A -1 Ft values are required to calculate the cumulative instantaneous mortality for fish that were alive before the surveys began. However, fishing mortal ity rates can be calculated only for years in which survey catch data are available because vulnerable biomass cannot be estimated prior to the first survey sample. This presents a problem if the fishery developed before the first survey occurred. The model can accommodate this situation by estimating an additional parameter: the initial average fishing mortality rate ( F0). This parameter represents the annual instantaneous fishing mortality for the years leading up to the collection of the fi rst survey. This is accomplished by setting the first A -1 values of the Ft vector equal to F0. This assumes that F was relatively constant for one generation time leading up to the first sample collection. The F0 parameter can be fixed at a value of zero if there is prior knowledge that the population was unf ished before the first sample was collected. Model Performance Model performance was evaluated by fitting the model to simulated data and estimating parameter bias for situations that would commonly occur in stock assessment situations. The

PAGE 25

25 model was used to generate data for all possibl e combinations of fish longevity (short-lived: A = 8 years; long-lived: A = 15 years), survey sampling durati on (i.e., years of data; short: 0.5* A yrs; long: A yrs), survey gear selectivity (asymptotic and dome-shaped), instantaneous fishing mortality rate (Fi = 0.2 and 0.7 yr-1) and fishing mortality trend (s table and increasing). For asymptotic sampling gear selectivity, the age at 50% selectivity was 0.3* L and fish attained 90% selectivity at 0.4* L. For dome-shaped gear selectivity, fish attained 50% selectivity at 0.3* L, maximum (100%) selectivity at 0.5* L, and 50% at L. For all scenarios, fishery gear selectivity was asymptotic with 50% selectivity at 0.6* L. The increasing F scenario allowed F to increase gradually from 0 to Fi over the years in which survey data were collected. Annual recruitment variation was log-norma lly distributed with a coeffici ent of variation (CV) of 50%, and observation errors on lengthand age-spec ific catches each year were drawn from a multinomial distribution. All other parameters we re held constant in the simulations. Monte Carlo simulation was used to generate data a nd estimate parameters for each combination of longevity, sampling duration, gear selectivity, an d fishing mortality schedule. Parameter bias was calculated for each Monte Carlo iteration by dividing the difference between the estimated and true parameter values by the true valu es. The median, 2.5%, 25%, 75%, and 97.5% quantiles for bias over 200 Monte Carlo iteration s were plotted for each parameter. Additional iterations (>200) resulted in no change in bias estimates. Application to Gizzard Shad The model was used to estimate parameters of a gizzard shad population that experienced a biomanipulation at Lake Dora, central Florida, USA. Bioman ipulation was achieved with an experimental commercial gill net fishery by the St. Johns Water Management District (SJRWMD). Prior to fish removal, the gizza rd shad population was unfished. Commercial

PAGE 26

26 fishers removed gizzard shad during March-April 2005 and agai n during January-March 2006. Gizzard shad were removed using gill nets with a minimum mesh size restriction of 102 mm, which selected for fish larger than approximately 300-mm tota l length. The SJRWMD used onboard observers to (1) record commercial catch-per-effort in 102-mm gill nets, and (2) measure a subsample of 100 harvested gizzard shad per week to characterize the length composition of the fishery harvest. The cumulativ e catch and total harvest (kg) each year were estimated from mandatory trip tickets, whic h were submitted to the SJRWMD daily by each fisher. An annual fishery-independent survey was conducted at Lake Dora to obtain data on the length-age composition of the gizzard shad populati on. Survey data were collected by setting 20 multi-panel floating gill nets at 20 fixed sites in late January or early February from 2005 to 2009. The 2005 sample was conducted prior to the initial biomanipulation and thus represented an unfished population size/age structure. The fi nal removal occurred in 2006, thus the 2008 and 2009 samples represented a rebuilding population. Gill nets were 2.4-m deep and contained eight, 15.3-m long panels of 38, 51, 64, 76, 89, 102, 114, and 127-mm stretch monofilament mesh. Each net was set for 2-3 hrs. Captured fish were measured for total length (mm) and counted, and otoliths were removed from a subs ample of 10 fish per 10-mm group for ageing. Otoliths were sectioned using a South Bay Tech Model 650 low-speed saw and read by three independent readers using a disse cting microscope at 40X magnification. Aged fish were extrapolated to the entire catch using an age-length key to estimate the age and length composition of the catch each year. Data inputs used in the model we re five years of survey length-age catch matrices, gizzard shad length distributions from the 2005 a nd 2006 fishery via onboard observers, and total

PAGE 27

27 harvested biomass in 2005 and 2006. The model was fitted using the optim() minimization function in program R. Parameter uncertainty was evaluated by inverti ng the Hessian matrix. Parameters that must be positive values (e.g., recruitments, M L) were constrained using a logarithmic transformation and parameters on a fixed interval from 0 to 1 (i.e., gs and gv) were fit using the logit transformation. Results Model Performance Median proportional bias was less than 0.1 fo r all parameters and scenarios except for the short-lived, short time period simulations, which had a slight upward recr uitment bias early in the time series (Figures 2-1, 2-2). Simulations with low fishing mortality (F = 0.2 yr-1; Figures 2-1 and 2-2) had greater bias than those with high er fishing mortality (F = 0.7; Figures 2-3 and 24). Median bias and bias uncertainty were greater for the short-lived than for the long-lived species (all Figures, panels a and b vs. c and d), and with a short ra ther than long time series of data (i.e., n = 0.5* A yrs; all figures 2, panels a and c vs. b and d). Bias was greater for scenarios with increasing F than with constant F (Figures 2-1 and 2-3 vs. Figures 2-2 and 2-4). Simulations that assumed a dome-shaped survey gear selectivity function exhibited greater bias than simulations with asymptotic selectivity, bu t the model could adequately determine the shape of the selectivity curve in most cases. Overall, the bias was relatively small, with parameters being biased by less than 5% in most cases but up to 40% in the worst case scenarios (i.e., R2 in Figure 2-2a). The amount of bias varied among parameters. Growth parameters had median bias of less than 2% and low uncertainty in bi as across all scenarios except for to which had a slight upward bias in the long-lived short time series constant high F scenario. However, the true value of the to parameter was near zero therefore very small absolute biases led to large proportional biases

PAGE 28

28 even though bias in mean length-at-age was negligible. Gear selectivity parameters also had low median bias and bias uncertainty across all scenar ios. Natural mortality, fishing mortality and recruitment parameters were the most biased due to moderate to high confounding among these parameters. Recruitment (i.e., particularly early in the time series) was correlated with natural mortality, and therefore overestimates in recrui tment led to overestimates in natural mortality because fish had to die more rapidly to explai n the catches. Recruitm ent overestimates would then lead to underestimates of fishing mortalit y because biomass was overestimated and because there would be an upward bias in natural mortal ity that would also reduce the impact of the removals on the population. Despite these correla tions, parameter uncertainty was low with CVs around recruitment estimates ranging from 10 to 35% and most other parameter CVs less than 10%. This indicated that the da ta contained enough information to reliably estimate all of the parameters except for the first few recruitment values. The first few recruitments produced the oldest ag e classes in the first year of survey catch data. These cohorts were captured in only one or two years of catch data and catches were very low due to cumulative mortality of the cohorts. Thus, the recruitment parameters for these cohorts were not well-defined in the data result ing in higher uncertainty in the model estimates. I used AIC to determine which of the early recruitment parameters were justifiably estimated by comparing AIC values from full models (all recruitment parameters estimated) with reduced models in which one or more recruitment values were fixed at the mean value of all other estimates. In general, the first recruitment parameter was not estimable for the short-lived species and the first three were not estimable for the long-lived species. Thus, these simulations identified the limitations of the model for esti mating early recruitments in a time series, but

PAGE 29

29 demonstrated that in nearly all other cases the model provided reliable estimates of the other parameters. Application to Gizzard Shad The total harvested biomass of gizzard shad was 124,989 kg (54 kg/ha) in 2005 and 135,095 kg (58 kg/ha) in 2006. The model fit th e survey length-age data reasonably well although there was some under-prediction of pro portional catches of age-1 fish in 2006, 2007 and 2009 (Figure 2-5). Modal length-at-age for age-1 fish was over-predicted in 2006 and 2009 suggesting slower than averag e pre-recruit growth in 2005 and 2008 (Figure 2-5). Length distributions from the fishery were pr edicted well by the model (Figure 2-6). Annual recruitment estimates varied from 0. 5 to 4 million age-1 gizzard shad and strong year classes occurred in 2000, 2004 and 2006 (Figure 2-7). Preliminary fits indicated that the first recruitment value ( R1998) was not estimable, and thus I fixed that value to the average of all other recruitment values. There was evidence for alternating strong and weak year classes although the pattern was not evident in 2001-2002 and 2008-2009 (Figure 2-7) Von Bertalanffy growth parameter and natural mortality estimates were precise with CVs less than 10% (Table 21). The length at 50% selectivity and (shape parameter) paramete rs were estimated precisely for all three gear sele ctivity functions (i.e., survey, 2005 fishery, 2006 fishery) with CVs less than 10%, but the steepness parameters were less certain (CV range: 15-36%; Table 2-1). The survey gear selectivity functi on was dome-shaped with selectiv ity increasing e xponentially up to 400 mm then peaking at 430 mm and declining to 0.75 at 450 mm (Figure 28). Gear selectivity curves for the fishery were asymptotic and suggest ed that length at 50% selectivity decreased in 2006 (Figure 2-8). This agreed wi th on-board observer data showi ng that fishers used mesh sizes

PAGE 30

30 of 102, 144 and 127 mm in 2005 but only used 102mm mesh in 2006 due to declining catches in the large mesh sizes. Discussion The model estimated recruitment, growth, natu ral mortality, and gear selectivity using only survey length-age and fishery length composition data. Parameter bias in these models should be lowest when fitting to long-lived species, by collecting data for at least one generation time of the species of interest, and if fishing mortality is relatively constant and moderate ( F = 0.7). However, performance was reasonable in other scenarios as well. The gizzard shad example represented what should be a relatively high-bias scenario because (1) gizzard shad are shor t lived, (2) only 5 years of data were collected (i.e., 63% of A ), (3) survey gear selectivity was dome shaped, and (4) F increased dramatically over two years. As a check on the gizzard shad model performa nce, I compared model-predicted exploitation rates to ones obtained via an in-season de pletion analysis. Catalano et al. ( in review ) estimated an annual exploitation rate of 0. 7 in 2005 and 0.65 in 2006 from a depletion of fisher catch per effort (kg per 100 m of net per hr) vs. cumulativ e catch of the fishery over the course of each annual harvest period. Model-predicted exploi tation rates were 0.67 in 2005 and 0.75 in 2006, which were similar in magnitude to the depletion estimates. As a second set of checks, gi zzard shad growth and mortal ity estimates were compared with literature values. Observed mean lengths -at-age were calculated directly from the agelength key data using methods of DeVries and Fr ie (1996) and these estimates were similar to model-predicted values (Figure 2-9). Mean lengths from the age-length key exceeded model predictions for older ages (ages 3-5 and 7). This was plausible considering that the survey gill nets should have selected for the largest individual s of these age classes due to an estimated steep selectivity curve from 300-400 mm (Figure 2-8). The model natura l mortality rate estimate of

PAGE 31

31 0.87 yr-1 was greater than estimates of 0.61 yr-1 from Paulys equation (Pauly 1980; M = f { L, K C}) and 0.53 yr-1 from Hoenig (1983; M = f { A }), but was similar to an estimate of 0.9 yr-1 from Jensen (1996; M = 1.5 K ). However, the Hoenig (1983) a nd Pauly (1980) values were derived from empirical models based on many fish populations and may not accurately represent gizzard shad life history characteristics. The mean and standard deviation of these literature M estimates could be used as priors in future uses of this model in a Bayesian framework. I also obtained M estimates using a simple catch curv e for unfished Lake Dora (2005) and a pooled estimate over five years from two near by unharvested control lakes (Lakes Eustis and Harris, 2005-2009) that had similar gizzard shad populations and were sampled with the same gear during the same time period as Lake Dora (Catalano et al. 2007). Catch curve M estimates were 0.6 yr-1 for Lake Dora in 2005 and 0.76 yr-1 from Lakes Eustis and Harris. Thus, the model M estimate was greater than the catch curve valu es suggesting that one of the two estimates was biased. Downward bias in the catch curve estimates could have resulted from an increase in survey gear selectivity with fish length, which w ould over-represent older fish in the catch. This selectivity trend was estimated to be the case by the model (Figure 2-8). Size selectivity is common for survey gears (Bayley and Austen 2002) but such biases are rarely considered when catch curves are estimated. The a dvantage of the model is that the M estimates account for gear selectivity. If M was in fact overestimated by the model, then there was likely a concurrent upward bias in recruitment values to produce a large enough population to explain the observed harvest. Taylor et al. (2005) reported that prior knowledge of the sh ape of the gear selectivity function was required for their model; however, my simulations suggested that the model could obtain unbiased estimates of the sh ape of the gear selectivity function with no prior information.

PAGE 32

32 The flexible selectivity function could accommodat e the dome-shaped survey gear selectivity for gizzard shad, which differed from the asymptotic selectivity assumption used by Taylor et al. (2005) for northern pikeminnow Ptychocheilus oregonensis Although the model showed a dome-shaped survey gear selectiv ity with peak selectivity at 430 mm, it should be noted that gizzard shad exceeding 430 mm should be extremel y rare in this population due to high natural mortality, an asymptotic length of 394 mm, and an estimated sta ndard deviation in length-at-age of age-5 to age-8 fish of around 30 mm. This dom e-shaped gear selectivity pattern could have been an artifact of size-dependent natural mortal ity rates in which faster-growing individuals had higher natural mortality rates. Nevertheless, failure to account for selectivity patterns could bias abundance estimates of large (older ) fish in the age structure. Hansen et al. (1997) found that total annual mortality of La ke Superior lake trout Salvelinus namaycush was underestimated by 20% when catch-age samples were not corrected for dome-shaped gear selectivity. This model could be very useful for evaluati ng populations where the shape of the gear selectiv ity function is unknown. The model used only survey catches at le ngth and age as well as fishery length distributions, but estimates could be improved with the inclusion of additional data types. For example, survey CPUE trend indices could be included to help define the magnitude of biomass reductions, which would provide information on the recruitment and mortality levels that would have been necessary to result in the observed catches and CPUE trend. Additional survey length-age catch matrices could al so be included, which reduced parameter bias and uncertainty in these models in preliminary simulation analys es (Catalano unpublished da ta). I did not allow for time, age, or cohort specific variation in gr owth, natural mortality, or gear selectivity. However, the model fits to the data suggested that mean age-1 length may have varied among

