Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Spatially explicit robust impact patterns : a new approach to account for uncertainties of long-term… Yip, Zheng Ki (Jackie) 2018

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

Item Metadata

Download

Media
24-ubc_2018_september_yip_zheng.pdf [ 180.4MB ]
Metadata
JSON: 24-1.0371129.json
JSON-LD: 24-1.0371129-ld.json
RDF/XML (Pretty): 24-1.0371129-rdf.xml
RDF/JSON: 24-1.0371129-rdf.json
Turtle: 24-1.0371129-turtle.txt
N-Triples: 24-1.0371129-rdf-ntriples.txt
Original Record: 24-1.0371129-source.json
Full Text
24-1.0371129-fulltext.txt
Citation
24-1.0371129.ris

Full Text

Spatially explicit robust impact patterns: a new approach to account for uncertainties of long-term sea-level rise impacts at the local level by  Zheng Ki (Jackie) Yip  M.Sc., McGill University, 2010 B.Sc. (Honours), University of Cape Town, 2007 B.Sc., University of Cape Town, 2006  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)  August 2018  © Zheng Ki (Jackie) Yip, 2018   ii The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, the dissertation entitled:  Spatially explicit robust impact patterns: a new approach to account for uncertainties of long-term sea-level rise impacts at the local level  submitted by Zheng Ki (Jackie) Yip in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Resource Management and Environmental Studies  Examining Committee: Prof. Stephanie E. Chang Supervisor  Prof. Karen Bartlett Supervisory Committee Member  Prof. Tim McDaniels Supervisory Committee Member Prof. Barbara Lence University Examiner Prof. Stephen Sheppard University Examiner      iii Abstract  While sea-level rise (SLR) is a certain effect of climate change, there are deep uncertainties about when and by how much. Uncertainties regarding how SLR can impact society at the local level - further compounded by changes in non-climatic drivers, cascading effects, and local contexts – act as a significant barrier to SLR adaptation. Recent literature has called for a shift from using best available predictions to find optimal adaptation options to using scenario-based approaches to find robust options that can perform reasonably under a range of possible futures.   In response, this dissertation develops a new approach, the Robust Impact Patterns (RIPs) method, to help decision-makers account for potential SLR impact under hundreds of future scenarios, assuming that no adaptation takes place. The method utilizes the pattern recognition capability of machine learning to transform thousands of modelled SLR impact maps into a small number of impact patterns that are robust across multiple futures, thereby processing an otherwise overwhelming volume of impact information in a spatially explicit and visualized manner. This method addresses the need to account for uncertainties in the early planning stage in order to inform selection of preliminary options to target robust impacts and avoid relying on generic or existing options.     An application to the City of Vancouver demonstrated the feasibility of the RIPs method and assessed its practical utility. Geospatial models assessed 14 potential impacts (e.g., business   iv disruption, sewage backup damage potential) in 336 plausible futures that account for uncertainties in future storm intensity, SLR, land-use, power infrastructure resilience, and structural integrity of buildings. The 14 impacts were selected to address the City’s information needs and to capitalize on the capabilities of the RIPs method. Results were synthesized into 16 robust impact patterns (RIPs). City officials and experts, as potential users, were engaged in a structured workshop to discuss the results and evaluate the RIPs method’s capability to support adaptation. The RIPs method was found to be a useful platform for convening multiple types of stakeholders to understand complex SLR impacts, which can facilitate the development of new adaptation ideas, partnerships, and resources for implementation.       v Lay Summary While sea-level rise (SLR) is a certain effect of climate change, how and when SLR can impact society is highly uncertain, acting as major barrier to SLR adaptation planning. This dissertation develops a new approach – the Robust Impact Patterns (RIPs) method - to help decision-makers understand the complexity of SLR impacts and how their severity and distribution across a community can vary under a wide range of possible futures. An application of the RIPs method at the City of Vancouver demonstrated its feasibility and showed that economic, social, and environmental impacts of SLR can affect population and assets located well beyond the flooded areas due to cascading effects. More broadly, the RIPs method was found to be useful for helping multiple types of stakeholders – from local residents to engineers - to understand the complexity and relevance of SLR impacts, which can help generate new adaptation ideas, partnerships, and resources for implementation.   vi Preface This dissertation is an original intellectual product of the author, Zheng Ki (Jackie) Yip. None of the text from this dissertation is taken directly from previously published or collaborative articles. The research presented in Chapter 6 that collected data through a survey, group discussions, and workshop is covered by the UBC Behavioural Research Ethics Certificate number H15-00597.   The analyses presented herein were conducted entirely by the author with the exception of the creation of selected flood depth layers and the sector coding of businesses in the City of Vancouver. The former was constructed by the Northwest Hydraulics Consultants, as noted in Section 4.5.3. The latter was conducted by the research group led by Dr. Stephanie E. Chang, as noted in Section 4.6.2.   vii Table of Contents  Abstract ......................................................................................................................................... iii	Lay Summary .................................................................................................................................v	Preface ........................................................................................................................................... vi	Table of Contents ........................................................................................................................ vii	List of Tables .............................................................................................................................. xiv	List of Figures ............................................................................................................................. xix	List of Abbreviations ............................................................................................................. xxxiv	Acknowledgements .................................................................................................................. xxxv	Dedication ............................................................................................................................. xxxviii	Chapter 1: Introduction ................................................................................................................1	1.1	 Background .......................................................................................................................... 1	1.2	 Problem statement ................................................................................................................ 3	1.3	 Research approach and thesis structure ................................................................................ 4	1.4	 Significance .......................................................................................................................... 8	Chapter 2: Background and Contexts .......................................................................................11	2.1	 Sea-level rise impacts and adaptation ................................................................................ 11	2.2	 Importance of local and spatial context in sea-level rise impacts ...................................... 14	2.3	 Deep uncertainties of sea-level rise and its representation in impact assessments ............ 16	2.4	 Approaches to adaptation in the face of deep uncertainties ............................................... 21	2.4.1	 Scenario planning ...................................................................................................... 21	  viii 2.4.2	 Scenario-based robust adaptation conceptual frameworks ....................................... 23	2.4.3	 Assumption-based Planning (ABP) .......................................................................... 24	2.4.4	 Robust Decision Making (RDM) .............................................................................. 25	2.4.5	 Adaptive policymaking (APM) ................................................................................. 27	2.4.6	 Adaptation Tipping Points (ATP), Adaptation Pathways (AP), and Dynamic Adaptive Policy Pathways (DAPP) ...................................................................................... 29	2.4.7	 Info-Gap (IG) ............................................................................................................ 31	2.4.8	 Scenario-neutral approach (SNA) ............................................................................. 33	2.5	 Research gap ...................................................................................................................... 33	Chapter 3: Development of the Robust Impact Patterns (RIPs) method ...............................36	3.1	 Method development – the thought process ....................................................................... 36	3.1.1	 Why Robust Impact Patterns? ................................................................................... 37	3.1.2	 How to identify the Robust Impact Patterns? ........................................................... 39	3.2	 The Robust Impact Patterns Method .................................................................................. 41	3.2.1	 Stage I - Future flood scenarios development ........................................................... 41	3.2.2	 Stage II - Flood impact modeling ............................................................................. 43	3.2.3	 Stage III - Robust impact pattern identification ........................................................ 44	3.2.3.1	 Phase A - Pre-processing ................................................................................... 46	3.2.3.2	 Phase B – Self-organizing Maps (SOMs) training ............................................ 46	3.2.3.3	 Phase C – Visualizing, evaluating and interpreting the SOMs .......................... 48	3.3	 Intended utility of RIPs in SLR adaptation ........................................................................ 53	Chapter 4: Application of the Robust Impact Patterns Method at the City of Vancouver – Data and Methods ........................................................................................................................57	  ix 4.1	 Introduction ........................................................................................................................ 57	4.2	 Study area – City of Vancouver ......................................................................................... 58	4.3	 Why apply the RIPs method at the CoV? .......................................................................... 60	4.4	 Summary of CoV’s current efforts in SLR adaptation planning ........................................ 62	4.5	 Future flood scenarios development .................................................................................. 66	4.5.1	 Impacts selection ....................................................................................................... 66	4.5.2	 Characterizing future scenarios ................................................................................. 68	4.5.3	 Inundation - storm intensity and sea-level rise ......................................................... 70	4.5.4	 Land-use - population and building distribution ....................................................... 76	4.5.5	 Power outage – resilience of electric power infrastructure ....................................... 87	4.5.6	 Building vulnerability to flood damage – stage-damage functions .......................... 94	4.6	 Impact models .................................................................................................................... 99	4.6.1	 Direct building damage and debris generated ......................................................... 101	4.6.2	 Business disruption ................................................................................................. 106	4.6.3	 Social Impacts ......................................................................................................... 114	4.6.4	 Sewage Back-up Damage Potential ........................................................................ 116	4.6.4.1	 Measuring sewage backup risk ........................................................................ 117	4.6.4.2	 Development of the Sewage Backup Damage Potential Index (SBDPI) ........ 118	4.6.4.3	 Sewage Backup Damage Potential Index (SBDPI) for the CoV ..................... 125	4.6.4.4	 Index validation ............................................................................................... 126	4.7	 Self-organizing Maps Analysis ........................................................................................ 127	4.7.1	 Training the SOMs .................................................................................................. 128	4.7.2	 Evaluating the SOMs .............................................................................................. 132	  x 4.8	 Summary .......................................................................................................................... 133	Chapter 5: Application of the Robust Impacts Patterns Method at the City of Vancouver -Results .........................................................................................................................................134	5.1	 Introduction ...................................................................................................................... 134	5.2	 Interpreting results by groups ........................................................................................... 134	5.3	 Business disruptions in the primary, secondary, and tertiary sectors ............................... 142	5.4	 Direct building damage – residential, commercial, and governmental buildings ............ 154	5.5	 Vulnerable population ...................................................................................................... 161	5.6	 Disaster response facilities – schools, emergency services, health care facilities, transportation points and social service facilities ........................................................... 164	5.7	 Debris ............................................................................................................................... 173	5.8	 Sewage backup damage potential .................................................................................... 176	5.9	 Conclusion ........................................................................................................................ 185	Chapter 6: Potential Utility of the Robust Impact Patterns Method ....................................190	6.1	 Introduction ...................................................................................................................... 190	6.2	 Methodology .................................................................................................................... 194	6.2.1	 Participants and recruitment process ...................................................................... 194	6.2.2	 Workshop structure and activities ........................................................................... 195	6.2.2.1	 Presentation ...................................................................................................... 196	6.2.2.2	 Survey .............................................................................................................. 197	6.2.2.3	 Group discussion .............................................................................................. 199	6.2.3	 Analysis method ...................................................................................................... 200	6.2.3.1	 High-level survey analysis ............................................................................... 200	  xi 6.2.3.2	 Qualitative analysis .......................................................................................... 201	6.3	 Results .............................................................................................................................. 202	6.3.1	 High-level analysis of survey responses ................................................................. 202	6.3.1.1	 Responses of subgroups ................................................................................... 205	6.3.2	 Qualitative analysis - potential applications of the RIPs to support adaptation ..... 209	6.3.3	 Generate new ideas and wider range of option types ............................................. 209	6.3.4	 Refine and target efforts .......................................................................................... 212	6.3.5	 More uncertainty-tolerant options ........................................................................... 212	6.3.6	 Long-range planning of modifications .................................................................... 214	6.3.7	 Prioritizing efforts ................................................................................................... 215	6.3.8	 Communicating SLR risk ....................................................................................... 217	6.3.9	 Justify to plan beyond 1m of SLR and provide better leverage .............................. 221	6.3.10	 Reframing issues as opportunities ........................................................................ 223	6.3.11	 Development of adaptation pathways ................................................................... 223	6.3.12	 Suggested modifications ....................................................................................... 225	6.4	 Conclusion ........................................................................................................................ 230	Chapter 7: Conclusion and Discussion ....................................................................................233	7.1	 Key findings and significance .......................................................................................... 233	7.1.1	 Development of the RIPs method ........................................................................... 233	7.1.2	 City of Vancouver case study ................................................................................. 235	7.1.3	 Evaluation of capability to support SLR adaptation ............................................... 237	7.2	 Limitations ....................................................................................................................... 241	7.2.1	 Conceptual limitations of the RIPs method ............................................................ 241	  xii 7.2.2	 Challenges in applying the RIPs method ................................................................ 243	7.2.3	 Limitation of the evaluation process ....................................................................... 243	7.3	 Future research ................................................................................................................. 244	7.3.1	 Future applications .................................................................................................. 244	7.3.2	 Modification of the key purpose ............................................................................. 246	References ...................................................................................................................................250	Appendix A ............................................................................................................................... 261	A.1	 Flood depth maps of the 21 inundation conditions ................................................... 262	A.2	 Power outage conditions associated with each inundation conditions ..................... 283	Appendix B ............................................................................................................................... 325	B.1	 Average number of units in each occupancy class of residential buildings in Hazus ………………………………………………………………………………………326	B.2	 Employments inside each occupancy class of non-residential buildings .................. 327	B.3	 Stage damage function (SDF) sources for each Hazus occupancy class of buildings ………………………………………………………………………………………335	B.4	 Equation for estimating debris generated from building damage ............................. 337	B.5	 Major sector categories of business types in the CoV’s business licensing data ...... 338	B.6	 Major sector categories of business sub-types in the CoV’s business licensing data ………………………………………………………………………………………343	B.7	 Sewage Backup Damage Potential Index result summary table ............................... 345	Appendix C ............................................................................................................................... 347	C.1	 Characteristics of scenarios represented by each RIP ............................................... 347	Appendix D ............................................................................................................................... 353	  xiii D.1	 Email invitation of the expert workshop and attached project summary .................. 354	D.2	 Workshop presentation slides ................................................................................... 356	D.3	 Workshop survey ...................................................................................................... 398	D.4	 Population pyramids of the Mann-Whitney U tests for subgroup pair A ................. 403	D.5	 Population pyramids of the Mann-Whitney U tests for subgroup pair B ................. 408	   xiv List of Tables  Table 2.1 Summary of SLR impacts on socio-economic sectors in coastal areas. (Table 2 from © Nicholls (2011) Planning for the impacts of sea level rise. Oceanography, 24(2), 144-157, Page 149. By permission from publisher.) ........................................................... 12	Table 3.1 Anticipated ways in which the RIPs method can support SLR adaptation planning ... 56	Table 4.1 Vulnerable infrastructure and services identified by the CoV’s CFRA (Table 2 from ©Lyle & Mills (2016). Assessing coastal flood risk in a changing climate for the City of Vancouver. Canadian Water Resources Journal, 41(1-2), 343-352, Page 349. Adapted by permission from publisher) ....................................................................... 63	Table 4.2 SLR impacts assessed in this CoV case study .............................................................. 66	Table 4.3 Inundation conditions for this CoV case study ............................................................. 72	Table 4.4 Occupancy classes of buildings defined in Hazus (Table 4.2 from Federal Emergency Management Agency (FEMA) (2009). Multi-hazard Loss Estimation Methodology, Flood Model Hazus-MH User Manual. Retrieved from https://www.fema.gov/media-library-data/20130726-1820-25045-8814/hzmh2_1_fl_um.pdf. By permission from publisher.) ..................................................................................................................... 78	Table 4.5 Change in population and dwelling units in the CoV by 2041 as projected in the Metro Vancouver’s Regional Growth Strategy (Metro Vancouver, 2013) ............................. 79	Table 4.6 Proportions to distribute new dwelling units across occupancy classes in the Compact land-use condition ........................................................................................................ 82	  xv Table 4.7 Proportion to distribute new dwelling units across occupancy classes in the Sprawl land-use condition ........................................................................................................ 82	Table 4.8 Flood damage models considered for modeling the building damage impact in this case study ............................................................................................................................. 95	Table 4.9 Summary of modeling approaches used for assessing each SLR impact ................... 101	Table 4.10 Examples of typical amount of debris generated by different flood depth for different occupancy class of buildings (Part of Table 11.1 from Federal Emergency Management Agency (2009). Multi-hazard Loss Estimation Methodology, Flood Model Hazus-MH Technical Manual. Retrieved from https://www.fema.gov/media-library-data/20130726-1820-25045-8292/hzmh2_1_fl_tm.pdf. By permission from publisher.) ................................................................................................................... 105	Table 4.11 Reclassification of building foundation types into two major foundation types (Table 11.2 from Federal Emergency Management Agency (2009). Multi-hazard Loss Estimation Methodology, Flood Model Hazus-MH Technical Manual. Retrieved from https://www.fema.gov/media-library-data/20130726-1820-25045-8292/hzmh2_1_fl_tm.pdf. By permission from publisher.) ....................................... 105	Table 4.12 The BD model estimates the operating status of businesses of these major sectors .................................................................................................................................. ..107	Table 4.13 Hypothetical example showing how the Hazus building damage count by different damage levels can be converted to cumulative probability distribution. ................... 110	Table 4.14 Levels of disruptiveness associated with each state of building damage ................. 111	Table 4.15 Level of disruptiveness from a power outage in each major sector (Table 2.4 from © Chang et al. (2008). Linking Lifeline Infrastructure Performance and Community   xvi Disaster Resilience: Models and Multi-stakeholder Processes. MCEER, Page 15. Adapt by permission from publisher) ......................................................................... 112	Table 4.16 Temporary business closure from multiple sources of disruption (Table 2.6 from © Chang et al. (2008). Linking Lifeline Infrastructure Performance and Community Disaster Resilience: Models and Multi-stakeholder Processes. MCEER, Page 19. Adapt by permission from publisher) ......................................................................... 113	Table 4.17 Classification of major sector categories into primary, secondary, and tertiary sector .................................................................................................................................... 113	Table 4.18 Social impact variables modeled, their descriptions, input data variables and sources. .................................................................................................................................... 115	Table 4.19 Potential indicators to measure the three components of sewage backup risk ......... 123	Table 4.20 Data sources for indicators used in the Sewage Backup Damage Potential Index (SBDPI) ...................................................................................................................... 123	Table 4.21 Elicited sewage backup vulnerability scores for each drainage setup ...................... 124	Table 4.22 Parameters used for the four SOMs trained using the modeled impacts for this case study. .......................................................................................................................... 131	Table 5.1 Business disruptions for three major types of sectors summarized in terms of the 3 groups of RIPs (see notes below table) ...................................................................... 145	Table 5.2 Direct building damage for three types of buildings summarized in terms of the 3 groups of RIPs (see notes below table). ..................................................................... 155	Table 5.3 Vulnerable population potentially affected by inundation and/or prolonged power outage summarized in terms of the 3 groups of RIPs (see notes below table) ........... 161	  xvii Table 5.4 Potentially affected schools, health care facilities and emergency services summarized in terms of the 3 RIPs groups (see notes below table) ............................................... 168	Table 5.5 Potentially affected transportation points and social service facilities summarized in terms of the 3 groups of RIPs (see notes below table). .............................................. 170	Table 5.6 Weight of flood debris from building damage summarized in terms of the 3 groups of RIPs (see notes below table). ..................................................................................... 174	Table 5.7 Sewage backup damage potential index (SBDPI) summarized in terms of the 3 groups of RIPs (see notes below table) .................................................................................. 177	Table 5.8 Weighted average of all assessed impacts in each RIP groups ................................... 186	Table 6.1 Ten anticipated ways in which the RIPs can be used to support SLR adaptation planning ...................................................................................................................... 193	Table 6.2 Expert workshop agenda ............................................................................................. 196	Table 6.3 List of different ways in which information was captured for this component of the dissertation .................................................................................................................. 200	Table 6.4 Statistical test (Mann-Whitney U test) for differences between responses of participants currently involved in CoV's adaptation planning and those who are not involved (Pair A). ....................................................................................................... 207	Table 6.5 Statistical test (Mann-Whitney U test) for differences between responses of CoV staff participants and those who are not CoV staff ............................................................. 208	Table B.1.1 The average number of dwelling units in each occupancy class of residential buildings in Hazus ...................................................................................................... 326	Table B.2.1 Type(s) of businesses, defined by the Standard Industrial Codes (SIC), expected in each Hazus occupancy classes of non-residential building (Table 3.1 from Federal   xviii Emergency Management Agency (2009). Multi-hazard Loss Estimation Methodology, Flood Model Hazus-MH Technical Manual. Retrieved from https://www.fema.gov/media-library-data/20130726-1820-25045-8292/hzmh2_1_fl_tm.pdf. By permission from publisher.) ....................................... 328	Table B.2.2 Hazus occupancy class matched to each BC Assessment Actual Use Codes ......... 329	Table B.3.1 The sources of SDFs used to model the direct damage of each Hazus occupancy class of buildings. Note: Dash in the last column indicates that the SDF used was the same as in Hazus (condition #1) ................................................................................. 333	Table B.5.1 Classification of CoV business types by major sector categories. Note: FIR – Finance and real estate; HTH – Health services; MAN - classified manually; MCT – Mining, construction, transport, and utilities; MFG – Manufacturing; TRD – Wholesale and retail; AGR – Agriculture; and SVC – All other services. ................ 338	Table B.6.1 Classification of CoV business sub-types by major sector categories. Note: FIR – Finance and real estate; HTH – Health services; MAN - classified manually; MCT – Mining, construction, transport, and utilities; MFG – Manufacturing; TRD – Wholesale and retail; AGR – Agriculture; and SVC – All other services. ................ 343	Table B.7.1 Sewage backup damage potential index results for two extreme scenarios summarized by CoV neighbourhoods ........................................................................ 345	   xix List of Figures  Figure 2.1 Cascade of uncertainties (Figure 1 from © Wilby & Dessai (2010) Robust adaptation to climate change. Weather, 65(7), 180-185, Page 181. By permission from publisher.) ...................................................................................................................................... 18	Figure 2.2 Upper limit projections published in the past 5 years for year 2100 global mean sea level. Studies included are listed below the chart. ....................................................... 19	Figure 2.3 Basic steps and flow of the Assumption-Based Planning framework (Figure 1.1 from © Dewar (2002). Assumption-based Planning: A Tool for Reducing Avoidable Surprises. Cambridge University Press. Page 2. By permission from publisher.) ....... 25	Figure 2.4 Four major steps of the RDM framework. (Figure 1 from © Lempert et al. (2013). Making Good Decisions Without Predictions Robust Decision Making for Planning Under Deep Uncertainty. RAND Corporation Research Briefs, RB-9701. By permission from publisher. Available at https://www.rand.org/pubs/research_briefs/RB9701.html) ......................................... 26	Figure 2.5 Steps within the 4 major stages of Adaptive Policymaking (APM) framework (Figure 1 from Kwakkel et al. (2010). Adaptive Airport Strategic Planning. European Journal for Transportation, Infrastructure Research, 10(3), 249-273, Page 258. By permission of publisher.) ................................................................................................................ 28	Figure 2.6 An example of an Adaptation Pathway. (Figure 2 from © Haasnoot et al. (2013). Dynamic adaptive policy pathways: A method for crafting robust decisions for a   xx deeply uncertain world. Global Environmental Change, 23, 485-498, Page 488. By permission from publisher.) ......................................................................................... 30	Figure 2.7 Iterative process of the Dynamic Adaptive Policy Planning (DAPP) framework (Figure 4 from © Haasnoot et al. (2013) Dynamic adaptive policy pathways: A method for crafting robust decisions for a deeply uncertain world. Global Environmental Change, 23, 485-498, Page 489. By permission of publisher) ............ 31	Figure 2.8 Flow diagram showing the process of the scenario-neutral approach to adaptation planning (Figure 3 from Wilby & Dessai (2010). Robust adaptation to climate change. Weather, 65(7), 180-185, Page 183. By permission from publisher.) ......................... 32	Figure 3.1 Summary of the 3 stages of the RIPs method .............................................................. 41	Figure 3.2 Schematic diagram showing the 3 phases of the process to use SOMs to extract Robust Impact Patterns (RIPs) ..................................................................................... 45	Figure 3.3 Three different shapes of surfaces in which reference vectors can be visualized collectively ................................................................................................................... 49	Figure 3.4 Schematic diagram showing the process to visualize the trained SOM reference vectors as RIPs ............................................................................................................. 50	Figure 3.5: Intended ways in which RIPs can be used to support SLR adaptation ...................... 55	Figure 4.1 Three key stages of the RIPs method (Same as shown in Chapter 3) ......................... 58	Figure 4.2 Map showing the geographic location of the City of Vancouver and its surrounding water bodies in the inset map ....................................................................................... 59	Figure 4.3 Variations of 4 different types of conditions are combined systematically to generate 336 future scenarios ...................................................................................................... 70	  xxi Figure 4.4 Flood depth maps for inundation condition a) C6 (1:10,000-year storm with 6m SLR) [top], and b) A1 (1:50-year storm with 1m of SLR) [bottom]. .................................... 75	Figure 4.5 General steps to derive the number of new residential buildings of occupancy class x per DB .......................................................................................................................... 79	Figure 4.6 Electric power transmission and distribution system .................................................. 88	Figure 4.7 Pessimistic/worse case power outage conditions associated with 1m of SLR and 1:500-year storm inundation condition. DBs shaded in green are expected to experience power outage. ............................................................................................. 92	Figure 4.8 General process of Hazus estimating the direct damage of a specific occupancy class of building. ................................................................................................................. 103	Figure 4.9 Schematic structure of the business disruption model (S.E. Chang et al., 2008) ...... 107	Figure 4.10: Comparing combined and separated sewer systems (Figure 1 from Potera (2015). After the Fall. Environmental Health Perspectives, 123(9), Page A243. Available at: https://ehp.niehs.nih.gov/123-a243/ Reproduced with permission) ........................... 121	Figure 4.11 A schematic diagram showing the three stages of the RIPs method implemented for the case study at the City of Vancouver ..................................................................... 133	Figure 5.1 The 16 RIPs of tertiary sector business disruption. The shading shows the affected businesses per hectare; the total number of affected businesses and the relative robustness (RR) are labelled above each RIP. ........................................................... 136	Figure 5.2 The level of SLR associated with impacts represented by the RIP at the corresponding location in the 2D array as shown in Figure 5.1. The x-axis is the percentage of scenarios represented by the RIP. ............................................................................... 137	  xxii Figure 5.3 An UMatrix produced by the SOM Toolbox to visualize the Euclidean distance (shading) between each RIP (grey blocks). ................................................................ 138	Figure 5.4 Bar charts showing the a) storm intensity, b) SLR, c) land-use, d) power outage, and e) vulnerability of buildings associated with the RIPs of Groups A, B, and C. The percentage of all scenarios for the case study represented by each group is shown in (f). ............................................................................................................................... 140	Figure 5.5 Neighborhoods in the City of Vancouver .................................................................. 141	Figure 5.6 The 16 RIPs of primary sector business disruption. The shading shows the number of affected businesses per hectare, while the total number of affected businesses and relative robustness (RR) of each RIP are labeled above it. ........................................ 143	Figure 5.7 The 16 RIPs of secondary sector business disruption. The shading shows the number of affected businesses per hectare, while the total number of affected businesses and relative robustness of each RIP are labeled above it. ................................................. 144	Figure 5.8 The 16 RIPs of tertiary sectors business disruptions. The shading shows the number of affected businesses per hectare, while the total number of affected businesses and relative robustness of each RIP are labeled above it. (Note: this is a larger version of Figure 5.1) .................................................................................................................. 147	Figure 5.9 The inundation conditions (left column) and power outage conditions (right column) for 3 different scenarios - 1m SLR with 1:50-year storm (top), 2m SLR with 1:500-year storm (middle), and 6m SLR with 1:10,000-year storm (bottom). .................... 148	Figure 5.10 The RIPs of tertiary sector business disruption of Group A (top), their associated land-use conditions (middle), and SDFs (bottoms). ................................................... 151	  xxiii Figure 5.11 The RIPs of tertiary sector business disruption of Group C (top-left), their associated SDFs (top-right), and land-use conditions (bottom) .................................................. 152	Figure 5.12 The distribution of new residential buildings in the Compact (top-left), Status Quo (top-right), and Sprawl (bottom) land-use conditions ................................................ 153	Figure 5.13 The 16 RIPs of residential buildings direct damage cost. The shading shows the cost of damage per hectare, while the total cost of damage and relative robustness of each RIP are labeled above it. ............................................................................................. 156	Figure 5.14 The 16 RIPs of commercial and industrial buildings direct damage cost. The shading shows the cost per hectare, while the total cost of damage and relative robustness of each RIP are labeled above it. .................................................................................... 157	Figure 5.15 The 16 RIPs of governmental and public buildings direct damage cost. The shading shows the cost per hectare, while the total cost in each RIP is labeled above it. ....... 160	Figure 5.16 The 16 RIPs of affected vulnerable population. The shading shows the affected population per hectare, while the total affected population in each RIP is labeled above it. ................................................................................................................................. 162	Figure 5.17 Population density of persons age 65 and above in the City of Vancouver based on 2011 Canadian Census data ........................................................................................ 163	Figure 5.18 The 16 RIPs of affected schools. The shading shows the number of affected schools per hectare, while the total number of affected schools and the relative robustness of each RIP are labeled above it. .................................................................................... 165	Figure 5.19 The 16 RIPs of the affected health care facilities. The shading shows the number of affected health care facilities per hectare, while the total number of affected facilities and the relative robustness of each RIP are labeled above it. .................................... 166	  xxiv Figure 5.20 The 16 RIPs of affected emergency service facilities. The shading shows the number of affected facilities per hectare, while the total number of affected facilities and the relative robustness of each RIP are labeled above it. ................................................. 167	Figure 5.21 The 16 RIPs of affected social service facilities. The shading shows the number of affected facilities per hectare, while the total number of affected facilities and the relative robustness of each RIP are labeled above it. ................................................. 171	Figure 5.22 The 16 RIPs of the affected transportation points. The shading shows the number of affected points per hectare, while the total number of affected points and the relative robustness of each RIP are labeled above it. .............................................................. 172	Figure 5.23 The 16 RIPs of debris generated. The shading shows the tons per hectare, while the total weight and the relative robustness of each RIP are labeled above it. ................ 175	Figure 5.24 The 16 RIPs of sewage backup damage potential index (SBDPI). The shading shows the damage potential per hectare, while the total damage potential in each RIP is labeled above it. .......................................................................................................... 179	Figure 5.25 Percent of ground-related homes that are constructed before the 1970s. Based on data from the 2011 National Household Survey (NHS) Profiles Files. ..................... 180	Figure 5.26 Percent of ground-related homes that are connected to combined sewer system and have sewage pump installed (top) and areas of prolonged power outage associated with inundation condition with 1:500-year storm and 3m of SLR (bottom). ............. 181	Figure 5.27 Percent of ground-related homes connected to the combined sewer system and have no sewage pump installed. .......................................................................................... 182	Figure 5.28 The RIPs of SBDPI in Group C (top-left) and its associated storm intensity (top-right), power outage conditions (bottom-left), and SLR (bottom-right). ................... 184	  xxv Figure 6.1 Range of City of Vancouver departments and organizations represented by the workshop participants ................................................................................................. 202	Figure 6.2 Number of years the participant spent in their current role ....................................... 203	Figure 6.3 Survey responses for Likert type questions related to the five ways to use the RIPs in supporting SLR adaptation options development ...................................................... 204	Figure 6.4 Survey responses for Likert type questions related to the five ways to use the RIPs in supporting access to resources and stakeholders support for implementing adaptation actions. ........................................................................................................................ 205	Figure A.1.1 Flood depth map of inundation condition of 1:50 year storm with 0m SLR ......... 262	Figure A.1.2 Flood depth map of inundation condition of 1:50 year storm with 1m SLR ......... 263	Figure A.1.3 Flood depth map of inundation condition of 1:50 year storm with 2m SLR ......... 264	Figure A.1.4 Flood depth map of inundation condition of 1:50 year storm with 3m SLR ......... 264	Figure A.1.5 Flood depth map of inundation condition of 1:50 year storm with 4m SLR ......... 266	Figure A.1.6 Flood depth map of inundation condition of 1:50 year storm with 5m SLR ......... 267	Figure A.1.7 Flood depth map of inundation condition of 1:50 year storm with 6m SLR ......... 268	Figure A.1.8 Flood depth map of inundation condition of 1:500 year storm with 0m SLR ....... 269	Figure A.1.9 Flood depth map of inundation condition of 1:500 year storm with 1m SLR ....... 270	Figure A.1.10 Flood depth map of inundation condition of 1:500 year storm with 2m SLR ..... 271	Figure A.1.11 Flood depth map of inundation condition of 1:500 year storm with 3m SLR ..... 272	Figure A.1.12 Flood depth map of inundation condition of 1:500 year storm with 4m SLR ..... 273	Figure A.1.13 Flood depth map of inundation condition of 1:500 year storm with 5m SLR ..... 274	Figure A.1.14 Flood depth map of inundation condition of 1:500 year storm with 6m SLR ..... 275	  xxvi Figure A.1.15 Flood depth map of inundation condition of 1:10,000 year storm with 0m SLR .................................................................................................................................. ..276	Figure A.1.16 Flood depth map of inundation condition of 1:10,000 year storm with 1m SLR .................................................................................................................................. ..277	Figure A.1.17 Flood depth map of inundation condition of 1:10,000 year storm with 2m SLR .................................................................................................................................. ..278	Figure A.1.18 Flood depth map of inundation condition of 1:10,000 year storm with 3m SLR .................................................................................................................................. ..279	Figure A.1.19 Flood depth map of inundation condition of 1:10,000 year storm with 4m SLR .................................................................................................................................. ..280	Figure A.1.20 Flood depth map of inundation condition of 1:10,000 year storm with 5m SLR .................................................................................................................................. ..281	Figure A.1.21 Flood depth map of inundation condition of 1:10,000 year storm with 6m SLR .................................................................................................................................. ..282	Figure A.2.1 Optimistic power outage condition associated with the inundation from 1:50 year storm with 0m SLR .................................................................................................... 283	Figure A.2.2 Pessimistic power outage condition associated with the inundation from 1:50 year storm with 0m SLR .................................................................................................... 284	Figure A.2.3 Optimistic power outage condition associated with the inundation from 1:50 year storm with 1m SLR .................................................................................................... 285	Figure A.2.4 Pessimistic power outage condition associated with the inundation from 1:50 year storm with 1m SLR .................................................................................................... 286	  xxvii Figure A.2.5 Optimistic power outage condition associated with the inundation from 1:50 year storm with 2m SLR .................................................................................................... 287	Figure A.2.6 Pessimistic power outage condition associated with the inundation from 1:50 year storm with 2m SLR .................................................................................................... 288	Figure A.2.7 Optimistic power outage condition associated with the inundation from 1:50 year storm with 3m SLR .................................................................................................... 289	Figure A.2.8  Pessimistic power outage condition associated with the inundation from 1:50 year storm with 3m SLR .................................................................................................... 290	Figure A.2.9 Optimistic power outage condition associated with the inundation from 1:50 year storm with 4m SLR .................................................................................................... 291	Figure A.2.10 Pessimistic power outage condition associated with the inundation from 1:50 year storm with 4m SLR .................................................................................................... 292	Figure A.2.11 Optimistic power outage condition associated with the inundation from 1:50 year storm with 5m SLR .................................................................................................... 293	Figure A.2.12 Pessimistic power outage condition associated with the inundation from 1:50 year storm with 5m SLR .................................................................................................... 294	Figure A.2.13 Optimistic power outage condition associated with the inundation from 1:50 year storm with 6m SLR .................................................................................................... 295	Figure A.2.14 Pessimistic power outage condition associated with the inundation from 1:50 year storm with 6m SLR .................................................................................................... 296	Figure A.2.15 Optimistic power outage condition associated with the inundation from 1:500 year storm with 0m SLR .................................................................................................... 297	  xxviii Figure A.2.16 Pessimistic power outage condition associated with the inundation from 1:500 year storm with 0m SLR ............................................................................................ 298	Figure A.2.17 Optimistic power outage condition associated with the inundation from 1:500 year storm with 1m SLR .................................................................................................... 299	Figure A.2.18 Pessimistic power outage condition associated with the inundation from 1:500 year storm with 1m SLR ............................................................................................ 300	Figure A.2.19 Optimistic power outage condition associated with the inundation from 1:500 year storm with 2m SLR .................................................................................................... 301	Figure A.2.20 Pessimistic power outage condition associated with the inundation from 1:500 year storm with 2m SLR ............................................................................................ 302	Figure A.2.21 Optimistic power outage condition associated with the inundation from 1:500 year storm with 3m SLR .................................................................................................... 303	Figure A.2.22 Pessimistic power outage condition associated with the inundation from 1:500 year storm with 3m SLR ............................................................................................ 304	Figure A.2.23 Optimistic power outage condition associated with the inundation from 1:500 year storm with 4m SLR .................................................................................................... 305	Figure A.2.24 Pessimistic power outage condition associated with the inundation from 1:500 year storm with 4m SLR ............................................................................................ 306	Figure A.2.25 Optimistic power outage condition associated with the inundation from 1:500 year storm with 5m SLR .................................................................................................... 307	Figure A.2.26 Pessimistic power outage condition associated with the inundation from 1:500 year storm with 5m SLR ............................................................................................ 308	  xxix Figure A.2.27 Optimistic power outage condition associated with the inundation from 1:500 year storm with 6m SLR .................................................................................................... 309	Figure A.2.28 Pessimistic power outage condition associated with the inundation from 1:500 year storm with 6m SLR ............................................................................................ 