Habitat Structures Rainbow Trout Population Dynamics Across Spatial Scales

University of Calgary PRISM: University of Calgary's Digital Repository

Graduate Studies The Vault: Electronic Theses and Dissertations

2018-12-19 Habitat Structures Rainbow Trout Population Dynamics Across Spatial Scales

Cantin, Ariane

Cantin, A. (2018). Habitat Structures Rainbow Trout Population Dynamics Across Spatial Scales (Unpublished doctoral thesis). University of Calgary, Calgary. AB. http://hdl.handle.net/1880/109388 doctoral thesis

University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission. Downloaded from PRISM: https://prism.ucalgary.ca UNIVERSITY OF CALGARY

Habitat Structures Rainbow Trout Population Dynamics Across Spatial Scales

Ariane Cantin

A THESIS

SUBMITTED TO THE FACULTY OF GRADUATE STUDIES

IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE

DEGREE OF DOCTOR OF PHILOSOPHY

GRADUATE PROGRAM IN BIOLOGICAL SCIENCES

CALGARY, ALBERTA

DECEMBER, 2018

Abstract

Density-dependent processes play an important role in structuring population dynamics – as the number of organisms within a set area increases, population vital rates and life history traits will also change arising from increased competition for limited food and space. In this thesis I explore how variation in the quantity of habitat available impacts size-structured animal populations using lake-dwelling rainbow trout (Oncorhynchus mykiss). First, I developed hypotheses on the biological processes by which habitat impacts population dynamics using a multi-habitat age-structured population model. This theoretical model showed that habitat limitations at any life stage can bottleneck the population and impact its dynamics, but that the timing of regulation influences population outcomes. Limited habitat in early life led to high early mortality, resulting in low overall population density of larger fish, while limited habitat in adult life led to high early survival and a high density of stunted fish populations. I then compared the model predictions to empirical data from 39 wild rainbow trout populations and to results of a harvest experiment. The field results corroborated the model predictions and showed that lakes with a higher early (stream) to late (lake) life stage habitat ratio presented higher number of recruits, later age at maturity and a smaller maximum size than lakes in which stream habitat was limiting. The density manipulation also supported the model predictions as the lake with the lowest habitat ratio presented the lowest compensatory reserve and showed a density-dependent growth response to harvest. Finally, I used the knowledge acquired at the lake-scale to predict rainbow trout production at the landscape-scale. I developed a methodology that describes rainbow trout distribution

ii based on stream network characteristics and connectivity. Then I used a landscape-scale proxy of stream habitat availability, stream order, and lake area to predict stream to lake habitat ratio and infer population dynamics. I combined this landscape-scale production information with recreational fishing demand to identify regions more prone to being impacted by overfishing or habitat perturbations. My research details how local habitat availability influences fish populations and can be used to predict population dynamics across a landscape of lakes providing a valuable tool to managers.

iii Preface

This thesis is an original work developed by the author (me), Ariane Cantin. The studies presented were done in collaboration with colleagues at the University of Calgary and fishery scientists and managers from the British Columbia Government and

Freshwater Fisheries Society of British Columbia (FFSBC). The contribution from those collaborators is described for each research chapter below.

A version of the research conducted in Chapter 2 has been published in the journal Ecology from the Ecological Society of America before the submission of this thesis. It is written with co-author John R. Post, my supervisor at the University of

Calgary, and therefore uses “we” instead of the “I” used in the rest of the thesis. The full citation is given in the thesis References section under Cantin and Post (2018). I developed the model presented in this chapter, performed all the analyses and writing.

John Post revised my findings and interpretations through discussions and edits to the manuscript.

The work presented in Chapter 3 is an empirical study I designed, led, analyzed and wrote. Kyle Wilson (University of Calgary) provided support for the Bayesian statistical analyses. Paul Askey (FFSBC) and John Post (University of Calgary) provided edits to the analyses and earlier versions of the chapter.

Chapter 4 is an experiment I designed, led, analyzed and wrote. John Post and

Stephanie Mogensen (University of Calgary) provided edits to my analyses and manuscript.

iv Chapter 5 is a landscape-scale modelling and application of the local-scale findings from the earlier chapters. I designed and led the study. Anne Farineau

(University of Calgary) provided GIS support and conducted preliminary spatial analyses. I was responsible for the final analyses, figures presented and writing of the manuscript. Anne Farineau, Darren Bender and John Post (University of Calgary) provided comments and edits to the analyses and edited the manuscript.

v Acknowledgements

The studies presented in this thesis were funded by Discovery and Collaborative

Research and Development Grants from the Natural Sciences and Engineering Research

Council of Canada (NSERC) and the Freshwater Fisheries Society of British Columbia

(FFSBC) to John R. Post. I also recognize the financial support of NSERC through a

Postgraduate Scholarship, the University of Calgary who awarded me a Silver

Anniversary Graduate Fellowship and the British Columbia Provincial Government.

I thank my supervisor, John Post, for taking me as his student and guiding me through this project. His hands-off supervisory style made me wonder at times if he would stop me from going in a wrong direction, but, in the end, I recognize that he always had my back and left me enough space to explore and learn on my own while still being completely present when I needed him. I am also thankful for my great supervisory committee, Paul Askey and Darren Bender, for raising important questions about my study design and work that greatly enhanced my project. Paul made me explore the management implications of my work, while Darren helped me situate my findings in a broader spatial scale.

I thank colleagues at BC FLNRO (Andrew Klassen, Steve Maricle, Russ

Bobrowski, Hillary Ward, Eric Hegerat), FFSBC (Theresa Godin, Kirstin Gale, Marcus

Boucher, Adrian Clarke) and BC Environment (Brett van Poorten, Eric Parkinson), that supported me in the development and planning of this project and offered tremendous field assistance. The data presented in this thesis would not have existed without the efforts of Eric Newton and Ron Bowron who were the core field team and spent

vi countless hours measuring fish with frozen hands, camping in the snow, driving on endless forestry roads (with the not so occasional flat tire), bushwacking with an electrofisher on their back and generating field stories that I will (for the most part) cherish and tell for many years to come. I also thank Marguerite Tibbles, Katrina Siebert,

Adam Hope, Eric Kukulowicz, Sean Boysen and Troy Machovec for their hard work in the field.

I thank my lab family: Kyle Wilson, Nilo Sinnatamby, Anne Farineau, Chris

Cahill, Jon Mee, Dylan Glaser and Fiona Johnston for their support that came in many forms (field, moral, statistical, pub). I am greatly indebted to two of my labmates, Hillary

Ward who mentored me and helped me develop the ideas that became this project, and

Steph Mogensen who has supported me and edited pretty much everything I’ve written in my “improving but still far from perfect” English in the past four years (not a small task!). I want to give a special “remerciement” to my dear friend and colleague Pascale

Gibeau who has been an almost daily virtual presence throughout this project and who many years ago gave me my first biology job. I must also thank fellow members of the

University of Calgary Biological Sciences Department for the great memories and friendships made, particularly but not limited to: Ella Bowles, Louise Hahn, Emma

Carroll and Analisa Lazaro-Côté.

My final thanks go to my friends and family who have encouraged me, took me skiing, made me dinner and listened to me throughout the completion of my degree even though they probably didn’t (or still don’t) understand what I’m doing or why I’m freaking out. I am particularly grateful to my husband Jay and my parents Serge and

Lorraine. Merci !

vii

Table of Contents

Abstract ...... ii Preface ...... iv Acknowledgements ...... vi Table of Contents ...... viii List of Tables ...... xi List of Figures and Illustrations ...... xiv List of Symbols, Abbreviations and Nomenclature ...... xix

Chapter 1: Introduction ...... 1

Chapter 2: Habitat Availability and Ontogenetic Shifts Alter Bottlenecks in Size- Structured Fish Populations ...... 8 Introduction ...... 8 Methods ...... 11 Model development...... 11 Stream habitat...... 12 Lake habitat...... 14 Migration schedule...... 15 Natural mortality...... 16 Harvest mortality...... 18 Growth...... 19 Fecundity...... 21 Simulations...... 22 Analysis...... 23 Comparison to empirical data...... 24 Results ...... 25 Population dynamics: Collapse...... 25 Population dynamics: Fixed-point equilibria...... 26 Population dynamics: Cycles...... 26 Assessing the importance of ecological processes...... 27 Parameter sensitivity analysis...... 28 Comparison to field data...... 29 Discussion ...... 30

Chapter 3: Interaction of Juvenile Stream Rearing and Adult Lacustrine Habitats on Growth and Demography of Wild Rainbow Trout: Inferring Processes from Field Observations...... 52 Introduction ...... 52 Methods ...... 54 Site selection and study design...... 54 Stream sampling...... 55 Stream habitat descriptors...... 56 Habitat and age-class associations...... 57

viii Lake sampling...... 58 Habitat capacity...... 59 Growth and maturation...... 60 Results ...... 64 Stream habitat descriptors...... 64 Stream capacity...... 65 Lake capacity...... 66 Growth and maturation...... 67 Discussion ...... 69

Chapter 4: Impact of Experimental Density Manipulation on Population Abundance and Growth of Rainbow Trout Populations with Contrasting Early Life Stage Habitat Conditions ...... 87 Introduction ...... 87 Methods ...... 90 Study sites...... 90 Experimental design...... 92 Data Analysis...... 93 Results ...... 96 Depletion...... 96 Temperature...... 97 Abundance and age structure...... 97 Growth...... 98 Discussion ...... 100

Chapter 5: Predicting Fish Populations’ Distribution, Production and Recreational Fishing Demand Based on Landscape Characteristics ...... 115 Introduction ...... 115 Methods ...... 118 Rainbow trout distribution...... 118 Species information...... 118 Data source...... 118 Categorization...... 119 Probability of presence...... 120 Rainbow trout population characteristics...... 123 Recreational fishing effort...... 125 Results ...... 126 Discussion ...... 130

Chapter 6: Conclusions ...... 150

References ...... 157

Appendix A: Age-Structured Model Sensitivity Analysis ...... 184

Appendix B: Study Lakes Description ...... 189

ix Appendix C: Stream Sampling ...... 192 Stream ...... 192 Reach Point ...... 192 Reach Overall ...... 193

Appendix D: Habitat Suitability Index (HSI) ...... 195 Habitat Suitability Index Calculations ...... 195 Weighted Useable Area Calculations ...... 196

Appendix E: Model Posterior Predictive Distribution ...... 200

Appendix F: Growth and Maturity Traits Correlation ...... 205

Appendix G: Life History Traits Model Selection ...... 206

x List of Tables

Table 2.1 Parameters used in the model and their source, for parameters that are calculated the associated equation number is provided...... 37

Table 2.2 Results of the modification of processes from those used in the baseline population (50 m2 stream area, 25% of stream habitat is spawning habitat, migration after one growing season, 1500 GDD) under no harvest (abundance of 2072 fish and size at age 3 of 17.97 cm) and harvest mortality rate of 0.7 (abundance of 1690 fish and size at age 3 of 21.90 cm). Simulations summarized using the resulting population status (collapse, fixed-point equilibrium [FPE], cycle), percentage change from the baseline in abundance of fish in the lake and mean size at age 3 (cm). When population cycle the abundance and size used are the mean values obtained over 10 years and are identified using an asterisk (*). The bold portions of the table represent a change in qualitative dynamics for the population status or a change in quantitative predictions greater than 25%...... 41

Table 3.1 An assessment of alternate models to predict stream habitat capacity based on estimated number of eggs (Neggs), tributary maximum width (Width), Habitat Suitability Index (HSI), weighted useable area (WUA), growing-degree- days (GDD) and presence/absence of other fish species (OtherSp). Values of R2, 2 2 adjusted R (adjR ), p-value, AIC and ΔAIC are presented, the most parsimonious model (lowest AIC) is in bold and models within two AIC points are in italic...... 77

Table 3.2 An assessment of alternate models to predict lake habitat capacity described using number of Age-2 fish in gillnet catch as a proxy for number of recruits (models 1-10) and total number of fish in gillnet catch as a proxy for lake abundance (models 11-20), based on number of juvenile fish caught in the stream (Nstream), stream weighted useable area (WUA), lake area (Lake.Area), WUA to lake area ratio (WUAratio), growing-degree-days (GDD), and presence/absence of other fish species (OtherSp). Values of R2, adjusted R2 2 (adjR ), p-value, AIC and ΔAIC are presented, the most parsimonious model (lowest AIC) is in bold and models within two AIC points are in italic...... 78

Table 3.3 Most parsimonious models to predict age at 50% maturity of females (A50f), length at 50% maturity of females (L50f), the Brody growth coefficient (log- transformed – ln(k)), asymptotic length (log-transformed – ln(L∞)) and maximum fish size observed (log-transformed – ln(MaxObsFL)), based on lake gillnet catch (Nlake), presence/absence of other fish species (OtherSp), WUA to lake area ratio (WUAratio), growing-degree-days (GDD). Values of R2, adjusted 2 2 R (adjR ) and p-value are presented, and a complete list of the models compared for each variable is presented in Appendix G: Table G.1...... 79

Table 4.1 Lake characteristics for the three study lakes, the total dissolved solids (TDS) value is the mean of the annual values calculated during the fall gillnetting (2012-2016), the growing-degree-days (GDD) value correspond to the decadal average (ClimateWNA - Wang et al. 2016), the Habitat Suitability Index

xi (HSI) and weighted useable area (WUA) values were calculated using the methodology described in Chapter 3...... 105

Table 4.2 An assessment of alternate models to predict annual growth rate based on length at first capture (L1), lake (Pantano, Stubby or Today) and depletion status 2 2 2 (PrePost). Values of R , adjusted R (adjR ), p-value, AIC and ΔAIC are presented, the most parsimonious model (lowest AIC) is in bold and models within two AIC points are in italic...... 105

Table 4.3 An assessment of alternate models to predict annual growth rate based on length at first capture (L1), estimated population density (Density – N/ha) and 2 2 2 climate (GDD). Values of R , adjusted R (adjR ), p-value, AIC and ΔAIC are presented, the most parsimonious model (lowest AIC) is in bold and models within two AIC points are in italic...... 106

Table 5.1 Lake categories determined by observation of rainbow trout (RB) and stocking history...... 137

Table 5.3 Comparison of five gradient thresholds (10-30%) using network connectivity of the validation subset (known presence/absence or rainbow trout population), for each threshold the number of categorized upstream lakes is presented...... 138

Table 5.4 Characteristics of the four lakes presented as an example in Figure 5.4...... 139

Table 5.5 Model comparison of predictors of lake tributaries’ weighted useable area (lnWUA) and their associated statistics; the model with the highest coefficient of variation (R2) is in bold...... 140

Table A.1 Sensitivity analysis results for a variation of +/- 5%, 10% and 25% of important parameters of the model and their impact on the baseline population (50 m2 stream area, 25% of stream habitat is spawning habitat, migration after one growing season, 1500 GDD) under no harvest (abundance of 2072 fish and size at age 3 of 17.97 cm) and under high harvest mortality rate of 0.7 (abundance of 1690 fish and size at age 3 of 21.90 cm). Simulations summarized using the resulting population status (collapse, fixed-point equilibrium [FPE], cycle), percentage change from the baseline in abundance of fish in the lake and mean size at age 3. When population cycles the abundance and size used are the mean values obtained over 10 years and are identified using an asterisk (*). Changes presented in bold represent situations for which the resulting population had either a different qualitative population dynamics or presented a change in abundance, or size at age that was of an equal or higher percentage than the associated change in parameter...... 184 xii Table B.1 Year sampled, location and description of the 39 waterbodies included in the different chapters of this thesis, all waterbodies had rainbow trout present, additional species are presented in bold in the species column (RB = rainbow trout, CSU = largescale sucker, LSU = longnose sucker, SU = sucker spp., NSC = northern pikeminnow, RSC = redside shiner), the chapters column refers to the thesis chapter in which each lake was used. Summary statistics (minimum, maximum and mean) of the variables elevation, lake area and GDD are presented at the bottom of the table...... 189

Table C.1 Summary statistics of the environmental variables collected at each stream reach sampled...... 194

Table D.1 Detailed calculations of the Habitat Suitability Index (HSI) variables used. 197

Table F.1 Table of the correlation between traits associated with maturity and growth calculated in the hierarchical Bayesian model, note that the fit for trait age at 50% maturity for males (A50m) was rejected due to its poor fit...... 205

Table G.1 List of all the models compared to predict the variation in life-history traits associated with female maturation age (A50f - models 1-12) and size (L50f - models 13-24), and log-transformed growth parameters k (models 26-36), asymptotic length (L∞- models 27-48) and observed maximum length (MaxObsFL - models 49-60). The most parsimonious model for each trait (lowest AIC) is in bold and models within two AIC points are in italic...... 206

xiii List of Figures and Illustrations

Figure 2.1 Schematic representation of the model, the input variables, important processes presented through the rainbow trout life cycle and output variables...... 44

Figure 2.2 Migration timing decision flowchart for juvenile fish in the stream (left) and associated migration schedule (right) presented with age at migration and corresponding proportion of the stream habitat (in percent) that is appropriate for Age-0 and Age-1 rearing, in the Age-1 shared scenario a range of values is presented since the percentage of habitat used will depend on the density of both age-classes and the saturation of their habitat...... 45

Figure 2.3 a) Survival rate in the lake varies with effective density and size (line thickness represents fish size) b) Growth in both the stream and the lake follows a von Bertalanffy growth curve which varies with fish density (percent habitat saturation in the stream and effective density in the lake) and growing-degree- days (GDD), density impacts the asymptotic size (L∞) while GDD impacts growth rate (k). The blue lines represent low density populations where juveniles migrate to the lake right after egg emergence while the orange lines represent high density populations where juveniles spend two growing seasons in the stream (from egg emergence to age 1), their growth trajectory is modified once they enter the lake. The solid lines represent high GDD while the dashed lines are low GDD...... 46

Figure 2.4 Schematic overview of the simulation results as they vary with stream habitat availability from no stream habitat available on the bottom to large stream with abundant rearing habitat at the top and impact of harvest to the right. Each box describes the dynamic outcomes (collapse, fixed-point equilibrium [FPE], cycle), population density and mean size at age in the lake...... 47

Figure 2.5 Simulation results presented as population abundance vs. maximum size from (a) all simulations that resulted in a fixed-point equilibrium without harvest presented by circles and outlined by a grey polygon, mean values for cycles presented by yellow polygon (low density cycles) and purple polygon (high density cycles). b-c) polygons that illustrate simulations from (a) as they relate to climate and stream habitat availability as follows: b) effect of low (yellow – 1000 GDD), intermediate (orange – 1500 GDD) and high (red – 2000 GDD) growing-degree-days; c) simulations results separated by stream habitat availability (red: stream area <10 m2, <15% of stream habitat is spawning/rearing habitat; blue: stream area 10 - 25 m2, < 25% of stream habitat is spawning/fry rearing habitat present, green: stream area > 25 m2, > 5% of stream habitat used for spawning/rearing habitat)...... 48

xiv stream. a) Low habitat availability (50 m2 stream and no rearing habitat available – first year/age 0 migration) leads to low density small amplitudes cycles in high harvest conditions and fixed-point equilibrium in no harvest conditions. b) High habitat availability (100 m2 stream and both fry and juvenile rearing habitat present – migration following two growing seasons) leading to high density cycles under no harvest conditions and fixed-point equilibrium in high harvest conditions...... 50

Figure 2.7 Field results from gillnet sampling illustrating gillnet catch (number of fish caught in standard gillnet set) as a proxy for population density and estimated asymptotic size as a measure of maximum size, the different point colours represent stream habitat index (limiting/orange: < -1.5 ; intermediate/blue: -1.5- 0.5; abundant/green: >0.5)...... 51

Figure 3.1 Schematic representation of the expected abundance and growth outcomes of variation in stream and lake habitat capacity. The lines are the range in habitat capacity (black for lake capacity, grey for stream capacity), the points are the assumed equilibrium values of stream and lake abundance, the shading represents the predicted maximum fish size attained (purple is smaller fish, yellow is larger fish and green is intermediate values). At high stream capacity (A, B), early survival is high leading to a large input of juveniles from the stream to the lake, resulting in a stunted growth population when lake capacity is low (A) and intermediate fish size when lake capacity is high (B). At low stream capacity (C, D), early survival is low resulting in intermediate lake abundance and fish growth at low lake capacity (D) while lake abundance is low and fish grow to larger sizes at high lake capacity (C)...... 80

Figure 3.2 Location of the sampling lakes (yellow points) and nearby population centres (purple) over satellite imagery in the province of British Columbia (BC). .. 81

Figure 3.3 Histograms of the frequency of (a) GDD values, (b) presence/absence of other fish species, (c-e) stream habitat descriptors, and (f-i) electrofishing catch- per-unit-effort (CPUE in N/sec) for the total sample and age-classes 0, 1 and 2. The number of sampling sites is 39 for all measures...... 82

Figure 3.4 Correlation biplot presenting the results of the RDA of CPUE of age- classes of rainbow trout (0, 1, 2 – purple lines) as a function of habitat variables (turquoise arrows) pool variability, mean stream width, velocity, percent of large woody debris (LWD), and percent of substrate that is cobble, boulder and sand/silt. The yellow dots are the different reaches sampled, their projection on the lines and arrows estimates their value. The angle between the different age- classes (purple lines) and habitat variables (turquoise arrows) represent the correlation between them, the closer the lines/arrows are the more correlated they are while lines/arrows pointing in opposite directions suggest that they are negatively correlated...... 83

xv Figure 3.5 Histograms of the frequency of (a) gillnet catch, (b) recruits (number of Age-2 fish in the gillnet catch), (c) growth parameters k and (d) L∞, (e) maximum observed size calculated as the mean FL of the five biggest fish caught in the gillnet, (f) fishing mortality rate estimated from the effort cameras, (g-h) female age and size at 50% maturity. The number of sampling sites is 39 for all measures...... 84

Figure 3.6 Graphical representation of the most parsimonious model for life-history traits related to (a) maturation and (b) growth. The traits are: female length at 50% maturity (in mm - a) and maximum size described as the estimated asymptotic size (L∞ in mm – b top) and maximum observed size calculated as the mean FL of the five biggest fish caught in the gillnet (in mm – b bottom). On the left side for each trait, plot of the linear relationship (solid black line) between the trait and the most significant predictor, total gillnet catch in a) and WUA/Lake area ratio in b). The dashed lines represent the 95%CI and the black dot the mean of the 39 populations sampled. On the right, the influence on each population parameter of predictor GDD (only part of the models in b) and presence/absence of other fish species (OtherSp) if all other variables remain constant is presented using the mean of all populations (black dot) and the influence on the parameter of a positive (+) and negative (-) change of one standard deviation of the variable of interest for GDD. For the presence/absence of other species variable the value resulting from the absence (0) and presence (1) of other species is presented, which is more than a standard deviation but is more realistic in the context of a binary variable. The asterisks at the top of each plot represents the level of significance for each predictor (blank > 0.1, ‘•’ ≤ 0.1, ‘*’ ≤ 0.05, ‘**’ ≤ 0.01, ‘***’ ≤ 0.001)...... 85

Figure 4.1 Flowchart describing the predictions of the population dynamic outcomes of a depletion (increased harvest) of adults from a population with a low juvenile (stream) to adult (lake) habitat ratio (left side) and a population with a high juvenile to adult habitat ratio (on the right)...... 107

Figure 4.2 Map of the study area with lakes and stream network, the three study lakes are presented (Pantano in yellow, Stubby in turquoise and Today in purple) and the inset map shows the location of the study area in the southern portion of the Province of British Columbia...... 108

Figure 4.3 Timeline of the sampling activities conducted on the three study lakes...... 109

Figure 4.4 Histograms of the number of fish caught in harvest gill nets by age-class during the two years of depletion (2014 and 2015) in Lakes Stubby and Today. .. 109

Figure 4.5 Variation in growing-degree-days (GDD) across study years (2012-2016, ClimateWNA - Wang et al. 2016) for the three study lakes (Pantano, Stubby and Today)...... 110

xvi Figure 4.6 Population abundance estimates for each study year in lakes Pantano (top – yellow), Stubby (middle – turquoise) and Today (bottom – purple), the grey polygons are the 95%CI of the estimates and the vertical black dotted lines are the occurrence of the depletion events in the two experimentally manipulated lakes...... 111

Figure 4.7 Estimates of number of fish in each age-class for each study year (2012- 2016) in lakes Pantano (yellow), Stubby (turquoise) and Today (purple), the black bars represent the 95%CI and the lake-year plots with a grey background are post-depletion...... 112

Figure 4.8 The relationship between growth rate and fish length at first capture before the depletion (Pre) and after the depletion (Post) and interaction with study lakes, only lake Stubby post-depletion (turquoise, right plot) presented a significant difference with the other lakes and with its pre-depletion growth rate (adjR2 = 0.48, p-value = 2.7 x 10-9)...... 113

Figure 4.9 Graphical representation of the relationship between growth rate, fish length at first capture, population density and growing-degree-days (adjR2 = 0.42, p-value = 3.6 x 10-7). The values of density and GDD presented correspond to the mean value observed plus/minus the standard deviation of the value (high/low)...... 114

Figure 5.1 Map of the province of British Columbia with all lakes larger than 4 ha categorized and coloured based on known rainbow trout observation and stocking history. Main population centres are represented by black triangles and merged Clearwater/North Thompson sub-sub-watersheds by a grey polygon...... 141

Figure 5.2 Map of the merged Clearwater/North Thompson sub-sub-watersheds with lakes larger than 4 ha categorized and coloured based on known rainbow trout observation and stocking history. The stream network connecting the lakes is shown in blue (only stream third order and over symbolized), while the inset map (black rectangle) delineates the data in Figure 5.4/Table 5.4 (all stream orders are symbolized in the inset)...... 142

Figure 5.3 Graphical representation of the logistic regression of the probability of rainbow trout presence by the distance to nearest rainbow trout observation. Black dots are data from the validation subset (known presence and absence of rainbow trout population), while the black curve is the model’s fit and the grey polygon is the 95% confidence interval (p-value < 0.01)...... 143

xvii 0.7) because there is no barrier and it is close to a known rainbow trout observation; b) the influence of stream order (represented as the thickness of the blue stream network), where lake B has large fourth order tributaries but lake C, of similar area, has small first order tributaries limiting the availability of early life stage habitat; and c) the different scenarios of accessibility, as lake D is connected to the road network (in grey) while lake C is over 100 m from a road and necessitates a hike-in. Note that the lake categories are simplified into “present” (wild and naturalized origins, in yellow), “present but stocked” (purple) and “absent” (brown). In example a) the black lake contour identifies lakes previously categorized as “unknown” and for which the probability of presence is predicted, in examples b) and c) lakes identified as “present” are lakes in which rainbow trout were observed or predicted (probability > 0.5) while “absent” lakes are where no rainbow trout were observed or predicted (probability < 0.5)...... 144

Figure 5.5 Maps of the merged Clearwater/North Thompson sub-sub-watersheds summarizing the predictions of a) rainbow trout distribution based on known observations and stream network connectivity predictions, where light yellow areas present a higher density of observed or predicted presence of rainbow trout, and darker orange and brown areas have limited rainbow trout observations or present gradient barriers limiting connectivity, only naturally reproducing populations are included (no “stocked” lakes); b) estimated stream to lake area ratio (WUA/lake area) based on lake order and known lake area where blue areas present larger stream for smaller lakes, beige areas are intermediate and red areas have smaller streams for larger lakes; and c) accessibility based on total travel time (min) from Kamloops, including estimated driving time and, when required, hiking time. Dark brown areas have the shortest travel time, lighter brown and beige areas are intermediate and turquoise areas are over 5 hours total travel from Kamloops...... 146

Figure 5.6 Graphical representation of the relationship between lake tributaries’ weighted useable area (WUA) and maximum Strahler stream order of lake’s tributaries. Black points are field data, the black line is the model fit and the grey polygon is the 95% confidence interval (adjR2 = 0.5, p-value = 3.5 x 10-7)...... 148

Figure 5.7 Map summarizing the ratio of the predictor of fish production (WUA/lake area in m2/ha) over recreational fishing demand (inverse of driving plus hiking time in min), low values (dark brown) present areas of concern where lakes present either or both low production and short travel time, while high values (blue) represent areas of lesser concern in which lakes are highly productive and/or presenting a long travel time...... 149

Figure E.1 Posterior predictive distribution of growth model for each lake, the circles represent each fish captured (mature in blue, immature in orange), the thick black line is the growth model’s predicted size at age and the bands are the credible intervals...... 200

xviii

List of Symbols, Abbreviations and Nomenclature

Symbol Definition

AIC Akaike Information Criterion BC British Columbia CPUE Catch-per-unit-effort FPE Fixed-point equilibrium FWA Freshwater Atlas GDD Growing-degree-days HSI Habitat Suitability Index LLC Lake Landscape-Context LWD Large woody debris PIT Passive integrated transponder PSRF Potential scale reduction factor RCC River Continuum Concept RDA Redundancy analysis TDA Total Dissolved Solids VBGF von Bertalanffy Growth Function WUA Weighted useable area

xix

Chapter 1: Introduction

Populations are regulated through feedbacks in reproduction and survival rates.

Variability in these vital rates within a population can be driven by both density independent, events such as a drought or other stochastic events, and density-dependent processes (Sinclair 1989). Density-dependent processes include predation and direct or indirect competition of individuals for limited resources. Population abundance and vital rates can also vary greatly across nearby populations due to their local biotic and abiotic conditions.

Population regulation in fish is often assumed to happen almost exclusively in the early life stages. Although there is evidence that strong regulatory forces, such as competition for food and predation, lead to low survival of young fish (Houde 1989), there are also indications that regulation can occur later in life, mostly through density- dependent growth and size-dependent mortality (Post et al. 1999, Ylikarjula et al. 1999,

Lorenzen and Enberg 2002). In addition to direct effects of the environment, vital rates can also be indirectly altered by changes to related traits. For example, changes to growth rate can have profound effects on both survival and reproduction. An increase in growth can lead to better survival through decreased vulnerability to predators and increased competitive advantage (Post and Evans 1989b, Sogard 1997). Fecundity is also impacted by growth as larger individuals tend to produce more and better quality eggs (Koops et al.

2004, Rollinson and Hutchings 2010). These indirect effects of growth and fecundity suggest that recruitment, like survival, is the composite result of a series of density- dependent processes that can impact one or multiple life stages. Therefore, a full

understanding of fish population dynamics needs to consider all life stages and the factors influencing their growth and survival.

A main driver structuring patterns of size dependent survival, recruitment and growth is habitat availability and quality. The habitat is the area of natural environment inhabited by organisms of a species, the habitat is characterized by the presence of both biotic and abiotic resources used by the species of interest. Most models linking fish populations to habitat limit their description of the fish population to its yield to the fishery or focus only on one life stage. One well known habitat model example is the morphoedaphic index which uses mean lake depth and total dissolved solids (TDS) to predict production (Ryder et al. 1974). Shuter et al. (1998) developed a more complex model linking habitat predictors (lake size and TDS) to life history parameters (growth and maturation) and fish vulnerability to recreational harvest to identify lake trout

(Salvelinus namaycush) populations at risk of overfishing. This model provides a description of the demography and harvest dynamics, but it does not detail the processes by which habitat impacts the life stages and resulting population dynamics. Although fisheries scientists have long recognized the importance of habitat for fish populations and communities, fisheries and habitat management have mostly been done separately, neglecting the links between habitat and population dynamics. Linking life stages to their habitat requirements is particularly important for species experiencing ontogenetic habitat shifts and where the dynamics of each stage need to be connected to obtain the overall population outcome and identify when density-dependence occurs in a population’s life cycle (Hayes et al. 2009, Andersen et al. 2017).

Identifying which life stage is most impacted by density-dependence gives insight into demography and size structure. This approach provides information fundamental for the sustainable management of fisheries, both commercial and recreational. The possibility of overfishing is well-known in commercial fisheries (Hilborn et al. 2003), but it has long been believed that recreational fisheries were immune to overfishing due to the lack of economic incentive to harvest declining populations (Johnson and Carpenter

1994, Hansen et al. 2000). However, recent evidence suggests that many recreational fisheries have been declining all over the world (Australia: McPhee et al. 2002, Canada:

Post et al. 2002, US: Coleman et al. 2004) and that many of the challenges previously solely associated with commercial fisheries also impact recreational fisheries (Cooke and

Cowx 2004, 2006).

In light of the threats to recreational fisheries there has been a move away from

“one size fits all” type of fishing regulations, while also recognizing that using a system- specific management approach, where each population is assessed and has its own set of regulations, is not realistic as most recreational fisheries managers have sparse resources to monitor large areas composed of hundreds of fish populations (Carpenter and Brock

2004). A broad-scale approach to management has been recommended, but it necessitates a thorough understanding of the drivers of the heterogeneity of both the fish population and angler dynamics across the landscape.

High quality research integrating the human dimension into fisheries models through social-ecological systems has been conducted in recent years (Johnston et al.

2010, 2013, Hunt et al. 2011, Carruthers et al. 2018). This research integrates information not only on the impact of angling on fish populations but also details angler motivations

and behaviour (Arlinghaus 2006, Beardmore et al. 2011, Ward et al. 2013b, Fenichel et al. 2013, Dabrowska et al. 2014). On the fish side, the effect of climate is one of the many abiotic factors that have been presented as filters to species distribution to explain changes in fish community composition at broad scales (Magnuson et al. 1979, Tonn

1990, Jackson et al. 2001). Temperature variation has also been used to describe fish growth for populations across landscapes (Venturelli et al. 2010, Lester et al. 2014, Ward et al. 2017, Varkey et al. 2018). Most of the landscape management models developed only present a general model of fish production or use stocked fish populations where density is manipulated (Post et al. 2008, Carruthers et al. 2018). Integration of human and fish population dynamics requires the ability to scale from within lake processes driven by habitat limitation to the scale of entire landscapes driving environmental productivity, connectivity, angler movement and behaviour.