PAGE 33

33 years suggesting time or cohort-sp ecific growth rates. Inclusi on of growth variation in the simulated data would likely have reduced model pe rformance. Uncertainty in growth parameters was very low in these trials, thus there may be enough information in the data to estimate variable growth rates by a dding additional time-varying growth parameters. As another example, it was assumed that instantaneous natura l mortality was constant across age classes. Work by Lorenzen (2000) suggests natural mortality declines with age inversely proportional to fish length. Thus, the model could be adapted to estimate timeor size-varying natural mortality rates as well. The Taylor et al. (2005) account ing structure used in the model was unique compared to other existing length-age models. Stock synthesis 2 (Methot 2005) uses a growth-type-group accounting method where the stock is divided into several growth morphs each with its own von Bertalanffy growth parameters. The dynamics of each morph are calculated separately through time as fish stay in the same morph throughout life. This method can be very efficient if only a few groups are needed but can become comput ationally demanding with many groups. The number of groups necessary has not been stud ied and likely varies depending on the species. Another approach is the matrix transition approach such as the Fleksibest model by Frysa et al. (2002) This accounting structure calculates probabilities of transition from a given length bin and age to other length bins at the next age and time step. Th ese calculations can be rather complicated because fish from a single length-age bin will transition to mu ltiple length bins in the next time step. The Taylor accounting structure can be thought of as a type of growth type group (GTG) model (Walters and Martell 2004) b ecause it assumes that fish grow along a growth trajectory although it does not explicitly calculate the dynamics of each growth morph through time. Instead, the dynamics of each le ngth-age bin are explicitly accounted for by

PAGE 34

34 calculating back in time rather than calculating fo rward as in the typical GTG formulation. In addition, population numbers at leng th and age are predicte d directly rather than aggregating fish into length-age bins from growth morphs (in th e typical GTG formulation) by assuming a length distribution for fish of a given age and morph. The most common model bias encountered in the simulation-estimation experiments was overestimation of recruitment values, particularly early in the time series. In these situations, M is often overestimated and there were trends in re cruitment bias with larger recruitment biases for older cohorts than for more recent ones. These ear ly recruitments represent cohorts that were not tracked fully through the age structure. The potential for these biases can be reduced by collecting data over a sufficiently long time series, by including additional survey data types, and by not estimating several of the earliest recruitment values but rather setting them equal to the average of the estimated recruitments. This would be problematic for short time series of data where few cohorts are tracked throu gh the entire age structure. The model requires an adequate number of aged fish. I simulated the aging of 2,000 fish annually from survey data and 200 lengths from the fishery harvest, which represents a substantial data investment but is not uncomm on for most high-profile marine fisheries (e.g., groupers and snappers). Taylor et al. (2005) found that at least 500 fish were needed for their model to perform adequately. Model performan ce also depends on the number and width of the length bins. The model performed best when le ngth bins widths were between 2 and 5 % of L. Parameter bias increased substantially when length bins were gr eater than 10% of L. This coarse level of discretization can cause some age classes to be represented by only 1-2 length bins, which can introduce parameter bias due to random sampling error on these

PAGE 35

35 underrepresented age classes. As with other le ngth-based models, care should be taken when setting the level of length discretization in the model. Demographic parameter estimates for gizzard shad are rare in the literature and the model estimates presented here could be useful for fishery management in eastern North America. Gizzard shad are hypothesized to control food webs in eutrophic lakes through complex middleout processes by larval overgrazing of zoopla nkton and adult detritivorous foraging (DeVries and Stein 1992). Therefore, gizzard shad have received attention from ma nagers as targets for biomanipulation (Kim and DeVries 2000) and as prey for recreational fishes (Cyterski and Ney 2005). However, there are few published estimates of gizzard shad growth, gear selectivity and natural mortality with which to inform these studies. Bodola (1965) estimated an L of 395 mm and K of 0.78 yr-1 for Lake Erie, and Perry et al. (2003) reported estimates of 370 for L and 0.58 yr-1 for K in the Ohio River, which are similar to my estimates. Gear selectivity estimates would be particularly useful for biomanipulation effort s that utilize gill nets as the removal method. Van den Avyle et al. (1995) estimated retention probabilities for gizzard shad in various gill net mesh sizes but their estimates assumed equal encounter rates among fish lengths and did not represent gear selectivities that could be used in a modeling context. I could find no peerreviewed estimates of natural mo rtality for adult gizzard shad. This model could be useful for estimating cri tical population parameters of fishes. There have been no peer-reviewed evaluations of the performance of age-length structured assessment models to date. I showed th at these models can provide unbiased estimates of population parameters under most conditions a nd explored situations in which biases could arise to test the limits of the approach. One important advantage of a fully age and length-structured model is the ability to directly estimate growth parame ters in the model. The models would likely

PAGE 36

36 outperform traditional age-based models when bias es in growth parameters are suspected due to gear selectivity and the cumulative effects of size-selective harvest on age/size distributions. Table 2-1. Point estimates and 95% confidence intervals for model parameters estimated from length-age data for gizzard shad at Lake Dora, Florida, USA. Parameter Description EstimateL95%CI U95%CI M (yr-1) instantaneous natural mortality 0.870.77 0.99 L (mm) asymptotic length 394.30390.00 398.70 K (yr-1) metabolic coefficient 0.600.58 0.63 t0 (yr) time at zero length 0.170.14 0.20 1 (mm) length-at-age scaling parameter 31.0430.47 31.63 2 length-at-age shape parameter 0.060.03 0.09 Ls 50 (mm) survey length at 50% selectivity 452.60441.60 463.80 s survey selectivity shape 0.850.73 0.92 s survey selectivity steepness 0.100.05 0.17 Lv50(2005) (mm) 2005 fishery length at 50% selectivity 336.16322.92 349.95 v(2005) 2005 fishery selectivity shape 0.000.00 0.02 v(2005) 2005 fishery selectivity steepness 0.060.04 0.08 Lv50(2006) (mm) 2006 fishery length at 50% selectivity 301.20297.10 305.40 v(2006) 2006 fishery selectivity shape 0.000.00 0.03 v(2006) 2006 fishery selectivity steepness 0.130.09 0.19

PAGE 37

37 R2R4R6R8R10M LK t012Ls 50bsgsLv 50bvgvF0 -11234 a R2R4R6R8R10R12R14M LK t012Ls 50bsgsLv 50bvgvF0 -11234 b R4R8R12R16R20R22M LK t012Ls 50bsgsLv 50bvgvF0 -11234 c R4R8R12R16R20R24R28M LK t012Ls 50bsgsLv 50bvgvF0 -11234 d ParameterProportional Bias Figure 2-1. Proportional bias of model parameters at constant low fishing mortality (F = 0.2 yr1) for a short-lived (panels a, b) and long-lived (panels c, d) species and for a short (panels a, c) and long (panels b, d) time series. Every fourth recruitment parameter estimate is shown for efficiency.

PAGE 38

38 R2R4R6R8R10M LK t012Ls 50bsgsLv 50bvgvF9F11 -11234 a R2R4R6R8R10R12R14M LK t012Ls 50bsgsLv 50bvgvF9F11F13F15 -11234 b R4R8R12R16R20R22M LK t012Ls 50bsgsLv 50bvgvF15F17F19F21 -11234 c R4R8R12R16R20R24R28M LK t012Ls 50bsgsLv 50bvgvF15F19F23F27 -11234 d ParameterProportional Bias Figure 2-2. Proportional bias of model parameters with fishing mortalit y increasing annually to a low level (F = 0.2 yr-1) for a short-lived (panels a, b) and long-lived (panels c, d) species and for a short (panels a, c) and long (panels b, d) time series. Every fourth recruitment parameter estimate is shown for efficiency.

PAGE 39

39 R2R4R6R8R10M LK t012Ls 50bsgsLv 50bvgvF0 -11234 a R2R4R6R8R10R12R14M LK t012Ls 50bsgsLv 50bvgvF0 -11234 b R4R8R12R16R20R22M LK t012Ls 50bsgsLv 50bvgvF0 -11234 c R4R8R12R16R20R24R28M LK t012Ls 50bsgsLv 50bvgvF0 -11234 d ParameterProportional Bias Figure 2-3. Proportional bias of model parameters at constant high fishing mortality (F = 0.7 yr1) for a short-lived (panels a, b) and long-lived (panels c, d) species and for a short (panels a, c) and long (panels b, d) time series. Every fourth recruitment parameter estimate is shown for efficiency.

PAGE 40

40 R2R4R6R8R10M LK t012Ls 50bsgsLv 50bvgvF0 -11234 a R2R4R6R8R10R12R14M LK t012Ls 50bsgsLv 50bvgvF0 -11234 b R4R8R12R16R20R22M LK t012Ls 50bsgsLv 50bvgvF0 -11234 c R4R8R12R16R20R24R28M LK t012Ls 50bsgsLv 50bvgvF0 -11234 d ParameterProportional Bias Figure 2-4. Proportional bias of model parameters with fishing mortalit y increasing annually to a high level (F = 0.7 yr-1) for a short-lived (panels a, b) and longlived (panels c, d) species and for a short (panels a, c) and long (panels b, d) time series. Every fourth recruitment parameter estimate is shown for efficiency.

PAGE 41

41 100200300400500 0.000.06 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 a 100200300400500 0.000.06 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 f 100200300400500 0.000.06 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 b 100200300400500 0.000.06 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 g 100200300400500 0.000.06 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 c 100200300400500 0.000.06 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 h 100200300400500 0.000.06 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 d 100200300400500 0.000.06 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 i 100200300400500 0.000.06 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 age 1 age 3 age 5 age 7 e 100200300400500 0.000.06 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 age 2 age 4 age 6 age 8 jTotal Length (mm)Catch Proportion Figure 2-5. Observed (points) and model-predicted (lines) le ngth-age survey catch proportions for gizzard shad at Lake Dora, Florida from January 2005 (a, f)to January 2009 (e, j). Odd ages are shown on the left column panels (a e)and even ages are on the right (f j).

PAGE 42

42 100 200 300 400 500 0.000.050.100.15 a observed predicted 100 200 300 400 500 0.000.050.100.15 Total Length (mm)Catch Proportionb observed predicted Figure 2-6. Observed (points) and model-predic ted (lines) gizzard shad length distributions from the 2005 (a) and 2006 (b) fishery at Lake Dora Florida. Odd ages are shown on the left column panels and even ages are on the right.

PAGE 43

43 20002002200420062008 02468 YearRecruitment (millions) Figure 2-7. Gizzard shad recruitment estimates (millions of age-1 recruits) from 1999 to 2009. Error bars represent 95 % confidence intervals.

PAGE 44

44 100150200250300350400450 0.0 0.20.4 0.6 0.8 1.0 Total Length (mm)Gear Selectivity Survey 2005 Harvest 2006 Harvest Figure 2-8. Model-estimated gear selectivity curves for the fishery-independent gill net survey (solid line), 2005 fishery (dashed line), and 2006 fishery (fine dashed line).

PAGE 45

45 2468 0 100 200 300 400 Age (yrs)Total Length (mm) Observed (+/95%CI) Predicted Predicted 95% CI Figure 2-9. Observed (points) and model-predicted (solid line) gizzard shad mean length-at-age from a fishery-independent gill net survey from 2005-2009 at Lake Dora, Florida. Error bars represent 95% confidence interv als for observed length-at-age and dashed lines signify the 95% confidence interval for model-predicted length-at-age.

PAGE 46

46 CHAPTER 3 DOES INCREASED PRE-RECRUIT SURVIVAL DRIVE FISH DENS ITY DEPENDENCE?: EVIDENCE FROM A WHOLE-LAKE EX PERIMENTAL DENSITY REDUCTION Introduction Compensatory density-dependence is a negative feedback on population growth rate via functional relationships between population density and vital rates. Understanding the strength and mechanisms of compensation is a pervas ive theme in ecology because these factors determine the ability of animal populations to withstand anthropogenic perturbations. Densityindependent processes are important in determinin g year-to-year variability in demographic rates of animals, but density-dependence must play a critical role in regulating population size (Murdoch 1994; Brooks and Bradshaw 2006); w ithout it, populations would increase unbounded or be driven to extinction stoc hastically. However, detection of density dependence in natural populations has been elusive (but see Brooks and Bradshaw 2006) due to inadequate analyses (Rotella et al. 2009), environmental noise, and the difficulty of experimentally manipulating animal population density at an appropriate spat ial or time scale. Compensation occurs through density-dependent changes in rates of survival reproduction or migrati on. Understanding the vital rates and life stage where density depe ndence occurs can provide insight into how populations might respond to pertur bations such as harvest and ch anges in habitat quality and quantity. There is strong support for the existence of compensatory density dependence in fish populations (Goodyear 1980; Myers et al. 1999; Rose et al. 2001). Recent meta-analyses of spawner-recruit data have confirmed that fish populations are subject to strong compensatory density-dependence via changes in reproduction (Myers and Barrowman 1996; Myers et al. 1999; Myers 2001; Myers 2002). Compensation result s in high per capita reproductive rates in fishes at low spawner abundance and relatively low reproductive rates at high abundance (Myers

PAGE 47

47 et al. 1999). The strength of density dependence varies widely across species and likely depends on life history strategy (Winemiller and Rose 1992) habitat structure, and species interactions (Rose and Cowan 2000). Such changes may result from shifts in availability of food and space due to relaxation of intraspecific competition. There is considerable debate about the re lative contributions of changes in various demographic rates to compensation in fish populations. Compensation may occur through changes in demographic rates duri ng early life (larval and juvenile) and adult life stage. Larval and juvenile life stages are pa rticularly important regulators of fish populations (Hjort 1914; reviewed by Heath 1992), and even small changes in these rates can cause substantial change in subsequent adult abundance (Houde 1989). Compensation during the early life phase can occur through changes in growth (Zijlstra and Witte 1 985; van der Veer 1986; Peterman and Bradford 1987) or variation in survival re lated to risk-sensitive foraging behaviors (Walters and Juanes 1993). Density dependence in ad ult life stages such as chan ges in maturation schedules, condition (leading to increased fecundity), egg size and quality, and growth can also regulate population growth (Trippel 1995; Rochet 1998; Rochet 2000). De spite extensive research on fish density dependence, few studi es have evaluated the relative importance of early versus adult life stages to compensation at the population level. A central problem in understanding compensation in animal populations is the difficulty in manipulating population densitie s on a large enough scale to measure density-dependent responses. Many studies have treated exploited fish populations as replicates from which to draw inferences regarding the strength and m echanisms of density dependence (Rijnsdorp 1993; Myers et al. 1999; Goodwin et al. 2006). However, these studies rarely include reliable data before fishing, and there are no observations fr om unmanipulated control systems against which

PAGE 48

48 population trends of the harvested populations could be compared. Moreover, management practices often aim to maintain exploited populations within bounds to maximize production, which can reduce contrast in population density and limit its utility for evaluating density dependence. Whole-lake experiments provide th e appropriate scale for detecting meaningful responses (Schindler 1998). Smith and Wa lters (1981) advocate large-scale adaptive management of ecosystems whereby individual fish stocks are treated as re plicates and subject to varying levels of exploitation, some of which cause population failure. Such studies are rarely feasible and are politically challenging becau se of the many individuals who depend on these stocks for their livelihoo ds (Walters and Martell 2004). Thus, few studies have evaluated density reductions in an experimental framewor k at the population (e.g., lake) scale. In Florida, U.S.A, the St. Johns River Wate r Management District (SJRWMD) has used biomanipulation of omnivorous gizzard shad Dorosoma cepedianum as a lake restoration tool to reduce phytoplankton abundance and improve water clarity at hypereutrophi c natural lakes. Biomanipulation is the selectiv e removal of fish to reduce grazing pressure on zooplankton thereby increasing water clarity vi a trophic cascades (Shapiro et al. 1975; Carpenter et al. 1987). The SJRWMD evaluated the efficacy of the gizzar d shad removal via a w hole-lake experimental density reduction of gizzard sh ad at Lake Dora in 2005 and 2006. The biomanipulation removed approximately 30% of total gizzard shad bioma ss from Lake Dora in 2005 and 2006 (Catalano et al. in review ) and larger amounts of adult gizzard shad because the fishery selectively removed large gizzard shad (> 300 mm). The removal pr ovided the opportunity to evaluate compensatory responses of the gizzard shad population followi ng density reduction. Here I assessed the relative importance of three m echanisms for compensatory density dependence of gizzard shad

PAGE 49

49 following whole-lake density-reduction. I measur ed changes in adult growth, maturation, and pre-recruit survival at an experimental lake (Lake Dora) and two unharvested control lakes. Study Site The study was conducted at lakes Dora, Eustis and Harris, Lake County, Florida, USA. Lake Dora was the experimental lake where gi zzard shad density reduction occurred, and the other lakes served as unmanipulated control syst ems. All lakes were large (>2,000 ha) shallow (3 m average depth) hypereutrophic lakes (mean chlorophyll a concentrat ions > 50 ug/L) with similar fish communities. Lake Dora is composed of two distinct basins: Lake Dora and the smaller Lake Beauclair. The two basins are separated by a 300-m long by 80-m wide canal. Gizzard shad harvest occurred in both basins, so they were treated as one system for these analyses and will be referred to collectively as Lake Dora hereafter. Narrow canals (<30 m wide, > 1.0 km long) also connected Lake Dora to Lake Eustis and Lake Harris to Lake Eustis. The degree to which gizzard shad moved among the lakes was unknown. Biomanipulation at Lake Dora was achieve d with a government-subsidized commercial gill net fishery by the SJRWMD. Prior to fish re moval, gizzard shad populations in both lakes were unfished. Commercial fish ers removed gizzard shad from Lake Dora during March-April 2005 and again during January-April 2006. Gizzard shad were removed using gill nets with a minimum mesh size restriction of 102 mm, which selected for gizzard shad larger than approximately 300-mm total length. Removal wa s carried out by commercial fishers, with an average of five boats setting 3-5 sinking gill nets per day, each net ranging in length from 75 to 600 m. The SJRWMD used onboard observers to obtain data on commerc ial catch-per-effort and catch composition.