310	Figure A.2.29 Optimistic power outage condition associated with the inundation from 1:10,000 year storm with 0m SLR ............................................................................................ 311	Figure A.2.30 Pessimistic power outage condition associated with the inundation from 1:10,000 year storm with 0m SLR ............................................................................................ 312	Figure A.2.31 Optimistic power outage condition associated with the inundation from 1:10,000 year storm with 1m SLR ............................................................................................ 313	Figure A.2.32 Pessimistic power outage condition associated with the inundation from 1:10,000 year storm with 1m SLR ............................................................................................ 314	Figure A.2.33 Optimistic power outage condition associated with the inundation from 1:10,000 year storm with 2m SLR ............................................................................................ 315	Figure A.2.34 Pessimistic power outage condition associated with the inundation from 1:10,000 year storm with 2m SLR ............................................................................................ 316	Figure A.2.35 Optimistic power outage condition associated with the inundation from 1:10,000 year storm with 3m SLR ............................................................................................ 317	Figure A.2.36 Pessimistic power outage condition associated with the inundation from 1:10,000 year storm with 3m SLR ............................................................................................ 318	Figure A.2.37 Optimistic power outage condition associated with the inundation from 1:10,000 year storm with 4m SLR ............................................................................................ 319	  xxx Figure A.2.38 Pessimistic power outage condition associated with the inundation from 1:10,000 year storm with 4m SLR ............................................................................................ 320	Figure A.2.39 Optimistic power outage condition associated with the inundation from 1:10,000 year storm with 5m SLR ............................................................................................ 321	Figure A.2.40 Pessimistic power outage condition associated with the inundation from 1:10,000 year storm with 5m SLR ............................................................................................ 322	Figure A.2.41 Optimistic power outage condition associated with the inundation from 1:10,000 year storm with 6m SLR ............................................................................................ 323	Figure A.2.42 Pessimistic power outage condition associated with the inundation from 1:10,000 year storm with 6m SLR ............................................................................................ 324	Figure C.1.1 Each bar chart shows the percentage of all scenarios represented by the given RIP that is characterized by the respective storm intensity condition. .............................. 348	Figure C.1.2 Each bar chart shows the percentage of all scenarios represented by the given RIP that is characterized by the respective sea-level rise (SLR) condition. ...................... 349	Figure C.1.3 Each bar chart shows the percentage of all scenarios represented by the given RIP that is characterized by the respective land-use condition. ........................................ 350	Figure C.1.4 Each bar chart shows the percentage of all scenarios represented by the given RIP that is characterized by the respective power outage condition. ................................ 351	Figure C.1.5 Each bar chart shows the percentage of all scenarios represented by the given RIP that is characterized by the respective building vulnerability condition. ................... 352	Figure D.4.1 Subgroup pair A population pyramids for survey question A1 that consists of the statement: “The impact patterns can help prioritize SLR adaptation efforts and resources” ................................................................................................................... 403	  xxxi Figure D.4.2 Subgroup pair A population pyramids for survey question A2 that consists of the statement: “The impact patterns can help generate new ideas of SLR adaptation options as they provide information on impacts that I had not considered before” ... 403	Figure D.4.3 Subgroup pair A population pyramids for survey question A3 that consists of the statement: “The impact patterns can help develop more refined and targeted adaptation options.” ..................................................................................................................... 404	Figure D.4.4 Subgroup pair A population pyramids for survey question A4 that consists of the statement: “The impact patterns can be useful tools for communicating the risk of sea-level rise to potentially affected population and organizations.” ............................... 404	Figure D.4.5 Subgroup pair A population pyramids for survey question B1a that consists of the statement: “By showing the potential impacts in futures beyond the common worse case scenario (2m of sea-level rise), the impact patterns can support long range planning of how to modify the options to respond to a worse situation (e.g. intense storms become more frequent).”  ............................................................................... 405	Figure D.4.6 Subgroup pair A population pyramids for survey question B1b that consists of the statement: “By showing the potential impacts in futures beyond the common worse case scenario (2m of sea-level rise), the impact patterns can support justification for planning for a worse scenario than those suggested in provincial guidelines.” ......... 405	Figure D.4.7 Subgroup pair A population pyramids for survey question B2a that consists of the statement: “The diversity of impacts shown in the patterns can support development of a wider range of adaptation options (e.g. commercial business continuity programs, backflow valve installation programs).” .................................................................... 406	  xxxii Figure D.4.8 Subgroup pair A population pyramids for survey question B2b that consists of the statement: “The diversity of impacts shown in the patterns can highlight new types of stakeholders to be engaged in SLR adaptation planning.”  ........................................ 406	Figure D.4.10 Subgroup pair A population pyramids for survey question B3b that consists of the statement: “The robustness of the impact patterns can provide better leverage to request for resources to support implementation.” ..................................................... 407	Figure D.5.1 Subgroup pair B population pyramids for survey question A1 that consists of the statement: “The impact patterns can help prioritize SLR adaptation efforts and resources” ................................................................................................................... 408	Figure D.5.2 Subgroup pair B population pyramids for survey question A2 that consists of the statement: “The impact patterns can help generate new ideas of SLR adaptation options as they provide information on impacts that I had not considered before” ... 408	Figure D.5.3 Subgroup pair B population pyramids for survey question A3 that consists of the statement: “The impact patterns can help develop more refined and targeted adaptation options.” ..................................................................................................................... 409	Figure D.5.4 Subgroup pair B population pyramids for survey question A4 that consists of the statement: “The impact patterns can be useful tools for communicating the risk of sea-level rise to potentially affected population and organizations.” ............................... 409	Figure D.5.5 Subgroup pair B population pyramids for survey question B1a that consists of the statement: “By showing the potential impacts in futures beyond the common worse case scenario (2m of sea-level rise), the impact patterns can support long range planning of how to modify the options to respond to a worse situation (e.g. intense storms become more frequent).” ................................................................................ 410	  xxxiii Figure D.5.6 Subgroup pair B population pyramids for survey question B1b that consists of the statement: “By showing the potential impacts in futures beyond the common worse case scenario (2m of sea-level rise), the impact patterns can support justification for planning for a worse scenario than those suggested in provincial guidelines.” ......... 410	Figure D.5.7 Subgroup pair B population pyramids for survey question B2a that consists of the statement: “The diversity of impacts shown in the patterns can support development of a wider range of adaptation options (e.g. commercial business continuity programs, backflow valve installation programs).” .................................................................... 411	Figure D.5.8 Subgroup pair B population pyramids for survey question B2b that consists of the statement: “The diversity of impacts shown in the patterns can highlight new types of stakeholders to be engaged in SLR adaptation planning.” ......................................... 411	Figure D.5.9 Subgroup pair B population pyramids for survey question B3a that consists of the statement: “The robustness of the impact patterns can support development of adaptation options that are more uncertainty-tolerant.” ............................................. 412	Figure D.5.10 Subgroup pair B population pyramids for survey question B3b that consists of the statement: “The robustness of the impact patterns can provide better leverage to request for resources to support implementation.” ..................................................... 412	    xxxiv List of Abbreviations AP – Adaptation pathways APM – Adaptive Policymaking ATP – Adaptation Tipping Points CoV – City of Vancouver CFRA – Coastal Flood Risk Assessment DAPP – Dynamic Adaptive Policy Planning DB – Dissemination block IG – Info-Gap RIPs – Robust impact patterns RDM – Robust decision-making RR – Relative robustness SBDPI – Sewage Backup Damage Potential Index SDF – Stage damage function SLR – Sea-level rise SNA – Scenario-neutral Approach SOMs – Self-organizing maps   xxxv Acknowledgements This research was funded by the Marine Environmental Observation, Prediction, and Response (MEOPAR) Network Centre of Excellence, UBC’s Bridge Program, and Pacific Institute for Climate Solutions (PICs).   This research would not have been possible without the in-kind support provided by a range of organizations and experts. I would like to thank the City of Vancouver’s Department of Planning, Urban Design, and Sustainability for their partnership on the research presented in Chapter 4, 5, and 6. This research has greatly benefited from the time and support provided by Tamsin Mills, Brad Badelt, and Angela Danyluk of the City’s Sustainability Group. I am particularly grateful for the knowledge and collaboration opportunity with Phil White and John Maciver of Development, Buildings and Licensing - their generous and whole-hearted support cannot be overstated. Many thanks to William Chow, Carol Wagner and Nicky Hastings for providing extensive technical support for Hazus-Canada, which created a critical foundation for the impact modelling effort in this research. I would also like to thank BC Hydro and staff members – Bill Wheeler; Garry Walls; Ed Burt; Ed Mah; Lana Gilpin-Jackson - for sharing their expert knowledge and data, which was critical in multiple ways. The impact modelling in this research has also benefited from the data provided by the Ministry of Family and Children.  Special thanks to all the expert participants of the Experts Workshop in the study presented in Chapter 6, as well as those that provided their expert knowledge that were critical for the development of several novel methodological components of this research. The success of the   xxxvi Expert Workshop would not have been possible without the invaluable assistance provided by Sara Muir Owen, Tugce Conger, and Alexa Tanner.  I am extremely grateful for my supervisor, Prof. Stephanie Chang, for guiding me through this Ph.D. journey with patience and trust. Her ability to think steps beyond the immediate and to uphold high scientific integrity throughout her research has inspired and made me a better scientist. I would also like to thank my committee members, Prof. Karen Bartlett and Prof. Tim McDaniels for sharing their unique perspectives to support my research and guiding me through this journey with kindness.  This Ph.D. journey at IRES was made up of many ups and downs, and I have been very fortunate to have a group of kind, bright, and inspiring friends to pick me up from the downs and celebrate the ups together: Ther Aung, Ghazal Ebrahimi, Tugce Conger, Guillaume Peterson St. Laurent, Nancy Silver, Manuel Colombo, Jin Qiu, Lucy Rodina, Alejandra Echeverri, Liz Williams, Alicia Lavalle, Olga Petrov, Emily Rugel, Angela Eykelbosh, Silja Hund, Alicia Speratti, Poushali Maji, Greg Oulahen, Hana Galal, Polly Ng, Erik Blair, Ruth Legg, Jennifer Romero, and many others. Your friendships have helped bring me to this point in one way or another – thank you!  Even though they are far away from Vancouver, I want to thank Emma Venters, Paloma Borque, Xue Meng Chen, Christopher Simmons, Lisa Coop, Oliver Crespo, Chris Jack, Rory Grandin, Christopher Lennard, and Jane Lennard-Battersby for their loving support and believing in me when I didn’t believe in myself.   xxxvii  I am deeply thankful for my parents for their ceaseless support, love, understanding, and wisdom. Thank you for encouraging me to challenge myself even when your instinct is to protect me from all adversities in life.  Finally, Kevin (Muppy), my husband and best friend who has never let go of my hand. Thank you for being my morning sunshine, teaching me ways to be good to myself, and always being my home no matter where we are in this world.   xxxviii Dedication For my late sister, Hazel,  who did not get a chance to live her dreams but always sailed beside me. 1 Chapter 1: Introduction 1.1 Background Sea-level rise (SLR) is one of the most concerning and certain effects of the warming climate. It is associated with significant economic, social, and environmental adverse impacts on coastal communities across the globe where there is high and growing concentration of population and assets (Meehl et al., 2007). A study by Hallegatte and colleagues (2013) assessed the current and future flood losses of 136 coastal cities around the world and found an eightfold increase in the current average global flood losses by 2050 due to increasing coastal concentration of assets, SLR, and subsidence - from approximately US$6 billion per year in 2005, to US$52 billion per year by 2050. While SLR is often framed as an impending peril, its effects are already observed in many cities where rising sea-levels have increased the magnitude of tidal floods, creating more frequent nuisance flooding. For example, many cities on the East Coast and Gulf Coast of the United States, such as Norfolk, Wilmington, and Miami, have stretches of roads that would flood regularly and sewers would backup onto streets in Charleston during high tides (Corum, 2016). In response, Miami is investing US$100 million to elevate existing structural flood protection, raise streets and install additional pumps (Flechas, 2017).  Although most industrialized countries are taking actions to mitigate the effects of climate change, due to the lag time in the global climate system, coastal cities around the world will still be subjected to the impact of SLR in the next few decades regardless of our actions today (Dessai & Hulme, 2001; Meehl et al., 2007; Pittock & Jones, 2000). Additionally, considering  2 how extensively SLR can impact the society and how such impacts are already observed, the impetus to plan for SLR adaptation is steadily growing.   In contrast with climate mitigation that has a global effect, the local level or municipal government is increasingly recognized to be in a good position to conduct climate adaptation for two key reasons (Dannevig & Aall, 2015; Hanssen, Mydske, & Dahle, 2013; Mukheibir, Kuruppu, Gero, & Herriman, 2013; Termeer et al., 2011). Firstly, the nature of the impacts (and many other climate impacts) are strongly shaped by the local contexts, which determine the community’s exposure, vulnerability, and capacity to deal with the impacts (Adger & Kelly, 1999; Cutter, Mitchell, & Scott, 2000; Radhakrishnan et al., 2018; Turner et al., 2003). In particular, the spatial variations of the local impacts play an important role in developing effective actions (Storch & Downes, 2011), given that the experience of SLR impact is often multidimensional, non-linear, and heterogeneous contextually and spatially within a community (Berkes, Colding, Folke, & Cambridge, 2003; Nicholls, 2011). Therefore, effective adaptation actions should be place-based and tailored with these local and spatial contexts in mind. Secondly and consequently, the local government often has the responsibility and legitimacy to structure actions to address the local impacts (Measham et al., 2011).  While there is a sound argument to conduct adaptation at the local level, it is also associated with many challenges. One of the biggest challenges lies in the deep uncertainties of SLR impacts at the local level (Klenk, MacLellan, Reeder, & Flueraru, 2018; Lawrence, Bell, Blackett, Stephens, & Allan, 2018). While there is strong scientific evidence that the sea-level will rise with the changing climate, there is deep uncertainty in how much the sea-level will rise and by  3 when. This uncertainty is also further compounded at the local level by place-specific factors (e.g., topography, infrastructure interdependencies) and long-term changes in non-climatic drivers (e.g., land-use change, population growth), all of which can significantly shape the magnitude and extent of SLR impacts (Wilby & Dessai, 2010).   1.2 Problem statement In response to the deep uncertainty of long-term local climate impacts, recent literature is encouraging the shift from the conventional predict-then-act approach that focuses on using the a small set of best available predictions to find the optimal strategy (i.e., lowest cost and highest benefit) to scenario-based approaches that focus on informing decisions under uncertain conditions to avoid surprises and identifying actions that are robust rather than optimal (e.g., Wilby & Dessai, 2010; Lempert et al., 2013). A robust strategy is one that can perform reasonably well (rather than optimally) over a wide range of plausible futures (Lempert, 2013; Wilby & Dessai, 2010), such that it is more likely to succeed even when the future digresses from the current prediction. To support the development of robust options or plans, there is now a growing host of conceptual frameworks that aim to improve the robustness of an initial set of preferred adaptation options (i.e., a basic plan) through a series of steps. By design, these frameworks are often applied in the options evaluation stage of adaptation planning.   While each of these conceptual frameworks has distinguishing qualities, they share the implicit assumption that the initial plan or preferred options are readily identifiable in the first step of the framework. However, the way to identify suitable or preferred adaptation options in an effective manner remains largely unclear. This can be a significant challenge since there is yet a  4 comprehensive list of adaptation options that decision-makers should consider in SLR adaptation planning and little is known about why communities adopt certain adaptation options over others (Brody, Kang, & Bernhardt, 2010), such that the basic adaptation plan may be composed of generic or extensions of existing options rather than options that are designed to address local impacts and vulnerability factors. Furthermore, in principle, selecting an initial set of options that recognizes local contexts and uncertainties of the multiplicity regarding long-term impacts in the early stage of planning would save the user from wasting time on appraisal and refinement of unsuitable options. From a theoretical point of view, such an initial plan would provide a better starting point for further refinement for robustness using one of the existing robust adaptation frameworks. However, there is no method as yet to allow users to efficiently understand the uncertainties by allowing the user to assess and consider the long-term climate impacts of a large range of plausible futures at the local level without being overwhelmed by the sheer amount of information.  1.3 Research approach and thesis structure A more detailed review of the relevant literature and a substantiated argument for the research gap is provided in Chapter 2. In the light of the above research gap, this research develops a new method – Robust Impact Patterns (RIPs) method – to assess the potential impacts of  SLR and storms under a large range of plausible futures, in a concise, spatially explicit, and visualized manner. Specifically, this dissertation addresses the following two research questions: 1. How can the multi-dimensional nature of flood impacts and the deep uncertainty of long-term SLR impacts be presented to support robust adaptation?  5 2. How can the Robust Impact Patterns (RIPs) method and results support SLR adaptation planning from the prospective users’ perspective? The multi-dimensionality of flood impacts in the first research question refers to how flood impacts can manifest in multiple sectors, contexts, and spatial scales, which is described in more detail in Section 2.1. To help structure the way to address the second research question, the researcher has identified a number of anticipated ways in which the RIPs can support SLR adaptation planning. These anticipated ways falls into two broad categories below, and their rationales are discussed in the last section of Chapter 3.  Adaptation options development: 1. Generate new ideas for SLR adaptation options  2. Generate more refined and targeted SLR adaptation options  3. Consider more diverse types of SLR adaptation options (e.g., soft, hard, combination) 4. Develop SLR adaptation options that are more uncertainty-tolerant 5. Facilitate for long-range planning of how to adjust current options to respond to a worse situation  Resources and support for implementation: 1. Prioritize SLR adaptation efforts and resources 2. Serve as a useful tool for communicating SLR risk  3. Identify new types of stakeholders to engage in planning  4. Provide justification for planning beyond the common worse case scenario 5. Provide better leverage to request for resources  6  To address these research questions, this dissertation consists of three integrative components. The first component addresses the first research question by developing a new approach – Robust Impact Patterns (RIPs) method - to assess and envision the potential impacts of coastal flooding in the context of SLR, under a large range of plausible futures where no risk reduction actions are taken. The goal is to help users start accounting for the uncertainties, local context, and multidimensionality of the potential impacts in the early stage of SLR adaptation planning where their understanding can shape their subsequent selection of preliminary adaptation options. Given that a key purpose of the RIPs method is to inform the selection of initial adaptation options, it aims to present a portfolio of potential SLR impacts, rather than predict the actual or residual impacts. The potential impact generally refers to impacts that can take place when no form of adaptation action takes place, including autonomous, reactive, or proactive actions, while actual or residual impact refers to the most likely way the impact will unfold, which would need to account for the most likely actions that will take place by the timeframe of the prediction (R. J. Nicholls, 2011). This study component is presented in Chapter 3, describing the 1) thought process behind the method development (Section 3.1); 2) the steps involved in implementing the RIPs method (Section 3.2), and 3) the intended utility of the method (Section 3.3).  The second component applies the RIPs method at the City of Vancouver (CoV), as a case study, to demonstrate the feasibility of the method and provide the basis on which to examine whether the RIPs method can fulfill its intended purpose of supporting more effective SLR adaptation. Therefore, this component in part addresses both research questions. Chapter 4 presents the methodology of this component by starting with an overview of the CoV and its current SLR  7 adaptation efforts (Section 4.1 to 4.3), which is followed by sections describing how each step of the RIPs method was implemented for the CoV (Section 4.5 to 4.7). Chapter 5 presents the resulting RIPs of each assessed impact, and concluding remarks that highlight the key findings and issues that one may need to manage when applying the RIPs method in practice.  The third component addresses the second research question by examining whether the RIPs method can support SLR adaptation planning from the prospective users’ point of view. This research component used the CoV case study as the basis for this investigation. Therefore the prospective users are a group of experts currently involved in SLR adaptation planning at the CoV or experts of the CoV's infrastructure and systems that are at imminent risk of being affected by SLR. Chapter 6 presents this component, starting with an overview of the approach taken to conduct this evaluation (Section 6.2). This is followed by the results from a high-level statistical analysis that provides a general sense of whether or not the participants think the RIPs method can support SLR adaptation planning (Section 6.3.1). A more rigorous qualitative data analysis is also provided (Section 6.3.2 to 6.3.12), which reveals more nuanced contexts and examples of how the RIP method can or cannot support adaptation planning. The concluding remarks of this chapter summarize the key findings and highlight the insights revealed through this research component, which include new limitations, opportunities, and modification suggestions (Section 6.4).  The conclusion and discussion of this research is presented in Chapter 7, which begins by outlining the key findings of each research component and highlighting their respective original contributions and significance in the broader context (Section 7.1). This is followed by a section  8 (Section 7.2) describing the key limitations of the RIPs method as a tool to help users account for uncertainties and local contexts of SLR impacts, as well as the general limitations of each research component. More case-specific and technical limitations are outlined in the chapters of each respective research components. The last section (Section 7.3) describes the future research directions generated from considering the identified limitations and opportunities of the RIPs method.  1.4 Significance By developing, demonstrating, and evaluating the RIPs method, these three research components collectively form an end-to-end process that contributes to the area of SLR adaptation in both practical and theoretical ways.   From a practical point of view, this research provides a new approach to pragmatically address the need for a concise and visual way to help users incorporate uncertainties into their understanding and planning of SLR adaptation at the local level. Furthermore, the application of the RIPs method at the City of Vancouver provides new information that significantly broadens and deepens our understanding about the potential impacts of coastal flooding and SLR in the City, which informs their ongoing SLR adaptation planning efforts.   As a theoretical contribution, this research adds to the adaptation literature, as well as the flood risk management literature. In the adaptation literature, scenario-based approaches are gaining momentum, as opposed to the conventional predict-then-act approaches where users rely on the single best available prediction to identify the optimal solution. These scenario-based approaches  9 are broadly represented by a growing group of conceptual frameworks (e.g., Robust Decision-Making, Dynamic Adaptive Policy Pathways, Scenario-neutral Approach, etc.) aiming to help users manage the deep uncertainty of impacts by improving the robustness of adaptation options or plans. Given the conceptual nature of these frameworks, a range of computational tools is used to implement these frameworks. The RIPs method adds to the small repertoire of tools (e.g., Stephens, Bell, & Lawrence, 2017) that are designed to support the implementation of these conceptual frameworks. Although the conceptual frameworks for developing robust adaptation plans have gained increasing interest in research, the integration of these concepts into adaptation in practice requires a paradigm shift from the traditional predict-then-act approach to those that help users manage uncertainties and plan for surprises rather than finding the lowest cost-highest benefit solution. The demonstration and evaluation of the RIPs method provides significant evidence to argue for this much-needed paradigm shift in the way climate adaptation planning should be conducted and thought of.   In the flood risk management literature, indirect impacts of flooding and SLR are less understood and seldom included in risk assessments in comparison to direct impacts (e.g., Carrera et al., 2015). By intentionally assessing a number of indirect economic, social, and environmental impacts of SLR in the case study at the CoV, this research also aims to improve our understanding of potential indirect flood impacts, which often accounts for a significant portion of the full cost of disasters. An example of indirect impacts is sewage backup, which has the potential to cause extensive damage but there is to date no published measure to quantify the potential damage from sewage backup associated with overland flooding. Therefore, the development of a new composite index to measure the relative damage potential of sewage  10 backup at the street block scale is a significant original contribution to the flood risk management literature as well as the repertoire of models to assess flood impacts at the local scale.  A more detailed discussion and additional original contributions specific to the research findings are presented in Section 7.1 of Chapter 7.    11 Chapter 2: Background and Contexts The purpose of this chapter is to provide an overview of the existing literature in a manner that leads up to the research gap being addressed by this dissertation research. Specifically, this chapter begins with an overview of the socio-economic impacts of SLR and how the nature of the impacts points to the need to account for their spatial variability and local context when planning for adaptation (Section 2.1 and Section 2.2). This is followed by the key challenges in planning for SLR adaptation, one of which is the core issue being addressed by this research - managing the deep uncertainties in the impacts of SLR in the long-term future. Section 2.3 describes the deep uncertainties surrounding SLR projections and impacts, and how it has been treated in impact assessments. Section 2.4 outlines the existing frameworks for developing robust adaptation plans, which can be applied to manage the deep uncertainties of SLR impacts. This provides the basis to define the research gap (Section 2.5).  2.1 Sea-level rise impacts and adaptation SLR is one of the most concerning effects of the warming climate. Although there are uncertainties around how much the sea-level will rise and by when, there is strong evidence indicating that SLR will pose severe challenges to coastal communities through direct and indirect adverse impacts (Meehl et al., 2007). Direct impacts of SLR include: increasing erosion and permanent inundation of infrastructure and ecosystems; more frequent and extreme coastal flooding; and salt-water intrusion within rivers and aquifers at the coastal areas (Nicholls et al.,  12 2007). These physical changes due to SLR also pose adverse socio-economic impacts on multiple sectors, as summarized by Nicholls (2011) in Table 2.1.    Table 2.1 Summary of SLR impacts on socio-economic sectors in coastal areas. (Table 2 from © Nicholls (2011) Planning for the impacts of sea level rise. Oceanography, 24(2), 144-157, Page 149. By permission from publisher.)   As shown in well-documented flood events, such as Hurricane Sandy (e.g., Henry et al. (2013); Lane et al. (2013)) and cases in the UK (e.g., Tapsell, 2001; Tunstall et al., 2006), direct flood impacts result in direct damage and losses, but they also cascade and result in a complex and diverse range of indirect impacts. For example, inundation of electrical power infrastructure, such as substations or transformers, results in power outages, which can in turn cause major disruptions to various critical infrastructures (e.g., water treatment, hospitals) and businesses. Therefore, rather than referring to how failure within a system can trigger subsequent failures, “cascading effect” in this context refers to how direct impacts of SLR (e.g., inundation) can consequently trigger other types of impacts, often across different systems or sectors.   13 Broadly speaking, the direct and indirect impacts of SLR fall into three major categories – economic impacts, social impacts, and environmental impacts. Economic impacts include the costs from repair and rebuild of damaged infrastructure, business disruptions, and loss of opportunities (e.g., erosion of beaches may severely reduce opportunities in tourism) (Henry et al., 2013). All of which can pose severe shock to the local economy. Although most risk assessment focuses on economic impacts, the damages from flooding also have strong social implications, such as displacement of vulnerable population, loss of access to health care (Alderman, Turner, & Tong, 2012) and other social services (e.g., addiction centres, free-meal locations). The experience of displacement and loss is also associated with high stress and exacerbation of mental illnesses (Carroll, Morbey, Balogh, & Araoz, 2009; Paranjothy et al., 2011). It is also known that financial losses can further exacerbate existing chronic illnesses when the financial stress causes one to cut back on receiving regular treatments (Kessler, 2007). SLR also affects the natural environment, as well as the human environment. The increase in inundated coastal areas is associated with increased erosion of shoreline systems and destruction of ecological habitats (e.g., wetlands), which contributes to the loss of biodiversity (Galbraith et al., 2002). In the human environment, SLR can adversely impact the human living environment through increasing risk of indoor flooding, sewage back-up inside homes (Sandink, 2016), and exposure to flood debris, which poses significant health risks by exposure to various contaminants carried in debris and sewage (Gooré Bi, Monette, Gachon, Gaspéri, & Perrodin, 2015; Lane et al., 2013). Furthermore, water-damaged homes have high risk of mould and bacteria growth, where human exposure through contact or inhalation is associated with exacerbated pre-existing allergic symptoms (e.g., fever, shortness of breath) and asthmatic symptoms, and potentially leads to new respiratory infections such as, upper respiratory tract  14 symptoms, cough, and wheeze in people with no pre-existing respiratory conditions (National Center for Environmental Health, 2009).  While most industrialized countries are taking actions to reduce greenhouse gas emission to mitigate the effects of climate change, due to the lag time in the global climate system, coastal cities around the world will still be subjected to the impact of SLR in the next few decades regardless of our actions today (Dessai & Hulme, 2001; Meehl et al., 2007; Pittock & Jones, 2000). Furthermore, considering how extensively SLR can impact the society and how the impacts are already being observed (e.g., Florida) (Florida Oceans and Coastal Council, 2010), there is an urgent need for implementing effective SLR adaptation.  2.2 Importance of local and spatial context in sea-level rise impacts In recent years, the local government (i.e., municipal government) is increasingly recognized as where adaptation planning needs to take place (Dannevig & Aall, 2015; Hanssen et al., 2013; Mukheibir et al., 2013; Termeer et al., 2011). The importance of planning adaptation at the local-level can be seen from the problem’s (i.e., SLR impacts) and the solution’s (i.e., adaptation actions) point of view. Starting with the impacts of SLR, as for many other impacts of climate change, are experienced locally. This means that the magnitude and extent of impact varies geographically and spatially, such that place-based actions are required and critically important (Adger & Kelly, 1999; Cutter et al., 2000; Radhakrishnan et al., 2018; Turner et al., 2003). Impacts are experienced locally not only because the level of SLR at the local-level is driven by different factors compared with the global mean sea-levels (Nicholls & Cazenave, 2010), but also because different places have different levels of exposure (what is exposed to the direct or  15 indirect impacts of flooding), vulnerability (propensity to suffer from losses), and capacity (ability to deal with impacts) (Cutter et al., 2008). The emphasis on adapting at the local-level also stems from how the local government is often the institution that has the responsibility and legitimacy to manage the impacts (Measham et al., 2011). Specifically, they are often in the position to structure the responses to local impacts; deliver and govern the resources for adaptation; and connect the individual and collective responses in a coherent manner (Agrawal, 2008).  Although adaptation planning at the local-level (municipal) is sound in principle and may produce more sensitive and effective actions, it will fall short if the spatial context of SLR impacts within the place is not accounted for. Spatial variations of flood impacts can play an important role in supporting adaptation planning and decision-making – from resource prioritization to evacuation and response. For example, spatial distribution of flood exposure is critical for developing effective land-use zoning, which is considered as an important strategy for sustainable adaptation (Storch & Downes, 2011). Spatially explicit flood impacts across a region of interest can also better capture the cascading, or otherwise hidden impacts caused by infrastructure interdependency within a city or region (Storch & Downes, 2011). For example, flooding of a small area causing inundation of a component of a power transmission system can result in power outage in areas beyond the flooded areas and cause additional disruption to the infrastructure and population in those areas (Eun Ho, Deshmukh, & Hastak, 2010; O'Rourke, 2007). Rather than considering a single outcome measure aggregated across the region (e.g., total population displaced), visualizing the spatial relationships and impacts through maps also facilitate for easier interpretation of the impacts, their causal relationships, and patterns  16 (Maceachren & Ganter, 1990). This may in turn be helpful in developing an appropriate response or adaptation strategy. Presenting the spatial information of the flood impact in a visual manner, such as in an impact map, is also important from the communication perspective, as it promotes engagement by helping the impacts to be more concrete to the stakeholders by providing the basis to envision the linkage between the impacts and the processes relevant to their respective lives (Sheppard et al., 2011). Another major challenge of local-level adaptation planning is the simplistic conception of the local community as a homogenous and spatially fixed population (Measham et al., 2011).   The experience of SLR impact is highly complex, non-linear, and heterogeneous within a community (Berkes et al., 2003; Nicholls, 2011). This complexity highlights the type of climate impact information that policy-makers and planners need in order to effectively plan for local adaptation. SLR projection information needs to be translated into local-level SLR impact information and presented as policy-relevant and spatially explicit indicators to fit well within existing spatial planning processes, which is where most adaptation takes place (Goosen et al., 2014; Measham et al., 2011).    2.3 Deep uncertainties of sea-level rise and its representation in impact assessments As illustrated in the previous section, effective local adaptation planning requires an understanding of SLR as a problem at the local level, which needs information that accounts for the spatial variability and multi-dimensionality of SLR impacts. Coincidentally, such information also represents another key challenge – the uncertainties of SLR impacts (Klenk et al., 2018; Lawrence et al., 2018). The risk or impact assessments of climate change effects, including SLR,  17 are subjected to a cascade of uncertainties (Wilby & Dessai, 2010) where the envelope and sources of uncertainty increases at each step of the assessment process [Figure 2.1]. Starting with long-term societal changes (e.g., land-use, population growth) that drive not only the exposure and vulnerability of the community, but also the long-term increase in the concentration of greenhouse gases (GHG). More uncertainty emerges in the subsequent steps in estimating how the climate will respond to the GHG emission; how much warming is associated; how will the ice sheets respond; and how will the global mean sea-level change. To bring the global mean sea-level down to the regional or local-level, additional local drivers of sea-level change (e.g., subsidence, post-glacial rebound) are included, which introduces additional uncertainties. From here, the uncertainties are further compounded with the uncertainties associated with the complexity and multi-dimensionality of SLR, rendering the uncertainties associated with long-term SLR impact information to be simply overwhelming.   This situation, where large uncertainty from multiple aspects that are only reducible with time, can be characterized as having deep uncertainty (Bakker, Louchard, & Keller, 2017), which is defined as “the condition in which analysts do not know or the parties to a decision cannot agree upon 1) the appropriate model to describe interactions among a system’s variables, 2) the probability distributions to represent uncertainty about key parameters in the models, and/or 3) how to value the desirability of alternative outcomes” (Lempert et al., 2003). This implies that a multiplicity of plausible futures representing how SLR can impact a community can be enumerated but it is not possible to determine their relative likelihood to eventuate in the future.    18 Figure 2.1 Cascade of uncertainties (Figure 1 from © Wilby & Dessai (2010) Robust adaptation to climate change. Weather, 65(7), 180-185, Page 181. By permission from publisher.)   To provide a rough sense of the uncertainty involved in the impact assessment of SLR, one can consider the range of SLR projections for the 21st century that has been published in the past 5 years. The projection of future SLR is a rapidly evolving and moving research topic where new studies are produced regularly (Slangen et al., 2017). Based on studies published in the past 5 years, the upper limit of global mean SLR projections for the timeframe of 2100 using the RCP 8.5 scenario (or equivalent amount of warming) ranges from 0.98m to 6m [Figure 2.2]. The question then is - which level of SLR should my community plan for? In principle, one can choose to use the projections produced by models that have better skills and credibility. However, opinions regarding the credibility of different models and approaches are divergent amongst scientists (Bakker et al., 2017). For example, there is no consensus about the relative reliability of the two major types of models used for projecting sea-level change – process-based  19 models and semi-empirical models, leaving limited confidence in the output of either type of models (Clark, Church, Gregory, & Payne, 2015). Figure 2.2 Upper limit projections published in the past 5 years for year 2100 global mean sea level. Studies included are listed below the chart.   a)	Church	et	al.	(2013)	(IPCC	5th	Assessment	Report)	 e)	Dutton	et	al.	(2015)	b)	Kopp	et	al.	(2014)	 f)	Hansen	et	al.	(2016)	c)	Horton	et	al.	(2014)	 g)	Jackson	and	Jevrejeva	(2016)	d)	Jevrejeva	et	al.	(2014)	 	 In light of the uncertainties associated with the impacts of SLR in the long-term future, it is not surprising to find that many decision-makers claim to need more accurate and precise predictions of the future climate in order for adaptation to be effective and successful (Dessai, Hulme, Lempert, & Pielke, 2009). If this is true and we rely on the hope for better climate predictions, we will face two problems. Firstly, in the near future, it is unlikely that the uncertainties in climate prediction will be reduced to the level reliable for adaptation decisions (Allen & Frame, 2007; Roe & Baker, 2007). Secondly, climate is only one of many processes that influences  20 climate change impacts, therefore even if more accurate and precise climate predictions are produced, there remain large uncertainties in the changes in other processes such as those that drive the socio-economic states (Adger et al., 2007). Therefore SLR adaptation must take place despite the limited accuracy and deep uncertainty of SLR projections and its associated impacts.   Although there is growing awareness of the deep uncertainty surrounding SLR projections (Bakker et al., 2017), many SLR impact assessments acknowledge the uncertainties (e.g., City of Surrey, 2018; Heberger, Cooley, Herrera, Gleick, & Moore, 2011; Northwest Hydraulics Consultants, 2014) but overlook the effects on the resulting impact estimation (Ruckert, Oddo, & Keller, 2017). Specifically, these assessments often rely on the mean, consensus (e.g., IPCC, high level government recommendations), or large quantile SLR projections, which can result in considerable underestimation of associated SLR impacts and overconfidence in adaptation investments. Using San Francisco Bay area of California as a case study, Ruckert and colleagues (2017) has found the differences between using the full distribution of SLR projection, and using the mean and 90% credible interval SLR projection, can result in 0.5m and 0.2m underestimation, respectively. This underestimation trickles down further to the flood exposure estimation. In comparison to using the mean SLR projection, using the full distribution increases the average flooded areas by a factor of 2 (Ruckert et al., 2017).  Nonetheless, some studies have used statistical approaches to incorporate the probabilistic uncertainties of SLR projections into various flood depth related estimates to support SLR impact assessments. For example, Buchanan et al. (2016) incorporated the uncertainties surrounding the rate of SLR by adding an  21 additional margin of uncertainty to the amount of freeboard1 for planning different infrastructure of various lifetimes. Furthermore, as described above, Ruckert et al. (2017) accounted for the full estimated distribution of SLR projections by generating a new exceedance probability function from the average probability of each flood depth (within a defined range) across a large number of stochastic simulations of global mean sea level.   2.4 Approaches to adaptation in the face of deep uncertainties While there is not as yet a standard approach for individual communities to account for the uncertainties of SLR projections in their impact assessments, there are more advances made in conceptual frameworks developed to help decision-makers account for SLR uncertainties in the options evaluation stage of adaptation planning. Many of these frameworks stem from the concept of scenario planning.  2.4.1 Scenario planning The conventional decision-analysis frameworks focus on the idea of predict-then-act that relies heavily on assimilating expert knowledge to derive a small set of best available predictions to find the optimal strategy given these predictions (Lindgren & Bandhold, 2009; Wilkinson & Eidinow, 2008). The predict-then-act framework works well only if the predictions are reasonably accurate and non-controversial (Lempert, 2013), which is clearly not the case in climate and impact predictions. The long-term future “presents a vast multiplicity of plausible futures” such that inferences made about the future based on one or small set of these predictions are very likely to be wrong (Lempert et al., 2003; Lindgren & Bandhold, 2009; Wilkinson &                                                 1 Freeboard is defined as the “amount of buffer in height to accommodate uncertainty in the estimated design flood level” US Army Corps of Engineers. (2015). Key USACE Flood Risk  22 Eidinow, 2008). To avoid this problem, rather than relying on predictions, scenario-based approaches are being used and considered as a key tool for decision-making under uncertainty (Jones et al., 2013). To better appreciate the difference in the two approaches, it is important to note the difference between predictions and scenarios in terms of the following two issues: 1. A scenario is not a prediction of the future (IPCC, 2013), but rather “a description of how the future may unfold” (Jäger, Rothman, Anastasi, Kartha, & van Notten, 2008). This distinction is important because scenarios are used to provide information for decision making about the future when predictions fails to do so (Jones, 2010). 2. Since the purpose of scenarios is to support decision-making, the scenarios must match the needs of the users and their framing. In contrast, predictions focus on forecasting the most likely future with little regard for the decision context.   More specific to climate adaptation, two key reasons have been suggested about why scenario-based approaches are considered to be better for planning and decision-making under uncertainty. One, in contrast to the conventional linear predictions that are largely based on extrapolation of historical trends (List, 2005), scenarios are less bounded by past experiences and facilitate envisioning of future implications that are outside our comfort zones, which may help in preparing us for surprises. This is of particular importance for climate adaptation planning since not only can climate change gradually exacerbate impacts but it can also push systems to their tipping points, leading to transformative changes that require transformative adaptations (Smith, 2013).  Two, given that the uncertainties in long-term climate projections are unlikely to be significantly reduced in the near future and that long-term future socio-economic changes are commonly deemed unpredictable, scenario planning’s focus on informing decision under  23 uncertain conditions can be more effective than prediction-based approaches where reducing uncertainty and forecasting the likely future is the main focus. The “use of plausible future scenarios to inform strategic planning and decision-making” is also known as scenario planning (Wiseman, Biggs, Edwards, & Rickards, 2011).   2.4.2 Scenario-based robust adaptation conceptual frameworks When adaptation to SLR is a necessity and uncertainty is so large in the projections and impacts of SLR, the literature suggests several additional qualities that we should look for in adaptation strategies. One is the win-win situation, referring to strategies that can produce some form of ancillary benefit for the community regardless of SLR (Klein, Schipper, & Dessai, 2005; Wilby & Dessai, 2010). However, in practice, such strategies are also associated with trade-offs and their co-benefits can be difficult to quantify (Klein et al., 2005). Another quality is robustness, where the strategies can perform reasonably well (rather than optimally) over a wide range of plausible futures (Lempert, 2013; Wilby & Dessai, 2010), also characterized as more ‘uncertainty-tolerant’. Lastly, the strategies should be adaptive. Since physical and social conditions are constantly changing such that strategies that are considered robust today may not remain so with time. Therefore, strategies also need to be adaptive – flexible to permit future adjustments and even reversibility to adapt to unforeseen changes over time (Haasnoot, Middelkoop, van Beek, & van Deursen, 2011).   The literature offers at least eight major scenario-based conceptual frameworks that aim to develop robust adaptation plans – 1) Assumption-based Planning (ABP), 2) Robust Decision Making (RDM), 3) Adaptive Policymaking (APM), 4) Adaptation Pathways (AP), 5) Adaptation  24 Tipping Point (ATP), 6) Dynamic Adaptive Policy Pathways (DAPP), 7) Info-gap, and 8) Scenario-Neutral Approach (SNA). The rest of this section will briefly describe these frameworks but a more detailed review of these frameworks can be found in Walker et al. (2013). Before delving into these frameworks, it is important to note that robustness can be achieved by static robustness or dynamic robustness, where the former aims to reduce vulnerability under the widest range of conditions (or plausible futures) enumerated at a point in time, while the latter aims to achieve robustness through flexibility where changes are planned to adjust to changing conditions with time (van Drunen, Leusink, & Lasage, 2009). Furthermore, a robust plan refers to one that can perform at an acceptable level in a wide range of plausible futures of the world, as opposed to an optimal plan where it may achieve the best outcome but only within a narrowly defined window of future conditions (Walker, Haasnoot, & Kwakkel, 2013). Lastly, the term ‘strategy’ and ‘plan’ is used interchangeably as a group of adaptation actions or options.  2.4.3 Assumption-based Planning (ABP) Developed at the RAND Corporation, this 5-step framework aims to improve the adaptability and robustness of an existing plan by reducing avoidable surprises (Dewar & Cambridge, 2002). Specifically, the avoidable surprises are identified through the assessment of load-bearing assumptions – the assumptions in the plan that drives the success of the plan - and consider what would happen if the assumption is no longer valid. It focuses on load-bearing assumptions that are most uncertain by considering a wide range of plausible events within a timeframe and identify 1) shaping actions to protect it from becoming invalid, 2) signposts – an event or threshold of when the assumption would fail, and 3) hedging actions to manage the  25 consequences when the assumption fails. Figure 2.3 illustrates the 5-step process of the framework.  