Describing fish population dynamics at the landscape scale first requires a thorough understanding of the processes through which fish density varies at the local

(lake) scale and how it is influenced by drivers such as habitat availability and climatic conditions. Then, the sources of variation of those drivers at the landscape-scale can be used to predict the broad scale distribution of different types of fish populations and their production. Classification systems based on landscape features describing the limnological characteristics of lakes have been described and recognized as useful for management and conservation of aquatic ecosystems (Soranno et al. 1999, 2010), but have not yet been developed for fish population dynamics.

My main objective with this thesis is to understand the density-dependent processes that shape fish populations and explain how neighbouring populations can vary

greatly in their density and size structure. My thesis is based on the hypothesis that life stage specific habitat availability impacts growth and survival of the specific life stage, but also influences the whole population, acting as a bottleneck determining when in the life cycle and how density-dependence occurs. I then use lake-scale drivers of population dynamics to describe fish populations at the landscape scale, a useful tool for the prioritization of monitoring and conservation initiatives.

Throughout the studies forming this thesis, I use the British Columbia (BC) rainbow trout (Oncorhynchus mykiss) fishery, but the general ideas presented are applicable to other species and fisheries. Rainbow trout is one of the most popular sport fish species in BC. The BC interior is the core native range of rainbow trout in Canada, the species has also been widely introduced across the country and the world (Scott and

Crossman 1973, McPhail 2007). This fishery in BC is made of two components, systems stocked annually and systems where rainbow trout reproduce naturally. Approximately

800 systems are stocked annually to provide fishing opportunities and lessen the fishing pressure on wild stocks, as rainbow trout is a species that has experienced declines in BC, particularly near population centres (Post et al. 2002, 2008). A large literature has developed characterizing many aspects of the biology and human dimensions of the stocked BC rainbow trout fishery (Cox and Walters 2002, Parkinson et al. 2004, Post et al. 2008, Askey et al. 2013, Ward et al. 2013a, Varkey et al. 2016, Mee et al. 2016,

Carruthers et al. 2018). Stocking in these systems artificially removes much of the biological feedback processes by controlling ‘recruitment’. On the other hand, studying naturally reproducing rather than stocked populations provides the opportunity to study the full range of complex dynamics within fish populations. Lake-dwelling rainbow trout

is an adfluvial species that experiences ontogenetic habitat shifts, uses tributaries for spawning and rearing of juveniles, and then migrate back to the lake. The use of those separate habitats makes this an ideal system to investigate the influence of multiple habitats on population dynamics. Also, using rainbow trout populations confined to small lakes (<100 ha) makes it logistically easier to follow fish throughout their life cycle. This approach would be difficult in larger systems, lakes or oceans.

This thesis is written in manuscript format and composed of four research chapters (Chapters 2-5). The first (Chapter 1) and final (Chapter 6) chapters are the bookends of the thesis presenting a general introduction situating my work and a conclusion that brings the thesis together and suggests future research ideas. Chapters 2,

3 and 4 all explore the influence of habitat features on population dynamics, using different approaches, while Chapter 5 details landscape-scale applications of results from the previous chapters that are useful for fishery and habitat management.

Chapter 2 is the theoretical framework orienting my thesis and is strongly informed by literature on salmonid population dynamics in both lakes and streams. Here,

I detail my hypotheses on the processes through which habitat influences rainbow trout population dynamics. I develop an age-and-size-structured population model that follows lake-dwelling rainbow trout from egg deposition in the stream, through the timing of juvenile stream to lake migration, to the adult stage in the lake, and close the loop with reproduction of eggs in the stream. For each life stage I detail how growth and survival are influenced by density-dependent and -independent processes, such as climate and harvest. I run the model through a series of simulations where habitat availability, climate

and recreational harvest vary. Stage- and size-structured interactions lead to complex dynamics that require a modelling approach to fully capture its outcomes.

The research in Chapter 3 uses an empirical approach where I compare populations presenting a gradient of the ratios of early (stream) to late (lake) habitat availabilities. I investigate the fine-scale associations of the stream habitat with juvenile fish and its implications for stream to lake migration. I explore the drivers of stream and lake capacities and compare the field observations to the Chapter 2 predictions using the demography, growth and maturation parameters of the populations sampled.

In Chapter 4, I use an experimental approach to assess the influence of habitat availability on the compensatory response following density manipulation. The manipulation imitates the effects of harvest on populations that vary in their early to late habitat availability ratios. Following on the two previous chapters, I compare the compensatory response of the experimental lakes to the predictions of the Chapter 2 model regarding the impact of harvest.

In Chapter 5, I move to a broader spatial scale and develop a methodology to model rainbow trout distribution using stream connectivity. I then use the local observations of the three previous chapters describing lake-scale processes to make predictions of fish production and population size structure at the landscape-scale. I combine this biological information with crude estimates of recreational fishing demand to assess the sensitivity of the different populations to overfishing and habitat changes, demonstrating how this approach can be used for fishery and habitat management.

Chapter 2: Habitat Availability and Ontogenetic Shifts Alter Bottlenecks in Size-

Structured Fish Populations

Introduction

Population dynamics are dictated by patterns of individual growth and survival.

For many organisms, survival is highly size-dependent (Sauer and Slade 1987, Sogard

1997, Mangel and Stamps 2001). Harvest for commercial or recreational purposes can be an important source of mortality that is almost always non-random as harvest selects based on large size (Fenberg et al. 2008). In addition, larger individuals within populations are often less susceptible to predation where mortality from gape-limited predators decreases with increases in individual size (Tonn and Paszkowski 1986).

Individual size is the result of growth rate which can be associated with environmental variables such as climate, for example temperature has been linked to different physiological processes that control growth in ectotherms leading to faster growth in warmer climates (Atkinson 1994). Compensatory individual growth can also result from reductions in population density, due to reduced competition for food resources.

Populations of the same species exposed to environmental variation (e.g. climate, habitat availability, harvest pressure) can express plasticity in individual growth leading to contrasting dynamical outcomes among populations due to the interplay between growth and survival (Werner 1988, Ebenman 1992).

Many indeterminate growing species experience growth of multiple orders of magnitude throughout their life cycle resulting in a size-structured population. This ontogenetic size variation also leads to changes in ecological niche that present

themselves as shifts in diet or habitat motivated by habitat segregation resulting from competitive interactions, predator avoidance, or resource seeking behaviour (Werner and

Gilliam 1984, Werner 1988, de Roos et al. 2003, de Roos and Persson 2013). Growth rate variation can influence the transition of organisms from one life stage to another modifying the timing of ontogenetic habitat shifts and impacting inter and intraspecific interactions (Werner and Gilliam 1984). For example, slow growth could prolong the time a cohort spends in a high predation risk habitat, impacting its survival. It can also influence the time spent in a habitat that overlaps with another competing life stage which can further decrease growth rate and increase mortality. Density-dependent processes, such as growth, are impacted by habitat or resource availability and quality, which can vary throughout a species life as different stages experience ontogenetic niche shifts in resource and habitat use.

Competitive and predatory interactions between and within size-classes, and the availability of the different habitats used by a species throughout its lifecycle can influence growth and survival of a specific stage resulting in a population bottleneck that limits the size and/or maximum number of individuals in that stage and has repercussion on the overall population (de Roos and Persson 2013). Harvest can also be an important external source of mortality that limits a life stage and impacts population abundance.

Bottlenecks can happen at any stage of a life cycle and can occur at different stages among populations of the same species depending on their environment. Since abundance is the result of survival through multiple stages, the carrying capacity of the system is not necessarily determined by the reproductive potential of the adult stage, but by the most limiting stage. Changes in growth impacting size structure can also impact carrying

capacity as individuals of different size will differ in their per biomass metabolic requirements, meaning that a habitat with a set amount of resources can support more or less fish depending on their individual size and associated energetic requirements.

Therefore, limiting the study of a population to a single life stage can lead to an incomplete understanding of the processes controlling population regulation (Halpern

2004). Hence, determining the right spatial and temporal scope of study is fundamental in the investigation of population dynamics.

Fish are a suitable organism to study the link between ontogenetic habitat shift and population dynamics as they typically experience size increase over several orders of magnitude and present a succession of life stages among which there is sufficient niche overlap to lead to intercohort competition (Werner and Gilliam 1984, Ebenman 1988).

Several studies have linked variability in physical habitat to population dynamics, particularly in salmonids (Milner et al. 2003, Lobón-Cerviá and Rincón 2004, Ayllón et al. 2010). Early life stages, where the two main causes of mortality are starvation and predation, are especially important in determining population recruitment (Houde 1989,

Cowan et al. 2000). Therefore, models that aim to describe fish population dynamics should explicitly account for the processes driving growth and survival of young fish.

Species like lake-dwelling rainbow trout (Oncorhynchus mykiss) are ideal to examine fish population dynamics as they do not perform large scale migrations like their anadromous counterparts but use distinct habitats at the adult (lake) and juvenile stages (spawning and rearing in the lake tributaries). Population dynamics have been extensively studied among salmonids (Grant and Kramer 1990, Elliott 1994, Cattanéo et al. 2002, Milner et al. 2003,

Grant and Imre 2005, Lobón-Cerviá 2007), including rainbow trout in lakes (Northcote

1962, Post et al. 1999, Cox and Walters 2002, Parkinson et al. 2004) and in streams

(Keeley and Slaney 1996, Keeley and McPhail 1998, Imre et al. 2004, Wood et al. 2012).

The large body of literature in this area provides a strong base on which to develop a cohesive multi-habitat model aimed at understanding bottlenecks to population growth.

The objective of this study is to assess the qualitative and quantitative outcomes of variation of habitat availability, density-dependence, climate and harvest on population bottlenecks and dynamics in spatially and size-structured fish populations. This will be done using populations simulated with an age-structured population model and by comparing the simulation outcomes to empirical data, with more detailed comparisons in

Chapters 3 and 4.

Methods

Model development.

We used an age-structured population model detailing processes in both the adult habitat (lake) and the early life stage habitat (stream) to follow rainbow trout through their life cycle and represent the impact of spawning and rearing habitat availability to fish population dynamics. The model was parameterized using, when possible, data from field and laboratory studies on rainbow trout or steelhead populations. When such data were not available, we used data from related salmonids.

The model time step is a year and results are presented as a snapshot of the population at the end of the growing season, in autumn. In general, lake-dwelling adult rainbow trout migrate to streams in the spring to spawn and then return to the lake (Scott and Crossman 1973, McPhail 2007). The eggs hatch in the stream early in the summer

and the juvenile fish spend a few weeks to two years in the stream before moving to the lake where they grow and mature (Northcote 1962, Scott and Crossman 1973, Rosenau

1991). Natural survival is stage dependent in the stream, but accompanied by density mediated migration behaviour (Slaney and Northcote 1974, Elliott 1994, Keeley and

McPhail 1998). While, in the lake, survival is density- and size-dependent (Post et al.

2008) and mature fish experience spawning mortality and harvest mortality resulting from recreational harvest (Parkinson et al. 2004). Growth rate varies with climate and density (Ward et al. 2017). Annual egg production varies with number of mature fish and their size. The egg production of year t becomes the number of eggs deposited in year t+1. The general framework of the model is presented in Figure 2.1 and Table 2.1 contains the full parameters set.

Stream habitat.

The stream can be considered as being divided into three habitats, characterized by their environmental condition (Raleigh et al. 1984, Bjornn and Reiser 1991, Keeley and Slaney 1996), and occupied by particular life stages: 1) spawning habitat used by adult fish to construct redds in well oxygenated gravel substrate where eggs are deposited and fertilized; 2) fry rearing habitat characterized by more cover (cobble, boulder, aquatic vegetation, woody debris) and shallow pools, particularly located in streamside for Age-0 fish; and 3) juvenile rearing habitat for immature fish older than age 1, similar to 2) but with deeper pools with more instream cover. In the model, spawning habitat varies in abundance but is always present, while rearing habitat can vary in abundance or be absent. In the wild, a situation where spawning habitat is available, but rearing is not

could occur if a stream does not present any fish cover, or if stream flow is highly seasonal and sufficient only for spawning in the spring but dries up later in the summer.

Fish density in the stream is presented as simple density of age-class i fish in their specific habitat (Di) and as percent habitat saturation (PHS). The PHS metric describes how much of an habitat is used based on fish density and individual territory size (Grant and Kramer 1990):

푛 (2.1) 푃퐻푆 = 100 × ∑푖=1 퐷푖/퐷푖_푚푎푥 where Di_max is the maximum density defined as:

(2.2) 퐷푖_푚푎푥 = 퐴푠푡푟푒푎푚 × 푃푖 ∕ 푇푖 where Astream is the total stream area, Pi is the proportion of stream area that is covered by

2 habitat suitable to age-class i and Ti is the territory area (in m ) for age-class i. The size of territory defended by salmonids is a power function of fish size (Grant and Kramer 1990,

Imre et al. 2004). Territory size has also been shown to vary with food abundance and population density through the effect of intruder pressure (Slaney and Northcote 1974,

Dill et al. 1981, McNicol and Noakes 1984, Keeley and McPhail 1998, Keeley 2000,

Wood et al. 2012). In our model, food abundance is assumed to be constant over a given area, thus as density increases food availability to individuals decreases. Territory size increases with fish size and shrinks with increases in density. The parameters used to calculate territory size are from an experiment conducted on juvenile steelhead trout by

Imre et al. (2004):

(2.3) log 푇푖 = −2.296 − 0.365 × log 퐷푖 + 2.027 × log 퐿푖 where Li is the fork size at age i assuming that all fish of a same cohort in the same habitat follow the same growth curve. Territory size at low densities is limited by the area

of habitat available. At high fish densities, minimum territory size is the minimum foraging area necessary for fish to meet their energy needs and results in territoriality playing an important role in limiting population density (Grant and Kramer 1990). The relationship between minimum territory size area (Aminterr) and fish size (Li) is:

푏푚푖푛푡푒푟푟 (2.4) 퐴푚푖푛푡푒푟푟 = 푎푚푖푛푡푒푟푟 × 퐿푖 where aminterr and bminterr are parameters obtained by calculating territory size using equation 2.3 for high fish densities that corresponded to a PHS of 150% (based on the salmonid interspecific fixed territory size equation developed by Grant and Kramer

1990).

Lake habitat.

The lake is treated in the model as a single homogenous habitat. Lake fish density is calculated using effective density (ED), a metric developed to represent competition between individuals of different size through their consumption rate. Effective density has been shown, using bioenergetics and empirical models, to better represent variation in density-dependent fish growth within size-structured populations than is numeric or biomass density (Walters and Post 1993, Post et al. 1999).

∑푛 퐿2 (2.5) 퐸퐷 = 푖=1 푖 A푙푎푘푒 where Li is individual fish length (mm) and Alake is lake area (ha). In the model fish length is calculated for each age-class (see growth section below) for individuals from each possible migration schedule (see migration schedule below).

Migration schedule.

The timing of migration of young rainbow out of their rearing streams and into lakes is highly variable, from as soon as a few days after emergence to one or two growing seasons after hatching (Northcote 1962, Scott and Crossman 1973, Rosenau

1991). The timing of migration can be highly stream specific as variation in migration age between fry emerging from tributaries of a same lake has been observed in New

Zealand lake rainbow trout populations (Rosenau 1991, Hayes 1995). In the model, the possibility of growing and overwintering in the stream depends on the availability of rearing habitat. The model captured the range in timing of migration observed in nature by simulating four scenarios of habitat availability and resulting migration schedule to the lake (see Figure 2.2 for a diagrammatic description): 1) no rearing habitat available: fish migrate to the lake immediately after emergence as would be observed in intermittent stream that dry up early in the summer for example (Erman and Leidy 1975); 2) only fry rearing habitat available: fish migrate to the lake after having spent one growing season in the stream; 3) both fry and juvenile rearing habitat available, but the habitats do not overlap: fish migrate to the lake after two growing season but fry and juvenile do not compete in the stream, limited intercohort competition would occur in highly heterogeneous stream habitat, for example where deep pools would be available for older fish to use (Ayllón et al. 2010); and 4) same habitat availability as in 3) but the different rearing stages compete for a shared portion of the homogeneous stream habitat. Here the high niche overlap leads to intercohort competition as observed in salmonids like brown trout (Crisp 1993, Nordwall et al. 2001, Lobón-Cerviá 2005, Grant et al. 2011), if juvenile fish are saturating their habitat (PHS ≥100%) they obtain all the shared habitat, if

fry are saturating their habitat and juvenile are not, then fry get the shared habitat, and, finally, if both life stages present a PHS below 100% each life stage obtains half of the shared habitat.

Once rearing habitat is saturated (PHS ≥100%), fish that are not able to secure a territory will be displaced (Grant and Kramer 1990) and complete an early migration to the lake, and lake conditions will determine their survival and growth for that year.

Therefore, a single cohort can have individuals in both the stream and lake habitats.

Natural mortality.

Egg to emergence survival is a function of egg density (Degg) and follows a

Beverton-Holt recruitment type curve where the number of fry that emerge (Dfry) reaches an asymptote once the carrying capacity of the habitat is reached:

푎퐵퐻×퐷푒푔푔 (2.6) 퐷푓푟푦 = 푏퐵퐻+퐷푒푔푔 where aBH is the asymptote or the maximum fish density determined by density- dependent mortality related to redd superimposition. Once the spawning habitat is saturated with redds newly arriving females will superimpose redds on top of others, destroying previously deposited eggs (Hayes 1987, Essington et al. 1998). Therefore, only the eggs deposited by the latest females should survive. This asymptote is calculated as:

푁̂푒푔푔 (2.7) 푎퐵퐻 = 퐷푟푒푑푑_푚푎푥 × 푁푓푒푚

1 (2.8) 퐷푟푒푑푑_푚푎푥 = 퐴̂푟푒푑푑

where Dredd_max is the maximum density of redds that can be dug and Negg/Nfem is the average number of eggs per female, while the numerator of equation 2.8 is the spawning habitat area and Aredd is the average redd area.

The second parameter in equation 2.6, bBH, is the egg density at 50% of the asymptote:

푎 (2.9) 푏 = 퐵퐻 × 푆 퐵퐻 2 emerg_max where Semerg_max is the maximum survival rate from egg to emergence which is the result of density-independent processes such as scouring and dewatering, its value corresponds to the highest survival rate observed in a steelhead emergence study by Ward and Slaney

(1993).

In both lake and stream habitats, natural mortality is modeled as an annual event that occurs between two growing seasons. Fry and juveniles that remain in the stream will survive at a fixed rate determined by their age (S0_max and S1_max). In the lake, natural survival (Si) for all life stages varies with length and effective density (ED) and is modeled using the method and parameters presented by (Post et al. 1999, Parkinson et al.

2004)):

(−푚푖퐸퐷) (2.10) 푆푖 = 푆푖푚푎푥 e

퐿푖−1.81 (2.11) 푆푖푚푎푥 = 0.49 [ ] 1+0.49 (퐿푖−1.81)

−6 −5 (2.12) 푚푖 = −3.27 × 10 ln(퐿푖) + 1.21 × 10 where Simax is the maximum survival for age-class i that varies with length (Li) and mi is an exponent also related to size (see Figure 2.3a). Maximum age was set at age 7, following that last year of life, natural survival was set to 0.

Maturity is modeled as age-determined and knife-edge. The age at maturity is fixed at three years old, maturation is usually later for female than male rainbow trout and most female reach maturity by age 3 (Cox 2000, Ward et al. 2017). We make the assumption that maturation is age determined as, for rainbow trout, size at maturity tend to be more variable than age at maturity. This is an important assumption and we assess the influence of this parameter on the model in a sensitivity analysis described further in this section. Mature fish are assumed to spawn every year and experience post-spawning mortality due to the energy required to migrate to the stream, prepare redds, spawn and defend redds. Post-spawning survival is fixed at 0.5 to represent the high spawning mortality experienced by rainbow trout, this value was chosen arbitrarily, and its impact is evaluated in the sensitivity analysis.

Harvest mortality.

Harvest is modeled by presenting recreational fishing mortality. Vulnerability to harvest is age-determined, knife-edge and fixed at three years old meaning that fish aged three and older have the same vulnerability. Harvest mortality is based on Cox and

Walters (2002) limited vulnerability (LV) model developed for rainbow trout recreational fishery in the province of British Columbia (BC) where most of the native range of rainbow trout in Canada is located. The LV model is driven by annual angling effort (E).

In this model, fish reactivity and angler and fish spatial distributions change rapidly leading to fish alternating between available and unavailable states. Fishing mortality rate

(f) is defined as:

푞푣퐸 (2.13) 푓 = 2푣+푞퐸 where q is a catchability coefficient of 0.09 ha x angler-days-1 and v is the instantaneous turnover rate (per year) between pools of available and unavailable fish, the harvest parameters are based on data collected in rainbow trout populations of small lakes in BC

(Cox 2000, Cox and Walters 2002) and assumes that all fish caught are harvested. This instantaneous mortality rate can then be used to obtain an annual harvest mortality rate

(U):

(2.14) 푈 = 1 − 푒(−푓)

Growth.

Size at emergence from the gravel is fixed at 2 cm (Murray 1980), based on the assumption that the size of females does not impact the size and survival of eggs, and that hatch timing is consistent across populations. Following emergence, fish growth rate is determined by climatic conditions and density of fish in their habitat following annual spawning, natural and harvest mortality (Figure 2.3b). The measure of density used is

PHS in the stream for each age’s habitat and total effective density in the lake.

Growth in both the stream and lake is modeled using a von Bertalanffy growth function (VBGF):

(−푘푖) (2.15) 퐿푖 = 퐿∞(1 − 푒 )

-1 where Li is length at age i, L∞ is the asymptotic length (cm) and k (yr ) is a metabolic constant. In this model, growth is influenced by both climate and density. Climate’s influence on growth was modeled as a positive linear relationship between growing-

degree-days above 5°C (GDD) and k, we assume that the relationship between GDD and k is linear within the temperature range modeled.

(2.16) 푘푐푙푖푚 = 푎푐푙푖푚 × 퐺퐷퐷푗 + 푏푐푙푖푚 where GDDj is the GDD value for lake j during a specific year, aclim is the slope and bclim is the intercept of a linear regression between the VBGF parameter k calculated for six rainbow trout lakes by Cox (2000) and annual growing-degree-days calculated for the appropriate year using ClimateBC 5.30 (Wang et al. 2012). Annual GDD corresponds to the difference in number of Celsius degrees between the mean temperature and the chosen base of 5°C. Climate, described using GDD above a base of 5°C, and density have both been linked to growth variation for rainbow trout in BC (Ward et al. 2017) and other fish species (Venturelli et al. 2010, Lester et al. 2014).

Density (PHS in the stream, effective density in the lake) affects growth through its impact on asymptotic length (L∞). In the stream, L∞ decreases with increases in saturation (PHS) following a negative power curve as described by Grant and Imre

(2005) who compiled data on six species of stream-dwelling salmonids and concluded that density-dependent growth was stronger at low density rather than large density.

−0.3 (2.17) 퐿∞_푠푡푟푒푎푚 = 64.96 × 푃퐻푆

In the lake, L∞ decreases linearly with increases in effective density following the parameters used by Parkinson et al. (2004) based on data by Post et al. (1999):

−5 (2.18) 퐿∞_푙푎푘푒 = 64.83 − 9.6 × 10 × 퐸퐷 where ED is the effective density (cm2/ha) in the lake.

The correlation between growth rate and maximum size is not explicitly modeled but is an emergent property of the density-dependent processes presented. By using

measures of density (PHS and effective density) that include the impact of fish size, an increase in growth rate leads to higher density and decrease of maximum size.

The VBGF parameters obtained for a specific climate and density were used to build Walford plots that represent size at age based on size at the previous age:

(2.19) 퐿푖+1 = 푎푊 + 푏푊퐿푖

(2.20) 푎푊 = 퐿∞(1 − 푏푊)

−푘 (2.21) 푏푊 = 푒 where Li is the length at age i, aW is the Walford intercept, and bW is the Walford slope.

The Walford plot parameters are then used in the model to calculate growth for each cohort from one age-class to the other. Table 2.1 presents the values of the parameters used to model the impact of density on the Walford plot intercept (aW and bW) directly instead of the relationship with L∞ presented above.

Fecundity.

Fecundity is calculated based on fish size. Parkinson et al. (2004) used data collected by local BC fish hatcheries to estimate the number of eggs produced per female

(Neggs_i) depending on female size (L):

−5 2.9 (2.22) 푁푒푔푔푠_퐿 = 2.49 × 10 × 퐿

The total number of eggs (Neggs) produced by the population for a year is:

푛 (2.23) 푁푒푔푔푠 = ∑푖=1(푁푖 × 푃푓푒푚 × 푁푒푔푔푠_푖) × 푃푒푔푔푠_푣푖푎 where Ni is the total number of fish in age-class i, Pfem is the proportion of females in that age-class and Peggs_via is the proportion of eggs that are fertilized and viable (Van Winkle

et al. 1998). Eggs produced during a year are then deposited in the stream in the following spring.

Simulations.

To simulate the impact of stream habitat area on the fish population, we ran the model using different values of stream area and proportion of stream used for spawning, fry rearing and juvenile rearing. For simplicity, the lake area was fixed at 1 ha for all simulations. The stream area varied from 0.5 to 500 m2 which, in relation to a 1 ha lake, is similar to stream to lake area ratios observed within the native range of rainbow trout in BC. Values ranging from 1 to 45% of total stream area were used for the proportion of stream used for spawning. The range of values used for rearing are presented in Table 2.1 and correspond to the migration scenarios explained above. GDD values ranging from

1000 to 2000 were used to simulate the climatic conditions across the native rainbow trout distribution in BC (range of 1036-1954 GDD observed by Ward et al. 2017).

Finally, harvest mortality rate ranging from 0 to 0.8 (corresponding to fishing effort 0 to

5000 angler-days x ha-1 x year-1) were simulated, harvest rate is usually lower on most rainbow trout lakes in BC (mean of ~0.6 following equation 2.14, Post et al. 2008), but extreme values were used to assess how high harvest mortality affects population dynamics. The variables used for the simulations are summarized in Table 2.1. A total of

6720 factorial combinations were simulated (7 stream areas x 6 spawning habitat proportions x 4 migration scenarios/rearing habitat proportions x 5 climatic conditions x

8 harvest levels). Each simulation ran over 200 years to ensure an equilibrium was reached.

Analysis.

Simulation results were first compared based on the status of the resulting population. After 200 years of simulations, we examined whether the population collapsed, and if so, whether it occurred as a result of overharvest or stream habitat limitations. We also assessed whether the population reached an equilibrium and if so, whether the equilibrium was a fixed-point, where abundance reached a unique value, or a bounded oscillation cycle, where abundance oscillates but repeats a cycling pattern.

Abundance, which is the same as density of fish per hectare since lake size was fixed, and mean size at maturity (fixed at age 3 in the model) were used to describe the resulting population characteristics.

To assess the impact of various processes on population characteristics, alternative processes were contrasted to a baseline model. The baseline model corresponds to intermediate values of stream habitat and climatic conditions under which the model consistently reached a fixed-point equilibrium (50 m2 stream area, 25% proportion spawning habitat, migration from streams after one growing season, 1500

GDD) under no harvest and harvest mortality rate of 0.7. The harvest values used represent a situation with no harvest and what is considered high harvest in the BC rainbow trout recreational fishery (corresponding to a fishing effort of 100 angler-days x ha-1 x year-1, Post et al. 2008). The processes assessed in this way are presented in Table

2.2.

A sensitivity analysis was conducted to assess the sensitivity of key parameters

(detailed in Appendix A: Table A.1) by varying them (+/- 5%, +/- 10%, +/- 25%), and contrasting to the population characteristics from the baseline model. The following

population characteristics were examined to test the sensitivity of the model parameters: population status (collapsed, fixed-point equilibrium or stable cycle), total abundance of fish in the 1 ha lake and mean size at maturity in the lake.

Comparison to empirical data.

The simulation results are compared to empirical data from 33 lake-dwelling rainbow trout populations located in the British Columbia interior collected between

2013 and 2016 (lake information detailed in Appendix B: Table B.1). The fish population data collected consist of autumn gillnet samples that follow British Columbia’s provincial guidelines (Ward et al. 2012), gillnet catch (number of fish per two net set) is used as a proxy for fish density in the lake and the asymptotic length (L∞) is estimated by fitting the

VBGF (equation 2.15) to the size at age data obtained from gillnet fork length measurements and otolith aging. L∞ is used to compare the size of fish between populations. Environmental data (lake characteristics such as location, size and depth) was acquired from provincial databases (British Columbia Ministry of Forests, Lands and

Natural Resource Operations; Freshwater Fisheries Society of British Columbia), and field measurements of stream habitat availability and fishing effort using trail cameras

(van Poorten et al. 2015). Spawning and rearing habitat availability were summarized by calculating a habitat index, the index consisted of the sum of the standardized estimated areas identified as stream spawning, Age-0 rearing, Age-1 rearing, and lake habitats.

Simulations and field results were compared on the base of the trends of population density and maximum size associated with early life stage habitat availability. The population status (collapse, cycle or fixed-point equilibrium) could not be compared since

the gillnet sampling were only done for one year which would not allow to detect population cycles, and sampling was only conducted in established rainbow trout populations, therefore no collapsed population would be observed in the field.

Results

The influence of increased stream habitat availability on population dynamics, abundance and size structure are assessed under no harvest and high harvest scenarios

(Figure 2.4). For those combinations of parameters that resulted in fixed-point equilibria,

Figure 2.5a presents the fish abundance and associated maximum size combinations and the regions of combinations covered by cycling populations. Higher GDD leads to larger size for the same abundance, unless abundance is high (Figure 2.5b). In general, as stream habitat quantity (area) and quality (availability of spawning and rearing habitat) increase, fish density increases and fish size decreases (see Figure 2.5c).

Population dynamics: Collapse.

A small number of the simulated populations (<1%) collapsed while there was no

2 harvest. Those were characterized by limited habitat where stream area was 0.5 m , of which 1% was spawning habitat and no stream rearing habitat was available. As such, the collapse occurred because the habitat limitation in the early life stage habitat (stream) results in individuals migrating to the lake immediately after emergence and experiencing high mortality in the lake. In these conditions, no matter the climatic conditions of the system (GDD), modeled fish populations could not sustain themselves. In 1.3% of simulations, collapses occurred due to a combination of medium-high harvest rate (≥ 0.6)

and low early life stage habitat availability (total stream area ≤ 2m2, spawning area ≤ 5% of total stream area and no stream rearing habitat available). As stream habitat and GDD increased, higher harvest was required to produce a collapse. No overharvest induced collapse were observed in simulations where streams were larger than 2 m2 for the 1 ha lake area used for all simulations.

Population dynamics: Fixed-point equilibria.

In most simulations (81% of the populations simulated) the populations reached a fixed-point equilibrium where population abundance and size structure is constant.

Population abundance increased as stream habitat quantity and quality improved, while size at age decreased. Increases in GDD led to large increases in lake fish size when stream habitat was limiting, but when stream habitat was abundant, the difference in fish size was minimal and fish abundance decreased (Figure 2.5b).

Population dynamics: Cycles.

For 17% of the simulations the population did not reach a stable fixed-point equilibrium, but a stable cycle emerged. The cycles are characterized by the presence of a strong cohort that dominates for a period of time. We observed two types of cycles

(Figure 2.5a). The first type, low-density cycles, results from populations with intermediate size streams, no rearing habitat and high GDD resulting in low fish abundance and large fish size (Figure 2.6a). This cyclic pattern is more prevalent under high harvest where older age-classes have high harvest mortality. Young fish recruiting into the lake have higher survival due to the low density of older conspecifics (i.e. low

intercohort competition). In larger streams, lower GDD and lower harvest are necessary to transition from fixed-point equilibria to cycling.

The second type of cyclic outcomes, high-density cycles, resulted from large streams with at least some Age-0 rearing habitat available leading to high abundance of small fish surviving their first year of life and migrating into the lake (Figure 2.6b). In this case, high harvest tends to switch cyclic behaviour to fixed-point equilibria.

Abundant stream habitat provides a refuge for juveniles for 1-2 years from the intense competition in the lake, producing a lag between egg production and impact of these young fish on density-dependent processes in the lake. This results in a strong cohort that survives better to the adult stage where it produces more eggs which feeds the cycle observed and produces a new strong cohort. The down part of the cycle occurs due to the intercohort competition between the adult strong cohort and the young fish entering the lake, resulting in low survival of the young fish to the adult stage, lessening competition at lake entry for the next strong cohort. With larger stream habitat and more spawning and rearing habitat, and higher GDD, more harvest mortality is required to drive the cycle to a fixed-point, and for streams larger than 50 m2, cycles were observed even with the highest level of harvest (mortality rate of 0.8) when the proportion of stream used for spawning and rearing habitat was high.

Assessing the importance of ecological processes.

The functional forms of biological processes represented here led to quantitative and qualitative changes in dynamics and demographic characteristics of the resultant fish populations (Table 2.2). The baseline population used led to a fixed-point equilibrium

under both low and high harvest pressure. The modification of four processes led to stable cycles, but only when there was no harvest. In all cases, harvest returned the population to a fixed-point similar to what was observed in the high density cycles described in the previous section where increased harvest mortality rate has a stabilizing effect. Modifying egg survival from an asymptotic relationship to a humped-shaped

Ricker type curve (Process 1, Table 2.2) or to a fixed survival (Process 2) led to increased emergence survival in the stream. Similarly, when fish growth in streams followed a linear decrease with density instead of a negative exponential function (Process 6) or both stream growth and survival were density independent (Process 8), more fish survived in the stream leading to the development of strong cohorts causing cycles. By making all adult fish vulnerable to harvest (Process 14) a high harvest mortality rate was enough to deplete a population, producing the only collapse as a result of a change in an important biological process.