PAGE 50

50Methods I evaluated density-dependence in growth, maturity, and pre-recruit survival using data from the density reduction at Lake Dora and two control lakes. There are two ways to analyze this type of control-impact design. One appr oach would be to conduct a before-after-controlimpacts analysis. In such cases, controls are trea ted strictly as a reference system for comparison to the impact system. However, I was interest ed in density effects on demographic rates, and density varied at the control la kes as well, although not as much as at Lake Dora. A second approach would be to include data from the cont rol lakes as well to take advantage of natural changes in density in those lakes due to recruitmen t variability. I chose this second approach and thus used all lake years of data as replicates in the analyses, such that the density reduction at Lake Dora served to increase contrast in the data set, but data from control lakes were also included in the assessment of density dependent pr ocesses. This approach improved the scope of inference for the study by including naturally fluc tuating populations as observations, along with the observations from the dens ity reduction at Lake Dora. Field Data Collection Data regarding the magnitude of the gizzard shad density reduction were collected by SJRWMD from a fishery-dependent onboard obser ver program. Onboard observers recorded fishery catch-per-effort in 102-mm gill nets an d measure a subsample of 100 harvested gizzard shad per week to characterize the length composition of the catch. The cumulative catch and total harvest (kg) each year were tabulated from mandatory trip tickets, which were submitted to the SJRWMD daily by each fisher. Gizzard shad demographic information (growt h, maturity, pre-recruit survival) was collected via annual fishery-independent gill ne t surveys conducted by the author (UF) and SJRWMD at each lake. The UF survey set multipanel floating gill nets at 20 fixed randomly-

PAGE 51

51 selected sites at each lake in January/February (all lakes: 2005 2009) a nd November (all lakes: 2004 2006; Lake Dora: 2009). The SJRWMD survey set multi-panel gill ne ts at 10 fixed sites at lakes Eustis and Harris and 20 sites at Lake Dora during January (Lake Dora: 2003, 2005 2009; Lake Eustis: 2003, 2006-2009; Lake Harris: 2003) Survey gill nets were 2.4-m deep and contained five, 15.3-m long panels of 76, 89, 102, 114, and 127-mm stretc h monofilament mesh and nets were set for 2 hours each. The UF gill nets had three additional panels of 38, 51, 64mm mesh to target age-1 fish. I collected information on size, age, and maturity for gizzard shad. Captured gizzard shad from both surveys were counted and measured fo r total length (mm). Gizzard shad from the UF survey were aged by analyzing otoliths from a s ubsample of 10 fish per 10-mm length interval; fish from SJRWMD surveys were not aged. At the lab, fish were measured, weighed, and otoliths were sectioned using a South Bay Tech Model 650 low-speed saw and aged by three independent readers using a disse cting microscope. The length and age composition of the UF survey data were estimated from the length distribution by multiplying the number of fish captured in each length interval by the proportion of fish at each age within that interval (i.e., age-length key method). Gender was determined on aged fish and the ovaries removed, weighed (g), and preserved in 10% buffere d formalin solution to assess age/size at maturity. To verify that the January/February survey was carried out wh en female gizzard shad were at or near peak spawning condition, additional gill nets (one to three nets) were set twice per month from January to May 2005-2007 at each lake. At least 30 adult females were collected per trip to assess temporal trends in the gonadosomatic i ndex (GSI; GSI = ovary weight/ovary free fish weight), which indicated the duration and peak of the spawning period. Recruitment

PAGE 52

52 Recruitment and other critical demographic pa rameters were estimated using the data collected above input into an ageand lengthstructured population model (Chapter 2). The model was fitted to gizzard shad data from Lake Dora and the two control lakes (Lakes Eustis and Harris) to estimate time-specific annual recruitment to age 1 for lake i ( Rt,i), age and timeinvariant instantaneous natural mortality ( Mi), von Bertalanffy growth parameters (asymptotic length, L i; metabolic parameter, Ki, and time-at-zero length, t0 i), and gear selectivity parameters (fishery and survey) using a mu ltinomial maximum likelihood func tion. Data inputs were (1) lengthand age-specific gill net catches from the November and January/February UF fisheryindependent surveys, (2) annual length dist ributions from the Ja nuary SJRWMD fisheryindependent gill net surveys, (3) gizzard shad length distributions from the 2005 and 2006 Lake Dora fishery from the onboard observers program, and (4) total harvested biomass at Lake Dora in 2005 and 2006. The model was conditioned on to tal harvested biomass (observed harvest was subtracted from predicted biomass in the model) and likelihood terms for each of the other three data sources were summed to calculate the tota l likelihood. Parameter uncertainty was evaluated by sampling from the posterior distribution of parameters with Markov Chain Monte Carlo simulation using the Metropolis-Hastings algo rithm (Hastings 1970). I simulated 250,000 iterations with a burn-in period of 25,000 and th inning interval of 250. The tuning parameter was set to obtain an acceptance rate of 0.25. Convergence of the chains was evaluated by inspecting trace plots. Sampling from the poste rior distribution of the length-age model parameters was used to assess uncertainty in density-dependent parameters of gizzard shad (see Pre-recruit Survival, below). The model scaled recruitment estimates (i.e., ag e-1 abundance) at Lake Dora such that they were large enough to explain th e observed harvested biomass in 2005 and 2006. Therefore,

PAGE 53

53 annual recruitments at Lake Dora could be freely estimated as parameters in the model. However, recruitments at lakes Eustis and Harri s had no scaling information because those lakes were unharvested. Thus, recruitments at lakes Eustis and Harris were estimated as lognormally distributed residuals ( t, i) around an average annual re cruitment value of 1.0: 2 iR, it,5.0 ,e RRit (3-1) and the variance was constrained using a penalty function that was added to the total likelihood value: t i it i iitP2 R, R R, R,,2 )ln()|(ln (3-2) where R,i is the standard deviation of the recruitment residuals (Maunder a nd Deriso 2003). This approach maintained an average recruitmen t of 1.0 and constrained the standard deviation of the recruitment anomalies to realistic values at lakes Eustis and Harris. Growth I tested for density dependence of growth ra tes by modeling associa tions between annual growth increments and population density across th e three lakes. Length and age data from the UF January gill net survey were used to calcula te mean length-at-age using methods of Devries and Frie (1996) for age-length keys. This appr oach produces unbiased means when aged fish are subsampled on fixed length intervals for an age-length key (i.e., 10 fish per 10-mm length interval). Growth increments were calculated as the difference in mean length from one year to the next for a given cohort and were loge transformed. Growth increments were obtained for the 2003 2007 cohorts and were limited to age-5 or younger fish because of low sample sizes of older age classes. Analysis of covariance was used to test for effect s of population density on logged growth increments using age as the concomitant variable and lake as a block factor in the

PAGE 54

54 model. Population density was the annua l total population biomass at the beginning of the year over which the growth increment was calculated. It was calculated as th e predicted numbers of fish in each 10-mm length interval multiplied by the m ean weight of fish of that interval using a lake and time invariant length-weight relationship. Density values were obtained as outputs from the length age model and were rescaled to a mean of zero. Model selec tion was carried out using Akaikes information criterion (AIC). Maturity I developed a relationship between GSI and matur ity using a subset of female gizzard shad. This allowed the use of GSI as a proxy for matur ity. Histological sections were prepared from formalin-preserved ovaries from lakes Dora and Eustis in late January to early March 2007 when fish were in peak spawning condition. Gizzard shad are batch spawners and reproduce over a 23 month period in central Florida (personal obser vation). Preliminary analyses of temporal trends in GSI from January to May indicated th at fish were in peak spawning condition from late-January to early March, and this pattern was relatively consistent across years. Thus only females collected during January-March were used in the analysis to minimize bias due to the timing of sampling relative to spawning. I sa mpled at least five females per 25-mm length interval. Histological sections were stained with hematoxylin a nd eosin, embedded in paraffin, sectioned, and mounted on a glass slide at the University of Florida College of Veterinary Medicine, Department of Tissue Pathology. Fe males were considered mature if histology showed the presence of vitelloge nic (yolked) oocytes. Maturity was modeled as a function of GSI using logistic regression and testing for la ke and lake length e ffects. Probability of maturity was estimated as a function of GSI for females from lakes and/or years with no

PAGE 55

55 histological information. Individuals with a model-predicted probability of maturity exceeding 0.5 were classified as mature and all others were classified as immature. I modeled maturation as a cohort-specific pr ocess, with each cohort potentially maturing according to its own cohort-specific ogive. I ev aluated two types of density effects on cohort maturity: intercohort and intracohor t. Intercohort effects were m odeled by including a term for the total biomass ( B2+) of all other age-cla sses when a given cohort recruited to age 1. Intracohort effects were modeled by including a term for cohort size, or the year class strength for a given cohort. Population density values we re obtained from the length-age model and were rescaled to a mean of zero as in Growth above. Cohort size was the annual recruitment estimate for each cohort from the length-age model and was also rescaled to a mean of zero. Maturity ogives (proportion of fish mature) were m odeled as a function of length, cohort size, B2+, lake, and lake length interactions wi th logistic regression. The lake length interaction tested whether the shape of the maturity schedule varied among lakes. Preliminary analyses indicated that length was a better predictor of maturity than age, but the tw o factors were highly collinear. Thus age was excluded from the models describi ng the maturity ogives. Model selection was carried out using AIC. Statis tically significant associations between cohort size or population density and maturity would indicate density dependence in maturation. Pre-recruit Survival Lakeand time-specific pre-recruit survival (St,i) was estimated from the length-age model by dividing annual estimates of recruitment (Rt,i) by the model-predicted total spawner biomass ( Bt-1 ,i) from the previous year. Spawne r biomass was calculated for lake i and time t as: 1 ,1,,,1,, ,1 t la litalital itHwmN B (3-3)

PAGE 56

56 where Nl,a,t-1 ,i is the model-predicted number of agea gizzard shad of length l in the population at time t ml,a,t-1 ,i is the lengthageand time-specific proportion of mature fish at lake i wl is the weight of a lengthl fish, and Ht-1 represents the spawner biom ass that was removed by the fishery just prior to the spawn in th e previous year. Population numbers ( Nl,a,t-1 ,i) were predicted from the model as a function of estimated parameters. Maturity was predicted from the logistic regression model relating maturity to length, age, cohort size, population density, and lake (see Maturity above). Weight is commonly used as a proxy for fish fecundity (Quinn and Deriso 1999) and was estimated from gizzard shad length data from the lakes using the lake-invariant allometric relationship wl = 6.97e-7 l 3.49. Harvest in 2006 began before gizzard shad spawned and this needed to be incorporated into the spawner biomass estimates. Examination of densities of yolked larval gizzard shad from biweekly larval fish tows sugge sted that approx imately half of the catch had been taken before the gizzard sh ad spawned. Thus I subtracted from the Bt an estimate of the spawner biomass that was removed just prior to the spawn in the previous year: 1 ,1,,,1,, ,15.0 tl la litalital ituvwmN H (3-4) where vl is the length based select ivity of the fishery and ut-1 is the proportion of vulnerable sized fish harvested the previous year (exploitation rate). Annual recruitments were scaled differently at Lake Dora (scaled to the observed harvest) than at lakes Eustis and Harris (scaled to mean of 1.0 fish). Pre-recruit survival was a quotient and was thus dimensionless and comparable amo ng lakes but spawner biomass was scaled to the annual recruitments and thus was not comparable Consequently, spawner biomass was rescaled to a mean unfished value of 1.0 kg at each lake prior to use in estimating density dependence in pre-recruit survival. The mean unfished spawner biomass at Lake Dora was the average of the 2003 to 2005 pre-density reduction estimates.