Figure 2.3 Basic steps and flow of the Assumption-Based Planning framework (Figure 1.1 from © Dewar (2002). Assumption-based Planning: A Tool for Reducing Avoidable Surprises. Cambridge University Press. Page 2. By permission from publisher.)  2.4.4 Robust Decision Making (RDM) RDM is a decision-making framework developed by Lempert, Popper and Banks at the RAND Corporation. Instead of improving an existing plan as in ABP, the RDM aim to evaluate candidate or preferred strategies to identify strategies that can perform reasonably well under a wide range of plausible futures as measured by a specific metric. The approach has been applied in several cases specific to SLR (e.g., Groves, Fischbach, Knopman, Johnson, & Giglio, 2014; Sriver, Lempert, Wikman-Svahn, & Keller, 2018). Drawing from Lempert et al. (2013), the RDM framework consists of four major steps that are designed to be iterated until the best strategy is identified [Figure 2.4]:   26 Figure 2.4 Four major steps of the RDM framework. (Figure 1 from © Lempert et al. (2013). Making Good Decisions Without Predictions Robust Decision Making for Planning Under Deep Uncertainty. RAND Corporation Research Briefs, RB-9701. By permission from publisher. Available at https://www.rand.org/pubs/research_briefs/RB9701.html)   Step 1: Decision Structuring  The process generally starts with decision structuring, where the candidate strategies to be considered are clarified with all decision-makers; the objectives of the decision (“what matters in this decision”) is elicited; and the key drivers affecting the performance of the evaluated strategy are identified. The candidate strategies can be a current policy or part of a proposed future plan (Matrosov, Woods, & Harou, 2013). Step 2: Scenario Generation  Given the context of what matters in this decision, it is the goal of this step to allow the decision-maker to envision the possible outcomes that can result from the candidate strategy under a wide spectrum of plausible futures. A computer model is usually employed to produce a large number of simulations showing the potential outcome of the strategy, where each simulation represents the outcome in a different plausible future. The input and output variables, are determined by the elicited key drivers and decision objectives, in step 1, respectively.   27 Step 3: Scenario Discovery  Provided the large database of modeled outcome of the candidate strategy, the goal of this step is to identify the set of plausible futures in which the proposed strategy fails to perform acceptably, hence revealing the future scenarios in which the strategy is robust (robust scenarios) and those in which it can fail.  Step 4: Trade-off analysis  Based on the scenarios where the strategy is found to be robust, the analyst and decision-makers can either examine the scenarios to develop a new strategy to address the newly identified vulnerabilities (and repeat step 1- 3) or they can feed the resulting scenarios into a trade-off analysis where different strategies are compared. The RDM process continues until decision-makers settle on a robust strategy. Between the two approaches to achieving robustness, the RDM takes the static robustness approach as adjustments to the strategy in the future to respond to changes is not an explicit component of the framework.  2.4.5 Adaptive policymaking (APM) In contrast to RDM, the APM framework aims to achieve robustness through dynamic robustness where adjustments to the plan are built-in from the get go as an explicit part of the process. Introduced by Walker, Rahman, and Cave (2001), also called Dynamic Adaptive Planning, the APM framework  consists of 5 stages [Figure 2.5]. The objectives and preferred adaptation options are identified in the first stage, which provides input to develop a basic adaptation plan in stage 2. The basic plan is then modified to become more robust in stage 3. Specifically, the vulnerabilities and opportunities in the plan are identified and addressed through 4 types of anticipatory actions: 1) mitigating actions – to reduce likely adverse effects, 2)  28 hedging actions – to reduce uncertain adverse effects, 3) seizing actions - to capitalize on likely opportunities, and 4) shaping actions – to reduce failure. Stage 4 sets up a monitoring program to monitor signposts - performance measures to be monitored and triggers – thresholds of the signposts to trigger reactive actions, including 1) reassessment of the plan, 2) corrective actions to adjust the plan for the change in conditions, 3) defensive actions to keep the plan on track, and 4) capitalizing actions to seize any opportunities from the changing conditions.  Figure 2.5 Steps within the 4 major stages of Adaptive Policymaking (APM) framework (Figure 1 from Kwakkel et al. (2010). Adaptive Airport Strategic Planning. European Journal for Transportation, Infrastructure Research, 10(3), 249-273, Page 258. By permission of publisher.)   29  2.4.6 Adaptation Tipping Points (ATP), Adaptation Pathways (AP), and Dynamic Adaptive Policy Pathways (DAPP) Although signposts and triggers in ABP and APM introduce the notion of time, it is the focus on the timing of the actions that distinguishes Adaptation Tipping Point (ATP), Adaptation Pathways (AP), and Dynamic Adaptive Policy Pathways (DAPP) from the other frameworks.   The ATP framework (Kwadijk et al., 2010) is motivated by the desire to design a strategy that is less dependent on when a given scenario would occur by focusing on conditions or magnitude of external change under which the current strategy will fail – the tipping points - and develop an alternative strategy to address that situation. This way, with time and availability of new information, the only adjustment to the strategy is the timing of the specific actions that falls under the alternative strategy. For example, in the case of SLR, the questions would be - until what level of SLR would the current drainage system remain effective? What actions must be taken after that level of SLR is reached?  The Adaptation Pathway (AP) framework (Haasnoot, Middelkoop, Offermans, van Beek, & Deursen, 2012; Haasnoot et al., 2011) emerged as the extension of the ATP, where a pathway is formed when more tipping points and new actions are sequentially generated as we consider different magnitudes of change in the external forces. The AP framework also aims to build flexibility into the overall strategy rather than individual actions through sequencing the action to allow the system to adapt to the changing conditions. By building in options for each action (i.e., alternative route) collectively, the adaptation pathway is similar to a roadmap for adaptation as  30 shown in the example in Figure 2.6. This example (Figure 2.6) shows the current situation represented by the gray line proceeds and starts to miss its target after 4 years where there are four options (A to D). Options A and D should be able to achieve the target for the next 100 years, while options B and C will require a change of path sometime within the next 100 years. If option B is selected, after reaching a tipping point around 5 years, one will need to shift to one of the other three options. If option C was selected and scenario X eventuates, then it needs to be shortened to one of the other 3 options to continue to reach the target. But option C will continue to meet the target in all other scenarios for the next 100 years.  Figure 2.6 An example of an Adaptation Pathway. (Figure 2 from © Haasnoot et al. (2013). Dynamic adaptive policy pathways: A method for crafting robust decisions for a deeply uncertain world. Global Environmental Change, 23, 485-498, Page 488. By permission from publisher.)       Integrating elements from APM, and AP, the Dynamic Adaptive Policy Pathways (DAPP) framework (Haasnoot et al., 2013) also develops adaptation pathways that are built from groups of actions that can meet the objectives and reduce vulnerabilities [Figure 2.7]. The key difference of the DAPP from other frameworks is how it extends and develops multiple adaptation  31 pathways and one or more preferred pathways and how their contingency plans collectively form the dynamic adaptation plan.  Figure 2.7 Iterative process of the Dynamic Adaptive Policy Planning (DAPP) framework (Figure 4 from © Haasnoot et al. (2013) Dynamic adaptive policy pathways: A method for crafting robust decisions for a deeply uncertain world. Global Environmental Change, 23, 485-498, Page 489. By permission of publisher)   2.4.7 Info-Gap (IG) Similar to ABP, the Info-Gap framework (Ben-Haim, 2006) also starts with a given set of actions rather than modifying or generating the actions through the process. However, instead of analyzing the assumptions of the plan, the IG framework computationally analyzes how the actions’ performance changes as uncertainty grows to estimate the actions’ robustness - minimum performance above the threshold for a given level of uncertainty – and opportuneness -  32 maximum level of performance above the threshold for a given level of uncertainty. The result is a visualization of the actions’ robustness and opportuneness as a function of uncertainty. Figure 2.8 Flow diagram showing the process of the scenario-neutral approach to adaptation planning (Figure 3 from Wilby & Dessai (2010). Robust adaptation to climate change. Weather, 65(7), 180-185, Page 183. By permission from publisher.)    33 2.4.8 Scenario-neutral approach (SNA) Similar to all the other frameworks above, the SNA framework (Wilby & Dessai, 2010) begins with identifying a set of adaptation options. However, the SNA framework specifically requires this initial inventory of options to be ‘low regret’ or ‘win-win situations’. The main purpose of SNA is to start with a large inventory of low-regret actions and take steps to sift out the options that are robust to climate change. This sifting process involves several steps [Figure 2.8]. Firstly, screening and appraisal is conducted to identify a subset of preferred options that can reduce vulnerability based on the current climate regime and are socially, technically, and economically feasible. If there are options that have a lifetime that spans through decades (e.g., new drainage systems), sensitivity testing of these options will be based on scenarios, which can be developed from: 1) upper and lower bounds in regional climate downscaled climate information, and/or 2) climate narratives, and/or 3) quantitative impact models. This is where non-climatic drivers can also be incorporated to assess how well these options reduced vulnerability. The options that pass the testing are deemed robust to climate change. With continuous and careful monitoring of environmental change, systematic appraisal of the options as new information become available can help the set of options to evolve and result in adaptation pathways with time. For example, an alternative option may be triggered by certain environmental change. This is similar to the notion of signpost and triggers in ABP and APM.   2.5 Research gap As outlined above, there are now conceptual frameworks that aim to guide the development of more robust and adaptive adaptation plans even when there is deep uncertainty in the future climate. While each of these frameworks has distinguishing conceptual qualities, they also share  34 similarities. One such similarity is their bottom-up resilience approach as opposed to top-down approaches that start from what can happen under different climate scenarios then develop the appropriate adaptation actions. These frameworks start on the opposite end by focusing on the actions or plan, then proceed to go through steps to assess the performance of the plan under different scenarios or futures, then identify changes or additional actions to increase the robustness of the plan either through static robustness or dynamic robustness. Therefore, they implicitly assume that such a basic plan or preferences are known or ready to be compiled as a basic plan in the first step of the framework; whereas in practice, these are not often known (Burton, Huq, Lim, Pilifosova, & Schipper, 2002). In fact, little is known about why communities adopt certain adaptation options over others (Brody et al., 2010), such that the basic plan may be composed of generic or existing options rather than adaptation options that are designed to account for local impacts and vulnerability factors. Although, an alternative is to start a list that is as complete as possible of all viable adaptation options and subsequently filter out those that are not robust as in the SNA framework (Wilby & Dessai, 2010). Furthermore, it is unlikely that uncertainty of the impacts – impacts under a wide range of plausible futures - is considered in the process of generating this basic plan since it is common practice to take the top-down approach where the impacts of a small number of future scenarios or projections are considered and adaptation measures are developed to address and meet key objectives (Kwadijk et al., 2010; Walker et al., 2013).   In principle the composition of the initial basic plan does not matter if the process of the framework for plan development will modify it in any case to increase its robustness. However, it can be argued that a basic plan that is made up of options that are selected through a process  35 that considers the uncertainties, and local and multi-dimensional contexts of SLR impacts can provide better starting points to the robust adaptation conceptual frameworks. Therefore, this thesis proposes a new SLR impact assessment approach – the Robust Impact Patterns (RIPs) method - to complement and support these conceptual frameworks by allowing the users to consider uncertainties at an earlier stage – the impact assessment stage – that can help them visualize and better understand the complexity and uncertainties in the local impacts of SLR, which may help in prioritizing their efforts and inform the identification of initial adaptation options that are specific to their community. Specifically, the RIPs method will provide information representing SLR impacts that can eventuate across a wide range of plausible futures – i.e., robust impacts - in a spatially explicit manner and at a resolution that can resolve within-community variability, as well as accounting for the multi-dimensionality and multi-sectorial nature of SLR impacts. These spatially explicit impact patterns are called Robust Impact Patterns (RIPs).   36 Chapter 3: Development of the Robust Impact Patterns (RIPs) method  3.1 Method development – the thought process This thesis aims to develop a computational approach that provides SLR impact information for analysts or decision-makers to consider at the early stage of adaptation planning where they need to understand the severity, spatial distribution, multi-dimensionality and uncertainties of the SLR socio-economic impacts in their community to inform their selection of an initial preferred set of adaptation options. The intention is for such preliminary options to become a basic plan to serve as input for one of the conceptual frameworks (e.g., Dynamic Adaptive Policy Pathways, Assumption-based planning) described in the previous chapter to be refined and modified for improved robustness. The developed approach is called Robust Impact Patterns (RIPs) method. This chapter begins by describing the thinking behind the development of the RIPs method and provides details about the steps involved in implementing the RIPs method, addressing questions such as – What is involved computationally? What issues should one be considering at each step of the process? Since this chapter will not describe the RIPs method in the context of a specific case, some details and issues about implementing the approach will only be discussed in the City of Vancouver case study where the RIPs method is applied and described in the next chapter.   37 3.1.1 Why Robust Impact Patterns? In comparison with the conventional impact assessment methods that consider the “most likely” future or a handful of plausible future scenarios, this approach assesses the potential SLR impact across a large range of plausible futures to allow the analysts or decision-makers to start accounting for the uncertainties of SLR impacts. As discussed in the previous chapter, besides the uncertainties, adaptation to SLR must also account for three key aspects of SLR impacts - 1) the multi-dimensionality, 2) local context, and 3) spatial distribution. With this in mind, the proposed approach will address these three aspects in the following ways:  Multi-dimensional and multi-sectorial nature of impacts To try to capture some of the complexity of SLR impacts on the coastal community or region, the proposed approach will assess the economic, social, and environmental impacts of SLR. Given that many of such impacts are indirect impacts resulting from direct impacts, this method will assess both direct and indirect impacts.  Local and spatial distribution Covering the spatial extent of the local government’s jurisdiction, multiple SLR impacts are modeled geospatially (i.e., spatially explicit) and at a resolution that can resolve variations within the community or region.  Uncertainties  To try to account for a wide variety of uncertainties, the method will model the potential SLR impacts over a large number of plausible future scenarios (hundreds) characterized by different  38 sources of relevant uncertainties to span the range in which SLR impacts can occur in the long-term futures. Given that the method assesses economic, social, and environmental impacts resulting directly or indirectly from coastal flooding, the future scenarios will include uncertainties that are relevant at different points along the impact pathway - starting from uncertainties in the drivers determining the severity of inundation (e.g., levels of SLR) to those that influence indirect impacts on the social systems (e.g., resilience of the electric power supply system to inundation).  In light of this, the method would produce a large number (hundreds or thousands) of potential SLR impact maps, each showing the estimated magnitude and spatial distribution of a type of impact in a certain plausible future scenario. Since it is unrealistic to expect analysts or decision-makers to consider this large amount of information in their planning, the method must present this impact information in a way that is more manageable and digestible. To do this, the method must summarize this large volume of impact maps by reducing them into a small number of spatially explicit predominant modes of each impact presented as maps. To envision this, consider for each type of impact, sorting the large volume of impact maps into different piles, putting similar maps in the same pile. By creating a hybrid impact map representing each resulting pile of impact maps, all the hybrid versions (representatives of each pile) then collectively represent the spectrum of how the impact can unfold across a range of plausible futures. These hybrid versions of impact maps are essentially spatial patterns of different SLR impacts that can eventuate across a defined range of plausible futures, which are called Robust Impact Patterns (RIPs). While the notion of robust impacts is not entirely new (e.g., Yu et al.  39 (2017)), to the best of our knowledge, it has yet to be applied in the context of climate adaptation planning.  3.1.2 How to identify the Robust Impact Patterns? To perform the summarizing process described above, one may consider using cluster analysis to sort the large volume of impact simulations (maps) into a small number of groups, where the RIPs (predominant modes) is represented by the mean or median of each group. However, we may face the problem of losing the information about the spatial distribution of the impact through the clustering process as well as by taking the mean or median of each resulting cluster to generate the RIPs. Principle components analysis (PCA) is another logical option, however, PCA is a linear method that is not ideal to identify patterns in nonlinear systems, such as the human systems within which socio-economic impacts take place. By comparison, the machine-learning algorithm, called self-organizing maps (SOMs) (Kohonen, 1982) is superior in terms of its ability to handle nonlinear data and preserving the information about the spatial variations in each impact map (Hewitson & Crane, 2002; Liu, Weisberg, & Christopher, 2006). This allows the spatial distribution of the impact in the resulting RIPs to be more representative of the most predominant spatial pattern rather than distorted by outliers. Therefore, the proposed method will use SOMs to synthesize the large number of impact maps into RIPs for each impact type. The SOMs is a good fit for this task for several reasons:  • SOMs has the ability to preserve spatial features in the clusters by preserving the topology of the data space (Yip & Yau, 2012)  40 • SOMs do not require prior knowledge about the condition of the data (e.g., data distribution, level of dispersion, outliers) to accurately determine cluster membership (Kohonen, 2001). • SOMs has been shown to outperform other traditional cluster analysis techniques for clustering large multivariate data that are “messy” (i.e., departure from the ideal condition of having compact isolated clusters), including hierarchical clustering (Astel, Tsakovski, Barbieri, & Simeonov, 2007; Mangiameli, Chen, & West, 1996) and principles component analysis (Astel et al., 2007).   SOMs is known for its advanced capability for pattern recognition, like the principle component analysis. While SOMs has been used to recognize patterns for various purposes, from facial recognition (Aly, Tsuruta, & Taniguchi, 2008) to network hacking activity identification (Ippoliti & Zhou, 2012), this will be the first time SOMs is applied for flood impact assessment. While it is certainly possible to write one’s own SOMs program in one of many existing types of computer programming languages (e.g., C, Python, R), there is a number of existing software that runs SOMs or has SOMs as one of its built-in tools. Some common software used for SOMs include: SOM PAK (Kangas, 2014); Package ‘som’ in the CRAN-R project (Yan, 2016); SOMs Toolbox for Matlab (Laboratory of Computer and Information Science (CIS), 2015). More detailed explanation of how the SOMs work and how it is used to identify RIPs will be discussed in the next section.   41 3.2 The Robust Impact Patterns Method The RIPs method consists of 3 main stages – 1) future flood scenarios development, 2) flood impact modeling, and 3) RIPs identification [Figure 3.1]. This section of the chapter describes the generic steps involved in each stage of the RIPs method in terms of what should be considered and what is involved computationally. Some of the more nuanced and detailed guidance on applying the method can be found in the following chapter about the application of the RIPs method at the City of Vancouver as a case study. Figure 3.1 Summary of the 3 stages of the RIPs method  3.2.1 Stage I - Future flood scenarios development The RIPs method begins with developing the range of plausible future scenarios to be considered in the application. Given the gradual nature of SLR, it is likely that the population and assets within permanent inundation areas due to SLR will be relocated as a form of reactive adaptation action. However, since the purpose of the RIPs method is to help users consider the range and spatial distribution of potential SLR impacts under a large range of future to inform their  42 prioritization and identification of initial adaptation options, the future scenarios here assumes no autonomous or planned adaptation have taken place in order to avoid such assumed reactive adaptation acting as a barrier to identifying other adaptation options. As a result, the RIPs method aims to estimate the potential impacts that are generally larger, rather than the actual or residual impacts.  The first step in scenario development is to decide what factors will characterize the future scenarios to develop the space in which the SLR impacts can be estimated. For example, the future scenarios can be characterized by factors such as the level of SLR, population density and distribution. Although the RIPs method can still be performed if the scenarios are only defined by physical drivers (e.g., level of SLR, storm intensity), the selection of the factors should be holistic rather than partial or sectorial to reasonably reflect the range of physical and socio-economic drivers of each impact to be assessed. For example, the cost of building damage from flooding at a given block may depend on factors such as depth and spatial distribution of inundation associated with different level of SLR, as well as building type, and building density. Taking a holistic approach to defining scenarios can support more comprehensive understanding and exploring of the complexity in physical and socio-economic impacts of SLR (Sheppard et al., 2011; Wiek & Walter, 2009). The factors characterizing the future scenarios should also be consistent with the spatial unit in which the impacts will be modeled (e.g., city block, neighbourhoods, or boundaries defined by first three digit of postal code). It is ideal if those factors can be defined at the spatial unit at which the impact is modeled in. Therefore, the decision regarding what impacts to assess and at what spatial coverage and scale should be made  43 in this first stage of the method. When selecting the type of impacts to assess, it is important to keep in mind that the impacts should:  • Be modeled as an indicator that is relevant and helpful to inform local adaptation planning, and  • Vary with variations of most if not all the factors characterizing the future scenarios in order to take advantage of the RIPs method.  Once the factors are selected, create variations of the factors to span the margin of uncertainty are created, aiming to include the best to the worst case or equivalent, based on existing literature or data. For example, based on the recent literature, the best case for the global mean SLR by 2100 is 0.98m (Meehl et al., 2007) to 6m (Dutton et al., 2015). Then variations of the SLR factor can be created – e.g., 1m, 1.5m, … to 6m, and presented geospatially as flood maps showing the depth and distribution of the inundation associated with each level of SLR. Once variations for each factor are developed, they can be systemically combined to create unique future scenarios. Before going into the next stage, the scenarios should be validated by checking for internal consistency. For example, in scenarios characterized by population growth and land-use patterns, a scenario that may not be internally consistent can be one that shows an increase in population in the city but for which land-use indicates a decrease in residential buildings or units.   3.2.2 Stage II - Flood impact modeling The objective of this step is to geospatially model each impact of interest in each of the future scenarios developed in Stage I. In the process of identifying the appropriate model to assess each  44 impact geospatially (i.e., model the impact within each spatial unit within the study area), one should consider the: • Study area and spatial unit in which the impact is modeled, • Data availability, and • Complexity of the model to avoid excessive run time, given that the impact will be modeled over a large number of scenarios. It is also ideal for the model to be sufficiently transparent to the analysts or decision-makers in order to foster trust and credibility of the results. It is important to ensure that each impact is modeled with the same study area and spatial unit, and over the same number of future scenarios.   3.2.3 Stage III - Robust impact pattern identification The output from Stage II should include x times y number of impact maps, where x is the number of impacts, and j is the number of future scenarios. This large volume of impact maps will serve as the data space (also called the training data) in which the machine learning algorithm – SOMs – will learn from to identify archetypal impact patterns (i.e., RIPs) for each impact. This process of using SOMs to identify RIPs involves three major phases: 1) pre-processing, 2) SOMs training, and 3) evaluating the SOMs [Figure 3.2].  45 Figure 3.2 Schematic diagram showing the 3 phases of the process to use SOMs to extract Robust Impact Patterns (RIPs) PHASE A: PRE-PROCESSING PHASE B: SOMs TRAINING PHASE C: SOMs EVALUATION  & INTERPRETATION 1 n P1 Pm Si Si Pm Step 2: Compare Input vector with reference vectors Compare  (Euclidean distances) Step 1: Initiate m number of reference vectors (nodes) Pm n 1 P1 Pm Step 3: Modify winning reference vector (Pwin) Step 4: Modify reference vectors in the neighborhood of winning node  P1 Pm Step 5: Repeat Steps 2 to 4 for all input vectors (Si), and for multiple iterations  Step 6: •  Assign input vector to best match reference vector (Pbm) to yield grouping/clusters  •  Return reference vectors (Pm) to impact maps format •  Compute error measures •  Visualize resulting RIPs •  Decide if training additional SOMs is necessary •  Determine range of future scenarios that can lead to impacts represented by each RIP  46 3.2.3.1 Phase A - Pre-processing In order for the impacts maps produced in Stage II to be useable in SOMs, the two dimensional map for an impact for a given future scenario must be converted into a one-dimensional vector, which has the length of n, where n is the number of spatial units within the study area [Figure 3.2].  In other words, the impact value of a given spatial unit (e.g., a block) becomes the value in an element of the vector. If more than one impact is modeled per future scenario, the other impact maps should also be converted and concatenated to form one continuous vector representing all the impacts for a given future scenario. This concatenated vector (Si) should have the length of the product of x and n. Therefore, at the end of the pre-processing, there should be y number of input vectors (Si), which serves as the SOMs training data.   3.2.3.2 Phase B – Self-organizing Maps (SOMs) training Figure 3.2 presents the schematic on the training of the SOMs, which involves six steps.   Step 1: Initiate a number (m) of reference vectors (Pm) (also called nodes) composed of random numbers, where m is the number of RIPs the user would like to identify. By the end of the training process, these reference vectors will become the RIPs, showing the predominant patterns for each impact as impact maps with the same resolution as they are modeled in Stage II.   Step 2: The training or learning then begins by comparing an input vector (Si) with each of the reference vectors (Pm) based on their absolute differences as measured by the  47 Euclidean distance between the two vectors. The reference vector (Pm) or node with the closest match to the input vector is the “winning node” (Pwin), where:  !!"# = !"# !! − !!     (3.1)  Step 3: The winning reference vector (Pwin) is modified to become a closer match to the input vector (Si) by an amount that is specified by the learning rate (α), which is a time-inverse function (i.e., the rate decreases with each learning cycle). Time is t.  !! ! + 1 = !! !  × ! !  × !! − !!(!)   (3.2)  Step 4: The reference vectors within a specified proximity (also called the neighbourhood radius) of the winning reference vector are also modified to become closer to the input vector (Si) by a fraction that is another time-inverse function (β) that is a Gaussian function inversely proportional to its Euclidean distance from the winning node (Pm; step 4 in Figure 3.2). Therefore, β defines the shape and size of the winning reference vector’s proximity (shaded grid cells in step 4 in Figure 3.2). The modification of a reference vector within the specified proximity to the winning reference vector at a certain training cycle (t) can be represented as: !! ! + 1 = !! !  × ! !  × !! − !!(!)   (3.3)    48 Considering each reference vector (Pm) in the n-dimensional space, by modifying the winning reference vector as well as its neighbouring reference vectors, SOMs is ensuring that reference vectors that are similar would be closer in the data space while distinctly different ones would be far apart to ultimately span across the space defined by the ensemble of future scenarios.  Step 5: Repeat steps 2 to 4 for all input vectors (Si), as one iteration of the training data. Multiple iterations are made through the training dataset until the reference vectors no longer get modified with continuing training.   Step 6: The last step of SOM training is to match each input vector to a reference vector simply according to their similarity, which yields the grouping of input vectors. Since each input vector represents the SLR impact(s) in a certain future scenario, the more input vectors are matched to a reference vector (RIPs), the more robust the impact pattern that it represents. Therefore, in the RIPs method, the robustness of a RIP refers to diverseness of futures where this impact pattern can occur, rather than the probability of that pattern occurring in the future.  3.2.3.3 Phase C – Visualizing, evaluating and interpreting the SOMs Visualizing the SOMs By the end of the last training iteration, the reference vectors should span across the data space to represent the spectrum of predominant modes. Given that a spectrum is by definition continuous, the collection of reference vectors in the n-dimensional data space can be considered as a surface  49 that spans across the data space. In this way, the SOMs is representing the distribution of high-dimensional data by a 2D projection where the scales of the horizontal and vertical directions of the surface should roughly comply with the extensions of the input-data distribution in the two orthogonal dimensions in which the variances of the data are largest (Kohonen, 2014). In training of the SOMs, the user can define the shape of this surface (2D projection), which is typically a sheet, but cylinder and toroid shapes are also available to accommodate circular data [Figure 3.3]. If the “sheet” shape is used, to visualize the resulting RIPs, the reference vectors can be projected onto a 2D array (with dimensions defined by user in the beginning of training) [Figure 3.4]. Since each reference vector contains the elements (map cells) of all assessed impacts, the RIPs of each individual impact can be visualized by extracting that segment of the reference vector and transforming it back into a geospatial map format [Figure 3.4].  Figure 3.3 Three different shapes of surfaces in which reference vectors can be visualized collectively    50 Figure 3.4 Schematic diagram showing the process to visualize the trained SOM reference vectors as RIPs   Evaluating the SOMs To determine whether the trained SOMs (i.e., RIPs) is of acceptable quality, two error measures are usually considered to provide some indications. One is the average quantization error (q), which is the average error between the input vectors and the reference vector it was assigned to in the last step of the training (i.e., best match reference vector) (Kohonen, 2001).   ! = !! !! − !!"!!!!      (3.4)  Another measure is the topographic error (k) measured as the percentage of input vectors for which the first- and second best match reference vector are not adjacent in the data space (k), indicating how well the reference vectors keeps the topography of the training data (Kohonen, 2001).   ! =  !! ! !!!!!!      (3.5)  51  Here µ(Si) = 0 if the second best match reference vector of an input vector (Si) is the closest to the best match reference vector, otherwise, µ(Si) = 1. For both error measures, lower value of the measure indicates better quality SOMs.  It is highly recommended to train multiple SOMs with different training parameters to compare their relative quality in terms of error measures and visually compare how well the resulting patterns represent the original data. The latter is made possible by assessing which input vector (i.e., impacts of a certain future scenario) is assigned to a certain reference vector that produces the RIPs.  The training parameters to vary with different SOMs can include: • Number of RIPs to extract (i.e., number of reference vectors to initiate) • Learning rates • Number of training cycles • Neighbourhood radius  Lastly, it is important to remember that there is no right or wrong size for a SOM (i.e., number of reference vectors) since it is essentially the degree to which one would like to reduce the data – the larger the SOM, the less the data is reduced. In other words, the larger the SOM, the more variations in the data can be captured, and vice versa. Therefore the size of the SOM should be driven by the purpose of its application – i.e., how much variation should be captured by the SOMs to provide useful information for the user?  52  Interpreting the resulting SOMs  Once a trained SOMs (referred as just SOMs from hereafter) is selected for use, the next step is to gather the information about its reference vectors to help interpret the resulting RIPs. There are two basic pieces of information about the reference vectors that would allow one to understand what the resulting RIPs are representing. These are:  1) Which input vector is assigned to a certain reference vector (i.e., RIP) in the last step of the training – i.e., the best matching reference vector (Pbm). Since in the RIPs method, each input vector represents the impacts under a certain future scenario, this piece of information tells you what range of future scenarios can result in impacts represented by the corresponding RIP. 2) Knowing which input vector is assigned to which reference vector, one can also tell how many future scenarios’ impacts are represented by a certain RIP. The more scenarios the RIP represent, the more robust it is. Since the nature of some impacts can be similar across a range of future scenarios, they may be represented by one RIP that would have relatively more input vectors assigned, while an RIP that has fewer vectors assigned may represent an outlying impact pattern.  Knowing what type of future scenarios can result in the impact represented by the different RIPs produced by the SOMs, further analysis and interpretation of the RIPs generally focuses on analyzing the distribution and magnitude of the impact in each RIP. For example, what is the total number of buildings affected in a certain RIP? Which neighbourhood is potentially affected in most RIPs? There are numerous ways to visualize and analyze the SOMs result, where certain  53 ways would be more suitable depending on the application and the type of questions one is trying to answer with the SOMs. Therefore, further discussions about interpreting SOMs results will be available in the next chapter where the RIPs method is applied to assess the impact of SLR at the City of Vancouver (CoV).  3.3 Intended utility of RIPs in SLR adaptation The key objective of the RIPs method is to help users consider the range and spatial distribution of potential SLR impacts under a large range of future to inform their prioritization and identification of initial adaptation options during the early stage of planning. Therefore, the resulting RIPs are not expected to be able to help users develop specific detailed adaptation options, which would require much more option-focused, detailed and rigorous analysis of its feasibility and performance, which is typically done in the options evaluation stage of planning. Furthermore, the future scenarios in the RIPs method assumes no reactive, autonomous or planned adaptation have taken place in order to avoid such assumed reactive adaptation acting as a barrier to envisioning other adaptation options. As a result, the RIPs method aims to represent the potential impacts rather than predicting the actual or residual impacts.  Ultimately, the RIPs are intended to support SLR adaptation planning by facilitating better understanding of the potential local SLR impact and subsequently achieve two goals:  1) To identify a better repertoire of preliminary adaptation options to build a basic plan that can be subsequently refined to increase robustness, but also  2) To provide evidence to give leverage to request for resources to support the implementation of adaptation.   54 The RIPs can help decision-makers and analysts achieve these two goal as they allow the users to visualize 1) the location of the potentially vulnerable population and assets, 2) how the indirect and cascading impacts of SLR can manifest, and 3) the range of spatial distribution and magnitude of potential SLR impacts that are plausible within the defined range of futures [Figure 3.5]. Specifically, users are expected to be able to the RIPs in 10 anticipated ways as listed in Table 3.1 to support their SLR adaptation planning.    55 Figure 3.5: Intended ways in which RIPs can be used to support SLR adaptation      56 Table 3.1 Anticipated ways in which the RIPs method can support SLR adaptation planning  Although Figure 3.5 shows a linear process, in practice, it is expected that these different ways to use the RIPs would overlap and interact to achieve the goals. The objective of the last component of this thesis, as described in Chapter 6, is to use the CoV case study results to help evaluate whether the RIPs can support SLR adaptation planning in these anticipated ways from the perspective of stakeholders at the CoV.  Ten	anticipated	ways	in	which	the	RIPs	can	be	used	to	support	SLR	adaptation	planning	Develop	adaptation	options	1. Generate	new	ideas	for	SLR	adaptation	options	2. Generate	more	refined	and	targeted	SLR	adaptation	options	3. Consider	a	wider	range	of	adaptation	types	(e.g.,	soft,	hard,	combination)	4. Development	of	SLR	adaptation	options	that	are	more	uncertainty-tolerant	5. Facilitate	for	long-range	planning	of	how	to	modify	current	options	to	respond	to	a	worse	situation	Access	to	resources	and	support	for	implementation	6. Prioritize	SLR	adaptation	efforts	and	resources	7. Serve	as	a	useful	tool	for	communicating	SLR	risk	8. Identify	new	types	of	stakeholders	to	engage	in	planning	9. Provide	justification	for	planning	beyond	the	common	worse	case	scenario	10. Provide	better	leverage	to	request	for	resources	  57 Chapter 4:  Application of the Robust Impact Patterns Method at the City of Vancouver – Data and Methods  4.1 Introduction The RIPs method is applied to the City of Vancouver (CoV or ‘the City’ from hereafter) as a case study to demonstrate its applicability as a method to assess potential impacts of SLR at the local level. Additionally, this case study serves as a platform on which to show the more nuanced issues and challenges that one would need to consider and address when using the method in practice. This chapter will describe the data and method used to implement the RIPs method for this case study, while the resulting RIPs and their implications are discussed in Chapter 5.  This chapter begins with an overview of the CoV (Section 4.2 and 4.3) and its current SLR adaptation efforts (Section 4.4). As described in Chapter 3, the RIPs method consists of three stages – 1) Future flood scenarios development, 2) Geospatial flood impact modeling, and 3) Robust impact pattern identification [Figure 4.1]. While the general process and thinking behind the RIPs method are described in Chapter 3, Sections 4.5, 4.6, and 4.7 describes how the steps within Stage 1, 2, and 3 are carried out respectively in the context of the CoV.   58 Figure 4.1 Three key stages of the RIPs method (Same as shown in Chapter 3)   4.2 Study area – City of Vancouver The CoV is a coastal city situated on the southern shoreline of British Columbia, Canada. It is the most populous city in the province of British Columbia, with a population of 631,486 (Statistics Canada, 2017). It is also the most densely populated city in Canada with over 5,400 people per square kilometer (Statistics Canada, 2017). While most of the City’s shoreline is coastal bordering the water of the Georgia Strait, the city also borders a section of the Fraser River where it meets the ocean [Figure 4.2]. Therefore in the event of coastal flooding and SLR, the entire stretch of the City’s shoreline can be affected.      59 Figure 4.2 Map showing the geographic location of the City of Vancouver and its surrounding water bodies in the inset map   Although the City of Vancouver has yet to experience a major flood event, studies (e.g., Hallegatte et al., 2013) have suggested it to become increasingly vulnerable to coastal flooding with climate change due to density of population and development at the waterfront areas. Acknowledging the deep uncertainties of SLR projections, Fisheries and Oceans Canada and the British Columbia Ministry of Environment have projected an upper range of 0.89 to 1.03m of SLR by 2100 (from year 2000 level) at the City of Vancouver (Thomson, Bornhold, & Mazzotti, 2008). Therefore the Government of British Columbia recommends planning for SLR of approximately 1.0m by the year 2100 (British Columbia Ministry of Environment, 2013). Additionally, certain areas within the City – including Southlands, part   60 of Fraser land across from Mitchell Island, False Creek, and Coal Harbour - are found to be experiencing ground subsidence of about 1 ± 0.5 mm/year, while the rest of the City is experiencing virtually no change in elevation (Hill et al., 2013).  4.3 Why apply the RIPs method at the CoV? There are several reasons for this study to choose the CoV as a case study community. First, it is timely. The CoV began their SLR adaptation planning with a coastal flood risk assessment in 2012, which is about the same time as when this study began. Therefore, a partnership was formed with the CoV for this case study, aiming to produce information that can supplement the City’s ongoing effort in SLR adaptation planning. Although the CoV has experienced some notable coastal flood events in the past that affected mostly beaches and park areas, it has yet to experience a major flood event. Therefore, flood protection has not been a priority in the CoV, and it has little empirical knowledge about how the City may be affected by a major flooding event. In this light, it is advantageous to apply the RIPs method to inform the City about the range of impacts it may experience under different level of flooding and future conditions.   Some recent studies have also assessed the potential impacts of SLR in the CoV, as a part of a larger regional level SLR impact assessment. These studies assessed a range of SLR impacts that go beyond identifying assets and population displaced by inundation. For example, the Lower Mainland Flood Management Strategy (Fraser Basin Council, 2016), led by the Fraser Basin Council, has assessed a range of potential SLR impacts in the Fraser River Basin, which covers 30 municipalities, including the CoV. This project has assessed some direct impacts (e.g., building damage, displacement of agricultural land, damage to critical infrastructure), as well as an indirect impact – disruption to freight shipment due   61 to rail traffic disruption. More indirect impacts were discussed but not assessed. Another example is the Coastal Cities at Risk (CCaR) Project (McBean, Cooper, & Joakim, 2017) that was funded by multiple Canadian federal government agencies to investigate climate change impacts on Greater Vancouver, which includes a similar set of municipalities as the Fraser River Basin. Besides assessing both direct and indirect economic impacts (e.g., impact on industrial outputs), this project also makes significant contribution to understanding the health impacts of SLR (Owrangi, Lannigan, & Simonovic, 2015), which is often overlooked in vulnerability assessments. While these studies have assessed a range of SLR impacts in the CoV, this application of the RIPs method will explicitly model flood-induced power outage that can result in indirect impacts that have not been assessed before (e.g., sewage backup damage, business disruption).   In comparison with cities that experience regular flooding, the City’s lack of existing flood protection infrastructure and experience also suggests that there may be less knowledge and constraints to shape preferences for different SLR adaptation options. Therefore, the RIPs method can be particularly useful in helping the City’s decision makers understand the associated uncertainties and local vulnerabilities to inform their selection of a preliminary set of adaptation options that can later be further refined for robustness. Lastly, the CoV has an established database of spatial data. As described in the previous chapter, the RIPs method requires a significant amount of data to model the potential flood impacts geospatially, over a large number of future scenarios.     62 4.4 Summary of CoV’s current efforts in SLR adaptation planning The CoV’s three-phased SLR adaptation planning began in 2012. Phase 1 started with a coastal flood risk assessment (CFRA) where detailed hydrologic-hydraulic modeling was conducted for 5 flooding scenarios (Northwest Hydraulics Consultants, 2014): • Scenario 1, Year 2013, 0.0 m SLR, 1:500-year storm hazard • Scenario 2, Year 2100, 0.6 m SLR, 1:500-year storm hazard • Scenario 3, Year 2100, 1.0 m SLR, 1:500-year storm hazard • Scenario 4, Year 2100, 1.0 m SLR, 1:10,000-year storm hazard • Scenario 5, Year 2200, 2.0 m SLR, 1:10,000-year storm hazard  Using the ocean levels produced by the modeling, flood extent and depth were estimated over land, which is subsequently used to geospatially identify what infrastructure and assets would be exposed to inundation. The identified vulnerable assets are summarized in Table 4.1.       63 Table 4.1 Vulnerable infrastructure and services identified by the CoV’s CFRA (Table 2 from ©Lyle & Mills (2016). Assessing coastal flood risk in a changing climate for the City of Vancouver. Canadian Water Resources Journal, 41(1-2), 343-352, Page 349. Adapted by permission from publisher)     64 To quantify some of the potential impacts on assets, infrastructure, and population within the inundated areas, the CoV has used the flood module of the Hazus-MH model (Hazus from hereafter) adopted for use in Canada (Nastev & Todorov, 2013) to calculate the potential losses in each flood scenario. Hazus is a tool developed by the U.S. Federal Emergency Management Agency (FEMA) to perform flood hazard and flood damage analysis. The specific impacts that the CoV modeled using Hazus include (Lyle & Mills, 2016): • Direct impacts o Damaged critical infrastructure buildings (e.g., hospitals) [number of facilities] o Direct building damage [percent damaged or repair cost in CAD$] o Loss of content in the buildings [CAD$] • Indirect impacts o Debris generated from damaged buildings [tons]  Using the outcome of Phase 1 of the CFRA, the CoV have raised their existing flood construction level (FCL)2 by 1m in 2014, to approximately 4.6m (Pander, 2014). This applies to all new building developments within designated floodplains.  The outcome of Phase 1 also facilitated for Phase 2 of the CoV’s adaptation planning, where plausible adaptation options are identified for 11 zones within the City that are considered to be at risk of coastal flooding. Using the Structured Decision Making approach (Gregory et al., 2012) to account for the                                                 2 As defined in the CoV’s administrative report regarding the recommendation to raise the FCL, the FCL is “… the minimum elevation of the underside of a floor system or the top of a concrete slab of a building used for habitation, business or storage of goods damageable by flood water.”   65 multiple objectives indicated by stakeholders, the trade-offs for options within 5 of the 11 zones are also assessed. It is worth noting that in Phase 2, only one future scenario out of five is considered – 1m SLR by 2100, with 1:500-year storm. The adaptation options considered in this phase ranges from structural engineering solutions (e.g., seawalls and sea barriers), to non-structural options (e.g., gradual retreat through density reduction). Being aware of the high degree of uncertainty around the long-term impacts of SLR, Phase 2 recommendations fall into 5 categories (Lyle, Long, & Beaudrie, 2015): • Preserve future options • Refine engineering design of specific short-term and/or big-ticket options • Implement short-term and “no-regrets” actions • Monitor developments and plan to actively adapt to them • Engage with communities, partners, and other levels of government A more detailed list of adaptation options considered in Phase 2 can be found in Lyle et al. (2015). Having completed Phase 2 in 2015, the ongoing Phase 3 focuses on providing detailed cost-benefit analysis and design work for the adaptation options (Pander, 2014).   Considering the SLR impact assessment approach taken by the CoV in the CFRA, the application of the RIPs method can complement the City’s process in the following ways: • Expand the range of future scenarios to consider in the planning by assessing the SLR impacts under a much wider range of future scenarios (hundreds) • Assess a wider range of direct and indirect impacts of SLR that expand upon those already included in the CFRA to account for potential indirect impacts resulting from exposed infrastructure (e.g., an inundated substation can lead to extensive and prolonged power outage in areas beyond the inundated areas and lead to other socio-economic impacts)   66 • Geospatially model the SLR impact over the same area and spatial units as used in the CFRA to facilitate easier integration of the information and comparison  4.5 Future flood scenarios development 4.5.1 Impacts selection To select the factors to characterize the future scenarios, the first step is to identify the set of SLR impacts to assess in this application. Table 4.2 lists the SLR impacts assessed in this case study. It is also important to note that these impacts are modeled with the spatial unit of a city block, which is delineated by the boundaries of dissemination blocks (DB from here forth) in the 2011 Canadian Census. This scale is the highest resolution at which impacts can be modeled without permitting identification of individuals, buildings, or businesses for ethical reasons.  Table 4.2 SLR impacts assessed in this CoV case study    67  Although the RIPs method can be used to assess other potential impacts of SLR, the selection of these impacts results from a balance of a number of principles to promote relevance to the City of Vancouver’s planning process as well as consistency with the fundamental guidelines of the RIPs method as described in Section 3.2.1 and 3.2.2. The specific principles guiding this section are as follow: 1) Impacts should align well with the impacts being considered by the CoV’s Coastal Flood Risk Assessment (CFRA) to facilitate easier comparison and integration of information with their planning.  