An increase of over 25% in size at age 3 (maturity) was observed when lake growth was density independent (Process 9) while a decrease of over 25% was observed when lake survival and stream and lake survival were density-independent (Processes 10 and 13). This decrease in size was accompanied by an increase in abundance. Density- independent lake growth and survival also lead to an increase in abundance (Process 11).

Parameter sensitivity analysis.

The majority of the parameter perturbations of +/- 5, 10 and 25% did not alter either the qualitative dynamic outcomes or the quantitative outcomes greater than the magnitude of the perturbations (Appendix A: Table A.1). In a few situations, the fixed-

point equilibria of the baseline simulation was changed to cycles. With harvest, when age at maturity (Agemat) is lowered or age at vulnerability to harvest (Agevuln) is increased, stable cycles were observed. Cycling also occurred when egg emergence survival is increased by increasing the asymptote of the Beverton-Holt recruitment function (aBH) by

25%. Increased stream growth rate, either through 25% increase in parameters determining how growth rate decrease with density in the stream (astream and bstream) or through a modification of how growth is impacted by climate (bclim), also led to cycles.

Finally, a decrease of spawning survival (Sspawn) also led to cycles.

Quantitatively, lake abundance increased greater than the parameter perturbation while size at age 3 decreased when: 1) age at maturity decreased; 2) age at vulnerability to harvest increased; 3) lake natural survival increased through a decrease in mi, the exponent of the relationship between density and survival that varies with fish size, the resulting abundance and size varied more than the change in the parameter for all levels of parameter perturbation (5, 10 and 25%) when there was no harvest but only at 25% with harvest mortality. Conversely, lake abundance decreased and size at age 3 increased with increased age at maturity and decreased age at vulnerability.

Comparison to field data.

Similar to the simulation results, the populations sampled in the field present smaller fish associated with high gillnet catch and larger fish associated with lower catch, none of the populations presented a high gillnet catch of large fish (Figure 2.7). Most of the populations that had limited early life stage habitat are in the lower right quadrant of the plot presenting low densities and large size. Populations from intermediate habitat

availabilities present a range of densities and size, while populations where high amounts of habitat is available mostly present small fish size. Because the site sampled presented negligible fishing effort, harvest mortality was not considered or presented for the empirical data.

Discussion

The model developed uses density-dependent ecological processes and spatial ontogeny to describe population bottlenecks resulting from variation in availability of habitat used for reproduction and juvenile growth. The addition of simple harvest dynamics makes this model relevant to management of species that experience commercial or recreational harvest. The simulation results illustrate how populations occupying similar adult habitat can differ in abundance and size structure when early life stage conditions differ. When early life stage (spawning and rearing) habitat is limited, populations are more sensitive to increases in mortality (presented as increases in harvest mortality in this model). The simulations also showed that a minimum quantity of spawning habitat is required to support a viable population, a threshold value has to be reached even in the absence of stochastic events. Population abundance increases as the early life habitat quantity and quality increases but, due to the effect of density- dependence in growth in the adult habitat, size at age tends to decrease leading to stunted populations. When large amounts of quality spawning and rearing habitat are available, the population bottleneck moves to the habitat of older stages.

The simulation results present a narrow range of combinations of population abundances and maximum individual sizes with size decreasing at high population

abundances as growth is modelled as a density-dependent compensatory process. A similar trend is also observed in the empirical data where populations with larger asymptotic length present lower density. Fish grow larger at low density, and growth is slowed at high density (Rose et al. 2001). The results are consistent with observations of density-dependent growth resulting from exploitative competition in lake-dwelling rainbow trout (Post et al. 1999).

The use of a multi-stage habitat model demonstrates the nature of the population bottlenecks impacting and shaping a population by linking limiting processes at different stages of the life cycle to the resulting population dynamics. Salmonid recruitment limitation has mainly been linked to bottlenecks occurring at the spawning (Knapp et al.

1998) or juvenile stages (Elliott 1985, Crisp 1993, Bond et al. 2008). Limitation at the adult stage is observed in many species (de Roos and Persson 2013), but could the lack of studies presenting limitations of the adult stage be an artifact of the complexity of salmonids life cycles? As presented by Milner et al. (2003), in anadromous fish, it is often difficult to study the processes occurring at sea and impacting the adult life stages.

Also, by its nature, the marine environment presents large available habitat that is less likely to present space limitation, but could freshwater salmonids be more likely to be limited at the adult stage? For example, Elliott (1985, 1994) has shown that the main source of regulation for anadromous brown trout of Black Brows Beck was at the very early life stage, but did not find any evidence of early life stage regulation in the resident brown trout population of Wilfin Beck and claimed that in that case the bottleneck was instead placed on the adult stage (Elliott and Hurley 1998). We argue that, given the range of habitat combinations presented in the simulations, population regulation of lake-

dwelling rainbow trout could come from all three presented stages (egg, juvenile, adult) and impact a population’s persistence, abundance and individual growth.

Population collapses were observed when spawning habitat availability was strongly limited, in this deterministic model this is the result of high natural mortality and insufficient compensation at low density, but it also corresponds to what would be expected based on stochastic dynamics alone in nature. At low spawning habitat availability, a bottleneck led to low egg to emergence survival resulting in low overall population densities as observed in stream-dwelling golden trout where the lack of appropriate substrate limits redd density (Knapp et al. 1998). The empirical data supports this model prediction as the largest fish were observed in systems with low fish density and limiting spawning and rearing habitat.

Limiting the available stream juvenile rearing habitat led to early lake migration, forcing young fish into less optimal lake habitat where they were in competition with older/larger fish, resulting in lower survival of the younger age-classes. Here, the stream functions as a nursery habitat where juvenile fish can find shelter from predation and competition (Beck et al. 2001). The lack in appropriate nursery habitat can impact whole population and communities by controlling the juvenile stage through habitat limitation as seen in marine fishes and invertebrates (Wahle and Steneck 1991, Beck 1995, Mumby et al. 2004, Sundblad et al. 2014). Nursery habitat’s positive impact on a population can also be the result of an increase in individual growth as in steelhead trout (anadromous rainbow trout) where access to estuarine nursery habitat has been linked to enhanced juvenile growth leading to a larger size at ocean entry that increases ocean survival (Bond et al. 2008).

When lake habitat is limiting (in this simulation framework where lake size is constant this is seen when stream habitats are abundant) the population bottleneck is in the adult stage. In the empirical data presented this occurs when lakes have large or many perennial tributaries offering abundant cover to young fish. The abundant stream habitat leads to high early age growth and survival, but once the juveniles migrate in the lake the high density in that habitat leads to lower growth rate resulting in slow growing stunted fish populations. Ylikarjula et al. (1999) explained that stunted populations are ultimately the result of resource limitations that can arise through increases in juvenile and/or adult survival or increases in intraspecific competition. In the simulations, stunting is indeed observed when adult survival is increased at low harvest, and when large rearing habitat is available, providing shelter to juvenile and increasing their survival.

In the first type of cycle, spawning habitat is limiting, stream rearing habitat is absent but GDD is high. The bottleneck on the early life stages (spawning and juveniles) leads to low stream and lake abundances, which intensifies individual growth and is further enhanced by the higher GDD. Similar to what has been observed by Borgstrørm et al. (1993) in brown trout, the early immigration to the lake leads to high intercohort competition between juvenile age-classes where one cohort becomes dominant. In the present model, this type of cycle was observed under high harvest which increased spawner mortality and periodically removed the older age-classes so that only one or two age-classes of spawner were present. A simulated 25% decrease in spawning survival caused a decrease in the number of spawners and the loss of older age-classes, resulting in cycling in a previously stable population. The high spawner mortality limits egg input and following year’s Age-0 cohort while also decreasing the competitive pressure on the

juvenile fish present in the lake, the following year that same advantaged juvenile cohort has less competition from both the Age-0 coming from the stream (deposited by previous year’s spawners) and from the fished adult population. Once that advantaged cohort becomes mature, they are subjected to harvest but their higher abundance and large size allow them to produce more eggs than the previous cohort producing another strong cohort and perpetuating the cycle. Older age-classes have the potential to dampen the cycles through their competitive interaction with early migrants, similar to the pressure applied by cannibalism which usually has a stabilizing effect (Claessen et al. 2000). Once the adults experience higher mortality, however, cycles arise.

Cycles of larger amplitude and longer period were observed in situations where abundant spawning and rearing habitat were available, these populations were characterized by high abundance in both the stream and the lake, with a small size at age in the lake. Here, the stabilizing force of early life stage mortality is absent since that habitat was not limiting. Hence, high growth and survival in the stream led to high intercohort competition once these abundant and large-bodied young fish enter the lake at

Age-1, creating a strong cohort. Parameter sensitivity analysis results support this explanation, cycling occurred when stream growth and survival were increased through changes in the functional form, density-dependence and parameters of the stream processes. The time spent by the young fish in the stream shelters them, for up to two years, from competition with larger/older fish in the lake. This additional delay occurring between reproduction and lake entry of the young fish adds to the complexity of the life cycle and further destabilizes the population. This type of cohort cycle resulting from intercohort competition is similar to what has been observed in vendace (Coregonus

albula - Hamrin and Persson 1986, Helminen and Sarvala 1994), roach (Rutilus rutilus -

Persson et al. 1998) and Eurasian perch (Perca fluviatilis - Persson et al. 1998, de Roos et al. 2002).

The presence of cycles is contrary to what was observed in an experiment on lake- dwelling rainbow trout by Post et al. (1999) who did not encounter situations where younger age-classes out-competed adults and speculated that rainbow trout populations should be stable rather than oscillate. This difference might come from the experimental systems used by Post et al. which only had two age-classes present (Age-0 and -1) and did not experimentally simulate the full age structure of a natural population.

Additionally, the input of young fish in their experiment was through stocking. They used a 10-fold variation in stocking density across treatments, which might not have resulted in sufficient density variation as generated by the different habitat combinations simulated here.

Variation in age at maturity and age at vulnerability to harvest, both fixed at three years old in the baseline simulations, impacted the model simulation results. The interaction between those two parameters is important. If maturity occurred before individuals become vulnerable to harvest there are now adults that are not subject to harvest mortality and have higher survival. In the sensitivity analysis this situation led to population cycles due to the higher population fecundity. Life-history theory suggests that earlier maturity could occur if early growth is accelerated (Ylikarjula et al. 1999,

Roff 2002), population bottlenecks leading to lower juvenile densities and higher growth rates could then lead to earlier maturation. For example, situations where fecundity is reduced due to high adult mortality and ample rearing habitat is available or where early

life stage mortality is high due to limited spawning habitat, but abundant habitat is available to older life stages. In contrast, if individuals become vulnerable to harvest before reaching maturity, adult density will decrease as fewer adults survive to older age, additionally decreasing population fecundity. Delayed maturation could also occur through slow growth, if adult survival is increased through a decrease in harvest mortality for example (Ylikarjula et al. 1999). In this model, age at maturity and age at vulnerability were both fixed to age 3 to simplify the model and its interpretation.

Allowing age at maturity and age at vulnerability to vary with size would likely impact population dynamics.

This multi-habitat age-structured population model links environmental conditions and habitat availability at different stages of a species life cycle to population dynamics outcomes. This information has direct applications in species management and habitat conservation. Our study species, rainbow trout, is exploited by recreational fisheries around the world meaning that our findings could provide insight to managers for this type of system but also for the management and conservation of other organisms that experience ontogenetic habitat shifts leading to population bottlenecks. Managers often oversee extensive areas composed of numerous systems that they cannot sample regularly. Information on the relative availability of habitats throughout ontogeny could assist in identifying likely bottlenecks in production and to identify systems at risk of being impacted by habitat loss or overharvest which would help managers orient their sampling strategy and allocation of resources. Knowledge of the importance of life history specific habitat requirements for recruitment, abundance and dynamics can also help prioritize initiatives of conservation and restoration of key habitats.

Table 2.1 Parameters used in the model and their source, for parameters that are calculated the associated equation number is provided. Parameter Value Description/Reference Physical Habitat and Density Stream 2 Astream 0.5-500 Stream total area (m ) Pspawn 0.01-0.45 Proportion of the Astream that is spawning habitat Pfry 0.25-1 Proportion of the Astream that is fry rearing habitat Pjuv 0.25-0.75 Proportion of the Astream that is juvenile rearing habitat PHS Eq. 2.1 Percent habitat saturation (Grant & Kramer 1990) Di_max Eq. 2.2 Maximum fish density, varies with territory size Individual fish territory size (m2), varies with L and density (Parameter values from T Eq. 2.3 i i Imre et al. 2004) 2 Aminterr Eq. 2.4 Individual fish minimum territory size (m ), vary with Li aminterr 0.0002 Parameters defining how minimum territory size varies with Li (calculated based on Aterr bminterr 3.1921 at high densities of PHS 150%)

Lake Alake 25 Lake area (ha) ED Eq. 2.5 Effective density (cm2/ha) Natural Survival Stream Dfry Eq. 2.6 Density of fry that have survived from egg to emergence stage 2 aBH 1982 Beverton-Holt recruitment a parameter (eggs/ m ), eq. 2.7 bBH 3303 Beverton-Holt recruitment b parameter, eq. 2.9 2 Âredd 0.3 Average redd size (m ) from Ottaway et al. 1981 2 Dredd_max 3.33 Maximum redd density (redds/ m ), eq. 2.8

Average egg produced per female (based on eq. 2.22 assuming average female size is 35 N /N 595 egg fem cm) Semerg_max 0.3 Maximum egg to emergence survival (Ward and Slaney 1993) S0_max 0.8 Maximum survival rate during first year of life in the stream S1_max 0.9 Maximum survival rate during second year of life in the stream

Lake Si Eq. 2.10 Survival rate in the lake, varies with fish length and ED (Parkinson et al. 2004) Si_max Eq. 2.11 Maximum survival rate, varies with Li (Parkinson et al. 2004) mi Eq. 2.12 Exponent of the relationship between ED and S, varies with Li (Parkinson et al. 2004) Harvest E 0-5000 Fishing effort (angler-days x ha-1 x year-1) Agevuln 3 Vulnerability to harvest is knife-edge and age determined Instantaneous turnover rate between pool of available and unavailable fish (year-1) from v 1.61 Cox and Walters 2002 q 0.09 Catchability coefficient (ha x angler-days-1) from Cox and Walters 2002 f Eq. 2.13 Instantaneous harvest mortality rate (Cox & Walters 2002) U Eq. 2.14 Annual harvest rate for fully vulnerable fish (Cox and Walters 2002) Growth Li Eq. 2.15 Length (cm) of fish at age i Lemerg 2 Length (cm) of fish at emergence (Murray 1980)

Climate Effect (Stream and Lake) k Eq. 2.16 Von Bertalanffy k, metabolic constant, varies with climate (GDD) GDD 1000-2000 Growing-degree-days > 5°C aclim 0.0002

Slope (a) and intercept (b) of the relationship between k and GDD (based on k values bclim 0.0122 from Cox 2000 and GDD obtained for lakes for the appropriate year from ClimateBC 2016) bW 0.66-0.97 Walford slope varies with climate, eq. 2.21

Density Effect Eq. 2.17- L Von Bertalanffy L , asymptotic length, varies with fish density ∞ 2.18 ∞ Stream Walford plot intercept, varies with PHS following a negative power curve (Imre et al. a 4-12 W_stream 2005), eq. 2.20 astream 20 Parameters of the decline (negative power curve) of aW_stream with increase in PHS bstream -0.3 Lake aW_lake 5-18 Walford plot intercept, varies with EDpost (Parkinson et al. 2004), eq. 2.20 alake 18.8 -5 Parameters of the decline of aW_lake with increase in EDpost (Parkinson et al. 2004) blake 2.8 x 10 Spawning Agemat 3 Age at maturity (knife-edge), value from Cox 2000 Sspawn 0.5 Survival rate post-spawning Pfem 0.5 Proportion of fish of each age-class that are female Neggs_i Eq. 2.22 Number of eggs produced per female, varies with female Li (Parkinson et al. 2004) Peggs_via 0.9 Proportions of eggs that are fertilized and viable (VanWinkle et al. 1998) -1 Neggs Eq. 2.23 Total number of eggs produced by the population (eggs x year )

Percent Change Harvest Population Size at Process Change from the Baseline Models Mortality Lake Dynamics age 3 Rate Abundance (cm)

0 Cycle -3* -1* (1) Egg survival follows a Ricker recruitment function 0.7 FPE -2 1

0 Cycle -8* 5* (2) Egg survival is constant (density independent) 0.7 FPE -17 7

0 FPE 0 6 (3) Territory size is density independent (size-dependent only) 0.7 FPE 7 2

(4) Early migrants from stream to lake die: in stream, fish without a territory 0 FPE 0 0 die instead of migrating to the lake 0.7 FPE 0 0

0 FPE 6 3 (5) Stream growth is density independent (climate-dependent only) 0.7 FPE 9 3 (6) Stream growth follows a linear decrease with density 0 Cycle -22* -5* 41

0.7 FPE -14 -2 (7) Stream migration behaviour is density independent (juveniles stay in the 0 FPE 14 -15 stream even if no territories are available and survive following age-specific survival rate) 0.7 FPE 16 -13

(8) Stream growth and migration behaviour are density independent, (5) and 0 Cycle -15* -19* (7) together 0.7 FPE 1 -16

0 FPE -6 25 (9) Lake growth is density independent (climate-dependent only) 0.7 FPE 3 5

0 FPE 1104 -39 (10) Lake survival is density independent (size-dependent only) 0.7 FPE 256 -40

0 FPE 1800 24 (11) Lake growth and survival are density independent, (9) and (10) together 0.7 FPE 1254 2

0 FPE 0 23 (12) Stream and lake growth is density independent, (5) and (9) together 0.7 FPE 7 1

(13) Stream migration behaviour and lake survival are density independent, 0 FPE 635 -49 (7) and (10) together 0.7 FPE 268 -38

0 FPE 0 0 (14) Adult fish are fully vulnerable to harvest 0.7 Collapse -100 NA

Figure 2.1 Schematic representation of the model, the input variables, important processes presented through the rainbow trout life cycle and output variables.

Figure 2.3 a) Survival rate in the lake varies with effective density and size (line thickness represents fish size) b) Growth in both the stream and the lake follows a von Bertalanffy growth curve which varies with fish density (percent habitat saturation in the stream and effective density in the lake) and growing-degree-days (GDD), density impacts the asymptotic size (L∞) while GDD impacts growth rate (k). The blue lines represent low density populations where juveniles migrate to the lake right after egg emergence while the orange lines represent high density populations where juveniles spend two growing seasons in the stream (from egg emergence to age 1), their growth trajectory is modified once they enter the lake. The solid lines represent high GDD while the dashed lines are low GDD.

Figure 2.6 Total abundance of juveniles (dashed lines) and adults (solid lines) in the lake in no harvest (blue) and high harvest (0.7 – orange) conditions over the last 15 years for four different habitat simulations. For all four simulations, the climatic variable is fixed at 1500 GDD and spawning habitat is 45% of the stream. a) Low habitat availability (50 m2 stream and no rearing habitat available – first year/age 0 migration) leads to low density small amplitudes cycles in high harvest conditions and fixed-point equilibrium in no harvest conditions. b) High habitat availability (100 m2 stream and both fry and juvenile rearing habitat present – migration following two growing seasons) leading to high density cycles under no harvest conditions and fixed-point equilibrium in high harvest conditions.

Chapter 3: Interaction of Juvenile Stream Rearing and Adult Lacustrine Habitats

on Growth and Demography of Wild Rainbow Trout: Inferring Processes from

Field Observations

Introduction

Availability of suitable habitat limits the carrying capacity of an environment.

However, species often go through ontogenetic habitat shifts where each stage-specific habitat could contribute to population regulation. If the lack of a specific habitat type reduces a population’s capacity to fully utilize habitats at other stages, then a production bottleneck is occurring. Such bottlenecks impact the growth and survival of the life stage that uses this specific habitat but will also have repercussion on the subsequent life stages

(de Roos and Persson 2013).

Population bottlenecks impact density-dependent processes and are often identified as the underlying cause of intraspecific variation in life history traits, such as growth and maturation, and demography of fish (Rose et al. 2001). For species that present great plasticity in growth, as do fish, increases in density have been linked to diminished growth rates resulting in stunted populations. Such density-dependent growth in fish results from resource limitations and constrained food consumption (Post et al.

1999, Jenkins et al. 1999, Lorenzen and Enberg 2002). Also, according to life history theory, changes in growth rate resulting in a stunted population can impact the timing of maturation. However, the relationship is complex, and the maturation outcome is dependent on the mechanisms causing high density (Stearns 1992, Ylikarjula et al. 1999,

Roff 2002).

Rainbow trout (Oncorhynchus mykiss) populations that reside in small lakes

(<100 ha) are ideal to study the effect of population bottlenecks on individual growth and maturation parameters. These populations use two clearly discrete habitats where the lake is used for adult rearing, and tributaries are used for spawning and juvenile rearing.

Furthermore, the adult habitat is relatively small creating greater variation in the ratio of adult to juvenile habitat and allowing for easier sampling than would be possible in large lakes, or the ocean, as is the case for other salmonids. In Chapter 2, I presented the idea that habitat limitations in specific life stages of rainbow trout can cause population bottlenecks that impact the overall population dynamics, and that the bottleneck severity and timing influence population demography and life history traits. This conclusion was the result of simulations of an age-structured population model for lake-dwelling rainbow trout in which the impact of habitat capacity limitation at the spawning/rearing

(tributaries) and adult stage (lake) on population dynamics. I described habitat capacity as abundance of physical habitat (e.g. area available, quality of substrate for a specific life stage), but also included the impact of variation in climate and harvest intensity. The model simulations revealed that limiting habitat capacity at the early life stage (low stream capacity/large lake capacity) led to high early mortality, resulting in low overall population density and larger individual maximum size. In contrast, abundant early life stage habitat and limiting capacity of the adult habitat (high stream capacity/low lake capacity) led to high survival through early life stages and high overall density stunted populations (Figure 3.1).

With this study, my main objective is to test the predictions of the complex dynamics presented in the Chapter 2 model and investigate empirically the influence of

habitat availability on capacity of both the stream and lake habitats, and how population bottlenecks impact demography, individual growth and maturation. To do so I collected a large field data set from wild rainbow trout in the province of British Columbia (Canada).

I describe populations as wild if they are naturally reproducing without active input from fish stocking. Some of the populations are wild and native, but others could be considered wild naturalized, meaning that they were established or impacted by historical stocking of fish of native or non-native origin, but are now maintaining themselves through natural reproduction. First, I will describe early (stream) and later (lake) life stage habitat capacity and assess the relationship between habitat capacity and population parameters.

Then, I will investigate the impact of population bottlenecks acting on habitat capacity, and varying in severity and timing, on demography, growth and maturation.

Methods

Site selection and study design.

I collected fish population and habitat data from 39 lakes and their tributaries were collected for this study (Appendix B: Table B.1, Figure 3.2). Study sites were located in the interior of British Columbia (BC) within the native range of rainbow trout in Canada (Figure 3.2). Sites were selected to maximize the gradient in: (1) lake and stream size, (2) elevation and (3) known recreational fishing effort (based on a combination of information from previous sampling, expert opinion from local anglers and biologists, and presence/absence of fishing lodges). Waterbodies also had to be accessible through ≤ 200 m hike for sampling logistics. Only lakes where no stocking had occurred in the past 20 years were selected, ensuring that the density observed was the

result of natural recruitment and not enhanced through stocking. Most sampled lakes were rainbow trout monocultures, but six lakes had other fish species such as: redside shiner (Richardsonius balteatus), largescale sucker (Catostomus macrocheilus), longnose sucker (Catostomus catostomus) and northern pikeminnow (Ptychocheilus oregonensis).

Each site was visited once for lake sampling and once for stream sampling in the same year between 2013 and 2016, resulting in sampling of fish size structure and demographic variation along broad environmental gradients. This snapshot approach assumes that each fish population has reached a stable abundance and age/size frequency distribution and stability in recruitment, growth, and natural and fishing mortality over time.

Stream sampling.

Data on juvenile trout size structure and stream habitat were collected simultaneously once for each system in the late summer (August 4-24 of each sampling year). Sampling was done by visiting every tributary (inlets and outlets) of the lakes.

Streams were divided into 100 m reaches starting from the lake and continuing until a major fish barrier was reached, or the stream went dry or flowed underground, or until a maximum of 15 reaches had been sampled. In each reach, fish were collected using a backpack electrofisher when water depth was sufficient and safe to wade in (20 cm to

100 cm). Sampling was done moving upstream (starting at the lake for inlets or at the furthest reach for outlets). Fish were identified, measured (fork length - FL in mm) and then released in their reach of origin. No lethal sampling of fish was conducted in the stream; therefore, age determination was done using length frequency distributions.

Juvenile rainbow trout are typically represented by few age-classes and have substantial size differences between ages at those early stages, which were verified using reported size at age for rainbow trout (Scott and Crossman 1973). Electrofishing effort was recorded using seconds fished per meter for each reach. At each reach, habitat descriptors consisting of stream width, depth, velocity, temperature, conductivity, habitat structure, substrate, cover and disturbance were measured. The habitat sampling protocol is modified from Barbour et al. (1999), and a complete list of the data collected and summary statistics are presented in Appendix C (Table C.1).

Stream habitat descriptors.

Several measures were used to describe stream habitat, they range in complexity from stream quantity consisting of stream mean wetted width (m) to more complex measures providing information on habitat quality. A Habitat Suitability Index (HSI) developed for rainbow trout (Raleigh et al. 1984) was used to describe stream habitat quality. The HSI ranges from 0.0 to 1.0 (where 1.0 is the highest possible quality) and represents habitat quality, which is assumed to be correlated with a system’s carrying capacity (US Fish and Wildlife Service 1981). The riverine rainbow trout HSI model follows a life-stage approach and is divided in five components (Raleigh et al. 1984):

Embryo (CE), Fry (CF), Juvenile (CJ), Adult (CA) and Other (CO). Each component is calculated using environmental variables related to the abundance of a specific life stage

(CE, CF and CJ) or to all life stages (CO). The CA component was not used in this analysis as lake-dwelling rainbow trout spend most of their adult life in the lake and the goal was to detail the stream habitat used by the immature life stages. The CO component contains

variables that relate to water quality and food supply and impact all life stages. The CE,

CF, CJ and CO components are included in the calculation of the total HSI index which is calculated for each reach sampled and averaged over all tributaries of each study lake.

The variables used, and their detailed calculations modified from Raleigh et al. (1984), are presented in Appendix D and Table D.1. The total HSI is multiplied by reach area to obtain a weighted useable area (WUA), a descriptor that summarizes both quality and quantity of habitat available. The resulting WUA is then summed over all reaches and streams of a specific lake to obtain a measure of the available quality stream habitat for that lake. Both the stream width and WUA descriptors are presented as total values when looking at stream capacity and as habitat ratios, divided by lake area to standardize across the different lake sizes, for the lake capacity analysis. HSI is presented as the mean index value across reaches for a given lake (ranging 0-1).

Habitat and age-class associations.

Associations between age-classes of fish present in stream segments and stream habitat characteristics were explored using a redundancy analysis (RDA) with 9999 permutations (Legendre and Legendre 2012). Number of fish caught per second (CPUE) of all age-classes of rainbow trout in stream was used to construct the dependent matrix.

A matrix of environmental variables constituted the explanatory portion of the analysis.

The best explanatory variables for the model describing CPUE of age-classes were then selected using forward selection and the function ordi2step of the R package vegan

(Oksanen et al. 2017). All variables were centered on their mean and standardized by 1σ prior to analyses. Each canonical axis was tested for significance to determine if they

explained more variation than random (Legendre and Legendre 2012). The results are plotted in reduced space (correlation biplot – scaling of type 2) which shows the relationships among environmental variables, age-classes and each other. The resulting age-classes to environment associations are compared to the habitat variables identified as important for different stages in the HSI literature (Raleigh et al. 1984).

Lake sampling.

Lake area estimates were obtained using the BC Government Fisheries Inventory

Data Queries (FIDQ). Fish sampling for each system consisted of one overnight standard gillnet set consisting of one sinking and one floating seven panel gillnet of varying mesh size (25, 76, 51, 38, 89, 64, 32 mm) following the BC provincial gillnet protocol described in Ward et al. (2012). Sampling was done in the autumn when decreasing water temperature optimizes net catches. For each fish caught weight (g), FL (mm), sex and maturity were recorded, and lapilli otoliths were collected and aged using transmitted light.

Density was described as numerical catch to define lake capacity in terms of the number of fish present. The adult egg production was presented using a proxy of the number of eggs (푁푒푔푔푠) deposited in the stream calculated using the length (FL in mm) of all mature females captured in the gillnet. The relationship between FL and egg production is based on data from Pennask Lake and Beaver Lake rainbow trout which are used as brood lakes for BC recreational stocking (Freshwater Fisheries Society of BC, unpublished data):

2.28 (3.1) 푁푒푔푔푠 = 0.0016 × 퐹퐿

Fishing effort density was estimated from cameras placed on each lake during at least one summer to capture the level of recreational fishing effort following methods in van Poorten et al. (2015). Cameras were set on shore at a location that maximized the portion of the lake captured and programmed to take hourly pictures during daylight.

Images were analyzed using software Timelapse2 (Greenberg and Godin 2015). To account for portions of the lake not captured by the camera and fishing effort that occurs outside of the camera sampling period (between the hourly pictures or earlier/later in the year) observations were converted to full season angler effort estimates following Askey et al. (2018).

Climate has been linked to rainbow trout growth in BC (Ward et al. 2017). The climatic conditions of each system was described using growing-degree-days > 5°C

(GDD), this metric corresponds to the difference in degrees Celsius between the mean air temperature and the chosen base of 5°C. GDD was estimated using the latitude- longitude-elevation for the centroid coordinates of each lake and the decadal average covering their sampling year using the ClimateBC 5.30 software (Wang et al. 2012).

Habitat capacity.

To determine what metric best represents habitat capacity in both streams and lakes, candidate models based on habitat availability and fish production were compared using an information-theoretic approach (Burnham and Anderson 2002). The most parsimonious models according to Akaike Information Criterion (i.e. lowest AIC and models within two AIC units from it) were selected to identify the best predictors of capacity for each habitat.

For stream habitat capacity, total electrofishing catch (all age-classes) was used as a dependent variable and several predictors relating to stream habitat (quantity: stream mean width; quality: mean HSI value; quality and quantity: total WUA) but also to lacustrine adult population egg production (푁푒푔푔푠), climate (GDD) and interspecific interactions (presence/absence of other fish species) were compared.

For lake habitat capacity, gillnet catch was used as a proxy for population abundance and catch of Age-2 fish as a proxy of recruitment, as this is the age at which most fish have migrated from the stream rearing habitat to the lake habitat. The predictors of lake habitat capacity relate to the input of juveniles from the stream (total electrofishing catch and the best descriptor of stream capacity), the quantity of lake habitat available (lake area), the habitat ratio of stream habitat available over lake habitat

(mean stream width and WUA over lake area ratios), the climatic conditions of the lake

(GDD) and the interspecific interactions in the lake (presence/absence of other fish species).

Growth and maturation.

Maturation and growth parameters were contrasted across populations of rainbow trout sampled. Growth was modeled following the von Bertalanffy growth equation:

−푘(푎−푡0) (3.2) 퐿푎 = 퐿∞(1 − 푒 ) where 퐿푎 is the length at age a, 퐿∞ is the asymptotic length, k is the Brody growth coefficient and 푡0 is the hypothetical age at length 0. Maturation was described separately for males and females using age at 50% maturity (퐴50), and 훿 which is the age after 퐴50 when 95% maturity is reached. Ward et al. (2017) took an alternate approach using a

biphasic growth model (Lester et al. 2004) to describe rainbow trout growth of stocked populations in BC. I attempted fitting this model as well, but it gave biologically unrealistic estimates of the asymptotic portion of the growth curve because, overall, too few age-classes of rainbow trout were present in the sample due to the short life span of the species and the selectivity of the sampling gear used which selected against immature age-classes (age 0-1).

Growth-associated traits were characterized for each population using a Bayesian hierarchical model in which a vector 휃푙 represents the lth population’s traits with a multivariate normal distribution (Helser and Lai 2004). The MVN has a mean vector 휇 composed of five traits relating to growth and maturation (퐿∞, k, 푡0, 퐴50 and 훿), it is used to describe the average trait performance for a population subjected to the average environment (hyperpriors), and a variance-covariance matrix ∑, which had a vague inverse-Wishart prior. Parameters 푘푙 and 훿푙 were assumed to be log-normally distributed, which improved numerical performance.

퐿∞푙 ln(푘푙) 푡 (3.3) 휃푙 = 0푙 ~푀푉푁(휇, Σ) for l = 1, 2, …, Nlakes, 퐴50푙 (ln(훿푙))

2 where the main diagonal of Σ (a 5×5 matrix) represents trait-specific variance (e.g., 휎퐿∞) and the off-diagonals represents covariance between different traits (e.g., 2 ). The 휎퐿∞,ln 푘 distributions of 퐿∞and 퐴50 were truncated positive. A normally-distributed likelihood

(truncated positive) with a multiplicative error structure was used to model size at age for each fish i across each population:

(3.4) 퐿푖,푙~푁표푟푚푎푙(퐿̂푎,푙, 퐿̂푖,푙푐푣퐿) with a constant coefficient of variation in size at age 푐푣퐿, and a corresponding standard deviation of 퐿̂푖,푙푐푣퐿. The maturation schedule 푃푀푎 was a function of age and was described using the following logistic function:

1 (3.5) 푃푀푎푖,푙 = (퐴푖,푙−퐴50 ) (− ln(19) 푙 ) 1+푒 훿푙 where 퐴50 was the age at 50% maturity, and 훿 was the age after 퐴50 to reach 95% maturity. The likelihood that a captured fish was mature was:

(3.6) 푀푎푖,푙~ 퐵푒푟푛표푢푙푙푖(푃푀푎푖,푙)

Fishing mortality (Fl) is calculated from estimated fishing effort (퐸푙) following a limited vulnerability model where q is a catchability coefficient fixed at 0.09 ha x angler- day-1 and v is the instantaneous turnover rate between pools of available and unavailable fish fixed at 1.61 year-1 (Cox 2000, Cox and Walters 2002, Parkinson et al. 2004).

