UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Determinants of West Nile Virus (WNV) incidence in select regions of Canada : an examination of climate… Roth, David Z 2017

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_february_roth_david.pdf [ 22.16MB ]
Metadata
JSON: 24-1.0362409.json
JSON-LD: 24-1.0362409-ld.json
RDF/XML (Pretty): 24-1.0362409-rdf.xml
RDF/JSON: 24-1.0362409-rdf.json
Turtle: 24-1.0362409-turtle.txt
N-Triples: 24-1.0362409-rdf-ntriples.txt
Original Record: 24-1.0362409-source.json
Full Text
24-1.0362409-fulltext.txt
Citation
24-1.0362409.ris

Full Text

Determinants of West Nile Virus (WNV)Incidence in Select Regions of CanadaAn examination of climate and ecological drivers inBritish Columbia and SaskatchewanbyDavid Z RothB.Sc., Queens University, 2001M.Sc., The University of Alberta, 2005A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Population and Public Health)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)December 2017c© David Z Roth 2017AbstractBackground: West Nile Virus (WNV) is a zoonotic arbovirus that has caused significant diseasein Canada, yet remains rare in British Columbia (BC). WNV is spread between avian reservoirsby Culex mosquitoes, and incidental transmission to humans can occur. Understanding tempo-ral and spatial changes in WNV risk is difficult because of the complexity of the transmissioncycle. Nonetheless, public health agencies require decision support in order to guide resourceallocation. Understanding the climate and ecological drivers of WNV disease can help with thedevelopment of such tools.Methods: Descriptive analyses were used to compare ecological and climate conditions in yearswith and without WNV transmission in BC from 2009 to 2015. Generalized linear mixed modelswere used to evaluate associations between WNV incidence in Saskatchewan (2003-2007) and:1) total irrigated landscape, and 2) avian community structure. Results were combined witha literature review to develop a WNV decision-support tool for BC which was evaluated via auser survey.Results: WNV activity in BC between 2009-2015 was limited to select locations during hot sum-mers when temperatures remained above key temperature thresholds during the amplificationperiod. In Saskatchewan, human incidence was positively associated with irrigated landscapesin 2003 but not 2007. Non-passerine species richness and abundance was positively associatedwith incidence from 2003-2007, and the dilution hypothesis was not supported. Heat was con-sistently associated with incidence, but other predictors had varying effects between years andmodels. The resulting decision support tool contained seven surveillance inputs and three haz-ard levels. Survey results supported the choice of surveillance inputs and the recommendedprevention measures.Discussion: Ecological complexity challenges quantitative risk prediction for WNV and ne-cessitates the use of simpler tools for public health decision support. Public health agenciesin low-incidence areas should invest public health resources in improving situational awarenessand preparedness instead of risk prediction. The decision support tool created here provides ageneral estimate of WNV amplification as a proxy of WNV hazard. Continued refinement ofthe tool as more is learned about the ecology of WNV would increase its utility.iiLay SummaryWest Nile Virus (WNV) has caused significant disease in Canada. WNV is spread from birds tohumans by mosquitoes. Rates of human WNV disease are affected by the interactions betweenweather, mosquitoes and birds, yet the relationship between these factors remains unclear inBritish Columbia (BC). Thus, I used weather, human cases of WNV illness, irrigation, mosquitoand bird abundance data from BC and Saskatchewan to better understand when and whereWNV makes people sick. WNV cases happened in hot years and locations, and irrigationand bird populations were linked to where the disease occurred. This work showed that therelationship between WNV, mosquitoes, birds and humans is very complex, making it difficultto predict how many people will get sick in the future. Study results were used to createa decision tool to help public health organizations prepare for future WNV activity and useprevention resources efficiently.iiiPreface• Components of the analysis of WNV in BC (2003-2015) was published in Emerging In-fectious Disease in 2010 (Volume 16(8): 1251-1258). Information from this publication isdistributed between Chapters 3 (Methods), Chapter 4 (Results), and Chapter 5 (Discus-sion). The analysis was greatly expanded for this thesis.• Dr. Bonnie Henry provided oversight and editing of the final manuscript. Sunny Mak,Allen Furnell, and Mieke Buller provided help with managing both the WNV surveillanceprogram and the resulting data. Dr. Muhummad Morshed ran the provincial WNV labtesting program during that time and contributed valuable information on lab interpre-tation and testing. David Roth was a member of the BC Centre for Disease ControlWNV surveillance team during this time and conducted all secondary analyses of datacollected through the surveillance program. David Roth wrote this article in its entirety,with reviews provided by co-authors.• The analysis of human WNV data from Saskatchewan was carried out under the approvalof the UBC Behavioural Research Ethics Board, certificate number H09-02248.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Literature Review: Environmental and Ecological Determinants of WNVTransmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.1 WNV Amplification Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.2 WNV Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.3 Reservoirs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.4 Vector Feeding Preference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122.5 Reservoir Community Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.6 Avian Immunity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.7 Role of Climate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.7.1 Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.7.2 Precipitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.8 Landscape Factors as Mediators of Disease . . . . . . . . . . . . . . . . . . . . . 212.8.1 Natural Landscapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222.8.2 Urbanization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232.8.3 Agriculture and Irrigation . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.9 Viral Introduction and Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262.10 Summary of Ecological and Climate Drivers of WNV Incidence . . . . . . . . . . 272.11 WNV Risk Prediction and Decision Support Models . . . . . . . . . . . . . . . . 28vTable of Contents3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.1 WNV Range Expansion and Activity in BC (2009-2015) . . . . . . . . . . . . . . 343.1.1 Broad Analytical Approach . . . . . . . . . . . . . . . . . . . . . . . . . . 343.1.2 Study Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.1.3 WNV Provincial Surveillance Data . . . . . . . . . . . . . . . . . . . . . . 373.1.4 Vector Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.1.5 Temperature Analysis and Degree-Day Calculations . . . . . . . . . . . . 403.2 Associations between WNV incidence and ecological conditions in Saskatchewan 413.2.1 Study Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413.2.2 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423.2.3 Modelling Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433.3 Decision Support Tool Framework . . . . . . . . . . . . . . . . . . . . . . . . . . 543.3.1 Hazard Categorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543.3.2 Decision Support Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . 563.3.3 Hazard Dependent Public Health Actions . . . . . . . . . . . . . . . . . . 573.3.4 Surveillance Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583.3.5 User Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 614 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 644.1 WNV Range Expansion into BC . . . . . . . . . . . . . . . . . . . . . . . . . . . 644.1.1 Overview of Provincial and Regional WNV Activity . . . . . . . . . . . . 644.1.2 Corvid Surveillance Trends . . . . . . . . . . . . . . . . . . . . . . . . . . 654.1.3 Vector Surveillance Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 694.1.4 Climatic Determinants of WNV Activity . . . . . . . . . . . . . . . . . . 734.1.5 Regional timing of temperature, precipitation, mosquito abundance inrelation to WNV spillover . . . . . . . . . . . . . . . . . . . . . . . . . . . 804.2 WNV incidence and ecological association in Saskatchewan, Canada . . . . . . . 884.2.1 Irrigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 884.2.2 Avian Community Structure . . . . . . . . . . . . . . . . . . . . . . . . . 1004.3 An Ecological Framework for WNV Decision Support in BC . . . . . . . . . . . 1094.3.1 Evaluation of Candidate Surveillance Inputs for WNV Decision Support . 1094.3.2 Draft Decision Support Tool . . . . . . . . . . . . . . . . . . . . . . . . . 1144.3.3 Hazard Dependent Public Health Action . . . . . . . . . . . . . . . . . . 1154.3.4 User Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1195 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1245.1 WNV Range Expansion and Activity in BC . . . . . . . . . . . . . . . . . . . . . 1245.2 Associations between WNV incidence and ecological conditions in Saskatchewan 1315.2.1 Irrigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1315.2.2 Avian Diversity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135viTable of Contents5.2.3 Issues Relevant to Both Analyses . . . . . . . . . . . . . . . . . . . . . . 1395.3 WNV Decision Support in BC . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1426 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155Appendices......................................................................................................................................181A Additional Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181A.1 Laboratory Case Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181A.1.1 Confirmed Case Diagnostic Test Criteria . . . . . . . . . . . . . . . . . . 181A.1.2 Probable Case Diagnostic Test Criteria . . . . . . . . . . . . . . . . . . . 181A.2 BBS Routes and Rural Municipalities . . . . . . . . . . . . . . . . . . . . . . . . 182A.3 Survey Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183B Additional results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186C WNV Prevention Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193viiList of Tables3.1 Measures of avian community structure including a description of each measure,the hypothesized association between the measure and WNV incidence as mea-sured in humans and/or mosquito vectors, and other studies previously using theselected measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483.2 Evaluation criteria for decision support tool. The individual or group doing theevaluating is also identified, as are broad methods of evaluation. . . . . . . . . . 634.1 Summary of BC WNV surveillance activities during the WNV seasons 2004-20014.1Note mosquito surveillance was reduced to the IHA only in 2013, andwas stopped completely in 2014. . . . . . . . . . . . . . . . . . . . . . . . . . . . 664.2 Summary of mosquito surveillance data for the IHA and the FHA between 2005and 2013. Data is shown for all traps within an HA, as well as for a subsetof stable traps that are consistent across years. Note mosquito surveillance wasreduced to the IHA only in 2013. . . . . . . . . . . . . . . . . . . . . . . . . . . . 684.3 Maximum Likelihood Estimates (MLE) of 2-week infection rates in Cx. tarsalismosquitoes, South Okanagan Valley, BC, 2009. Here the southern OkanaganValley represents all traps south of Penticton (Figure 3.1). . . . . . . . . . . . . . 724.4 Cumulative DDs between January 1 − August 31, 2003-2015∗ for select BC com-munities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 754.5 Summary table showing the mean (and SD) acres of irrigated land per RM foreach irrigation type in both 2003 and 2007. In addition, the table shows thetotal acres of irrigated land across all RMs in 2003 and 2007. Irrigation data wasprovided by the Irrigation Branch of the Saskatchewan Ministry of Agriculture(Branch, 2011). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 904.6 Effect estimates for a negative binomial model of the association between irriga-tion and WN disease incidence at the level of Rural Municipality in Saskatchewanfor both 2003 and 2007. Total irrigation includes surface irrigation, 200mm dutybacklog, miscellaneous back flood irrigation, pivot system irrigation, linear sys-tem irrigation, miscellaneous sprinkler irrigation, and remaining irrigation types.Results are presented for the full dataset, and a reduced dataset with the bottom10% of population size removed to evaluate potential effects of inflated incidence. 98viiiList of Tables4.7 Fixed effect parameter estimates of the negative binomial model of the associationbetween irrigation and WN disease incidence at the level of Rural Municipalityin Saskatchewan for both 2003 and 2007. Two forms of irrigation were evaluated:surface irrigation (left hand side of table) and sprinkler irrigation (right hand sideof table). Total surface irrigation included surface irrigation, 200mm duty back-log, and miscellaneous back flood irrigation. Total sprinkler irrigation includedpivot system irrigation, linear system irrigation, and miscellaneous sprinkler irri-gation. The category of "remaining irrigation types" was not included in eithersub category. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 994.8 Parameter estimates of the fixed effects of a poisson random effect model of therelationship between avian diversity and WNV incidence at the level of ruralmunicipality. AIC values quantify the fit of the model to the observed data. . . . 1064.9 Hazard-specific public health responses. Region reflects the suitable spatial unitfor a given surveillance input (HSDA, LHA, community, etc). Descriptions ofindividual actions can be found in Section 4.3.3. . . . . . . . . . . . . . . . . . . . 1174.10 Summary table showing the mean risk (and SD) for each scenario, as well as theproportion of respondents selecting a prevention measure identified in the survey.Scenarios are grouped according to the stage of the transmission cycle. . . . . . . 1204.11 Modified hazard-specific public health responses based on user feedback. Regionsreflects the suitable spatial unit for a given surveillance input (HSDA, LHA,community, etc). Descriptions of individual actions can be found in Section 4.3.3. 123ixList of Figures2.1 Changing levels of WNV in vector and reservoir communities for temperate lo-cations throughout the WNV season. The amplification cycle is split into threestages: suitable conditions, enzootic transmission, and spillover. Modified fromChilds (2007). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.2 Hypothesized causal pathway between observed patterns of WNV incidence andecological, environmental, and weather conditions based on the literature review. 293.1 Select cities (lower case) in BC, Canada, and regional HAs (upper case). Eachregional HA undertakes WNV surveillance under the guidance and recommenda-tions of the BC Centre for Disease Control. The Interior Health Authority andthe Fraser Health Authority are shaded darker grey because they had the mostintense WNV surveillance programs. The dashed oval encompasses the Okana-gan Valley, which has been the primary focal point of WNV activity in BC. WA,Washington, USA; ID, Idaho, USA; MT, Montana; AK, Alaska, USA . . . . . . . 363.2 Spatial linkage approach for Breeding Bird survey (BBS) routes and Saskatchewancensus data. Multiple RMs can be linked to a single BBS route. Shaded areasrepresent a single RM-BBS grouping. Census information and case counts areaggregated up to the level of these spatially aggregated units. . . . . . . . . . . . 463.3 Schematic outlining the linkages between data sources for the analysis of associ-ations between avian community structure and WNV incidence at the RM levelin Saskatchewan. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533.4 Framework for the creation of a WNV decision support tool for low incidencesettings, including: 1) identification of climatic or ecological inputs, 2) evalua-tion of surveillance inputs in relation to regional surveillance data sources, 3)creation/choice of a decision support tool, 4) hazard classification and 5) hazard-specific prevention recommendations. Sections of the framework are also linkedto the situational awareness (SA) levels as identified by Endsley (1995). . . . . . 553.5 A.) Proposed hazard classification scheme which is linked to stages of WNVamplification and transmission (Section 3.3.1). Color (yellow-to-red) reflect levelsof WNV hazard. B.) Proposed template decision tool structure. Surveillanceinputs define the rows and discrete blocks of time define the columns. . . . . . . 58xList of Figures4.1 WNV surveillance data from Washington State, 2002-2015. The vertical dottedlines represent years with confirmed WNV in British Columbia, Canada. . . . . 674.2 Nightly average catch for Cx. pipiens and Cx. tarsalis mosquitoes in the FHA(red) and in the IHA (blue) of BC, Canada, during 2005-2013, using all traps(solid line) and stable traps only (dashed line). Provincial vector surveillancedata were aggregated by week beginning January 1, and the dates provided usedSunday as the first day of each Epidemiological Week. . . . . . . . . . . . . . . . 714.3 Cumulative DDs (14.3◦base) over the preceding 14 days for select communitiesin British Columbia. The horizontal lines represents 109 and 150 cumulativeDDs, with the former estimated as number required to complete the extrinsicincubation period of WNV in Cx. tarsalis (Reisen et al., 2006b). Thicker linesare used to denote years with confirmed regional detection of WNV. . . . . . . . 764.4 Days above 22◦C (grey) and 26.7◦C (black) for select communities in BC (2007-2015). White blocks represent periods below 22◦C. Biological mechanisms behindtemperature thresholds described in Hartley et al. (2012). . . . . . . . . . . . . . 774.5 Minimum daily temperature for Osoyoos, Penticton and Kelowna BC, Canada inyears with (2009, 2010, 2013) and without (2012, 2014, 2015) WNV activity. Linesrepresent LOESS smoothers using a span parameter value of 0.4. The horizontaldotted line at 14.3◦C represents minimum estimated temperature required for Cx.tarsalis mosquito development and transmission (Reisen et al., 2006b). . . . . . . 784.6 Cumulative DDs (14.3◦base) over the preceding 14 days for years with (2009,2010, 2013) and without (2012, 2014, 2015) WNV detected by provincial surveil-lance. Communities are identified by color, with the thicker lines used to representcommunities from the Okanagan Valley. The horizontal line represents 109 cu-mulative DDs, which is estimated to be the number of DDs required to completethe extrinsic incubation period of WNV in Cx. tarsalis (Reisen et al., 2006b).Thicker lines are used to the denote the three communities found in the Okanaganvalley. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 794.7 Heat maps showing Z-scores for total monthly precipitation (mm) for Osoyoos,Penticton and Kelowna. Z-scores represent the number of standard deviationsoff the 20-year average value, with red representing values above the mean andblue representing values below the mean. Vertical dashed lines highlight July andAugust. Horizontal dashed lines represent 2009, 2010 and 2013, the years withdocumented endemic WNV transmission in BC. . . . . . . . . . . . . . . . . . . 81xiList of Figures4.8 Minimum daily temperatures, daily precipitation and average Cx. tarsalis trapcatch in 2009 WNV season in Osoyoos and Kelowna, BC. The vertical line is es-timated exposure of human case. Dashed lines in figure a) are LOESS smoothers(span=0.4) of daily minimum temperature for 2009 (red) and the average dailymin temperatures (blue) (1997-2015). The horizontal dotted line represents14.3◦C. For mosquito abundance, the red line represents the average trap catchfor all traps in the Osoyoos region, the green line represents the max trap catchfor an individual trap during that week, and the dashed blue line represents theoverall average trap catch for the entire IHA. . . . . . . . . . . . . . . . . . . . . 834.9 Minimum daily temperatures, daily precipitation and average Cx. tarsalis trapcatch in 2010 WNV season in Osoyoos and Kelowna, BC. The vertical line is es-timated exposure of human case. Dashed lines in figure a) are LOESS smoothers(span=0.4) of daily minimum temperature for 2010 (red) and the average dailymin temperatures (blue) (1997-2015). The horizontal dotted line represents14.3◦C. For mosquito abundance, the red line represents the average trap catchfor all traps in the Osoyoos region, the green line represents the max trap catchfor an individual trap during that week, and the dashed blue line represents theoverall average trap catch for the entire IHA. . . . . . . . . . . . . . . . . . . . . 844.10 Minimum daily temperatures, daily precipitation and average Cx. tarsalis trapcatch in 2011 WNV season in Osoyoos and Kelowna, BC. Dashed lines in figurea) are LOESS smoothers (span=0.4) of daily minimum temperature for 2011(red) and the average daily min temperatures (blue) (1997-2015). The horizontaldotted line represents 14.3◦C. For mosquito abundance, the red line representsthe average trap catch for all traps in the Osoyoos region, the green line representsthe max trap catch for an individual trap during that week, and the dashed blueline represents the overall average trap catch for the entire IHA. . . . . . . . . . 854.11 Minimum daily temperatures, daily precipitation and average Cx. tarsalis trapcatch in 2012 WNV season in Osoyoos and Kelowna, BC. Dashed lines in figurea) are LOESS smoothers (span=0.4) of daily minimum temperature for 2012(red) and the average daily min temperatures (blue) (1997-2015). The horizontaldotted line represents 14.3◦C. For mosquito abundance, the red line represent theaverage trap catch for all traps in the Osoyoos region, the green line representsthe max trap catch for an individual trap during that week, and the dashed blueline represents the overall average trap catch for the entire IHA. . . . . . . . . . 86xiiList of Figures4.12 Minimum daily temperatures, daily precipitation and average Cx. tarsalis trapcatch in 2013 WNV season in Osoyoos and Kelowna, BC. The vertical line is es-timated exposure of human case. Dashed lines in figure a) are LOESS smoothers(span=0.4) of daily minimum temperature for 2013 (red) and the average dailymin temperatures (blue) (1997-2015). The horizontal dotted line represents14.3◦C. For mosquito abundance, the red line represent the average trap catchfor all traps in the Osoyoos region, the green line represents the max trap catchfor an individual trap during that week, and the dashed blue line represents theoverall average trap catch for the entire IHA. . . . . . . . . . . . . . . . . . . . . 874.13 Total WNV cases (WN non-NS and WNNS) for province between 2003-2009 inCanada. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 894.14 Total WNV incidence (combined WN non-NS and WNNS) for Saskatchewan forthe outbreak years of 2003 and 2007. . . . . . . . . . . . . . . . . . . . . . . . . . 904.15 Spatial patterns of WNV incidence in the province of Saskatchewan in the out-break years of 2003 and 2007. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 914.16 Total acres of irrigated land in each RM in Saskatchewan in 2003 and 2007.Data was provided by the Irrigation Branch of the Saskatchewan Ministry ofAgriculture (Branch, 2011). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 924.17 WNV incidence in relation to ecoregions in the province of Saskatchewan, Canada.Cell shading represents WNV incidence, with darker shading representing higherincidence. The table provides the total number of cases in each ecoregion for 2003and 2007. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 934.18 Boxplots of distribution of key covariates in groups characterized by zero, low(<12/10000 population), medium (12-32/10,000 population), and high (>32.8/10,000population) WVN incidence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944.19 Monthly mean precipitation (mm) in RM's across Saskatchewan in 2003 and 2007.Horizontal line at 50mm is added for easy comparison. . . . . . . . . . . . . . . . 954.20 Cumulative DDs (14.3◦base) over the preceding 14 days for select communitiesin Saskatchewan. The horizontal line represents 109 cumulative DDs, which wasestimated to be the number of DDs required to complete the extrinsic incubationperiod of WNV in Cx. tarsalis (Reisen et al., 2006b). Large WNV outbreaksoccurred in Saskatchewan in 2003 (947 cases) and 2007 (1456 cases). . . . . . . . 964.21 Individual level avian community measures for RM-BBS groupings with no WNVincidence, WNV incidence between 0 and 10.7/10,000 population, and greaterthan 10.7/10,000 population. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1014.22 Species level avian community measures for RM-BBS groupings with no WNVincidence, WNV incidence between 0 and 10.7/10,000 population, and greaterthan 10.7/10,000 population. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102xiiiList of Figures4.23 Avian community structure measures for RM-BBS groupings with no WNV inci-dence, WNV incidence between 0 and 10.7/10,000 population, and greater than10.7/10,000 population. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1034.24 Feasibility evaluation for decision support inputs. Criteria include spatial andtemporal coverage, analytical requirements, and cost. . . . . . . . . . . . . . . . . 1144.25 Decision matrix characterizing the relationship between surveillance inputs, time,and hazard. Colors represent hazards categories (yellow=low, orange=medium,and red=high). The typical amplification and spillover periods are overlaid ontop of the decision matrix. See Section 3.5 for a description of how to use thedecision matrix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116A.1 Rural municipalities (RM) and breeding bird survey routes (black lines) run be-tween 2003-2007 in Saskatchewan. . . . . . . . . . . . . . . . . . . . . . . . . . . . 182B.1 Average trap catch between Epidemiological Week 25 and 26 for 2004-2013 forstable traps in the Fraser Health Authority (FHA) and the Interior Health Au-thority (IHA). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186B.2 Minimum daily temperature for Osoyoos BC, Canada in 2009, 2010 and 2013.The horizontal dotted line at 14.3◦C represents minimum estimated temperaturerequired for Cx. tarsalis mosquito development and transmission (Reisen et al.,2006b). The vertical dashed line in 2009 represents the estimated exposure datefor human cases and the collection date for the first positive mosquito pools. . . . 187B.3 Minimum daily temperature for Kelowna and Penticton BC, Canada in 2009, 2010and 2013. The horizontal dotted line at 14.3◦C represents minimum estimatedtemperature required for Cx. tarsalis mosquito development and transmission(Reisen et al., 2006b). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188B.4 Minimum daily temperature for Osoyoos from 2003-2008. The horizontal dot-ted line at 14.3◦C represents minimum estimated temperature required for Cx.tarsalis mosquito development and transmission (Reisen et al., 2006b). . . . . . . 189B.5 Model diagnostics for irrigation negative binomial models 2003. Numerical labelsreflect the RM number. Points are labeled to help identify stable outliers in bothoutbreaks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190B.6 Model diagnostics for irrigation negative binomial model 2007. Numerical labelsreflect the RM number. Points are labeled to help identify stable outliers in bothoutbreaks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191B.7 Model residuals from best fit avian model, that including non-passerine abun-dance, with and without 2 high leverage points removed. . . . . . . . . . . . . . . 192xivAcknowledgementsI would like to thank the following people and organizations for helping complete my degree.• Lorena Mota for dealing with my ever increasing stress levels. Your support gave me theconfidence I needed for the final push.• My parents, sister, and friends for providing the support I needed to stick this through tothe end.• Dr. Bonnie Henry for taking a chance on me and introduction into public health. Thereis absolutely no way I would have completed this degree without your guidance.• Dr. David Patrick, Dr. Craig Stephen, and Dr. Muhammad Morshed for providing keycontent expertise and guidance needed to navigate the challenges of academia.• The Environment, Public Health and Safety division of the Government of Saskatchewanfor data sharing and support. Special thanks to Phil Curry for all his guidance and helpwith understanding WNV in Saskatchewan.• The Bridge Program for providing an amazing interdisciplinary academic experience aswell as financial support. The professors and students involved in this program were someof the most amazing people I met during my time at UBC. This program provided anincredible introduction to the intersecting fields of public health, engineering and policy,while also providing a much needed support system.• The British Columbia Centre for Disease Control for providing me with an internshipopportunity that blossomed into an amazing work experience. Thank you to the ClinicalPrevention Services management for giving me the time needed to completing my thesis,and to my team members team for covering for me while I was away on academic leaves.• Canadian Institute for Health Research for providing me with a Canadian Graduate Schol-arship. This thesis would not have been possible without this financial support.• Dr. Michael Otterstatter and Dr. Rob Balshaw for their help with statistical analyses.• The Osoyoos Indian Band for help with vector surveillance in the Southern Okanagan.xvChapter 1IntroductionClinical medicine focuses on the human-centred proximal determinants of disease and has tradi-tionally failed to recognize how climate and ecology affect human health (Chivian & Bernstein,2004). In contrast, infectious disease epidemiologists are explicitly confronted with the im-portant role climate and ecology play in driving diseases like cholera (Lipp et al., 2002) andthose diseases spread by, or originating in, non-human animals (Wilson, 2001). Zoonotic dis-eases are caused by bacteria, viruses or parasites that originated in animals but also causeillness in humans (Karesh et al., 2012). Zoonotic diseases cause approximately one billioncases of human disease and more than a million deaths annually (Karesh et al., 2012), and areestimated to represent 58% of all human pathogens (Woolhouse & Gowtage-Sequeria, 2005).Zoonoses are disproportionately represented in novel diseases (Taylor et al., 2001), includingsevere acute respiratory syndrome (SARS), avian influenza, pandemic H1N1, Middle East Res-piratory Syndrome (MERS-CoV), Nipah, Zika and Ebola (Halpin et al., 2007; Karesh et al.,2012; Weissenböck et al., 2010).1Zoonoses can be acquired via direct contact, airborne, or oral routes, but approximately 40%of them are transmitted to humans via arthropod vectors like ticks and mosquitoes (KreuderJohnson et al., 2015; Loh et al., 2015). Vector-borne zoonoses (VBZs) include malaria, denguefever, lyme disease, and Zika (Fauci & Morens, 2016; Gubler, 2009). VBZs can be maintainedin a vector-human transmission cycle, or in a vector-reservoir-human cycle whereby viral ampli-fication in non-human hosts (enzootic transmission) leads to incidental dead-end transmissionto humans. The relationship between weather or regional ecology and VBZ incidence is morecomplex than the relationship between weather or regional ecology in human-to-human spreaddiseases because non-human species play important roles as both vectors and reservoirs: condi-tions impacting the abundance of, or interaction between, vectors, reservoirs, and non-reservoirhosts therefore affect human population health (Eisenberg et al., 2007; Loh et al., 2015; Patzet al., 2004).Arthropod-borne viruses, or arboviruses, are a specific subgroup of viral vector-borne dis-eases (VBDs) that include chikungunya, Japanese encephalitis, St. Louis encephalitis (SLE),dengue virus, yellow fever, and many other diseases (Gubler, 1998). Arboviruses are transmit-ted by blood feeding arthropods like mosquitoes, ticks, or biting flies, and have a nearly globaldistribution (Weaver & Reisen, 2010). Some arboviruses with zoonotic origins like dengue or1Despite the media attention given to novel zoonotic infection, the majority of zoonotic mortality and mor-bidity results from traditional diseases like Tuberculosis (Lozano et al., 2012)1Chapter 1. Introductionchikungunya have evolved from enzootic amplification to exist solely in human populationsand cause large outbreaks in dense cities, with transmission driven by the urban vector Aedesaegypti (Weaver & Reisen, 2010); Dengue alone causes 50-100 million cases annually (Endyet al., 2010). Other arboviruses like Japanese Encephalitis (JE) cause 30-50,000 cases annually(Solomon, 2006) and are spread between Culex mosquitoes, birds and pigs (Weaver & Reisen,2010). SLE was historically the most common arbovirus in the US where it was transmittedbetween Culex mosquitoes and birds, with humans acting as dead-end hosts. Although SLE iscurrently rare (20-50 cases annually in the US) (Davis & Peterson, 2008), it has caused largeoutbreaks, such as the 2800 cases that were reported across the 31 US states in 1975 (Calisher,1994).In Canada, VBDs and arboviruses have historically caused limited disease because coldweather limits the number of poikilothermic arthropod vectors (Artsob, 2000). Mosquito-bornearboviruses like SLE, Western Equine Encephalitis (WEE), Eastern Equine Encephalitis (EEE),and snowshoe hare virus are detected sporadically yet remain rare (Calisher, 1994; Kettylset al., 1972; McClean et al., 1969). Tick- borne disease like Lyme disease, relapsing fever,tularemia, bartonella, plague, Colorado tick fever and babesiosis have also been detected inCanada (Artsob, 2000; Banerjee et al., 1998), although they have historically caused limiteddisease2. However, the impact of VBDs in Canada increased in the early 2000s after the NorthAmerican introduction of West Nile Virus (WNV) to New York City in 1999 (Anonymous, 1999;Nash et al., 2001). By 2016, 5414 clinical WNV cases had been diagnosed in Canada (PublicHealth Agency Canada, 2009). The large number of cases and rapid spread of WNV in Canadaprovides a unique opportunity to study how ecological and climatic factors promote or preventthe transmission of arboviruses and VBZs in both Canada and BC. Gaining an understandingof the complex role that climate and non-human species play in the transmission of WNV isrequired to improve regional disease prevention and control efforts.WNV was first isolated in Uganda in 1937, with subsequent sporadic outbreaks of mildfebrile illness documented in Africa, Europe, Australia and Asia (Hayes, 2001; Kramer et al.,2007). Observable increases in the severity of WNV illness appeared in Mediterranean outbreaksduring 1990s, with severe neurological disease occurring in some infected individuals (Krameret al., 2007). Increased disease severity was also seen in subsequent outbreaks in Romania (Hanet al., 1999), Russia (Dinu et al., 2015; Platonov, 2001), Israel (Paz, 2006; Weinberger et al.,2001), Italy (Barzon et al., 2013) and Greece (Pervanidou et al., 2014). The virus first appearedin North America in New York City in 1999 when 59 patients were hospitalized; 63% of thesesuffered from encephalitis and 7 died of the infection (Anonymous, 1999; Hayes, 2001; Komaret al., 2003; Mostashari et al., 2001; Nash et al., 2001). Prior to the emergence of human cases,abnormal avian mortality was reported in-and-around New York zoos (Nash et al., 2001). It wasinitially believed that this outbreak was caused by SLE (Calisher, 2000), but dead bird autopsies2The incidence of Lyme disease in Canada has increased 4-fold in the last 10-years, making it currently themost important VBZ in Canada (Public Health Agency of Canada, 2016)2Chapter 1. Introductionquickly confirmed the presence of WNV, an arbovirus not previously detected in North America(Nash et al., 2001).Ominously, the increased neurological disease and avian mortality seen in the outbreaks ofEurope and Russia also occurred in North America (Blitvich, 2008).3Although 80% of infectedhumans showed no symptoms, 20% experienced a mild form of the disease which is termedWest Nile Virus non-Neurological Syndrome (WN non-NS) in Canada (Public Health Agencyof Canada, 2008) and West Nile fever in the US (Mostashari et al., 2001). WN non-NS ischaracterized by at least 2 of the following: fever, myalgia, arthralgia, headache, fatigue, lym-phadenopathy, or maculopapular rash (Public Health Agency of Canada, 2008). Approximately1% of those infected with WNV developed West Nile Neurological Syndrome (WNNS) whichis characterized by the development of encephalitis, viral meningitis, acute flaccid paralysis,movement disorders, Parkinson-like conditions, or other neurological conditions (Public HealthAgency of Canada, 2008). Risk of neuroinvasive disease increases with age (Jean et al., 2007;O'Leary et al., 2004) and can be as high as 1 in 54 infections amongst those over the age of 65(1998-2008) (Carson et al., 2012); the mortality rate in this group is approximately 10% (Pe-tersen et al., 2012a). Risk factors for the development of neurological disease compared to WNnon-NS include alcohol use, diabetes hypertension, being male, and being older than 64 yearsof age (Bode et al., 2006; Jean et al., 2007; O'Leary et al., 2004). It was originally believed thatthe majority of patients with WN non-NS recovered fully. In Canada, most of those infectedwith WNV returned to normal health within 1-year of infection, although recovery was longerfor those with pre-existing conditions or those aicted with the severe form of the disease (Loebet al., 2008). However, even the mild form of the disease can have long-term health consequences(12-18 months), with long-term sequelae including muscle weakness, muscle ache, and fatigue,as well as cognitive difficulties like a lack of concentration (Cook et al., 2010; Klee et al., 2004;Sejvar et al., 2008)WNV quickly spread from New York City after the initial outbreak in 1999, a pattern thatcontrasted with the epidemiology of the disease in Africa and Europe where epidemics weretypically followed by periods of inactivity (Chancey et al., 2015). WNV introduction in New Yorkwas followed by confirmed disease incidence in the southeast US in the following years (Florida,Georgia, Louisiana, Maryland) (n=66 total US cases), and in 2002 the expanding wave of WNVexploded westward into the centre of the US (n=4156), and cases were sporadically reportedeven in California. WNV had become the most widespread arbovirus in the world only 5 yearsafter arriving in North America (Kramer et al., 2007; Petersen & Hayes, 2004). This continentalrange expansion was followed by the largest outbreak of WNV in the summer of 2003 with 98623The observed increase in disease severity may have resulted from infection in immunologically naive popu-lations (Briese & Bernard, 2005). However, similar patterns in Europe and concurrent rises in avian mortalitysuggest the possibility of increasing virulence (Blitvich, 2008). This theory is supported by the emergence ofa novel WNV strain in 2002 (WN02) which soon out-competed the original NY99 strain (Moudy et al., 2007;Snapinn et al., 2007). Viral evolution has continued in North America, as ≥3 genetic strains of WN02 are nowco-circulating in select regions of Texas (although no phenotypic differences have yet been identified) (McMullenet al., 2011).3Chapter 1. Introductionreported cases (29% neuroinvasive) (Centers for Disease Control and Prevention, 2016, hereafterCDC, 2016). During this outbreak, Colorado (n=2947), Nebraska (n=1942), and South Dakota(n=1039) were most strongly impacted, with New York and New Jersey only having 71 and34 reported cases, respectively. Over the next decade (1999-2010), the virus became focusedin the Great Plains. The highest cumulative incidence occurred in South Dakota, Wyoming,North Dakota, Nebraska, and Colorado (Petersen et al., 2012a), with additional high-incidencepockets occurring in Louisiana and Mississippi (Lindsey et al., 2008) (CDC, 2016). Californiaalso experienced significant WNV activity during the same time frame (Barber et al., 2010;Kwan et al., 2010). After this initial spread, viral activity slowed, with only small outbreaksand sporadic cases occurring in the US between 2004-2011 (Petersen et al., 2012b); fewer than1400 total cases were seen between 2008 and 2011. However, a resurgence of WNV infectionoccurred in 2012 with 5674 (51% neuroinvasive) reported cases - the most since 2003 (CDC,2016). The largest regional focal point in 2012 was Texas, with 1868 cases in the Dallas areaalone (Murray et al., 2013). Additional regional outbreaks were seen in California (n=479),Louisiana (n=335), Michigan (n=202), Mississippi (n=247), Oklahoma (n=191), and SouthDakota (n=203) (CDC, 2016).Patterns of WNV spread in Canada, the northernmost range of WNV in North America,mirror that seen in the US. The virus was first detected in Canada in 2001, with confirmedcases occurring in birds and mosquitoes in Ontario and Quebec. The following year, 394 humancases occurred in Ontario and 20 in Quebec (Public Health Agency Canada, 2009), and thepublic health system was generally still unprepared for WNV (Aiken, 2003). The virus quicklyspread westward into the Prairie Provinces, mirroring the westward movement seen in the US(Petersen & Hayes, 2004). In 2003, Canada experienced its first large scale WNV outbreak(n=1481), with 947 confirmed cases in Saskatchewan, 63 of which were WNNS (SaskatchewanMinistry of Health, 2016); 144 in Manitoba (35 WNNS) (Manitoba Health, Seniors and ActiveLiving, 2017), and; 275 (48 WNNS) in Alberta (Alberta Health, 2016). In 2003, 9.8% of 501individuals participating in a sero-survey in Saskatchewan tested positive for WNV, with anattack rate of 93/100,000 population in the province at that time (Schellenberg et al., 2006). InAlberta in 2003, rates of infection were as high as 46/1000 population in the southern Palliserregion (Ivan et al., 2005). A second major outbreak occurred in 2007 (n=2215), again centredin Alberta, Manitoba, and Saskatchewan. In 2007, 1,456 (113 WNNS) cases were confirmed inSaskatchewan (Saskatchewan Ministry of Health, 2016); 587 (72 WNNS) in Manitoba (ManitobaHealth, Seniors and Active Living, 2017) and; 318 (21 WNNS) in Alberta (Alberta Health, 2016).WNV incidence in Canada decreased between 2007 and 2012, with fewer than 115 cases per year,with the exception of 2012 (n=428). Between 2002 and 2014, 5,454 human cases (symptomaticand asymptomatic) were reported to the Public Health Agency of Canada, with an estimated18,000-27,000 total infections occurring in this period (Zheng et al., 2014).Despite the presence and rapid spread of large WNV outbreaks in the US and Canada, not allareas were affected equally. Incidence was low for Oregon andWashington State in the US Pacific4Chapter 1. IntroductionNorthwest, at 0.01-0.24/100,000 population from 1990-2015 (Lindsey et al., 2008) (CDC, 2016).Similarly, no local WNV transmission was detected within British Columbia (BC) from 2002-2008, despite the disease having been transmitted in Alberta since 2003 (Public Health AgencyCanada, 2009). In response to neighbouring WNV activity, the BC Centre for Disease