2) Target assets that were found to be vulnerable in the CFRA but were not further assessed 3) Include both direct and indirect impacts 4) Impacts should span across economic, social, and environmental impacts 5) Aim to include impacts that vary at similar spatial scale so that they can all be resolved adequately when modeled at the same spatial units 6) Aim to include impacts that are influenced by similar drivers (e.g., land-use, level of SLR, access to electricity), such that all the impacts can vary under each future scenarios characterized by these drivers3 7) Data and model availability  Although there is always a limited number of impacts an assessment can address, a number of impacts were considered but were ultimately not included for various reasons. For example, the ecological                                                 3 More on this issue will be discussed in the results of this case study in Chapter 5   68 impacts of coastal flooding, such as the impact of saltwater intrusion on wetlands and biodiversity, and the impact of erosion and inundation on endangered species, would require additional analyses of detailed variables (e.g., salinity of flood water, rate of erosion, fine-scale distribution of different species) to reasonably model the impact. Other environmental hazard impacts induced by flooding, such as contamination of drinking water (Delpla, Jung, Baures, Clement, & Thomas, 2009) and dispersion of hazardous material (Sengul, Santella, Steinberg, & Cruz, 2012), were also not included here due to the lack of critical information. Drinking water can be contaminated if the inflow of contaminated floodwater or toxins occurs at the air vault locations of the underground water distribution pipelines that often may not have as much inflow control (Val-matic, 2011). Due to the lack of information about inflow control measures taken at various air vaults in the CoV’s water distribution network, it is currently not possible to include this impact in the case study. While the location data for various hazmat sources in the CoV are available, there is insufficient information regarding the vulnerability to flood-induced leakage for each hazmat source. Nonetheless, the environmental impacts included in this case study can serve as a starting point to bring attention to the environmental impacts of flooding, which is an aspect overlooked in impact assessments (Gautum & van der Hoek, 2003).  4.5.2 Characterizing future scenarios To account for major sources of uncertainties in each impact listed in Table 4.2, the future scenarios are composed of the following 4 different types of conditions representing 4 key sources of uncertainty in the long-term impacts of SLR being assessed in this case study [Figure 4.3]:  1. Inundation (depth and spatial distribution) – 21 different inundation conditions are formed by combinations of 7 different levels of SLR (in meters) and 3 different levels of storm intensity (as measured by return periods)   69 2. Land-use – four different land-use conditions are formed by four different regimes to distribute the growth of the population and buildings (only residential and commercial) 3. Power outage (which blocks have prolonged outage) – two different power outage conditions are formed by two different sets of assumptions (optimistic and pessimistic) about the resilience of the power infrastructure (e.g., substation) 4. Buildings’ vulnerability to damage from inundation – two different buildings vulnerability conditions are defined by using two different sets of stage-damage functions (SDFs) to estimate the amount of building damage from a given level of inundation Systematically combining the variations of inundation, land-use, power outage, and building vulnerability conditions define a total of 336 future scenarios for this case study [Figure 4.3]. The following sub-sections outline how each type of condition is developed.   70 Figure 4.3 Variations of 4 different types of conditions are combined systematically to generate 336 future scenarios   4.5.3 Inundation - storm intensity and sea-level rise The spatial distribution and magnitude of the inundation associated with a flood event is the key contributor to all of the impacts assessed in this case study as it determines where flooding occurs and by how much. The variations in inundation (i.e., different inundation conditions) are generated by varying a) storm intensity as measured by its return period and b) levels of SLR. Three levels of storm intensity are considered here: • 1:50-year storm (equivalent to a King Tide in the current climate) Stage damage functions Power outage extent Population and land-use distribution Sea-level rise Storm return period 2050 - 2100Flood	event	with	no	adaptation	action	1:50	0m	Current	land-use	Optimistic	HAZUS	default	MCM	Pessimistic	Status	Quo	Compact	Sprawl	1m	2m	3m	4m	5m	6m	1:500	1:10000	1	Land-use Power outage Building vulnerabilityInundation  71 • 1:500-year storm (i.e., rare storm based on the current climate) • 1:10,000-year storm (i.e., very rare storm based on the current climate)  The CoV also considered the 1:500-year and 1:10,000-year storms in their CFRA. But since the CoV has yet to experience either of those storm intensities, the 1:50-year storm is also included to provide a baseline that the CoV has experienced before to help envision the associated level of impacts and severity. In comparison with the range of SLR included in the CoV’s CFRA, this case study will expand to cover the range of projected SLR by 2100 based on current published studies, which is from 0.64m to 6m (see Figure 2.2 in Chapter 2). Acknowledging that there is also deep uncertainty around when the projected amount of SLR would occur, the scenarios do not have any explicitly assigned probability and only resemble a hypothetic and plausible future in the 2nd half of the 21st century. The systematic combination of different levels of SLR and storm intensity results in 21 different inundation conditions [Table 4.3].    72 Table 4.3 Inundation conditions for this CoV case study  Since detailed hydrologic-hydraulic modeling was already been conducted in the CFRA (by consultants contracted by the CoV to conduct Phase 1 of the CFRA) and produced high-resolution4 flood depth maps for their 5 inundation scenarios, this case study builds upon those data to generate the flood depth maps representing the 21 inundation scenarios here. Specifically,                                                 4 The overland model used a mesh instead of a grid to allow different element sizes (e.g., smaller for higher resolution) for different locations in the covered area. The mesh elements were generated based on a node spacing specification of 5m Northwest Hydraulics Consultants. (2014). City of Vancouver Coastal Flood Risk Assessment: City of Vancouver.  Inundation	condition	ID	 Storm	intensity	(return	period)	Level	of	SLR	A0	 1:50-year		(Equivalent	to	King	tide)	0m	A1	 1m	A2	 2m	A3	 3m	A4	 4m	A5	 5m	A7	 6m	B0	 1:500-year		 0m	B1	 1m	B2	 2m	B3	 3m	B4	 4m	B5	 5m	B6	 6m	C0	 1:10,000-year	 0m	C1	 1m	C2	 2m	C3	 3m	C4	 4m	C5	 5m	C6	 6m	  73 the flood depth maps for scenarios B0~C6 were generated by building upon the CFRA's modeled inundation for 1:500-year and 1:10,000-year storms, using the following approach.  Firstly, the flood depth estimates for each of their 5 scenarios accounts for several contributing factors (Northwest Hydraulics Consultants, 2014):  a) High tides,  b) Storm surge,  c) Local wind set up,  d) Local wave setup, e) Level of sea-level rise, and  f) Freeboard of 0.6m (to account for uncertainties in flood hazard analysis)  In the CFRA, the water level contributed by the first 4 factors was estimated through hind-casting simulations and subsequently used in the 2D overland model to estimate the flood depth and extent associated with the 1:500-year and 1:10,000-year storms (Northwest Hydraulics Consultants, 2014). Then the level of SLR and the freeboard amount is spatially added to generate the final flood depth maps for each scenario. Since the inundation scenarios for this case study uses different levels of SLR than those in the CFRA, the SLR component (1m) of the CFRA’s flood depth map data for their scenario #3 (1.0 m SLR, 1:500-year storm hazard) and #4 (1.0 m SLR, 1:10,000-year storm hazard) was subtracted using raster analysis in GIS (Geospatial Information Systems), to serve as the flood depth data for the 1:500-year storm and 1:10,000-year storms, respectively.  By adding different levels of SLR onto the flood depth data for each storm scenario, the flood depth maps for inundation conditions B0 to C6 are generated [Table   74 4.3]. For example, Figure 4.4a shows the flood depth map for the worst inundation condition (i.e., condition C6, 1:10,000-year storm with 6m of SLR) in this defined range of conditions.   Since the 1:50-year storm was not one of the scenarios included in the CFRA, the flood depth map for inundation conditions A0 to A6 [Table 4.3] are developed using the bathtub method in GIS to estimate the flood depth and extent for a water level of 2.66m, which is the maximum water level of the flood event on 21st December 2012 (Tinis, 2012). The bathtub method is a common and simple method that assumes that an area with an elevation less than a projected flood level will be flooded like a “bathtub” where the water is assumed to be flat (i.e., no account of hydrodynamics) (Neumann & Ahrendt, 2013). In the GIS environment, the bathtub method determines the areas of inundation through a simple calculation where the elevation above sea-level in each cell is compared against a certain water level such that all cells with values lower than the predicted sea level are considered flooded (Yunus et al., 2016). The elevation used in this calculation was from the 1m resolution Digital Elevation Model (DEM) that the CoV used in their CFRA flood hazard analysis. While the literature has documented several limitations of the bathtub method, such as overestimation due to the omission of hydrodynamics (e.g., flow direction, bed friction) (Seenath, Wilson, & Miller, 2016), the bathtub method is a satisfactory approach for this case study as it is only used here as a baseline scenario and the purpose is to demonstrate the application of the RIPs method. As an example, Figure 4.4b shows the flood depth map for the inundation condition A1 (1:50-year storm with 1m of SLR). Flood depth maps for the other inundation conditions of this case study can be found in Appendix A1.      75 Figure 4.4 Flood depth maps for inundation condition a) C6 (1:10,000-year storm with 6m SLR) [top], and b) A1 (1:50-year storm with 1m of SLR) [bottom].     76 4.5.4 Land-use - population and building distribution  The spatial distribution of population and buildings (residential and commercial) characterizes the land-use aspect of the future scenarios. Specifically, four different land-use conditions are developed, one representing the current land-use, while the other three represent distinct ways in which the additional population and buildings in the future are distributed across the City. It is important to note that the total magnitude of population and building growth in those three conditions are the same but distributed differently across the City. Given that each impact assessed in this case study will be modeled at the Census dissemination block (DB) level (approximately the same as city blocks in the CoV), the land-use conditions are also defined at the DB level where the population, number of residential buildings, and commercial buildings [Table 4.4] are defined. The general idea of each land-use condition is as follow: a) Current (baseline condition) The number and distribution of population and buildings as it was in 20115 b) Status-Quo Additional population and buildings from future growth are distributed with the same proportion as in the current land-use of 2011 c) Compact Additional population and buildings from future growth are distributed to blocks within the CoV that have higher population density relative to the rest of the City                                                 5 The 2011 census data was the latest available dataset at the time of analysis.   77 d) Sprawl Additional population and buildings from future growth are distributed to blocks within the CoV that have lower population density relative to the rest of the City 	Population	and	residential	buildings	The population, number and distribution of residential buildings per DB are defined in the four different land-use conditions using the method proposed by Tse (2011), which is a study assessing how injuries and casualties resulting from a hypothetical earthquake in Metro Vancouver would vary with different land-use and population density conditions. The number and distribution of commercial buildings per DB are defined using a different approach, which will be described in the next sub-section.  i. Current land-use condition Since the Current land-use condition is based on current population and land-use distribution, the 2011 Canadian Census data was used for population at the dissemination block (DB) level, which was the latest available data at the time this analysis was conducted. The number of different types of residential and commercial buildings are drawn from the building inventory data that comes with the Canadian version of the Hazus model (Nastev & Todorov, 2013). Therefore, in the Current land-use condition, the number of different residential and commercial buildings are defined by the number of buildings that belong to different residential and commercial occupancy classes in the Hazus inventory, as shown in Table 4.4. “RES” and “COM” denote residential and commercial building occupancy classes respectively.    78 Table 4.4 Occupancy classes of buildings defined in Hazus (Table 4.2 from Federal Emergency Management Agency (FEMA) (2009). Multi-hazard Loss Estimation Methodology, Flood Model Hazus-MH User Manual. Retrieved from https://www.fema.gov/media-library-data/20130726-1820-25045-8814/hzmh2_1_fl_um.pdf. By permission from publisher.)    ii. Status-quo land-use condition This land-use condition represents the distribution of future population and buildings; therefore the total population would be larger than those in the Current land-use condition. Specifically, the population in all three future land-use conditions (Status-quo, Compact, and Sprawl) are based on the Metro Vancouver’s Regional Growth Strategy, Metro Vancouver 2040 - Shaping our Future (Metro Vancouver, 2013). This Regional strategy projects trend in population growth   79 and the number of dwelling units by 2041 for each municipality within the Metro Vancouver region, CoV being one of them [Table 4.5].  Table 4.5 Change in population and dwelling units in the CoV by 2041 as projected in the Metro Vancouver’s Regional Growth Strategy (Metro Vancouver, 2013) 	 2011	 2041	 Growth	Population	 603,502	 740,000	 136,498	(22.6%)	Dwelling	units	 264,575	 339,500	 74,925	(28.3%)	  The following approach, also outlined in Figure 4.5, is used to derive how the new population and dwelling units are distributed amongst the different DBs in the CoV.  Figure 4.5 General steps to derive the number of new residential buildings of occupancy class x per DB   First, the number of additional dwelling units distributed to a given DB in the CoV (dnew) is determined using:    80 !!"# = !!"#×!!"#$$     (4.1)  where DCoV is the projected increase in dwelling units for the CoV as a whole (i.e., 74,925) and pdwell is the percentage of CoV’s total dwelling units in 2011 that are locate at this given DB. But since we only have data for the number of residential buildings per DB (from the Hazus inventory), the number of dwellings in each DB is derived using the average number of dwelling units in each Hazus occupancy class of residential building, which can be found in Appendix B1.   After computing the number of new dwelling units in this DB (dnew), they are distributed across different occupancy classes of residential buildings. The number of new dwelling units located inside each residential buildings of occupancy class x (dnew_x) in the given DB is:  !!"!_! = !!"#$ × !!"#     (4.2)  pRESx is the percentage of the total existing dwelling units in the DB that are in occupancy class x buildings. This way, the new dwelling units are distributed across different building occupancy classes with the same proportions as they were in 2011. Then, to translate this in terms of the number of new class x residential buildings (Bnew_x), the number of new dwelling units in class x residential buildings is divided by the average number of units in existing class x residential buildings.  !!"#_! = !!"#_!!!!_! ;  !!"#_! = 0 !" !!"#_!!!!_! < 1    (4.3)   81  The way the new population is distributed across the DBs in the city is also defined using a similar approach where the number of new residents in a DB (Rnew) is:  !!"# = !!"#  × !!"!     (4.4)  where the PCOV is the projected population increase in the CoV (i.e., 136,498) and ppop is the percentage of CoV’s population living in the DB in 2011, which is simply the population of the DB in 2011 divided by the total population in CoV in 2011. The number of new residents that belongs to a certain age group (less than 16 years of age, between 16 and 65, and over 65) is also distributed within the DB with the same proportion as in 2011. This definition of age groups facilitates modeling the number of potentially affected children and elderly in each flood scenario. The definition of how many new residential buildings of different occupancy classes are in each DB also facilitates the modeling of direct building damage cost of residential buildings under different scenarios.   iii. Compact land-use condition The distribution of new population and residential buildings in this land-use condition is defined using the same approach as for the Status-quo condition, except that the new population and residential buildings are distributed only to DBs where more than 50% of its residential buildings are of high-density type (i.e., multi-family residential buildings with four or more dwelling units). In terms of occupancy classes, these include RES3B-F. The proportion in which the new   82 dwelling units are distributed across buildings of different occupancy classes is shown in Table 4.6, where new dwelling units are mostly allocated to low-rise and high-rise buildings. Table 4.6 Proportions to distribute new dwelling units across occupancy classes in the Compact land-use condition  Table 4.7 Proportion to distribute new dwelling units across occupancy classes in the Sprawl land-use condition  iv. Sprawl land-use condition This land-use condition is defined using the same approach as above except that the new population and residential buildings are only distributed into the low-density DBs and using the proportions in Table 4.7, where new units are mostly allocated to ground-relate homes (i.e., Residential	building	occupancy	classes	Description	 Percentage	of	new	dwelling	units	allocated	[ρRESx]	RES1,	RES2,	and	RES3A	 Single	family	home	and	duplexes	 0%	RES3B	 Townhouses	 5%	RES3C	 5%	RES3D	 Apartment	with	less	than	5	storeys	30%	RES3E	 Apartment	with	more	than	5	storeys	30%	RES3F	 30%	RES4	 Temporary	lodging	 0%	Residential	building	occupancy	classes	Description	 Percentage	of	new	dwelling	units	allocated	[ρRESx]	RES1	 Single	family	homes	 42.5%	RES2	and	RES4	 Mobile	homes	and	temporary	lodging	2.5%	RES3A	 Duplexes	 42.5%	RES3B	 Townhouses	 5%	RES3C	 5%	RES3D	 Apartment	with	less	than	5	storeys	0%	RES3E	 Apartment	with	more	than	5	storeys	0%	RES3F	 0%	  83 single family homes, duplexes, and townhouses). Low-density DBs are those that have more than 50% of its residential buildings of the low-density type (i.e., occupancy classes RES1-3A and RES4).   Non-residential	buildings	Although the distribution of new non-residential buildings across the different DBs in the CoV is also defined using the same conceptual approach as above, the data and assumptions used to derive the distribution are different. Similar to the increase in residential buildings, the number of new non-residential buildings in each DB is inferred from the number of new employments in the CoV as projected by the Metro Vancouver’s Regional Growth Strategy, which is 89,000 by year 2041 (Metro Vancouver, 2013). The underlying assumption is that an increase in employment implies an increase in businesses, which in turn can lead to an increase in non-residential buildings.   i. Status-quo land-use condition The following process is used to define the number of non-residential buildings in the Status-quo land-use condition. The number of new employment in a certain DB (enew) is:  !!"# = !!"#×!!"#      (4.5)  where ECOV is the projected number of new employment in CoV and pemp is the percentage of the CoV’s employment that is in this DB. Since the National Household Survey only have employment count data at the Dissemination Area level and not at the DB level, pemp is derived   84 from the business licensing data from the City of Vancouver in 2011, which provides the following information for each licensed business to help define pemp: • Business physical address • Business type and sub-type (descriptions of the business activity) • Number of employees  By geocoding each business address in GIS, it is then possible to determine which of the businesses belong to each DB in the CoV. pemp  of a certain DB can then be derived by summing the number of employees each business in that DB. Since pemp was computed using the business licensing data, which does not include employment in the public sectors, ECOV here is the projected number of new employments in CoV minus 16%, because public sector employment accounts for approximately 16% of CoV’s employments (i.e., 74,760).  Then, at a given DB, the number of new employment for an occupancy class x of non-residential building (enew_x) is: !!"#_! = !!"#  × !!     (4.6)  where px is the percentage of that DB’s employment hosted within buildings of occupancy class x:  !!"#$ = !!/!!"      (4.7)    85 Here, eDB is the total number of employment in the given DB. The number of employment inside non-residential buildings of occupancy class x (ex) is determined following several steps that are described in Appendix B2.   Using the above information, the average employment per occupancy class and the average number of businesses within each occupancy class of non-residential building can be computed to estimate the average number of employments in a business that resides in a building of occupancy class x (eav_x). Therefore the number of new non-residential buildings of occupancy class x in a DB (Cnew_x) is:  !!"#_! = !!"!_!!!!_! ;  !!"#_! = 0 !" !!"!_!!!!_! < 1    (4.8)  ii. Compact and Sprawl land-use condition The non-residential buildings in these two land-use conditions are determined using the same approach as for the Status-quo condition, except that for commercial buildings, which are classified as occupancy classes COM1 to COM10 [Table 4.4]. This is based on the assumption that the increase in dwelling units is associated with an increase in commercial businesses in the same area. Therefore in these conditions, employment increase in commercial buildings is defined to only increase where dwelling units increase. In that case, increase in non-residential buildings of occupancy classes COM1 to COM10 would be different between the three future land-use conditions (i.e., Status-quo, Compact, and Sprawl). Specifically, the increase in employment in commercial buildings (i.e., occupancy classes COM1 to COM10) in a given DB (enew_com) is:   86 !!"#_!"# = !!"#_!"#×!!"#$$_!"#    (4.9)  Where ECoV_com is the total number of new employments in commercial buildings in the CoV computed using the steps described in Appendix B2, and pdwell_new is the percentage of new dwelling units in the CoV located in the given DB.  Limitations	There are a number of major limitations in the way non-residential buildings are defined in the future land-use conditions: • The calculations above rely on average employment by occupancy classes of buildings and the average number of businesses by building occupancy classes that were calculated using the business inventory data, which only captures 170000 out of the 360000 jobs. Therefore, the averages are based on a sample of the employments in CoV rather than a complete dataset. • Since the number of new non-residential building of a certain occupancy class is determined by the number of new employment that would reside in that class of building, it is assuming that new buildings are made up of only one type of business – the business type expected to reside in that class of building. While most of the commercial occupancy class buildings are likely to be made up of their own type of businesses (e.g., COM10 parking, COM2 wholesale trade), in reality, some will have mixed-use in some places like Downtown Vancouver. Therefore this limitation should be considered when interpreting the results for businesses disruption impact for tertiary sectors in Downtown Vancouver.   87 • Another limitation that should be considered especially when interpreting results in the Downtown area is the assumption that the number of non-residential buildings increases with increase of businesses. In reality, the increase in the number of businesses is more likely to increase the number of floors of buildings rather than the number of non-residential buildings, especially on land that is already densely developed.   4.5.5 Power outage – resilience of electric power infrastructure Power outage is a key contributor of cascading impacts of flooding as a wide range of infrastructure and services has strong dependency on electric power. To develop the power outage conditions associated with different inundation conditions in this case study, the researcher elicited expert opinions from a number of staff members at BC Hydro – the electric power supplier of the CoV, as well as the rest of BC. Two key BC Hydro staff served as experts of the CoV’s electric power infrastructure – one is a consultant for the Transmission and Distribution Asset Investment Management of BC Hydro who has extensive experience with the CoV’s substations (25 years); another is a Planning Engineer for Distribution Planning of BC Hydro who has 12 years of experience working with the distribution assets of the power system in the CoV. In this section, the information about the potential impact of flooding on the electric power system and operation are elicited expert knowledge unless cited otherwise.  A storm or flooding event can lead to power outage through wind damage and/or water damage of transmission and distribution assets of the electric power system. Transmission assets are infrastructures that bring electric power from the generator to terminal stations, which connects power to the distribution assets [Figure 4.6]. The distribution assets include step-up transformers,   88 high voltage transmission lines, and terminal stations [Figure 4.6]. Once the power reaches the terminal station, it is connected to either large industrial users or to distribution substations where the power voltage is stepped down and further distributed to residential, commercial, and small and medium industrial customers.   Figure 4.6 Electric power transmission and distribution system   Although power outage in the CoV can be due to damage to assets located outside the boundaries of the CoV (e.g., damage to generator or transmission lines), the power outage conditions defined here will only account for power outage due to inundation of assets within the case study area based on the inundation conditions defined for this case study. Geospatial data showing the location of all transmission and distribution system assets located in the CoV was provided by BC Hydro. Using GIS, the assets potentially exposed to inundation under each of the 21   89 inundation conditions were identified as those that intersect the inundated areas. Asset types potentially at risk include: • Distribution cables (primary and secondary) • Manholes • Distribution transformers • Substations Conceptually, the DBs that would have power outage due to the flood damage of electric power system assets can be identified by mapping the area(s) the damaged asset serves. However, the applicability of this concept varies with different asset types. Inundation at locations of distribution cables does not necessarily damage the cable, since they are located on tall wooden poles, hanging at least 3m above ground. While storms are often associated with wind damage of trees that fall and damage power cables, causing power outage, this case study does not account for wind associated with each storm condition, as this requires high-resolution wind modeling that would be beyond the scope of this study.   Inundation at manholes can lead to inundation of assets located underground but it is difficult to estimate whether that would result in power outage since assets located underground are installed and built to withstand exposure to water.  Inundation of distribution transformers located above ground can result in power outage in areas that it serves. But the areas that a given transformer serves changes regularly as the configuration of the system changes based on demand and other circumstances. Furthermore, the associated outage is unlikely to last more than 24 hours due to the redundancy of the system in the CoV.     90 In contrast, inundation of a substation is likely to result in extensive power outage spatially and temporally, and the service area of a substation is defined and less likely to change. Another mechanism in which flooding can lead to power outage is when power is turned off in inundated areas to prevent electrocution and further damage to equipment.   In consultation with BC Hydro staff, it was determined that given how the substations are the only type of potential exposed asset that permit reasonable estimate of outage areas due to overland flooding, the use of a network model to parameterize and estimate likelihood of power outage due to overland flooding is unnecessary and unlikely to produce meaningful results. Therefore, the power outage conditions in this case study is characterized by two sets of simple assumptions about the substations’ operation status during a flood event and availability of power in inundated areas.   Two	sets	of	power	outage	conditions	Two sets of power outage conditions – Optimistic and Pessimistic (also called Worst-case) - were defined for this case study. Each shows the spatial extent of the potential power outage and assumed duration. The estimation of power outage extent is supported by the substation service area data and the following assumptions based on expert opinions from the BC Hydro key informants: • A substation in the CoV would most likely no longer be operational when it is subjected to 1.5m or more of inundation, • Power would most likely be turned off for areas that are inundated as a safety measure until inundation is cleared,    91 • Service areas of the Dal Grauer substation would be out of power if the Murrin substation is no longer operational, and service areas of the George Dickie substation would be out of power if the Kidd 1 substation is no longer operational (see Figure 4.7 for locations of the substations). The two power outage conditions are defined as follow:   Since there are 21 inundation conditions and two sets of power outage rules (optimistic and pessimistic), there are 42 power outage conditions defined for this case study. As an example, Figure 4.7 shows the pessimistic power outage condition associated with inundation from 1m of SLR and 1:500-year storm. The spatial distribution of the power outage conditions will be discussed further in Chapter 5. Maps of other power outage conditions can be found in Appendix A2. The BC Hydro staff that provided expert knowledge to develop the conditions reviewed "Optimistic/Likely" outage condition for a given inundation condition (PO) • All DBs with inundation would have no power, and • If a substation: o Has no inundation - no additional DB without power o Has inundation < 1.5m - no additional DB without power o Has inundation >= 1.5m - DBs in its service area would have no power o Duration - 24-72 hours "Pessimistic/Worse case" outage condition for a given inundation condition (PW) • All DBs with inundation would have no power • If a substation: o Has no inundation - no additional DB without power o Has inundation < 1.5m - DBs in its service area would have no power o Has inundation >= 1.5m - DBs in its service area would have no power o Duration - more than 72hrs    92 these power outage conditions, and a formal information-sharing agreement was made between the researchers and BC Hydro.  Figure 4.7 Pessimistic/worse case power outage conditions associated with 1m of SLR and 1:500-year storm inundation condition. DBs shaded in green are expected to experience power outage.  	Limitations	It is important to note that these power outage conditions are hypothetical and are by no means predictions of future power outages or flooding, thus, there is no information about the likelihood ####### ##MurrinCamosunKidd #1SperlingMainwaringDal GrauerGeorge DickieMount PleasantCathedral SquareWorst-case outage scenario of B1 flood scenario  93 of these conditions occurring in the future. These conditions are also highly simplified with various assumptions and limitations.   The extent of power outage represented by these conditions is likely to be an underestimation for several reasons. Firstly, broad range of power outage durations can be experienced during a flood event, from a few hours to weeks or months. But only prolonged power outage is defined in these outage conditions because shorter duration outages from damages to distributional assets, such as distribution cables, are not accounted for. This issue is considered to have minor impact on the resulting outage extent since all areas with any amount of inundation would be assumed to have no power in any case. Secondly, wind damage to power supply assets is not accounted for6, even though wind damage to wooden poles holding distribution cables is a common cause of power outage in this region. Lastly, the assumed duration of the outages resulting from inundation at a substation is likely to be significantly longer than 72 hours, up to weeks of outage. However, the duration is assumed to be more than 72 hours without specifying an upper limit due to high uncertainty.   On the other hand, there are also ways in which these power outage conditions can overestimate the outage extent. For example, inundation at the Murrin substation is the cause of extensive outages in the Downtown area in many outage conditions. However, BC Hydro is planning to decommission that substation in the near future and replace the load with a new substation in Mount Pleasant, which is less likely to be exposed to inundation. Furthermore, many commercial                                                 6 Wind damage was not accounted for as it would require high resolution modeling of future wind patterns, which is beyond the scope of this case study   94 buildings may have a backup power supply, such that the outage extent defined in these conditions only accounts for power outage from grid power supply and in reality, the building may not be as affected. However, backup power supply typically can provide power up to only 3 days. Therefore buildings or areas estimated to experience prolonged outages as defined in these conditions might still be affected after backup power fuel is spent and not refilled.  4.5.6 Building vulnerability to flood damage – stage-damage functions The last aspect characterizing the future scenarios in this case study is the vulnerability of different buildings to flood damage. In flood impact modeling, this is often estimated by stage-damage functions (SDFs), also called depth-damage curves, which determines the amount of damage a building is expected to have due to a given amount of inundation (Jongman et al., 2012). The expected damage can be measured in terms of percentage of the building that is damaged or in monetary terms. Usually, different SDFs are used for estimating damage in buildings of different sectors (e.g., residential, industrial) but different sets of SDFs may have different levels of refinement (i.e., microscale, mesoscale, and macroscale) (Prahl, Rybski, Boettle, & Kropp, 2016). The different scales are defined as follows (Merz, Kreibich, Schwarze, & Thieken, 2010): 1. Micro-scale - Damages are calculated for each affected object (building, infrastructure object, etc.) to help estimate the damage at the community level 2. Meso-scale – Damages are assessed at spatially aggregated units, such as land-use units (e.g., residential areas, industrial areas) 3. Macro-scale – Damages are assessed at large-scale spatial units, such as municipalities, regions, nations   95 There is now a broad range of flood damage models available, using different sets of SDFs. Some SDFs are only based on empirical data from observed flood event(s), while others are developed combining empirical data and/or expert judgment or literature review of existing studies (Scorzini & Frank, 2017).   Table 4.8 lists the flood damage models that have been considered for assessing direct building damage in this case study. While there are more flood damage models available than those listed in Table 4.8, these models are considered because 1) they are developed at cities and countries with relatively similar to CoV economically, 2) they are developed and used by government and academic research, and 3) they are well-documented.  Table 4.8 Flood damage models considered for modeling the building damage impact in this case study  Besides the uncertainties from how much assets and population may be exposed to flooding in the long-term future, the SDFs are the largest source of uncertainty in flood impact modeling Flood	damage	models	 Scale	 Data	source	types	Rhine	Atlas	(Rhine	Basin	Germany)		(ICPR,	2001)	Meso-scale	 Empirical	data	and	expert	judgment	FLEMO		(Germany)	(Kreibich,	Merz,	Seifert,	&	Thieken,	2010)	Meso-scale	 Empirical	data	and	expert	judgment	Damage	Scanner	(Netherlands)	(Klijn,	Baan,	De	Bruijn,	&	Kwadijk,	2007)	Meso-scale	 Empirical	data	and	expert	judgment	Flemish	Model	(Belgium)	(Vanneuville	et	al.,	2006)	Meso-scale	 Empirical	data	and	expert	judgment	JRC	Model	(Europe)	(Consultants,	2007)	Meso-scale	 Empirical	data	and	literature	review	Multi-Coloured	Manual	(MCM)	(UK)	(Penning-Rowsell	et	al.,	2005)	Micro-scale	 Empirical	data	Hazus-MH	(USA)	(Scawthorn	et	al.,	2006)	Micro-scale	 Empirical	data	and	expert	judgment	  96 (e.g., Bubeck et al., 2011, Cammerer et al., 2013). The reliability of the function would depend on how similar the situation being assessed (e.g., nature of flood event and types of buildings) is to the situation(s) from which the functions were developed (Cammerer, Thieken, & Lammel, 2013). A number of studies have shown that SDFs developed for a given place performs better than those transferred from elsewhere (e.g., Cammerer et al., 2013, Scorzini and Frank, 2017). Nonetheless, many places do not have local SDFs and resort to using SDFs transferred from other areas, which result in additional uncertainty in the modeling process. Since many impacts assessed in this case study are influenced by whether and to what extent buildings are damaged in the flood event, it is important to account for the uncertainty introduced by the SDFs used to assess building damage. SDFs are also a key source of uncertainty as they often do not account for other factors that can significantly influence the amount of damage buildings experience from flooding (e.g., inundation duration, flow velocity, precaution measures taken) (Kreibich et al., 2009; Thieken, Müller, Kreibich, & Merz, 2005).   Given that the CoV has yet to experience a major flood event, the City does not have SDFs that are specific to the city or the region. In this case study, building damage is assessed using the Canadian version of the Hazus model (previously introduced in the first section of this chapter), which has a default set of SDFs that is at micro-scale and suitable for community-level flood impact assessment [Table 4.8]. However, this set of SDFs was developed for communities in the mid-western and southern US (U.S. Federal Emergency Management Agency, 2009), which can be quite different to a highly urbanized coastal city in the Pacific Northwest, such as the CoV. An additional shortcoming of using the Hazus default SDFs is the omission of flow velocity and wave damage in the functions. However, the impacts in this case study may not be as sensitive to   97 this omission since building damage is less sensitive to wave damage in comparison to flood impacts on roads (Kreibich et al., 2009), and only a few areas in the CoV are expected to have significant waves during the flood events (Lyle & Mills, 2016). Nonetheless, the Hazus default SDFs are based on data from the 1980s and are limited in terms of their ability to resolve damage from shallow flooding, which is common in urban settings. Therefore, in addition to the Hazus default SDFs, building damage is also assessed using another set of SDFs in the hopes of compensating for the shortcomings of the Hazus SDFs. As shown in Table 4.8, there are various options of SDFs. The SDFs in the Multi-Coloured Manual (MCM)7 that was based on data from the UK, is selected for this case study for several reasons: • The MCM SDFs are one of the most updated sets of SDFs amongst those that were available at the time of selection. Originally developed in 2005 and the non-residential SDFs were completely revised in 2013. • In comparison to the SDFs in Hazus, MCM SDFs are based on flood damage data in more urbanized communities. • MCM SDFs can better resolve damage of shallow water flooding by providing estimated damage at smaller increments of flood depths. • Besides Hazus-MH, MCM is the only other flood damage model that is designed for assessing flood impacts at the microscale, which is ideal for community-level assessment.                                                  7 To access the MCM database that includes the SDFs, a 1-year educational license was purchased, which was valid from 2015 to 2016. Since the SDFs are proprietary, they are not shown this in dissertation.   98 To use the MCM SDFs to assess the building damage in this case study, the SDFs are imported into the Hazus flood model such that the only difference in the modeled building damage is due to the different SDFs. Several modifications were made to the MCM SDFs to fit the Hazus environment: a. Building classifications Although both Hazus and MCM have SDFs for a wide range of buildings with a similar level of specificity, there are discrepancies in the way buildings are classified. For example, building occupancy class RES1 represent a single detached home with sub-categories of those with and without basements. This is equivalent to the building categories Detached and Bungalow. Appendix B3 shows how each Hazus building occupancy classes are matched with an MCM building category. b. Damage values Since the MCM SDFs are absolute damage functions rather than relative damage functions, the estimated damage is measured in terms of British Pounds instead of percentage of the building damaged. Therefore the MCM SDFs needed to be translated into relative terms (i.e., percentage of damage) by dividing the estimated damage cost at each flood depth by the maximum damage cost in the corresponding SDF. However, since the maximum flood depth in the MCM SDFs is 3m (UK standard), this conversion method may result in overestimation of building damage in high-rise buildings.  It is also important to note that not all Hazus occupancy classes of buildings are equivalent to those in the MCM SDFs. These are indicated as “-” in Appendix B3. While SDFs for many of these categories of buildings are also not available in other flood damage models [Table 4.8],   99 SDFs are available for occupancy classes IND6 (Construction) and AGR1 (Agriculture) in the DamageScanner model, which is imported into Hazus to model the damage for these two building occupancy classes. Since the DamageScanner SDFs are measuring relative damage, the conversion made with the MCM SDFs is not required. The SDFs from DamageScanner are included in the MCM SDFs building vulnerability condition. By using two different sets of SDFs, two conditions of buildings vulnerability are defined for the future scenarios in this case study – 1) Hazus SDFs, and 2) MCM SDFs.   4.6 Impact models The second stage of the RIPs method is to geospatially model the SLR impacts in each future scenario defined in Stage 1. Therefore this section describes the different models that are employed to estimate the 14 SLR impacts and their respective data requirements and limitations.  To minimize the inter-model discrepancies in terms of structure and assumptions, it is ideal to use one flood impact model to estimate all the different types of impacts of interest. Since in principle the flood module of the Canadian version of Hazus has the capability to model direct and indirect impacts of flooding, each impact of interest is first checked to see if it may be reliably estimated in Hazus before considering other models. Hazus was first introduced in Section 4.4 where it was described as the GIS-based flood damage model that the CoV has used in their CFRA to geospatially identify population and infrastructure at risk of inundation. Besides being adapted specifically for use in Canada, Hazus is also chosen as the primary impact assessment tool in this case study because it has been shown to be generally superior to a number of other available flood impact models in terms of spatial resolution, as well as capability and   100 usability of flood impact assessment functions (Banks, Camp, & Abkowitz, 2014). Specific advantages include: a. Built-in SDFs from credible sources b. Built-in inventory of buildings and critical infrastructure c. Ability to use data imported by user (e.g., imported flood depth maps) d. Ability to perform and support spatial data viewing and processing  Nonetheless, many of the 14 impacts of interest here cannot be modeled directly within Hazus, except for direct building damages and the amount of debris generated. The other 12 impacts are modeled using either built-in functions of ArcGIS or specific impact models are implemented within the GIS environment. As shown in Table 4.9, some impact variables are assessed using similar modeling approaches. Therefore the following sub-sections group impacts that are modeled using the same or similar approach and describe how the models are implemented for this case study.    101 Table 4.9 Summary of modeling approaches used for assessing each SLR impact   4.6.1 Direct building damage and debris generated Both direct building damage and the amount of debris generated for each future scenario are modeled using Hazus. The first step is to import the scenarios defined for this case study into Hazus. The digital elevation model (DEM) data provided by the CoV and the inundation conditions associated with different storm return periods and SLR (i.e., flood surfaces) are imported into Hazus in raster format using the Flood Information Tool (U.S. Federal Emergency Management Agency, 2009). These data are then integrated into the Hazus’ flood model process by reading the flood depth at different locations in the CoV. Although the Canadian version of Hazus has built-in CoV building inventory database, the building inventory is edited to create three additional building inventories to represent the three other land-use conditions (i.e., Compact, Sprawl, and Status Quo) that are described in Section 4.5.4. Since one of the building 	 Impact	variable	 Modeling	approach	Economic	 Direct	building	damage:	Residential	Commercial	Public	&	governmental	Hazus	Canada	Business	disruption:		Primary	sectors	Secondary	sectors	Tertiary	sectors	Business	disruption	model	developed	by	Chang	and	colleagues	(S.E.	Chang,	Pasion,	Tatebe,	&	Ahmad,	2008)	Social	 Potentially	affected:	Vulnerable	population	Schools	Health	care	facilities	Social	service	facilities	Transportation	points	Emergency	services	GIS	spatial	analyses	to	quantify	population	or	infrastructure	within	inundated	areas	and/or	power	outage	areas	Environmental	 Debris	generated	 Hazus	Canada	Sewage	backup	damage	potential	 Sewage	Backup	Damage	Potential	Index	(SBDPI)	developed	by	researcher	  102 vulnerability conditions is defined by the Hazus SDFs, the MCM SDFs are also imported into Hazus as the other vulnerability condition to estimate building damage (as described in Section 4.5.6).  Direct	building	damage	The Hazus flood model allows the user to utilize its default general building inventory to estimate the direct physical damages of building structure and contents due to flooding and the consequential direct economic losses. The built-in building inventory consists of residential, commercial, industrial, agricultural, religious, government and education buildings. These buildings are classified according to the 33 occupancy classes [Table 4.4] and each class is described in terms of:  o Square footage (floor area);  o Full replacement value; and  o Building count o Mapping to their default building types (e.g., wood, concrete, masonry) As shown in Figure 4.8, to estimate the direct damage of a given occupancy class of building at a given DB, Hazus uses the first three attributes of the building occupancy class as well as the flood depth at that DB as input to the SDFs for that specific occupancy class. For each occupancy class, a different SDF is used for estimating building structural damage and building contents or inventory losses for the cases of industrial and commercial buildings. The resulting structural and content damage of that given occupancy class of building is then produced in terms of percentage of replacement cost, which can also be translated into cost in dollars.    103 Figure 4.8 General process of Hazus estimating the direct damage of a specific occupancy class of building.   In this case study, this method is used to model the direct building damage (sum of structural damage and content damage) at each DB in the CoV under each of the 336 future scenarios. Although Hazus models the direct building damage for each occupancy class, here the resulting damage per DB are aggregated into 3 general occupancy types (see Table 4.4 for the list of 33 occupancy classes):  1. Residential – All occupancy classes starting with “RES” 2. Commercial and industrial buildings – All occupancy classes starting with “IND” or “COM” 3. Governmental and public buildings - All occupancy classes starting with “GOV” or “REL” or “EDU” 	Debris	generated	Debris disposal can be a significant and costly impact of flooding. In contrast to earthquake debris from damaged buildings that includes both structural and non-structural components, most flood debris is building contents (e.g., furniture, appliances, etc.) and finishes (e.g., carpeting,   104 drywall, etc.). Even flood-fighting efforts and the floodwaters themselves can add additional debris, such as sandbags, mud, and sediment. However, the Hazus flood debris model only accounts for building-related debris (i.e., building finishes and structural components) and does not address building contents or additional debris loads from natural sources (e.g., vegetation). Therefore, the flood debris estimation in Hazus is only driven by the flood depth and the occupancy classes of buildings in the DB.   The flood debris model in Hazus estimates the amount of debris generated at each DB in terms of weight in tons. The total weight of debris can be made up of three components of the buildings – 1) building finishes, 2) structural components, and 3) foundation materials of buildings. Similar to the methodology for direct building damage, the amount of debris generated is estimated deterministically. First, the flood depth distribution throughout the DB and the square footage of a given occupancy class of building are used to determine how much of that class of building’s floor area are exposed to different flood depths. Then, the total amount of debris generated is determined based on the typical amount of debris generated from the three components of a given occupancy class of building (see Table 4.10 for examples). Whether a building of a given occupancy class has slab or footing foundation depends on its default foundation type defined in the Hazus built-in building inventory [Table 4.11].  The equation used by Hazus to estimate the amount of debris (in tons) generated from each occupancy class of building in a DB can be found in Appendix B4.       105 Table 4.10 Examples of typical amount of debris generated by different flood depth for different occupancy class of buildings (Part of Table 11.1 from Federal Emergency Management Agency (2009). Multi-hazard Loss Estimation Methodology, Flood Model Hazus-MH Technical Manual. Retrieved from https://www.fema.gov/media-library-data/20130726-1820-25045-8292/hzmh2_1_fl_tm.pdf. By permission from publisher.)  Table 4.11 Reclassification of building foundation types into two major foundation types (Table 11.2 from Federal Emergency Management Agency (2009). Multi-hazard Loss Estimation Methodology, Flood Model Hazus-MH Technical Manual. Retrieved from https://www.fema.gov/media-library-data/20130726-1820-25045-8292/hzmh2_1_fl_tm.pdf. By permission from publisher.) Foundation	type	by	occupancy	class	Reclassified	as	“slab-on-grade”	foundation	type	Reclassified	as	“footings”	foundation	type	Basement/garden-level	 ✔	 	Slab-on-grade	 ✔	 	Pile	 	 ✔	Solid	wall	 	 ✔	Pier/post	 	 ✔	Crawlspace	 	 ✔	Fill	 	 ✔	 Occupancy	 Depth	of	Flooding	Debris	Weight	(tons/1000	sq.	ft.)	Finishes	 Structure	Foundations	Footing	 Slab	on	Grade	RES1	(without	basement)	 0’	to	4’	 4.1	 	 	 		 4’	to	8’	 6.8	 	 	 		 8’	+	 6.8	 6.5	 12.0	 25.0	RES1	(with	basement)	 -8’	to	-4’	 1.9	 	 	 		 -4’	to	0’	 4.7	 	 	 		 0’	to	6’	 8.8	 	 	 		 6’+	 10.2	 32.0	 12.0	 25.0	RES2	 0’	to	1’	 4.1	 	 	 		 1’	+	 6.5	 10.0	 12.0	 25.0	RES3		(small	1	to	4	units)	 0’	to	4’	 4.1	 	 	 		 4’	to	8’	 6.8	 	 	 		 8’	+	 10.9	 6.5	 12.0	 25.0	  106 4.6.2 Business disruption Past disasters, such as Hurricane Sandy and Katrina, highlighted that businesses are highly vulnerable to disruption-related losses. For example, the total business losses after Hurricane Sandy was estimated to be USD$8.3B in the state of New Jersey, which includes about 19,000 small businesses that sustained more than USD$250,000 worth of damage (U.S. Congress, 2013). Affected businesses represent a wide variety of industries and losses were mostly due to temporary closure (Henry et al., 2013). While business closures following a disaster are a result of many reasons, the loss of lifelines (e.g., electric power, water) is consistently found to be the primary contributor (Chang, 2003; Tierney & Nigg, 1995). Therefore, business disruption in each future flood scenario in this case study is measured in terms of the number of businesses expected to experience temporary closure due to lifeline disruption and/or direct building damage induced by the flooding.   