The model was implemented in JAGS run through R in rjags and run.jags

(Plummer 2003, Denwood 2016, R Core Team 2017) using 4 MCMC chains. Each chain took 15,000 samples of the posterior, with a burn-in period of 50%, and thinning rate of

50 for a total chain length of 1,650,000 iterations. Starting parameter values were jittered for each chain. Model suitability and MCMC chain convergence to a common posterior mode was validated in several complementary ways. First, the convergence of the

MCMC chains was inspected visually on traceplots. I ensured that parameters had an effective sample size ≥1,000 and low percent error for the Markov Chain relative to the parameter’s standard deviation (Gelman et al. 2013). The Gelman-Rubin diagnostic test was used on each parameter to determine whether independent chains converged to a

common posterior distribution, with potential scale reduction factors (PSRF) < 1.1 suggesting convergence. Graphical posterior predictive checks were used to test for model misspecification and assessed bias with standardized residuals by comparing posterior mean lengths at age to observed lengths at age.

The estimated von Bertalanffy growth parameters k and 퐿∞ were used to test for differences in growth patterns across populations. An observed measure of maximum fish size (MaxObsFL), the mean of the five biggest fish caught in each lake, was also used to describe maximum size without constraining it through the von Bertalanffy growth model and its assumption of asymptotic size. For maturation, I used female and male-specific age at 50% maturity (A50) and associated length at 50% maturity (L50), calculated using the growth parameters obtained. I built candidate models to explore relationships between these parameters and the descriptors identified as important for stream and lake capacity.

The impact of climate (GDD) and interspecific interactions (presence/absence of other species) on population parameters was assessed. Posthoc linear regression was used to estimate relationships between observed environmental data and posterior mean estimates of the growth and maturation parameters. Variables k, 퐿∞ and MaxObsFL did not follow a normal distribution and were transformed using a natural logarithm.

Candidate models were contrasted using an information-theoretic approach (Burnham and Anderson 2002). Models including GDD and presence/absence of other species as explanatory variables were also included. The most parsimonious models according to

Akaike Information Criterion (i.e. lowest AIC model and models within 2 AIC values) were then selected to identify the best set of predictors for each growth and maturation parameter. The effect size of all explanatory variables was calculated using regression

coefficients and percentage of variance explained. Unless otherwise mentioned, all analyses were conducted using R (R Core Team 2017).

Results

Stream habitat descriptors.

Data from stream sampling suggested large variation in fish density, age structure and habitat availability. Stream electrofishing CPUE ranged from 0-2.08 fish/sec with a mean of 0.74 fish/sec, and fish size (FL) ranged between 35 and 217 mm with a mean of

69 mm. Based on modes from length frequency distributions, fish <70 mm were attributed to Age-0, fish 70-150 mm to Age-1 and fish >150 mm to Age-2. Stream wetted width presented a 12-fold variation (range: 0.7 to 8.20 m). HSI values ranged from 0 to

0.75, while overall WUA values ranged from 0 to 1445 m2. A summary of stream electrofishing catch and habitat descriptors is presented in Figure 3.3. Only one site presented water temperature over the 25°C lethal limit for rainbow trout (Raleigh et al.

1984) and many juveniles were observed at that location.

The RDA identified habitat variables associated with the different young rainbow trout age-classes and the results corroborated the HSI methodology. The forward selection of environmental stream variables associated to CPUE of age-classes, of rainbow trout in streams selected the variables velocity, pool variability, stream width, percentage of cover of large woody debris, and three variables associated with substrate cover: percentage of boulder, percentage of cobble and percentage of sand and silt. These seven explanatory variables explained approximately 28% of the variation in age-class

CPUE constrained on the first two redundancy axes (RDA 1 = 17% and RDA 2 = 11% -

Figure 3.4). The RDA associated Age-0 fish to high percentage of large woody debris and cobble. This corresponds to the HSI variables associated with fry habitat (CF): a substrate dominated by gravel/cobble and low percentage of fines. Age-1 and Age-2 fish were associated to a substrate more composed of boulders, larger stream width, higher stream velocity and higher pool variability (particularly for Age-2). This follows the HSI habitat variables identified as important for juvenile rainbow trout (CJ): availability of instream cover and pool classes’ variability. High percentage of sand and silt were negatively correlated to all three age-classes, which follows the HSI criterion for rainbow trout. The RDA age-classes/habitat associations support the use of the HSI methodology and the associated measures of stream habitat quality and quantity (WUA) to describe the availability of early life stage habitat for rainbow trout populations.

Stream capacity.

Stream juvenile abundance, described using total electrofishing catch, was best predicted by the stream WUA (Table 3.1), meaning that the stream habitat quality and quantity rather than egg input from the adult population dictates stream abundance. A model also including the number of eggs was within two AIC points, but the number of eggs coefficient in that model was not significant (p-value = 0.0848), most of the variation was explained by WUA. The relationship between stream abundance and WUA can be described by:

(3.7) Stream juvenile abundance = 10.100 + 0.067 WUA

Lake capacity.

The habitat ratio (stream WUA over lake area - WUAratio) was the best predictor of both gillnet catch of Age 2 fish in the lake (proxy for the number of recruits) and total gillnet catch (proxy for total lake abundance). For both recruits and total lake abundance the best model, or a model within two AIC points, also included the predictors GDD and presence/absence of other fish species (Table 3.2). The most parsimonious model for recruits is described as:

(3.8) Recruits = −83.436 + 0.229 WUAratio + 0.129 GDD − 9.997 OtherSP

Climate (GDD) presents a positive relationship with recruits, in warmer conditions more fish recruit to the lake, while the presence of other fish species negatively impacts recruitment.

For total lake abundance, the lowest AIC model is:

(3.9) Lake abundance = 96.459 + 0.360 WUAratio

Following the stream capacity results, WUA is a good predictor of the juvenile input from the stream. Dividing WUA by lake area, making it a habitat ratio, divides the stream input by the lake habitat availability, the habitat ratio increases as more stream habitat is available for less lake habitat. The habitat ratio had a positive relationship with the number of recruits and a weaker positive relationship with the total lake abundance. A model composed only of lake area was within two AIC units of that model, suggesting that, lake area is an important driver of lake abundance.

Growth and maturation.

Gillnet catch in the lakes varied 18-fold across systems (range: 19-341 fish/overnight gillnet set – Figure 3.5a). Fish captured in the net ranged from Age-1 to

Age-7 and in size from 113 to 565 mm. Fishing mortality was not included in the analysis because it is so low in the sampled systems that total mortality was assumed to be almost exclusively composed of natural mortality. Of the 39 lakes sampled four of the lakes had no fishing effort, 29 lakes less than 1 angler-day per hectare (AD/ha), and the maximum effort estimated was 13 AD/ha corresponding to an instantaneous fishing mortality rate

(F) of 0.43 yr-1 (see associated mortality rate in Figure 3.5f). This highest fishing effort estimated is on the low end of what has been observed in other studies on rainbow trout in the same region, Cox (2000) reports 27-114 AD/ha while van Poorten et al. (2015) report ~4-196 AD/ha (transformed to AD from 17-785 AH/ha).

Most of the growth and maturity parameter estimates passed diagnostic and

MCMC convergence checks, except for the age at 50% maturity for males. Male age at

50% maturity did not present a good fit, the distribution was bimodal with some fish maturing around age 1 and others maturing around age 2-3. This precocious male maturation is common in salmonids (Fleming 1998) and has been observed in rainbow trout (Liley et al. 2002). It would have been interesting to separate the samples between those two types of male life histories and include their maturation schedule in this analysis, particularly since the incidence of precocious male maturation has been linked to habitat characteristics (DeFilippo et al. 2018) and growth patterns (Hutchings and

Jones 1998, Utrilla and Lobon-Cervia 1999), but sample size was insufficient to do so.

Therefore, only maturation of females was used in the analysis. All other global and

population specific traits passed the Gelman-Rubin diagnostic with PSRFs <1.1 which indicates convergence to a common posterior distribution and effective sample size was

>1,000 for all traits, with the exception of lake Hardy Lower for the lake specific growth parameters (푘 and 퐿∞). Most size at age samples were contained within the 95% credible intervals of the posterior predictive distribution suggesting no systemic bias associated with the growth model (Appendix E: Figure E.1).

Growth and maturation parameters varied across systems (Figure 3.5). Growth parameters 퐿∞ and k varied 3- and 12-fold respectively (Figure 3.5c-d) while the maximum observed size varied 2-fold (Figure 3.5e). For female maturation, age and length at 50% maturity (A50f) varied 2 and 1.5-fold respectively (Figure 3.5g-h). 퐿∞, k and A50f are all positively correlated to one another (Appendix F: Table F.1). Theses correlations suggest that in systems were fish reach larger sizes (퐿∞), they get to the asymptotic size faster (k), but females of those populations tend to reach maturity at a later age (A50f).

The habitat ratio (WUAratio) was the top predictor of asymptotic growth while the total gillnet catch (proxy of lake abundance) was better at predicting female length at

50% maturity. Both predictors support the importance of density-dependent processes in fish growth and maturation. The two measures of maximum growth, estimated asymptotic size (퐿∞) and maximum observed size (MaxObsSize), were both best described by the same model composed of negative relationships with habitat ratio, GDD and presence of other fish species (16% and 43% of variance explained respectively for each measure of maximum growth, Table 3.3 and Figure 3.6b). As more early life stage habitat is available for less adult habitat (higher habitat ratio), maximum fish size tends to

decrease while warmer climate (higher GDD) and presence of other fish species further decrease maximum fish size. The growth parameter k and age at 50% maturity did not present significant relationships with any of the predictors. Female length at 50% maturity presented a negative relationship with total gillnet catch and a positive but non- significant relationship with the presence of other fish species (30% of the L50f variance explained, Table 3.3 and Figure 3.6a) suggesting that as lake density increases, females reach maturity at a smaller size, but not necessarily at a different age.

Predictions on abundance and size structure of rainbow trout populations were made using variables relating to stream and lake physical habitat (habitat ratio), GDD and presence/absence of other fish species. Only the prediction of maturation schedule was improved by adding actual fish data (gillnet catch). The predicted recruitment and maximum fish size follow the size-number tradeoff presented in Chapter 2. Populations with high recruitment/abundance have smaller fish while low recruitment/abundance leads to larger maximum fish size and no high density large maximum size populations were predicted or observed.

Discussion

These field results suggest early life stage habitat capacity limitation is linked to lower recruitment and larger adult body size. In contrast, abundant juvenile habitat

(streams) in relation to the availability of adult habitat (lakes) leads to high overall population density causing density-dependent reductions in adult growth, resulting in stunted populations. This is empirical evidence of the complex dynamics linking habitat limitations at one life stage to population bottlenecks impacting overall population

density and growth patterns as presented in Chapter 2. This study’s results provide an example of the importance of population bottlenecks, but also demonstrates they can occur at different life stages and that their consequences can be predicted using metrics of habitats used throughout ontogeny. Weighted useable area (WUA), a measure of both habitat quality and quantity, was identified as the best descriptor of stream habitat availability and used to predict stream juvenile input into the adult lacustrine habitat.

Stream WUA in relation to lake area was used to create a habitat ratio variable that predicts fish growth patterns across systems that vary in characteristics of the two key habitats. This link between habitat and growth shows that rainbow trout population size structure and abundance can be predicted using physical stream and lake habitat characteristics, climatic information (GDD) and species composition.

The stream sampling results of juvenile fish and their habitat support the habitat suitability curves presented in the HSI methodology (Raleigh et al. 1984) and its life stage to habitat associations. These habitat associations also support assumptions made in

Chapter 2 regarding the linkages between age-class specific habitat availability and migration behaviour. In most reaches, if no age-class specific habitat was present, that age-class was not captured and therefore was assumed to have migrated in the lake. The strong relationship between WUA and stream juvenile abundance suggests that juvenile fish numbers are dictated by habitat availability and density-dependent mortality rather than by egg input. Systems with much higher fishing pressure than was observed here may result in limited recruitment due to egg limitation (Myers and Barrowman 1996).

One caveat of the HSI methodology is that it does not directly describe the productivity of the system, information on nutrient levels, primary production and prey density would

likely better describe the processes structuring the stream ecosystem and improve predictions.

In the lakes, recruitment is driven by the availability of early life stage habitat in relation to the lake habitat availability (habitat ratio). The positive relationship between stream habitat availability (habitat ratio) and recruitment supports the hypothesis that limitation of early life stage habitat leads to high early mortality, when habitat is abundant, more fish survive to migrate into the lake and reach recruitment age. This result follows what is commonly described in fish ecology as one of the main sources of population regulation, density-dependent survival at the pre-recruit stage (Elliott 1994,

Milner et al. 2003). Total gillnet catch was used as a proxy for lake abundance, it also presented a positive, but weaker, relationship with habitat ratio and lake area. This study established the link between stream habitat availability, stream capacity and recruitment, but the lake dynamic appears more complicated. In both Chapter 2 and this empirical analysis, lake is described in a simplistic manner, reducing it strictly to its area, temperature (GDD) and fish species assemblage. It is likely that other lake characteristics such as basin morphology, nutrient availability and trophic state impact the fish populations, as has been demonstrated in other populations such as lake trout (Salvelinus namaycush) in Ontario (Shuter et al. 1998). As such, a more detailed description of lake characteristics could improve this analysis. Overall, the lakes in this study present low concentrations of total dissolved solids (TDS < 250 mg/L) which would categorize them as low productivity. The three lakes presenting the highest fish densities (Hardy Lower,

Wollaston and Curtis – Figures 3.6a) are clustered in the Okanagan region, at high elevation and present low TDS values. All three have high habitat ratios which, according

to the relationships described, would lead to high fish density as is observed, but other environmental factors could be at play here, further increasing fish recruitment and abundance.

While this study has likely not captured all the lake variables impacting fish populations, its prediction of the importance of habitat availability and bottlenecks on structuring populations is supported by the observed relationships between habitat predictors and growth parameters. The size-number tradeoff observed follows the results of simulations made using a theoretical model of rainbow trout populations (Chapter 2).

Maximum size (퐿∞ and MaxObsSize) was related to habitat availability variations with high habitat ratio, associated with high densities, leading to smaller fish. The maximum observed FL in the catch (MaxObsSize) was better explained by habitat variation than asymptotic size (퐿∞). This difference between the two measures of fish length is likely due to the low numbers of older age-classes and the shape of the size at age relationship in some of the lakes sampled where it does not look like an asymptotic length was reached (Appendix E: Figure E.1). Therefore, for some lakes, the fitted 퐿∞ might present a biased growth trajectory. Strong density-dependent growth has been observed in rainbow trout (Post et al. 1999) and in other fish species (Lorenzen and Enberg 2002).

Although many explanations exist for the mechanisms responsible for stunting in fish populations, the most common process presented relates to resources limitation leading to lower food consumption rates (Ylikarjula et al. 1999). This work supports the Chapter 2 model prediction that limitation of habitat at the adult stage, when juvenile habitat is abundant in relation to adult habitat, leads to regulation through density-dependent growth rather than mortality. Andersen et al. (2017) suggested that in small habitats, like

lakes, and particularly in the absence of predators, density-dependence will occur late in life and often lead to stunted growth. While likely often the case, even with limited additional mortality (from predator or harvest), density-dependence can occur at the early life stages in small lakes if the juvenile stream rearing to adult lacustrine habitat ratio is not balanced.

Female age at 50% maturity varied between ages two and four with most females reaching maturity at age 3, which is consistent with data from other rainbow trout populations from the same region (Cox 2000), but this variation was not explained by any of the descriptors assessed. This lack of relationship between age at maturity and density predictors is contrary to what would have been expected based on life history theory, which predicts that when juvenile survival rate is high, stunting will occur, and fish will mature at a later age (Ylikarjula et al. 1999). The female length at 50% maturity presented a negative relationship with density, which would support the hypotheses of density-dependent growth and age based, rather than size based, maturation for rainbow trout. If age at maturity is fixed, fish in stunted slower growing population will reach maturity at a smaller size. The correlation between 퐿∞ and 퐴50푓 has been observed previously in rainbow trout (Cox 2000) and is supported by life history theory which predicts that size at maturity is associated with lifetime growth patterns where iteroparous fish maximize lifetime fecundity (Shuter et al. 1998).

Growing-degree-days (GDD) has been linked to enhanced individual growth in rainbow trout populations (Ward et al. 2017, Varkey et al. 2018), which is contrary to what is observed here. In this study, higher GDD led to smaller estimated asymptotic

(퐿∞) and observed maximum size. However, the range of GDD in the sampled lakes is

small, 840-1177 GDD for this study vs 1062-1964 GDD for Ward et al. (2017), which might have limited my results. Another explanation for these discrepancies is that the estimated asymptotic size is the result of multiple conditions including climate but also density. The theoretical model (Chapter 2) does predict a lower asymptotic size at higher

GDD in high density populations due to the complex dynamic between early growth and survival. Abundant early life stage habitat and high GDD would lead to higher early growth and survival leading to greater recruitment and higher density in the lake which would, ultimately limit asymptotic size through density-dependent growth at the adult stage. In the model, there is a density threshold (Chapter 2: Figure 2.5b) under which density is so low that GDD does lead to a direct increase in growth resulting in a larger asymptotic size as was observed by Ward et al. (2017). My study’s results support that hypothesis as there was a positive relationship between GDD and recruitment. Both Ward et al. (2017) and Varkey et al. (2018) used stocked rainbow trout populations were fish density is controlled externally, the different results I obtained might be a consequence of using populations that are naturally reproducing and where a complex dynamic of climate and density-dependent processes are at play. Further observational and experimental studies across broader ranges of both density and GDD would be required to disentangle these complex interactions at the population level.

The presence of other fish species reduced growth rates and recruitment. This would point toward interspecific competitive interactions for lake resources that limit growth. The decrease in recruitment could be the result of predation by adults of other fish species on rainbow trout juveniles. For example, northern pikeminnow have been known to prey on juvenile salmonids (Zimmerman 1999) and are present in two of the

lakes sampled. However, the low number of multispecies sites (6) and the lack of population density information for other species limits the inferences that can be drawn from this relationship.

One caveat of this study is the uncertainty associated with the measures of density and recruitment. The use of a snapshot approach, in which number of study sites was prioritized over number of sampling events, increased the uncertainty of individual estimates, but increased the ability to capture variation in density and population dynamics among systems. For these reasons, the recruitment values are treated as relative measures for comparison among systems. A second caveat relates to the limited range of both fishing effort and GDD in the sample. Higher harvest pressure would be expected to relieve stunted populations from density-dependent growth, but I could not test that hypothesis directly. When planning this study design, I tried to sample sites with a wide range of fishing pressure but found that most lakes popular with anglers were also part of stocking programs. Similarly, trying to sample warmer lakes (higher GDD) proved to be difficult as most warmer lakes are lower in elevation and tend to be more easily accessible, which makes them more popular with anglers and has, over the years, lead to stocking programs. The last issue encountered is also linked to site selection and field sampling. Only a limited number of the lakes selected were not rainbow trout monoculture (6 out of 39). I did attempt to sample more lakes with fish species assemblages but realized that the sampling method selected was not always appropriate as gillnet saturation by the other fish species and low rainbow trout catch were observed.

For this reason, and to minimize the sampling bias associated with net saturation by other

species, only lakes for which at least 30 rainbow trout were captured were used in the analysis.

In conclusion, this study provides empirical evidence of the importance of habitat availability for population dynamics and describes how the interplay between the habitats used throughout ontogeny impacts fish growth and demography. In addition, the close alignment between these empirical patterns, which were observed among natural populations, and the parallel modelling analysis presented in Chapter 2, adds credence to the inferences drawn among habitat characteristics and bottlenecks in growth and population dynamics. This is adding to a growing body of literature that supports the idea that density-dependence can occur at different life stages in freshwater fish and that population regulation is not limited to early life stages as it is often assumed in marine fish population models (Walters and Martell 2004, Persson et al. 2014, Andersen et al.

2017). Understanding the mechanisms responsible for population regulation is important in the context of management, it provides an evaluation of a population’s capacity to compensate for increased mortality associated with harvest or habitat perturbations and can help managers prioritize monitoring and conservation effort.

Table 3.1 An assessment of alternate models to predict stream habitat capacity based on estimated number of eggs (Neggs), tributary maximum width (Width), Habitat Suitability Index (HSI), weighted useable area (WUA), growing-degree-days (GDD) and presence/absence of other fish species (OtherSp). Values of R2, adjusted R2 2 (adjR ), p-value, AIC and ΔAIC are presented, the most parsimonious model (lowest AIC) is in bold and models within two AIC points are in italic.

2 2 Model R adjR p-value AIC ΔAIC 1 Neggs 0.01 -0.02 5.8E-01 461.43 19.30 2 Width 0.3 0.28 3.5E-04 448.09 5.96 3 HSI 0.11 0.09 3.6E-02 457.07 14.94 4 WUA 0.4 0.38 1.8E-05 442.13 0.00 5 Width + Neggs 0.3 0.26 1.7E-03 449.99 7.86 6 HSI + Neggs 0.13 0.08 8.2E-02 458.33 16.20 7 WUA + Neggs 0.4 0.36 1.1E-04 444.09 1.96 8 Width + GDD + OtherSp 0.33 0.28 2.5E-03 449.98 7.85 9 HSI + GDD + OtherSp 0.16 0.09 1.1E-01 459.06 16.93 10 WUA + GDD + OtherSp 0.41 0.36 2.8E-04 444.9 2.77 11 Width + Neggs + GDD + OtherSp 0.33 0.25 6.8E-03 451.95 9.82 12 HSI + Neggs + GDD + OtherSp 0.17 0.07 1.7E-01 460.6 18.47 13 WUA + Neggs + GDD + OtherSp 0.41 0.35 9.1E-04 446.9 4.77

Table 3.2 An assessment of alternate models to predict lake habitat capacity described using number of Age-2 fish in gillnet catch as a proxy for number of recruits (models 1-10) and total number of fish in gillnet catch as a proxy for lake abundance (models 11-20), based on number of juvenile fish caught in the stream (Nstream), stream weighted useable area (WUA), lake area (Lake.Area), WUA to lake area ratio (WUAratio), growing-degree-days (GDD), and presence/absence of other fish species 2 2 2 (OtherSp). Values of R , adjusted R (adjR ), p-value, AIC and ΔAIC are presented, the most parsimonious model (lowest AIC) is in bold and models within two AIC points are in italic.

2 2 Model R adjR p-value AIC ΔAIC 1 Nstream 0 -0.03 9.60E-01 395.74 10.17 2 WUA 0.01 -0.02 5.90E-01 395.43 9.86 3 Lake.Area 0.1 0.08 4.50E-02 391.43 5.86 4 WUAratio 0.17 0.15 8.80E-03 388.41 2.84 5 GDD 0.09 0.07 5.90E-02 391.94 6.37 6 OtherSp 0 -0.02 7.50E-01 395.63 10.06 7 Nstream + GDD + OtherSp 0.11 0.04 2.30E-01 395.06 9.49 8 WUA + GDD + OtherSp 0.11 0.03 2.50E-01 395.18 9.61 9 Lake.Area + GDD + OtherSp 0.18 0.11 6.70E-02 391.89 6.32 10 WUAratio + GDD + OtherSp 0.3 0.24 4.90E-03 385.57 0 11 Nstream 0.01 -0.02 6.20E-01 457.59 3.82 12 WUA 0.01 -0.01 4.90E-01 457.33 3.56 13 Lake.Area 0.06 0.04 1.20E-01 455.29 1.52 14 WUAratio 0.1 0.08 5.10E-02 453.77 0 15 GDD 0.03 0 3.00E-01 456.69 2.92 16 OtherSp 0.02 -0.01 4.10E-01 457.11 3.34 17 Nstream + GDD + OtherSp 0.06 -0.02 5.10E-01 459.32 5.55 18 WUA + GDD + OtherSp 0.06 -0.02 5.00E-01 459.24 5.47 19 Lake.Area + GDD + OtherSp 0.11 0.03 2.60E-01 457.4 3.63 20 WUAratio + GDD + OtherSp 0.17 0.1 9.00E-02 454.71 0.94

Table 3.3 Most parsimonious models to predict age at 50% maturity of females (A50f), length at 50% maturity of females (L50f), the Brody growth coefficient (log-transformed – ln(k)), asymptotic length (log-transformed – ln(푳∞)) and maximum fish size observed (log-transformed – ln(푴풂풙푶풃풔푭푳)), based on lake gillnet catch (Nlake), presence/absence of other fish species 2 2 2 (OtherSp), WUA to lake area ratio (WUAratio), growing-degree-days (GDD). Values of R , adjusted R (adjR ) and p-value are presented, and a complete list of the models compared for each variable is presented in Appendix G: Table G.1.

Model 2 2 R adjR p-value

퐴50푓 ~ 3.04 - 0.0008 Nlake 0.02 -3.1E-03 0.35

퐿50푓 ~ 296.47 - 0.19 Nlake + 17.89 OtherSp 0.34 0.30 5.9E-04

ln(푘) ~ -1.69 - 0.0003 Nlake + 0.46 OtherSp 0.08 0.03 0.23

ln(퐿∞) ~ 7.08 - 0.001 WUAratio - 0.0008 GDD - 0.24 OtherSp 0.22 0.16 0.03

ln(MaxObsFL) ~ 6.48 - 0.001 WUAratio - 0.0007 GDD - 0.1 OtherSp 0.47 0.43 4.8E-05

Figure 3.2 Location of the sampling lakes (yellow points) and nearby population centres (purple) over satellite imagery in the province of British Columbia (BC).

Figure 3.3 Histograms of the frequency of (a) GDD values, (b) presence/absence of other fish species, (c-e) stream habitat descriptors, and (f-i) electrofishing catch-per- unit-effort (CPUE in N/sec) for the total sample and age-classes 0, 1 and 2. The number of sampling sites is 39 for all measures.

Figure 3.4 Correlation biplot presenting the results of the RDA of CPUE of age- classes of rainbow trout (0, 1, 2 – purple lines) as a function of habitat variables (turquoise arrows) pool variability, mean stream width, velocity, percent of large woody debris (LWD), and percent of substrate that is cobble, boulder and sand/silt. The yellow dots are the different reaches sampled, their projection on the lines and arrows estimates their value. The angle between the different age-classes (purple lines) and habitat variables (turquoise arrows) represent the correlation between them, the closer the lines/arrows are the more correlated they are while lines/arrows pointing in opposite directions suggest that they are negatively correlated.

Figure 3.5 Histograms of the frequency of (a) gillnet catch, (b) recruits (number of Age-2 fish in the gillnet catch), (c) growth parameters k and (d) 푳∞, (e) maximum observed size calculated as the mean FL of the five biggest fish caught in the gillnet, (f) fishing mortality rate estimated from the effort cameras, (g-h) female age and size at 50% maturity. The number of sampling sites is 39 for all measures.

Figure 3.6 Graphical representation of the most parsimonious model for life-history traits related to (a) maturation and (b) growth. The traits are: female length at 50% maturity (in mm - a) and maximum size described as the estimated asymptotic size (푳∞ in mm – b top) and maximum observed size calculated as the mean FL of the five biggest fish caught in the gillnet (in mm – b bottom). On the left side for each trait, plot of the linear relationship (solid black line) between the trait and the most significant predictor, total gillnet catch in a) and WUA/Lake area ratio in b). The dashed lines represent the 95%CI and the black dot the mean of the 39 populations sampled. On the right, the influence on each population parameter of predictor GDD (only part of the models in b) and presence/absence of other fish species (OtherSp) if all other variables remain constant is presented using the mean of all populations (black dot) and the influence on the parameter of a positive (+) and negative (-) change of one standard deviation of the variable of interest for GDD. For the presence/absence of other species variable the value resulting from the absence (0) and presence (1) of other species is presented, which is more than a standard deviation but is more realistic in the context of a binary variable. The asterisks at the top of each plot represents the level of significance for each predictor (blank > 0.1, ‘•’ ≤ 0.1, ‘*’ ≤ 0.05, ‘**’ ≤ 0.01, ‘***’ ≤ 0.001).

Chapter 4: Impact of Experimental Density Manipulation on Population

Abundance and Growth of Rainbow Trout Populations with Contrasting Early Life

Stage Habitat Conditions

Introduction

When faced with additional sources of mortality, whether from predation or harvest, animal populations regulate themselves through processes that act as a negative feedback on population abundance, maintaining carrying capacity. Such mechanisms are associated with changes in survival, fecundity, growth and movement and are described as compensatory if they lead to numerical increases at low density (Rose et al. 2001). The main driver of those density-dependent processes is usually the reduced competition for resources associated with decreased density. Examples of direct compensation include changes in survival rate and fecundity that produce a direct numerical response, a change in population abundance. Density-dependent growth and changes in movement or behaviour are described as indirect compensation, processes that are only considered compensatory if they lead to a survival or reproductive success change (Rose and Cowan

2000). For examples, at low population densities when resources are abundant, juvenile fish would change their behaviour from foraging in high predation risk habitat to stay in refugia and limit exposure to predators, resulting in an increase in survival (Walters and

Juanes 1993).

Density-dependent growth has been recognized as an important factor regulating fish populations (Lorenzen and Enberg 2002). The indirect impact of growth on compensation can be associated with increased survival due to larger size, particularly for

Age-0 fish, as larger size at age has been linked to predator avoidance and lower risk of starvation (Post and Evans 1989a, Sogard 1997, Byström and Garcia-Berthou 1999).

Higher growth rates can increase fecundity through earlier maturation (Ylikarjula et al.

1999, Roff 2002), but also because larger females tend to produce more or better-quality eggs (Koops et al. 2004, Rollinson and Hutchings 2010).

The compensatory reserve describes a system’s capacity to offset additional mortality, this concept is fundamental to sustainable fish and wildlife management where it is described as the surplus that can be harvested (Bartmann et al. 1992, Rose et al.

2001). Compensation implies that, within some limited densities, the population can increase its abundance at a greater rate than replacement (Goodyear 1993). Examples of compensatory response have been presented for many fish species (Myers et al. 1999,

Rose et al. 2001), but the strength and type of compensation varies across species and populations due to local habitat and food web structure (Rose and Cowan 2000, Rose

2005).

In the previous chapters I presented the importance of early life stage habitat availability and its influence in determining population bottlenecks using a theoretical age-and-size-structured population model (Chapter 2) and empirical data from the rainbow trout fishery of British Columbia (Chapter 3). Based on the simulations presented in Chapter 2, I showed that a constant harvest pressure can have different impacts depending on the population’s different habitat availabilities (Chapter 2, Figure

2.4). The Chapter 2 model would predict that high density stunted growth populations, resulting from high early life stage habitat availability, would have a greater compensatory reserve than a lower density population inhabiting a system with limited

spawning and rearing habitat (Figure 4.1). The low habitat ratio/low density population would be more vulnerable to an increased harvest of adults as it would have the potential to reduce the number of eggs deposited in the stream to a level where eggs would no longer saturate the habitat, leading to a decrease in the number of juveniles present. This decrease could result in a compensatory response in growth and survival (Figure 4.1, left side). While in the high habitat ratio population, the same increase in adult harvest would not lead to reduction in the abundance of juveniles if there are still enough spawners present to saturate the stream habitat with eggs. This would produce the same number of juveniles as in the pre-harvest conditions (Figure 4.1, right side). In this high habitat ratio scenario, the increased harvest could even lead to an increase in the number of juvenile fish reaching maturity (defined here as recruitment) due to the decreased competition with older age classes in the lake which are subjected to harvest. These predictions could not be tested empirically in Chapter 3 because none of the 39 populations studied presented strong enough recreational fishing pressure to drive a compensatory response of rainbow trout populations to harvest. In this study (Chapter 4), I present the results of an experimental depletion of two rainbow trout (Oncorhynchus mykiss) populations with contrasting stream to lake habitat ratios.

Whole-system manipulations offer the advantage of incorporating the effects of the natural variation of physical, chemical and biological processes that can impact the outcome of experiments (Schindler 1998). An experimental approach also provides the ability to maximize contrast and obtain extreme conditions that are not easily observable in nature. In fish populations, manipulations for the purpose of observing density- dependent processes can be done through fish stocking (Post et al. 1999) or removal

(Healey 1980, De Gisi 1994, Catalano and Allen 2011), but can also be done indirectly by modifying fishing regulations. For example the transition to a catch-and-release only for a bull trout (Salvelinus confluentus) fishery in Alberta led to an increase in fish density that resulted in decreased growth rates and increased age at maturity (Johnston et al. 2009). In rainbow trout populations, experimental manipulations of population density and size structure has provided important information on the impact of density-dependent processes on intercohort interactions (Post et al. 1999) and on habitat use and foraging behaviour (Landry et al. 1999, Biro et al. 2003a, 2003b). Such manipulations help understand the biological processes underlying compensation and how populations respond to exploitation.

The main objective of this chapter is to assess the compensatory response of rainbow trout populations to experimental harvest. I will do this using mark-recapture data to compare population abundance, age structure and individual growth rates between a control and two experimental lakes before and after a density manipulation through harvest. The influence of population density and climatic variation on growth rate will also be assessed to describe the mechanisms underlying the compensatory response.

Methods

Study sites.