Control(BCCDC) and provincial health partners developed an extensive multi-jurisdictional surveillanceprogram in 2003 to determine risk for WNV and to prepare for its occurrence in BC. However, noWNV was detected provincially until 2009 when transmission was confirmed along the southernborder of the province (Morshed et al., 2011). WNV had been causing illness and death inCanada for almost six years prior to this expansion in range and the public health communitywas beginning to understand the complex and unique epidemiology of the disease in NorthAmerica. Researchers clearly understood which mosquito species were involved in transmission(Sardelis et al., 2001; Turell et al., 2001b) and recognized the role of birds in amplifying the virus(Komar et al., 2003; McLean et al., 2001). Yet, many unanswered questions remained, includingwhy some areas experienced significant transmission and human outbreaks while other areasremained disease-free. There were clear links between WNV outbreaks and weather (Huhn et al.,2003; Reisen et al., 2004), but the intensity, timing, and sequence of weather events required fortransmission remained unclear. Furthermore, the relative importance of climate and ecologicalconditions, and the interaction between these conditions, remain difficult to discern to this day.It was during this period that this thesis was designed. Despite several severe outbreaksin Canada, no WNV had yet been detected in BC. Questions were raised as to the value ofmaintaining an intensive surveillance program for a disease that was causing such limited regionalhealth impacts. However, halting WNV surveillance was not possible without understandingthe disease's potential for regional activity, and public health organizations desired estimatesof the regional likelihood of transmission. Extensive research had been conducted for specificregions of the US, yet not BC and Washington State. This regional research gap was problematicfor public health institutions because results from one area may not be applicable to regionswith differing ecological and environmental conditions. In order to rationalize the continuationor cessation of BCCDC-led WNV surveillance programs, the public health system needed toimprove its understanding of the specific combination of environmental and ecological conditionsthat would facilitate or limit the introduction or establishment of the virus in BC, and whatsuch conditions implied for future WNV activity in the province.The research comprising this thesis was undertaken to fill this specific knowledge gap. Theoverarching objective of this work was to improve the situational awareness of public healthorganizations in BC so that defensible decisions relating to WNV management and resourceallocation could be made on the basis of uncertain and incomplete information. To do this,I evaluated WNV predictors identified as being important to transmission in other locationsto determine if they also impacted transmission in BC, and if they could be useful in guidingregional decision-making. This objective was undertaken under the hypothesis that a certainset of predictors could be consistently associated with WNV incidence in BC or Saskatchewan,5Chapter 1. Introductionand that these predictors could be identified using epidemiological methods.In the first chapter of this thesis, the WNV literature on ecological and climatic drivers ofdisease was reviewed with a specific focus on North America. A literature summary is providedfor 1) the ecological conditions that facilitate WNV amplification and how those factors varybetween regions, 2) regional contextual factors like climate and landscape that impact variationin the incidence of WNV, 3) mechanisms of WNV spread and their implications for viral rangeexpansions, and 4) approaches to WNV risk prediction and decision-making in North America.Two broad statistical analyses were then carried out. First, WNV surveillance data fromBC (2009 to 2015) was evaluated in relation to ecological and environmental thresholds asso-ciated with disease in other locations in order to elucidate the conditions that facilitated therange expansion into the province and limited subsequent activity. Reference thresholds wereidentified from the literature with a specific emphasis placed on those with a biological basis,as such thresholds were felt to be more transferable across locations than were those derivedon correlation alone. This evaluation took advantage of 10-years of surveillance data collectedby the BCCDC and its Health Authority (HA) partners and included vector abundance andinfection rates, dead corvid surveillance, and passive reporting for both humans and horse cases.Historical temperature and precipitation data was used to characterize years with and withoutconfirmed WNV activity in order to identify the factors driving inter-year variability in regionaltransmission.The absence of wide-spread WNV transmission in BC limited both the scope of possibleanalyses and the number of disease determinants that could be examined. The second analysiswas done using WNV case data from Saskatchewan during the 2003-2007 period in order toevaluate two ecological factors hypothesized to affect patterns of WNV incidence: irrigation andavian reservoir community structure. Gaining a better understanding of the relative explanatorypower of these ecological factors in Saskatchewan relative to climate variables may improve ourunderstanding of WNV risk in BC despite the ecological dissimilarity between provinces if therelative importance of determinants remains consistent. At the time this thesis was developed,no published work had evaluated the link between landscape features andWNV in Saskatchewan.WNV activity has been focused in agricultural regions in North America, and irrigation washypothesized to be positively associated with mosquito-borne diseases (MBD) by providingbreeding sites for developing vectors (Harrus & Baneth, 2005; Molyneux, 2003; Organization,1997). With this hypothesis in mind, generalized linear models (GLM) were used to examinethe association between agricultural irrigation in Saskatchewan and WNV incidence during itsmajor 2003 and 2007 outbreaks. The independent effects of sprinkler and surface irrigation wereevaluated in relation to the hypothesis that surface irrigation will be more strongly associatedwith disease because it results in a greater amount of standing water.WNV incidence may also be affected by reservoir biodiversity and avian community struc-ture. Key empirical studies suggested negative associations between avian diversity and WNVincidence (Ezenwa et al., 2006; Swaddle & Calos, 2008). However, the consistency and general-6Chapter 1. Introductionizability of this pattern remained unclear (Loss et al., 2009b; Randolph & Dobson, 2012). Thisinconsistency emphasized the importance of evaluating avian community structure and WNVincidence across regions with unique ecologies, especially along the northern extent of the viralrange. The second part of this statistical analysis used data from the North American BreedingBird Survey (BBS) to evaluate associations between avian community structure and WNV inci-dence in Saskatchewan. Passerines, or `perching birds' (Campbell & Branch, 1990), are typicallythought to be more competent WNV reservoirs than are non-passerines. It was hypothesizedthat WNV incidence would be negatively associated with broad measures of avian biodiversity,and negatively associated with the number of non-passerine avian species. Revealing relation-ships between avian community structure and incidence could inform our understanding of WNVtransmission in BC.The final component of this thesis aimed to improve the situational awareness of WNV inBC by synthesizing the study results and the literature review to create a decision supporttool in order to help with WNV decision-making in the province. Provincial public healthorganizations specifically required tools that could provide guidance on when and where tofocus resources for WNV surveillance, prevention and messaging, since maintaining expensivesurveillance and prevention programs in years with low risk was an inefficient use of resources.The thesis results informed both the selection of surveillance inputs for hazard tracking andtheir thresholds. The structure of the tool was similar to those used in other regions, but itwas specifically parameterized to fit the unique features of WNV transmission in BC. A usersurvey was conducted to determine whether the ecological framework on which the tool wasbased conformed with how regional WNV experts viewed regional hazard. Finally, a discussionwas presented as to the generalizability of the tool to other VBD threats that could impact BCin the future, given global climate change.In summary, this research aimed to evaluate the following specific research questions as theyrelate to two areas of Canada.1. What were the climatic, ecological and landscape factors that explained: 1) the delayedpresence of WNV in BC, 2) the expansion of the virus into BC in 2009, and 3) the lowlevels of activity since 2009?2. How was the total amount of regional irrigation associated with WNV incidence at therural municipality (RM) level in Saskatchewan in 2003 and 2007? More specifically, diddifferent irrigation methods have differential associations with disease incidence?3. Was WNV incidence at the rural municipality level associated with avian community struc-ture in Saskatchewan from 2003-2007? More specifically, which measures of communitystructure were most strongly associated with WNV incidence?4. Can our understanding of the complex interplay of climate and ecological factors be syn-thesized into a tool to aid practical decision support for WNV prevention in low incidencesettings like BC?7Chapter 2Literature Review: Environmental andEcological Determinants of WNVTransmission2.1 WNV Amplification CycleWNV is a member of the Flaviviridae family, and is closely related to Japanese and St. LouisEncephalitis (SLE) (Mackenzie et al., 2004). Like SLE, WNV is maintained in an enzootic4cyclebetween avian reservoirs and mosquito vectors (Campbell et al., 2002). WNV is an ecologicalgeneralist that can replicate in a wide variety of hosts, having been identified in 65 mosquitospecies and 326 bird species in North America alone as of 2013 (Petersen et al., 2013). Trans-mission of WNV requires that mosquitoes take at least two blood meals: the first to contractthe virus from an infected reservoir, and the second for transmission to a new avian host. WNVamplification is the process whereby a virus is transmitted horizontally between vectors andreservoir hosts (Day et al., 2015), with infection rates subsequently increasing in both vectorsand reservoirs over the transmission season. In temperate climates, amplification is typicallyinitiated in the early spring via the successful overwintering of the virus in infected mosquitoes(Dohm & Turell, 2001; Goddard et al., 2003; Nasci et al., 2001) or through viral introductionvia the migration of infected reservoirs from regions with year round circulation (Dusek et al.,2009; Rappole & Hubalek, 2003). Viral amplification occurs between spring and early fall (Huhnet al., 2003), with the length of the season decreasing in northern locations (Hayes et al., 2005);in southern Manitoba up to 75% of key vectors have entered diapause by the middle of August(Buth et al., 1990), although hibernation can be delayed until September if temperatures arewarm (Mackay, 1996). Viral rates in both vectors and reservoirs are typically low, but canincrease in response to environmental conditions (Day et al., 2015). Mammals do not serve asamplifying hosts, and incidental transmission to humans, which is termed spillover, occurs onlyduring periods of intense regional amplification (Power & van Marle, 2004) (Figure 2.1). Intemperate climates, amplification is most intense in late summer (Campbell et al., 2002), withhuman cases documented between mid-to-late June to the end of November (O'Leary et al.,2004).The incidence of WNV is clearly both spatially and temporally heterogeneous in North4Enzootic means that the transmission cycle is primarily focused in animals82.2. WNV VectorsAmerica. Certain areas remain free of endemic WNV activity despite significant outbreaks innearby regions. Furthermore, the heterogeneity in incidence seen between regions is mirroredby heterogeneity within regions. Disease incidence varies significantly between counties withina state (Lindsey et al., 2008) (CDC, 2016), and even between census blocks within individualcounties (Winters et al., 2010) or communities (Nielsen et al., 2008). Disease incidence alsovaries between years in single locations in both the US (Lindsey et al., 2008) and Canada(Public Health Agency Canada, 2009).Spatial and temporal patterns of WNV incidence are driven by multiple factors. Demo-graphic and human behavioural characteristics such as age and population density modify dis-ease risk (Bode et al., 2006), while mosquito control efforts like larviciding and adulticiding areassociated with reductions in both mosquito populations (Baker & Yan, 2010) and disease inci-dence (Carney et al., 2008; Haley, 2012). Behavioural and societal factors also affect patterns ofincidence (eg. evening outdoor activity, mosquito avoidance, adulticiding) (Gujral et al., 2007;Reisen, 2013), as do the distribution and intensity of surveillance efforts (Gu et al., 2008; Hadleret al., 2015). However, WNV is also strongly affected by ecological and environmental conditionsbecause of the importance of non-human vectors and reservoirs (Eisenberg et al., 2007). Gaininga systems-level understanding of the factors that affect WNV amplification and transmissioncan allow for: 1) more accurate tracking of periods of high and low incidence, thereby reducingthe application of pesticides and allowing for messaging focused on avoidance measures, and 2)a better understanding of which drivers of disease are most amenable to interventions.2.2 WNV VectorsAlthough WNV has been detected in 65 mosquito species (Petersen et al., 2013), the efficacyof viral replication and transmission varies between vector species. Many mosquitos becomeinfected with WNV in laboratory settings, including species from the genus Aedes, Ochlerotatusand Culex (Sardelis et al., 2001; Turell et al., 2005, 2001a). However, laboratory-derived vec-tor competence5is not sufficient to implicate a vector in WNV transmission; a vector specieswith low abundance, or one that feeds on non-competent reservoirs, will play a limited role inWNV transmission (Turell et al., 2005). Vector competence can vary within a single speciesdepending on the circulating WNV strain (Moudy et al., 2007). In addition, spatial variation invector competency can also result from broader environmental conditions or population genetics(Kilpatrick & Pape, 2013).In North America, Culex mosquitoes drive the majority of transmission with up to 95% ofWNV positive mosquitoes collected in the US belonging to this genus (Andreadis, 2012). Cx.pipiens and Cx. quinquefasciatus (both members of the Cx. pipiens complex) are believedto collectively be responsible for 80% of human infection in the eastern US (Kilpatrick et al.,2005), while Cx. tarsalis is the primary driver of transmission in the western US (Andreadis,5Vector competency in lab settings was determined by allowing mosquitoes to feed on infected hosts and thenevaluating the efficiency with which vectors infect birds in subsequent feedings (Turell et al., 2001a).92.2. WNV VectorsFigure 2.1: Changing levels of WNV in vector and reservoir communities for temperate loca-tions throughout the WNV season. The amplification cycle is split into three stages: suitableconditions, enzootic transmission, and spillover. Modified from Childs (2007).2012). In the southern US Cx. quinquefasciatus is the primary vector (Andreadis, 2012), withCx. nigipalis driving transmission regions where Cx. quinquefasciatus is absent (Godsey et al.,2013).Similar patterns of vector distribution have been seen in Canada, with Cx. tarsalis andCx. pipiens causing the majority of infection nationally. Cx. tarsalis is the dominant vector inthe Canadian prairies, but is rare east of the Mississippi River (Conly & Johnston, 2007). Incontrast, Cx. pipiens is absent in the Prairie Provinces, yet drives transmission in Ontario andQuebec, and is also found in BC (Conly & Johnston, 2007). Both Cx. tarsalis and Cx. pipiensare found throughout BC - as far west as Vancouver Island (Stephen et al., 2006) - making BCthe only province with both species. Cx. quinquefasciats, Cx. nigipalis, and Cx. salinarius donot play a significant role in WNV transmission in Canada.Cx. tarsalis is more competent than Cx. pipiens in laboratory settings (Turell et al., 2005).Cx. pipiens is generally common in urban areas where it thrives in water with a high degreeof organic material, like storm drains, catch basins, bird baths, and ditches (Andreadis, 2012;Nielsen et al., 2008). In contrast, Cx. tarsalis is the dominant vector species in the rural areasof western North America where it breeds in clear standing surface water like irrigated fields,roadside ditches, shallow ponds, and water-filled hoof prints (Andreadis, 2012; Curry, 2004;Eisen et al., 2010; Nielsen et al., 2008). However, Cx. tarsalis has increased in abundance insome urban California centres due to increases in abandoned homes and neglected swimming102.3. Reservoirspools (Reisen et al., 2008b).2.3 ReservoirsBirds are the natural reservoir of WNV, yet not all species are affected equally by infection.Reservoir competency is determined by a combination of susceptibility, mean daily infectious-ness, and the duration of infectiousness (Komar et al., 2003), which all vary between species.Birds infected with WNV typically show non-specific clinical signs 5-6 days after being infected,and generally respond in one of three ways: 1) major organ failure and death, 2) prolongednon-acute clinical infection, and 3) low-level chronic persistence of the virus (Pérez-Ramírezet al., 2014). High viremia typically causes death, but some species like the American robin(Turdus migratorius), northern cardinal (Cardinalis cardinalis), house sparrow (Passer domesti-cus), and raptors experience high viremia without the associated mortality (Komar et al., 2003;Pérez-Ramírez et al., 2014; Wheeler et al., 2009). A threshold viremia of 104-106 pfu/ml ofblood is necessary for the infection of feeding Culex species (Komar et al., 2003), and 28 of 77bird species (from 13 orders and 29 families, respectively) were experimentally shown to haveviral levels above this threshold (Pérez-Ramírez et al., 2014).The numerous competent reservoir species suggests that multiple species play a role in viralamplification. Passerines and Charadriiforme birds (waders, gulls and auks) are thought to beespecially competent WNV reservoirs (Komar et al., 2003). Sparrows and finches were importantamplifying hosts in some areas due to their elevated viremia and high abundance (Kilpatricket al., 2007; Komar et al., 2001; Molaei et al., 2010; Nemeth et al., 2009; O'Brien et al., 2010b;Reisen et al., 2006b), but this was not the case in all areas (Reisen & Brault, 2007). Similarfindings were shown for some small songbirds like tufted titmouse (Baeolophus bicolor) andCarolina wrens (Thryothorus ludovicianus) (Kilpatrick et al., 2007). Corvids have historicallybeen associated with the North American introduction of WNV because their high viremiaand frequent mortality (Brault et al., 2004; Komar et al., 2003; Yaremych et al., 2004). Thenumber of dead crows reported in a region was correlated with the intensity of WNV risk inhuman populations in the years immediately after the introduction of WNV to the New Yorkarea (Eidson et al., 2001), case-control studies show that the odds WNV-positive mosquitoesat a residence were 19.75 times greater if dead corvids were also present (Nielsen & Reisen,2007), and both mosquito and human infection rates were greater near large crow roosts insome areas (Reisen et al., 2006b). Yet there is controversy with regards to whether crows playan important role in WNV amplification because of their high and rapid mortality (Kilpatrick,2011), or whether they are simply conspicuous marker of circulating virus.112.4. Vector Feeding Preference2.4 Vector Feeding PreferenceCompetent vectors and reservoirs are a baseline requirement for transmission, and it is theinteraction between the two that drives the intensity of WNV amplification, despite evidenceof non-vector mediated acquisition via oral routes (Komar et al., 2003). Mosquito feedingbehaviour and host preference affect the timing and intensity of disease transmission (Day,2005). Generally, Cx. pipiens is ornithophilic and feeds almost exclusively on birds, makingthis species important for the maintenance of viral circulation in avian populations (Appersonet al., 2002; Molaei et al., 2006; Thiemann et al., 2012; Turell et al., 2005). In contrast, Cx.tarsalis is a highly competent bridge vector in the western US because it feeds on both birds andmammals (Thiemann et al., 2012; Turell et al., 2005). Yet, these differences in host preferencesmay not be as simple as they first appear, as Cx. pipiens is a more competent bridge vectorthan originally thought because it also feeds on humans (Hamer et al., 2008; Kilpatrick et al.,2006). In North America, Cx. pipiens is a hybrid of two old-world subtypes: one that feedsprimarily on avian species and one that feeds primarily on mammals (Fonseca et al., 2004).This hybridization makes Cx. pipiens an ideal bridge vector and may be one of the reasons whythe severity of outbreaks has increased more in North America than in Europe (Fonseca et al.,2004; Kilpatrick et al., 2006).Mosquito species not only vary in their relative preference for birds versus mammals, butthey also shown meaningful preferences for specific avian species. The proportional compositionof mosquito blood meals are often dominated by individual avian species (Kilpatrick et al., 2006).For example, Cx. pipiens and Cx. restuans in urban and residential settings in Maryland andWashington fed disproportionately on robins (Turdus migratorius), making them a regionallyimportant disease reservoir (Kilpatrick et al., 2006). Cx. pipiens's preference for robins has beensupported by other studies (Hamer et al., 2008), and this avian species was also preferentiallyfed upon by Cx. tarsalis in Colorado (Kent et al., 2009).Other vectors, however, appear to feed more opportunistically. In some areas of California,Cx. tarsalis fed more on house sparrows than on American robins, mourning doves or housefinches (Lura et al., 2012). Cx. tarsalis has been shown to generally avoid passerines in somelocations (Kent et al., 2009), while preferring to feed on American robins and sparrows (Kentet al., 2009; Kilpatrick et al., 2006). In some locations in California, Cx. tarsalis fed on ardeids(e.g. herons and bitterns) during the summer, and yellow-billed magpies (Pica nuttalli) duringthe winter (Thiemann et al., 2012). This regional, and even site specific, variation in blood mealcomposition for a single vector species indicates that Cx. tarsalis opportunistic feeding patternsin at least some locations (Reisen et al., 2013; Thiemann et al., 2012).Host preference can even vary between regions for a single vector species, suggesting thatfeeding preference may be partially dependent on contextual factors. For example, Americancrows were one of the most common blood meal sources for Cx. tarsalis in multiple studysites in California (Thiemann et al., 2012), and for Cx. pipiens in select sites in Maryland andWashington (Kilpatrick et al., 2006). However, an analysis of Cx. pipiens blood meals in Illinois122.5. Reservoir Community Structureindicated corvids played a lesser role in regional amplification than their laboratory-derivedreservoir competency would suggest (Hamer et al., 2009; Kilpatrick et al., 2006). Similarly,sparrows have been implicated as important reservoirs in some locations (Hamer et al., 2008;Molaei et al., 2006), yet are avoided by key vectors in other areas (Kilpatrick et al., 2006; Komaret al., 2001).Vector host preference also changes throughout the WNV season. These changes can impactspillover if vectors switch from 1) avian hosts to mammal hosts, or 2) a primarily avian host toanother species with lesser/greater reservoir competency. The fraction of blood meals taken frommammals in relation to birds increased in Cx. pipiens after the dispersal of American robinsin the north-east and north-central regions of the US, and similar patterns were seen for Cx.tarsalis in California and Colorado (Kilpatrick et al., 2006). However, such host switching wasnot observed in all locations and varies spatially and temporally depending on the availabilityof mammal hosts (Lampman et al., 2013). The idea of host switching as a driver of spilloveris disputed by work done in Chicago that is indicative of an opportunistic feeding strategy forCx. pipiens which switched from feeding from robins to other bird species (Hamer et al., 2009).In fact, Cx. pipiens fed less frequently on mammals during the period of estimated spillover,suggesting that spillover resulted not from changes in feeding patterns but from a ramp-upin amplification that occurred late in the WNV season (Hamer et al., 2009). Similarly, Cx.tarsalis in Colorado fed preferentially on robins in the early summer and on communal roostsof sparrows in the later season, while maintaining a constant preference for doves (Kent et al.,2009). Cx. tarsalis did, however, show a marked increase in the proportion of blood mealstaken from mammals in the later summer months, although human blood meals remainedrare (approximately 1.5%). This suggests that this trend resulted not from altered feedingpreferences but instead from changes in mosquito density and avian community structure (Kentet al., 2009). Finally, changes in host feeding preference do not always favour spillover, as hostswitching from robins early in the season to the less competent northern cardinal in the lateseason was hypothesized to actually lower human risk of WNV transmission (Levine et al.,2016).In summary, regional differences in feeding preferences may explain differences in WNVdisease intensity between locations, and further research is needed to evaluate the consistencyof such patterns in Canada. However, care must be taken when linking within-season feedingshifts to spillover into human populations, as the timing of such shifts may also result fromhuman behavioural changes such as increased evening activity in late summer (Bolling et al.,2009).2.5 Reservoir Community StructureObserved differences in vector feeding preferences, coupled with variations in avian reservoircompetency, mean that the regional composition of avian communities may affect WNV in-132.5. Reservoir Community Structurecidence (Keesing et al., 2006; Ostfeld & Keesing, 2000a). The 'dilution effect' describes howlower vector infection rates may occur in regions with diverse reservoir communities (Ostfeld &Keesing, 2000a; Schmidt & Ostfeld, 2001). This was hypothesized to occur if biodiverse com-munities contained a higher proportion of low competency reservoirs that represent a sourceof dead-end blood meals (see Keesing et al. (2006) for review). Some avian species thereforeinhibit WNV amplification (Levine et al., 2016). Several criteria must be met for the dilutioneffect to occur: a generalist vector species must be present, there must be variations in reser-voir competency, a dominant and competent host must feed the majority of vectors, and vectorinfection must occur primarily through host feeding (Ostfeld & Keesing, 2000a). The dilutioneffect has been postulated to occur for Lyme disease (LoGiudice et al., 2003; Ostfeld & Keesing,2000a) and hantavirus (Peixoto & Abramson, 2006), and has been supported empirically forLyme disease (Ostfeld & Keesing, 2000a) and rodent-borne viral hemorrhagic diseases (Mills,2006).The dilution effect may affect WNV systems given the significant variation in reservoircompetency between bird species (Kilpatrick et al., 2006; Komar et al., 2003; Wheeler et al.,2009). Several empirical studies have shown associations between avian community structureand WNV infection at local and regional scales. The first published report showed a negativecorrelation between the biodiversity of non-passerine species, a group of birds typically thoughtto be poor WNV reservoirs, and mosquito infection rates at a local level, as well as humaninfection rates at the county level in Louisiana (Ezenwa et al., 2007). This was suggested toindicate that the non-competent reservoirs were dampening WNV amplification in the region.However, there were no associations seen between passerine diversity or abundance and WNVinfection. Furthermore, avian diversity and vector infection were not determined at the samepoint in time, and a bias may exist because of wide-scale avian mortality in the years afterviral introduction (LaDeau et al., 2007). A more recent study used a novel county-level contrastbetween adjacent counties with-and-without WNV to evaluate associations between nationalbird survey data and country level human incidence in the eastern US (Swaddle & Calos, 2008).This study's findings showed a negative association between avian biodiversity, as measured byspecies richness (number of unique species), and human infection rates in the eastern US, withavian community structure6explaining up to 50% of the variation in human WNV incidence.This study also differentiated between community evenness (the relative abundance of eachspecies in the community) and species richness (the absolute number of unique species), withthe latter showing stronger statistical associations with disease incidence than did the former.However, the evenness of non-passerine species was positively associated with human infection,suggesting that this group may be better reservoirs than originally believed (Swaddle & Calos,2008). These patterns were consistent before and after endemic WNV activity, indicating arobust pattern even after significant avian die-off (LaDeau et al., 2007). A negative association6Community structure here refers to the relative abundance of key reservoir species within regional avianpopulations142.5. Reservoir Community Structurebetween WNV incidence at the county-level and avian species diversity was also evaluatedover the entire continental US, with a negative association between avian diversity and WNVincidence detected for all three years of the study period (Allan et al., 2008). Furthermore, thisstudy controlled for urbanization and mosquito abundance, thereby providing a better estimateof the true impact of diversity (Allan et al., 2008). Significantly, this study was the first toemploy a composite measure that accounted for both the relative abundance of species andtheir competency. The measure is termed community competence, and represented the sumof the product of abundance and competences as defined in lab studies (Komar et al., 2003,2005). Significant associations were seen between community competence and human infection,although the strength of those associations were weaker than seen between diversity and WNVincidence (Allan et al., 2008).Despite support for the dilution effect on large spatial scales, questions were raised as to themechanism driving this effect. Specifically, were the observed patterns driven by biodiversity orby other characteristics of avian community structure (Randolph & Dobson, 2012)? A meta-analysis evaluating the role of biodiversity in driving zoonotic diseases indicated a weak effectand suggested that publication bias preferred studies that favoured support of the dilutionhypothesis (Salkeld et al., 2013). For WNV specifically, recent empirical work failed to showassociations between avian diversity and infection in Culex species or seroprevalence in birdswhen these associations were measured at a fine scale in the Chicago region, which suggests thatpatterns seen at large scales may not be causative (Loss et al., 2009b), or alternatively that thesite of exposure was not related to the site of residence.Relationships between avian community structure and WNV incidence may be driven notby biodiversity per se, but instead driven by differences in relative abundance of key species(Randolph & Dobson, 2012). A true relationship with biodiversity would require disease ratesto vary across sites that have different relative abundances of key reservoir species, and the sameoverall reservoir abundance (Randolph & Dobson, 2012). It may therefore be the case that forWNV, mosquito infection and human risk were driven more by community composition or thepresence of specific avian species. For example, highly competent reservoirs like robins, bluejays (Cyanocitta cristata), grackles (Quiscalus quiscula), house sparrows and finches are moreprevalent in areas with low overall avian diversity (Allan et al., 2008; Miller et al., 2003; Smith,2003), although these findings are most applicable to urban settings. Negative associationsbetween disease and diversity could possibly be driven by a greater abundance of these keyspecies. Similarly, regional reductions in WNV incidence seen after several large hurricanes inFlorida were attributed to reductions in the presence of blue jays, common grackles, mourningdoves and northern cardinals (Day et al., 2015), which supports the claim that the total numberof reservoirs is more important than the relative number.In addition, it must be clearly understood that any impacts of avian community structure onpopulation level incidence are filtered through the feeding behaviours of mosquito vectors (seeSection 2.4). Novel measures like the community force of infection - which use lab information152.6. Avian Immunityon reservoir competency to identify the per capita rate at which susceptible vectors acquire in-fection, and express it for an entire avian community - may provide better measures of the trueassociation between avian communities and infection rates (Hamer et al., 2011). Using thesemeasures identified that American robins and house sparrows accounted for 95.6% of all Cx.pipiens infection, indicating that community measures like avian diversity and richness maybe poorly correlated with WNV risk given the disproportionate role that select species play.This finding was further expanded upon by work from Colorado that showed the abundanceof high amplification species, specifically blue jays, American robins, house finches, and Ameri-can kestrels (Falco sparverius), were significantly associated with human incidence in Colorado(McKenzie & Goulet, 2010). This provided additional support for the idea that a small subset ofspecies may drive WNV amplification in many regions. There is little doubt that the structuralmakeup of reservoir communities affects disease rates, regardless of the mechanism by whichthis process occurs.2.6 Avian ImmunityIn many regions, annual WNV incidence follows a three-year cycle with low activity in theyear of initial detection, explosive increases the following year, and subsidence in year three(Day et al., 2015; Hayes et al., 2005; Kwan et al., 2012; Nolan et al., 2013; Reisen & Brault,2007; Reisen, 2013). This cyclic behaviour is caused by changes in avian herd immunity. In LosAngeles, the seroprevalence of peridomestic passerines (house finches and house sparrows) in thepreceding winter (January-March) was negatively associated with the intensity of spillover thefollowing summer (Kwan et al., 2012). WNV outbreaks were initiated when the herd immunitydropped below 10%, and collapsed once acquired immunity rose above 25% (Kwan et al., 2012;Reisen, 2013). The pattern repeated once community seroprevalence became diluted below keythresholds by the addition of naive chicks and fledglings, usually after a years time. This lossof avian immunity was speculated to be responsible for the large 2012 outbreak in Texas aftertwo years of limited transmission (Nolan et al., 2013; Reisen, 2013).2.7 Role of ClimateClimate and weather conditions are the most important determinants of the timing of WNVoutbreaks, with combinations of prior temperature and precipitation rates explaining up to 80%of the variation in Cx. pipiens infection rates in Chicago (Ruiz et al., 2010). Climate can impactWNV transmission on both large and small spatial scales, as broad climatic conditions drivespecies distribution, while conditions over smaller scales impact species interactions and focalpoints of WNV transmission (Cohen et al., 2016).162.7. Role of Climate2.7.1 TemperatureTemperature is strongly associated with the intensity of annual WNV transmission. WNVepidemics in both the US and Canada occurred during periods of above average summer tem-peratures (Canada, 2003, 2007; Reisen et al., 2006a), including in Saskatchewan (Reisen, 2013).A case-crossover study evaluating the effects of weather on 16,298 US WNV cases showed thata 5◦C increase in mean maximum weekly temperature was associated with a 32-50% increase inWNV incidence (Soverow et al., 2009). In California, transmission was associated with averagetemperatures being 2-5◦C above the 30-year average (Reisen et al., 2006a). Heat is also a knowndriver of WNV activity in Europe and surrounding countries (Marcantonio et al., 2015; Savage,1999), as well as in Israel (Paz, 2006).This positive association between temperature and WNV incidence is driven by the effectsof temperature on the developmental rates of both vector and virus. Insects are ectothermsand external temperature affects both the development and survival rate of the aquatic larvaland terrestrial adult stages (Ciota et al., 2014; Dodson et al., 2012). The relationship betweentemperature and development or survival rates varies between species (Ciota et al., 2014) andbetween locations for a single species (Ruybal et al., 2016). Cx. pipiens, Cx. restuans andCx. quinquefasciatus developed 2.9 times faster when the temperature was increased from 16to 24◦C, and a 2.3-fold increase was seen when the temperature increased from 24 to 32◦C(Ciota et al., 2014). The relationship between temperature and mosquito development alsovaried between locations for the same species, as Cx. pipiens development rates varied up to84% between locations in the eastern US (Ruybal et al., 2016).The increased rate of mosquito development with increasing temperatures translates into agreater number of mosquito generations per season, which increases the intensity of viral am-plification and the resulting human risk. However, temperatures above 32◦C are detrimental tomosquito development (Lafferty, 2009) and can lead to increased mortality (Epstein & Defilippo,2001). Larval mortality increased for Cx. tarsalis when reared at 31◦C in comparison to 19 or25◦C (Dodson et al., 2012), while Cx. restuans was negatively affected by temperatures above24◦C. Culex species have also been shown to avoid temperatures exceeding 25◦C (Thompsonet al., 1996).Temperature also determines the rate of viral development within vectors. Specifically,increasing the temperature shortens the extrinsic incubation period (EIP), which is the timerequired for WNV to replicate and disseminate within a mosquito, which causes the vectorto become infectious (Reisen et al., 2006a). Furthermore, the impact of temperature on viraldevelopment varies between species. WNV transmission by Cx. pipiens was significantly lowerat 18◦C than at 30◦C, with disseminated infection detected 4 days after infection in mosquitoeskept at 30◦C, as compared to 25 days after infection at 18◦C (Dohm et al., 2002).Positive associations between temperature, viral development and transmission have alsobeen observed in Cx. tarsalis. The minimum temperature requirements for the development ofthe NY99 strain in Cx. tarsalis is 14.3◦ Celsius, with greater titres and shorter EIP occurring172.7. Role of Climatewhen vectors are reared at 30◦C rather than at 18◦C (Reisen et al., 2006a). The importance ofthe impact of temperature on WNV transmission may be even greater than originally thought, asthe transmission of WN02 by Cx. pipiens was related to temperature through a T 4 relationship(instead of a linear relationship) and also invaded mosquito salivary glands more efficientlythan the NY99 strain at warmer temperatures (Kilpatrick et al., 2008). Models relying onlinear association between temperature and WNV transmission may, therefore, significantlyunderestimate the importance of temperature (Kilpatrick et al., 2008).The frequency at which mosquitoes take blood meals is also positively associated with tem-perature. The length of the gonotrophic cycle - the time required to oviposit after taking ablood meal - decreases as temperatures increases resulting in more host contact (Reisen et al.,2006b). Temperature is therefore directly related to the number of mosquito generations perseason and total mosquito abundance (Becker, 2008). This relationship between temperature,the virus's EIP, and vector's gonotrophic period, is key, as the virus's failure to replicate beforemosquito death will halt amplification. This is especially relevant for temperate climates, wherethe relationships drive the seasonality of the disease (Hartley et al., 2012). The ratio betweenthe EIP and the gonotrophic cycle has been termed the number of bites to transmission (BT),and modelling has shown that BT decreases with increasing temperature and to be inverselyrelated to the reproductive number for WNV in California (Hartley et al., 2012). In California,55-58% of seroconversion in chickens was documented in regions with BT values between 2 and3 (Hartley et al., 2012).It is the cumulative heat experienced by mosquitoes that determines the EIP. Degree days(DD) are a concept that captures the association between viral development and fluctuatingtemperatures. DDs were originally created to determine development times for crops and croppests (Wilson & Barnett, 1983), and use the product of temperature and time to estimate thecumulative energy required for an organism to develop (Wilson & Barnett, 1983). The measurehas been successfully co-opted to describe the EIP for viral development within mosquito vectors.Laboratory work has estimated that 109 cumulative DDs (base 14.3◦C) were required for thecompletion of the WNV EIP in Cx. tarsalis (Reisen et al., 2006a). The relationship betweenDDs and development may change over the course of a season, as model optimization suggeststhat while the 109 DD estimate worked early in the season, an EIP of 89 DD days provided abetter match late in the season (Chen et al., 2013). Variation from the 109 DD threshold wasalso suggested in California, with an improved model fit for mosquito transmission occurringwith an EIP estimate of 75 DD (Konrad et al., 2009).WNV transmission and spillover into human populations is also affected by temperaturethrough more subtle ecological mechanisms. Warm temperatures early in the WNV seasonhave been linked with early initiation of the amplification cycle (Shand et al., 2016). Theaccumulation of heat over the WNV season may also impact the timing of vector crossover,whereby the abundance of key avian feeding mosquitoes decrease relative to bridge vectors thatfeed on both birds and mammals, including humans (Kunkel et al., 2006). Warm early season182.7. Role of Climatetemperatures may have therefore facilitated higher WNV incidence by accelerating the initialamplification cycle, thereby extending the period of transmission to humans by allowing forearlier vector crossover.Temperatures also likely define the northern limit of the viruses transmission range (Reisenet al., 2006a). Climate change is hypothesized to drive range increases for key VBDs in northernregions like Canada or northern Europe, increase the frequency of future outbreaks (Epstein& Defilippo, 2001), and extend the duration of the WNV season (Brown et al., 2015). Theimpact of warmer temperatures