Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Understanding and assessing cumulative impacts to coastal ecosystem services Singh, Gerald Gurinder 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2016_november_singh_gerald.pdf [ 3.14MB ]
Metadata
JSON: 24-1.0319134.json
JSON-LD: 24-1.0319134-ld.json
RDF/XML (Pretty): 24-1.0319134-rdf.xml
RDF/JSON: 24-1.0319134-rdf.json
Turtle: 24-1.0319134-turtle.txt
N-Triples: 24-1.0319134-rdf-ntriples.txt
Original Record: 24-1.0319134-source.json
Full Text
24-1.0319134-fulltext.txt
Citation
24-1.0319134.ris

Full Text

UNDERSTANDING AND ASSESSING CUMULATIVE IMPACTS TO COASTAL ECOSYSTEM SERVICES by  Gerald Gurinder Singh  B.Sc., The University of Alberta, 2008 M.Sc., The University of British Columbia, 2010  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Resource Management and Environmental Studies)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  October 2016  © Gerald Gurinder Singh, 2016 ii  Abstract Anthropogenic impacts to the environment are often co-occurring and cumulative. While research on cumulative environmental impacts has historically focused on biophysical attributes, anthropogenic activities also pose risks to ecosystem services. This dissertation evaluates the state of environmental impact assessment, particularly the characterization of cumulative impacts, and pilots new methods in two sites with varying data availability: coastal British Columbia (relatively high data availability) and Tasman and Golden Bays, New Zealand (relatively low data availability).  First, I assessed the state of cumulative impact assessment in its most common legally mandated form, Environmental Impact Assessments (EIAs), from seven nations. EIAs generally identified a large number of impacts, though a consistently minute subset was deemed “significant” for decision-making. Many EIAs considered spatiotemporal scopes smaller than justifiable, presumed mitigations effective without justification, and determined significance by consultants (paid by developers) with minimal stakeholder input.  Next, I piloted two novel cumulative impact assessment procedures for contexts with available data. These procedures combined spatial analysis with expert elicitation for the first, and used Bayesian networks for the second. First, I found that some ecosystem services in British Columbia face higher risk from global stressors, while others face higher risk from local stressors. Changes to ecosystem service access and perceived quality may be as important as changes to biophysical attributes. Second, I show that management plans for the herring fishery are likely ineffective because important impacts were unaddressed. iii  Finally, I piloted a novel expert elicitation approach to characterize and quantify impacts on ecosystem services in data-poor contexts. Local New Zealand experts were tasked with estimating impact before and after group deliberations, and describe causes of impact. This methodology simultaneously reduces the variability among experts’ best estimates, while also increasing individual uncertainty. Despite high uncertainty of individual stressor impact, cumulative impacts were consistently high across ecosystem services. The key stressor was sedimentation, caused by interacting climate change and activities based on land and in the water. As a whole, this dissertation advances the nascent state of cumulative impact assessment for ecosystem services, and pioneers diverse methods to synthesize understanding of these crucial considerations for management and policymaking.   iv  Preface The work that follows here benefitted from the time and effort of numerous individuals. This dissertation reports on analyses of four different data-sets, of which two required expert elicitation. These elicitations were approved by UBC’s Behavioural Research Ethics Board (certificate number H12-01868 for Chapter 3, and certificate number H14-00042 for Chapters 5 and 6).  Chapter 2 is the result of a working group I chaired, also consisting of Professors Kai Chan and Terre Satterfield, as well as students Jackie Lerner, Bernardo Ranieri, Guillaume Peterson St-Laurent, Janson Wong, Alice Guimaraes, Gustavo Yunda-Guarin, and researchers Megan Mach and Cathryn Clarke Murray. I designed the database, with refinement from Jackie Lerner, Bernardo Ranieri, Guillaume Peterson St-Laurent, Megan Mach, Cathryn Clarke Murray, Kai Chan, and Terre Satterfield. Data collection was carried out by all members of the working group except the professors. Additional data collection was carried out by Adrian Semmelink, Raoul Wieland, and Calum Watt. I carried out all analysis and writing, with considerable comments and input from all working group members. Chapter 3 benefitted from a foreign study fellowship I acquired to work with members from the National Center for Ecological Analysis and Synthesis. I designed the research instrument to collect expert responses, with input from Kai Chan, Terre Satterfield, and Ben Halpern. I collected the spatial data with Ian Eddy. I conducted most of the analysis, with support from Ian Eddy and Rabin Neslo. I wrote the manuscript with comments and input from Kai Chan, Terre Satterfield, and Ben Halpern. v  Chapter 4 is a result of a research fellowship I had with the Centre of Excellence for Environmental Decisions. I used a risk assessment I helped develop with Cathryn Clarke Murray, Megan Mach, Rebecca Martone, and Kai Chan. I performed the analysis with guidance from Jonathan Rhodes, Eve McDonald-Madden, Hugh Possingham, Edd Hammill, and Kai Chan. I wrote the manuscript, which benefitted greatly from the input of Cathryn Clarke Murray, Rebecca Martone, Kai Chan, Jonathan Rhodes, Edd Hammill, Terre Satterfield, and Ben Halpern. Chapters 5 and 6 are a consequence of a collaboration with the Cawthron Institute. I designed the research, with invaluable input by Kai Chan, Terre Satterfield, Jim Sinner, and Milind Kandlikar. I received help with organizing and facilitating an expert workshop from Kai Chan, Jim Sinner, Sarah Klain, Paige Olmsted, Dana Clark, and Mark Newton. I conducted the analysis and writing, and benefitted from comments and input from Joanne Ellis, Jim Sinner, Milind Kandlikar, Terre Satterfield, Kai Chan, and Ben Halpern. With the exception of Chapters 1 and 7, all chapters were written with the intention of publication in various research journals. None have yet been published.     vi  Table of Contents Abstract .......................................................................................................................................... ii Preface ........................................................................................................................................... iv Table of Contents ......................................................................................................................... vi List of Tables .............................................................................................................................. xiii List of Figures ............................................................................................................................. xvi Acknowledgements .................................................................................................................. xxiii Dedication ...................................................................................................................................xxv Chapter 1: Introduction ................................................................................................................1 1.1 Relating Ecosystem Services to Cumulative Impact ...................................................... 2 1.2 Relating Cumulative Impact Assessment to Ecosystem Services .................................. 5 1.3 Problem Statement .......................................................................................................... 8 1.4 Existing Frameworks to Build From............................................................................... 9 1.4.1 Ecosystem Models .................................................................................................... 11 1.4.2 Ecological Risk Assessments .................................................................................... 11 1.4.3 Stressor Footprints .................................................................................................... 12 1.4.4 Interaction Determination ......................................................................................... 13 1.4.5 Cumulative Impact Mapping .................................................................................... 13 1.4.6 Process Based Models............................................................................................... 15 1.4.7 Network Analysis...................................................................................................... 17 1.4.8 Stressor Scale Frameworks ....................................................................................... 18 1.4.9 Cumulative Impact Planning..................................................................................... 19 vii  1.4.10 Past Impact Determination .................................................................................... 19 1.5 Gaps in Frameworks and Opportunities for Development ........................................... 20 1.6 Considering Data Needs ............................................................................................... 21 1.6.1 Proxy Data ................................................................................................................ 22 1.6.2 Models....................................................................................................................... 22 1.6.3 Literature-Based Measures ....................................................................................... 23 1.6.4 Expert Elicitation ...................................................................................................... 24 1.7 Case Studies for the Dissertation .................................................................................. 25 1.8 Structure of the Dissertation ......................................................................................... 26 Chapter 2: Scientific Shortcomings in Environmental Assessments ......................................31 2.1 Introduction ................................................................................................................... 31 2.2 Methods......................................................................................................................... 33 2.3 Results and Discussion ................................................................................................. 37 2.3.1 Different Places, Same Bottom Line ........................................................................ 37 2.3.2 Narrowly Addressed Impacts .................................................................................... 39 2.3.3 Overconfidence in Mitigation Measures ................................................................... 42 2.3.4 Suspected Bias in Significance Determination ......................................................... 44 2.4 Conclusions ................................................................................................................... 47 Chapter 3: Mapping Cumulative Impacts to Coastal Ecosystem Services in British Columbia .......................................................................................................................................49 3.1 Introduction ................................................................................................................... 49 3.2 Methods......................................................................................................................... 52 3.2.1 Spatial Representation of Ecosystem Services ......................................................... 53 viii  3.2.2 Spatial Representation of Impacting Activities ........................................................ 54 3.2.3 Impact Mapping ........................................................................................................ 54 3.2.4 Risk Scores................................................................................................................ 55 3.2.5 Cumulative Impact Model ........................................................................................ 59 3.2.6 Understanding Mechanisms of Impact ..................................................................... 60 3.3 Results ........................................................................................................................... 60 3.3.1 Impacts to Ecosystem Services ................................................................................. 60 3.3.2 Components of Risk to Ecosystem Services ............................................................. 66 3.3.3 Pathways of Effects................................................................................................... 68 3.4 Discussion ..................................................................................................................... 69 3.4.1 Including Social Criteria Lead to Greater Accounting of Impact ............................. 69 3.4.2 Different Ecosystem Services Are Impacted in Different Ways .............................. 70 3.4.3 Social Dimensions of Risk Are Important for Impact Characterization ................... 72 3.4.4 Modeling Impact Should Account for Pathways of Impact...................................... 73 3.4.5 Limitations and Opportunities .................................................................................. 75 3.5 Conclusion .................................................................................................................... 76 Chapter 4: Prioritizing Management for Cumulative Impacts ...............................................78 4.1 Introduction ................................................................................................................... 78 4.2 Methods......................................................................................................................... 80 4.2.1 Case Study – North Coast of British Columbia ........................................................ 81 4.2.2 Bayesian Belief Networks......................................................................................... 81 4.2.3 Calculating Risk ........................................................................................................ 82 4.2.4 Creating the Conditional Probability Table .............................................................. 85 ix  4.2.5 Identifying Leverage Nodes ...................................................................................... 86 4.2.6 Scenarios ................................................................................................................... 87 4.2.7 Estimating the Number of Species and Ecosystem Service Provision at Risk ......... 88 4.3 Results ........................................................................................................................... 88 4.3.1 Identifying Leverage Nodes ...................................................................................... 88 4.3.2 Leverage Nodes for Multiple Species Management ................................................. 90 4.3.3 Comparing Revealed and Assumed Leverage Nodes in Herring Management ....... 91 4.3.4 Consequences for Species ......................................................................................... 94 4.3.5 Consequences for Ecosystem Service Provision ...................................................... 96 4.4 Discussion ..................................................................................................................... 98 4.4.1 Managing Multiple, Diverse Risks ......................................................................... 101 4.4.2 Limitations .............................................................................................................. 103 4.4.3 Applications ............................................................................................................ 104 4.4.4 Working with the EBM Toolbox ............................................................................ 105 4.5 Conclusion .................................................................................................................. 106 Chapter 5: Group Elicitations Yield More Consistent, Yet More Uncertain Experts ........108 5.1 Introduction ................................................................................................................. 108 5.2 Methods....................................................................................................................... 112 5.2.1 Study Site ................................................................................................................ 112 5.2.2 Online Survey ......................................................................................................... 113 5.2.3 Individual Interviews .............................................................................................. 114 5.2.4 Group Workshop ..................................................................................................... 115 5.2.5 Analysis................................................................................................................... 116 x  5.3 Results ......................................................................................................................... 118 5.3.1 Among Expert Consistency .................................................................................... 118 5.3.1.1 Ordinal Consistency ........................................................................................ 118 5.3.1.2 Numerical Consistency ................................................................................... 120 5.3.2 Within-Expert Subjective Uncertainty .................................................................... 121 5.4 Discussion ................................................................................................................... 125 5.4.1 Reducing Variability Among Experts ..................................................................... 125 5.4.2 Increasing Subjective Uncertainty Within Experts ................................................. 128 5.5 Conclusion .................................................................................................................. 129 Chapter 6: Mechanisms and Risk of Cumulative Impacts to Ecosystem Services: An Expert Elicitation Approach ..................................................................................................................130 6.1 Introduction ................................................................................................................. 130 6.2 Methods....................................................................................................................... 133 6.2.1 Tasman and Golden Bays ....................................................................................... 134 6.2.2 Expert Elicitation .................................................................................................... 136 6.2.3 Data Analysis .......................................................................................................... 140 6.3 Results ......................................................................................................................... 143 6.3.1 Ranking ................................................................................................................... 143 6.3.2 Mechanisms of Cumulative Impact ........................................................................ 144 6.3.2.1 Activities/Drivers of Change .......................................................................... 144 6.3.2.1.1 Climate Change ......................................................................................... 144 6.3.2.1.2 Commercial Fishing .................................................................................. 144 6.3.2.2 Stressors .......................................................................................................... 145 xi  6.3.2.2.1 Sedimentation ............................................................................................ 145 6.3.2.2.2 Pollution .................................................................................................... 146 6.3.3 Impacts on Ecosystem Services .............................................................................. 146 6.3.3.1 Biodiversity ..................................................................................................... 146 6.3.3.2 Shellfish Aquaculture...................................................................................... 149 6.3.3.3 Fishing............................................................................................................. 151 6.3.3.4 Marine Recreation ........................................................................................... 153 6.3.4 Cumulative Impacts ................................................................................................ 155 6.4 Discussion ................................................................................................................... 155 6.4.1 Using Ecosystem Services as a Basis for Management, Even with Data Paucity .. 161 6.4.2 Addressing Mechanisms of Cumulative Impact ..................................................... 162 6.4.2.1 Prioritizing Research and Management .......................................................... 164 6.4.3 Cross-scale impacts ................................................................................................. 165 6.5 Conclusions ................................................................................................................. 165 Chapter 7: Conclusion ...............................................................................................................167 7.1 Summary of Key Findings and Discussion ................................................................. 167 7.2 Potential Applications and Future Research ............................................................... 175 7.3 Strengths and Limitations ........................................................................................... 177 7.3.1 Strengths ................................................................................................................. 177 7.3.2 Limitations .............................................................................................................. 179 7.4 Revisiting Our Dual Role as Agents of Risk and Beneficiaries ................................. 183 Bibliography ...............................................................................................................................184 Appendices ..................................................................................................................................213 xii  Appendix A Supplementary Material for Chapter 2 ............................................................... 213 A.1 Supplementary Tables ............................................................................................. 213 Appendix B Supplementary Materials for Chapter 3 ............................................................. 217 B.1 Ecosystem Service Models ..................................................................................... 217 B.2 Human Impacts ....................................................................................................... 222 B.3 Supplementary Figures ........................................................................................... 231 Appendix C Supplementary Methods for Chapter 4............................................................... 232 C.1 Scoring exposure and consequence ........................................................................ 232 C.2 Scoring Uncertainty ................................................................................................ 235 C.3 Supplementary Figures and Tables ......................................................................... 237 Appendix D Supplementary Methods for Chapter 6 .............................................................. 244  xiii  List of Tables Table 1.1 Dimensions of cumulative impacts to ecosystem services that existing research frameworks represent and emphasize. .......................................................................................... 10 Table 4.1 The leverage nodes for species, grouped by sector (fisheries, sea, land, and long term impacts). The number in brackets indicates the number of leverage nodes revealed through our analysis for each species. .............................................................................................................. 93 Table 5.1 The list of ten activities and stressors initially provided for experts to rank, with the additional eight suggested by experts. ........................................................................................ 114 Table 5.2 Modeled rank of prominent risks to ecosystem services in Tasman and Golden Bays, New Zealand. The degree of expert homogeneity in ranking is provided under each ecosystem service and time (at interview stage or after workshop), and risks are ranked from greatest to lowest risk assessed. There is a different number of ranks among ecosystem services because there were a different number of risks assessed among the times for each ecosystem service. . 122 Table 5.3 Summaries of mixed-effect paired t-tests comparing the consistency in best estimate and interval for risks to the four ecosystem services before and after the workshop. The column “experts” refers to the number of experts in the analysis, while “comparisons” refers to the number of before-after comparisons included in the analysis (these were treated as nested random effects in the models). The column “difference” records the estimate of the difference score and the interval length of expert responses before the workshop compared to after the workshop. The column “df” corresponds to the degrees of freedom in the analysis. Positive “best estimate” scores indicate that the individual best estimates were more similar to the average after xiv  the workshop (than before); positive “interval” scores indicate that the range of estimates shrunk after the workshop....................................................................................................................... 123 Table 6.1 Modelled ranks for the impact of 12 stressors on four ecosystem services, from highest impact to lowest. ......................................................................................................................... 143 Table A.1 Composition of the EISs evaluated for each jurisdiction by types of developments considered. .................................................................................................................................. 213 Table A.2 List of species selected and associated area measurements (in km2)......................... 214 Table A.3 Studies on longevity of Acid Mine Discharge (AMD) impacts ................................ 215 Table A.4 A typology of consultation with stakeholders used in our study. .............................. 216 Table B.1 Erosion risk to different coastal classes ..................................................................... 218 Table B.2 Data files, sources, and resolution used to map impacts and ecosystem services ..... 223 Table B.3 Descriptions of human activities and stressors provided to experts to assess risk..... 227 Table B.4 Descriptions of exposure criteria given to experts to assess risk ............................... 228 Table B.5 Descriptions of consequence criteria given to experts to assess risk ......................... 229 Table C.1 Scoring of variables: a) Spatial Scale, b) Temporal Scale, c) Load and d) Consequence..................................................................................................................................................... 234 Table C.2 Scoring definition of the uncertainty of risk scores. .................................................. 235 Table C.3 Nodes ranked by influence in maintaining low risk to herring. The columns identify the probabilities of low, medium, and high risk according to each scenario. Nodes in bold are assumed leverage nodes in the herring IFMP ............................................................................. 238 Table C.4 Human activities (in bold) and associated stressors. The numbers beside each activity and stressor corresponds to the nodes in Figure 4.2 ................................................................... 240 xv  Table C.5 The provision of ecosystem services (in bold) and their associated species. The numbers beside each ecosystem service and species corresponds to the nodes in Figure 2. ...... 243 Table D.1 A list of activities and stressors that experts verified at the workshop ...................... 244 xvi  List of Figures Figure 1.1 Conceptual model showing cumulative impacts to ecosystem services. Ecosystem services are an interaction between the biophysical and social environment, and can be considered along different points. Supply considers only the biophysical system that produces the service of interest, while service consider the access of people to that supply, and value considers people’s preferences. Realizing the benefits of ecosystem services relies on all three (represented by blue arrows). Human activities can impact ecosystem services through each of the three points (represented by red arrows). This model modifies a model of the ecosystem service production chain found in Tallis et al. (2012)................................................................................. 4 Figure 2.1 The number of potential impacts, residual impacts, and significant impacts reported in EISs for A) project-specific impacts and B) cumulative impacts. Bars represent bootstrap 95% confidence interval of the medians, and the red lines represent the global medians. EISs were selected from a single state or province within the countries marked with an asterisk (*) in the legend. ........................................................................................................................................... 39 Figure 2.2 Density histograms showing the overlap of spatial scale in EISs (in red) and current scientific data (in grey) on A) spatial range of populations and species assessed in EISs and B) time after mine decommissioning that acid mine drainage impacts ecosystems. Note the logarithmic scale used in A). That the spatial and temporal scales of assessments were often smaller/shorter than the ranges of affected species/duration of relevant impacts suggests that the scoping was insufficient to address cumulative effects. ............................................................... 41 Figure 2.3  The A) proportion of mitigation measures written in ambiguous and unenforceable language in each jurisdiction (bars represent 95% bootstrap CI of the median and red line xvii  represents global median) and B) the proportion of EISs in all jurisdictions that have explicit analysis of mitigation effectiveness and consider uncertainty of mitigation effectiveness. No single EIS considered both mitigation effectiveness and uncertainty. EISs were selected from a single state or province within the countries marked with an asterisk (*) in the legend. ............. 43 Figure 2.4 The proportion EISs reporting on each jurisdiction that consulted stakeholders to various degrees. EISs were selected from a single state or province within the countries marked with an asterisk (*) in the legend. Refer to Table A.4 for a description of the different categories of consultation in the legend. ........................................................................................................ 45 Figure 3.1 Cumulative impact maps for four ecosystem services (aesthetics, coastal protection, commercial demersal and commercial pelagic fisheries), with associated bar graphs of causes of impact. Maps display the summed impact of all drivers and stressors to each ecosystem service; bar graphs show total impact values for each activity or stressor. Red bars indicate impact only accounting for biophysical risk criteria, and black bars indicate impact accounting for all risk criteria. Coastal protection is not to scale to allow for visibility. ................................................. 62 Figure 3.2 Cumulative impact maps for four ecosystem services (recreation, energy, finfish and shellfish aquaculture), with associated bar graphs of causes of impact. Maps display the summed impact of all drivers and stressors to each ecosystem service; bar graphs show total impact values for each activity or stressor. Red bars indicate impact only accounting for biophysical risk criteria, and black bars indicate impact accounting for all risk criteria. Aquaculture sites are not to scale to allow for visibility........................................................................................................ 63 Figure 3.3 Density histograms of per-cell Ic values for each ecosystem service. Ecosystem services are: A) aesthetics, B) coastal protection, C) commercial demersal fisheries, D) commercial pelagic fisheries, E) marine recreation, F) potential wave and tidal energy, G) finfish xviii  aquaculture, and H) shellfish aquaculture. Red histograms indicate impact only accounting for biophysical risk criteria, and black histograms indicate impact accounting for all risk criteria. .. 65 Figure 3.4 The risk posed by future climate change risks and oil spills on six ecosystem services, compared with current climate change risks. Points represent mean risk scores, error bars represent 25th and 75th percentiles, and lines connecting points demonstrate the trajectory of risk from current conditions to future conditions. ............................................................................... 66 Figure 3.5 The perceived importance of risk criteria to exposure and consequence. Points and error bars represent mean and standard deviations of the distribution of relative importance of risk criteria. ................................................................................................................................... 67 Figure 3.6 The proportion of each type of impact pathways (direct, both direct and indirect, indirect, no impact, and unsure) from four categories of activities and stressors to the eight ecosystem services, as indicated by experts. ................................................................................ 69 Figure 4.1 Three basic structures of impact networks. Leverage nodes are outlined in red. Faded nodes represent nodes that are minor contributors to risk. ........................................................... 90 Figure 4.2 The impact network linking drivers, stressors, species and ecosystem services of the categories listed on the right (a given category may have multiple items, e.g., the driver “fisheries” includes seine fisheries, gillnet fisheries, and others). Numbers in the nodes correspond to the codes on Supplementary Tables C.2 and C.3. Nodes focused on by the Herring Integrated Fisheries Management Plan are outlined in black, while the top three revealed leverage nodes are outlined in blue. .............................................................................................. 92 Figure 4.3 The probabilities of low, medium, and high risk for four species depending on the five different scenarios. White represents low risk, medium grey represents medium risk, and dark grey represents high risk. .............................................................................................................. 95 xix  Figure 4.4 The number of A) species and B) ecosystem services predicted to be at low, medium, and high risk from the five different scenarios. ............................................................................ 97 Figure 5.1 Homogeneity of expert rankings. Bars represent homogeneity scores before (grey) and after (red) the group workshop. .................................................................................................. 119 Figure 5.2 Probability distribution functions (PDFs) representing expert-derived estimates of impact from interviews (grey), and following the group workshop (red). Following the workshop, experts were more likely to provide wider intervals estimating impact from specific risks to ecosystem services compared to when interviewed in isolation. Each row represents the paired estimates for a single expert (1-14). Experts are grouped by the ecosystem service of their expertise. ..................................................................................................................................... 124 Figure 6.1 The location of Tasman and Golden Bays, at the north end of the South Island of New Zealand. ....................................................................................................................................... 135 Figure 6.2 A network of pathways of impacts from climate change and commercial fishing to the four ecosystem service types. A) This hive plot shows the driver of change (climate change, at top), leading to various stressors (lower right axis) shown in orange edges, which then impact the four ecosystem service types (lower left axis) shown in blue edges. B) This hive plot shows the driver of change (commercial fishing, again at top), leading to various stressors (lower right axis) shown in orange edges, which then impact the four ecosystem service types (lower left axis) shown in blue edges. The thickness of each edge represents how many experts mentioned each link. The nodes along each axis are organized by ranking the nodes with the highest number of linked edges to the lowest (highest number of links on the outside). ..................................... 145 Figure 6.3 A network of pathways of impacts from various drivers through sedimentation and pollution to the four ecosystem service types. A) This hive plot shows the drivers of xx  sedimentation (at top), leading to the stress (sedimentation, lower right axis) shown in orange edges, which then impact the four ecosystem service types (lower left axis) shown in blue edges. B) This hive plot shows the drivers of pollution (again at top), leading to the stress (pollution, lower right axis) shown in orange edges, which then impact the four ecosystem service types (lower left axis) shown in blue edges. The thickness of each edge represents how many experts mentioned each link. The nodes along each axis are organized by ranking the nodes with the highest number of linked edges to the lowest (highest number of links on the outside). ........... 147 Figure 6.4 Elicited impact curves of various stressors to (existence value from) biodiversity from the 15 experts that attended the workshop. Violin plots show density histograms overtop of boxplots indicating median (white dot), first and third quartile (box) and max and minimum value (whiskers). Impact scores are on the y-axis, and individual expert number on the x-axis. Violins coloured green represent biodiversity experts, while yellow represents fisheries experts, grey represents aquaculture experts, and blue represents marine recreation experts. ................ 148 Figure 6.5 Elicited impact curves of various stressors to shellfish aquaculture from the 15 experts that attended the workshop. Violin plots show density histograms overtop of boxplots indicating median (white dot), first and third quartile (box) and max and minimum value (whiskers). Impact scores are on the y-axis, and individual expert number on the x-axis. Violins coloured green represent biodiversity experts, while yellow represents fisheries experts, grey represents aquaculture experts, and blue represents marine recreation experts. ......................... 150 Figure 6.6 Elicited impact curves of various stressors to fisheries from the 15 experts that attended the workshop. Violin plots show density histograms overtop of boxplots indicating median (white dot), first and third quartile (box) and max and minimum value (whiskers). Impact scores are on the y-axis, and individual expert number on the x-axis. Violins coloured green xxi  represent biodiversity experts, while yellow represents fisheries experts, grey represents aquaculture experts, and blue represents marine recreation experts. .......................................... 152 Figure 6.7 Elicited impact curves of various stressors to marine recreation from the 15 experts that attended the workshop. Violin plots show density histograms overtop of boxplots indicating median (white dot), first and third quartile (box) and max and minimum value (whiskers). Impact scores are on the y-axis, and individual expert number on the x-axis. Violins coloured green represent biodiversity experts, while yellow represents fisheries experts, grey represents aquaculture experts, and blue represents marine recreation experts. .......................................... 154 Figure 6.8 Cumulative impact curves aggregated from individual impact curves elicited from each expert for each of the four ecosystem service types. Density histograms are depicted on top of boxplots indicating median (white dot), first and third quartile (box) and maximum and minimum value (whiskers). Impact scores are on the y-axis, and individual expert number on the x-axis. Violins coloured green represent biodiversity experts, while yellow represent fisheries experts, grey represent aquaculture experts, and blue represent marine recreation experts. The numbers under each violin indicate how many activities and stressors contributed to each cumulative impact curve. ............................................................................................................ 156 Figure B.1 Side by side comparison of impact maps considering all risk criteria (maps on the left) versus only considering biophysical criteria of risk which only assesses impact to ecosystem service supply (maps on the right). Map pairs are for A) aesthetics, B) coastal protection, C) commercial demersal fisheries, D) commercial pelagic fisheries, E) coastal recreation, F) potential renewable energy, G) finfish aquaculture, and H) shellfish aquaculture. .................... 231 xxii  Figure C.1 Food web and habitat relationships between the species in the coastal British Columbia case study. Habitat relationships are represented by dashed lines and trophic relationships are represented by solid lines. ............................................................................... 237  xxiii  Acknowledgements I don’t know where to begin. I began my doctoral studies inheriting a supportive department and peer group, as I conducted my studies in the same department I completed my Master’s in, and developed a larger support group throughout my tenure. I would like to thank my supervisory committee for their steadfast mentorship and encouragement, and for challenging me to develop and communicate my ideas beyond the extent I would have alone. My supervisor Dr. Kai Chan provided unbelievable time and dedication to me, and trusted me enough to pursue my varied research interests in ways I thought suitable. I benefitted immensely from the guidance of Dr. Terre Satterfield, whose professionalism, perspective, and honest feedback I’ve admired since I started my graduate career. Dr. Ben Halpern strengthened the work in this dissertation substantially, and even played the contrarian in order to force me to anticipate counter-arguments. He’s also an incredibly calm, down-to-earth person considering the level of responsibility he’s weighed with. Thanks to all my coauthor for their hard work, including Janson Wong, Alice Guimaraes, Gustavo Yunda-Guarin, Megan Mach, Cathryn Clarke Murray, Rebecca Martone, Edd Hammill, Hugh Possingham, Jonathan Rhodes, Eve McDonald-Madden, Jim Sinner, Joanne Ellis, and Milind Kandlikar. This work stands as it is because of you. I am indebted to my graduate school cohort, many of whom I knew since my Master’s, for their support as colleagues and friends. Thanks to Andres Cisneros, Sarah Klain, Sara Elder, Emily Anderson, and Jordan Levine for their years of support. More recent colleagues whom I benefitted from include Guillaume Peterson St-Laurent, Bernardo Ranieri, Johnnie Manson, Joey xxiv  Bernhardt, Paige Olmsted, Adrian Semmelink, Ian Eddy and Laura Dee. I have also benefitted from being part of Dr. Kai Chan’s lab group and taking part in various research discussions with many bright people. I would like to single out Jordan Tam and Jackie Lerner for their advice, support, and status as two of my best friends.  In some ways my greatest growth, both intellectually and emotionally, came outside the University. I cannot imagine reaching this stage without the care, compassion, strength, and patience of Camila Romero Fujiwara. I can’t think of anyone else who could withstand my venting while challenging me as she does. She has introduced me to more new ideas, purviews, and philosophies than any other person I have ever met.  Finally, my gratitude goes out to my family and friends (in and outside UBC) who are too numerous to name without it being onerous. Thank you for the seemingly endless hours of companionship, consoling, and celebration through my program. Funding, in-kind contributions, and data agreements supporting for this work came from various sources. I am grateful to the Natural Science and Engineering Research Council of Canada, the Pacific Institute for Climate Solutions, the University of British Columbia, the Centre of Excellence for Environmental Decisions for grant support. The Cawthron Institute, the Bren School at the University of California, Santa Barbara, and the National Center for Ecological Analysis and Synthesis kindly provided work-space and in-kind support for research. Thanks also to the World Wildlife Fund Canada for providing important data for Chapter 3 of this dissertation.  xxv  Dedication A few months ago, at a small market stall, I unexpectedly ran into an emeritus professor at UBC, selling home-made soaps. He asked me about my research, and after I responded, recounted how his contemporaries saw him as an oddity when he started his research career “because no one was doing human ecology at the time, or knew what a human ecologist was”.  I have never cared to label my identity, and less so with conviction. I have always found the fascination with membership to academic disciplines unsettling and unnecessarily constraining: it’s more than a profession, it’s an identity. Yet here was a soap-vender, whose past life helped raise the academy’s interest in my research topics, proudly asserting his research identity and considering me a colleague. Much of my past and current work investigates the human relationship with the environment. If an artisan can self-identify as a scientist, perhaps my concerns of constraints were excessive. If he wants to consider me a colleague, I can comply. So now when I’m asked the inevitable “what kind of researcher are you?” I say “a human ecologist”, though somewhat begrudgingly. At least I have a story to recount now. To anyone who is uncertain about their identity while their coworkers probe for labels: you aren’t the only one, and you’ll figure it out.     1  Chapter 1: Introduction For each man kills the thing he loves, Yet each man does not die. - Oscar Wilde, The Ballad of Reading Gaol That humans must often live as both the beneficiaries and agents of risk to those things we value seems simultaneously an irony and truism. Humans as a whole are thriving under increasing environmental impact, yet some populations are negatively affected and continued environmental impact may continue to threaten human well-being (MA 2005; Raudsepp-Hearne et al. 2010). Humans have and continue to cause such impact to the environment that some geologists petition to call the current epoch “The Anthropocene” (Crutzen 2006). The environmental force of humanity is the result of cumulative impacts rather than any single one (Sanderson et al. 2002; Halpern et al. 2008b; Doney 2010).  Cumulative impact assessment is a field of accelerated growth, advancing both methodologically and conceptually (Halpern et al. 2008b; Teck et al. 2010; Kelble et al. 2013; Brown et al. 2014). However, impact assessment has historically only focused on biophysical characteristics of the environment, such as species and habitats, unnecessarily limiting its value for research and policy (Rounsevell et al. 2010; Kelble et al. 2013). The natural environment renders diverse benefits to people, providing necessary resources for life, supporting and regulating a comfortable environment, and providing opportunities for recreation and community building (MA 2005). These processes are often referred to as ecosystem services. As ecosystem services are characteristics of our relationship with the natural environment, they can be changed by the same processes that impact the natural environment, but also through changing the social 2  considerations that contribute to ecosystem services (Tallis et al. 2012; Kelble et al. 2013). This dissertation seeks to understand cumulative impacts to ecosystem services, and advance methodologies to assess these impacts. 1.1 Relating Ecosystem Services to Cumulative Impact The widely understood human dependency on ecosystems has recently been studied formally in the field of ecosystem services (Ehrlich and Mooney 1983). The ecosystem services concept gained notoriety in studies seeking to quantify total value of ecosystems to people (Costanza et al. 1997; Loomis et al. 2000; Boyles et al. 2011). These studies were successful in terms of raising awareness of the diverse values people have associated with the environment and a sense of the scale of the benefits people derive from the environment. However, these studies were also criticized for their accounting methodologies, and their limited ability to contribute to environmental decision-making and address policy tradeoffs (Rees 1998; Bateman et al. 2011; Fisher and Naidoo 2011).  Moving beyond the initial emphasis of static valuation of nature, recent studies of ecosystem services have sought to understand how they change as a result of changes in human activity or infrastructure (Chan et al. 2006a; Naidoo and Ricketts 2006). Multiple frameworks have been proposed to understand change to ecosystem services in a causal, sequential series of steps (Rounsevell et al. 2010; Atkins et al. 2011; Kelble et al. 2013; Mach et al. 2015). These frameworks are direct adaptations of frameworks of change to ecosystems, highlighting the impacts of human activities through the natural environment as the primary (or sole) regulators of change to ecosystem services (Kristensen 2004; Niemeijer and de Groot 2008).  3  The representation of change to ecosystem services has improved considerably over the past decade. A prominent method to characterize changes in ecosystem services is using proposed “production functions”, whereby biological and social variables act as inputs in a mathematical model to “produce” ecosystem services (Chan and Ruckelshaus 2010a). These approaches have been successful in exploring consequences of particular policy options, usually regulating the change in a particular input (or suite of inputs), and exploring consequences for ecosystem services (Guerry et al. 2012; Boumans et al. 2015). The factors that are considered to be inputs into production functions are few and emphasize specific biophysical aspects of ecosystems, neglecting potential effects to the broader environment and potentially neglecting impacts to the service metrics (measures related to people’s ability to acquire ecosystem services) and value metrics (measures related to the preferences people have for different ecosystem services) of ecosystem services (Tallis et al. 2012, Figure 1.1). These production functions therefore do not allow for a systematic identification of prominent drivers of impact nor a full accounting of impact. As they are usually focused on particular policies, they are often suited for exploring particular impacts to ecosystems (Nelson et al. 2009), and less for exploring cumulative impacts on ecosystem services. 4   Figure 1.1 Conceptual model showing cumulative impacts to ecosystem services. Ecosystem services are an interaction between the biophysical and social environment, and can be considered along different points. Supply considers only the biophysical system that produces the service of interest, while service consider the access of people to that supply, and value considers people’s preferences. Realizing the benefits of ecosystem services relies on all three (represented by blue arrows). Human activities can impact ecosystem services through each of the three points (represented by red arrows). This model modifies a model of the ecosystem service production chain found in Tallis et al. (2012).   Despite recent improvements, more effort is needed to characterize how services are at risk due to the simultaneous action of many human activities and structures (Allan et al. 2013; Kelble et 5  al. 2013; Cook et al. 2014a). An inability to address this kind of change not only limits the utility of existing treatments of ecosystem service change, but also limits the utility of the ecosystem service concept itself in influencing environmental policy. Without an exploration of cumulative impacts to ecosystem services, analysis runs the risk of misallocating attention away from prominent drivers of change (Jackson et al. 2001), or underestimating the total change resulting from diffuse causes (Cook et al. 2014a). Without an ability to address cumulative impacts to ecosystem services, environmental management may be limited in protecting and sustaining ecosystem services. 