The study was conducted in three lakes (Pantano, Stubby and Today) located in the Bonaparte Plateau within the North Thompson River watershed in British Columbia

(Figure 4.2). Stubby flows into Today which drains into a series of second order streams and lakes that flow into Skull Creek, while Pantano is located ~2 km away and drains

into a second order stream which also flows into Skull Creek. Skull Creek flows east into

Fishtrap Creek, a tributary of the North Thompson River. All three lakes are only accessible through hiking. The remoteness of the lakes and the abundance of good quality fishing opportunities with easier access nearby results in minimal fishing pressure.

During the five years of this study the field team only witnessed evidence of one visit from anglers to Lake Today and none on the other lakes. Similarly, in a study conducted in the early 2000’s, Askey et al. (2006) only witnessed one visit by anglers on Lake

Today over a year of study of all three lakes. Therefore, I considered fishing mortality on those lakes negligible. All three lakes are composed of rainbow trout monoculture that have been previously stocked and manipulated in other studies, but none of the lakes have been supplemented by stocking since at least 2005 (Askey et al. 2006, 2007). The lakes’ populations are maintaining themselves through natural reproduction and the genetic origin and influence of past stocking is unknown.

The three lakes have a similar small size (< 7 ha), climate regime and total dissolved solids concentration, a measure often used to predict lake productivity (Table

4.1 and Appendix B: Table B.1 for detailed locations). They differ in the number and size of their tributaries. Lake Pantano has one ephemeral inlet that provides minimal fish habitat, its second inlet and its outlet are both second order streams that flow year-round and are used by juvenile rainbow trout for rearing. Lake Stubby has one inlet and one outlet to Lake Today, both run dry during the summer months and offer limited spawning habitat in the spring. Although Lakes Stubby and Today are connected through their shared stream there is no evidence of fish movement between the two lakes, to my knowledge no fish tagged in one lake have been recaptured in the other, therefore I

assumed that the two populations are independent. Lake Today has the most tributaries of the three lakes with three inlets and one outlet, most inlets have limited rearing availability due to low water level in the summer, but the outlet flows year-round, and juvenile fish have been observed rearing in it. Temperature variation during the five study years was followed using growing-degree-days above 5°C metric estimated from historical data in ClimateWNA (Wang et al. 2012, 2016).

Experimental design.

The three fish populations were monitored throughout the study using a mark- recapture framework composed of three summer fyke net assessments and one fall gillnet sampling occasion each year (see timeline in Figure 4.3). Two fyke nets size were used, the small ones had a hoop sizes of 0.7 m with 0.8 cm mesh and the large ones had 1 m hoops with 2 cm mesh, nets of different sizes were moved between lakes to ensure that the same sampling effort was applied during each fyke net capture occasion. Fyke net capture occasions were composed of 4 days during which the nets were set overnight each day. The three fyke net capture occasions occurred during the months of July and

August of each sampling year, and each 4-day occasion was at least 14 days after the previous sampling to leave time for the fish to mix in the lake. Each fish captured in the fyke net was measured in fork length (FL – mm) and weighted (g), fish over 100 mm were marked by clipping their adipose fin and tagged using a passive integrated transponder (PIT). This double marking technique is used to calculate PIT tag loss, when fish captured have their adipose fin clipped but do not have a PIT tag. In the field, fish that had lost their PIT tag were implanted with a new tag. At each capture occasion fish

were checked for marks and deformities. Some recaptures occurred during the summer fyke netting, but most recaptures happened in the fall gillnet sampling. The gillnet sampling used one floating and one sinking net set overnight following the BC gillnetting standards, methodology described in Ward et al. (2012). Fish caught in the gillnet were measured (FL – mm), weighted, their sex and maturity were recorded, they were checked for marks and PIT tags and their lapilli otoliths were collected and aged using transmitted light.

In the spring of 2014 and 2015, depletion events were conducted on Lakes Stubby and Today, while Pantano was kept as a control (detailed timeline in Figure 4.3). The depletion events consisted of intensive sampling where five gillnets (one floating, four sinking) were set for five consecutive nights. The gillnets were the standard nets described above, but with the two smaller mesh (25 and 32 mm) removed to target the spawning adults of the population. For both years, the depletion events occurred right after ice-off. All fish captured were sampled following the protocol described above for fall gillnet sampling.

Data Analysis.

To assess the effect of the depletion on population abundance, total population and age specific abundance estimates were compared between years. For each year sampled (2012-2016), the capture history of PIT tagged fish across the three fyke net sampling occasions and the fall gillnet sampling (4 capture occasions) was used to calculate a Schnabel population abundance estimate for the whole population and for each age-class. The Schnabel method is a multiple recaptures closed-population model

used for samples collected over a short period of time, as it assumes that the population is constant, experiencing no immigration, migration, recruitment and mortality over the sampling period (Hayes et al. 2007):

푡 ̂ ∑푖=2 푛푖푀푖 (4.1) 푁 = 푡 ∑푖=2 푚푖+1 where 푁̂ is the estimate, so in this case either the total population abundance or the abundance of each age-class, t is the number of sampling occasions, ni is the number of fish caught in the ith sample, mi is the number of fish with marks caught in the ith sample and Mi is the number of marked fish present in the population at the ith sample (marked in previous captures).

The estimated abundance of each age-class was based on the ageing of the gillnet caught fish. For fish that were only captured in the fyke nets, age was estimated using a lake and year specific age-length key. The age-length key is a probability matrix of the proportion of fish of each age in 25 mm length bin in the gillnet, it is used to categorize each fyke net caught fish following that probability (Isermann and Knight 2005). To account for the growth that occurs between the fyke net sampling (July-August) and the gillnet sampling (October), the length used for age attribution to fyke net caught fish was adjusted (퐿푎푑푗) for each lake-year based on an estimated daily relative growth rate (퐺퐿푌 ).

The daily growth rate is estimated based on the relationship between initial length (퐿1) and relative growth rate calculated using j number of PIT tagged fish that were captured at least twice within the sampling year.

(4.2) 퐿푎푑푗 = 퐿푓푦푘푒 + 퐿푓푦푘푒((푎퐿푌퐿푓푦푘푒 + 푏퐿푌)(푡푔푖푙푙 − 푡푓푦푘푒))

(4.3) 퐺퐿푌 = 푎퐿푌퐿1 + 푏퐿푌

퐿푗.2− 퐿푗,1 (4.4) 퐺푗,퐿푌 = 퐿푗,1(푡푗,2−푡푗,1) where 퐿푓푦푘푒 is the length at fyke capture, 푎퐿푌 and 푏퐿푌 are the fitted parameters of the relationship between daily growth rate and fish length at first capture, 푡푔푖푙푙 is the Julian day at which the gillnet sampling occurred, 푡푓푦푘푒 is the Julian day at which the fish was caught in the fyke net, 퐺푗,퐿푌 is the relative growth rate for PIT tagged fish j, 퐿푗.2 is the size at the last recapture of fish j, 퐿푗,1 is the size at first capture of fish j, 푡푗,2 is the Julian day at last recapture of fish j and 푡푗,1 is the Julian day at first capture of fish j. The adjusted age structure of each population was graphically compared across years. Due to the lack of independence between sampling years (same lakes) and absence of replicates, other statistical methods were not appropriate for this analysis.

Annual growth rates were calculated using tagged fish captured across years, their growth is expressed on a daily time step to account for the timing of their captures. The calculations follow equation 4.4 above. Due to the low number of fish recaptured across more than one year, the years were pooled together into pre and post-depletion categories for Lakes Stubby and Today. Years of the control lake, Pantano, were placed in the pre- depletion group, since it was not manipulated. Regression models were built to assess the impact of the inter-lake variation and of the depletion manipulation (pre/post) on the growth rate. My first hypothesis was that there would be a difference in growth rate between the different populations before the experiment, and that the lake with the least early life stage habitat, Stubby, would present lower fish population density and higher growth rates (across lake variation pre-depletion). I also hypothesized that the manipulation would reduce population density resulting in higher growth rates post-

depletion in the experimental lakes (Stubby and Today pre vs post-depletion). I also included the fish length at first capture variable in the models as individual fish growth rate decreases with age/size. The linear models I compared include the variables length at first capture, pre/post-depletion status and lake name (Pantano, Stubby and Today) and the interaction between the variables, sample size is 71 fish and the models are detailed in

Table 4.2.

To further investigate the growth rate variation and link it directly to the density manipulation, models describing the impact of estimated population density (abundance calculated above divided by lake area) on growth rate were also assessed. As in the previous analysis, models included the variable fish length at first capture. The growing- degree-days above 5°C (GDD) variable was also included because GDD has been associated with variation in fish growth in rainbow trout in British Columbia (Ward et al.

2017, Varkey et al. 2018) and other fish species (Neuheimer and Taggart 2007, Venturelli et al. 2010).

Results

Depletion.

The spring depletion manipulation in Stubby removed a total of 301 fish in 2014 and 245 fish in 2015. Many fish were marked recaptures, and in both years, they represented 39% of the fish marked that year that had not been previously harvested in the fall gillnet. In Today, 413 and 330 fish were harvested in the spring depletions of

2014 and 2015, respectively. In 2014, the marked recaptures were 27% of the total marks, and in 2015 that number went up to 41%. In both lakes and both years, no Age-1 fish

were harvested during the depletion (Figure 4.4). In Stubby, mostly Age-3 fish were removed from the population in 2013, while in 2014 most fish were Age-3 and -4. In

Today, Age-4 fish were the most represented age-class in the 2014 depletion, while in

2015 mostly Age-3 and -4 fish were removed.

Temperature.

Climatic conditions varied among the study years for all three lakes, with GDD values ranging from 812 to 1033 GDD (Figure 4.5). The GDD values obtained from the

ClimateWNA model for the lakes are similar due to the proximity of the three lakes, the climate model uses historical data from weather stations to estimate temperature by interpolating between stations and accounting for the effect of elevation (Wang et al.

2016). The coldest year was 2012 where GDD values for all three lakes are under 850

GDD. The following years were warmer, reaching a maximum GDD in 2015 where all lakes were above 1000 GDD. There is a ~100 GDD decrease in 2016 (Figure 4.5).

Abundance and age structure.

The mark recapture sampling provided population and age-class abundance estimates on all three lakes (Figures 4.6 and 4.7). Tag loss across all three lakes ranged between 0 and 2.6% of tag implanted, with a mean of 0.82%. Age-1 fish are often smaller than 150 mm and have not fully recruited in the gillnet (Askey et al. 2006), meaning that they are less vulnerable to the gear used than older age-classes. This might explain the low and variable numbers of Age-1 fish captured across lakes and years (Figure 4.7). The control lake, Pantano, presents the most variation in estimated population abundance of

all three lakes (Figure 4.6), but a consistent age structure for Age-2+ fish across years

(Figure 4.7 – top row). In both Lakes Pantano and Today, year 2012 presents a lower population abundance estimate, which could either be the result of environmental variation or be an indication that the populations are not at equilibrium. In Lake Stubby, there is a slight decline following the first depletion event (2014); that decline continues more dramatically following the second depletion (2015) and is still apparent in the following year (2016 – Figures 4.6 and 4.7 – middle row). In the other experimental lake,

Today, there is a slight decline following the first depletion (2014), but the population abundance increases to pre-depletion level in the following years (2015-2016 – Figure

4.6). The age structure in Today changes following the second depletion (2015), with higher ratios of Age-3 and -4 over Age-2 fish (2015-2016 – Figure 4.7 – bottom row),

Age-4 fish were only captured post-depletion, which might indicate that Age-2+ fish are experiencing greater survival rates post-depletion in that lake. According to the variation in the population estimates obtained, the depletion manipulation removed 39% of the total population in Lake Stubby in 2014 and 34% in 2015. In Lake Today, 27% of the total population was removed in 2014 and 33% in 2015. If age at maturity is assumed to be 3 years old, spawners removal in Lake Stubby was 40% in 2013 and 37% in 2014, while in Lake Today 20% of spawners were removed in 2013 and 64% in 2014.

Growth.

Growth rate variation was best described by a model that included the fish length at first capture (L1) and the interaction between lake and depletion status (Table 4.2), meaning that both the slope and intercept of the relationship between growth rate and

initial length varied following the depletion for at least one of the lakes. Only the coefficients associated with the effect of L1 and the interaction between Lake Stubby and depletion status were significant (p-value < 0.001), meaning that in Stubby fish grew faster after the depletion, but that in Today, growth rate did not change following the depletion (Figure 4.8).

The variation in growth across lakes and years was best explained by a model that included fish initial size (L1), lake-year specific density (abundance estimate divided by lake area) and climatic variable (GDD – Table 4.3). Similar to the previous analysis comparing between lakes and pre/post depletion, the relationship with initial size was negative, supporting the idea that older larger fish grow slower. Growth rates decreased with increases in fish density and increased with warmer temperatures (Figure 4.9). The model can be described by:

(4.5) 퐺 = 0.1082 − 0.0012 퐿1 − 0.0010 퐷푒푛푠 + 0.0004 퐺퐷퐷 where G is the growth rate (%/day), L1 is the fish length at first capture (mm), Dens is the lake-year specific density (fish/ha) and GDD is the lake-year specific growing-degree- days value. The coefficients for the variables L1 and Dens were highly significant (p- value < 0.0001), while the coefficient for GDD presented a lower level of significance (p- value < 0.05). Another model including an interaction between the variables density and

GDD was within two AIC points of this model (Table 4.3) and should also be considered.

The direction of relationships between the variables is the same as what is described above, but the effect of GDD varies negatively with increases in density, meaning that the positive effect of GDD on growth rate lessens at higher density.

Discussion

This experiment showed that the effects of an additional source of mortality (such as harvest mortality) can have very different repercussions for population abundance, demography and growth of neighbouring populations due to their specific environmental conditions. As presented in Chapter 2, the ratio of early life stage habitat (stream) to adult habitat (lake) is an important factor impacting population density and size structure that can also impact the magnitude and type of compensatory response of harvested populations (Figure 4.1). Low habitat ratios associated with lower fish density populations would be expected to have a lower compensatory reserve and present a density-dependent growth response at lower harvest rates than higher density populations. In this experiment, the three study lakes were similar in size and climate but differed in habitat ratios. The lake where early life stage habitat was limiting, Stubby, had a habitat ratio 19% smaller than Lake Today, the other experimental lake. Lake Stubby showed an overall decrease in population abundance accompanied by density-dependent growth following the depletion, as would be expected by the Chapter 2 model predictions

(Figure 4.1, left side). The higher density lake that had better spawning and rearing habitat, Today, did not experience a large variation in its total abundance following the manipulation. This lack of variation appears to be the result of an increase in recruitment that compensated for the experimentally harvested fish, again this is consistent with model predictions (Figure 4.1, right side). Growth rate across all lakes (control and experimental) and years presented a negative relationship with density and a positive relationship with the climate variable GDD.

100

Pantano, the lake that was kept as a control throughout the experiment, presented the largest variation in population abundance estimates across years. Experimental lake

Today also presents variation that follows the control lake with lower population abundances in 2012 and 2014. These variations could be associated with natural environmental fluctuations impacting either the population dynamics or the sampling method used. For example, 2012 was a colder year, there could have been higher overwintering mortality, or the low temperature might have impacted the efficiency of the sampling gear used. This variation in both population abundance estimates and GDD across years might also suggest that the systems are not at equilibrium.

The 2014 decrease in Lake Today could also be a direct consequence of the first depletion event that took place in the spring of that year; if it is the case the population rebounded the following year. Despite the removal of 743 fish in 2014 (second depletion event), the total population of Lake Today increased by ~40% in 2015 with the number of spawners (Age-3+) increasing by over 240%. The age structure appears to have changed from a dominance of the Age-2 class to a more even distribution of Age-2 -3 -4 following the second depletion event (Figure 4.7, years 2015-2016), even though those older age- classes were the most targeted by the experimental harvest. This pattern suggests that the juvenile and young recruits that were not caught during the depletion experienced an increase in survival rate, likely related to lower intraspecific competition. This increased survival rate compensated for depletion-induced mortality of older fish, which returned the population abundance to its pre-depletion level. Such increases in juvenile survival is similarly described by Rose et al. (2001) as the main compensatory mechanism in fish. A similar outcome was observed in a whole-lake density manipulation of gizzard shad

101

(Dorosoma cepenadium) density where pre-recruits survival was negatively related to spawner biomass (Catalano and Allen 2011). Similar to what I observed in Lake Today,

Catalano and Allen (2011) did not observe a growth variation in gizzard shad associated with their density manipulation. Rather, they reported a ~70% reduction in spawners biomass while in Lake Today the depletion removed ~20% of spawners in 2014 and

~64% in 2015. In both Catalano and Allen’s study and this experiment, it is possible that the depletion was not large enough to lower density sufficiently to elicit a compensatory growth response.

In the other experimental lake, Stubby, the two depletion events resulted in a steady decline of overall population abundance, the population declined ~10% after the first depletion (2013-2014) then ~50% following the second experimental harvest (2014-

2015) and kept declining with a ~40% decrease one year following the last harvest (2015-

2016). The population’s age structure reflects that overall decrease and, in 2016, the ratio of Age-3 over Age-2 fish in Lake Stubby is similar to what is observed in Lake Today in

2015-2016, again suggesting an increase in juvenile survival rate. The number of fish

Age-3+ decreased during the depletion years (from 2013 to 2015) and then more than quadrupled between 2015 and 2016 (one year after the last experimental harvest). In

Stubby only, the post-depletion low density is associated with an increase in growth rates, congruent with density-dependent growth as presented in various systems and fish species (Post et al. 1999, Jenkins et al. 1999, Lorenzen and Enberg 2002). Increased growth rate could have consequences for both survival and fecundity that would accelerate the recovery of a depleted population. The increase of Age-3+ fish in 2016 is likely the result of increased survival due to lower intercohort competition, as predicted

102

using Chapter 2’s model and simulations (Figure 4.1). A fecundity response that would increase the number of eggs and recruits to pre-depletion level would require more time

(3+ years post-depletion) to be observed. Observation of a fecundity response would require the faster growing post-depletion cohort to reach maturity and its progeny to recruit into the sampling gear, which could explain why it was not observed here only one year post-depletion.

My Chapter 2 and 3 results suggest that the difference in initial density between the populations of Lakes Stubby and Today are related to habitat ratios and that the limited early life stage habitat (stream) of Stubby is responsible for its population’s lower density and, ultimately, limits its compensatory response. This variation of compensatory response within a species has been observed for other freshwater fish species. For example, in lake trout (Salvelinus namaycush) populations, lake size and TDS have been linked to life-history parameters variation (growth and maturation), and small low-TDS lakes have been identified as having less compensatory reserve (Shuter et al. 1998).

Another example comes from striped bass (Morone saxatilis) where, similar to rainbow trout, populations present compensatory responses that vary with their local conditions and by which life stage is the most impacted by density-dependent processes (Rose and

Cowan 2000).

The negative impact of density and the positive impact of GDD on growth rate follows the results of other studies on lake-dwelling rainbow trout in BC (Ward et al.

2017, Varkey et al. 2018). The range of GDD observed over the course of this study was smaller than in other studies, but the climatic variation was observed over the same waterbody across years qinstead of across a landscape of lakes. The effect of GDD was

103

weak but warrants more research into the influence of the interplay between temperature and density variation on fish growth rates.

The use of a whole-lake manipulation provides added realism to this study, but also potentially introduces confounding factors. For example, the impact of GDD was only added after realizing that climatic conditions had varied considerably over the different years of study and was a variable that could be assessed post hoc. The large variation of abundance in the control lake, Pantano, suggest that either other environmental factors are at play or that the population is not at equilibrium, which limits its value in the study.

The findings of this study support the predictions of Chapter 2 and the observations of Chapter 3 on the influence of habitat availability on population dynamics.

A population that is limited by its early life stage habitat availability presents a lower compensatory reserve than another for which the population bottleneck occurs later in life. This has important consequences for fisheries management as lakes with less capacity to compensate for harvest are more at risk of overfishing and should be monitored more closely.

104

Max. TDS Lake Area (ha) GDD HSI WUA Depth (m) (ug/L) Pantano 4.56 6 30 872 0.50 141 Stubby 5.88 8 40 841 0.58 219 Today 6.22 12 35 857 0.37 274

2 2 Model R adjR p-value AIC ΔAIC

1 L1 0.29 0.28 1.5E-06 -168.10 19.27 2 PrePost 0.02 0.00 2.6E-01 -145.45 41.93 3 Lake 0.20 0.17 5.5E-04 -157.82 29.55

4 L1 + Lake 0.45 0.43 8.6E-09 -182.69 4.68

5 L1 * Lake 0.49 0.45 1.7E-08 -183.88 3.49

6 L1 + PrePost 0.29 0.27 9.5E-06 -166.28 21.09

7 L1 * PrePost 0.30 0.26 3.0E-05 -164.99 22.38

8 L1 + Lake + PrePost 0.46 0.43 2.5E-08 -181.79 5.59

9 L1 + Lake * PrePost 0.51 0.48 3.7E-09 -187.37 0.00

10 L1 * PrePost + Lake 0.47 0.43 6.3E-08 -180.89 6.49

105

Table 4.3 An assessment of alternate models to predict annual growth rate based on length at first capture (L1), estimated population density (Density – N/ha) and climate 2 2 2 (GDD). Values of R , adjusted R (adjR ), p-value, AIC and ΔAIC are presented, the most parsimonious model (lowest AIC) is in bold and models within two AIC points are in italic.

2 2 Model R adjR p-value AIC ΔAIC

1 L1 0.26 0.25 4.4E-05 -133.47 13.48 2 GDD 0.00 -0.02 7.5E-01 -116.14 30.81 3 Density 0.18 0.16 1.0E-03 -127.35 19.60

4 L1 + GDD 0.26 0.23 2.6E-04 -131.48 15.47

5 L1 + Dens 0.41 0.39 5.1E-07 -144.58 2.37 6 Dens + GDD 0.24 0.22 4.5E-04 -130.28 16.67 7 Dens * GDD 0.27 0.23 6.3E-04 -130.39 16.56

8 L1 + Dens + GDD 0.45 0.42 3.6E-07 -146.95 0.00

9 L1 + Dens * GDD 0.47 0.43 8.3E-07 -146.35 0.60

106

107

108

Figure 4.3 Timeline of the sampling activities conducted on the three study lakes.

Figure 4.4 Histograms of the number of fish caught in harvest gill nets by age-class during the two years of depletion (2014 and 2015) in Lakes Stubby and Today.

109

Figure 4.5 Variation in growing-degree-days (GDD) across study years (2012-2016, ClimateWNA - Wang et al. 2016) for the three study lakes (Pantano, Stubby and Today).

110

Figure 4.6 Population abundance estimates for each study year in lakes Pantano (top – yellow), Stubby (middle – turquoise) and Today (bottom – purple), the grey polygons are the 95%CI of the estimates and the vertical black dotted lines are the occurrence of the depletion events in the two experimentally manipulated lakes.

111

Figure 4.7 Estimates of number of fish in each age-class for each study year (2012-2016) in lakes Pantano (yellow), Stubby (turquoise) and Today (purple), the black bars represent the 95%CI and the lake-year plots with a grey background are post- depletion.

112

113

Figure 4.9 Graphical representation of the relationship between growth rate, fish length at first capture, population density and growing-degree-days (adjR2 = 0.42, p- value = 3.6 x 10-7). The values of density and GDD presented correspond to the mean value observed plus/minus the standard deviation of the value (high/low).

114

Chapter 5: Predicting Fish Populations’ Distribution, Production and Recreational

Fishing Demand Based on Landscape Characteristics

Introduction

Management of habitats and animal populations often requires detailed local information to inform decisions made over large areas. Such information usually necessitates in situ data collection that requires large amounts of sampling effort, but much of this knowledge can also be obtained using landscape-level variables that explain the processes responsible for variation in the physical structure of habitats and population dynamics. Landscape ecology describes how patterns of spatial heterogeneity emerge and impact ecological processes at different scales. This discipline has been more commonly used for terrestrial habitats, although its use in aquatic ecosystems is growing. Wiens

(2002) is responsible for “taking landscape ecology in the water” and using its principles to describe riverine systems, recognizing that aquatic ecologists were already using landscape ideas by linking the physical structure of aquatic systems and their drainage to explain the function and structure of their ecosystems (e.g. the River Continuum Concept,

Vannote et al. 1980). The connectivity and dendritic organization of stream networks makes them obvious candidates to apply landscape ecology principles to aquatic systems.

Lakes were not initially included in the riverscape idea as they are often seen as distinct systems or patches isolated from one another, but they are linked together and interact with their tributaries in that same stream network. Soranno et al. (2009) presented the influence of the hierarchy of aquatic, terrestrial and human landscape features on the limnological characteristics of lakes through the Lake Landscape-Context framework

115

(LLC). It is suggested that the position a lake occupies in the watershed will influence its limnological conditions, its species richness and assemblage (Martin and Soranno 2006,

Mehner et al. 2007).

Recreational fisheries management has also been adopting a landscape approach where patterns of angler demand are described over large areas. Anglers can be seen as predators that prey on patches (lakes) and, by their movement, connect the different patches together (Carpenter and Brock 2004, Post et al. 2008, Hunt et al. 2011, Ward et al. 2013b, Dabrowska et al. 2014, Carruthers et al. 2018). Through that connection, the outcome of fishing on one lake, i.e. the perceived quality of fishing (Parkinson et al.

2004, Askey et al. 2013, Wilson et al. 2016), can impact the angler’s decision to stay on the patch they are exploiting or move on to another lake and impact its fish population.

Therefore, in exploited systems, fish population dynamics is not independent among lakes. The vulnerability of specific populations to overfishing across the landscape can then be evaluated by coupling the fishing demand with information on fish production

(Post and Parkinson 2012). Fishing demand can be described simply as the number of potential anglers, but details on angler behaviour and spatial distribution of different angler types can improve our understanding of the human dimension of recreational fisheries (Hunt et al. 2011, Ward et al. 2013b, Dabrowska et al. 2014). An accurate description of the spatial structure of the system, the corridors (roads and trails) connecting anglers to fishing opportunities and the elements that facilitate or impede angler use or desire to use the resource can greatly improve predictions of angler demand distribution across a landscape (Carruthers et al. 2018).

116

In studies of recreational fisheries, fish production variation at the regional scale has primarily been investigated via the lens of climate and its impact on fish growth.

Growing-degree-days (GDD) has been suggested as a strong predictor of fish growth and can easily be calculated across large areas (Neuheimer and Taggart 2007, Venturelli et al.

2010, Ward et al. 2017, Varkey et al. 2018). Most landscape-scale studies on recreational fisheries focus on populations maintained through stocking programs (Post and Parkinson

2012, Askey et al. 2013, Carruthers et al. 2018), and the drivers of natural density variation are rarely investigated, even though the importance of density-dependent processes at the lake-scale is clear (Byström and Garcia-Berthou 1999, Post et al. 1999,

Lorenzen and Enberg 2002, Arranz et al. 2016). An understanding of the environmental variables influencing natural production would greatly improve the ability to predict the production of wild fish populations, particularly at broad spatial scales. The previous chapters have described how the variation in local stream and lake characteristics can impact population density, growth parameters (Chapter 2-3) and compensatory reserve

(Chapter 4). In this chapter I will use that local scale information to identify fish production patterns at a broader spatial scale.

The main objective of this chapter is to develop a theoretical framework to predict broad, landscape-scale fish production. Using the British Columbia (BC) rainbow trout as a case study (Oncorhynchus mykiss), I will illustrate how this approach can be used for broad-scale prediction of fish population attributes such as occupancy, density and size distribution. To do so I will: 1) describe the distribution of rainbow trout in BC and predict the probability of presence of rainbow trout across a region; 2) use environmental variables available at the landscape scale to infer rainbow trout population dynamics

117

based on the local processes presented in the previous chapters; and 3) use proximity to human population centres to assess the potential recreational fishing effort across rainbow trout populations of this same landscape.

Methods

Rainbow trout distribution.

Species information.

The rainbow trout is a salmonid native to the Pacific basin in North America and northeastern Siberia. In Canada, the species is native in lakes and streams west of the continental divide that flow toward the Pacific Ocean, with the exception of the headwaters of the Peace, Liard and Athabasca drainages that flow into the Mackenzie

River toward the Arctic Ocean (McPhail and Carveth 1992, McPhail 2007). The native distribution of rainbow trout in Western Canada is the result of post-glacial immigration from the Columbia and Bering ice-free regions (McPhail and Lindsey 1986). Rainbow trout are very popular with anglers, and stocking events have been reported in Western

Canada since the early 20th century with accidental and unreported stocking also influencing the dispersal of rainbow trout within and outside of its native range.

Data source.

The current understanding of the distribution of rainbow trout across the province of BC was established using historical fish observations and stocking records. To obtain the location and characteristics of lakes and rivers I used the EauBC Lakes layer (British

Columbia Ministry of Environment and Climate Change Strategy - Knowledge

118

Management 2017a), which is based on the original 1:50,000 Freshwater Atlas (FWA;

Gray 2010, British Columbia Ministry of Forests Lands Natural Operations and Rural

Development - GeoBC 2017). To capture all waterbodies stocked in BC, 19 lakes from the 1:20,000 FWA layer and one lake/wetland polygon from the 1:20,000 Wetlands layer

(British Columbia Ministry of Forests Lands Natural Operations and Rural Development

- GeoBC 2017b) were added to the EauBC Lakes Layer. Rainbow trout observation data comes from the Known BC Fish Observations and BC Fish Distributions layer (British

Columbia Ministry of Environment and Climate Change Strategy - Knowledge

Management 2017b). Stocking records used were provided by the Freshwater Fisheries

Society of British Columbia (FFSBC) using queries of their private Small Lakes

Database (Freshwater Fisheries Societry of British Columbia 2017); additional stocking data was provided by provincial and federal governments (British Columbia Ministry of

Forests Lands Natural Operations and Rural Development 2017, Parks Canada 2018).

Categorization.

Lakes were classified into five categories based on species observation (rainbow trout observed, not observed but other species were or no information on species assemblage) and historical stocking (Table 5.1). For both the “wild” and “naturalized” categories fish are naturally reproducing, but in the “naturalized” category there is evidence that the population might be of hatchery origin or at least has been previously supplemented by hatchery product. Large uncertainty is associated with the “apparently absent” category as it is often difficult to confirm species absence (McArdle 1990, Kéry

2002, Mackenzie et al. 2002). For example, sampling effort could have been insufficient,

119

or the methods used did not target rainbow trout, leading to false absences. Information on specific sampling protocols of all known fish observation points used is limited. Since rainbow trout is one of the most popular freshwater fish species for recreational harvest in

BC, it is assumed that most sampling events would have targeted that species. Finally, if there was no known fish observation and no stocking reported, lakes were classified as

“unknown”.

Probability of presence.

Since rainbow trout presence/absence is unknown for many BC waterbodies, stream network connectivity was used to estimate the probability of rainbow trout presence in lakes categorized as “unknown”. To develop this methodology, BC watersheds with the most detailed field observations — the Clearwater and North

Thompson Rivers sub-sub-watersheds (area highlighted in Figure 5.1 and presented in

Figure 5.2) — were used. Both rivers originate from glaciers in the Cariboo Mountains.

Most of the Clearwater River flows in Wells Gray Provincial Park and drains into the

North Thompson River, the northern branch of the Thompson River, largest tributary of the Fraser River, which flows into the Pacific Ocean. As the Fraser system was completely iced-over during the last glaciation, the current species distribution is the result of postglacial dispersal coming primarily from the Columbia system (McPhail and

Lindsey 1986). Rainbow trout is a fairly ubiquitous species that can establish populations in a wide range of abiotic conditions, which limits the ability to predict its distribution from predictive habitat models based on local conditions alone, but models using largescale geomorphic variables have been used with success for this species (Porter et al.

120

2000). The presence of within-basin biogeographic barriers to upstream fish movement has also been identified as an important limiting factor to fish distribution at the watershed scale (Kruse et al. 1997, Latterell et al. 2003, Fransen et al. 2006, Rahel 2007).

Stream network connectivity is described using the presence of barriers to upstream fish movement and distance to nearest known fish observation using a graph- theoretic approach as presented for landscape connectivity (Urban and Keitt 2001) and in streams seen as both directed and dendritic networks (Schick and Lindley 2007,

Campbell Grant et al. 2007). The directed graph of the BC stream network is built using the NetworkX Python module (Hagberg et al. 2008). The stream segments are the edges of the network, while lakes and stream segment connections where tributaries enter a stream system are the nodes. The directed graph is built under the assumption that downstream movement is always possible but that upstream fish movement may be limited by barriers, such as waterfalls, cascades and steep slopes, which have been identified as important biogeographic elements structuring fish distribution (Rahel 2007).

A second assumption on upstream fish movement is that proximity to an established population of rainbow trout (i.e., source) increases the probability of colonization and observed presence.

To describe both the barrier threshold and the probability of presence based on proximity to nearest observation, I used a validation subset of the Clearwater/North

Thompson lakes classified as “apparently absent” and “wild” (presence not related to stocking). The validation subset’s connectivity along the directional stream network to

“wild” and “naturalized” rainbow trout populations was assessed and each lake of the validation subset was categorized as connected to a rainbow trout observation or not.

121

Barriers to fish movements were introduced in the network using a GIS stream network layer that summarizes known blockades (e.g. cascades, dams, obstructions) and stream gradients ranging from 10 to 30% (this layer is still in development and was generously shared by the BC Ministry of Environment; for an overview see Methodology: Phase II in

Mount et al. 2011). For each lake in the validation subset and each gradient barrier level

(10, 15, 20, 25 and 30%), connectivity to known “wild” or “naturalized” populations was assessed to determine what stream gradient represents a barrier to upstream fish movement.

I used a logistic regression to describe the colonization probability, the validation subset is the dependent variable (presence = 1, absence = 0) and distance to nearest

“wild” or “naturalized” population along the stream network is the independent variable.