The business disruption model (referred as “BD model” hereafter) developed by Chang and colleagues (2008) is used to assess the business disruption in each flood scenario. Although this model was developed based on loss data from earthquake events rather than flooding, it is designed to be readily applicable to other types of hazards because the model inputs (i.e., direct building damage, lifeline disruptions) can be estimated for specific hazard events (Chang et al., 2008). The model was adopted for this case study due to its empirical basis and explicit treatment of lifeline disruptions, which is the primary cause of business closures during flood events. Besides the regions of Memphis (Chang & Shinozuka, 2004; Shinozuka, 1995) and Los Angeles (Chang et al., 2008), this BD model has also been adapted and applied to estimate the business losses associated with different earthquake scenarios in the District of North Vancouver   107 (Lotze, 2014). While there are conceptual frameworks to guide the assessment of flood impact on commercial buildings (e.g., Bhattacharya et al., 2013), these are not based on empirical damage data and do not provide a numerical or statistical function to quantify the potential loss.   This BD model estimates whether a business of a given sector would experience temporary closure due to the overall level of disruptiveness from 3 sources - building damage (Ds), water outage (Dw), and power outage (De) [Figure 4.9]. Table 4.12 lists the major sectors included in this model.  Figure 4.9 Schematic structure of the business disruption model (S.E. Chang et al., 2008)   Table 4.12 The BD model estimates the operating status of businesses of these major sectors Major	sector	categories	Agriculture	(AGR)	Mining,	construction,	transport,	communication,	and	utilities	(MCT)	Manufacturing	(MFG)	Wholesale	and	retail	trade	(TRD)	Finance,	insurance,	and	real	estate	(FIR)	Health	services	(HTH)	All	other	services	(SVC)	   108 To do this, the model consists of temporary closure probabilities for each level of disruptiveness based on empirical business disruption data derived from surveys of more than 2,000 businesses after the 1994 Northridge and 1989 Loma Prieta earthquakes. For example, the model uses the probability distribution of manufacturing sector businesses that close temporarily to estimate whether a given manufacture business would close temporarily given a certain level of building damage, power outage, and water outage associated with the given hazard event. To account for the inherent uncertainty of disaster events, a Monte Carlo approach is used to determine the most likely operating status of the business. This will be further explained below.  Although this BD model accounts for disruptiveness from building damage, as well as power and water outage, this case study assumes that there is no water outage as there is insufficient evidence pointing to flooding in the CoV that is associated with disruptions to the water supply. Another limitation of this model is how it treats businesses as independent entities. Besides their dependence on various lifelines (e.g., electrical power, water), they are also vulnerable to flood-induced losses through various other issues, including supply chain disruption, reduction in customer traffic and change in demand post-disaster, and reduction in employee productivity (Tierney, Webb, & Dahlhamer, 2000).  Since this BD model estimates business temporary closure on the basis of individual businesses, the model must be first applied to individual businesses in the CoV, then aggregated by DB to estimate the total number of businesses per DB that would be disrupted. Data of individual businesses in the CoV is drawn from the CoV’s business licensing data in 2011-2012, which includes attributes such as business address, name, business type and sub-types, and the number   109 of employees. These data were also used for developing the number and distribution of new non-residential buildings under each land-use conditions, as described in Section 4.5.4.  Before the Hazus direct building damage results can be used as input for this BD model, they must be translated from building damage count into building damage probability distributions by specific occupancy classes. Hazus models the direct building damage for a given DB as the number of buildings that would sustain different levels of damage (i.e., 0%, 1-10%, 11-20%, 21-30%, 31-40%, 41-50%, 51-60%, 61-70%, 71-80%, 81-90%, and 91-100% damaged). Table 4.13 shows a hypothetical example of the modeled building damage count for the occupancy class COM1 at a given DB and the corresponding translated building damage probability distribution. To convert this to building damage probability of different states of damage, the building counts for damage levels 1-10% and 11-20% are summed and divided by the total number of buildings in the DB to determine the proportion of buildings that have none to minor damage (i.e., 1-20% damage). The same is done for other combinations of damage levels. Then the cumulative probability distribution across different damage states is computed by summing the proportions cumulatively for each level of damage.      110 Table 4.13 Hypothetical example showing how the Hazus building damage count by different damage levels can be converted to cumulative probability distribution. Hazus	damage	levels	Number	of	occupancy	class	COM1	buildings	Building	damage	states	for	BD	model	Cumulative	probability	for	a	DB	with	total	building	count	of	102	COM1	buildings	1-10%	 10	 Minor	to	none	 0.127	11-20%	 3	21-30%	 16	 Slight		 0.284	31-40%	 0	41-50%	 31	 Moderate		 0.784	51-60%	 20	61-70%	 11	 Extensive		 0.941	71-80%	 5	81-90%	 0	 Complete		 1.000	91-100%	 6	 To implement this BD model for an individual business in the CoV under a given future scenario in this case study, the following steps are carried out: Step 1: Identify the DB and occupancy class of building that the business resides in. Since building damage (by occupancy classes) and power outage are modeled and defined at the DB level, respectively, the first step is to identify which DB and in which occupancy class building the business resides in. This will help to determine what level of building damage the business is expected to sustain and whether it will have a power outage in the given future scenario. The DB in which the business resides in is identified using the spatial join function in GIS to determine the business address within which the DB falls. The occupancy class of building to which the business belongs to is determined based on the business’s address, its building’s primary use, and the type of businesses that Hazus is expected to have inside this type of building (more details on this process can be found in Appendix B2).   111 Step 2: Determine the damage state of the business’s building and whether there is a power outage in the business’s DB. The level of damage sustained by the business’s building is determined using a generated random number (between 0 and 1) and the building damage probability distribution for the occupancy class to which the business belongs. Using the COM1 occupancy class example in Table 4.13, with a random number of 0.36, the business building would be assigned a moderate damage. Then the level of disruptiveness associated with the assigned level of damage is determined using Table 4.14. In this example, level of disruptiveness will be “disruptive” (D).   Table 4.14 Levels of disruptiveness associated with each state of building damage      The power outage condition associated with this future scenario is then used to determine whether the DB in which this business reside is expected to have a power outage. If the DB is not expected to have a power outage, then the business is assigned the level of disruptiveness of “not at all disruptive” (NAA). If it is expected to have a power outage, then based on the business’s sector, the level of disruptiveness is determined using Table 4.15. The business types and sub-type attributes in the CoV’s business licensing data are used to determine to which major sector the business belongs to. The major sector category corresponding to each business type and sub-type can be found in Appendix B5 and B6, respectively. The sector category the business belongs to is first determined based on its business type, and the sub-type is only used if the Building	damage	state	 Disruptiveness	Level		None	 Not	at	all	disruptive	(NAA)	Slight	 Not	very	disruptive	(NV)	Moderate	 Disruptive	(D)	Extensive	 Very	disruptive	(VD)	Complete	 Very	disruptive	(VD)	  112 business type is not clear. The matching of business types and sub-types to different sector categories is conducted based on a different project led by Chang (2014), where the matching is based on the descriptions of the major sector categories, business types, and sub-types. As for the disruptiveness of building damage, a random number is generated to determine which level of disruptiveness the business may experience from a power outage [Table 4.15].  Table 4.15 Level of disruptiveness from a power outage in each major sector (Table 2.4 from © Chang et al. (2008). Linking Lifeline Infrastructure Performance and Community Disaster Resilience: Models and Multi-stakeholder Processes. MCEER, Page 15. Adapt by permission from publisher)      Step 3: Determine whether the business will be temporarily closed based on the cumulative level of disruptiveness from each source using Table 4.16. This table shows the probability of temporary closure for a business at different level of disruptiveness and the sample size of disrupted businesses in the empirical business disruption data that was used to generate the probabilities. First, determine which case of overall disruptiveness (A-F) the business is experiencing based on the level of disruptiveness from a power outage and/or building damage. For example, if the business is assigned to be “not very disruptive” due to power outage and “very disruptive” due to building damage, then it will fall under category C because there is one source of disruption that is very disruptive to the business. Then generate another random number between 0 and 1. If the random number is greater than the percentage of businesses Disruptiveness	level	Major	sector	categories	AGR	 MCT	 MFG	 TRD	 FIN	 HTH	 SVC	NAA	 0.09	 0.11	 0.03	 0.04	 0.04	 0.04	 0.07	NV	 0.31	 0.33	 0.19	 0.21	 0.19	 0.23	 0.26	D	 0.62	 0.59	 0.37	 0.48	 0.45	 0.42	 0.46	VD	 1.00	 1.00	 1.00	 1.00	 1.00	 1.00	 1.00	  113 closed due to that level of overall disruptiveness, then it will be assigned a deterministic state of “closed”. Otherwise, it will be assigned “open”.   Since the Monte Carlo approach is employed here, steps 1 to 3 are carried out 100 times for each business to determine the most likely business operating status (i.e., open or close), where a business is deemed closed if more then 50 simulations indicate the business closes due to the cumulative level of disruption.  Table 4.16 Temporary business closure from multiple sources of disruption (Table 2.6 from © Chang et al. (2008). Linking Lifeline Infrastructure Performance and Community Disaster Resilience: Models and Multi-stakeholder Processes. MCEER, Page 19. Adapt by permission from publisher)  Although the outcome at this point is the number of businesses in each major sector that may experience temporary closure, for this case study the results are aggregated by primary, secondary, and tertiary sectors as shown Table 4.17.  Table 4.17 Classification of major sector categories into primary, secondary, and tertiary sector Major	sector	categories	 Sectors	Primary	sector	businesses	 AGR,	MCT	Secondary	sectors	businesses	 MFG	Tertiary	sector	businesses	 FIN,	HTH,	SVC,	TRD		Case	 Number	of	sources	in	each	disruptiveness	category	Percent	closed	Sample	size		(empirical	business	disruption	data)	NAA	 NV	 D	 VD	A	 	 	 	 2+	 90%	 115	B	 	 	 1+	 1	 80%	 55	C	 	 	 0	 1	 63%	 114	D	 	 	 1+	 0	 54%	 133	E	 	 1+	 0	 0	 30%	 128	F	 3	 0	 0	 0	 4%	 268	  114 4.6.3 Social Impacts The six social impacts modeled for this case study represent the key social assets and vulnerable population that can be affected by prolonged power outage and/or flooding. Potentially affected assets are those that lie within DB with a power outage or inundation. Table 4.18 outlines these six social impacts, their descriptions, and data sources. Except for the vulnerable population potentially being affected, all social assets have specific locations and are geocoded within the ArcGIS environment to identify which asset is located within inundated areas and/or DBs with power outage using the spatial join function8. The same approach is used to estimate the amount of vulnerable population potentially affected, except that the vulnerable population data do not require geocoding as the data are aggregated at the DB level. All social impact variables are modeled for each future scenario and aggregated at the DB level.                                                   8 The spatial join function is a built-in function within the ArcGIS’s Spatial Analyst extension   115 Table 4.18 Social impact variables modeled, their descriptions, input data variables and sources. Social	impact	variable	Description	 Data	variables	 Data	source	Vulnerable	population	affected	Vulnerable	population	potentially	affected	by	flood	water	exposure	and/or	prolonged	power	outage	Pop	age	65+		 Census	20119	Pop	age	under	16	Pop	under	poverty	line	(income	under	$10000)	Schools	affected	Number	of	schools	affected	by	floodwater	exposure	and/or	prolonged	power	outage	Schools	 City	of	Vancouver	Open	Data	Catalogue10	Health	care	facilities	affected	Number	of	health	care	facilities	affected	by	floodwater	exposure	and/or	prolonged	power	outage	Hospitals	 DataBC	11	Walk-in	clinics	Pharmacies	Social	services	affected	Number	of	social	service	facilities	affected	by	floodwater	exposure	and/or	power	outage.		Homeless	shelters	 City	of	Vancouver	Open	Data	Catalogue	Free	meal	locations	Senior	centres	Community	centres	Non-market	housing	Childcare	and	preschool	centres	Requested	from	Ministry	of	Family	and	Children	Transportation	 Number	of	transportation	points	affected	by	floodwater	exposure	and/or	power	outage.		Bus	stops	 Translink	via	UBC	Abacus	Skytrain	stations	 City	of	Vancouver	Open	Data	Catalogue	Gas	stations	 Canmap	Content	Suite	2015	12	Sea	bus	terminals	 DataBC	and	False	Creek	Ferries	                                                9 https://www12.statcan.gc.ca/datasets/Index-eng.cfm 10 http://data.vancouver.ca/datacatalogue/index.htm 11 https://data.gov.bc.ca/ 12 2015-11, "CanMap Content Suite, v2015.3", <a href="http://hdl.handle.net/11272/P66YQ">hdl:11272/P66YQ</a> DMTI Spatial, Inc. [Distributor] V1 [Version]   116 Social	impact	variable	Description	 Data	variables	 Data	source	Emergency	services	Number	of	emergency	service	facilities	affected	by	floodwater	exposure	and/or	power	outage	Ambulance	services	 Exported	from	Google	Map	in	KML	format	Fire	halls	 City	of	Vancouver	Open	Data	Catalogue	Police	stations	 Extracted	from	CanMap	Content	Suite	2013	enhanced	POI	file	13	 4.6.4 Sewage Back-up Damage Potential Based on statistics from Insurance Bureau Canada, water damage associated with basement flooding and plumbing failure is one of the most significant causes of home insurance claims in Canada (Insurance Bureau of Canada, 2014). In ground-related homes14, basement flooding can result from three major mechanisms during an overland flood event (Sandink, 2016): a. Infiltration or seepage through cracks in the foundation wall when groundwater levels exceed the lowest level of the basement. b. Overland stormwater surcharge when the stormwater management system is overloaded. The surcharge enters the home through above-grade openings, such as doors and windows. c. Sewage backup from surcharging underground sewer systems, especially in combined systems where stormwater and wastewater is connected and conveyed through the same network. During an overland flood event with excessive                                                 13 2013, "Enhanced Points of Interest, v2013.3", <a href="http://hdl.handle.net/11272/10095">hdl:11272/10095</a> DMTI Spatial, Inc. [Distributor] V2 [Version] 14 Includes detached, semi-detached or townhomes with basements or foundations at or below grade, and excludes high-rise residences, including apartments.    117 stormwater load, combined systems can surcharge and sewage can flow back into homes through basins, shower drains, toilets, and bath drains.  Besides building damage from floodwater inundation, secondary exposure to contaminated floodwater or untreated sewage from sewage backup can lead to various adverse health impacts. Poor air quality and mould growth can subsequently exacerbate respiratory health conditions (Ivers & Ryan, 2006; Taylor et al., 2011). Exposure of occupants to sewage can result in faecal-oral transmission of disease (Ahern, Kovats, Wilkinson, Few, & Matthies, 2005) and contamination of building materials facilitating growth of human pathogens (Taylor et al., 2011).  While it is ideal to assess the risk of basement flooding from all three mechanisms, only sewage backup is assessed in this case study due to the lack of data for a number of critical variables. Basement flooding from infiltration and overland stormwater strongly depends on the specific structural conditions of the homes, such as cracks in foundation walls in the basement, and the grading of the lot around the home, respectively (Sandink, 2009). Such building-specific data are currently not available unless the home has recently been assessed. Sewer backup, on the other hand, is influenced by many factors that either have data available for the CoV or can be inferred from reasonable proxies.  	4.6.4.1 Measuring sewage backup risk Since sewer backup due to overland flooding is a result of dynamic interactions between multiple factors, the associated damage or risk cannot be easily assessed using simple spatial analysis. There are two existing measures that specifically measure sewage backup risk. The Insurance Bureau of Canada developed the Municipal Storm and Sanitary Infrastructure Risk Assessment   118 Tool (MRAT) to quantify the probability of municipal water infrastructure failure using a 20 variable risk formulae (Tremblay, 2011). This tool is now the property of Tesera Systems Inc. Another is the Sewer Backup Risk Score developed by CoreLogic®, a global real estate information analytics service provider based in the United States (Guyton & Hurst, 2015). The Sewer Backup Risk Score measures the potential risk of sewage backup based on location-specific data, including hydrological information, flood risk, neighbourhood characteristics, property characteristics, terrain, and other geological information (Guyton & Hurst, 2015). Unfortunately, both of these measures of sewage backup risk are proprietary and cannot be employed for this case study. Furthermore, as far as the researcher is aware, there are currently no such measures published in the academic literature that can estimate location-specific sewage backup risk or damage associated with overland flooding. Therefore a new composite index - Sewage Backup Damage Potential Index (SBDPI) - is developed for this case study to estimate the potential risk of sewage backup at each DB in the CoV. The following sub-sections will describe the development process of this index and how it is operationalized in this case study.  4.6.4.2 Development of the Sewage Backup Damage Potential Index (SBDPI) This section describes the process involved in developing the Sewage Backup Damage Potential Index (SBDPI) for ground-related residential buildings15 in the CoV. While sewage backup can be caused by a number of different ways, such as sewer main blockage by tree roots or improper                                                 15 Includes detached, semi-detached or townhomes with basements or foundations at or below grade, and excludes high-rise residences, including apartments.     119 flushing, this index aims to measure the relative risk of overland flood-induced sewage backup in ground-related residential buildings (referred as homes hereafter) at each DB in the City.  	Theoretical framework The widely recognized concept that risk is a function of hazard, exposure and vulnerability (see Equation 4.10) (Birkmann, 2006; Cuny, 1984; UNDRO, 1980; UNISDR, 2004; van Aalst, 2006) forms the basis and starting point for the theoretical framework of this index.   !"#$ = ℎ!"!#$ × !"#$%&'! × !"#$%&'()#)*+     (4.10)  where hazard refers to the intensity or recurrence of the hazard event; exposure refers to the inventory of elements in an area in which hazard events may occur; vulnerability refers to the propensity of exposed elements such as human beings, their livelihoods, and assets to suffer adverse effects when impacted by hazard events. To apply this concept to measure the relative risk of flood-induced sewage backup at a DB:  • Hazard refers to the extent to which the flooding event is overwhelming the sewer system of a building;  • Exposure refers to the number of homes potentially exposed to sewage backup in the DB; and  • Vulnerability refers to the propensity of the homes within the DB to be adversely affected by the hazard (i.e., propensity of sewage entering the interior space of the building)    120 To identify suitable indicators to represent the above three components of sewage backup risk, a literature review was conducted to better understand how overland flooding can lead to sewage backup in homes, and factors that influence the likelihood of sewage backup occurring in a building during a flood event at, or nearby, the building location. The likelihood of sewage backup at a given home is strongly influenced by the likelihood of the sewer system at being overloaded, which can be estimated using network modeling. However, once the sewer system surcharges, the propensity of sewage backing up into a home strongly depends on the structural design of the home and wastewater policies adopted by the municipality, both of which vary considerably across Canada and cities (Sandink, 2016). Therefore findings in the literature review must be supplemented and verified with local experts. An expert opinion elicitation session was conducted with a manager and a supervisor of the City of Vancouver’s Plumbing, Gas and Sprinkler Inspections. These two City staff are referred to as City informants from here onwards. The sub-sections below describe the findings.  Mechanisms of sewage backup in the CoV Most urban homes are either served by a combined municipal drainage system (i.e., the sewer and stormwater drainage system is connected and sharing the load) or a separated system (i.e., the sewer and stormwater systems are not connected) [Figure 4.10]. In the combined sewer system, the additional stormwater load can overload the system causing the wastewater to flow back into the homes through various drainage connections, including toilets, bath and basin drains, and floor drains (Sandink, 2009).     121 Although flood-induced sewage backup is more commonly associated with combined sewer systems, sewage backup can also occur in a separated sewer system (Sandink, 2016).In either case, the sewer can be overloaded, but whether the wastewater can enter the home (i.e., is vulnerable to sewage backup) depends on whether there are mechanical means to stop the backflow. In the City of Vancouver, the common mechanical means that can prevent backflow of sewage is a sewage pump, which mechanically pumps wastewater from the home into the municipal sewer system. The sewage pump is normally powered electrically by grid electricity. Installing a backup battery to the pump is optional. Although the pump may stop operating during a power outage, the pump can still serve as a backflow preventer even when it is not actively pumping sewage.   Figure 4.10: Comparing combined and separated sewer systems (Figure 1 from Potera (2015). After the Fall. Environmental Health Perspectives, 123(9), Page A243. Available at: https://ehp.niehs.nih.gov/123-a243/ Reproduced with permission)  Whether a ground-related home has a sewage pump depends mostly on:   122 • Whether the building needs to have a sewage pump according to the City’s fixture restriction policy (City of Vancouver, 2008), which specifies that: o A ground-related home with an elevation of lowest slab below the fixture restriction elevation must have a sewage pump to actively eject wastewater from all sanitary fixtures into the municipal sewer system. o A home with elevation of lowest slab above the fixture restriction is not required to install a sewage pump and its sewage is drained to the municipal sewer by gravity • Whether the homeowner obeys the restriction policy; and • Year of construction – buildings built before the 1970s are unlikely to have mechanical means since the fixture restriction was not yet created. While a home located within the area of an overland flood is at risk of sewage backup, homes located outside the area can also be at risk given that the sewer system is a connected network. Therefore, all homes served by the surcharging regions of the sewer system are potentially exposed to the risk of sewage backup.   Factors influencing likelihood of sewage backup  Based on the understanding about the way sewage backup can occur in ground-related homes during a flood event in the City of Vancouver, the researcher and the City informants agreed upon the list of variables that are potential indicators to measure the hazard, exposure, and vulnerability components of flood-induced sewage backup risk at a ground-related home [Table 4.19].     123 Indicator data Data for all the indicators in Table 4.19 are available except for the sewage system load during a flood as there were insufficient data to conduct a reasonable drainage system model to estimate the potential load. Therefore, the surcharge potential due to sewage system load will need to be measured by the nearest flood depth by proxy. The indicator and its data sources are described in Table 4.20.  Table 4.19 Potential indicators to measure the three components of sewage backup risk Hazard	 Exposure	 Vulnerability	• Flood	depth	(m)	• Sewage	system	load	during	flood	(%)	• Distance	from	the	flooded	area	or	surcharged	parts	of	the	sewer	system	(m)	Ratio	of	number	of	ground-related	homes	inside	the	dissemination	block	to	the	total	number	of	ground-related	homes	in	CoV	• Sewage	pump	installed	(yes/no)	• Power	outage	(yes/no)	• Served	by	a	combined	sewer	system	(yes/no)	• Age	of	building	 Table 4.20 Data sources for indicators used in the Sewage Backup Damage Potential Index (SBDPI) Indicator	 Unit	 Data	source	Flood	depth	at	the	home’s	location	m	 Inundation	conditions	defined	for	each	the	future	scenario	Distance	from	the	flooded	area	m	 Inundation	conditions	defined	for	each	the	future	scenario	Sewage	pump	installed	 Yes/no	 Inferred	from	whether	the	City’s	fixture	restriction	applies	to	the	home.	Fixture	restriction	data	was	attained	from	the	City’s	Department	for	Permits	and	Licensing		Power	outage	 Yes/no	 Power	outage	conditions	defined	for	the	future	scenario	Combined	sewer	system	type		Yes/no	 City’s	Department	for	permits	and	licensing	  124 Indicator	 Unit	 Data	source	Number	of	ground	related	residential	buildings	#	 Census	Canada	2006	16	and	HAZUS	Canada	General	Building	Stock	Inventory	Number	of	ground	related	residential	buildings	constructed	before	1970s	#	 Census	Canada	2006	 Relative weighting of vulnerability indicators While the hazard and exposure indicators are quantifiable, vulnerability indicators are qualitative. Furthermore, it is not clear how much each of the vulnerability indicators contributes to the home’s vulnerability to sewage backup. Therefore the City informants were asked to assign relative vulnerability scores to each combination of the vulnerability indicators as shown in Table 4.21.   Table 4.21 Elicited sewage backup vulnerability scores for each drainage setup                                                 16 http://www12.statcan.gc.ca/census-recensement/2006/ref/rp-guides/housing-logement-eng.cfm Combination	ID	Drainage	system	[Separate/Combined]	Power	outage	(More	than	1	week)	[Yes/No]	Sewage	pump	[Yes/No]	Vulnerability	score	(p)	[Very	low	(0)/	Low	(1)/	Med	(2)/	High	(3)/	Very	high	(4)]	1	 Combined	 Yes	 Yes	 2	2	 Combined	 Yes	 No	 3	3	 Combined	 No	 Yes	 0	4	 Combined	 No	 No	 3	5	 Separate	 Yes	 Yes	 2	6	 Separate	 Yes	 No	 1	7	 Separate	 No	 Yes	 0	8	 Separate	 No	 No	 1	9	 Ground	related	home	constructed	pre-70s	 4	  125 4.6.4.3 Sewage Backup Damage Potential Index (SBDPI) for the CoV The indicators are aggregated based on the theoretical framework discussed earlier in this section where the risk is a function of the hazard, exposure and vulnerability. Since risk is commonly considered to include a notion of probability, which is only implied by different return periods of each inundation condition of the scenarios, this index measures the damage potential of sewage back up rather than the absolute risk. Starting by considering the sewage backup risk at a home located within a flooded area. The SBDPI (rf) (Equation 4.11) for a given home located within a flooded area would only be a function of flood depth and the homes’ vulnerability score based on Table 4.21, as there would be no distance from the flooded area and number of homes exposed would equal to one. !! =  ! × !       (4.11) where F is the maximum flood depth of the flooded DB in metres, normalized using the maximum and minimum flood depth across all inundation conditions from the defined scenarios  and p is the vulnerability score based on Table 4.21.  To modify the index to measure the sewage backup damage potential for the entire flooded DB, the index must account for how many homes in the DB are exposed and their different vulnerability scores. Equation 4.12 represents the SBDPI for a DB located within the flooded area (Rf).  !! = ! × ! ×  !!!!      (4.12) where F is the normalized flood depth as defined above, q is the number of ground-related homes in the DB normalized by the max-min number of ground-related homes in each DB in the   126 City; p is the vulnerability score based on Table 4.21; and u is the percent of ground-related homes in the DB with a given vulnerability score p.  To modify the index for DBs outside the flooded area, the hazard component of the index is modified to account for the distance from the flooded area. Given the expert opinion that the risk should decrease with distance from the flooded area and increase with flood depth, the SBDPI for DBs located outside the flooded area [Rnf] is represented in Equation 4.13. !!" = (!!) × ! ×  !!!!       (4.13)   ! = |! − 1|       (4.14)  where C is the closeness of the DB to the nearest flooded DB and D is the distance (in meters) from the nearest flooded DB, normalized by the maximum and minimum distance across all included flood scenarios17. For this case study, the SBDPI was computed using Python and visualized in GIS. 	4.6.4.4 Index validation In principle, it is ideal to validate the index by comparing the result with observed data, such as sewage backup insurance claims or complaints received by the City from residents that experienced sewage backup due to malfunctions in the municipality-owned portion of the sewer system. However, these types of empirical data at the CoV would not serve the purpose of                                                 17 The distance is measured by the centroids of each dissemination blocks   127 validation as very few of the associated sewage backup cases would have been caused by an overland flooding event. Therefore, the City informants that have contributed to the index development reviewed the index results as a form of verification and an opportunity to identify shortfalls. Instead of asking the City informants to review the index results for all 336 scenarios, they were asked to review the best and worst cases (i.e., 0m SLR with 1:50-year storm, and 6m of SLR with 1:10,000-year storm). This review process involved the researcher presenting the maps of the index results as well as each input variable (e.g., power outage conditions, ground-related homes per DB), and explaining the possible factors attributable to various hotspots. For example, some areas with high SBDPI may be due to the high concentration of homes constructed before the 1970s, while others may be due to the relatively high concentration of homes relying on sewage pumps coinciding with a power outage. The City Staff were also provided with a summary table outlining the index results by neighbourhoods, which can be found in Appendix B7.  4.7 Self-organizing Maps Analysis Stage 3 of the RIPs method can begin after all impacts are modeled geospatially in all future scenarios defined in Stage 1. Training the SOMs with the resulting impact maps as its learning data identifies the RIPs of each impact. For this case study, the SOMs Toolbox in Matlab (Version R2015b) is used to train and evaluate the SOMs, the Python programming language (Version 2.7-3.0) is used for processing the input and output of the SOMs, and the resulting RIPs are mapped in ArcGIS and automated by Python programming. While the structure of the SOMs algorithm is described in detail in Section 3.1.2 of Chapter 3, this section describes the specific   128 parameters used to train the different sets of SOMs to identify the most suitable set for this case study.   4.7.1 Training the SOMs The user must specify a number of parameters to train the SOMs to identify the desired number of RIPs. As mentioned in Chapter 3, there are no right or wrong values for these parameters as they largely depends on the application purpose and the optimal values are determined through trial and error (i.e., training multiple SOMs with different configurations of parameters). The following describes the parameters that must be specified for a basic SOMs training and some guidance based on the Technical Manual for the Matlab SOM Toolbox (Vesanto, Himberg, Alhoniemi, & Parhankangas, 2000):  • Number of reference vectors to initiate This parameter determines the number of RIPs to extract from the impact maps (i.e., size of the SOMs). The lower the number of RIPs (or reference vectors), the less variations of the data are represented by the RIPs collectively. The user should specify this parameter based on its application purpose. In this case study, the researcher chose to first train the SOMs with 16 reference vectors based on the thinking that it is probably the maximum number of RIPs that users can meaningfully comprehend when it is further reduced to 3-4 groups for interpretation18.  • Shape and lattice of the array of reference vectors in the data space Practically, the shape of the array (e.g., rectangular, hexagonal) determines how the                                                 18 The purpose and method for this grouping will be discussed in Section 5.2 in Chapter 5   129 reference vectors are located relative to each other in the data space. To allow the reference vectors to span across the data space effectively, the shape should correspond to the data manifold. Therefore, the “sheet” shape is commonly used while “toroid” and “cylinder” shapes are only recommended when the data are found to be circular. The lattice can be rectangular or hexagonal. An hexagonal lattice can produce a SOM with smoother variations from one extreme pattern to another but this effect is not apparent when the size of the SOMs is small, as in this case study. If a rectangular shape is chosen, the resulting RIPs should be arranged in a grid with the most similar RIPs being closer on the grid while the most dissimilar RIPs should be on opposite corners. • Learning rates This parameter determines how much the reference vectors are modified to become more similar to the winning input vector at each training cycle. Therefore, to balance the need to save computing time and to capture both coarse and fine variations in the input data (i.e., impact maps), it is common to begin the training with higher learning rate to capture coarse variations and then decrease the rate with each training cycle to another lower specified learning rate to capture the finer variations. Therefore, the user should specify the “coarse” and “fine” learning rate. By default, the coarse rate is 0.5 and the fine is 0.05. • Number of training cycles Knowing that reference vectors are trained using different learning rates, the coarse and fine training cycles specify how many cycles are conducted in the training is done using coarse and fine learning rates. One training cycle is complete when all input vectors have been presented and compared with the reference vectors once.  Each time an input vector   130 is presented counts as a training step, and typically the number of training steps should be at least ten times the number of reference vectors, but there is no maximum. • Neighbourhood radius This parameter determines the size of the neighbourhood in which the reference vectors will be modified at each training step. It is recommended that the neighbourhood radius starts from one fourth of the SOMs size during the coarse training and reduces to one fourth of that until the last training cycle of coarse training. During the fine training phase, the neighbourhood radius starts from where it stopped in previous phase, and goes to one.  To develop the optimal set of RIPs for the application purpose, a number of different SOMs with different configurations of parameters should be trained. Table 4.22 shows the parameter specifications for the four different SOMs trained for this case study. The parameters used in SOM #1 represent mostly default values selected based on the above guidelines and aim to include all 14 SLR impact variables. SOM #2 differs from #1 by only including the eight impact variables across more scenarios than the other six. For example, by the design of the respective impact models, business disruption varies with different inundation, land-use, power outage, and building vulnerability conditions. However, the number of affected emergency services would only vary with different inundation and power outage conditions because the number of emergency services is the same across different land-use conditions and they are not accounted for in the different building vulnerability conditions. Therefore SOM #2 is trained to test whether this difference would produce patterns of the eight impacts that are different from those produced by SOM #1.    131 Table 4.22 Parameters used for the four SOMs trained using the modeled impacts for this case study.  Parameters	SOMs	trained	#1	 #2	 #3	 #4	Impact	variables	 14	 8	 14	 14	Shape	 Sheet	 Sheet	 Sheet	 Sheet	Lattice	 Rectangle	 Rectangle	 Rectangle	 Rectangle	Dimensions	 4	x	4	 4	x	4	 3	x	3	 4	x	4	Number	of	reference	vectors	 16	 16	 9	 16	Learning	rate	(coarse)	 0.5	 0.5	 0.5	 0.5	Learning	rate	(fine)	 0.05	 0.05	 0.05	 0.05	Neighborhood	radius	(coarse)	 3	to	1	 3	to	1	 3	to	1	 3	to	1	Neighborhood	radius	(fine)	 1	 1	 1	 1	Training	cycles	(coarse)	 1000	 1000	 1000	 5000	Training	cycles	(fine)	 500	 500	 500	 2500	Av.	quantization	error	at	beginning	of	training	(q0)	42.7715	 38.8872	 42.8131	 42.7715	Av.	quantization	error	at	end	of	training	(q)	15.3761	 14.8371	 17.1689	 15.2063	Topographic	error	at	beginning	of	training	(k0) 0.8571	 0.6488	 0.5119	 0.8571	Topographic	error	at	end	of	training	(k) 0.1042	 0.1667	 0.0506	 0.0982	 The resulting patterns for the eight impacts were not notably different but their relative locations on the array were somewhat different. In other words, the RIPs for the impact variables that vary with more scenarios does not differ notably between the 8-variable SOM and 14-variable SOM. Given that including the impact variables that have less variation across scenario does not seem to affect the resulting patterns of other impact variables, two additional 14-variable SOMs are trained (SOM #3 and #4). As shown in Table 4.22, SOM #3 is different from SOM #1 by using a 3 by 3 dimension, while SOM #4 differs by having five times more training cycles.     132 4.7.2 Evaluating the SOMs To determine which of the four trained SOMs is most appropriate for this case study, their results are compared to see which set has the lowest error (quantified by two error measures – q and k) while also providing the optimal level of variation to address the questions. The error measures for the four different SOMs are also in Table 4.22. As expected, the 3 by 3 dimension of SOM #3 has a higher quantization error value (q) because it is reducing the data to a smaller number of patterns. Going further in visualizing its resulting RIPs, it is found that the 3 by 3 dimension is not a good fit for these data. This observation is made based on the principle that RIPs across the 2D array represent a continuum of modes, such that RIPs that are more similar should be closer to each other in the array and the most different ones should be on opposite ends of the array. Looking at the range of scenarios represented by each RIP in SOM #3, RIPs representing impacts from scenarios of 6m SLR are located in between RIPs representing scenarios of 0m to 1m of SLR, indicating that this 3 by 3 structure may not be permitting the effective spread of the range of patterns in this data.   The comparison between error measures of all four SOMs shows that the larger number of training cycles in SOM #4 do reduce error values but only by a little. Nonetheless, SOM #4 has the lowest error values and its results have a reasonably continuous distribution of patterns across the array where RIPs representing 0-1m of SLR and those of 6m SLR are located on the opposite corners of the 4 by 4 array. Therefore, the results from SOM #4 are carried forward for further analysis, which is described in the next chapter.     133 4.8 Summary In summary, the application of the RIPs method for this case study involved [Figure 4.11]:  • Developing 336 future scenarios characterized by different conditions of inundation, population and land-use, power outages, and building vulnerability • Geospatially modeled 14 SLR impacts for each of the 336 future scenarios, which produced a total of 4,704 impact maps • Identified 16 RIPs for each of the impacts using the SOMs method  Figure 4.11 A schematic diagram showing the three stages of the RIPs method implemented for the case study at the City of Vancouver  134 Chapter 5: Application of the Robust Impacts Patterns Method at the City of Vancouver -Results 5.1 Introduction As a result of the analyses described in Chapter 4, a total of 16 RIPs were identified for each of the 14 SLR impacts that were modeled for the City of Vancouver. These 16 RIPs represent impact patterns that are predominant across the 336 plausible future scenarios defined for this case study. This chapter begins by explaining how to go about interpreting the RIPs (Section 5.2). Then the RIPs are interpreted in Sections 5.3 to 5.8 in terms of the magnitude and spatial distribution of the impacts and the characteristics of future scenarios that can lead to those impact patterns. Collectively, this shows how the impacts can change under different range of future scenarios – from those that represent our current sea-levels and land-uses, to those representing extreme but plausible futures. Some concluding remarks about this case study are in Section 5.9.   5.2 Interpreting results by groups Although understanding 16 patterns is much more manageable than understanding 336 individual impact maps, it is still not practical for decision-makers to understanding 16 individual patterns for 14 different impacts. Therefore, instead of considering each of the 16  135 RIPs individually, it is more efficient to group the RIPs by how similar they are and interpret the results by groups. Before explaining how these groups are formed, it may be helpful to first see Figure 5.1, which shows the set of 16 RIPs for one of the impacts (business disruption in tertiary sectors) as an example. As mentioned in Section 3.2.3, the 16 RIPs are arranged on a grid with more similar patterns located closer together while very different patterns are at opposite ends of the grid. This manner of presenting the robust impact maps follows the visualization concept called small multiples introduced by Tufte (1990), which is a common way to present rich, multi-dimensional data without forcing too much information into a single map or chart. Figure 5.1 also shows how the 16 RIPs have been divided into three groups (A, B, and C). Each RIP is denoted by its coordinate on the 2D array. For example, the 3rd RIP from the left on the top row is denoted as RIP 1:3.   136 Figure 5.1 The 16 RIPs of tertiary sector business disruption. The shading shows the affected businesses per hectare; the total number of affected businesses and the relative robustness (RR) are labelled above each RIP.     AC B12341 2 3 4Disrupted businesses per hectare  137 Figure 5.2 The level of SLR associated with impacts represented by the RIP at the corresponding location in the 2D array as shown in Figure 5.1. The x-axis is the percentage of scenarios represented by the RIP.  The grouping of the RIPs is based on both objective information about the RIPs and the researcher’s judgment about how the RIPs can be presented to promote easier integration into adaptation planning and thought processes. Specifically, two pieces of information were used to divide the RIPs into 3 groups. One is the range of SLR associated with each RIP [Figure 5.2], which is determined by retracing which input vectors were mapped to each reference vector. Figure 5.2 shows that RIPs in Groups A, B, and C mostly represents impacts associated with SLR of high (~4-6m), medium (~2-4m), and low (~0-2m) range of SLR, respectively. Similarly, the range of storm intensity, land-use distribution, power outage, and vulnerability of buildings corresponding to each RIP can also be determined, and their charts similar to Figure 5.2 are available in Appendix C. Although the grouping can also be made based on those other AC B12341 2 3 4 138 characteristics of the scenarios, such as land-use, the grouping is based on SLR range to address the priority question of how the impacts can look like under different levels of SLR in the future. Another source of information that guided the grouping is the level of similarity between each RIP as measured by their Euclidean distance indicated by the shading between the grey blocks in Figure 5.3. The highest Euclidean distances are between RIPs in row 2 and 3, as well as those in the lower half between column 2 and 3 [Figure 5.3]. This corresponds to the delineation of Groups A, B, and C, except for RIP 2:1 that would have been allocated to Group A if the grouping was solely guided by the Euclidean distances. But for more intuitive interpretation, RIP 2:1 was allocated to Group C to fall within the lower range of SLR grouping.  Figure 5.3 An UMatrix produced by the SOM Toolbox to visualize the Euclidean distance (shading) between each RIP (grey blocks).    139 Another important piece of information shown in Figure 5.1 and Figure 5.3 is the relative robustness (RR) of each RIP, which is measured by the percentage of the 336 future scenarios that are matched to each RIP. For example, RIP 4:1 accounts for the variations in SLR impacts across 14.29% (approximately 48 out of 336) of the future scenarios defined in this case study. It is also the most robust RIP in Group C. RIP 4:4 is the most robust in Group B. Rather than having one or two highly robust RIPs, the RIPs in Group A share similar RR except for 2:2, 3:2, and 2:3. This indicates that there are more different modes of impacts when the level of SLR is in the higher range. The RIPs with low percentage (i.e., low relative robustness) can be considered as outliers - impact patterns that can occur under a small number of scenarios. The RR of each RIP will be labeled above each RIP from hereafter.   One may also notice that RIPs 3:1 and 2:4 have zero matched scenarios. This is because SOMs assumes the data are continuous and thus trains the reference vectors to span across the data space (i.e., SLR impacts across 336 scenarios), each representing a dominant mode and collectively representing a continuous spectrum of modes. Therefore, there are instances where SOMs identify a mode that does not exist in the training data but SOMs created it as a logical transition in-between two modes that does existing in the training data in order for the modes to be continuous. These RIPs will not be considered in further analysis.  To summarize characteristics of the scenarios represented by the 3 groups of RIPs, Figure 5.4 shows the relative proportion of scenarios represented by each group of RIPs. Overall, the variations between groups are most pronounced in terms of SLR, while others are more evenly spread across the three groups [Figure 5.4]. Besides representing impacts associated with the  140 higher range of SLR (4-6m SLR), Group A RIPs are also associated with the most scenarios with 1:10,000-year storm and pessimistic power outage conditions [Figure 5.4], while Group C is essentially the opposite and Group B lies in-between these groups. With respect to land-use and population distributions, all four land-use conditions are represented in the three groups, except that Group A has the least scenarios with Compact land-use, while Group C has the most.   Figure 5.4 Bar charts showing the a) storm intensity, b) SLR, c) land-use, d) power outage, and e) vulnerability of buildings associated with the RIPs of Groups A, B, and C. The percentage of all scenarios for the case study represented by each group is shown in (f).  7	(a)	 (b)	(c)	 (d)	(f)	(e)	 141  Since the difference between the three groups of RIPs are the most pronounced by the range of SLR they are associated with, the following sections will describe the impacts associated with each group (i.e., each SLR range) in terms of the impact magnitude and spatial distribution (e.g., hotspots). The top five affected neighborhoods in the most robust RIPs of each group are used to summarize the hotspots. The neighborhoods shown in Figure 5.5 are delineated by the CoV. Collectively, this provides an overview of how each impact may look like for the different ranges of SLR. Besides the level of SLR, the influences of other factors (e.g., power outage, land-use) are also highlighted wherever applicable.  Figure 5.5 Neighborhoods in the City of Vancouver   142 5.3 Business disruptions in the primary, secondary, and tertiary sectors Business disruption refers to the number of businesses in the given sector(s) that may be temporarily closed due to the flood event. Here, the business disruption is modeled separately for primary, secondary, and tertiary sectors. Primary sectors include agriculture, mining, construction, transport, communication, and utilities. Secondary sectors include only businesses in manufacturing. Tertiary sectors include wholesale, retail trade, finance, insurance, real estate, health services, and all other services.  The highest and lowest number of affected primary sector businesses are found in RIPs 1:4 and 4:1 with 404 (~4% of the primary sectors) and 60 (~1% of primary sectors) businesses affected respectively [Figure 5.6]. As shown in Table 5.1 and Figure 5.6, the spatial distribution of the affected businesses changes significantly as SLR increases. With 0-2m of SLR, the affected businesses are concentrated in Sunset, Marpole, and Downtown, but the hotspots expand to include new areas such as Grandview-Woodlands and Strathcona with high levels of SLR. A similar change in magnitude and spatial distribution is also observed for businesses in the secondary sectors [Table 5.1]. Given that most manufacturing businesses are located along the Fraser River shoreline area and northeast edge of the City, even when the power outage and inundation areas expands with high levels of SLR, the areas with affected businesses do not change as notably as the magnitude [Figure 5.7].   Overall, the positive trend between the number of affected businesses and level of SLR is also observed in tertiary sector businesses. However, the number of affected tertiary sector businesses has the widest range (approx. 500 to 4500) in comparison with the change in primary and secondary sector business [Table 5.1]. 143 Figure 5.6 The 16 RIPs of primary sector business disruption. The shading shows the number of affected businesses per hectare, while the total number of affected businesses and relative robustness (RR) of each RIP are labeled above it.   AC B12341 2 3 4Disrupted businesses per hectare  144 Figure 5.7 The 16 RIPs of secondary sector business disruption. The shading shows the number of affected businesses per hectare, while the total number of affected businesses and relative robustness of each RIP are labeled above it.  AC B12341 2 3 4Disrupted businesses per hectare  145 Table 5.1 Business disruptions for three major types of sectors summarized in terms of the 3 groups of RIPs (see notes below table)  Notes: 1. The weighted mean refers to the total number of affected businesses averaged across the RIPs of the group weighted by the percentage of scenarios each RIP represent (i.e., relative robustness),  2. The range refers to the highest and lowest total number of affected businesses in the RIPs of the group 3. The top five hotspots of the impact in each group are denoted by the list of neighbourhoods. Hotspots for each group is based on the hotspots in the RIP that represents the highest number of scenarios, which is 1:3 for Group A, 4:4 for Group B, and 4:1 for Group C. The adjacent percentage refers to the amount of the total affected businesses that are located in that neighbourhood.  4. 404 and 60 primary sector business is approximately 4% and 1% of the City’s primary sector respectively according to 2011-2012 business licensing data 5. 299 and 74 secondary sector business is approximately 28% and 7% of the City’s secondary sector respectively according to 2011-2012 business licensing data 6. 4516 and 511 tertiary sector business is approximately 12% and 1% of the City’s tertiary sector respectively according to 2011-2012 business licensing data  	 A			(4-6m	SLR)	B		(2-4m	SLR)	C	(0-2m	SLR)	Business	disruption		(Primary	sectors)	 	Weighted	mean:	375	Range:	283	–	404	4		Downtown	 					 	22%	Strathcona		 						 14%	Marpole		 						 13%	Grandview	 						 12%	West	end		 					 	7%	Weighted	mean:	215	Range:	179	–	269			Marpole	 	 24%	Downtown	 	 18%	Strathcona	 	 12%	Sunset		 	 11%	Kerrisdale		 							 10%	Weighted	mean:	93	Range:	60	–	173	4		Sunset		 	 29%	Marpole		 	 22%	Downtown		 	 21%	Kerrisdale		 	 6%	Strathcona		 	 5%	Business	disruption		(Secondary	sectors)	Weighted	mean:	278	Range:	228	–	299	5		Strathcona	 						 35%	Grandview		 						 15%	Marpole		 						 13%	Sunset		 	 11%	Mount	pleasant					 10%	Weighted	mean:	168	Range:	129	–	213			Strathcona		 							 27%	Marpole		 							 22%	Sunset		 							 20%	Grandview	 						 	7%	Downtown		 						 	6%	Weighted	mean:	96	Range:	74	–	150	5		Sunset		 	 38%	Marpole		 	 29%	Downtown		 	 8%	Strathcona		 	 6%	Kerrisdale		 	 5%	Business	disruption		(Tertiary	sectors)	Weighted	mean:	4137	Range:	3114	–	4516	6		Downtown	 	 45%	Strathcona		 	 9%	West	end		 	 8%	Grandview	 	 8%	Marpole		 	 6%	Weighted	mean:	2124	Range:	1680	-	2844			Downtown	 	 38%	Marpole	 	 13%	Strathcona	 	 9%	Kerrisdale	 	 8%	Oakridge	 	 6%	Weighted	mean:	891	Range:	511	–	1809	6		Downtown	 	 33%	Fairview	 	 13%	Marpole		 	 12%	Sunset		 	 12%	Kitsilano	 	 7%	 146 Given that the Downtown area has the highest concentration of tertiary sector businesses and that it is affected significantly by both inundation and power outage in many scenarios, it is not surprising to find that the highest number of affected businesses are in Downtown in all three groups of RIPs [Figure 5.8]. However, the other hotspots change notably with the different range of SLR. In the lower range, the impact concentrates around Downtown, False Creek, and Marpole. With higher levels of SLR, the effect of the power outage in the Southern areas and east of Downtown is more pronounced where the impact hotspots shifts northwards and eastwards to include Oakridge and Strathcona. But with further SLR increses, more disruptions are concentrated in the northern side of the city - Downtown, Strathcona, Grandview, and West End [Table 5.1]. Comparing the disruptions to these three major groups of businesses, the secondary sectors appear to have the highest portion of their businesses potentially affected, ranging from 7% when the level of SLR is 0-1m up to 28% when the level of SLR is 5-6m [Table 5.1].  