1.2 Relating Cumulative Impact Assessment to Ecosystem Services In response to the widespread recognition of many diverse drivers of environmental change, a suite of cumulative impact frameworks have been proposed (Beanlands and Duinker 1983; Duinker et al. 2012). Many frameworks for assessing cumulative impacts have been proposed in a variety of ecosystems (e.g. aquatic systems, forests, etc.), and more recently literatures devoted to understanding cumulative impacts generally (Halpern et al. 2008b; Niemeijer and de Groot 2008; Stelzenmüller et al. 2010; Brown et al. 2014). Studies of cumulative impact have generally focused on impacts to biophysical components of systems (species or habitats), falling short of ecosystem services. (Niemeijer and de Groot 2008; Halpern et al. 2009; Kappel et al. 2012; Knights et al. 2013a; Micheli et al. 2013; Knights et al. 2015). Conceptual treatments have emphasized the linkage between drivers (human activities), stressors (the mechanisms by which activities cause impact), and the biophysical environment (Kristensen 2004; Atkins et al. 2011). Empirical investigations typically quantify the 6  consequence of multiple stressors on species (Crain et al. 2008; Darling and Côté 2008; Halpern et al. 2008a). Mapping and modeling approaches have also typically focused on representing total impact on different habitat types (Halpern et al. 2009; Kappel et al. 2012; Murray et al. 2015a; Clark et al. 2016). But the relationship between the biophysical environment, whether representing habitats, species, or biological communities, and ecosystem services is not straightforward (Duffy 2008). By limiting a representation of change to the biophysical environment, most existing treatments have limited applicability to address change to ecosystem services (Tallis et al. 2012).  Management plans with goals beyond biodiversity conservation (such as for resource conservation, managing aesthetically pleasing landscapes, and maintaining recreational areas) are better served by assessing impact to ecosystem services than the biophysical environment alone (Granek et al. 2009). Assessments that only focus on biophysical impact fails to distinguish between ecosystems and the valued services that derive from them. At best, they assume that ecosystem change is a suitable proxy for ecosystem service change, which is a questionable assumption. Representing change to the biophysical environment can only capture change to the supply metrics of ecosystem services, and neglects changes to the service and value metrics (Tallis et al. 2012). There are many human activities and stressors that can negatively affect the delivery of services to people, such as limiting people’s access to a service (Chan et al. 2012a; Chan et al. 2012b). For example, zoning a particularly productive part of the coast for aquaculture can limit people’s ability to fish there, as trespass laws may prevent them from accessing that area (Wieland et al. 2016). There are also mechanisms that can affect people’s ability to enjoy an ecosystem service, regardless of whether or not the supply of the service or 7  people’s ability to access them is affected (Chan et al. 2012a; Chan et al. 2012b). If a pollutant changes the taste of shellfish without affecting its growth, and it does not initiate a harvest ban, the pollutant may decrease people’s enjoyment of the shellfish.  In initial attempts to move beyond impact assessments to biophysical ecosystems, some studies have begun to address cumulative impacts to ecosystem services (Allan et al. 2013; Cook et al. 2014a; Biggs et al. 2015). These studies have advanced cumulative impact assessment by representing ecosystem services as the recipient of impact. In the case of Allan et al (2013), which produced hotspot maps of cumulative impacts, maps of ecosystem services were produced where traditionally this approach captured impacts to different ecosystem types. The study by Cook et al. (2014) related multiple human activities to ecosystem services. Biggs et al. (2015) develop a database to study multiple drivers of ecosystem collapse and regime shifts with impacts to the supply of ecosystem services. Analyses on this database have produced important insights, such as linking multiple drivers to regime shifts and ecosystem services, and providing evidence for cascading regime shifts – where any particular regime shift can amplify the impacts from drivers of other regime shifts (Rocha et al. 2015a; Rocha et al. 2015b). Downing et al. (2014) created a model of social-ecological system of Lake Victoria to identify drivers of change to the ecosystem that affects ecosystem services. However, these treatments relied on impact frameworks that assume that impact to ecosystem services is regulated primarily through changes to the biophysical environment (Teck et al. 2010; Kelble et al. 2013; Biggs et al. 2015). Frameworks to understand and assess cumulative impacts to ecosystem services requires a treatment of impact that goes beyond representing change to the biophysical environment. 8  1.3 Problem Statement The issue of cumulative impacts to ecosystem services gets to the heart of environmental management, spatial planning, ecosystem based management, and many other applications of environmental science. Currently, these fields of management are operating without an understanding of how ecosystem services are affected by multiple human activities that are being managed (Mach et al. 2015). Many ecosystems support multiple uses, and coastal systems in particular, are prone to multiple uses because these areas typically have the highest density of humans (UNEP 2012b). In response, multiple fields have emerged to address planning for multiple uses. Marine spatial planning aims to regulate the activities that occur in certain places to alleviate cumulative impacts (Foley et al. 2010). Conservation planning hopes to establish systems of protected areas to protect ecosystems and biodiversity, and should avoid areas under high cumulative impact (Margules and Pressey 2000). Ecosystem based management is an approach to management that considers the suite of key interactions between humans and the environment, including cumulative impacts that threaten important environmental components (McLeod and Leslie 2009). Successful application of any of these approaches to environmental management will benefit from effective approaches to assess cumulative impacts.     Establishing effective assessment procedures of cumulative impacts to ecosystem services can benefit from an understanding of how effective current assessment methods are, as well as advancing our understanding of the dynamics of cumulative impacts and ecosystem services. How effective are our current methods of assessing impact? How important are social dimensions of risk relative to biophysical? Are there prominent threats across ecosystem services, or is service threatened by a unique set of threats? How can we plan research and 9  management in the face of cumulative impacts to ecosystem services? These are the primary questions that this dissertation seeks to address. 1.4 Existing Frameworks to Build From Advancing the nascent field of cumulative impacts to ecosystem services is dependent on advancing research frameworks and methodologies to generate insights. While existing frameworks may not fully capture the dynamics to understand cumulative impacts to ecosystem services, they provide benchmarks that can be modified to generate new insights (Rounsevell et al. 2010; Atkins et al. 2011). Though there are numerous limitations with each approach that prevents them from relating cumulative impacts to ecosystem services, methods developed in this dissertation adopt and modify existing frameworks explained below.   Frameworks for addressing cumulative impacts are necessary to highlight linkages between multiple activities, their inputs and impacts to important aspects of the ecosystem, the stakeholders likely to be affected, and the environmental and political boundaries that define scales of governance (Rounsevell et al. 2010; Atkins et al. 2011; Kelble et al. 2013). They provide a common structure upon which to base projects and planning. All frameworks, whether explicit or not, emphasize some aspect of cumulative impacts while understating others. Some are specific to particular aspects of cumulative impact, such as exploring management actions in response to impact interaction, while others are larger in scope. Table 1.1 organizes many prominent frameworks by their ability to represent the various components of impact to ecosystem services. 10  Table 1.1 Dimensions of cumulative impacts to ecosystem services that existing research frameworks represent and emphasize.   Framework Impact PathwaysRecipient of Impact - BiophysicalRecipient of Impact - SocialResponse PlanningTemporal - Past ImpactsTemporal - FutureTemporal - PeriodicitySpatial - cross scale stressorsSpatial - overlapping StressorsImpact AccumulationImpact Interaction Key ReferencesCumulative Impact mappingHalpern et al. (2008); Teck et al. (2010); Ban et al. (2010); Murray et al. (2015)Process Based Models - Atkins et al. (2011) Atkins et al (2011)EBM-DPSIR Kelble et al. (2013)FESP Rounsevell et al. (2010)Process Based Models - Mach et al. (2015) Mach et al (2015)Stressor Mapping Sanderson et al. (2002)eDPSIR Niemeijer and De Groot (2008)EBM-DPSIR network Cook et al. (2014)Networks - Knights et al. (2013) Knights et al. (2013)Interaction Modeling Brown et al (2013)Species modeling Kaplan et al (2010)Risk assessments - Samhouri and Levin (2012) Samhouri and Levin (2012)Risk assessment - Knights et al. (2015) Knights et al. (2015)Hierarchical Response Framework Smith et al. (2009)Press-Pulse Dynamics Collins et al. (2011)Cumulative Impact planning Spyce et al. (2009)Cultural Loss Framework Turner and Turner (2008)11  1.4.1 Ecosystem Models While there are many system-specific models of ecosystems and impact to them, there are fewer modeling frameworks to investigate cumulative impacts. Two prominent marine ecosystem modeling frameworks are Ecopath with Ecosim (Christensen and Walters 2004) and Atlantis (Sainte-Marie et al. 2010). These modelling frameworks allow for researchers to incorporate species interactions in cumulative impact assessment by incorporating environmental forcing functions that influence the population-level variables associated with specific species. Ecosystem models typically do not represent the causal processes by which human activities and stressors cause impact, have limited temporal dynamics, and do not account for the social criteria of impact. Additionally, ecosystem models typically do not represent interactive impacts (synergistic and antagonistic impacts), are not typically spatial in representation, and do not explicitly consider management responses to impacts (Gregr and Chan 2014). In this dissertation I1 incorporate species interactions as part of a risk assessment to address indirect processes that cumulative impacts can affect species through their ecological dependencies.  1.4.2 Ecological Risk Assessments Ecological risk assessments are typically done in contexts of high uncertainty, useful to understand emerging stressors and other situations where high quality data may not exist. Risk assessments are another category of framework that have many species and system specific                                                  1 Though this dissertation as a whole is my work, I have benefitted from numerous and diverse collaborations through my tenure as a graduate researcher. In the introduction and conclusion chapters I use the first person singular pronoun (“I”), but in the primary research chapters I use the first person plural pronoun (“we”) to acknowledge the coauthorship of these components, which varies across chapters. 12  models, though Samhouri and Levin (2012c) proposes a risk assessment framework that is adaptable to any species of value that faces multiple stressors. By scoring exposure and sensitivity (with levels of uncertainty) of a given species to various human activities, this framework allows for an assessment of risk across activity types. Knights et al. (2015) expands on this risk assessment methodology by linking risk and measures of the recovery lag time to specific causal chains of impact (and not broadly to human activities). The ecological risk assessment processes described here do not exclusively consider impacts across space and time, do not account for impact interaction, do not calculate total impact, do not account for social criteria of impact, and do not explicitly consider responses to impacts. In this dissertation, I use a risk assessment approach inspired by Samhouri and Levin (2012) and Knights et al. (2015) to generate risk weights as part of an analysis to identify components of a system of cumulative impacts to prioritize for management (Clarke Murray et al. 2016).  1.4.3 Stressor Footprints Stressor footprint analysis is the simplest spatial analysis of cumulative impacts. In fact, it does not represent impacts but rather the stressors that contribute to impacts. Using proxies of stressors such as human population density and human activity, stressors are mapped and their overlap and range quantified. This approach analyzes areas with the greatest stressor overlay. It has been applied with mapped output, allowing for determination of where stressors are most felt (Sanderson et al. 2002), and without mapping, providing estimates of spatial coverage of activities and stressors (Leu et al. 2008). This framework is limited as it does not account for causal processes of impact, nor does it represent temporal dimensions of impact, social criteria of impact, impact interaction, or explicitly consider the responses to impacts. In this dissertation I 13  use the overlay approach as part of an analysis to identify areas in a map where different ecosystem services are most impacted. 1.4.4 Interaction Determination This research methodology is principally interested in exploring the consequences of different management actions to cumulative impacts under different scenarios of impact interaction (additive, synergistic, antagonistic) on a particular impact receptor. These different interactions are incorporated into mechanistic models of impact with multiple stressors, generating results of total predicted impact and predicted impact under specific management actions to restrict particular stressors (Brown et al. 2013). This method has been used in multiple seagrass studies (Brown et al. 2013; Brown et al. 2014). This is a very specific framework that does not account for the causal processes of impacts, social criteria of impact, explicit consideration of responses to impact, and temporal dimensions of impact. In this dissertation I use an interactive model of impact as part of an analysis to calculate cumulative impacts where total impact does not equal the sum of individual impacts, under conditions where ecosystem services have a threshold of collapse. 1.4.5 Cumulative Impact Mapping Cumulative impact mapping is one of the most widely used and developed cumulative impact frameworks. Applied first to global oceans (Halpern et al. 2008b), cumulative impact mapping has been applied regionally, including the California current (Halpern et al. 2009), the Mediterranean and Black sea (Micheli et al. 2013), British Columbia (Ban et al. 2010), New Zealand (Clark et al. 2016), Australasia (Brown et al. 2014), Hawaiian islands (Selkoe et al. 14  2009), Massachusetts (Kappel et al. 2012), the Baltic sea (Korpinen et al. 2012), the Great Lakes (Allan et al. 2013), the North sea (Heinanen et al. 2013), among others. Cumulative impact mapping has an impressive pedigree, often with each successive application advancing the methods. The original application built on stressor mapping by overlaying maps of human activities and stressors with different habitat types, and applying a vulnerability score of each habitat to each stressor type (Halpern et al. 2008). This method was developed to assess where impacts are felt most, what habitats are under greatest impact, and what human activities and stressors contribute to impact. In the years following the global analysis, researchers added to and enhanced the applicability of the method. Decision science tools were used to generate widely applicable vulnerability measures (Teck et al. 2010), zones of influence based on likely area of impact from a given stressor replaced physical footprint (Ban et al. 2010), interactive impacts have been modeled in place of additive ones (Brown et al. 2014), and ecosystem services modeled in place of habitats by using human uses as proxies of ecosystem services (Allan et al. 2013). This mapping approach has even been used to assess impact from planned developments, thus addressing future impacts (Murray et al. 2015b). Despite these refinements, the method seems applicable for regional scale (i.e. the California current) and global scale assessments and not local scale. At the larger scales (i.e. the Baltic Sea) the results of cumulative impact mapping have been verified by measures of biodiversity and ecosystem status (Andersen et al. 2015), but no relationship was found between mapping results and local environmental condition in Tauranga Harbour, New Zealand, likely reflecting the often poor quality of spatial data at local scales (Clark et al. 2016).  15  Cumulative impact mapping is a data-intensive approach to cumulative impact assessment, and the results will depend on the availability and quality of data input (Halpern and Fujita 2013). It is also primarily a spatial approach, and will be able to answer questions about space with greater resolution than other dimensions. For example, even though future impacts have been assessed with this method, only projects with approvals can be incorporated because spatial data exists for these (Murray et al. 2015a). Cumulative impact mapping has still not been able to account for other likely future projects that have not yet been approved. This framework does not account for the causal processes of impact, nor does it consider social criteria of impact, nor does it explicitly consider the responses to impact. I rely heavily on this framework in this dissertation. Chapter 3 of this dissertation advances cumulative impact mapping directly, and given the prominence of this framework I compare other methods I develop to this framework, and consider how other methods I use can be used in conjunction with cumulative impact mapping. 1.4.6 Process Based Models Process based models are another impact framework with widespread uptake and gradual development. Instead of a spatial focus, process models are systems-based analyses focused on the interactions of people and the environment. The most widely known formulation for assessing impacts is the DPSIR framework (Drivers of change – Pressures on the environment – State change in the environment – resulting Impact – Response to impact). Originally adopted by the European Environmental Agency as an extension to the pressure-state-response model developed by the Organization for Economic Cooperation and Development (OECD), the DPSIR framework has been applied by researchers and government agencies worldwide (Kristensen 2004; Atkins et al. 2011; Kelble et al. 2013).  16  As a conceptual and organizing model, DPSIR has limited direct application to spatial and temporal analysis, though it may be possible to incorporate past and future impacts into its structure. Many major advances in DPSIR modeling have come through an expansion of what the “state change” variable is. Multiple independent frameworks exist adapting the DPSIR approach to ecosystem services. Atkins et al. (2011) expands the original DPSIR cycle by linking biophysical ecosystem change to ecological processes that produce ecosystem services. Kelble et al. (2013) introduces the EBM-DPSIR framework to link changes to the environment to effect (both positively and negatively) different categories of ecosystem services. Rounsevell et al. (2010) proposes the FESP (Framework for Ecosystem Service Provision) framework which embeds ecosystem services within the DPSIR cycle, explicitly considering ecosystem service beneficiaries (ESBs) and ecosystem service providers (ESPs). According to this framework, impact to ecosystem services is dependent on who is considered as ESBs, as the makeup of ESBs will affect the demand for ESPs. Similarly, Mach et al. (2015) propose a modified DPSIR model that addresses impacts to service production that are demanded by beneficiaries, who in turn modify management based on their value of services. While these models provide a detailed relational perspective of particular human pressures to and human benefits from the environment, they have limited explicit scope in other ways. The model processes typically evaluate a particular stressor on a particular ecosystem service, in detail (Niemeijer and de Groot 2008). While researchers acknowledge that multiple process pathways exist in unison, typically these frameworks are more amenable to assessing individual causal processes (Kelble et al. 2013). These frameworks do not address the social criteria of impacts, nor do they account for spatial or temporal dimensions of impact. They are not frameworks that 17  focus on impact accumulation or impact interaction. In this dissertation I adapt the sequential, causal ontology of this framework as part of analyses to understand prominent causes of impact to ecosystem services in many of my chapters.  1.4.7 Network Analysis Network analysis is a method that specifically analyses the connections among various elements in a system. It is often used to identify important nodes in a system that regulate system level behavior (Niemeijer and de Groot 2008). Network analysis has been used to link interacting stressors on coral reef communities to determine the dominance of climate change stressors relative to others (Ban et al. 2014a). But network analysis as also been adopted to analyze multiple processes of impact concurrently.  Researchers have developed network analysis to describe impact pathways using the ontology of the DPSIR framework, thus adapting a well-known causal framework for systems analysis. This approach effectively expands process based models by allowing for analysis on multiple receptors by multiple drivers and pressures, though in practice it often deemphasizes detailed analysis on any given impact pathway. Niemeijer and De Groot (2008) proposes the enhanced-DPSIR (eDPSIR) model to collate individual impact process pathways into an impact network, with the hope of identifying key nodes within the network. Cook et al. (2014a) builds networks using the EBM-DPSIR model to highlight the most prominent activities to ecosystem services. Knights et al. (2013a) develops network analysis to group similar impact mechanisms in order to highlight common stressors to allow management to target groups of stressors simultaneously. management can target with simple measures. Similarly, some authors borrow from the resilience literature to use causal loop diagrams (CLD) to characterize co-occurring drivers of 18  cumulative impacts (Biggs et al. 2015). These CLDs allow for an analysis of primary drivers of ecosystem change linked to ecosystem service impact (Rocha et al. 2015a; Rocha et al. 2015b). Downing et al (2011) used qualitative network analysis to characterize change in Lake Victoria based on multiple causal pathways of impact, with consequences to ecosystem services. Current network analysis frameworks do not address the social criteria of impact, nor do they account for temporal and spatial dimensions of impact, nor impact interactions. In this dissertation I use network analysis as part of larger studies to understand prominent causes of cumulative impacts and identify the activities and stressors to prioritize for management to regulate impact.  1.4.8 Stressor Scale Frameworks Stressors operate over different scales, and cumulative impacts often have discrete “pulse” stressors causing impact in the context of continuous “press” stressors (Smith et al. 2009; Collins et al. 2010). Smith et al. (2009) proposes the Hierarchical Response Framework (HRF) to organize “pulse” and “press” stressors according to the scale at which the affect ecosystems (from individual change to species change within ecosystems to species loss among ecosystems). Collins et al. (2010) borrows the temporal framing of the HRF to propose the Pulse-Press Dynamics (PPD) framework of socioecological systems. The PPD is meant as a multidisciplinary research organization framework, linking biophysical and social disciplines and methods to address key issues in environmental change. Similar to process based frameworks discussed previously, the PPD relates changes to the biophysical environment, which in turn have consequences for ecosystem services that affect human well-being resulting in changes to policy which cycle to influence stressors. The PPD focusses on the interaction of pulse and press stressors to affect the environment. Being temporal-scale frameworks, these frameworks are not 19  explicitly spatial, and provides no immediate framework to quantitatively estimate cumulative impact. These frameworks also do not account for social criteria of impact. In this dissertation I consider press disturbances, mostly in the form of climate change, alongside anthropogenic activities that often contribute to pulse impacts, across my analyses in multiple chapters. This is a conscious inclusion to understand how climate change contributes to cumulative impacts, as some prominent treatments of cumulative impacts do not incorporate climate change impacts (Ban et al. 2010).   1.4.9 Cumulative Impact Planning Few frameworks exist in the published literature to explicitly evaluate stakeholder preferences of development trajectories in the context of cumulative impacts, but Spyce et al. (2012) proposes the use to choice experiments to determine stakeholder preference. Choice modeling involves preference ranking among competing scenarios, with tradeoffs among criteria among the scenarios. This framework allows for an assessment of the types of scenarios preferred, as well as an assessment of what criteria are preferred by what stakeholder groups. This framework does not explicitly consider impacts (temporal, spatial, accumulation, interaction, or social). This framework inspired this dissertation to consider cumulative impacts beyond characterizing impacts and consider management prioritization to respond to cumulative impacts.  1.4.10 Past Impact Determination While often not described as frameworks, some studies explicitly understand how past environmental change has shaped the current environment. Many studies use the framing of “shifting baselines” to focus on the fact that current ecosystems, even ones we consider 20  “pristine”, have often had severe human-caused changes (Pauly 1995; Pinnegar and Engelhard 2008). Studies on shifting baselines often compare current ecosystems with past ones, separated by the existence of some major human stressor, or compare areas with and without human stressors, to provide data on the past state of the environment and the contribution of human stressors to this change (Jackson et al. 2001; Knowlton and Jackson 2008). Salomon et al. (2007) was able to construct a timeline of human pressure and environmental change to intertidal Alaskan resources by integrating traditional ecological knowledge, historical fisheries data, archaeological data, and modern field methods. Turner and Turner (2008) propose an organizing framework of cultural loss. This framework does not easily lend itself to quantitative assessment, but is noteworthy as a framework that often considers past and cumulative impacts to cultural identity, community, and health and knowledge. This framework does not address causal processes of impact, nor does it represent impact interaction or accumulation, and it does not explicitly consider responses to cumulative impact. This dissertation is inspired by these frameworks to consider past ecosystem change in understanding current contexts of cumulative impacts. Though this dissertation does not exhaustively explore cultural impact assessment, these frameworks have inspired this framework to explore social criteria of impact in multiple chapters. 1.5 Gaps in Frameworks and Opportunities for Development There are many frameworks that consider specific aspects, and some that take a broad view of cumulative impacts. Frameworks that are explicitly spatial (i.e. mapping approaches) address broader spatial dimensions, and explicitly temporal frameworks address broader temporal dimensions. Integrative frameworks either focus on spatial scope, temporal scope, or 21  mechanisms of impact, and not all explicitly consider cumulative impact (Table 1.1). The few frameworks that account for impact on ecosystem services only consider changes to ecosystem services as a function of changes to biophysical components that produce ecosystem services, and do not allow for impacts to occur through lack of access or degradation to the perceived quality of an ecosystem service at a site. Additionally, there are no identified frameworks that calculate total cumulative impact across multiple causal processes. This dissertation seeks to fill in some of these conceptual gaps by directly address cumulative impacts to ecosystem services, explore the spatial and temporal dimensions of impact, understanding the major criteria for impact, the causes of impact, and addressing management options to reduce impact. 1.6 Considering Data Needs As with organizing frameworks to research, selecting methods to measure impacts to ecosystem services can borrow from the separate literatures of ecosystem services and cumulative impacts. Assessing cumulative impacts to ecosystem services is a budding field, and as such is often faced with extreme data paucity. A lack of data does not necessarily limit a researcher’s ability to perform an assessment, however. What is affected is the quality of information, and will often shift the focus of study away from accurate estimates to understanding limitations and uncertainty of an assessment (Cooke 1991). In lieu of direct data for assessment, researchers are left with proxy data, models, literature-based semi-quantitative scales, and expert elicitation. These can often be used in conjunction.  22  1.6.1 Proxy Data Proxy data is commonly used in mapping studies to represent both stressors and valued components of the environment (Halpern and Fujita 2013). There is rarely exact physical data on the location and extent of stressors on particular receptors, though sometimes there exists data on the activities that cause stressors. When representing ecosystem services, proxies in the form of human activities associated with an ecosystem are often used (Halpern et al. 2009; Ban et al. 2010). While proxies are associated with data of interest, they also only ever partly address the data of interest. For example, activities may not occupy the same space as associated stressors, and footprints of human use may not accurately represent the total area that humans enjoy a service (Ban et al. 2010; Halpern and Fujita 2013). In this dissertation I utilize proxy data of stressors by mapping human activities, with associated modeled areas of influence, as part of an analysis to create maps of cumulative impact on ecosystem services.  1.6.2 Models Models may represent stressors and ecosystem services with greater depth than proxies, though models have associated assumptions. Most research studies model cumulative impact as additive processes, ignoring the possibility of interactive impact (Halpern and Fujita 2013), though Brown et al. (2013; 2014) utilize interactive models for impact. The InVEST models (Integrated Valuation of Ecosystem Services and Tradeoffs) allow for the spatial modeling of diverse ecosystem services, requiring information on human use as well as physical phenomena to represent the location and measure of a given service (Guerry et al. 2012; Sharp et al. 2014). Similarly, the Multiscale Integrated Model of Ecosystem Services (MIMES) allows for the dynamic modeling of multiple ecosystem services, allowing for scenario analysis and valuation 23  (Boumans et al. 2015). Diaz et al. (2011) promote a data framework linking information on ecosystem service providers with information on ecosystem service beneficiaries to generate information on ecosystem services. In this dissertation I use the InVEST models to create maps of ecosystem services as part of an analysis to create maps of cumulative impacts to ecosystem services. 1.6.3 Literature-Based Measures When direct measurements are not available, researchers can often turn to the literature. However, as direct data of impact on ecosystem services is often non-existent, direct reliance on the literature is often unhelpful. In this case an index can be created to represent the data of interest with the literature filling the role of providing justification for index scores. For example, Samhouri and Levin (2012c) use the literature to  score criteria associated with risk (on low-medium-high scales) to assess risk to different species. Willemen et al. (2008) develop a methodological framework to spatially represent ecosystem function and services based on the availability of data. If both direct and proxy data are unavailable, they suggest using decision rules based on literature reviews to decide what landscape data can be used to estimate ecosystem services. In this dissertation I appeal to the literature to compare existing impact assessments from environmental impact assessments with published information relevant to specific impacts. I use this information to assess how diligently existing assessment approaches account for cumulative impacts. I also utilize a risk assessment that calculates risk based on literature derived measures, as well as calculates associated uncertainty based on the availability and certainty of literature to calculate these measures, as part of an analysis to identify priorities for management. 24  1.6.4 Expert Elicitation When data is sparse while analysis is needed for imminent decisions, expert elicitation can provide valuable information (Burgman et al. 2011b). However, as human conduits of information, experts are subject to an impressive arsenal of biases and mental shortcuts that can undermine data collection (Tversky and Kahneman 1974; Gilovich et al. 2002). Often expert based data are more useful as a means to understand uncertainty in a problem than as a way to answer a problem – a claim that may be quite unpopular with managers requiring an answer to a problem (Cooke 1991). A well designed expert elicitation process, however, can gather diverse information with reasonable confidence. In some cases they can provide the only data that can be collected to address a particular problem (Aspinall 2010; Burgman et al. 2011a; Martin et al. 2012). Teck et al. (2010) use expert elicitation to quantify the vulnerability of different ecosystems to different activities, and Cook et al. (2014a) use expert elicitation to understand what activities and stressors interact with what environments to impact ecosystem services. Altman et al. (2010) use a hierarchical decision process with experts to generate relative impact scores of different human activities to ecosystem services. In this dissertation I rely on expert elicitation across multiple studies. I used expert elicitation to quantify risk of multiple human activities and stressors to multiple ecosystem services, adapting an elicitation framework to assess ecosystem vulnerability (Teck et al. 2010). I also develop an elicitation strategy to understand the causal processes of dominant impacts to multiple ecosystem services, and calculate cumulative impact to these ecosystem services. I also advance expert elicitation methods for cumulative impact studies by assessing the effects of group elicitation strategies on expert judgments.  25  1.7 Case Studies for the Dissertation This dissertation contains studies on cumulative impacts on ecosystem services in coastal British Columbia, Canada, and Tasman and Golden Bays, New Zealand. These geographically disparate places nevertheless have socio-ecological similarities: both are temperate coastal ecosystems with similar governance structures and colonial histories, with diverse indigenous, settler, and immigrant populations, and have similar coastal uses (Fisher 1980). But these sites do have important differences, such as being coastal systems in different oceans, with British Columbia having a highly productive coast and Tasman and Golden Bays having a relatively low flow, with low productivity (Ware and Thomson 2005; Handley 2006). These sites also have culturally distinct indigenous and settler populations (Fisher 1980).  The two study systems also represent incredibly different cases of data availability. British Columbia has relatively rich repositories of spatial data to generate models of ecosystem services (BCMCA 2016; GeoBC 2016), and an interest in cumulative impacts, which has generated availability of risk assessments and data on cumulative impacts (O et al. 2015). In contrast, New Zealand has little data available to model or analytically assess impacts to ecosystem services, leaving few sources of data but experts at my disposal. Cumulative impacts to ecosystem services occur all around the world, both in cases where data exists and does not exist, and elicitation strategies should exist to cater to both situations (Tallis et al. 2010). Taken together I believe my case study systems provide a powerful argument that informative cumulative impacts assessments on ecosystem services can be performed in a variety of settings. 26  1.8 Structure of the Dissertation This dissertation is organized into seven chapters and an appendix. The core of the work in this dissertation can be found in chapters two through six, which are manuscripts in the process of submission to research journals (at the time of writing they are either already submitted or being prepared for submission). Given that they are acting both as dissertation chapters as well as stand-alone papers they will contain redundant background and discussion among themselves, and may vary in their format. However, each chapter contains a unique research challenge, and most chapters (save two) have unique datasets and pilot unique (or advance existing) methods.  Chapter 2 addresses the current dominant policy context for assessing cumulative anthropogenic impacts, evaluating the standard of this common practice. Across much of the world, environmental impact assessments (EIAs) are legally mandated to predict, evaluate and suggest mitigation measures in response to environmental impacts (Wood 2003). EIA was formalized by the United States National Environmental Policy Act (NEPA) in 1970, propagated quickly, and is now practiced around the world (Wood 2003). While most EIAs are not required to evaluate impact to ecosystem services, an evaluation of EIAs will provide insight into the current strengths and limitations that pervade impact assessment worldwide, with an explicit consideration of how cumulative impacts are considered. I explore the assessment process in British Columbia (Canada), California (USA), Veracruz (Mexico), Brazil, England and Wales, Queensland (Australia), and New Zealand. A separate literature on EIAs exists, which discusses the limitations of EIAs (Duinker and Greig 2006; Duinker et al. 2012) but a quantitative study of EIAs on this scale is unprecedented. I assess the number of impacts assessed in each EIA and show that the proportion deemed significant for development decisions is negligible. I 27  demonstrate that important steps in the EIA process do not cater to standards of scientific rigor. I provide recommendations for EIAs to improve and confront its most systematic problems. Chapter 3 combines an advanced approach for mapping cumulative impacts to habitat with a leading approach for mapping and modeling the provision and delivery of ecosystem services. Existing analyses have pinpointed hotspots of impacts to marine habitats, and sites that are important for several ecosystem services, but few analyses have assessed the cumulative impacts to multiple ecosystem services, and none have explored the contribution of social criteria of impact (impacts to service and value metrics) to ecosystem services. Using coastal British Columbia as a case study, I generate models of ecosystem services using the marine InVEST modeling platform, and paired this with maps of human activity and expert derived estimates of risk. I assess geographic variation in impacts across ecosystem services, the relative importance of social versus biophysical criteria for ecosystem service risk, and the unique data needs that mapping impact to ecosystem services have over mapping impacts to ecosystem types (the current norm). I show that different ecosystem services can occupy different areas, affecting their exposure to different impacting human activities. I also demonstrate that ecosystem services can be at risk from changes to access and perceptions of service quality, and these kinds of impacts can be as important as impacts to the provision of services. I question the use of standardized activity and stressor spatial data in cumulative impact mapping when assessing ecosystem services, arguing – with empirical support – that different ecosystem services face different risks from the same human activities and stressors.  Whereas Chapter 3 was virtually devoid of ecological interactions, Chapter 4 expands upon an ecological risk analysis framework to incorporate such interactions and pinpoint—using network 28  approaches—leverage points for management to reduce risk to specific valued environmental components. Whereas management often targets few stressors or activities, presumed to be the most effective at managing risk to stocks or species, such stressors and activities have not yet been identified through a structured analysis. To create the networks, I adapt a prominent impact framework the DPSIR impact process model – in a network of impact pathways, to understand prominent mechanisms of risk to coastal species and ecosystem service provision, and prioritize management actions, in British Columbia. Existing network frameworks use the number of connections between process nodes to identify important nodes, I pilot an approach with estimated weights of importance. I use a DPSIR based risk assessment conducted on seventeen species in British Columbia to construct the network and generate estimates of risk propagation, mediation, and accumulation through the network, and utilize a Bayesian Belief Network to analyze various herring management scenarios. I show that current integrated herring management plans fail to remedy stressors that contribute high risk to herring, which could lead to ineffective management results. I propose the methods I pilot be used to prioritize management actions in contexts of cumulative impacts. Whereas Chapters 3 and 4 had rich sources of data to assess cumulative impacts and prioritize response, Chapters 5 and 6 assess cumulative impacts to ecosystem services in Tasman and Golden Bays, where data are sparse. Expert judgment served as for a crucial method to quantify vulnerability of habitats to specific impacts, but when experts are required to identify prominent activities and stressors as well as quantify impact, effects of elicitation design on expert judgements are not well known. In Chapter 5 I investigate the effects of group deliberation in an expert elicitation methodology focused on combating prominent biases in experts on expert 29  responses. Research on expert elicitation is replete in methodologies that counter expert overconfidence when the context of a problem is fully known (Speirs‐Bridge et al. 2009; Morgan 2014). In contexts of cumulative impacts, estimating levels of impact to the most important activities and stressors can only be done when the most important activities and stressors are known. I show that group elicitation methods can lead to more consistent answers by experts regarding the relative importance of specific activities and stressors as well as in the estimation of their impact through reducing linguistic uncertainty and moderating extreme opinions. At the same time, group elicitation effectively reduces overconfidence in individual experts. I suggest that group elicitation methods be a staple for elicitation strategies addressing cumulative impacts because of the dual beneficial effect it has on uncertainty among and within experts. Chapter 6 uses the expert elicitation methodology assessed in Chapter 5 to identify prominent causes of impact and estimate total impact to multiple ecosystem services in Tasman and Golden Bays, New Zealand. There is a pressing need to understand cumulative impacts in contexts where data is limited, and expert elicitation provides a flexible approach when data is sparse and decisions urgent. Using both expert interviews and group elicitation strategies, experts provide causal process pathways of impact to the different ecosystem services and estimate the level of impact from individual activities and stressors. Despite high uncertainty in expert estimates of impact from particular activities and stressors, experts provided consistently high values of cumulative impact across ecosystem services. I show that across the various ecosystem services, a core set of interacting activities and stressors featured prominently. Of particular importance is sedimentation, which is amplified by climate change, as well driven by land clearing and ocean based activities. This chapter suggests that a rich understanding of cumulative impacts to 30  ecosystem services, with concrete policy recommendations on particular stressors to target, can be acquired even when experts are the only source of information available. The dissertation concludes with Chapter 7, wherein I reflect on the dissertation as a whole. I highlight the substantive and methodological contributions of this dissertation, both to the fields of cumulative impact assessment and ecosystem services, and which I suggest can be applied to the research and practice of ecosystem-based management. I further consider the limitations of the dissertation as a whole, as well as ponder the future research directions for which this dissertation provides early steps.    31  Chapter 2: Scientific Shortcomings in Environmental Assessments 2.1 Introduction Large–scale development is a hallmark of the modern world, providing society with things humans value, but potentially compromising environmental outcomes (Crutzen 2006). Trying to navigate this tradeoff, many governments rely on the process of environmental impact assessment (EIA) to inform the decision-making process by providing an accurate accounting of a development’s adverse impacts on valued aspects of the environment, with the ultimate goal to facilitate long-term sustainability (Wood 2003). EIA was initiated by the US National Environmental Policy Act (NEPA) in 1970, and has propagated around the world (Wood 2003; NEPA 2007). Proponents of EIA refer to it as a “robust,” “science-based” approach, terms which carry connotations of credibility and objectivity (Killingsworth and Palmer 2012). How firmly is EIA actually rooted in scientifically rigorous standards of evidence and analysis? To answer this question, we examined the main outputs of the EIA process, written reports commonly referred to as environmental impact statements (EIS), from seven mature regulatory jurisdictions: British Columbia (Canada), California (USA), Veracruz (Mexico), Brazil, England and Wales, Queensland (Australia), and New Zealand. Each EIS is written by a multidisciplinary team that typically 1) establish the spatiotemporal scope for the study, 2) consult relevant stakeholders, 3) determine the potential impacts of the project to valued environmental components (including impacts that may occur in concert with other past, present, and future projects, called cumulative impacts), 4) propose mitigation to avoid, reduce, and remedy potential impacts, 5) determine the residual impacts that will likely persist after mitigations are 32  applied, 6) and finally determine the importance – or significance – of these residual impacts (Wood 2003). Exceptions occur in Mexico, Brazil, and England and Wales, where significance is sometimes determined before (or both before and after) mitigations are proposed. Significance determination is arguably the “bottom line” of an EIS, supplying decision-makers with a final account of the impacts to be weighed against development benefits.  As published works within a practice characterized as scientific, we might expect EISs to be guided by rigorous standards of evidence and analysis within relevant disciplines. For example, setting an appropriate spatiotemporal scope of analysis for affected species could be informed by wildlife biology (Long and Nelson 2012). Determining the likely lag effects from decommissioned mines could apply established methods from environmental toxicology (Demchak et al. 2004). Designing consultation methods to reflect and respond to stakeholder concerns could rely on research into risk and deliberative methods (Pidgeon et al. 2005; Fishkin 2009). The effectiveness of mitigation measures could be judged on evidence from restoration ecology (Quigley and Harper 2006). In this paper, we evaluate how well current practices align with these expectations: whether the practice exemplified by EISs reflects the current state of relevant scientific fields. Specifically, we address the following questions: 1) How consistently are potential impacts found to be significant across countries and project types? 2) Does the scope of EISs reflect the current state of science? 3) How scientifically robust are the proposed mitigation measures? 4) How does stakeholder consultation influence EISs, and how is significance determined? 33  2.2 Methods We compiled a database of EISs from seven different jurisdictions of the world from diverse continents (excluding Asia and Africa) by convenience of language proficiency of authors and public availability of EISs. These included British Columbia (Canada), California (USA), Veracruz (Mexico), Brazil, England and Wales, Queensland (Australia), and New Zealand. We have a mix of governance levels (states/provinces and countries) because we chose the most local level at which decisions are made. Though the EIA process in Mexico is nation-wide (there are no state-level EIA processes), we focused on Veracruz to geographically pair with Brazil on the Atlantic coast to compare differences in jurisdiction while trying to minimize differences due to geography. We selected the ten most recent EISs from each jurisdiction to focus on current regulatory processes. Most EISs were initiated between 2012 and 2015 (one in British Columbia was from 2010 and one from New Zealand was from 2011). The paucity of EISs in New Zealand led us to review only 7 EISs from there, and the high number of EISs in Queensland led us to review 11 EISs. A breakdown of types of projects in each jurisdiction can be found in Table A.1. We looked at official guidance documents for each jurisdiction on how to prepare an EIS to ensure that the studies were conducted in a similar fashion (from predicting impacts, proposing mitigations, and evaluating significance of impacts). After we were satisfied this was largely the case, we counted the number of impacts identified in each EIS to estimate the proportion of impacts that were deemed “significant”; distinguishing between recorded project-specific impacts, residual impacts, cumulative impacts, and significant impacts by relying on the EIS to accurately differentiate these (that is, we took the reports at their word and did not interpret types of impacts for them). This allowed us to answer how consistent the determination of significance 34  was across EISs. We also classified the methods by which significance was determined in broad categories (technical, collaborative, reasoned) as outlined by Lawrence (2007). Because of the highly skewed nature of the data on impact frequencies, we used bootstrap 95% confidence intervals (using the BCa method) of the median to determine significant differences between jurisdictions. We calculated a global median from all jurisdictions included in our analysis. Where bootstrapped confidence intervals cross the global median, this indicates that there is no significant difference between the jurisdiction and the global median. Analysis was conducted using the boot package in R (Canty and Ripley 2015). To determine the robustness of EIS scope, we compared the spatial and temporal dimensions of each EIS with the spatial ranges of prominent species the EISs are supposed to assess impacts on as well as the duration of prominent impacts. We also highlight the frequency that EISs consider impact interactions relative to the importance of these kinds of impacts in the scientific literature. To determine the spatial dimensions for each EIS, we determined largest area investigated by the EISs to assess cumulative impacts (the largest area assessed for all valued components). Where only maps were provided (and data not provided in-text), we calculated area measures from the maps using DataThief (Tummers 2006) and ImageJ (Abràmoff et al. 2004). To assess the suitability of these spatial areas we compared these areas against the published ranges of species that EISs in each jurisdiction consider. We haphazardly sampled a list of valued components assessed by our sampled EISs in each jurisdiction (we chose 6 common species per jurisdiction) and used publicly available resources to acquire data on species ranges (Table A.2). We matched the scale of species ranges to the scale at which EISs claim to assess them. For example, if EISs claimed to assess impacts to specific populations, subspecies, or species, we looked up range 35  data at that scale. We made sure that each EIS described species scales consistently when the species was introduced as a valued component and when impacts to the species were described.  Where possible we used government online resources from each jurisdiction, or online resources that the government sites provided. Where this was not possible, we relied on IUCN and Natureserve online resources. To assess temporal scale, we recorded the number of years estimated for construction of the project, the number of years the project was projected to be operational, and the total number of years for which the EIS assessed impacts. The difference between the number of years for impact evaluation and the number of years for operation and construction constituted the number of years that impacts from the project in each study will contribute to cumulative impacts. As we noted that mining EISs had the longest post-closure time periods, we focused our analysis on this subset of EISs (N=11). We then collected peer reviewed published data on the number of years post mine closure the effects of Acid Mine Discharge (AMD) have been recorded (Table A.3). We contrasted this data with the temporal scope of mining EISs. The studies found AMD impacts at sufficiently long times (decades) past mine closure, while commenting that AMD can persist centuries past mine closure.  