PAGE 57

57 The strength of density-depende nt recruitment at the lakes was evaluated by modeling prerecruit survival as a function of spawner biomass and environmenta l factors using the linear form of the Ricker stock-recruit function: ittiti ittwBb S, ,)ln()ln( (3-5) where is the maximum pre-recruit survival at ve ry low population density (initial slope of recruitment vs. spawner biomass relationship) and was the parameter of primary interest in this model, b describes the strength of density depe ndence at high spawner biomass, the wt terms represent annual environmental effects on pre-recruit survival that act on all of the lakes. Including these shared environmental affects in th e model may help ameliorate bias due to serial autocorrelation in pre-recruit survival and spawner biomass (Walters and Martell 2004). The mechanism for these environmental effects was not of interest but visual examination of temporal trends in survival suggested that the lakes were affected by a shared environmental influence on year class strength, which is not uncommon for geographi cally proximate fish populations (Maceina and Stimpert 1998). Myers et al. (1999) conclude d that the Ricker model wa s appropriate for evaluating density-dependent recruitment for a range of spec ies when the primary parameter of interest is However, the magnitude of is not comparable among populations unless it is compared to prerecruit survival in an equilibrium unharvested po pulation. Thus, the more valuable measure of density-dependence of pre-recruit survival is the maximum lifetime reproductive rate ( ; Myers et al. 1999), which is also referred to as the Goodyear compensation ra tio (Goodyear 1980). This value represents the ratio of j uvenile survival at low populati on density to survival in the unfished condition and is a standardized measure of density dependence that is comparable across populations (Myers et al. 1999). I calculated for each lake as:

PAGE 58

58i i ,0 (3-6) where i ,0 is the equilibrium lifetime spawner bioma ss per recruit for gizzard shad at lake i : a iaaia iMws, ,0, (3-7) where sa,i is the survivorship to age a wa is the average weight, and ma,i is the proportion mature to age a Uncertainty in the maximum lifetime re productive rate was as sessed by repeatedly fitting the stock-recruitment model to survival and spawner biomass estimates taken from posterior samples of the parameter distributi ons obtained via the MCMC simulation of the length-age model. Observation error in spawner biomass and seri al correlation between pre-recruit survival and spawner biomass can cause overestimates of (reviewed by Walters and Martell 2004). Monte Carlo simulation was used to explore these potential biases I simulated a fish population using length-age model estimates of gizzard shad population parame ters, simulated a time series of random recruitments and subseque nt spawner biomass assuming a known sampled from stock and recruit pairs with observation error, fit the Ricker model and evaluated bias in estimates. Results Recruitment Estimated recruitment time series showed some degree of temporal synchrony in year class strength among lakes (Figure 3-1). Lake Do ra had strong age-1 re cruitment in 2000 and 2006 (1999 and 2005 year classes; Figu re 3-1a). Lake Eustis had above average recruitment in 2000 as well, but also had high recruitment in 1999 an d 2009, as did Lake Harris (Figure 3-1b,c). The 2006-2008 post-manipulation year classes at Lake Dora showed no decline following density reduction but rather were near the long-term average recruitment for the time series, suggesting

PAGE 59

59 that the density reduction did not substantially affect recruitment (Figure 3-1a). All other model parameter estimates are listed in Table 3-1. Growth Annual growth increments differed among ag es, but not among lakes or with population density (Table 3-2). The model with age only had the best AIC support (Table 3-2; intercept = 5.64.08; slope (age) = -0.51.02; df = 55; R2 = 0.89). Fitting additional parameters for population density and lake was not justified based on AIC (Table 3-2). The best model (age) fit the data considerably better than the single parameter (null) model (Table 3-2). Thus I conclude that growth was not density dependent and re mained relatively constant throughout the time period at each of the lakes. Maturity Maturity was strongly related to GSI and th ere were no significant lake or lakelength effects. The best model had two parameters (i ntercept = -9.2.53, slope (GSI) = 4.6.28) on 94 residual degrees of freedom. The GSI (%) at which the model-predicted probability of maturity was 0.5, was 1.99%. Correct classification rates of mature and immature females were high. Ninety-three percent (3/48) of females cl assified by the model as mature were in fact mature as indicated by histology. Likewise, 93% (3/48) of females classified as immature were in fact immature. Hence, female gizzard shad ar e likely to be mature if their GSI exceeds 2%. Because of the high classification rates, I was co mfortable extrapolating the model to other lakes and years to estimate maturity of females for which ovarian histology was not analyzed. Length-at-maturity was weakly related to popula tion density and the dire ction of the effect was opposite of my prediction (Table 3-2). The minimum AIC model was an additive model with length, lake, and po pulation density (Table 3-3). However, there were seven other models

PAGE 60

60 with nearly equiva lent AIC support ( AIC < 5), each of which included lake (Table 3-3). Thus maturity varied among lakes (Figure 3-2); models excluding lake had AIC values near 50. Including density resulted in marginal impr ovements in model parsimony over the model with just length and lake ( AIC = 3) and including cohort size resulted in no improvement in AIC (Table 3-3). The population densit y coefficient was positive in th e model suggesting that gizzard shad matured earlier at higher population densities but the size of the effect was relatively small; 50% increase in population density w ould result in a 20-mm decrease in length at maturity. Thus maturity was not strongly related to density und er the range of variation in density that I observed in the study lakes. Of the seven mode ls with nearly equiva lent AIC support, the simplest included the factors length and lake (i ntercept = -11.29; length = 0.041.002; lake = 1.52.23; df = 1166), and was selected as the most pars imonious, biologically plausible model, and was used in subsequent calculat ions of spawner biomass (see below). Pre-Recruit Survival Spawner biomass at Lake Dora decreased to 28% of the average unharvested biomass in 2006 following the second year of harvest (Figure 3-3a). This reduction exceeded the natural variation in spawner biomass obser ved at control lakes (Figure 3-3). Spawner biomass decreased steadily from 2003-2009 at control la kes due to natural mortality of large year classes in 1999 and 2000 (Figure 3-3b,c). Pre-recruit survival wa s greatest at Lake Dora in 20052007 just after density reduction as recruitment was near the long-term average despite substantially reduced spawner biomass (Figure 3-4). However, pre-re cruit survival at Lake Dora after density reduction was exceeded by survival rates at control lakes in 2009 (Figure 3-4). Juvenile survival was negatively related to spawner biomass across lakes and years (Figure 3-4). The most parsimonious model included additive spawner biomass and year effects (Table

PAGE 61

61 3-4). The point estimate of (maximum pre-recruit survival at low spawner abundance) was 0.07 at the maximum likelihood estimates for recr uitment and spawner biom ass from the lengthage model. Lake-specific point estimate s of the maximum lifetime reproductive rate, were 10.0 at Lake Dora, 9.2 at Lake Eustis, and 6.9 at Lake Harris. Variability in among lakes was due to variation in equilibrium lifetime spawners per recruit,0 among lakes (Lake Dora: 153.8; Lake Eustis: 140.7; Lake Harris: 106.4). This va riation resulted primarily from differences in length at maturity, with Lake Dora having the yo ungest length at 50% matu rity and therefore the largest0 which resulted in a large estimate. The point estimates of were higher than the average estimates across the posterior parameter samples from the MCMC simulation of the length-age model (Lake Dora mean: 7.3; lake Eus tis mean: 7.0; Lake Harris mean: 5.5; Figure 35). Thus, values less than the point estimates were more likely than higher values across the posterior parameter samples from MCMC simulation. A value for of 1.0 results in a linear relations hip between pre-recr uit survival and spawner biomass and signifies a lack of density dependence in pre-recruit survival. Examination of 95% confidence intervals indicate d that a value of 1.0 was not c ontained in the interval for any lake, indicating that a lack of density dependenc e in juvenile survival was unlikely for these populations (Lake Dora 95%CI: 1. 9-16.5; Lake Eustis 95%CI: 1. 8-16.0; Lake Harris 95%CI: 1.4-13.3; Figure 3-5). Monte Carlo simulations used to assess potential biases in suggested that in the case of gizzard shad from these three lakes, time seri es bias and observation error in stock biomass would cause an underestimation of rather than overestimation. These estimates of the maximum lifetime reproductive rate should th erefore be viewed as conservative, which is preferable to an overly-optimistic estimate from a management perspective.

PAGE 62

62Discussion There are many peer reviewed studies showi ng increased pre-recruit survival, increased growth and reduced size/age at maturity at low po pulation densities in fish es, but there is debate about the relative importance of these mechanisms in fish comp ensation. Density-dependent changes in juvenile survival ha ve been considered the primary mechanism for compensation in fish populations (Rose et al. 2001). For exampl e, individual-based-model simulations by Cowan et al. (2000) indicated th at density dependent feedbacks on re cruitment are most likely during the late larval to early juvenile phase because of peak total cohor t consumption rates during that phase. Conversely, Lorenzen and Enberg (2002 ) suggested that density dependence in adult growth alone could explain observed compensation in 15 exploited fish populations. They further postulated that density dependence in growth may be most important under moderate reduction in density but that incr eased pre-recruit survival would be the dominant compensatory mechanisms at very low population sizes. My re sults disagree with Lo renzen and Enbergs (2002) findings and indicate that under a modera te change in population density, pre-recruit survival increased substantially at Lake Do ra whereas growth and maturation schedules remained relatively unchanged. This suggests th at changes in pre-recr uit survival may be important under moderate as well as se vere reductions in population density. Recent meta-analyses have made major advanc es in our understanding of compensation in fish populations. Myers et al. (1999) and Goodwin et al. (2006) estimated the maximum lifetime reproductive rate (i.e., compensation ratio), for 237 and 54 stocks of commercially exploited fishes, respectively. I calculated an average of 47 (95%CI: 10-84) ac ross all stocks included in both studies. Clupeids had below-average maximum reproductive rates at 19.3 (95%CI: 13.425) across stocks, and no gizzard shad stocks were included in their analyses. The mean estimate

PAGE 63

63 for clupeids was greater than the upper 95% c onfidence interval for the maximum lifetime reproductive rate for gizzard shad from my st udy. Gizzard shad may have relatively weak compensation when compared to other clupeids, or alternately, the estimates from other species could be biased. For example, most estimates of come from stock recruitment data generated from stock assessment models, which contain subs tantial uncertainty and possible biases from serial autocorrelation, error in spawner biomass estimates, and lack of contrast in the data (Walters and Martell 2004). I was able to use an experimental density reduction with control systems to evaluate and quantified the uncertainty in this parameter, which may provide less biased estimates than those obtained from traditional stock assessments. Goodwin et al. (2006) identifie d associations betw een life history characteristics and the strength of compensation. They found that fishes fall along a conti nuum of long-lived highlyfecund species with low annual recruitment and strong compensa tion (survivors, eg., sturgeon) to short-lived, early-maturing species with high annua l recruitment and weak density dependence (;highly productive e.g., clupeids). The survivors group exhi bits a bet hedging strategy to reproduce over many years wherea s the survivors are adapted to quickly invade and exploit highly variable resources (Stearns 1992). My data suggested that gizzard shad fall toward the highly productive end of the spectrum with fast growth, early maturation, and relatively weak density dependence in recruitment compensation. My study assessed the relative importance of density dependence of several demographic rates, but was unable to assess specific mechan isms influencing changes in those rates. I observed increased pre-recruit survival following density reduction but th is change could have been due to several mechanisms. Walters and Ju anes (1993) proposed that reduced survival at high juvenile densities results fr om increased risk taking at sma ll spatial and temporal scales by

PAGE 64

64 individuals attempting to procure scarce resources in a competitive environment. For example, juveniles may be forced to leave food-poor re fugia and spend more time in predator-dense feeding zones in order to maintain adequate growth rates when density of conspecifics is high. Density dependent growth rates of pre-recruits may also affect survival ra tes. Numerous studies have shown increased predation ri sk for slower growing individuals within a cohort (reviewed by Sogard 1997). The bigger is better hypothesi s (Shepherd and Cushing 1980) proposes that larger age-0 individuals have lo wer rates of mortality, because faster growth decreases the duration of exposure to stages where mort ality is high (Houde 1987; Miller et al. 1988; Hovenkamp 1992; Sogard 1997). Additional growth and survival studies on pre-recruits are needed within the context of experimental dens ity reduction to evaluate mechanisms for densitydependent survival that could not be addressed by my study. I expected adult demographic rates such as maturity and growth to respond following density reduction. Substantial research has shown changes in th ese rates with changing density. Age at maturity, for example, generally decrea ses with increased expl oitation (Trippel 1995). Rjinsdorp (1993) considered the ex ploitation of the North Sea plaice Pleuronectes platessa an experimental density manipulati on and reported decreases in lengt h and age at maturity since 1900, and Beacham (1983) documented decreased age and size at maturity following exploitation. Shifts in maturation schedules may be controlled by f eeding conditions during nutritionally limited periods in ga metogenesis (Burton 1994). Popul ations with size-dependent maturation schedules may also undergo changes in ag e at maturity via increases in growth rate (Trippel 1995). Somatic growth typically increases when populat ion density decreases due to decreased intraspecific competition. Many studies have documented increases in growth related to exploitation of fish stocks (Millner and Whiting 1996; R ijnsdorp and van Leeuwen 1996;

PAGE 65

65 Helser and Almeida 1997). Kim and Devries (2 000) reported substantia l increases in age-0 gizzard shad growth following density reduction at Walker Count y Lake, Alabama, and Schaus et al. (2002) found increased grow th of gizzard shad at Acton Lake, Ohio, in years with lowpopulation density. Thus, gizzard shad have clearly exhibited plasticity in growth in other systems. The strength of manipulation s hould be a consideration in an y whole-lake experiment and researchers should strive for larg e perturbations to elicit system responses (Carpenter 1989). In my study, size selective remova l of gizzard shad reduced spawner biomass by approximately 70%. This corresponds to a spawning potential ratio (SPR) of 0.3, which would put many species at risk for recruitmen t overfishing (Mace 1994; Clark 2002). However, changes in total population biomass were modera te (30%; Catalano et al. in review ) due to high estimated natural mortality which caused a large proportion of the population to resi de in young age classes that were invulnerable to harvest. Contrast in total population biomass was less than contrast in spawner biomass, which may have dampened grow th and maturation responses. The change in total population biomass may not have been enou gh to elicit strong responses in growth and maturation. Thus, the lack of change in growth and maturation may have been an artifact of the relatively weak total density reduction. Nevertheless, the experiment resulted in a substantial reduction in spawner biomass, which allowed estimation of density-depen dent changes in prerecruit survival. Future dens ity reduction studies should ach ieve stronger total biomass reductions so that changes in all de mographic rates can be evaluated. There are few examples of whole-lake densit y manipulations to test fish compensatory responses. Healey (1978; 1980) manipulated lake trout ( Salvelinus nameycush ) and lake whitefish densities in a Canadian shield lake and reported compensatory changes in growth,

PAGE 66

66 recruitment, and fecundity, but the magnitude of these changes was not proportional to the amount of density reduction. Experiment ally-manipulated m ountain brook trout ( Salvelinus fontinalis ) populations have provided good insights into compensation. DeGisi (1994) manipulated seven lakes in the Sierra Neva da Mountains, California, and found that the maximum lifetime reproductive rate in these pop ulations was approximately 19 (Myers 2002). At a smaller scale, Silliman and Gutsell (1958 ) and Silliman (1968), experimentally reduced laboratory populations of guppies (Lebistes reticulatus) and examined trends in abundance and yield at varying exploitation a nd feeding rates. They found dome-shaped relations between exploitation rate and yield that conformed to theoretical compensa tory predictions, but they did not explicitly assess the streng th of density dependence (Beverton and Holt 1957). However, results of these small-scale studies may not be appropriate for making conclusions about responses at larger scales and in more comple x environments (Schindler 1998). Experimental manipulation of the Lake Dora gizzard shad pop ulation contributes subs tantially to the body of experimental research on compensation and is uni que in that changes in demographic rates in the adult as well as pre-recruit phase were assessed to evaluate the relative importance of the two life stages in density dependence.