was expected to be more pronounced in regions with meansummer temperatures 16-24◦C, as compared to regions with current summer temperatures above24◦C (Ciota et al., 2014). Warmer winters may also lead to early release from hibernacula,leading to higher winter mortality if temperatures subsequently cool (Ciota et al., 2011). Thereremains much to be learned regarding the complex associations between temperature and VBDtransmission (Kilpatrick & Randolph, 2012; Reiter, 2008). Simplistic climate envelope modelsthat predict elevated disease activity with increasing temperatures should be evaluated criticallyin light of the known ecological complexities of the system (Davis et al., 1998). In addition, thepredictions of many climate studies were based on mean daily or monthly temperatures, butfailed to account for changes to daily temperature fluctuations that can increase the susceptibilityof ectotherms to altered climatic conditions, especially at the high end of the ectotherms' thermalrange (Blanford et al., 2013).2.7.2 PrecipitationStanding water is required for mosquito larval development, leading to the assumption thatprecipitation promotes mosquito-borne diseases (Chase & Knight, 2003). However, the effect ofprecipitation on WNV transmission is location specific and at times counterintuitive (Soverowet al., 2009). Unlike the direct effects of temperature on viral and mosquito development,precipitation indirectly effects mosquito abundance either by altering the number of mosquitobreeding habitats, or by indirectly affecting the contact between vector and reservoir. In addi-tion, the impacts of precipitation on WNV incidence are affected by temperature: precipitationhas limited effects on disease incidence when temperatures are low and a greater effect whenwater levels are low (Shand et al., 2016).Positive associations between precipitation and WNV incidence have been seen in locationswhere mosquito abundance was limited by the availability of development sites (Landesmanet al., 2007). Recent precipitation was important for container-breeding mosquitoes like Aedes,which explain positive associations between precipitation and VBDs like malaria (Krefis et al.,2011) and dengue (Colon-Gonza et al., 2013). In theory, similar processes could occur for WNVin urban settings where the container and storm-drain developing Cx. pipiens drives WNVtransmission. Rainfall greater than 50mm was associated with WNV cases 1-2 weeks later inthe US (Soverow et al., 2009). However, across the continental US, annual precipitation wasnegatively associated with WNV activity, with wet areas like the Pacific Northwest generally192.7. Role of Climatereporting a limited WNV incidence (CDC, 2016). Excessive precipitation may negatively af-fected mosquito survival (Jones et al., 2012), or abundance when runoff flushed mosquito larvafrom sewer and catch basin breeding sites (Paaijmans et al., 2007; Su et al., 2003). However,the impacts of precipitation were regionally specific, likely in relation to the distribution of keyvectors, as annual precipitation was positively associated with incidence in most western regionsbetween 2004 and 2012 (Hahn et al., 2015).The timing of precipitation also matters. Spring drought has been associated with manyWNV outbreaks (Epstein & Defilippo, 2001). In New Jersey, increased temperatures and de-creased precipitation in June-July were strongly correlated with WNV infection rates over the2003-2011 period (Johnson & Sukhdeo, 2013). In Chicago, low precipitation was the bestpredictor of the spatial distribution of high mosquito infection, and a dry spring followed byprecipitation was seen in some, but not all years of WNV activity (Ruiz et al., 2010). Twoto four months of drought preceded outbreaks of both SEV and WNV in the United States(Epstein & Defilippo, 2001), while in Florida, WNV incidence was correlated with the occur-rence of drought 2-6 months prior followed by ground wetting 0.5-1.5 months prior (Shamanet al., 2005). This positive relationship between drought and WNV was also seen in outbreaksin Russia, Romania, and Israel (Epstein & Defilippo, 2001). In non-urban settings, droughtmay spatially focus vectors and reservoirs around remaining patches of standing water (Shamanet al., 2002). This increases the interaction between vector and reservoir, thereby facilitatingincreased amplification. Drought may also reduce river and stream flow, increasing the abun-dance of Cx. tarsalis breeding sites (Lafferty, 2009; Landesman et al., 2007). Finally, droughtcan also increase the organic content in remaining water, a condition favoured by Cx. pipiens(Shaman et al., 2010).Associations between precipitation and WNV activity are dependent on regional ecologicaland environmental conditions. Winter and spring precipitation (including snowpack) have beenpositively associated with Cx. tarsalis abundance in California, with the spring release of snow-pack affecting the prevalence of mosquito habitats (Reisen et al., 2008a). In eastern Colorado,wet springs and dry summers were associated with human disease (Shaman et al., 2010). How-ever, WNV incidence was more strongly associated with drier springs and summer conditions inthe mountainous western part of the state because spring snow melt makes mosquito develop-ment sites less reliant on spring precipitation (Shaman et al., 2010). These findings clearly showthat the impact of precipitation is modified by the regional hydrology. Precipitation may playa larger role in regions with limited water than in areas with abundant natural water sources.Despite the somewhat consistent association between drought occurring over the precedingsix months and WNV activity, the impacts of drought during the preceding WNV season differsbetween western and eastern regions of North America (Landesman et al., 2007). Above averageprecipitation in the preceding WNV season was positively associated with WNV activity in theeastern US and negatively associated with WNV activity in western locations (Landesmanet al., 2007). Differences were again likely driven by the distribution of the key vectors and202.8. Landscape Factors as Mediators of Diseasetheir differing habitat preferences. In the urbanized eastern US, elevated precipitation in theprevious year may increase the number of container breeding sites for Culex pipiens, subsequentlyincreasing the abundance of overwintering mosquitoes (Lafferty, 2009; Landesman et al., 2007).In the western parts of the continent, however, drought in the preceding months or years ishypothesized to affect aquatic mosquito predators more severely than mosquito larva, leadingto competitive release and increases in vector populations the following year because mosquitopopulations rebound faster than their predators (Chase & Knight, 2003; Landesman et al.,2007). Finally, it must be noted that this east-west breakdown in the effect of previous years'precipitation was not universal as negative association between incidence and the precedingyears' precipitation have been reported in Chicago where Cx. pipiens is the dominant vector(Ruiz et al., 2010), as well as in Mississippi where drought in the previous year was positivelyassociated with WNV incidence (Wang et al., 2010).Humidity, independent of precipitation, can also affect the distribution of arboviruses (Gubler,2002), mosquito survival (Yé et al., 2007) and has been associated with MBDs like malaria (Liet al., 2013). However, evaluations of the role of humidity in driving WNV transmission arelacking and the exact nature of the association remains unclear. Cx. tarsalis and Cx. pipienshave been shown to preferentially move towards areas with higher relative humidity (Kessler &Guerin, 2008; Service, 1993), yet the current months average humidity was negatively correlatedwith Cx. tarsalis abundance in South Dakota, potentially because high humidity was correlatedwith precipitation (Chuang et al., 2011). Relative humidity has also been linked to human WNVdisease, being positivity correlated with WNV hospitalizations in Israel (Paz, 2006) and withillness in Europe (Paz et al., 2013). However, temperature remained a stronger predictor in bothinstances (Paz & Semenza, 2013). Positive associations between humidity and WNV disease inthe US is speculated to result from links with increased vector breeding sites and not from directimpacts on mosquito survival (Soverow et al., 2009). Humidity may well be related to WNVtransmission, yet its effects on transmission are difficult to separate from those of precipitationor temperature.2.8 Landscape Factors as Mediators of DiseaseThe presence of key vectors and reservoirs likely define the basic requirements for WNV trans-mission. However, land cover (biophysical processes), patterns of land-use (economic classifica-tions), and active anthropogenic landscape change all affect disease risk (Lambin et al., 2010;Patz et al., 2004). Landscape characteristics, defined here as the type and spatial distribu-tion of habitats within an area, affect VBD transmission by impacting: 1) the distribution andabundance of disease reservoirs and vectors, 2) the contact rate between disease vectors andreservoirs, and/or 3) the contact rate between vectors and human hosts (Eisenberg et al., 2007;Wilson, 2001). Landscape therefore represents the interface between social and environmentalprocesses (Turner, 1989).212.8. Landscape Factors as Mediators of Disease2.8.1 Natural LandscapesWetlands are hypothesized to facilitate WNV activity during drought by spatially focusing birdsand mosquito around the remaining standing water (Shaman et al., 2002). The relationship be-tween wetlands and WNV is affected both by weather conditions and the specific characteristicsof a given wetland. Wetlands were negatively associated with county-level human WNV inci-dence in Iowa (DeGroote et al., 2008), and the prevalence of infection in both humans and vectorswas lower near wetlands in Louisiana (Ezenwa et al., 2007). However, this negative associationbetween wetlands and WNV infection appears dependent on wetland size, with small wetlands(<15 ha) associated with higher vector infection rates than were large wetlands (>100 ha) inNew Jersey (Johnson et al., 2012). Similar findings in the northeastern and midwestern USshowed 100% greater WNV incidence in counties with small wetlands and Cx. tarsalis was theprimary vector than in counties with large wetlands (Skaff & Cheruvelil, 2016). Transient andpermanent wetlands also differed in their mediating effects. WNV incidence was 300% higherwhen drought occurred in areas with Cx. tarsalis and a high abundance of semi-permanent wet-lands than in drought-affected counties with few semi-permanent wetlands (Skaff & Cheruvelil,2016). Such differences were likely driven by several ecological conditions. First, large wetlandshave greater avian diversity and infection may be lower because of a higher abundance of non-competent reservoirs (Ezenwa et al., 2007; Johnson et al., 2012). Secondly, mosquito predatorswere more abundant in large and well-connected wetlands than they were in small isolated wet-lands, and small wetlands may therefore have larger vector populations because of competitiverelease (Chase & Knight, 2003; Chase & Shulman, 2009). In addition, birds and mosquitoesaggregate around areas of remaining water when semi-permanent wetlands disappear duringperiods of drought, thereby increasing vector-avian contact and facilitating viral amplification(Shaman et al., 2005; Skaff & Cheruvelil, 2016). The impacts of drought discussed in previoussections were likely partially driven by the presence or absence of wetlands, a hypothesis whichmerits more research.Associations have also been shown between WNV and other landscape features like moun-tains and forests. Mountains were negatively associated with WNV incidence in Georgia, whichthe authors attribute to a combination of lower temperatures, lower vector abundance, anddifferences in avian community structure in mountainous settings (Gibbs et al., 2006). The linkbetween lower temperatures at elevations and limited vector abundance has also been seen formalaria in the highlands of Africa (Bødker et al., 2003). Similarly, mosquito community diver-sity and Cx. tarsalis abundance were low above 1600m in Colorado, with the species rare infoothills or mountainous regions (Eisen et al., 2008). Climatic conditions at elevation also likelyinhibited viral amplification as no WNV positive Culex mosquitoes were detected above 1600mduring a year of intense WNV activity in Colorado (Bolling et al., 2009). Nearby mountains,however, may be positively associated with WNV due to the impacts of snow-pack on mosquitobreeding sites (Shaman et al., 2010).Forests were negatively associated with WNV in the northeastern US, with counties con-222.8. Landscape Factors as Mediators of Diseasetaining the lowest quartile of forest cover 4.4 times more likely to have above-median WNVincidence than counties with >70% forest cover (Brown et al., 2008). Forest patch size was alsonegatively associated with vector index in the Atlanta area by Lockaby et al. (2016). These au-thors hypothesized that WNV risk increased as forest cover become lower because hard surfacesincrease water levels in sewage outflows, leading to overflow and the pooling of nutrient richwater, which serves as a mosquito breeding habitat. In addition, negative associations betweenWNV risk and forest cover in Atlanta were attributed to a higher relative abundance of lesscompetent WNV reservoirs (Levine et al., 2016). However, greater forest cover has also beennegatively associated with WNV risk in humans in Saskatchewan (Epp & Waldner, 2009). Incontrast, in Europe strong associations were seen between WNV incidence and populated forestcharacterized by low population density and a mix of forest, human habitation, transitionalhabitat and farmland (Marcantonio et al., 2015). These associations, coupled with the recentfocus on WNV activity in the Great Plains (Lindsey et al., 2008) and Prairie Provinces (Pub-lic Health Agency Canada, 2009) suggest that low lying agricultural lands or grasslands favourWNV activity in human populations, likely because irrigation plays a potentially important rolein providing a habitat for the prairie mosquito Cx. tarsalis (Gates & Boston, 2009).The spatial distribution of disease risk may be dictated not only by the vector-producingpotential of a given landscape, but also by specific small-scale habitat features that favourvector-reservoir interaction. Cx. tarsalis in California were known to aggregate in ecotones withstanding vegetation, but were less common in areas with low lying vegetation, sand spits andopen water (Lothrop & Reisen, 2001). Such fine-scale habitat preferences may have resulted frommosquitoes aggregating in areas with abundant avian reservoirs. Habitat preferences thereforedictate the contact rate between vector and reservoir (Lothrop & Reisen, 2001). In turn, thehabitat usage of avian species may affect their risk of WNV infection if their favoured habitat issituated close to mosquito habitat. Consequently, habitat choice may effectively modify reservoircompetency of any given avian species2.8.2 UrbanizationUrbanization has driven disease emergence worldwide (Patz et al., 2004), and affects the trans-mission of some mosquito-borne diseases (Gubler et al., 2001; Mackenzie et al., 2004). This isbecause it alters ecological communities, often favouring generalist vectors and reservoirs thatthrive in human-dominated landscapes (Bradley & Altizer, 2007). Vector-borne and zoonoticdisease risk can, therefore, vary across time and space in relation to patterns of human devel-opment. One area at particular risk for disease emergence is the boundary between natural andhuman driven landscapes where species from different ecological niches interact (Despommieret al., 2006).Initial outbreaks of WNV in North America occurred in urban areas in the north-easternUS (Gubler, 1998), and variations in WNV activity between northern New England (lowerWNV activity) and the more urbanized mid-Atlantic (higher WNV activity) was attributed to232.8. Landscape Factors as Mediators of Diseasegradients of urbanization (LaDeau et al., 2008). The observed association between urbanizationand WNV infection in the eastern US was likely driven by the habitat requirements of keyregional vectors (Kilpatrick et al., 2005). These habitat requirements also cause variation invector community structure along gradients of urbanization at smaller scales. The proportionof Cx. pipiens relative to Cx. tarsalis was greater in urban sites than in suburban sites, and hasbeen shown to be greater in suburban sites than in exurban sites in the Seattle area (Pecoraroet al., 2007). Similar patterns have been observed in New Mexico, where Cx. tarsalis abundancewas higher in rural settings, with Cx. salinaris and Cx. pipiens quinquefasciatus more commonin urban areas (DiMenna et al., 2006).The relationship between urbanization and vector community structure has important con-sequences for WNV transmission. WNV seroprevalence rates in adult birds were up to 2.5 timeshigher in urban sites than in non-urban sites in Atlanta, Georgia (Bradley et al., 2008). WNVprevalence in avian populations was also greater in suburban and urban areas of Atlanta thenin rural areas (Gibbs et al., 2006), and infection rates in small mammals were greater in moreurbanized areas in Maryland and Washington (Gómez et al., 2008). The association betweenurbanization and WNV infection rates in human populations is not well-studied, but does in-dicate that human-made environments can modify risk. For example, suburbs with moderatevegetation cover and moderate housing density had higher WNV risk during the 2002 outbreakin Chicago (Ruiz et al., 2007). Urban and suburban development were also positively associatedwith WNV infections (Pecoraro et al., 2007; Ruiz et al., 2007), while road density, stream den-sity, slope grades, and vegetation cover within an urban settings have also been linked to WNVincidence in Mississippi (Cooke et al., 2006). Similar patterns have occurred in Ohio, whereWNV cases were clustered in suburban and urban areas, with lower disease incidence occurringin areas with parks, agricultural land, and non-forested wetlands (LaBeaud et al., 2008). InAtlanta, significant positive associations were also shown between combined sewer outflows andhuman, mosquito, and avian infection (Vazquez-Prokopec et al., 2010).While WNV incidence appears associated with urban characteristics in the eastern US, thispattern does not hold in the western US (Bowden et al., 2011). Since the initial westward spreadof WNV, the highest disease incidence in the US has been focused in the Great Plains region(Petersen et al., 2012a), with a similar pattern being seen in Canada (Public Health AgencyCanada, 2009). WNV incidence was higher in rural areas of the US (Gates & Boston, 2009;Wimberly et al., 2008) and in parts of Canada (Schellenberg et al., 2006). In Saskatchewan,Canada, seroprevalence in rural residents was 16.8% compared to 3.2% in urban residents (OR:6.13 (CI: 2.82-13.34)) (Schellenberg et al., 2006). Similarly, in the US, positive cases identifiedfrom blood bank samples are 1.6 times (adjusted OR: 1.3-2.0) more likely to come from ruralareas than from urban centres (Orton et al., 2006). No relationship between urbanization andWNV at a continental level has been reported in Europe, possibly resulting from urban planningstrategies that are different from those in North America (i.e. US suburbs typically having moregreen space than European suburbs) (Marcantonio et al., 2015).242.8. Landscape Factors as Mediators of Disease2.8.3 Agriculture and IrrigationAgriculture has been associated with elevated VBD risk worldwide (Harrus & Baneth, 2005;Molyneux, 2003; Organization, 1997). For example, agriculture has been linked to Bancroftianfilariasis (Dzodzomenyo & Simonsen, 1999; Harb et al., 1993; Thompson et al., 1996), Rift Val-ley fever (Wilson, 1994), Japanese Encephalitis (Keiser et al., 2005), and malaria7(Eisenberget al., 2007; Gratz, 1988; Qunhua et al., 2004). The link between agriculture and VBD inci-dence is hypothesized to result from the landscape modification that accompanies agriculturalproduction. Not all water sources are equally suitable for mosquito development; slow movingor stagnant standing water is most conducive to vector development (Norris, 2004). Changesto local hydrological conditions that increase the amount of standing water - like irrigation -therefore increase the number of vector development sites (Mackenzie et al., 2004; Norris, 2004;Patz et al., 2004; Service, 1991). This has been shown for WNV vectors like Cx. tarsalis, whichthrives in wet pasture lands or irrigated landscapes in prairie regions (Luby et al., 1969; Rapp,1985; Reisen, 2002) and in California (Nielsen et al., 2008) (see Chuang & Wimberly (2012) forcontrasting findings).The association between WNV incidence and irrigation, and WNV and agriculture moregenerally, can be difficult to demonstrate due to the complex web of interacting factors thataffect VBD transmission. However, the spatial distribution of WNV incidence in humans pointstoward the mediating effect of agriculture and irrigation. WNV incidence in the western US hasbeen more strongly associated with agricultural land than it was with urban land cover (Bowdenet al., 2011), and in Illinois the proportion of agricultural landscape was one of the best predictorshuman WNV cases (Liu et al., 2008). Similarly, census divisions in Iowa with WNV activityhad a higher proportion of agriculture landscapes than did divisions with no WNV activity(DeGroote et al., 2008). Indirect measures of agricultural activity, such as total county cropsales, have also predicted human WNV incidence rates in Colorado, Nebraska, Louisiana andPennsylvania (Miramontes et al., 2006). Positive associations between agriculture and WNVwere also seen across Europe (Marcantonio et al., 2015; Pradier et al., 2008).Multiple characteristics of agricultural land may facilitate WNV transmission, however, ir-rigation has been specifically linked to WNV transmission by studies using a variety of method-ologies. The percentage of irrigated landscape has been associated with incidence in the GreatPlains (Wimberly et al., 2008) and Colorado (Eisen et al., 2010), with distance to irrigated landalso positively associated with disease in the latter location. In Texas, the greatest risk forhuman WNV in 2003 occurred in regions with abundant irrigation (Warner et al., 2006), whilea geo-referenced case-control study in in the same state found that human cases living within327m of irrigation canals had a greater risk of developing WNV than those living >7396m fromirrigation canals (OR: 5.7, CI: 2.8-11.9) (Cardenas et al., 2011). This trend was consistent inthe continental US where the incidence of both human and equine WNV cases increased by 50%7This relationship is likely confounded in part by concurrent increases in socioeconomic conditions or historicimmunity252.9. Viral Introduction and Spreadand 63% respectively with each 0.1% increase in total crop lands irrigated at the county levelin the continental US (Gates & Boston, 2009).While a body of work has accumulated suggesting meaningful associations between agri-culture and WNV incidence in humans and animals in the US, it is possible that this trendresults not from an inherently greater hazard of WNV infection in agricultural settings, butfrom unrelated behavioural patterns that independently increase WNV risk (i.e. greater timespent doing outdoor activities or reduced use of WNV protection) (Gujral et al., 2007; Reisen,2013). However, the summation of research is strongly suggestive of links between irrigation andWNV disease, at least in some areas. No studies evaluating irrigation and WNV disease havebeen carried out in Canada despite the occurrence of large WNV outbreaks in Canada's prairieregions. Specific work is also needed evaluating how differing types of irrigation differentiallyimpact WNV incidence (Eisen et al., 2010).2.9 Viral Introduction and SpreadThe introduction of WNV into North America may have occurred through several mechanisms.Bird migration is one possibility. White storks that died on a migratory stopover in Israel in1998 were shown to have the same WNV strain that caused disease in New York city in 1999(Giladi et al., 2001; Lanciotti et al., 2000; Malkinson et al., 2002)8. Several species of migratingpasserines exhibit migratory behaviours while infectious in laboratory settings (Owen et al.,2006), and viral introduction into North America via migration or storm-driven bird movementhas been hypothesized (Rappole & Hubalek, 2003). The human-mediated importation of in-fected birds from Israel has been hypothesized because of the generally short viremic periodfor most bird species (Rappole & Hubalek, 2003; Zeller & Schuffenecker, 2004). Other possiblemodes of introduction include transcontinental mosquito movement by airplanes or ship (Giladiet al., 2001; Rappole et al., 2000; Zeller & Schuffenecker, 2004), or via infected humans travellingfrom Israel (Gubler, 2007). However, this latter mode of introduction is unlikely, given humans'low viremia.Once WNV arrived in North America, is likely spread cross-continental via a combinationof bird and mosquito movement. WNV infection has been found in birds migrating along theAtlantic and Mississippi fly-ways (Dusek et al., 2009), and the continued westward movement ofthe virus roughly matched the distribution of known migratory flyways within the continentalUS (Gubler, 2007). However, the rate of expansion was less than would be expected by migratingbirds that can travel up to hundreds of kilometres in a day, and the observed cross-continentalmovement was perpendicular to individual north-south migration routes (Rappole & Hubalek,2003). It has therefore been suggested that the initial western expansion was driven regionallyby corvids, with the southern expansion driven by migrating passerines (Artsob et al., 2009).8Subsequent phylogenetic analysis indicate that the strain from Israel did not necessarily seed the NorthAmerican outbreak, but instead that both strains shared a common but unknown source location (May et al.,2011)262.10. Summary of Ecological and Climate Drivers of WNV IncidenceThe common grackle may have also been involved in the continued expansion of WNV in 2002(Artsob et al., 2009). Mosquitoes may also move the virus regionally, as movement within andbetween three genetically distinct groups of Cx. tarsalis matched known waves of viral dispersal(Venkatesan & Rasgon, 2010). Single Cx. tarsalis females are known to travel several kilometresin a single night (Reisen WK, 1990), and the long-distance wind-borne movement of Cx. specieshas been implicated in the introduction of Japanese encephalitis (Ritchie & Rochester, 2001).However, the primary direction of wind in North America is west-to-east, which contradicts thishypothesis (Reisen, 2013).2.10 Summary of Ecological and Climate Drivers of WNVIncidenceArbovirus transmission requires the co-occurrence of viruses, vectors, and reservoirs. Theseconditions determine the baseline regional susceptibility for WNV transmission (Day, 2001).However, arbovirus amplification and transmission are determined by both the distributionof, and contact between, vectors, hosts, and susceptible populations. Landscape and habitatcharacteristics affect the spatial distribution of vectors, hosts and susceptible populations and theinteraction of these factors, while meteorological conditions facilitate rapid viral amplification bysynchronizing mosquito and avian behaviours (Day, 2001). In the same way that environmentalfactors mediate a genetic predisposition towards disease at an individual level, stochastic climaticprocesses mediate broad-scale ecological risk to trigger disease outbreaks in a given population.The interplay between individual risk factors therefore becomes as important as the individualfactors themselves. Traditional risk factor based methods may not be able to provide systemslevel understanding, and alternative ways of viewing and responding to the system may berequired.Gaining an understanding of the interacting drivers of WNV the system is further challengedbecause of species-specific environmental responses, as well as ecological complexities related toavian and human immunology (Kwan et al., 2012), reservoir die-off (LaDeau et al., 2007), andhuman behaviour (Gujral et al., 2007; Reisen, 2013). This complexity means that care mustbe taken when attempting to apply research findings from one location to another location.Consistent continental drivers of WNV human incidence may exist, yet the body of pertinentresearch reveals a significant variation in the intensity and timing of WNV transmission be-tween regions and across time. Many climatic or ecological associations with WNV incidenceare modified by setting-specific factors. The importance of such context-specific complexitieshas been recognized in the field of wildlife health for years (Hanson, 1988) and has importantimplications for public health because disease forecasting techniques developed and parameter-ized in one region, or at a single point in time, may have limited utility in regions with differentecological conditions, or at another point in time (Chevalier et al., 2013; Gray & Webb, 2014).This is of particular importance for Canada since most WNV research comes from the US,272.11. WNV Risk Prediction and Decision Support Modelswith several regions overrepresented (e.g. Chicago, California, Atlanta) in that body of work.In contrast, Canada is sparsely represented in the WNV literature and additional research onWNV in Canada is sorely needed.This literature review has provided a glimpse into a potential causal pathway for WNVincidence (Figure 2.2), and the studies included in this thesis will investigate whether the pro-posed causal pathway holds in the Western Canada. Although alternative system level analysesmay be required to fully understand the intricacies of transmission, we first need to identifythe regionally important factors that drive amplification in western Canada if we are to makeinformed public health decisions.2.11 WNV Risk Prediction and Decision Support ModelsThe observed variability in WNV incidence, both within and between years, indicates that hu-man risk is not constant. However, determining when and where WNV will emerge is difficultand tools or approaches are required to guide prevention messaging and resource allocations.One approach uses predictive modelling to estimate the risk of future outbreaks. These riskprediction approaches require a detailed understanding of system dynamics, or large compre-hensive datasets on which to parameterize predictive models. A second, more reactive approach,focuses on tracking current conditions to estimate the intensity of regional amplification in orderto inform decision making.Risk of WNV can be thought of as the product of exposure and hazard, where hazard isdefined as the intensity of viral amplification within a region, and exposure is determined bythe contact between infected vectors and humans (Rousseau et al., 2017). A true WNV riskassessment would therefore require detailed information on the distribution, abundance, andinfection rates for key vector species, as well as information on human behaviours that mod-ify exposure. Behavioural factors that affect exposure are nearly impossible for governmentalagencies to monitor in real time. In contrast, WNV disease hazard can be estimated by robustvector surveillance programs.The most proximal representation of hazard is the abundance of infected vectors. Mosquitoinfection rates are typically quantified using either minimum infection rates (MIR) or maximumlikelihood estimates (MLE) of infection (Bustamante & Lord, 2010). MIR estimates use theratio of positive pools to total number of mosquitoes tested, and represent the lower range ofpossible infection rates (Gu et al., 2003). The formula for MIR is:MIR = 1000 x (# positive mosquito pools) / (number of mosquitoes tested)This is contrasted by MLEs which provide the most probable infection rate using testing dataand a probabilistic model (Gu et al., 2003). MLE is considered a better estimate than MIR,which underestimates infection rates when true infection rates are high (Gu et al., 2003).282.11. WNV Risk Prediction and Decision Support ModelsFigure 2.2: Hypothesized causal pathway between observed patterns of WNV incidence andecological, environmental, and weather conditions based on the literature review.292.11. WNV Risk Prediction and Decision Support ModelsVector infection rates alone do not directly capture human risk because they fail to considerthe number of infectious mosquitoes (not all infected mosquitoes are infectious), and reflect onlyone aspect of a complex transmission cycle (Bustamante & Lord, 2010). The density of infectedmosquitoes combines vector abundance with the infection rate for an individual vector species(Gu et al., 2008), while the Vector Index (VI) is a composite measure that combines vectorabundance with maximum likelihood estimates (MLE) of infection rates for multiple species(Nasci & Biggerstaff, 2005).The formula for VI is:∑(Key Species) [Mean Number Collected per Night] x [Proportion Infected]Models suggest that the VI underestimates true infection rates in mosquito populations (Bus-tamante & Lord, 2010). However, the measure has been used to trigger public health actionin Manitoba (Ellis, 2005). It was also associated with human incidence in Colorado (Barkeret al., 2009; Bolling et al., 2009; Kilpatrick & Pape, 2013), California (Kwan et al., 2012), andSaskatchewan, with a lag of 1-2 weeks between peak VI and human incidence (Bolling et al.,2009; Colborn et al., 2013). Retrospective analyses have shown the VI to accurately predict pe-riods of low risk areas in California, but it was only moderately successful at predicting periodsof high risk (Kwan et al., 2012).VI measures do, however, require fine-scale mosquito trapping data, and this can be pro-hibitively expensive for some regions (Hadler et al., 2015). High precision hazard estimatesrequire a sufficient trap density. Evaluations in California show that the precision of infectionestimates improve up to 12-traps /100km2(Healy et al., 2014), and the estimated average per-trap cost per week in California was $72 US dollars (Healy et al., 2014). These costs may beeven higher in low incidence regions because a large number of vectors must be tested to detectviruses when infection rates are low (Gu et al., 2008).While mosquito infection rates remain correlated with hazard, they are not the only signalof circulating virus. Approaches that augment limited mosquito trapping and testing withenvironmental and epidemiological surveillance data may improve our ability to coarsely trackhuman risk and guide public health decision-making. Additional surveillance inputs include deadbird collection and testing, the passive reporting of infected horses, temperature evaluations, andpassive human case reporting (Centers for Disease Control and Prevention, 2013). Tools thatestimate WNV hazard based on a combination of these, and other, environmental and ecologicalconditions provide cost-effective alternatives to large-scale vector-based estimates. In general,these tools can be classified according to the input data they require: only a climatic inputs,a single ecological inputs, or a combination of both. For example, the Dynamic Continuous-Area Space-Time system (DYCAST), which was developed in New York City, estimates humanrisk based solely on the presence of dead birds. This system successfully predicted areas thatdeveloped human cases 13-days prior to illness onset (Theophilides et al., 2003). In California,in 2005, the DYCAST system had an 80.8% sensitivity and 90.6% specificity for predicting302.11. WNV Risk Prediction and Decision Support Modelshuman disease, with high risk cells having an RR 39X greater than low risk cells (Carney et al.,2011). However, the sensitivity and positive predictive value of this system did decrease afterinitial viral establishment, limiting its long-term value (Carney et al., 2011; Kwan et al., 2012).Some researchers have attempted to estimate risk based on climatic data alone. In a cross-city comparison, (El Adlouni et al., 2007) attempted to link the occurrence of WNV in 2002to climate conditions, but showed that temperature anomalies were insufficient to explain earlypatterns of WNV on the east Coast of Canada and in central Canada. Temperature valuesobserved in Toronto in 2000 were well within climate norms and consistent with weather con-ditions that occur every 2-years. This was unlike many other of the hardest hit cities, whichshowed above normal temperatures. DD models have also been applied to Wyoming WNV ac-tivity (2003-2005) in order to estimate periods with temperatures above the 109 DD thresholdfor extrinsic incubation of the virus in Cx. tarsalis (Reisen et al., 2006a; Zou et al., 2007).Researchers were able to predict WNV activity within counties in Wyoming with 91.3%, 65.2%,and 78.3% sensitivity in 2003, 2004 and 2005, respectively. Similarly, DD-based approaches havebeen used as part of WNV surveillance at the BCCDC to produce fine-scale estimates of thespatial distribution of risk, and correctly predicted the Okanagan Valley in BC as a high-riskarea (British Columbia Centre for Disease Control., 2013, hereafter BCCDC, 2013). Regionalclimate models developed in the Chicago region combine temperature and precipitation data topredict the MIR of circulating mosquitoes (Ruiz et al., 2010; Shand et al., 2016). The best modelfit included MIR estimates from 2 weeks previous, but weather-only models explained 66% of thevariation in weekly MIR using temperature and precipitation exclusively (Shand et al., 2016).Mechanistic models that link temperature to viral EIP and the vector gonotrophic period showthat ≥55% of sentinel chicken seroconversions in California fell within areas meeting key dailymean temperature thresholds of 22◦C and 26.7◦C (Hartley et al., 2012). These climate-onlymodels are important approaches for risk estimation because they only require freely availabletemperature data. However, risk estimates are unlikely to have sufficient specificity or positivepredictive value because they only address a single aspect of a complex amplification process.Others risk models have incorporate both ecological and weather conditions into risk estima-tion. Models developed in Saskatchewan combined climate data (temperature and precipitation)and landscape features like tree coverage, landscape coverage and wetland coverage. Using dis-criminate analyses, Epp & Waldner (2009) were able to accurately classify spatial units intohigh, medium or low risk with between 44-67% accuracy, depending on the year (Epp & Wald-ner, 2009). Although this approach was developed to determine the spatial distribution of risk,the researchers noted that it could support decision-making by using early season data to es-timate later season risk. Another Canadian group in Quebec attempted to utilize a complexgeo-simulation approach that simulated the entire transmission process by including climate,vectors, reservoirs, and municipal control data. This approach allowed users to evaluate the im-pact of differing intervention scenarios with some success, even though simplified assumptionsregarding the transmission cycle were required to be made (Bouden et al., 2008).312.11. WNV Risk Prediction and Decision Support ModelsSuch complex geo-simulation approaches may technically become feasible as computationalpower increases. Nonetheless, precisely predicting WNV risk even months in advance may benearly impossible due to the complex multi-species transmission cycle and the important rolethat stochastic weather events play in driving arbovirus transmission (Petersen et al., 2012b). Amore practical and realistic approach for organizations with limited resources may be to moveaway from prediction, and to instead focus on ensuring situational awareness and improvingpreparedness should outbreaks occur (Stephen et al., 2015). The situational awareness model isideally suited for dynamic systems and focuses on 1) understanding the determinants of diseasein space and time (gathering data), 2) comprehending the current situation (synthesizing andinterpreting the collected data), and 3) being able to anticipate future states based on thisinformation (Endsley, 1995; Stubbings et al., 2012). In order for a public health agency todevelop situational awareness, the right data is being collected at the right time, must go tothe right person or group for analysis, and must be applied in useful ways (Toner, 2009). Asituational awareness approach that involves monitoring a series of environmental inputs andthresholds correlated with disease, in combination with expert opinion, may be sufficient toguide the timing and intensity of public health interventions like mosquito larval treatment ormessaging the public. Multi-input based tools may be useful in such a scenario. The CaliforniaMosquito-Borne Virus Risk Assessment provides a strong example of a multi-input tool thattracks changing risk through the WNV season. This tool combines environmental and enzooticinputs to estimate risk and initiate public health action (California Department of Public Health,2017). Specific inputs include weekly temperatures, vector abundance, mosquito infection rates,conversion rates in sentinel chickens, and human/equine cases. Thresholds are assigned to eachinput, with an associated score of 1-5 for each level. Inputs are tracked throughout the seasonand the average across-input score is used to identify a risk-appropriate set of interventions. Thiscomposite risk approach consistently prompted emergency planning prior to human infectionand provided a better estimate of predicted risk than did vector surveillance alone (Kwan et al.,2012). However, epidemic thresholds were often not met until after human cases were detected(Kwan et al., 2012). Similar composite approaches were used in the Florida Mosquito ControlArbovirus Response Plan to WNV (FMCARP-WNV) (Florida Department of Health, 2014),which combines vector sampling data, dead bird surveillance, temperature monitoring, sentinelflock monitoring, and passive case reporting in order to estimate potential human incidencein differing scenarios (Day et al., 2015). The use of such an approach in Florida is especiallyrelevant to BC, given the sporadic inconsistent pattern of WNV in this state (Day et al., 2015).Although there are a range of risk models that aim to provide decision support, we cur-rently lack within-season decision-making support tools for low-incidence areas like BritishColumbia.The tools described above are undoubtedly valuable for the regions in which theywere developed. However, while broad approaches may be shared across regions, the complexityof the WNV system and the importance of context-specific factors necessitates that decisionsupport tools be parameterized in the region where they will be implemented (Gray & Webb,322.11. WNV Risk Prediction and Decision Support Models2014; Ruiz et al., 2010). This is especially important for BC and Washington State because thePacific Northwest region has unique ecological conditions not found elsewhere in North Amer-ica. In addition, many risk assessment tools quantify levels of post-spillover risk by effectivelydifferentiating between small and large outbreaks. However, in BC it is necessary to predict thelikelihood of any activity versus an absence of activity; probabilistic estimates focusing on inci-dence are unnecessary because of the rarity of disease (Hayes et al., 2005). Coarsely identifyingthe current stage of the transmission cycle via select surveillance inputs is sufficient to guidepublic health resource allocation for disease prevention. Currently, no