To assess the interaction categories of cumulative impacts, we analyzed the EISs’ methodology sections and noted how cumulative impacts were described and assessed. If there was any mention of interaction type (e.g. additive, synergistic, antagonistic) we recorded that EIS as having considered that specific interaction type. We also recorded whether the EIS did not specify types of cumulative impacts (but still described their methodology) and whether the EIS did not describe their methodology at all.  36  To look at the importance of mitigations in significance determination, we analyzed EISs that consider significance before and after mitigations. When a significant impact remained significant post-mitigation stage, we noted whether this was due to a lack of any mitigation applied to the specific significant impact or to an ineffective mitigation (e.g. some significant impacts on visual amenity in England and Wales had no mitigations proposed). We used this information to compute the odds of a mitigation reducing the significance level of an impact and recorded how mitigations reduced significance overall. Additionally, we counted the total number of mitigation measures indicated in each report. For each mitigation measure, we looked to see if the associated language established a baseline level needed for the mitigation to be effective and enforceable. To do this, we looked to see if the language of the mitigations were sufficiently vague as to not hold the mitigation action as an enforceable commitment, and recorded the number of mitigations with vague language around implementation or execution. Examples of vague mitigation language include “to the extent possible”, “where feasible”, “if practical”, “will attempt”, “explore the possibility of”, and “plan to create a plan to mitigate”. We also made note of whether the EIS provided evidence for mitigation effectiveness, assessed the effectiveness of proposed mitigations or acknowledged uncertainty in the proposed mitigations.  We collected data from each EIS on stakeholder consultation to examine the role of consultation in each EIS. We reviewed each EIS and recorded the level of public engagement that was undertaken according to the typology of participation developed by Hughes (1998), described in Table A.4. We recorded the most inclusive form of consultation undertaken on behalf of the 37  project. We also recorded the types of stakeholders and affected parties involved in consultation, according to the categories from Hughes (1998).  Multiple members of the author list participated in collecting data. To ensure that we minimized among-collector variation, we took measures to make sure all data collectors were approaching data collection in a similar way. First, all data collectors took part in a short workshop to communally collect data from the initial EIS. Second, one of the data collectors who is fluent in all relevant languages (English, Portuguese, and Spanish) either directly collected data, or supervised the collection of data, for all regions and performed quality control on the completed database. To promote greater consistency, a second workshop was conducted after data collection had begun to help collectors calibrate their approaches with one another. Finally, the group member responsible for the database re-checked the data, paying attention to any data points that seemed to stand out. This helped ensure that similar approaches to data collection were taken in all contexts. 2.3 Results and Discussion 2.3.1 Different Places, Same Bottom Line The number of potential, cumulative, residual, and residual cumulative impacts reported in EISs varied considerably across jurisdictions.2 This variation likely follows idiosyncratic norms of                                                  2 No EIS included additional mitigation measures to address cumulative impacts, thus the number of potential cumulative impacts is equal to the number of residual cumulative impacts. 38  assessment (e.g., sometimes impacts are discussed individually and sometimes they are grouped together under broader headings) and disparity in the types of projects being assessed (e.g., mining projects realistically have more impacts than a small extension to a highway). We found fewer residual impacts than potential impacts, as mitigation measures reduce impact severity.  However, despite the aforementioned variation in number of impacts, the proportion of significant impacts was uniformly small across jurisdictions (all bootstrap 95% CIs overlap the global medians of 2 significant project-specific impacts and 0 significant cumulative impacts, Figure 2.1). If the determination of impact significance was robust, we might expect that jurisdictions with similar types of projects and environmental settings would have relatively similar proportions of potential impacts considered significant. At least, we might expect that jurisdictions with a high ratio of proposals for large-scale projects requiring vast landscape transformation (e.g., surface mining), such as Queensland, would consider a relatively higher proportion of potential impacts to be significant, compared to jurisdictions with a high ratio of proposals for smaller scale development (e.g. renewable energy), such as England and Wales (Table A.1). Instead, regardless of jurisdiction, a consistently small proportion of potential impacts was considered significant. One explanation for this pattern is that the EIA process leading up to the EIS is a systematic barrier to projects that will likely contribute to significant impacts, allowing only relatively benign projects to undergo significance determination.                                                  As mitigations are central to EIAs, the lack of mitigations for cumulative impacts may reflect a general lack of attention to cumulative impacts (Wood 2003).  39  Additionally, systematic bias against positive significance determination could be occurring in the EIA process. Below, we review characteristics of EIA that may contribute to such a bias.   Figure 2.1 The number of potential impacts, residual impacts, and significant impacts reported in EISs for A) project-specific impacts and B) cumulative impacts. Bars represent bootstrap 95% confidence interval of the medians, and the red lines represent the global medians. EISs were selected from a single state or province within the countries marked with an asterisk (*) in the legend. 2.3.2 Narrowly Addressed Impacts The spatial scope of EISs are not ecologically sufficient. For example, to adequately consider cumulative impacts to a valued wildlife species, the spatial scope of an EIS should be at least as large as the range of that species (João 2002; BCEAO 2013). In practice, spatial scopes are generally considerably smaller than the ranges of a collection of species (or specific populations) 40  purportedly assessed in the sampled EISs. Our analysis indicates that only a minority of EISs consider spatial scales comparable to the ranges of species or population units assessed (Figure 2.2A). Ecologically appropriate EISs would not necessarily assess impact on scales of entire species (e.g. there are discrete population units within the 8.75 million km2 of jaguar range), because this coarser taxonomic scale may mask concerns for specific population units important to stakeholders (although many EISs purport to consider impacts at the species scale).  The temporal scopes set in EISs were similarly too short to be justifiable ecologically. Some projects can affect the environment long past decommissioning, causing lag impacts (Collins et al. 2010). In practice, we found EISs routinely restrict analyses to well before impacts are likely to cease. The case of mining is illustrative. Mining EISs in our sample assessed impacts further past decommissioning than other EISs, but even these temporal scopes were generally far shorter than published durations of ecological impacts from acidic mine discharge (AMD) after mine closures. Whereas most mining EISs limited their assessment to a period of between zero and four years after mine closure (Figure 2.2B), independent ecological studies emphasized that AMD can last decades to centuries past mine closure, even accounting for modern techniques (Demchak et al. 2000; Demchak et al. 2004; Moncur et al. 2006). Out of 26 mining EISs sampled from Queensland, Brazil, and British Columbia, one British Columbian mine used an appropriate temporal scale for AMD: 250 years past decommissioning. Narrow temporal scoping is doubly insidious because it limits the number of residual impacts considered for a given EIS, and will prevent future EISs from including relevant residual impacts in their cumulative impact assessment. 41   Figure 2.2 Density histograms showing the overlap of spatial scale in EISs (in red) and current scientific data (in grey) on A) spatial range of populations and species assessed in EISs and B) time after mine decommissioning that acid mine drainage impacts ecosystems. Note the logarithmic scale used in A). That the spatial and temporal scales of assessments were often smaller/shorter than the ranges of affected species/duration of relevant impacts suggests that the scoping was insufficient to address cumulative effects. The assessment of potential interactions among impacts was similarly limited and generally opaque. Research suggests that cumulative impacts are commonly synergistic (where total 42  impact is greater than the sum of individual impacts) or antagonistic (where total impact is less than the sum of individual impact), yet only 4% of EISs explicitly considered possible non-additive impacts (Crain et al. 2008; Darling and Côté 2008). In contrast, 15% of EISs only addressed additive impacts, 53% had unclear methods and 28% provided no methods (including every EIS investigated from New Zealand). Where methods were given, EISs tended to treat cumulative impacts as entire projects overlapping assessed areas, rather than outlining mechanistic processes and specific stressors investigated (only 3% of EISs explicitly explore cumulative impacts using mechanistic processes, such as explicitly documenting tanker traffic and underwater noise associated with nearby energy projects). The limited scope of EISs in space, time, and interactions across impacts all contribute to an avoidably narrow assessment of impacts (Lenzen et al. 2003).  2.3.3 Overconfidence in Mitigation Measures Several results demonstrate the high confidence placed in the effectiveness of mitigation measures, and that this confidence is likely undeserved. Many EISs in England, Wales and Brazil - and a few in Queensland and California - consider significance before and after application of mitigation measures, and the resulting change in characterization of significance indicates the EIS authors’ confidence in the proposed measures. Out of 505 impacts deemed significant prior to mitigation, 80 were ultimately characterized as significant after considering all mitigations, of which only 22 involved specific mitigations. In other words, for those 447 significant impacts that had associated mitigation measures, 425 were deemed effectively resolved by that mitigation and 22 deemed still significant (a 19.3:1 ratio). This high confidence in mitigation measures is 43  questionable, because mitigation proposals in EISs are 1) sometimes not enforceable and 2) often not scientifically verified (Hollick 1981; Duinker and Greig 2006; Duinker et al. 2012).   Figure 2.3  The A) proportion of mitigation measures written in ambiguous and unenforceable language in each jurisdiction (bars represent 95% bootstrap CI of the median and red line represents global median) and B) the proportion of EISs in all jurisdictions that have explicit analysis of mitigation effectiveness and consider uncertainty of mitigation effectiveness. No single EIS considered both mitigation effectiveness and uncertainty. EISs were selected from a single state or province within the countries marked with an asterisk (*) in the legend. We found that 5-11% (bootstrap 95% CI) of mitigation measures across jurisdictions were expressed in ambiguous language that made it unclear what actions would be taken, if any (e.g. “where applicable, mitigation X will be installed”; “to the extent possible, mitigation X will be explored”; Figure 2.3A). The consequence of this considerable faith in mitigation is that some 44  impacts with severe negative consequences may be written off as not significant (Duinker et al. 2012).  Furthermore, we found no EIS that transparently assessed both mitigation effectiveness and uncertainty on impact reduction (Figure 2.3B). Mitigation measures are routinely treated as effective despite research demonstrating them to be ineffective (e.g. fish habitat compensation, (Quigley and Harper 2006), or despite a lack of research into specific mitigation effectiveness which underscores great uncertainty in effectiveness (Duinker et al. 2012).  2.3.4 Suspected Bias in Significance Determination In our sampled EISs, significance was determined by consultants (who were paid by project developers), with minimal input from other stakeholders, and often without any transparent justification. Consultation with stakeholders other than developers is necessary to judge significance in any case where biophysical impacts have social or cultural implications (Canter and Canty 1993; Briggs and Hudson 2013). Yet in all but one EIS, stakeholders had no input in the determination of significance. In the outlier, a New Zealand EIS, a team of Maori stakeholders assessed cultural impacts. In some EISs, non-development stakeholders were simply told of a planned development without being given the option to voice concerns. Most commonly, non-developer stakeholder concerns were documented (with no documented follow-up) or responded to in facilitated meetings with no further opportunity to influence the design of the project or determine if their values were factored into significance determination (Figure 2.4).  45   Figure 2.4 The proportion EISs reporting on each jurisdiction that consulted stakeholders to various degrees. EISs were selected from a single state or province within the countries marked with an asterisk (*) in the legend. Refer to Table A.4 for a description of the different categories of consultation in the legend. In fact, stakeholder groups that may be the only people who can legitimately judge significance for specific cultural values were often not consulted in EIAs. In our sample, community organizations were consulted in only 51% of EISs, indigenous groups in 58%, and environmental groups in 70%. Other groups were consulted regularly: business and political groups were consulted in 88% and 97% of EISs, respectively. There are a few potential explanations for these disparities in representation. First, cultural, aboriginal, and environmental groups may lack the capacity to represent their interests to the same extent as business or political groups. Second, these less represented groups may not have elected to participate in the consultation process as much as the more represented groups, for various reasons including not having a stake in proposed development sites (however, in two jurisdictions where there are strong legal 46  requirements for First Nation consultation—Canada and New Zealand—we found 100% consultation). Finally, environmental and aboriginal groups may be underrepresented even when they have a stake. Our findings cannot distinguish among these explanations, and invite further research on this gap in consultation. The literature documents many case studies of EIAs suppressing concerns of local groups and those who might be against development (O'Faircheallaigh 2010). The limited consultation with indigenous groups has extra consequence, as indigenous groups often have dependencies on and histories linked to the environment not shared by others (Stevenson 1996; Banerjee 2000). EIAs may exacerbate a power imbalance in environmental decisions that has contributed to cultural loss for indigenous people worldwide (Banerjee 2000). In lieu of stakeholder consultation, significance in our sample was overwhelmingly determined according to the professional judgment of the EIS authors, often without documentation of methods, assumptions, or reasoning. Though quantitative thresholds were sometimes consulted in determining the significance of an impact (48% of EISs used quantitative thresholds for a subset of impacts, and 42% of EISs did for cumulative impacts), we found that every EIS relied on the consultants’ judgement for the majority if not all determinations of impact significance.  Relying on professional judgement to determine impact significance opens the door for biased assessment. Across the countries we investigated, developers pay the consultants who prepare EISs; finding few (or no) significant impacts often fulfills the financial interest of the developer and the consultant (Hollick 1984). While we cannot measure how this system affects EIS conclusions drawn, the normalization of this EIS practice invites bias (Hollick 1984). Judgements are easily influenced by affiliation with interested partisans (Moore and Loewenstein 47  2004; Moore et al. 2010). Bias may help explain why some EISs in England and Wales and New Zealand highlighted approximately as many “significant beneficial” impacts as significant adverse impacts (such beneficial impacts are not called for in the guidance documents in any jurisdiction we assessed). Our aim is not to accuse consultants of poor intentions, but to point out that potential institutional bias introduced by conflicts of interest is problematic even when EIS authors believe in the scientific rigor of their work (Moore et al. 2010).  2.4 Conclusions Our findings suggest that around the world, EISs rely on questionable scientific evidence and analysis that may contribute bias against determining impacts as significant, which may ultimately sway decision-makers in favor of developments. Six major types of changes could salvage EISs as a legitimate science-based regulatory tool for environmental protection in a context of natural resource development: 1) The spatial and temporal scope of assessments should be ecologically justifiable and explicitly consider cumulative impacts, encompassing the ranges of ecosystem components affected and the duration of demonstrated lag impacts from relevant literatures (Shepherd and Ortolano 1996; Duinker et al. 2012); 2) Interactions among impacts should be explicitly considered and in reference to available evidence, acknowledging that interactive, non-additive effects are the norm (Crain et al. 2008); 3) Mitigation actions should be stated in ways that are enforceable, and all mitigations should require an analysis of their degree of effectiveness, with uncertainty acknowledged, and contingencies for potential failure (Hollick 1981; Duinker et al. 2012). Mitigations should only allow for the down-grading of significant impacts when planned mitigations have a demonstrated effectiveness in appropriate contexts; 4) Stakeholder consultation in determination of impact significance must 48  be ensured whenever environmental impacts may have local, social or cultural consequences – likely the majority of cases (O'Faircheallaigh 2010); 5) Policies should force developers to comply with 1-4, empower regulators to reject any EIS that does not, and make environmental audits compulsory to ensure the compliance of projects to their mitigation commitments; and 6) The inherent conflict of interest in EIS authorship must be eliminated, e.g. by having developers pay into a common fund, administered by governments, to hire  independent consultants (Hollick 1984; Moore et al. 2010). To be a truly transparent and robust tool of environmental protection, EIA needs to embrace current evidence and practices when relying on science. Failing to improve the regulation and practice allows EISs to obscure and facilitate important environmental impacts more often than they reveal and prevent them.    49  Chapter 3: Mapping Cumulative Impacts to Coastal Ecosystem Services in British Columbia 3.1 Introduction Humanity’s great and growing influence on the planet demands an increased understanding of how multiple activities cumulatively affect the human benefits and values associated with the environment (Arkema et al. 2006; Granek et al. 2009). In part due to their ease of comprehension and display of multiple human stressors at once, impact mapping has gained much traction in environmental science and management (Halpern et al. 2008a; Halpern et al. 2009; Selkoe et al. 2009; Ban et al. 2010; Allan et al. 2013; Halpern and Fujita 2013; Micheli et al. 2013; Murray et al. 2015b; Clark et al. 2016). However, impact mapping studies generally reflect how human activities affect species, communities and habitats, neglecting thus far how multiple activities cumulatively affect ecosystem services (the processes by which nature renders benefits for people, however see Allan et al. 2013). Understanding impacts on ecosystem services would allow for a representation of multiple societal benefits from the environment, enabling targeted management on specific ecosystem services.  Existing approaches also may not capture spatial patterns of human impacts to ecosystem services. In particular, because delivery of ecosystem services to people requires both the provision of services through biophysical means and delivery to people that demand those services (Tallis et al. 2012), maps of ecosystem services may be more restricted in space than maps of total service supply (each with potentially unique spatial patterns).  The few mapping studies that address impacts to ecosystem services represent ecosystem services through 50  landscape proxies, and quantify impact through changes to the underlying ecosystem, which may lead to an inaccurate representation of, and incomplete accounting of impacts on services (Allan et al. 2013; Angradi et al. 2016). Put differently, if the underlying landscape proxy does not account for human use of an ecosystem service, resulting maps may resemble ecosystem service potential rather than the actual benefits accrued to people. Ultimately, highlighting how ecosystems change as a response to cumulative impacts implicitly downplays the important social dimensions in ecosystem services related to service delivery and enjoyment.  Mapping impacts to ecosystem services requires understanding where human activities co-occur (i.e., the ‘footprint’ of human activities) and the risk each activity poses to each ecosystem service within that footprint (Rounsevell et al. 2010; Mach et al. 2015). Risk to ecosystem services can be understood at each step in the ecosystem services ‘cascade’ (Haines-Young and Potschin 2010), with human impacts potentially affecting supply (the biophysical components that produce ecosystem services), service (the ability of people to access and benefit from a service), and value (people’s preferences for ecosystem services, Tallis et al. 2012). However, the relative importance of these factors in regulating impact to ecosystem services is not known. Existing frameworks of impact to ecosystem services characterize change in the underlying ecosystem as driving impact, with human beneficiaries of services largely subject to changes in ecosystem service supply (Collins et al. 2010; Rounsevell et al. 2010; Kelble et al. 2013; Mach et al. 2015). Recently, spatial models have been created that utilize production functions for ecosystem services, relating landscape features important for ecosystem services, as well as spatial social data on human use of the environment, to generate maps of ecosystem services on the coast (Guerry et al. 2012; Sharp et al. 2014). These models allow for a more complete 51  mapping of ecosystem services. If biophysical production of ecosystem services largely regulate impact to ecosystem services, existing maps of impact to species and habitats, encompassing only supply of ecosystem services, may approximate impact to ecosystem services. If, however, explicit spatial consideration of the use of ecosystem services, as well as the changes to people’s access and the perceived quality of services (given people’s preferences) are important to understand impact to ecosystem services, then mapping studies should explicitly include these considerations to ecosystem services, and frameworks of impact to ecosystem services should be updated to reflect these types of impacts (Chan et al. 2012a; Wieland et al. 2016). For example, pollution might not affect shellfish growth in aquaculture, but it may lead to tenure closure for health concerns, or could affect the taste of shellfish caught at polluted sites. Changes to the enjoyment of shellfish aquaculture, in this case, is not a result of changes in the biophysical supply of the service but in the change to either access or quality of the service. In what follows here, we model human impacts to specific ecosystem services on coastal British Columbia to identify areas of high impact considering the ecosystem service cascade, and advance the understanding of risks to ecosystem services. Coastal British Columbia is an area renowned for its scenery and productivity, contributing greatly to the economy, sense of place and other values important to residents and visitors (Dawson 2005; Klain and Chan 2012). Maps of cumulative impacts to coastal British Columbia ecosystems have been produced (Ban et al. 2010; Murray et al. 2015a; Murray et al. 2015b). This work, alternately, does so for ecosystem services themselves. We ask: 1) What ecosystem services face the greatest impact in coastal British Columbia?; 2)What human activities pose the greatest threat to what ecosystem services?; 3) Where are services under greatest threat? 4) Do the answers to the first three questions change 52  when social dimensions of impact are included or left out?; 5) How are likely future impacts thought to affect ecosystem services?;  6) What is the relative importance of impact to metrics of service supply, service, and value?; 7) What are the important causal pathways that coastal ecosystem services in British Columbia are threatened by? Together, addressing these questions builds on established methods to map cumulative impacts using geospatial data and expert derived estimates of ecological vulnerability (Halpern et al. 2009; Teck et al. 2010). We use spatially explicit models of ecosystem services, overlapped with maps of multiple human activities with expert derived estimates of risk (including risk to the underlying ecological community that supply services, as well as risk to change in access and change in service quality).  3.2 Methods  The analysis consisted of five main steps. First, we mapped eight ecosystem services using InVEST models and spatial data available for the region. Second, we assembled spatial data for 21 potentially impacting human activities and stressors. Third, we derived risk scores for each service-activity (risk of activity x on service y) combination via an expert elicitation process. Fourth, we combined the above three steps spatially to assess the cumulative impacts of all available activities on each service. The resulting maps allowed us to answer where ecosystem services were under greatest impact, and tracking the impact data allowed us to answer which ecosystem services were most impacted and by what human activities or stressor. The expert scores allowed us to consider separately biophysical impacts on service supply and social impacts on service delivery, as well as potential future risk. We contrast maps of total impact with maps that only incorporate biophysical impacts to explore the importance of social 53  dimensions of risk. Finally, also through expert elicitation, we explore the causal pathways of impacts to ecosystem services. We detail each step below. 3.2.1 Spatial Representation of Ecosystem Services The InVEST models addess and mapped eight different ecosystem services (Guerry et al. 2012; Sharp et al. 2014): commercial demersal fisheries, commercial pelagic fisheries, finfish aquaculture, shellfish aquaculture, marine recreation, coastal aesthetics, coastal protection, and potential wave and tidal energy generation. We model “potential” energy generation because British Columbia currently does not have wave and tidal energy operations, but there is interest in harnessing this energy supply. The InVEST tool has tiered models for mapping ecosystem services based on different levels of data availability. InVEST is capable of modeling the quantity and even monetary value per given area of ecosystem services within the area that people use them (to represent ecosystem service supply, service, and value, Guerry et al. 2012). Due to data limitations, we were prevented from modeling ecosystem service quantity and value within a given area for all eight services across coastal BC, but we could produce maps of the extent of human use of ecosystem services for all eight. We used the base InVEST models for fisheries, aquaculture, and recreation maps whereby overlapping maps of different activities create the resulting service model. We modeled coastal aesthetics by calculating the viewshed from sites of recreation and human habitation. This model takes into account bathymetry and topography to calculate the viewshed.  By mapping the spatial extent of ecosystem services by areas where people use them, we capture spatial representation of the supply and service components of ecosystem services (without representing value). We modeled coastal protection by mapping the parts of the coast protected by vegetation, kelp, and 54  erosion-resistant substrate (not mapped are areas of the coast without protection). We did not use InVEST to map potential renewable energy, as we had publicly available spatial data on wave and tidal energy areas of interest along the BC coast, so we mapped this data. See Appendix B for detailed descriptions of ecosystem service model parameterization.  3.2.2 Spatial Representation of Impacting Activities We assembled spatial data layers for 21 different activities and stressors, including activities and stressors related to fisheries, coastal commercial industries, land based pressures, and climate change impacts (these broad categories derived from Ban et al. 2010). Many of the data layers were adapted from a previous cumulative impact study by Ban et al. (2010) supplemented with data from British Columbia Marine Conservation Analysis (BCMCA 2016) and GeoBC (GeoBC 2016), while compiling the 25 fisheries into five categories (demersal destructive, demersal non-destructive, pelagic low bycatch, pelagic high bycatch, recreational fishing). Reducing the number of fisheries categories was done to account for the fact that the number of data layers influences the overall cumulative impact scores (Ban et al. 2010), and we did not want to overly bias impact based on fisheries scores. This dataset includes the footprint of human activities as well as zones of influence, with each zone’s distance dependent on prominent stressors associated with the activity. We also included climate impacts adapted from Halpern et al. (2008b)’s global map (see Table B.2 for data sources). 3.2.3 Impact Mapping Following the cumulative impact mapping approach first demonstrated by Halpern et al. (2008), we overlay maps of impacting activities on ecosystem services, and incorporate quantitative 55  estimates of activity intensity and the risk that each unit of an activity has on an ecosystem service. The quantitative estimates of risk and activity intensity are described below, followed by a description of how cumulative impacts are calculated. 3.2.4 Risk Scores To calculate the risk scores (μi,j) we relied on expert judgement, due to pervasive data gaps (see Table S3 for a description of activities and stressors for risk quantification). We adapted the expert survey used in Teck et al. (Teck et al. 2010), and focused on ecosystem services. We used an online platform because the diversity of ecosystem services and the large number of μi, values precluded individual surveys, workshops, and other elicitation methods (McBride and Burgman 2012). We invited 437 experts to take part in the survey and received responses from 220, though not all experts fully completed their survey (50.3% response rate). We allowed participants to self-organize for chosen ecosystem services (some indicating their expertise for multiple ecosystem services), and they provided responses for all ecosystem services they presumed themselves experts on. Risk estimates were compiled for commercial fisheries in aggregate (instead of demersal and pelagic commercial fisheries) as well as commercial aquaculture in aggregate (instead of shellfish and finfish aquaculture) because only a small minority of experts indicated their expertise and responses as fitting these specific categories.  Experts were tasked with quantifying risk according to seven criteria adapted from and expanded on Teck et al. (2010). The criteria encompass exposure (area of influence, frequency of impact and recovery time, Table B.4) and consequence (magnitude of impact on ecosystem service production, ecological extent of impact, effects to access and effects to perceived quality, Table B.5). For potential energy generation, only one expert provided these quantitative measurements 56  (though others provided other information on potential energy generation) so quantitative results for this ecosystem service should be considered tentative. To partially assess future risks to ecosystem services, experts were asked to quantify risk to two global stressors and one regional stressor of high concern, given the changing climate and development trajectory of British Columbia. Experts were asked to quantify risk from sea surface temperature rise and ocean acidification according to projections for the year 2100 (3°C increase and 0.3 pH decrease, respectively, (Stocker et al. 2014), and to quantify risk from a major oil spill (>40 000 m3, Peterson 2003). All criteria were normalized so that the resulting expert scores were scaled between 0 – 1. Contrary to Teck et al (2010), who considered ecosystem vulnerability as a linear combination of exposure and consequence considerations, we chose to model ecosystem service risk according to a technical model of risk as , = , × , where Pi,j is the exposure of ecosystem service i  to an individual occurrence of activity or stressor j and Ci,j is the consequence of an individual occurrence of activity or stressor j on ecosystem service i. This approach allows for the risk to an ecosystem service to be zero if the corresponding Ci,j is zero. We treat Pi,j and Ci,j as linear combinations of exposure and consequence variables, including both biophysical and social components of consequence. The risk criteria for exposure include area of influence of the activity, frequency of impact, and recovery time of the ecosystem service to the activity. The risk criteria for consequence include the magnitude of risk to the environment that produces the ecosystem service, the extent to 57  which the biophysical community is affected, how the activity limits people’s access to the service, and the extent to which people perceive a loss of quality in the service (see Appendix B for further explanation of the criteria). The mean expert derived risk scores were used in the cumulative impact model. We calculate Pi,j and Ci,j as , =    × ,, and , =    × ,, where Wk and Wl are the weights of risk criteria k and l (where ∑  = 1  and ∑  = 1 ), pi,j,k is the risk value of activity or stressor i on criterion k for ecosystem service j, and ci,j,k is the risk value of activity or stressor i on criterion k for ecosystem service j. Similar to previous mapping efforts, we assume that the weights are similar for all combinations of i and j, allowing for a single model to be applied to all ecosystem services, in turn allowing for direct comparison between them (Halpern et al. 2009; Teck et al. 2010). To assess the suitability of only considering biophysical criteria in cumulative impacts to ecosystem services, the consequence (Ci,j) scores were also calculated with only the biophysical risk criteria included. To determine risk criteria weights, we asked experts to complete discrete choice exercises using a ‘revealed importance’ method (Neslo 2011). This is because it is often unreliable to assess relative criterion importance directly through statements (e.g. asking, “how important is criterion x to variable y”), as this does not force experts to consider multiple criteria simultaneously. We 58  thus asked experts to rank hypothetical scenarios of anthropogenic impacts with plausible values for criteria, split between two ranking exercises. The first ranking exercise focused on the exposure criteria, and the second focused on the consequence criteria. Splitting the criteria this way allowed us to isolate the importance of criteria within the two components of risk while limiting the possibility for cognitive overload. Here, we are referring to cognitive research suggesting  that people can only receive and process between 3 and 9 variables at once, with recent research emphasizing the lower end of this spectrum (Miller 1956; Gross 2012). Asking experts to judge scenarios with 3-4 criteria therefore seemed a more reliable technique than asking them to judge scenarios with 7 criteria. Resulting ranking data was produced using probabilistic inversion to generate weights. When faced with so-called “inverse problems” where observational performance data is known and information on the structure that generates this data is sought, probabilistic inversion determines a joint distribution of criteria weights that models the distribution of expert preferences given the observed data (Teck et al. 2010; Neslo 2011). For the purposes of cumulative impact (Ic), we use the mean weights from the distribution that best fit the population of expert preference rankings, such that large weights reflect criteria that contribute to consistently high ranking. The same mean weights were used in impact maps that only consider biophysical risk criteria to allow for direct comparison with maps including social risk criteria. But because the method produces a joint distribution of weights across the distribution of expert preferences, we report on the mean and standard deviation of each criterion to visualize the spread of relative importance among experts, especially the relative importance of biophysical versus social criteria. We report the standard deviation rather than the standard error of the mean because we are interested in the 59  distribution among experts, and because probabilistic inversion uses scenario modeling with thousands of scenarios to generate joint distributions, rendering standard error estimates near zero (White et al. 2014). We used 20 000 scenarios to generate modeled joint distributions for exposure criteria, and 40 000 scenarios for consequence criteria. Probabilistic inversion was completed using the program UNIVERSE (Neslo 2011). 3.2.5 Cumulative Impact Model After all ecosystem services were modeled, the overlap of all activity and stressor layers with the individual ecosystem services were generated in maps with 500x500m cell resolution. The specific ecosystem services served as the boundary for each overlapped map. All activity and stressor layers have associated data related to their level of activity intensity (e.g. density of ships). In order to make the disparate intensity measures comparable, and not influenced by highly skewed intensity data, all intensity scores were log transformed and normalized according to the largest intensity value in each activity and stressor dataset to generate a dimensionless 0-1 intensity scale (Halpern et al. 2008).  Cumulative impact Ic was calculated for each pixel according to the cumulative impact map formula  =   ×  × , where Di is the log-transformed and normalized intensity scores for activity or stressor i, Ej is the presence or absence of ecosystem service j, and μi,j is the risk of individual occurrences of activity or stressor i on ecosystem service j (Halpern et al. 2008). This cumulative impact model 60  assumes impacts are independent, non-negative, and additive in nature, which is an acknowledged limitation of the cumulative impact mapping framework, but allows for an estimate of total impact where great uncertainties persist regarding where, when, and under what conditions non-additive impacts occur (Halpern and Fujita 2013). Cumulative impacts were calculated both including and excluding social risk criteria to examine the contribution of social dimensions of risk on ecosystem services. 3.2.6 Understanding Mechanisms of Impact We asked experts in the risk survey to indicate whether or not the given activities and stressors affected their chosen ecosystem service directly or indirectly (or neither or both), with an optional follow-up to describe the pathways of impact.  3.3 Results 3.3.1 Impacts to Ecosystem Services Across all ecosystem services, total and per-pixel impact scores were more severe when including social risk criteria (expressed here as access and quality) in impact calculations than excluding them (Figures 3.1, 3.2, and 3.3). Resulting maps show greater overall impact across the spatial range of all ecosystem services when these social criteria are included on top of biophysical criteria (Figure B.1). Including these social criteria had the greatest proportional increase in per-pixel Ic for potential renewable energy generation (2.07 times greater than only considering impact on biophysical production of the service), followed by coastal protection (1.97), commercial demersal fisheries (1.70), commercial pelagic fisheries (1.69), recreation (1.64), aesthetics (1.42), finfish aquaculture (1.24), and shellfish aquaculture (1.16). Considering 61  total Ic values, including social impact criteria had the greatest proportional increase for recreation (2.50), followed by potential renewable energy generation (2.07), coastal protection (1.97), commercial demersal fisheries (1.70), commercial pelagic fisheries (1.69), aesthetics (1.42), finfish aquaculture (1.28), and shellfish aquaculture (1.19). The only case where considering impacts on access and quality did not add to impact estimates was the impact of shellfish aquaculture on itself (Figure 3.2).  Our results indicated that all modeled ecosystem services are impacted across most – if not all – of their range (Figure 3.1). Controlling for total range, the average per-cell impact was highest for commercial demersal fisheries (Ic = 0.43), followed by commercial pelagic fisheries (0.41), shellfish aquaculture (0.36), finfish aquaculture (0.35), potential renewable energy (0.32), marine recreation (0.31), coastal protection (0.18), and aesthetics (0.06). Considering only biophysical criteria, the position of potential energy and recreation are switched, otherwise this ranked list of ecosystem services facing impacts is consistent with the list considering social criteria. However, all ecosystem services vary greatly in their relative impacts (Figure 3.3). Most ecosystem services have per-cell Ic values that range from ~0-0.8, except aesthetics, which only has Ic values ~0-0.4. This ranking is largely consistent with a ranked list of ecosystem services facing impact only considering biophysical criteria. 62   Figure 3.1 Cumulative impact maps for four ecosystem services (aesthetics, coastal protection, commercial demersal and commercial pelagic fisheries), with associated bar graphs of causes of impact. Maps display the summed impact of all drivers and stressors to each ecosystem service; bar graphs show total impact values for each activity or stressor. Red bars indicate impact only accounting for biophysical risk criteria, and black bars indicate impact accounting for all risk criteria. Coastal protection is not to scale to allow for visibility. 63   Figure 3.2 Cumulative impact maps for four ecosystem services (recreation, energy, finfish and shellfish aquaculture), with associated bar graphs of causes of impact. Maps display the summed impact of all drivers and stressors to each ecosystem service; bar graphs show total impact values for each activity or stressor. Red bars indicate impact only accounting for biophysical risk criteria, and black bars indicate impact accounting for all risk criteria. Aquaculture sites are not to scale to allow for visibility. Considering total summed impact across the range of an ecosystem service, commercial demersal fisheries face the highest impact (aggregate Ic = 1.57×105), followed by commercial pelagic fisheries (1.27×105), aesthetics (3.54×104), marine recreation (3.28×104), potential renewable energy generation (8.24×103), coastal protection (1.24×103), finfish aquaculture (59.9) and shellfish aquaculture (55.9). Considering only biophysical criteria, the position of finfish 64  aquaculture and shellfish aquaculture are switched, otherwise this ranked list of ecosystem services facing impacts is consistent with the list considering social criteria. This ranking largely follows the ranking in spatial range of the ecosystem services themselves. For many ecosystem services, higher levels of impact were found on the south of the coast, between Vancouver Island and the mainland (for finfish and shellfish aquaculture and potential energy generation, and coastal protection), and the north coast (for aesthetics, coastal protection, demersal and pelagic fisheries, and marine recreation). Major hotspots of impact are similar when considering social dimensions of impact versus not considering them. Different groups of drivers and activities generate prominent impacts for different ecosystem services (Figures 3.1 and 3.2). Climate related stressors contributed high levels of impact to demersal and pelagic fisheries, marine recreation, finfish aquaculture and shellfish aquaculture. Ocean acidification was the main climate related stressor contributing to impact in these ecosystem services. Climate related stressors have the highest spatial range across all ecosystem services (occupying all map cells). Land-based activities contributed high levels of impact to aesthetics, coastal protection, and both aquaculture categories. Human settlements and onshore mining contributed the most impact to most of these ecosystem services. Coastal commercial activities contributed high levels of impact to finfish aquaculture. Aquaculture was seen as a prominent activity impacting itself, as experts scored biological and social criteria of risk high for aquaculture, and multiple experts described the self-harmful practices and invasive and disease problems of aquaculture. They also cited the poor public attitude towards aquaculture as a high risk to itself. Fisheries contributed high levels of impact to potential tidal and wave 65  energy. Experts scored social criteria of risk from fisheries to potential energy generation high, specifically the effects of fisheries on access to good renewable energy sites.   Figure 3.3 Density histograms of per-cell Ic values for each ecosystem service. Ecosystem services are: A) aesthetics, B) coastal protection, C) commercial demersal fisheries, D) commercial pelagic fisheries, E) marine recreation, F) potential wave and tidal energy, G) finfish aquaculture, and H) shellfish aquaculture. Red histograms indicate impact only accounting for biophysical risk criteria, and black histograms indicate impact accounting for all risk criteria. Considering future impacts, it experts perceive that some ecosystem services are at greater risk from some future climate stressors than potential major oil spill, while others are at greater risk from potential major oil spills (Figure 3.4). Aesthetics, coastal protection, and potential energy generation were all perceived to be at higher risk from a major oil spill on the coast than rising 66  sea temperature or ocean acidification. Coastal protection and potential energy generation were perceived to be at high risk from sea level rise, but we did not have spatial data for this stressor so we do not represent it here. In contrast, fisheries, aquaculture, and marine recreation all appeared to be at higher risk from future ocean acidification and sea surface temperature rise, and particularly ocean acidification.  Figure 3.4 The risk posed by future climate change risks and oil spills on six ecosystem services, compared with current climate change risks. Points represent mean risk scores, error bars represent 25th and 75th percentiles, and lines connecting points demonstrate the trajectory of risk from current conditions to future conditions. 3.3.2 Components of Risk to Ecosystem Services Based on expert ranking, risk to ecosystem services is dependent on diverse criteria of exposure and consequence, without a clearly dominant criteria influencing risk (Figure 3.5). For exposure 67  criteria, experts considered the spatial extent of individual occurrence of activities to be most important, followed by the recovery time of an ecosystem service to an impact, and finally the frequency at which an ecosystem experiences an activity. For consequence criteria, experts considered the magnitude of change to the biophysical processes that produce the ecosystem service to be most important, followed by how perceived quality of an ecosystem service changes in response to an impacting activity, the extent to which the environment is impacted (from individual species to entire ecosystems), and finally the changes to access to an ecosystem service. However, simple rankings mask the finding that experts perceived all criteria to contribute non-trivially to risk (the mean weights from the best predictive ranking model includes the frequency to contribute 20% towards exposure and access to contribute 19% to consequence criteria), and that there was a diversity of weights considered across our experts (Figure 3.5).  Figure 3.5 The perceived importance of risk criteria to exposure and consequence. Points and error bars represent mean and standard deviations of the distribution of relative importance of risk criteria. 68  3.3.3 Pathways of Effects Experts suggested diverse prominent pathways of effect from anthropogenic activities and stressors among the ecosystem services (Figure 3.6). Across all types of impact, including fisheries impacts, coastal commercial activities, land based activities and climate stressors, some ecosystem services have consistent impact pathway types (according to experts). Most aesthetics experts suggested that impact pathways to aesthetics are direct, with some specifically suggesting that the physical footprint of the activity is often all that matters for aesthetics.  Renewable energy potential was an ecosystem service that many experts suggested was not affected by any activity or stressor, though a sizeable minority suggested that fisheries affected it directly through restricting access and climate change affecting it both directly and indirectly through changing sea levels and affecting energy demand (which affects the infrastructural needs and suitability of locations for energy sites). Coastal protection was most often thought to be directly affected by activities through physical damage to kelp and seagrass beds and through pollution, and some suggested that recreational fishing vessels crowd estuaries and fjords, destroying habitat that support wave attenuation, and themselves generate additional wake that can risk coastlines. Most experts suggested that aquaculture is predominantly directly affected by some activities (such as land based runoff) but indirectly through others (such as invasive and disease spread from fishing vessels and ships). Fisheries and recreation were both suggested to face both direct and indirect impacts according to experts. Many experts suggested that changes to foodwebs and other ecological dynamics result in indirect impacts along with direct impacts from all types of human impacts. 69   Figure 3.6 The proportion of each type of impact pathways (direct, both direct and indirect, indirect, no impact, and unsure) from four categories of activities and stressors to the eight ecosystem services, as indicated by experts.  3.4 Discussion 3.4.1 Including Social Criteria Lead to Greater Accounting of Impact Considering social criteria of risk in addition to biophysical criteria leads to more severe cumulative impact scores and a greater diversity of impact pathways. However, our results indicate that considering social criteria does not greatly affect the rank of ecosystem services facing impact or the hotpots of areas facing impact. Our results suggest that impact maps of ecosystem services that only consider biophysical criteria may accurately generate conclusions about what services face greatest impact and where they face greatest impact. Ours is an initial 70  investigation into the importance of social criteria for ecosystem service impact, and expert scores of risk criteria may fail to emphasize social criteria because of two important biases. First, many of the experts taking part in our survey have ecological and biophysical training. Second, most prominent frameworks of ecosystem service change represent impacts as mediated solely through the biophysical community (Collins et al. 2010; Rounsevell et al. 2010; Kelble et al. 2013), which may affect how experts think about impacts. In cases where there are important impacts that overwhelmingly impact ecosystem services through social criteria, excluding these criteria may lead to different rankings of threatened ecosystem services and different map hotspots. Determining how prevalent these cases are in different settings remains to be seen. Our results indicate that studies based only on biophysical criteria may underrepresent the processes that generate impact to ecosystem services. Considering the ecosystem service cascade from service supply through service delivery through satisfying values (Haines-Young and Potschin 2010) may lead to detailed understanding about impacts and potential responses to these impacts.  3.4.2 Different Ecosystem Services Are Impacted in Different Ways Impact on ecosystem services is a function not only of spatial overlap with concurrent activities and stressors, but the biophysical and social risk of those activities and stressors to ecosystem services as well. Many ecosystem services face high impact in the area between Vancouver Island and the mainland, which is an area that previous studies focused on impacts to habitats also indicated as areas of high impact (Ban and Alder 2008; Halpern et al. 2009; Ban et al. 2010; Murray et al. 2015a; Murray et al. 2015b). Not all ecosystem services have impact hotspots here, however, reflecting the importance of accurately mapping ecosystem services. While our study alongside previous ones may share similar patterns of human activity, the distribution of 71  ecosystem services themselves is important in determining where areas of high impact are. The marine InVEST models use data of environmental process and human activity to spatially represent ecosystem services, allowing us to directly model ecosystem services (Guerry et al. 2012).  