PAGE 67

67 Table 3-1. Parameter estimates (95% confidence intervals) from the length age model for lakes Dora, Eustis and Harris. Parameter Dora Eustis Harris M 0.94 (0.86 1.02) 1.02 (0.92 1.13) 1.07 (0.99 1.16) K 0.61 (0.60 0.63) 0.70 (0.66 0.74) 0.76 (0.73 0.79) L 387.89 (385.08 390.73) 406.85 (401.87 411.89) 389.67 (386.91 392.45) t0 0.16 (0.13 0.18) 0.34 (0.29 0.40) 0.39 (0.34 0.44) 1 30.86 (30.43 31.29) 33.80 (32.86 34.77) 30.39 (29.17 31.66) 2 0.06 (0.04 0.09) -0.19 (-0.24 -0.14) -0.11 (-0.16 -0.05) L 50UF 462.45 (445.64 479.89) 440.15 (428.78 451.82) 471.28 (460.62 482.19) UF 0.84 (0.71 0.97) 0.65 (0.56 0.74) 0.81 (0.62 1.00) UF 0.09 (0.04 0.2) 0.06 (0.05 0.08) 0.14 (0.05 0.40) L 50SJRWMD 319.46 (312.66 326.42) 395.05 (374.20 417.07) SJRWMD 0.00 (0.00 0.00) 0.00 (0.00 0.00) SJRWMD 0.04 (0.03 0.04) 0.03 (0.03 0.03) R 0.61 (0.40 0.92) 0.77 (0.50 1.18) L 5005 357.80 (340.36 376.13) 05 0.04 (0.03 0.06) 05 0.00 (0.00 0.00) L 5006 300.86 (297.06 304.70) 06 0.15 (0.11 0.21) 06 0.00 (0.00 0.00)

PAGE 68

68 Table 3-2. Delta AIC values for competing mo dels describing associat ions between growth increments and age, lake, and population density (i.e., total population biomass). Model AIC AIC age 5.5 0.0 age+lake 7.2 1.7 age+density 7.4 1.9 intercept only (null) 130.7 125.2 Table 3-3. Delta AIC values for competing mode ls describing associations between gizzard shad maturity and lake, population density (Bt), cohort size, year, and cohort. Model AIC AIC length+lake+density 614.1 0.0 length+lake+density+length:lake 615.4 1.3 length+lake+density+cohort size 616.0 1.9 length+lake 617.1 3.0 length+lake+cohort size+lengthlake 617.2 3.1 length+lake+cohort size 617.4 3.3 length+lake+lakelength 618.5 4.4 length+lake+cohort size+lengthlake 618.5 4.4 length+density 662.9 48.8 length 664.1 50.0 length+cohort size 664.2 50.1 intercept only (null) 1501.5 887.4 Table 3-4. Delta AIC values for competing mode ls describing associations between gizzard shad pre-recruit survival and spaw ner biomass (SB) and year. Model AIC AIC SB+year 26.0 0.0 SB 31.9 5.9 intercept only (null) 37.6 11.6

PAGE 69

69 19982000 2002200420062008 02468 a 19982000 2002200420062008 0123 b 19982000 2002200420062008 01234567 cRecruitmentYear Figure 3-1. Annual age-1 recruitment estimates (+ /95%CI) for lakes Dora (a), Eustis (b) and Harris (c) from the length-age model.

PAGE 70

70 100150200250300350400450 0.0 0.2 0.4 0.6 0.8 1.0 a 2003 2004 2005 2006 100150200250300350400450 0.0 0.2 0.4 0.6 0.8 1.0 b 2003 2004 2005 2006 100150200250300350400450 0.0 0.2 0.4 0.6 0.8 1.0 c 2003 2004 2005 2006 Total Length (mm)Probability Figure 3-2. Cohort-specific maturity ogives for gizzard shad at lakes Dora (a), Eustis (b) and Harris (c) with respect to fish length.

PAGE 71

71 200320042005200620072008 0.00.51.01.5a 200320042005200620072008 0.00.51.01.5b 200320042005200620072008 0.00.51.01.5cSpawner BiomassYear Figure 3-3. Time series of predicted spawner bioma ss for 2003 2008 at lakes Dora (a), Eustis (b), and Harris (c) from the length-age model.

PAGE 72

72 0.40.60.81.01.21.41.6 -5.5-5.0-4.5-4.0-3.5 Spawner Biomassln Pre-recruit Survival Dora Eustis Harris2003 2004 2005 2006 2007 2008 Figure 3-4. Loge pre-recruit surv ival as a function of spawner bi omass at lakes Dora (circles), Eustis (triangles) and Harris (plus). Surv ival and spawner biomass observations are point estimates from the length-age model. Cohort years are indicated for the Lake Dora cohorts.

PAGE 73

73 0510152025 0.00 0.05 0.10 0.15 ^Density Dora Eustis Harris Figure 3-5. Kernel density of maximum lifetime reproductive rate estimates for lakes Dora (solid line), Eustis (dashed line) and Harri s (fine dashed line). Estimates were obtained by repeated fitting of a stock-re cruit model across 1,000 parameter sets of the length-age model that were simulated from the posterior parameter distributions using Markov Chain Monte Carlo simulation.

PAGE 74

74 CHAPTER 4 EXPLORING FISH REMOVAL STRATE GIES FOR BIOM ANIPULATION THAT ACCOUNT FOR UNCERTAINTY IN THE STRENGTH OF DENSITY DEPENDENCE OF TARGET SPECIES Introduction Fish biomanipulation has been used to re duce phytoplankton biomass and improve water transparency in eutrophic lakes via increased grazing pressure by zooplankton on phytoplankton (Shapiro et al. 1975; DeMelo et al. 1992; Ha nsson et al. 1998; Dre nner and Hambright 1999; Meijer et al. 1999). This has been achieved by addition of predatory fish species to increase predation on planktivorous fish or direct removal of planktivores themselves (Carpenter et al. 1987; Meronek et al. 1996; Drenne r and Hambright 1999). Direct planktivore removals typically are achieved with a large-scale commercial fish ery to rapidly and efficiently remove large amounts of planktivore biomass (Drenner and Ha mbright 1999). Consequently, the population dynamics of planktivorous fish species may have important consequences for the efficacy of direct planktivore removals (Romare and Bergman 1999). Compensatory density dependence is an impor tant life history characteristic of fish populations. The strength of compensation determines a populations ability to withstand increased mortality rates and therefore defines th e limits of harvest (Myers et al. 1999), which could have implications for biom anipulation fisheries. Bioman ipulation fisheries could have unintended consequences for lake ecosystems depending on the shape and strength of density dependence of the target species. For ex ample, fish species with dome-shaped overcompensatory relationships (e .g., Ricker recruitment) between spawner abundance and age-1 recruits could become more abundant following moderate removals that reduce the population size to a state of optimal pr oductivity (Zipkin et al. 2008). Thus removals may release populations from density-dependent suppression of recruitment due to competition between

PAGE 75

75 adults and juveniles. Such compensatory responses could lead to increased rather than decreased grazing of zooplankton, which would have the opposite effect on phytop lankton abundance that is desired (Romare and Bergman 1999). Species with asymptotic relationships between spawner biomass and recruitment (e.g., Beverton-Holt recruitment) may maintain relatively constant recruitment despite reductions in spawner biom ass, which would reduce the efficacy of removal efforts. Accounting for the strength of density depende nce of target organisms in biomanipulation studies could help improve the efficacy of remo val programs. Such studies could guide removal strategies by suggesting removal methods that achieve maximum density reduction or recommending the discontinuation of programs that are unlikely to achieve large enough biomass reductions to reduce phytoplankton biomass. In this study, I evaluated the efficacy of removal strategies for biomanipulation while accoun ting for uncertainty in the strength of density dependence of the target species. I eval uated the effect of exploitation rate (u ), gill net mesh size, and harvest interval (years between rem ovals) on total population biomass and spawning potential ratio (SPR) of the target species. My study species was the gizzard shad in hypereutrophic Florida lakes, and this species has been the target of biom anipulation in eastern North America (DeVries and Stein 1990; Kim and DeVries 2000; Catalano et al. in review ). Study Site Gizzard shad biomanipulation was conducted at Lake Dora, Lake County, Florida, USA. Gizzard shad populations at two nearby unmanipulat ed lakes were also studied to serve as unfished control populations. The lakes were part of the Harris Chain of Lakes (HCL) in central Florida, USA. The lakes were shallow (3 m average depth) hypereutrophic lakes (mean chlorophyll a concentrations > 50 ug /L) with similar fish communitie s. Biomanipulation at Lake Dora was achieved with a government-subsidi zed commercial gill net fishery by the regional

PAGE 76

76 water management agency, the St. Johns River Wa ter management District (SJRWMD). Prior to fish removal, gizzard shad populations in all la kes were unfished. Comm ercial fishers removed gizzard shad from Lake Dora during MarchApril 2005 and again during January-April 2006. Gizzard shad were removed using gill nets with a minimum mesh size restriction of 102 mm, which selected for gizzard shad larger than approximately 300-mm total length. Removal was carried out by commercial fishers, with an averag e of five boats setting 3-5 sinking gill nets per day, each net ranging in length from 75 to 600 m. The total harvest was 125,000 kg in 2005 and 135,000 kg in 2006. Catalano et al. ( in review ) estimated exploitation rates ( u ; annual proportion of vulnerable sized fish removed) of 0.71 in 2005 and 0.65 in 2006. The total biomass reduction was estimated at 30% through the two years of fishing (Catalano et al. in review ). Methods I evaluated the influence of exploitation rate, gi ll net mesh size, and harvest interval on the percent total biomass reduction and spawning poten tial ratio of gizzard shad at the HCL using a simulation model. Percent biomass was ev aluated to assess the degree to which a biomanipulation target of 75% biomass reduction (Meijer et al. 1999) was met by a given harvest strategy. Spawning poten tial ratio (SPR) is a measure of th e potential spawner biomass under a given harvest rate relative to the unfished condition and was used to assess the potential for recruitment overfishing (Mace 1994). Simulated population responses to the harvest regime accounted for uncertainty in the strength of density dependence for gizzard shad via a parametric bootstrap procedure. Gear Selectivity Understanding gear selectivity is essential for simulating potential effects of harvest on fish populations. I evaluated the efficacy of five gillnet mesh sizes: 51, 64, 76, 89, and 102 mm for gizzard shad removal. Estimates of a gear se lectivity function were ne eded for each of these

PAGE 77

77 mesh sizes as input parameters for the simulation model. Catalano (Chapter 2) estimated survey and fishery gear selectivity for the gizzard shad removal at Lake Dora. However, the gizzard shad removal had a minimum mesh size restrictio n of 102 mm and consequently gear selectivity for smaller mesh sizes was unknown and could not be estimated. Survey gear selectivity estimates from Catalano (Chapter 2) were also not useful because the survey nets had multiple panels of different mesh sizes and the estimated selectivity function applied to all of the panels collectively. To obtain gear selectivity estimates for each mesh size, I estimated the selectivity parameters using a length and age structured pop ulation model (see Chapters 2 and 3 for model details). Catalano (Chapter 3) es timated natural mortality, growth and recruitment time series for gizzard shad at lakes Dora, Eustis, and Harris. Using these parameter estimates as model inputs, I estimated gear selectivity for the 51 102 mm mesh sizes by fitting the model to lengthspecific gill net catch data from UF and SJRWMD annual January fishery-independent gill net surveys. The survey gill net data were sepa rated by mesh size, and a three parameter gear selectivity function was fit to the data from each mesh size to estimate mesh-specific gear selectivity parameters. This approach assumed that the point estimates of growth, mortality, recruitment that were used as model inputs were the true values for the lakes and thus the estimates of mesh-specific gear selectivity did not account for uncerta inty in these input parameters. Lake-specific gear selectivity parameters were not estimated because I was interested in obtaining average gear selectivity curv es across all of the la kes for use as inputs in the simulation model. Therefore the simulation model represented a generic system with similar fishery characteristics to the Harris Chain of Lakes.

PAGE 78

78 Gill net catch data were obtained from a nnual fishery-independent gill net surveys conducted by the author (UF) and SJRWMD in Ja nuary/February at each lake. The UF survey set multi-panel floating gill nets at 20 fixed randomly-selected sites at each lake in January/February (all lakes: 2005 2009). The SJ RWMD survey set multi-panel gill nets at 10 fixed sites at lakes Eustis a nd Harris and 20 sites at Lake Do ra (Lake Dora: 2003, 2005 2009; Lake Eustis: 2003, 2006-2009; Lake Harris: 2003). Survey gill nets were 2.4-m deep and contained five, 15.3-m long panels of 76, 89, 102, 114, and 127-mm stretc h monofilament mesh and nets were set for 2 hours each. The UF gill nets had an additional panel of 51-mm mesh to target age-1 fish. All captured fish were measured in the field and placed in 10-mm length bins to construct length distributions to wh ich the length-age model was fit. Uncertainty in Density Dependence The maximum lifetime reproductive rate is a standardized measure of the strength of density dependence for a given fish population an d is comparable across species (Myers et al. 1999; Goodwin et al. 2006). This parameter is also known as the Goodyear compensation ratio and describes the ratio pre-recruit survival at very low population density to pre-recruit survival in an unfished population (Goodwin et al. 2006; Walters et al. 2006). An estimate of the mean and standard deviation of for gizzard shad was obtained from Catalano (Chapter 3). Catalano (Chapter 3) found that the average for gizzard shad was 6.6 with a 95% confidence interval of 1.7 to 15.2 across lakes Dora, Eustis, and Harris, and this degree of compensation was due primarily to density-dependent changes in pre-recr uit survival and not to changes in growth or maturity. Uncertainty in estimates were used to estimate uncertainty in population biomass and SPR as a function of gill net mesh size, exploitation rate and harvest interval using a simulation model.

PAGE 79

79Simulations I constructed a simple population model to si mulate the efficacy (i.e., percent biomass reduction and SPR) of gizzard shad removal over a ra nge of exploitation rate gill net mesh size, and harvest interval. The model was of the form: )1(1,1 ta M tatauveNN (4-1) where Na,t is the number of fish in the population at age a in year t M is the instantaneous natural mortality rate, va is the age-specific gear selectivity term ranging from 0 to 1, and ut is the finite annual fishing mortality rate. Fishery gear selectivity, va, was estimated using the function (Thompson 1994): )( )(50 501 1 1 1LL LL ae e v (4-2) where L is the mean length at age a from the von Bertalanffy growth model, is the shape parameter that determines the shape, describes the steepness, and L50 is the length at 50% selectivity. This is a flexible selectivity f unction that produces either a dome shaped or sigmoidal curve, depending on parameter values. Values of are bounded between 0 and 1. The functional form becomes sigmoidal (i .e., knife edge selectivity) as approaches 0 and increasingly dome-shaped as approaches 1. Estimated selectivity functions that were dome-shaped were converted to asymptotic curves by setting equal to zero (Figure 4-1). This was done to mimic occasional fisher use of larger mesh sizes than the minimum to exploit la rge individuals that were invulnerable to small mesh sizes due to dome-shaped selectivity. Using the dome-shaped selectivity curves would have been unrealistic because fishers would not have used a small mesh if large fish were available in the population and coul d be caught with a larger mesh.

PAGE 80

80 The equilibrium model had deterministic recr uitment predicted as a function of spawner biomass using the Beverton-Holt stock-recr uitment model (Walters et al. 2006): 1 00 1 0 ,11 1 t t tE R E N (4-3) where 0 is the equilibrium lifetime spawner biomass per recruit in the absence of fishing, R0 is the equilibrium unfished age-1 recrui tment set to a value of 1, and Et-1 is the total population spawner biomass from the previous year. The Be verton-Holt model is an asymptotic model such that recruitment is relatively constant across a wide range of spawner biomass. This is a different model than the Ricker f unction that was used to obtain the estimates (Catalano, Chapter 3). Few stock-recruit data sets contain enough observati ons at extremely high spawner biomass to differentiate between Ricker and Beverton-Holt models (M yers et al. 1999). Estimates of from the Ricker model are relatively r obust to varying assumptions regarding the shape of the function (asymptotic or dome-shaped) and should be appropriate for use in the Beverton-Holt model (Myers et al. 1999). Stock-recr uit data from Catalano (Chapter 3) were too sparse to differentiate between the two models. In the absence of knowledge on the shape of the stock recruit function for gizzard shad at the Harr is Chain of Lakes, the Beverton-Holt model is appropriate given the life histor y characteristic of gizzard shad. Beverton-Holt recruitment dynamics are typically associated with pelagic fish with plankti vorous diets and ontogenetic diet or habitat shifts such that negative interactions between adults and juveniles are weak. As such, the Beverton-Holt model is appropriate for simu lating gizzard shad population dynamics and is a more conservative approach because it does not allow for overcompensatory recruitment dynamics such as increased recruitment followi ng moderate reduction in spawner biomass.