Although the affected areas are expected to expand with larger areas of inundation associated with higher levels of SLR and storm intensity, spatial extent of impacts does not necessarily change linearly with SLR. Instead, it can be strongly influenced by cascading effects. For example, Figure 5.9 shows that the horizontal flood extent does not change as dramatically as the change in locations of affected businesses when sea-level increases from 0-2m to 4-6m [Figure 5.8]. In addition to the increased level of SLR, Figure 5.9 also shows that the change in the horizontal extent of power outage as SLR increases, influences the change in areas of affected businesses more strongly.     147 Figure 5.8 The 16 RIPs of tertiary sectors business disruptions. The shading shows the number of affected businesses per hectare, while the total number of affected businesses and relative robustness of each RIP are labeled above it. (Note: this is a larger version of Figure 5.1) AC B12341 2 3 4Disrupted businesses per hectare  148 Figure 5.9 The inundation conditions (left column) and power outage conditions (right column) for 3 different scenarios - 1m SLR with 1:50-year storm (top), 2m SLR with 1:500-year storm (middle), and 6m SLR with 1:10,000-year storm (bottom).       149 Besides inundation and power outage conditions, business disruption is also influenced by conditions of land-use and vulnerability of buildings to flood damage. However, both of these factors do not vary dramatically between the three groups of RIPs. Therefore, to see how they have contributed to the resulting RIPs, we look at the variations within groups. For example, as shown in Figure 5.8, the RIPs of tertiary sector business disruption in Group A shows that RIPs 2:2 and 3:2 have a notably lower number of affected businesses (~30%) than other RIPs in Group A, even though RIPs 2:2 and 3:2 are associated with mostly 5m and 6m of SLR respectively. To understand why these two patterns have a lower level of business disruption, we can consider how land-use and building vulnerability influence those RIPs.   Looking at the land-use conditions associated with RIPs in Group A [Figure 5.10], RIPs 2:2 and 3:2 are associated with only Compact land-use conditions. But when we focus on RIPs of Group C [Figure 5.11], the opposite is found; where the RIP associated with Compact land-use conditions have higher number of affected businesses than other RIPs in the group. This suggests that the way the new buildings are allocated in the Compact land-use condition may increase the number of affected businesses when the level of SLR is within the lower range, but may have the opposite effect when the level of SLR is in the higher range.   This effect is made more clear when we see that the Compact land-use conditions allocates new buildings (residential and commercial) in much more concentrated pockets rather than being more evenly spread across the city, as in the Status quo or Sprawl land-use conditions [Figure 5.12]. Specifically, almost half of those concentrated pockets in the Compact land-use condition are near the shoreline, while the rest are outside the floodplains and power outage areas even when the level of SLR is 6m. Therefore, in the Compact land-use conditions, the impact of planning those new buildings in the  150 shoreline pockets is quickly observable when the sea-level begins to rise. However, as the sea-level continues to rise, the impact is not much as in other land-use conditions where new buildings are more evenly distributed such that continued SLR will continue to increase affected businesses. This demonstrates the potential influence of land-use planning on the magnitude and spatial distribution of SLR impacts.  Although both RIPs 2:2 and 3:2 are associated with scenarios with Compact land-use conditions, the total number of affected businesses in 2:2 is higher than in 3:2 [Figure 5.10]. Looking further at the building vulnerability (as represented by different sets of stage-damage functions, SDFs) used to model the impacts represented by those two RIPs [Figure 5.10], we find that 2:2 are only associated with MCM SDFs while RIP 3:2 is modeled with only Hazus SDFs. Given that MCM SDFs tend to reach maximum damage at 7ft of inundation, while HAZUS SDFs only hits maximum damage at 10ft of inundation, this may explain why the impacts in RIP 2:2 are notably higher than in RIP 3:2 especially when the level of SLR is in the higher range. Although it is not possible to tell which set of SDFs can better represent the vulnerability of buildings in the CoV, this shows the amount of influence the SDFs have on the modeled impacts even when it is not the sole input variable.     151 Figure 5.10 The RIPs of tertiary sector business disruption of Group A (top), their associated land-use conditions (middle), and SDFs (bottoms).  AC B12341 2 3 4Disrupted businesses per hectare AC B12341 2 3 4AC B12341 2 3 4 152 Figure 5.11 The RIPs of tertiary sector business disruption of Group C (top-left), their associated SDFs (top-right), and land-use conditions (bottom) AC B12341 2 3 4AC B12341 2 3 4AC B12341 2 3 4 153 Figure 5.12 The distribution of new residential buildings in the Compact (top-left), Status Quo (top-right), and Sprawl (bottom) land-use conditions 154 5.4 Direct building damage – residential, commercial, and governmental buildings The direct building damage was modeled separately for three types of buildings – 1) residential buildings, 2) commercial and industrial buildings, and 3) governmental and public buildings. The damage refers to the cost of replacing the damaged structural components of the buildings as well as the loss of contents. As shown in Table 5.2, the cost of direct damage to residential buildings and commercial buildings are highly comparable. With 4-6m of SLR, the replacement cost ranges from CAD$1.4 to 3.6 billion for residential buildings, and CAD$1.5 to 3.6 billion for commercial and industrial buildings [Table 5.2]. On the other end, with 0-2m of SLR, the replacement cost ranges from CAD$0.1 to 0.8 billion for residential buildings, and CAD$0.2 to 0.9 billion for commercial and industrial buildings [Table 5.2]. Since both residential and commercial buildings are widespread across the city, the spatial extent of the damage is also similar and tends to follow the inundation pattern [Figure 5.13 and Figure 5.14]. The most notable difference between the damage areas of residential and commercial buildings are at the area near the shoreline of the Fraser River where the area immediately by the shoreline is mostly dominated by industrial and commercial buildings. Downtown has the highest concentration of building damage for both residential and commercial buildings across all three range of SLR [Table 5.2]. However, other hotspots of residential building damage concentrate around False Creek area (e.g., Fairview and Mount Pleasant) while the damage is mostly focused in the South side of the city for commercial buildings due to the higher density of industrial buildings at the shoreline of the Fraser River.     155 Table 5.2 Direct building damage for three types of buildings summarized in terms of the 3 groups of RIPs (see notes below table). Notes: 1. The weighted mean refers to the total cost of building damage averaged across the RIPs of the group weighted by the percentage of scenarios each RIP represents (i.e., relative robustness),  2. The range refers to the highest and lowest total cost of building damage in the RIPs of the group 3. The top five hotspots of the impact in each group are denoted by the list of neighbourhoods. Hotspots for each group is based on the hotspots in the RIP that represents the highest number of scenarios, which is 1:3 for Group A, 4:4 for Group B, and 4:1 for Group C. The adjacent percentage refers to the amount of the total affected businesses that are located in that neighbourhood.  	 A			(4-6m	SLR)	B		(2-4m	SLR)	C	(0-2m	SLR)	Residential	building	damage	Weighted	mean:		2.7	billion	$	Range:	1.4	–	3.6	billion	$		Downtown						 	 56%	Fairview					 	 11%	Mount	pleasant								 8%	Kitsilano						 	 7%	Killarney	 	 4%	0	Weighted	mean:		1.2	billion	$	Range:	0.6	–	2%	billion	$		Downtown						 	 49%	Mount	pleasant							 13%	Fairview						 	 9%	Dunbar-Southlands	 8%	Kitsilano	 	 7%	Weighted	mean:		0.3	billion	$	Range:	0.1	–	0.8	billion	$		Downtown						 	 34%	Dunbar-Southlands	 23%	Killarney						 	 10%	Mount	pleasant							 10%	Kerrisdale						 	 9%	Commercial	building	damage		Weighted	mean:		2.8	billion	$	Range:	1.5	–	3.6	billion	$		Downtown						 	 45%	Marpole					 	 13%	Strathcona						 	 11%	Fairview					 	 10%	Kitsilano						 	 5%	Weighted	mean:		1.4	billion	$	Range:	0.7	–	2.2	billion	$		Downtown						 	 42%	Marpole					 	 16%	Strathcona						 	 11%	Fairview						 	 9%	Sunset							 	 6%	Weighted	mean:		0.4	billion	$	Range:	0.2	–	0.9	billion	$		Downtown						 	 24%	Marpole					 	 24%	Dunbar-Southlands					11%	Kitsilano						 	 9%	Sunset							 	 9%	Governmental	building	damage	Weighted	mean:		126	million	$	Range:	63	–	170	million	$		Downtown						 	 43%	Strathcona						 	 37%	Dunbar-Southlands					9%	Kitsilano						 	 3%	Fairview						 	 3%	Weighted	mean:		59	million	$	Range:	28	–	89	million	$		Downtown						 	 43%	Strathcona						 	 32%	Dunbar-Southlands					18%	Kitsilano						 	 4%	Marpole						 	 2%	Weighted	mean:		18	million	$	Range:	8.7	–	41	million	$		Dunbar-Southlands					47%	Downtown						 	 33%	Kitsilano						 	 8%	Strathcona								 7%	Hastings-Sunrise								 3%	 156 Figure 5.13 The 16 RIPs of residential buildings direct damage cost. The shading shows the cost of damage per hectare, while the total cost of damage and relative robustness of each RIP are labeled above it.  AC B12341 2 3 4Building damage cost [CAD$] per hectare  157 Figure 5.14 The 16 RIPs of commercial and industrial buildings direct damage cost. The shading shows the cost per hectare, while the total cost of damage and relative robustness of each RIP are labeled above it. AC B12341 2 3 4Building damage cost [CAD$] per hectare  158 Public and governmental buildings include buildings for religious purposes (e.g., churches, temples), government-owned buildings (e.g., non-market housing, community centers), and public educational buildings (e.g., public schools). The replacement cost of governmental buildings associated with 0-2m and 4-6m of SLR, ranges from CAD$8.7 to 41 million and CAD$63 to 170 million respectively [Table 5.2]. The estimated replacement cost for governmental building damage is significantly lower than those for residential and commercial building largely due to their lower values but the number of affected governmental buildings is not necessarily lower. This is partly demonstrated by the RIPs of the number of potentially affected social service facilities that are mostly owned by the government, which will be discussed Section 5.6. As for other building types, the highest concentration of damage for governmental building is also the Downtown area [Figure 5.15]. However, as the level of SLR increases, the concentration shifts more towards the east side of Downtown, into the Strathcona area where there is high concentration of City-owned properties, such as non-market housing and shelters [Figure 5.15].  Besides the inundation condition, building vulnerability and land-use conditions also influence direct building damage. The effect of Compact land-use conditions observed in business disruption is also observable here where higher damage is observed in RIPs associated with Compact land-use, but reverses with higher levels of SLR. The same effect of different SDFs is also shown here where impacts in RIPs associated with MCM SDFs tend to estimate higher damage, especially in the higher range of SLR. However, these effects should not be applied to the governmental building damage impact because the number of governmental buildings is defined to not change with different land-use conditions. Therefore this effect on governmental building damage is only an artifact of SOMs and our study design. As many impact variables (eight out of 14) are influenced by land-use, power outage and inundation conditions, SOMs identified different patterns (i.e., RIPs) to represent the variations that are  159 predominantly driven by the different levels of SLR and land-use conditions. But because all elements in a reference vector (i.e., all impacts in a given scenario) are modified during the SOM training process, even the impacts that are not influenced by the land-use conditions are modified with those that are influenced by land-use conditions. Therefore the effect of Compact land-use on governmental building damage should be considered as an artifact. When considering the RIPs of governmental building damage, one should only consider those that are robust across all land-use conditions. For example, with the lower range of SLR (i.e., Group A), RIPs 4:1 and 4:2 should be considered while RIP 2:1 that represents impact of only scenarios with Compact land-use should be omitted. This is a good example to demonstrate a key limitation of this method and the importance of using the RIPs method to assess impacts that are influenced by similar sources of uncertainty. This was also described in Section 4.3.1.1 as a principle to help guide the selection of impacts to include in the assessment and the factors to use to characterize each future scenario.  160 Figure 5.15 The 16 RIPs of governmental and public buildings direct damage cost. The shading shows the cost per hectare, while the total cost in each RIP is labeled above it. AC B12341 2 3 4Building damage cost [CAD$] per hectare  161 5.5 Vulnerable population This impact refers to the number of vulnerable populations that are potentially affected by inundation and/or prolonged power outage. This population group includes persons of age 65 or above, children of age 16 or below, and those who have an annual income of CAD$10,000 or less. Even with the lower range of SLR, the potentially affected vulnerable population ranges from 4,004 to 21,411 [Table 5.3]. The highest estimate is 56,592 (approximately 7% of projected population in 2041) with 6m of SLR. While the impact is concentrated in the south side of the city with lower to mid-range SLR, the concentration shifts notably to the north side, especially the Downtown and West End with SLRs of 4m or more [Figure 5.16]. This is attributable to the high elderly population density in West End [Figure 5.17] and 4m of SLR is the threshold at which the Murrin substation is affected and stops providing power to its service areas that include the West End.    Table 5.3 Vulnerable population potentially affected by inundation and/or prolonged power outage summarized in terms of the 3 groups of RIPs (see notes below table) Notes: 1. The weighted mean refers to the total affected vulnerable population averaged across the RIPs of the group weighted by the percentage of scenarios they represent,  2. The range refers to the highest and lowest total affected vulnerable population in the RIPs of the group 3. The top five hotspots of the impact in each group are denoted by the list of neighbourhoods. Hotspots for each group is based on the hotspots in the RIP that represents the highest number of scenarios, which is 1:3 for Group A, 4:4 for Group B, and 4:1 for Group C. The adjacent percentage refers to the amount of the total affected businesses that are located in that neighbourhood. 	 A			4-6m	SLR)	B		2-4m	SLR)	C	0-2m	SLR)	Vulnerable		Population	Weighted	mean:	52620	Range:	15996	–	56592		West	end							 22%	Downtown					 20%	Grandview							12%	Marpole					 10%	Kerrisdale						 9%	Weighted	mean:	24549	Range:	21459	-	26903		Marpole					 	 22%	Kerrisdale					 	 19%	Downtown						 	 17%	Oakridge						 	 12%	Dunbar-Southlands						5%	Weighted	mean:	16809	Range:	4004	–	21411		Downtown						 	 25%	Dunbar-Southlands					19%	Kerrisdale					 	 15%	Killarney						 	 11%	West	Point	Grey							 6%	 162 Figure 5.16 The 16 RIPs of affected vulnerable population. The shading shows the affected population per hectare, while the total affected population in each RIP is labeled above it. AC B12341 2 3 421411	Affected vulnerable population per hectare  163 Figure 5.17 Population density of persons age 65 and above in the City of Vancouver based on 2011 Canadian Census data  Given that this impact accounts for vulnerable population potentially affected by inundation and/or power outage, the areas with affected vulnerable population correspond strongly with the areas of power outage in different scenarios and the concentration of elderly population. This is attributable to how the areas with power outage overlap more with the areas of elderly population density than to the areas of high concentration of population that lives under the poverty line. Since population distribution is defined differently for each land-use conditions, the effect of Compact land-use observed related to business disruption and building damage is also applicable here. However, building vulnerability does not influence it. Therefore the anomaly of much lower estimated affected vulnerable population in RIP 3:2 should be considered as an artifact, as for the case with governmental building direct damage described in the previous section. 164 5.6 Disaster response facilities – schools, emergency services, health care facilities, transportation points and social service facilities This sub-section describes the RIPs of several social impacts of SLR, which focuses on the effect of inundation and/or power outage on facilities in the city that play a critical role in the City’s capacity to respond to a disaster. These facilities include: • Schools • Emergency services (ambulance, fire halls, police stations) • Health care facilities (hospitals, walk-in clinics, and pharmacies) • Transportation points (Bus stops, Skytrain stations, gas stations, and sea bus or ferry terminals) • Social service facilities (Community centres, senior centers, child care centers or pre-schools, homeless shelters, free-meal locations, and non-market housing)  The impacts on these different types of facilities and services are modeled separately as 5 different social impacts. Since none of these facilities are defined differently across land-use conditions, the range of impact magnitude summarized in Table 5.4 and Table 5.5 does not account for the impact magnitude of RIPs that are only associated with Compact land-use (i.e., RIPs 2:2, 3:2 (Group A), 3:3 (Group B) and 2:1 (Group C), and only include RIPs that are robust across all land-use conditions. The RIPs of the impact of inundation and power outage on these 5 types of facilities are shown in Figure 5.18 to Figure 5.21.  165 Figure 5.18 The 16 RIPs of affected schools. The shading shows the number of affected schools per hectare, while the total number of affected schools and the relative robustness of each RIP are labeled above it.  AC B12341 2 3 4Affected	schools	per	hectare	 166 Figure 5.19 The 16 RIPs of the affected health care facilities. The shading shows the number of affected health care facilities per hectare, while the total number of affected facilities and the relative robustness of each RIP are labeled above it.  AC B12341 2 3 4Affected	health	care	facilities	per	hectare	 167 Figure 5.20 The 16 RIPs of affected emergency service facilities. The shading shows the number of affected facilities per hectare, while the total number of affected facilities and the relative robustness of each RIP are labeled above it. AC B12341 2 3 4Affected	emergency	service	facilities	per	hectare	 168 Table 5.4 Potentially affected schools, health care facilities and emergency services summarized in terms of the 3 RIPs groups (see notes below table) Notes: 1. The weighted mean refers to the total affected facilities averaged across the RIPs of the group weighted by the percentage of scenarios they represent. Since these facilities do not change with different land-use conditions, the mean and range are calculated excluding RIPs that are only associated with Compact land-use (i.e., RIPs 2:2, 3:2 (group a), 3:3 (group b) and 2:1 (group c)  2. The range refers to the highest and lowest total affected facilities in the RIPs of the group, excluding those that are only associated with Compact land-use. 3. The top five hotspots of the impact in each group are denoted by the list of neighbourhoods. Hotspots for each group is based on the hotspots in the RIP that represents the highest number of scenarios, which is 1:3 for Group A, 4:4 for Group B, and 4:1 for Group C. The adjacent percentage refers to the amount of the total affected businesses that are located in that neighbourhood. 4. 28 schools is about 24% of all schools (elementary and secondary) in the CoV  5. 25 health care facilities is about 28% of health care facilities in the CoV 6. About 23% of emergency service facilities in the CoV 	 A			(4-6m	SLR)	B		(2-4m	SLR)	C	(0-2m	SLR)	Schools	 Weighted	mean:	24	Range:	25	–	28	4		Grandview	 	 25%	Kerrisdale					 	 11%	Marpole					 	 11%	Oakridge						 	 11%	West	End						 	 11%	Weighted	mean:	13	Range:	12	–	15		Kerrisdale					 	 20%	Marpole					 	 20%	Oakridge						 	 20%	Shaughnessy					 14%	Downtown								 7%	Weighted	mean:	1	Range:	1	–	3		Downtown					 		 99%		Health	care	facilities	Weighted	mean:	21	Range:	21	–	25	5		Downtown		 	 44%	West	End		 	 24%	Sunset		 	 8%	Kerrisdale		 	 8%	Grandview		 	 8%	Weighted	mean:	8	Range:	7	–	11		Downtown		 	 42%	Sunset		 	 23%	Kerrisdale		 	 23%	Mount	Pleasant		 12%	Weighted	mean:	2	Range:	2	–	5		Sunset		 	 86%	Mount	Pleasant		 7%	Downtown		 	 6%	Emergency	Services	Weighted	mean:	20	Range:	23	6		Downtown						 	 40%	Strathcona						 	 17%	West	End							 	 13%	Grandview	 	 13%	Marpole						 	 9%	Weighted	mean:	8	Range:	7	–	9		Downtown						 	 33%	Strathcona						 	 33%	Marpole							 	 22%	Kerrisdale		 	 11%	Mount	Pleasant						 1%	Weighted	mean:	1	Range:	1	–	4		Strathcona		 	 82%	Downtown		 	 17%	 169 As shown in the RIPs for health care facilities, schools, and emergency services, these facilities are sparsely distributed across the city. Therefore the listing of hotspots in Table 5.4 is somewhat less relevant than for those that are more widely distributed (i.e., social service facilities and transportation points) in Table 5.5. Nonetheless, they provide some guidance to where the affected facilities are located in the City. Their relatively smaller numbers also allow us to retrace and briefly describe them in more details. For example, most of the affected health care facilities are walk-in clinics and pharmacies, and only one hospital is affected when SLR is 4m or more. Amongst the affected emergency services, many of which are police services (approximately 4 with 0-2m of SLR and 15 with 4-6m SLR), and as many as three are fire halls and ambulance services.  Overall, for all five types of facilities, the affected locations shifts in the same manner as the change in power outage areas with the increase in sea-level – starting with areas near the shoreline with low levels of SLR, then the Oakridge-Marpole area with mid-range SLR and with additional outage areas in the West End and Strathcona-Grandview. Variations in affected areas between the different types of facilities are driven by their different distribution across the city. For example, about 40% of the City’s social service facilities are located in the neighborhoods of Downtown, Strathcona, and Grandview-Woodlands, which are located in the north side of the CoV. Therefore, when the level of SLR reaches the threshold where the power outage extent expands into those neighborhoods, the highest concentration of affected facilities quickly shifts from the south side of the City to those areas in the north side [Table 5.5]. Although the majority of affected transportation points are bus stops that do not rely on electricity, many of the affected bus stops are along bus routes of trolley busses, which do rely on grid power (e.g., route 17 along Oak Street). Even with 1m of SLR, the hotspots of affected transportation points [Table 5.5] also include a significant number of affected gas stations  170 (approximately 16 just in the Oakridge-Marpole area) and five out of 22 Skytrain stations across the City. Another hotspot is by the Waterfront Skytrain station in Downtown that serves as a terminal for several modes of transportation used by those commuting between North Vancouver and the CoV.  Table 5.5 Potentially affected transportation points and social service facilities summarized in terms of the 3 groups of RIPs (see notes below table). Notes: 1. The weighted mean refers to the total affected facilities averaged across the RIPs of the group weighted by the percentage of scenarios they represent. Since these facilities do not change with different land-use conditions, the mean and range are calculated excluding RIPs that are only associated with Compact land-use (i.e., RIPs 2:2, 3:2 (group a), 3:3 (group b) and 2:1 (group c)  2. The range refers to the highest and lowest total affected facilities in the RIPs of the group, excluding those that are only associated with Compact land-use.  3. The top five hotspots of the impact in each group are denoted by the list of neighbourhoods. Hotspots for each group is based on the hotspots in the RIP that represents the highest number of scenarios, which is 1:3 for Group A, 4:4 for Group B, and 4:1 for Group C. The adjacent percentage refers to the amount of the total affected businesses that are located in that neighbourhood. 4. 412 social service facilities is about 50% of the social service facilities in the CoV 5. 617 transportation points is about 30% of transportation points in the CoV  	 A			(4-6m	SLR)	B		(2-4m	SLR)	C	(0-2m	SLR)	Social	service	facilities	Weighted	mean:	345	Range:	104	–	412	4		Downtown						 	 29%	Grandview		 	 18%	Strathcona						 	 18%	West	end								 	 9%	Marpole						 	 8%	Weighted	mean:	140	Range:	113-181		Downtown						 	 39%	Marpole					 	 20%	Fairview						 	 9%	Mount	Pleasant								 7%	Oakridge							 	 6%	Weighted	mean:	27	Range:	27	–	58		Downtown						 	 26%	Fairview					 	 23%	West	Point	Grey						 21%	Mount	Pleasant							 12%	Marpole						 	 8%	Transportation	points	Weighted	mean:	516	Range:	596	–	617	5		Downtown	 	 18%	Marpole	 	 13%	Strathcona	 	 11%	Kerrisdale	 	 9%	Oakridge	 	 8%	Weighted	mean:	366	Range:	344	–	437		Marpole	 	 20%	Downtown	 	 15%	Kerrisdale	 	 14%	Oakridge	 	 13%	Sunset		 	 7%	Weighted	mean:	70	Range:	70	–	147		Downtown	 	 27%	Marpole	 	 18%	W	Point	Grey	 	 16%	Sunset		 	 16%	West	End	 	 6%	 171 Figure 5.21 The 16 RIPs of affected social service facilities. The shading shows the number of affected facilities per hectare, while the total number of affected facilities and the relative robustness of each RIP are labeled above it.   AC B12341 2 3 4Affected	social	service	facilities	per	hectare	 172 Figure 5.22 The 16 RIPs of the affected transportation points. The shading shows the number of affected points per hectare, while the total number of affected points and the relative robustness of each RIP are labeled above it.   AC B12341 2 3 4	Affected	transportation	points	per	hectare	 173 5.7 Debris This impact refers to the weight of debris generated by the flood, measured in tons. This only accounts for debris generated from building damage (i.e., building finishes and structural components), and does not include building contents or additional debris loads from natural sources (e.g., vegetation).   The amount of debris increases rather linearly with the level of SLR, starting with about 22,000 tons to 110,000 tons with 0-2m of SLR, up to 424,224 – 775,444 tons with 4-6m of SLR [Table 5.6]. Since Hazus estimates the flood debris based on the debris generated from three specific components of the building structure – 1) building finishes, 2) structural components, and 3) foundation materials, it does not use stage-damage functions (SDFs) that were used for direct building damage, which estimates the overall building damages in terms of structure and content rather than specific structural components. Nonetheless, the hotspots of debris are similar to those of direct damage to residential buildings, which includes Downtown, Fairview, Mount Pleasant, and Dunbar-Southlands.   This impact varies with different inundation and land-use conditions. Given that the horizontal extent of inundation does not change dramatically between 1 and 6m of SLR, the spatial distribution of this impact also does not change much with different range of SLR [Figure 5.23 and Table 5.6]. However, the effect of Compact land-use observed in other impact is also applicable here, where the higher impact is represented by RIPs associated with Compact land-use conditions when the range of SLR is of the lower end, and reverse with upper range of SLR.     174 Table 5.6 Weight of flood debris from building damage summarized in terms of the 3 groups of RIPs (see notes below table). Notes: 6. The weighted mean refers to the total weight of debris averaged across the RIPs of the group weighted by the percentage of scenarios they represent,  7. The range refers to the highest and lowest total weight of debris in the RIPs of the group 8. The top five hotspots of the impact in each group are denoted by the list of neighbourhoods. Hotspots for each group is based on the hotspots in the RIP that represents the highest number of scenarios, which is 1:3 for Group A, 4:4 for Group B, and 4:1 for Group C. The adjacent percentage refers to the amount of the total affected businesses that are located in that neighbourhood.	 A			4-6m	SLR)	B		2-4m	SLR)	C	0-2m	SLR)	Debris	 Weighted	mean:	495,606	tons	Range:	424,224	–	775,444	tons		Downtown		 	 53%	Fairview		 	 13%	Mount	Pleasant		 9%	Dunbar-Southlands		 6%	Kitsilano		 	 6%	Weighted	mean:	112,624	tons	Range:	56835	–	237,453	tons		Downtown		 	 43%	Dunbar-Southlands		 13%	Mount	Pleasant		 12%	Fairview		 	 10%	Kitsilano		 	 8%	Weighted	mean:	49,498	tons	Range:	22942	–	111,676	tons		Downtown		 	 49%	Mount	Pleasant		 17%	Dunbar-Southlands		 10%	Kitsilano		 	 10%	Fairview		 	 5%	 175 Figure 5.23 The 16 RIPs of debris generated. The shading shows the tons per hectare, while the total weight and the relative robustness of each RIP are labeled above it.  AC B12341 2 3 4Debris [Tons] per hectare  176 5.8 Sewage backup damage potential Sewage backup (also called sewer backup) refers to the inflow of untreated wastewater from toilets and drains in the lowest floor. While sewage backup can be caused by a number of different ways, such as blockage due to tree roots, the sewage backup damage assessed here refers specifically to those that are due to overland flooding where sewage backup is caused by sewer system surcharging. To assess this impact, the Sewage Backup Damage Potential Index (SBDPI) developed by the researcher is used. Details about the development of this index can be found in Sections 4.6.4.2 and 4.6.4.3. This index is not a measure of the probability or likelihood of sewage backup happening within a block. Instead, it is measuring the relative amount of population and asset inside a city block (i.e., DB) that have the potential to be affected by sewage backup in the event of overland flooding, and it is relative to the rest of the city. Therefore, unlike other impacts assessed in this case study; such as the number of business closures, it is not measuring the absolute risk of sewage backup. Furthermore, it only accounts for sewage backup in ground-related homes (e.g., single detached homes, duplexes, townhomes).   In comparison with the other impacts, this impact has rather different spatial distribution because it is a result of a complex combination of multiple factors besides flood depth, land-use, and power outage. Specifically, the SBDPI depends on the following variables at a given DB:  • Maximum flood depth in the DB (m) • Distance from nearest flooded area (Equal to zero if the DB is flooded) • Power outage (Yes/No) • Amount of ground-related homes in DB (Percent of the City’s ground-related homes located in this DB)  177 • Vulnerability score of each ground-related home in the DB, which depends on 1) whether a sewage pump is installed, and the type of sewer system the home to which is connected (combined or separate sewer system) • Amount of old homes in the DB (number of ground-related homes constructed before the 1970s Figure 5.24 shows the 16 RIPs of the SBDPI. Although many of the above variables are defined by the inundation conditions, land-use conditions, and power outage conditions, the other variables characterizing the sewer systems and vulnerability at the household level (e.g., sewage pump, sewer system connected) also plays a significant role in this impact. This is particularly apparent when one compares the spatial distribution of SBDPI to those of other impacts that typically concentrate in areas of inundation and/or power outage (e.g., Downtown, Marpole, Dunbar-Southlands, Fairview). The highest SBDPI is found to be in the Hastings-Sunrise neighbourhood in all three ranges of SLR [Table 5.4]. With 0-2m of SLR, the other hotspots of SBDPI are found in Kitsilano, West Point Grey, and Dunbar-Southlands. With mid-range SLR, the hotspots shift slightly with higher concentration in Kerrisdale and Marpole, while Hastings-Sunrise remains the neighbourhood with the highest SBDPI [Figure 5.24].  Table 5.7 Sewage backup damage potential index (SBDPI) summarized in terms of the 3 groups of RIPs (see notes below table) 	 A			4-6m	SLR)	B		2-4m	SLR)	C	0-2m	SLR)	Sewage	backup		damage	potential	Weighted	mean:	160	Range:	104	–	185		Hastings-sunrise								18%	Kerrisdale					 											12%	West	point	grey							 7%	Marpole						 	 7%	Dunbar-Southlands	 7%	Weighted	mean:	113	Range:	100	–	122		Hastings-sunrise							 19%	Kerrisdale					 	 11%	Dunbar-Southlands	 11%	Kitsilano					 	 11%	West	point	grey							 8%	Weighted	mean:	95	Range:	83	–	99			Hastings-sunrise							 17%	Kitsilano					 	 11%	Dunbar-Southlands	 8%	Grandview		 	 8%	West	point	grey							 7%	 178 Notes: 1. The weighted mean refers to the total damage potential index averaged across the RIPs of the group weighted by the percentage of scenarios they represent,  2. The range refers to the highest and lowest total damage potential index in the RIPs of the group 3. The top five hotspots of the impact in each group are denoted by the list of neighbourhoods. Hotspots for each group is based on the hotspots in the RIP that represents the highest number of scenarios, which is 1:3 for Group A, 4:4 for Group B, and 4:1 for Group C. The adjacent percentage refers to the amount of the total affected businesses that are located in that neighbourhood  Unlike more straightforward impacts, such as the number of health care facilities affected by inundation and/or power outage, identifying the attributing factors for high magnitudes of SBDPI is somewhat more complex. Hotspots can be contributed by different factors. For example, the high SBDPI at Hastings-Sunrise and part of Grandview-Woodlands are mostly due to two factors – 1) their high concentration of pre-70s ground-related homes [Figure 5.25], which are assumed to not have any measures implemented for inflow control and thus have the highest vulnerability score, and 2) the area is close to areas of high flood depth.   On the other hand, the high SBDPI in Kerrisdale and Marpole area when the level of SLR is more than 3m (e.g., RIP 4:4 in Figure 5.24) is largely due to areas with high density of homes with sewage pumps coinciding with the areas of prolonged power outage that extends from Marpole into Kerrisdale and Oakridge area [Figure 5.26]. The higher SBDPI in West Point Grey and Kitsilano is not attributable to power outage, but rather their proximity to the shoreline, concentration of homes constructed before the 1970s, and concentration of homes that are connected to combined sewer system and have no pumps installed [Figure 5.27]. Although it is not listed as one of the hotspots in Table 5.7, the strip of homes in southwest of Killarney also has high SBDPI. This block of the CoV has the highest density of pre-70s ground-related homes [Figure 5.25] and is within the floodplain in various scenarios.   179 Figure 5.24 The 16 RIPs of sewage backup damage potential index (SBDPI). The shading shows the damage potential per hectare, while the total damage potential in each RIP is labeled above it.  AC B12341 2 3 4SBDPI per hectare  180 Figure 5.25 Percent of ground-related homes that are constructed before the 1970s. Based on data from the 2011 National Household Survey (NHS) Profiles Files.  181 Figure 5.26 Percent of ground-related homes that are connected to combined sewer system and have sewage pump installed (top) and areas of prolonged power outage associated with inundation condition with 1:500-year storm and 3m of SLR (bottom).  182 Figure 5.27 Percent of ground-related homes connected to the combined sewer system and have no sewage pump installed.  183 Other interesting attributing factors are the land-use conditions. As shown in Figure 5.28, the RIPs associated with Compact land-use conditions (i.e., RIPs 2:1, 2:2, 3:2) have lower SBDPI than those of that are robust across different land-use conditions but associated with similar levels of SLR. Rather than due to the way new buildings are allocated in the Compact land-use condition, this is attributable to the type of new buildings in the Compact land-use condition, where most new dwelling units are defined to be inside multi-family residential buildings rather than ground-related homes (e.g., duplex, single detached). Therefore, regardless of the level of SLR, the SBDPI is lower in Compact land-use conditions. The influence of different power outage conditions is also apparent in the RIPs of SBDPI. Since the spatial distribution of power outage areas is the most different between the optimistic and pessimistic power outage conditions at 1m and 4m of SLR, this difference is sufficiently large to be reflected in the RIPs of SBDPI. For example, even though RIP 4:2 is associated with higher levels of SLR and stronger storms than RIP 4:1, the SBDPI in RIP 4:2 is lower because it is only associated with optimistic power outage conditions [Figure 5.28]. Similarly, both RIP 1:1 and 3:4 are associated with 3-4m of SLR, but RIP 1:1 has higher SBDPI as it is only associated with pessimistic power outage conditions while RIP 3:4 is mostly associated with optimistic power outage conditions.  184 Figure 5.28 The RIPs of SBDPI in Group C (top-left) and its associated storm intensity (top-right), power outage conditions (bottom-left), and SLR (bottom-right).  185 5.9 Conclusion The RIPs method was applied to the CoV to demonstrate the method as a new approach to assess the potential socio-economic impacts of SLR under a large number of future scenarios. A total of 14 types of impacts, including economic, social, and environmental impacts, were assessed across 336 future scenarios characterized by different conditions of storm intensity, level of SLR, land-use and population distribution, power outage, and vulnerability of building integrity. A total of 16 RIPs for each impact were identified showing the magnitude and spatial distribution of each pattern predominant amongst the 336 future scenarios.   To allow more efficient interpretation, the 16 RIPs for each impact were divided into three groups, representing impacts associated with low, medium, and high range of SLR. The impact patterns were described in terms impact magnitude and how the impact varies across the city. Table 5.8 summarizes the average magnitude of each assessed impact represented by the RIPs in the three groups, to show how the impacts change with different ranges of SLR. However, it is important to remember that the 16 RIPs are not all equally robust. By considering the number of scenarios that each RIP is representing, it is possible to determine their relative robustness (RR). Therefore the average impact magnitude of each group is weighted by the relative robustness of the RIPs in the group. As expected, the impacts intensify with higher levels of SLR – from Group C (0-2m SLR) to Group A (4-6m SLR). However, even with low end SLR, the associated impacts can still be quite concerning. For example, even with less than 2m of SLR, one can expect almost a billion dollars worth of direct building damage and 2% of the city’s businesses to experience temporary closure.    186 Table 5.8 Weighted average of all assessed impacts in each RIP groups Business	disruption	(Primary	sectors)	[#	businesses	closed]	375	 215	 93	Business	disruption	(Secondary	sectors)	[#	businesses	closed]	278	 168	 96	Business	disruption	(Tertiary	sectors)	[#	businesses	closed]	4137	 2124	 892	Direct	damage	cost	(Residential)	[CAD$]	 2.7	billion	 1.2	billion	 0.38	billion	Direct	damage	cost	(Commercial)	[CAD$]	2.8	billion	 1.4	billion	 0.4	billion	Direct	damage	cost	(Governmental)	[CAD$]	126	million	 59	million	 18	million	Affected	vulnerable	population	[#	persons	elderly/children/	income	<	$10,000]	52620	 24550	 16809	Affected	schools	[#	facilities]	 24	 13	 1	Affected	health	care	facilities	[#	facilities]	 21	 8	 2	Affected	social	services	facilities	[#	facilities]	 516	 366	 70	Affected	transportation	points	[#	points]	 345	 140	 27	Affected	emergency	services	[#	facilities]	 20	 8	 1	Debris	[Weight	in	tons]	 495606	 112624	 49498	Sewage	backup	damage	potential	[Unit	less	composite	index]	160	 113	 95	   Impact	 Group	A	4-6m	SLR	Group	B	2-4m	SLR	Group	C	0-2m	SLR	 187 Besides interpreting the results by the 3 groups of RIPs (i.e., in terms of the 3 ranges of SLR), some interesting insights were also drawn from the variations between RIPs within each group. For example, RIPs associated with only Compact land-use conditions are found to be notably different than others but the direction of the difference is not necessarily the same for each impact variables. The influence of Compact land-use had an amplifying effect on business disruption and building damage when the level of SLR is within the lower range of 0-2m but has the opposite effect with higher levels of SLR. This was attributable to the way Compact land-use allocates new buildings in concentrated pockets of the city such that more buildings can be affected with initial SLR but do not necessarily continue to amplify with further SLR, while more even distribution of new buildings (as in Status-Quo or Sprawl) facilitates damage increasing more consistently with SLR. This is a good example to demonstrate how non-climatic drivers can introduce nonlinearity to the impacts. On the other hand, in the Compact land-use condition, the higher proportion of new buildings being high-rises rather than ground-related homes have a reducing effect on sewage back-up damage potential. This is largely due to the fact that the SBDPI only measures the damage potential in ground-related homes. However, this effect is not necessarily detached from reality, as only the lowest floor of a high-rise building would be directly exposed to the sewage should sewage backup occur in high-rise buildings. Therefore, unless sewage backup affects access to the building, high-rise buildings are less vulnerable to sewage backup than ground-related homes when considering the number of dwelling units that would be directly affected should sewage backup occur in both types of buildings.   188 Overall, this case study clearly demonstrates that assets, population, and environment located outside the potential floodplains can also be impacted in multiple ways. This was made possible by accounting for future changes in non-climatic factors (e.g., power outage, land-use) in addition to hydrological factors (SLR, storm intensity), which facilitated the assessment of both direct and indirect impacts. Furthermore, the dynamics between the factors characterizing the scenarios and the spatial distribution of each impact subject (e.g., ground-related homes, vulnerable population groups) bring about nonlinearity, cascading effects, and uncertainties in SLR impacts that are captured in this method but not in conventional flood consequence assessments that often do not account for indirect impacts.  This case study has shown that the RIPs method can identify RIPs for impacts of different levels of complexity – from more straightforward impacts such as the number of healthcare facilities affected by power outage and inundation to sewage backup damage potential that is influenced by multiple factors that are less hazard-related. It has also demonstrated the importance of selecting impact variables that are influenced by similar major factors that are used to characterize the future scenarios. By including variables, such as affected health care facilities, the variations in RIPs created by factors that do not by design influence that specific impact variable would be considered as artifacts. Although including variables that vary with significantly different factors do not seem to change the RIPs for other impact variables, it is perhaps better to include those variables in a separate application of the RIPs method (i.e., training a different SOMs for this set of variables) to avoid confusion. In principle, the RIPs provide a new way for users to account for more uncertainties of SLR impacts in an efficient manner where the possible impacts of SLR under a broad range of futures are visualized and  189 summarized as spatially explicit predominant patterns. However, this exercise does not address the question of whether the RIPs can support SLR adaptation planning in reality. This will be addressed in the following chapter where the RIPs method and the resulting RIPs from this case study are evaluated with experts at the CoV, who are either currently involved in the City’s SLR adaptation planning or are experts of the City’s infrastructure and systems that are imminent of being affected by flooding. Broader implications of this case study will be discussed in the Conclusion chapter (Chapter 7).     190 Chapter 6: Potential Utility of the Robust Impact Patterns Method 6.1 Introduction This chapter describes the last major component of this dissertation that addresses the question - How can the Robust Impact Patterns (RIPs) method and results support SLR adaptation from the prospective users’ perspective?  The methodological approach to this research component is described in the next section. The results and concluding remarks can be found in Section 6.3 and Section 6.4 respectively.  The quality of newly developed models or tools is traditionally and widely acknowledged to be measured based on their scientific integrity. The performance aspect is typically analyzed by verification of whether the tool is working according to its intended specifications, while comparing the model output to empirical data and sensitivity analysis can assess the validity aspect. However, whether the information produced by the tool is useful for anticipated users – the value of the information – is not part of standard practice until recently (Feldman & Ingram, 2009). For example, to support emergency response planning, a new model may be developed to estimate the likelihood of hazardous material (HAZMAT) release during a tropical storm. To demonstrate the feasibility of the model, it can be applied to a certain city as a case study to assess the risk of HAZMAT release in different scenarios of storms with various intensities. The performance of the model could be verified by comparing its results to observed data of past  191 cases of HAZMAT release that were caused by a storm. However, this model is unlikely to be examined for the value of the information it produces – do the users find the model output useful for its intended purpose of supporting emergency response planning?   Traditionally, the provision of tools and information to support climate change decision-making followed the “loading-dock” model where scientists produce models or information without consulting with prospective users to understand their needs but expect the users to find the information useful (Cash, Borck, & Patt, 2006).  Schmolke and colleagues (2010) highlighted three key barriers that contribute to the ‘loading-dock' model. The lack of involvement of users in model development or evaluation could be due to the lack of standardized approach for users to contribute their input and users or decision-makers usually do not have the appropriate training or time to figure it out. The mismatch of the researcher's research agenda and the scope of the policy issue at hand provide little incentive for the researcher to go beyond the standard practice. Lastly, the inconsistent terminology related to the quality of models or products, such as verification, validation, and uncertainty, also serves as a barrier to effectively involve users in the model development and evaluation process. Evidence shows that this model of scientific information production cannot support decisions (Cash et al., 2006) and the need to bridge the gap between science and practice has grown significantly in the recent years, especially in the fields related to climate adaptation and mitigation (Kirchhoff, Carmen Lemos, & Dessai, 2013; Lemos, Kirchhoff, & Ramprasad, 2012; Wall, Meadow, & Horganic, 2016). This evolution away from the ‘loading-dock’ model is driven by both the increasing evidence of adverse societal impacts of climate change as well as the growing demand from funding agencies for evidence  192 showing the impact of the research (Feldman & Ingram, 2009; Lemos, Kirchhoff, & Ramprasad, 2012; Thornton, 2006).   Besides validating the utility of the information produced by the tool, assessing the value of the information from the users’ point of view can benefit both users and researchers. For example, such assessment can help researchers build trust with the users, which can influence the users’ perception of the information’s credibility and salience (Cash et al., 2006; Kirchhoff, 2010) and identify barriers to its uptake (Feldman & Ingram, 2009; Thornton, 2006). For decision-makers and planners, the process can help them in choosing which tools to adopt and better understand how to employ them in an effective manner (Rice, Woodhouse, & Lukas, 2009).   To facilitate for the end-to-end usefulness of the RIPs method, this component of the dissertation aims to investigate in what ways can the RIPs or the RIPs method support SLR adaptation planning from the potential users’ perspective. Specifically, this component will assess whether the RIPs method can support SLR adaptation planning in the ten anticipated ways listed in Table 6.1. The RIPs method was motivated by the need to help users account for uncertainties and local contexts when identifying an initial set of adaptation options to further refine for robustness at a later stage. It is anticipated that the RIPs can help users develop adaptation options in five different ways (#1 to 5 in Table 6.1) by visualizing the nature of different impacts and how it changes (or not change) in space under different futures. For example, in comparison to designing an adaptation option to address an impact of a certain magnitude and distribution according to one specific future scenario, the user may develop a more uncertainty-tolerant adaptation option by designing it to address the impact represented by an RIP that is robust  193 across different land-use conditions and a range of SLR from 1 to 2m. However, besides providing information to help develop adaptation options, the RIPs are also expected to be able to support adaptation by serving as a way to gather resource and stakeholders support for adaptation. For example, literature (e.g., Kirchhoff et al., 2013) suggests that the visual and spatially explicit nature of climate information, as provided by the RIPs, can make it a useful tool to communicate the potential impacts to a range of stakeholders. Furthermore, using the previous example of more uncertainty-tolerant adaptation options, stakeholders and funding agents may see that as better investments and provide support and resources for implementing that option.  Table 6.1 Ten anticipated ways in which the RIPs can be used to support SLR adaptation planning  Ten	anticipated	ways	in	which	the	RIPs	can	be	used	to	support	SLR	adaptation	planning	Develop	adaptation	options	1. Generate	new	ideas	for	SLR	adaptation	options	2. Generate	more	refined	and	targeted	SLR	adaptation	options	3. Consider	a	wider	range	of	adaptation	types	(e.g.	soft,	hard,	combination)	4. Development	of	SLR	adaptation	options	that	are	more	uncertainty-tolerant	5. Facilitate	for	long-range	planning	of	how	to	modify	current	options	to	respond	to	a	worse	situation	Access	to	resources	and	support	for	implementation	6. Prioritize	SLR	adaptation	efforts	and	resources	7. Serve	as	a	useful	tool	for	communicating	SLR	risk	8. Identify	new	types	of	stakeholders	to	engage	in	planning	9. Provide	justification	for	planning	beyond	the	common	worse	case	scenario	10. Provide	better	leverage	to	request	for	resources	 194 6.2 Methodology To address the research question, the researcher hosted an expert workshop on 31st October 2017 with the potential users of the RIPs as produced in the case study at the City of Vancouver (CoV). The objective is to present the RIP method and the RIPs from the City of Vancouver case study to the potential users and assess whether they think the RIPs can potentially support SLR adaptation planning. An expert workshop approach was chosen for several reasons. Firstly, since the RIPs method is new, somewhat complex, and significantly different to conventional impact assessment methods, it is important to allow the participants to easily ask questions about the results and how to use it. Therefore a workshop environment that encourages idea exchange and exploratory dialogues is important. Secondly, a large amount of information must be communicated to the participants to help them understand the RIPs method and what the RIPs from the case study represent. Therefore the workshop option is preferred since it can ask for more time from the participants than what a one-on-one interview typically would permit. Lastly, in comparison to a one-on-one interview, the group setting of a workshop may allow participants to feel a sense of anonymity to their ideas and less pressure to implement the possible ways in which they may use the RIPs in their work related to SLR adaptation. This can allow participants to think more freely and facilitate for thinking “outside the box”.  