Accurately representing the overlap of activities and stressors on ecosystem services generate additional insights. Knowing where ecosystem services are at highest risk can allow managers to assess impact relative to areas of high demand (Wieland et al. 2016). A high concentration of high risk areas to coastal protection was found in areas close to population centers (in the southern Strait of Georgia), partly because the human activities that might benefit the most from coastal protection – human settlements – also provide the largest impact to coastal protection. Spatial representation also allows for an understanding of whether an ecosystem service faces high risk on account of large spatial range though it faces low per-area impact (such as aesthetics), versus ecosystem services that face low total impact because of limited geographic range despite having high per area impact (such as shellfish and finfish aquaculture). Aesthetics was found to be the least impacted area-specific ecosystem service, indicating that a beautiful coast may mask a highly impacted coast. Explicit inclusion of risk criteria is important because activities and stressors with extensive spatial range and high overlap with ecosystem services do not necessarily generate high impact. Similar to a recent cumulative impact mapping study on coastal ecosystems in British Columbia (Murray et al. 2015b), we found climate change impacts to be important stressors, highlighting the importance of their inclusion in analysis, and cautioning results from mapping studies that do not include them (Ban et al. 2010). Climate change stressors exist across the entire marine 72  system along the British Columbia coast and consequently fully overlap with every ecosystem service we modeled, yet impact on some ecosystem services is driven largely by climate impacts (such as fisheries and aquaculture) while others are largely indifferent to climate change stressors (such as aesthetics, coastal protection and potential energy generation). Indeed, kelp and seagrasses associated with coastal protection may benefit with ocean acidification (Kroeker et al. 2010; Kroeker et al. 2013). Sea-level rise was indicated as a high risk stressor especially to coastal protection, but we did not have spatial data for sea level rise.  This dichotomy between global and regional impacts may exacerbate in the future, as experts suggested that future climate change impacts (specifically warming and ocean acidification) will be a higher risk to those climate-sensitive ecosystem services compared to current conditions, while climate-insensitive ecosystem services will face similar risk levels. For these latter ecosystem services, potential future development may pose a greater cause for concern. Future oil spill potential related to planned developments of oil and gas with associated marine shipping poses a significant risk to these ecosystem services. Previous efforts to compare climate change impacts with future developments in British Columbia indicated that climate change has greater regional scale impact across ecosystem types but lower local impact (Murray et al. 2015b). We show that some ecosystem services – in contrast to ecosystem types – show varying degrees of risk to different types of stressors, leading to insensitivity to climate change stressors for some ecosystem services at local and regional scales. 3.4.3 Social Dimensions of Risk Are Important for Impact Characterization Experts in our survey treated individual social criteria with comparable importance to biophysical criteria when ranking scenarios. Social dimensions are definitional to ecosystem 73  services, yet are often overlooked in quantitative assessments. Relying on physical proxies of ecosystem services or considering ecological components is shortsighted for two reasons. First, any quantitative measure of impact is likely to be an underestimate (Tallis et al. 2012). Social factors, such as how people perceive an ecosystem service, can regulate the extent to which people enjoy and benefit from the ecosystem service (Chan et al. 2012a). For example, open-pen finfish aquaculture practices are perceived negatively by many people in British Columbia (Klain and Chan 2012), creating a self-stigmatized industry. Whether public perceptions on finfish aquaculture are warranted or not, it affects aquaculture as the aquaculture industry has launched marketing campaigns to fight its reputation (www.bcsalmonfacts.ca). Second, many ecosystem services can be impacted largely (even solely) through social means. Experts indicated that potential wave and tidal energy production face risk from fisheries and ports partly through the competition for space, as access to suitable power generation sites can be blocked or zoned out by competing interests for the area. Failing to consider these social pathways of impact signals an epistemic failing of analysts concerned with impacts to ecosystem services.   3.4.4 Modeling Impact Should Account for Pathways of Impact Ecosystem services may require different data and representation techniques than ecosystem types. Unlike ecosystem types, ecosystem services are not taxonomic variants of a geographical theme; ecosystem services do not only exist on a landscape but are related to people’s values and ability to obtain them (Chan et al. 2012a; Chan et al. 2012b). Ecosystem services face risk to the underlying ecosystem that supplies them, as well as risk from processes limiting people’s ability to obtain and enjoy them. The same activity may have different impact pathways on two different ecosystem services because one ecosystem service is primarily impacted through a 74  change in species density and another is impacted primarily because the activity restricts people to a region through property rights and trespassing laws. The greater diversity of potential pathways of impact that ecosystem services face arguably puts greater emphasis on understanding the causal processes of impact for ecosystem services than for ecosystem types. Given the diverse kinds of ecosystem services that exist, a common spatial representation of specific human activities and stressors across ecosystem services may produce misleading results in two important ways. First, the impact pathway important for the ecosystem service should dictate the size of the zone of influence (Ban et al. 2010). Many experts in our study suggested that aesthetics are directly impacted from most activities and stressors, and that what matters is the physical footprint of any activity. We have onshore mining spatially represented to account for acid mine drainage and tailings that occur kilometers away from mines themselves. This area of influence is likely appropriate when mapping impact to ecosystem services affected by these processes, such as fisheries and aquaculture, but it may lead to overestimated overlap of mining impacts and aesthetics. Future efforts to map impacts on ecosystem services should match the spatial representation of activities with relevant impacts. Second, not all experts understand the impact pathways the same, which means they do not answer the same questions. The precedent set here using expert surveys, in conjunction with impact mapping, asks experts to assess vulnerability/risk to an activity “considering all relevant impacts”. This open question framing allows for a tractable survey, yet our results suggest that what is considered in “all relevant impacts” may vary from expert to expert for a given human activity. What’s hidden in our resulting maps is a significant epistemic uncertainty that can be reduced with appropriate elicitation strategies (Regan et al. 2002). Future expert elicitation processes should emphasize 75  specific pathways when assessing risk, even if it means batching surveys into sets of different impact pathways so different experts quantify risk to different impacts.  3.4.5 Limitations and Opportunities While we present advancement in cumulative impact mapping – namely representing ecosystem services and accounting for impact along biophysical and social dimensions – and recommend data considerations specifically for ecosystem services, we must also acknowledge persistent limitations of impact mapping. Most importantly these include a static representation of impact and a simplistic model of cumulative impacts (Ban et al. 2010; Halpern and Fujita 2013; Murray et al. 2015a). Though experts considered temporal criteria of exposure as less important than area of influence, they were still important components of risk, showing that temporal considerations are essential. Spatial models are often snapshots in time, and though we include some temporal dynamics (assessing risk of foreseen impacts) there are many important temporal aspects of impact that are not captured. We do not represent future impacts spatially (but see Murray et al. 2015 for a spatial analysis of proposed projects), though understanding future impacts would be highly valuable to managers. We also do not account for historic impacts. By focusing on contemporary impacts we set a contemporary benchmark and do not consider change from ecosystem service states that may be more ideal, such as times in the past when overfishing was not as prevalent (Pauly 1995; Pinnegar and Engelhard 2008).  The cumulative impact model we employ assumes an additive, relative model of impact with no upper bound. Both activity intensity and risk scores were normalized between 0-1, so components of the model have measurement boundaries, but the cumulative impact can aggregate indefinitely. Empirical studies have shown additive cumulative impacts to occur in a 76  minority of situations (Crain et al. 2008; Darling and Côté 2008). Synergistic impacts – when the total impact is greater than the sum of component impacts – occur often, especially when more than two impacts co-occur (Harley et al. 2006; Crain et al. 2008; Halpern et al. 2008a). Antagonistic impact – when total impact is less than the sum of component impacts – are also prevalent, and have been shown when global impacts interact with local impacts (Brown et al. 2013; Brown et al. 2014). The theoretically limitless measure of impact produced by this model also assumes that impacts can accumulate indefinitely, and that thresholds do not exist (Halpern and Fujita 2013). These are obviously false assumptions, but this model can still provide broad insights into the relative impact faced by multiple ecosystem services. Despite modeling limitations, mapping cumulative impacts to ecosystem services allows for unique planning opportunities. Ocean managers can use this approach to assess the spatial feasibility of potential coastal uses, as we show for potential wave and tidal power generation. By mapping areas of potential energy generation, we see that the areas of lowest threat to energy generation are the central coast and some areas between Vancouver Island and the mainland. These are relatively unpopulated areas, which may mean higher infrastructure costs to establish turbines, but these costs may be worth avoiding impediments in more populous areas. 3.5 Conclusion By mapping cumulative impacts to ecosystem services, we can better steward our ecosystems and  understand the dual relationship of humans to the environment: as agents of change and beneficiaries of services (MA 2005). We have demonstrated the kinds of rich insights that can be gained from mapping impacts to ecosystem services, including discovering where, and by what means, different ecosystem services face greatest impact, determining what ecosystem services 77  are comparatively worse (or better) off under current conditions, understanding the ways in which impacts manifest, and assess spatial feasibility for new ocean uses. We have also demonstrated the importance of considering social criteria in assessing impact. We argue that considering social criteria is not only important to more fully understand impact, but also to plan effective management responses. We have also pointed to areas of future methodological refinement, and encourage greater innovation in cumulative impact mapping. Ecosystem services can be highly location specific (Chan et al. 2012a), so future risk assessments are warranted in new places.  Understanding risk and impact to ecosystem services should be an essential management priority to maintain the flow of services we benefit from.    78  Chapter 4: Prioritizing Management for Cumulative Impacts 4.1 Introduction Ecosystems face multiple anthropogenic stressors (Sanderson et al. 2002; Halpern et al. 2008b), requiring management strategies that address the potential cumulative impact of multiple stressors to meet environmental goals, such as preservation, ecosystem services, or others (McLeod and Leslie 2009; Brown et al. 2014). For such management to be effective and efficient requires understanding the processes by which cumulative impacts occur, and locating key intervention points (Niemeijer and de Groot 2008). Yet few management plans regulate stressors across sectors and account for ecological complexity across an ecosystem (Arkema et al. 2006). Because multiple stressors arise from many sectors, managing for cumulative impacts across multiple species often translates into managing human stressors across sectors. However, single sector management may be justified when the most important stressors occur within a given sector. When and where single- versus cross-sectoral management is needed has not often been shown empirically (Halpern et al. 2010; Fulton et al. 2014). We propose using network analysis to address this need and help identify and prioritize key activities and stressors for management. Understanding the mechanisms of impact and their individual importance to cumulative impacts should allow for a more effective prioritization of management targets (Niemeijer and de Groot 2008). Networks based on causal pathways of impact have been used to analyze the structure of cumulative impact pathways, including identifying nodes with the most connections to other nodes, and to understand mechanistically similar pathways (Niemeijer and de Groot 2008; Knights et al. 2013a), but they do not account for the magnitude of impacts. More recently, 79  networks have been constructed with associated impact weights, characterizing cumulative impacts in marine systems (Cook et al. 2014a). However, impact weights have only been calculated to identify the nodes that contribute impact across an ecosystem, and as of yet no analysis has determined what nodes are most important in regulating impact to environmental components of management concern.  Environmental managers routinely, if implicitly, assume that certain nodes within an impact network disproportionately regulate the most risk to environmental components, offering the greatest leverage for management action. But do such leverage nodes exist, or is risk to environmental components distributed relatively evenly across an impact network? If such leverage nodes exist, then management can benefit from prioritization schemes that identify and target them. If instead they are rare or do not exist, then management should regulate risk across an impact network.  Here we use Bayesian Belief Networks (BBNs) to test if leverage impact nodes exist and whether they are general to multiple conservation targets or unique for each one. We analyze impact networks consisting of nodes possessing quantitative weights representing the contribution of risk to diverse marine species. We build upon a recent pilot risk analysis for seventeen marine species contributing to fifteen ecosystem services in coastal British Columbia (Clarke Murray et al. 2016). This analysis uses a mechanistic pathway structure to quantify risk (with associated uncertainty) across numerous individual causal pathways, incorporating indirect risk through ecological dependencies, and calculates total risk with an additive model. Causal pathways in the risk assessment link human activities (or long term pressure) to specific stressors to species to ecosystem services, and we use these stages to create networks with nodes for 80  activities, stressors, species, and ecosystem services (see Appendix C for more information). Although the pilot study was not designed for a network analysis, we converted it to a BBN network for this study by calculating the contribution of activity and stressor nodes to risk (or consequences of risk on species and ecosystem services) and relating these nodes to each other according to the network structure (see Methods). We demonstrate the utility of this approach by analyzing herring (Clupea pallasii) management in coastal British Columbia (BC), Canada to test where leverage nodes exist. We then test how these revealed nodes compare with the current intervention points in herring management. We first characterized the prevalence of leverage nodes by identifying nodes that contribute over twice the average total change in risk probabilities from activities and stressors associated with a change in risk probabilities to guarantee target species or species groups are categorized as low risk. We then explored four different management scenarios to help identify leverage nodes and assess optimal management strategies: 1) target SAR, a multi-species approach for herring and species at risk (SAR), 2) habitat protection, an approach aimed at important habitat forming species, 3) target herring, a single-species approach, and 4) IFMP, a single-species herring approach that assumes leverage nodes from the herring Integrated Fisheries Management Plan (IFMP). Finally, we assessed indirect effects of different management scenarios on all 17 species to see which approach best serves ecosystem-wide objectives, and on the supply of 15 ecosystem services to see which approach best reduces risk across ecosystem service supply.  4.2 Methods To explore impact networks and determine leverage nodes, we employ the following steps. First, we create an impact network based on a risk assessment for coastal BC. Next, we use the risk 81  assessment to calculate conditional probability tables for a BBN. We then use the BBN to identify leverage nodes for the species in the network, and scenarios of multiple species. We compare the revealed leverage nodes for herring against the assumed leverage nodes in the herring IFMP. Finally, we assess the consequences across the biological community and supply of ecosystem services from the various scenarios we assess. 4.2.1 Case Study – North Coast of British Columbia The Pacific North coast of British Columbia, Canada, is a site of extensive natural resource development, is the territorial home of coastal First Nations people, and renowned as a destination for ecotourism (Dawson 2005). The region has unique species and habitats, such as glass sponge reefs (Cook et al. 2008), seabird populations (Smith and Hyrenbach 2003), culturally significant salmon, eulachon, and herring (Turner and Clifton 2009; Chan et al. 2012a), and both resident and transient orca populations (Ford et al. 2000). The region supports a broad suite of human activities and redevelopment that take advantage of diverse ecosystem services. Sea-based activities include fishing (commercial, First Nations food fish, and recreational), aquaculture, tourism, utilities and transportation. Coastal activities include human settlements, ports and marinas, and log storage. Land-based activities affect the coast through runoff and include forestry, agriculture, mining and pulp and paper mills. There are also persistent and global drivers of change including climate change and persistent organic pollutants  4.2.2 Bayesian Belief Networks Given that impact networks are directional and hierarchical (that is, there are no feedback loops) they can be analyzed with Bayesian Belief Networks (BBNs) to identify leverage nodes. So long 82  as nodes in the network can be parameterized with probabilities of states (e.g. a node for a species faces 40% probability of low impact and 60% high impact) a BBN can be constructed to relate the probability states of nodes by their connections through conditional probability tables (Marcot et al. 2001). A conditional probability table describes the probability of a set of nodes meeting some criteria contingent on the state of other nodes (Jensen 1996; Marcot et al. 2001). With a BBN, therefore, a researcher can set the impact level on a species (or set of species) at a desired level and see what activity and stressor nodes change most from initial probabilities to identify the leverage nodes. We use a risk assessment that additively accumulates risk scores to multiple species and species groups across multiple causal pathways based on the coast of British Columbia to generate an impact network, and use Bayesian Belief Network (BBN) populated with risk estimates to identify leverage nodes within the network. 4.2.3 Calculating Risk  We adapted a pilot cumulative risk assessment on 17 different species and species groupings on coastal British Columbia, Canada (Clarke Murray et al. 2016; Clarke Murray et al. in prep). Briefly, this risk assessment considered direct risk and indirect risk (risk experienced by one species is carried over to another species through ecological dependencies). Direct risk pathways are represented in Figure 4.2 and ecological dependencies (that regulate indirect risk) are represented in Figure C.1. Human activities and stressors are listed in Table C.2, and species and species groups are listed in Table C.3. After the species were selected, all relevant stressors were listed and explicitly connected to the species based on literature reviews. The human activities connected to theses stressors were linked to the stressors, to create a network of pathways of effect on these species (a network of 83  human activities →stressor→species). Risks to species come from ocean activities, such as shipping, fisheries, terrestrial activities, such as land use, and long term stressors, such as climate change.  After the network was assembled, a cumulative risk assessment was undertaken to quantify risk from all pathways to the species. Risk scores were literature derived, providing scores and measurements of uncertainty for each pathway. The risk calculation was made up of characteristics of exposure and consequences of impacts to species, with each score having an associated uncertainty measure (see Appendix C).  In this risk assessment, total risk (TR) to each species i from each impact pathway j is a product of the criteria Temporal Scale (TS) and Spatial Scale (SS) of a stressor from an activity, Intensity (I) of the stress from an activity, and Consequence (C) of the impact pathway according to the formula   =   , × , × , × , Because exposure criteria (TS, SS, and I) are on the scales 0-4, 0-3, and 0-3, respectively, and the consequence criterion (C) is on a 0-6 scale, C is squared to give exposure and consequence equal weighting. To incorporate indirect risk, we assumed that risk flowed through trophic links of species according to 10% rule, so that predators accumulated 10% risk of prey (Lindeman 1942). We also assumed that habitat-providers transfer 10% of their risk to habitat-receptors (Figure C.1).  84  We also linked the species to 15 ecosystem services they provide (Table S3). We treated risk to ecosystem service provision (ESR) to be a sum of the risk of the species that contribute to each ecosystem service, normalized by the number of species contributing to ESR, according to the formula  =  1 , where m is the ecosystem service. We stress that this risk is on ecosystem service provision alone, and not the full risk to ecosystem services, as we only capture risk to the ecological side of ecosystem services, and do not account for the social dimensions (Tallis et al. 2012). Every criterion had an associated uncertainty score, and we used this score to incorporate uncertainty across the network (DFO 2015). The uncertainty scores are on a relative scale, and we use these scores to generate distributions of criteria, assuming a normal distribution with the standard deviation related to the uncertainty score (see Appendix C for more information on uncertainty scales). Using a Monte Carlo resampling procedure, we could randomly select values of Temporal Scale, Spatial Scale, Intensity, and Consequence from their distributions to generate a distribution of Total Risk for every species and ecosystem service (DFO 2015). Because we knew the structure of each impact pathway and the risk from each pathway, we could also calculate how much risk originates in each activity (OR), and how much risk flows through each stressor (RF), according to the formula  =   , × , × , × , 85  and  =   , × , × , × , Where g is the human activity, h is the stressor, k is the mechanistic pathway from stressor to species, and l is the intersection of activity and species through stressor h. Though this risk assessment provided us with the best available data to explore impact networks and leverage nodes, this risk assessment was originally intended to characterize current risk to species, with uncertainty measures characterizing certainty in particular measures of risk. We use the final risk distribution derived from these scores of risk criteria and associated uncertainty as a proxy for the scope of change in risk throughout the network, which is necessarily smaller than the actual scope of change for risk. As a result, our findings should not be taken as conclusive for specific management but rather illustrative of the possibilities of our methods. 4.2.4 Creating the Conditional Probability Table The risk scores are relative in nature, and meant to make sense in relation to other nodes in a stage (O et al. 2015). Using the 33rd and 66th percentile of risk scores within each level (activity, stressor, species, ecosystem service), we set categories of low, medium, and high risk. For example, if for a particular iteration of the Monte Carlo procedure the stressor “oil spill” has a risk score that falls above the 66th percentile of the risk scores in the stressor category, then “oil spill” would be categorized as having high relative risk for that iteration. If, for the next iteration, the risk score for “oil spill” fell between the 33rd and 66th percentile of the risk scores in the stressor category, then “oil spill” would be categorized as having medium relative risk. Using a 86  Monte Carlo process to generate these relative risk states among all nodes concurrently allowed us to create conditional probability tables to populate a BBN.  We use these categories and cutoffs for illustrative purposes only, as we recognize that they are arbitrary designations and thresholds, however they are useful in examining changes to risk probabilities broadly and according to the same criteria. Setting the risk levels this way meant that in some cases a node invariably faces risk at a certain level, preventing us from manipulating that node. For example, killer whales (Orcinus orca) consistently face comparatively high levels of risk based on our risk data, and finfish aquaculture invariably produces risk at high levels. BBN analysis was conducted using the R package gRain (Højsgaard 2012). 4.2.5 Identifying Leverage Nodes Leverage nodes for management should reflect what management strategies will drive outcomes from current conditions to goal states (Arkema et al. 2006). With a BBN therefore, we can set the risk level of a species (or set of species) at a desired goal level and determine what associated activity and stressor nodes change most from initial probabilities (current conditions) to identify the leverage nodes. We define leverage nodes as those nodes that contribute over twice the average total change in risk probabilities from activity and stressor nodes when comparing initial conditions to conditions where target species are guaranteed to be at low risk. Once leverage nodes were identified for all species, we could determine if they exist within different sectors (fisheries, sea-based activities, land-based activities, or long term stressors) based on the categories of the risk assessment (Clarke Murray et al. 2016). 87  4.2.6 Scenarios Herring are an important commercial, recreational, and cultural fishery in BC (27). We identify the leverage nodes in the network by determining the nodes that contribute the most to reducing risk to herring (and keeping herring at a low risk state) from initial conditions. We set the herring node as 100% probable of being at a low risk state and determining the activities and stressors whose risk state probabilities deviate the most from initial conditions (we call this the target herring scenario). Next, we compare the revealed leverage nodes as identified by the target herring scenario with an existing management plan, Fisheries and Oceans Canada’s (DFO) Integrated Fisheries Management Plan (Kanno 2015). The goals of the strategy are to protect herring, while acknowledging concerns for species at risk (SAR), including Cassin’s Auklet, Stellar sea lion, humpback whales and killer whales. This IFMP outlines implicit assumptions about leverage nodes in an impact network by indicating the activities and stressors it will manage. Specifically, it assumes that the leverage nodes in the impact network are three activities (seine fisheries, gillnet fisheries, recreational fisheries), and one stressor (bycatch). We represent an ideal of this scenario (the IFMP scenario), assuming that management actions will be completely effective. To represent the IFMP scenario, we set the risk state for these assumed leverage nodes to be 100% low risk. The resulting change to the herring node, SAR, and all species from initial conditions represent the consequences of this management plan. We compare the revealed leverage nodes identified by the target herring scenario with these assumed leverage nodes to see how closely management assumptions resemble actual leverage nodes. We also assess 88  predicted results to risk state probabilities for the provision of ecosystem services, paying particular attention to risk to the ecosystem service wild harvest as the IFMP is a fisheries plan. Finally, we investigate two additional scenarios. The first is a habitat provisioning scenario, whereby the species nodes for habitat forming species are kept at low levels (the habitat protection scenario). Though not explicitly spatial, this scenario can provide insight for a protected area plan meant to protect habitat. The second scenario explores the revealed leverage nodes for managing herring and SAR at low risk levels (the target SAR scenario). We explore how these scenarios compare with the herring-specific scenarios. 4.2.7 Estimating the Number of Species and Ecosystem Service Provision at Risk Using the resulting probabilities of species and the provision of ecosystem services at each risk level from the BBN analysis, we established a random resampling procedure to select risk levels for each species and ecosystem service across 1000 iterations. Using these data on risk levels we calculate bootstrap 95% confidence intervals (using the BCa method) of the mean (Shiue et al. 1993). 4.3 Results 4.3.1 Identifying Leverage Nodes In total there are 86 nodes in the network. The number of activity and stressor nodes associated with individual species ranges from 35-48. In every species and species group we investigated (11 species), we found leverage nodes. The number of leverage nodes per species varied from 1 to 11 (comprising 3-23% of total nodes contributing to risk), with humpback whales having the 89  fewest and seagrasses having the most (Table 4.1). The mean number (± standard error) of leverage node in our sample is 6.91 (±1.02). In every case, leverage nodes are found across sectors, and in 64% of species we found leverage nodes across all four sectors that generate impact (fisheries, sea-based activities, land-based activities, and long term stressors). The SAR in our sample (humpback whale, Cassin’s auklet, Stellar sea lion) have relatively few leverage nodes (1, 2, and 3 respectively), each of which can be managed by focusing on a subset of sectors. Based on our results, we categorize three kinds of impact networks based on the location of leverage nodes within a network (Figure 4.1).  We found 18 leverage nodes in stressor nodes and 1 leverage node in activity nodes across our sample of species. For example, if a species is primarily affected by three stressors, harvesting, bycatch, and habitat alteration, and these three stressors are primarily driven by a single activity, such as bottom trawl fisheries, then the activity node for bottom trawl fisheries will be a leverage node, because this fishery regulates the impacts. We call this system a “top-down leverage network”. In contrast, leverage nodes can also be found at the stressor stage. For example, if a species is primarily affected by acoustic stress, and there are multiple shipping, fisheries, and other activities contributing to acoustic stress, the node of acoustic stress is a leverage node. Potential management actions on this node could be limiting boat activity where the species occurs, or applying vessel quieting technologies to the vessels that contribute to acoustic noise (Simmonds et al. 2014). We consider this a “mid-point leverage network”. Finally, there are systems where leverage nodes can be found at the activity stage as well as the stressor stage. We consider these as “mixed leverage networks”. 90   Figure 4.1 Three basic structures of impact networks. Leverage nodes are outlined in red. Faded nodes represent nodes that are minor contributors to risk. 4.3.2 Leverage Nodes for Multiple Species Management In order to manage SAR along with herring at low risk, there are 5 leverage nodes (10% of the 48 nodes contributing risk to these species). There are fewer leverage nodes in the target SAR scenario than the target herring scenario, but the leverage nodes common to both require a greater decrease in risk for the target SAR scenario. For example, oil spills and acoustic stress have more weight as leverage nodes in the target SAR scenario compared to the target herring (as indicated by a greater change in risk state probabilities from initial conditions compared with the target herring scenario).  To protect habitat-forming kelp and seagrass, there are 9 leverage nodes across fisheries, sea-based impacts, land based impacts, and long term impacts (19% of the 48 nodes contributing risk to these species). The leverage nodes are direct capture, nutrient input, incidental mortality (from 91  small vessels, large vessels, and fisheries), marine debris, oil spill, temperature change, and changes to water flow. The diffuse leverage nodes for the habitat protection scenario do not have risk probabilities as different from initial conditions as other scenarios (often only being 5-13% different). 4.3.3 Comparing Revealed and Assumed Leverage Nodes in Herring Management The BBN analysis of the target herring scenario allowed us to compare revealed leverage nodes with the assumed leverage nodes of the IFMP scenario. We found that the risk level probabilities of 35 of 54 driver and stressor nodes in the impact network changed from initial conditions to the target herring scenario. Of these, we find 7 leverage nodes for herring (Table 4.1): direct capture, acoustic stress, oil spill, change in water flow, marine debris, temperature change, and nutrient input. There is no prominent driver for direct capture, oil spills, or nutrient input, so no prominent driver was identified as a leverage node. According to the pilot risk assessment, acoustic stress on herring is driven by finfish aquaculture. Based on our classification of risk levels, finfish aquaculture was invariably a driver of high risk comparable to other drivers (its risk profile is 100% high, see Methods for how we classified risk levels), so it is a node that cannot vary among scenarios. The inelastic nature of the aquaculture driver node means that the stressor node is the leverage node. Marine debris and temperature change have similar inelastic (or nearly inelastic) driver activities. The target herring scenario is a mid-point leverage network. 92   Figure 4.2 The impact network linking drivers, stressors, species and ecosystem services of the categories listed on the right (a given category may have multiple items, e.g., the driver “fisheries” includes seine fisheries, gillnet fisheries, and others). Numbers in the nodes correspond to the codes on Supplementary Tables C.2 and C.3. Nodes focused on by the Herring Integrated Fisheries Management Plan are outlined in black, while the top three revealed leverage nodes are outlined in blue.  93  Table 4.1 The leverage nodes for species, grouped by sector (fisheries, sea, land, and long term impacts). The number in brackets indicates the number of leverage nodes revealed through our analysis for each species. Species (number of leverage nodes) Fisheries Sea Land Long Term Herring (7) Direct Capture Acoustic stress; Oil spill; Change in water flow; Nutrient input Nutrient input Marine debris; Temperature change Geoduck Clam (8) Habitat disturbance; Sedimentation; Direct capture Large vessel invasive species; Habitat disturbance; Invasive species; Nutrient input Change in freshwater flow; Sedimentation; Nutrient Input Marine debris Cassin's Auklet (2) Small vessel acoustic Small vessel acoustic   Marine debris Cold Water Coral (10) Sedimentation; Small vessel incidental mortality Large vessel incidental mortality; Incidental mortality; Small vessel incidental mortality; Nutrient input; Oil spill; Invasive species; Change in water flow Sedimentation; Nutrient input Marine debris; Temperature change Humpback Whale (1) Small vessel acoustic Small vessel acoustic     Kelp (9) Direct Capture; Sedimentation; Small vessel incidental mortality; Habitat disturbance Invasive species; Small vessel incidental mortality; Habitat disturbance; Oil spill; Large vessel incidental mortality Sedimentation  Marine debris; Sea level rise Prawn (9) Direct capture; Sedimentation; Seine Fisheries  Nutrient input; Change in water flow; Oil spill Nutrient input; Sedimentation; Change in freshwater flow Temperature change; Marine debris Stellar Sea Lion (3) Small vessel acoustic Small vessel acoustic; Large vessel incidental mortality; Acoustic stress     Seagrasses (11) Direct capture; Small vessel incidental mortality; Sedimentation Large vessel incidental mortality; small vessel incidental mortality; Nutrient input; Incidental mortality; Oil spill; Change in water flow Nutrient input; Sedimentation  Marine debris; Temperature change; Sea level rise Sponges (9) Small vessel incidental mortality; Sedimentation Large vessel incidental mortality; Incidental mortality; Nutrient input; Oil spill; Small vessel incidental mortality; Change in water flow Nutrient input; Sedimentation Marine debris; Temperature change Zooplankton (7)   Oil spill; Change in water flow; Nutrient input Nutrient input Temperature change; Marine debris; Ocean acidification; Persistent organic pollutants Target SAR scenario (5) Direct capture; Small vessel acoustic Small vessel acoustic; Acoustic stress; Oil spill; Nutrient input Nutrient input   Habitat Protection scenario (9) Direct capture; Small vessel incidental mortality Nutrient input; Large vessel incidental mortality; Small vessel incidental mortality; Incidental mortality; Oil spill; Change in water flow Nutrient input; Marine debris; Temperature change 94  In comparing the revealed leverage nodes of the target herring scenario with the assumed leverage nodes of the IFMP, we find little overlap. First, the IMFP assumes the leverage nodes to be three drivers (seine fisheries, recreational fisheries, gillnet fisheries), and one stressor, bycatch. Our analysis ranks these assumed leverage nodes as the 10th, 13th, 14th, and 28th most influential nodes for herring, respectively (out of 54 total, Table C1). The IFMP effectively reduces direct capture by keeping three types of fisheries low, but does not address many of the other influential nodes, which may be important priorities to ensure management efficacy (Figure 4.2). 4.3.4 Consequences for Species The target herring scenario and the target SAR scenario predefine herring at low risk. In contrast, the probabilities of herring being at low or high risk in the IFMP and the habitat protection scenarios are 15% and 13%, respectively. Both extreme risk categories are more probable in these scenarios compared to initial conditions (11% low and 11% high). In turn these two scenarios have reduced probability of medium risk (73% and 73%, respectively) compared to initial conditions (79% medium risk, Figure 4.3).  95   Figure 4.3 The probabilities of low, medium, and high risk for four species depending on the five different scenarios. White represents low risk, medium grey represents medium risk, and dark grey represents high risk. The target SAR scenario sets all SAR species to low risk. The target herring, IFMP, and habitat protection scenarios have comparable risk level probabilities for Stellar sea lion (~60% high and ~40 medium risk), humpback whale (~25% high, ~55% medium, and ~20% low risk) and Cassin’s auklet (~15% high, ~35% medium, and ~50% low risk). These scenarios are also comparable to initial conditions. The IFMP and target herring scenarios have slightly more favourable risk probabilities for Stellar sea lions and humpback whales than the habitat 96  protection scenario, and the habitat protection scenario has slightly more favourable risk probabilities for Cassin’s auklet then the target herring and IFMP scenarios. Because risk levels were determined in relation to the risk scores across species, initial settings for the BBN predict an equal number of species at low, medium, and high risk relative to each other (Figure 4.4A). The scenarios with the most species at low risk are target SAR, habitat protection, and target herring. All scenarios except for the IFMP scenario are predicted to manage more species at low risk compared to initial conditions, based on 95% confidence intervals. The target SAR scenario is the only scenario predicted to manage fewer species at high risk compared to initial conditions, with other scenarios managing more species at high risk. All scenarios result in fewer species at medium risk compared to initial conditions. 4.3.5 Consequences for Ecosystem Service Provision Specifically considering the risk to the provision of wild fisheries harvest, the IFMP and habitat protection scenarios perform comparably to initial conditions. In contrast, the target herring and target SAR scenarios are predicted to be less risky for wild fisheries harvest, as herring is one of the fish species contributing to wild harvest. Both of these scenarios have a 78% probability that the provision of this ecosystem service is at low risk, a 15% chance at medium risk, and a 7% chance at high risk.  Considering risk across the provision of multiple ecosystem services, the IFMP is predicted to perform similarly to initial conditions (95% confidence intervals overlap across all risk levels, Figure 4.4B). The habitat protection and target SAR scenarios are predicted to manage the provision of the fewest ecosystem services at high risk. The target herring scenario is predicted 97  to manage the provision of fewer ecosystem services at high risk than initial conditions. The target herring, habitat protection, and target SAR scenarios are all predicted to manage the provision of more ecosystem services at low and medium risk than the IFMP, with the habitat protection scenario managing the most at low risk.  Figure 4.4 The number of A) species and B) ecosystem services predicted to be at low, medium, and high risk from the five different scenarios. 98  4.4 Discussion We found leverage nodes within the impact network for coastal BC for every species we investigated, and for every multi-species scenario we investigated. We found between 1 and 11 leverage nodes for species, and did not find that targeting multiple species increases the number of leverage nodes compared to the individual species within the multispecies assemblage. Though there may be cases where managing multiple species has a larger number of leverage nodes, where there are few common prominent activities and stressors that regulate risk among species, management can focus attention on reducing risk from a few activities and stressors. The widespread existence of leverage nodes in our study of the pilot risk assessment suggests that management targeted on a few nodes can offer considerable leverage for reducing risk. Encouragingly, we found relatively few leverage nodes for SAR (1-3 leverage nodes), indicating that effectively managing these species may pose a relatively simple management challenge. Because species at low population sizes are often more vulnerable to diverse impact than at high populations (Fagan and Holmes 2006), this finding is surprising but may reflect the fact that SAR in our context are species that happen to receive few prominent pathways of risk. The existence of leverage nodes supports the approach of many management plans, such that management is focused and not diffuse across the entire impact network (Arkema et al. 2006). However, in all cases of determining leverage nodes to individual and groups of species, we found that leverage nodes exist across sectors, and in many cases across all broad categories of sectors we assess (fisheries, sea based activities, land based activities, and long term stressors). To the extent that our finding that many leverage nodes must be managed across sectors is 99  universal, it may challenge sector specific approaches  that management plans either choose to employ or are constrained to (Arkema et al. 2006).   Our results suggest that prioritizing management on leverage nodes revealed by impact analysis and BBNs can provide better management outcomes than managing based on assumptions and jurisdictional constraints. When leverage nodes exist at the stressor level, as indicated in two out of the three types of impact networks (Figure 4.1), effectively managing these nodes requires managing all prominent activities that drive this stressor or directly mitigating this stressor. The IFMP is a case in point. Through BBNs and a comparative risk assessment, we show that the IFMP does effectively reduce risk from the stressor “direct capture”, but does not account for other influential stressors that impact herring. Herring face greater probabilities of both low and high risk under the IFMP compared to initial conditions. BBNs are not deterministic models and cannot represent ceteris paribus changes in ecosystems, but relate nodes by their conditional probabilities of co-occurring at various states, which may be a useful analogue to the real world (McCann et al. 2006). Our result of a higher probability of herring at high risk in the IFMP demonstrates that other drivers of risk can more than compensate for the reduction of risk to herring from direct capture. This result indicates that managing direct capture is not enough to be certain of controlling risk to herring. Considering risk to all assessed species and the provision of ecosystem services, the IFMP is indistinguishable from initial conditions, except that there are a higher number of species predicted at high risk and a lower number of species at medium risk. By increasing the probability of both low and high risk in herring, and having a higher number of species predicted at high risk, this analysis suggests that the IFMP might even be a less desirable management scenario for risk adverse managers than leaving things as they are (Slovic 1987). 100  Our analysis suggests that prioritization based on revealed leverage nodes can have beneficial outcomes across common management goals, even when prioritizing environmental components indirectly related to a specific management goal. When herring themselves are not targeted, but management for habitat-forming species is prioritized, the risk profile for herring is comparable to the IFMP with better outcomes for other species (though not SAR) and ecosystem service provision. Seagrasses and kelp act as habitat that multiple species either directly or indirectly depend on (Orth et al. 1984; Trebilco et al. 2015), so when risk levels for activities and stressors are lowered to ensure low risk levels for seagrasses and kelp, they simultaneously contribute to a lowering of risk for other species and provision of ecosystem services. The leverage nodes to maintain habitat-formers are stressors that affect multiple species, including oil spill risk, so mitigating the risk from these nodes has a broad effect on reducing risk to multiple species and the resulting ecosystem service provision. However, managing the wild harvest of fish as an ecosystem service requires managing the leverage nodes that effectively manage the specific fish species that contribute to wild harvest, including herring. Our results indicating the predominance of leverage nodes in species management may be generalizable depending on the types of interaction among stressors. Though the risk assessment employed in our analysis uses an additive cumulative impacts model, evidence suggests multiple stressors on species can  interact additively, antagonistically – where total impact is less than the sum of component impacts – or synergistically – where total impact is less than the sum of component impacts (Crain et al. 2008; Darling and Côté 2008). There are possible situations where effective management species will require managing most stressors, such as when antagonisms persist and most stressors contribute similarly to impact (Crain et al. 2008). There 101  are also possible situations where reducing impact from any individual stressors will provide favourable management results beyond what an additive model may predict, such as when synergisms are present (Crain et al. 2009). However, a growing literature highlights the role of asymmetric and dominant stressors in regulating impact to species (Folt et al. 1999; Darling et al. 2010; Brown et al. 2013; Ban et al. 2014a). Leverage nodes will occur in additive interactions where dominant stressors occur (such as in this study), which would be predicted in a straightforward manner from additive models (Crain et al. 2008; Darling and Côté 2008). Where antagonisms are found, a dominant stressor can drive the magnitude of combined effects, which emphasizes the importance of identifying these dominant stressors (Darling et al. 2010). Even in contexts of synergistic impacts, asymmetries in impact contribution can point to dominant stressors (Halpern et al. 2008a). 4.4.1 Managing Multiple, Diverse Risks In every case we investigated, addressing leverage nodes requires diverse management. In only one case did we find a single leverage node to manage a species (for humpback whales). However, in this case the leverage node is acoustic stress from small vessels, which is a stressor that multiple activities contribute to, including different fisheries, marine tourism, log handling, aquaculture operation, and other boating activity. Though this is a single node, actually managing risk from this node will require regulation across multiple economic sectors.  Our results indicate management plans that account for the most important leverage node but do not address others may not be successful. Despite the IFMP effectively reducing risk from the most important leverage node (direct capture), our analysis suggests that other stressors, including acoustic stress and oil spills, may need to be effectively managed to manage herring at 102  low risk (Gerlotto and Fréon 1992; Hose et al. 1996). Traditionally, management for fisheries relies on regulating catch, with little consideration of other stressors on fish populations (Pikitch et al. 2004). Other stressors, left unchecked, can elicit a comparatively high risk to herring. The focus on selective versus nonselective fishing practices in the fisheries management literature (Zhou et al. 2010) may neglect other important stressors important for fisheries management from changing ocean conditions (Kaufman et al. 2004).  Though potentially costlier than managing single stressors, managing diverse stressors is more likely to reduce risk to multiple species than managing single, or even multiple, similar stressors (Fulton et al. 2014). Where diverse stressors are managed in our scenarios, all show better risk outcomes for the ecological community than the IFMP scenario where functionally similar stressors contributing to catch are managed. We believe two complementary processes contribute to this finding. First, effective scenarios in our study reduced the threat posed by some key stressors that are mechanistically linked to many species (e.g. oil spill and acoustic stress).  Second, reducing impacts from common stressors will likely reduce risk to similar species with similar vulnerabilities. Alternatively, reducing impacts from diverse stressors will likely reduce risk to a greater diversity, and greater number, of species in an ecosystem. Species diversity also predictably contributes to the provision of ecosystem service diversity through trait diversity (Cadotte et al. 2009). The logic that reducing multiple risks supports higher diversity is supported by our findings: scenarios that reduce risk from diverse stressors reduce risk for more species and for the supply of more ecosystem services. Reducing risk to multiple species (especially SAR) and ecosystem services is in line with the aims of the IFMP in terms of 103  coinciding with commitments to many management plans. More generally, it is in line with the system-focus of ecosystem-based management (EBM, McLeod and Leslie 2009). 4.4.2 Limitations Our study links human activities to ecosystem service provision. We are cautious to address risk to ecosystem services in total, because we only capture the biophysical aspect of ecosystem services (Tallis et al. 2012). Services also have social dimensions (such as human access to services and enjoyment) that are prone to risks independent from biophysical risks (Chan et al. 2012a; Chan et al. 2012b). We did not address these social dimensions because of the lack of available data. We used mechanistic pathways of risk through biophysical pathways, but so long as social impacts can be described through pathways, theoretically they can be represented in networks.  As in all BBN approaches, our results are dependent on the network structure we supplied. Other network structures (such as differences in links between nodes or other nodes included) might change the results, so confidence in the results depends on confidence in the input network and data. The risk assessment scoring and network connections used here were created as part of a pilot project and were intended as illustrative for the potential of risk assessment and not intended for use directly by managers without further vetting. Additionally, the pilot risk assessment was focused on characterizing risk as it currently exists, and the uncertainty in the measures used to calculate risk. Though the risk assessment data was the best available, using measurement uncertainty as a proxy to define the scope of risk estimates in the network is incomplete (McCann et al. 2006). The result is that the amount of risk regulated by any particular node may be constrained. This constrained scope likely contributed to the fact that our analysis 104  has some nodes at fixed risk levels (they cannot vary), and may have affected our ability to identify some leverage nodes in some cases. We stress the illustrative nature of our findings to demonstrate the utility of our approach without making specific management recommendations. 