The logistic model’s statistical significance was assessed using a chi-square test (alpha =

0.05). The model was evaluated using repeated K-fold cross validation in the caret R package (Kuhn et al. 2018), 10 folds were repeated over 10 iterations, meaning that for each iteration the validation set of lakes was shuffled and separated in 10 groups and that one-by-one each group was held-out (test set) while the remaining groups were used to predict its probability of rainbow trout presence (training set). The performance of the model is then described using kappa, a measure of accuracy or of the agreement between the predictions of the different folds (Landis and Koch 1977).

The probability of rainbow trout presence for all lakes categorized as “unknown” was then calculated using the barrier threshold selected and the calculated relationship with distance to nearest observation. If a barrier was present, no upstream movement was possible and lakes that were not connected had a probability of 0. For lakes that were

122

connected to known observations, the probability of rainbow trout presence was calculated using the logistic relationship determined using the validation subset. The logistic regression probability was then used to categorize lakes as having rainbow trout

“present” or “absent” using a threshold of 0.5, lakes over the threshold were characterized as “present” while lakes under the threshold were characterized as “absent”.

Rainbow trout population characteristics.

In the previous chapters, I linked population dynamics and growth parameters to local environmental characteristics, particularly life stage specific habitat quality and availability. In Chapter 3, weighted useable area (WUA), calculated from the Habitat

Suitability Index (HSI) methodology, was identified as a good predictor of early life stage habitat availability; when coupled with lake area to make a habitat ratio, WUA can be used to predict population density and growth parameters. WUA is a detailed measure of habitat quality and quantity that requires intensive sampling and is only available for a small number of fish populations. In contrast lake area is usually easily available in databases and GIS layers. I searched for covariates of stream WUA that were readily available through databases and GIS layers and could be used to predict WUA and associated population characteristics at the landscape scale. I used WUA data collected for the 39 rainbow trout populations presented in Chapter 3 to compare the different proxies. Detailed information on each lake used is presented in Appendix B (Table B.1), stream data collection protocol is presented in Chapter 3 and Appendix C (Table C.1), and the WUA calculation methodology is detailed in Chapter 3 and Appendix D (Table

D.1).

123

The landscape-scale variables were all obtained or calculated from GIS layers contained in the FWA. Lake area and perimeter length describe the size of the lake and were included because I assumed that bigger lakes tend to have more, bigger inlets and these measures would be the simplest to describe habitat availability. The shoreline development index (DL) describes the shape of the lake and was calculated as:

퐿 (5.1) 퐷 = 퐿 2 휋 퐴 where L is the lake perimeter and A is the lake area (Wetzel 2001). The index cannot be smaller than 1 and the closer the index is to 1 the more circular the lake is. A high index could suggest the presence of more bays associated with tributaries and potentially more stream habitat. The number of inlets and outlets was calculated using the number of stream segments that intersected with each lake polygon, lakes that present more tributaries would likely have more stream habitat. Lake order refers to the Strahler order

(Strahler 1957) and was attributed to each lake based on the maximum order of streams intersecting with each lake polygon. As stream order increases, streams are expected to get bigger (Hughes et al. 2011) and present more complex physical conditions (Vannote et al. 1980), which could provide more early life stage stream habitat.

The different landscape-scale predictors were compared using univariate linear regression, assuming that WUA is log-normally distributed. The adjusted coefficient of variation (adjR2) was used to describe how much of the variation in stream WUA was explained by each predictor. The predictor presenting the highest coefficient of variation was then selected and used to predict stream WUA.

124

Recreational fishing effort.

Recreational fishing pressure has been linked to population centre proximity as driving time limits the movement of anglers over a landscape of lakes (Post et al. 2008).

To characterize the potential angling pressure on each rainbow trout population, the distance to the city of Kamloops, the main population centre near the Clearwater/North

Thompson watersheds, was calculated. This distance also encompasses the movement of anglers from large population centres to the west (Vancouver and lower mainland) and the east (Calgary, Alberta), as the main road through the watersheds of interest, highway

5, connects to the TransCanada highway at Kamloops.

The first variable calculated for each lake was access; it was assumed that a lake had no road access if it was more than 100 m from any road. Hiking distance was calculated as the straight-line distance between lake and nearest road using the Near tool in ArcMap. Hiking time was estimated assuming a speed of 4km/h. The driving distance and time was then calculated for all lakes using an Origin-Destination Matrix (Network

Analyst) based on the stream network which details the length (km) and speed limit

(km/h) of each road segment. A ratio relating fish production to recreational fishing demand was calculated by dividing the predicted WUA to lake area ratio (production) by the inverse of total travel time calculated for all lakes (demand). To summarize information on rainbow trout across the watersheds of interest, surfaces generated by kriging interpolation methods were created for a) the probability of presence, b) WUA to lake area ratio, c) total travel time and d) ratio of fish production and fishing demand. All spatial data manipulation and analyses were done in ArcMap 10.4 (ESRI 2011) and

125

Python for Windows extension (PythonWin 2.7, Hammond et al. 2008). Statistical analyses were done in R (R Core Team 2017).

Results

A total of 27,402 lakes with a sum of 20,540 km2 of water distributed across the

944,735 km2 area of BC were categorized into five categories describing historical observations and stocking of rainbow trout (Table 5.2, Figure 5.1). Over 80% of all BC lakes were in the unknown category, meaning that there were no fish surveys reported and no record of rainbow trout stocking for these lakes. These waterbodies represented

~15% of the total area occupied by lakes over 4 ha, which suggests that there are many lakes for which no information is available but most of them are small, while larger lakes tend to be better documented. In the merged Clearwater and North Thompson sub-sub- watersheds, 924 lakes over 4 ha were documented in the database, and 546 of them were classified as “unknown”. Similar to the whole province, a disproportionally large number of small lakes are included in the “unknown” category, but they represent a small portion of the total lake area (59% of the total number of lakes and 10% of the total lake area for the region - Table 5.2, Figure 5.2).

Using a validation subset composed of all the lakes in the Clearwater/North

Thompson classified as “apparently absent” or “wild” in the past 20 years (total of 44 lakes), I established a stream network structure that can be used to predict rainbow trout distribution. No rainbow trout populations were observed upstream of any of the gradient thresholds considered (10-30%, Table 5.3), while a gradient of 10% was the threshold associated with the highest number of lakes for which species assemblage was known

126

and no rainbow trout were observed (“apparently absent” category). This result suggests that even the smallest gradient (10%) can be considered a barrier to upstream movement of rainbow trout. The decreasing probability of presence of rainbow trout (Ppres) as distance to nearest observation increases was described using a logistic regression (Figure

5.3, p-value = 0.00973) and described by the equation:

−4 푒1.35 −1.21 ×10 푑푛푒푎푟 (5.2) 푃 = 푝푟푒푠 1+ 푒1.35 −1.21 ×10−4 푑푛푒푎푟 where dnear is the distance (m) to the nearest known rainbow trout observation of the same subset using the “apparently absent” lake category as absence (0) and “wild” category as presence (1). Model K-fold cross validation yielded a kappa value of 0.43 which is described as an agreement of moderate strength between the different folds

(Landis and Koch 1977).

Lakes identified as “unknown” and upstream of a gradient barrier were given a 0 probability of rainbow trout presence, while for lakes connected to observations, the probability of presence was based on distance to the nearest observation and calculated using equation 5.2. A distance of ~11 km corresponds to a probability of 0.5, meaning that lakes further than 11 km from the nearest rainbow trout observation were estimated as having less than a 50% probability of presenting a rainbow trout population and were predicted as “absent”, while lakes above that threshold were predicted as “present”.

Figure 5.4a) and Table 5.4 present an example of how presence probabilities were estimated along the stream network, and Figure 5.5a) shows the overall predicted distribution of rainbow trout across the Clearwater/North Thompson region. The upper reaches of the watershed in the northern portion of the region present a low probability of

127

rainbow trout presence. This area is mostly located within Wells Gray Provincial Park, which contains steep terrain resulting in high gradient streams (i.e. barriers), but also contains low numbers of fish observations and appears to not have been sampled as intensely as the rest of the region. Areas of high probability rainbow trout presence mostly follow the North Thompson River valley, while higher elevation portions of the watershed generally correspond with lower presence probabilities, particularly in the case of headwater lakes.

Local stream habitat quantity and quality characteristics, described as the WUA of stream for rainbow trout using detailed data collected from 39 rainbow trout populations, was best explained by the position of the lake in the watershed using stream order (Table

5.5). The lakes used to describe the relationship range from first to fifth order, while across the whole Clearwater/North Thompson the maximum stream order is eight. As stream order increases the quantity and quality of stream habitat (WUA) also increases

(Figure 5.6, R2 = 0.5, p-value = 3.5 x 10-7):

(5.3) ln 푊푈퐴 = 3.85 + 0.81 푆푡푟 where lnWUA is the log-transformed stream weighted useable area (m2) and Str is the maximum Strahler stream order associated with that lake. An example of a small area of the Clearwater/North Thompson where lakes of similar size have contrasting positions in the watershed is presented in Figure 5.4b) and Table 5.4. According to the relationship presented in equation 5.3, the higher order lake would have the highest WUA, with ~106 m2 for a first order stream and ~534 m2 for a third order stream. Based on the observations of Chapter 3, higher WUA for identical lake area would correspond to a greater ratio of stream to lake habitat, leading to increased recruitment and lake density.

128

If high enough, this ratio can restrict individual growth and result in a stunted population.

For each lake, stream order is used to calculate predicted WUA (equation 5.3) which is then divided by lake area, providing an estimate of WUA to lake area ratio, with values ranging from 0.7 to 1377 m2/ha. The WUA to lake area ratio across the region is summarized in Figure 5.5b). In general, higher elevation areas present lower order streams and result in lower WUA/lake area ratio values.

Lake accessibility was calculated as total travel time (sum of driving and hiking time) from the main population centre of Kamloops to all the lakes in the

Clearwater/North Thompson region, ranging from 25 to over 1000 minutes. Additional accessibility information includes number of lakes (with known stocked and observed or predicted rainbow trout presence) within 10 km of each lake (stocked range: 0-18; present lakes range: 0-74 lakes). An example of how accessibility information was compiled is presented in Figure 5.4c) and Table 5.4. Areas containing protected areas tend to require greater travel time because, due to their remoteness, their road density is generally lower and many of their lakes require hiking. For example, Wells Gray and

Taweel Provincial Parks in the northwest and Dunn Peak in the east are all over five hours total travel time from Kamloops (Figure 5.5c).

Areas of greater conservation concern, either due to low predicted fish production

(low WUA/lake area ratio) or high recreational fishing demand (short travel time), are located primarily in the southern portion of the region near Kamloops and located near main highways (Figure 5.7). The probability of rainbow trout presence was not considered in the development of Figure 5.7, but most areas presenting a low probability

129

of rainbow trout presence (brown areas in Figure 5.5a) correspond to areas of low concern in Figure 5.7 (blue areas – high fish production to fishing demand ratio).

Discussion

In this component of my thesis research, I described the current distribution of rainbow trout in BC, and a more detailed occurrence and probability of presence of the species in the Clearwater and North Thompson Rivers watersheds. Lake watershed position, described using Strahler stream order, was identified as the best proxy for availability of early life stage habitat, which, once coupled with lake area, is a predictor of fish production and general population dynamics, as presented in previous chapters.

This chapter (Chapter 5) describes a methodology predicting the probability of rainbow trout presence, infers population dynamics from estimated habitat ratios, and describes potential fishing effort based on landscape characteristics and travel time from population centres. I provide a supply-demand analysis combining fish production with recreational fishing demand, which can be used to identify areas of concern containing populations either more sensitive to habitat disturbance or more at risk of overfishing.

My research demonstrated that there are likely still many, many small lakes for which the presence of rainbow trout has not been confirmed, even though the species has been extensively studied in British Columbia and is one of its most popular freshwater sport fish species. Local knowledge on those small and often remote waterbodies likely exists, but the information is not captured in any official database. No rainbow trout population was observed upstream of any stream segments identified as a barrier. The data set only had 21 lakes for which rainbow trout were identified as “apparently absent”

130

and, of those lakes, four were above identified stream gradient barriers. This lack of populations above barriers supports the idea that geomorphic barriers can limit upstream colonization and persistence (Nelson et al. 1992). The limited dataset of known absences of rainbow trout from lake fish species assemblages suggests known obstructions and a stream gradient of 10% are sufficient to limit rainbow trout distribution.

As the streams presented in the analysis were of small size and almost no obstructions were recorded in the databases (such as cascades or dams), most of the barriers encountered were related to steep stream gradient. Due to the small stream size, it is plausible that many absolute barriers, like cascades, or partial barriers, like shallow riffles at low flow or beaver dams that can change drastically over relatively short time periods, are not reported in the database used. Little sampling is done on small streams, particularly if they are remote and not impacted by road crossings. Gradient and other correlated variables, such as stream size and elevation, have been identified as good predictors of upstream limit of stream-dwelling cutthroat and rainbow trout distribution

(Kruse et al. 1997, Latterell et al. 2003, Fransen et al. 2006). Kruse et al. (1997) suggest a stream gradient ≥ 10% limits Yellowstone cutthroat trout distribution in northwestern

Wyoming streams as was observed for rainbow trout in this analysis. Similarly, my study shows that high stream gradients block upstream fish movement on most lower order streams and result in low probability of rainbow trout presence, as demonstrated in the northwest portion of the Clearwater/North Thompson watersheds corresponding to the

Clearwater Valley in Wells Gray Provincial Park. These results could be attributable to a lack of information, because not many of Wells Gray backcountry lakes have been sampled, but there is evidence that most of the area’s lakes contained no salmonids prior

131

to introductions by stocking programs (Goward and Hickson 1995). The presence of other fish species, such as largescale suckers (Catostomus macrocheilus) and redside shiners (Richardsonius balteatus), but no rainbow trout, is likely due to changes in drainage patterns during the deglaciation of this area that resulted in two colonization waves from the Columbia drainage. The first wave made it farther toward the headwaters, transporting coarse scale suckers and redside shiners, while rainbow trout came with the second wave, yet stopping at volcanic activity that produced Osprey Falls at the end of

Clearwater Lake (McPhail and Lindsey 1986, Coombes 1991). Additional data on known rainbow trout presence and absences, ground truthing of absence predictions for example, would help confirm the relationship with stream barriers and distance to closest known observation, in addition to defining rainbow trout distribution limitations with more confidence in this watershed and the rest of the province. Detailed information on the impact of different types of barriers would provide the information required to produce a more elaborate model describing the cumulative passage probability through different types of barriers, similar to what has been done on the Truckee River in Nevada (McKay et al. 2013)

Barriers to colonization are a largescale biogeographic filter of rainbow trout, but finer-scale filters relating to spawning habitat and local biotic and abiotic conditions can also impact fish distribution and population dynamics (Quist et al. 2005). Lake-dwelling rainbow trout require access to tributaries for spawning as shoal spawning is largely unsuccessful (Scott and Crossman 1973, McPhail 2007). Flow changes and obstructions, by beaver dams for example, whether temporary or permanent, can limit population persistence and impact access to early life stage habitat. Also, small lakes are often

132

shallow and more at risk of winterkill, which can lead to large population declines and even extinction. In both those cases, a lake could be or have been connected to other populations that might recolonize it but cannot support its own rainbow trout population.

These additional explanations of rainbow trout absence can explain the lakes categorized as “apparently absent” that are not upstream of a barrier and close to rainbow trout observation (probability of 0 and < 5 km from an observation in Figure 5.3).

I observed a relationship between lake watershed position, as defined by lake order, and availability of early life stage habitat (WUA). Lake order is a variable that can easily be calculated with readily available stream network data and can be used as a proxy of stream habitat availability to young rainbow trout, but which can also inform us of the limnological conditions of their adult habitat. Lake watershed position has been identified as a descriptor of lake limnological conditions such as morphometry and concentration of ions (Kratz et al. 1997, Riera et al. 2000, Quinlan et al. 2003, Martin and

Soranno 2006). These previous studies have all supported the idea that lakes should be considered as part of their landscape, and that their connection through dendritic stream networks influences their physical, chemical and biological conditions as has been described for stream networks (Vannote et al. 1980).

In general, lower order headwater lakes get most of their water input from precipitation, while groundwater contributes more lower in the watershed (Kratz et al.

1997). The magnitude of the streams’ input to the lake then drives water chemistry with, for example, higher ion concentrations in higher order lakes (Riera et al. 2000, Quinlan et al. 2003, Martin and Soranno 2006). Variation in ions and nutrient concentration then impacts primary productivity and species richness (Riera et al. 2000). In a study of 86

133

lakes in south-central Ontario, Quinlan et al. (2003) found that higher order lakes tended to be deeper with higher hypolimnetic dissolved oxygen concentrations, providing better habitat for lake trout (Salvelinus namaycush). Rainbow trout are not as specifically linked to a lake habitat characteristic as are lake trout, but because of their use of stream habitat during early life stages, variation in the size and type of streams highly influences their population dynamics. As lake order increases, tributaries get bigger and present more complex habitats that provide cover for invertebrates and fish, increasing the stream’s capacity to support more young of the year fish and older juveniles. Lake order remains a very coarse categorical measure of lake watershed position, and as sum, the development of a finer-scale quantitative metric would improve the predictions of stream habitat availability.

Landscape-level studies on recreational fisheries, and the BC rainbow trout fishery in particular, have focused on the prediction of fishing effort distribution (Post et al. 2008, Post and Parkinson 2012, Ward et al. 2013c, Dabrowska et al. 2014, Carruthers et al. 2018) and fishing quality as it relates to the trade-off between fish number and size

(Askey et al. 2013, Wilson et al. 2016, Varkey et al. 2018). These studies recognize the importance of density-dependent processes but present density as the result of stocking programs or, if using wild populations, do not investigate the processes responsible for density variations. This study explicitly accounts for natural density-dependent processes by linking habitat variables to fish production beyond the sole impact of climate or nutrients on growth. By building on the relationships between habitat availability and population characteristics from the previous chapters, I describe here how landscape- scale environmental variables can be used to predict local habitat availability responsible

134

for population density and life history parameter variation. This link to processes allows for a more accurate description of the dynamics of wild rainbow trout populations at the landscape scale. For example, the relationship between lake order and WUA suggests early life stage habitat availability increases from the headwaters to the higher orders of a watershed.

To account for recreational demand, I used a crude estimate based largely on travel time from the main population centre. Travel distance has been presented as one of the important decision factors on recreational angler effort allocation (Post et al. 2008).

Using travel time instead of distance accounts for the differences in road type (paved vs gravel) and travel type (driving vs hiking). This method can easily be reproduced using appropriate road network parameters. The method can also be expanded to include more complex predictions of recreational fishing effort that involve complex utility models and decision making, as has been done for the stocked rainbow trout fishery of BC

(Carruthers et al. 2018). Integrating recreational fishing demand with predictions of fish production and population characteristics helps identify hotspots for conservation where production is low, and demand is high. This information can also inform management objectives, such as identifying populations having the potential to produce large fish and which could potentially be used as trophy fisheries.

The findings I have presented here could be improved by adding more validation data to understand the uncertainties and limits of the occupancy model, but also data from other watersheds to allow for predictions of rainbow trout production across BC. This added data would help better understand which factors are most limiting the distribution and production of rainbow trout while also adding further validation to the relationships

135

observed here. This methodology describing directional network connectivity and integrating habitat conditions could also be used to predict invasion by non-native species or probability of genetic purity of native species exposed to introgression with non- natives. A broad-scale approach that investigates landscapes, riverscapes and lakescapes has great potential to help predict conditions and threats across a large area, but it is also interesting from a more theoretical standpoint as it suggests the ecological basis of many processes.

136

Table 5.1 Lake categories determined by observation of rainbow trout (RB) and stocking history. Category Observed Stocked

Wild Yes Never

Yes, historical Naturalized Yes (> 20 yrs) Yes, recent Stocked Yes (< 20 yrs) Apparently No RB, but other Never absent species observed Fish surveys Unknown Never absent

Table 5.2 Lakes categorized by rainbow trout historical data in BC and in the combined sub-sub-watersheds of Clearwater/North Thompson. Predicted values assign “unknown” lakes to either “present” (“wild” and “naturalized” combined) or “apparently absent” categories. Clearwater/North Thompson Province of British Watersheds Columbia Category Observed Predicted N. of Area N. of Area N. of Area lakes (ha) lakes (ha) lakes (ha) Wild 1667 802225 207 20462 360 39365 Naturalized 717 396775 54 17954 Stocked 885 134500 96 10855 96 10855 Absent 1318 306170 21 4233 468 9096 Unknown 22815 414267 546 5812 Total 27402 2053936 924 59316 924 59316

137

Gradient Number of lakes upstream of barrier Threshold (%) Present Absent 10 0 4 15 0 2 20 0 2 25 0 1 30 0 0

138

Table 5.4 Characteristics of the four lakes presented as an example in Figure 5.4.

Lake number in 10km Access Pred. WUA buffer Category Area Lake Order Lake Ratio Total (Prob) (ha) 2 Road Hike (m /ha) travel time All RB Stocked RB (min) (min) (min) A Absent (0%) 4.17 1 25.34 98 0 98 47 8

B Present (7%) 4.06 4 295.8 105 4 109 41 5

C Present (7%) 4.49 1 23.53 97 4 101 38 5

D Present (7%) 9.63 4 124.6 94 0 94 50 9

139

Table 5.5 Model comparison of predictors of lake tributaries’ weighted useable area (lnWUA) and their associated statistics; the model with the highest coefficient of variation (R2) is in bold.

Model R2 p-value

Stream order 0.51 3.5E-07

Number of streams 0.09 0.07

Lake perimeter length 0.02 0.45

Lake area 0.03 0.30

Shoreline development index 4.0E-03 0.69

140

141

142

143

Figure 5.4 Subsets of the merged Clearwater/North Thompson sub-sub-watersheds used to illustrate: a) the predicted probability of rainbow trout presence based on gradient barriers (red stream segments) and distance to nearest observation (flow direction presented as arrows on the blue stream network), where, in this first example, lake A is categorized as “absent” (probability of 0) because it is upstream of a barrier, while lake B is categorized as “present” (probability of 0.7) because there is no barrier and it is close to a known rainbow trout observation; b) the influence of stream order (represented as the thickness of the blue stream network), where lake B has large fourth order tributaries but lake C, of similar area, has small first order tributaries limiting the availability of early life stage habitat; and c) the different scenarios of accessibility, as lake D is connected to the road network (in grey) while lake C is over 100 m from a road and necessitates a hike-in. Note that the lake categories are simplified into “present” (wild and naturalized origins, in yellow), “present but stocked” (purple) and “absent” (brown). In example a) the black lake contour identifies lakes previously categorized as “unknown” and for which the probability of presence is predicted, in examples b) and c) lakes identified as “present” are lakes in which rainbow trout were observed or predicted (probability > 0.5) while “absent” lakes are where no rainbow trout were observed or predicted (probability < 0.5).

144

145

146

147

148

149

Chapter 6: Conclusions

The main objective of this thesis was to investigate the influence of habitat availability on population dynamics. The four research chapters have provided evidence of the structuring role of habitat and its impact on density-dependent processes happening throughout a population’s life cycle and from which population abundance and size structure arise. The studies described in the chapters of this thesis build upon and complement one another to demonstrate the importance of habitat limitation, first at a fine scale through its impact on specific life stages’ growth, survival and behaviour within a population; then to showcase how habitat characteristics are the drivers of variation among populations; and, finally, to describe trends in populations at a landscape spatial scale. In this concluding chapter, I highlight the contribution of my work to the fields of population and landscape ecology and its applicability to habitat and fishery management. As I summarize my results, I will present some of the limitations of my findings and offer suggestions for future research directions.

In Chapter 2, I assessed the importance of habitat limitation using a theoretical approach. In order to create quantitative predictions, I built an age-structured population model that accounted for interactions at all life stages and through the different habitats used by a species experiencing ontogenetic habitat shifts. The model was parameterized based on density-dependent processes in both stream and lake salmonid populations. I simulated the population outcomes for a gradient of habitat availabilities, climatic conditions and harvest. This approach predicted that habitat limitation at the early life stages (stream) would lead to low population densities but high maximum size, while

150

habitat limitation of the adult stage (lake) would result in high density stunted growth populations. Across all simulations, there was evidence of a size-number tradeoff, no high-density population presented a high maximum size. For all habitat ratios, high harvest pressure led to compensatory growth and increased juvenile survival. Population collapses only occurred when early life stage habitat was limited, suggesting that early population bottlenecks lead to lower compensatory reserve. This theoretical chapter highlighted that density-dependence can occur and impact populations at any life stage.

The model created could easily be re-parameterized to follow size-structured populations of other species through ontogenetic habitat shifts, and it would be interesting to see if the same patterns arise.

In Chapter 3, I presented a large-scale field study, consisting of 39 populations, where I assessed the role of habitat on population dynamics. The results supported the theoretical framework of Chapter 2 by showing that low early life stage (stream) to adult

(lake) habitat ratio led to low density populations where fish reach larger maximum size and mature at a larger size. In contrast, abundant early life stage habitat (high habitat ratio) was associated with high density stunted growth populations with lower size at maturity. The same size-number tradeoff as in the Chapter 2 model was observed. Finer- scale observations at the stream level supported the use of the Habitat Suitability Index

(HSI) methodology in this context, particularly when used to describe weighted useable area (WUA) and showed its implications for habitat capacity. Observed associations of stream habitat to specific age-classes validated assumptions made in the model regarding stream to lake migratory behaviour that were based on territoriality relationships from stream-resident salmonids ecology.

151

The results of Chapter 3 highlighted an important variable not included in the

Chapter 2 model, the presence of other fish species. Monocultures are very simple fish communities that are not that common. Chapter 3 shows that the presence of other fish species reduced growth rates and recruitment. Including the impact of more complex species assemblages in the Chapter 2 model would make that framework applicable to a wider range of species by accounting for the influence of interspecific interactions.

Expanding my data collection, analysis and modelling to include more interspecific interactions would broaden the inferences to a greater number of fish communities at both the lake and landscape scale.

In Chapter 4, I assessed the importance of habitat-mediated processes using the finer resolution of a 5-year time series for three lakes using mark-recapture analyses. My goal was to use an experimental approach with a control to assess the compensatory response to density reduction for two populations which differed in habitat availability.

This study provided an experimental test of the Chapter 2 theoretical framework predictions related to the effects of harvest. As predicted by the model, the population presenting the lowest habitat ratio had an initial lower density and a lower compensatory reserve. Following the harvest experiment this population expressed a density-dependent growth response, while it appeared that the higher habitat ratio/ density population compensated for the increased mortality rapidly through increased juvenile survival and no compensatory growth was observed.

Most of the studies and experiments investigating the impact of density- dependence on rainbow trout have used stocked fish and mostly followed one or two age- classes. By using naturally reproducing populations and exploring density-dependent

152

processes from a fine habitat specific scale to a broader population scale, I showed that there are important feedback loops operating that could not be captured without considering the whole life cycle and accounting for the impact of fecundity variation and intercohort interactions on populations. It provides support to the idea that recruitment should be described as the outcome of multiple underlying density-dependent processes

(Rose et al. 2001).

The impact of climate variation is presented in Chapter 2-4 using growing-degree- days (GDD) as a proxy, but the impact of GDD on fish growth is not fully clear. Other fish studies on walleye (Venturelli et al. 2010) and rainbow trout in BC (Ward et al.

2017, Carruthers et al. 2018) have presented results linking high growth rates to low fish density and high GDD. This is also what I observed in Chapter 4, temperature variation between years showed higher growth rates on warmer years. But the results associated with climate in Chapter 2 and 3 are contradictory. In the empirical study of Chapter 3, lakes presenting higher GDD tend to have lower maximum size. It is important to repeat that the range in GDD observed in both the Chapter 3 and 4 studies are small, which explains why I do not feel confident in drawing conclusions from these observations. The

Chapter 2 simulations suggest that high GDD and abundant early life stage habitat would lead to high juvenile growth rates, but that in association with an adult stage bottleneck this high early growth rates would accelerate stunting resulting in a smaller observed maximum size. More research on the potential interaction between habitat bottlenecks, density and GDD is required to determine if this theoretical prediction is observed in nature.

153

The lake-scale habitat information gathered can be useful to management of both habitat and fisheries. Knowledge on the amount of habitat required to support a population can be used in habitat remediation projects or to predict the impact of habitat perturbation on populations. Information on habitat availability ratios can be valuable to fishery managers in recognizing populations with lower compensatory reserves that are more at risk from both the habitat perturbation stand point but also to overfishing.

Managers trying to provide a range of fishing opportunities can orient the management objective of populations based on habitat availability identifying populations that have the most potential for the development of a trophy fishery (yielding large fish) or a family fishery (priority to high catch rates) for example.

In Chapter 5, I provide more provincial-scale information on the distribution of wild rainbow trout populations and a finer spatial scale analysis of the North Thompson-

Clearwater watersheds for which a stream network connectivity methodology is developed to “fill the blanks” in rainbow trout distribution by providing probability of presence for lakes in which species composition is unknown. The results presented would benefit from more field validation, but the method developed presents promising opportunities. First, it could be used to predict the probability of native vs stocked origin.

As explained earlier in the thesis, the populations studied are naturally reproducing but not necessarily indigenous, they could have come from nearby stocked systems. Using a metapopulation approach, the probability of colonization from nearby historically stocked populations could be assessed. Stream network connectivity can also be informative in the context of invasion by non-native species, to assess proximity of native populations to known invasive observations, which can help identify populations more at risk of

154

invasion or prioritize the conservation of populations that have a higher probability of genetic purity in cases of invaders that can hybridize with native stocks. An understanding of how existing stream network structure has preserved native populations from competition or introgression can also support the development of novel management techniques such as the placement of barriers to limit contact between native populations and invasive genotypes or species.

The broad-scale approach of Chapter 5 provides more applications of the lake- scale information presented in Chapters 2-4 for the management of wild stocks at the landscape scale. Linking habitat characteristics to watershed position (e.g., Strahler order) allows for landscape-scale predictions of fish population characteristics which has obvious implications for management in facilitating the identification of areas containing populations more at risk of decline through habitat modifications (from industrial use for example) or overfishing. Information on areas more at risk of decline can help prioritize monitoring and conservation initiatives. This is also a first step in producing the information necessary to the inclusion of wild stocks in social-ecological models of recreational fisheries, such as the model developed by Carruthers et al. (2018) that only included stocked rainbow trout populations.

This broad-scale approach is also an opportunity to integrate fish population characteristics in the context of landscape (or riverscape) ecology. Although stream order is a very coarse predictor, it shows that, as it has been demonstrated for stream ecosystems (River Continuum Concept - Vannote et al. 1980) and limnological lake characteristics (Lake Landscape-Context framework - Soranno et al. 2009), fish populations vary following watershed organization. Better and more quantitative

155

landscape-level predictors of habitat characteristics should be identified, perhaps using remote sensing, to provide more accurate estimates of the availability of both habitat quantity and quality for fish populations at landscape scales. It would also be interesting to integrate the potential effects of climate change on stream habitats to predict how climatic variation may impact habitat availability and result in population dynamics changes.

My thesis integrates density-dependent processes across life stages, habitats, generations, climates and harvest pressures to describe how habitat availability influences populations dynamics at both the local and landscape scales. By building this novel framework that explicitly details the likely processes acting on populations, I learned the importance of the feedback loops operating across life stages for the prediction of population dynamics as many of the outcomes could not be predicted using a life stage or habitat in isolation. The local scale model predictions are validated using variation of life history parameters among populations using both an empirical and experimental approach to then be used to predict landscape-scale dynamics in species distribution and production. This landscape-scale modelling presents how habitat characteristics and connectivity are valuable in aquatic habitat and fisheries management. This integration across life stages, habitats and spatial scales is essential to accurately represent within and among population variation.

156

References

Andersen, K. H., N. S. Jacobsen, T. Jansen, and J. E. Beyer. 2017. When in life does

density dependence occur in fish populations? Fish and Fisheries 18:656–667.

Arlinghaus, R. 2006. On the apparently striking disconnect between motivation and

satisfaction in recreational fishing: The case of catch orientation of german anglers.

North American Journal of Fisheries Management 26:592–605.

Arranz, I., T. Mehner, L. Benejam, C. Argillier, K. Holmgren, E. Jeppesen, T. L.

Lauridsen, P. Volta, I. J. Winfield, and S. Brucet. 2016. Density-dependent effects

as key drivers of intraspecific size structure of six abundant fish species in lakes

across Europe. Canadian Journal of Fisheries and Aquatic Sciences 73:519–534.

Askey, P. J., E. A. Parkinson, and J. R. Post. 2013. Linking fish and angler dynamics to

assess stocking strategies for hatchery-dependent, open-access recreational fisheries.

North American Journal of Fisheries Management 33:557–568.

Askey, P. J., J. R. Post, E. A. Parkinson, E. Rivot, A. J. Paul, and P. A. Biro. 2007.

Estimation of gillnet efficiency and selectivity across multiple sampling units: A

hierarchical Bayesian analysis using mark-recapture data. Fisheries Research

83:162–174.

Askey, P. J., S. A. Richards, J. R. Post, and E. A. Parkinson. 2006. Linking angling catch

rates and fish learning under catch-and-release regulations. North American Journal

of Fisheries Management 26:1020–1029.

Askey, P. J., H. Ward, T. Godin, M. Boucher, and S. Northrup. 2018. Angler effort

estimates from instantaneous aerial counts: Use of high-frequency time-lapse

157

camera data to inform model-based estimators. North American Journal of Fisheries

Management 38:194–209.

Atkinson, D. 1994. Temperature and organism size-A biological law for ectotherms?

Advances in Ecological Research 25:1–58.