PAGE 81

81 Equilibrium lifetime unfished spawner biomass per recruit was calculated as: a aaawms0, (4-4) where sa is the survivorship to age a ma is the age-specific propor tion of fish mature, and wa is the age-specific mean weight. Maturity was estimated using the mean age at 50% maturity from lakes Dora Eustis and Harris from Catalano (Chapt er 2). Weight at age was estimated using the allometric relationship: b acLw (4-5) where c and b are allometric coefficients obtained by f itting the model to gizzard shad collected in UF survey gillnets from 2005-2009. Annual spawner biomass Et was estimated as: a aaa twmNE (4-6) The model predicted total popul ation biomass and spawning pot ential ratio (SPR) as a function of exploitation rate, gill net minimum allowable mesh size, and harvest interval (number of years between harvests). Ot her model inputs were set based on literature values. The model simulated eight age classes and an average unfished recruitment ( R0) of 1.0. Instantaneous natural mortality ( M ), asymptotic length (L), the von Bertalanffy me tabolic coefficient ( K ), and time at zero length ( t0) were taken from Catalano (Chapter 3). Each parameter value was obtained by averaging over lake-specific estimates from lakes Dora, Eustis, and Harris ( M = 1.01 yr-1, L = 394, K = 0.69, t0 = 0.3). Length at maturity was obtained from a l ogistic regression model by Catalano (Chapter 3) a nd was used to estimate the pr oportion of females mature at each age ( ma).

PAGE 82

82 The maximum lifetime reproductive rate is an important term because it defines the degree of compensation in the popula tion and thus determines the lim its of harvest. Populations with high will maintain relatively constant recruitment across a wide range of adult population sizes (i.e., large declin es), compared to low which indicates that reductions in adult population sizes cause declines in aver age recruitment. Thus, I varied using the uncertainty from Chapter 3 in a parametric bootstrap analysis. For each combin ation of exploitation rate (0.1 to 1.0 by 0.1), mesh size (51, 64, 76, 89, 102 mm) and harvest interval (every 1, 2, 3 and 4 years), I drew 1,000 random lognorma lly-distributed deviates for assuming a mean of 6.6 and standard deviation of 3.7 from Catalano (Chapter 3). I calculated tota l population biomass and SPR for each value of across each possible combination of ex ploitation rate, gill net mesh size, and harvest interval. Biomass and SPR for each value was calculated by averaging the last 50 model years after a 150-yr burn-in period to a llow the population to reac h equilibrium. The average equilibrium population biomass and SPR was calculated for each possible combination of harvest frequency, mesh size, and exploi tation rate by averagi ng over the 1,000 bootstrap estimates. Uncertainty in biomass and SPR was estimated by calculati ng the 2.5% and 97.5% quantiles of biomass and SPR across the 1,000 bo otstrap estimates. Total population biomass was calculated as: a aawNB (4-7) SPR was calculated as: ffR R SPR 00 (4-8) where Rf and f are the equilibrium recruitment and spawner biomass per recruit, respectively, under a given harvest scenario.

PAGE 83

83 To further explore uncertainty in biomass and SPR, I calculated the probability that a given harvest regime would result in a biomass that was less than 25% of the equilibrium unfished value (i.e., 75% biomass reduction) and an SPR of less than 25%. The target level of 75% reduction in total gizzard shad biomass was used to indicate harvest st rategies (i.e., fishing frequency, gill net mesh, and exploitation rate) that achieve rates likely to cause changes in lake phytoplankton abundance (Hansson et al. 1998; Meijer et al. 1999). Fishing mortality rates that result in SPR less than 40% increase the risk fo r recruitment overfishing (i.e., fishing at a rate that prevents a stock from replacing itself; Mace 1994), but this cutoff may be lower for highly productive species such as the gizzard shad (Cla rk 2002). Thus, I chose 25% as a target SPR to indicate which harvest scenarios presented th e greatest probability of causing recruitment overfishing for gizzard shad since gizzard shad are short lived and likely to withstand substantial harvest because of high natural mo rtality and rapid growth rates. Results Gear selectivity functions were dome-shaped for the 51 to 76-mm gill net mesh and were sigmoidal for the 89 and 102-mm mesh (Figure 4-1a). Lengths at 50% selectivity (L50) ranged from 166 mm for the 51-mm mesh to 339 mm for th e 102-mm mesh (Figure 41a,b; Table 4-1). Predicted catches of gizzard shad tightly fit the observed length distributions for each mesh size (Figure 4-2). Equilibrium population biomass was sensitive to changes in gill net mesh size. The 51mm mesh and a one-year harvest interval dr ove the population to extinction when the exploitation rate exceeded 0.8 (Fi gure 4-3a). None of the other mesh sizes reduced the average population biomass to less than 35% of the unfishe d value even at an exploitation rate of 1.0 (Figure 4-3a,b,c). When account ing for uncertainty in the maxi mum lifetime reproductive rate, the probability of reducing the population biomass to less than 25% of the unfished value was

PAGE 84

84 less than 0.15 for all meshes except 51-mm when the exploitation rate was 0.8 or less (Table 42). The 102-mm mesh was the least effective at reducing biomass; the average biomass was 72% of the unfished value at an exploitation rate of 1.0 (Figure 4-3e) and the probability of achieving a biomass of 25% was 0.01 at an exploita tion rate of 0.8. Biomass remained greater than 50% of the unfished value for all mesh si zes except 51-mm when the harvest interval was two or more years, regardless of exploitation rate. Equilibrium SPR was reduced more than total population biomass due to the size-selective nature of the fishery. The 51-mm mesh and a one -year harvest interval reduced SPR to near zero when the exploitation rate ex ceeded 0.8 (Figure 4-4a), wh ich the model suggested would eliminate recruitment and drive the population to extinction. The 64, 78, and 89-mm mesh resulted in SPR of 30-50% (Figure 4-4b,c,d). The probability of reducing SPR to less than 25% exceeded 0.85 when exploitation rate exceeded 0. 2 for the 51-mm mesh (Table 4-3). The 64 and 76-mm mesh resulted in a less than 0.25 pr obability of an SPR dropping below 25% when exploitation rate was 0.8 (Tab le 4-3). The 102-mm mesh was ineffective at reducing SPR (Figure 4-4e). Reducing SPR to less than 25% was highly unlikely with a two-year harvest interval for all mesh sizes except 51 mm (Table 4-3). Discussion Gizzard shad removals at the Harris Chain of Lakes using gill nets are unlikely to achieve large (75%) reductions in bioma ss or SPR unless a 51-mm mesh si ze is used, a high exploitation rate is achieved, and fish are harvested every year Larger mesh sizes le ft a large proportion of the population biomass invulnerable to harvest due to the selective properties of the gear. High estimated natural mortality rate also dampened the effects of fishing on total biomass of this population because a large proportion of the population resided in young, invulnerable, age classes and most fish died naturally before they could be harvested. Failure to reduce SPR to

PAGE 85

85 less than 25% indicates that recruitment overfishing is unlikely in this system. Thus, recruitment failure is unlikely and the population would have to be harvested annually to maintain biomass reductions, which would in crease removal costs. These findings have implications for gizzard shad biomanipulation. The SJRWMD has been harvesting gizzard shad since 1995 to reduce phytoplankton biomass in hypereutrophic Florida lakes using a 102-mm me sh size restriction to reduce bycatch of black crappie, a recreationally important fish species. My findings s uggest that this approach is unlikely to attain substantial biomass reduction of gizzard shad in the long term given curren t exploitation rates of 60-70% (Catalano et al. in review ). Use of the smallest mesh size (51 mm) would increase biomass reductions and the likelihood for recr uitment overfishing but would also increase bycatch of black crappie, which could reduce th e value of an important recreational fishery (Dotson et al. In press ). In such situations resource mana gers and stakeholders will need to carefully explore the tradeoffs be tween gizzard shad biomanipulati on and black crappie fisheries. There are several assumptions of my analysis that should be addressed. I chose a 75% biomass reduction target from the literature because meta-analyses have shown that this level of reduction is associated with higher biomanipula tion success rates (Hansson et al. 1998; Meijer et al. 1999). However, these studies are based primar ily on planktivore removals. Gizzard shad are omnivores capable of consuming zooplankton as well as benthic organic detritus. Gizzard shad benthivory may provide a source of new nutrients to the phytopl ankton that were previously unavailable in the sediments (Schaus et al. 1997; Gido 2002). Thus, gizza rd shad may affect phytoplankton biomass via top-down grazing and bottom-up direct nut rient enrichment (DeVries and Stein 1992). Horpilla et al (1998) reported substantial reduction in phytoplankton biomass following 79% biomass reduction of omnivorous roach ( Rutilus rutilus ). The biomass reduction

PAGE 86

86 that would reduce phytoplankton bi omass at Lake Dora is unknow n and may be more or less than the 75% target identified from plankti vore removals. My data do not address the applicability of this valu e to Lake Dora. However, my data suggest that long-term total gizzard shad biomass reductions are unlikely to exceed 40 -50% at Lake Dora or similar lakes without substantial increases in the exploitation ra te and decreases in gill net mesh size. Second, I accounted for uncertain ty only in the strength of de nsity dependence. The results of the simulation model are dependent on many ot her parameters such as growth and natural mortality, which I assumed were known without erro r in the model. This approach isolated the effects of recruitment compensation on harvest po licies, which in this case was desirable given that I only found changes in juvenile fish survival after fishing. However, fish stock assessment models can be very sensitive to error in natu ral mortality estimates (Mertz and Myers 1997; Clark 1999; Quinn and Deriso 1999). Upward bias in natural mortality in my analysis would underestimate biomass reduction because the model would overestimate the number of fish dying naturally before reaching a harvestable size. Similarly, overestimates of growth rates would underestimate biomass reduction and SPR b ecause fish would reach a larger size more rapidly, which would increase stoc k productivity. Nevertheless, in corporating uncertainty in the strength of density dependence is an advance ove r previous analyses of fish removals and the gizzard shad population at the HCL was sensitive to the assumed value for For example, the estimated biomass reduction for the 76-mm mesh at an exploitation rate of 1.0 and an annual harvest interval ranged from 0.18 to 0.7, depending on the assumed value. Third, I assumed gear selectivity functions we re asymptotic when in fact each mesh size had a dome shaped gear selectivity curve. Using the estimated dome-shaped selectivity functions would have been unrealis tic because it is unlikely fisher s would have exclusively used

PAGE 87

87 a particular mesh size if they had the option to use larger mesh sizes. By assuming asymptotic selectivity functions, I assumed th at fishers would occasionally choos e to use large mesh sizes to exploit fish that had grown large enough to es cape minimum mesh size, thus resulting in approximately asymptotic selectivity. This was an attempt to mimic the process of fishers fishing down the population and eventually settling on the sma llest mesh size allowed after catches in large meshed declined. Moreover, allowing the dome-shaped selectivity function for each mesh would have underestimated the biomass reduction for the smallest mesh sizes and at lower exploitation rates because large fish would have been invulnerable to capture. As such, my gear selectivity assumptions were reasonable and provided the most realistic estimates of biomass reduction and SPR. My simulations showed that the 51-mm mesh wa s most likely to cause substantial biomass reduction but it is not known whether that mesh size would be acceptable to gill net fishers. Smaller mesh sizes are more labor-intensive to process and bycatch of undesirable species would have increased (Dotson et al. In press ). If biomass were reduced substantially and SPR was reduced enough to cause recruitmen t failures, then catches would decline drastically. In this case, it would be difficult for fishers to mainta in adequate catches to cover costs and they may choose to use a larger mesh size or leave the fish ery. Increased price subsidy may be required as catches decline to maintain high harvest rates. Populations of species like the gizzard shad would likely recover rapidly if exploitation were relaxed. Indeed, biomanipulation programs often must be continued indefinitely to maintain changes in phytoplankton biomass due to planktivore removal (McQueen 1998). The strength and functional form of compensatory density dependence in a population defines the limits of harvest and has important im plications for removal efforts. Zipkin et al.

PAGE 88

88 (2008) found that experimental removal of sma llmouth bass at Little Moose Lake, NY, resulted in increased recruitment, suggesting a Ricker form of the stock-recruitment relationship. They identified several mechanisms that could explain the increase such as high per capita recruitment at low population size combined with high juveni le survivorship and high maturation rates of age-4 smallmouth bass. Meijer et al. (1999) re ported increased age-0 fish abundance following several fish removals. Kim and DeVries (2000) found strong compensatory growth and maturation of gizzard shad at Walker Count y Lake, Alabama following partial piscicide treatment. In their study, mean length of age-0 gizzard shad in fall was >60% larger at low gizzard shad densities than at high densities. These compensato ry responses allowed the gizzard shad population to return to pre-manipulation abu ndance within one year of treatment. Romare and Bergman (1999) reported a 20-fold increase in juvenile fish abundanc e following planktivore removal at Lake Ringsjn, Sweden. Thus, compen satory responses of target species have been observed following fish removals. Despite the importance of compensation in the efficacy of fish removals, only one study to date (Zipkin et al. 2008) has expl ored the potential eff ects of compensation. The strength of manipulation is a key consideration for bioman ipulation and many studies fail to adequately address this issue (DeVries and Stein 1990). At minimum, the degree to which population biomass was reduced should be quantified to as sess biomanipulation strength. However, this analysis is not completed in many cases, thus le ading to uncertainty in mechanisms that impact the results of biomanipulation e fforts (Meronek et al. 1996). Id eally, simulations such as those presented here should be carried out prior to manipulation to de termine the levels of biomass reduction that could be expected given current knowledge of mortality, growth, and gear selectivity. Such simulations should attempt to incorporate compensatory density dependence.

PAGE 89

89 One problem with this approach is that the strength of densit y dependence is often not known for a particular population and likely depends on many f actors that could be sp ecific to a particular system (Walters and Martell 2004). However, there is a growing body of knowledge on general patterns in density dependence across species that could be used to inform simulations that account for compensation (Myers et al. 1999; Goodwin et al. 2006). Goodwin et al. (2006) related the strength of density de pendence to life history characteris tics of 54 fish stocks. Thus, one could choose a range of values of based on these patterns.