4.4.3 Applications EBM is predicated on management across sectors, and though we show that managing few specific nodes in an impact network can be effective, our findings also show that these species with few leverage nodes can still demand management across ecosystem boundaries (such as the land and sea). EBM is also concerned with ecosystem resilience, which is thought to be dependent on the diversity of responses to environmental change in the biological community (Arkema et al. 2006; McLeod and Leslie 2009; Halpern et al. 2010). Though the ecological literature stresses the importance of maintaining response diversity for functioning ecosystems (Elmqvist et al. 2003), there are few tools that allow managers to plan for response diversity in an ecosystem. Using BBNs and impact networks, we propose that managers can plan for response diversity by determining leverage nodes for diverse species and enacting management to reduce risk to them. Our results indicate that management plans may fail even if important aspects of EBM are embraced. The IFMP embraces certain EBM characteristics, such as cross-fisheries management, multi-species consideration, and attention to the human relationship with the environment, but based on the risk assessment data and our analysis, the IFMP may be ineffective. Our results do not discredit EBM properties, but offer warning to those who implicitly promote EBM based on principle alone. EBM can fail management goals when effort 105  and resources are spent inefficiently on managing suboptimal components of a socio-ecological system (Evans et al. 2011; Game et al. 2013; Brown et al. 2015). The emerging emphasis of strategic environmental assessments (SEA) and cumulative impact assessment (CIA) in many regulatory regimes of the world are another avenue for impact networks and leverage nodes to be an important tool (Bina 2007; Fischer 2010). Proposing mitigations are one of the most important steps in the environmental assessment process (Wood 2003), and discovering leverage nodes in impact networks can help determine where mitigations are most needed for environmental protection, and where mitigations provide the best outcomes to diminish impacts from development. Determining what mitigations to apply will be context-specific, but when resources are limited, prioritization can help ensure mitigation is effective. 4.4.4 Working with the EBM Toolbox Engaging with whole-ecosystem management such as EBM and SEA requires a suite of collaborative tools and approaches to operationalize. Much is written to promote these approaches for environmental management, and much is written to outline the primary characteristics of such approaches, but the creation of tools and approaches to carry out these concepts is lagging (Leslie and McLeod 2007; McLeod and Leslie 2009; Curtice et al. 2012). Matching general characteristics to functional tools and approaches is key to help operationalize these concepts, and understand how they can work together. Currently, spatial analysis of cumulative impacts provides important tools to aid in zone planning (Halpern et al. 2008a; Halpern et al. 2009; Halpern et al. 2012a). We see impact networks and leverage nodes complementing impact-mapping tools by prioritizing human activities and stressors that should be excluded from some areas.  106  Our study explores leverage nodes of a scenario whereby habitat-providing species are protected at low levels of risk. This scenario demonstrates the extent that impact networks can reflect protected area planning. Alone, the impact network is limited in planning for zoning as it is not spatial, however where impact mapping analysis is done to prioritize areas to zone for protection, impact networks can help determine which human activities are most important to limit within protected areas. Networks and leverage nodes can also complement structured decision processes tasked with determining management goals and targets by providing an analytical tool to help achieve those goals (Niemeijer and de Groot 2008; Gregory et al. 2012). Conversely, as determining leverage nodes is fully dependent on the network structure and risk data used, elicitation techniques and structured decision processes can feed into leverage node analysis to ensure that the best information and well-informed expertise guides the network analysis. In determining what nodes to target for management interventions and mitigation, however, impact network analysis cannot provide recommendations on what management interventions and mitigation measures to apply. A well-developed research agenda on the efficacy of mitigation actions can help fill this gap. 4.5 Conclusion Effective management of cumulative anthropogenic impacts on the environment will benefit from prioritization of management actions in the context of growing human pressure on the planet. Understanding cumulative impacts as a network of mechanistic pathways and analyzing the most prominent leverage nodes provides an effective, efficient approach to prioritize management actions. Management plans with an ecosystem-based framework, that attempt to manage cumulative impacts, can be ineffective if management actions are misallocated (Pikitch 107  et al. 2004; Game et al. 2013). Instead of assuming leverage nodes, we advocate for an analysis of revealed leverage nodes in a target-driven management plan. Our research provides a practical basis to aid emerging and ongoing environmental management.    108  Chapter 5: Group Elicitations Yield More Consistent, Yet More Uncertain Experts 5.1 Introduction Policy and management concerns on issues as diverse as health, technology, and the environment require scientific input to understand the mechanisms and consequences of emerging risks (Aspinall 2010; Morgan 2014). A rapidly changing world has spurred research to assess the severity and uncertainty of novel diseases and technologies, such as the risk posed by disease incursions and nanoparticles on human health, and climate change on ecosystems (Morgan et al. 2001; Kandlikar et al. 2007; Gale et al. 2010). Emerging risks are often associated with sparse data to inform policy. A lack of data can lead to a reliance on expert elicitation by default. Elicitation of expert judgement can yield useful and prompt information in the face of novel, data-poor problems that require expeditious management decisions (Burgman et al. 2011b; Morgan 2014). Expert judgements, if elicited carefully, carry the weight of experience, potentially providing policymakers with a rich understanding of estimates, uncertainties, and tradeoffs of variables of interest (Burgman 2005; Kandlikar et al. 2007; Burgman et al. 2011a; Morgan 2014). Cumulative anthropogenic impacts to the environment are one type of emerging and increasingly serious issue for policy, often requiring management when data is sparse, requiring reliable expert elicitation processes to generate useful information. In order to increase the quality of information derived from experts, researchers relying on expert judgement require knowledge about effective elicitation strategies (Burgman 2005; McBride and Burgman 2012; Morgan 2014). Experts are, of course, human, and as such they are also 109  susceptible to cognitive biases and unreliable mental shortcuts as non-experts (Tversky and Kahneman 1974; Fischhoff et al. 1982; Gilovich et al. 2002; Slovic 2011). Thankfully, research has established effective techniques to generate reliable quantitative answers to management-relevant science questions. For example, to address overconfidence and other cognitive biases among experts, group elicitation procedures are often recommended. In groups, experts can challenge and corroborate each other’s ideas while clarifying vague terminology, and they are often forced to reevaluate the confidence placed in their individual elicited quantitative estimates (Kandlikar et al. 2005; Burgman et al. 2011b; Sutherland and Burgman 2015). However, we do not know how techniques designed to question confidence then affect the judgement of experts when they are asked to parameterize a priori undetermined variables in an underspecific context.  This problem is thus: Humans are imperfect intuitive statisticians, relying on cognitive heuristics to make judgements when confronting uncertainty (Tversky and Kahneman 1974; Gilovich et al. 2002; Kynn 2007). Instead of calculating the probability of events, people often make judgements based on the ease by which instances are recalled (the availability heuristic), and provide quantitative estimates influenced by previous suggestions (anchoring bias). Experts are also often recognized as privileged holders of information and can display overconfidence in their judgements, reporting confidence intervals for estimates that do not encompass the “true” answer (Speirs‐Bridge et al. 2009; Morgan 2014). Group elicitation approaches (often modified from Delphi group approaches) can mitigate these biases, and specific approaches can even mitigate the influence of dominant voices drowning out less assertive ones (Burgman et al. 2011b; McBride and Burgman 2012). Group settings allow experts to challenge each other and are more likely to induce an individual expert to rationally reassess their judgements (Kandlikar 110  et al. 2005). Diverse groups of experts are also more likely to challenge in-group thinking that can come from similar disciplines of experts reinforcing each other’s judgements (Fish et al. 2009; Sunstein and Hastie 2015). However, group decisions can easily lead to “groupthink”, where groups tend towards uniformity and censorship, particularly when groups are made of like-minded individuals that crowd out dissenting voices (Sunstein and Hastie 2015).  Groups can be effective, therefore, when they effectively challenge the confidence any expert has in their initial judgment while ensuring diverse participants that mitigates groupthink. This group-discussion effect has been tested and understood under contexts where experts are asked deliberate questions to parameterize their answers – where the context is fully specified for them to enable parameterization (Burgman 2005). There are many cases in environmental management where there is insufficient specificity of context that expert input is valuable in first specifying this context and then parameterizing variables given this context (Fischhoff and Morgan 2013). For example, when managers are interested in effectively combating the causes of cumulative impacts, direction regarding which risks to prioritize for management is rarely clear a priori (MacDonald 2000). When faced with a two-part problem of trying to sort and prioritize the most important risk, and understand the uncertainty in quantitatively understanding the level of impact those important risks generate, the problem of how members of the group affect each other might be louder still. How does group elicitation affect expert responses? When experts are asked to first prioritize important risks in addition to estimate parameters, the difficulty of the elicitation is increased, which increases the probability that experts will desire conformity and commit groupthink (Hart 1991; Baron 2005). If group processes lead to unreliable judgements when the elicitation context is difficult, elicitation processes using 111  individuals may be more useful for understanding highly uncertain problems such as cumulative impacts. Specifically, we ask: 1) Is the group-level aggregation of best impact estimates more variable after a group process than before?; and 2) Do individual experts express greater subjective uncertainty in impacts after a group deliberation process than before? If the experts suffer the consequences of groupthink, we might expect that expert responses are more certain both in individual and group estimates because groupthink minimizes conflict and decreases critical evaluation in experts while increasing confidence (Baron 2005). However, if group deliberations significantly shake the confidence of expert judgements, we may see greater variation among and subjective uncertainty within expert responses. There is also the possibility that expert responses show signs of group polarization, whereby opinions among experts become more extreme and confident as like-minded subgroups reinforce each other and strongly disagree with other subgroups (Yardi and Boyd 2010). This mechanism would result in greater variability in group-level expert responses, and increases subjective certainty in individual estimates. Finally, ostensible initial disagreements among experts may be the result of misunderstanding and variation in terminology, which critical group elicitation can clarify while still challenging individual experts (Kandlikar et al. 2005; Carey and Burgman 2008) . This mechanism can lead to a context where group-level expert responses are less variable and individual responses are more uncertain.  This study seeks to address these questions, given a case where experts are asked specify the most important risks, then score the impact from them, in the context of eliciting judgments about cumulative impacts on Tasman and Golden Bays in New Zealand. Specifically, we address these questions through an expert elicitation study designed to understand prominent risks to 112  ecosystem services in coastal New Zealand. In this case, experts were asked to provide individual responses before and after a workshop, to understand how group deliberation affected expert consistency/variation and subjective uncertainty. 5.2 Methods Our case study in Golden Bay and Tasman Bay (New Zealand) included four major methodological steps. First, we identified local experts and solicited their expertise about major risks to four ecosystem services via an online survey platform. Second, we interviewed all of these experts individually, wherein we asked additional questions about mechanisms and magnitude of impacts. Third, we held a Delphi-like group workshop involving most interviewed experts, which involved a group deliberation and re-scoring of impacts. Our design was focused on training and guiding experts through the problem context and data collection exercises. The opportunity to enable expert learning/training for tasks is a major advantage of expert elicitation (McBride and Burgman 2012; Morgan 2014). Fourth, we analyzed the magnitude and uncertainty of impacts within and across experts. In the five subsections below, we describe the study site and these four major steps. 5.2.1 Study Site Tasman and Golden Bays are situated at the northern end of the South Island of New Zealand. We focus on impacts to fisheries, shellfish aquaculture, marine recreation, and existence values of biodiversity because these are all important uses/benefits in the bays. Tasman and Golden Bays have a history of human alteration. Trawl fisheries have historically contributed to the transformation of the benthic habitat in these areas from a structurally complex 113  environment with thick mussel beds and oyster reefs to a flat silty bottom (Handley 2006). Increased land use in the form of agriculture, forestry, and urban centres in the latter half of the nineteenth century have also affected the bays (Handley 2006).  5.2.2 Online Survey We recruited experts through recommendation by a leading research organization in the area (the Cawthron Institute), considering their acknowledged expertise. Competence was gauged based on years worked and studied within the bays (most had over a decade of experience), and recognition among the expert community through snowball sampling (Ban et al. 2014b). Experts were people who worked and recreated in, studied, or advocated for the area, which was important given the site specificity of both the ecosystem services and the anthropogenic impacts that occur there (Burgman et al. 2011a; Martin et al. 2012; McBride and Burgman 2012). In total 42 experts were contacted and 20 took part in the elicitation (47.6% recruitment rate). Choosing experts with area-specific knowledge is important, as the opinions of experts without such knowledge can misrepresent the specific characteristics of the two bays (Burgman 2005; Murray et al. 2009). Local experts are still prone to cognitive biases, but structured elicitation processes (such as cross-examination by facilitators and other experts) can effectively circumvent or ameliorate these biases (Burgman et al. 2011a).  Overall 20 experts took part in the study, including five fisheries experts, three aquaculture experts, four recreation experts, and seven biodiversity experts (emphasizing that we were interested in the existence value of biodiversity). First, each expert was asked to complete an online survey consisting of one question asking them to provide a ranked list of up to five risks they regarded as threatening a given ecosystem service. In order to do so, they were given a list of risks to choose from, with the option of adding others 114  not indicated (Table 5.1). The pre-defined list consisted of ten broad activities and stressors, adapted from the global and regional cumulative impact assessments (Halpern et al. 2008, Halpern et al. 2009), emphasizing broad categories. Five experts identified eight additional activities and stressors not on our initial list.   Table 5.1 The list of ten activities and stressors initially provided for experts to rank, with the additional eight suggested by experts. List of Activities and Stressors Initial List Agriculture Pollution Coastal Structures Commercial Shipping Invasive Species Aquaculture Recreational Fishing Commercial Fishing Climate Change Human Trampling Additional List Disease Ocean Acidification Sedimentation Social Licence Nutrient Input Forestry Land Clearing Poor Regional Planning  5.2.3 Individual Interviews All experts who took part in the online survey were then interviewed (step 2). Experts were first asked to confirm their ranking, then asked why they ranked the way they did. As the interest was in impacts to ecosystem services, experts were reminded that humans can cause impact through 115  biophysical as well as socioeconomic means. Emphasis was placed on understanding the processes by which risks cause impact, and what aspects of risk they emphasized in their ranking (such as the scale of impact, if the ecosystem service was particularly vulnerable to it, etc.). Experts were then asked to provide estimates of impact of their chosen risks to the bays under current levels of activity for the immediate future (0-5 years). Experts were asked to provide a range of values they believe captures the level of impact to their specific ecosystem service, between scores of 0-1. A score of 0 indicates that the ecosystem service is unimpacted, while a score of 1 indicates that the ecosystem service is rendered unavailable for human enjoyment. They were also asked to provide a best estimate of impact. After completing these steps, experts were asked to indicate how confident they were in their judgement (from 0-100%), and asked to make sure they were at least 50% confident. This step forces experts to reevaluate the range they provided, and often forces them to expand their range (Speirs‐Bridge et al. 2009). Then experts were introduced to probability distribution functions (PDFs), and asked to draw the function, with the ends of the interval indicating the extent of the function, and the best estimate indicating the mode. We had a facilitator train and assist experts in drawing PDFs, and provide feedback, to ensure that expert responses are reliable (Burgman 2005; O'Hagan et al. 2006). 5.2.4 Group Workshop Finally, we convened an expert workshop adapted from Burgman (2011b), Fischhoff (2013), Fish (2009), and Speirs-Bridge (2009) designed to reduce prominent biases in expert elicitation, including availability, anchoring and adjustment, dominance, and overconfidence. Fifteen experts attended a workshop following the interviews (6 biodiversity experts, 3 aquaculture experts, 3 fisheries experts, and 3 marine recreation experts). All experts save one were 116  previously interviewed; one fisheries expert was asked to fill in on behalf of another expert who could not make it to the workshop but was briefed on the activities and background. This expert’s data was not analyzed for this study because we did not have data for them before the workshop. The workshop was structured into four sessions, with each session dedicated to one ecosystem service. Each session was based on a Delphi method approach with facilitators prompting experts to defend their views to each other about the most important risks. All ecosystem service experts were invited to partake in each discussion; this was designed to reduce disciplinary bias and expand the scope of the problem being considered (Fish et al. 2009). At the end of each session, experts were asked to again rank up to five risks and provide probability distribution functions for their level of impact and indicate their confidence in the intervals, providing intervals they were at least 50% confident in.  5.2.5 Analysis Expert consistency was analyzed in two ways. First, importance of risks within each expert domain was modeled to predict what the experts as a group would rank using the Insertion Sorting Rank (ISR) algorithm for analyzing ranking data (Biernacki and Jacques 2013; Grimonprez and Jacques 2014). Associated with the model is a measure of expert homogeneity from 0.5 to 1, where 1 indicates total certainty that the model captures expert ranking as a group and 0.5 indicating complete uncertainty in ranking among experts captured by the model. This is a predictive analysis based on observed expert ranks and not a comparative test among ranking, and it is not subject to sample-size effects. However, we use this analysis to compare homogeneity of experts. Small differences in the homogeneity score (~0.03) can represent differences in expert homogeneity, and we use the standard deviation of the distance between the 117  final estimate of the homogeneity score and the score at each step in the algorithm as a measure of score variation (Jacques et al. 2014). To further assess expert consistency (or variability), we also analyzed the best estimates (modes of the elicited expert distribution) of impact from all experts before and after the workshop, and compared the consistency in scores. To do this, we collated the best estimates for each specific type of risk for each ecosystem service, and calculated the difference of individual scores from the mean score to generate a measure of difference among experts within each ecosystem service. We compared differences in best estimates by subtracting the difference scores post-workshop from the difference scores pre-workshop. According to the impact scale experts used to judge activities and stressors, differences from the mean on the order of 0.1 represent a view that is 10% different from the mean expert judgement. Because our samples before and after the workshop were not independent and were shared with experts, we conducted a mixed-effect paired t-test of these differences for estimates before the workshop and after (Baayen et al. 2008). The paired before-after risk category nested within experts were treated as random effects, and the difference score treated as a fixed effect. Because 14 experts were shared between the interview and workshop stages, data analysis was restricted to these experts, and only on risks that they repeatedly scored before and after the workshop. Some risks were named differently before and after the workshop by experts, but based on the reasoning given in the interviews (and/or confirmed with specific experts following the workshop), some risks were combined. For example, many experts in interviews scored impacts from agriculture because of its contribution to sedimentation, and after the workshop these experts scored impacts from sedimentation. In these cases, we included comparisons between agriculture and sedimentation. 118  Sample sizes for these analyses are 14 comparisons for aquaculture, 5 comparisons for fisheries, 9 comparisons for marine recreation, and 19 comparisons for biodiversity. Subjective uncertainty in individual expert judgements was analyzed by comparing the intervals of impact for a given risk on a given ecosystem service by a given expert. Similar to the analysis on best estimates, we only include results from the 14 experts shared between the interview and workshop stages. We compared the range of intervals before and after the workshop by subtracting the range of post-workshop estimates from the range of pre-workshop estimates and using a mixed-effect paired t-test. We treated the before-after pair nested within an expert as a random effect and interval range as a fixed effect. In contrast to analyzing best estimates which compared individual estimates against a group mean, we could include ranges for activities and stressors that were not considered by multiple experts, meaning that our samples sizes were larger for analyses on ranges than on best estimates. Sample sizes for analyses on ranges are 14 comparisons for aquaculture, 7 comparisons for fisheries, 11 comparisons for marine recreation, and 20 comparisons for biodiversity. Mixed-effect paired t-tests were conducted using the nlme package in R (Pinheiro et al. 2007) and the rank modeling was conducting using the R package rankcluster (Jacques et al. 2014). 5.3 Results 5.3.1 Among Expert Consistency 5.3.1.1 Ordinal Consistency Following the expert workshop, experts displayed a more homogenous rank of risks across all four ecosystem services compared to when experts were interviewed alone (Figure 5.1). The 119  largest changes in consistency were found in marine recreation experts (homogeneity score change from 0.91 to 0.96) followed by fisheries experts (0.93 to 0.97), followed by biodiversity experts (0.91 to 0.94), and finally aquaculture experts (0.93 to 0.95). Variation in these estimates were considerably smaller (ranging from 6.0×10-4-2.0×10-3).  Figure 5.1 Homogeneity of expert rankings. Bars represent homogeneity scores before (grey) and after (red) the group workshop. After the workshop, experts were more likely to rank specific stressors highly, such as sedimentation and pollution, whereas before the workshop most experts pointed to activities that contribute to these stressors, providing a diverse list of activities such as agriculture, forestry, 120  and land clearing. One expert during interviews highlighted sedimentation as an important risk not included on the pre-set list, indicating that many activities are primarily a risk through sedimentation. Indeed, many experts who raises concerns of agriculture, forestry, land clearing, and even dredge-based fisheries suggested sedimentation as the primary reason they ranked these risks highly, with some suggesting runoff-based pollution and nutrient input a major consideration for the inclusion of land based activities. Through facilitated discussion, terminology was agreed on and experts were able to clarify their meaning and focus their ranking on more specific stressors (Table 5.2). In some cases, experts included entirely different risks in their ranks after deliberation with experts in other fields. For example, in interviews marine recreation experts did not consider land-based drivers important (compared to their other top five activities and stressors), but when confronted with expertise in fisheries and biodiversity (who ranked sedimentation as the chief concern), recreation experts included sedimentation as one of the top 5 risks to marine recreation. 5.3.1.2 Numerical Consistency After workshops, experts were more likely to provide more consistent best-estimate impact scores than when interviewed individually (Table 5.3). This finding was found among all four ecosystem services, and always found to be statistically significant. The greatest difference in consistency between the interviews and workshop was for biodiversity (the mean±SE difference from the mean of pre-workshop scores was 0.076±0.027 greater than post workshop, p=0.012), followed by marine recreation (0.072±0.024, p=0.018), fisheries (0.06±0.02, p=0.041), and finally aquaculture (0.044±0.014, p= 0.009).  In very few cases did experts provide best estimates more in line with average estimates before the workshop. In two out of fourteen cases 121  aquaculture experts provided best estimates post-workshop that were more different from mean estimates than pre-workshop. For biodiversity experts this situation occurred in four out of nineteen cases, and for marine recreation experts this situation occurred in one out of nine cases. 5.3.2 Within-Expert Subjective Uncertainty The mean interval that experts believe captures the true impact score of specific risks was larger post-workshop than in individual interviews, for all four ecosystem services (Table 5.3). This finding was always found to be statistically significant. The size of the interval increased the most for biodiversity experts (the mean±SE of post-workshop intervals was 0.159±0.063 greater than pre-workshop intervals, p= 0.021), followed by fisheries experts (0.121±0.049, p= 0.047), marine recreation experts (0.118±0.051, p= 0.043), and finally aquaculture experts (0.091±0.042,p=  0.047).  In few cases did experts provide smaller ranges post-workshop than pre-workshop (Figure 5.2). In three out of fourteen cases aquaculture experts provided smaller ranges post-workshop than pre-workshop. For biodiversity experts this situation occurred in two out of twenty cases, and for marine recreation experts, this situation occurred in two out of eleven cases.   122  Table 5.2 Modeled rank of prominent risks to ecosystem services in Tasman and Golden Bays, New Zealand. The degree of expert homogeneity in ranking is provided under each ecosystem service and time (at interview stage or after workshop), and risks are ranked from greatest to lowest risk assessed. There is a different number of ranks among ecosystem services because there were a different number of risks assessed among the times for each ecosystem service.   Aquaculture Fisheries Marine Recreation Biodiversity   Interview Workshop Interview Workshop Interview Workshop Interview Workshop Homogeneity 0.93 0.95 0.93 0.97 0.91 0.96 0.91 0.94 1 Pollution Invasive Species Forestry Sedimentation Commercial Fishing Pollution Commercial Fishing Sedimentation 2 Agriculture Climate Change Agriculture Climate Change Pollution Human Trampling Coastal Structures Coastal Structures 3 Invasive Species Pollution Aquaculture Aquaculture Human Trampling Shipping Pollution Pollution 4 Climate Change Nutrient Input Commercial Fishing Commercial Fishing Recreational Fishing Sedimentation Invasive Species Invasive Species 5 Disease Social License Pollution Pollution Invasive Species Climate Change Climate Change Commercial Fishing 6 Commercial Fishing Sedimentation Coastal Structures Recreational Fishing Aquaculture Social License Agriculture Social License 7 Aquaculture Aquaculture Climate Change Social License Nutrient Input Nutrient Input Forestry Human Trampling 8 Human Trampling Disease Invasive Species Agriculture Shipping Commercial Fishing Aquaculture Climate Change 9 Ocean Acidification Shipping Recreational Fishing Human Trampling Climate Change Recreational Fishing Human Trampling Forestry 10 Recreational Fishing Recreational Fishing Sedimentation Forestry Social License Aquaculture Recreational Fishing Shipping 11 Sedimentation Human Trampling Social License Shipping Sedimentation Invasive Species Shipping Nutrient Input 12 Social License Commercial Fishing Shipping Invasive Species Coastal Structures Coastal Structures Land Clearing Poor Regional Planning 13 Shipping Coastal Structures Human Trampling Nutrient Input   Sedimentation Land Clearing 14 Coastal Structures Ocean Acidification Nutrient Input Coastal Structures   Nutrient Input Agriculture 15 Nutrient Input Agricultures     Social License Recreational Fishing 16             Poor Regional Planning Aquaculture 123  Table 5.3 Summaries of mixed-effect paired t-tests comparing the consistency in best estimate and interval for risks to the four ecosystem services before and after the workshop. The column “experts” refers to the number of experts in the analysis, while “comparisons” refers to the number of before-after comparisons included in the analysis (these were treated as nested random effects in the models). The column “difference” records the estimate of the difference score and the interval length of expert responses before the workshop compared to after the workshop. The column “df” corresponds to the degrees of freedom in the analysis. Positive “best estimate” scores indicate that the individual best estimates were more similar to the average after the workshop (than before); positive “interval” scores indicate that the range of estimates shrunk after the workshop. Ecosystem Service Test Experts Comparisons Difference Standard Error df t-value P-value Aquaculture Best Estimate 3 14 0.044 0.014 13 3.095 0.009   Interval 3 14 -0.091 0.042 13 -2.194 0.047 Fisheries Best Estimate 2 5 0.060 0.020 4 2.979 0.041   Interval 2 7 -0.121 0.049 6 -2.497 0.047 Marine Recreation Best Estimate 3 9 0.072 0.024 8 2.978 0.018   Interval 3 11 -0.118 0.051 10 -2.316 0.043 Biodiversity Best Estimate 5 19 0.076 0.027 18 2.782 0.012   Interval 6 20 -0.159 0.063 19 -2.524 0.021 124     Figure 5.2 Probability distribution functions (PDFs) representing expert-derived estimates of impact from interviews (grey), and following the group workshop (red). Following the workshop, experts were more likely to provide wider intervals estimating impact from specific risks to ecosystem services compared to when interviewed in isolation. Each row represents the paired estimates for a single expert (1-14). Experts are grouped by the ecosystem service of their expertise.  125  5.4 Discussion Our findings support the well-documented effect of group settings to reduce expert over-confidence and non-overlapping estimates among experts (Speirs‐Bridge et al. 2009; Burgman et al. 2011b; Morgan 2014). Moreover, our findings support the hypothesis that group elicitations can clarify misunderstandings and linguistic uncertainty, challenge extreme estimates, and assuage expert overconfidence (Carey and Burgman 2008; Speirs‐Bridge et al. 2009). Despite the increased risk of unfavourable group dynamics (such as groupthink and group polarization) in difficult elicitation settings, our results suggest that these group dynamics did not dictate expert judgements. Structured, facilitated discussions may help avoid unfavourable outcomes. 5.4.1 Reducing Variability Among Experts Some authors have proposed group processes as a mechanism to make expert responses more consistent, but such processes have sometimes been shown to yield consistency at the expense of accuracy via a phenomenon with the Orwellian title “groupthink” (Sunstein and Hastie 2015). Groupthink often occurs when conformity within the group is encouraged through suppressing dissent and when members avoid conflict. Dominant personalities can also control the discussion and direct responses towards their point of view, introducing bias, as can expertise dominated by a particular field (Fish et al. 2009; Burgman et al. 2011b). However, groupthink (as well as group polarization) also leads to increased confidence in judgements because of corroboration from other group members (Sunstein and Hastie 2015). We did not find this symptom of groupthink in our results. By including pre-workshop individual responses as anchors, encouraging debate and defense of opinion among experts from diverse professions and fields, and forcing experts to 126  provide post-workshop scores individually, we have shown that a healthy level of variability among experts can be retained after the group process, though this variation is smaller than when expert responses are elicited without group deliberation.  The design of our workshop was specifically tailored to counteract groupthink. First, allowing experts to provide their estimates individually reduces the chance for pressure from dominant voices to sway their reporting (Burgman 2005). Second, effective facilitators can skillfully reduce the time taken up by any individual voice, and allow others to assess claims and judgements (Aspinall 2010). Third, giving experts an opportunity to listen to each other, assess and cross examine judgements in a structured facilitated setting reduces the effect of individual dominance and improves average performance (Kandlikar et al. 2005; Burgman et al. 2011b). Exposing subject experts to diverse yet complementary expertise provides a more holistic consideration of the situation, and reduces discipline dominance (Fish et al. 2009). For example, we found that marine recreation experts reassessed their ranking to include climate change as a top five concern when exposed to fisheries, aquaculture, and biodiversity experts. Marine recreation in Tasman and Golden Bays is dependent on marine organisms, which was a specialty of the experts of the other three subjects.  The potential list of human impacts on coastal environments is immense and diverse, without an established nomenclature, and made up of bundled activities (such as agriculture, fisheries, etc.) and specific stressors (such as direct capture, ship strikes, sediment runoff, etc.). Experts tasked with identifying the more important risks must wade through this list, and individual experts may have different ways of describing and identifying risks, even when describing the same phenomena (Regan et al. 2002). To experts more accustomed to thinking about landscape level 127  issues, broader activities may be more salient and immediately available for recall, whereas experts accustomed to investigating direct harm to wildlife may be more comfortable considering specific stressors. Navigating the resulting linguistic uncertainty among experts (where uncertainty is embedded in the terms used) can be mitigated through experts having the opportunity to clarify meaning and understanding (McBride and Burgman 2012). We found that in multiple cases, different experts identified diffuse activities as prominent risks because of their contribution to a particular stressor before workshops, and when they were involved in facilitated discussion, were able to agree on consistent terminology by pointing to the dominant stressor. For example, individual experts pointed to agriculture, forestry, general land clearing (and other land uses), as well as dredging activities as important risks because of their role in sedimentation and estimated impact from these relatively vague risks, but workshops allowed experts to agree on consistent, specific terminology, highlighting the importance of sedimentation itself as a major risk across all ecosystem services. Our finding of increased numerical consistency among experts suggests that the workshop was effective beyond clarifying language to reveal latent agreement. Rather, the workshops helped experts reassess their individual judgements about specific risks. More extreme views about the severity (or lack of severity) of impact from risks were moderated when confronted with challenges from other experts (Kandlikar et al. 2005). Though pressure from dominant voices might also have this effect, asking experts to provide their estimates individually likely mitigated this possibility (Burgman 2005). The universality of our finding of increased numerical consistency among experts after a workshop will likely depend on the makeup of experts involved, including whether experts with mutual pre-existing animosity are included, and 128  whether contrasting advocates are involved in the workshop (Burgman 2005). Any factor that increases the chance that the presentation of evidence or contrasting judgement serves to buttress pre-existing views could counter the effect of converging judgements among experts, and contribute to polarization among group members (Yardi and Boyd 2010). The design of expert elicitations is important to reduce the probability that biases regulate expert responses (McBride and Burgman 2012; Morgan 2014). Well-designed workshops are an important component of effective expert judgement elicitation (Burgman et al. 2011b; Morgan 2014).  5.4.2 Increasing Subjective Uncertainty Within Experts One of the most prevalent problems with eliciting expert advice is contending with over-confidence (Speirs‐Bridge et al. 2009; Morgan 2014). Experts are highly regarded individuals with high levels of training, qualifications, and experience. Society, and experts themselves, expect heightened performance as a result (Burgman et al. 2011b). Status and expectations can lead to experts displaying excessive certainty in the accuracy of their beliefs. In expert elicitations, overconfidence is often found when experts who provide estimate intervals reflecting a high degree of subjective confidence do not capture the true value they are estimating (Speirs‐Bridge et al. 2009). Given the prevalence of this form of overconfidence, methods to coax experts to provide larger intervals for a given level of confidence are useful.  Some specific interventions applied to individual experts, such as asking experts to provide confidence intervals that meet a confidence cutoff, which we applied here, can force experts to reassess their intervals and provide larger intervals (Speirs‐Bridge et al. 2009). Similarly, experts exposed to each other in workshop settings as we did here are known to force a reevaluation of 129  estimate intervals, increasing the estimate range (Burgman et al. 2011b; Sutherland and Burgman 2015). This study utilizes interventions at both individual and group level, and because data was collected before and after the workshop, provides evidence that the effects of group workshops to reduce overconfidence occur in addition to the effects of interventions at the individual level.  5.5 Conclusion Expert elicitation is a valuable tool in situations where data is sparse and decisions urgent, and have the promise to be more valuable to tackle problems where the context is not well known beforehand. Where experts are needed both to specify context and provide estimates, evidence-based designs are needed to ensure that high quality expert judgements are captured. We show that in understanding cumulative impacts on ecosystem services, where prominent risks are not known beforehand, group deliberation in an expert workshop can both increase consistency in experts’ identification of risks, and decrease overconfidence in estimates of impact. We argue that given their favourable effects on expert uncertainty, expert workshops are an important tool to deal with highly uncertain problems.   130  Chapter 6: Mechanisms and Risk of Cumulative Impacts to Ecosystem Services: An Expert Elicitation Approach 6.1 Introduction Many human uses of coastal ecosystems degrade and convert coastal ecosystems through infrastructure development, resource extraction, tourism, and other human activities (Halpern et al. 2008, Doney 2010). People harvest and grow food from, recreate in, and transport goods through coastal systems. Coastal ecosystems are some of the most populated ecosystems on Earth, with half of the global population and three quarters of major cities within 60 km of a coastline (Kennish 2002; UNEP 2007; UNEP 2012a). Future projections of human population and movement trends suggest that these figures will continue to rise, increasing the pressure on coastal environments. Coastal systems are being converted to "use" environments and are projected to experience greater stress (Kennish 2002) as some 60% of the world’s population (~6 billion) will live in coastal areas by 2025 (UNEP 2007). Given the current and future context, there is a need to understand how human activities and stressors impact ecosystems and the services they provide.  The concept of ecosystem services has been used to understand human uses and values associated with many environments (Carpenter et al. 2009; Chan and Ruckelshaus 2010b; Kareiva et al. 2011; Queiroz et al. 2015). Most ecosystem service research to date, through mapping and valuation, emphasizes the benefits humans derive from the natural environment (De Groot et al. 2002; Chan et al. 2006b; Costanza et al. 2014). The analysis of ecosystem services includes three steps: the supply (production of services by the biophysical environment), 131  the service (the service actually used by people), and the value (the preferences for different services, Tallis et al. 2012). All of these may be affected by human activities. Thus, understanding risk to ecosystem services as the relationship between people and the environment (including human influence on the environment, as well as the environment’s influence on people) is key to a balanced treatment of management concerns (Raymond et al. 2013a). Effective management would explicitly address how and why valued aspects of the environment are at risk, including ecosystem services at risk from cumulative impacts of interacting local and global stressors (Allan et al. 2013). Researchers have advanced various ways of studying anthropogenic effects on coastal systems, but few address ecosystem services specifically. Some researchers have begun mapping the distribution of biological communities and human impact in response to human pressures on coastal environments (Rodrigues et al. 2004). Other methods attempt to understand the process by which human activities contribute to ecosystem impact (i.e. the DPSIR approach, Curtin and Prellezo 2010). Recent literature has called for research linking anthropogenic activities and stressors with their impact on ecosystem state (Brown et al. 2014). These various threads of research on environmental impacts have, however, rarely linked process (mechanism) with the size of impact, instead focusing on either process or impact. The current and changing nature of coastal systems demands not only planning for conservation set-asides, but a consideration of human use alongside ecosystem processes.  To prioritize management actions based on delivery of ecosystem services, tools are needed to identify how human drivers impact ecosystem services (Cook et al. 2014b). One response to this need is to focus on the drivers of change (human activities and global forces that instigate 132  impacts), and stressors (the mechanisms by which drivers cause impact) that have the greatest impact on management goals. Conversely, failing to understand the severity of cumulative impact, as well as the mechanistic pathway of impact, can lead to misallocated management efforts (Brown et al. 2013; Cook et al. 2014b). These considerations underline the importance of linking pathways of effects (the processes by which human and large scale environmental processes contribute to impacts) to the most prominent impacts.  Research on cumulative impacts faces the ever-present problem of data paucity, and this is especially true for impacts to ecosystem services. In many cases data are non-existent. In such cases the best (often only) option is to rely on expert elicitation (Burgman 2005; Altman et al. 2010). Expert judgment has benefits over traditional data collection. First, expert data is often time-integrated knowledge, where field collected data is often a snapshot, or is often prohibitively costly to acquire (Burgman et al. 2011a; Martin et al. 2012; McBride and Burgman 2012; Morgan 2014). Second, for measuring "impact" on ecosystem services, there often are no clear metrics that are easily measured in the field (because of lack of agreement on how to measure ecosystem services and uncertainty in how human activities pose risk), whereas an expert may still score impact by interpolating or extrapolating based on their experiences and acquired instinct (Burgman 2005; Teck et al. 2010; Sagoff 2011; Cook et al. 2014b). Experts can also assess tradeoffs and uncertainties (and provide a logical defense of their judgements) in a way that is not possible otherwise (Kandlikar et al. 2007). Relying on expert intuition may also introduce extra uncertainty through diverse linguistic and epistemic understandings of “impact” (Regan et al. 2002), but elicitation can be designed to reduce this uncertainty by forcing experts to defend their understandings to each other (Martin et al. 2012).  Expert judgment is a collection 133  of various uncertainties and mixed biases regarding what matters for impact, and multiple experts may not translate the problem in the same way, but a well-designed elicitation process can help alleviate some of these challenges to analysis (Martin et al. 2012; Morgan 2014). We use Tasman and Golden Bays, New Zealand as a case study to explore cumulative impacts on ecosystem services by using expert elicitation to identify which ecosystem services are at risk by what human activities in what ways. Specifically, we ask, 1) how severe are cumulative impacts on ecosystem services? 2) are threats to ecosystem services evenly distributed across activities and stressors, or do few dominate? 3) Do prominent activities mainly operate through direct stressors, or do they often exacerbate other impacts?  6.2 Methods In two New Zealand coast areas, we quantified the risks to the four ecosystem services roughly as follows. We assembled a team of experts for each service, and used a survey instrument to provide a ranked list of primary activities and stressors acting upon each service. We then interviewed each expert individually to derive impact scores and pathways for each designated activity or stressor, characterizing uncertainty parameters for each resulting in ‘impact profiles’. We then invited all experts to a group workshop (Delphi-style), in which experts exchanged views about their impact scores and pathways, and again provided impact profiles. We used these impact profiles and pathways to calculate cumulative effects (across relevant stressors), bounded within a meaningful range, and to generate networks of causal impact pathways describing the activities and stressors affecting ecosystem services mentioned by all experts. We provide more context about the study site and explain each methodological step below. 134  6.2.1 Tasman and Golden Bays Tasman and Golden Bays are situated at the northern end of the South Island of New Zealand (Figure 6.1). We focus on impacts to fisheries, shellfish aquaculture, marine recreation, and existence values of biodiversity because these are all primary uses/benefits in the bays. Tasman and Golden Bays have a history of human activities that have altered the physical environment. Trawl fisheries have historically contributed to the transformation of the benthic habitat in these areas from a complex 3-dimensional environment with thick mussel beds and oyster reefs to a flat silty bottom (Handley 2006). The terrestrial catchments of the bays have also witnessed a significant increase in urban, agricultural (sheep and beef, horticulture and dairy) and forestry activity since European settlement in the latter half of the nineteenth century (Handley 2006).  135   Figure 6.1 The location of Tasman and Golden Bays, at the north end of the South Island of New Zealand. Tasman BayGolden Bay0 25 50 75 10012.5Kilometers136  6.2.2 Expert Elicitation We recruited experts through recommendation by a leading research organization in the area (the Cawthron Institute) considering their high levels of competence regarding one of four stated ecosystem services in Tasman and Golden Bays. Competence was gauged based on years worked and studied within the bays (most had over a decade of experience), and recognition among the expert community through snowball sampling (Ban et al. 2014b). We selected people who worked and recreated in, studied, or advocated for the area, which was important given the site specificity of both the ecosystem services and the anthropogenic impacts that occur there (Burgman et al. 2011a; Martin et al. 2012; McBride and Burgman 2012). In total 42 experts were contacted and 20 took part in the elicitation (47.6% recruitment rate). Choosing experts with area-specific knowledge is important, as the opinions of experts without such knowledge can misrepresent the specific characteristics of the two bays (Burgman 2005; Murray et al. 2009). The elicitation process followed a three-step procedure focused on priming and training experts for specific tasks. The opportunity to enable expert learning/training for tasks is a major advantage of expert elicitation (McBride and Burgman 2012; Morgan 2014). First, experts were asked to fill out an online survey consisting of one question, where experts ranked the most important stressors to the ecosystem service in question (for which they had expertise), listing up to 5 stressors (e.g. biodiversity experts listed stressors for the existence value of biodiversity). Experts could choose from a list of pre-defined stressors or identify stressors not in the list. The pre-defined list consisted of ten broad activities and stressors, adapted from the global and regional cumulative impact assessments (Halpern et al. 2008, Halpern et al. 2009), emphasizing 137  broad categories. Five experts identified eight additional activities and stressors not on our initial list.   After survey results were collected, each expert who filled out a survey was interviewed independently.  Interviews occurred one week after the survey results were collected. The interviews were guided by the individual survey results, as experts were asked to explain why they ranked stressors as they did.  After describing the particular aspects of risk that justified any one expert’s inclusion, each expert was asked to draw the mechanistic pathway by which the stressor posed a threat to the ecosystem service as path diagrams on paper. We were primarily interested in capturing pathways as a three-step procedure, linking activities to stressors to ecosystem services (similar to Knights et al. 2013). For example, an aquaculture expert might relate climate change (as an impact driver) to shellfish aquaculture by suggesting that climate change impacts aquaculture through ocean acidification, increased temperatures, increasing storm intensity and frequency increasing mortality and increasing rates of sedimentation. In this example, we were primarily interested in linking climate change, climate change stressors (acidification, temperature, storm intensity, sedimentation), and shellfish aquaculture. Each expert did this for each of the five stressors on an individual ecosystem service. For the final part of the interview, experts were asked to characterize magnitude of impact.  