decision-making tool in-corporating both environmental and ecological risk factors has been designed for the uniquecontext of WNV in BC - a gap that this thesis hopes to fill.33Chapter 3Methods3.1 WNV Range Expansion and Activity in BC (2009-2015)3.1.1 Broad Analytical ApproachSurveillance and environmental data were combined to frame an exploration of the spatial andtemporal patterns of WNV in BC during the 2009 range expansion and the years following (2010-2015). This analysis represents a secondary analysis using data collected by several agencies.Provincial WNV surveillance data is courtesy of the BCCDC, mosquito abundance and testingwas quantified by the BC Centre for Disease Control Public Health Laboratory (BCCDC-PHL),and temperature and precipitation data were obtained from the Canadian National WeatherService, Environment Canada (Environment Canada, 2016). This analysis takes advantageof data collected as part of WNV surveillance and provincial climate monitoring, but all thesummary measures that are presented here were calculated from raw data and did not entailany analyses or calculations performed by others.The objectives of this analysis was to compare environmental and ecological conditions inyears with-and-without WNV in BC in order to identify: 1) the determinants of transmissionalong the northern and western range of WNV transmission, and 2) the factors that facilitatedthe range expansion into BC in 2009. The absence of historic transmission in BC limited thecapacity of multivariate modelling to identify environmental and ecological conditions associatedwith disease. Instead, basic descriptive epidemiology of data on humans, horses, and infectedvectors was used to characterize WNV in BC (Szklo et al., 2007). The first 7 years of provincialWNV activity are described with each year between 2009 and 2015 classified as having, or nothaving, regional transmission. Spatial and temporal trends in regional activity were evaluated inrelation to: 1) mosquito abundance and infection rates, and 2) climate conditions, with specificcomparisons made between years and locations with and without detected transmission. Themajority of the analysis focused 2009, when the range expansion occurred, and the years thatfollowed. Limited analyses were conducted on the period 2003-2008 to help understand theconditions that prevented WNV establishment prior to 2009.3.1.2 Study AreaThe province of BC is an ecologically, climatically, and geomorphologically diverse area covering947,000 km2that include a lengthy coastline, high mountain ranges, and a desert region (Figure343.1. WNV Range Expansion and Activity in BC (2009-2015)3.1). BC is the most geologically, climatically, and biologically diverse province in Canada (Far-ley, 1979), and shares more features in common with Washington State than with neighbouringAlberta. The province is dominated by vast regions of temperate forests in mountainous areas>1,000 m above sea level (Campbell & Branch, 1990). The coast and mountains ecoprovince9, has warm summers and rainy winters with the mildest temperatures in Canada. Mean dailytemperatures usually remain above freezing year-round (Environment Canada, 2016). Coastalregions receive >1,100 mm of rain per year as moisture-laden air from the Pacific Ocean risesabove the Coast Mountain Range, resulting in orographic precipitation. Most of the majorpopulation centres in BC are found in the Georgian Depression ecoprovince, which is locatedbetween the southern coastal mountains and the northern cascades (Demarchi, 2011). Thisregions is heavily urbanized relative to the rest of the province. The northern half of BC iscomprised of sub-boreal interior, boreal plains, northern boreal mountains, and taiga plains, allof which are characterized by forested landscape and cooler temperatures than the province'ssouthern locations. In contrast, the southern interior of the province is part of the semiaridsteppe highlands (Demarchi, 2011), which have near desert-like conditions, including hot, drysummers, cool winters, and an average rainfall of 260 mm per year (Environment Canada, 2016).The total population of BC in 2016 was 4,516,000, with 2,504,000 living in the Greater Van-couver area (Statistics Canada, 2012). BC is split into 5 geographic Health Authorities (HAs):Vancouver Island Health (VIHA), Vancouver Coastal Health (VCH), Fraser Health Authority(FHA), Northern Health Authority (NHA), and the Interior Health Authority (IHA) Figure3.1). The most populated communities in BC include Vancouver and Surrey (VCHA), Victoriaand Nanaimo (VIHA), Abbotsford, White Rock, Chilliwack (FHA), Kelowna, Kamloops andVernon (IHA), and Prince George (NHA).BC contains two bird migration routes. The Pacific Flyway flows through the coastal andcentral regions of the province, and the eastern edge of the Central Flyway touches on BC'seastern boarder (Campbell & Branch, 1990). Hot spots of summer avian diversity occur in theGeorgia Depression (Nanaimo Lowlands and Fraser Lowlands), the southern Interior (OkanaganValley, Vernon and Kamloops), the southern interior mountains, and parts of the central Interior(Williams Lake) (Campbell & Branch, 1990). Both flyways are oriented north-south, althoughthese routes reflect only approximations and bird movement is not restricted to them (Boere &Stroud, 2006). Some species like sea ducks engage in significant east-west movement (Sea DuckJoint Venture, 2015).9The hierarchical Ecoregion classification scheme splits BC into 208 individual units according to macroclimateprocesses and physiography. These units include 4 ecodomains, 7 ecodivisions, 11 ecoprovinces, 47 ecoregions,and 139 ecosections according to macroclimatic processes and physiography (Demarchi, 2011). The ecoprovincewas used here, and categorizes regions based on "consistent climatic processes, oceanography, relief and regionallandforms" (Demarchi, 2011)353.1. WNV Range Expansion and Activity in BC (2009-2015)Figure 3.1: Select cities (lower case) in BC, Canada, and regional HAs (upper case). Eachregional HA undertakes WNV surveillance under the guidance and recommendations of the BCCentre for Disease Control. The Interior Health Authority and the Fraser Health Authorityare shaded darker grey because they had the most intense WNV surveillance programs. Thedashed oval encompasses the Okanagan Valley, which has been the primary focal point of WNVactivity in BC. WA, Washington, USA; ID, Idaho, USA; MT, Montana; AK, Alaska, USA .363.1. WNV Range Expansion and Activity in BC (2009-2015)3.1.3 WNV Provincial Surveillance DataThe BC Centre for Disease Control (BCCDC) and the BCCDC-PHL partnered with regionalHAs, municipalities, and regional governments to conduct human surveillance, mosquito sam-pling, and dead corvid surveillance between 2003 and 2015.Between 2003 and 2008, all HAs ran mosquito surveillance, and traps were placed as farnorth as Fort St. John (Latitude: 56.25, Longitude: -120.85). CDC light traps (Model 512;John W. Hock Company, Gainesville, FL, USA) baited with dry ice were run 1 or 2 nights perweek from June through September. Mosquito traps were placed two meters above the ground inedge habitat in locations with as many of the following characteristics as possible: near a watersource, in or near a wooded or bushy area, in or near a populous area, near bird roosting sites,in an area safe from vandalism (British Columbia Centre for Disease Control, 2005). Gravidtraps were used at select locations in 2003, 2004 and 2005, but data from these traps was notincluded in this analyses. In 2008, in response to the prolonged absence of the virus, in 2008mosquito traps were spatially restricted to at or below 50◦N latitude, and included traps inVIHA, FHA, VCHA and IHA (Figure 3.1). The primary northern locations were Kamloops,Salmon Arm (IHA) and Courtney (VIHA) (see British Columbia Centre for Disease Control.(2007, 2008, 2009, 2010b, 2011, 2012, 2013, 2014) for maps of trapping locations). These areaswere identified as having higher WNV risk than the rest of the province (British Columbia Centrefor Disease Control., 2009, hereafter BCCDC, 2009). In 2009, an additional 16 traps were runin the southern Okanagan Valley for this thesis work in order to supplement the 91 traps theprovince was already operating in the area. Two traps were placed in each trapping site and runtwice per week in order to improve the regional sampling intensity (see British Columbia Centrefor Disease Control. (2009) for enhanced trapping locations). This targeted trapping effortwas conducted in collaboration with the Osoyoos Indian Band (OIB), with trapping locationschosen based on local knowledge provided by the band members. In 2010, the provincial trappingschema was again shifted with traps restricted to the IHA and FHA (British Columbia Centrefor Disease Control., 2010b, hereafter BCCDC, 2010). In 2011, a modification to the trappingmethodology was implemented in five locations in the Fraser Valley with multiple traps run ina single location and one of these elevated within the canopy of a tree (BCCDC, 2011). Thiswas done specifically to improve the catch of Cx. pipiens (Anderson et al., 2004; Drummondet al., 2006). In 2012, vector sampling in the Fraser Valley was more intensely focused in areasof suspected risk and with abundant vector abundance (British Columbia Centre for DiseaseControl., 2012, hereafter BCCDC, 2012). Vector surveillance was stopped entirely in BC in2014.All mosquitoes collected as part of the BCCDC-led provincial surveillance were sent to theBCCDC-PHL, where they were sorted by sex, identified to genus and/or species, and pooled toa maximum of 50 mosquitoes per pool. All pools of female Culex mosquitoes were homogenized,and RNA was extracted using a QIAamp Viral RNA Mini Kit (QIAGEN, Valencia, CA, USA);other mosquito species were not tested for WNV. RNA extracts were subjected to an in-house-373.1. WNV Range Expansion and Activity in BC (2009-2015)developed TaqMan real-time reverse transcription PCR (RT-PCR) specific for the 3' non-codingregion and the nonstructural protein 5' of the WNV genome. Positive pools were confirmed byusing a second TaqMan real-time RT-PCR specific for the WNV envelope protein (Eisler et al.,2004; Lanciotti et al., 2000). The TaqMan assay has been shown to have a 100% sensitivity toLineage 1 WNV for positive mosquito pools, although no information on specificity was provided(Lanciotti et al., 2000).Passive dead corvid surveillance was conducted by regional HAs and included: 1) online re-porting of dead bird sightings by the public, and/or 2) the collection of dead corvids, which werethen submitted for testing at the BC Ministry of Agriculture and Lands Animal Health Centre(AHC). Oropharyngeal swabs from dead birds were screened for WNV by using the VecTest(Microgenics Corporation, Fremont, CA, USA); RT-PCR was used as the confirmatory test onpooled tissues from birds that were suspected to be positive (Stone et al., 2004). The sensitivityof this test varies by species, and it performs well for American Crows, Blue Jays, and HouseSparrows, but poorly for raptors, Mourning Doves (Zenaida macroura), Fish Crows (Crovus os-sifragus) and American Robins. Specificity was nearly 98%, with false positives in Gray Catbirds(Dumatella carolinensis) and Green Herons (Butorides virescens) accounting for the remaining2% (Stone et al., 2004). Testing of corvids submitted as part of the BCCDC-driven provincialWNV surveillance stopped in 2014, but online reporting by community partners continued until2015. WNV continues to be a reportable animal disease under the Reportable and NotifiableDisease Regulation of the Animal Health act, and the provincial Chief Veterinary Officer (CVO)can share WNV reports with the Provincial Health Officer under the Information Sharing ofZoonotic Communicable Disease Reports (British Columbia Centre for Disease, 2015). In ad-dition, birds with clinical illness compatible with WNV continue to be reported through theCanadian Wildlife Health Cooperative (CWHC) National WNV Surveillance Program (Cana-dian Wildlife Health Cooperative, 2017) and in some instances dead corvids are still submittedto the Animal Health Centre (AHC), located in Abbotsford, BC.WNV infection in humans is a reportable disease in BC and information about probablehuman cases is communicated to the requesting physician and to public health officials. Then, acase questionnaire is administered by BCCDC epidemiologists to collect information on symp-toms, travel history, and the likely mode of transmission. Cases are classified as West Nilenon-NS or WNNS according to the Public Health Agency of Canada's case definition (PublicHealth Agency Canada, 2009). Cases are further categorized as probable or confirmed based onthe associated laboratory testing (see Appendix A.1.1 for case definitions). All suspected humanwere tested for WNV immunoglobulin M (IgM) and IgG by using ELISA (FOCUS Technologies,Cypress, CA, USA) and acute-phase and convalescent-phase serum samples; in house hemag-glutination inhibition (HI) tests were conducted when needed (Lanciotti et al., 2000). Thisapproach had a documented sensitivity of 98% for Lineage 1 in human tissue (Lanciotti et al.,2000). Positive test results from the BCCDC PHMRL are sent to the National MicrobiologyLaboratory in Winnipeg, Manitoba, Canada, for confirmatory plaque reduction neutralization383.1. WNV Range Expansion and Activity in BC (2009-2015)testing. All organ transplant donors have cerebrospinal fluid (CSF), plasma and organ sam-ples tested by reverse transcriptase-polymerase chain reaction (PCR). Cerebrospinal fluid frompatients admitted to hospital for encephalitis or meningoencephalitis was also tested for WNVby PCR. Canadian Blood Services provided the BCCDC with a weekly summary of all blooddonations and associated tests to help identify asymptomatic infection (Busch et al., 2006). TheBCCDC in turn provided Canadian Blood Services (CBS) with information on all individualstested for WNV (BCCDC, 2011).Historically, confirmed horse infections were reported to the BCCDC through a mutualagreement between the CVO and the AHC. Since 2010, the AHC has offered serological testingfor horses with compatible clinical symptoms (BCCDC, 2011). WNV in horses has been identi-fied by veterinarians using ELISA, serum neutralization, and/or plaque-reduction neutralizationtest. Horses suspected of dying from WNV were brought to the AHC for diagnostic necropsy.As of 2015, WNV became a reportable disease in animals under the Animal Health Act (BritishColumbia Government, 2014), with cases reported by local veterinarians to the CVO and inturn to the PHO (British Columbia Centre for Disease, 2015). Although equine vaccinationsare available in BC, their application is not widespread, with the exception of horses that travelto the United States and must be vaccinated according to US requirements. Horse surveillancehas been identified as a useful indicator of spillover in regions with limited human populations,but is impacted by equine vaccination rates (Epp et al., 2008).3.1.4 Vector Data AnalysisAverage trap catch for each Epidemiological Week, and for each season, was calculated as thetotal number of female mosquitoes of each species caught in an epidemiological week or season,divided by the total trap nights for that given period. This method ensured comparabilityin instances where a trap was run more than once per week. An Epidemiological Week is astandardization measure allowing for cross-year comparisons (Arias, 2006). EpidemiologicalWeeks were assigned using the Epi Week R package, with Sunday chosen as the first day ofthe week (Zhao, 2016). Summary statistics were calculated for weeks 25-36 (approximately thelast week of June to the first week of September) to control for variable start dates for seasonaltrapping. In addition, this period encompasses the months during which viral amplificationand WNV spillover occur in temperate climates (Campbell et al., 2002). Focusing on femalemosquitoes for surveillance purposes has been evaluated in other regions (Reisen & Lothrop,1995) and makes sense biologically as only female mosquitoes seek blood meals.Quantifying the mean, median and max weekly trap catch helps quantify variations betweentraps and across time. The maximum weekly trap catch was calculated to help identify localizedincreases in vector abundance that may not be reflected in regional averages. Weekly averagetrap catch was calculated for: 1) all traps run in BC, and 2) separately for the IHA and theFHA. These two areas received special attention in the data analysis because they generallyhad the largest vector populations (British Columbia Centre for Disease Control., 2010b), the393.1. WNV Range Expansion and Activity in BC (2009-2015)largest number of traps, and the longest duration of consistent sampling across the surveillanceperiod. Mean, median, and max weekly mosquito catch were also calculated for a subset oftraps in the Okanagan Valley and Fraser Valley that sampled a consistent location for at least5 years between 2005-2013. This stable trap subset was selected to minimize the impact of themicro-climate when evaluating the yearly averages (Godsey et al., 2013). A consistent set oftrapping locations (n=10-12 depending on year) were identified for the Fraser Valley, however,changes to mosquito surveillance programs in 2011 resulted in two separate groups of stabletraps: one set of traps for the period 2005-2009 (n=13), and one for 2010 and 2013 (n=9).Vector abundance data was combined with lab testing data to estimates population vectorinfection rates using minimum infection rates (MIR) and maximum likelihood rates (MLE)(Bustamante & Lord, 2010). Both measures were calculated over a two-week periods using freesoftware created at the CDC, Atlanta (Biggerstaff, 2017). Aggregating to 2-week periods isthought to improve the precision of vector infection rates because of the low number of vectorscaught in some weeks, especially in 2013. VI was also calculated for regions and years withpositive mosquito pools. Additional details on MIR, MLE and VI calculations was described inSection 2.11).3.1.5 Temperature Analysis and Degree-Day CalculationsSeveral analytic strategies were used to compare and characterize temperature conditions be-tween locations, within one season, and across years. These include 1) degree day calculationsthat quantify the cumulative heat experienced by mosquitoes in relation to the extrinsic in-cubation period of WNV in Cx. tarsalis (Reisen et al., 2006b; Wilson & Barnett, 1983), 2)a graphical comparison of minimum daily temperatures in relation to key viral developmentalthresholds (Reisen et al., 2006b), 3) an analysis of yearly precipitation anomalies in relationto 20-years of historic data, visualized via the use of heat maps, and 4) an evaluation of meandaily temperatures in relation to key thresholds associated with a semi-validated relationshipsbetween viral development and biting activity (Hartley et al., 2012).The single-sign method (Allen, 1976), which provides the most accurate degree-day quantifi-cation when daily temperatures are below the minimum development threshold (Pruess, 1983),was used with a 14.3◦C base for all DD calculations. Two DD summaries were calculated. First,the total accumulated DDs between January 1 and August 1 for the years 2003−2009 were calcu-lated for the following BC communities; Abbotsford (FSA) , Cranbrook (IHA), Creston (IHA),Kamloops (IHA), Kelowna (IHA), Penticton (IHA), Osoyoos (IHA), Prince George (NHA), Van-couver (VCHA), and Victoria (IHA) (see Figure 3.1). Community comparisons helps identifythe spatial distribution of cumulative heat across the province. Second, cumulative DDs overthe preceding 14 days were calculated and compared to the 109-DD threshold (Reisen et al.,2006a), which has been used elsewhere to delineate spatial risk (Zou et al., 2007).Daily minimum temperatures were visualized for key locations using locally weighted least-squares regression smoothers (LOESS) to compare daily minimum temperature to that of the 20-403.2. Associations between WNV incidence and ecological conditions in Saskatchewanyear average. LOESS smoothers are a form of locally weighted scatter plot smoothers (Cleveland& Devlin, 1988). The span parameter, which determines the smoothness of the line, was set at0.8. LOESS analyses focused on the Okanagan communities of Osoyoos, Penticton and Kelownabecause these locations represent the closest primary population centres to the documented focalpoints of WNV activity in BC in 2009, 2010 and 2013. Monthly average minimum temperaturesand total monthly precipitation anomalies were visualized using heat maps that graphicallyshow the difference between monthly average temperatures and the 20-year average through theuse of Z-scores (Zar, 1999). Heat maps have been used for anomaly visualization in other fieldsincluding climate science and genetics (Pleil et al., 2011; Sebastião et al., 2009).Mean daily temperatures were compared across years and locations in relation to biologicalthresholds identified by Hartley et al. (2012). Data was visualized using a novel graphing schemato identify periods with temperatures below key thresholds. Mean daily temperatures wereplotted across the WNV season using a grid structure, with color identifying when daily meantemperatures were above threshold values. This graphical approach clearly identified periodswith below threshold temperatures that may be hidden when evaluating aggregate temperaturedata over weeks or months.Finally, daily estimates of temperature and precipitation, and weekly estimates of vectorabundance, were plotted along a common axis for Osoyoos and Kelowna in order to evaluatethe relative timing of temperature, daily precipitation and Cx. tarsalis abundance in relationto dates of positive cases or mosquito pools. Temporal correlation measures using varying lagperiods could have been utilized, but the lack of WNV in the regions provided insufficient datawith which to run such models.All temperature and precipitation data came from the Canadian National Weather Service,Environment Canada (Environment Canada, 2016). Where possible, a single weather stationwas used across all years for a given location. Airport weather stations were chosen whenpossible. In some years, a specific weather station was unavailable or suffered from long periodsof missing data. In such instances, the nearest weather station was chosen as a replacement. Allcalculations and figure creations were performed using the R Statistical package (R Core Team,2016).3.2 Associations between WNV incidence and ecologicalconditions in Saskatchewan3.2.1 Study AreaSaskatchewan is sparsely populated with an average population density of only 1.7 individualsper square kilometre (Anderson, 2011) and a total population of 1,049,701 people as of October1st, 2010, and 1,098,352 as of October 1st 2016 (Government of Saskatchewan, 2012). The largestcommunities in Saskatchewan are Saskatoon (2006 pop: 202,340, 2011 pop: 222,246, 2016: 246,376 ) and Regina (2006 pop: 179,246, 2011 pop: 193,150, 2016 pop: 215,106 ) (Government of413.2. Associations between WNV incidence and ecological conditions in SaskatchewanSaskatchewan, 2012). Saskatchewan is the most agriculturally dominant province in Canada,containing 38.5% of all farms in the country (Statistics Canada, 2008), and has been one of theprimary focal points of WNV activity in Canada since 2005 (Public Health Agency Canada,2009). The combination of high WNV incidence and abundant agriculture made Saskatchewanan ideal location in which to evaluate links between irrigation and WNV incidence. In addition,Saskatchewan is split into 296 approximately equal sized Rural Municipalities (RM) that aresimilar in size to US counties, allowing for a comparison with previous research in the US(Ezenwa et al., 2007; Miramontes et al., 2006; Swaddle & Calos, 2008). The use of similar sizedanalytical units minimizes problems stemming from analyzing spatial units of varying sizes.Saskatchewan is a relatively ecologically uniform province. Southern Saskatchewan is domi-nated by 3 ecoregions: moist mixed grasslands, mixed grasslands, and cypress uplands (Marshallet al., 1999). The central portion of the province contains mid-boreal uplands, mid-boreal low-lands, and the boreal transition, while northern Saskatchewan is dominated by the ChurchillRiver Uplands and the Athabasca plain (Marshall et al., 1999). Saskatchewan has warm sum-mer temperatures and severely cold winters, with temperatures varying by up to 65◦C in asingle year (Cote, 2011). Saskatchewan is also a relatively dry province with precipitation ac-cumulating primarily in the summer months, although there can be significant year-to-yearvariability in precipitation (Cote, 2011). Between 2000 and 2010, the minimum reported cu-mulative precipitation for a RM between April and Sept was 68mm, while the maximum was865mm (Environment Canada, 2016).3.2.2 Descriptive StatisticsHistograms of primary explanatory variables were used to check basic assumptions of normality.Right skewed variables were log transformed prior to their inclusion in statistical models. Keybivariate relationships between predictors and outcome variables were examined using scatter-plots. Key predictors were also evaluated between groups defined by WNV incidence. ThreeWNV incidence categories were used for the irrigation analysis, and two for the avian analysis.This difference stems from lower sample size in the latter analysis. For the irrigation analysis,incidence categories were defined as disease absence, and low, medium and high incidence weredefined by tertiles of observed incidence for RMs with at least one case of disease. For the aviananalysis, incidence categories were defined as disease absence, and low and high incidence weredefined by the median of observed incidence for RMs with at least one case of disease. A cor-relation matrix of key predictors was used to identify highly correlated variables when buildingstatistical models. It is important to note that the addition of highly correlated variables to amodel may result in artificially elevated Standard Errors (SE).Maps showing the distribution of WNV incidence and key predictors at the RM level werecreated using ArcMap (ERSI, 2011). Such mapping allows for a coarse evaluation of changingspatial disease patterns over time in relation to key ecological predictors. Maps were specificallycreated to show spatial variation in WNV incidence for both outbreak years, the distribution of423.2. Associations between WNV incidence and ecological conditions in Saskatchewanirrigation within the province, and the distribution of incidence rates and case counts in relationto ecoregions (Ecological Stratification Working Group, 1995).3.2.3 Modelling ApproachWNV incidence in humans at the RM level was modelled as count data using a combinationof Generalized Linear Models (GLM) and Generalized Linear Mixed Models (GLMM). Thesemodels provided a general form of multiple linear regression that allowed for non-normal dis-tributions of the error terms (Nelder & Wedderburn, 1972). Poisson and negative binomialregression were used to evaluate the relationship between WNV incidence at the RM level and:1) total amount of irrigated landscape and 2) various measures of avian community structure.Variable effect estimates in these regression approaches represented the expected increase in thelog count for a one-unit increase in the predictor variable (Szklo et al., 2007). RM populationwas used as an offset variable to control for the variation in exposure or for scaling populationvariation in the case of disease incidence (Yiannakoulias et al., 2006).In contrast to GLMs, GLMMs allow for the inclusion of random effects, which are variablesthat can be considered to have been drawn from a population of potential values, or structuralfactors related to the experimental design (eg. blocking variables, identification of repeatedmeasures, etc) (Bickel, 2007). The inclusion of random effects was used to control of the non-independence or correlation between points resulting from data clustering - a failure to accountfor this type of correlation results in incorrect standard errors and p values (Bickel, 2007; Bolkeret al., 2009). Random effects were used to account for the correlation between measurementstaken at the same location in multiple years. Random effects were incorporated into modelsusing the LMe4 package (Bates et al., 2015) in the R statistical language (R Core Team, 2016).Overdispersion was determined by comparing the residual deviance to the residual degreesof freedom. Models were validated using model diagnostics that included plots of residualdeviance against fitted values, qqplots of the standardized deviance residuals, and influenceestimates derived from Cook Distance (Cook, 1977). Points with high influence or leverage wereinvestigated to determine whether they had characteristics that warranted their exclusion fromthe model. High influence observations have a disproportionate influence on model estimates(Cook, 1977). High leverage points effectively reflect outliers, and are points found far from thegeneral mass of the data and may have significant influence if they differ from the general bestfit regression line. (Rousseeuw & van Zomeren, 1990). High leverage points were only removedif there was a significant biological rationale for doing so.An estimation of variance explained by a given GLM was calculated by comparing theresidual and null deviance with the following formula:1− Null DevianceResidual Deviance(3.1)In both the irrigation and avian modelling exercises, multiple models were created with the433.2. Associations between WNV incidence and ecological conditions in Saskatchewanaim of comparing subtle measures of select ecological characteristics. Akaike information criteria(AIC) (Akaike, 1974) was used to guide model selection. AIC is favoured for model selectionover the historic use of R2and the significance values of individual variables, since the formeris a description parameter that provides no information relating to model selection, while thevariable significance approach is inappropriate for model selection when comparing a series ofmodels (Burnham & Anderson, 2002). Key predictors (not covariates) were not included ifthe addition of a single measure reduced AIC values by less than 2 points, as there is limitedevidence to warrant their inclusion (Burnham & Anderson, 2002). AIC has been used previouslyfor model selection in studies of WNV (Diuk-Wasser et al., 2006; McKenzie & Goulet, 2010) andheat mortality in the US (Curriero et al., 2002). The Vuong's non-nested hypothesis test wasused to compare models with different forms, specifically standard poisson, negative binomial,and zero inflated regression (Vuong, 1989).Outcome Measures: WNV Incidence and Human Case DataThe primary outcome of this analysis was the annual incidence of non-travel related WNV atthe RM level during the 2003-2007 period. WNNS is believed to provide the most unbiased rep-resentation of disease patterns because the severe manifestations of the disease are less likely tobe associated with patterns of hospital access or hospital-seeking behaviours, and because physi-cians are less likely to report or test for the presence of the virus in those with WN non-NS only(Kwan et al., 2012). However, the rarity of this condition limits the power of statistical analyses,especially for sub-analyses looking at avian community structure. In addition, focusing only onWNNS will potentially misrepresent the true disease patterns given the known distribution ofWNNS to WN non-NS to asymptomatic infection, especially in RMs with small populations. Acombined measure of WNV infection was therefore used that includes all case of WNNS, WNnon-NS and asymptomatic WNV reported to Saskatchewan Public Health. Similar compositemeasures were used previously for evaluations of environment-disease associations (Wimberlyet al., 2008).Incidence measures were derived using a line-listing of WNV cases reported to the SaskatchewanMinistry of Health that linked each case to the nearest community (Saskatchewan Ministry ofHealth, 2017). Cases were plotted to community of residence and then aggregated to the RMlevel using a spatial join function in ArcGIS (ERSI, 2011). While it remains impossible to en-sure that exposure occurred in the community identified in the line-listing, analyzing incidenceat the RM level increases the likelihood that exposure has occurred in the relevant analyticalunit (Winters et al., 2010). Geographic information systems (GIS), specifically ArcMap (ERSI,2011) and the R Statistical Package (R Core Team, 2016), were used to link the 2006 census dataand the WNV case data. All aggregation was performed on secure servers at the SaskatchewanMinistry of Health. GIS layers containing the census units and associated population data weredownloaded from the Canadian Census webpage (Statistics Canada, 2012). Human case countswere aggregated up to the identified census units. Cases identified as coming from communities443.2. Associations between WNV incidence and ecological conditions in Saskatchewannot found in the 2006 Census were spatially identified using Google Maps and manually matchedto their associated RM.Measures of IrrigationThe total acres of irrigated landscape within each RM for 2003-2007 was provided by the Irri-gation Branch of the Saskatchewan Ministry of Agriculture (Branch, 2011). Irrigation in eachRM is categorized as follows: wheel-move systems, pivot systems, linear systems, miscellaneoussprinklers, surface irrigation, 200mm duty back flood, miscellaneous back flood, remaining irri-gation types, and the total area of irrigation. Total acreage of irrigated landscape was used asthe primary exposure measure for this analysis. Sub-analyses were performed comparing the ef-fects of 1) sprinkler based irrigation, and 2) surface flooding irrigation. Surface based irrigationsystems are known to result in more severe waterlogging than sprinkler-based systems (Branch,2011). Combined sprinkler-based irrigation was classified as the combined total acres of wheelmove systems, pivot systems, linear systems, and miscellaneous sprinkler systems. Total surfaceirrigation was classified as the combined acres of surface irrigation, 200mm duty back flood ir-rigation, and miscellaneous back flood irrigation. The category of remaining irrigation systemswas not included in either of the sub-analyses because detailed information is lacking on whatis included in this irrigation class.Measures of Avian Community StructureAvian data comes from the Breeding Bird Survey (BBS), which is an annual continent widebird survey carried out between May 28th and July 7th (USGS Patuxent Wildlife ResearchCentre, 2001). Approximately 4,100 routes were run in North America - including 88 historicroutes in the province of Saskatchewan - although not all of these routes were run in each year(Environment Canada, 2012). Routes are 24.5 miles long, and volunteers carried out 3-minutepoint counts every 0.5 miles. Inter-observer reliability (Sauer et al., 1994) and potential biasesrelated to sampling along road routes (Keller & Scallan, 1999) were both of concern, althoughthe BBS represents the best large-scale data source for this type of work and has been usedpreviously to examine relationships between avian community structure and WNV incidence(Allan et al., 2008; McKenzie & Goulet, 2010; Swaddle & Calos, 2008). Furthermore, the BBSis carried out during the summer months when WNV amplification is occurring, in contrast toother large scale surveys such as the winter bird count.Spatial data on the location of each BBS route came from detailed .pdf maps provided by theUSGS (USGS Patuxent Wildlife Research Centre, 2001). BBS routes were digitally traced usingArcMap, with the provided maps and Google Earth used as reference points. Avian communitymeasures were linked to 2006 census data by spatially overlaying each BBS routes with thepolitical unit of interest. An ArcMap spatial join was used to link the census data from eachRM to its overlapping BBS route. This resulted in a single BBS route being linked to multipleRMs in some instances, creating a RM-BBS grouping (Figure 3.2). The choice to link BBS453.2. Associations between WNV incidence and ecological conditions in SaskatchewanFigure 3.2: Spatial linkage approach for Breeding Bird survey (BBS) routes and Saskatchewancensus data. Multiple RMs can be linked to a single BBS route. Shaded areas represent a singleRM-BBS grouping. Census information and case counts are aggregated up to the level of thesespatially aggregated units.routes to RMs reduces the pseudo-replication that would result if BBS route information wasapplied to multiple RMs (Hurlbert, 1984). When this occurred, census data and case countswere aggregated across all RMs associated with a given BBS route, resulting in a single measureof WNV incidence for the area surrounding each BBS route for each year. A total of 137 routeswere run from 2003 through 2007, resulting in 137 WNV RM-BBS linkages from 39 unique BBSroutes over the five year period.All avian species present along the BBS routes for Saskatchewan were classified as passerineor non-passerine using information provided by the eBird/Clements Checklist of Birds of theWorld (Clements et al., 2012). Avian community structure along each BBS route was summa-rized by calculating several measures of avian community structure (Table 3.1). These measureswere chosen based on the theory of the dilution effect (Schmidt & Ostfeld, 2001) (Ostfeld &Keesing, 2000b; Randolph & Dobson, 2012; Schmidt & Ostfeld, 2001) (see Section 2.5).Species richness measures the number of species within a given area, but says nothing abouttheir abundance (Hubalek, 2000). Species richness of key subgroups like non-passerines hasbeen evaluated for WNV previously in other locations (Ezenwa et al., 2006). In contrast, theShannon diversity index combines both the number of species with information on the 'evenness'of the community (Hubalek, 2000; Levin et al., 2009). For example, a community with 100individuals of species A and 100 of species B would be considered more even than a communitywith 10 individuals from species A and 1000 from species B. Shannon's measure of diversity ismore sensitive to rare species than other measures (Hubalek, 2000), but was used previously toevaluate the impacts of diversity on WNV (Allan et al., 2008; Hamer et al., 2011). The Shannonindex of diversity is calculated as:H‘ = −∑(pi ∗ ln(pi)) (3.2)463.2. Associations between WNV incidence and ecological conditions in Saskatchewanwhere pi is the proportion of a given species (Hubalek, 2000).To more specifically evaluate the role of these specific avian groups in relation to WNV trans-mission, both abundance and species richness were evaluated for passerines and non-passerines.In addition, the ratio of individual passerines to non-passerines was evaluated, as well as theratio of passerine species to non-passerine species. Doing so evaluated whether the absolute orrelative number of the preferred host has a greater impact on feeding rates, and in turn, onhuman incidence. The independent effect of total birds along a BBS route was also evaluated toensure that observed patterns for individual passerines or non-passerines were not reflective ofthe total avian abundance irrespective of species. This helped to determine if observe patternsresulted from density-dependent feeding patterns, or biodiversity per se. The total number ofrobins was evaluated independently as this species is believed to be especially important toWNV amplification in some settings (Hamer et al., 2008; Kent et al., 2009; Kilpatrick et al.,2006).Finally, community competence - which is a summary measure that quantifies the reservoirpotential of a given avian community by combining information on species abundance and speciesreservoir competency (Allan et al., 2008) - was also evaluated. This measure is calculated as:H =n∑i=0ni ci (3.3)where ni is the number of individuals of a given species, and ci is the reservoir competence ofthat particular species according to Komar et al. (2003) or Komar et al. (2005). It is salientto note that these measures are based only on published literature, and so not all regionallyrelevant species have been necessarily evaluated.The hypothesized effect for each of the described measures can be found below in Table 3.1.473.2.AssociationsbetweenWNVincidenceandecologicalconditionsinSaskatchewanTable 3.1: Measures of avian community structure including a description of each measure, the hypothesized association betweenthe measure and WNV incidence as measured in humans and/or mosquito vectors, and other studies previously using the selectedmeasures.Measure Description Hypothesis PreviousStudySpecies Richness (S) # of species Inc. ↓ as species richness ↑ (Allanet al., 2008;Ezenwaet al., 2006)Shannon's Diversity Index Combines species evenness and species di-versityInc. ↓ as Shannon diversity ↑ (Allan et al.,2008)# Non-Passerine Sp. and #Passerine Sp.Non-passerines are generally poorer reser-voirs than are passerine speciesInc. ↓ as number of non-passerine sp. ↑ (Ezenwaet al., 2006)Ratio # Passerine Sp to Non-Passerine SpRelative measure of passerines sp. to non-passerines sp.Inc. ↑ as proportion of passerine to non-passerine species ↑# Ind. Non-Passerines and #Ind. PasserinesAbundance of competent and less compe-tent WNV reservoirs.Inc. ↓ as abundance non-passerines ↑Ratio Ind. Pass. to Ind. Non-Pass.Measure of the feeding options of themosquito vectorInc. ↑ as ratio passerines to non-passerines↑(Ezenwaet al., 2007)Community Competence1Amplification potential of avian commu-nity based on species reservoir competencyand abundance.Inc. ↑ as cumulative competency ↑ (Allan et al.,2008)1Reservoir competency is based on laboratory work by Komar et al. (2003), and is the product of susceptibility, mean infectiousness, andmean duration (days)483.2. Associations between WNV incidence and ecological conditions in SaskatchewanCovariates/ConfoundersIt was not feasible to control for all factors associated with disease incidence because of thecomplexity of the WNV transmission cycle. The small sample size also limited the numberof variables that could be used in some models, which is a consequence of working at a largespatial scale. In addition, only variables with quality data could be included. As a result, onlythe following confounders or covariates were included: human population density, temperature,cumulative precipitation, population over 50 years of age, and WNV vector control efforts atthe community level.Population density for each RM was calculated by dividing the yearly population estimatesof each RM provided by the Saskatchewan Ministry of Health's Covered Population data set(Government of Saskatchewan, 2010) by estimates of total land area provided by Canada's 2006Agricultural Census (Statistics Canada, 2008).The ratio of the population over 50 was used to control for the effects of age instead of totalpopulation over 50, since the latter is correlated with total RM population (offset variable)and population density. The ratio of the population over 50 was calculated as the sum of allage categories greater than 50 divided by the total number of the under 50 population. Allage data was provided by the Saskatchewan Ministry of Health's Covered Population data set(Government of Saskatchewan, 2010).DDs were computed using a baseline of 14.3◦C (Reisen et al., 2006a) and the single signmethod (Zalom et al., 1983) (see Section 3.1.5). Daily temperature data came from weatherstations spread between Saskatchewan, Alberta, Manitoba, North Dakota, and Montana. Cana-dian temperature data came from Environment Canada (Environment Canada, 2016), whileUS data came from the United States Historical Climatology Network (Menne et al., 2011).Weather stations outside of Saskatchewan were used