6.2.1 Participants and recruitment process The potential users of the RIPs produced in the case study at the CoV are the subjects of this research component. To be included in this study, the participants have met at least one of the following criteria: • Currently and directly involved in SLR adaptation planning for the CoV  195 • An expert in a critical infrastructure or service in the CoV that can potentially be affected by SLR according to the RIPs of the case study • An expert in SLR as a problem in the CoV or the region (Metro Vancouver) Most participants were identified through the process of working in partnership with the CoV on the case study, while others were identified by their work title as shown on the CoV’s official staff directory. Besides the City Planners, the CoV’s adaptation planning process typically involves a diverse range of stakeholders, including consultants that specializes in decision-making and/or flood management; City staff of various related departments; and staff of different utilities (e.g., electricity providers). Therefore, the workshop also aims to ensure experts from each of those groups that meet the criterion are invited. Each participant was invited to the workshop via an email invitation with a one-page summary of the project and workshop objective. This email invitation and attached project summary can be found in Appendix D1. If the invitee cannot attend, the invitee is asked to suggest another person. The maximum number of participants to attend the workshop was set to be 20, which is large enough to include a diverse range of experts but also small enough to promote more intimate plenary and group discussions.  6.2.2 Workshop structure and activities The expert workshop was a half-day workshop led by the researcher. As shown in Table 6.2, the workshop consisted of three key activities – 1) presentation of the RIPs method and RIPs of the case study, 2) survey, and 3) group discussions.   196 Table 6.2 Expert workshop agenda  Three research assistants were also present to assist with facilitating group discussions and note-taking. The research assistants were prepared for this task through a two-hour meeting one week before the workshop. The meeting involved 1) a dry-run of the workshop presentation to ensure they have a clear understanding of the RIPs and the RIP method, 2) discussion of the workshop agenda and 3) discussion on how they prefer to facilitate group discussions. The latter provided useful insights for designing the group discussion component of the workshop.  6.2.2.1 Presentation Before the presentation, about ten minutes were used to welcome the participants and to describe the workshop objective, agenda, and highlight that the workshop is an exploratory exercise where there are no right or wrong answers, and thinking “outside the box” is encouraged. The presentation started with a brief outline of why the RIP method was developed, how it works, and why it was applied to the CoV as a case study. The rest of the presentation focused on the RIPs of three different types of impacts – 1) business disruption in the tertiary sectors, 2) social services facilities potentially affected, and 3) sewage backup damage potential. Although 14 types of impacts were modeled in the case study, due to limitation of time and concern of Time	 Activity	8:30	–	9:00	 Breakfast	and	registration	9:00	–	9:15	 Welcome	9:15	–	10:15	 Presentation	and	Q&A	10:15	–	10:30	 Survey	10:30	–	11:00	 Break	with	refreshments	11:00	–	12:00	 Group	discussions	12:00	–	12:15	 Time	for	survey	revision	12:15	 Lunch	and	closing	remarks	 197 information overload, one impact from each major categories (i.e., economic, social, and environmental) were presented in more detail to allow the participants to understand how to interpret the results and get a good idea of what the RIPs method can provide. The presentation of the RIPs of each impact started with a high-level summary of the impact magnitude and spatial variation in the three groups of patterns, as described in Chapter 5, followed by highlighting some interesting insights that can be gathered from examining the variations between RIPs within each group. This demonstrated to the participants that the RIPs can be used at different levels of detail. The presentation session ended with 15 minutes for questions from the participants to provide an opportunity for participants to ask any immediate questions and comments. The slides of the presentation can be found in Appendix D2.  6.2.2.2 Survey Immediately after the question and answer period, the participants were asked to complete a short paper-based survey without discussing with other participants, aiming to capture the individual participant’s perspective without the potential influence of a participant(s) that dominates a group discussion. The full survey can be found in Appendix D3. The first page of the survey is an informed consent form where the participant is asked to read through thoroughly but is not required to sign. The survey consists of 17 questions, five of which ask for demographic information, including job title, affiliation, and whether they are currently involved in CoV’s SLR planning. The rest of the questions focus on assessing: a. Whether the participant thinks the RIPs can support SLR adaptation planning in the ways listed in Table 6.1,  198 b. What features of the RIP (i.e., spatial explicitness, robustness, diversity range of impacts assessed) are attributable to the way the RIPs can be used, and c. If there are other ways in which they think the RIPs can be used to support SLR adaptation planning.  The anticipated ways in which the RIPs can support planning are presented as statements and participants were asked to what degree they agreed with the statement using a Likert scale from 1 to 5, where 1 is ‘strongly disagree’, 2 is ‘agree’, 3 is neutral, 4 is ‘agree’ and 5 is ‘strongly agree’. To address question b listed above, the survey explicitly asks the participants to answer the questions with those three features of RIPs in mind. Furthermore, the questions are specifically designed to investigate whether a specific way to use the RIPs is attributable to a certain feature of the RIPs method. Many statements in the survey questions specify which feature facilitates for that specific way to use the RIPs. For example, one of these statements was "the robustness of the impact patterns can support the development of adaptation options that are more uncertainty-tolerant". Here the specific attributing feature is the “robustness” of the RIPs. However, no specific attributing features were suggested in the questions about four out of the ten ways listed in Table 6.1 – prioritizing efforts; generating new ideas; generating more refined and targeted options; and serving as tool for communicating SLR risk. Instead, there was a space provided for the participant to provide comments or examples. These four ways are treated differently because they are likely to be attributable to more than one feature of the RIPs method and they are broader concepts relative to the others. The last two questions of the survey are open-ended to provide space for the participant to indicate any other ways in which the RIPs can be used and any general comments.    199 6.2.2.3 Group discussion After the survey, participants were assigned to specific groups and asked to address the two questions below in their groups: 1. Overall, do you think the robust impact patterns method can support adaptation planning for sea-level rise (other climate impacts), and why? 2. Please provide examples of potential ways in which the robust impact patterns can support adaptation planning or risk reduction efforts  The main purpose of this group exercise is to encourage participants to think collectively and collaborate on generating examples of how the RIPs or the RIPs method can or cannot support SLR adaptation. Therefore, the participants were assigned to groups in a way that would allow each group to have a diverse range of experts or stakeholders of different departments and levels of experience, in order to support development of diverse ideas and opinions. In addition to addressing the two questions above, this activity also serves as an opportunity for participants to voice any thoughts triggered by the survey.  In response to the two discussion questions, each participant is asked to write down their answers on a sticky-note, and share their answer or idea with the rest of the group as they add their sticky-note to a large sheet of paper at the centre of the table. This design helps ensure that each participant gets a chance to articulate and share his or her thoughts with others. In comparison to an unstructured group discussion, this design helps avoid the situation where one or more participants may dominate the conversation. This design also helps relieve the moderator from some burden of note-taking and allow the moderators to focus on moderating the discussion and capture insights that are not captured in the participants’ sticky-notes. In plenary, each group was  200 asked to report back on highlights from their respective discussions. To capture any new ideas triggered by the group discussion or thoughts that the participant does not want to share in a group setting, participants were given 15 minutes before lunch as an opportunity to revise their survey response if they wished to.  6.2.3 Analysis method The workshop outcome is made up of information collected in the ways listed in Table 6.3. The workshop outcomes were analyzed through a two-level approach – a high-level, quantitative approach, and a detailed qualitative approach.   Table 6.3 List of different ways in which information was captured for this component of the dissertation   6.2.3.1 High-level survey analysis This analysis aims to provide a general idea of the participants’ response to the RIPs and the RIPs method – whether they think the RIPs can be used in the anticipated ways. The data from the Likert type questions were analyzed with descriptive statistics and the responses between subgroups of participants were tested for statistically significant differences. Specifically, the responses from two pairs of subgroups (Pair A and B) were tested for differences. First, the responses of participants that are currently involved in CoV’s SLR adaptation planning are tested against those who are not involved (Pair A), to investigate whether having prior knowledge From	the	written	survey:	1. Demographic	data	about	the	participant		2. Likert	type	data		3. Written	comments	and	examples	on	open-ended	survey	questions	From	the	group	discussions:	4. Answers	written	by	participants	on	sticky	notes	5. Notes	written	by	the	group	moderator	to	provide	additional	insights	 201 about CoV’s SLR planning efforts changes their perspective on the RIPs’ utility. Second, the responses from participants that are CoV staff are tested against those who are not CoV staff (Pair B), to investigate whether the RIPs’ utility is dependent on the CoV context of the RIPs. The subgroups were identified using responses to the demographic questions in the survey. Since the data is categorical and ordinal, parametric tests cannot be used to test the difference between the sub-groups. Furthermore, given the small sample size and only two subgroups are compared at a time, the Mann-Whitney U test was employed as a suitable non-parametric test.   6.2.3.2 Qualitative analysis This analysis aims to identify: 1) the ways in which the participants think they can use the RIPs or the RIP method to support SLR adaptation planning, 2) what feature of the RIPs is attributable to how useful it is, 3) specific contextual examples of how the RIPs can be used, and 4) how can the RIPs or the RIPs method be modified to become more useful for SLR adaptation planning. Although participants were not explicitly asked to comment on how the RIPs can be modified to become more useful, participants suggested various ideas during the group discussions as well as in their survey responses.  The qualitative data (i.e., all workshop outcome, except the survey responses from the Likert type questions) was first imported into Nvivo to be coded and subsequently grouped by themes. In Nvivo, “nodes” define themes. Coding nodes were defined before and during the coding process. The pre-defined nodes represent the anticipated ways in which the RIPs can support SLR adaptation planning. Three additional nodes were defined to represent new themes there were identified during the coding process, including:  202 • How the RIPs can support the reframing of the SLR issue as opportunities • How to modify the RIPs in the case study to become more useful for SLR adaptation planning at the CoV • How to modify the RIPs method, in general, to become more useful for SLR adaptation planning  6.3 Results 6.3.1 High-level analysis of survey responses Fifteen out of twenty invited experts attended the workshop and participated in all workshop activities. The participants represent a diverse range of experiences, sectors, and departments in the CoV [Figure 6.1 and Figure 6.2]. Furthermore, their backgrounds include engineering, planning, landscape architecture, environmental sciences, emergency management, and decision support.   Figure 6.1 Range of City of Vancouver departments and organizations represented by the workshop participants      203 Figure 6.2 Number of years the participant spent in their current role  As described in the previous section, participants were asked in the survey to indicate to which degree to they agree that the RIPs can be used to support SLR adaptation in ten different ways. These ten ways [Table 6.1] are broadly categorized into two groups, those that: 1) Support the development of adaptation options or plans, and  2) Support the access to resource and support for implementing adaptation  Based only on the responses for these Likert type questions, more than 60% of participants responded positively with either ‘agree’ or ‘strongly agree’ to the five ways in which the RIPs can be used to support adaptation options development [Figure 6.3]. About 25% of the participants indicated ‘neutral’ or ‘disagree’ with two out of the five ways being:  1) refinement of adaptation options and 2) supporting long-range modification of options [Figure 6.3]. Given that both of these ways are about modifying specific or existing options, the fewer participants in support of these applications may be attributable to the lack of direct linkage to adaptation options in the RIPs.   204 Figure 6.3 Survey responses for Likert type questions related to the five ways to use the RIPs in supporting SLR adaptation options development  In comparison to supporting adaptation options development, more participants (80% or more) responded positively to the ways in which the RIPs can support accessing resources and stakeholders’ support to implement adaptation actions [Figure 6.4]. Out of the 10 anticipated ways, the only one that had unanimous agreement was using the RIPs as a tool to communicate SLR risk [Figure 6.4].   205 Figure 6.4 Survey responses for Likert type questions related to the five ways to use the RIPs in supporting access to resources and stakeholders support for implementing adaptation actions.  6.3.1.1 Responses of subgroups While the responses were generally positive, the responses from sub-groups of participants are compared to further investigate whether the participants’ perspective on the RIPs’ utility for SLR adaptation planning is:   a) Influenced by the participants’ prior knowledge about CoV’s SLR planning efforts, assuming participants that are not involved in the CoV’s SLR adaptation planning have little to no knowledge about the CoV’s SLR planning efforts.   206 b) Dependent on the CoV context of the RIPs, assuming that CoV staff values the CoV context more than non-CoV staff.  To investigate these two issues, the survey responses of subgroup Pair A and B are compared respectively for statistically significant differences: • Pair A: Participants currently involved in the CoV’s SLR adaptation planning (n = 8) vs. those who are not involved (n = 7) • Pair B: CoV staff (n = 8) vs. non-CoV staff participants (n = 7)  Table 6.4 and Table 6.5 shows the results from the Mann-Whitney U tests conducted for subgroup Pair A and B respectively. In both cases, the null hypothesis is that there is no difference between the responses of the two sub-groups and it is tested with the Mann-Whitney U test with significance level of 0.05. The test results indicated that there are no statistically significant differences between the subgroups’ responses for all 10 Likert type questions [Table 6.4 and Table 6.5]. This indicates that prior knowledge of CoV’s SLR adaptation planning efforts and the CoV context of the case study RIPs does not have a statistically significant influence on the participants’ perspective on the RIPs’ utility for SLR adaptation planning. However, in comparison to the two-sample t-test, the Mann-Whitney U test is known to have lower ability to detect differences. Therefore future investigations should aim to identify indicators that are more suitable to quantify the participants’ perspectives on the RIP utility that would permit more rigorous statistical testing. 207 Table 6.4 Statistical test (Mann-Whitney U test) for differences between responses of participants currently involved in CoV's adaptation planning and those who are not involved (Pair A).  Notes: 1. If the shapes of the response distribution between the two groups are similar, then we can use the Mann-Whitney U test results to infer about the medians of the two groups. If the shapes are dissimilar, an inference can only be made in terms of their mean ranks. This was determined by visual inspection of the respective population pyramids, which can be found in Appendix D4 2. The p-value here is the asymptotic p-value rather than the exact p-value because the exact p-value tends to be inflated when more than 2 participants in a group have identical answers, which is common in the case of ordinal type dependent variables. In this case, it is better to report the asymptotic p-value (an approximation of the exact p-value, which becomes closer to the real p-value as sample size increases)   Questions	Similar	distribution	shape?	1	Median	or	mean	rank	Test	statistic	 P-value	2	Involved		 Not	involved	Adaptation	options	development	New	ideas	of	adaptation	options	 N	 7.12	 9.00	 21.0	 .296	Refine	adaptation	options	 Y	 4	 4	 24.5	 .657	Wider	range	of	options	 N	 6.69	 9.50	 17.5	 .143	More	uncertainty	–tolerant	options	 Y	 4	 4	 25.0	 .677	Long-range	modifications	 N	 7.25	 8.86	 22.0	 .458	Access	to	resources	and	support	for	implementation	Prioritize	efforts	 N	 8.06	 7.93	 28.5	 .947	Communicate	risk	of	SLR	 Y	 4.5	 4	 30.0	 .789	Identify	new	stakeholder	types	 Y	 4	 4	 18.5	 .427	Justify	planning	beyond	1m	SLR	 N	 8.62	 7.29	 33.0	 .509	Better	leverage	for	resources	 Y	 4	 4	 27.0	 .386	 208 Table 6.5 Statistical test (Mann-Whitney U test) for differences between responses of CoV staff participants and those who are not CoV staff  New	ideas	of	adaptation	options	 N	 7.88	 8.14	 27.0	 .881	Refine	adaptation	options	 N	 7.25	 8.86	 22.0	 .446	Wider	range	of	options	 N	 8.25	 7.71	 30.0	 .780	More	uncertainty-tolerant	options	 Y	 4	 4	 25.0	 .677	Long-range	modifications	 N	 8.12	 7.86	 29.0	 .902	Prioritize	efforts	 N	 9.44	 6.36	 39.5	 .128	Communicate	risk	of	SLR	 Y	 4.5	 4	 30.0	 .789	Identify	new	stakeholder	types	 Y	 4	 4	 29.5	 .427	Justify	planning	beyond	1m	SLR	 N	 7.25	 8.86	 22.0	 .428	Better	leverage	for	resources	 Y	 4	 4	 27.0	 .386	 Notes: 1. If the shapes of the response distribution between the two groups are similar, then we can use the Mann-Whitney U test results to infer about the medians of the two groups. If the shapes are dissimilar, an inference can only be made in terms of their mean ranks. This was determined by visual inspection of the respective population pyramids, which can be found in Appendix D5 2. The p-value here is the asymptotic p-value rather than the exact p-value because the exact p-value tends to be inflated when more than 2 participants in a group have identical answers, which is common in the case of ordinal type dependent variables. In this case, it is better to report the asymptotic p-value (an approximation of the exact p-value, which becomes closer to the real p-value as sample size increases)   Questions	 Similar	distribution	shape?	1	Median	or	mean	rank	 Test	statistic		P-value	2	CoV	staff	1		 Not	CoV	staff	Adaptation	options	development	Access	to	resources	and	support	for	implementation	 209 6.3.2 Qualitative analysis - potential applications of the RIPs to support adaptation This analysis used data drawn from the open-ended questions in the survey responses, the sticky-notes written by the participants for the group discussions, and notes from the moderators in the workshop. For each of the ten anticipated ways in which RIPs can support SLR adaptation planning [Table 6.1], the following sections will discuss the key themes in the participants’ responses that were identified using Nvivo, in terms of: 1) whether the participants think the RIPs and RIPs method can support adaptation in the given way and wherever possible, 2) provide the reasons why, and 3) practical examples. Sections 6.3.3 to 6.3.6 addresses the results for the anticipated ways in which the RIPs can support adaptation options development, while sections 6.3.7 to 6.3.9 addresses the anticipated ways to support accessing resources and support for implementation. Sections 6.3.10 and 6.3.11 describes the unanticipated ways that were suggested by the participants. Section 6.3.12 describes the participants’ suggestions on how to modify the RIPs and the RIPs method to become more useful for supporting SLR adaptation planning.  6.3.3 Generate new ideas and wider range of option types Most participants agree (11/15 agreed, 3/15 strongly agreed, 1/15 neutral) that the RIPs in this application can support SLR adaptation by supporting the generation of new ideas for SLR adaptation options. Furthermore, most participants (10/15 agreed; 4/15 strongly agreed; 1/15 neutral) also agreed that the diversity of impacts shown in the RIPs could support the development of a wider range of adaptation options. Responses suggest that the RIPs can help generate new ideas by the following ways:    210 Bring	together	private	and	public	sectors	Since the assessed impacts are relevant to both public and private sectors, the results can bring together public and private sectors, which can create new discussions that may lead to new ideas. “Spurs public and private sectors innovation and leads to a new way at doing business” [Group discussion sticky note] “Enabling access to broad range of stakeholders means innovation in adaptation” [Group discussion sticky note] New	conversations	New ideas may be generated by allowing the users to have conversations beyond the normal conditions (e.g., common SLR scenario of 1m rise).  “Conversations out [sic]19 normal confines (more impact scenarios)” [Moderator note] “Gives a wide visual and breaks it down to 0-6, the spectrum of impacts is important because it provides a big picture to prevent going down the middle path. Helps get people thinking about other impacts.” [Moderator note] Reveal	new	demands	Visualizing the spatial extent of impacts that the users have not considered before can also generate the demand for new adaptation ideas. Some participants have associated the generation of new adaptation ideas with the RIPs of two particular types of impacts. One is the RIPs of the SBDPI because it can help highlight that SLR impacts can affect population and assets located outside the inundation areas and this can generate the new demand for ways to protect the neighbourhoods.                                                 19 The correct phrases should have been “out of normal confines…”   211 “The sewer impacts map in particular helps to illustrate impacts beyond shorelines and can help identify ways to protecting neighbourhoods” [Participant #1, Consultant] Another is the RIPs of the potentially affected social service facilities. Realizing that community centers that currently serve as disaster hubs can also be affected by SLR, can generate demand for new ideas. “Consider how social service facilities such as community centres are impacted and develop design ideas to mitigate effects of SLR. Noting that in emergency scenario these are places where people gather” [Participant #2, CoV Planning] In relation to the above point, a participant has also suggested that the RIPs can be used to identify “safe spots” to allocate supplies and muster zones. Other new ideas generated include: • Reconsidering construction material based on debris generated • Use the RIPs of the SBDPI as policy lever for sewer upgrade • Provide economic investment and incentives for adaptation “Particularly with respect to construction material (debris), policy levers (sewer upgrades) and economic investment and incentives” [Participant #9, CoV Emergency Management] “Implement the patterns for design of the sewer/storm infrastructure i.e. backwater valves on the arc connections” [Participant #11, CoV Development and Permitting] "The patterns could support the adaptation of the consequences of drinking water reservoirs and the implication of saltwater contamination etc." [Participant #11, CoV Development and Permitting]  212 6.3.4 Refine and target efforts Although 10/15 participants agree that the RIPs can help develop more refined and targeted adaptation options, this anticipated way of using the RIPs to support SLR adaptation is the least agreed upon, with 4 participants indicated neutral and 1 disagreed. This could be related to some participants' comments about how the impact variables in this application are too high-level to help refine existing options. On the other hand, many participants indicated that they believe the RIPs can help refine adaptation efforts because of the block-level resolution of the RIPs. Specifically, impacts shown at the block-level was suggested to be able to support refinement in the following ways: § Customize plans for neighbourhoods – tailor the actions to the needs and vulnerabilities of the neighbourhoods § Support more refined understanding of the impacts - can spur discussions about what matters to the community and help tailor actions to protect what matters. "Neighbourhood scale cumulative impacts can help spark further refinement of understanding of consequences and provides access points for discussion based on assets that matter to community" [Participant #9, CoV Emergency Management] “Illustrates the importance of a community/district level approach” [Participant #13, CoV Planning]  6.3.5 More uncertainty-tolerant options The key aspect of the RIPs method that aims to help users manage uncertainties in SLR adaptation planning is the robustness of the RIPs. Most participants (12/15) agreed that the  213 robustness of the RIPs can support development of adaptation options that are more uncertainty-tolerant. Additionally, participants also suggested that the RIPs method could help manage uncertainties in the following ways: § Help identify adaptation options by allowing users to “search for options that can protect against 0-1m as well as 4-6m” [Group discussion sticky note] of SLR. § Support and motivate the consideration of a wider range of scenarios and impacts “The methodology has the capacity to make evident a large portfolio of plausible scenario and associated uncertainties” [Group discussion sticky notes] “Allow users to consider uncertainties in a comfortable way by showing a spectrum of outcomes rather than focusing on probability. However, this may not be effective for decision-making as some of the details required for decision-making may be lost in the broad range of outcomes” [Moderator note] “Support more confident decisions in the face of uncertainty…” [Participant #12, Consultant] “Gives a wide visual and breaks it down to 0-6, the spectrum of impacts is important because it provides a big picture to prevent going down the middle path. Helps get people thinking about other impacts.” [Moderator note] “Picture is so important to the assessment ability to evaluate and compare scenarios, manually you need to select and pick. When done automatically, you can see many more patterns and strengthen decision-making (not individual selection)” [Moderator note] § Highlight the need to account for interdependencies and cascading impacts in adaptation planning  214 § Help ventilate the planners’ anxiety of addressing “all the scenarios” – planners may be under the pressure to address all possible scenarios and this method may provide a way to address this need. “Captures planners’ anxiety of ‘all the scenarios’ – which can point or shape business case for land-use and infrastructure decisions” [Group discussion sticky notes]  6.3.6 Long-range planning of modifications Similar to refining adaptation options, using scenarios beyond the conventional worse case (e.g., 2m of SLR) to plan for long-term modifications of options is one of the least agreed upon ways of how RIPs can support adaptation, with 3/15 indicating neutral and 1/15 disagreed.  Some participants’ comments relating the RIPs to planning may help reveal why they may find it difficult to use the RIPs to support long-range planning of SLR adaptation. An aspect of the RIPs that may not help long-range planning is the lack of a time aspect to each RIP. The timing of actions is an important part of planning. Since each RIP does not suggest when each pattern may eventuate, it can be an obstacle to planning. Furthermore, although the RIPs in this application shows the potential impact over a wide range of SLR, short-term adaptation actions to be implemented may change the distribution and magnitudes of impacts. Therefore, applying the RIPs method to a lower range of SLR with finer increments (e.g., 0.2m to 2m) may be more useful to support short to medium term adaptation. These short-term options can then be integrated into the future scenarios of another RIP application with the higher levels of SLR.  On the other hand, other participants find that being able to see the potential impacts over a wide range of futures is particularly helpful for planning or supporting the adaptation of infrastructure  215 that either take a long time to implement or have a long life cycle. These include changing building codes, upgrading sewer system, and relocating or upgrading electrical supply systems (e.g., substations).   "Influence regulation for building control at National, Provincial, and Municipal level. Not only does this apply to new building but also upgrades to existing buildings going through the planning process (e.g. policy statement) or rezoning. Input this data early on into the planning process." [Participant #2, CoV Planning] "(Can help assess impact on) critical infrastructure: bridges, tunnels, hospitals, schools, and electrical generation. Timeline for repair or mitigation as well as relative cost per person or square area over a timeframe.” [Participant #4, CoV Sustainability] Typically, the nature of these infrastructures does not support short-term actions that can be taken to respond to incremental changes, but rather they can rely on different contingency plans. The broad range of scenarios in the RIPs can support development of such contingency plans.  6.3.7 Prioritizing efforts Most participants agree that the RIPs can support adaptation efforts by helping users prioritize their efforts or resources. Specifically, RIPs can support prioritization by helping users: • Identify hotspots of impacts • Identify safe-spots • Balance the multiplicity of impacts to the number of vulnerable people exposed • Refine existing priorities 		 216 Identify	hotspots	The most commonly suggested way in which RIPs can support prioritization is how the RIPs can help users identify hotspots of impacts, which is attributable to the spatial explicitness and spatial scale of the RIPs.  “Help different level of government budget and realize which areas to prioritize resources” [Participant #8, CoV Development, and Permitting]  Additionally, the City-wide coverage and block-level resolution can help highlight unexpected hotspots – those beyond the inundated areas. "On a citywide scale, it highlights areas where SLR should be a consideration beyond the usual suspects, e.g. around False Creek. On a more local scale, identifying individual sites could be helpful." [Group discussion sticky-note] “Drive attention to high impact areas. But also brought attention to areas you wouldn’t think of – sewage and the cascading effects” [Moderator note] 	Identify	safe-spots	On the other hand, the RIP can also help identify ‘safe spots’ – areas that may be less affected relative to the rest of the city. This can support efforts to allocate critical infrastructure and emergency supplies to reduce vulnerability. Suggested examples include: “(Help planning for) risk hubs for post-event mustering – power, water, sanitation, communication” [Group discussion sticky notes] “What are the critical pieces of infrastructure and how long down? Who and where are vulnerable?” [Moderator note]  217 	Balancing	the	multiplicity	of	potential	impacts	with	number	of	vulnerable	people	It is intuitive to prioritize efforts to areas with the highest number of vulnerable people. However, a participant has found that by seeing the multiple types of impacts in a spatially-explicit manner can help balance efforts between prioritizing areas with high concentration of socially-vulnerable population with areas that are exposed to many different types of impacts. For example, an area potentially affected by one type of impact and have a high concentration of population may be considered to be as high in priority as an area less populated but may be facing multiple impacts (e.g., sewage back-up, building damage, and social services disruptions). "Seeing the layers of risk helps to prioritize actions to balance with number of people of vulnerability" [Participant #4, CoV Sustainability] 	Refine	existing	priorities	While a community may have already identified their priorities in SLR adaptation, the impact maps (or the spatial aspect) can help refine the existing priorities.  “There could be an opportunity to align impact maps or patterns with existing planning priorities to help establish a higher priority to help geographical prioritization for comprehensive planning initiatives.” [Participant #13, CoV Planning]  6.3.8 Communicating SLR risk All participants agreed (8/15 agreed, 7/15 strongly agreed) that the RIPs could serve as a useful tool for communicating the risk of SLR to potentially affected population and organizations. Participants suggested that the RIPs could help communicate SLR risk in the following ways:  218 	Educate	and	raise	awareness	in	communities	to	provide	support	to	public	sector	leadership	in	adaptation	As the impacts are shown at the block-level resolution, stakeholders of different neighbourhoods can see what impacts are specific to them. Therefore can better attain community input and support, which is needed by public sector leadership.  “Helps to diversify the risk. Not just people that live near water that are affected. Helps to put it on people's radar.” [Participant #2, CoV Planning] 	Spur	conversations	and	better	understanding	of	cascading	effects		The RIPs allow the user to visualize the horizontal extent of the direct and indirect impact, which facilitates a better understanding of cascading effects of SLR and how the impact of SLR can reach beyond the normal confines, in terms of scenarios and spatial distribution – i.e., beyond the shoreline.  "The work around disruption (electrical and sewage) is extremely useful to broaden the conversation" [Participant #7, Consultant] 	Gather	neighbourhood-level	input	Awareness of neighbourhood-specific impacts may motivate behavioural change in individuals. Furthermore, a better understanding of neighbourhood-level impacts can facilitate conversation about what matters to the neighbourhood can be affected by flooding.   219 “Neighbourhood scale cumulative impacts can help spark further refinement of understanding of consequences and provides access points for discussion based on assets that matters to community” [Participant #9, CoV Emergency Management] To build upon the point above, participants also suggest that by overlaying other City indicators, such as walkability, bike lanes, and parks, can help identify what matters to people.  Communicate	risk	of	SLR	to	multiple	types	of	end-users	Participants have compared communicating the impacts via maps to using a universal language, where multiple types of stakeholders would find it understandable and relevant. Specific types of stakeholders suggested include, homeowners, developers, planners, and insurance companies. "Mapping is an excellent way of translating complex info in an easy to understand format” [Participant #13, CoV planning]  Besides being understandable to multiple types of stakeholders, many participants (8/15 agreed; 3/15 strongly agreed; 3/15 neutral; 1/15 no answer) agreed that the diversity of the impacts shown in the RIPs could highlight new types of stakeholders to engage in SLR adaptation. However, it is not clear which new types were identified.  	Communicate	impacts	of	different	adaptation	alternatives		The mapping presentation of the assessed impacts can help transform numbers into stories. Therefore, if adaptation options are integrated into the future scenarios where RIPs are based upon, the RIPs can serve as a basis to develop narrative-based scenarios that represent the different adaptation planning alternatives.   220 “It may be worth considering the development of narratives that represent different adaptation pathway decisions (optimization vs. trade-offs) to help communicate the compounding impacts of SLR (economic/social/environmental) to different stakeholder audiences” [Participant #6, NRCan] Furthermore, the RIPs can be refined to include impact variables that show the cost per hectare or the number of people impacted per hectare and time required to help prioritize and communicate the trade-offs of different adaptation options.  Concerns	and	suggestions	around	using	RIPs	as	communication	tool	Some participants also raise some concerns about using the RIPs to communicate SLR risks, as indicated by the following quotes from survey responses: “It depends on how the information is presented. The maps may need to be refined further and contextualized because they imply no adaptation and convey message of significant impact. City of Vancouver has similar maps and the impact patterns convey that information but just different. These maps would be helpful generally speaking.” [Participant #3, CoV Sustainability] “Communicating "risk" might not be the correct term since risk is defined by chance of occurrence, which I believe you are intentionally not wanting to consider? Perhaps communicating potential outcomes, vulnerability, priorities, etc.” [Participant #12, Consultant]  221 “Helps to show how far reaching the impacts can be. Will be important to tone down the score [sic]20 factors with 6m SLR, which may drive bad decisions.” [Participant #1, Consultant]  Although many ways in which the RIPs can help communicate SLR risk is attributable to its spatial explicitness and scale (neighbourhood level variations), it is worth noting how the broadness of the range of future scenarios also played a significant role. For example, the single worse case scenario of SLR impacts can be made less overwhelming and easier to communicate to the general public when it is presented as RIPs (i.e., impacts that may occur under a range of scenarios) rather than a single prediction. “Practically, the worse case scenario is something we don't want to make even more terrible and this tool could support this“ [Participant #7, Consultant] Other modifications that may help the RIPs to be even better communication tools include: 1) making the model 3D, and 2) use the RIPs in conjunction with other communication tools, such as storytelling narratives, to help bring awareness and understanding  6.3.9 Justify to plan beyond 1m of SLR and provide better leverage Most participants (13/15) agreed that the RIPs could support justification for planning for a worse scenario than the one suggested by the provincial guidelines (i.e., 1m of SLR by 2100). As discussed in the subsection above, the RIPs may be particularly helpful for planning adaptation of infrastructure or services that take a long time to implement or have a long life cycle. The long life cycle of these infrastructures also limits the adaptiveness of the actions. Therefore their                                                 20 Based on the context of the discussion, the participant has meant “scare” instead of “score”  222 adaptation planning should consider a wider margin of uncertainties. By presenting the RIPs associated with up to 6m of SLR to decision-makers can help justify the need to plan for scenarios beyond 2m of SLR.  For other aspects of the city that permits more incremental changes, the RIPs beyond 2m of SLR may not be as helpful. Since the provincial guideline recommends cities to adapt to 1m SLR that has higher probability of occurrence, it will be difficult to argue for adaptation beyond 2m of SLR if one relies on resources from the province. “Yes, they can support consideration but the business case and probability of occurrence will have significant weight because local government has limited resources.” [Participant #3, CoV Sustainability] Nonetheless, most participants (14/15) agreed that the robustness of the RIPs can provide better leverage to request for resources to support implementation.  “Put forward the case to direct some money towards this in planning or rezoning projects” [Participant #2, CoV Planning] This could be attributable to the following benefits of the RIPs as suggested by participants: • Provide better understanding of costs, vulnerabilities, and who are the potentially affected population • Support conversations regarding ownership of potentially affected assets – e.g., Metro Vancouver vs. City of Vancouver sewer lines – that may provide leverage to request for resources from other sources or motivate cost-sharing. Specifically, it was suggested that by defining the issues in space could support the conversation around who should bear the costs.  223 • Highlights the need to be proactive  6.3.10 Reframing issues as opportunities Although not anticipated and was not mentioned in the workshop presentation or survey, using the RIPs to help reframe SLR risk as opportunities have surfaced as another way the RIP method can support adaptation. Specifically, participants suggested the following ways to use RIP in reframing SLR impacts as opportunities: § The spatial explicitness of the RIPs helps highlight safe zones, which can be considered as areas for better investments § By planning according to patterns that can occur across a broad range of scenarios, the RIPs can help demonstrate how such planning can lead to better investments (“bigger bang for your buck” was the phrase used by a participant), which may help avoid certain costs and expenses in the future (e.g. rebuilding infrastructure). “Support more confident decisions in the face of uncertainty (not so much supporting more options)” [Participant #12, Consultant]  6.3.11 Development of adaptation pathways As a specific way to use the RIPs to support the development of adaptation actions, a participant suggested that the RIPs method has the potential to allow backcast modeling where decision-makers are shown a portfolio of impact patterns associated with different future scenarios and they are asked to think ‘backwards’ to generate potential adaptation pathways to address the impacts they see in the different patterns. It is worth noting that adaptation pathways are different to adaptation options, as they represent a sequence of adaptation actions over time. The aspect in  224 the RIPs that facilitates this specific application may be the broadness in the range of scenarios as required in the method. "Backcast modeling presents a portfolio of plausible future scenarios and asks decision makers to identify pathways that either 1) optimize for planning objective, or 2) seek to balance trade-offs between choices and consequences. This might be a suitable framing of the robust impact pattern methodology" [Participant #6, NRCan] It was further suggested that after an adaptation pathway is elicited, it can be built into a new set of future scenarios to develop another set of RIPs such that the RIPs can then be used to develop a narrative representing that adaptation pathway to communicate the compounding impacts of SLR and trade-offs of different pathways to different types of stakeholders. It was also suggested that the RIPs could be useful in combination with other communication tools, such as storytelling narratives, to help raise awareness and better understanding. “It may be worth considering the development of narratives that represent different adaptation pathway decisions (optimization vs. trade-offs) to help communicate the compounding impacts of SLR (economic/social/environmental) to different stakeholder audiences” ” [Participant #6, NRCan] “Use in conjunction with other communication including storytelling narratives to help bring awareness and understanding.” [Participant #5, NRCan] It is interesting to note that this idea of using RIPs to support the development of adaptation pathways is, in essence, the same as the original intended purpose of the RIPs method, as described in Chapter 3. However, the intention was not to have the RIPs used throughout the  225 adaptation pathway development but rather to serve as an approach to inform the selection of a better pool of preliminary options as the building blocks of the adaptation pathways.   6.3.12 Suggested modifications While most participants agreed that the RIPs or the RIPs method could support SLR adaptation planning in the ways suggested in the survey, participants also provided various ideas to modify the application and the method itself that may make it more useful for adaptation planning.   Case	study-specific	modification	suggestions	Modeling impacts resulting from less than 1m of sea-level rise It was suggested that it would helpful to include SLR values that are less than 1m as it is more relevant for planning near-term adaptation actions, which is the City’s current priority. It was further suggested that this may allow one to see what kind of adaptation options can be robust in scenarios of SLR less than 1m as well as 4-6m. "Would be great to see impacts from <1m SLR, and then learn which protection measures are robust for both <1m and 4-6m. Will need to highlight near-term implications of SLR to be more tangible for Planners" [Participant #1, Consultant]  Including additional impact types Ecological impacts of SLR, such as coastal squeeze, are impacts that the City would be interested in including. While such an indicator would be suitable for measuring ecological impact, the change of coastal squeeze across different scenario will not resolve well at the city-wide scale of this application.  226 “Possible indicators for ecological impact - coastal squeeze - because landscape ecology may be best indicator bucket vs. species or yield).” [Participant #3, CoV Sustainability] Another impact that the City is interested in is the recreational and cultural impact of SLR and different adaptation options. For example, the recreational impacts of seawalls, as an adaptation option, on the community. A cultural impact of flooding on heritage or old buildings (e.g., in Chinatown) is another key concern. A CoV planner participant suggested that including such impacts could provide a significant political lever for adaptation action. “Also should consider assessing the recreational impact of options such as seawalls on community.” [Participant #3, CoV Sustainability] There are also concerns over the health impacts of flooding, such as hazardous material and mould, and financial impacts. Specifically, it was suggested that quantifying the impacts in monetary terms might support better leverage to request for resources to support implementation.  Include additional layers to provide context It was suggested that additional map layers could be included to provide more contexts to the users. For example: o Historical information, such as fire insurance maps, historical shorelines, and waterways o Recreational information, such as walkability, bike lanes, and parks, to help attain input from residents and identify what matters to them.     227 Longer power outage scenarios It was suggested that the current assumption of 1-week long power outage associated with substation damage should be modified to be more lengthy since repair of substation would take longer than 1 week. “Power outage to be more lengthy and extensive than indicated” [Participant #15, BC Hydro]  More detailed and rigorous information Some participants suggested that the impact information should be more detailed and rigorous to be helpful for decision-making. Furthermore, it may be helpful to refine the RIPs by providing contexts about the adaptation actions that the City is currently considering. “It depends on how the information is presented. The maps may need to be refined further and contextualized because they imply no adaptation and convey message of significant impact. City of Vancouver has similar maps and the impact patterns convey that information but just different. These maps would be helpful generally speaking.” [Participant #3, CoV Sustainability]  Modifications	for	the	RIPs	method	in	general Integrate adaptation options or narratives into scenarios Various participants saw the potential of the RIPs method as an approach to help evaluate different adaptation options. In general, some participants suggest that by integrating adaptation options into the future scenarios will create an explicit link between impacts and options, which would support options evaluation and decision-making.   228 “Can add in several adaptation scenarios – e.g. mix of dikes and other measures, model how these hold back water, and how different adaptation scenarios are effective at protecting against different scenarios” [Group discussion sticky note] “By adding adaptation options into the scenario this could help inform planning or decision-making” [Participant #2, CoV Planning]  Provide a user-guide Given that the RIPs can be used to communicate SLR risk to multiple types of stakeholders, it was suggested that it would be helpful to provide users with a handbook to describe: 1) how and when to use the RIPs, 2) data sources used, and 3) limitations, in lay terms to allow different user groups to use the RIPs properly.  Include cultural impact as another pillar of sustainability framework Although the RIP method allows the user to select the impact types to be included in the analysis, it is encouraged to include impact types from the three pillars of sustainability (economic, social, and environmental impacts). It was suggested that a fourth pillar in the concept should be added to include “cultural” impacts, such as impacts on cultural assets, to strengthen the impact narrative and political support for adaptation.  “I would include a fourth pillar of sustainability to the impact list - "culture". Could map using cultural assets. Establish a more robust impact narrative and political support” [Participant #13, CoV Planning]    229 Include other hazard types The ability of the RIPs to represent compounding impacts also brought about the suggestion that the RIP method can be used to identify areas that can potentially be affected by multiple types of hazards by including additional hazard types in the future scenarios.   Open source and adaptable Various participants indicated that the RIP method should be made open-source to allow easy access to the tool and made adaptable by different users (e.g., adding different impact types). “Suggest building on the impacts, make the application open source and adaptable.” [Participant #5, NRCan] Similarly, participants have suggested that the RIPs method could be built as a live model where users can provide planning data to support the analysis. This may help collate existing data that can support adaptation planning in one place and demonstrate its use through the RIP results.    230 6.4 Conclusion An expert workshop was hosted by the researcher to convene 15 experts in either SLR adaptation planning of the region or the CoV, or in the City's systems or infrastructure that may potentially be affected by SLR. The objective of the workshop was to present the RIPs method and RIPs from the CoV case study to the experts and assess their perspective on whether and how this type of information can support SLR adaptation planning. More specifically, through a written survey and group discussions, this research component assessed whether the RIPs method can support SLR adaptation in ten anticipated ways that broadly falls under two categories – 1) adaptation options development and 2) accessing resources and support to implement adaptation. The analysis has shown that more than 60% of participants agreed that the RIPs could support SLR adaptation options development through the five anticipated ways. But even more participants (more than 80%) agreed that the RIPs can help provide leverage to access resources and stakeholders support to implement adaptation actions. Nonetheless, it is worth noting that about 25% of the participants were neutral or disagreed with two out of the ten anticipated ways to use the RIPs – 1) refinement of adaptation options and 2) supporting long-range modification of options. Given that both of these are about modifying specific or existing options, the fewer participants in support of these applications may be attributable to the lack of direct linkage to adaptation options in the RIPs.   Further statistical analyses of the survey responses were conducted to investigate whether the participants’ view of the RIPs method’s utility depends on a) their prior knowledge about the CoV’s current SLR adaptation planning efforts and b) the CoV context of the RIPs in this case  231 study. The results indicated that the participants’ opinions about the RIPs method’s capability to support SLR adaptation are not dependent on either of those factors.  