To determine magnitude and uncertainty of impacts from specific stressors to specific ecosystem services, experts were asked to consider the condition of the bays in reference to human activity in the near future (0-5 years). Determining magnitude and uncertainty in interviews was intended to train individual experts for the final phase of the elicitation. For each stressor-ES combination, experts were asked to provide a relative impact score from 0 to 1, with 0 representing the 138  ecosystem service as unimpacted by the stressor and 1 rendering the ecosystem service unavailable for human benefit. Experts were also asked to estimate the lowest and highest plausible impact levels, and to provide a best estimate (Morgan 2014). They were then asked to provide an estimate (in percentage) of their confidence that the plausible range they provide captures the “true” value. We asked experts to provide a range they were at least 50% confident in. Requiring this level of confidence explicitly reduces overconfidence (i.e. providing too narrow an interval) by forcing experts to revisit and evaluate their stated interval (Speirs‐Bridge et al. 2009; Burgman et al. 2011b). Finally, respondents were asked to draw a probability distribution function (PDF) that described the probability of impact scores within this range (with best estimates being the most probable in the distribution). We had a facilitator train and assist experts in drawing PDFs, and provide feedback, to ensure that expert responses are reliable (Burgman 2005; O'Hagan et al. 2006). For the final phase of the elicitation process, all interviewed experts were invited to a workshop (n=20). In total 15 experts attended, with multiple experts representing each of the four ecosystem services present. Six biodiversity experts were present, as well as three fisheries experts, three aquaculture experts, and three marine recreation experts. The workshop was structured into four sessions, with each session dedicated to one ecosystem service. The sessions were round-table, Delphi-style discussions with experts, in which facilitators prompted experts to talk out loud about ranking stressors and choose impact scores (Burgman et al. 2011b). All experts were allowed to participate in each session even if the session did not match their thematic expertise. By allowing all experts to take part in all sessions, regardless of their specific expertise, we designed each session to incorporate multiple thematic and epistemological 139  backgrounds to help reduce bias and prevent narrow scope (Burgman 2005; Fish et al. 2009). A diversity of experts provides multiple perspectives and can broaden an otherwise narrow focus on any single dimension of the problem (e.g., fisheries experts have knowledge of fisheries practices that may challenge questionable assumptions other experts hold about fisheries impacts on biodiversity that might otherwise influence the data collected). Experts verified a list and definitions of activities and stressors derived from the survey and interviews that captured all important impacts. This list captured a common and agreed upon set of terms for experts, clarifying language and reducing linguistic uncertainty among experts (see Table D.1 for the list and definitions of activities and stressors experts assessed in the workshop). They were also primed with summaries of impact scores (arithmetic means of best estimate, lower estimate, and upper estimates) from the interviews. Facilitators encouraged experts to challenge each other and defend their positions, which is another method to reduce overconfidence and linguistic and epistemic uncertainty (Kandlikar et al. 2007). At the end of each session all experts, regardless of expertise, were asked to provide their updated ranking of the top stressors (up to five) on the ecosystem service, and rescore the impact (with PDFs, repeating the exercise in the interviews, described above). Experts were again asked to provide a range for which they were at least 50% confident that it contained the "true" value (Speirs‐Bridge et al. 2009). Because some activities/drivers of change can modulate the impact of some stressors indirectly (i.e. they are not independent), we asked experts to score impact from activities/drivers of change excluding the impacts from those specific stressors on interest, and asked that they consider the contribution of the amplifying effect on the stressor. For example, if climate change amplifies the impact of sedimentation, we asked that experts consider this impact when considering sedimentation, and not climate change. Separating impact from activities and stressors this way allowed us to limit 140  double counting of impact, and reminded experts to consider impact under near-future conditions (e.g. sedimentation impact scores should consider runoff rates under near-future climate change). We elicited PDFs from experts by asking them to draw the PDFs on paper with pencils. 6.2.3 Data Analysis Three types of analysis were performed on these data. First, a statistical model was created to predict the ranking of each stressor to each ES by the experts as a group. This analysis helped us determine how prevalent common activities and stressors were across ecosystem services or if a diverse suite affected diverse ecosystem services. This analysis was done using rank cluster analysis with the R package Rankcluster (Grimonprez and Jacques 2014), assuming a single expert cluster, as our focus was on how stressors were ranked across all experts and not on how different experts clustered based on ranking. For each model, a homogeneity score was calculated, giving the probability that the expert group ranked stressors similarly according to the predicted ranks (scored from 0.5 to 1, with higher values indicating more support for similar ranking within the cluster of experts, Grimonprez and Jacques 2014). The second analysis was to map the pathways by which stressors impact ecosystem services. Understanding prominent impact pathways allowed us to determine if activities impact ecosystem services directly or not. The impact pathways from the interviews were coded so that they resemble the following sequence   Human activity → Stressor → Impact on Ecosystem Service For example, where experts describe a situation in which agricultural runoff leads to sedimentation that smothers and kills benthic biota, the data was coded as  141    Agriculture → Sedimentation → Biodiversity Different experts used similar terminology, making coding relatively straightforward. Where there were variants, experts were asked if what they meant resembled terms used by other experts. Every pathway described by experts was coded for each ecosystem service, and all pathways from all experts were collated to create a network of pathways. For example, a given stressor (such as pollution) may be a product of multiple drivers of change (e.g., recreational activities and industry), just as a given driver of change may contribute to multiple stressors (such as commercial fishing contributing to bycatch and habitat destruction). Analogous to a weight of evidence approach used to highlight most likely explanations in contexts of multiple competing hypotheses, we used a “weight of expertise” approach to highlight pathways with most expert attention. The number of experts that described a specific pathway was recorded, and these data were also incorporated into the network to show how often a given link was mentioned by experts. This data was then used to describe how various stressors pose risks to the various ecosystem services. For the final analysis, the PDFs drawn by the experts were digitized using the program PlotDigitizer (Huwaldt 2014). The digitized data were combined to create density histograms of the impact scores for every driver and stressor on every ecosystem service across all experts. For each expert, cumulative scores were calculated (across all of their listed stressors) using a resampling procedure that selected impact scores from the individual PDFs to accumulate the scores while maintaining uncertainty (Clarke Murray et al. 2016). For our resampling procedure we used 10 000 iterations. 142  To maintain the impact scale and its associated meaning, we aggregated impact scores to respect that a cumulative score of 1 is the upper limit (i.e., 100% loss of service). To do this we calculated the extent to which each ES is not impacted by each stressor and multiplied these scores together to calculate how much of each ES is free from all impact. We then subtracted this product from 1 to generate the total impact score. Cumulative impact scores on a given ecosystem service, as estimated by individual experts, were calculated with the following equation  = 1 (1 ,) where CI is the cumulative impact score from an individual expert for a given ecosystem service j, and ES is the impact score that expert assigned to individual stressor i on ecosystem service j. For example, if stressor i1 has an impact of 0.3 and stressor i2 has an impact of 0.5, the cumulative impact would be = 1 – (1-0.3)*(1-0.5) = 1 – 0.7*0.5 = 0.65. This model allowed us to assess the severity (and uncertainty) of cumulative impacts on ecosystem services. This style of impact accumulation is markedly different from how cumulative impact scores are often treated in the literature, as they are often generated using a linear additive model with no upper bound of impact (Halpern et al. 2008b; Halpern et al. 2009; Ban et al. 2010; Allan et al. 2013; Halpern and Fujita 2013). Though Allan et al. (2013) have an upper bound on their impact score, the meaning of the upper score is not apparent (it is a standardization). Our model is also a generalization and assumes that the cumulative impact is less than the sum of component impacts, but allows us to accumulate scores while maintaining the meaning of our upper bound: 143  once a service is entirely lost, it cannot be further degraded. All data was processed and analyzed using the computer program R (R Core Team 2015). 6.3 Results 6.3.1 Ranking The overall top 5 drivers and stressors for each ecosystem service by rank are listed in Table 6.1. Each resulting ranking has a homogeneity score of 0.85 or above, confirming the treatment of all experts as being part of a homogenous cluster that ranks similarly. Biodiversity has a homogeneity score of 0.9, shellfish aquaculture has a homogeneity score of 0.88, fisheries 0.88 and marine recreation 0.85. The two drivers of change consistently ranked within the top five are climate change and commercial fishing, while the top two stressors are sedimentation and pollution. We outline the causal pathways of impact of these prominent activities and stressors across ecosystem services below. Table 6.1 Modelled ranks for the impact of 12 stressors on four ecosystem services, from highest impact to lowest. Rank Biodiversity Shellfish Aquaculture Fisheries Marine Recreation 1 Sedimentation Nutrient Input Sedimentation Pollution 2 Commercial Fishing Invasive Species Commercial Fishing Sedimentation 3 Pollution Social Licence Recreational Fishing Commercial Fishing 4 Invasive Species Climate Change Climate Change Shipping 5 Climate Change Pollution Pollution Human Trampling 6 Recreational Fishing Shipping Nutrient Input Aquaculture 7 Coastal Structures Sedimentation Invasive Species Climate Change 8 Social Licence Aquaculture Aquaculture Recreational Fishing 9 Aquaculture Coastal Structures Human Trampling Social Licence 10 Human Trampling Commercial Fishing Social Licence Nutrient Input 11 Nutrient Input Human Trampling Coastal Structures Invasive Species 12 Shipping Recreational Fishing Shipping Coastal Structures  144  6.3.2 Mechanisms of Cumulative Impact 6.3.2.1 Activities/Drivers of Change 6.3.2.1.1 Climate Change Experts indicated a variety of stressors driven by climate change (Figure 6.2A). Some are direct consequences of climate change, such as acidification, warming, changes to ENSO, sea level rise, changes to currents, and increases in storm frequency (this last stressor is the one that was most often suggested by experts). However, many experts also linked climate change with regional stressors, most notably sedimentation and pollution (which are also some of the most important impacts considered by experts). Experts also point out numerous other impacts that climate change indirectly contribute to, such as changes to species distributions and community dynamics, invasive species and disease spread, effects to ENSO events, and a general reduction in ecosystem resilience that can affect biodiversity.  6.3.2.1.2 Commercial Fishing In contrast to climate change, many experts linked commercial fishing with numerous direct stressors, and fewer indirect impacts (Figure 6.2B). Notable direct impacts are overfishing and habitat destruction, with fewer experts mentioning impacts through bycatch and disruptions to the marine community through fish removal (noted as "community effects"). Much of the commercial fishing in this region is done by trawling, which has an impact on benthic habitats. A number of experts also pointed to commercial fishing also contributing to sedimentation, as the ships and benthic trawls can re-suspend sediment that has settled in the bays. Some experts also link commercial fishing to limiting access to other ES, such as other types of fishing, recreation, 145  and aquaculture. Other indirect risks of commercial fishing indicated by experts include the spread of invasive species, and impacts to fish recruitment by removing adults.  Figure 6.2 A network of pathways of impacts from climate change and commercial fishing to the four ecosystem service types. A) This hive plot shows the driver of change (climate change, at top), leading to various stressors (lower right axis) shown in orange edges, which then impact the four ecosystem service types (lower left axis) shown in blue edges. B) This hive plot shows the driver of change (commercial fishing, again at top), leading to various stressors (lower right axis) shown in orange edges, which then impact the four ecosystem service types (lower left axis) shown in blue edges. The thickness of each edge represents how many experts mentioned each link. The nodes along each axis are organized by ranking the nodes with the highest number of linked edges to the lowest (highest number of links on the outside). 6.3.2.2 Stressors 6.3.2.2.1 Sedimentation Experts identified prominent land-based, sea-based, and global-change drivers of sedimentation (Figure 6.3A). Agriculture and forestry were the most often cited sources of sediment, with few experts citing coastal structures and a few using the general terms "land use", and "terrestrial 146  runoff". Dredging and commercial fishing were the prominent marine contributors of sedimentation, with some experts listing aquaculture (as a more localized source), shipping and "vessels". Climate change is also considered to amplify sedimentation through increasing runoff rates.  6.3.2.2.2 Pollution In contrast to sedimentation, the causes of pollution were perceived to be numerous and diffuse (14 causes are listed, the most of any stress, Figure 6.3B). Diffuse attention was paid to the contributions of pollution, from industrial chemical runoff and human structures on the coast, pollution in runoff from agriculture and other land use, fuel from ships, and trash thrown off the side of ships, fishing vessels, and recreationalists on the coast. No source of pollution stands out as having been selected by a prominent number of experts. 6.3.3 Impacts on Ecosystem Services 6.3.3.1 Biodiversity Experts provided PDFs for ten different stresses on biodiversity, with experts consistently scoring mid-range to high levels of impact (Figure 6.4). The most prominent drivers of change that experts identified impacting biodiversity are commercial fishing and climate change. Many experts indicated high levels of impact from commercial fishing and relatively less impact from climate change. Biodiversity experts especially indicated high impact from commercial fishing. The most prominent stressors that impact biodiversity that experts indicated are sedimentation, pollution, and invasive species. Sedimentation has consistently high impact scores across experts, with other stresses providing mid-range levels of impact. Fewer experts provided PDFs 147  for recreational fishing and coastal structures, which have mid-range impact scores. A few stresses (trampling, social license, and nutrient input) are considered to have very high impact by a few experts. Social license here, as suggested by a few experts, indicates that growing social indifference to biodiversity decreases the value of the existence of biodiversity. This translates into few protections for biodiversity and/or less resistance to developments and policy that threaten biodiversity.   Figure 6.3 A network of pathways of impacts from various drivers through sedimentation and pollution to the four ecosystem service types. A) This hive plot shows the drivers of sedimentation (at top), leading to the stress (sedimentation, lower right axis) shown in orange edges, which then impact the four ecosystem service types (lower left axis) shown in blue edges. B) This hive plot shows the drivers of pollution (again at top), leading to the stress (pollution, lower right axis) shown in orange edges, which then impact the four ecosystem service types (lower left axis) shown in blue edges. The thickness of each edge represents how many experts mentioned each link. The nodes along each axis are organized by ranking the nodes with the highest number of linked edges to the lowest (highest number of links on the outside). 148   Figure 6.4 Elicited impact curves of various stressors to (existence value from) biodiversity from the 15 experts that attended the workshop. Violin plots show density histograms overtop of boxplots indicating median (white dot), first and third quartile (box) and max and minimum value (whiskers). Impact scores are on the y-axis, and individual expert number on the x-axis. Violins coloured green represent biodiversity experts, while yellow represents fisheries experts, grey represents aquaculture experts, and blue represents marine recreation experts. 149  6.3.3.2 Shellfish Aquaculture Experts provided PDFs for ten different stresses on shellfish aquaculture, with most experts scoring mid-range impacts (Figure 6.5). The most prominent driver of change identified was climate change, and prominent stressors were invasive species, social license, pollution, and nutrient input. Social license here refers to the permitting restrictions to continuing and expanding aquaculture due to public wariness around the practice. Some experts also indicated impacts from aquaculture itself limiting aquaculture due to space competition, and aquaculture practices related to crowding and limited disease control enhancing impacts of invasive species and disease. Fewer experts provided scores for shipping, coastal structures and commercial fishing. These practices are considered to enhance impacts from invasive species by acting as vectors and enhance sedimentation by affecting shoreline. A few experts provided high impact scores for sedimentation impacting aquaculture via smothering. 150   Figure 6.5 Elicited impact curves of various stressors to shellfish aquaculture from the 15 experts that attended the workshop. Violin plots show density histograms overtop of boxplots indicating median (white dot), first and third quartile (box) and max and minimum value (whiskers). Impact scores are on the y-axis, and individual expert number on the x-axis. Violins coloured green represent biodiversity experts, while yellow represents fisheries experts, grey represents aquaculture experts, and blue represents marine recreation experts. 151  6.3.3.3 Fishing Experts identified impacts from ten stresses on fishing, with most experts scoring mid-to-high-range impact (Figure 6.6). The most prominent driver of change experts identified were commercial fishing, climate change, and recreational fishing, with experts providing high range impact curves for commercial fishing and climate change and mid-range curves for recreational fishing. The most prominent stressors are sedimentation and pollution, with sedimentation mostly having high-range curves and pollution mostly with mid-range curves. Fewer experts provided impact curves for shellfish aquaculture and shipping because of limits to access, and some experts provided mid-range impact curves for nutrient input and invasive species. A single expert provided an impact curve for social license, which the expert explained as the negative public opinion of fishing limits the ability of people to enjoy fishing.  152   Figure 6.6 Elicited impact curves of various stressors to fisheries from the 15 experts that attended the workshop. Violin plots show density histograms overtop of boxplots indicating median (white dot), first and third quartile (box) and max and minimum value (whiskers). Impact scores are on the y-axis, and individual expert number on the x-axis. Violins coloured green represent biodiversity experts, while yellow represents fisheries experts, grey represents aquaculture experts, and blue represents marine recreation experts. 153  6.3.3.4 Marine Recreation Experts provided impact curves for eleven stresses on marine recreation, with most experts scoring mid-to-high-range impact curves (Figure 6.7). The most prominent drivers of change are commercial fishing, recreational fishing and climate change, and the most prominent stressors were sedimentation, pollution, and physical trampling. Fewer experts provided impact curves for coastal structures modifying shorelines, shipping and aquaculture for enhancing invasive species, and nutrient input and invasive species spread. 154   Figure 6.7 Elicited impact curves of various stressors to marine recreation from the 15 experts that attended the workshop. Violin plots show density histograms overtop of boxplots indicating median (white dot), first and third quartile (box) and max and minimum value (whiskers). Impact scores are on the y-axis, and individual expert number on the x-axis. Violins coloured green represent biodiversity experts, while yellow represents fisheries experts, grey represents aquaculture experts, and blue represents marine recreation experts. 155  6.3.4 Cumulative Impacts All ecosystem services featured cumulative impacts of 0.5 or greater in Tasman and Golden Bays (as calculated across experts in every ecosystem service type; median cumulative impact score often approached 1, Figure 6.8). This consistently high cumulative impact is found despite sometimes great variation in the impact scores for individual stressors. Many experts indicated that biodiversity might be the most impacted, but all ecosystem services had high cumulative impact scores as perceived by experts. 6.4 Discussion  Cumulative impact scores were uniformly considered to be mid-to-high across ecosystem services (generally 0.5-1), suggesting that there is considerable room for ecosystem service enhancement through management, conservation, and restoration. This result seems surprising given that the bays have a widespread reputation as spots valued for their naturalness and biodiversity (Klain et al. in prep). Some activities act as both benefit and risk, with fishing being a human activity that is both a prominent ecosystem service and risk to ecosystem services. Experts highlighted the diverse, sometimes indirect and amplifying nature of impacts on ecosystem services, and noted four prominent activities and stressors thought to be the most impactful across the four ecosystem services in this study. 156   Figure 6.8 Cumulative impact curves aggregated from individual impact curves elicited from each expert for each of the four ecosystem service types. Density histograms are depicted on top of boxplots indicating median (white dot), first and third quartile (box) and maximum and minimum value (whiskers). Impact scores are on the y-axis, and individual expert number on the x-axis. Violins coloured green represent biodiversity experts, while yellow represent fisheries experts, grey represent aquaculture experts, and blue represent marine recreation experts. The numbers under each violin indicate how many activities and stressors contributed to each cumulative impact curve. Two stressors appeared dominant, affecting all ecosystem services in this study: sedimentation and pollution. Both stressors are modulated by climate change and are a result of processes across the land-sea interface. In tracking the drivers of these stressors, we found that sedimentation has prominent land-based drivers (agriculture and forestry), sea based stressors 157  (commercial fishing and dredging), and a global stressor (climate change). In contrast, pollution has a diffuse set of drivers, with none that stand out as prominent. The variation in responses among experts in impact score as well as the diffuse pathways could indicate that pollution is simply too broad of a topic of impact. The indeterminacy of the term may have led the diverse experts to consider “pollution” in broad ways among themselves (Regan et al. 2002; Fish et al. 2009). Splitting the topic of "pollution" into various types (e.g. "inorganic pollution from runoff", "trash") could lead to a more precise treatment, potentially leading to clearer leverage points for management. Climate change and commercial fishing are consistently dominant drivers of change, and also amplify other indirect stressors. Experts identified many impacts from climate change as amplifying local stressors in these systems, such as sedimentation. The predominance of indirect impacts from climate change found in this study highlights a missing element from many cumulative impact mapping studies, which by their nature focus on direct impacts of climate change (e.g. warming, ocean acidification, e.g. Halpern et al. 2009). Though commercial fishing was mostly associated with the direct impacts of its activities (catch, bycatch, habitat destruction from dredging), it too is a prominent indirect driver of impact by re-suspending land-based sediment that builds up in the bays. Here again prominent indirect pathways of impact by an activity are often not accounted for in prominent treatments of cumulative impacts, this time considering explicitly the mechanisms of impacts traversing the land-sea interface. Where indirect pathways may play a prominent role, considering the mechanism by which impacts occur may be very important. Failing to consider indirect impacts can underrepresent the threats presented by key drivers. 158  Regardless of whether experts were considering impacts of drivers or stressors, for any of the ecosystem services considered, expert provided impact curves around mid-range to high-range scores (0.5-1). Resulting cumulative impact curves made up of these individual curves are very high, approaching (and often butting up to) the upper boundary of impact. This indicates that experts think that the ecosystem services are largely unavailable for human benefit. On its face this result is spurious, as fishing, aquaculture, and tourism are among the key industries in Tasman district. Perhaps experts provided impact curves that are too high for individual stresses, or the cumulative impact equation does not properly capture the way that experts conceive of interactions between impacts. Asking experts to score impacts individually rather than collectively may lead to overestimates of cumulative impacts as people are poor at assigning fractional scores (Bateman et al. 1997). A third possibility is that we double count impact among experts. Though we tried to diminish the possibility of double-counting by asking experts to score impact per stressor only once, experts may not have followed through with this. Those experts that considered impact from climate change to come primarily through indirect processes may have been more likely to score impact from sedimentation and pollution both through these stressors and again from climate change. At the same time, antagonistic impacts (where the total effect is less than expected from an additive model) are common and found in cases of cross-scale impacts, and our model of impact accumulation may not have captured the degree to which some impacts are antagonistic. Experts may all have been advocates trying to highlight the impact of the bays, resulting in overestimation. However, multiple experts representing industry (fisheries and aquaculture) were adamant in the relative benign effects of their industry relative to others, and their cumulative impact scores were still relatively high.  159  A final explanation is that current common beliefs about, and uses dependent on the integrity of the Bays belie very degraded environments compared to their historic potentials, meaning that there is great potential for restoration (Handley 2006). Experts, with historic understanding, may not suffer the shifting-baseline effect that current common opinion may suffer from (Knowlton and Jackson 2008). Many experts in the interview stage highlighted the difference in the bays today compared to historic conditions, indicating that the bays have suffered regime shifts (Rocha et al. 2015a; Rocha et al. 2015b). Responses of interviewed experts may indicate that experts had historic conditions in mind, or were primed by historic levels, when answering questions. Future attempts using the methods of this paper may directly ask experts to provide impact profiles for specific activities and stressors, as well as provide total cumulative impact scores. Using expert estimates of the parts and the whole, an analyst may then: 1) correct individual impact scores to retain their relative weight but sum to the cumulative score, 2) adjust both individual and cumulative impact scores to adhere to a specific model, or 3) confront experts with the apparent contradiction and ask them to justify their thinking and re-evaluate their scores to understand their individual models of impact accumulation. Also, having a clear baseline from which to gauge impact is also needed. Experts may have considered historical baselines when scoring impact, but what this baseline was among experts (or whether all experts agreed) is unclear. Similar to how we set a standard to think about activity levels in the bays, setting a standard to think about baselines of change will help reduce these uncertainties. Despite these limitations and unknowns, we believe our model of cumulative impact scoring offers some insights against more common methods in the literature for two reasons. First, our 160  method allows for uncertainty of individual impacts to directly influence the resulting cumulative impact curve. The distinct advantage here is that the shape of the cumulative impact curve can be analyzed, such as determining whether there are possible high impact-low probability scenarios. (That scenario is rare in this study, since the peaks of the cumulative impact curves are mostly in the high impact region, indicating the prevailing expert view that high impacts are most likely.) Where facilitators train experts and provide feedback on PDFs, as we did during interviews and the workshop, expert distributions can be highly informative (O'Hagan et al. 2006). A second insight gained by this method is that it includes an upper bound of impact. Almost all studies of regional cumulative impacts use an additive model for accumulating impact, with no upper limit of impact scores and no absolute sense of what impact levels mean (Halpern et al. 2009; Halpern et al. 2012a; Allan et al. 2013; Halpern and Fujita 2013; Micheli et al. 2013; Brown et al. 2014; Cook et al. 2014a). Our study has a relative scale of impact to ecosystem services, with a meaningful upper limit. This is effectively modelling a system with breaking points, as many systems cannot face endless stress without responding with mortality or collapse (Groffman et al. 2006; Brown et al. 2013; Halpern and Fujita 2013). As a consequence, when impacts were accumulated, our results show a consistently high impact – with relatively low uncertainty – across ecosystem services. This finding persists even in the presence of high uncertainty regarding individual impacts, except when all contributing stressors had low impact scores (an infrequent occurrence). A pure additive treatment of cumulative impacts may place undeserved attention on accumulating impact scores where in reality a threshold is passed and impact may not continue to accrue. However, identifying thresholds in marine systems is notoriously difficult (Barange et al. 2008; Halpern and Fujita 2013), and we cannot preclude the 161  possibility that experts have mischaracterized the proximity of the current state of ecosystem services to their breaking points.  6.4.1 Using Ecosystem Services as a Basis for Management, Even with Data Paucity Managing coastal systems is complex because of the various interactions of people and the environment, but cumulative impacts need not be paralyzing. We propose that linking ecosystem services and cumulative impact assessment can suggest priorities for management (such as identifying sedimentation as a prominent stressor then identifying the prominent causes of sedimentation for management to regulate). Many prominent drivers of impact in this study operate outside the marine context, such as agriculture, forestry, and climate change. Using an ecosystem service approach allowed us to investigate social impacts as well, particularly changes to access and a lack of social license which can limit the development of shellfish aquaculture. Cross-system impacts occur in multiple places around the world . Given the siloed nature of environmental management, cross-system impacts can be distressing as it raises the question, “what can management do about the stressors?” The answer, we suggest, is to move towards ecosystem-based management (Arkema et al. 2006; Halpern et al. 2008a; Granek et al. 2009; McLeod and Leslie 2009; Tallis et al. 2010), applying our approach of linking mechanistic processes with scales of impact. Though attempting ecosystem based management is intimidating because of persistently high uncertainty and data paucity, our approach offers a set of methods to broadly understand a socio-ecological system with low data availability and assess uncertainty.  While we show that there can be a common set of prominent stressors that impact various ecosystem services (sedimentation, pollution, commercial fishing, and climate change, in this 162  case), their mechanisms of action on different ecosystem services vary. For example, the main concerns identified by our experts as mediating the effect of climate change on marine recreation involve spreading invasive species, increasing pollution-laden runoff, and increasing storm frequencies. The main concern about climate change for shellfish aquaculture, according to our experts, was ocean acidification.  The same human activities can both enhance and degrade ecosystem services (Raymond et al. 2013a). For example, fisheries activities enable the benefit from fish stocks (an ecosystem service) but simultaneously pose a risk to other ecosystems services. Human activities can restrict access to ecosystem services (e.g., aquaculture and marine recreation were suggested to restrict where fisheries can operate), and can degrade habitat for biodiversity and for fisheries (e.g., many experts described fisheries as "self-destructive"). 6.4.2 Addressing Mechanisms of Cumulative Impact Much of the current literature on cumulative impacts has focused on identifying where impacts occur and characterizing the magnitude and nature of the cumulative effect (i.e. whether it is additive, synergistic, or antagonistic; Crain et al. 2008, Darling and Côté 2008, Halpern et al. 2008b, Halpern et al. 2009, Ban et al. 2010, Brown et al. 2013, Halpern and Fujita 2013, Micheli et al. 2013). While this focus has yielded important advances, it does not foster understanding of mechanisms through which impacts occur. The cumulative impact mapping approach originally developed in Halpern (2008) is arguably the most influential technique to assess cumulative impacts in the literature, and has developed its own pedigree with associated strengths (e.g. spatial focus) and pitfalls (e.g. static representation; Halpern et al. 2008b, Halpern et al. 2009, Ban et al. 2010, Halpern and Fujita 2013). Here we showcase another approach to advance 163  cumulative impact studies that focuses instead on linking magnitude of impact with causes of impact. Our method has its own strengths (e.g. linking causes and consequence, disentangling direct and indirect pathways of impact) and weaknesses (e.g. no spatial resolution within the study area). While other approaches exist to address the process of impact generation (most commonly DPSIR - Driver Pressure State Impact Response - approaches), often separate impact pathways are treated in isolation and not in concert (Atkins et al. 2011). When multiple stressors are considered, they are often considered in the context of impacts to a single component of the environment (i.e. seagrasses, Grech et al. 2011); marine landscapes, Stelzenmüller et al. 2010); corals, Selkoe et al. 2009). Like Kelble et al. (2013) and Cook et al. (2014b) we emphasize the links between impact and multiple ecosystem services, but unlike these other efforts we are interested in exploring the network of processes (the diversity of ways that activities cause impacts and how they interact) linking human activities to stresses to impacts on ecosystem services, and understanding the full (direct and indirect) impact of stressors on ecosystem services. Focusing on mechanistic networks in concert with impact magnitude has potential to advance the field of cumulative impact studies. Combining multiple pathways of impact into networks of impacts can showcase causal networks and point to prominent drivers and/or potential routes for further research and potential entry points for management and recovery (Niemeijer and de Groot 2008). To our knowledge this is the first study to explicitly explore the multiple mechanistic pathways by which human activities cause impacts, and to link these pathways to impact magnitude. Other studies that use a network approach with associated impact scores link activities (e.g. commercial fishing) to states of the environment (e.g. protected species) without 164  addressing stressor mechanisms (Kelble et al. 2013; Cook et al. 2014b). Overlooking the prominent stressors associated with human activities deemphasizes the multi-causal paths of impact that connect individual activities to individual ecosystem services. 6.4.2.1 Prioritizing Research and Management Sedimentation provides an illustrative example of how the approach presented here can lead to targeted management action. Agriculture, forestry, dredging and commercial fishing were all suggested by many experts to contribute to sedimentation, which was one of the most important stressors across ecosystem services by rank and impact score. Effective management of sedimentation will require consideration and regulation of not only the sources of sedimentation (land-based practices of agriculture and forestry), but also to activities that re-suspend and re-animate the sediment that already exists in the bays (e.g. commercial dredge and trawl fishing, Eriksson et al. 2004). Limiting the human-caused land-based inputs of sedimentation will not negate the problem of resuspension because sediment is not only introduced from runoff inputs within the bays. Alternatively, aggressive restoration efforts to re-establish historic oyster reefs and mussel beds in the Bays may help control erosion and sedimentation (Coen et al. 2007). Historically, expansive oyster reefs and mussel beds may have provided this function until dredging and other human disturbance brought an alternative stable state (Jackson et al. 2001; Handley 2006; Coen et al. 2007). Managing cumulative impacts requires management accounting for the causal network of impact.  165  6.4.3 Cross-scale impacts Scenarios of cumulative impacts often involve interactions of drivers and stressors that cross scales (Brown et al. 2013; Brown et al. 2014). In this case study, climate change is identified as a major global stressor across ecosystem services, but some of the prominent drivers and stressors associated with climate change are actually indirectly linked to local or regional stressors. Climate change may be a major stressor largely because it amplifies local stressors (Harley et al. 2006). An often mentioned example from the experts in this case study is that climate change increases the intensity and variability of storms and runoff, which increases sedimentation and pollution. By managing local stressors that climate change contributes to, such as sedimentation, global stressors may be effectively mitigated to some degree (Brown et al. 2013; Brown et al. 2014). Some treatments of cumulative impacts in the literature effectively ignore global changes because they argue that local and regional management cannot prevent global stressors (Ban et al. 2010), but our results show that global change should be taken into account anyway because they interact with local stressors (Brown et al. 2013; Brown et al. 2014) 6.5 Conclusions An ecosystem service focus to cumulative impact analysis can provide a basis for understanding multiple, competing uses, and provide a series of clear management goals (Granek et al. 2009). Studying cumulative impacts is fraught with data paucities, and this is especially true for cumulative impacts to ecosystem services. However, the techniques used in this study can be applied with extreme data limitations to identify prominent drivers and stressors of impact, understand their mechanistic networks, and link these to magnitude of impact can be used to point to specific targets of activities to monitor and regulate, even in cases of global change and 166  across ecosystem boundaries.  As researchers of cumulative impact studies, we believe there are opportunities to effectively manage for cumulative impacts, even recover and restore socioecological systems to render greater benefits from ecosystem services. We argue that the first step to manage cumulative impacts is to understand how impacts operate (Guerry 2005; Granek et al. 2009; McLeod and Leslie 2009; Altman et al. 2010); the second is for management institutions to operate in an ecosystem-based management framework.    167  Chapter 7: Conclusion 7.1 Summary of Key Findings and Discussion The human relationship with the natural environment is broadly bi-directional: we effect change to the environment and face environmental consequences. Though this type of relationship is true for all species, humanity is changing the environment (with consequences to ecosystem services) to an unparalleled extent. Humans as threats to their own well-being has been recognized across societies and time periods, reflecting the specific interactions between people and the environment in those contexts. The ancient Mesopotamians recognized the role of forest clearing on soil retention and runoff (Yasuda et al. 2000), pre-colonial aboriginal peoples of modern Washington, British Columbia, and Alaska recognized the role of overfishing for their harvest of salmon (Trosper 2002), research points to cumulative impacts from fisheries, invasive species, and land use to fisheries in the African Great Lakes (Ogutu-Ohwayo et al. 1997), and we are now understanding humanity’s collective contribution to global change, potentially affecting all ecosystem services (Sanderson et al. 2002; Parmesan and Yohe 2003; MA 2005; Estes et al. 2011). Effective management of ecosystems and their life-sustaining contributions requires rigorous study to understand and respond to cumulative impacts, yet explicitly connecting impact and services is a burgeoning field of study (Halpern and Fujita 2013). Considering ecosystem services means considering social dimensions at a level well beyond what might be needed to characterize impacts to habitat (Tallis et al. 2012). That is, social factors contribute to impacts to ecosystem services via degradation of the perceived quality of an ecosystem service at a site and ability of people to benefit from an ecosystem service at a site. 168  But social considerations are also important in assessing, mediating and responding to impacts on ecosystem services. In this dissertation I assess the practices of institutions that predict and plan for emerging impacts (Chapter 2), provide tools to help respond to existing impacts (Chapter 4), assess the importance of social dimensions of impact by exploring the causal mechanisms of impact (Chapters 3 and 6), and explore reliable ways to elicit information in structured social settings (through expert elicitation, Chapter 5). The primary legal mechanisms by which cumulative impacts are predicted and addressed is environmental impact assessment (Wood 2003). In Chapter 2, I showed that, around the world, anthropogenic impacts are routinely considered unimportant for decision-making while various aspects of the process bias towards this finding. Assessment scopes are often inadequate to address cumulative impacts given the ranges of affected ecosystem components (usually species or populations) and the documented duration of residual effects (e.g., acid mine drainage). These assessments are carried out by consultants paid for by developers, and stakeholders are rarely given the chance to determine the importance of an impact (Beder 1993). EIAs often highlight project modification, leading many researchers to claim that mitigations are the most important contribution of EIAs (Wood 2003; Ott et al. 2012). However, I show that mitigation measures are often justified without reference to existing studies and without assessments of effectiveness or uncertainty; furthermore, the language used to describe mitigations frequently renders them unenforceable. The lack of attention paid to stakeholders may reflect a lack of consideration of how people benefit from the environment, which an ecosystem service framing, alongside a consultation to determine prominent environmental values, may help rectify (Klain and Chan 2012; Raymond et al. 2013b).  169  In Chapter 3 I use cumulative impact mapping to assess cumulative impacts to diverse ecosystem services in British Columbia. Results from this chapter suggest that the spatial overlap of ecosystem services with activities and stressors varies greatly across ecosystem services. This chapter also provides evidence that social criteria of risk (specifically perception of quality loss and changes to access) contribute importantly to ES impacts, and in can expand the consideration of the most pertinent mechanisms of impact. Linking these findings with those of Chapter 2, the importance of social considerations and processes of impact suggests that environmental assessment processes should include stakeholder input on ecosystem services to assess impact on, as well as determine social impact from development (Kasperson 2006).  Results from Chapter 3 also highlight the importance of understanding causal processes in characterizing impact to different ecosystem services, because the data needed to estimate impacts and determine the zone of influence are specific to the causal process at play, and these processes vary across services. This finding cautions against any researcher thinking about directly acquiring spatial data from one study to apply to another when assessing impact to different ecosystem services. Though using spatial data from existing cumulative impact assessments is common practice in cumulative impact mapping literature, especially when acquiring data from a study at a larger scale  (Halpern et al. 2009; Kappel et al. 2012; Murray et al. 2015a), careful consideration regarding the relationship of the zone of influence of specific activities and stressors with specific ecosystem services is required to meaningfully represent impact.   In Chapter 4, I pilot new methods to prioritize stressors and activities that offer management leverage to target with mitigation, using Bayesian Belief Networks. The literatures of EIA and 170  restoration ecology highlight the mitigation hierarchy to prioritize strategies to achieve impact avoidance first, then impact minimization, then restore and offset, and finally when all else is impossible, compensate for impacts felt (Tinker et al. 2005; Quintero and Mathur 2011). However, the scale and complexity of cumulative impacts can make them near impossible to fully mitigate, requiring methods to simplify and prioritize mitigations strategies necessary to achieve management goals (Knights et al. 2013a). Despite this need, most studies on cumulative impacts are concerned with characterizing impact: where, how, and to what extent they occur. Few studies explicitly address the management response to cumulative impacts. Knights et al. (2013a) suggests methods to group similar causal impacts in order for managers to manage many impacts with few mitigations, though it does not prioritize management based on what will lead to the greatest reduction in impact. The methods I pilot in Chapter 4 directly prioritize management and mitigation to target prominent activities and stressors that drive risk to valued species. Process based models of impact and network frameworks have advanced our understanding of cumulative impacts by emphasizing causal mechanisms, though they neglect the magnitude of impact associated with individual (and interacting) causal pathways (Rounsevell et al. 2010; Atkins et al. 2011; Kelble et al. 2013). The results of Chapter 4 explicitly link impact process and magnitude in an effort to plan management responses to impacts. Similar methods can be employed for management agencies and environmental assessments to prioritize mitigation. Whereas expert judgment elicitation is a widely recognized crucial tool for addressing the data limitations that pervade cumulative impacts research, Chapter 5 addresses two important concerns regarding the estimation of uncertainty in expert elicitation. First, to counter the widely 171  recognized problem of expert overconfidence, I showed that group deliberation leads to individual experts provide larger confidence bounds around their estimate. Second, reducing linguistic uncertainty and extreme response, experts provide more consistent ranks of prominent activities and stressors with similar best estimates of impact following group elicitations than in individual settings. Expert elicitation is a tool with a rich literature establishing best practices, but established practices do not account for the structural uncertainties that characterize many contexts of cumulative impacts -namely uncertainty about which parameters to estimate (Fischhoff and Morgan 2013). Research on expert elicitation procedures often focus on parameter estimation and representing subjective uncertainty when all important parameters have been defined (Burgman 2005). Though results from this chapter may suggest that experts can be subject to groupthink bias following group deliberation (because of more consistent responses among experts), the inclusion of diverse experts should mitigate this risk (Fish et al. 2009; Sunstein and Hastie 2015). Group processes may help ensure experts are considering the problem context in similar ways, reducing the linguistic uncertainty among experts (Regan et al. 2002; Fish et al. 2009). Individual experts may have multiple terms to describe the same phenomena and group settings may allow them to come to a common terminology as well as reassess their assumptions. Using humans as a source of data is fraught with challenges because of the many sources of biases that can influence expert responses (Gilovich et al. 2002; Regan et al. 2002; McBride et al. 2012). When relying on expert judgment, impact assessments should do all they can to limit introgression of bias, and my results indicate that using group deliberations can be important. 172  Chapter 6 addresses two major concerns prominent in cumulative impact assessment, and does so in a context of sparse data: estimating total impact faced by ecosystem services and identifying prominent activities and stressors responsible for impacts. Using a model for the accumulation of impacts that sets a meaningful upper bound of impact, I found that cumulative impact across ecosystem services was reliably high despite high uncertainty in individual impacts. At the same time, a consistent few activities and stressors were responsible for impacts across ecosystem services, with important land (e.g. sedimentation), sea (e.g. commercial fishing), and global change (e.g. increased storm intensity) variables driving them.  Many cumulative impact studies utilize additive models of impact without an upper bound, focusing on total impact (Halpern et al. 2009; Ban et al. 2010; Kappel et al. 2012; Allan et al. 2013; Brown et al. 2014; Murray et al. 2015a). The results of this chapter suggest that relying on purely additive models may ignore important thresholds of impact to ecosystem services. The common set of activities and stressors that drive impact likely follow the fact that the ecosystem services investigated here (fisheries, aquaculture, marine tourism, and existence of biodiversity) all functionally depend on common coastal biological community primarily composed of shellfish and finfish (though marine recreation also considers landscape characteristics and other species such as birds).  On the surface, results from Chapters 3 and 6 contradict each other. In British Columbia (Chapter 3), impact to diverse ecosystem services is driven by diverse activities and stressors, while in New Zealand (Chapter 6), impact to diverse ecosystem services is driven by a common set of activities and stressors. However, many more ecosystem services were investigated in British Columbia than in New Zealand. Comparing the ecosystem services in common between 173  the two case studies (fisheries, aquaculture, and marine recreation), there were common prominent activities and stressors driving impact in British Columbia and New Zealand (largely climate change related stressors). In both locations, marine recreation, fisheries and aquaculture are often functionally dependent on marine species (Dawson 2005; Cisneros-Montemayor and Sumaila 2010). Only when a greater diversity of ecosystem services are considered (including potential energy generation – primarily dependent on the physical environment, aesthetics – primarily dependent on landscape, and coastal protection – primarily dependent on primary producers, and bathymetry) does it become clear that a larger variety of activities and stressors are important for impact to ecosystem services. Like portfolio theory in ecology, the functional diversity of ecosystem services may regulate the sensitivity of ecosystem services to common impacts (Schindler et al. 2010).  