Ayllón, D., A. Almodóvar, G. G. Nicola, and B. Elvira. 2010. Ontogenetic and spatial

variations in brown trout habitat selection. Ecology of Freshwater Fish 19:420–432.

Barbour, M. T., J. Gerritsen, B. D. Snyder, and J. B. Stribling. 1999. Rapid bioassessment

protocols for use in streams and wadeable rivers: periphyton, benthic

macroinvertebrates and fish, Second Edition. Page US Environmental Protection

Agency Office. Washington, DC.

Bartmann, R. M., G. C. White, and L. H. Carpenter. 1992. Compensatory mortality in a

Colorado mule deer population. Wildlife Monographs 121:3–39.

Beardmore, B., W. Haider, L. M. Hunt, and R. Arlinghaus. 2011. The importance of trip

context for determining primary angler motivations: Are more specialized anglers

more catch-oriented than previously believed? North American Journal of Fisheries

Management 31:861–879.

Beck, M. W. 1995. Size-specific shelter limitation in stone crabs: A test of the

demographic bottleneck hypothesis. Ecology 76:968–980.

Beck, M. W., K. L. Heck, K. W. Able, D. L. Childers, D. B. Eggleston, B. M. Gillanders,

B. Halpern, C. G. Hays, K. Hoshino, T. J. Minello, R. J. Orth, P. F. Sheridan, and M.

P. Weinstein. 2001. The identification, conservation, and management of estuarine

and marine nurseries for fish and invertebrates. BioScience 51:633.

Biro, P. A., J. R. Post, and E. A. Parkinson. 2003a. Density-dependent mortality is

158

mediated by foraging activity for prey fish in whole-lake experiments. Journal of

Animal Ecology 72:546–555.

Biro, P. A., J. R. Post, and E. A. Parkinson. 2003b. From individuals to populations: Prey

fish risk-taking mediates mortality in whole-system experiments. Ecology 84:2419–

2431.

Bjornn, T., and D. Reiser. 1991. Habitat requirements of salmonids in streams. Pages 83–

138 Influences of forest and rangeland management on salmonid fishes and their

habitat. American Fisheries Society Special Publication 19.

Bond, M. H., S. A. Hayes, C. V. Hanson, and R. B. MacFarlane. 2008. Marine survival of

steelhead (Oncorhynchus mykiss) enhanced by a seasonally closed estuary.

Canadian Journal of Fisheries and Aquatic Sciences 65:2242–2252.

Borgstrørm, R., J. Heggenes, and T. G. Northcote. 1993. Regular, cyclic oscillations in

cohort strength in an allopatric population of brown trout. Salmo trutta L. Ecology of

Freshwater Fish 2:8–15.

British Columbia Ministry of Environment and Climate Change Strategy - Knowledge

Management. 2017a. EAUBC Lakes. https://catalogue.data.gov.bc.ca/dataset/eaubc-

lakes.

British Columbia Ministry of Environment and Climate Change Strategy - Knowledge

Management. 2017b. Known BC Fish Observations and BC Fish Distributions.

https://catalogue.data.gov.bc.ca/dataset/known-bc-fish-observations-and-bc-fish-

distributions.

British Columbia Ministry of Forests Lands Natural Operations and Rural Development.

2017. Fish stocking records for British Columbia. Unpublished raw data.

159

British Columbia Ministry of Forests Lands Natural Operations and Rural Development -

GeoBC. 2017a. Freshwater Atlas Lakes.

https://catalogue.data.gov.bc.ca/dataset/freshwater-atlas-lakes.

British Columbia Ministry of Forests Lands Natural Operations and Rural Development -

GeoBC. 2017b. Freshwater Atlas Wetlands.

https://catalogue.data.gov.bc.ca/dataset/freshwater-atlas-wetlands.

Burnham, K. P., and D. R. Anderson. 2002. Model selection and multimodel inference: a

practical information-theoretic approach. Second edition. Springer-V. New York,

USA.

Byström, P., and E. Garcia-Berthou. 1999. Density dependent growth and size specific

competitive interactions in young fish. Oikos 86:217.

Campbell Grant, E. H., W. H. Lowe, and W. F. Fagan. 2007. Living in the branches:

Population dynamics and ecological processes in dendritic networks. Ecology

Letters 10:165–175.

Carpenter, S. R., and W. A. Brock. 2004. Spatial complexity, resilience, and policy

diversity: Fishing on lake-rich landscapes. Ecology and Society 9:8.

Carruthers, T. R., K. Dabrowska, W. Haider, E. A. Parkinson, D. A. Varkey, H. Ward, M.

K. McAllister, T. Godin, B. Van Poorten, P. J. Askey, K. L. Wilson, L. M. Hunt, A.

Clarke, E. Newton, C. Walters, and J. R. Post. 2018. Landscape scale social and

ecological outcomes of dynamic angler and fish behaviours: processes, data, and

patterns. Canadian Journal of Fisheries and Aquatic Sciences.

Catalano, M. J., and M. S. Allen. 2011. A whole-lake density reduction to assess

compensatory responses of gizzard shad Dorosoma cepedianum. Canadian Journal

160

of Fisheries and Aquatic Sciences 68:955–968.

Cattanéo, F., N. Lamouroux, P. Breil, and H. Capra. 2002. The influence of hydrological

and biotic processes on brown trout (Salmo trutta) population dynamics. Canadian

Journal of Fisheries and Aquatic Sciences 59:12–22.

Claessen, D., A. M. de Roos, and L. Persson. 2000. Dwarfs and giants: Cannibalism and

competition in size‐structured populations. The American Naturalist 155:219–237.

Coleman, F. C., W. F. Figueira, J. S. Ueland, and L. B. Crowder. 2004. The impact of

United States recreational fisheries on marine fish populations. Science 305:1958–

1960.

Cooke, S. J., and I. G. Cowx. 2004. The role of recreational fishing in global fish crises.

Bioscience 54:857–859.

Cooke, S. J., and I. G. Cowx. 2006. Contrasting recreational and commercial fishing:

Searching for common issues to promote unified conservation of fisheries resources

and aquatic environments. Biological Conservation 128:93–108.

Coombes, D. M. V. 1991. A reconnaissance survey of Hobson Lake. British Columbia,

Canada.

Cowan, J. H., K. A. Rose, and D. R. DeVries. 2000. Is density-dependent growth in

young-of-the-year fishes a question of critical weight? Reviews in Fish Biology and

Fisheries 10:61–89.

Cox, S. P. 2000. Angling quality, effort response, and exploitation in recreational

fisheries: field and modeling studies on British Columbia rainbow trout

(Oncorhynchus mykiss) lakes. University of British Columbia, Vancouver, British

Columbia.

161

Cox, S. P., and C. Walters. 2002. Modeling Exploitation in Recreational Fisheries and

Implications for Effort Management on British Columbia Rainbow Trout Lakes.

North American Journal of Fisheries Management 22:21–34.

Crisp, D. T. 1993. Population densities of juvenile trout (Salmo trutta) in five upland

streams and their effects upon growth, survival and dispersal. The Journal of

Applied Ecology 30:759.

Dabrowska, K., W. Haider, and L. Hunt. 2014. Examining the impact of fisheries

resources and quality on licence sales. Journal of Outdoor Recreation and Tourism

5–6:58–67.

DeFilippo, L. B., D. E. Schindler, J. L. Carter, T. E. Walsworth, T. J. Cline, W. A.

Larson, and T. Buehrens. 2018. Associations of stream geomorphic conditions and

prevalence of alternative reproductive tactics among sockeye salmon populations.

Journal of Evolutionary Biology 31:239–253.

Denwood, M. J. 2016. runjags : An R package providing interface utilities, model

templates, parallel computing methods and additional distributions for MCMC

models in JAGS. Journal of Statistical Software 71.

Dill, L. M., R. C. Ydenberg, and A. H. G. Fraser. 1981. Food abundance and territory

size in juvenile coho salmon (Oncorhynchus kisutch). Canadian Journal of

Fisheries and Aquatic Sciences 59:1801–1809.

Ebenman, B. 1988. Dynamics of age-and size-structured populations: intraspecific

competition. Pages 127–139 in B. Ebenman and Persson, editors. Size-structured

populations. Springer-Verlag, Berlin, Germany.

Ebenman, B. 1992. Evolution in organisms that change their niches during the life cycle.

162

The American Naturalist 139:990–1021.

Elliott, J. M. 1985. Population regulation for different life-stages of migratory trout

Salmo trutta in a lake district stream, 1966-83. Journal of Animal Ecology 54:617–

638.

Elliott, J. M. 1994. Quantitative ecology and the brown trout. Oxford University Press,

Oxford, UK.

Elliott, J. M., and M. A. Hurley. 1998. Population regulation in adult, but not juvenile,

resident trout (Salmo trutta) in a lake district stream. Animal Ecology 67:280–286.

Erman, D. C., and G. R. Leidy. 1975. Downstream movement of rainbow trout fry in a

tributary Sagehen Creek, under permanent and intermittent flow. Transactions of the

American Fisheries Society 104:467–473.

ESRI. 2011. ArcGIS Desktop: Release 10. Environmental Systems Research Institute,

Redlands, California.

Essington, T. E. W., P. W. Sorensen, and D. G. Paron. 1998. High rate of redd

superimposition by brook trout (Salvelinus fontinalis) and brown trout (Salmo trutta)

in a Minnesota stream cannot be explained by habitat availability alone. Canadian

Journal of Fisheries and Aquatic Sciences 55:2310–2316.

Fenberg, P. B., K. Roy, S. Diego, and L. Jolla. 2008. Ecological and evolutionary

consequences of size-selective harvesting: How much do we know? Molecular

Ecology 17:209–220.

Fenichel, E. P., J. K. Abbott, and B. Huang. 2013. Modelling angler behaviour as a part

of the management system: Synthesizing a multi-disciplinary literature. Fish and

Fisheries 14:137–157.

163

Fleming, I. A. 1998. Pattern and variability in the breeding system of Atlantic salmon

(Salmo salar), with comparisons to other salmonids. Canadian Journal of Fisheries

and Aquatic Sciences 55:59–76.

Fransen, B. R., S. D. Duke, L. G. McWethy, J. K. Walter, and R. E. Bilby. 2006. A

logistic regression model for predicting the upstream extent of fish occurrence based

on geographical information systems data. North American Journal of Fisheries

Management 26:960–975.

Freshwater Fisheries Societry of British Columbia. 2017. Small Lakes Database - PARIS

RELEASES. Unpublished raw data.

Gelman, A., J. B. Carlin, H. S. Stern, D. B. Dunson, A. Vehtari, and D. B. Rubin. 2013.

Bayesian Data Analysis, 3rd edition. Chapman and Hall/CRC Press.

De Gisi, J. S. 1994. Year class strength and catchability of mountain lake brook trout.

University of British Columbia.

Goodyear, C. P. 1993. Spawning stock biomass per recruit in fisheries management:

foundation and current use. Pages 67–82 in S. J. Smith, J. J. Hunt, and D. Rivard,

editors. Risk evaluation and biological reference points for fisheries management.

Canadian Special Publication of Fisheries and Aquatic Sciences.

Goward, T., and C. Hickson. 1995. Fish. Pages 20–22 Nature Wells Gray, The

Clearwater Valley. The Friends of Wells Gray Park, Kamloops, British Columbia,

Canada.

Grant, J., E. Kvingedal, and S. Einum. 2011. Intracohort and intercohort spatial density

dependence in juvenile brown trout (Salmo trutta). Canadian Journal of Fisheries

and Aquatic Sciences 68:115–121.

164

Grant, J. W. A., and I. Imre. 2005. Patterns of density-dependent growth in juvenile

stream-dwelling salmonids. Journal of Fish Biology 67:100–110.

Grant, J. W. A., and D. L. Kramer. 1990. Territory size as a predictor of the upper limit to

population density of juvenile salmonids in streams. Canadian Journal of Fisheries

and Aquatic Sciences 47:1724–1737.

Gray, M. 2010. Freshwater Water Atlas User Guide, GeoBC Integrated Land

Management Bureau. Victoria, British Columbia.

Greenberg, S., and T. Godin. 2015. A tool supporting the extraction of angling effort data

from remote camera images. Fisheries 40:276–287.

Hagberg, A. A., D. A. Schult, and P. J. Swart. 2008. Exploring network structure,

dynamics, and function using NetworkX. Pages 11–15 in G. Varoquaux, T. Vaught,

and J. Millman, editors. Proceedings of the 7th Python in Science Conference

(SciPy2008). Pasadena, CA, USA.

Halpern, B. S. 2004. Habitat bottlenecks in stage-structured species: hermit crabs as a

model system. Marine Ecology-Progress Series 276:197–207.

Hammond, M., S. da Silva, J. R. Coombs, V. Cole, L. Immisch, R. Upole, T. Heller, T.

Golden, and T. Mick. 2008. Python for Windows Extensions.

Hamrin, S. F., and L. Persson. 1986. Asymmetrical competition between age classes as a

factor causing population oscillations in an obligate planktivorous fish species.

Oikos 47:223–232.

Hansen, M. J., T. D. Beard, and S. W. Hewett. 2000. Catch rates and catchability of

walleyes in angling and spearing fisheries in Northern Wisconsin lakes. North

American Journal of Fisheries Management 20:109–118.

165

Hayes, D. B., J. R. Bence, T. J. Kwak, and B. E. Thompson. 2007. Abundance, biomass,

and production. Pages 327–374 in C. S. Guy and M. L. Brown, editors. Analysis and

interpretation of freshwater fisheries data. American Fisheries Society, Bethesda,

Maryland, USA.

Hayes, D., M. Jones, N. Lester, C. Chu, S. Doka, J. Netto, J. Stockwell, B. Thompson, C.

K. Minns, B. Shuter, and N. Collins. 2009. Linking fish population dynamics to

habitat conditions: Insights from the application of a process-oriented approach to

several Great Lakes species. Reviews in Fish Biology and Fisheries 19:295–312.

Hayes, J. W. 1987. Competition for spawning space between brown (Salmo trutta) and

rainbow trout (S. gairdneri) in a lake inlet tributary, New Zealand. Canadian

Journal of Fisheries and Aquatic Sciences 44:40–47.

Hayes, J. W. 1995. Importance of stream versus early lake rearing for rainbow trout fry in

Lake Alexandrina, South Island, New Zealand, determined from otolith daily growth

patterns. New Zealand Journal of Marine and Freshwater Research 29:409–420.

Healey, M. C. 1980. Growth and recruitment in experimentallly exploited lake whitefish

(Coregonus clupeaformis) populations. Canadian Journal of Fisheries and Aquatic

Sciences 37:255–267.

Helminen, H., and J. Sarvala. 1994. Population regulation of vendace (Coregonus albula)

in Lake Pyhäjärvi, southwest Finland. Journal of Fish Biology 45:387–400.

Helser, T. E., and H. L. Lai. 2004. A Bayesian hierarchical meta-analysis of fish growth:

With an example for North American largemouth bass, Micropterus salmoides.

Ecological Modelling 178:399–416.

Hilborn, R., T. A. Branch, B. Ernst, A. Magnusson, C. V Minte-Vera, M. D. Scheuerell,

166

and J. L. Valero. 2003. State of the world’s fisheries. Annual Review of Environment

and Resources 28:359–399.

Houde, E. D. 1989. Subtleties and episodes in the early life of fishes. Journal of Fish

Biology 35:29–38.

Hughes, R. M., P. R. Kaufmann, and M. H. Weber. 2011. National and regional

comparisons between Strahler order and stream size. Journal of the North American

Benthological Society 30:103–121.

Hunt, L. M., R. Arlinghaus, N. Lester, and R. Kushneriuk. 2011. The effects of regional

angling effort, angler behavior, and harvesting efficiency on landscape patterns of

overfishing. Ecological Applications 21:2555–2575.

Hutchings, J. A., and M. E. Jones. 1998. Life history variation and growth rate thresholds

for maturity in Atlantic salmon, Salmo salar. Canadian Journal of Fisheries and

Aquatic Sciences 55:22–47.

Imre, I., J. W. A. Grant, and E. R. Keeley. 2004. The effect of food abundance on

territory size and population density of juvenile steelhead trout (Oncorhynchus

mykiss). Oecologia 138:371–378.

Isermann, D. A., and C. T. Knight. 2005. A computer program for age–length keys

incorporating age assignment to individual fish. North American Journal of

Fisheries Management 25:1153–1160.

Jackson, D. A., P. R. Peres-Neto, and J. D. Olden. 2001. What controls who is where in

freshwater fish communities – the roles of biotic, abiotic, and spatial factors.

Canadian Journal of Fisheries and Aquatic Sciences 58:157–170.

Jenkins, T. M., S. Diehl, K. W. Kratz, and S. D. Cooper. 1999. Effects of population

167

density on individual growth of brown trout in streams. Ecology 80:941–956.

Johnson, B. M., and S. R. Carpenter. 1994. Functional and numerical responses: A

framework for fish-angler interactions? Ecological Applications 4:808–821.

Johnston, F. D., R. Arlinghaus, and U. Dieckmann. 2010. Diversity and complexity of

angler behaviour drive socially optimal input and output regulations in a

bioeconomic recreational-fisheries model. Canadian Journal of Fisheries and

Aquatic Sciences 67:1897–1898.

Johnston, F. D., R. Arlinghaus, and U. Dieckmann. 2013. Fish life history, angler

behaviour and optimal management of recreational fisheries. Fish and Fisheries

14:554–579.

Johnston, F. D., J. R. Post, and B. Sciences. 2009. Density-dependent life-history

compensation of an iteroparous salmonid. Ecological Applications 19:449–467.

Keeley, E. R. 2000. An experimental analysis of territory size in juvenile steelhead trout.

Animal Behaviour 59:477–490.

Keeley, E. R., and J. D. McPhail. 1998. Food abundance, intruder pressure, and body size

as determinants of territory size in juvenile steelhead trout (Oncorhynchus mykiss).

Behaviour 135:65–82.

Keeley, E. R., and P. A. Slaney. 1996. Quantitative measures of rearing and spawning

habitat characteristics for stream-dwelling salmonids: Guidelines for habitat

restoration. Page Watershed Restoration Project No. 4, Ministry of Environment,

Lands and Parks, Vancouver, Canada.

Kéry, M. 2002. Inferring the absence of a species: a case study of snakes. The Journal of

Wildlife Management 66:330–338.

168

Koops, M. A., J. A. Hutchings, and T. M. McIntyre. 2004. Testing hypotheses about

fecundity, body size and maternal condition in fishes. Fish and Fisheries 5:120–130.

Kratz, T. K., K. E. Webster, C. J. Bowser, J. J. Magnuson, and B. J. Benson. 1997. The

influence of landscape position on lakes in northern Wisconsin lakes. Freshwater

Biology 37:209–217.

Kruse, C. G., W. A. Hubert, and F. J. Rahel. 1997. Geomorphic influences on the

distribution of Yellowstone cutthroat trout in the Absaroka Mountains, Wyoming.

Transactions of the American Fisheries Society 126:418–427.

Kuhn, M., J. Wing, S. Weston, A. Williams, C. Keefer, A. Engelhardt, T. Cooper, Z.

Mayer, B. Kenkel, R. C. T. R. Null, M. Benesty, R. Lescarbeau, A. Ziem, L.

Scrucca, Y. Tang, C. Candan, and T. Hunt. 2018. caret: Classification and regression

training. R package version 6.0-80.

Landis, J. R., and G. G. Koch. 1977. The measurement of observer agreement for

categorical data. Biometrics 33:159.

Landry, F., J. R. Post, and E. A. Parkinson. 1999. Spatial ontogeny of lentic age-0

rainbow trout, Oncorhynchus mykiss :whole-lake manipulations of population size

structure. Canadian Journal of Fisheries and Aquatic Sciences 56:1916–1928.

Latterell, J. J., R. J. Naiman, B. R. Fransen, and P. A. Bisson. 2003. Physical constraints

on trout (Oncorhynchus spp.) distribution in the Cascade Mountains: a comparison

of logged and unlogged streams. Canadian Journal of Fisheries and Aquatic

Sciences 60:1007–1017.

Legendre, P., and L. Legendre. 2012. Numerical ecology 3rd Edition. Page Journal of

Chemical Information and Modeling. Elsevier, Amsterdam, NL.

169

Lester, N. P., B. J. Shuter, and P. A. Abrams. 2004. Interpreting the von Bertalanffy

model of somatic growth in fishes: The cost of reproduction. Proceedings of the

Royal Society B: Biological Sciences 271:1625–1631.

Lester, N. P., B. J. Shuter, P. Venturelli, and D. Nadeau. 2014. Life-history plasticity and

sustainable exploitation: A theory of growth compensation applied to walleye

management. Ecological Applications 24:38–54.

Liley, N. R., P. Tamkee, R. Tsai, and D. J. Hoysak. 2002. Fertilization dynamics in

rainbow trout (Oncorhynchus mykiss): effect of male age, social experience, and

sperm concentration and motility on in vitro fertilization. Canadian Journal of

Fisheries and Aquatic Sciences 59:144–152.

Lobón-Cerviá, J. 2005. Spatial and temporal variation in the influence of density

dependence on growth of stream-living brown trout (Salmo trutta). Canadian

Journal of Fisheries and Aquatic Sciences 62:1231–1242.

Lobón-Cerviá, J. 2007. Density-dependent growth in stream-living Brown Trout Salmo

trutta L. Functional Ecology 21:117–124.

Lobón-Cerviá, J., and P. A. Rincón. 2004. Environmental determinants of recruitment

and their influence on the population dynamics of stream-living brown trout Salmo

trutta. Oikos 105:641–646.

Lorenzen, K., and K. Enberg. 2002. Density-dependent growth as a key mechanism in the

regulation of fish populations: evidence from among-population comparisons.

Proceedings of the Royal Society B: Biological Sciences 269:49–54.

Mackenzie, D. I., J. D. Nichols, G. B. Lachman, S. Droege, J. Andrew, and C. a

Langtimm. 2002. Estimating site occupancy rates when detection probabilities are

170

less than one. Ecology 83:2248–2255.

Magnuson, J. J., L. B. Crowder, and P. A. Medvick. 1979. Temperature as an ecological

resource. Integrative and Comparative Biology 19:331–343.

Mangel, M., and J. Stamps. 2001. Trade-offs between growth and mortality and the

maintenance of individual variation in growth. Evolutionary Ecology Research

3:583–593.

Martin, S. L., and P. A. Soranno. 2006. Lake landscape position: Relationships to

hydrologic connectivity and landscape features. Limnology and Oceanography

51:801–814.

McArdle, B. H. 1990. When are rare species not there? Oikos 57:276.

McKay, S. K., J. R. Schramski, J. N. Conyngham, and J. Craig Fischenich. 2013.

Assessing upstream fish passage connectivity with network analysis. Ecological

Applications 23:1396–1409.

McNicol, R. E., and D. L. G. Noakes. 1984. Environmental influences on territoriality of

juvenile brook charr, Salvelinus fontinalis, in a stream environment. Environmental

Biology of Fishes 10:29–42.

McPhail, J. D. 2007. The freshwater fishes of British Columbia. The University of

Alberta Press, Edmonton, AB, Canada.

McPhail, J. D., and R. Carveth. 1992. A foundation for conservation: The nature and

origin of the freshwater fish fauna of British Columbia. Page UBC Fish Museum

Report. Vancouver.

McPhail, J. D., and C. C. Lindsey. 1986. Zoogeography of the freshwater fishes of

Cascadia (the Columbia system and rivers north to the Stikine). Pages 615–637 in C.

171

H. Hocutt and E. O. Wiley, editors. The zoogeography of North American

freshwater fishes. Wiley, New York, USA.

McPhee, D. P., D. Leadbitter, and G. A. Skilleter. 2002. Swallowing the bait: Is

recreational fishing in Australia ecologically sustainable? Pacific Conservation

Biology 8:40–51.

Mee, J. A., J. R. Post, H. Ward, K. L. Wilson, E. Newton, and A. Cantin. 2016.

Interaction of ecological and angler processes: Experimental stocking in an open

access, spatially structured fishery. Ecological Applications 26:1693–1707.

Mehner, T., K. Holmgren, T. L. Lauridsen, E. Jeppesen, and M. Diekmann. 2007. Lake

depth and geographical position modify lake fish assemblages of the European

“Central Plains” ecoregion. Freshwater Biology 52:2285–2297.

Milner, N. J., J. M. Elliott, J. D. Armstrong, R. Gardiner, J. S. Welton, and M. Ladle.

2003. The natural control of salmon and trout populations in streams. Fisheries

Research 62:111–125.

Mount, C., S. Norris, R. Thompson, and D. Tesch. 2011. GIS modelling of fish habitat

and road crossings for the prioritization of culvert assessment and remediation.

Streamline Watershed Management Bulletin 14:7–13.

Mumby, P. J., A. J. Edwards, J. E. Arias-Gonzalez, K. C. Lindeman, P. G. Blackwell, A.

Gall, M. I. Gorczynska, A. R. Harborne, C. L. Pescod, H. Renken, C. C. C. Wabnitz,

and G. Llewellyn. 2004. Mangroves enhance the biomass of coral reef fish

communities in the Caribbean. Nature 427:533–536.

Murray, C. B. 1980. Some effects of temperature on zygote and alevin survival, rate of

development and size at hatching and emergence of Pacific salmon and rainbow

172

trout. Master of Science Thesis, University of British Columbia, Vancouver, British

Columbia.

Myers, R. A., and N. J. Barrowman. 1996. Is fish recruitment related to spawner

abundance? Fishery Bulletin 94:707–724.

Myers, R. A., K. G. Bowen, and N. J. Barrowman. 1999. Maximum reproductive rate of

fish at low population sizes. Canadian Journal of Fisheries and Aquatic Sciences

56:2404–2419.

Nelson, R. L., W. S. Platts, D. P. Larsen, and S. E. Jensen. 1992. Trout distribution and

habitat in relation to geology and geomorphology in the North Fork Humboldt River

Drainage, Northeastern Nevada. Transactions of the American Fisheries Society

121:405–426.

Neuheimer, A. B., and C. T. Taggart. 2007. The growing degree-day and fish size-at-age:

the overlooked metric. Canadian Journal of Fisheries and Aquatic Sciences 64:375–

385.

Nevada, S., M. Lakes, S. Barbara, P. Southwest, R. A. Knapp, V. T. Vredenburg, and K.

R. Matthews. 1998. Effects of stream channel morphology on golden trout spawning

habitat and recruitment. Ecological Applications 8:1104–1117.

Nordwall, F., I. Näslund, and E. Degerman. 2001. Intercohort competition effects on

survival, movement, and growth of brown trout (Salmo trutta) in Swedish streams.

Canadian Journal of Fisheries and Aquatic Sciences 58:2298–2308.

Northcote, T. G. 1962. Migratory behaviour of juvenile rainbow trout, Salmo gairdneri,

in outlet and inlet streams of Loon Lake, British Columbia. Journal of the Fisheries

Research Board of Canada 19:201–270.

173

Oksanen, J., F. G. Blanchet, M. Friendly, R. Kindt, P. Legendre, D. McGlinn, P. R.

Minchin, R. B. O’Hara, G. L. Simpson, P. Solymos, M. H. H. Stevens, E. Szoecs,

and H. Wagner. 2017. vegan: Community Ecology Package.

Parkinson, E. A., J. R. Post, and S. P. Cox. 2004. Linking the dynamics of harvest effort

to recruitment dynamics in a multistock, spatially structured fishery. Canadian

Journal of Fisheries and Aquatic Sciences 61:1658–1670.

Parks Canada. 2018. Fish stocking records for the mountain parks. Unpublished raw data.

Persson, L., A. Van Leeuwen, and A. M. de Roos. 2014. The ecological foundation for

ecosystem-based management of fisheries: Mechanistic linkages between the

individual-, population-, and community-level dynamics. ICES Journal of Marine

Science 71:2268–2280.

Persson, L., K. Leonardsson, A. M. de Roos, B. Christensen, A. M. De Roos, M.

Gyllenberg, and B. Christensen. 1998. Ontogenetic scaling of foraging rates and the

dynamics of a size-structured consumer-resource model. Theoretical Population

Biology 54:270–293.

Plummer, M. 2003. JAGS: A program for analysis of Bayesian graphical models using

Gibbs sampling. Page 125 Proceedings of the 3rd International Workshop on

Distributed Statistical Computing. van Poorten, B. T., T. R. Carruthers, H. G. M. Ward, and D. A. Varkey. 2015. Imputing

recreational angling effort from time-lapse cameras using an hierarchical Bayesian

model. Fisheries Research 172:265–273.

Porter, M. S., J. Rosenfeld, and E. A. Parkinson. 2000. Predictive models of fish species

distribution in the Blackwater Drainage, British Columbia. North American Journal

174

of Fisheries Management 20:349–359.

Post, J. R., and D. O. Evans. 1989a. Size-dependent 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.

Post, J. R., and D. O. Evans. 1989b. Experimental evidence of size-dependent predation

mortality in juvenile yellow perch. Canadian Journal of Zoology 67:521–523.

Post, J. R., and E. A. Parkinson. 2012. Temporal and spatial patterns of angler effort

across lake districts and policy options to sustain recreational fisheries. Canadian

Journal of Fisheries and Aquatic Sciences 69:321–329.

Post, J. R., E. A. Parkinson, and N. T. Johnston. 1999. Density-dependent processes in

structured fish populations: Interaction strengths in whole-lake experiments.

Ecological Monographs 69:155–175.

Post, J. R., L. Persson, E. A. Parkinson, and T. Van Kooten. 2008. Angler numerical

response across landscapes and the collapse of freshwater fisheries. Ecological

Applications 18:1038–1049.

Post, J. R., M. Sullivan, S. Cox, N. P. Lester, C. J. Walters, E. A. Parkinson, A. J. Paul,

L. Jackson, and B. J. Shuter. 2002. Canada’s recreational fisheries: The invisible

collapse? Fisheries 27:6–17.

Quinlan, R., A. M. Paterson, R. I. Hall, P. J. Dillon, A. N. Wilkinson, B. F. Cumming, M.

S. V. Douglas, and J. P. Smol. 2003. A landscape approach to examining spatial

patterns of limnological variables and long-term environmental change in a southern

Canadian lake district. Freshwater Biology 48:1676–1697.

Quist, M. C., F. J. Rahel, and W. A. Hubert. 2005. Hierarchical faunal filters: An

175

approach to assessing effects of habitat and nonnative species on native fishes.

Ecology of Freshwater Fish 14:24–39.

R Core Team. 2017. R: A language and environment for statistical computing. R

Foundation for Statistical Computing, Vienna, Austria.

Rahel, F. J. 2007. Biogeographic barriers, connectivity and homogenization of freshwater

faunas: It’s a small world after all. Freshwater Biology 52:696–710.

Raleigh, R. F., T. Hickman, and R. C. Solomon. 1984. Habitat suitability information:

Rainbow trout. Fish and Wildlife Service. FWS/OBS-82/10.60. Page U. S.

Department of Interior, Fish and Wildlife Service.

Riera, J. L., J. J. Magnuson, T. K. Kratz, and K. E. Webster. 2000. A geomorphic

template for the analysis of lake districts applied to the Northern Highland Lake

District, Wisconsin, U.S.A. Freshwater Biology 43:301–318.

Roff, D. 2002. Life history evolution. Sinauer Associates, Inc., Sunderland,

Massachusetts, USA.

Rollinson, N., J. A. Hutchings, and N. Rollinson. 2010. Why does egg size increase with

maternal size? Effects of egg size and egg density on offspring phenotypes in

Atlantic salmon (Salmo salar). Evolutionary Ecology Research 12:949–960. de Roos, A. M., K. Leonardsson, L. Persson, and G. G. Mittelbach. 2002. Ontogenetic

niche shifts and flexible behavior in size-structured populations. Ecological

Monographs 72:271–292. de Roos, A. M., and L. Persson. 2013. Population and community ecology of ontogenetic

development. Page Monographs in population biology. Princeton University Press. de Roos, A. M., L. Persson, A. M. De Roos, and L. Persson. 2003. Competition in size-

176

structured populations: Mechanisms inducing cohort formation and population

cycles. Theoretical Population Biology 63:1–16.

Rose, K. A. 2005. Lack of relationship between simulated fish population responses and

their life history traits: inadequate models, incorrect analysis, or site-specific

factors? Canadian Journal of Fisheries and Aquatic Sciences 62:886–902.

Rose, K. A., and J. H. Cowan. 2000. Predicting fish population dynamics: compensation

and the importance of site-specific considerations. Environmental Science & Policy

3:433–443.

Rose, K. A., J. H. Cowan, K. O. Winemiller, R. A. Myers, and R. Hilborn. 2001.

Compensatory density dependence in fish populations: importance, controversy,

understanding and prognosis. Fish and Fisheries 2:293–327.

Rosenau, M. L. 1991. Natal‐stream rearing in three populations of rainbow trout in Lake

Taupo, New Zealand. New Zealand Journal of Marine and Freshwater Research

25:81–91.

Ryder, R. A., S. R. Kerr, K. H. Loftus, and H. A. Regier. 1974. The morphoedaphic

index, a fish yield estimator-review and evaluation. Journal Fisheries Research

Board of Canada 31:663–688.

Sauer, J. R., and N. A. Slade. 1987. Size-based demography of vertebrates. Annual

Review of Ecology, Evolution, and Systematics:71–90.

Schick, R. S., and S. T. Lindley. 2007. Directed connectivity among fish populations in a

riverine network. Journal of Applied Ecology 44:1116–1126.

Schindler, D. W. 1998. Replication versus realism: The need for ecosystem-scale

experiments. Ecosystems 1:323–334.

177

Scott, W. B., and E. J. Crossman. 1973. Freshwater fishes of Canada. Page Bulletin of the

Fisheries Research Board of Canada. Bulletin of the Fisheries Research Board of

Canada. No. 184, Ottawa, Canada.