PAGE 90

90 Table 4-1. Gear selectivity parameter estimate s (95% confidence interval) for each gill net mesh size from the from the length age model. Po int estimates of L50 and b were used in the simulation model to evaluate competing removal scenarios with varying exploitation rate, mesh size, and harvest in terval. Parameter subscripts denote the mesh size. Parameter Estimate L5051 165.86 (163.99 167.74) 51 0.07 (0.05 0.10) 51 0.28 (0.22 0.35) L5064 217.25 (214.26 220.28) 64 0.13 (0.11 0.16) 64 0.13 (0.12 0.15) L5076 251.32 (247.84 254.84) 76 0.05 (0.03 0.07) 76 0.10 (0.09 0.11) L5089 290.27 (287.92 292.64) 89 0.00 (0.00 0.00) 89 0.07 (0.07 0.08) L50102 340.09 (336.59 343.62) 102 0.00 (0.00 0.00) 102 0.05 (0.05 0.06)

PAGE 91

91 Table 4-2. Probability that to tal population biomass is less than 25% of equilibrium unharvested value for a one and two year harvest in terval, a range of exploitation rates ( ), and five gill net mesh sizes ranging from 51 to 102 mm. Gill Net Mesh Size Interval (yrs) 51647689 102 1 0.4 0.440.030.020.01 0.00 0.6 0.970.070.050.02 0.00 0.8 1.000.140.090.03 0.01 2 0.4 0.040.010.000.00 0.00 0.6 0.120.020.020.01 0.00 0.8 0.430.060.040.02 0.00 Table 4-3. Probability that transitional spawning potential ra tio (SPR) is less than 25% of for a one and two year harvest interval a range of exploitation rates ( ), and five gill net mesh sizes ranging from 51 to 102 mm. Gill Net Mesh Size Interval (yrs) 51647689 102 1 0.4 0.840.030.030.01 0.00 0.6 1.000.090.070.02 0.00 0.8 1.000.230.150.04 0.01 2 0.4 0.110.010.000.00 0.00 0.6 1.000.040.030.01 0.00 0.8 1.000.150.110.02 0.00

PAGE 92

92 100150200250300350400450 0.00.20.40.60.81.0 Total Length (mm)Gear Selectivity a 100150200250300350400450 0.00.20.40.60.81.0 Total Length (mm)Gear Selectivity b Figure 4-1. Estimated gear selectivity curves for 51, 64, 76, 89, and 102-mm gill net mesh sizes (panel a). The dome-shaped curves were converted to asymptotic functions for the simulation analysis to simulate the choice by fishers to occasionally use of larger mesh sizes than the minimum when large fi sh are available for capture. (panel b). Length at 50% selectivity is identical be tween the two panels for each mesh size.

PAGE 93

93 100200300400500 0.000.100.200.30 a observed predicted 100200300400500 0.000.100.200.30 b 100200300400500 0.000.100.200.30 c 100200300400500 0.000.100.200.30 d 100200300400500 0.000.100.200.30 eTotal Length (mm)Catch Proportion Figure 4-2. Observed (points) and predicted (lin es) length distributions of catches for each gill net mesh size. Observations and model predictions were summer over lakes and years.

PAGE 94

94 0.20.40.60.81.0 0.00.40.8 a ) 51 mm 0.20.40.60.81.0 0.00.40.8 b ) 64 mm 0.20.40.60.81.0 0.00.40.8 c ) 76 mm 0.20.40.60.81.0 0.00.40.8 d ) 89 mm 0.20.40.60.81.0 0.00.40.8 e ) 102 mmExploitation RatePopulation Biomass Figure 4-3. Total population biomass as a function of exploitation rate and gill net mesh size for an annual harvest interval. Panels are a rranged as follows: panel a, 51-mm mesh; panel b, 64-mm mesh; panel c, 76-mm mesh ; panel d, 89-mm mesh; panel e, 102-mm mesh. Dashed lines represent the 95% conf idence intervals obtained via parametric bootstrap of the maximum lifetime reproductive rate.

PAGE 95

95 0.20.40.60.81.0 0.00.40.8 a ) 51 mm 0.20.40.60.81.0 0.00.40.8 b ) 64 mm 0.20.40.60.81.0 0.00.40.8 c ) 76 mm 0.20.40.60.81.0 0.00.40.8 d ) 89 mm 0.20.40.60.81.0 0.00.40.8 e ) 102 mmExploitation RateSPR Figure 4-4. Spawning potential ra tio (SPR) as a function of expl oitation rate and gill net mesh size for an annual harvest interval. Panels are arranged as follows: panel a, 51-mm mesh; panel b, 64-mm mesh; panel c, 76-mm mesh; panel d, 89-mm mesh; panel e, 102-mm mesh. Dashed lines represent the 95% confidence inte rvals obtained via parametric bootstrap of the ma ximum lifetime reproductive rate.

PAGE 96

96 CHAPTER 5 SYNTHESIS AND FUTURE RESEARCH Gizzard sh ad populations appear to be regulated by density dependence in juvenile survival at least at the level of density reduction achieved by the experime ntal removal at Lake Dora. This is not surprising considering the abundance of literature suggesting that density likely plays a key role in the survival of pr e-recruits. One clear weakness of the study was the strength of the total biomass reduction. The reduction in sp awner biomass was substantial and provided good contrast in the data to evaluate changes in juvenile survival, but total biomass reduction was moderate due to the size-selective nature of the fishery and high es timated natural mortality rate. The lack of density dependence in adult growth and maturity may have been due to the relatively weak total density reduction. Fu ture studies should achieve str onger total bioma ss reductions to fully assess the relative importance of compensa tion among life stages. I would expect large total biomass reductions to result in changes in growth and maturity of gizzard shad based on data from gizzard shad biomanipulation in Alabama (Kim and DeVries 2000) and apparent density related effects on growth at Acton lake, Ohio (Schaus et al. 2002 ). Thus, I view my estimates of the strength of density dependence as conservative. Biomanipulation studies could provide info rmation on compensatory density dependence of target species. However, analysis of dens ity reduction levels and compensatory responses are rarely rigorous enough for such inve stigations as the focus of bi omanipulation is usually on the responses of phytoplankton and zooplankton communitie s. With some additional effort, I argue that more detailed data should be collected fr om these manipulations. These could provide a very large sample size of density reduction expe riments to make strong inferences about fish compensation.

PAGE 97

97 My study was not able to examine specific mechan isms for changes in pr e-recruit survival. Understanding these mechanisms would be in teresting from an ecological and population dynamics perspective. Future density reduction st udies should collect da ta on the juvenile life stage to assess possible mechanisms such as grow th or risk-sensitive foraging behavior. These studies will be difficult because of problems in captu ring small juvenile fishes in such a way that is comparable throughout the first year of life. This is challenging because juvenile fishes grow rapidly and therefore vulnerability to sampling gear s is constantly changing. In addition, these animals undergo complicated ontogenetic shifts in habitat and diet, and the processes controlling density dependence may occur at very fine spatia l and temporal scales. Controlled pond studies may provide the best means for studying mechan isms of juvenile fish density dependence. I introduced a maximum likelihood model to esti mate a recruitment time series along with mortality, growth, and gear selectivity. This model was useful for exploring gizzard shad population dynamics. The model provided an altern ative to other existing length-age fish stock assessment models. In addition, my analysis evaluated potential biases in parameter estimates, which to my knowledge has yet to be published for length-age structured models. Future work should compare estimates from my model with those of other length-a ge models such as Flexibest and Stock Synthesis 2. In addition, investigators shou ld assess model efficiency in terms of the amount of computation time required for estimation. My model was computationally demanding but likely could have b een run more efficiently with a program that uses automatic differentiation such as AD Model Builder (ADMB). ADMB is much more efficient than program R in estimating complicated models with many parameters. Thus, my model will likely perform better using the ADMB platform.

PAGE 98

98 I introduced a simple simulation study show ing how uncertainty in density dependence could be incorporated into forecasts of the efficacy of fish removal for biomanipulation. This study could serve as a guide for planning biom anipulation experiments. In my view, biomanipulation studies could do a better job of quantitatively assess ing how the proposed removal could affect the target population. Thes e modeling efforts could be extended further to include ecosystem dynamics to attempt to pred ict effects of biomanipulation on phytoplankton biomass and other ecosystem components. These models would represent testable hypotheses of system function and would promote more struct ured thinking regardi ng when, where and why biomanipulation could be successful. Current ecosystem models such as ECOPATH/ECOSIM could be used in these explorat ory modeling exercises. However, other models may need to be developed for omnivorous fish biomanipulation that incorporate bottom-up nutrient enrichment due to benthivory, as well as t op-down effects of zooplankton grazing by juvenile omnivores on phytoplankton dynamics. These dynamics could be quite complicated and therefore a modeling exercise would force investigators to clarify our understanding of system function as well as identify data needs that could resolve uncertainties. Few biomanipulation studies to date have used basic fish population dynamics methods to understand the effects of harvesti ng on target populations. Incor poration of estimates of gear selectivity and natural mortality could clarify how many fi sh should be removed to attain biomanipulation targets. Thes e studies should be conducted befo re biomanipulation occurs to ensure that funds are not wasted by removing only a relatively small part of the population, which would have minimal impacts on water qualit y. These analyses need not be complicated and could provide excellent guidance in planning of biomanipulation projects.

PAGE 99

99 LIST OF REFERENCES Allee, W C., A. E. Emerson, T. Park, and K. P. Schmidt. 1949. Principles of Animal Ecology. Saunders, Philadelphia. Allen, M. S., M. V. Hoyer, and D. E. Canfield, Jr. 2000. Factors related to gizzard shad and threadfin shad occurrence and abundance in Fl orida lakes. Journal of Fish Biology 57:291302. Baccante, D. A., and D. M. Reid. 1988. Fecundity changes in two exploited walleye populations. North American Journal of Fisheries Management 8:199-209. Bayley, P. B., and D. J. Austen. 2002. Capture effi ciency of a boat electrofisher. Transactions of the American Fisheries Society 131:435-451. Beacham, T. D. 1983. Variability in median size and age at sexual maturity of Atlantic cod, Gadus morhua, on the Scotian shelf in the northw est Atlantic Ocean. Fishery Bulletin 81:303-321. Bertalanffy, L. v. 1938. A quantitative theory of organic growth. Human Biology 10:181-213. Beverton, R. J. H., and S. J. Holt. 1957. On th e dynamics of exploited fi sh populations. Fisheries Investment Series 2, volume 19. U.K. Mini stry of Agriculture and Fisheries, London. Bodola, A. 1965. Life history of the gizzard shad Dorosoma cepedianum (LeSueur), in western Lake Erie. Fishery Bulletin 65:391-425. Brooks, B. W., and C. J. A. Bradshaw. 2006. Strength and evidence for density dependence in abundance time series of 1198 species. Ecology 87:1445-1451. Brophy, D., and B. S. Danilowicz. 2003. The infl uence of pre-recruitment growth on subsequent growth and age at first spawning in Atlantic herring ( Clupea harengus L.). ICES Journal of Marine Science 60:1103-1113. Burton, M. P. M. 1994. A critical period for nutr itional control of early gametogenesis in female winter flounder, Pleuronectes americanus (Pisces: Teleostei). Journal of the Zoological Society of London 233:405-415. Carpenter, S. R. 1989. Replication and Treatment Strength in Whole-Lake Experiments. Ecology 70:453-463. Carpenter, S. R., and coauthors. 1987. Regul ation of lake primary productivity by food web structure. Ecology 68:1863-1876. Catalano, M. J., M. J. Allen, and L. DeBrab andere. 2007. Assessing effects of gizzard shad removal on nutrient cycling and gizzard shad population dynamics. Final Report to the St. Johns River Water management District. University of Florida, Gainesville.

PAGE 100

100 Catalano, M. J., and coauthors. in review Lack of short-term biomanipulation effects on a subtropical lake demonstrates trade offs among management goals. Fisheries Management and Ecology. Clark, W. C. 2002. F35% revisited ten years later. Nort h American Journal of Fisheries Management 22:251-257. Clark, W. G. 1999. Effects of an erroneous natu ral mortality rate on a simple age-structured stock assessment. Canadian Journal of Fisheries and Aquatic Sciences 56:1721-1731. Cowan, J. H., Jr., K. A. Rose, and D. R. DeVries. 2000. Is density-d ependent growth in youngof-the-year fishes a question of critical we ight? Reviews in Fish Biology and Fisheries 10:61-89. Cushing, D. H. 1990. Plankton production and ye ar-class strength in fish popoulations: an update of the match/mismatch hypothesis. A dvances in Marine Biology 26:249-293. Cyterski, M. J., and J. J. Ney. 2005. Availability of clupeid prey to primary piscivores in Smith Mountain Lake, Virginia. Transactions of the American Fisheries Society 134:1410-1421. DeGisi, J. S. 1994. Year class strength and catch ability of mountain lake brook trout. Master's Thesis. University of British Columbia, Vancouver, British Columbia. DeMelo, R., R. France, and D. J. McQueen. 1992. Biomanipulation: hit or myth. Limnology and Oceanography 37:192-207. DeVries, D. R., and R. V. Frie. 1996. Dete rmination of age and growth. Pages 483-512 in B. R. Murphy, and D. W. Willis, editors. Fisheries techniques. American Fisheries Society, Bethesda, Maryland. DeVries, D. R., and R. A. Stein. 1990. Manipulating shad to enhance sport fisheries in North America: an assessment. North American J ournal of Fisheries Ma nagement 10:209-223. DeVries, D. R., and R. A. Stein. 1992. Comple x interactions between fish and zooplankton: quantifying the role of an open-water planktivore. Canadian Journal of Fisheries and Aquatic Sciences 49:1216-1227. Dotson, J. R., M. S. Allen, W. E. Johnson, and J. Benton. In press Impacts of commercial gill net bycatch and recreational fi shing on a Florida black crappi e population. North American Journal of Fisheries Management. Drenner, R. W., and K. D. Hambright. 1999. Revi ew: biomanipulation of fish assemblages as a lake restoration technique. Ar chiv fr Hydrobiologie 146:129-165. Fogarty, M. J., A. A. Rosenberg, and M. P. Si ssenwine. 1992. Fisheries risk assessment: a case study of Georges Bank haddock. Environm ental Science and Technology 26:440-447.

PAGE 101

101 Fournier, D. A., J. R. Sibert, and M. Terciero. 1991. Analysis of length frequency samples with relative abundance data for the Gulf of Maine northern shrimp ( Pandalus borealis ) by the MULTIFAN method. Canadian Journal of Fisheries and Aquatic Sciences 48:591-598. Frysa, K. G., B. Bogstad, and D. W. Skagen. 2002. Fleksibest: an age-length structured fish stock assessment model. Fi sheries Research 55:87-101. Gido, K. B. 2002. Interspecific comparisons an d the potential importance of nutrient excretion by benthic fishes in a large re servoir. Transactions of th e American Fisheries Society 131:260-270. Gilliam, J. F., and D. F. Fraser. 1987. Habitat se lection under predation hazard: test of a model with foraging minnows. Ecology 68:1856-1862. Goodwin, N. B., A. Grant, A. L. Perry, N. K. Dulvy, and J. D. Reynolds. 2006. Life history correlates of density-dependent recruitment in marine fishes. Canadian Journal of Fisheries and Aquatic Sciences 63:494-509. Goodyear, C. P. 1980. Compensation in fish populations. Pages 253-280 in C. H. Hocutt, and J. R. Stauffer, editors. Biological monitori ng of fish. D. C. Heath and Company. Gotelli, N. J. 1995. A Primer of Ecology. Sina uer Associates, Inc., S underland, Massachusetts. Hansen, J. M., and G. A. Chouinard. 1992. Eviden ce that size-selective mo rtality affects growth of Altantic cod (Gadus morhua L.) in the sout hern Gulf of St. Lawrence. Journal of Fish Biology 41:31-41. Hansen, M. J., C. P. Madenjian, J. H. Selgeby, and T. E. Helser. 1997. Gillnet selectivity for lake trout ( Salvelinus namaycush ) in Lake Superior. Canadian Journal of Fisheries and Aquatic Sciences 54:2483-2490. Hansson, L.-A., and coauthors. 1998. Biomanipulat ion as an application of food-chain theory: constraints, synthesis, and recommendations for temperate lakes. Ecosystems 1:558-574. Hastings, W. K. 1970. Monte Carlo sampli ng methods using Markov chains and their applications. Biometrika 57:97-109. Healey, M. C. 1978. Fecundity changes in expl oited populations of lake whitefish (Coregonus clupeaformis) and lake trout (Salvelinus nama ycush). Journal of th e Fisheries Research Board of Canada 35:945-950. Healey, M. C. 1980. Growth and recruitment in experimentally exploited lake whitefish (Coregonus clupeaformis) populations. Canadi an Journal of Fisheries and Aquatic Sciences 37:255-267. Heath, M. R. 1992. Field investigations of the early life stages of mari ne fish. Advances in Marine Biology 28:1-174.