Beyond substantive empirical contributions to the cumulative impact and ecosystem services literatures, this dissertation advances these disciplines methodologically. First, Chapter 3 adapts the cumulative impacts mapping approach commonly used to assess impact to ecosystem type and addresses cumulative impacts to ecosystem services (Halpern et al. 2009; Kappel et al. 2012; Allan et al. 2013; Brown et al. 2014; Murray et al. 2015a). Using the marine InVEST modeling platform, eight diverse ecosystem services were modeled to assess impact (Guerry et al. 2012; Sharp et al. 2014). Concurrently, I modified an expert survey to investigate the vulnerability of different ecosystems to various activities and stressors to assess risk to ecosystem services (Teck et al. 2010). This modification of a widely used assessment methodology can be adopted to assess cumulative impact in many parts of the world. However, this is a data intensive methodology that will not be suitable everywhere. 174  Second, I advance network modeling of cumulative impacts to account for total risk to different species in order to prioritize causal pathways of impact for management and mitigation in Chapter 4. Using a risk assessment procedure created to propagate uncertainty and accumulate levels of risk (O et al. 2015), this method employs a Bayesian Belief Network to determine the nodes in the network that contribute the most risk to given species. This method successfully marries process-based network modeling with quantitative estimates of risk (Halpern et al. 2012c; Cook et al. 2014b; Knights et al. 2015). It is also one of the few analytical methods currently developed to explicitly address the management response to cumulative impacts (but also see Knights et al 2013). Though this is a method that can be very data intensive, the data can come from the literature and expert input. Finally, this dissertation describes a novel expert elicitation strategy to understand causal processes of impact and to estimate total impact, given data paucity (Chapters 5 and 6). Given the technical responses asked of experts (i.e. providing probability distribution functions of impact scores), as well as the depth of information asked of experts (i.e. identifying prominent activities and stressors, outlining their relationships with ecosystem services, and estimating their impact), this elicitation design incorporates ongoing training and feedback to ensure that experts know how to carry out tasks and to minimize translation error between experts and analysts (Aspinall 2010; McBride and Burgman 2012; Morgan 2014; Sutherland and Burgman 2015). As I argue in Chapter 5, this elicitation strategy has favorable effects on expert uncertainty. Additionally, it borrows the philosophy of building a weight of evidence to increase confidence in a networks of causal impact pathways (Linkov and Satterstrom 2006). This expert elicitation strategy for understanding cumulative impacts can be applied to any context with willing experts. 175  7.2 Potential Applications and Future Research As a dissertation with considerable methodological innovation, the potential applications of this dissertation for future research and environmental management are broad. The analysis on EIAs can inspire further research and (hopefully) lead to change in EIA practice. The methods and findings of Chapters 3-6 can all promote greater research effort into cumulative impacts and ecosystem services, and have the potential to be used as methods in various research, management, and regulatory processes for the environment. Instead of suggesting an entirely new system to supplant current EIA practice, I recommend improvements be implemented to EIA practice that already has global reach. The results of Chapter 2 have important implications for EIAs as practiced in many nations, as indicated in “Summary of Key Findings and Discussion” above.  The consideration of future impacts within cumulative impact assessment was unexplored by my study, but is a ripe topic for future research leading to recommendations for practice. Though ecosystem services are starting to enter the EIA process, many nations do not directly appeal to the concept. However, some EIAs describe specific values associated with environmental components that impacts are assessed on. Future research exploring the latent attention paid to certain classes of ecosystem services (i.e. provisioning, supporting, regulating, cultural) will provide insight into what types of ecosystem services are more likely to be assessed and what types are more often neglected. Chapter 3 demonstrates the use of applying spatial ecosystem service models to generate maps of ecosystem services (Guerry et al. 2012), but any method capable of mapping ecosystem services may be suitable for the task of mapping impact to ecosystem services. Future research can determine if specific types of ecosystem services are generally at risk through social causal 176  process, as well as if some types of ecosystem services generally face greater impact than others. The value of ecosystem services can be highly site-dependent, and because perceptions can affect enjoyment of services, collecting data relevant to explicit beneficiaries of ecosystem services may provide results most relevant to these beneficiaries (Wieland et al. 2016).  Results from Chapters 3 and 6 suggest that assessments of impact to ecosystem services that do not address social pathways of impact can underestimate impact. This challenges many accepted representations of how humans benefit from, and cause impact to, ecosystem services. Prominent frameworks of ecosystem services hold that humans derive benefit from the environment purely through changes in the biophysical environment, and that impacts to ecosystem services derive purely a consequence to changes in the production of ecosystem services (Collins et al. 2010; Rounsevell et al. 2010; Atkins et al. 2011; Kelble et al. 2013; Cook et al. 2014b). The recognition that ecosystem services can change as a result of how people perceive and respond to impact reopens the question of how people relate to the biophysical environment (Chan et al. 2012a; Chan et al. 2012b). Clearly, understanding how people benefit through ecosystem services, and how we affect those services is in need of greater theoretical and empirical study.  The methods used in Chapters 4-6 can be applied in assessment and environmental management contexts to understand the causes and consequences of impact, and determine leverage points for management and mitigation. Further research can help determine if leverage nodes (the parts of the network that regulate risk to valued species and ecosystem services) in an impact network generally occur at activity or stressor levels, and under what conditions are leverages found at each level. Additional research can greatly improve expert elicitation strategies to address cumulative impacts. How impacts accumulate quantitatively is often an open question (Brown et 177  al. 2013; Brown et al. 2014), and I made assumptions about how experts considered impact accumulation in Chapter 6. Similarly, I piloted methods to build impact pathways based on accumulating expert judgment, and further research is needed to further develop this method to ensure results are robust and repeatable in similar ways that Chapter 5 showcases how group elicitation can reduce bias and uncertainty in expert responses. The methods employed in this dissertation can be combined with other existing methodologies. Different methods have set strengths and weaknesses, and set focuses and limitations, so combining methodologies has the potential to expand the scope and insight of a project. One potential application of combined methods is to use cumulative impact mapping with impact network analysis. Cumulative impact mapping is usually not a strong method to consider causal pathways of impact to prioritize management and mitigation action, and network impact analysis is not explicitly spatial (Niemeijer and de Groot 2008; Halpern and Fujita 2013). Combing these approaches can potentially allow for an analysis to generate spatially explicit management and mitigation priorities to combat cumulative impacts. 7.3 Strengths and Limitations 7.3.1 Strengths As an emerging field, this dissertation has forged new paths across many issues at the intersection of cumulative impacts and ecosystem services. This dissertation’s simultaneous assessment of prominent processes to understand impacts to ecosystems (EIA) alongside studies that promote methods to advance impact assessment in the future, simultaneously characterizes and advances the state of the field. It highlights the importance of social dimensions of impact to 178  ecosystem service to enrich both impact assessment and basic understanding of ecosystem services (Chan et al. 2012a; Chan et al. 2012b; Tallis et al. 2012). By connecting research on causal processes of impact with quantitative estimates of impact, this dissertation also attempts to transition research towards considering management responses to cumulative impacts (Knights et al. 2013a). The multinational scope of Chapter 2 and its multiple lines of evidence demonstrate that there are widespread problems with EIA, and that they are common across diverse jurisdictions. By comparing self-reported EIA statistics with data related to EIA goals (e.g. relating spatial scale of assessments with spatial scale of assessed species) as well as calculating unique metrics to characterize EIAs (e.g. the proportion of unenforceable mitigations), I provide an evidentiary basis for critique on a widespread policy tool for environmental protection. This style of analysis can be applied to other institutions and policies, and may reveal similar insights. The greatest strength of Chapter 3 was its integration of cutting-edge methods for characterizing impacts to ecosystems with sophisticated methods for mapping ecosystem services. By combining methods from cumulative impact literature with ecosystem services literature I was able to highlight the importance of representing ecosystem services explicitly and to explore both the biophysical and social impacts that accrue to ecosystem services. The strength of Chapter 4 was the development of a method to evaluate management plans and suggest priority management and mitigation measures to protect species and ecosystem services that are threatened by cumulative impacts. Because impact networks are directional and modular (Uusitalo 2007), the network analysis method I pilot has many future potentials. A further extension of this network method would be to link ecosystem services to diverse beneficiaries. 179  Various groups of people do not homogenously benefit from ecosystem services, a characteristic only now gaining some attention among ecosystem service researchers (Wieland et al. 2016). To date, impact studies to ecosystem services have not related impact to ecosystem services to different beneficiaries, though some existing frameworks include beneficiaries (Rounsevell et al. 2010; Mach et al. 2015). The methods in Chapter 4 could be used to prioritize management actions to protect ecosystem services for specific groups of people. In terms of methodological innovation, perhaps the greatest strength of this dissertation is the successful piloting on methods that can be applied in a variety of data contexts. The expert elicitation structure in Chapters 5 and 6, which built on previous expert elicitation literature, provides a way for researchers and managers to understand the causal pathways of prominent activities and stressors along with total impact (with associated uncertainty) with access to sparse data. Using this approach, research can suggest management actions to combat cumulative impacts with experts being the only available data source. 7.3.2 Limitations As a dissertation striving to advance an emerging field, this dissertation has many limitations. First, my analysis of EIAs took only a sample of reports from a sample of countries worldwide. I cannot conclusively relate my findings globally, though there are reasons to believe that similar patterns might hold elsewhere (i.e. the diasporic spread of the EIA model around the world from a common place, Wood 2003). Furthermore, because I consulted only the environmental impact statements and not audits or other independent studies of proposed projects, I cannot conclude that impacts occur at greater frequencies than predicted from EIAs, I can only surmise that they 180  might. All results from this chapter are based on self-reported indices of the reports, so I have propagated any inaccuracies in the reports. To generate maps of ecosystem services in Chapter 3, I relied on the marine InVEST modeling platform (Guerry et al. 2012; Sharp et al. 2014), which confers several limitations. InVEST can be used to create spatial models of ecosystem services, producing spatial extents but also magnitudes of service provision. However, due to data limitations on a number of modeled ecosystem services, I could not generate these measures of magnitude. If the needed data were available, I could have related impact to areas weighted based on amount of service in a given area, providing even richer maps. Impacts on areas of high service value will be of greater concern to some than impacts on areas of low service value. As in most cumulative impact mapping studies, I relied on an additive model of cumulative impact, assuming no threshold responses by any ecosystem service to impact (Halpern and Fujita 2013), which is analytically tractable but sure to misrepresent impacts. While we did not have data on the particular interactions among multiple impacts to our ecosystem services, there is strong reason to believe that a universal additive model is incorrect (Crain et al. 2008; Darling and Côté 2008). Similarly, in Chapter 3 I followed previous studies and represented ecosystem services as able to withstand limitless amounts of impact (i.e. no threshold), which is clearly simplistic (Halpern and Fujita 2013). These are recognized limitations of the cumulative impact mapping framework, and there are not yet widely accepted methods to correct for them. The expert elicitation survey used in Chapter 3 was based on a previous survey used to generate vulnerability scores of ecosystems to activities and stressors (Teck et al. 2010). As such, my survey propagated some limitations from the initial survey. Experts had little opportunity to 181  clarify questions and could not benefit from group processes, potentially contributing to error in the results (Burgman 2005). This survey format also asks experts to quantify risk criteria across a large number of activities and stressors, potentially contributing to survey fatigue in experts (Porter et al. 2004). The major limitation in Chapter 4 was that I could not assess risk to ecosystem services as enjoyed by people, but rather as supplied by the ecosystem. This was done because the risk assessment I had access to was focused on risk to ecosystem components (in many cases, species) and not to ecosystem services specifically (O et al. 2015). A similar risk assessment explicitly for ecosystem services is a massive undertaking beyond the scope of this dissertation, but an important next step.  Also for Chapter 4, I relied on a Bayesian Belief Network to analyze risk networks. While allowing for complex conditional probability calculations and explicitly analyzing uncertainty, Bayesian network analysis is completely dependent on the set structure of the network (Uusitalo 2007). If there is error in the structure of the network, that error is unacknowledged and will affect the results, such as if a key stressor that really impacts a species is not represented in the network. The risk assessment used to parameterize the Bayesian Belief Network was focused on characterizing risk as it currently exists, and using the uncertainty from that assessment to create the conditional probability tables for the analysis assumes that uncertainty in current risk represents the scope of risk possible under different management plans. Bayesian networks also only allow  directed networks (no feedbacks), limiting the dynamics that can be represented in the analysis (Uusitalo 2007; Choy et al. 2009). I tried to incorporate foodweb dynamics among species in the analysis by having foodweb linkages influence risk to the species nodes, but even 182  with this addition the network structure is a static, fixed representation of what is actually a dynamic network. Conducting the analysis using multiple iterations of network structure may help alleviate the problems associated with static network structure (Uusitalo 2007). For Chapter 5, I used a before-after design to understand the influence of group elicitation on expert responses. Though this provides evidence of influence, a before-after-control-impact (BACI) design provides stronger evidence (Underwood 1991). By comparing against a control group, I could rule out potential competing explanations for my results (that experts are more consistent but individually more unsure in their judgment after group deliberation), such as individual experts having more time to reflect on the questions. I did not have access to a sufficient number of willing experts to conduct this experiment, however. Fortunately, notes from participants and facilitators in the workshop corroborate my inference that the change was largely a result of the group process. The major limitation of Chapter 6 was that I assumed the model of how impacts accumulate. The combination of this additive model (with an upper threshold) and high estimated individual impacts resulted in consistently high impact assessments across experts. Although this model allowed me to demonstrate the influence of an upper threshold on impact scores even in the context of high uncertainty in individual impacts, individual experts may have had very different models of cumulative impact (Aven and Guikema 2011). To explore the usefulness of a chosen accumulation model, future studies can deliberately elicit cumulative impact scores from experts to compare against different models of impact accumulation. The method I used to generate impact networks was also piloted, having no previous studies to provide an evidentiary basis to support it. 183  7.4 Revisiting Our Dual Role as Agents of Risk and Beneficiaries Humanity’s relationship with the natural environment is a constant tension between our desires and effects (Raymond et al. 2013a). Because we pose risks to the processes that benefit us, our tension is often with ourselves, usually with consequences to some groups while benefits accrue to others. Successful environmental management is thus management of human activities (McLeod and Leslie 2009). Understanding the various ways in which humans plan for, cause, and respond to impacts is crucial to achieve successful management of human impact (Arkema et al. 2006; Altman et al. 2010; Curtin and Prellezo 2010). I hope this dissertation, as both a study on its own as well as a developer and proponent of methodologies, can contribute to our understanding of our impacts. With time, and with knowledge, perhaps we can diminish our role as agents of risk and enhance our role as beneficiaries.   184  Bibliography Abràmoff, M., P. Magalhães and S. Ram (2004). "Image processing with ImageJ." Biophotonics International 11(7): 36-42. Allan, J. D., P. B. McIntyre, et al. (2013). "Joint analysis of stressors and ecosystem services to enhance restoration effectiveness." Proceedings of the National Academy of Sciences 110(1): 372-377. Altman, I., A. M. Blakeslee, G. C. Osio, C. B. Rillahan, S. J. Teck, J. J. Meyer, J. E. Byers and A. A. Rosenberg (2010). "A practical approach to implementation of ecosystem-based management: a case study using the Gulf of Maine marine ecosystem." Frontiers in Ecology and the Environment 9(3): 183-189. Andersen, J. H., B. S. Halpern, S. Korpinen, C. Murray and J. Reker (2015). "Baltic Sea biodiversity status vs. cumulative human pressures." Estuarine, Coastal and Shelf Science 161: 88-92. Angradi, T. R., J. J. Launspach, D. W. Bolgrien, B. J. Bellinger, M. A. Starry, J. C. Hoffman, A. S. Trebitz, M. E. Sierszen and T. P. Hollenhorst (2016). "Mapping ecosystem service indicators in a Great Lakes estuarine Area of Concern." Journal of Great Lakes Research 42(3): 717-727. Arkema, K. K., S. C. Abramson and B. M. Dewsbury (2006). "Marine ecosystem-based management: from characterization to implementation." Frontiers in Ecology and the Environment 4(10): 525-532. Aspinall, W. (2010). "A route to more tractable expert advice." Nature 463(7279): 294-295. 185  Atkins, J. P., D. Burdon, M. Elliott and A. J. Gregory (2011). "Management of the marine environment: integrating ecosystem services and societal benefits with the DPSIR framework in a systems approach." Marine Pollution Bulletin 62(2): 215-226. Aven, T. and S. Guikema (2011). "Whose uncertainty assessments (probability distributions) does a risk assessment report: the analysts' or the experts'?" Reliability Engineering & System Safety 96(10): 1257-1262. Baayen, R. H., D. J. Davidson and D. M. Bates (2008). "Mixed-effects modeling with crossed random effects for subjects and items." Journal of Memory and Language 59(4): 390-412. Ban, N. and J. Alder (2008). "How wild is the ocean? Assessing the intensity of anthropogenic marine activities in British Columbia, Canada." Aquatic Conservation: Marine and Freshwater Ecosystems 18(1): 55-85. Ban, N. C., H. M. Alidina and J. A. Ardron (2010). "Cumulative impact mapping: Advances, relevance and limitations to marine management and conservation, using Canada's Pacific waters as a case study." Marine Policy 34(5): 876-886. Ban, S. S., N. A. Graham and S. R. Connolly (2014a). "Evidence for multiple stressor interactions and effects on coral reefs." Global Change Biology 20(3): 681-697. Ban, S. S., R. L. Pressey and N. A. J. Graham (2014b). "Assessing interactions of multiple stressors when data are limited: A Bayesian belief network applied to coral reefs." Global Environmental Change 27: 64-72. Banerjee, S. B. (2000). "Whose land is it anyway? National interest, indigenous stakeholders, and colonial discourses. The case of the Jabiluka uranium mine." Organization & Environment 13(1): 3-38. 186  Barange, M., G. Beaugrand, R. Harris, R. I. Perry, M. Scheffer and F. Werner (2008). "Regime shifts in marine ecosystems: detection, prediction and management." Trends in Ecology & Evolution 23(7): 402-409. Baron, R. S. (2005). "So right it's wrong: Groupthink and the ubiquitous nature of polarized group decision making." Advances in Experimental Social Psychology 37: 219-253. Bateman, I., A. Munro, B. Rhodes, C. Starmer and R. Sugden (1997). "Does part–whole bias exist? An experimental investigation." The Economic Journal 107(441): 322-332. Bateman, I. J., G. M. Mace, C. Fezzi, G. Atkinson and K. Turner (2011). "Economic analysis for ecosystem service assessments." Environmental and Resource Economics 48(2): 177-218. BCEAO (2013). Guideline for the selection of valued components and assessment of potential effects, British Columbia Environmental Assessment Office. 45 pp. BCMCA. (2016). "BC Marine Conservation Analysis."   Retrieved 2016, from http://bcmca.ca/. Beanlands, G. E. and P. N. Duinker (1983). An ecological framework for environmental impact assessment in Canada, Dalhousie University, Canada 142 pp. Beder, S. (1993). "Bias and credibility in environmental impact assessment." Chain Reaction 68: 28-30. Biernacki, C. and J. Jacques (2013). "A generative model for rank data based on insertion sort algorithm." Computational Statistics & Data Analysis 58: 162-176. Biggs, R. O., G. D. Peterson and J. C. C. Rocha (2015). "The regime shifts database: A framework for analyzing regime shifts in social-ecological systems." bioRxiv: 018473. 187  Bina, O. (2007). "A critical review of the dominant lines of argumentation on the need for strategic environmental assessment." Environmental Impact Assessment Review 27(7): 585-606. Boumans, R., J. Roman, I. Altman and L. Kaufman (2015). "The Multiscale Integrated Model of Ecosystem Services (MIMES): Simulating the interactions of coupled human and natural systems." Ecosystem Services 12: 30-41. Boyles, J. G., P. M. Cryan, G. F. McCracken and T. H. Kunz (2011). "Economic importance of bats in agriculture." Science 332(6025): 41-42. Briggs, S. and M. D. Hudson (2013). "Determination of significance in ecological impact assessment: past change, current practice and future improvements." Environmental Impact Assessment Review 38: 16-25. Brown, C. J., M. Bode, O. Venter, M. D. Barnes, J. McGowan, C. A. Runge, J. E. M. Watson and H. P. Possingham (2015). "Effective conservation requires clear objectives and prioritizing actions, not places or species." Proceedings of the National Academy of Sciences 112(32): E4342. Brown, C. J., M. I. Saunders, H. P. Possingham and A. J. Richardson (2013). "Managing for Interactions between Local and Global Stressors of Ecosystems." PLoS ONE 8(6): e65765. Brown, C. J., M. I. Saunders, H. P. Possingham and A. J. Richardson (2014). "Interactions between global and local stressors of ecosystems determine management effectiveness in cumulative impact mapping." Diversity and Distributions 20(5): 538-546. Burgman, M. (2005). Risks and Decisions for Conservation and Environmental Management, Cambridge University Press. 504 pp. 188  Burgman, M., A. Carr, L. Godden, R. Gregory, M. McBride, L. Flander and L. Maguire (2011a). "Redefining expertise and improving ecological judgment." Conservation Letters 4(2): 81-87. Burgman, M. A., M. McBride, R. Ashton, A. Speirs-Bridge, L. Flander, B. Wintle, F. Fidler, L. Rumpff and C. Twardy (2011b). "Expert status and performance." PLoS ONE 6(7): e22998. Cadotte, M. W., J. Cavender-Bares, D. Tilman and T. H. Oakley (2009). "Using phylogenetic, functional and trait diversity to understand patterns of plant community productivity." PloS ONE 4(5): e5695. Canter, L. and G. Canty (1993). "Impact significance determination—basic considerations and a sequenced approach." Environmental Impact Assessment Review 13(5): 275-297. Canty, A. and B. Ripley (2015). boot: Bootstrap R (S-Plus) Functions. R package version 1.3-17. Carey, J. M. and M. A. Burgman (2008). "Linguistic uncertainty in qualitative risk analysis and how to minimize it." Annals of the New York Academy of Sciences 1128(1): 13-17. Carpenter, S. R., H. A. Mooney, et al. (2009). "Science for managing ecosystem services: Beyond the Millennium Ecosystem Assessment." Proceedings of the National Academy of Sciences 106(5): 1305-1312. Casiot, C., M. Egal, F. Elbaz-Poulichet, O. Bruneel, C. Bancon-Montigny, M.-A. Cordier, E. Gomez and C. Aliaume (2009). "Hydrological and geochemical control of metals and arsenic in a Mediterranean river contaminated by acid mine drainage (the Amous River, France); preliminary assessment of impacts on fish (Leuciscus cephalus)." Applied Geochemistry 24(5): 787-799. 189  Chan, K. M., A. D. Guerry, et al. (2012a). "Where are cultural and social in ecosystem services? A framework for constructive engagement." BioScience 62(8): 744-756. Chan, K. M. and M. Ruckelshaus (2010). "Characterizing changes in marine ecosystem services." F1000 Biology Reports 2(54). Chan, K. M., T. Satterfield and J. Goldstein (2012b). "Rethinking ecosystem services to better address and navigate cultural values." Ecological Economics 74: 8-18. Chan, K. M., M. R. Shaw, D. R. Cameron, E. C. Underwood and G. C. Daily (2006). "Conservation planning for ecosystem services." PLoS Biology 4(11): e379. Choy, S. L., R. O'Leary and K. Mengersen (2009). "Elicitation by design in ecology: using expert opinion to inform priors for Bayesian statistical models." Ecology 90(1): 265-277. Christensen, V. and C. J. Walters (2004). "Ecopath with Ecosim: methods, capabilities and limitations." Ecological Modelling 172(2): 109-139. Cisneros-Montemayor, A. M. and U. R. Sumaila (2010). "A global estimate of benefits from ecosystem-based marine recreation: potential impacts and implications for management." Journal of Bioeconomics 12(3): 245-268. Clark, D., E. Goodwin, J. Sinner, J. Ellis and G. Singh (2016). "Validation and limitations of a cumulative impact model for an estuary." Ocean & Coastal Management 120: 88-98. Clarke Murray, C., M. E. Mach, R. G. Martone, G. G. Singh, M. O and K. M. A. Chan (2016). "Supporting Risk Assessment: Accounting for indirect risk to ecosystem components." PLoS ONE 11(9): e0162932. Clarke Murray, C., M. E. Mach and M. O (2016). Pilot ecosystem risk assessment to assess cumulative risk to species in the Pacific North Coast Integrated Management Area (PNCIMA). DFO, DFO Can. Sci. Advis. Sec. Res. Doc.: vii + 61 p. 190  Coen, L. D., R. D. Brumbaugh, D. Bushek, R. Grizzle, M. W. Luckenbach, M. H. Posey, S. P. Powers and S. Tolley (2007). "Ecosystem services related to oyster restoration." Marine Ecology Progress Series 341: 303-307. Collins, S. L., S. R. Carpenter, et al. (2010). "An integrated conceptual framework for long-term social-ecological research." Frontiers in Ecology and the Environment 9(6): 351-357. Cook, G. S., P. J. Fletcher and C. R. Kelble (2014a). "Towards marine ecosystem based management in South Florida: investigating the connections among ecosystem pressures, states, and services in a complex coastal system." Ecological Indicators 44: 26-39. Cook, G. S., P. J. Fletcher and C. R. Kelble (2014b). "Towards marine ecosystem based management in South Florida: Investigating the connections among ecosystem pressures, states, and services in a complex coastal system." Ecological Indicators 44(0): 26-39. Cook, S. E., K. W. Conway and B. Burd (2008). "Status of the glass sponge reefs in the Georgia Basin." Marine Environmental Research 66: S80-S86. Cooke, R. M. (1991). Experts in uncertainty: opinion and subjective probability in science. Oxford University Press. 336 pp. Costanza, R., R. d'Arge, et al. (1997). "The value of the world's ecosystem services and natural capital." Nature 387: 253-260. Costanza, R., R. de Groot, P. Sutton, S. van der Ploeg, S. J. Anderson, I. Kubiszewski, S. Farber and R. K. Turner (2014). "Changes in the global value of ecosystem services." Global Environmental Change 26: 152-158. Crain, C. M., B. S. Halpern, M. W. Beck and C. V. Kappel (2009). "Understanding and managing human threats to the coastal marine environment." Annals of the New York Academy of Sciences 1162(1): 39-62. 191  Crain, C. M., K. Kroeker and B. S. Halpern (2008). Interactive and cumulative effects of multiple human stressors in marine systems, Wiley-Blackwell. 11: 1304-1315. Crutzen, P. J. (2006). "The 'anthropocene'”. Earth System Science in the Anthropocene, Springer: 13-18. Curtice, C., D. C. Dunn, J. J. Roberts, S. D. Carr and P. N. Halpin (2012). "Why ecosystem-based management may fail without changes to tool development and financing." BioScience 62(5): 508-515. Curtin, R. and R. Prellezo (2010). "Understanding marine ecosystem based management: A literature review." Marine Policy 34(5): 821-830. Darling, E. S. and I. M. Côté (2008). "Quantifying the evidence for ecological synergies." Ecology Letters 11(12): 1278-1286. Darling, E. S., T. R. McClanahan and I. M. Côté (2010). "Combined effects of two stressors on Kenyan coral reefs are additive or antagonistic, not synergistic." Conservation Letters 3(2): 122-130. Dawson, M. (2005). Selling British Columbia: tourism and consumer culture, 1890-1970, UBC Press. 288 pp. De Groot, R. S., M. A. Wilson and R. M. Boumans (2002). "A typology for the classification, description and valuation of ecosystem functions, goods and services." Ecological Economics 41(3): 393-408. Demchak, J., J. Skousen, G. Bryant, P. Ziemkiewicz and Icard (2000). Comparison of water quality from fifteen underground coal mines in 1968 and 1999. Littleton, Soc Mining Metallurgy and Exploration Inc. 192  Demchak, J., J. Skousen and L. M. McDonald (2004). "Longevity of acid discharges from underground mines located above the regional water table." Journal of Environmental Quality 33(2): 656-668. 24 pp. DFO (2015). Application of an ecological risk assessment framework to inform ecosystem-based management for SGaan Kinghlas-Bowie Seamount and Endeavour Hydrothermal Vents Marine Protected Areas, DFO Can. Sci. Advis. Sec. Sci. Advis. 15 pp. Díaz, S., F. Quétier, D. M. Cáceres, S. F. Trainor, N. Pérez-Harguindeguy, M. S. Bret-Harte, B. Finegan, M. Peña-Claros and L. Poorter (2011). "Linking functional diversity and social actor strategies in a framework for interdisciplinary analysis of nature's benefits to society." Proceedings of the National Academy of Sciences 108(3): 895-902. Doney, S. C. (2010). "The growing human footprint on coastal and open-ocean biogeochemistry." Science 328(5985): 1512-1516. Duffy, J. E. (2008). "Why biodiversity is important to the functioning of real-world ecosystems." Frontiers in Ecology and the Environment 7(8): 437-444. Duinker, P. N., E. L. Burbidge, S. R. Boardley and L. A. Greig (2012). "Scientific Dimensions of Cumulative Effects Assessment: Toward Improvements in Guidance for Practice." Environmental Reviews 21(1): 40-52. Duinker, P. N. and L. A. Greig (2006). "The impotence of cumulative effects assessment in Canada: ailments and ideas for redeployment." Environmental Management 37(2): 153-161. Ehrlich, P. R. and H. A. Mooney (1983). "Extinction, substitution, and ecosystem services." BioScience 33(4): 248-254. 193  Elmqvist, T., C. Folke, M. Nyström, G. Peterson, J. Bengtsson, B. Walker and J. Norberg (2003). "Response diversity, ecosystem change, and resilience." Frontiers in Ecology and the Environment 1(9): 488-494. Eriksson, B. K., A. Sandström, M. Isæus, H. Schreiber and P. Karås (2004). "Effects of boating activities on aquatic vegetation in the Stockholm archipelago, Baltic Sea." Estuarine, Coastal and Shelf Science 61(2): 339-349. Estes, J. A., J. Terborgh, et al. (2011). "Trophic downgrading of planet earth." Science 333(6040): 301-306. Evans, M. C., H. P. Possingham and K. A. Wilson (2011). "What to do in the face of multiple threats? Incorporating dependencies within a return on investment framework for conservation." Diversity & Distributions 17(3): 437-450. Fagan, W. F. and E. Holmes (2006). "Quantifying the extinction vortex." Ecology Letters 9(1): 51-60. Fischer, T. B. (2010). The Theory and Practice of Strategic Environmental Assessment: Towards a More Systematic Approach, Routledge. 208 pp. Fischhoff, B. and G. Morgan (2013). "The science and practice of risk ranking." Risk Analysis and Human Behavior 26: 379. Fischhoff, B., P. Slovic and S. Lichtenstein (1982). "Lay foibles and expert fables in judgments about risk." The American Statistician 36(3b): 240-255. Fish, R., M. Winter, D. M. Oliver, D. Chadwick, T. Selfa, A. L. Heathwaite and C. Hodgson (2009). "Unruly pathogens: eliciting values for environmental risk in the context of heterogeneous expert knowledge." Environmental Science & Policy 12(3): 281-296. 194  Fisher, B. and R. Naidoo (2011). "Concerns about extrapolating right off the bat." Science 333(6040): 287-288. Fisher, R. (1980). "The impact of European settlement on the indigenous peoples of Australia, New Zealand, and British Columbia: Some comparative dimensions." Canadian Ethnic Studies= Etudes Ethniques au Canada 12(1): 1. Fishkin, J. (2009). When the people speak: Deliberative democracy and public consultation, OUP Oxford. 256 pp. Foley, M. M., B. S. Halpern, et al. (2010). "Guiding ecological principles for marine spatial planning." Marine Policy 34(5): 955-966. Folt, C., C. Chen, M. Moore and J. Burnaford (1999). "Synergism and antagonism among multiple stressors." Limnology and Oceanography 44(3): 864-877. Ford, J. K., G. M. Ellis and K. C. Balcomb (2000). Killer whales: the natural history and genealogy of Orcinus orca in British Columbia and Washington, UBC Press. 104 pp. Fulton, E. A., A. D. Smith, D. C. Smith and P. Johnson (2014). "An integrated approach is needed for ecosystem based fisheries management: insights from ecosystem-level management strategy evaluation." PLoS ONE 9(1): e84242. Gale, P., A. Brouwer, V. Ramnial, L. Kelly, R. Kosmider, A. Fooks and E. Snary (2010). "Assessing the impact of climate change on vector-borne viruses in the EU through the elicitation of expert opinion." Epidemiology and Infection 138(02): 214-225. Game, E. T., P. Kareiva and H. P. Possingham (2013). "Six Common Mistakes in Conservation Priority Setting." Conservation Biology 27(3): 480-485. GeoBC. (2016). "GeoBC."   Retrieved 2016, from http://geobc.gov.bc.ca/. 195  Gerlotto, F. and P. Fréon (1992). "Some elements on vertical avoidance of fish schools to a vessel during acoustic surveys." Fisheries Research 14(4): 251-259. Gilovich, T., D. Griffin and D. Kahneman (2002). Heuristics and Biases: The Psychology of Intuitive Judgment, Cambridge University Press. 882 pp. Granek, E. F., S. Polasky, et al. (2009). "Ecosystem Services as a Common Language for Coastal Ecosystem‐Based Management." Conservation Biology 24(1): 207-216. Grech, A., R. Coles and H. Marsh (2011). "A broad-scale assessment of the risk to coastal seagrasses from cumulative threats." Marine Policy 35(5): 560-567. Gregory, R., L. Failing, M. Harstone, G. Long, T. McDaniels and D. Ohlson (2012). Structured Decision Making: A Practical Guide to Environmental Management Choices, John Wiley & Sons. 312 pp. Gregr, E. J. and K. M. Chan (2014). "Leaps of faith: how implicit assumptions compromise the utility of ecosystem models for decision-making." BioScience 65(1): 43-54. Grimonprez, Q. and J. Jacques (2014). "Rankcluster: Model-based clustering for multivariate partial ranking data." Journal of Statistical Planning and Inference 149: 201-217. Groffman, P. M., J. S. Baron, et al. (2006). "Ecological Thresholds: The Key to Successful Environmental Management or an Important Concept with No Practical Application?" Ecosystems 9(1): 1-13. Gross, H. J. (2012). "The magical number four: A biological, historical and mythological enigma." Communicative & Integrative Biology 5(1): 1-2. Guerry, A. D. (2005). "Icarus and Daedalus: conceptual and tactical lessons for marine ecosystem-based management." Frontiers in Ecology and the Environment 3(4): 202-211. 196  Guerry, A. D., M. H. Ruckelshaus, et al. (2012). "Modeling benefits from nature: using ecosystem services to inform coastal and marine spatial planning." International Journal of Biodiversity Science, Ecosystem Services & Management 8(1-2): 107-121. Haines-Young, R. and M. Potschin (2010). "The Links Between Biodiversity, Ecosystem Services and Human Well-Being." Ecosystem Ecology: A New Synthesis: 110-139. eds. David G. Raffaelli and Christopher L. J. Frid. Cambridge University Press. Halpern, B. S., J. Diamond, et al. (2012a). "Near-term priorities for the science, policy and practice of Coastal and Marine Spatial Planning (CMSP)." Marine Policy 36(1): 198-205. Halpern, B. S. and R. Fujita (2013). "Assumptions, challenges, and future directions in cumulative impact analysis." Ecosphere 4(10): art131. Halpern, B. S., C. V. Kappel, et al. (2009). "Mapping cumulative human impacts to California Current marine ecosystems." Conservation Letters 2(3): 138-148. Halpern, B. S., S. E. Lester and K. L. McLeod (2010). "Placing marine protected areas onto the ecosystem-based management seascape." Proceedings of the National Academy of Sciences 107(43): 18312-18317. Halpern, B. S., C. Longo, et al. (2012b). "An index to assess the health and benefits of the global ocean." Nature 488(7413): 615-620. Halpern, B. S., C. Longo, et al. (2012c). "An index to assess the health and benefits of the global ocean." Nature 488(7413): 615-620. Halpern, B. S., K. L. McLeod, A. A. Rosenberg and L. B. Crowder (2008a). "Managing for cumulative impacts in ecosystem-based management through ocean zoning." Ocean & Coastal Management 51(3): 203-211. 197  Halpern, B. S., S. Walbridge, et al. (2008b). "A Global Map of Human Impact on Marine Ecosystems." Science 319(5865): 948-952. Handley, S. (2006). An analysis of historical impacts and composition of the benthic environment of Tasman and Golden Bays. Nelson, National Institute of Water and Atmospheric Research Ltd: 28 pp. Harley, C. D. G., A. Randall Hughes, K. M. Hultgren, B. G. Miner, C. J. B. Sorte, C. S. Thornber, L. F. Rodriguez, L. Tomanek and S. L. Williams (2006). "The impacts of climate change in coastal marine systems." Ecology Letters 9(2): 228-241. Hart, P. t. (1991). "Irving L. Janis' victims of groupthink." Political Psychology: 247-278. Heinanen, S., M. Mannerla, M. Vinther, J. H. Andersen and A. Stock (2013). Human uses, pressures and impacts in the eastern North Sea. DCE – Danish Centre for Environment and Energy. 136 pp. Technical Report from DCE – Danish Centre for Environment and Energy No. 18. Højsgaard, S. (2012). "Bayesian networks in R with the gRain package." Journal of Statistical Software 46(10): 1-26. Hollick, M. (1981). "Enforcement of mitigation measures resulting from environmental impact assessment." Environmental Management 5(6): 507-513. Hollick, M. (1984). "Who should prepare environmental impact assessments?" Environmental Management 8(3): 191-196. Hose, J. E., M. D. McGurk, G. D. Marty, D. E. Hinton, E. D. Brown and T. T. Baker (1996). "Sublethal effects of the (Exxon Valdez) oil spill on herring embryos and larvae: morphological, cytogenetic, and histopathological assessments, 1989 1991." Canadian Journal of Fisheries and Aquatic Sciences 53(10): 2355-2365. 198  Hughes, R. (1998). Environmental impact assessment and stakeholder involvement, IIED London, UK. 11 pp. Huwaldt, J. A. (2014). Plot Digitizer [Computer software]. Retrieved from www.sourceforge.net Jackson, J. B., M. X. Kirby, et al. (2001). "Historical overfishing and the recent collapse of coastal ecosystems." Science 293(5530): 629-637. Jacques, J., Q. Grimonprez and C. Biernacki (2014). "Rankcluster: An R package for clustering multivariate partial rankings." The R Journal 6(1): 101-110. Jensen, F. V. (1996). An Introduction to Bayesian Networks, UCL Press London. 186 pp. João, E. (2002). "How scale affects environmental impact assessment." Environmental Impact Assessment Review 22(4): 289-310. Kandlikar, M., G. Ramachandran, A. Maynard, B. Murdock and W. A. Toscano (2007). "Health risk assessment for nanoparticles: A case for using expert judgment." Nanotechnology and Occupational Health 9(1): 137-156. Kandlikar, M., J. Risbey and S. Dessai (2005). "Representing and communicating deep uncertainty in climate-change assessments." Comptes Rendus Geoscience 337(4): 443-455. Kanno, R. (2015). Herring Integrated Fisheries Mangeement Plan, DFO. 151 pp.  Kappel, C. V., B. S. Halpern and N. Napoli (2012). "Mapping cumulative impacts of human activities on marine ecosystems." Sea Plan, Boston, Massachusetts. 107 pp. Kareiva, P., H. Tallis, T. H. Ricketts, G. C. Daily and S. Polasky (2011). Natural Capital: Theory and Practice of Mapping Ecosystem Services: Theory and Practice of Mapping Ecosystem Services, OUP Oxford. 392 pp. 199  Kasperson, R. E. (2006). "Rerouting the stakeholder express." Global Environmental Change 16(4): 320-322. Kaufman, L., B. Heneman, J. T. Barnes and R. Fujita (2004). "Transition from low to high data richness: an experiment in ecosystem-based fishery management from California." Bulletin of Marine Science 74(3): 693-708. Kelble, C. R., D. K. Loomis, S. Lovelace, W. K. Nuttle, P. B. Ortner, P. Fletcher, G. S. Cook, J. J. Lorenz and J. N. Boyer (2013). "The EBM-DPSER Conceptual Model: Integrating Ecosystem Services into the DPSIR Framework." PLoS ONE 8(8): e70766. Kennish, M. J. (2002). "Environmental threats and environmental future of estuaries." Environmental Conservation 29(01): 78-107. Khaska, M., C. L. G. La Salle, P. Verdoux and R. Boutin (2015). "Tracking natural and anthropogenic origins of dissolved arsenic during surface and groundwater interaction in a post-closure mining context: Isotopic constraints." Journal of Contaminant Hydrology 177: 122-135. Killingsworth, M. J. and J. S. Palmer (2012). Ecospeak: Rhetoric and environmental politics in America, SIU Press. 328 pp. Klain, S. C. and K. M. A. Chan (2012). "Navigating coastal values: Participatory mapping of ecosystem services for spatial planning." Ecological Economics 82(0): 104-113. Knights, A. M., R. S. Koss and L. A. Robinson (2013a). "Identifying common pressure pathways from a complex network of human activities to support ecosystem-based management." Ecological Applications 23(4): 755-765. 200  Knights, A. M., R. S. Koss and L. A. Robinson (2013b). "Identifying common pressure pathways from a complex network of human activities to support ecosystem‐based management." Ecological Applications 23(4): 755-765. Knights, A. M., G. J. Piet, et al. (2015). "An exposure-effect approach for evaluating ecosystem-wide risks from human activities." ICES Journal of Marine Science: Journal du Conseil: fsu245. Knowlton, N. and J. B. C. Jackson (2008). "Shifting Baselines, Local Impacts, and Global Change on Coral Reefs." PLoS Biology 6(2): e54. Korpinen, S., L. Meski, J. H. Andersen and M. Laamanen (2012). "Human pressures and their potential impact on the Baltic Sea ecosystem." Ecological Indicators 15(1): 105-114. Kristensen, P. (2004). "The DPSIR framework." National Environmental Research Institute, Denmark. 10 pp. Kroeker, K. J., R. L. Kordas, R. Crim, I. E. Hendriks, L. Ramajo, G. S. Singh, C. M. Duarte and J.-P. Gattuso (2013). "Impacts of ocean acidification on marine organisms: quantifying sensitivities and interaction with warming." Global Change Biology 19(6): 1884-1896. Kroeker, K. J., R. L. Kordas, R. N. Crim and G. G. Singh (2010). "Meta‐analysis reveals negative yet variable effects of ocean acidification on marine organisms." Ecology Letters 13(11): 1419-1434. Kynn, M. (2007). "The ‘heuristics and biases’ bias in expert elicitation." Journal of the Royal Statistical Society: Series A (Statistics in Society) 171(1): 239-264. Lambert, D. C., D. A. Dzombak and W. W. Aljoe (2000). Evaluation of Natural Amelioration of Acidic Deep Mine Discharges in the Uniontown Syncline, Pennsylvania. Proc. of the 21st West Virginia Surface Mine Drainage Task Force Symp., Morgantown, WV. 201  Lawrence, D. P. (2007). "Impact significance determination—back to basics." Environmental Impact Assessment Review 27(8): 755-769. Lenzen, M., S. A. Murray, B. Korte and C. J. Dey (2003). "Environmental impact assessment including indirect effects—a case study using input–output analysis." Environmental Impact Assessment Review 23(3): 263-282. Leslie, H. M. and K. L. McLeod (2007). "Confronting the challenges of implementing marine ecosystem-based management." Frontiers in Ecology and the Environment 5(10): 540-548. Leu, M., S. E. Hanser and S. T. Knick (2008). "The human footprint in the west: a large-scale analysis of anthropogenic impacts." Ecological Applications 18(5): 1119-1139. Lindeman, R. L. (1942). "The trophic-dynamic aspect of ecology." Ecology 23(4): 399-417. Linkov, I. and F. K. Satterstrom (2006). "Weight of evidence: What is the state of the science?" Risk Analysis 26(3): 573. Long, J. A. and T. A. Nelson (2012). "Time geography and wildlife home range delineation." The Journal of Wildlife Management 76(2): 407-413. Loomis, J., P. Kent, L. Strange, K. Fausch and A. Covich (2000). "Measuring the total economic value of restoring ecosystem services in an impaired river basin: results from a contingent valuation survey." Ecological Economics 33(1): 103-117. Lupton, M. K., C. Rojas, P. Drohan and M. A. Bruns (2013). "Vegetation and soil development in compost‐amended iron oxide precipitates at a 50‐year‐old acid mine drainage barrens." Restoration Ecology 21(3): 320-328. MA (2005). Ecosystems and Human Well-Being, Island press Washington, DC. 155 pp. 202  MacDonald, L. H. (2000). "Evaluating and managing cumulative effects: process and constraints." Environmental Management 26(3): 299-315. Mach, M. E., R. G. Martone and K. M. Chan (2015). "Human impacts and ecosystem services: Insufficient research for trade-off evaluation." Ecosystem Services 16: 112-120. Marcot, B. G., R. S. Holthausen, M. G. Raphael, M. M. Rowland and M. J. Wisdom (2001). "Using Bayesian belief networks to evaluate fish and wildlife population viability under land management alternatives from an environmental impact statement." Forest Ecology and Management 153(1): 29-42. Margules, C. R. and R. L. Pressey (2000). "Systematic conservation planning." Nature 405(6783): 243-253. Martin, T. G., M. A. Bugman, F. Fidler, P. M. Kuhnert, S. Low-Choy, M. McBride and K. Mendersen (2012). "Eliciting expert knowledge in conservation science." Conservation Biology 26(1): 29-38. McBride, M. F. and M. A. Burgman (2012). "What is expert knowledge, how is such knowledge gathered, and how do we use it to address questions in landscape ecology?" Expert Knowledge and its Application in Landscape Ecology: 11-38. eds. Ajith H. Perera, C. Ashton Drew, and Chris J. Johnson. Springer. McBride, M. F., S. T. Garnett, et al. (2012). "Structured elicitation of expert judgments for threatened species assessment: a case study on a continental scale using email." Methods in Ecology and Evolution 3(5): 906-920. McCann, R. K., B. G. Marcot and R. Ellis (2006). "Bayesian belief networks: applications in ecology and natural resource management." Canadian Journal of Forest Research 36(12): 3053-3062. 203  McLeod, K. and H. Leslie (2009). Ecosystem-based management for the oceans, Cambridge Univ Press. 392 pp. Micheli, F., B. S. Halpern, S. Walbridge, S. Ciriaco, F. Ferretti, S. Fraschetti, R. Lewison, L. Nykjaer and A. A. Rosenberg (2013). "Cumulative human impacts on mediterranean and black sea marine ecosystems: assessing current pressures and opportunities." PLoS ONE 8(12): e79889. Miller, G. A. (1956). "The magical number seven, plus or minus two: some limits on our capacity for processing information." Psychological Review 63(2): 81-97. Moncur, M. C., C. J. Ptacek, D. W. Blowes and J. L. Jambor (2006). "Spatial variations in water composition at a northern Canadian lake impacted by mine drainage." Applied Geochemistry 21(10): 1799-1817. Moore, D. A. and G. Loewenstein (2004). "Self-interest, automaticity, and the psychology of conflict of interest." Social Justice Research 17(2): 189-202. Moore, D. A., L. Tanlu and M. H. Bazerman (2010). "Conflict of interest and the intrusion of bias." Judgment and Decision Making 5(1): 37-53. Morgan, M. G. (2014). "Use (and abuse) of expert elicitation in support of decision making for public policy." Proceedings of the National Academy of Sciences 111(20): 7176-7184. Morgan, M. G., L. F. Pitelka and E. Shevliakova (2001). "Elicitation of expert judgments of climate change impacts on forest ecosystems." Climatic Change 49(3): 279-307. Murray, C. C., S. Agbayani, H. M. Alidina and N. C. Ban (2015a). "Advancing marine cumulative effects mapping: An update in Canada’s Pacific waters." Marine Policy 58: 71-77. 204  Murray, C. C., S. Agbayani and N. C. Ban (2015b). "Cumulative effects of planned industrial development and climate change on marine ecosystems." Global Ecology and Conservation 4: 110-116. Murray, J. V., A. W. Goldizen, R. A. O’Leary, C. A. McAlpine, H. P. Possingham and S. L. Choy (2009). "How useful is expert opinion for predicting the distribution of a species within and beyond the region of expertise? A case study using brush‐tailed rock‐wallabies Petrogale penicillata." Journal of Applied Ecology 46(4): 842-851. Naidoo, R. and T. H. Ricketts (2006). "Mapping the economic costs and benefits of conservation." PLoS Biology 4(11): e360. Nelson, E., G. Mendoza, et al. (2009). "Modeling multiple ecosystem services, biodiversity conservation, commodity production, and tradeoffs at landscape scales." Frontiers in Ecology and the Environment 7(1): 4-11. NEPA (2007). Guidelines for conducting environmental impacts assessment, National Environment and Planning Agency. 60 pp. Neslo, R. E. J. (2011). Discrete Decisions with Model Validation using Probabilistic Inversion, TU Delft, Delft University of Technology. 169 pp. Niemeijer, D. and R. S. de Groot (2008). "Framing environmental indicators: moving from causal chains to causal networks." Environment, Development and Sustainability 10(1): 89-106. O'Faircheallaigh, C. (2010). "Public participation and environmental impact assessment: Purposes, implications, and lessons for public policy making." Environmental Impact Assessment Review 30(1): 19-27. 205  O'Hagan, A., C. E. Buck, A. Daneshkhah, J. R. Eiser, P. H. Garthwaite, D. J. Jenkinson, J. E. Oakley and T. Rakow (2006). Uncertain judgements: eliciting experts' probabilities, John Wiley & Sons. 338 pp. O, M., R. Martone, L. Hannah, L. Greig, J. Boutillier and S. Patton (2015). An Ecological Risk Assessment Framework (ERAF) for Ecosystem-based Oceans Management in the Pacific Region. D. C. S. A. S. R. D. 2014/072.vii: 59. Ogutu-Ohwayo, R., R. E. Hecky, A. S. Cohen and L. Kaufman (1997). "Human impacts on the African great lakes." Environmental Biology of Fishes 50(2): 117-131. Orth, R. J., K. L. Heck and J. van Montfrans (1984). "Faunal communities in seagrass beds: a review of the influence of plant structure and prey characteristics on predator-prey relationships." Estuaries 7(4): 339-350. Ott, K., F. Mohaupt and R. Ziegler (2012). Environmental Impact Assessment. Encyclopedia of Applied Ethics (Second Edition). C. Editor-in-Chief: Ruth. San Diego, Academic Press: 114-123. Parmesan, C. and G. Yohe (2003). "A globally coherent fingerprint of climate change impacts across natural systems." Nature 421(6918): 37-42. Pauly, D. (1995). "Anecdotes and the shifting baseline syndrome of fisheries." Trends in Ecology & Evolution 10(10): 430. Peterson, C. H., S. D. Rice, J. W. Short, D. Esler, J. L. Bodkin, B. E. Ballachey and D. B. Irons (2003). "Long-term ecosystem response to the Exxon Valdez oil spill." Science 302(5653): 2082-2086. 