Shuter, B. J., M. L. Jones, R. M. Korver, and N. P. Lester. 1998. A general, life history

based model for regional management of fish stocks: the inland lake trout

(Salvelinus namaycush) fisheries of Ontario. Canadian Journal of Fisheries and

Aquatic Sciences 55:2161–2177.

Sinclair, A. R. E. 1989. Population regulation in animals. Pages 197–241 in J. M.

Cherrett, editor. Ecological Concepts: The Contribution of Ecology to an

Understanding of the Natural World. The 29th Symposium of the British Ecological

Society, 12–13 April 1988, London. Blackwell Scientific Publications, Oxford, UK.

Slaney, P. A., and T. G. Northcote. 1974. Effects of prey abundance on density and

territorial behavior of young Rainbow Trout (Salmo gairdneri) in laboratory stream

channels. Journal of the Fisheries Research Board of Canada 31:1201–1209.

Sogard, S. M. 1997. Size-selective mortality in the juvenile stage of teleost fishes: A

review. Bulletin of Marine Science 60:1129–1157.

Soranno, P. A., K. S. Cheruvelil, K. E. Webster, M. T. Bremigan, T. Wagner, and C. A.

Stow. 2010. Using landscape limnology to classify freshwater ecosystems for multi-

ecosystem management and conservation. BioScience 60:440–454.

Soranno, P. A., K. E. Webster, K. S. Cheruvelil, and M. T. Bremigan. 2009. The lake

landscape-context framework: linking aquatic connections, terrestrial features and

human effects at multiple spatial scales. Verh. Internat. Verein. Limnol. 30:695–700.

Soranno, P. A., K. E. Webster, J. L. Riera, T. K. Kratz, J. S. Baron, P. A. Bukaveckas, G.

178

W. Kling, D. S. White, N. Caine, R. C. Lathrop, and P. R. Leavitt. 1999. Spatial

variation among lakes within landscapes: Ecological organization along lake chains.

Ecosystems 2:395–410.

Stearns, S. C. 1992. The evolution of life histories. Oxford University Press, Oxford, UK.

Strahler, A. N. 1957. Quantitative analysis of watershed geomorphology. Transactions,

American Geophysical Union 38:913.

Sundblad, G., U. Bergström, A. Sandström, and P. Eklöv. 2014. Nursery habitat

availability limits adult stock sizes of predatory coastal fish. ICES Journal of Marine

Science 71:672–680.

Tonn, W. M. 1990. Climate change and fish communities: A conceptual framework.

Transactions of the American Fisheries Society 119:337–352.

Tonn, W. M., and C. A. Paszkowski. 1986. Size-limited predation, winterkill, and the

organization of umbra–perca fish assemblages. Canadian Journal of Fisheries and

Aquatic Sciences 43:194–202.

Urban, D., and T. Keitt. 2001. Landscape connectivity: a graph-theoretic perspective.

Ecology 85:1205–1218.

US Fish and Wildlife Service. 1981. Standards for the development of habitat suitability

index models. Washington, D.C.

Utrilla, C. G., and J. Lobon-Cervia. 1999. Life-history patterns in a southern population

of Atlantic salmon. Journal of Fish Biology 55:68–83.

Vannote, R. L., G. W. Minshall, K. W. Cummins, J. R. Sedell, and C. E. Cushing. 1980.

The River Continuum Concept. Canadian Journal of Fisheries and Aquatic Sciences

37:130–137.

179

Varkey, D. A., M. K. McAllister, P. J. Askey, E. Parkinson, A. Clarke, and T. Godin.

2016. Multi-criteria decision analysis for recreational trout fisheries in British

Columbia, Canada: A Bayesian network implementation. North American Journal

of Fisheries Management 36:1457–1472.

Varkey, D. A., B. Van Poorten, M. K. Mcallister, T. Godin, E. Parkinson, and R. Kumar.

2018. Describing growth based on landscape characteristics and stocking strategies

for rainbow trout. Fisheries Research 208:229–238.

Venturelli, P. A., N. P. Lester, T. R. Marshall, and B. J. Shuter. 2010. Consistent patterns

of maturity and density-dependent growth among populations of walleye (Sander

vitreus): application of the growing degree-day metric. Canadian Journal of

Fisheries and Aquatic Sciences 67:1057–1067.

Wahle, R., and R. Steneck. 1991. Recruitment habitats and nursery grounds of the

American lobster Homarus Americanus: A demographic bottleneck? Marine

Ecology-Progress Series 69:231–243.

Walters, C. J., and F. Juanes. 1993. Recruitment limitation as a consequence of natural

selection for use of restricted feeding habitats and predation risk taking by juvenile

fishes. Canadian Journal of Fisheries and Aquatic Sciences 50:2058–2070.

Walters, C. J., and S. J. D. Martell. 2004. Fisheries ecology and management. Princeton

University Press, Princeton, NJ, USA.

Walters, C. J., and J. R. Post. 1993. Density-dependent growth and competitive

asymmetries in size-structured fish populations: a theoretical model and

recommendations for field experiments. Transactions of the American Fisheries

Society 122:34–45.

180

Wang, T., A. Hamann, D. Spittlehouse, and C. Carroll. 2016. Locally downscaled and

spatially customizable climate data for historical and future periods for North

America. PLoS ONE 11:1–17.

Wang, T., A. Hamann, D. L. Spittlehouse, and T. Q. Murdock. 2012. ClimateWNA-high-

resolution spatial climate data for western North America. Journal of Applied

Meteorology and Climatology 51:16–29.

Ward, B. R., and P. A. Slaney. 1993. Egg-to-smolt survival and fry-to-smolt density

dependence of Keogh River steelhead trout. Pages 209–217 in R. J. Gibson and R.

E. Cutting, editors. Production of juvenile Atlantic salmon, Salmo salar, in natural

waters. Canadian Special Publication Fisheries and Aquatic Science.

Ward, H. G. M., P. J. Ashley, and J. R. Post. 2013a. A mechanistic understanding of

hyperstability in catch per unit effort and density-dependent catchability in a

multistock recreational fishery. Canadian Journal of Fisheries and Aquatic Sciences

70:1858.

Ward, H. G. M., P. J. Askey, J. R. Post, D. A. Varkey, and M. K. McAllister. 2012. Basin

characteristics and temperature improve abundance estimates from standard index

netting of rainbow trout (Oncorhynchus mykiss) in small lakes. Fisheries Research

131–133:52–59.

Ward, H. G. M. M., J. R. Post, N. P. Lester, P. J. Askey, and T. Godin. 2017. Empirical

evidence of plasticity in life-history characteristics across climatic and fish density

gradients. Canadian Journal of Fisheries and Aquatic Sciences 74:464–474.

Ward, H. G. M., M. S. Quinn, and J. R. Post. 2013b. Angler characteristics and

management implications in a large, multistock, spatially structured recreational

181

fishery. North American Journal of Fisheries Management 33:576–584.

Werner, E. E. 1988. Size, scaling, and the evolution of complex life cycles. Pages 60–81

in B. Ebenman and L. Persson, editors. Size-structured populations. Springer-

Verlag, Berlin, Germany.

Werner, E. E., and J. F. Gilliam. 1984. The ontogenetic niche and species interactions in

size-structured populations. Annual Review of Ecology and Systematics 15:393–425.

Wetzel, R. G. 2001. Limnology: Lake and river ecosystems, 3rd edition. Academic Press,

New York, USA.

Wiens, J. A. 2002. Riverine landscapes: Taking landscape ecology into the water.

Freshwater Biology 47:501–515.

Wilson, K. L., A. Cantin, H. G. M. Ward, E. R. Newton, J. A. Mee, D. A. Varkey, E. A.

Parkinson, and J. R. Post. 2016. Supply – demand equilibria and the size – number

trade-off in spatially structured recreational fisheries. Ecological Applications

26:1086–1097.

Van Winkle, W., H. I. Jager, S. F. Railsback, B. D. Holcomb, T. K. Studley, and J. E.

Baldrige. 1998. Individual-based model of sympatric populations of brown and

rainbow trout for instream flow assessment: model description and calibration.

Ecological Modelling 110:175–207.

Wood, J. L. L. A., J. W. W. A. Grant, and M. H. Belanger. 2012. Population density and

territory size in juvenile rainbow trout, Oncorhynchus mykiss: implications for

population regulation. Canadian Journal of Fisheries and Aquatic Sciences

69:1121–1128.

Ylikarjula, J., M. Heino, and U. Dieckmann. 1999. Ecology and adaptation of stunted

182

growth in fish. Evolutionary Ecology 13:433–453.

Zimmerman, M. P. 1999. Food habits of smallmouth bass, walleyes, and northern

pikeminnow in the Lower Columbia River Basin during outmigration of juvenile

anadromous salmonids. Transactions of the American Fisheries Society 128:1036–

1054.

183

Appendix A: Age-Structured Model Sensitivity Analysis

No harvest High harvest

Param Change Population Lake Size at Population Lake Size at

Dynamics Abundance age 3 Dynamics Abundance age 3

+1 year FPE -9 2 FPE -35 14 Agemat -1 year FPE 8 -1 Cycle 32* -13*

+1 year FPE 0 0 Cycle 19* -14* Agevuln -1 year FPE 0 0 FPE -22 18

+5% FPE 1 -1 FPE 2 -1

+10% FPE 2 -1 FPE 3 -1

* * aBH +25% Cycle 5 -3 FPE 7 -3

-5% FPE -1 1 FPE -2 1

-10% FPE -3 2 FPE -4 2

184

-25% FPE -7 4 FPE -10 4

+5% FPE -1 1 FPE -2 1

+10% FPE -2 1 FPE -3 1

+25% FPE -4 2 FPE -7 3 bBH -5% FPE 1 0 FPE 1 -1

-10% FPE 2 -1 FPE 3 -1

-25% FPE 5 -3 FPE 8 -3

+5% FPE 0 0 FPE 0 0

+10% FPE 0 0 FPE 0 0

+25% FPE 0 0 FPE 0 0 aminterr -5% FPE 0 0 FPE 0 0

-10% FPE 0 0 FPE 0 0

-25% FPE 0 0 FPE 0 0

+5% FPE 0 0 FPE 0 0

+10% FPE 0 0 FPE 0 0

+25% FPE 0 0 FPE 0 0 bminterr -5% FPE 0 0 FPE 0 0

-10% FPE 0 0 FPE 0 0

-25% FPE 0 0 FPE 0 0

+5% FPE 1 -1 FPE 2 -1

+10% FPE 3 -1 FPE 4 -1

S0_max +25% FPE 7 -3 FPE 9 -3

-5% FPE -2 1 FPE -2 1

-10% FPE -3 1 FPE -4 1

185

-25% FPE -8 4 FPE -11 4

+5% FPE 0 0 FPE 0 0

+10% FPE 0 0 FPE 0 0

+25% FPE 0 0 FPE 0 0 S1_max -5% FPE 0 0 FPE 0 0

-10% FPE 0 0 FPE 0 0

-25% FPE 0 0 FPE 0 0

+5% FPE -2 -1 FPE -2 0

+10% FPE -4 -1 FPE -4 -1

+25% Cycle -10* -3* FPE -10 -2 astream -5% FPE 2 1 FPE 2 1

-10% FPE 3 1 FPE 4 1

-25% FPE 6 3 FPE 8 3

+5% FPE -2 -1 FPE -2 0

+10% Cycle -4* -1* FPE -4 -1

+25% Cycle -16* -6* FPE -11 -2 bstream -5% FPE 2 1 FPE 2 0

-10% FPE 3 1 FPE 3 1

-25% FPE 6 3 FPE 7 2

+5% FPE -5 2 FPE -4 1

+10% FPE -9 3 FPE -8 2

mi +25% FPE -21 8 FPE -18 5

-5% FPE 5 -2 FPE 4 -1

-10% FPE 11 -3 FPE 9 -2

186

-25% FPE 31 -9 FPE 26 -6

+5% FPE 0 -2 FPE -1 -1

+10% FPE 0 -4 FPE -2 -3

+25% FPE -1 -9 FPE -5 -6 blake -5% FPE 0 2 FPE 1 1

-10% FPE 0 4 FPE 2 3

-25% FPE -2 12 FPE 4 8

+5% FPE 0 0 FPE 0 0

+10% FPE 0 -1 FPE 0 0

+25% FPE 0 -2 FPE 1 -1 Sspawn -5% FPE 0 0 FPE 0 0

-10% FPE 0 1 FPE -1 0

-25% Cycle 0* 2* FPE -1 1

+5% FPE 0 0 FPE 0 0

+10% FPE 0 0 FPE 0 0

+25% FPE -1 0 FPE 0 0 aclim -5% FPE 0 0 FPE 0 0

-10% FPE 0 0 FPE 0 0

-25% FPE 0 0 FPE 0 0

+5% FPE -2 1 FPE -1 1

+10% FPE -3 1 FPE -2 1

* * bclim +25% Cycle -10 0 FPE -6 3

-5% FPE 1 -1 FPE 1 -1

-10% FPE 3 -1 FPE 2 -2

187

-25% FPE 7 -4 FPE 2 -6

+5% FPE 0 0 FPE 0 0

+10% FPE 0 0 FPE -1 0

+25% FPE 0 0 FPE -2 1 q -5% FPE 0 0 FPE 0 0

-10% FPE 0 0 FPE 1 0

-25% FPE 0 0 FPE 2 -1

188

Appendix B: Study Lakes Description

Table B.1 Year sampled, location and description of the 39 waterbodies included in the different chapters of this thesis, all waterbodies had rainbow trout present, additional species are presented in bold in the species column (RB = rainbow trout, CSU = largescale sucker, LSU = longnose sucker, SU = sucker spp., NSC = northern pikeminnow, RSC = redside shiner), the chapters column refers to the thesis chapter in which each lake was used. Summary statistics (minimum, maximum and mean) of the variables elevation, lake area and GDD are presented at the bottom of the table.

Lake Name Year Sampled Latitude Longitude Elevation (m) Area (ha) GDD Species Chapters Beartrack 2015 51.95556 -120.75059 1037 10 1170 RB, SU 3, 5 Beauregard 2016 51.20359 -120.33222 1282 12 1017 RB, RSC, CSU, NSC 3, 5 Big Bear 2014 52.08134 -121.10574 1365 37 915 RB 2, 3, 5 Boundary 2015 51.72984 -120.55177 1072 12 1100 RB 2, 3, 5 Cannine 2013 51.14062 -120.38720 1415 6 919 RB 2, 3, 5 Coffee 2014 52.02040 -121.09427 1067 58 1156 RB, LSU 3, 5 Corsica 2014 51.83221 -120.41528 1341 72 904 RB 2, 3, 5 Curtis 2016 50.11959 -119.01653 1325 11 1112 RB 2, 3, 5 Dunsapie 2015 51.18519 -120.33916 1316 12 998 RB, SU 3, 5 Ejas 2014 51.84799 -120.32937 1374 51 876 RB 2, 3, 5 Galena 2015 49.80688 -120.27519 1443 7 977 RB 2, 3, 5 Grassy 2013 51.55341 -120.64964 1261 4 977 RB 2, 3, 5

189

Hardcastle 2013 51.56798 -120.32993 1258 17 974 RB 2, 3, 5 Hardy Lower 2016 49.71912 -119.01891 1634 2 917 RB 2, 3, 5 Hidden 2015 50.13425 -119.17526 1445 9 1072 RB 2, 3, 5 Kaiser Bill 2015 50.15092 -119.20499 1368 5 1127 RB 2, 3, 5 Kitty Ann 2014 51.79623 -120.45542 1293 7 931 RB 2, 3, 5 Lost Horse 2013 51.57047 -120.40668 1373 38 871 RB 2, 3, 5 Lower Barge 2016 49.65095 -119.18832 1378 19 1086 RB 2, 3, 5 Meadow 2013 51.56258 -120.40106 1365 11 880 RB 2, 3, 5 Meridian 2015 51.59494 -120.56567 1230 22 971 RB 2, 3, 5 Mystery 2016 49.72581 -120.56197 1369 5 1030 RB, NSC, RSC 3, 5 No Name 1 2015 51.57485 -120.41930 1387 52 855 RB 2, 3, 5 Pantano 2013 51.16484 -120.38635 1479 5 872 RB 2, 3, 4, 5 Parky 2013 51.15114 -120.33198 1348 7 1011 RB 2, 3, 5 Phyllis 2014 51.81755 -120.43267 1346 12 903 RB 2, 3, 5 Rock Island 2013 51.54017 -120.33010 1267 65 998 RB 2, 3, 5 No Name 2 2015 52.00747 -120.54182 1093 24 1127 RB 2, 3, 5 Shillings 2016 51.31252 -120.32636 1319 15 966 RB 2, 3, 5 Sicily 2014 51.79690 -120.41456 1254 21 954 RB 2, 3, 5 Sock 2014 51.76594 -120.06757 1282 16 1006 RB 2, 3, 5 Solco 2015 49.35619 -119.27817 1675 8 859 RB 2, 3, 5 Stubby 2013 51.14556 -120.40828 1506 6 841 RB 2, 3, 4, 5

190

Surprise 2014 51.72742 -120.27527 1392 27 875 RB 2, 3, 5 Tahoola 2016 51.58290 -120.48342 1410 39 843 RB 2, 3, 5 Today 2013 51.15053 -120.39983 1490 6 857 RB 2, 3, 4, 5 Weller 2015 51.94304 -120.74605 1029 17 1177 RB, SU 3, 5 Whitewood 2014 51.09568 -120.34107 1337 19 1046 RB 2, 3, 5 Wollaston 2015 50.05872 -119.10503 1312 11 1130 RB 2, 3, 5 Min 1029 2 841 Max 1675 72 1177 Mean 1332 20 982

191

Appendix C: Stream Sampling

List of the environmental data collected during stream sampling at the stream level (once per stream to describe the whole stream - Stream) and for each 100 m reach within a stream at the start point (Reach Point) and throughout the reach (Reach Overall). Summary statistics for the reach variables collected are in table C.1.

Stream

− Coordinate at beginning and end of − Dominant aquatic vegetation sampling (categorical: Rooted emergent, Rooted − Time at beginning and end of submerged, Rooted floating, Free sampling floating, Floating algae, Attached − Cloud cover % algae) − Air temperature (°C) − Aquatic vegetation stream cover (%) − Water temperature (°C) − Sediment odour (categorical: Normal, − Water conductivity (uS) Chemical, Sewage, Anaerobic, − Water turbidity (categorical: Clear, Petroleum) Opaque, Stained, Turbid) − Sediment/Substrate oil (categorical: − Surrounding land use (categorical: Absent, Slight, Moderate, Profuse) Forest, Field/Pasture, Agriculture, − Sediment/Substrate deposits Residential, Industrial) (categorical: Sludge, Sawdust, Paper − Riparian vegetation (categorical: Fiber, Sand, Relict shells) Tree, Grasses, Shrubs, Herbaceous)

Reach Point

− Wetted width (m) − Maximum depth (m) − Bankful width (m) − Velocity (m/sec)

192

Reach Overall

− Large woody debris cover (%) − Habitat score (20 points each, *10 − Canopy cover (%) points per bank, details in Barbour et − Undercut (%) al. 1998) − Flow type (% - total 100%): Riffle, 1. Epifaunal Substrate/Available Run Pool Cover − Inorganic subtrate type (% - total 2. Pool Substrate Characterization 100%): Bedrock, Boulder, Cobble, 3. Pool Variability Gravel, Sand, Silt, Clay 4. Sediment Deposition − Organic substrate type (%): Detritus, 5. Channel Flow Status Muck/Mud, Marl 6. Channel Alteration 7. Channel Sinuosity 8. Bank Stability* 9. Vegetative Protection* 10. Riparian Vegetative Zone Width*

193

Table C.1 Summary statistics of the environmental variables collected at each stream reach sampled.

Category Variable Mean SD Min Max Width (m) 3.20 2.33 0 12 Reach start Bankfull Width (m) 6.08 5.33 1 35 point Maximum Depth (m) 0.27 0.17 0 1 Velocity (m/s) 0.13 0.12 0 1 Stream Large Woody Debris 14.66 12.97 0 55 characteristics Canopy Cover 25.09 22.47 0 95 (%) Undercut Banks 39.11 31.72 0 100 Stream Riffle 29.82 25.42 0 90 morphology Run 52.37 25.73 5 100 type (% - total 100%) Pool 17.78 15.98 0 80 Bedrock 2.00 7.10 0 50 Boulder 24.54 22.10 0 80 Inorganic Cobble 21.35 20.24 0 90 substrate type Gravel 24.74 19.41 0 80 (% - total 100%) Sand 12.67 14.31 0 65 Silt 13.55 27.28 0 100 Clay 1.23 9.22 0 100 Detritus 28.31 17.38 2 100 Organic Muck/Mud 20.77 27.97 0 95 Substrate (%) Marl 0.00 0.00 0 0 Epifaunal Substrate 16.34 2.86 6 20 Pool Substrate 14.94 3.55 0 20 Pool Variability 11.53 5.26 0 19 Sediment Deposition 14.92 3.71 1 20 Habitat score Channel Flow 12.63 2.96 0 19 (1-20 points) Channel Alteration 19.38 1.51 13 20 Channel Sinuosity 15.74 4.94 2 20 Bank Stability 18.39 2.62 10 36 Vegetative Protection 19.44 1.48 12 20 Riparian Vegetation 19.19 1.94 10 20

194

Appendix D: Habitat Suitability Index (HSI)

Habitat Suitability Index Calculations

Habitat Suitability Index variables used, and calculations modified from the suitability curves presented in Raleigh et al. (1984), the habitat scores (0-20) refer to the scores described in Barbour et al. (1998).

1⁄ (S3.1) 퐻푆퐼 = (퐶퐸 × 퐶퐹 × 퐶퐽 × 퐶푂 ) 4

1⁄ (S3.2) 퐶퐸 = (푉7 × 푉16푎) 2 where V7 is a value described by the average size of substrate (cm) V16b is the percent of substrate that is fines (substrate < 3 mm) which both relate to the size of substrate that is the most suitable for spawning in terms of water exchange rates and proper redd construction.

1⁄ 1⁄ (S3.3) 퐶퐹 = (푉10 × (푉8 × 푉16푏) 2) 2 where all three variables are associated with fry cover: V10 is the percent pools, V8 is the availability of substrate associated with fry cover (gravel, cobble, boulders), V16b is the percent fines.

푉 + 푉 +푉 (S3.4) 퐶 = 6 10 15 퐽 3 where V6 is the availability of instream cover, V10 is the percent of pools and V15 is the pools classes’ variability. All three variables describe the availability of cover for juveniles.

1 (푉 × 푉 ) ⁄2+ 푉 1 1 (S3.5) 퐶 = ( 9 16푏 11 × (푉 × 푉 ) ⁄2) ⁄2 푂 2 12 14

195

where V9 is the predominant substrate type associated with high insect numbers and optimal production, V16b is percent fines, V11 is the average percent of terrestrial vegetation cover which provides allochthonous material to the stream, V12 is the cover that provides adequate erosion control and V14 describes the water level of the stream as flow variations can affect habitat quality.

Weighted Useable Area Calculations

HSI values were calculated for each stream segment i. These values were then used to calculate Weighted Useable Area (WUA) to describe both the quality and quantity of habitat available (Schamberger et al. 1982).

(S3.6) 푊푈퐴푖 = 퐻푆퐼푖 × 퐴푖

(S3.7) 퐴푖 = 퐿푖 × 푊̅푖

2 where Ai is a segment’s area (in m ) calculated using the segment’s length (Li in m) and mean wetted width (푊̅푖 in m). The total area for the whole stream was calculated by summing the different segments.

(S3.8) 푊푈퐴푠푡푟푒푎푚 = ∑ 푊푈퐴푖

196

Table D.1 Detailed calculations of the Habitat Suitability Index (HSI) variables used.

Variable Description of x Calculations

V6 Epifaunal If x ≤ 8 then V6 = 0.1143x + 0.2

Substrate/Available cover If x > 8 then V6 = 1.0

score (0-20) – for

percentage instream cover

in Raleigh et al. (1984)

V7 Average size of substrate If x ≤ 15 cm then V7 = 0.067x

(cm) If 15 cm < x ≤ 60 cm then V7 = 1.0

If 60 cm < x ≤ 100 cm then V7 = -0.0225x + 2.35

If x > 100 cm then V7 = 0.1

V8 Percentage of substrate that If x ≤ 10% then V8 = 0.1x

is gravel, cobble or boulder If x > 10% then V8 = 1.0

(%)

V9 Predominant substrate type If cobble/boulder or vegetation ≥ 50% cover V9 = 1.0

(categorical) If gravel or cobble/boulder/fines ≥ 50% V9 = 0.6

If bedrock or fines ≥ 50% and cobble and gravel ≤

25% then V9 = 0.3

197

V10 Percentage of pools (%) If x ≤ 35% cm then V10 = 0.02x + 0.3

If 35% < x ≤ 65% then V10 = 1.0

If x > 65% then V10 = -0.014x + 1929

V11 Vegetative protection cover If x ≥ 18 then V11 = 1.0

score (0-20) – for average If 12 ≤ x < 18 cm then V11 = 0.6

percent vegetational If x < 12 then V11 = 0.3

ground cover and canopy

in Raleigh et al. (1984)

V12 Bank stability score (0-20) If x ≥ 18 then V12 = 1.0

– for average percent If 12 ≤ x < 18 cm then V12 = 0.6

rooted vegetation and If x < 12 then V12 = 0.3

stable bank in Raleigh et al.

(1984)

V14 Channel flow status score If x ≥ 11 then V14 = 1.0

(0-20) – for average annual If 6 ≤ x < 10 cm then V14 = 0.6

base flow If x < 6 then V14 = 0.1

V15 Pool variability score (0- If x ≥ 16 then V15 = 1.0

20) – for pool class rating If 11 ≤ x < 16 cm then V15 = 0.6

in Raleigh et al. (1984) If x < 11 then V15 = 0.3

198

V16a Percentage of fines (sum of If x ≤ 5% cm then V16a = 1.0

sand, silt and clay) *For If 15% < x ≤ 25% then V16a = -0.04x + 1.2

spawning areas If 25% < x ≤ 75% then V16a = -0.004x + 0.31

If x > 75% then V16a = 0.05

V16b Percentage of fines (sum of If x ≤ 15% cm then V16b = 1.0

sand, silt and clay) *For If 15% < x ≤ 60% then V16b = -0.018x + 1.27

other areas If x > 60% then V16b = 0.2

199

Appendix E: Model Posterior Predictive Distribution

200

201

202

203

204

Appendix F: Growth and Maturity Traits Correlation

퐴50푓 퐴50푚 푡0 k 퐿∞

퐴50푓 1.00

퐴50푚 0.39 1.00

푡0 0.05 -0.03 1.00

k 0.37 0.29 0.68 1.00

퐿∞ 0.53 0.14 0.06 0.34 1.00

205

Appendix G: Life History Traits Model Selection

Table G.1 List of all the models compared to predict the variation in life-history traits associated with female maturation age (푨ퟓퟎ풇 - models 1-12) and size (푳ퟓퟎ풇 - models 13-24), and log-transformed growth parameters k (models 26-36), asymptotic length (푳∞- models 27-48) and observed maximum length (MaxObsFL - models 49-60). The most parsimonious model for each trait (lowest AIC) is in bold and models within two AIC points are in italic.

2 2 Model R adjR p-value AIC

1 퐴50푓 ~ WUAratio 1.4E-03 -0.03 0.82 48.20

2 푨ퟓퟎ풇 ~ Lake abundance 0.02 -3.1E-03 0.35 47.34

3 퐴50푓 ~ WUAratio + Lake abundance 0.03 -0.02 0.56 49.01

4 퐴50푓 ~ WUAratio + GDD 0.01 -0.04 0.81 49.79

5 퐴50푓 ~ Lake abundance + GDD 0.04 -0.01 0.49 48.70

6 퐴50푓 ~ WUAratio + Lake abundance + GDD 0.05 -0.03 0.60 50.18

7 퐴50푓 ~ WUAratio + OtherSp 0.02 -0.03 0.65 49.31

8 퐴50푓 ~ Lake abundance + OtherSp 0.04 -0.01 0.48 48.69

9 퐴50푓 ~ WUAratio + Lake abundance + OtherSp 0.05 -0.03 0.62 50.31

10 퐴50푓 ~ WUAratio + GDD + OtherSp 0.03 -0.05 0.79 51.11

11 퐴50푓 ~ Lake abundance + GDD + OtherSp 0.05 -0.03 0.62 50.32

12 퐴50푓 ~ WUAratio + Lake abundance + GDD + OtherSp 0.06 -0.05 0.70 51.78

13 퐿50푓 ~ WUAratio 0.07 0.05 0.09 377.69

14 퐿50푓 ~ Lake abundance 0.29 0.27 3.9E-04 367.26

15 퐿50푓 ~ WUAratio + Lake abundance 0.30 0.26 1.5E-03 368.63

16 퐿50푓 ~ WUAratio + GDD 0.10 0.05 0.15 378.61

17 퐿50푓 ~ Lake abundance + GDD 0.29 0.25 2.0E-03 369.18

18 퐿50푓 ~ WUAratio + Lake abundance + GDD 0.31 0.25 4.8E-03 370.45

19 퐿50푓 ~ WUAratio + OtherSp 0.14 0.10 0.06 376.62

20 푳ퟓퟎ풇 ~ Lake abundance + OtherSp 0.34 0.30 5.9E-04 366.58

21 퐿50푓 ~ WUAratio + Lake abundance + OtherSp 0.35 0.29 1.7E-03 368.04

22 퐿50푓 ~ WUAratio + GDD + OtherSp 0.19 0.13 0.05 376.25

23 퐿50푓 ~ Lake abundance + GDD + OtherSp 0.35 0.29 1.7E-03 368.02

24 퐿50푓 ~ WUAratio + Lake abundance + GDD + OtherSp 0.36 0.29 3.6E-03 369.24 25 푙푛(푘) ~ WUAratio 4.3E-05 -0.03 0.97 79.51 26 푙푛(푘) ~ Lake abundance 0.01 -0.02 0.64 79.27 27 ln(푘) ~ WUAratio + Lake abundance 0.01 -0.05 0.88 81.23

206

28 ln(푘) ~ WUAratio + GDD 0.01 -0.04 0.83 81.11 29 ln(푘) ~ Lake abundance + GDD 0.02 -0.04 0.71 80.76 30 ln(푘) ~ WUAratio + Lake abundance + GDD 0.02 -0.06 0.85 82.65 31 푙푛(푘) ~ WUAratio + OtherSp 0.08 0.03 0.24 78.38 32 퐥퐧(풌) ~ Lake abundance + OtherSp 0.08 0.03 0.23 78.36 33 푙푛(푘) ~ WUAratio + Lake abundance + OtherSp 0.08 8.6E-04 0.40 80.27 34 푙푛(푘) ~ WUAratio + GDD + OtherSp 0.08 -4.4E-04 0.41 80.32 35 푙푛(푘) ~ Lake abundance + GDD + OtherSp 0.08 8.2E-04 0.40 80.27 ln(푘) ~ WUAratio + Lake abundance + GDD + 36 OtherSp 0.08 -0.03 0.55 82.14

37 ln(퐿∞) ~ WUAratio 0.06 0.03 0.14 24.63

38 ln(퐿∞) ~ Lake abundance 0.03 -8.4E-04 0.33 25.93

39 ln(퐿∞) ~ WUAratio + Lake abundance 0.07 0.01 0.30 26.30

40 ln(퐿∞) ~ WUAratio + GDD 0.16 0.11 0.05 22.30

41 ln(퐿∞) ~ Lake abundance + GDD 0.09 0.04 0.17 25.05

42 ln(퐿∞) ~ WUAratio + Lake abundance + GDD 0.16 0.08 0.11 24.28

43 ln(퐿∞) ~ WUAratio + OtherSp 0.16 0.12 0.04 21.97

44 ln(퐿∞) ~ Lake abundance + OtherSp 0.13 0.08 0.08 23.38

45 ln(퐿∞) ~ WUAratio + Lake abundance + OtherSp 0.18 0.11 0.07 23.22

46 ln(퐋∞) ~ WUAratio + GDD + OtherSp 0.22 0.16 0.03 21.00

47 ln(퐿∞) ~ Lake abundance + GDD + OtherSp 0.17 0.10 0.09 23.83 ln(퐿∞) ~ WUAratio + Lake abundance + GDD + 48 OtherSp 0.23 0.14 0.06 22.78 49 ln(MaxObsFL) ~ WUAratio 0.15 0.12 0.02 -31.84 50 ln(MaxObsFL) ~ Lake abundance 0.02 -0.01 0.45 -26.22 51 ln(MaxObsFL) ~ WUAratio + Lake abundance 0.15 0.10 0.06 -29.84 52 ln(MaxObsFL) ~ WUAratio + GDD 0.42 0.39 4.7E-05 -45.17 53 ln(MaxObsFL) ~ Lake abundance + GDD 0.23 0.19 0.01 -34.00 54 ln(MaxObsFL) ~ WUAratio + Lake abundance + GDD 0.44 0.39 1.4E-04 -44.04 55 ln(MaxObsFL) ~ WUAratio + OtherSp 0.26 0.22 4.8E-03 -35.17 56 ln(MaxObsFL) ~ Lake abundance + OtherSp 0.11 0.07 0.11 -28.34 ln(MaxObsFL) ~ WUAratio + Lake abundance + 57 OtherSp 0.26 0.19 0.01 -33.27 58 ln(MaxObsFL) ~ WUAratio + GDD + OtherSp 0.47 0.43 4.8E-05 -46.52 59 ln(MaxObsFL) ~ Lake abundance + GDD + OtherSp 0.27 0.21 0.01 -34.06 ln(MaxObsFL) ~ WUAratio + Lake abundance + GDD 60 + OtherSp 0.48 0.42 1.5E-04 -44.95

207