in this interpolation to minimize edge ef-fects. The R statistical package TimeSeries was used to carry out linear data interpolation fora single weather station with missing daily maximum or minimum temperature data (RmetricsCore Team et al., 2015). DD calculations for each station were analyzed to determine if greaterthan 5% of data was missing between April and September. Minimal DDs were accumulatedprior to April 1stfor any of the referenced weather stations, and missing data from January,February, or March was not considered to be biologically relevant. Similarly, WNV transmis-sion is typically completed by the end of August, and missing temperature data for October,November, or December was also not relevant to within-season mosquito development or viraldevelopment. Therefore, stations with greater than 10 days of missing data in any particularmonth between April-September were removed from the analysis. Finally, only weather stationsbetween 54◦C degrees latitude and 45◦C longitude were included in this analysis in order tolimit the impact of weather stations far from the study location. A total of 159 and 104 weatherstations were used in this spatial interpolation in 2003 and 2007, respectively.Spatial interpolation was used to create a smooth data layer estimating cumulative DDsacross Saskatchewan, allowing a unique DD estimate to be calculated for each RM. This data493.2. Associations between WNV incidence and ecological conditions in Saskatchewaninterpolation provided a coarse estimate of cumulative temperature differences between locationsand between years. Spline interpolations create a mathematical function describing a smoothlayer that fits through the observed data points (Childs, 2004). They require fewer data pointsthan other interpolation methods, result in a more uniform data layer than inverse distanceweighting or kriging (Anderson, 2002), and are ideal for smoothly varying natural conditions liketemperature (Childs, 2004). Each spline data layer was created using the Tension approach, witha weight parameter of 20: high values of this parameter result in a data layer with minimum andmaximum values equivalent to those in the original data set (ERSI, 2011). A separate data layerwas created for each year, and data from the spatial interpolation was summarized for each RMusing the mean, median, minimum, and maximum raster cell values as calculated using the ZonalTable summarization tool in ArcMap (V. 10) (ERSI, 2011). Cumulative DD were also calculatedover a 14-day window for the communities of Estevan, Lloydminster, MeadowLake, Melfort,Moose Jaw, North Battleford, Prince Alberta, Regina, Saskatoon, Swift Current, Weyburn,and Yorkton. These communities were chosen because of their spatial distribution across theprovince and the presence of consistent weather station data. The cumulative DDs were thengraphically analyzed in relation to the 109-DD threshold for the extrinsic incubation period ofWNV in Cx. tarsalis (Reisen et al., 2006a).Monthly cumulative precipitation estimates between April and October were provided foreach RM by the Saskatchewan Governments Agricultural Specialist Branch (Agriculture Knowl-edge Centre, 2011). Not all RM's reported precipitation values in all years, and inverse distanceweighting (IDW) spatial interpolation was used to estimate missing values (Childs, 2004). Cen-troids were calculated for each RM using ArcMap, and IDW referenced the nearest 6 data pointsto create a smoothed data layer estimating cumulative precipitation using a power of 2. Themean raster cell value within each RM was taken as an estimate of total precipitation. Giventhe small sample size, the potential errors introduced by spatial interpolation are outweighed bythe loss of power that would result from eliminating over 150 missing observations. It should benoted that the spatial interpolation methods used here differ from those used for DD interpola-tion because of differences in the underlying nature of the data. The temperature interpolationattempted to create a smooth layer between point locations (weather stations), whereas theinterpolation done here filled in for missing RMs using nearest neighbours.Communities employing dedicatedWNV prevention measures were identified by the SaskatchewanMinistry of Health (Saskatchewan Ministry of Health, 2017). A dichotomous variable represent-ing the presence or absence of community level mosquito control was created for these analyses:RMs containing a community employing dedicated WNV control programs (either larvicidingor adulticiding) were coded 1, with communities lacking a WNV control program coded 0.Irrigation-specific modellingWNV outbreaks occurred in Saskatchewan in 2003 and 2007, and data was analyzed for eachyear individually. The primary explanatory variable of interest in this analysis was acres of503.2. Associations between WNV incidence and ecological conditions in Saskatchewanirrigated landscape. Initial models evaluated associations between incidence and the total acresof irrigated land in an RM. For the initial analysis of total irrigation, poisson, negative binomialand zero inflated models were fit, and AIC and Voung's model selection criteria used to identifythe most appropriate model form (Vuong, 1989). The best fit model was then validated andevaluated using the previously described model diagnostics. Subsequent sub-analyses examinedthe differential effects of surface irrigation compared to sprinkler-based irrigation. Finally, asensitivity analysis was carried out using a dataset excluding RMs with the lowest 10% of totalpopulation in order to ensure that study results were not biased by inflated incidence in RMswith small population sizes. In all models, fixed effects included total precipitation between Apriland October; the cumulative DD estimate using the spline interpolation; the log of the averagepopulation density in each RM; the ratio of the population over 50; community mosquito controlefforts; and; the log value of the irrigation variable of interest. These variables were selecteda priori based on both our current understanding of their role in disease transmission and onprevious research (see Section 3.2.3). This approach minimizes data dredging, which can leadto the creation of a models tailored to the individual data set and thereby minimize the models'generalizability (Burnham & Anderson, 2002). Cumulative DDs were scaled by dividing by 100,and total precipitation was scaled by dividing by 10.Avian-specific modellingThe association between avian community structure and WNV incidence was analyzed using theyears 2003 to 2007 to minimize the impact of small sample size resulting from the low numberof BBS routes run in most years (Figure 3.3). The primary predictor(s) of interest were thepreviously discussed measures of avian community structure. A single model was created foreach community measure, with environmental and epidemiological covariates included to controlfor confounding. The same covariates were included in each model, with only the measure ofcommunity structure varying between candidate models. AIC and the multi-model approach laidout by Burnham & Anderson (2002) were used to select between candidate models. A differenceof two AIC points was used to indicate that the addition of a specific measure of avian communitystructure provides a better model fit than a covariate-only model. A random effect for each RM-BBS grouping was used to account for the correlation between points resulting from the use ofthe same BBS routes across years. The fixed effects in the model include the unique biodiversitymeasure (Shannon diversity index, species richness, number of non-passerine species, the ratioof passerine/non-passerine species, number non-passerines, ratio passerines/non-passerines, andcommunity competence); the ratio of the proportion of the population over 50; the interpolatedmean DD value for a given analytical unit; total precipitation between April and October;and population density. Year was also included as a fixed effect in order to control for inter-annual variation in WNV incidence resulting from unmeasured climatic, immunological andecological variables. Cumulative DDs, precipitation, ratio of the population over 50, total birds,total passerines, total non-passerines, and community competence were scaled to improve model513.2. Associations between WNV incidence and ecological conditions in Saskatchewanconvergence (Bates et al., 2015). Scaling was done by subtracting the mean and dividing by thestandard deviation. No sub-analysis excluding small RMs was performed because of concernsabout limited sample size. Small population size was expected to be a lesser issue than inprevious analyses because of the aggregation of RMs linked to BBS routes. However, residualsand measures of influence were evaluated to determine sufficient model fit. Influence statisticsfor GLMMs were created using the Influence.ME package (Nieuwenhuis et al., 2012).523.2. Associations between WNV incidence and ecological conditions in SaskatchewanFigure 3.3: Schematic outlining the linkages between data sources for the analysis of associationsbetween avian community structure and WNV incidence at the RM level in Saskatchewan.533.3. Decision Support Tool Framework3.3 Decision Support Tool FrameworkThe rarity of WNV in BC, coupled with the complexity of the system and known regionalknowledge gaps, means that model based risk prediction was infeasible. We focused instead ondesigning a tool that can help improve the situational awareness of regional public health agen-cies with respect to WNV spillover in BC. The following framework was used to guide the cre-ation of a WNV decision support tool and is suitable for areas with sporadic WNV transmission(Figure 3.4). First, the proposed hazard classification and general framework were described.This was followed by a discussion of general considerations pertinent to the application of theresearch findings to decisions support input selection, with criteria also identified with which toevaluate surveillance input feasibility. After clarifying the general approach, a limited plan ofevaluation for the resulting decision support tool was presented that takes the historically lowlevels of WNV in BC into account. The tool was then parameterized to the specific context ofWNV in BC. Finally, a scenario-based survey that was distributed to provincial WNV expertsin order to gauge their understanding of the ecological nuances of transmission, determine howtheir perception of risk relates to the underlying amplification cycle, and ascertain their viewon appropriate prevention measures for varying levels of risk.3.3.1 Hazard CategorizationHere the term hazard is used to reflect the regional intensity of the WNV amplification, andhence the environmental viral load in reservoirs and vectors. A simple hierarchical hazardclassification was used that explicitly links hazard to the underlying WNV amplification cycle(Figure 2.1). Note that this hazard schema does not differentiate between varying levels ofpost-spillover risk.The four hazard categories were defined as:• Level 1: Environmental Predisposition : Both the previous years' environmental con-ditions and the severity of previous WNV activity in a region affect current amplificationand transmission through complex community-level ecological processes (e.g. competitiverelease, acquired immunity, etc). The previous years' activity therefore may impact theexpected severity of a given season's WNV activity.• Level 2: Suitable Environmental and Ecological Conditions: No circulating virushas been identified through laboratory confirmation but environmental conditions, specifi-cally temperature and to a lesser degree precipitation, are sufficient for viral amplificationand transmission.• Level 3: Confirmed Circulating Virus: Enzootic transmission has been confirmed bylaboratory confirmation of the virus in birds or mosquitoes. No spillover has been detectedin humans or horses. This is the period in which prevention has its greatest impact onlimiting human disease (Reisen & Brault, 2007).543.3. Decision Support Tool FrameworkFigure 3.4: Framework for the creation of a WNV decision support tool for low incidence settings,including: 1) identification of climatic or ecological inputs, 2) evaluation of surveillance inputsin relation to regional surveillance data sources, 3) creation/choice of a decision support tool,4) hazard classification and 5) hazard-specific prevention recommendations. Sections of theframework are also linked to the situational awareness (SA) levels as identified by Endsley(1995).553.3. Decision Support Tool Framework• Level 4: Confirmed Viral Spillover : The level of circulating virus is sufficient forspillover to dead-end hosts like human and horses (Childs, 2007; Childs & Gordon, 2009;Power & van Marle, 2004). Prevention measures implemented after spillover are lesseffective than those implemented prior to spillover (Figure 2.1).Directly relating hazard to stages of the transmission cycle was appropriate for guiding preven-tion within the BC context for several reasons. First, the combination of data limitations andthe stochastic nature of WNV transmission effectively precluded a precise estimation of risk(Petersen et al., 2012b). Second, clearly linking hazard to distinct stages of the transmissioncycle allowed for clear messaging and a simple coherent prevention strategy. However, this cat-egorization does not differentiate levels of hazard or risk after spillover as is the case for mostrisk indices (Barker et al., 2003; Bolling et al., 2009; California Department of Public Health,2017; Carney et al., 2011; Florida Department of Health, 2014; Kwan et al., 2012; Winters et al.,2010). Such post-spillover differentiation was not needed in BC because the region has only asporadic incidence of WNV.3.3.2 Decision Support MatrixThe proposed decision support tool required that surveillance inputs be linked and summarizedin order to estimate current hazard, with public health responses being tied to the resultinghazard estimates (Figure 3.4). This tool effectively implements the second stage of situationalawareness, and attempts to synthesize information from each surveillance input in order toprovide "comprehension of the current status" of the system (Endsley, 1995). A decision supportmatrix links surveillance input values with the hazard classifications. Similar decision supporttools are used for WNV in California (California Department of Public Health, 2017) and Florida(Florida Department of Health, 2014); the drought response in BC (BCMinistry of Environment,2016), and; the management of complex ecological systems like fisheries (Rice & Rochet, 2005;Rychetnik et al., 2002). However, the proposed matrix approach was simplified to match theproposed hazard classification scheme.The proposed matrix contains surveillance inputs (rows), discrete periods of the WNV season(columns), and input thresholds (within cells). To assess the current stage of amplification andthe associated hazard, each surveillance input should be tracked and observed values shouldbe compared to identified thresholds. Each cell of the matrix is color coded to reflect theestimated hazard level (Figure 3.5) (see Section 3.3.1): light yellow represent conditions in theprevious year that are believed to increase hazard in the upcoming season; yellow indicatessufficient conditions for WNV amplification; orange indicates confirmed zoonotic circulation,and; red represents confirmed viral spillover. The number of thresholds required for transitionbetween hazard categories depends on the current hazard level: movement from no hazard tolow hazard requires at least two thresholds to be met (pale yellow to dark yellow cells), whiletransitions from low to medium hazard, or medium to high hazard requires only a single medium563.3. Decision Support Tool Frameworkor high threshold value be met (orange or red). These differential threshold requirements are setbecause movement from no-to-low hazard depends on the cumulative nature of environmentalconditions favouring transmission, with more positive surveillance inputs representing a greaterrisk of WNV transmission. In contrast, a single positive enzootic surveillance input is deemedsufficient to confirm circulation in vectors or reservoirs (medium hazard), while confirmation ofa single human or equine case is sufficient to confirm spillover (high hazard).Surveillance inputs are to be tracked over the course of the WNV season. The frequency ofsurveillance input monitoring should be determined by the timeliness of key data inputs and bythe resources of the end user. The hazard associated with a specific surveillance input dependsnot only on the chosen threshold value, but also on when it occurs during the WNV season (seeSection 3.3.4). Time is categorized into discrete blocks reflecting the regional ecology of thedisease. The tool, as currently constructed, focuses on spring (June) as a period of early viralramp-up, with summer (July and August) reflecting the peak period of WNV amplification.Conditions in the previous summer were included because they modify baseline transmissionrisk (see Section 2.7.2 and 2.6). However, these antecedent conditions primarily exert their effectsvia more proximal surveillance inputs like mosquito abundance, and are included only to provideearly season insight into the potential severity of the upcoming season. These surveillance inputsmay be removed if those using the tool do not value their inclusion.The tool is envisioned to be scale independent and the spatial resolution was intentionallyleft undefined. Surveillance input values should be averaged over the chosen spatial unit whenappropriate. For example, mosquito abundance data could be averaged over all traps within agiven HA if one were estimating regional-level hazard. The tool could be applied to the entireprovince, individual municipalities, or even communities, although hazard assessment is mostmeaningful at small spatial scales (Winters et al., 2010). However, trade-offs exist between thevalue of fine-scale hazard assessment and the cost of running fine resolution surveillance inputs.The practical reality is that the appropriate spatial scale will be determined by the nature ofinput data and not necessarily be a choice of the user.3.3.3 Hazard Dependent Public Health ActionsA modified and reduced version of a framework used for Lyme disease risk prediction (Quineet al., 2011) was used to clearly link hazard and public health actions. This approach defines:1) what action to take and 2) who/where to focus the prevention measure for each hazardlevel. Similar approaches have been used in the Florida response plan for WNV (Day et al.,2015). The public health interventions presented here were developed through a review ofhistoric intervention measures that have been used provincially, as well as a review of additionalapproaches suggested by the literature (see Appendix C for details). Finally, the resultingdecision tool was not intended to trigger the immediate implementation of the identified publichealth actions. Instead, human risk must be weighed against the costs of potential preventivemeasures, the alternative uses public health uses of those funds that will potentially be allocated573.3. Decision Support Tool FrameworkFigure 3.5: A.) Proposed hazard classification scheme which is linked to stages of WNV ampli-fication and transmission (Section 3.3.1). Color (yellow-to-red) reflect levels of WNV hazard.B.) Proposed template decision tool structure. Surveillance inputs define the rows and discreteblocks of time define the columns.to these measures, and stakeholder response to the implementation of such any measures.3.3.4 Surveillance InputsIdentification of Candidate Surveillance InputsSurveillance inputs are used to identify the current stage of the amplification cycle (Figure 2.1).However, choosing surveillance inputs and associated thresholds requires an understanding ofthe factors that drive amplification, transmission and spillover (Figure 2.1). The literaturereview in Chapter 2 helped to identify the causes of variations in disease incidence, while theresulting causal diagram clarified the complex interactions between predictive factors and disease(Plowright et al., 2008) (Figure 2.2). In addition, surveillance inputs used in decision supporttools from other regions helped further inform the selection of surveillance inputs.Surveillance inputs must be evaluated based on their biological relevance, the methodologicalvalidity of supporting studies, and the logistics of using them for decision support purposes. Thisevaluation is especially important because of WNV's complex causal pathway and the potentialfor residual confounding when evaluating associations between disease and conditions measuredat a large spatial scale. Hill's Criteria can guide the selection of surveillance inputs (Hill, 1965),and the application of these criteria to emerging disease is summarized elsewhere (Plowrightet al., 2008). Several of these criteria have specific relevance to VBD research, particularlythe methods employed to identify key associations, and the role of ecology in determining thegeneralizability of previous research.The ecological nature of WNV transmission makes it difficult to identify the drivers of dis-ease using the traditional risk factor-based epidemiological approaches. The large spatial scaleat which ecological and climate associations manifest means that randomized control studies583.3. Decision Support Tool Frameworkare often impractical (Brighton et al., 2003), sample sizes are small, and true replication orrandomization is difficult (Hobbs, 2003). Retrospective cohort study designs are also oftenimpractical for emerging infectious diseases, at least in the immediate years after viral introduc-tion. Furthermore, the rarity of WNV in many locations limits the types of analyses that canbe done without cross-jurisdictional collaboration and data sharing (Petersen et al., 2012b), yetprivacy implications make such collaboration difficult. Consequently, ecological or case-controlstudies are often used to identify disease drivers, despite their limited ability to differentiatecausative association from residual confounding or ecological bias. This reliance on less rigorousepidemiological designs increases the importance of supporting studies given that the likelihoodof causality for a given association increases with the number of supportive studies. This is es-pecially true if laboratory confirmations or experimental field manipulations are used to confirmfindings from observational studies (Hill, 1965). A 'triangulation' approach combining evidencefrom laboratory studies, field studies and/or model exploration therefore provides the strongestbasis for causality (Plowright et al., 2008).Consideration must also be given to the importance of context for ecological diseases. Incon-sistencies in study findings between locations typically suggest the absence of true associations(Hill, 1965). However, associations identified in a single location may lack generalizability notbecause the relationship is false, but because the ecology of the disease varies according to re-gional context (Ciota et al., 2014; Hanson, 1988; Lambin et al., 2010; Reisen et al., 2006a; Turner,1989). Regional differences can be difficult to quantify and are often masked by the data ag-gregation that is performed in response to the sporadic nature of the disease. The continuedevolution of the disease also means that some associations are specific to unique periods. Theintroduction of WNV into previously unexposed avian and human populations is homologousto a recent infection in the human body, as the way in which ecosystems change in response towidespread viral introduction is similar to how the human body transitions through immuno-logical stages when fighting infection. This may occur via broad avian die-offs which can affectavian community structure (George et al., 2015; Kilpatrick et al., 2007; LaDeau et al., 2007,2008), or in response to the changing immunological status of disease reservoirs (Kwan et al.,2012). Associations evaluated during the early stages of introduction may therefore differ fromthose observed after a disease becomes endemic.Several criteria suggest a transferability of associations across settings. First, experimen-tally validated associations acting at the level of individual reservoirs or vector species (eg.associations between temperature and WNV (see Section 2.7.1) are expected to have greatergeneralizability than distal factors supported solely by observational evidence. This further im-plies that regions with shared vectors and reservoirs will be more likely to have shared drivers ofdisease. However, shared vectors do not guarantee the generalizability of a disease associationbecause of the role of contextual factors like vector and reservoir immunological history, straindifferences, and landscape-modified species interactions (Hanson, 1988).593.3. Decision Support Tool FrameworkEvaluation of Surveillance Input FeasibilitySurveillance inputs must be feasible for use in public health settings, and numerous evaluationcriteria exist (Rice & Rochet, 2005). However, the evaluation of surveillance inputs for hazardassessment should focus on: 1) data quality, 2) the timeliness of data collection and availability,3) spatial coverage, 4) ease of use, and 5) cost (Figure 3.4). Data quality reflects samplingmethodology and acceptable levels of error, and special consideration should be given to theconsistency and appropriateness of sampling methodologies. Changes in sampling approachesover time will affect our ability to monitor changing hazards, while the spatial sampling regimewill determine the geographic range and spatial resolution over which hazard can be assessed.Data timeliness should be evaluated to ensure that data is available for real-time hazard assess-ment. Finally, ease-of-use reflects the logistical constraints for a given data source. Users shouldconsider the data form (e.g. file structure and variable types), as well as the bureaucratic re-quirements for data use. Only costs beyond that dedicated towards baseline WNV surveillancewere considered.Spatial and temporal coverage was classified as Excellent, Sufficient, and Insufficient. Insuf-ficient included surveillance inputs for which an absence of observations limited our ability tomake estimates within each HA. A sufficient surveillance input was one for which an accuratemeasurement could be estimated for each authority, but for which better meaningful varianceestimates might be limited at a HA level. Finally, an excellent surveillance input was one forwhich variance estimates could be calculated for each HA, indicating the possibility of evaluat-ing hazard at a sub-HA level. Analytical requirements were classified as low, medium or high.Low analytical costs reflected inputs that require presence/absence reporting only (eg. posi-tive human case). Medium analytical costs reflected those requiring summarization of observedvalues in relation to historic averages. Finally, high analytical requirements represented datainterpolation or modelling (e.g. degree day calculations). Costs were categorized as free, low orhigh.Threshold SelectionThe sensitivity and specificity of surveillance input thresholds must be considered, where sen-sitivity reflects the consistency with which transmission is confirmed when a surveillance inputthreshold is met (Szklo et al., 2007). An ideal surveillance input would also be specific in that thedisease is absent when when input thresholds are not met (Plowright et al., 2008). The relatedconcept of surveillance inputs being necessary but not sufficient for transmission should also beconsidered. A sufficient surveillance input, if present, guarantees that the disease is present.Conversely, a necessary but insufficient surveillance input would be one that is required fortransmission, but which also requires additional factors to be present.Threshold values are selected to differentiate levels of hazard. Identification of surveillanceinput thresholds relies on information from the literature review, regional decision support toolsused elsewhere, or the analytical results presented here. Input thresholds were selected and603.3. Decision Support Tool Frameworkthen categorized as high, medium, or low threshold certainty depending on the strength ofevidence supporting the selected threshold. High certainty thresholds have values that 1) reflecta lab-confirmed presence or absence of the virus, or 2) are supported by strong evidence andidentified in the region where the decision tool is being applied. Medium certainty exists forthreshold variables that 1) are supported by strong laboratory or experimental evidence, butthat have been parameterized in different ecosystems, or 2) lack experimental evidence but arecorrelated with disease in many locations. In contrast, low certainty thresholds are those withweak evidence, primarily a correlation with low sample size, in the region of interest.3.3.5 User SurveyThe decision support tool presented here is not designed to be predictive. Validating the tool byquantifying the sensitivity, specificity or positive predictive value is therefore not a requirementfor its use, nor is it possible in BC because of the limited enzootic transmission in the province.Other features can, however, be evaluated to identify if further refinement is required. Literaturefrom both public health and environmental science guided the choice of evaluation criteria forboth the tool and for individual inputs (Centers for Disease Control and Prevention, 2012;Jackson et al., 2000). Table 3.2 shows the chosen evaluation criteria as well as the groupsresponsible for the evaluation. Criteria identified as being appraised by the BCCDC/PhD inTable 3.2 were evaluated using the methods identified in the last column.This decision support tool was evaluated with respect to both its practicality and its suit-ability for public health recommendations. Surveillance inputs feasibility was evaluated usingthe criteria identified in the previous paragraph. In addition, identifying the views of provincialpublic health staff can help refine the tool and provide limited validation of specific tool com-ponents. Feedback was elicited from those involved in provincial decision-making with respectto WNV or with the implementation of WNV prevention. This group was described as regionalexperts (RE) and included provincial Medical Health Officers, and Environmental Health offi-cers, as well as provincial epidemiologists and mosquito control experts. Criteria identified asbeing evaluated by REs (4b and 5b) in Table 3.2 were appraised using a scenario-centred survey.Each scenario described a unique set of environmental and ecological conditions that mapped toone of the four hazard levels identified in Section 3.3.1. The primary aim of the survey was toelucidate if respondents perceived risk differently for each unique stage of the WNV transmis-sion cycle. In addition, survey scenarios were designed to identify if respondents considered thefollowing when evaluating provincial hazard: 1) the role of nearby regional WNV transmission,and 2) within-season timing of positive surveillance inputs.For each scenario (see Appendix A.3), respondents were asked to score the perceived riskof human infection (scale: 1-10), and to identify if sufficient information was provided to makea prevention recommendation; if not, respondents identified what additional information wasrequired (Criteria 5a). When possible, the REs were asked to choose their recommended pre-vention measure(s) from a list in order to validate the risk-recommendations linkage identified in613.3. Decision Support Tool Frameworkthe initial draft of the tool (Criteria 4b). The list includes: additional vector surveillance, addi-tional larviciding, personal prevention messaging, corvid surveillance messaging, restrictions tooutdoor activities, and adulticiding. Respondents could suggest alternative prevention measuresthat they considered to be necessary.The survey was reviewed prior to distribution by two ex-Medical Health Officers (MHOs).Once it had been reviewed and approved, it was distributed to REs using the Provincial HealthServices Authority's (PHSA) approved "Fluid Surveys" (Fluid Surveys, 2014). The survey wasdistributed via email to the BCCDC's WNV communication distribution list, which is comprisedof MHOs, EHOs, epidemiologists, and mosquito control contractors involved in the provincialWNV program. A reminder email was distributed 1.5 months after the initial distribution.Survey results were summarized using descriptive statistics. The mean and standard devia-tion (SD) of respondents' risk scores were calculated for each scenario. The SD was calculatedto provide a measure of variation between respondent responses. The proportion of respondentsrecommending a given prevention measure was also calculated for each scenario. Preventionmeasures selected by fewer than 50% of respondents were removed from consideration. Scenar-ios were grouped into categories according to the stage of the amplification cycle and a meanscore was calculated for each group. The category of enzootic transmission in nearby locationswas used to specifically address the value of including surveillance inputs of nearby regionalactivity.623.3.DecisionSupportToolFrameworkTable 3.2: Evaluation criteria for decision support tool. The individual or group doing the evaluating is also identified, as are broadmethods of evaluation.Evaluation Criteria Evaluator Method of Evaluation1) Conceptual and Scientific Relevancea) Appropriateness of Chosen Surveillance Inputs PhD/BCCDC Strength of evidence (Chapter 2, Section 4.3.1)2) Feasibility)a) Spatial Coverage PhD/BCCDC Spatial resolution/extent of datab) Temporal Coverage PhD/BCCDC Prospective and retrospective resolutionc) Analytical Requirements PhD/BCCDC Complexity of required analyticsd) Cost PhD/BCCDC Relative to base surveillance3) Sensitivity/Specificity n/a4) Interpretationa) Chosen Thresholds PhD/BCCDC/REs Strength of evidenceb) Linkage to Recomm. REs Survey, Expert Opinion5) Acceptability and Utilitya) Sufficiency for Prev. Recomm. REs Survey, Expert Opinion63Chapter 4Results4.1 WNV Range Expansion into BC4.1.1 Overview of Provincial and Regional WNV ActivityIn early August 2009, two residents of Kelowna (Latitude: 49.88, Longitude: -119.47) presentedwith symptoms compatible with WNNS; (Table 4.1; Figure 3.1). WNV was initially detectedin both cases by RT-PCR, and 10-days later by enzyme-linked immunosorbent assay (ELISA)and hemagglutination inhibition test (FOCUS Technologies) (Morshed et al., 2011). Both caseswere subsequently confirmed by plaque reduction neutralization test done by the National Mi-crobacteriology Laboratory (NML) in Winnipeg, Canada. Travel histories indicated that neitherperson had been outside of interior BC during the period of potential exposure and that eachhad traveled recently in the southern Okanagan Valley near Osoyoos (Latitude: 49.03, Lon-gitude: -119.46), which is 70−80 km south of Kelowna (Figure 3.1). During the same week,BCCDC-led vector surveillance detected a positive mosquito pool (real-time PCR); 7 more pos-itive mosquito pools were detected over the subsequent 2 weeks. All positive pools came fromthe southern Okanagan Valley and were located up to 35 km apart. Three WNV-positive horseswere reported to the chief veterinarian and the AHC in early September: 2 from the southernOkanagan Valley and 1 from the more eastern Fraser Valley (Figure 3.1). None of the horseshad traveled during their exposure period. With the exception of BC, WNV activity in Canadain 2009 was among the lowest recorded, with only 13 human cases reported nationwide (PublicHealth Agency Canada, 2009). Washington, however, had its greatest WNV activity on recordin 2009 (38 cases in humans and 73 cases in horses), up from previous highs of 3 cases in hu-mans and 41 cases in horses or other mammals in 2008 (Table 4.1, Figure 4.1) (WashingtonState Department of Health, 2017)). Only two human cases were detected in Alberta in 2009.In 2010, a single human case of WNNS was detected in Kelowna, BC; lab confirmation wasagain done by the BCCDC PHL and the NML. Exposure estimated to have occurred in thecentral Okanagan Valley near Penticton (Latitude: 49.50, Longitude: 119.59) (approximately70−80 km north of the WNV activity in 2009). In addition, five corvids submitted from Kelownatested positive for WNV, the first birds ever to be found positive in BC. No positive mosquitopools or horses were detected in BC in 2010. The number of WNV cases in Canada was lowin 2010, with only 4 human cases reported (Public Health Agency Canada, 2009). WashingtonState had only 1 endemically acquired human case in 2010, 2 positive birds, and 126 positivemosquito pools (Washington State Department of Health, 2017). A single human case of WNV644.1. WNV Range Expansion into BCwas detected in Alberta in 2010.A single positive horse was detected in 2011, however questions were raised as to the pos-sibility of travel related infection.This uncertainty, coupled with the absence of all other WNVsurveillance inputs leads to the possibility that no true endemic WNV was detected in 2011.For this reason this year will be excluded from comparisons between seasons with and withoutWNV activity. No positive WNV indicators were detected in 2012. Regional WNV activitywas low in 2011-2012, as Washington State had 5 positive mosquito pools and no human casesin 2011, and 5 positive mosquito pools, 1 infected horse, and 2 in state WNV cases reportedin 2012 (Washington State Department of Health, 2017). In Canada, there were 101 and 428WNV cases in 2011 and 2012, respectively, with 0 human cases in Alberta in 2011 and 9 in 2012(Alberta Health, 2016).In 2013, WNV activity was reported again with a single human case detected in the SouthernOkanagan near Osoyoos BC (Latitude: 49.03, Longitude: -119.47); in addition, one positivecorvid and one positive mosquito pool were also detected in the same region. The presenceof a positive mosquito pool, despite a nearly 10 fold decrease in the number of submittedmosquito pools, coupled with the positive corvid suggest significant regional amplification in2013. In 2013, 18 positive mosquito pools were detected in Washington State, with 2 horsecases reported. Increases were also seen in Alberta, with 21 human cases detected in 2013(Alberta Health, 2016).WNV was not detected in 2014 or 2015, despite continued increases seen in WashingtonState (2014: 12 human cases, 5 horse cases, 80 positive mosquito pools; 2014: 24 human cases,36 positive horses, 7 positive birds and 157 positive mosquito pools). No human cases weredetected in Alberta in 2014. It should be noted that mosquito and corvid surveillance wassignificantly reduced in BC starting in 2013 and stopped completely in 2014.4.1.2 Corvid Surveillance TrendsA total of 6,681 corvids were tested for WNV during 2003-2009; none were positive (Table 1).In 2010, five positive corvids were detected in Kelowna (British Columbia Centre for DiseaseControl., 2010b; Tam & Tsuji, 2016), followed by a single positive corvid detected in 2013 (BC-CDC, 2013). The number of dead corvid submissions generally decrea sed over the evaluationperiod, with a temporary increase in the number of corvids submitted for testing seen in 2010(n=233) after the initial viral detection in the province. However, submissions continued to de-crease after with only 40, 22, and 5 submitted in 2011, 2012, and 2013 respectively. Anecdotalevidence suggests that public participation in dead bird surveillance decreased in response tothe absence of the virus. In addition, regional surveillance strategies were re-evaluation in 2006leading to a decrease in messaging focused on dead bird surveillance.654.1.WNVRangeExpansionintoBCTable 4.1: Summary of BC WNV surveillance activities during the WNV seasons 2004-20014.1Note mosquito surveillance wasreduced to the IHA only in 2013, and was stopped completely in 2014.Type 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014Mosq.Traps 145 189 148 155 98 91 101 75 52 12 10Mosq. 52.657 198.228 394,047 242,215 202,460 181,942 203,753 388,082 175,748 84,458 97,061Mosq. Pools22,980 6,631 2,329 2,568 1,873 2,482 2,092 2,282 1,912 290 236Pos. Pools 0 0 0 0 0 8 0 0 0 1 0Provincialx Cx. pipiens 9.2 7.2 15.4 20.4 28.8 32.0 (28.8) 33.1 38.3 56.3 26.7 x Cx. tarsalis 1.5 2.6 8.9 5.4 3.8 18.2 (16.4) 6.9 13.0 14.7 42.0 IHA3x Cx. pipiens 4.1 4.4 5.8 5.0 2.6 4.0 (6.9) 3.6 15.0 4.7 26.7 x Cx. tarsalis 2.2 6.0 19.5 8.3 1.9 29.3 (12.8) 7.5 33.5 25.4 42.0 FHAx Cx. pipiens 15.0 10.0 23.4 33.6 37.8 44.2 47.4 54.1 68.6 - - x Cx. tarsalis 1.5 1.4 4.5 4.8 4.7 13.9 7.5 6.8 12.1  Bird - Dead CorvidsSighted 1,292 740 605 562 458 398 355 267 238 155 140Submitted 1,437 1,058 803 740 205 144 233 40 22 5 0Positive 0 0 0 0 0 0 5 0 0 1 0Human: BCTested 481 755 239 805 530 379 325 415 438 862 836Positive40 0 0 19 (0) 0 3 (1) 1 0 0 1 0Human: Regional4Washington 0 0 3 0 3 38 2(1) 0 4(2) 1(1) 12(10)Alberta 1 10 40 320 1 2 1 0 9 21 01WNV, West Nile Virus2Values represent average trap catch. Provincial, average trap catch for all traps in BC; IHA, average trap catch from the Interior HealthAuthority, which includes the southern Okanagan Valley; FHA, average trap catch from the Fraser Health Authority (Figure 3.1).3Values represent average trap catch from all traps run in the southern Okanagan Valley in 2009, including 16 traps run in the area aspart of the research project. Values within parentheses represent average trap catch without research traps.4# human cases detected (# locally). Washington cases from (Washington State Department of Health, 2017), Alberta cases from(Alberta Health, 2016).664.1. WNV Range Expansion into BCFigure 4.1: WNV surveillance data from Washington State, 2002-2015. The vertical dotted linesrepresent years with confirmed WNV in British Columbia, Canada.674.1.WNVRangeExpansionintoBCTable 4.2: Summary of mosquito surveillance data for the IHA and the FHA between 2005 and 2013. Data is shown for all traps withinan HA, as well as for a subset of stable traps that are consistent across years. Note mosquito surveillance was reduced to the IHA onlyin 2013.Type 2005 2006 2007 2008 2009 2010 2011 2012 2013IHA: Cx pipiensAll trapsMean (Median) 4.4 (1.6) 5.8 (1.6) 5.0 (1.2) 2.6 (1.0) 4.0 (1.9) 3.6 (2.0) 15.0 (5.0) 4.7 (3.1) 26.7 (24.9)Weekly Trap Max 178 144 100 34 79 55 288 47 240Stable TrapsMean (Median) 5.0 (5.1) 3.5 (1.2) 4.0 (1.5) 2.7 (0.7) 5.9 (3.9) 3.2 (2.4) 21.1 (7.3) 3.3 (2.6) 9.6 (9.0)Weekly Trap Max 50 78 33 34 79 31 288 47 120IHA: Cx. tarsalisAll TrapsMean (Median) 6.0 (2.2) 19.5 (4.0) 8.3 (3.8) 1.9 (1.3) 29.3 (5.3) 7.5 (1.9) 33.5 (4.5) 25.4 (5.5) 42.0 (18.2)Weekly Trap Max 563 640 227 21 842 270 1323 530 924Stable TrapsMean (Median) 13.3 (2.6) 31.7 (9.6) 7.4 (2.1) 1.9 (1.3) 4.8 (2.7) 15.9 (7.6) 20.4 (5.0) 44.3 (7.4) 27.0 (6.6)Weekly Trap 563 576 100 21 39 270 262 530 462FHA: Cx pipiensAll trapsMean (Median) 10.0 (3.5) 23.4 (8.5) 33.6 (14.9) 37.8 (18.8) 44.2 (20.2) 47.4 (21.3) 54.1 (40.0) 68.6 (36.9) Weekly Trap Max 281 345 460 635 766 1452 978 1580 Stable TrapsMean (Median) 21.0 (14.4) 42.9 (29.2) 65.7 (69.0) 68.5 (52.4) 66.4 (38.9) 89.7 (83.8) 68.3 (53.6) 100.9 (42.5) Weekly Trap Max 281 345 460 560 766 571 913 920 FHA: Cx. tarsalisAll trapsMean (Median) 1.4 (0.5) 4.5 (3.0) 4.8 (2.5) 4.7 (2.5) 13.9 (4.9) 7.5 (2.9) 6.8 (4.2) 12.1 (5.1) Weekly Trap Max 61 129 135 110 640 266 143 260 Stable TrapsMean (Median) 3.6 (2.2) 7.9 (8.6) 10.4 (9.6) 11.04 (9.5) 33.3 (15.2) 17.5 (12.4) 9.3 (5.9) 7.7 (5.1) Weekly Trap Max 61 77 135 110 470 266 143 71 684.1. WNV Range Expansion into BC4.1.3 Vector Surveillance DataGeographic and Temporal Patterns in Vector abundanceThe most abundant mosquito species caught via the BCCDC-led mosquito surveillance programbelonged to the genus Aedes, followed