PAGE 102

102 Helser, T. E., and F. P. Almeida. 1997. Density-d ependent growth and sexual maturity of silver hake in the north-west Atlantic. Journal of Fish Biology 51:607-623. Henderson, B. A., J. L. Wong, and S. J. Neps zy. 1996. Reproduction of wa lleye in Lake Erie: allocation of energy. Canadian Journal of Fisheries and Aquatic Sciences 53:127-133. Hjort, J. 1914. Fluctuations in the great fisherie s of northern Europe viewed in light of biological research. Rapports et Proces-verbaux des Reunions du Conseil International pour l'Exploration de la Mer 19:1-228. Hoenig, J. M. 1983. Empirical use of longevity da ta to estimate mortality rates. Fishery Bulletin 82:898-903. Horppila, J., H. Peltonen, T. Malinen, E. Luokkanen, and T. Kairesalo. 1998. Top-down or bottom-up effects by fish: issues of concern in biomanipulation of lakes. Restoration Ecology 6:20-28. Houde, E. D. 1987. Fish eary life dynamics and recruitment variability. American Fisheries Society Symposium 2:17-29. Houde, E. D. 1989. Subtleties and episodes in the early life history of fishes. Journal of Fish Biology 35:29-38. Hovenkamp, F. 1992. Growth-dependent mortality of larval plaice Pleuron ectes platessa in the North Sea. Marine Ecology Progress Series 82:95-101. Jensen, A. L. 1996. Beverton and Holt life histor y invariants result from optimal trade-off of reproduction and survival. Canadian Journal of Fisheries and Aquatic Sciences 53:820822. Kim, G. W., and D. R. DeVries. 2000. Effects of a selectively reduced gizzard shad population on trophic interactions and age0 fishes in Walker County Lake, Alabama. North American Journal of Fisheries Management 20:860-872. Koslow, J. A., J. Bell, P. Virtue, and D. C. Sm ith. 1995. Fecundity and its variability in orange roughy: effects of population density, condition, egg size, and senescence. Journal of Fish Biology 47:1063-1080. Kristiansen, T. S., and T. Svsand. 1998. Effect of size-selective mortality on growth of coastal cod illustrated by tagging data and an individual-based growth and mortality model. Journal of Fish Biology 52:688-705. Lorenzen, K. 2000. Allometry of natural mortality as a basis for assessing optimal release size in fish-stocking programmes. Canadian Journal of Fisheries and Aquatic Sciences 57:23742381.

PAGE 103

103 Lorenzen, K., and K. Enberg. 2002. Density-dep endent growth as a key mechanism in the regulation of fish populations: evidence from among-population comparisons. Proceedings of the Royal Society of London: B 269:49-54. Mace, P. M. 1994. Relationships between common bi ological reference points used as thresholds and targets of fisheries management strategies. Canadian Journal of Fisheries and Aquatic Sciences 51:110-122. Maceina, M. J., and M. R. Stimpert. 1998. Rela tions between reservoir hydrology and crappie recruitment in Alabama. North American J ournal of Fisheries Ma nagement 18:104-113. Marshall, C. T., and K. T. Frank. 1999. The e ffect of interannual variation in growth and condition on haddock recruitment. Canadian J ournal of Fisheries and Aquatic Sciences 56:347-355. Maunder, M. N., and R. B. Deriso. 2003. Estima tion of recruitment in catch-at-age models. Canadian Journal of Fisheries and Aquatic Sciences 60:1204-1216. McQueen, D. J. 1998. Freshwater foood web mani pulation: a powerful tool for water quality improvement, but maintenance is required. Lakes and Reservoirs: Research and Management 3:83-94. Meijer, M.-L., I. de bois, M. Scheffer, R. Por tielje, and H. Hosper. 1999. Biomanipulation in shallow lakes in the Netherlands: an ev aluation of 18 case studies. Hydrobiologia 408/409:13-30. Meronek, T. G., and coauthors. 1996. A review of fish control projects. North American Journal of Fisheries Management 16:63-74. Mertz, G., and R. A. Myers. 1997. Influence of error in natural morta lity estimates in cohort analysis. Canadian Journal of Fisher ies and Aquatic Sciences 54:1608-1612. Methot, R. D. 2005. Technical description of th e Stock Synthesis II assessment program version 1.17. NOAA Fisheries, Seattle Washington. Miller, T. J., L. B. Crowder, J. A. Rice, and E. A. Marschall. 1988. Larval size and recruitment mechanisms in fishes: toward a conceptual fr amework. Canadian Journal of Fisheries and Aquatic Sciences 1657-1670. Millner, R. S., and C. L. Whiting. 1996. Longterm changes in growth and population abundance of sole in the North Sea from 1940 to presen t. ICES Journal of Ma rine Science 53:11851195. Murdoch, W. W. 1994. Populat ion regulation in theory and practice. Ecology 75:271-287. Murray, B. G. 1994. On density dependence. OIKOS 69:520-523.

PAGE 104

104 Myers, R. A. 2001. Stock and recruitment: generalizations about maximum reproductive rate, density dependence, and variability using me ta-analytic approaches. ICES Journal of Marine Science 58:937-951. Myers, R. A. 2002. Recruitment: understanding density dependence in fish populations. Pages 123-148 in P. J. B. Hart, and J. D. Reynolds editors. Handbook of fish biology and fisheries: fish biology, volume 1. Bl ackwell Science, Mald en, Massachusetts. Myers, R. A., and N. J. Barrowman. 1996. Is fi sh recruitment related to spawner abundance? Fishery Bulletin 94:707-724. Myers, R. A., N. J. Barrowman, J. A. Hutchings, and A. A. Rosenberg. 1995. Population dynamics of exploited fish stocks at low population levels. Science 269:1106-1108. Myers, R. A., K. G. Bowen, and N. J. Barrowman. 1999. Maximum reproduc tive rates of fish at low population sizes. Canadian Journal of Fisheries and Aquatic Sciences 56:2402-2419. Oskarsson, G. J., and C. T. Taggart. 2006. Fecundity variation in Icelandic summer-spawning herring and implications for reproductive potential. ICES Journal of Marine Science 63:493-503. Pauly, D. 1980. On the interrelationships betwee n natural mortality, growth paramters, and mean environmental temperature in 175 fish stoc ks. Journal du Conse il International pour l'Exploration de la Mer 39:175-192. Perry, E., G. Seegert, J. Vondruska, T. Lohner, and R. Lewis. 2003. Modeling possible cooling water intake system impacts on Ohio River fish populations. Dixon, Veil, and Wisniewski, editors. Defining and assessing adverse e nvironmental impact from power plant impingement and entrainment of aquatic organisms. Peterman, R. M., and M. J. Bradford. 1987. Dens ity-dependent growth of age 1 English sole ( Parophtys vetulus ) in Oregon and Washington coasta l waters. Canadian Journal of Fisheries and Aquatic Sciences 44:48-53. Post, J. R., and D. O. Evans. 1989. Size-depe ndent overwinter mortality of young-of-the-year yellow perch (Perca flavescens): laboratory, in situ enclosure, and field experiments. Canadian Journal of Fisheries and Aquatic Sciences 46:1958-1968. Quinn, T. J., II, and R. B. Deriso. 1999. Quantitative fish dynamics. Oxford University Press, New York, New York. Rijnsdorp, A. D. 1993. Fisheries as a largescale experiment on lif e-history evolution: disentangling phenotypic and genetic effects in changes in maturation and reproduction of North Sea plaice, Pleuronectes platessa L. Oecologia 96:391-401. Rijnsdorp, A. D., and P. I. van Leeuwen. 1996. Changes in growth of North Sea plaice since 1950 in relation to density, eu trophication, beam-trawl effo rt, and temperature. ICES Journal of Marine Science 53:1199-1213.

PAGE 105

105 Rochet, M.-J. 1998. Short-term effects of fishing on life history traits of fishes. ICES Journal of Marine Science 55:371-391. Rochet, M. 2000. Does the concept of spawning pe r recruit make sense? IC ES Journal of Marine Science 57:1160-1174. Romare, P., and E. Bergman. 1999. Juvenile fi sh expansion following bi omanipulation and its effects on zooplankton. H ydrobiologia 404:89-97. Rose, K. A., and J. H. Cowan, Jr. 2000. Pred icting fish population dynamics: compensation and the importance of site-specific considerati ons. Environmental Science and Policy 3:S433S443. Rose, K. A., J. H. Cowan, Jr., K. O. Wine miller, R. A. Myers, and R. Hilborn. 2001. Compensatory density dependence in fish populations: importance, controversy, understanding and prognosis. Fi sh and Fisheries 2:293-327. Rotella, J. J., and coauthors. 2009. An evaluation of density-dep endent and density-independent influences on population growth rates in Weddell seals. Ecology 90:975-984. Sauer, J. R., and N. A. Slade. 1987. Size-based demography of vertebra tes. Annual Review of Ecology and Systematics 18:71-90. Schaefer, M. B. 1954. Some aspects of the dynamcis of populations important to the management of commercial marine fisheries. Bulletin of the Inter-American Tropical Tuna Commission 1:27-56. Schaus, M. H., M. J. Vanni, and T. E. Wissi ng. 2002. Biomass-dependent diet shifts in omnivorous gizzard shad: implications for growth, food web, and ecosystem effects. Transactions of the American Fisheries Society 131:40-54. Schaus, M. H., and coauthors. 1997. Nitrogen an d phosphorus excretion by detritivorous gizzard shad in a reservoir ecosystem. Limnology and Oceanography 42:1386-1397. Scheuerell, M. D. 2005. Influence of juvenile size on age at maturity of individually marked wild Chinook salmon. Transactions of the American Fisheries Society 134:999-1004. Schindler, D. W. 1998. Replication versus realis m: the need for ecosystem-scale experiments. Ecosystems 1998:323-334. Shapiro, J., V. Lamarra, and M. Lynch, editors 1975. Biomanipulation: an ecosystem approach to lake restoration, Gainesville, Florida. Shepherd, J. G., and D. H. Cushing. 1980. A mech anism for density-dependent survival of larval fish as the basis of a stock-recruitment relati onship. Journal du Cons eil International pour l'Exploration de la Mer 39:160-167.

PAGE 106

106 Silliman, R. P. 1968. Interaction of food level and exploitation in experimental fish populations. Fishery Bulletin 66:425-439. Silliman, R. P., and J. S. Gutsell. 1958. Expe rimental exploitation of fish populations. Fishery Bulletin 58:213-252. Sinclair, A. F., D. P. Swain, and J. M. Hanson. 2002. Measuri ng changes in the direction and magnitude of size-selective mortality in a comm ercial fish population. Canadian Journal of Fisheries and Aquatic Sciences 59:361-371. Smith, A. D. M., and C. J. Walters. 1981. Adapti ve management of stock-recruitment systems. Canadian Journal of Fisheries and Aquatic Sciences 38:690-703. Sogard, S. M. 1997. Size-selective mortality in th e juvenile stage of te leost fishes: a review. Bulletin of Marine Science 60:1129-1157. Stearns, S. C. 1992. The Evoluti on of Life Histories. Oxford University Press, Oxford. Stein, R. A., D. R. DeVries, and J. M. Dettmers. 1995. Food-web regulation by a planktivore: exploring the generality of the trophic cascade hypothesis. Ca nadian Journal of Fisheries and Aquatic Sciences 52:2518-2526. Taylor, N. G., C. J. Walters, and S. J. D. Martell. 2005. A new likelihood for simultaneously estimating von Bertalanffy growth parameters gear selectivity, and natural and fishing mortality. Canadian Journal of Fish eries and Aquatic Sciences 62:215-223. Thompson, G. G. 1994. Confounding of gear selec tivity and the natural mortality rate in cases where the former is a nonmonotone function of age. Canadian Journal of Fisheries and Aquatic Sciences 51:2654-2664. Trippel, E. A. 1995. Age at maturity as a stre ss indicator in fisheries. Bioscience 45:759-771. Van Den Avyle, M. J., J. Boxrucker, P. Mich aletz, B. Vondracek, and G. R. Ploskey. 1995. Comparison of catch rate, lengt h distrubution, and precison of six gears used to sample reservoir shad populations. North American Journal of Fisheries Management 15:940-955. van der Veer, H. W. 1986. Immigration, settleme nt, and density-dependent mortality of a larval and early post-larval 0-group plaice ( Pleuronectes platessa ) population in the western Wadden Sea. Marine Ecology Progress Series 29:223-236. Vanni, M. J., and coauthors. 2006. Nutrient cycl ing by fish supports relatively more primary production as lake productivity increases. Ecology 87:1696-1709. Walters, C. J., and F. Juanes. 1993. Recruitment limitation as a consequen ce of natural selection for use of restricted feeding ha bitats and predation risk taking by juvenile fishes. Canadian Journal of Fisheries and A quatic Sciences 50:2058-2070.

PAGE 107

107 Walters, C. J., and S. J. D. Martell. 2004. Fisheries Ecology and Management. Princeton University Press, Princeton, New Jersey. Walters, C. J., S. J. D. Martell, and J. Korm an. 2006. A stochastic appr oach to stock reduction analysis. Canadian Journal of Fisher ies and Aquatic Sciences 63:212-223. Winemiller, K. O., and K. A. Rose. 1992. Patte rns of life-history diversification in North American fishes: implications for population re gulation. Canadian Journal of Fisheries and Aquatic Sciences 49:2196-2218. Zijlstra, J. J., and J. I. Witte. 1985. On the re cruitment of age-0 group plaice in the North Sea. Netherlands Journal of Zoology 35:360-376. Zipkin, E. F., and coauthors. 2008. Overco mpensatory response of a smallmouth bass (Micropterus dolomieu) population to harvest: release from competition? Canadian Journal of Fisheries and Aquatic Sciences 65:2279-2292.

PAGE 108

108 BIOGRAPHICAL SKETCH Matt is f rom Lorain, Ohio and grew up within a stones throw of Lake Erie. He received a Bachelor of Science degree in Zoology from Miami University (Ohio) in 1997. After graduation, he worked as a field technician at Yellowst one National Park where he assisted in a study of the effects of non-native lake trout on the cutthroat trout popul ation in Yellowstone Lake. He also spent two years working as a fish eries technician at the Illinois Natural History Survey in Salem, Illinois. Matt received his Ma ster of Science degree in 2002 at the University of Wisconsin-Stevens Point where he evaluated the effects of low-head dam removal on fish communities in the Baraboo River, Wisconsin. He then worked for two years as a research biologist with the Wisconsin De partment of Natural Resources in Madison, Wisconsin. After working as a fisheries research biologist for one year under Mike Allen at the University of Florida, Matt began his Ph.D. working on popul ation dynamics of gizzard shad in 2005.