Besides assessing the potential utility of the RIPs method, the workshop has also revealed a number of useful insights. Firstly, the suggestions made by various participants on how to modify the case study have improved our understanding of some barriers that the users may face in adopting the information or tool. For example, the lack of smaller increments of SLR levels in the scenarios is a barrier to using the information for planning short-term adaptation actions. Suggested modifications to the RIPs method itself helps refine the framing of the method purpose and scope. For example, the suggested modification to integrate adaptation options into the future scenarios to facilitate for options evaluation reveals two different views about what is needed to support adaptation planning. One that sees the RIPs must have direct linkage to adaptation options in order to support planning, and the other view that considers RIPs to be useful for planning even without explicit link to options. This variation could be related to how different participants may require different type of information to help envision future actions. Since the workshop presentation did not specify when (i.e., which stage of planning) the RIPs should be used for adaptation planning, it is also possible that it is due to the difference between participants in terms of which stage of adaptation they anticipate to use the RIPs method. Some may envision themselves using the RIPs method during initial risk assessment stage while others may prefer to use them during the options evaluation stage. This highlights the need to clearly articulate the purpose of the method and how it is designed to be used, but also the potential of the RIPs method to serve more than one purpose if the appropriate modifications are made.    232 Secondly, the workshop result does not provide evidence that the RIPs method can support adaptation decision-making regarding adaptation options selection, which aligns well with the intended purpose of the method as described in Chapter 3. On the other hand, the positive responses to the anticipated ways the RIPs can support adaptation planning in the boarder sense indicates how the RIPs method can change the way decision-makers and planners would approach this issue – e.g., considering impacts beyond the usual suspects, considering more future scenarios and uncertainties, and collaborating with wider range of stakeholders, and placing priority on neighbourhood-level details and actions. By comparison, this goes somewhat beyond the original intended purpose of the RIPs method, which is to help account for uncertainties in earlier stages of planning by showing where and what can be affected under a wide range of futures. Specifically, to identify a pool of adaptation options that account for vulnerabilities and uncertainties specific to the given community so that it can be used in at a later stage in frameworks that further refines them for better robustness (e.g., Adaptation Pathway, Dynamic Adaptive Policy Pathways, Robust Decision-Making).  Although the expert workshop addressed the research question and provided additional insights, it does not facilitate for any quantifiable measures for the value of the information that the RIPs method can provide. Without a baseline for comparison, the workshop can only assess whether the RIPs can support adaptation in various ways but it cannot suggest whether it can provide support that is better than other existing methods. While the workshop outcomes show that all three aspects of the RIPs – 1) Spatial explicitness, 2) Robustness, 3) Diverse impact types – play important roles in the RIP method’s ability to support SLR adaptation, it is impossible to infer their relative importance and whether those individual aspects would be useful on their own.  233 Chapter 7: Conclusion and Discussion The research in this dissertation comprises of three sequential components – development of the RIPs method; application of the RIPs method at the City of Vancouver as a case study; and evaluation of the method's potential utility from the prospective users' point of view. This chapter will provide a summary of the key findings and significance of each of these components, their respective limitations, and proposes directions for future research.  7.1 Key findings and significance  Concluding remarks from each of the three components in Chapter 3, 4, 5, and 6 respectively are summarized in the following sections where the significance of their respective contributions is discussed in relation to their implications for adaptation planning and originality.  7.1.1 Development of the RIPs method The RIPs method was developed (Chapter 3) and demonstrated (Chapter 4 and 5) to address the first research question - How can the multi-dimensional nature of flood impacts and the deep uncertainty of long-term SLR impacts be presented to support robust adaptation planning? This method represents a new approach to assess and present a portfolio of potential SLR impacts under a large range of plausible futures, in a way that is spatially explicit and concise. The intention was to allow users to start considering the associated uncertainties, local context, and spatiality of SLR impacts in the early stage of planning (e.g., impact assessment stage) rather than the options evaluation stage. From the practical perspective, this is an attempt to ensure that  234 a wide range of plausible futures and local-contexts are incorporated into the planning from the beginning such that the options being considered and refined in the rest of the planning process are better starting points to become robust adaptation plans. From the theoretical perspective, the intended purpose of the RIPs method is to complement the burgeoning host of conceptual frameworks that aim to guide users in developing robust and/or adaptive adaptation plans. These frameworks often assume the initial set of preferred options are easily identifiable or already exist while in fact they are often unknown and users' have little guidance. Therefore, the RIPs method can help users identify the initial set of preferred options in a way that is mindful of uncertainties, local contexts, and spatial variations. However, it is not intended for developing and decision-making about detailed and specific adaptation options, which would require more rigorous performance analysis specific to the option.   The RIPs method is an original contribution in several ways. Its key innovation premise is its ability to allow users to process and understand an otherwise large and overwhelming volume of impact information. It takes advantage of the machine learning algorithm – Self-organizing maps (SOMs) – by using it to transform thousands of SLR impact maps, which represents the potential socio-economic impacts over a large range of futures, into a small number of impact maps that represent the predominant patterns of each impact types in this defined range of futures. Although SOMs is by no means a novel tool, the RIPs method as a framework represents a new approach to assess SLR impacts that draws the users’ attention to the associated uncertainties in a way that is understandable by multiple stakeholder types and requires users to take a holistic approach in selecting the type of impacts and impact drivers to consider. The way uncertainties are represented in the RIPs method also provides an alternative to presenting uncertainties in  235 terms of ensembles or quantiles. Sources of uncertainties are defined and represented by variations in how they are defined in the scenarios as opposed to being aggregated into a statistic. This may support more comprehensive understanding of the associated uncertainties and thus, develop strategies to manage them. The RIPs method also contributes to the repertoire of tools (Stephens, Bell, & Lawrence, 2017) that facilitates the implementation robust adaptation development frameworks (e.g., Robust Decision-Making; Dynamic Adaptive Policy Pathways, Adaptation Pathways). Lastly, it is also the first to apply the notion of robust impact to climate risk assessment and adaptation planning.   7.1.2 City of Vancouver case study The RIPs method was applied to assess the potential economic, social, and environmental impacts of SLR at the CoV as a case study (Chapter 4 and 5) that demonstrates the feasibility of the RIPs method. In addition, the case study has revealed the issues in implementing the method, as well as providing the basis to evaluate the RIPs method as an approach to support SLR adaptation planning. Hence, this research component helps to partially address both research questions. This application of the RIPs method assessed 14 types of SLR impacts across 336 plausible futures and identified 16 RIPs for each impact. The resulting RIPs represent the range of predominant modes of each impact that spans the defined range of plausible futures, where some modes are more robust than others. For example, one RIP may be robust across futures with 0-1m of SLR, while another is robust across futures of 3-4m of SLR. As each RIP shows the robust impact pattern as an impact map labeled with the total magnitude of the impact, the user can identify the areas in the City that can be affected by that given impact and at what magnitude across the range of futures that this RIP is robust in. This capability of the RIPs  236 facilitates for the users to use the RIPs in adaptation planning in a number of anticipated ways, which were examined in the third research component presented in Chapter 6.  The RIPs in this case study improves our understanding of the potential impacts of SLR at the CoV – a city yet to experience a major flood event. Although the CoV have conducted their own extensive assessments, this case study made original contributions by revealing new information on several aspects of SLR impacts in the CoV.  First, the set of assumptions developed with the support of BC Hydro staff represents a new approach to estimate areas that are potentially at risk of prolonged power outage associated with overland flooding in the CoV. The spatially explicit power outage conditions allowed the City to consider the indirect impacts on population, assets, and infrastructure of the City as a result of the flood-induced power outage. These indirect impacts resulting from the cascading effects have highlighted how some SLR impacts can affect areas far beyond the projected floodplains.   Second, this research developed a new composite index – Sewage Backup Damage Potential Index (SBDPI) - specifically for this case study with the support of staff at the CoV. Since existing measures to assess the risk of sewage backup are proprietary, this new index allows the City to start considering the sewage backup as an indirect SLR impact that has the potential of causing extensive damage in ground-related homes. Besides assessing an indirect impact of SLR that has not been assessed previously, this index is also an original contribution as it is the first sewage backup damage index to be published that accounts for risk factors in both sides of the property - the private property (e.g., sewage pump installed in home) as well as in the municipal  237 property (e.g., municipal drainage system). As a first attempt of such measure, it has much room for improvement, including the use of a more rigorous approach (e.g., a network model of the sewer system) to spatially estimate the differential surcharging potential of the sewer system. Nonetheless, the approach taken has the advantage of incorporating place-specific risk factors that are shaped by the City’s policies and in a manner that is more transparent than simply integrating into a complex network model. Although the index was developed specifically for the CoV, it provides a conceptual basis for further modification and improvements to better assess a rather complex phenomenon.  Third, the 16 RIPs collectively shows how the spatial distribution and magnitudes of different impacts changes with factors that are essentially beyond the government’s control (e.g., levels of SLR) as well as factors that are within their reach and provide room for adaptation (e.g., land-use regimes, population distribution, power outage extents, and vulnerability of buildings in the City). Therefore, the RIPs method also represents an impact assessment approach that not only facilitates for a more realistic impact assessment but also provides the room for users to see the opportunities for actions.  7.1.3 Evaluation of capability to support SLR adaptation This research component addresses the second research question - How can the Robust Impact Patterns (RIPs) method and results support SLR adaptation planning from the prospective users’ perspective? This research focuses on two aspects of adaptation planning: 1) developing adaptation options and 2) attaining resources and support for implementation. The RIPs identified in the CoV case study provide the basis to address this research question through an  238 evaluation process with a range of experts in SLR adaptation or infrastructure and systems at the CoV. Specifically, a group of 15 experts participated in a half-day experts workshop hosted by the researcher. The RIPs method and selected RIPs from the case study were presented and the experts were asked to participate in a written survey and group discussions to provide their opinion on whether the RIPs can support SLR adaptation planning in the ten anticipated ways (Chapter 6).   Overall, the responses were highly positive with the majority of experts agreeing or strongly agreeing that the RIPs can support SLR adaptation in those anticipated ways, and also suggesting additional ways to use the RIPs. However, a small number of experts have indicated concerns about two ways in which the RIPs can help users: 1) refine adaptation options and 2) support long-range planning of modifications to the adaptation options. The former concern was related to the high-level nature of the assessed impact variables, while the latter was associated with how short-term adaptation actions may subsequently change the nature of the impacts represented by the RIPs associated with higher levels of SLR. On the other hand, using the RIPs as a tool to communicate SLR risk to the potentially affected population and organization received a unanimously positive response. The survey outcomes were further analyzed to investigate whether their opinions were dependent on the CoV context of the case study or having prior knowledge about the SLR impact in the CoV. This was done by testing for statistically significant differences between responses of two pairs of subgroups (CoV staff vs. non-CoV staff, and experts involved in SLR adaptation planning vs. not involved), and the differences were not significant.   239 Besides the general level of agreement as shown by the survey results, additional insights and ways in which the RIPs can support adaptation, and specific practical examples were generated in the group discussions. For example, by showing the potential impacts that are robust across a range of plausible futures, which is wider than the range typically considered, can help prevent users from going down the "middle path" and makes it easier for users to consider a large number of plausible futures. While impact assessments typically focus on the adverse effects, participants also found the RIPs can help demonstrate the positive sides, such as making better investments and making the worst-case scenario seem less daunting. To help explain the unanimously positive response to using the RIPs to communicate SLR risks, participants attributed this to the spatial explicitness (i.e., map format) of the RIPs as it makes the impacts easily interpretable and relevant to a wide range of stakeholders in the CoV. This, in turn, facilitates engagement across a wider range of stakeholders and organizations, which can trigger new conversations, generate new adaptation ideas, and gain more leverage for requesting resources. This is a good example of how the discussion outcomes revealed the dynamics between the different ways in which the RIPs can support SLR adaptation planning.  The spatial-explicitness of the RIPs were often attributed for various ways in which RIPs can support SLR adaptation planning and the outcomes in the workshop provides additional evidence to support the implications of spatial differentiation in flood risk adaptation as suggested by Douglas et al. (2017). The implications supported by the workshop outcome include 1) increasing spatial flood hazard awareness; 2) reducing the differences between different groups of stakeholders to promote an integrated understanding of flood risk, and 3) improving flood risk perception and awareness through street-level (block level) spatial differentiation.  240  A key innovation of this research component is the design of an end-to-end process where the CoV staff and stakeholders were involved in the development process as well as the evaluation process to ensure the RIPs method can serve its intended purpose. Besides verifying the capability of the approach, in this case, this type of evaluation also helped identify any limitations of the RIPs method that has been overlooked and ways in which the method can be improved. But one of the most important workshop outcomes is that although the RIPs method has the potential to support SLR adaptation in various ways, the question of whether it can complement the conceptual frameworks for developing robust adaptation plans is perhaps more complicated. Rather than directly helping users identify an initial set of adaptation options that account for uncertainties and local contexts, the RIPs seem to better support the different steps or issues involved in identifying specific adaptation options. For example, helping users spatially and contextually prioritize their efforts and better communicate SLR risk to relevant parties contributes to the efforts in selecting suitable adaptation options. Therefore, besides addressing the second research question, this research component also made a contribution by demonstrating the value of this type of end-to-end process in academic studies when researchers often do not have the incentives to design their studies in this way. In this case, this process has generated insights that call for rethinking the assumption that the RIPs method can directly help users develop adaptation options and highlight the broader and additional uses of the RIPs method.    241 7.2 Limitations The research in this dissertation was constrained by various specific limitations that were described in the chapters of each respective component, but there are also some general limitations that are discussed in this section.   7.2.1 Conceptual limitations of the RIPs method Despite the successful application of the RIPs method at the CoV and positive responses from the experts, the RIPs method itself has several limitations as a long-term impact assessment approach. First, although one can backtrack to see what kind of scenarios are represented by each RIP to understand what kind of future can result in a certain robust pattern, it is difficult to determine the relative contribution of each input to the resulting impact pattern and clustering. For example, we can determine that a certain RIP x, that has a lower impact magnitude than RIP y and z, is representing impacts from future scenarios characterized by 2m of SLR, 1:500-year and 1:10,000-year storms, Compact land-use regime, and pessimistic power outage. However, it is not easy to determine whether RIP x's lower magnitude is due to its Compact land-use or other input variables. Furthermore, the machine learning part of the RIPs method's conceptual premise is somewhat complex. Therefore, some participants at the expert workshop have suggested that a user-guide should be developed to ensure appropriate and effective use of the method in the future.   Another limitation is demonstrated in the CoV case study where the inclusion of impact variables that do not vary with as many factors defining the scenarios as others can create artifacts in the resulting RIPs. Although such artifacts did not seem to have affected the RIPs of the other  242 impact variables, it can be a source of confusion for the users. Therefore, not all impact variables of interest to the user can be included in an application of the RIPs method. Users are recommended to include the impact variables that vary with fewer factors in a separate application of the RIPs method so that their scenarios can be defined with the appropriate number of factors. However, the RIPs from the two separate applications cannot be compared directly as they would be resulting from scenarios that are defined differently.  As described in Chapter 2, the robustness of adaptation plans (or options) can be improved through static robustness and dynamic robustness (van Drunen et al., 2009). Static robustness can be achieved by ensuring a plan can perform reasonably well in as many scenarios as possible, while dynamic robustness is achieved by increasing the flexibility of the plan such that it is adaptive to ongoing change. Since a notion of time is not explicitly incorporated into the RIP method, the RIPs can only promote identification of adaptation options that are more robust through static robustness. Similarly, but from a practical perspective, some workshop participants suggested that the RIPs may not be able to support long-range planning of modifications in adaptation options because the RIPs are static and do not account for how implemented adaptation options can subsequently change the magnitude and distribution of impacts. However, by design, it is possible for adaptation options to be integrated into the scenarios defined in Phase 1 of the RIPs method so that the resulting RIPs can account for the effects of the adaptation actions and provide insights to the robustness of different trade-offs and benefits of the adaptation options.   243 7.2.2 Challenges in applying the RIPs method As a major city that has yet to experience major flooding but is likely to become more vulnerable to SLR in the future, the CoV is a good case study community partner. The outcome of the RIPs method can improve their understanding of the potential impacts of coastal flooding in their city. Although the application of the RIPs method for the CoV was overall successful in demonstrating the feasibility of the RIPs method, it has also revealed issues pertaining to its ease of use. For example, the CoV has little to no empirical flood impact data or knowledge to help verify impact model outcomes. Therefore, in some cases (e.g., sewage backup damage potential), the verifications and model development relied upon expert knowledge elicitation, which can be time-consuming and can be a barrier to prospective users of the method. Furthermore, defining future scenarios and geospatial modeling the impacts over a large number of scenarios can be costly in terms of time and data. This contributed to one of the key challenges in the case study where there was a mismatch between the project timeline of the CoV and this research study. The RIPs could have been incorporated into the CoV’s adaptation planning process at an earlier stage if the impact modeling required less time. Less time would have been required if the appropriated data and models were already identified for each impact to be assessed. Therefore, this case study's application of the RIPs method can serve as a template for more efficient future applications.  7.2.3 Limitation of the evaluation process The expert workshop outcomes have shown that the RIPs method is a promising approach for supporting SLR adaptation planning, but it was not able to measure how much more informed would the users be if they have used the RIPs method because there is no baseline for  244 comparison. Conducting the evaluation process soon after the CoV’s own stakeholder engagement as part of their Coastal Flood Risk Assessment would have provided a good condition to use the CoV’s SLR impact information as the baseline for comparison. However, this was not possible since the amount of time between the CoV’s engagement workshop and the expert workshop of this case study was too long (i.e., more than 1 year) to serve as a reliable comparison. Another question that this evaluation process did not address is whether the RIPs method can address its theoretical purpose – help users identify a better initial set of adaptation options to serve as input for robust adaptation development frameworks. To examine this question may require comparing the robustness of adaptation plans resulting from one of the conceptual frameworks with and without using the RIPs method to identify the initial set of options. This study set up is currently beyond the scope of this study but forms a logical and interesting direction for future research.   7.3 Future research Based on the findings and limitations of this dissertation, this section suggests several future research directions, which are grouped into two categories: future applications of the RIPs method, and modification of the key purpose of the RIPs method.  7.3.1 Future applications A number of future applications of the RIPs method are potentially fruitful for assessing the validity of the RIPs method as an approach or framework for impact assessment and identifying an initial set of adaptation options that account for uncertainties, local contexts, and spatial context.   245  First, as discussed in the last limitation section, the ability of the RIPs method to address its intended purpose is still to be verified. This is possible by collaborating with users of one of the robust adaptation development frameworks (e.g., Dynamic Adaptation Policy Pathways, Scenario-neutral Adaptation) to use the RIPs method as one of the tools to implement the framework. Comparing the robustness of the resulting adaptation plan with and without using the RIPs method may provide some insights into whether the RIPs method can indeed complement those conceptual frameworks as intended.   Second, the RIPs method should also be applied to a community that has experienced major or regular flooding in the past to re-evaluate the utility of the method and see if it can still provide new and useful information (e.g., identify new types of stakeholders to engage, new ideas for adaptation options) to the users.   Lastly, application of the RIPs method at a regional scale can help examine whether the municipal level is the only scale at which to use the RIPs method. The experts workshop outcome suggested that the RIPs can be particularly useful for informing adaptation decisions for infrastructure or systems that have a long lifetime or where changes tend to take a long time to implement (e.g., drainage systems, building codes). Such infrastructure or systems tend to have jurisdictional or physical boundaries that span across multiple municipalities. Therefore, an application at the regional level, such as the region of Metro Vancouver, with a focus on SLR impacts on these specific infrastructures or systems can potentially demonstrate additional capabilities of the RIPs method.   246  7.3.2 Modification of the key purpose The outcome of the expert workshop suggested that the RIPs method has good potential to support the tasks involved in identifying adaptation options (e.g., prioritizing efforts, communicating risks to different types of stakeholders) but its capability of directly helping users to identify specific adaptation options may be questionable. Therefore, besides further analysis that focuses on the RIPs method's capability pertaining to options identification, the use of the RIPs method explicitly as a tool for other purposes - communicating SLR risk and adaptation options evaluation and pathway development – is potentially worthwhile.   To verify the unanimously positive response to how the RIPs method can be a useful tool for communicating SLR risk, applying the method explicitly as a communication tool can help examine its potential benefits. As suggested by workshop participants, these benefits can include: helping neighbourhoods identify what matters to them that can be affected and increase their support for adaptation actions; visualize impacts in a universally comprehensible way to bring together different groups of stakeholders to develop an integrated understand and generate new ideas and opportunities; and improve understanding of ownership to promote cost-sharing. Although a single, shared vision is not necessarily a pre-requisite for adaptation pathway development, engagement of a broad range of stakeholders can enrich participatory processes and strategy development (van der Voorn, Quist, Pahl-Wostl, & Haasnoot, 2017).   In principle, the scenarios developed in the first phase of the RIPs method can be defined to explicitly incorporate adaptation options, such as sea walls or coastal buyout, to provide users  247 with information about the potential effects and trade-offs of the adaptation options. This has the potential to serve as a tool to help implement decision support frameworks, such as Structured Decision-Making (Gregory et al., 2012), by providing decision-makers with visualization of how different alternatives can performance under different futures in order to identify ways to optimize the alternatives. Similarly, conducting this type of application for short-term adaptation options and subsequently for longer-term options generated from the findings of the previous application can potentially support the development of adaptation pathways. Furthermore, the static nature of the RIPs as applied in the case study can potentially be improved by integrating dynamic elements into the scenarios; for example, by incorporating reactive and concurrent adaptation actions, in addition to anticipatory actions. The incorporation of reactive adaptation actions can also be included as one way to start accounting for human behaviour, which is consistently an omitted element in flood risk assessments (Aerts et al., 2018).   As suggested by one of the workshop participants, the visual aspect of the RIPs also provides a good basis to develop narratives for different adaptation options. Using the RIPs to derive narratives can also further support communication of SLR impacts. A narrative can be more comprehensible to different types of stakeholders and help them envision what kind of consequences is associated with each adaptation option, but more importantly, the associated uncertainties. Uncertainties are often communicated quantitatively or statistically, in terms of probabilities, which can be difficult to understand and subsequently incorporate into one's decision-making. Visualizations of the impacts in the RIPs and the narratives that can be derived from them can contribute towards the growing interest in providing a good alternative to communicating uncertainties to non-experts in adaptation planning (e.g., Dessai et al., 2018).   248  By design, the main objective of the RIPs method is to help users to understand and account for the deep uncertainties in the long-term impacts of SLR on the outset of adaptation planning in a manner that is less overwhelming to avoid positioning the uncertainties as a barrier to implementing adaptation. It is ideal that the RIPs can help users generate an initial set of adaptation options that are cognizant of uncertainties and local contexts to effectively use one of the conceptual frameworks to further improve their robustness. However, even if it does not succeed in that purpose, this research has shown that the RIPs method can still make significant contributions by addressing the two key challenges in planning for SLR adaptation, as described by Lawrence, Bell, Blackett, Stephen, and Allan (2018). One is the need for the adaptation planning to account for ongoing change and deep uncertainties to prevent lock-in exposure by being mindful of how the uncertainty bounds are widening and compounded, especially in the upper-range SLR should the Antarctic ice sheet become unstable. Another is the need for adaptation planning to understand the needs and values of stakeholders that are likely to be affected, in order to support the design of context-specific options, as well as to facilitate adaptation cost-sharing that is anticipatory rather than in an after-the-fact and inevitable manner. The RIPs method can help address these challenges by providing a concise format for users to learn about the often-ignored uncertainties and their significance to their decisions, and also serve as a communication tool to engage a wider range of stakeholders to develop an integrated understanding and vision. Facilitating such learning in the early stage of adaptation planning would represent a key step towards the much-needed shift from the traditional predict-then-act approach that aims to find the optimal solution to scenario-based approaches that aim to develop  249 more robust adaptation plans for SLR impacts as one of the most concerning and yet deeply uncertain climate impacts.  250 References  Adger, W. N., Agrawala, S., Mirza, M. M. Q., Conde, C., O’Brien, K., Pulhin, J., et al. (2007). Assessment of  adaptation practices, options, constraints and capacity. In M. L. Parry, O. F. Canziani, J. P. Palutikof, P. J. van der Linden & C. E. Hanson (Eds.), Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution ofWorking Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (pp. 717-743). Cambridge, UK: Cambridge University Press. Adger, W. N., & Kelly, P. M. (1999). Social Vulnerability to Climate Change and the Architecture of Entitlements. Mitigation and Adaptation Strategies for Global Change, 4(3), 253-266. Aerts, J. C. J. H., Botzen, W. J., Clarke, K. C., Cutter, S. L., Hall, J. W., Merz, B., et al. (2018). Integrating human behaviour dynamics into flood disaster risk assessment. Nature Climate Change, 8(3), 193-199. Agrawal, A. (2008). The role of local institutions in adaptation to climate change. Papers of the Social Dimensions of Climate Change Workshop. Washington, DC: The World Bank. Ahern, M., Kovats, R. S., Wilkinson, P., Few, R., & Matthies, F. (2005). Global health impacts of floods: epidemiologic evidence. Epidemiologic reviews, 27(1), 36-46. Alderman, K., Turner, L. R., & Tong, S. (2012). Floods and human health: a systematic review. Environment international, 47, 37-47. Allen, M. R., & Frame, D. J. (2007). Call off the Quest. Science, 318(5850), 582-583. Aly, S., Tsuruta, N., & Taniguchi, R.-I. (2008). Face recognition under varying illumination using Mahalanobis self-organizing map. Artificial Life and Robotics, 13(1), 298-301. Astel, A., Tsakovski, S., Barbieri, P., & Simeonov, V. (2007). Comparison of self-organizing maps classification approach with cluster and principal components analysis for large environmental data sets. Water Research, 41(19), 4566-4578. Bakker, A. M. R., Louchard, D., & Keller, K. (2017). Sources and implications of deep uncertainties surrounding sea-level projections. Climatic Change, 140(3), 339-347. Banks, J. C., Camp, J. V., & Abkowitz, M. D. (2014). Adaptation planning for floods: a review of available tools. Natural Hazards, 70(2), 1327-1337. Ben-Haim, Y. (2006). Chapter 1 - Overview Info-Gap Decision Theory (Second Edition) (pp. 1-8). Oxford: Academic Press. Berkes, F., Colding, J., Folke, C., & Cambridge, B. (2003). Navigating social-ecological systems: building resilience for complexity and change. Cambridge;New York;: Cambridge University Press. Bhattacharya, N., Lamond, J., Proverbs, D., & Hammond, F. (2013). Development of conceptual framework for understanding vulnerability of commercial property values towards flooding. International Journal of Disaster Resilience in the Built Environment, 4(3), 334-351. Birkmann, J. (2006). Measuring Vulnerability to Natural Hazards – Towards Disaster Resilient Societies. Tokyo, Japan: United Nations University Press. British Columbia Ministry of Environment. (2013). Sea level rise adaptation primer - a toolkit to building adaptive capacity on Canada's South Coast.  251   Brody, S. D., Kang, J. E., & Bernhardt, S. (2010). Identifying factors influencing flood mitigation at the local level in Texas and Florida: the role of organizational capacity. Natural Hazards, 52(1), 167-184. Bubeck, P., de Moel, H., Bouwer, L. M., & Aerts, J. C. J. H. (2011). How reliable are projections of future flood damage? Natural Hazards and Earth System Science, 11(12), 3293-3306. Buchanan, M. K., Kopp, R. E., Oppenheimer, M., & Tebaldi, C. (2016). Allowances for evolving coastal flood risk under uncertain local sea-level rise. Climatic Change, 137(3), 347-362. Burton, I., Huq, S., Lim, B., Pilifosova, O., & Schipper, E. L. (2002). From impacts assessment to adaptation priorities: the shaping of adaptation policy. Climate Policy, 2(2-3), 145-159. Cammerer, H., Thieken, A. H., & Lammel, J. (2013). Adaptability and transferability of flood loss functions in residential areas. Natural Hazards And Earth System Sciences, 13(11), 3063-3081. Carrera, L., Standardi, G., Bosello, F., & Mysiak, J. (2015). Assessing direct and indirect economic impacts of a flood event through the integration of spatial and computable general equilibrium modelling. Environmental Modelling & Software, 63, 109-122. Carroll, B., Morbey, H., Balogh, R., & Araoz, G. (2009). Flooded homes, broken bonds, the meaning of home, psychological processes and their impact on psychological health in a disaster. Health & Place, 15(2), 540-547. Cash, D. W., Borck, J. C., & Patt, A. G. (2006). Countering the Loading-Dock Approach to Linking Science and Decision Making: Comparative Analysis of El Niño/Southern Oscillation (ENSO) Forecasting Systems. Science, Technology, & Human Values, 31(4), 465-494. Chang, S. E. (2003). Evaluating Disaster Mitigations: Methodology for Urban Infrastructure Systems. Natural Hazards Review, 4(4), 186-196. Chang, S. E., & Lotze, A. (2014). Infrastructure contribution to business disruption in earthquakes: Model and application to North Vancouver, Canada. Paper presented at the National Conference in Earthquake Engineering, Anchorage, AK. Chang, S. E., Pasion, C., Tatebe, K., & Ahmad, R. (2008). Linking Lifeline Infrastructure Performance and Community Disaster Resilience: Models and Multi-Stakeholder Processes Multidisciplinary Centre for Earthquake Engineering Research (MCEER) Technical Report, (07-004), 9-32 Chang, S. E., & Shinozuka, M. (2004). Measuring Improvements in the Disaster Resilience of Communities. Earthquake Spectra, 20(3), 739-755. Church, J. A., Clark, P. U., Cazenave, A., Gregory, J. M., Jevrejeva, S., Levermann, A., et al. (2013). Sea level change Climate Change 2013. The Physical Science Basis. Contribution of Working Group I to the 5th Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge: Intergovernmental Panel on Climate Change. City of Surrey. (2018). Surrey Coastal Flood Adaptation Strategy (CFAS). Primer Part I: Coastal Flooding in Surrey.: City of Surrey. City of Vancouver. (2008). Fixture Restriction guidelines for contractors and Homeowners. City of Vancouver. (2009). Open Data Catalogue -  Building footprints   2015, from http://data.vancouver.ca/datacatalogue/buildingFootprints.htm  252 Clark, P. U., Church, J. A., Gregory, J. M., & Payne, A. J. (2015). Recent Progress in Understanding and Projecting Regional and Global Mean Sea Level Change. Current Climate Change Reports, 1(4), 224-246. Consultants, H. (2007). Flood damage functions for EU member states. Final Report for JRC: Institute for Environment and Sustainability. Corum, J. (2016, 3 September 2016). A Sharp Increase in "Sunny Day" Flooding. The New York Times. Retrieved from https://www.nytimes.com/interactive/2016/09/04/science/global-warming-increases-nuisance-flooding.html Cuny, F. C. (1984). Disaster and Development. New York: Oxford University Press. Cutter, S. L., Barnes, L., Berry, M., Burton, C., Evans, E., Tate, E., et al. (2008). A place-based model for understanding community resilience to natural disasters. Global Environmental Change, 18(4), 598-606. Cutter, S. L., Mitchell, J. T., & Scott, M. S. (2000). Revealing the Vulnerability of People and Places: A Case Study of Georgetown County, South Carolina. Annals of the Association of American Geographers, 90(4), 713-737. Dannevig, H., & Aall, C. (2015). The regional level as boundary organization? An analysis of climate change adaptation governance in Norway. Environmental Science & Policy, 54, 168-175. Delpla, I., Jung, A. V., Baures, E., Clement, M., & Thomas, O. (2009). Impacts of climate change on surface water quality in relation to drinking water production. Environment International, 35(8), 1225-1233. Dessai, S., Bhave, A., Birch, C., Conway, D., Garcia-Carreras, L., Gosling, J. P., et al. (2018). Building narratives to characterize uncertainty  in regional climate change through expert   elicitation   Sustainability Research Institute: Working papers,  Dessai, S., & Hulme, M. (2001). Climatic Implications of Revised IPCC Emissions Scenarios, the Kyoto Protocol and Quantification of Uncertainties. Integrated Assessment, 2(3), 159-170. Dessai, S., Hulme, M., Lempert, R., & Pielke, R. J. (2009). Climate prediction: a limitation to adaptation? In W. N. Adger, I. Lorenzoni & K. L. O’Brien (Eds.), Adapting to Climate Change: Thresholds, Values, Governance (pp. 64-78). Cambridge, UK: Cambridge University Press. Dewar, J. A., & Cambridge, B. (2002). Assumption-based planning: a tool for reducing avoidable surprises. New York;Cambridge;: Cambridge University Press. Douglas, H., Wing, C., Victoria, B., David, F., Richard, M., Brett, F. S., et al. (2017). The Influence of Hazard Maps and Trust of Flood Controls on Coastal Flood Spatial Awareness and Risk Perception. Environment and Behavior, 0013916517748711. Dutton, A., Carlson, A. E., Long, A. J., Milne, G. A., Clark, P. U., DeConto, R., et al. (2015). Sea-level rise due to polar ice-sheet mass loss during past warm periods. [10.1126/science.aaa4019]. Science, 349(6244). Eun Ho, O., Deshmukh, A., & Hastak, M. (2010). Disaster impact analysis based on inter-relationship of critical infrastructure and associated industries. International Journal of Disaster Resilience in the Built Environment, 1(1), 25-49. Federal Emergency Management Agency. (2009). Multi-hazard Loss Estimation Methodology,  Flood Model,   Hazus®-MH Technical Manual     253 Feldman, D. L., & Ingram, H. M. (2009). Making Science Useful to Decision Makers: Climate Forecasts, Water Management, and Knowledge Networks. Weather, Climate, and Society, 1(1), 9-21. Flechas, J. (2017, 28 January 2017). Miami Beach to begin new $100 million flood prevention project in face of sea level rise. Miami Herald. Retrieved from http://www.miamiherald.com/news/local/community/miami-dade/miami-beach/article129284119.html Florida Oceans and Coastal Council. (2010). Climate Change and Sea-Level Rise in Florida: An Update of “The Effects of Climate Change on Florida’s Ocean and Coastal Resources.”. Tallahassee, Florida: Florida Oceans and Coastal Council. Fraser Basin Council. (2016). Lower Mainland Flood Management Strategy - Phase 1 Summary Report  Galbraith, H., Jones, R., Park, R., Clough, J., Herrod-Julius, S., Harrington, B., et al. (2002). Global Climate Change and Sea Level Rise: Potential Losses of Intertidal Habitat for Shorebirds. Waterbirds: The International Journal of Waterbird Biology, 25(2), 173-183. Gautum, K. P., & van der Hoek, E. E. (2003). Literature Study on Environmental Impact of Flood: GeoDelft. Gooré Bi, E., Monette, F., Gachon, P., Gaspéri, J., & Perrodin, Y. (2015). Quantitative and qualitative assessment of the impact of climate change on a combined sewer overflow and its receiving water body. Environmental Science and Pollution Research, 22(15), 11905-11921. Goosen, H., de Groot-Reichwein, M. A. M., Masselink, L., Koekoek, A., Swart, R., Bessembinder, J., et al. (2014). Climate Adaptation Services for the Netherlands: an operational approach to support spatial adaptation planning. Regional Environmental Change, 14(3), 1035-1048. Gregory, R., Failing, L., Harstone, M., Long, G., McDaniels, T., & Ohlson, D. (2012). Structured Decision Making : A Practical Guide to Environmental Management Choices. Hoboken, United Kingdom: John Wiley & Sons, Incorporated. Groves, D. G., Fischbach, J. R., Knopman, D., Johnson, D. R., & Giglio, K. (2014). Strengthening Coastal Planning: How Coastal Regions Could Benefit from Louisiana's Planning and Analysis Framework Retrieved from http://www.rand.org/pubs/research_reports/RR437.html Guyton, L., & Hurst, A. (2015). CoreLogic Launches Risk Analytics for Sewer Backup. 2016, from http://www.corelogic.com/about-us/news/corelogic-launches-risk-analytics-for-sewer-backup.aspx Haasnoot, M., Kwakkel, J. H., Walker, W. E., & ter Maat, J. (2013). Dynamic adaptive policy pathways: A method for crafting robust decisions for a deeply uncertain world. Global Environmental Change, 23(2), 485-498. Haasnoot, M., Middelkoop, H., Offermans, A., van Beek, E., & Deursen, W. P. A. (2012). Exploring pathways for sustainable water management in river deltas in a changing environment. Climatic Change, 115(3-4), 795-819. Haasnoot, M., Middelkoop, H., van Beek, E., & van Deursen, W. P. A. (2011). A method to develop sustainable water management strategies for an uncertain future. Sustainable Development, 19(6), 369-381. Hallegatte, S., Green, C., Nicholls, R. J., & Corfee-Morlot, J. (2013). Future flood losses in major coastal cities. [Letter]. Nature Clim. Change, 3(9), 802-806.  254 Hansen, J., Sato, M., Hearty, P., Ruedy, R., Kelley, M., Masson-Delmotte, V., et al. (2016). Ice melt, sea level rise and superstorms: evidence from paleoclimate data, climate modeling, and modern observations that 2 °C global warming could be dangerous. Atmos. Chem. Phys., 16(6), 3761-3812. Hanssen, G. S., Mydske, P. K., & Dahle, E. (2013). Multi-level coordination of climate change adaptation: by national hierarchical steering or by regional network governance? Local Environment, 18(8), 869. Heberger, M., Cooley, H., Herrera, P., Gleick, P. H., & Moore, E. (2011). Potential impacts of increased coastal flooding in California due to sea-level rise. Climatic Change, 109, 229-249. Henry, D. K., Cooke-Hull, S., Savukinas, J., Yu, F., Elo, N., & Van Arnum, B. (2013). Economic Impact of  Hurricane Sandy -   Potential Economic Activity Lost and Gained in   New Jersey and New York. Retrieved from http://www.esa.doc.gov/sites/default/files/sandyfinal101713.pdf. Hewitson, B. C., & Crane, R. G. (2002). Self-organizing maps: applications to synoptic climatology. Climate Research, 22(1), 13-26. Hill, P. R., Butler, R. W., Elner, R. W., Houser, C., Kirwan, M. L., Lambert, A., et al. (2013). Impacts of sea level rise on Roberts Bank (Fraser Delta, British Columbia). Horton, B. P., Rahmstorf, S., Engelhart, S. E., & Kemp, A. C. (2014). Expert assessment of sea-level rise by AD 2100 and AD 2300. Quaternary Science Reviews, 84, 1-6. ICPR. (2001). The Rhine Atlas (flood hazard and risk maps of the International River Basin District ‘Rhine’). 2017, from https://www.iksr.org/en/documentsarchive/rhine-atlas/ Insurance Bureau of Canada. (2014). Facts of the Property and Casualty Insurance Industry: Insurance Bureau of Canada. Ippoliti, D., & Zhou, X. (2012). A-GHSOM: An adaptive growing hierarchical self organizing map for network anomaly detection. Journal of Parallel and Distributed Computing, 72(12), 1576. Ivers, L. C., & Ryan, E. T. (2006). Infectious diseases of severe weather-related and flood-related natural disasters. Current Opinion in Infectious Diseases, 19(5), 408-414. Jackson, L. P., & Jevrejeva, S. (2016). A probabilistic approach to 21st century regional sea-level projections using RCP and High-end scenarios. Global and Planetary Change, 146, 179-189. Jäger, J., Rothman, D., Anastasi, C., Kartha, S., & van Notten, P. (2008). Training Module 6: Scenario development and analysis GEO Resource Book: A training manual on integrated environmental assessment and reporting. Canada and Nairobi: IISD and UNEP. Jevrejeva, S., Grinsted, A., & Moore, J. C. (2014). Upper limit for sea level projections by 2100. Environmental Research Letters, 9(10), 104008. Jones, R. (2010). The use of scenarios in adaptation planning: managing risks in simple to complex settings. VCCCAR Scenarios for Climate Adaptation Working Paper. Retrieved from http://www.vcccar.org.au/sites/default/files/vcccar/Jones%20scenarios%20presentation%2011-11-10.pdf Jones, R. N., Patwardhan, A., Cohen, S., Dessai, S., Lammel, A., Lempert, R., et al. (2013). Foundations for Decision Making. Climate Change 2014: Impacts, Adaptation and Vulnerability Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change.  255 Jongman, B., Kreibich, H., Apel, H., Barredo, J. I., Bates, P. D., Feyen, L., et al. (2012). Comparative flood damage model assessment: towards a European approach. Natural Hazards and Earth System Sciences, 12, 3733-3752. Kangas, J. (2014). SOM PAK: The self-organizing map program package. 2017, from https://www.researchgate.net/publication/2596686_SOM_PAK_The_self-organizing_map_program_package Kessler, R. C. (2007). Hurricane Katrina's impact on the care of survivors with chronic medical conditions. Journal of general internal medicine, 22(9), 1225-1230. Kirchhoff, C. J. (2010). Integrating Science and Policy: Climate Change Assessments and Water Resources Management. University of Michigan. Kirchhoff, C. J., Carmen Lemos, M., & Dessai, S. (2013). Actionable Knowledge for Environmental Decision Making: Broadening the Usability of Climate Science. Annual Review of Environment and Resources, 38(1), 393-414. Klein, R. J. T., Schipper, E. L. F., & Dessai, S. (2005). Integrating mitigation and adaptation into climate and development policy: three research questions. Environmental Science & Policy, 8(6), 579-588. Klenk, N. L., MacLellan, J. I., Reeder, K., & Flueraru, D. (2018). Local Knowledge Co-production, Emergent Climate Adaptation Publics and Regional Experimentalist Governance: An Institutional Design Case Study. In W. Leal Filho (Ed.), Climate Change Impacts and Adaptation Strategies for Coastal Communities (pp. 261-281). Cham: Springer International Publishing. Klijn, F., Baan, P., De Bruijn, K. M., & Kwadijk, J. (2007). Overstromingsrisico’s in Nederland in een veranderend klimaat: verwachtingen, schattingen en berekeningen voor het project Nederland Later, WL Delft Hydraulics Report: WL Delft Hydraulics. Kohonen, T. (1982). Self-organized formation of topologically correct feature maps. Biological Cybernetics, 43(1), 59-69. Kohonen, T. (2001). Self-organizing maps (Vol. 30.). New York: Springer. Kohonen, T. (2014). MATLAB Implementations and Applications of the Self-Organizing Map. Helsinki, Finland: Aalto University, School of Science. Kopp Robert, E., Horton Radley, M., Little Christopher, M., Mitrovica Jerry, X., Oppenheimer, M., Rasmussen, D. J., et al. (2014). Probabilistic 21st and 22nd century sea‐level projections at a global network of tide‐gauge sites. Earth's Future, 2(8), 383-406. Kreibich, H., Merz, B., Seifert, I., & Thieken, A. (2010). Application and validation of FLEMOcs - a flood-loss estimation model for the commercial sector. Hydrological Sciences Journal, 55(8), 1315-1324. Kreibich, H., Piroth, K., Seifert, I., Maiwald, H., Kunert, U., Schwarz, J., et al. (2009). Is flow velocity a significant parameter in flood damage modelling? Natural Hazards And Earth System Sciences, 9(5), 1679-1692. Kwadijk, J. C. J., Haasnoot, M., Mulder, J. P. M., Hoogvliet, M. M. C., Jeuken, A. B. M., van der Krogt, R. A. A., et al. (2010). Using adaptation tipping points to prepare for climate change and sea level rise: a case study in the Netherlands. Wiley Interdisciplinary Reviews-Climate Change, 1(5), 729-740. Kwakkel, J. H., Walker, W. E., & Marchau, V. (2010). Adaptive Airport Strategic Planning. European Journal Of Transport And Infrastructure Research, 10(3), 249-273. Laboratory of Computer and Information Science (CIS), H. U. o. T. (2015). SOM Toolbox 2.0. from http://www.cis.hut.fi/somtoolbox/  256 Lane, K., Charles-Guzman, K., Wheeler, K., Abid, Z., Graber, N., & Matte, T. (2013). Health effects of coastal storms and flooding in urban areas: a review and vulnerability assessment. Journal of environmental and public health, 2013, 913064-913064. Lawrence, J., Bell, R., Blackett, P., Stephens, S., & Allan, S. (2018). National guidance for adapting to coastal hazards and sea-level rise: Anticipating change, when and how to change pathway. Environmental Science & Policy, 82, 100-107. Lemos, M. C., Kirchhoff, C. J., & Ramprasad, V. (2012). Narrowing the climate information usability gap. [Review Article]. Nature Climate Change, 2, 789. Lempert, R. (2013). Scenarios that illuminate vulnerabilities and robust responses. Climatic Change, 117(4), 627-646. Lempert, R. J., Popper, S. W., Bankes, S. C., Publications, R., Reports, R., & Ebrary Academic Complete Subscription, C. (2003). Shaping the next one hundred years: new methods for quantitative, long-term policy analysis. Santa Monica, CA: Rand Corporation, The. Lempert, R. J., Popper, S. W., Groves, D. G., Kalra, N., Fischbach, J. R., Bankes, S. C., et al. (2013). Making Good Decisions Without Predictions Robust Decision Making for Planning Under Deep Uncertainty RAND Corporation Research Briefs, 6. Retrieved from http://www.rand.org/pubs/research_briefs/RB9701.html Lindgren, M., & Bandhold, H. (2009). Scenario planning: the link between future and strategy. Basingstoke, Hampshire: Palgrave Macmillan Ltd. List, D. (2005). Scenario Network Mapping: The Development of a Methodology for Social Enquiry. University of South Australia. Liu, Y., Weisberg, R. H., & Christopher, N. K. M. (2006). Performance evaluation of the self-organizing map for feature extraction. Journal of Geophysical Research - Oceans, 111(C5), C05018. Lotze, A. E. (2014). Addressing risk in research and practice : business earthquake vulnerability in North Vancouver. University of British Columbia Dissertation. Lyle, T. S., Long, G., & Beaudrie, C. (2015). City of Vancouver Coastal Flood Risk Assessment Phase II, Final Report: Compass Resource Management Ltd. Lyle, T. S., & Mills, T. (2016). Assessing coastal flood risk in a changing climate for the City of Vancouver. Canadian Water Resources Journal / Revue canadienne des ressources hydriques, 41(1-2), 343-352. Mangiameli, P., Chen, S. K., & West, D. (1996). A comparison of SOM neural network and hierarchical clustering methods. European Journal of Operational Research, 93(2), 402-417. Matrosov, E. S., Woods, A. M., & Harou, J. J. (2013). Robust Decision Making and Info-Gap Decision Theory for water resource system planning. Journal of Hydrology, 494(0), 43-58. McBean, G., Cooper, R., & Joakim, E. (2017). Coastal Cities at Risk (CCaR): Building Adaptive Capacity for Managing Climate Change in Coastal Megacities. Final Technical Report: International Development Research Centre. Measham, T. G., Preston, B. L., Smith, T. F., Brooke, C., Gorddard, R., Withycombe, G., et al. (2011). Adapting to climate change through local municipal planning: barriers and challenges. Mitigation and Adaptation Strategies for Global Change, 16(8), 889-909. Meehl, G. A., Stocker, T. F., Collins, W. D., Friedlingstein, P., Gaye, A. T., Gregory, J. M., et al. (2007). Global Climate Projections. Cambridge, UK: Intergovernmental Panel on Climate Change.  257 Merz, B., Kreibich, H., Sch