by Coquillettidia purturbans and Cx. pipiens (BritishColumbia Centre for Disease Control., 2010b). Mosquito abundance varied spatially in BC,with IHA having the greatest total catch per year because of the abundance of Aedes mosquitoes(Figure B.1).The distribution of key WNV vectors also varied spatially in BC. Cx. pipiens was mostabundant in the Fraser Valley with the average nightly trap catch increasing every year since2004 (Table 4.1, Figure 4.2). In contrast, Cx. tarsalis was most abundant in the provincialinterior and outnumbered Cx. pipiens in regions of BC where WNV has been detected. Cx.pipiens was consistently more abundant in the FHA than Cx. tarsalis was in the provincialinterior.Spatial variation in vector abundance also exists at a smaller spatial scale. Mean trap catchwas almost always greater than median trap catch, meaning select traps captured a greater thanaverage number of mosquitoes (Zar, 1999); this trend was more pronounced for Cx. tarsalis thanfor Cx. pipiens, and greater in FHA than in IHA (Table 4.2). Spatial hotspots of Cx. tarsalisabundance were observed in the Okanagan Valley, with the Osoyoos region consistently havinga greater abundance of this vector than other areas of the Okanagan Valley (Figure 4.2). This isillustrated by comparing averages for the entire IHA, the Osoyoos region alone, and max catchfrom a single trap. In 2009, a large peak in Cx. tarsalis abundance occurred in early August;average trap catch was 82.7 females per night for the entire IHA, 238.8 in the Osoyoos area only,with the greatest single trap catch that week of 842.0 Cx. tarsalis females. In contrast, in 2010the peak period for Cx. tarsalis abundance was the week beginning June 27th, with the averageCx. tarsalis trap catch of 27.4 per trap in the IHA, 139.5 per trap in the Osoyoos region, andthe max single trap catch of 270 mosquitoes.Temporal variation in vector abundance was also observed. The average nightly trap catchfor Cx. pipiens in the Fraser Valley increased every year between 2005 and 2014 (Table 4.2).Within a WNV season, the abundance of this vector consistently increased in mid-June anddecreased at the end of August (Figure 4.2). Trap catch of Cx. tarsalis was generally lowerin the FRA than in the IHA, yet the abundance of this species was greater in 2009 relativeto other years. In contrast to the consistent population patterns for Cx. pipiens in the FHA,Cx. tarsalis showed more temporal variation both between and within years. The average Cx.tarsalis trap catch in 2013 was the highest on record in BC with 2009 showing the third highestaverage (Table 4.1). In 2009, during week 26, 27 and 31, Cx. tarsalis accounted for 31%,49%,and 82% of all mosquitoes caught in the Osoyoos area. However, 16 additional traps placed inthe Okanagan Valley in 2009 likely explains the elevated numbers in that year. Cx. tarsalisalso exhibits more sporadic short-lived peaks during the summer in the IHA than in the FHA694.1. WNV Range Expansion into BC(Table 4.1, Figure 4.2). Peaks in average Cx. tarsalis abundance in the IHA typically occurredin July (Figure 4.2), however, short term increases were also seen in June with counts of 744,628, and 576 observed from select traps in 2006. August peaks in Cx. tarsalis abundance arerare, having only occurred in 2009 and 2005 (568 Cx. tarsalis caught in a single trap in thefirst week of August). Both 2007 and 2008 are unique for not having a noticeable peak in Cx.tarsalis during the WNV season.Mosquito estimates from stable traps showed similar patterns to all trap-averages, althoughaverage abundance was typically greater for stable traps than for all traps (Figure 4.2). Dis-crepancies between all-trap and stable-trap averages were common for Cx. tarsalis, especiallyin the provincial interior. The most meaningful discrepancies between the stable and all-trapaverages occurred in 2009 and 2011, with a late peak in Cx. tarsalis abundance in 2009, and anearly peak in 2011, being missed when measured using stable traps only. The absence of thislarge peak in 2009 drove differences between all-trap and stable-trap estimates differing for Cx.tarsalis (Figure 4.2, Table 4.2).Vector Infection RatesNo WNV positive Culex mosquitoes were detected in BC prior to 2009. Pools of Cx. tarsalistested positive for the virus in both 2009 and 2013. In 2009, WNV reached detectable levelsin Cx. tarsalis populations in late July and peaked between August 9th to August 22nd, 2009(Table 4.3). VI estimate were greater during the July 26-Aug 8 period than in the Aug 9-Aug22 period because of higher vector abundance.Despite the detection of WNV in mosquito vectors in 2009, no positive mosquito poolswere detected in BC in 2010, 2011, 2012, or 2014. However, a single positive mosquito pool wasdetected in BC in 2013 despite only 290 total provincial pools tested. This positive mosquito poolwas collected on August 1st, 2013, approximately 1 week earlier than the positive mosquito poolin 2009. Two week MLE estimates for the period July 28-Aug 10 show MLE infection estimatesin the Southern Okanagan that were higher than peak estimates seen in 2009. Associated VIestimates were also greater in 2013 than in 2009 (Table 4.3).704.1.WNVRangeExpansionintoBCFigure 4.2: Nightly average catch for Cx. pipiens and Cx. tarsalis mosquitoes in the FHA (red) and in the IHA (blue) of BC, Canada,during 2005-2013, using all traps (solid line) and stable traps only (dashed line). Provincial vector surveillance data were aggregated byweek beginning January 1, and the dates provided used Sunday as the first day of each Epidemiological Week.714.1.WNVRangeExpansionintoBCTable 4.3: Maximum Likelihood Estimates (MLE) of 2-week infection rates in Cx. tarsalis mosquitoes, SouthOkanagan Valley, BC, 2009. Here the southern Okanagan Valley represents all traps south of Penticton (Figure3.1).Year Week Individuals Pools Positive pools MIR MLE (95% CL) AverageTrapCatch VI2009Jun 28−Jul 11 1,542 55 0 0.00 0.00 49.8 0.00Jul 12−Jul 25 672 31 0 0.00 0.00 33.6 0.00Jul 26−Aug 8 3,402 98 4 1.18 1.20 (0.39−2.90) 77.4 0.09Aug 9−Aug 22 966 70 4 4.14 4.46 (1.44−10.87) 15.3 0.07Aug 23−Sep 5 143 7 0 0.00 0.00 17.1 0.002013Jun 30−Jul 13 668 21 0 0.00 0.00 83.5 0.00Jul 14−Jul 27 563 18 0 0.00 0.00 70.4 0.00Jul 28−Aug 10 179 10 1 3.16 5.83 (0.34−30.91) 25.6 0.15Aug 11−Aug 24 19 7 0 0.00 0.00 2.7 0.00Aug 25−Sep 7 8 4 0 0.00 0.00 2.0 0.00MLE, maximum-likelihood estimate; MIR, minimum infectious rate estimate; CL, confidence interval. Estimates werecalculated using the software PooledInfRate from Centres for Disease Control and Prevention (Biggerstaff, 2017) andrepresent the number of positive mosquitoes per 1,000 tested. Values within parentheses represent 95% CI.724.1. WNV Range Expansion into BC4.1.4 Climatic Determinants of WNV ActivityTemperatureThe most cumulative heat between January 1-August 31st consistently occurred in the GreaterOkanagan Valley (Table 4.4). Osoyoos accumulated the greatest average number of DDs, fol-lowed by Kamloops and Penticton. Coastal communities like Victoria and Vancouver accu-mulated, on average, less than half the DDs accumulated by communities in the OkanaganValley.Years with positive surveillance indicators were warmer during July and August than thosewithout. More DDs were accumulated in 2009 than any year since 2003-2004 for most com-munities in the province (Table 4.4). For the southern Okanagan communities, only 2009 hadabove average monthly temperatures in June, July and August. Warm weather continued intothe fall and winter of 2009-2010, with average minimum temperatures +2.8, +3.7, +0.6, and+1.5◦C above average in January, February, March and April in Osoyoos (Figure B.2 and B.3).However, spring and summer temperatures returned to normal in 2010 and Osoyoos accumu-lated the fewest DDs in the previous 8 years (n=759). Cooler temperatures continued in 2011and 2012 (Table 4.4); 2011 was particularly cool, with both June and July temperatures inOsoyoos well below the 20-year average (Figure B.2). The summer of 2013 was warmer thanthe 20-year average, with climate patterns similar to 2009 (Figure B.2); minimum temperaturesrose above the historic norm in May in Osoyoos and April in Kelowna, although below averagetemperatures seen in Penticton for much of the summer months (Figure B.3). The hot summerof 2013 was followed by an abnormally cool fall and winter in Osoyoos (Figure B.2). Both 2014and 2015 had a greater than average amount of cumulative heat (Table 4.4); in 2015, mostcommunities, with the exception of Prince George, had the greatest accumulation of DDs in theevaluation period.In most years, the 109 DD threshold for the 14-day cumulative DD moving window wassurpassed in July and August by Penticton, Osoyoos, and Kamloops (Figure 4.3). In fact, 2009saw a period in late July-early August where the 14-day cumulative DD total approached 170 inOsoyoos and Kamloops (Figure 4.3). A threshold of 150 cumulative DDs over a 14-day periodwas surpassed by Osoyoos in 2009, 2014 and 2015, and only briefly in 2012 and 2013. The109-DD threshold was not met by the communities of Victoria, Vancouver, and Abbotsford inmost years.The Okanagan Valley had a greater accumulation of heat in 2014 and 2015 than many otheryears; however 2014 and 2015 both had stretches during the key July-August amplificationwindow where mean daily temperature was below 22◦C (Figure 4.4). In contrast, Osoyoos in2009, 2010 and 2013 had consistent periods of heat above the 22◦C threshold from mid-Julythrough the beginning of August. (Figure 4.4, Figure 4.5).When evaluated in aggregate, those years without WNV activity (2012, 2014, 2015) werecharacterized by both a lower max cumulative DDs and a bimodal pattern with a trough at the734.1. WNV Range Expansion into BCbeginning of August (Figure 4.6). Years with WNV lacked this trough pattern, and also showedgreater daily minimum temperatures from mid-July through mid-August (Figure 4.5).744.1.WNVRangeExpansionintoBCTable 4.4: Cumulative DDs between January 1 − August 31, 2003-2015∗ for select BC communities.Year Cranbrook Creston Osoyoos Penticton Kelowna Kamloops Abbotsford Vancouver Victoria Prince George∗2003 599 700 962 765 779 820 485 408 375 2832004 479 668 993 757 727 880 562 485 425 3612005 409 581 850 676 593 738 481 386 357 2752006 542 700 851 703 692 821 469 366 347 3332007 561 757 859 688 726 738 419 344 311 2732008 475 611 811 651 556 729 386 311 202 2652009 477 661 919 765 649 860 519 420 365 3402010 399 515 759 617 548 684 389 324 277 2852011 401 524 683 556 493 605 340 283 251 1652012 471 589 799 632 560 722 370 318 278 2852013 515 691 912 705 615 795 464 447 331 3602014 542 645 923 729 643 794 483 406 360 3552015 618 797 1027 851 747 917 604 462 447 347Average 499 649 879 700 641 777 459 382 333 302DDs are calculated by using the single-sine method (Pruess, 1983) with a 14.3◦C base (Reisen et al., 2006b). SeeFigure 3.1 for location of selected communities.754.1.WNVRangeExpansionintoBCFigure 4.3: Cumulative DDs (14.3◦base) over the preceding 14 days for select communities in British Columbia. The horizontal linesrepresents 109 and 150 cumulative DDs, with the former estimated as number required to complete the extrinsic incubation period ofWNV in Cx. tarsalis (Reisen et al., 2006b). Thicker lines are used to denote years with confirmed regional detection of WNV.764.1.WNVRangeExpansionintoBCFigure 4.4: Days above 22◦C (grey) and 26.7◦C (black) for select communities in BC (2007-2015). White blocks represent periods below22◦C. Biological mechanisms behind temperature thresholds described in Hartley et al. (2012).774.1. WNV Range Expansion into BC(a) Osoyoos (b) Penticton(c) KelownaFigure 4.5: Minimum daily temperature for Osoyoos, Penticton and Kelowna BC, Canada inyears with (2009, 2010, 2013) and without (2012, 2014, 2015) WNV activity. Lines representLOESS smoothers using a span parameter value of 0.4. The horizontal dotted line at 14.3◦Crepresents minimum estimated temperature required for Cx. tarsalis mosquito development andtransmission (Reisen et al., 2006b).784.1.WNVRangeExpansionintoBCFigure 4.6: Cumulative DDs (14.3◦base) over the preceding 14 days for years with (2009, 2010, 2013) and without (2012, 2014, 2015)WNV detected by provincial surveillance. Communities are identified by color, with the thicker lines used to represent communitiesfrom the Okanagan Valley. The horizontal line represents 109 cumulative DDs, which is estimated to be the number of DDs required tocomplete the extrinsic incubation period of WNV in Cx. tarsalis (Reisen et al., 2006b). Thicker lines are used to the denote the threecommunities found in the Okanagan valley.794.1. WNV Range Expansion into BCPrecipitationThe year 2009 was classified as a drought year by the BC Ministry of the Environment (BCMinistry of Environment, 2016), although July and August both had more precipitation thannormal (Figure 4.7). Similar precipitation patterns were seen in both Penticton and Kelownain that year. Total precipitation in Osoyoos in 2010 was below average in the winter months,above average in the early spring, and below average in July and August (Figure 4.7). Belowaverage precipitation was seen in both July and August in both Penticton and Kelowna. Pre-cipitation was above average in both the winter and spring of 2011 and 2012 in both Pentictonand Kelowna, although the pattern was less pronounced in Osoyoos. In Osoyoos in 2013, pre-cipitation levels were below average for the winter, spring and summer. Penticton and Kelownareceived more precipitation in the spring of 2013 than did Osoyoos, but showed below averageprecipitation in July and August.The total average monthly precipitation differed between years with and without WNV.In Osoyoos, the years with WNV had below average precipitation in January through Marchwith greater than above average precipitation in May-August. The discrepancy between yearswith-and-without WNV is particularly noticeable in August; the average precipitation in WNVyears (2009, 2010, 2013) was 27.4mm, compared to 5.5mm in years without WNV. However,this trend is entirely driven by the abnormally wet August of 2009; wet summers did not occurin either 2010 or 2013 in Osoyoos (Figure 4.7). Above average precipitation is, however, seen inboth Penticton and Kelowna in the spring of 2013.4.1.5 Regional timing of temperature, precipitation, mosquito abundancein relation to WNV spilloverCx. tarsalis abundance in the southern Okanagan appeared to generally peak 1-2 weeks afterlocalized precipitation, with human cases detected between 1-4 weeks after observed peaks invector abundance. This pattern is seen in 2009, 2010 and 2013, but also appears in 2014 and2015 in the absence of viral detection.In 2009, late July precipitation was followed by a 2-week period of elevated heat, during whichCx. tarsalis abundance peaked. Detection of the first positive mosquito pool coincided with thispeak in vector abundance, as did the estimated exposure date for the human cases (Aug 6th).No peaks in Cx. tarsalis abundance were seen in Kelowna. In 2010, Cx. tarsalis abundanceagain peaked after precipitation in early June, with daily minimum temperatures rising abovethe 20-year average in late July (Figure 4.9). The estimated date of human exposure in thisyear was approximately 1-month after peak Cx. tarsalis abundance in the South Okanagan andconcurrent with a smaller peak in vector abundance. Similarly, in 2013, late June precipitationwas followed by increases in Cx. tarsalis abundance, with above average temperatures seenthroughout the June-August period (Figure 4.12). Human exposure was estimated to haveoccurred approximately 3 weeks after initial increases in vector abundance and coincided with804.1. WNV Range Expansion into BC(a) Osoyoos (b) Penticton(c) KelownaFigure 4.7: Heat maps showing Z-scores for total monthly precipitation (mm) for Osoyoos,Penticton and Kelowna. Z-scores represent the number of standard deviations off the 20-yearaverage value, with red representing values above the mean and blue representing values belowthe mean. Vertical dashed lines highlight July and August. Horizontal dashed lines represent2009, 2010 and 2013, the years with documented endemic WNV transmission in BC.814.1. WNV Range Expansion into BCpeak vector abundance. The abundance of both key vectors was lower in Kelowna than in theOsoyoos, yet Cx. tarsalis abundance in was greater than in previous years.No locally circulating virus was detected in 2011 or 2012, despite localized peaks in vectorabundance in the Southern Okanagan in both years. In Osoyoos, Cx. tarsalis abundance wasgreater and more prolonged in 2011 than in years with detected WNV despite low precipitationduring the summer months (Figure 4.10): however, this region did have above average precipita-tion in May (not shown in figure) (Figure 4.7). Vector abundance in 2012 was also greater thanin 2009 or 2011 (Figure 4.10, 4.11), with high amounts of precipitation also observed. Dailyminimum temperatures in both 2011 and 2012 were, however, at or below historic norms withthe exception of a short period of heat in mid July of 2012.Mosquito surveillance was stopped in 2014, so no estimates of vector abundance exist for2014 or 2015.824.1. WNV Range Expansion into BC(a) Osoyoos: Daily minimum temperature (b) Kelowna: Daily minimum temperature(c) Osoyoos: Daily precipitation (mm) (d) Kelowna: Daily precipitation (mm)(e) Osoyoos: Avg. Cx. tarsalis nightly trap catch (f) Kelowna: Avg. Cx. tarsalis nightly trap catchFigure 4.8: Minimum daily temperatures, daily precipitation and average Cx. tarsalis trap catchin 2009 WNV season in Osoyoos and Kelowna, BC. The vertical line is estimated exposure ofhuman case. Dashed lines in figure a) are LOESS smoothers (span=0.4) of daily minimumtemperature for 2009 (red) and the average daily min temperatures (blue) (1997-2015). Thehorizontal dotted line represents 14.3◦C. For mosquito abundance, the red line represents theaverage trap catch for all traps in the Osoyoos region, the green line represents the max trapcatch for an individual trap during that week, and the dashed blue line represents the overallaverage trap catch for the entire IHA.834.1. WNV Range Expansion into BC(a) Osoyoos: Daily minimum temperature (b) Kelowna: Daily minimum temperature(c) Osoyoos: Daily precipitation (mm) (d) Kelowna: Daily precipitation (mm)(e) Osoyoos: Avg. Cx. tarsalis nightly trap catch (f) Kelowna: Avg. Cx. tarsalis nightly trap catchFigure 4.9: Minimum daily temperatures, daily precipitation and average Cx. tarsalis trap catchin 2010 WNV season in Osoyoos and Kelowna, BC. The vertical line is estimated exposure ofhuman case. Dashed lines in figure a) are LOESS smoothers (span=0.4) of daily minimumtemperature for 2010 (red) and the average daily min temperatures (blue) (1997-2015). Thehorizontal dotted line represents 14.3◦C. For mosquito abundance, the red line represents theaverage trap catch for all traps in the Osoyoos region, the green line represents the max trapcatch for an individual trap during that week, and the dashed blue line represents the overallaverage trap catch for the entire IHA.844.1. WNV Range Expansion into BC(a) Osoyoos: Daily minimum temperature (b) Kelowna: Daily minimum temperature(c) Osoyoos: Daily precipitation (mm) (d) Kelowna: Daily precipitation (mm)(e) Osoyoos: Avg. Cx. tarsalis nightly trap catch (f) Kelowna: Avg. Cx. tarsalis nightly trap catchFigure 4.10: Minimum daily temperatures, daily precipitation and average Cx. tarsalis trapcatch in 2011 WNV season in Osoyoos and Kelowna, BC. Dashed lines in figure a) are LOESSsmoothers (span=0.4) of daily minimum temperature for 2011 (red) and the average daily mintemperatures (blue) (1997-2015). The horizontal dotted line represents 14.3◦C. For mosquitoabundance, the red line represents the average trap catch for all traps in the Osoyoos region,the green line represents the max trap catch for an individual trap during that week, and thedashed blue line represents the overall average trap catch for the entire IHA.854.1. WNV Range Expansion into BC(a) Osoyoos: Daily minimum temperature (b) Kelowna: Daily minimum temperature(c) Osoyoos: Daily precipitation (mm) (d) Kelowna: Daily precipitation (mm)(e) Osoyoos: Avg. Cx. tarsalis nightly trap catch (f) Kelowna: Avg. Cx. tarsalis nightly trap catchFigure 4.11: Minimum daily temperatures, daily precipitation and average Cx. tarsalis trapcatch in 2012 WNV season in Osoyoos and Kelowna, BC. Dashed lines in figure a) are LOESSsmoothers (span=0.4) of daily minimum temperature for 2012 (red) and the average daily mintemperatures (blue) (1997-2015). The horizontal dotted line represents 14.3◦C. For mosquitoabundance, the red line represent the average trap catch for all traps in the Osoyoos region,the green line represents the max trap catch for an individual trap during that week, and thedashed blue line represents the overall average trap catch for the entire IHA.864.1. WNV Range Expansion into BC(a) Osoyoos: Daily minimum temperature (b) Kelowna: Daily minimum temperature(c) Osoyoos: Daily precipitation (mm) (d) Kelowna: Daily precipitation (mm)(e) Osoyoos: Avg. Cx. tarsalis nightly trap catch (f) Kelowna: Avg. Cx. tarsalis nightly trap catchFigure 4.12: Minimum daily temperatures, daily precipitation and average Cx. tarsalis trapcatch in 2013 WNV season in Osoyoos and Kelowna, BC. The vertical line is estimated exposureof human case. Dashed lines in figure a) are LOESS smoothers (span=0.4) of daily minimumtemperature for 2013 (red) and the average daily min temperatures (blue) (1997-2015). Thehorizontal dotted line represents 14.3◦C. For mosquito abundance, the red line represent theaverage trap catch for all traps in the Osoyoos region, the green line represents the max trapcatch for an individual trap during that week, and the dashed blue line represents the overallaverage trap catch for the entire IHA.874.2. WNV incidence and ecological association in Saskatchewan, Canada4.2 WNV incidence and ecological association in Saskatchewan,Canada4.2.1 IrrigationDescriptive analyses and Summary ResultsSaskatchewan had more WNV cases between 2003 and 2009 than any province in Canada(Figure 4.13), with the majority of cases occurring during the outbreaks of 2003 and 2007. Atotal of 941 confirmed cases of WNV were detected in Saskatchewan in 2003, of which 63 wereneurological, 858 were non-neurological, and 10 were asymptomatic. In 2007, a total of 1390confirmed cases of WNV were detected in Saskatchewan, of which 113 were neurological, 1167were non-neurological, and 168 were asymptomatic. Note that only 102 cases of WNV werereported in Saskatchewan for all non-outbreak years between 2003-2009.The total number of cases within an RM in a given year varied between 0 and 273 in thetwo outbreak years. WNV cases were distributed over a broader area of the province in 2007,than in 2003 (Figure 4.15), with 152 RMs having at least 1 case in 2003 and and 202 RMsin 2007. The histogram of RM WNV incidence shows many RMs with low WNV incidence,especially in 2003 (Figure 4.14). Population in RMs varied between 59 and 227,740 individuals(median population = 963.50); RM size varied between 445.25 acres and 6473.60 acres (median= 842.34 ha), with geographically larger RMs generally clustered in the northern part of theprovince (Figure A.1). Combined WNV incidence in Saskatchewan varied across RMs from 0to 21.68 cases/1,000 population over both years (Figure 4.14). In 2003, the mean incidencewas 20.43/10,000 population (median: 3.1, max=217.00), while in 2007 the mean incidenceacross RMs was 20.71/10,000 population (median:16.2, max=183.5). RMs with elevated WNVincidence are clustered primarily in the Mixed Grasslands in 2003, and in the Mixed Grasslands,Moist Mixed Grasslands, and Aspen Parklands ecoregions in 2007 (Figure 4.17). Case totalsdecreased from the outbreak of 2003 to that of 2007 in the Mixed Grasslands, while increasingin both the Aspen Grasslands and the Moist Mixed Grassland ecoregions.A total of 329,195.19 acres of land in Saskatchewan was irrigated in 2003, (median: 209.54,IQR: 50.52-872.01). In 2007, 335,351.09 total acres of land were irrigated (median: 161.30, IQR:14.57-625.48). A total of 262 out of 295 RMs had some irrigation in 2003, and 233 had someirrigation in 2007. Although a greater total number of acres were irrigated in 2007 than in 2003,it was restricted to fewer RMs. Irrigation was primarily focused in the south-western cornerof Saskatchewan in both years, and only minor differences in the spatial pattern of irrigationwere seen between 2003 and 2007 (Figure 4.16). Sprinkler irrigation was more abundant inSaskatchewan than was surface based irrigation in both years, although the difference was smallerin 2007 than in 2003 (Table 4.7).Examinations of the cumulative 14-day DDs over the years 2003-2010 show that only theyears 2003 and 2007 cross the 109 DD threshold (Figure 4.20). In 2003, the 109 threshold was884.2. WNV incidence and ecological association in Saskatchewan, Canadamet (or nearly met) for most of August, while 2007 crossed the threshold in mid-July throughmid-August. The maximum 14-day cumulative DDs are greater in 2007 than in 2003, nearlyreaching 175 cumulative DD in early August (Figure 4.20).Figure 4.13: Total WNV cases (WN non-NS and WNNS) for province between 2003-2009 inCanada.Covariates/confounders are evaluated across categories of WNV severity (Figure 4.18). Nostrong trend is seen for the relationship between population density and WNV incidence in2003, although population density in 2007 was highest in RMs with low WNV incidence. Totalprecipitation was lower in RMs with medium and high WNV incidence, with the pattern beingmore pronounced in 2007 than in 2003. Total acres of irrigated landscape was more abundantin RMs with medium and high WNV than in RMs with low WNV incidence for both 2003and 2007, although irrigation was greater in RMs with no WNV than in RMs with low WNVin 2007. The ratio of the population over 50 was greater in RMs with medium or high WNVincidence than in RMs with low WNV incidence in both 2003 and 2007. Finally, RM's withmedium or high WNV incidence had a greater number of accumulated DDs than did RM's withlow WNV incidence. The total cumulative DDs in 2003 were greater than that in 2007.894.2. WNV incidence and ecological association in Saskatchewan, CanadaFigure 4.14: Total WNV incidence (combined WN non-NS and WNNS) for Saskatchewan forthe outbreak years of 2003 and 2007.Table 4.5: Summary table showing the mean (and SD) acres of irrigated land per RM for eachirrigation type in both 2003 and 2007. In addition, the table shows the total acres of irrigatedland across all RMs in 2003 and 2007. Irrigation data was provided by the Irrigation Branch ofthe Saskatchewan Ministry of Agriculture (Branch, 2011).Type Year Mean SD TotalSurface 2003 494.88 1519.90 145989.292007 472.38 1521.25 139353.30Sprinkler 2003 610.78 2389.07 180180.702007 509.25 2401.78 150229.90Remaining Irrigation 2003 9.95 52.18 2935.902007 9.42 51.40 2782.77Total Irrigation 2003 1115.62 3060.21 329106.502007 991.07 3094.63 292365.90904.2. WNV incidence and ecological association in Saskatchewan, CanadaFigure 4.15: Spatial patterns of WNV incidence in the province of Saskatchewan in the outbreakyears of 2003 and 2007.914.2. WNV incidence and ecological association in Saskatchewan, CanadaFigure 4.16: Total acres of irrigated land in each RM in Saskatchewan in 2003 and 2007. Datawas provided by the Irrigation Branch of the Saskatchewan Ministry of Agriculture (Branch,2011).924.2. WNV incidence and ecological association in Saskatchewan, CanadaFigure 4.17: WNV incidence in relation to ecoregions in the province of Saskatchewan, Canada.Cell shading represents WNV incidence, with darker shading representing higher incidence. Thetable provides the total number of cases in each ecoregion for 2003 and 2007.934.2.WNVincidenceandecologicalassociationinSaskatchewan,Canada(a) Log Population Density (b) Total Precipitation (c) Total Irrigation (acres)(d) Ratio of Population Over 50 (e) Cumulative DDsFigure 4.18: Boxplots of distribution of key covariates in groups characterized by zero, low (<12/10000 population), medium (12-32/10,000population), and high (>32.8/10,000 population) WVN incidence.944.2. WNV incidence and ecological association in Saskatchewan, CanadaFigure 4.19: Monthly mean precipitation (mm) in RM's across Saskatchewan in 2003 and 2007.Horizontal line at 50mm is added for easy comparison.954.2.WNVincidenceandecologicalassociationinSaskatchewan,CanadaFigure 4.20: Cumulative DDs (14.3◦base) over the preceding 14 days for select communities in Saskatchewan. The horizontal line represents109 cumulative DDs, which was estimated to be the number of DDs required to complete the extrinsic incubation period of WNV in Cx.tarsalis (Reisen et al., 2006b). Large WNV outbreaks occurred in Saskatchewan in 2003 (947 cases) and 2007 (1456 cases).964.2. WNV incidence and ecological association in Saskatchewan, CanadaModel ResultsTotal Yearly IrrigationPoisson, negative binomial and zero inflated poisson models were created for each year indi-vidually. For both 2003 and 2007, AIC estimates indicate that the negative binomial modelprovided the best fit. For 2003, AIC values were 1696.96, 1054.93, and 1820.33 for the poisson,negative binomial, and zero inflated models, respectively. In 2007, the associated AIC valuesare 1139.52, 1095.76, 1179.16 for the poisson, negative binomial, and zero inflated model fit,respectively. In 2003, Vuong's test of non-nested models also identify the negative binomial toprovide better fit than the poisson model (Z statistic=-3.49, p=0.00023), and the zero inflatedmodel (Z statistic=2.34, 0.0095).Modelling the combined incidence of all forms of WNV disease over both the 2003 and 2007outbreaks shows that irrigation is significantly positively associated with the combined WNVincidence in Saskatchewan in 2003 only (Table 4.6). In 2003, a 1 unit increase in the log ofthe total acres of irrigated land in an RM was associated with a 0.18 increase in the log WNVcount (incidence rate ratio (IRR): 1.19 (CI: 1.11-1.30)). Irrigation was, however, not associatedwith WNV combined incidence for 2007. Total precipitation was negatively associated withWNV combined incidence in 2007 only, although the effect size of precipitation is the samein both years (Table 4.6). Estimates of the total cumulative DDs within a given RM werepositively associated with the log count of WNV incidence in both 2003 and 2007, with theeffect estimate greater in 2003 (Table 4.6). In 2003, an increase in 100 DDs over the course ofthe WNV season was associated with a 0.76 unit increase in the log case count (IRR: 2.14 (CI:1.53-3.05)). In comparison, in 2007 an increase in 100 DDs over the course of the WNV seasonwas associated with a 0.41 unit increase in the log case counts (IRR: 1.50 (CI: 1.30-1.73)). Theratio of the population over the age of 50 within a given RM, as well as the presence of vectorcontrol, were both significantly positively associated with WNV incidence in 2003 only (Table4.6). Population density was negatively associated with WNV incidence in 2003 but not in 20074.6), with a 1 unit increase in the log population density associated with a 0.33 decrease to thelog case counts (IRR: 0.72 (CI: 0.59-0.89)).The irrigation model for 2003 explained 23.1 % of the total model deviance and 19.3%in 2007. Residual deviance divided by the residual degrees of freedom was 0.97 for the 2003model, and 1.16 for the 2007 model, suggesting no overdispersion. An examination of modeldiagnostic plots revealed several high influence and high leverage points in both years. Thesepoints were identified and evaluated with regards to their covariates, with specific attentionpaid to the consistency of high influence points across years. Two particular points showed highleverage across both years, and analysis reveals these points to be the RMs containing Saskatoonand Regina, the only communities in Saskatchewan with populations over 50,000 people. Asubsequent model was run with these points removed, however, model effect estimates remainessentially unchanged indicating that such points had high leverage but not high influence. No974.2. WNV incidence and ecological association in Saskatchewan, CanadaTable 4.6: Effect estimates for a negative binomial model of the association between irrigationand WN disease incidence at the level of Rural Municipality in Saskatchewan for both 2003and 2007. Total irrigation includes surface irrigation, 200mm duty backlog, miscellaneous backflood irrigation, pivot system irrigation, linear system irrigation, miscellaneous sprinkler irriga-tion, and remaining irrigation types. Results are presented for the full dataset, and a reduceddataset with the bottom 10% of population size removed to evaluate potential effects of inflatedincidence.2003 Full Sample Reduced SampleVariable Estimate Std. Error Pr(> |z|) Estimate Std. Error Pr(> |z|)Intercept -11.83 1.27 <0.0001** -12.25 1.36 < 0.0001**Total Precip. -0.03 0.022 0.15 -0.03 0.02 0.21Cumulative DD 0.76 0.15 <0.0001** 0.84 0.16 < 0.001**log(Pop. Dens. ) -0.33 0.11 0.0017 -0.37 0.11 < 0.001**RatioOver50 1.40 0.63 0.027 1.38 0.66 0.037Control 0.81 0.34 0.016 0.82 0.33 0.016log(Irrigation) 0.18 0.042 <0.0001* 0.18 0.043 <0.001**2007 Full Sample Reduced SampleVariable Estimate Std. Error Pr(> |z|) Estimate Std. Error Pr(> |z|)Intercept -7.58 0.60 <0.0001** -7.50 0.61 <0.0001**Total Precip. -0.03 0.01 0.0039** -0.3 0.011 <0.001**Cumulative DD 0.41 0.07 <0.001** 0.40 0.072 <0.001**log(Pop. Dens. ) 0.07 0.06 0.29 - 0.06 0.064 0.41RatioOver50 0.15 028 0.58 0.22 0.29 0.44Control -0.17 0.19 0.36 -0.16 0.18 0.38log(Irrigation) 0.019 0.020 0.35 0.016 0.021 0.42strong patterns were seen in the characteristics of other remaining high influence points, and nobiological or epidemiological reason could be attributed to the points that would warrant theirexclusion.A total of 29 RMs containing the lowest 10% of RM total population were removed fromboth the 2003 and the 2007 dataset in order to test for the possibility of inflated incidence fromRMs with low sample size. This limited sensitivity analysis shows that model results remainessentially unchanged for models created using the full and reduced dataset (Table 4.6 and4.7). The similarity of model findings indicate that observed results were not driven by inflatedincidence. Effect estimates for irrigation parameters were identical in both the full and reduceddata sets for both 2003 and 2007. Even the effect size and significance of the association betweenpopulation density and incidence measured remained similar in both years, despite removingRMs that have low populations densities (Table 4.6 and 4.7). Model AIC values were notpresented for the reduced model because AIC values were comparable only between modelscreated using the identical data sets (Burnham & Anderson, 2002).984.2. WNV incidence and ecological association in Saskatchewan, CanadaTypes of Irrigation Sub-AnalysesSurface irrigation was more strongly associated with WNV incidence than is sprinkler basedirrigation in both 2003 and 2007, although the association between both types of irrigation andincidence was significant only in 2003 (Table 4.7). Using the 2003 dataset, a 1% increase inthe log acres of surface irrigated land was associated with a 0.18 increase in the log case counts(IRR: 1.18 (CI: 1.12-1.32)), with no significant association seen in 2007. Similarly, a 1% increasein acres of sprinkler irrigated land was associated with a 0.09 increase in the log counts in 2003(IRR 1.11 CI: 1.04-1.20) compared to 0.021 in 2007 (IRR: 1.03 (CI: 1.00-1.08).AIC estimates indicate that in 2003 the surface irrigation model provided a better fit tothe observed data than do models containing either sprinkler based irrigation or with totalirrigation (AIC_total_irrigation: 1054.9, AIC_Surface: 1051.5, AIC_sprinkler: 1067.4). How-ever, in 2007 the sprinkler based model provided the best fit (AIC_total_irrigation: 1095.8,AIC_Surface: 1093.7, AIC_sprinkler: 1092.3). In 2003, the surface irrigation model explained24.2% of the total deviance, while the sprinkler irrigation model explained 19.3% of the totaldeviance. In 2007, the sprinkler model explained 18.7% of the total deviance, while the sprinklermodel explained 19.4% of the total deviance.Table 4.7: Fixed effect parameter estimates of the negative binomial model of the associationbetween irrigation and WN disease incidence at the level of Rural Municipality in Saskatchewanfor both 2003 and 2007. Two forms of irrigation were evaluated: surface irrigation (left handside of table) and sprinkler irrigation (right hand side of table). Total surface irrigation includedsurface irrigation, 200mm duty backlog, and miscellaneous back flood irrigation. Total sprinklerirrigation included pivot system irrigation, linear system irrigation, and miscellaneous sprinklerirrigation. The category of "remaining irrigation types" was not included in either sub category.2003 Surface Irrigation (AIC: 1051.5) Sprinkler Irrigation (AIC: 1067.4)Variable Estimate Std. Error Pr(> |z|) Estimate Std. Error Pr(> |z|)Intercept -11.94 0.23 <0.0001** -10.78 1.28 <0.001**Total Precip. -0.02 0.02 0.34 -0.040 0.02 0.075Cumulative DD 0.75 0.14 <0.0001** 0.76 0.15 <0.001**log(Pop. Dens. ) -0.27 0.11 0.011 -0.37 0.11 0.0006RatioOver50 1.50 0.61 0.014 1.05 0.65 0.10Control 0.74 0.34 0.028 0.89 0.35 0.009*log(Irrigation) 0.18 0.039 <0.001 ** 0.090 0.04 0.011*2007 Surface Irrigation (AIC: 1093.7) Sprinkler Irrigation (AIC: 1092.3)Variable Estimate Std. Error Pr(> |z|) Estimate Std. Error Pr(> |z|)Intercept -7.58 0.10 <0.0001** -7.60 0.60 <0.001**Total Precip. -0.03 0.061 0.0057* -0.03 0.010 0.0048*Cumulative DD 0.38 0.066 <0.0001** 0.40 0.071 <0.001**log(Pop. Dens. ) 0.08 0.062 0.21 0.06 0.061 0.31RatioOver50 0.16 0.06 0.58 0.17 0.28 0.54Control -0.19 0.18 0.31 -0.16 0.18 0.37log(Irrigation) 0.036 0.021 0.085 0.021 0.019 0.26994.2. WNV incidence and ecological association in Saskatchewan, Canada4.2.2 Avian Community StructureDescriptive analysesWNV Incidence in the RMs linked to a BBS route (RM-BBS grouping, see Figure 3.2) run inSaskatchewan between 2003 and 2007 ranged between 0.00 and 13.9/1000 population, with amean of 0.649/1000 population. The maximum number of cases associated with a given RM-BBS grouping was 266, with a mean of 4.43 cases grouping. A total of 30, 27, 30, 28, and22 BBS routes were run in 2003, 2004, 2005, 2006, and 2007, respectively. The case totalsassociated with these routes in each of the five years of BBS from 2003-200 were 162, 1, 9, 6,424 in 2003-2007.RM-BBS groupings with high WNV incidence had a lower number of total passerines, ahigher number of non-passerines (with the exception of 2006), and a lower number of totalbirds. Similarly, RM-BBS groupings with high WNV had a slightly lower ratio of passerines tonon-passerines (Figure 4.21). Comparing species measures across categories of disease incidenceshowed that RM-BBS groupings with high WNV incidence were characterized by a low numberof passerine species, a slightly higher number of non-passerines, and a lower number of totalspecies (species richness) (Figure 4.22). Groupings with high WNV incidence also had a lowerratio of passerine to non-passerine species. RM-BBS groupings with higher WNV incidence hada higher Shannon diversity in 2003, but a lower diversity in 2007 (Figure 4.23). No clear patternwas seen when evaluating community competence. Finally, RM-BBS groupings with high WNVincidence had a greater number of robins in 2003 than low incidence areas, but this pattern wasreversed in 2007, mirroring the pattern seen for Shannon diversity (Figure 4.23). Populationdensity, the ratio of individual passerines to non passerines, and the ratio of the number ofpasserine species to non-passerine species were all positively skewed and hence log-transformedprior to entry into statistical models.Model ResultsInitial model fitting for the analysis of associations between avian community structure andWNV incidence was challenged by difficulties with achieving maximum likelihood convergenceof for select models. This limits model stability for some sub-models. Convergence issues werelikely a consequence of limited sample size and the added complexity of including a randomeffect to control for repeated measures across years. Evaluation of residuals from base modelsthat would converge showed the presence of two outliers that differed significantly from theremaining residuals. Removing these two BBS routes greatly improved convergence for theremaining models. These data points were removed not because of any concerns about theirbiological validity, but to facilitate the statistical stability of the models. In addition, the modelstructure was shifted from a negative binomial model to a simplified poisson regression modelto further facilitate convergence as some error messages indicated that estimating the thetaparameter for the negative binomial model is one cause of convergence issues. Only minor1004.2. WNV incidence and ecological association in Saskatchewan, Canada(a) # Passerines (b) # Non Passerines(c) Total Birds (d) Ratio passerines/non-passerinesFigure 4.21: Individual level avian community measures for RM-BBS groupings with no WNVincidence, WNV incidence between 0 and 10.7/10,000 population, and greater than 10.7/10,000population.1014.2. WNV incidence and ecological association in Saskatchewan, Canada(a) # Passerine Species (b) # Non Passerines Species(c) Total Species (d) Ratio passerine species/non-passerines speciesFigure 4.22: Species level avian community measures for RM-BBS groupings with no WNVincidence, WNV incidence between 0 and 10.7/10,000 population, and greater than 10.7/10,000population.1024.2. WNV incidence and ecological association in Saskatchewan, Canada(a) Shannon Diversity (b) Community Competence(c) Number of RobinsFigure 4.23: Avian community structure measures for RM-BBS groupings with no WNV in-cidence, WNV incidence between 0 and 10.7/10,000 population, and greater than 10.7/10,000population.1034.2. WNV incidence and ecological association in Saskatchewan, Canadadifferences were seen between negative binomial results and those from the simplified poisson.Model findings indicate that non-passerines were more strongly associated with increasedWNV incidence than are passerines in Saskatchewan. AIC values show that only four measuresof avian community structure improved model fit beyond a simplified model with covariatesonly, and all contained some measure of non-passerine abundance or richness. The four modelsselected by AIC were the ratio of individual passerines to non-passerines, non-passerine abun-dance, the ratio of passerines species to non-passerines, and the number of non-passerine species(Table 4.8). The addition of passerine abundance, passerine richness, Shannon diversity index,total species richness, community competence, total bird abundance, or the abundance of robinsall failed to improve model fit sufficient to warrant their inclusion. In addition, none of thesemeasures were individually significantly associated with WNV incidence.The best fit model was one including non-passerine abundance (Table 4.8). In this model,a 1 unit increase in the log of the scaled mon-passerine abundance translated to a IRR of 1.54(CI: 1.24-1.91) (Table 4.8). Fixed effects in this model explained 35% of total variation, whilethe total model explained 42% of observed variation. The best fit species richness estimatewas provided by the inclusion of the ratio of passerine species to non-passerine species (Table4.8). The effect estimates for the association between incidence and this ratio is -0.57, whichtranslated to a rate ratio of 0.56 (CI: 0.42-0.76) for a 1 unit increase in the ratio of passerinesto non-passerines. The model including the ratio of passerine species also explained 35% ofvariation with fixed effects, but only 40% of variation with both fixed and random effects.Within