206  Pidgeon, N. F., W. Poortinga, G. Rowe, T. Horlick‐Jones, J. Walls and T. O'Riordan (2005). "Using surveys in public participation processes for risk decision making: The case of the 2003 British GM nation? Public debate." Risk Analysis 25(2): 467-479. Pikitch, E. K., C. Santora, et al. (2004). "Ecosystem-based fishery management." Science 305(5682): 346-347. Pinheiro, J., D. Bates, S. DebRoy and D. Sarkar (2007). Linear and Nonlinear Mixed Effects Models. R package version 3 - 57. http://CRAN.R-project.org/package=nlme Pinnegar, J. K. and G. H. Engelhard (2008). "The ‘shifting baseline’phenomenon: a global perspective." Reviews in Fish Biology and Fisheries 18(1): 1-16. Porter, S. R., M. E. Whitcomb and W. H. Weitzer (2004). "Multiple surveys of students and survey fatigue." New Directions for Institutional Research 2004(121): 63-73. Queiroz, C., M. Meacham, K. Richter, A. V. Norström, E. Andersson, J. Norberg and G. Peterson (2015). "Mapping bundles of ecosystem services reveals distinct types of multifunctionality within a Swedish landscape." Ambio 44(1): 89-101. Quigley, J. T. and D. J. Harper (2006). "Effectiveness of fish habitat compensation in Canada in achieving no net loss." Environmental Management 37(3): 351-366. Quintero, J. D. and A. Mathur (2011). "Biodiversity offsets and infrastructure." Conservation Biology 25(6): 1121-1123. R Core Team (2015). R: A Language and Environment for Statistical Computing. Vienna, Austria, R Foundation for Statistical Comuting. Raudsepp-Hearne, C., G. D. Peterson, M. Tengö, E. M. Bennett, T. Holland, K. Benessaiah, G. K. MacDonald and L. Pfeifer (2010). "Untangling the environmentalist's paradox: why is 207  human well-being increasing as ecosystem services degrade?" BioScience 60(8): 576-589. Raymond, C. M., G. G. Singh, et al. (2013). "Ecosystem services and beyond: Using multiple metaphors to understand human–environment relationships." BioScience 63(7): 536-546. Rees, W. E. (1998). "How should a parasite value its host?" Ecological Economics 25(1): 49-52. Regan, H. M., M. Colyvan and M. A. Burgman (2002). "A taxonomy and treatment of uncertainty for ecology and conservation biology." Ecological Applications 12(2): 618-628. Rocha, J., J. Yletyinen, R. Biggs, T. Blenckner and G. Peterson (2015a). "Marine regime shifts: drivers and impacts on ecosystems services." Phil. Trans. R. Soc. B 370(1659): 20130273. Rocha, J. C., G. D. Peterson and R. Biggs (2015b). "Regime shifts in the Anthropocene: drivers, risks, and resilience." PLoS ONE 10(8): e0134639. Rodrigues, A. S., H. R. Akcakaya, et al. (2004). "Global gap analysis: priority regions for expanding the global protected-area network." BioScience 54(12): 1092-1100. Rounsevell, M. D. A., T. Dawson and P. Harrison (2010). "A conceptual framework to assess the effects of environmental change on ecosystem services." Biodiversity and Conservation 19(10): 2823-2842. Sagoff, M. (2011). "The quantification and valuation of ecosystem services." Ecological Economics 70(3): 497-502. Sainte-Marie, B., I. C. Kaplan, P. S. Levin, M. Burden and E. A. Fulton (2010). "Fishing catch shares in the face of global change: a framework for integrating cumulative impacts and 208  single species management." Canadian Journal of Fisheries and Aquatic Sciences 67(12): 1968-1982. Salomon, A. K., N. M. Tanape Sr and H. P. Huntington (2007). "Serial depletion of marine invertebrates leads to the decline of a strongly interacting grazer." Ecological Applications 17(6): 1752-1770. Sanderson, E. W., M. Jaiteh, M. A. Levy, K. H. Redford, A. V. Wannebo and G. Woolmer (2002). "The Human Footprint and the Last of the Wild: the human footprint is a global map of human influence on the land surface, which suggests that human beings are stewards of nature, whether we like it or not." BioScience 52(10): 891-904. Schindler, D. E., R. Hilborn, B. Chasco, C. P. Boatright, T. P. Quinn, L. A. Rogers and M. S. Webster (2010). "Population diversity and the portfolio effect in an exploited species." Nature 465(7298): 609-612. Selkoe, K. A., B. S. Halpern, C. M. Ebert, E. C. Franklin, E. R. Selig, K. S. Casey, J. Bruno and R. J. Toonen (2009). "A map of human impacts to a “pristine” coral reef ecosystem, the Papahānaumokuākea Marine National Monument." Coral Reefs 28(3): 635-650. Sharp, R., H. Tallis, et al. (2014). "InVEST user’s guide." The Natural Capital Project, Stanford. 371 pp. Shepherd, A. and L. Ortolano (1996). "Strategic environmental assessment for sustainable urban development." Environmental Impact Assessment Review 16(4–6): 321-335. Shiue, W.-k., C.-w. Xu and B. C. Rea (1993). "Bootstrap confidence intervals for simulation outputs." Journal of Statistical Computation and Simulation 45(3-4): 249-255. 209  Simmonds, M. P., S. J. Dolman, M. Jasny, E. Parsons, L. Weilgart, A. J. Wright and R. Leaper (2014). "Marine noise pollution-increasing recognition but need for more practical action." Journal of Ocean Technology 9(1), 71-90. Skousen, J., L. McDonald, B. Mack and J. Demchak (2006). "Water quality from above-drainage underground mines over a 35-year period." ICARD: Lexington, American Society for Mining and Reclamation: 2044-2054. Slovic, P. (1987). "Perception of risk." Science 236(4799): 280-285. Slovic, P. (2011). "The Feeling of Risk: New Perspectives on Risk Perception." Energy & Environment 22(6): 835-836. Smith, J. L. and K. D. Hyrenbach (2003). "Galapagos Islands to British Columbia: seabird communities along a 9000 km transect from the tropical to the subarctic eastern Pacific Ocean." Marine Ornithology 31(2): 155-166. Smith, M. D., A. K. Knapp and S. L. Collins (2009). "A framework for assessing ecosystem dynamics in response to chronic resource alterations induced by global change." Ecology 90(12): 3279-3289. Speirs‐Bridge, A., F. Fidler, M. McBride, L. Flander, G. Cumming and M. Burgman (2009). "Reducing overconfidence in the interval judgments of experts." Risk Analysis 30(3): 512-523. Spyce, A., M. Weber and W. Adamowicz (2012). "Cumulative effects planning: finding the balance using choice experiments." Ecology and Society 17(1): 22. Stelzenmüller, V., J. Lee, A. South and S. I. Rogers (2010). "Quantifying cumulative impacts of human pressures on the marine environment: a geospatial modelling framework." Marine Ecology Progress Series 398: 19-32. 210  Stevenson, M. G. (1996). "Indigenous knowledge in environmental assessment." Arctic: 278-291. Stocker, T. F. (2014). Climate change 2013: the physical science basis: Working Group I contribution to the Fifth assessment report of the Intergovernmental Panel on Climate Change, Cambridge University Press. 1308 pp. Sunstein, C. R. and R. Hastie (2015). Wiser: Getting Beyond Groupthink to Make Groups Smarter, Harvard Business Press. 272 pp. Sutherland, W. and M. Burgman (2015). "Policy advice: Use experts wisely." Nature 526(7573): 317-318. Tallis, H., S. E. Lester, et al. (2012). "New metrics for managing and sustaining the ocean's bounty." Marine Policy 36(1): 303-306. Tallis, H., P. S. Levin, M. Ruckelshaus, S. E. Lester, K. L. McLeod, D. L. Fluharty and B. S. Halpern (2010). "The many faces of ecosystem-based management: Making the process work today in real places." Marine Policy 34(2): 340-348. Teck, S. J., B. S. Halpern, et al. (2010). "Using expert judgment to estimate marine ecosystem vulnerability in the California Current." Ecological Applications 20(5): 1402-1416. Tinker, L., D. Cobb, A. Bond and M. Cashmore (2005). "Impact mitigation in environmental impact assessment: paper promises or the basis of consent conditions?" Impact Assessment and Project Appraisal 23(4): 265-280. Trebilco, R., N. K. Dulvy, H. Stewart and A. K. Salomon (2015). "The role of habitat complexity in shaping the size structure of a temperate reef fish community." Marine Ecology Progress Series 532: 197-211. 211  Trosper, R. L. (2002). "Northwest coast indigenous institutions that supported resilience and sustainability." Ecological Economics 41(2): 329-344. Tummers, B. (2006). DataThief III. http://datathief.org/ Turner, N. J. and H. Clifton (2009). "“It's so different today”: Climate change and indigenous lifeways in British Columbia, Canada." Global Environmental Change 19(2): 180-190. Turner, N. J. T. N. and K. L. T. K. Turner (2008). "“Where our women used to get the food”: cumulative effects and loss of ethnobotanical knowledge and practice; case study from coastal British Columbia This paper was submitted for the Special Issue on Ethnobotany, inspired by the Ethnobotany Symposium organized by Alain Cuerrier, Montreal Botanical Garden, and held in Montreal at the 2006 annual meeting of the Canadian Botanical Association." Botany 86(2): 103-115. Tversky, A. and D. Kahneman (1974). "Judgment under Uncertainty: Heuristics and Biases." Science 185(4157): 1124-1131. Underwood, A. (1991). "Beyond BACI: experimental designs for detecting human environmental impacts on temporal variations in natural populations." Marine and Freshwater Research 42(5): 569-587. UNEP (2007). GEO 4 Global Environmental Outlook. Nairobi: 540 pp. UNEP (2012). GEO 5 Global Environment Outlook, Progress Press Ltd Malta: 528 pp. Uusitalo, L. (2007). "Advantages and challenges of Bayesian networks in environmental modelling." Ecological Modelling 203(3): 312-318. Ware, D. M. and R. E. Thomson (2005). "Bottom-up ecosystem trophic dynamics determine fish production in the Northeast Pacific." Science 308(5726): 1280-1284. 212  White, J. W., A. Rassweiler, J. F. Samhouri, A. C. Stier and C. White (2014). "Ecologists should not use statistical significance tests to interpret simulation model results." Oikos 123(4): 385-388. Wieland, R., S. Ravensbergen, E. J. Gregr, T. Satterfield and K. M. Chan (2016). "Debunking trickle-down ecosystem services: The fallacy of omnipotent, homogeneous beneficiaries." Ecological Economics 121: 175-180. Willemen, L., P. H. Verburg, L. Hein and M. E. van Mensvoort (2008). "Spatial characterization of landscape functions." Landscape and Urban Planning 88(1): 34-43. Wood, C. (2003). Environmental Impact Assessment: A Comparative Review, Pearson Education. 430 pp. Yardi, S. and D. Boyd (2010). "Dynamic debates: An analysis of group polarization over time on twitter." Bulletin of Science, Technology & Society 30(5): 316-327. Yasuda, Y., H. Kitagawa and T. Nakagawa (2000). "The earliest record of major anthropogenic deforestation in the Ghab Valley, northwest Syria: a palynological study." Quaternary International 73: 127-136. Zhou, S., A. D. Smith, et al. (2010). "Ecosystem-based fisheries management requires a change to the selective fishing philosophy." Proceedings of the National Academy of Sciences 107(21): 9485-9489.    213  Appendices Appendix A  Supplementary Material for Chapter 2 A.1 Supplementary Tables Table A.1 Composition of the EISs evaluated for each jurisdiction by types of developments considered. Jurisdiction Percentage of EISs Agriculture Energy Fishing Manufacturing Marine Construction Mining Oil and Gas Tourism Transportation Water Management British Columbia 0 30 0 0 0 30 40 0 0 0 California 0 40 0 0 20 0 20 10 10 0 Veracruz 0 20 0 0 10 0 30 0 20 20 Brazil 0 40 0 0 10 20 10 0 20 0 England and Wales 0 90 0 0 0 0 0 0 10 0 Queensland 9 0 0 9 0 55 9 0 18 0 New Zealand 0 14 14 0 0 0 0 0 72 0        214  Table A.2 List of species selected and associated area measurements (in km2).  Jurisdiction species of concern range (km2) Area Measurement Resource British Columbia grizzly bear 2980000 area of occupancy BC Species Explorer1 British Columbia mountain goat 20000 area of occupancy BC Species Explorer1 British Columbia silverhaired bat 2500000 range extent BC Species Explorer1  British Columbia western toad 20000 area of occupancy BC Species Explorer1 British Columbia steelhead trout 50000 area of occupancy BC Species Explorer1 British Columbia northern caribou 20000 area of occupancy BC Species Explorer1 California desert tortoise 50000 area of occupancy Natureserve2 California Mohave ground squirrel 50000 area of occupancy Natureserve2 California California brown pelican 200000 extent of occurrence Natureserve2 California tidewater goby 500 area of occupancy Natureserve2 California arroyo toad 200000 extent of occurrence Natureserve2 California coastal California gnatcatcher 20000 extent of occurrence Natureserve2 Veracruz Swainson's hawk 988000 extent of occurrence IUCN3 Veracruz scissor tailed flycatcher 305000 extent of occurrence IUCN3 Veracruz red lored amazon 876000 extent of occurrence IUCN3 Veracruz northern crested caracara 3890000 extent of occurrence IUCN3 Veracruz Morelet's crocodile 396455 area of occupancy IUCN3 Veracruz reddish egret 349000 extent of occurrence IUCN3 Brazil Brazilian three banded armadillo 937000 extent of occurrence IUCN3 Brazil jaguar 8750000 extent of occurrence IUCN3 Brazil maned three toed sloth 1000 area of occupancy IUCN3 Brazil oncilla 11000000 extent of occurrence IUCN3 Brazil Brazilian teal 8740000 extent of occurrence IUCN3 Brazil curl crested jay 2900000 extent of occurrence IUCN3 England and Wales badger 180000 extent of occurrence Joint Conservation Committee4 England and Wales water vole 106000 extent of occurrence Joint Conservation Committee4 England and Wales sand lizard 8850 extent of occurrence Joint Conservation Committee4 England and Wales otter 130800 extent of occurrence Joint Conservation Committee4 England and Wales greater crested newt 157749 extent of occurrence Joint Conservation Committee4 England and Wales common pipistrelle 230249 extent of occurrence Joint Conservation Committee4 Queensland black throated finch 5000 area of occupancy Department of Environment5 Queensland red goshawk 200000 area of occupancy Department of Environment5 Queensland regent honeyeater 600000 extent of occurrence Department of Environment5 Queensland Squatter pigeon (southern distribution) 10000 area of occupancy Department of Environment5 Queensland large eared pied bat 9120 area of occupancy Department of Environment5 Queensland border thick tailed gecko 1000 area of occupancy Department of Environment5 New Zealand New Zealand pigeon 208000 extent of occurrence  IUCN3 New Zealand grey warbler 269000 extent of occurrence  IUCN3 New Zealand paradise shelduck 269000 extent of occurrence  IUCN3 New Zealand giant kokopu 348 area of occupancy IUCN3 New Zealand wrybill 7700 extent of occurrence IUCN3 New Zealand New Zealand bush falcon 149000 extent of occurrence IUCN3 1http://a100.gov.bc.ca/pub/eswp/ 2http://explorer.natureserve.org/ 3http://www.iucnredlist.org/ 4http://jncc.defra.gov.uk/ 5http://environment.gov.au/  215  Table A.3 Studies on longevity of Acid Mine Discharge (AMD) impacts Type of Mine Years after mine closure Zinc-copper (Moncur et al. 2006) 50 gold and arsenic (Khaska et al. 2015) 8 lead and zinc (Casiot et al. 2009) 45 Coal (Lupton et al. 2013) 62 Coal (Skousen et al. 2006) 50 Coal (Skousen et al. 2006) 65 Coal (Skousen et al. 2006) 65 Coal (Skousen et al. 2006) 60 Coal (Skousen et al. 2006) 55 Coal (Skousen et al. 2006) 55 Coal (Skousen et al. 2006) 53 Coal (Skousen et al. 2006) 55 Coal (Skousen et al. 2006) 55 Coal (Skousen et al. 2006) 65 Coal (Skousen et al. 2006) 55 Coal (Skousen et al. 2006) 52 Coal (Skousen et al. 2006) 70 Coal (Skousen et al. 2006) 70 Coal (Skousen et al. 2006) 55 Coal (Skousen et al. 2006) 58 Coal (Skousen et al. 2006) 65 Coal (Skousen et al. 2006) 70 Coal (Skousen et al. 2006) 53 Coal (Skousen et al. 2006) 57 Coal (Lambert et al. 2000) 40 Coal (Lambert et al. 2000) 65       216  Table A.4 A typology of consultation with stakeholders used in our study.  Typology of Participation (adapted from Hughes 1998) Example 1. Inform stakeholders Consultant or extension worker appears in village and tells villagers that an irrigation scheme will be constructed to "improve" crop yield. 2. Collect data from stakeholders Consultant or extension worker appears in village and asks for information about their crops, and about seasonal water flows. Records their answers and leaves. 3. Hear stakeholder views Consultant or extension worker explains that crop yields need to be improved, and that the government intends to build an irrigation scheme. They seek the views and responses of villagers (for example, how they feel it might increase soil erosion), and then leave. 4. Respond to stakeholder concerns Consultants or extension workers inform villagers that they intend to construct an irrigation project. The consultants then facilitate the development of a village committee to discuss particular aspects of the project (such as minimizing soil erosion, downstream impacts on fisheries; or to agree on arrangement for water management). 5. Stakeholders assess impacts Local villagers identify their own needs, and external facilitators work with them to assist in finding solutions to potential negative impacts and improving positive effects. In some cases, new institutions will develop at the local level, which might then play a role in the management of their own project and its impacts. Villagers then have a real stake in maintaining structures or practices. 6. Stakeholders self-mobilize Villagers plan and identify their own irrigation structures, perhaps learning from experience in a nearby village. They may develop contacts with external institutions for resources and technical advice they need, but retain control over how resources are used.       217  Appendix B   Supplementary Materials for Chapter 3 B.1 Ecosystem Service Models Aquaculture: While the InVEST aquaculture model projects biomass production over time of aquaculture facilities, the data needed to do so was not readily available. Given that only the location of aquaculture facilities was required for this project, the InVEST aquaculture model was unnecessary. Vector s showing the location of shellfish and finfish aquaculture facilities were obtained from GeoBC, and this was sufficient to map the ecosystem service.  Recreation: The InVEST Overlap analysis model was run using the ‘Grid the Seascape’ method, with a 1000 metre grid. The Inputs were kayaking and pleasure craft routes, dive sites, recreational fishing sites, coastal parks, and marinas. The raw data for recreational fishing sites distinguished between different catches, but was combined so as not to bias the output toward recreational fishing. The coastal campsites and marinas points were buffered by 150 and 2000 metres respectively.  Coastal Protection:  The Invest Coastal Vulnerability model was run with the following inputs: the shorezone file, described in Table B.1 below, with a rank assigned representing erosion risk dependent upon the coastal class of the shorezone data (ranks listed below), natural vegetation layers consisting of dunes, eelgrass, and kelp, a continental shelf layer provided by InVEST, the SRTM DEM from the USGS, and an average depth of 2000 metres. The output produces a raster called “Eros Index” that includes a vulnerability index as well as a ‘protection from vegetation’ score.  A coastline segment with no protection from vegetation scores a 5 in this category. Therefore, all cells with scores lower than 5 were extracted, and displayed according to their 218  vulnerability index. Additionally, the natural vegetation scores were determined by the NaturalHabitat.csv table in the vulnerability table. The original values offered by InVEST were used; a protection distance of 1500 m for kelp, 500 m for eelgrass, and 300 metres for dunes. The ‘protection rank’ of dunes, eelgrass, and kelp were 2, 4 and 4, respectively. It should be noted that all three natural vegetation files are incompletely and likely underestimate the extent of the habitats. This is especially true of the dunes layer, which is confined exclusively to the Salish sea.   Table B.1 Erosion risk to different coastal classes Shorezone Erosion Risk (1 is low, 5 high) Rank COASTAL_CLASS_NAME 3 Undefined 1 Rock Ramp, wide > 30m 1 Rock Platform, wide > 30m 1 Rock Cliff, narrow < 30m 1 Rock Ramp, narrow < 30m 1 Rock Platform, narrow < 30m 4 Rock Ramp with Gravel Beach, wide > 30m 4 Rock Platform with Gravel Beach, wide > 30m 3 Rock Cliff with Gravel Beach, narrow < 30m 4 Rock Ramp with Gravel Beach, narrow < 30m 4 Rock Platform with Gravel Beach, narrow < 30m 219  Shorezone Erosion Risk (1 is low, 5 high) Rank COASTAL_CLASS_NAME 5 Rock Ramp with Sand and Gravel Beach, wide > 30m 5 Rock Platform with Sand and Gravel Beach, wide > 3 4 Rock Cliff with Sand and Gravel Beach, narrow < 30 5 Rock Ramp with Sand and Gravel Beach, narrow < 30m 4 Rock Platform with Sand and Gravel Beach, narrow < 5 Rock Ramp with Sand Beach, wide > 30m 5 Rock Platform with Sand Beach, wide > 30m 5 Rock Platform with Sand Beach, wide > 30m 5 Rock Ramp with Sand Beach, narrow < 30m 5 Rock Platform with sand Beach, narrow < 30m 4 Gravel Flat, wide > 30m 4 Gravel Beach, narrow < 30m 4 Gravel Flat or Fan, narrow < 30m 5 Sand and Gravel Flat or Fan, wide > 30m 5 Sand and Gravel Beach, narrow < 30m 5 Sand and Gravel Flat or Fan, narrow < 30m 220  Shorezone Erosion Risk (1 is low, 5 high) Rank COASTAL_CLASS_NAME 5 Sand Beach, wide > 30m 5 Sand Flat, wide > 30m 5 Mud Flat, wide > 30m 5 Sand Beach, narrow < 30m 4 Estuary (Organics/Fines) 2 Man made, permeable 1 Man made impermeable 4 Channel 4 Hanging Lagoon  Aesthetic Quality:  The InVEST model functions by determining areas that are visually impacted and delineating the locations from which these areas can be seen. It then grids the area of interest and calculates the number of impacted sites visible for each cell in grid. However, this assumes areas with no impact are visually pleasing, and requires a subjective determination of what constitutes a visual impact. Rather than use this approach, we opted to adapt it by determining the visible extent from areas that are frequented by people, using the recreation inputs from the Overlap model as well as the InVEST provided population data. This assumes that an underlying driver of recreational activities is the view.  221  The InVEST tool requires point inputs, therefore the kayak, recreational boating, population, and recreational fishing layers had to be converted to points. Recreational fishing, kayak and pleasure craft boating routes were converted using the ‘Create random points’ tool.  For the fishing polygons, inputs were 3000 “randomly generated” points, spaced at a minimum of 750 metres apart. The tool works by adding up to the specified amount of points without crossing the minimum distance.  However, the model was unable to fit all 3000 points for any of the fishing polygons. This ensures that each point is the minimum 750 m distance from at least one other point, but does not guarantee an equal distance among all points. While imperfect, it does ensure that the viewshed completely includes the fishing polygon, and that the visibility of areas outside of the fishing polygon is determined at approximately 750 metre intervals around the perimeter.   The kayak and pleasure craft routes were converted to points with a minimum distance of 1000 metres.  The points along a single segment are all ~1000 m apart, and where lines overlap they are close as the minimum distance only applies to points within the same segment. The population data was converted to puts using the ‘Raster to point’ tool. This creates a point for every raster cell. However, only cells that were within a 5 km buffer of the Shorezone coastline were included. In addition, cells with a value of 1 or 0 were excluded (i.e. areas with population densities less than 2 people/25 ha). Camping and dive sites were also included.  The other inputs included the SRTM 7.5 arc-second DEM and a special AOI that was created specifically for this model. The AOI simply followed the contours of the viewshed of an earlier run (with a coarser DEM) to shorten the computing time. The output is a raster that models whether each 500 x 500 m cell visible from one of the input points.   222  Renewable Energy The Wave Energy model was not completed with InVEST. The BCMCA has publicly available spatial data on wave and tidal energy areas of interest in a 1x1 km grid format. The cells have been weighed by a panel of experts according to their relative importance in interest or promise of wave/tidal energy development. The original data was not clipped to the shoreline, as this would affect the precision by which they were drawn as well as their ultimate score (each expert drew polygons on maps at scales of 1:700,000 to 1:900,000). However, they were clipped regardless and scaled using the ArcGIS Slice tool, as it was determined that this would not significantly affect accuracy at the regional scale. This data was used because it incorporates political feasibility when determining relative importance, which would be challenging to account for with the InVEST model.  Commercial Fisheries The commercial fisheries model was created in the same manner as the recreational model, using the gridded seascape. The inputs consist of various commercial fisheries spatial files provided by the Fisheries and Oceans Canada (DFO), the Province of BC, and Parks Canada. They were separated based on demersal or pelagic fisheries. Though each grid cell is 1 by 1 km, the actual resolution is coarser, at 4-10 km grids depending upon the fishery.  B.2 Human Impacts Spatial extent of human impacts was modeled using data from the Ban et al. (2010). However, to keep impact and ecosystem service data consistent, the recreational fishery data was not used, as we had already obtained our own for the recreation ecosystem service model. We also acquired 223  our own shipping data from BCMCA that gave a more complete province-wide map of shipping activity on a 200x200m cell grid. This data counted the number of types of ship in a cell (the actual data on shipping intensity per cell was not publicly available though a map of this data was), but we found this to be a suitable proxy of shipping intensity as the resulting map largely mirrored the shipping intensity map.  Commercial fisheries were separated based on Ban et al. (2010) classification of demersal destructive, demersal non-destructive low by-catch, pelagic high by-catch, and pelagic low by-catch. These were then combined into 4 rasters. Regional climate impact data from Halpern et al (2008) consisting of 1 by 1 km cells were used to model aragonite saturation state, sea surface temperature change, and UVb change. The other impacts were provided in vector format and converted to 500m rasters. A description of data files is shown in Table B.2. Table B.2 Data files, sources, and resolution used to map impacts and ecosystem services Data Description Resolution Campsites_coastal coastal British Columbia campsites 1: 40,000 Can_EEZ Exclusive economic zone of Canada 1: 1, 000 000 Coastal_parks_WGS Parks in Coastal British Columbia not listed coastal_protection shoreline protected by  vegetation  224  Data Description Resolution coastal_protection_veg eelgrass and kelp within 500/1500 m of shore  Continental_shelf* continental shelf off of BC coast  cs_protect finished ES data  Divesites_BCMCA dive sites  1 :40,000 Dunes_3  used in protection ES model. Limited extent 1: 20,000 Eelgrass_1 eelgrass input for coastal protection 1: 40 000 EEZ_Bathymetry bathymetry raster. Low res at greater depths 100 m Fetch_cmb produced with InVEST fetch tool  Finfish_aquaculture finfish aqua. and associated structures not listed Kayak_BCMCA kayak routes 1: 40,000 Kayak_points kayak routes in point form  225  Data Description Resolution Kelp_2 kelp input for coastal protection 1: 40,000 Land_BC* land from global land polygon  Marinas point file of marinas 1: 40,000 Population_raster* global population raster 2.5 arcminutes Rec_boating pleasure craft boating 1: 40,000 Recreational_Crab recreational crab fishing polygons 1: 70-120 000 RecreationaL_Finfish recreational finfish polygons 1: 70-120 000 Recreational_Groundfish recreational groundfish polygons 1: 70-120 000 Recreational_Shrimp recreational shrimp polygons 1: 70-120 000 Shellfish_aquaculture shellfish aqua. and associated structures 1: 40,000 Shorezone GeoBC shorezone database  1: 20,000 SRTM_DEM SRTM topography data 7.5 arc second Tidal_energy areas of interest for future tidal energy dev. 1: 700-900 000 226  Data Description Resolution Wave_energy areas of interest for future wave energy dev. 1: 700-900 000 WaveWatchIII* Wave watch III data 1 x 1.25 degree              227  Table B.3 Descriptions of human activities and stressors provided to experts to assess risk Activity/Stressor Type Description Fishing One fishing vessel of the specific fishing type indicated and its associated risks such as catch, bycatch and lost fishing gear Finfish Aquaculture One aquaculture facility and its associated effects such as disease and nutrient transmission Shellfish Aquaculture One aquaculture facility including its risks to intertidal habitats and risks of invasive species release Large Boat Traffic One large boat or ship and associated risks such as strikes/collisions, acoustic impacts, and illegal dumping of oil wastes and greywater Ports, Marinas and Harbours One marina/port/harbour, including the dredging required to create it and associated risks of contaminants and breakwaters Small Docks, Ramps, and Wharves One dock/ramp/wharf, and associated risks such as shading and contaminants Log Dumping, Handling and Storage One log-dumping site and associated risks such as scouring and leachates Ocean Dumping One dump event in a site designated for ocean dumping of nontoxic materials Industry One industrial building (e.g. factory) including associated risks such as pollutants Pulp and Paper One pulp mill and associated risks including toxic effluent and leaks Onshore Mining One mining pit and associated risks from discharge and drainage Human Settlements One human dwelling and its associated risks such as pollutants Agriculture and Forestry One farm and its associated risks including silt and pesticide runoff, or one forest cutblock and associated risks including sediment runoff Climate Change For each specific stressor, consider an event to cover the entire phenomenon of climate change and only the current resulting effects Potential Risk: Future Climate Change For each specific stressor, consider an event to cover the entire phenomenon of climate change with a 3 °C  in temperature and a decrease of 0.3 in ocean pH (corresponding to projections for 2100) Potential Risk: Oil Spill One large oil spill (>40 000 m3 of oil spilled, approximately the size of the Exxon Valdez oil spill in 1989)    228  Table B.4 Descriptions of exposure criteria given to experts to assess risk Dimension Description Area of Influence The spatial influence of a single event of an activity on the area where a service is provided, where a risk may be direct or indirect, measured in kilometers squared (km2)  This represents the impact of a single event of an activity, not the aggregate or cumulative presence of the activity across the seascape. For example, trawling, in total, may impact thousands of square kilometres, but a single trawling event may cover less than 1 to 10 km2. It is the latter in which we are interested. If onshore mining negatively impacts the view of an entire bay, then the spatial influence of onshore mining to aesthethic quality is the area of the entire bay.   Unit of Measurement: Kilometres squared (km2) Frequency The average annual frequency of individual events of an activity at a particular location, measured in days per year.  Frequency is not a measure of duration, but rather, how many times an activity occurs in an area in a given year. For example, if fishing groundfish occurs everywhere in a region, but on average only occurs at a given location three times a year, the frequency would be 3 days/yr. Duration will be captured by "Recovery Time". In cases were an activity is ongoing (i.e. there is no pinpoint "event") the frequency of that activity should be counted as every day - 365 days/yr. Fractions represent frequencies less than once per year (e.g. 0.1 days/yr represents once per decade).   Unit of Measurement: Days per year (days/yr) Recovery Time The average time required for the affected ES to return to its former level of provision, following disturbance by a given activity, measured in years.  Fractions represent times shorter than a single year. Recovery of the ES is related to the resilience of the system to a type of risk. It relates to recovery of a system within a site as well as the necessary species or environmental components that make up the biophysical producers/production of an ES from surrounding areas to recolonize a site. "Biophysical producers" indicates the specific species responsible for producing the ES under consideration.    Unit of Measurement: Years (yrs)  229  Table B.5 Descriptions of consequence criteria given to experts to assess risk Dimension Description Magnitude of Risk The degree to which an ES is at risk due to potential negative impacts of an activity on the biophysical producers/production of that ES, measured in percent (%).  For renewable energy, consider the risk to potential renewable energy generated by wind or waves; for fishery-related ES, consider the risk to fishery yield; for marine carbon sequestration, consider the change in the amount of carbon sequestered; for coastal protection, consider the change in erosion risk. Magnitude of risk to ES affected by an activity also addresses the diversity of producers responsible for the ES. If a risk affects a high proportion of ES producers, that ES is likely more vulnerable and less resistant to the risk.   Unit of Measurement: Percent (%)  Guidelines: 0% = No risk to ES producers or production 100% = Complete loss of ES producers or production Community Extent The extent of risk on the underlying ecological community responsible for producing an ES, measured in a 0-3 score.  Human activities can affect the primary biophysical producers of ES directly or indirectly through associated species and habitats. More extensive risks may also affect primary biophysical producers of ES and their associated species and habitats.  For example, shipping would score a 1 for marine recreation if considering that ship-strikes impact sought-after marine mammals. Fishing however, would score a 2 if marine mammal abundance and distribution are at risk indirectly through a decreased prey base. Larger phenomena such as climate change could rank 3 for altering coastal habitat and thereby affecting the abundance of prey species.   Unit of Measurement: Integer from 0 to 3  0 = An activity might occur on some part of the region, but does not affect the biophysical producers of the ES  1 = The activity affects the biophysical producers of the ES (directly or indirectly)  2 = The activity affects the biophysical producers and supporting species  230  Dimension Description 3 = The activity affects the biophysical producers, supporting species and surrounding habitat structure Access to Service The change in the ability and rights of people to access an area so that they may benefit from the ES in question, measured as percent (%)  For example, if log dumping and handling restricts access to subsistence fishing in an area by 50%, than this would be scored as 50. If it completely closes an area that people use to fish, than this would be scored as 100.   Unit of Measurement: Percent (%)  Guidelines: 0% = Positive or no effect on the accessibility of the ES 100% = Complete loss of access to the ES Quality of Service The change in the enjoyment or benefit that people derive from an ES given the same quantity of good or experience, measured as percent (%)  Assuming that there are no limitations in accessibility, evaluate how an activity would affect the enjoyment of that service. This is intended to capture the risk of human activities to the intangible benefits that people derive from ES. For landscape aesthetics and recreation, consider the change in the quality of the scenery or enjoyment. For fisheries and aquaculture, if agricultural runoff reduces the water quality of an area such that shellfish harvest poses a major health hazard, this would be scored as a 100.   Unit of Measurement: Percent (%)  Guidelines:  0% = Positive or no effect on the quality of service 100% = Complete loss of enjoyment or benefit provided by the ES   231  B.3 Supplementary Figures   Figure B.1 Side by side comparison of impact maps considering all risk criteria (maps on the left) versus only considering biophysical criteria of risk which only assesses impact to ecosystem service supply (maps on the right). Map pairs are for A) aesthetics, B) coastal protection, C) commercial demersal fisheries, D) commercial pelagic fisheries, E) coastal recreation, F) potential renewable energy, G) finfish aquaculture, and H) shellfish aquaculture. 232  Appendix C  Supplementary Methods for Chapter 4 C.1 Scoring exposure and consequence The data on risk scores for this project were based on a risk assessment conducted for coastal British Columbia. Below, we outside the prominent components of risk, their values, and calculating final risk scores from these criteria (with associated uncertainty). Temporal Scale (TS) refers to the frequency of the event, rather than its duration.  Consideration was given to how often the stressor occurs, rather than how long the effect is felt by the SEC (which in practice was included in the Consequenceij scoring). Scoring is described in Table C.1A. Spatial Scale (SS) is the scale or spatial extent of the impact from the stressor.  For example, under sedimentation from trawl fisheries, consideration was given to how far sediment is carried from the site of the trawl. Scoring for the dive fishery considered the size of the footprint of habitat disturbance from a single dive.  Scoring is described in Table C.1B. Intensity (I) is a measure of the density and persistence of the stressor. Depending on the stressor or activity in question, Intensity can refer effort, density, amount of an activity, or the amount or strength of a stressor (e.g. quantity or concentration of a pollutant or harmful species, rate of change for climate change) across the entire study area. For example, load for finfish aquaculture evaluates how many finfish farms are there in British Columbia and how often and how much area is covered by finfish farms. Scoring described in Table C.1C. 233  Consequence (C) is the impact of the stressor on the individual environmental component and therefore must be scored for each environmental component by each mechanistic impact pathway. It is scored from 1 to 6 ranging from negligible to intolerable consequence and indicates the impact of the stressor on the individual SEC, as described in Table C.1D.  Consequenceij scoring is based on the subcomponent (population size, geographic range, behaviour, etc) but most commonly Consequenceij was scored on the population size or geographic range subcomponent.  If information was available about more than a single subcomponent, the most sensitive subcomponent was used to assign the score. In choosing the most sensitive subcomponent, consideration should be given to the subcomponent most important for long-term persistence and/or the subcomponent that is the most sensitive to the stressor being scored. Uncertainty was also included for the Consequenceij score; see Table C.2 for uncertainty categories and scores.         234  Table C.1 Scoring of variables: a) Spatial Scale, b) Temporal Scale, c) Load and d) Consequence (a) Temporal Frequency Scale Score Effect Definition 1 Rare Every several years – Decadal 2 Relatively Often Quarterly – Annually 3 Frequent Weekly – Monthly 4 Continuous Daily occurrences or continuous (b) Spatial Scale Score Effect Definition 1 Few restricted locations 1-10 kilometres 2 Localized 10-100 kilometres 3 Widespread >100 kilometres (c) Load – Density/Persistence Score Effect Definition 1 Low Low density and low persistence 2 Moderate High density or persistence 3 High High density and persistence  (d) Consequence Score Effect Definition 1 Negligible Negligible impact on population/habitat/community 2 Minor Minimal impact on population/habitat/ community structure or dynamics 3 Moderate Maximum impact that still meets an objective (e.g. sustainable level of impact such as a full exploitation rate for a target species; maintaining levels of critical habitat) 4 Major Wider and longer term impacts (e.g. long-term decline in CPUE) 5 Severe Very serious impacts occurring, with a relatively long time period likely to be needed to restore to an acceptable level (e.g. serious decline in spawning biomass limiting population increase) 6 Intolerable Widespread and permanent/irreversible damage or loss will occur – unlikely to ever be fixed (e.g. local extinction)       235  Table C.2 Scoring definition of the uncertainty of risk scores. Uncertainty Score Literature Definition 1 Extensive Extensive scientific information; peer-reviewed information; data specific to the location; supported by long-term datasets (10 years or more) 2 Substantial Substantial scientific information; non-peer-reviewed information; data specific to the region; supported by recent data (within the last 10 years) or research 3 Moderate Moderate level of information; data from comparable regions or older data (more than 10 years) from the area of interest 4 Limited Limited information; expert opinion based on observational information or circumstantial evidence 5 Little to None Little or no information; expert opinion based on general knowledge  C.2 Scoring Uncertainty An uncertainty propagation exercise was completed to include the uncertainty of the qualitative risk scoring in the final and cumulative risk scores.  Each risk variable (Temporal Scale, Spatial Scale, Intensity and Consequence) was assigned as the mean of a normal distribution with standard deviation set according to the level of uncertainty assigned (Table C.2). The distribution was bounded by the minimum and maximum scores for each risk variable so that the scores could not be higher or lower than the variable’s scale (e.g., the intensity score cannot be lower 236  than 1 or higher than 3).  The score of each risk variable was then randomly sampled from this distribution with 3000 replicates. The final risk score for each mechanistic impact pathway was a product of the four risk variable arrays (Risk = SS x TS x I x C2), where the first score generated from each variable array is multiplied across all four risk variables, followed by the second, and so on for all 3000 replicates and resulting in a final risk array of 3000 scores.       237  C.3 Supplementary Figures and Tables  Figure C.1 Food web and habitat relationships between the species in the coastal British Columbia case study. Habitat relationships are represented by dashed lines and trophic relationships are represented by solid lines. 238  Table C.3 Nodes ranked by influence in maintaining low risk to herring. The columns identify the probabilities of low, medium, and high risk according to each scenario. Nodes in bold are assumed leverage nodes in the herring IFMP Rank Class Node Low Risk Medium Risk High Risk       Initial Target Herring IFMP PA SAR Initial Target Herring IFMP PA SAR Initial Target Herring IFMP PA SAR 1 Stressor Direct Capture 0.17 0.40 0.43 0.23 0.38 0.83 0.60 0.57 0.77 0.62 0.00 0.00 0.00 0.00 0.00 2 Stressor Acoustic 0.13 0.22 0.13 0.13 0.27 0.87 0.78 0.87 0.87 0.73 0.00 0.00 0.00 0.00 0.00 3 Stressor Oil Spill 0.00 0.00 0.00 0.00 0.00 0.11 0.19 0.11 0.14 0.25 0.89 0.81 0.89 0.86 0.75 4 Stressor Change in water flow 0.92 0.84 0.92 0.89 0.86 0.08 0.16 0.08 0.11 0.14 0.00 0.00 0.00 0.00 0.00 5 Stressor Marine Debris 0.75 0.67 0.75 0.71 0.71 0.25 0.33 0.25 0.29 0.29 0.00 0.00 0.00 0.00 0.00 6 Stressor Temperature Change 0.09 0.16 0.09 0.12 0.14 0.91 0.84 0.91 0.88 0.85 0.00 0.00 0.00 0.00 0.00 7 Stressor Nutrient Input 0.00 0.00 0.00 0.00 0.00 0.74 0.67 0.74 0.70 0.62 0.26 0.33 0.26 0.30 0.38 8 Stressor Sedimentation 0.00 0.00 0.00 0.00 0.00 0.33 0.38 0.33 0.35 0.41 0.67 0.62 0.67 0.65 0.59 9 Stressor Persistent Organic Pollutants 0.00 0.00 0.00 0.00 0.00 0.97 0.93 0.97 0.96 0.90 0.02 0.07 0.02 0.03 0.10 10 Driver Seine Fisheries 0.19 0.22 1.00 0.20 0.22 0.68 0.64 0.00 0.66 0.63 0.13 0.15 0.00 0.13 0.15 11 Driver Trolling 0.25 0.27 0.31 0.26 0.27 0.65 0.61 0.56 0.64 0.60 0.10 0.12 0.13 0.11 0.13 12 Driver Hook and Line 0.23 0.25 0.25 0.24 0.25 0.65 0.61 0.63 0.64 0.60 0.12 0.14 0.13 0.12 0.14 13 Driver Recreational Fisheries 0.28 0.30 1.00 0.29 0.30 0.63 0.60 0.00 0.62 0.59 0.09 0.11 0.00 0.09 0.11 14 Driver Gillnet Fisheries 0.08 0.10 1.00 0.09 0.11 0.66 0.63 0.00 0.65 0.62 0.26 0.27 0.00 0.26 0.27 15 Stressor Large Vessel Contaminants 0.02 0.05 0.02 0.03 0.08 0.98 0.95 0.98 0.97 0.92 0.00 0.00 0.00 0.00 0.00 16 Stressor Small Vessel Acoustic 0.00 0.00 0.00 0.00 0.00 0.26 0.29 0.47 0.27 0.43 0.74 0.71 0.53 0.73 0.57 17 Driver Trap Fisheries 0.30 0.32 0.33 0.31 0.32 0.62 0.59 0.58 0.61 0.58 0.08 0.09 0.09 0.08 0.10 18 Stressor Change in freshwater flow 0.58 0.55 0.58 0.56 0.55 0.42 0.45 0.42 0.44 0.45 0.00 0.00 0.00 0.00 0.00 19 Driver Small Vessel Use 0.40 0.43 0.40 0.41 0.45 0.55 0.52 0.55 0.54 0.50 0.06 0.05 0.06 0.05 0.05 239  Rank Class Node Low Risk Medium Risk High Risk       Initial Target Herring IFMP PA SAR Initial Target Herring IFMP PA SAR Initial Target Herring IFMP PA SAR 20 Driver Long Range Contamination 0.97 0.94 0.97 0.96 0.93 0.03 0.06 0.03 0.04 0.07 0.00 0.00 0.00 0.00 0.00 21 Stressor Debris 0.01 0.02 0.01 0.01 0.03 0.99 0.96 0.99 0.98 0.94 0.01 0.02 0.01 0.01 0.02 22 Driver Large Vessel Use 0.00 0.00 0.00 0.00 0.00 0.15 0.17 0.15 0.16 0.18 0.85 0.83 0.85 0.84 0.82 23 Driver Dive Fisheries 0.39 0.39 0.39 0.39 0.39 0.56 0.54 0.56 0.56 0.54 0.05 0.07 0.05 0.06 0.07 24 Stressor Invasive Species 0.00 0.00 0.00 0.00 0.00 0.46 0.47 0.46 0.48 0.48 0.54 0.53 0.54 0.52 0.52 25 Stressor Small Vessel Incidental Mortality 0.00 0.00 0.00 0.00 0.00 0.41 0.43 0.50 0.45 0.47 0.59 0.57 0.50 0.55 0.53 26 Stressor Large Vessel Nutrient Input 1.00 0.99 1.00 0.99 0.99 0.00 0.01 0.00 0.01 0.01 0.00 0.00 0.00 0.00 0.00 27 Driver Human Settlement 0.78 0.77 0.78 0.78 0.77 0.21 0.22 0.21 0.22 0.22 0.00 0.01 0.00 0.00 0.01 28 Stressor Bycatch 0.00 0.00 0.00 0.00 0.00 0.51 0.52 1.00 0.51 0.50 0.49 0.48 0.00 0.49 0.50 29 Driver Land-based Activities 0.70 0.69 0.70 0.69 0.69 0.30 0.30 0.30 0.30 0.30 0.01 0.01 0.01 0.01 0.01 30 Driver Log Handling 0.00 0.00 0.00 0.00 0.00 0.23 0.24 0.23 0.24 0.25 0.77 0.76 0.77 0.76 0.75 31 Stressor Incidental Mortality 0.87 0.87 0.87 0.84 0.83 0.13 0.13 0.13 0.16 0.17 0.00 0.00 0.00 0.00 0.00 32 Driver Climate Change 0.99 0.99 0.99 0.99 0.99 0.01 0.01 0.01 0.01 0.01 0.00 0.00 0.00 0.00 0.00 33 Stressor Sea Level Rise 0.92 0.92 0.92 0.90 0.93 0.08 0.08 0.08 0.10 0.07 0.00 0.00 0.00 0.00 0.00 34 Stressor Ocean Acidification 0.97 0.97 0.97 0.96 0.97 0.03 0.03 0.03 0.04 0.03 0.00 0.00 0.00 0.00 0.00 35 Stressor Habitat Disturbance 0.00 0.00 0.00 0.00 0.00 0.43 0.43 0.43 0.45 0.44 0.57 0.57 0.57 0.55 0.56  240  Table C.4 Human activities (in bold) and associated stressors. The numbers beside each activity and stressor corresponds to the nodes in Figure 4.2 Fisheries Sea Land LongTerm 1. Dive Fisheries 10. Finfish Aquaculture 17. Human Settlement 19. Climate Change 21. Direct Capture 25. Acoustic 26. Contaminants 36. Ocean Acidification 22. Habitat Disturbance 26. Contaminants 30. Debris 37. Sea level rise 46. Small Vessel_Acoustic 27. Fish Escapement 28. Nutrient Input 34. Temperature Change 47. Small Vessel_Contaminants 28. Nutrient Input 24. Sedimentation 20. Long Range Contamination 48. Small Vessel_Incidental Mortality 29. Predatory Control 18. Land-based Activities 38. Marine Debris 49. Small Vessel_Invasive Species 46. Small Vessel_Acoustic 35. Change in freshwater flow 39. Persistent Organic Pollutants 50. Small Vessel_Nutrient Input 47. Small Vessel_Contaminants 26. Contaminants   51. Small Vessel_Oil Spill 48. Small Vessel_Incidental Mortality 28. Nutrient Input   2. Gillnet Fisheries 49. Small Vessel_Invasive Species 24. Sedimentation   23. Bycatch 50. Small Vessel_Nutrient Input     21. Direct Capture 51. Small Vessel_Oil Spill     46. Small Vessel_Acoustic 11. Large Vessel Use     47. Small Vessel_Contaminants 25. Acoustic     48. Small Vessel_Incidental Mortality 26. Contaminants     49. Small Vessel_Invasive Species 52. Incidental Mortality     50. Small Vessel_Nutrient Input 53. Invasive Species     51. Small Vessel_Oil Spill 28. Nutrient Input     3. Hand Digging 54. Oil Spill     21. Direct Capture 12. Log Handling     4. Hook and Line 26. Contaminants     23. Bycatch 30. Debris     21. Direct Capture 22. Habitat Disturbance     46. Small Vessel_Acoustic 28. Nutrient Input     47. Small Vessel_Contaminants 46. Small Vessel_Acoustic     48. Small Vessel_Incidental Mortality 47. Small Vessel_Contaminants     49. Small Vessel_Invasive Species 48. Small Vessel_Incidental Mortality     50. Small Vessel_Nutrient Input 49. Small Vessel_Invasive Species     51. Small Vessel_Oil Spill 50. Small Vessel_Nutrient Input     5. Recreational Fishing 51. Small Vessel_Oil Spill     23. Bycatch 13. Marine Tourism     21. Direct Capture 31. Disruption of Wildlife     241  Fisheries Sea Land LongTerm 46. Small Vessel_Acoustic 22. Habitat Disturbance     47. Small Vessel_Contaminants 40. Large Vessel_Acoustic     48. Small Vessel_Incidental Mortality 41. Large Vessel_Contaminants     49. Small Vessel_Invasive Species 43. Large Vessel_Incidental Morality     50. Small Vessel_Nutrient Input 45. Large Vessel_Invasive Species     51. Small Vessel_Oil Spill 44. Large Vessel_Nutrient Input     6. Seine Fisheries 42. Large Vessel_Oil Spill     23. Bycatch 46. Small Vessel_Acoustic     21. Direct Capture 47. Small Vessel_Contaminants     46. Small Vessel_Acoustic 48. Small Vessel_Incidental Mortality     47. Small Vessel_Contaminants 49. Small Vessel_Invasive Species     48. Small Vessel_Incidental Mortality 50. Small Vessel_Nutrient Input     49. Small Vessel_Invasive Species 51. Small Vessel_Oil Spill     50. Small Vessel_Nutrient Input 14. Ports, Marinas, Harbours     51. Small Vessel_Oil Spill 32. Change in water flow     7. Trap Fisheries 26. Contaminants     23. Bycatch 22. Habitat Disturbance     21. Direct Capture 40. Large Vessel_Acoustic     46. Small Vessel_Acoustic 41. Large Vessel_Contaminants     47. Small Vessel_Contaminants 43. Large Vessel_Incidental Morality     48. Small Vessel_Incidental Mortality 45. Large Vessel_Invasive Species     49. Small Vessel_Invasive Species 44. Large Vessel_Nutrient Input     50. Small Vessel_Nutrient Input 42. Large Vessel_Oil Spill     51. Small Vessel_Oil Spill 28. Nutrient Input     8. Trawling 46. Small Vessel_Acoustic     23. Bycatch 47. Small Vessel_Contaminants     21. Direct Capture 48. Small Vessel_Incidental Mortality     22. Habitat Disturbance 49. Small Vessel_Invasive Species     24. Sedimentation 50. Small Vessel_Nutrient Input     50. Small Vessel_Nutrient Input 51. Small Vessel_Oil Spill     46. Small Vessel_Acoustic 15. Shellfish Aquaculture     47. Small Vessel_Contaminants 53. Invasive Species     242  Fisheries Sea Land LongTerm 48. Small Vessel_Incidental Mortality 33. Shading     49. Small Vessel_Invasive Species 46. Small Vessel_Acoustic     51. Small Vessel_Oil Spill 47. Small Vessel_Contaminants     9. Trolling 48. Small Vessel_Incidental Mortality     23. Bycatch 49. Small Vessel_Invasive Species     21. Direct Capture 50. Small Vessel_Nutrient Input     46. Small Vessel_Acoustic 51. Small Vessel_Oil Spill     47. Small Vessel_Contaminants 16. Small Vessel Use     48. Small Vessel_Incidental Mortality 25. Acoustic     49. Small Vessel_Invasive Species 26. Contaminants     50. Small Vessel_Nutrient Input 52. Incidental Mortality     51. Small Vessel_Oil Spill 53. Invasive Species       28. Nutrient Input       54. Oil Spill              243  Table C.5 The provision of ecosystem services (in bold) and their associated species. The numbers beside each ecosystem service and species corresponds to the nodes in Figure 2. 55. Aquaculture 58. Charismatic Species 61. Spiritual 63. Habitat Provision 68. Contaminant Filtration 70. Geoduck  81. Sponges 78. Seagrasses 78. Seagrasses 78. Seagrasses 56. Wild Harvest 80. Cold Water Coral 79. Kelp 79. Kelp 79. Kelp 71. Salmon 71. Salmon 80. Cold Water Coral 80. Cold Water Coral 86. Phytoplankton 72. Lingcod 82. Humpback Whale 81. Sponges 81. Sponges 70. Geoduck  73. Spiny Dogfish 83. Killer Whale 70. Geoduck  70. Geoduck  69. Nutrient Cycling 70. Geoduck 84. Stellar Sea Lion 75. Dungeness Crab 64. Primary Production 78. Seagrasses 75. Dungeness Crab 85. Cassin's Auklet 76. Pacific Herring 86. Phytoplankton 79. Kelp 76. Pacific Herring 59. Medicinal 77. Prawn 78. Seagrasses 80. Cold Water Coral 77. Prawn 79. Kelp 71. Salmon 79. Kelp 81. Sponges 57. Science 80. Cold Water Coral 72. Lingcod 65. Sediment Retention 70. Geoduck 78. Seagrasses 71. Salmon 73. Spiny Dogfish 78. Seagrasses 75. Dungeness Crab 79. Kelp 73. Spiny Dogfish 82. Humpback Whale 79. Kelp 76. Pacific Herring 80. Cold Water Coral 70. Geoduck 83. Killer Whale 80. Cold Water Coral 77. Prawn 81. Sponges 60. Aesthetic 84. Stellar Sea Lion 81. Sponges 71. Salmon 70. Geoduck 78. Seagrasses 85. Cassin's Auklet 70. Geoduck  72. Lingcod 75. Dungeness Crab 79. Kelp 62. Recreation 66. Water Retention 73. Spiny Dogfish 76. Pacific Herring 82. Humpback Whale 78. Seagrasses 78. Seagrasses 82. Humpback Whale 77. Prawn 83. Killer Whale 79. Kelp 79. Kelp 83. Killer Whale 71. Salmon 84. Stellar Sea Lion 80. Cold Water Coral 80. Cold Water Coral 84. Stellar Sea Lion 72. Lingcod 85. Cassin's Auklet 81. Sponges 81. Sponges 85. Cassin's Auklet 73. Spiny Dogfish  70. Geoduck  70. Geoduck  86. Phytoplankton 82. Humpback Whale  75. Dungeness Crab 67. Climate Regulation 87. Zooplankton 83. Killer Whale  76. Pacific Herring 86. Phytoplankton  84. Stellar Sea Lion  77. Prawn 78. Seagrasses  85. Cassin's Auklet  71. Salmon 79. Kelp  86. Phytoplankton  72. Lingcod   87. Zooplankton  73. Spiny Dogfish     82. Humpback Whale     83. Killer Whale     84. Stellar Sea Lion     85. Cassin's Auklet        244  Appendix D  Supplementary Methods for Chapter 6 Table D.1 A list of activities and stressors that experts verified at the workshop Stressor Type Description Commercial Fishing All associated risks such as catch, pollution, bycatch and lost fishing gear Recreational Fishing All associated risks such as catch, pollution, bycatch and lost fishing gear Shellfish Aquaculture All associated risks to intertidal habitats and risks of pollution and invasive species and disease release Commercial Shipping All associated risks such as strikes/collisions, acoustic impacts, and illegal dumping of oil wastes and greywater Coastal Structures All associated risks associated with construction, including the dredging required to create it and associated risks of contaminant release and breakwaters Sedimentation All associated risks of sediment input from land based activities Nutrient Input and Disease All associated risks of nutrient input and disease from land based activities Social Licence The risk that social opposition to an activity limits its development or risks existing facilities Climate Change All associated risks of current climate change including warmer temperatures, acidification, and storm intensity and frequency Invasive Species The risk posed by current levels of invasive species and diseases in the bays, including the potential for current species to have impacts Human Trampling The risks associated with people and vehicles causing physical damage and harm, and pollution  Industrial and Urban Pollution All associated risks associated with sewage, fuels, and other contaminants originating from human use on or near the bays. Not including sedimentation effects from forestry and agriculture  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0319134/manifest

Comment

Related Items