the four avian models selected by AIC, effect estimates of the covariates showedgeneral consistency in direction of the effect, although effect estimates did vary between mod-els (Tables 4.8). The cumulative number of DDs was significantly positively associated withWNV incidence in all models, with effect estimates varying between 0.72 and 1.03 in the fourselected models. In the best fit model, a one unit increase in scaled DDs had a IRR of 2.78 (CI1.47-5.28). Precipitation was not significantly associated with WNV incidence for any of thecandidate models although the effect size was consistently negative in all models. Effect sizes forprecipitation were consistent with effect estimates seen in the irrigation model despite the lack ofsignificance. WNV incidence was negatively associated with increasing population density in allavian models, but the relationship was not statistically significant. Effect estimates for the logof population density were also similar to those seen in 2003 in the irrigation model. The ratioof the population over 50 to that under 50 was significantly negatively associated with WNVincidence in three of the four models, with effect estimates for a one unit increase in the scaledratio of -0.44 to -0.49 (Table 4.8). In the best fit model, a 1 unit increase in the scaled ratioof the population over 50 had an IRR of 0.62 (CI: 0.41-0.92). This finding contrasted with theresults of the irrigation analysis in which the association between the ratio of population over 50was significantly positively associated with WNV incidence. Year is included in all avian modelsas a fixed effect, and each year was significantly associated with disease incidence compared tothe 2003 reference category. Effect estimates for 2004 ranged between -2.88 and -3.34, for 20051044.2. WNV incidence and ecological association in Saskatchewan, Canadabetween -1.21 and 1.67, for 2006 between -3.35 and 3.49, and for 2007 between 1.43 and 1.75.Model diagnostics for the best fit model containing non-passerine abundance were furtherevaluated to evaluate model appropriateness (Figure B.7). The best fit model was not overdis-persed, as the residual deviance divided by the residual degrees of freedom was 2.52. Cooksdistance values did, however, reveal several high influence points. Evaluation of these pointsshowed one with high population abundance, while the second had a low abundance but a highnumber of cases. The model was re-run with both points removed to determine their impact oneffect estimates. Population density became significant when evaluated on this reduced datasetwith effect estimates similar to that seen in the 2003 irrigation model. All other variables re-tained similar effect estimates although standard errors and associated p values were smaller inthe reduced data model.1054.2. WNV incidence and ecological association in Saskatchewan, CanadaTable 4.8: Parameter estimates of the fixed effects of a poisson random effect model of the relationshipbetween avian diversity and WNV incidence at the level of rural municipality. AIC values quantify thefit of the model to the observed data.Model Fixed Effects Estimate Std.Error Pr AICNo Community Mea-sure343.2INDIVIDUALABUNDANCE- - - -Total Birds Intercept -8.88 0.49 <0.001 342.2Total Birds 0.25 0.14 0.083DegreeDays 1.01 0.35 0.0046Precipitation -0.11 0.24 0.63RatioOver50 -0.43 0.23 0.06log(PopDensity) -0.29 0.24 0.21Year: 2004 -3.23 1.19 0.0007Year: 2005 -1.39 0.58 0.0016Year: 2006 -3.41 0.49 <0.001Year: 2007 1.68 0.27 <0.001Passerines Intercept -8.62 0.49 <0.001 345.1Ind. Passerines -0.06 0.18 0.75DegreeDays 0.86 0.33 0.016Precipitation -0.097 0.24 0.68RatioOver50 -0.32 0.23 0.16log(PopDensity) -0.23 0.24 0.33Year: 2004 -3.70 1.13 0.0010Year: 2005 -1.82 0.56 0.0014Year: 2006 -3.45 0.48 <0.001Year: 2007 1.56 0.27 <0.001Non-Passerines Intercept -8.87 0.44 <0.001 330.6Ind. Non-Pass. 0.43 0.11 <0.001DegreeDays 1.03 0.33 0.0023Precipitation -0.20 0.25 0.42RatioOver50 -0.48 0.21 0.02log(PopDensity) -0.30 0.20 0.16Year: 2004 -2.88 1.22 0.0019Year: 2005 -1.21 0.56 0.030Year: 2006 -3.49 0.50 <0.001Year: 2007 1.59 0.26 <0.001Ratio Pass./Non-PassIntercept -7.88 0.49 <0.001 333.8Ratio Pass./Non-Pass -0.12 0.18 0.75Continued on next page1064.2. WNV incidence and ecological association in Saskatchewan, CanadaTable 4.8  Continued from previous pageModel Fixed Effects Estimate Std.Error Pr AICDegreeDays 0.72 0.33 0.016Precipitation -0.36 0.24 0.68RatioOver50 -0.42 0.23 0.16log(PopDensity) -0.23 0.24 0.33Year: 2004 -3.34 1.13 0.0010Year: 2005 -1.60 0.56 0.0014Year: 2006 -3.41 0.48 <0.001Year: 2007 1.43 0.27 <0.001Robins Intercept -8.77 0.49 <0.001 344.7Robins 0.012 0.016 0.45DegreeDays 0.88 0.35 0.012Precipitation -0.089 0.24 0.71RatioOver50 -0.33 0.23 0.14log(PopDensity) -0.23 0.24 0.32Year: 2004 -3.66 1.17 0.0018Year: 2005 -1.79 0.54 <0.001Year: 2006 -3.48 0.49 <0.001Year: 2007 1.56 0.26 <0.001NUMBER OFSPECIES- - - -# Passerine Sp. Intercept -7.04 0 1.14 <0.001 342.7# Passerine Sp. -0.04 0.03 0.12DegreeDays 0.70 0.37 0.061Precipitation -0.085 0.24 0.72RatioOver50 -0.28 0.22 0.20log(PopDensity) -0.16 0.23 0.48Year: 2004 -4.03 1.23 0.0011Year: 2005 -2.18 0.63 <0.001Year: 2006 -3.48 0.49 <0.001Year: 2007 1.47 0.28 <0.001# Non - Passerine.Sp.Intercept -10.04 0.64 <0.001 335.2# Non - Pass.. Sp. 0.06 0.018 0.0018DegreeDays 1.00 0.36 0.0038Precipitation -0.19 0.24 0.43RatioOver50 -0.49 0.22 0.026log(PopDensity) -0.29 0.22 0.19Year: 2004 -2.88 1.20 0.016Year: 2005 -1.27 0.54 0.019Year: 2006 -3.35 0.49 <0.001Continued on next page1074.2. WNV incidence and ecological association in Saskatchewan, CanadaTable 4.8  Continued from previous pageModel Fixed Effects Estimate Std.Error Pr AICYear: 2007 1.75 0.27 <0.001Ratio PasserineSpeciesIntercept -7.56 0.54 <0.001 331.4Ratio Pass. Species -0.57 0.15 <0.001DegreeDays 0.78 0.33 0.0018Precipitation -0.24 0.24 0.31RatioOver50 -0.44 0.21 0.033log(PopDensity) -0.22 0.21 0.28Year: 2004 3.18 1.09 0.0036Year: 2005 -1.67 0.52 0.0012Year: 2006 -3.38 0.48 <0.001Year: 2007 1.60 0.25 <0.001COMMUNITYCOMPOSITION- - - -Shannon Diversity Intercept -8.34 1.67 <0.001 345.2Shannon Diversity -0.10 0.49 0.83DegreeDays 0.89 0.36 0.013Precipitation -0.096 0.24 0.69RatioOver50 -0.33 0.23 0.14log(PopDensity) -0.24 0.24 0.32Year: 2004 -3.61 1.16 0.0020Year: 2005 -1.72 0.57 0.0024Year: 2006 -3.44 0.49 <0.001Year: 2007 1.59 0.27 <0.001Community Compe-tenceIntercept -8.74 1.67 <0.001 344.0Community Competence 0.14 0.12 0.26DegreeDays 0.89 0.36 0.013Precipitation -0.074 0.24 0.76RatioOver50 -0.34 0.23 0.13log(PopDensity) -0.28 0.24 0.24Year: 2004 -3.61 1.13 <0.001Year: 2005 -1.83 0.55 <0.001Year: 2006 -3.32 0.50 <0.001Year: 2007 1.70 0.29 <0.0011084.3. An Ecological Framework for WNV Decision Support in BC4.3 An Ecological Framework for WNV Decision Support in BC4.3.1 Evaluation of Candidate Surveillance Inputs for WNV DecisionSupportCandidate surveillance inputs were evaluated on their ability to track the changing intensity ofenzootic amplification in North America using the criteria identified in Section 4.3.1. Surveil-lance inputs were not selected for their ability to delineate spatial patterns of disease. Theliterature review and resulting causal diagram (Chapter 2, Figure 2.2), provided the basis forsurveillance input selection. The tool was designed to be used by regional public health officialsto guide within-season decision making for WNV resource allocation, and this evaluation there-fore focused on surveillance inputs that were components of regional surveillance programs;including weather data (temperature and precipitation), vector abundance, reservoir surveil-lance, and spillover host infection (humans and horses). Landscape features were not includedbecause they are expected to be static within a given season. Surveillance inputs were orderedfrom proximal to distal in the causal chain of human WNV disease (Figure 2.2), and thresholdswere chosen when possible.Horses and HumansRationale: Recent locally acquired infection in humans or horses confirms that amplificationhas reached levels sufficient for spillover. Passive reporting of both human and horse cases hasa high specificity and low sensitivity because mild or asymptomatic infection is not always re-ported. Blood donation surveillance can minimize under-reporting in humans (Vamvakas et al.,2006).Feasibility: WNV in humans has been a reportable disease in BC since 2003, and both theBCCDC and regional MHOs are notified of all WNV cases. In addition, WNV was added tothe routine test panel for blood donations during the WNV season. Horses suspected of havingWNV are currently reported to the Chief Veterinary Officer, who then informs the ProvincialHealth Officer (British Columbia Government, 2017). Hence, spillover reporting adds no ad-ditional costs to provincial surveillance programs. The spatial coverage of equine reporting isaffected by the distribution of provincial veterinarians, patterns of equine testing, reportingtrends, and historic patterns of vaccination. The spatial and temporal resolution of WNV in-fection in horses is therefore lower than that of WNV infection in humans.Threshold: Confirmed non-travel acquired infection in horses or humans indicates high WNVhazard. A single infected blood donation sample from a confirmed resident without recent travelto a known endemic area should raise the risk level to high.1094.3. An Ecological Framework for WNV Decision Support in BCMosquito Abundance and Vector Infection RatesRational: The presence of a single positive mosquito pool indicates regional transmission (un-less movement of infected vectors is suspected) and the number of positive mosquito pools waspredictive of WNV hazard in some settings (Brownstein et al., 2004). VI is the best proximalmeasure of WNV hazard (Section 2.11), but is not feasible for jurisdictions without a vectortesting program (Petersen et al., 2012a). Vector abundance has been considered to be a hazardpredictor in the absence of viral testing. Transmission is unlikely to occur below a currentlyundefined threshold value of vector abundance (Reeves, 1965) and increased vector abundancehas been associated with human disease in some locations (Bolling et al., 2009). However, highvector density does not always result in human disease (Eldridge, 1987) and vector abundancemay not be accurately measured by low intensity vector trapping (Service, 1993). High vectorabundance as measured by mosquito surveillance is neither necessary nor sufficient to drive viralspillover into humans (Hill, 1965) and should not be included in decision support tools.Feasibility: Recent reductions in the spatial coverage and intensity of vector surveillance re-duced our ability to both detect circulating virus and minimize the impacts of stochastic weatheron trap catch. Despite these limitations, VI provides the most proximal estimate of hazard andshould be used to guide decision support when possible. Although the trapping scheme op-erated in BC until 2014 was not ideal, it provided important data on within-season changesto mosquito abundance and likely had sufficient sampling intensity to detect circulating virusin the provincial interior and in the Fraser Valley. However, inconsistent trapping locationsbetween years confound historical comparisons because of the impact of microclimates on trapyield (Drummond et al., 2006; Godsey et al., 2013).Threshold: The presence of a single positive mosquito pool suggests localized transmission andraises hazard from low to medium. Similar thresholds were sufficient for predicting human casesin urban settings in California (Kwan et al., 2012). High VI thresholds suggest the possibilityof undetected infection in humans, horses, or other mammal species, even in the absence ofconfirmed spillover, and may be sufficient to raise hazard from medium to high. A threshold VIof 4 has been used previously in Manitoba and preceded human infection by 2-3 weeks in 2003in Saskatchewan (Ellis, 2005).Reservoirs - Community Structure and Dead Bird SurveillanceRationale: Avian community structure and vector abundance combine to determine the eco-logical potential for amplification. However, significant uncertainty exists as to both the truerelationship between avian community structure and human disease, and the best measure(s)with which to quantify it (see Section 2.5). Dead bird surveillance in BC has historically focusedon corvids, yet confirmation of the virus in any bird species suggest enzootic transmission10.Dead bird surveillance should be included in a decision support tool where feasible, although10The presence of virus in a single dead bird is suggestive of enzootic transmission, but does not confirm localtransmission because of the possibility of birds migration while infected1104.3. An Ecological Framework for WNV Decision Support in BCthe sensitivity of this surveillance input may vary over time with acquired immunity in avianpopulations (Carney et al., 2011; Kwan et al., 2012). Rural-urban biases may limit the valueof this surveillance input in the Southern Okanagan Valley (Ward et al., 2006). Despite theselimitations, confirmation of virus in dead birds indicates likely enzootic circulation (Mostashariet al., 2003), although bird movement after infection elsewhere is impossible to rule out.Feasibility: Quality estimates of reservoir community structure are unavailable at the resolu-tion required for regional decision support. Only 56, 63, and 73 Breeding Bird Survey (BBS)routes were run in all of BC in 2013, 2014, and 2015 respectively (no data is available for 2016)(Environment Canada, 2012). The timeliness of BBS data is also insufficient for real-time riskmonitoring because data is not available until after the WNV transmission season. Dead corvidsreported to the BCCDC via the online reporting tool have not historically been collected forviral testing. While viral detection in any species strongly indicates circulating virus in theenvironment, avian testing is an additional cost to provincial laboratories already suffering frombudget constraints. The provincial dead corvid testing program was therefore stopped in 2015.The public can, however, still report observed dead corvids to public health at the BCCDC,and dead birds can also be reported to the Canadian Wildlife Health Cooperative (CanadianWildlife Health Cooperative, 2017).Threshold: Confirmation of a positive dead bird in the spring or summer suggests enzootictransmission and raises the hazard from low to medium. Predictive density thresholds for deadbird clusters have been identified elsewhere (Eidson et al., 2001), but should not be includedhere without confirmatory testing because bird death may result from other infections.TemperatureRationale: Temperature is the most consistent factor positively associated with WNV activ-ity (see Section 2.7.1). DD estimates represent a cost-efficient method with which to capturethe association between cumulative temperature and the biology of viral development in themosquito vector (Allen, 1976; Zalom et al., 1983; Zou et al., 2007). However, DD thresholds arenecessary but not sufficient for WNV activity and do not reflect the presence of circulating virusin the absence of additional positive surveillance inputs. More importantly, the value of DDsas a surveillance input may vary across the WNV season (Chen et al., 2013). Key temperaturethresholds are met in most years in the Okanagan Valley and alternative threshold values maybe needed (Section 4.1). In regions where the temperature remains above key thresholds formuch of the season, the primary value of temperature analyses may be their ability to identifyperiods when temperature and DD thresholds were insufficient for viral and mosquito replica-tion.Feasibility: Environment Canada operated 217 weather stations in BC as of 2015 (Environ-ment Canada, 2016). The majority of the provincial weather stations were in or near populationcentres, which is appropriate given the focus on human risk. Temperature tracking is feasiblegiven the daily frequency of data collection, the spatial coverage, and the fact that the data is1114.3. An Ecological Framework for WNV Decision Support in BCavailable for free. Timely analysis of this data requires the automation of data managementand DD calculations, which can be done using free software such as R. Prospective temperatureanalysis has been conducted historically at the BCCDC (BCCDC, 2012).Threshold: Multiple temperature surveillance inputs are recommended. First, cumulative DDover a 14-day moving window should be tracked relative to the cumulative heat needed tocomplete the EIP of the WVN in Cx. tarsalis (109 DDs) (Reisen et al., 2006a). However, themajority of high risk communities in BC surpass this threshold in most years. A threshold of 150DDs appears correlated with WNV activity in both BC and Saskatchewan (Chapter 2, Figure4.3), but this threshold is based on limited data and lacks laboratory conformation. A specialfocus for the 14-day moving window analyses should be applied to the mid-July to early-Augustperiod (Section 4.1) because of its apparent importance within the BC context. Further, meandaily temperature should be tracked in relation to thresholds identified in California (Hartleyet al., 2012). Prolonged periods (>1wk) during the summer months with mean temperaturesbelow 22◦C may be sufficient to halt transmission and should be used to lower the currenthazard level.PrecipitationRationale: The association between precipitation and WNV transmission is complex, contextspecific, and mediated by 1) temperature, 2) the amount of regional standing water (e.g. irriga-tion, wetlands, melting snowpack), and 3) regional vector species (See Chapter 2, Section 2.7.2).In addition, the link between precipitation and incidence occurs primarily via the effects of vec-tor breeding sites on vector abundance. Locations with robust vector surveillance programs willhave a direct estimate of vector abundance and therefore do not need proxy measures exceptin non-surveilled areas (see Winters et al. (2010) for example). In contrast, drought impactstransmission in other ways (Shaman et al., 2002, 2010) and retains a predictive value even inthe presence of robust vector surveillance programs.Feasibility: Current drought conditions in BC can be identified through the "Living WaterSmart" agency at the level of Local Health Area (LHA) (British Columbia Government, 2012).This spatial resolution should be sufficient for regional decision support. Localized precipitationdata is freely available from Environment Canada (Environment Canada, 2016).Threshold: The Living Water Smart agency data should be tracked for identification of "verydry" or "drought" classifications. Documentation of current drought is used to indicate suffi-cient conditions, or low risk. If a region is experiencing drought, then local weather in selecthigh risk regions could be monitored for bursts of precipitation which may facilitate mosquitodevelopment and viral amplification.Nearby WNV Activity - Washington State and AlbertaRationale: The lack of evidence confirming the overwintering of infected vectors in BC meansthat the yearly re-introduction for the virus remains a possible explanation of observed patterns1124.3. An Ecological Framework for WNV Decision Support in BCof transmission (Roth et al., 2010). A correlation in inter-year WNV activity between Wash-ington and BC may result from 1) Washington State serving as a source of infected mosquitoesor birds for yearly WNV introduction, and/or 2) climate conditions common to both BC andWashington State that favour viral amplification (Roth et al., 2010). Early detection of WNVactivity in Washington State represents the potential for extended periods of amplification andan increased likelihood of late season spillover. Detection of positive mosquito pools early in theseason is correlated with late season viral activity in other locations (Ginsberg et al., 2010).Feasibility: Epidemiological summaries of WNV data from Washington State are freely avail-able via the Department of Health web site (Washington State Department of Health, 2017).This includes data on positive humans, horses, mosquito pools, and birds by country.Thresholds: A review of historical records fromWashington State in relation to observed WNVactivity in BC suggests the following input thresholds: 1) confirmed WNV in mosquitoes, birds,or mammals prior to Epidemiological Week 26, or 2) >50 cumulative positive mosquito poolsreported in a given season. Little to no WNV activity in Washington state has occurred whenWNV was first detected after the 29th Epidemiological Week, and early detection of positivemosquito pools is a known predictor of subsequent within season spillover (Ginsberg et al.,2010). In 2009, the first positive surveillance input, a positive mosquito pool, was reported atthe end of May, nearly 7 weeks earlier then the first positive pool in previous years (2002-2009)(BCCDC, 2009), (Washington State Department of Health, 2017). Similar early WNV activityin Washington State was also seen in 2010 and 2013, but not in 2011. Week 26 has been conser-vatively chosen as the temporal threshold with which to elevate hazard level from none to low.Surpassing either of these thresholds will only increase the hazard level to low. However, thesethresholds are again identified based on limited data and should be viewed as unvalidated untiladditional data can be collected.Surveillance Input Evaluation SummaryThe majority of surveillance inputs identified here were feasible when BC had a WNV surveil-lance program (Figure 4.24). This was to be expected because only those surveillance inputcollected as part of WNV surveillance were included. Climate analyses have minimal analyticalrequirements and are easily implemented by most public health departments. Furthermore, thedata is timely, free and covers most of the province. Reporting of spillover in humans is provin-cially mandated, and data on equine infection is readily shared between veterinarians and publichealth officials. Data on WNV activity in Washington State is also timely and freely available(Washington State Department of Health, 2017). The primary concern for hazard identificationpurposes was the limited spatial coverage of both vector and avian surveillance, which stemmedfrom the high logistical and laboratory costs of mosquito trapping and corvid testing. Althoughthis tool was not specifically designed to delineate spatial risk, the spatial coverage of key datawill dictate where it can be applied. In recent years, vector surveillance in BC was intentionallyfocused on regions with elevated risk, so the tool retained value in the areas where it is most1134.3. An Ecological Framework for WNV Decision Support in BCneeded. However, the 2015 termination of the mosquito surveillance program has limited deci-sion support inputs, as mosquito trapping and testing provide the most proximate indicator ofWNV risk.Limitations do exist for several key threshold values. The VI threshold and thresholds forsurveillance inputs from Washington State are based on limited data and should be viewed asunvalidated until additional data is collected. The remaining thresholds have greater certaintybecause they do not represent cutoff points in continuous variables, but instead are simple pres-ence/absence flags of confirmed viral circulation and hence do not require additional validation.Figure 4.24: Feasibility evaluation for decision support inputs. Criteria include spatial andtemporal coverage, analytical requirements, and cost.4.3.2 Draft Decision Support ToolThe final decision support matrix was tailored to both the regional ecology of BC and to theprovincial surveillance program (Figure 4.25). Rows represent the six select surveillance inputs(Section 4.3.1), with threshold values identified in each box . Human and horse surveillancewere separated to account for the inclusion of blood donation data for human cases. Columnsrepresent the four periods of previous summer, winter, spring, and current summer.The final tool contains three unique threshold values that indicate a transition from no riskto low risk: a positive surveillance input from Washington State prior to July 1st; the 109and 150 two-week cumulative DD threshold, and; the presence of documented regional drought.Only three surveillance inputs reflecting medium risk were identified: the presence of a singlepositive mosquito pool, the presence of an infected corvid (or other dead bird), and significant1144.3. An Ecological Framework for WNV Decision Support in BCactivity in Washington state. A positive mosquito pool detected in May or June indicates highspillover risk, even in the absence of disease in humans or horses, because it would indicatethe potential for prolonged amplification and spillover. There are 3 additional threshold valuesthat indicate a high risk of spillover. First, spillover and transition to high risk is confirmed bya regionally acquired horse or human case, or by detection of a positive WNV blood sample.Second, a VI of 4 detected in any region should raise the hazard level to high, as this thresholdhas been used in other jurisdictions to trigger action and was included because a VI estimate ator above the threshold indicates a high probability of undocumented spillover. Third, a uniquetemperature threshold was also included to lower the risk level; all other thresholds are point-in-time triggers that raise the hazard level, but WNV hazard will fluctuate in response to changingweather conditions. A period of at least 7-days between July 15 and August 15th with meandaily temperatures below 22◦C (Hartley et al., 2012) will lower the threat level from medium tolow, or from high to medium. It is recognized that this may have little practical impact if theintervention for a given hazard level has already been implemented, but documenting reducedhazard is of value because interventions could subsequently be stopped or messaging about theneed for personal prevention changed.4.3.3 Hazard Dependent Public Health ActionA modified and reduced version of a framework used for Lyme disease risk prediction (Quineet al., 2011) was used to clearly link hazard and public health action. This approach defines:1) what action to take and 2) who/where to focus the prevention measure for each hazardlevel. Similar approaches have been used in the Florida response plan for WNV (Day et al.,2015). The public health interventions presented here were developed by reviewing historicintervention measures that have been used provincially and referencing additional approachesthat are suggested by the literatures and have been used elsewhere (see Appendix C for details).The proposed measures included community and individual level prevention measures, as wellas real-time changes to existing provincial surveillance. The prevention measures describedhere assume the presence of a minimum WNV surveillance program that is comprised of vectorsurveillance, public reporting of dead corvids, information sharing between animal and humanhealth agencies, and mandatory reporting of human cases.Low hazard1. Inform the HAs of elevated hazard - Initial hazard assessments will likely be con-ducted by the BCCDC WNV surveillance team. Key contacts in each HA, as designatedby the provincial WNV communications plan, should be made aware of elevated hazardonce suitable conditions are met. Effective communication with regional HAs is importantgiven their role in WNV surveillance and control.2. Messaging : Dead corvid reporting - Public health messaging about the importance1154.3. An Ecological Framework for WNV Decision Support in BCFigure 4.25: Decision matrix characterizing the relationship between surveillance inputs, time,and hazard. Colors represent hazards categories (yellow=low, orange=medium, and red=high).The typical amplification and spillover periods are overlaid on top of the decision matrix. SeeSection 3.5 for a description of how to use the decision matrix.1164.3.AnEcologicalFrameworkforWNVDecisionSupportinBCTable 4.9: Hazard-specific public health responses. Region reflects the suitable spatial unit for a given surveillance input (HSDA, LHA,community, etc). Descriptions of individual actions can be found in Section 4.3.3.Hazard Level Action Where WhoInform HAs of elevated regional hazard Agreed upon surveillance leads for provincialHAsLow HazardMessaging : Dead bird reporting reminders The public in regions with elevated hazard con-ditionsIncrease intensity of mosquito surveillance Regions with elevated hazard conditionsSecond round of targeted larviciding HAs mosquito contractors, regions with elevatedhazardMediumHazardMessaging : Personal protection All public, but focused on small children andelderly.Adulticiding and Implementation of Order inCouncilCommunity or region with >3 human or horsecasesHigh HazardRestricting outdoor evening activities Regions with >0 confirmed human or horse in-fections1174.3. An Ecological Framework for WNV Decision Support in BCof dead corvid surveillance is appropriate at the low hazard level prior to viral detection,as viral detection in birds is strongly suggestive of zoonotic transmission.Medium hazard1. Increase intensity of mosquito surveillance in high hazard areas - Mosquitosurveillance intensity should increase with rising hazard to improve the sensitivity of vi-ral detection by: 1) running additional traps, or 2) increasing the frequency of mosquitocollection (e.g. twice per week). This allows for the maintenance of surveillance infras-tructure and saves costs in years with cool temperatures. Locations for additional trapplacement should be chosen based on viral activity and mosquito abundance.2. Additional larviciding - Increase larviciding intensity should be considered if early sea-son circulation is confirmed in order to disrupt the amplification process. Reducing vectorpopulations at seasons end will lower the probability of successful viral overwintering.Additional larviciding should only be considered after local entomologists and mosquitocontractors have been consulted to gauge the maturation of the mosquito population inrelation to diapause (Buth et al., 1990).3. Messaging : Personal Protection - Targeted messaging focused on 1) personal pre-vention and 2) behavioural changes should be implemented in medium hazard regions.Messaging related to prevention measures was strongly associated with a reduced risk ofSLE infection in Florida (Meehan et al., 2000). Consistent messaging during the summermonths should be avoided to prevent messaging fatigue.High hazard1. Increase communication with the CVO - Daily communication with the CVO shouldbe implemented to improve the timeliness of the reporting of infected horses or othermammal species, which helps elucidate the spatial scope of disease spread within theprovince.2. Restricting outdoor evening activities - Restriction of public outdoor gatherings likesporting events or outdoor concerts should be considered in high hazard areas resulted ina four-fold decrease in risk of SLE infection during outbreaks in Florida (Meehan et al.,2000). Messaging targeting areas with elevated hazard should be provided to stakeholdersin high hazard locations if such closures are not possible.3. Consideration of Adulticiding - Implementation of the adulticiding response planshould be considered only if numerous human cases are detected in any given LHA ortown. However, no clear incidence or case-total threshold exists for adulticiding and trig-ger values will need to be determined by the regional public health. The benefits of adul-ticiding must be weighed against its financial costs and stakeholder concerns. The time1184.3. An Ecological Framework for WNV Decision Support in BCrequired for stakeholder consultation must also be considered, which may be greater thanthe spillover period in locales with temperate climates. Adulticiding currently requiresmodifying the provincial 'Integrated Pest Management' plan, which requires 45 days forconsultation unless an 'Order in Council' is used to facilitate the plan's implementation(British Columbia Centre for Disease Control., 2010a).4.3.4 User SurveyA total of 18 respondents started the survey, and only 12 of these completed the survey. Resultsare presented only for those individuals who provided responses to all 10 scenarios. The groupthat completed the survey included 3 MHOs, 1 EHO, and 8 additional WNV staff, includingepidemiologists, mosquito control experts, and ecologists. The respondents who worked in publichealth (n=10) had 16.1 years of work experience on average.Respondents perceived increases in risk when conditions changed from non-suitable for trans-mission to suitable for transmission, then to confirmed zoonotic transmission, and finally toconfirmed spillover (Table 4.10). The scenario with multiple symptomatic human infections wasperceived as having the highest risk, scoring almost 2 full points more than a positive non-travelrelated blood donation. Perceived risk was lower for late-season confirmed enzootic transmission(Aug 28th) than for early-season enzootic transmission (July 15th). Perceived risk for early sea-son enzootic transmission was similar to that of confirmed spillover in horses, which itself hada perceived risk equivalent to asymptomatic human cases. The scenarios eliciting the greatestvariability in risk perception between respondents included the one describing early enzooticactivity and the one describing a single asymptomatic case detected via blood donation. Only66% of respondents indicated they had sufficient information to make a recommendation for thescenario with the asymptomatic case detected via blood supply. Nearby activity in WashingtonState without local transmission was seen as having greater hazard than scenarios describingconfirmed regional enzootic transmission.Most respondents were comfortable recommending prevention measures based on the infor-mation provided in the scenario descriptions. The exception was Scenario 9 (1 asymptomaticcase) in which 25% (n=3) of respondents were unable to make a public health recommendationbecause they lack sufficient information. Respondents requested additional regional surveillancedata, as well as additional epidemiological information from Washington State, prior to recom-mending a specific intervention for that scenario. Other information that was requested for otherscenarios included: information on historic regional activity, the location of positive mosquitopools relative to human population centres, and more detailed case histories (eg. additionaltravel information, location of primary residence, occupation, etc).This survey also aimed to gain user feedback on the appropriateness of proposed inter-ventions. Scenarios were aggregated for each stage of the transmission cycle and used a 50%response rate as a selection cutoff in order to identify acceptable prevention recommendations(Table 4.11). Respondents indicated that no prevention measures should be implemented if1194.3.AnEcologicalFrameworkforWNVDecisionSupportinBCTable 4.10: Summary table showing the mean risk (and SD) for each scenario, as well as the proportion of respondents selecting aprevention measure identified in the survey. Scenarios are grouped according to the stage of the transmission cycle.Scenario Mean Risk(SD)IncreaseVectorSurveillanceIncrease Lar-vicidingAdulticiding Messaging:Corvid.RepMessaging:PersonalPreventionRestrictOutdoorActivityScn1 - Unsuitable Conditions 1.00 (0.95) 0.22 0.11 0.00 0.00 0.44 0.00Not Suitable 1.00 0.22 0.11 0.00 0.00 0.44 0.00Scn2 - Suitable 3.08 (1.75) 0.36 0.09 0.00 0.18 0.55 0.00Suitable 3.08 0.36 0.09 0.00 0.18 0.55 0.00Scn4 - Enzootic: Early 5.50 (1.98) 0.78 0.78 0.11 0.67 1.00 0.33Scn5 - Enzootic: Corvid 5.08 (1.31) 0.82 0.55 0.09 0.64 0.91 0.18Scn6 - Enzootic: Late 4.75 (1.76) 0.55 0.27 0.00 0.36 0.91 0.27Enzootic 5.11 0.52 0.71 0.06 0.55 0.94 0.22Scn3 - Suitable+Nearby 5.67 (1.15) 0.82 0.64 0.09 0.64 0.91 0.36Suitable: Nearby 5.67 0.82 0.64 0.09 0.64 0.91 0.36Scn7 - Spillover: Equine 5.58 (1.56) 0.73 0.73 0.00 0.55 1.00 0.45Scn8 - Spillover: Asymptomatic 5.42 (1.83) 0.60 0.70 0.00 0.60 1.00 0.40Scn9 - Spillover: Symptomatic 5.83 (1.64) 0.67 0.78 0.11 0.44 1.00 0.22Scn10 - Spillover: 3 Cases 7.58 (1.38) 0.60 0.70 0.10 0.50 1.00 0.40Spillover 6.10 0.65 0.73 0.05 0.52 1.00 0.231204.3. An Ecological Framework for WNV Decision Support in BCconditions are unsuitable for WNV transmission, likely to avoid messaging fatigue. Once con-ditions become suitable for WNV, only messaging directed at personal prevention was deemedappropriate. Messaging promoting personal prevention was the most popular recommendation,being selected by >80% of users in all enzootic or spillover scenarios. Personal prevention mes-saging was favoured by 55% of respondents once conditions were suitable for transmission, butfavoured by only 44% of respondents when conditions were unsuitable for transmission (Table4.10). Respondents indicated that the following measures should be implemented in a scenarioof enzootic transmission: increased vector surveillance, increased larviciding, messaging aboutcorvid surveillance, and personal prevention measures. Additional larviciding was recommendedless if enzootic detection occurred late in the WNV season (27%), when it has limited value be-cause vector abundance is already decreasing. Adulticiding was only recommended by a singlerespondent and should be removed from the set of recommended public health responses be-cause of the negative public opinion about insecticide use in BC and elsewhere (Petersen et al.,2012b), and because it lacks benefit due to the short WNV season.The recommended public health interventions did not differ for enzootic transmission andconfirmed spillover, despite differing levels of perceived risk (Table 4.10). This finding is incontrast to the position taken in the original draft of the decision support tool, which proposedunique prevention measures for medium and high hazard (Table 4.10). No additional preventionmeasures were recommended for spillover beyond those identified for enzootic transmission.The original tool suggested that both adulticiding and restrictions on outdoor activities werepossible prevention measures in scenarios of confirmed spillover. However, only 40% and 10%of respondents recommended restricted outdoor activity and adulticiding, respectively, for thehighest risk scenario, and these recommendations were made even less for all the other scenarios.Respondents also identified two additional measures to be considered for addition to the updatedtool (Table 4.11).• Location-based prioritization of viral testing in vectors (Enzootic Transmis-sion) - As hazard increases, batch processing of mosquito samples should stop in favourof location specific testing to improve our ability to determine regional hotspots of viralcirculation. Locations with a high hazard should be given elevated priority for laboratorytesting to improve the timeliness of viral detection. Increases in lab capacity may berequired.• Implementation of single unit testing (SUT) of blood supply (Spillover) - Cana-dian Blood Services typically batch tests blood samples in groups of 6 (Vamvakas et al.,2006). SUT should be implemented after confirmed spillover to minimize transfusionrelated infection. Individual-donation nucleic acid testing (ID-NAT) can detect approxi-mately 30% more cases than batch testing in some scenarios (O'Brien et al., 2010a), andwas cost-effective during select epidemics in the US (Busch et al., 2006).User feedback was incorporated into the updated list of hazard specific prevention measures1214.3. An Ecological Framework for WNV Decision Support in BC(Table 4.11). Messaging promoting corvid surveillance was included in the tool, despite respon-dents' preferences, because dead corvid reporting can signal enzootic transmission (mediumhazard). Recommended prevention measures did not generally differ between enzootic trans-mission and confirmed spillover, even though the mean perceived risk was higher for the spilloverscenarios. The lone prevention measure selected for transitioning to high hazard was a transitionto single unit testing of the blood supply in regions with confirmed spillover. This approachcoincides with current CBS standard protocols (Vamvakas et al., 2006).1224.3.AnEcologicalFrameworkforWNVDecisionSupportinBCTable 4.11: Modified hazard-specific public health responses based on user feedback. Regions reflects the suitable spatial unit for a givensurveillance input (HSDA, LHA, community, etc). Descriptions of individual actions can be found in Section 4.3.3.Hazard Level Action Where WhoMessaging : Dead corvid reporting reminders The public in regions with elevated hazard con-ditionsSuitableConditions Messaging : Personal protection All public, but focused on small children and el-derly.Increase intensity of mosquito surveillance Regions with elevated hazard conditionsMediumhazardTransition to regional or trap-specific analysis ofvector dataBCCDC; Regional surveillance units in HAs withelevated hazardSecond round of targeted larviciding HAs mosquito contractors, regions with elevatedhazardHigh hazard Single unit testing of blood supply Canadian Blood Services;123Chapter 5DiscussionThe primary objective of this thesis work was to improve public health agencies' situationalawareness regarding WNV transmission in BC with the specific goal of producing a practicaldecision support tool to guide WNV decision-making. As part of this, we evaluated whetherthere was a consistent set of weather and ecological predictors that were associated with WNVtransmission in western Canada, and that could be generalized to the BC context. Study resultsprovide a glimpse of the climate and ecological conditions that determine patterns of WNVincidence in BC and Saskatchewan, including consistently positive associations with heat. Yetthe dominant characteristic of the results presented in this thesis is the inconsistency of findingsbetween models, between years and between regions. Model inconsistencies may partially resultfrom methodological limitations related to working at large spatial scales, however, they may alsobe inherent characteristics of complex ecological systems that are characterized by non-linearity,multiple stable states, processes working over multiple scales, and feedback loops (Costanzaet al., 1993). Traditional epidemiological approaches like those used here were common for thistype of analyses when this research began and remain widely accepted tools in field of publichealth. However, such approaches are likely insufficient to capture the full dynamics of thecomplex ecological systems. The complexity and unpredictability of ecological systems challengeprobabilistic risk estimation (Parrott, 2010), and public health agencies in low-incidence areasshould not invest resources attempting to predict future WNV incidence. Instead, the focusshould be placed on identifying the broad environmental features that facilitate transmission toimprove situational awareness, and then investing in preparedness for the rare instances whensignificant regional transmission will occur.5.1 WNV Range Expansion and Activity in BCThe first seven years of WNV activity in BC are characterized by the following facts: 1) WNV didnot occur at detectable levels in BC prior to 2009 despite activity in neighbouring jurisdictions,2) the first occurrence of WNV in BC occurred in the south Okanagan Valley during a summerwith abnormally warm conditions, regional peaks in Cx. tarsalis abundance, and abnormallyhigh WNV incidence in nearby Washington State, and 3) low levels of sporadic endemic WNVactivity continued in the Okanagan Valley in the years after initial detection, even in yearswith normal temperatures and comparatively low mosquito abundance. Coming to a completeunderstanding of the factors driving WNV activity in BC is unlikely. However, the location and1245.1. WNV Range Expansion and Activity in BCtiming of the initial WNV activity, coupled with the observed patterns of activity in the yearsfollowing the initial detection, provide clues as to potential drivers of this range expansion, andimprove our regional situational awareness of the WNV threat.The detection of WNV in the southern Okanagan Valley in 2009 proved that select regionsin BC have climate and ecological conditions suitable for WNV amplification, at least in someyears. However, after the virus was initially detected, it remained unclear whether the observedrange expansion along this northern limit would be followed by: 1) spatial expansion and anincrease in disease incidence, 2) rare instances of sporadic disease, or 3) a complete disappearanceof viral activity. Concerns about increasing incidence the following year were not unfounded,given the proven ability of the WNV to overwinter in adult mosquitoes (Nasci et al., 2001) oravian species (Reisen, 2013; Reisen et al., 2006c) and the explosive three-year WNV cycle seenin Colorado, California and North Dakota (Day et al., 2015; Hayes et al., 2005; Nolan et al.,2013; Reisen & Brault, 2007; Reisen, 2013). These fears proved unfounded, as limited WNVactivity was detected in 2010 and again in 2013. The detection of a human case in the centralOkanagan in 2010 suggested a northward movement of the virus, but the absence of spillover inthe following years, coupled with a return of the virus in Osoyoos in 2013, continues to suggestthat the Southern Okanagan remains the provincial focal point of WNV activity.The location of the initial WNV detection should not have been a surprise, as provincialrisk maps created by the BCCDC prior to 2009 identified that it had a high WNV risk becauseof climate and ecological conditions that made it uniquely susceptible to WNV amplification(see Section 4.1.4) (BCCDC, 2009). The southern Okanagan Valley is the hottest region inBC during the summer months and accumulates more DDs than other areas of the province,especially in July and August. Communities outside of the Okanagan Valley like Creston, Cran-brook, and Abbotsford also surpassed key temperature thresholds in select years, but the heatexperienced in these areas is often inconsistent with sub-threshold temperatures reached duringthe amplification season. Periods of cooler temperatures are believed to break the amplificationcycle such that the EIP does not complete before vector death or diapause, effectively haltingthe WNV amplification process in both the Fraser Valley and Kootenay regions. Temperatureconditions alone will likely restrict WNV to select regions of the province for the foreseeablefuture barring significant changes to regional weather conditions (Harrigan et al., 2014).The geology and ecology of the southern Okanagan Valley also favour viral introduction andestablishment. Mountains are generally unsuitable for WNV amplification and transmission(Bolling et al., 2009; Gates & Boston, 2009), which means that the province has limited habitatssuitable for WNV amplification and transmission. Furthermore, the Canadian Rockies likelyacted as a barrier to WNV introduction from Alberta (Barker et al., 2009), similar to the barrierposted by the Great Divide in the continental US (Venkatesan & Rasgon, 2010). Northward viralmovement, however, may have been facilitated by mountain valleys running from WashingtonState into BC. These valleys may provide a path of least resistance for vector- or reservoir-mediated introduction of the virus from Washington (Bailey et al., 1965; Rappole et al., 2006),1255.1. WNV Range Expansion and Activity in BCor introduction by migrating birds along the Pacific Flyway (Rappole & Hubalek, 2003). TheOkanagan and Kootenay valleys are both migration routes for many avian species (Campbell &Branch, 1990). Long distance wind-borne movement of the Culex species was implicated as theintroductory mechanisms for Japanese encephalitis (Ritchie & Rochester, 2001), and movementof infected vectors is posited as a mechanism of WNV introduction elsewhere in North America(MacKenzie et al., 2000; Venkatesan & Rasgon, 2010). The southern Okanagan Valley alsocontains a higher abundance of Cx. tarsalis than other parts of the province (Table 4.1), (Tachiiriet al., 2006), as well as abundant irrigated landscapes that cluster with human habitation nearrivers, wetlands, and lakes to provide a favourable habitat that facilitates interaction betweenvectors, reservoirs, and humans (Shaman et al., 2005).Three years of confirmed WNV transmission in the south Okanagan valley suggest thatbaseline conditions in the area are sufficient for WNV amplification; inter-annual variation inincidence is therefore likely driven by weather and time-varying ecological factors like vectorabundance. Above average heat in the southern Okanagan Valley appeared to be a drivingfactor of observed WNV activity in 2009 and 2013. In California, transmission effectivenesswas elevated when temperatures were 2-5◦C above the 30-year average (Reisen et al., 2006b),while the transmission of WEE was also more common at the upper end of regional temperaturenorms (Barker et al., 2010b). Above average daily minimum temperatures in June, July andAugust may be a useful predictor of favourable conditions for amplification, albeit one that isnot sufficient for transmission as similar conditions were seen in 2014 and 2015 with no positivesurveillance inputs. In addition, waiting to determine if August temperatures are above averagewould preclude the implementation of timely prevention measures.The positive correlation between above average temperatures and elevated WNV incidencehas been well-documented (see Section 2.7.1). July and August temperatures were specificallypositively correlated with WNV transmission intensity in some years in Illinois (Lampman et al.,2006), which supports the findings presented here. What remains unclear is how temperatureaffects WNV transmission along the upper edge of the acceptable temperature gradient, espe-cially along the virus's northern range, where the amplification season is short (Hayes et al.,2005). Mosquito mortality increases above certain temperatures (Ciota & Kramer, 2013), andalthough thermal adaptation in vectors has been suggested, it is not supported by current re-search (Ruybal et al., 2016). To the contrary, the results presented here suggest that aboveaverage temperatures accelerate amplification even in desert regions where temperatures aregenerally above key thresholds. The sporadic and low levels of observed WNV in the Okana-gan may, however, result from the tension between increased vector mortality and faster viralamplification that occurs when heat is extreme (Reisen, 2013). Further research on the relativeimportance of these competing factors is needed to fully understand the dynamics of WNV inthe southern Okanagan Valley.The hot desert conditions in the Southern Okanagan Valley are likely suitable for WNVtransmission in most years, even in the absence of above average heat. In 2010, daily minimum1265.1. WNV Range Expansion and Activity in BCtemperatures were slightly below average in the summer months, yet mean daily temperaturesremained above the 22◦C threshold for extended periods in Osoyoos, and intense amplificationlikely occurred in late July when mean temperatures were above 26.7◦C (Hartley et al., 2012)(Figure 4.3). The WNV activity detected in 2010 may also have been facilitated by aboveaverage temperatures in the preceding winter that increased the survival rate of overwinteringinfected mosquitoes (Mitchell, 1979), leading to earlier seasonal amplification (Reisen, 2013).Winter temperature anomalies were a more consistent predictor of WNV incidence than summertemperature anomalies in the US in the northern Great Plains, the upper midwest and southcentral states (Wimberly et al., 2014).Although temperature may be the primary factor driving transmission, amplification andspillover are driven by a combination of environmental and ecological conditions. In 2009, thesouthern Okanagan Valley not only experienced above average temperatures, but also experi-enced a sequence of events that may have promoted viral amplification after introduction. Thefirst positive mosquito pool was detected a week after heavy rainfall and immediately following aperiod of extreme heat during which nightly temperatures were well above the 14.3◦C threshold(Reisen et al., 2006a) (Figure B.2 and Figure 4.8). The Ministry of the Environment classified2009 as a drought year (BC Ministry of Environment, 2016), and the rainfall events seen in Julyand August likely increased the abundance of vector development sites, while the ensuing periodof high temperature facilitated rapid mosquito development, virus amplification, and subsequenttransmission. Summer precipitation may also have resulted in vectors and reservoirs aggregatingaround remaining bodies of water (Shaman et al., 2002). A similar sequence of events was seenin the Okanagan in 2013, with late-June precipitation followed by above average temperaturesand a subsequent peak in Cx. tarsalis abundance. Precipitation followed by heat and droughthas been identified as a potential driver of SLE outbreaks in the past (Day, 2001). The elevatedabundance of Cx. tarsalis - a known bridge vector (Kent et al., 2009) - at the end of July andbeginning of August in 2009 and throughout the summer in 2013 (Figure 4.8, 4.2) also likelyfacilitated spillover transmission from birds to humans and horses.The apparent low abundance of Cx. tarsalis from 2003-2008 (with the exception of 2006)may have been one factor that prevented wide-scale amplification. However, Cx. tarsalis wassporadically abundant in select parts of the province in 2009, 2010 and 2013, and these localizedpeaks may have driven focal points of amplification. In all three years, peaks in Cx. tarsalisabundance occurred near the beginning of July, with human cases documented in early August.This provides limited support for the 4-week lag between vector abundance and human infectionhypothesized by others (Barker et al., 2010a; Bolling et al., 2009). However, peaks in vectorabundance also appeared to coincide with the period of human infection in 2009 and 2013,making it difficult to comment on the importance of the timing of these peaks relative to humaninfection. Furthermore, localized peaks were seen in years and in regions without documentedWNV transmission, mirroring inconsistent associations between vector abundance and diseaseseen elsewhere (Kilpatrick & Pape, 2013). Care must be taken when attributing spillover to1275.1. WNV Range Expansion and Activity in BCsuch peaks, as little is known about the ecology of Cx. tarsalis mosquitoes in the OkanaganValley, and vector surveillance may have missed short-duration increases in vector abundance ifthey didn't coincide with the chosen trap night. To enable focused prevention efforts, specificinformation is needed regarding habitat preferences, areas with high mosquito production, andoverwintering behavior.The potential explanation for the initial detection of WNV in BC in 2009, and the reasonfor its prolonged absence, are partially based on routinely collected surveillance data, and theexplanations for the observed patterns may be biased due to the limitations of those datasources. This is particularly relevant to vector abundance data because of evolving surveillancemethodology. Vector surveillance in BC started in 2003 in response to the national spread ofWNV, and trapping methods improved as regional experts learned about trap placement andthe ecology of the Okanagan Valley. Systemic improvements to vector surveillance approachesmay be responsible for year-to-year increases in Cx. pipiens abundance from 2003-2014 andannual variation in Cx. tarsalis abundance. Such changes do not, however, explain within-season patterns of vector abundance. Targeted high-intensity trapping was run in the southernOkanagan in 2009, but not in 2010. While the absence of positive mosquito pools in 2010 couldindicate that levels of circulating virus were lower than they were in 2009, it could also haveresulted from the reduced sampling intensity. This illustrates the potential value of focusingsurveillance effort in areas perceived to be at high risk for transmission, since doing so willimprove viral detection when circulation is limited (Gu et al., 2008). The targeted approachwas used again in 2013 as traps were limited to the Okanagan Valley which proved sufficient todetect circulating virus when coupled with increasing trapping intensity.Similarly, positive dead corvids were detected in 2010 but not 2009, with the greatest numberof positive corvid submissions (n=5) occurring four years after initial viral detection. Corvidsubmissions decreased every year after the implementation of the surveillance program, sim-ilar to trends seen elsewhere (Hadler et al., 2015; Kwan et al., 2012). The effectiveness ofcorvid surveillance is known to decrease after regional WNV establishment due to broad avianmortality, heard immunity, and public reporting apathy (Carney et al., 2011; Gu et al., 2008;Hochachka et al., 2004; Kwan et al., 2012). The absence of positive dead corvids in 2009, whichoccurred despite confirmed infection in humans, horses and mosquitoes, was not unprecedented.US surveillance data from 2001 and 2002 revealed that many jurisdictions with human casesreported no WNV-positive dead corvids (Guptill et al., 2003). Furthermore, dead corvids maygo unnoticed in the rural Okanagan because of low human population density and the fast con-sumption of carrion by wild animals (Ward et al., 2006). The reduced effectiveness of corvidsurveillance over time, coupled with the rural focus of WNV in BC, validated the decision tostop corvid surveillance in BC in 2014.While study results suggest a plausible explanation for the location and timing of the WNVrange expansion into BC, it remains unclear what factor, or combination of factors, preventedWNV activity in BC between 2003 and 2009 despite activity in nearby areas. Amplification1285.1. WNV Range Expansion and Activity in BCin a naive area requires a series of discrete events with independent probabilities to occur (seeSection 2.9), and several general explanations exist that delineate these events. First, theRocky Mountains and the northern Cascades may have limited viral introduction by impedingthe movement of birds or mosquitoes (Barker et al., 2009; Moore, 2008; Rappole & Hubalek,2003). While viremic birds can show migratory behaviour in test settings (Owen et al., 2006),the health impacts of WNV infection make cross-mountain migration unlikely for many species(Wheeler et al., 2009). Movement of the virus into the province is therefore not necessarily ayearly occurrence. Second, viral introduction could have occurred prior to 2009, but unsuitableecology or weather at the introduction site could have prevented viral establishment, persistence,and amplification. Environmental and ecological conditions are unsuitable for amplification inmuch of BC, and even the Okanagan Valley has insufficient conditions in some years. The years2004, 2005, 2006 and 2008 each had significant periods where temperatures were below keythresholds during the key summer months (Figure 4.4). Cx. tarsalis abundance was generallylow in the provincial interior in 2003, 2004, 2005, 2007 and 2008, although this pattern mayhave resulted from changing vector surveillance methodology. Finally, regional activity waslimited in both Washington State and Alberta in 2004, 2005, or 2008, reducing the probabilityof short distance mosquito-borne introduction, although avian migratory introduction remainedpossible. It therefore appears that 2009 may have been the first year since the virus reached theprovincial border with above average temperature, elevated vector abundance, and the nearbyWNV activity required for introduction and amplification sufficient to be detected.One additional factor may explain the limited WNV activity seen in BC. Minimum arearequirements define the minimum habitat size required for the long-term maintenance of agiven species (MacArthur & Wilson, 1967); if the area of suitable habitat is too small, thenstochastic events may result in population extinction. The mountain valleys of southern BCcontain the conditions necessary for amplification, but are narrow, with limited suitable terrain.In regions with large expanses of suitable habitat, acquired immunity and fine-scale weatherevents in one location may be offset by naive populations or suitable weather conditions innearby regions. Similar to metapopulation or metacommunity theory and the maintenance ofregional populations (Leibold et al., 2004; Wiens et al., 1993), WNV may operate accordingto a version of source-sink dynamics whereby the combined effects of interacting pockets ofamplification allow for longterm maintenance of provincial transmission. For example, localizedprecipitation may lead to increased vector abundance sufficient to seed nearby areas throughregional movement. Similarly, if amplification is halted due to a local cold weather system, anearby region may have sufficient weather to maintain amplification. The Canadian Prairies,with their large areas of suitable habitat, may therefore provide multiple regional opportunitiesfor the correct alignment of conditions required for amplification and transmission - in essenceminimizing the role of stochastic events when evaluated at a larger scale. These small pocketsof localized amplification, combined with regional vector or avian movement, could maintainviral transmission in ways that would not occur in the Okanagan. While researchers in spatial1295.1. WNV Range Expansion and Activity in BCand infectious disease epidemiology are beginning to consider the impact of landscape processeson disease and disease transmission (Lambin et al., 2010; Ostfeld et al., 2005), no studies havespecifically examined the role of minimum area requirements for a VBD system.This analysis of the ecological and environmental correlates of WNV activity in BC hasimportant implications for regional public health. First, WNV has had a limited impact onhuman health in BC since 2009. This trend will likely continue unless the virus expands inthe more heavily populated Fraser Valley, where Cx. pipiens could drive regional transmission.Second, the analysis highlights the challenges of using surveillance data to inform our under-standing of disease emergence and transmission. The purpose of surveillance differs from thatof pure research and sampling methodologies can reflect this difference. Shifting the objectiveof vector surveillance from identifying the spatial extant of the virus to targeting viral detectionin high-risk areas likely confounded between year comparisons. Attempts were made to controlfor this by focusing on a subset of traps at stable locations, yet even minor changes in traplocation can affect mosquito catch because of the importance of the microclimate (Drummondet al., 2006; Godsey et al., 2013). Vector abundance was therefore given only limited weight inthis analysis. Despite these limitations, public health authority run surveillance data remainsvaluable for understanding the ecology of VBZs over a large area, and over long periods of time.Perhaps more importantly, this descriptive analysis further points toward the complexity ofthe WNV amplification and transmission process. Comparing years with and without detectedWNV has revealed some potential causes of the historically limited activity, but also suggeststhat the web of causative factors for WNV amplification and transmission will not be easilyquantified. Simple climate or ecological thresholds that cleanly differentiate periods with-and-without transmission are unlikely to exist. Instead, variations in annual incidence will likelybe determined by interacting factors that facilitate or dampen transmission. As we may neverfully understand the complex interplay of disease determinants at a level needed for prediction,regional public health will need to continue making decisions based on incomplete information.In summary, the introduction and within-season amplification of WNV in 2009 represented along-delayed range expansion. Although the cause of the delay remains uncertain, it is hypoth-esized that a perfect storm of environmental and ecological conditions in 2009 facilitated bothintroduction and establishment. The high levels of WNV in Washington State in 2009 provideda sufficient nearby source of WNV that facilitated its northward introduction into BC throughcross-border mountain valleys. Given that the Rocky Mountains limit migration of infectedbirds or mosquitoes into BC from Alberta, the high WNV activity seen in 2009 in WashingtonState was likely a particularly important driver of the 2009 introduction. This introductionoccurred during a summer with prolonged periods of heat that included the key amplificationperiod and peaks of Cx. tarsalis abundance, to present a convergence of favourable events thatallowed the virus to gain an ecological foothold in the region and, for the first time, to reachlevels of infection in avian and mosquito populations sufficient for the successful overwintering ofthe virus. The WNV seasons after 2009 provided insights into the nature of viral expansion and1305.2. Associations between WNV incidence and ecological conditions in Saskatchewantransmission along the borders of BC, although key questions remain. While the detection ofWNV transmission in 2010 and 2013 confirms that select regions of the province have sufficientconditions for transmission at least in some years, the sporadic nature of provincial surveillancemakes it difficult to determine whether the virus has fully established and overwintered in BC,or is sporadically re-introduced.5.2 Associations between WNV incidence and ecologicalconditions in Saskatchewan5.2.1 IrrigationThe results presented here add to the growing body of research linking irrigation to elevatedWNV incidence in agricultural settings (see Section 2.8.3). The results suggest that differentmethods of irrigation have differential impacts on disease incidence, and that the effects ofirrigation are potentially dependent on regional weather. Total acres of irrigated landscapewithin an RM was positively associated with WNV incidence in Saskatchewan in 2003, and thisassociation is consistent with both the initial hypothesis and with previous research highlightingthe association between irrigation and MBDs like WNV. However, irrigation was not associatedwith WNV disease during the 2007 outbreak. The absence of an association in this year suggeststhat either 1) the effect of irrigation may be mediated by precipitation or variation in otherhydrology-related landscape features, 2) that irrigation is important in driving transmission inonly select locations of the province, 3) that irrigation is not associated with disease and thepatterns observed in 2003 are driven by another unexamined factor, or 4) that associations withirrigation are non-linear. All four options are possible, and yet our inability to determine thecause of observed variation limits the usefulness of this predictor for decision support purposes.The shifting significance of associations between irrigation and WNV incidence between2003 and 2007 may be driven by changes in the relative importance of irrigation as a sourceof vector development sites. Only 2% of farms in Saskatchewan irrigate (Poirier, 2009), andirrigation is limited to a small percentage of all land in the province (approximately 0.5% in2006) (Saskatchewan Irrigation Projects Association, 2008). Alternative sources of standingwater may therefore reduce the association between irrigation and incidence. WNV incidencemay be more strongly associated with irrigation in dry years, when precipitation-based standingwater is rare. Total cumulative precipitation was similar in both 2003 and 2007, but totalmonthly precipitation in May and June of 2007 was greater than total monthly precipitation in2003 (Figure 4.19). In addition, drought conditions were seen in Saskatchewan in 2001 and 2002which may have limited the amount of precipitation-based standing water in 2003 (Agricultureand Agri-Food Canada, 2016), thereby facilitating a stronger association between irrigation anddisease incidence the following year. In contrast, spring and early summer precipitation in 2007(Artsob et al., 2009) may have provided abundant vector breeding sites, thereby minimizingthe importance of irrigation as a driver of disease. The years between 2003 and 2007 were1315.2. Associations between WNV incidence and ecological conditions in Saskatchewanconsidered wet years, and precipitation is suggested to have increased the abundance of wetlandsin Saskatchewan, which may have in turn provided alternative breeding locations during thesecond outbreak (Epp & Waldner, 2009). The weaker effect of irrigation in 2007 may be furtherexplained by the differing distribution of incidence in 2003 and 2007. Irrigation was primarilyclustered in the south-west corner of the province in those years, and WNV expanded northwardinto non-irrigated landscapes in 2007. Contrasting precipitation patterns in 2003 and 2007 likelycombined with this viral range expansion to reduce the relative importance of irrigation as aprovincial driver of disease in the second outbreak.The limited effect of total irrigation on incidence in 2007 suggests that the previous year'sprecipitation may modify the impacts of irrigation, but sub-analyses suggest that specific typesof irrigation remain near significant in 2007 despite observed precipitation patterns. Twobroad classes of irrigation are used in Saskatchewan: sprinkler-based and surface irrigation(Saskatchewan Agriculture, Food, and Rural Revitalization, 2003). Surface irrigation resultsin greater amounts of standing water than sprinkler-based systems (Saskatchewan IrrigationProjects Association, 2008) and was therefore hypothesized to be more strongly associatedwith WNV. This is mildly supported as the effect size for surface irrigation was greater thansprinkler-based irrigation in both years. In addition, AIC values indicate that surface irriga-tion is a stronger driver of observed patterns of incidence than sprinkler irrigation. While thisdifferential effect was especially strong in 2003, surface irrigation was nearly significant in the2007 model, when it again had a greater effect estimate than sprinkler irrigation. The non-significance of total irrigation in 2007 - despite the near significance of surface irrigation - maypartially result from the miscellaneous catch-all variable in the total irrigation measure, whichis not included as part of either the surface or sprinkler irrigation variables (see "RemainingIrrigation" in Table 4.5). This catch-all category could mask the association between irrigationand disease in the overall model if it does not produce suitable mosquito development sites.However, additional work is required to tease apart the causes of these discrepant findings.Total cumulative DDs was the most significant variable associated with WNV incidenceat the RM level in Saskatchewan during both outbreaks, as it was positively associated withdisease in all models. The significant positive association between heat and WNV incidencein Saskatchewan was expected, given our understanding of the WNV transmission system (seeSection 2.7.1). Temperature was above average during key transmission periods in both 2003 and2007 (Reisen, 2013), and the impact of temperature in Saskatchewan may have been even greaterif it were evaluated across outbreak and non-outbreak years, given that cooler temperatures likelydampen transmission in some non-outbreak years. Analyses also suggest that surpassing 109DDs over a 14-day period threshold is necessary for large-scale outbreaks in Saskatchewan, asthis threshold was only surpassed in the outbreak years of 2003 and 2007 (Figure 4.20). Trackingcumulative DDs in relation to this threshold may have risk prediction value for public healthdecision-making and prevention planning in the region.Unlike the association between temperature and WNV, the relationship between precipita-1325.2. Associations between WNV incidence and ecological conditions in Saskatchewantion and WNV incidence is varied both in the literature (see Section 2.7.2) and in the modelscreated in this thesis. The association was negatively associated with disease only in 2007.This negative association does not support suggestions that precipitation increases incidenceby providing mosquito breeding locations, but is consistent with other research (Hahn et al.,2015; Harrigan et al., 2014) and may have several causes. First, water-filled hoof prints areknown breeding locations for Cx. tarsalis in Saskatchewan, and these pools may be washedaway given significant precipitation (Curry, 2004; Epp & Waldner, 2009). Second, abundantprecipitation may reduce vector-reservoir contact by providing multiple breeding sites and lim-iting the impact of focal points of aggregation (Shaman et al., 2002, 2005). Third, the negativeassociation between precipitation and incidence may result from the cooler temperatures thatoccur during periods of rain. Cumulative DDs are a stronger predictor of WNV incidence thantotal cumulative precipitation in all models, supporting similar findings from the Chicago area(Shand et al., 2016). Finally, precipitation may reduce vector-human contact as people avoidoutdoor activities when it is raining.The unexpected effects of precipitation indicated by the above results may also have resultedfrom the choice of analytical approaches. The impacts of precipitation on WNV incidence wereevaluated over a limited range of possible values because 2003, and to a lesser degree 2007,were comparatively dry years (Figure 4.19). The use of cumulative precipitation also failsto account for the timing of precipitation over the course of the WNV season (see Section2.7.2). This may be particularly important because precipitation has a unimodal non-linearrelationship with transmission in other prairie regions, with 200mm of precipitation in the Mayand June period showing the strongest association with disease incidence in the Northern GreatPlains of the US (Wimberly et al., 2008). The non-linear effects of precipitation were notevaluated in this analysis, but elevated monthly precipitation in Saskatchewan in May and Juneof 2007 tentatively support the findings of Wimberly et al. (2008), although the pattern wasabsent in 2003. Teasing apart the complex interaction between precipitation, irrigation, andspatial patterns of incidence requires additional work that focuses on the timing of precipitationrelative to observed changes in vector abundance, infection rates, and WNV incidence in humanpopulations.The variation of model results seen in 2003 and 2007 - for both key predictors and for de-mographic covariates - may have resulted because the geographic distribution of WNV-positiveRMs differed between two outbreaks with the virus expanding northward into the aspen grass-lands and moist mixed grassland ecoregions in 2007 (Figure 4.17). This expansion into novelregions of the province could explain the variation in demographic and ecological covariates ifit led to the inclusion of regions with differing population density and age characteristics. Thenorth-east expansion of WNV may also explain the decreased importance of irrigation in 2007 ascompared to 2003. Wetlands are a known risk factor for WNV transmission where Cx. tarsalisis the dominant vector, and this species is particularly abundant near wetlands in Saskatchewan(Curry, 2004). Wetlands are comparatively rare in the south-west corner of Saskatchewan, and1335.2. Associations between WNV incidence and ecological conditions in Saskatchewanincrease in abundance at more northern and eastern locales (Natural Resources Canada, 1999).The expansion of the virus into regions with more wetlands, coupled with the increased wet-land capacity seen in 2007 as compared to 2003, may have reduced the impact of irrigationas a driver of observed patterns of incidence because non-irrigation related mosquito breedingsites were abundant. A more explicit spatial analyses of the distribution of cases in 2003 and2007 is warranted, with specific attention paid the underlying demographic characteristics anddistribution of key landscape features in RMs with incidence increases in 2007.The positive association between vector control and incidence in 2003 was somewhat unex-pected, but is unlikely to be causative. This study did not measure changes in within-seasonincidence, nor did it compare vector populations before and after the implementation of larvi-ciding. The positive association between mosquito control and incidence instead likely resultsfrom the targeted application of larviciding programs in areas with high mosquito abundanceand a high risk of WNV disease. Results do, however, indicate that community mosquito con-trol did not reduce vector populations enough to minimize transmission, which is not surprisingas vector control is primarily focused in urban areas, while Cx. tarsalis favours rural habitats(Curry, 2004). The variation in the relative importance of environmental predictors betweenoutbreaks has been seen previously in Saskatchewan, with wetland and landscape variables beingless associated with high risk areas in 2007 than in 2003 (Epp & Waldner, 2009). Ecological andweather conditions within a single location are expected to change over shorter timespan thenare the demographic characteristics of human populations. However, the expansion of WNVinto new regions of Saskatchewan in 2007 compared to 2003 brought with it a change in theenvironmental context in which transmission was occurring, which may explain the observedheterogeneity in the effect of select predictors.Human population density is one of the few variables to vary during the sensitivity anal-yses, as it was consistently non-significant in the original analysis, but significantly negativelyassociated with incidence in both the combined model and in 2003 after small populations wereexcluded. This fluctuation was not entirely unexpected, given that RMs were excluded basedon total population, which effectively truncates the lower end of the population density range.This indicates the possibility of non-linear effects of population density which have significantassociations with incidence only above certain thresholds. Furthermore, the negative associa-tion between population density and WNV in the reduced model aligns with previous researchshowing higher provincial infection rates in more rural areas (Schellenberg et al., 2006). Thisnegative association with population density is likely related to the absence of Cx. pipiens inthe province (Curry, 2004) and rural habitat preference of Cx. tarsalis. The fluctuating impactsof population density may also occur in part because the province lacks the variation in urban-ization required to consistently detect such patterns. This possibility is worthy of additionalstudy, as population density may be an important driver of disease incidence in Saskatchewanonly in areas above certain population thresholds.Multivariate regression modelling indicated that WNV incidence in Saskatchewan was driven1345.2. Associations between WNV incidence and ecological conditions in Saskatchewanby a combination of factors related to temperature, precipitation and irrigation. However, thedistribution of predictors across categories of WNV incidence suggested that RMs with andwithout observed reported cases may differ with respect to uncontrolled for confounder(s). Thisis indicated by instances when the covariate summary statistics for RMs with no cases deviatefrom clearly increasing or decreasing patterns across categories of low, medium, and high diseaseincidence. For example, RMs with disease differed fundamentally from those without diseasewith respect to total precipitation in 2007, population density in 2007, and total irrigation in2007 (Figure 4.18). A minimum threshold may exist in some years and in some locations thatdetermines the capability of an area to support WNV transmission. A second series of factorsmay then affect the variation in disease incidence if key prerequisites are met. For example, thepresence of key vector and reservoir species may limit where WNV can occur, with differences intemperature, local precipitation, or human behaviours driving the variation in disease incidencegiven the presence of the vector. Additional work evaluating the characteristics of those RMsthat are consistently free from disease will help clarify the climatic and ecological range limitsof the virus in Canada, while also helping to explain the prolonged absence of the disease in BC.This type of transmission dynamics could be represented well by zero inflated models, whichhave the ability to describe both the causes of an overabundance of zeroes and the separatecause of observed counts. Zero inflated models were tested during the model building process,although the negative binomial models provided a better fit. Nonetheless, zero inflated modelsshould be considered for future analyses of WNV incidence to better control for, and explain,the overabundance of zeros seen in many years.In summary, the association between irrigation and WNV incidence in Saskatchewan is sug-gestive, but the evidence is inconclusive because the pattern varies between years. It is possiblethat the lack of an association between irrigation and incidence seen in 2007 is the true pattern,and that the pattern seen in 2003 is an anomaly driven by uncontrolled-for confounders. How-ever, the sub-analysis and greater effect of surface irrigation as compared to sprinkler irrigationprovides biologically consistent support for the importance of irrigation as a driver of patternsof WNV incidence in some areas - albeit a less important role than temperature. Additionalevaluations are needed which explicitly focus on landscape features like wetlands, since irrigationmay interact with both precipitation patterns and wetlands to modify the effects of irrigationin some years and in some locations.5.2.2 Avian DiversityStudy findings from the avian analyses provide additional support for the importance of eco-logical variables and climate in driving patterns of WNV incidence. The positive associationbetween WNV incidence and non-passerines is consistent across models and is detected usingmeasures of both abundance and species richness. However, the positive associations betweenincidence and non-passerines refute the study hypotheses and are unexpected given previouspositive associations between passerines and incidence, and because passerines are generally1355.2. Associations between WNV incidence and ecological conditions in Saskatchewanconsidered more competent reservoirs than non-passerines (Ezenwa et al., 2006; Komar et al.,2003). The lack of associations between incidence and species richness or Shannon diversity con-tradicts the dilution hypothesis (Ostfeld & Keesing, 2000a) and contrasts negative associationsbetween WNV incidence and avian diversity found in other research (Allan et al., 2008; Swaddle& Calos, 2008). The absence of an association between WNV incidence and measures combiningavian abundance and reservoir competency fails to support patterns seen at a continental levelin the US (Allan et al., 2008). Despite these contrasting findings, the positive associations be-tween non-passerines and incidence found by this analysis were unlikely to solely be a modellingartifact because the trend is observable in descriptive analyses showing community structureplotted against levels of WNV incidence (Figure 4.21, 4.22, 4.23).There are several explanations for the contrasting findings, specifically with regards to theabsence of associations with biodiversity. First, the lack of support for the dilution hypothe-sis may result from the relative ecological uniformity of Saskatchewan. The southern half ofSaskatchewan is comprised entirely of temperate and semi-arid plains (United States Environ-mental Protection Agency, 2016; U.S. Environmental Protection Agency, 2003). BBS routes inSaskatchewan are found in four different ecoregions, yet all are grassland ecosystems or transi-tion zones between grassland and boreal forest and hence share some ecological characteristics.These regions may not contain the requisite range of ecological diversity, and hence the variationin avian community structure, needed to detect patterns between biodiversity and disease. Thelack of variation in avian community structure would not, however, explain the positive asso-ciations between incidence and non-passerines birds found in Saskatchewan. The associationbetween non-passerines and incidence could be driven by a positive association with total birdabundance if amplification is greater in regions with more reservoirs irrespective of classification(passerine vs. non-passerine). Such an association could be indicative of density-dependenttransmission patterns (Begon, 2009). Neither the abundance of individual passerines nor totalbirds was significantly associated with WNV incidence and neither measure improved modelfit sufficiently to warrant their inclusion. This could suggest frequency-dependent transmissionpatterns wherein contact between vector and host is independent of host density - which ac-cording to mathematical modelling is the condition needed for biodiversity to be protective inmosquito driven systems (Dobson, 2004).Another potential explanation for the contrasting findings is that certain non-passerinespecies may play a more important role in driving amplification in Saskatchewan than an-ticipated. Passerines are especially competent reservoirs (Komar et al., 2003)11, and have beensingled out in previous studies that looked at the association between reservoir communitystructure and WNV infection (Ezenwa et al., 2007; Swaddle & Calos, 2008). However, positiveassociations between non-passerine evenness and disease incidence were present in the east-11In addition, the majority of reservoir competency work was done in response to the initial WNV outbreak of1999 (Nash et al., 2001) and focuses on species that are important in the northeastern US. While many reservoirsare shared between locations, no published studies have explicitly evaluated the competency of reservoirs orreservoir communities in Saskatchewan or in Canada.1365.2. Associations between WNV incidence and ecological conditions in Saskatchewanern US, indicating that the role of non-passerines in disease amplification is under-appreciated(Swaddle & Calos, 2008). Charadriiformes (wading birds) have been identified as highly compe-tent reservoirs (Komar et al., 2003), yet to my knowledge this group has not been identified asa favored source of blood meals for WNV vectors. The composition of vector blood meals alsovaries regionally depending on the makeup of the host communities, especially for Cx. tarsalis(Lampman et al., 2013; Thiemann et al., 2011). There are 32 species of Charadriiformes inSaskatchewan that may play a role in regional amplification, especially in wetland landscapes(Gratto-Trevor, 2011), which have been linked to WNV incidence in other locations (Johnsonet al., 2012; Skaff & Cheruvelil, 2016). Ardeid birds, which include wetland species like herons,are implicated as WNV reservoirs in non-urban areas of California (Thiemann et al., 2011) andin India (Rodrigues et al., 1981), and are key maintenance reservoirs for Japanese Encephalitis(van den Hurk et al., 2009). Consequently, their regional importance in Saskatchewan shouldalso be evaluated. The contribution of un-examined reservoirs to regional amplification mayexplain the lack of an association between incidence and community competence because calcu-lating this measure requires published lab-derived reservoir competency estimates (Allan et al.,2008; Komar et al., 2003) and regionally important birds not yet tested would therefore beignored. Regional vector feedings studies are required to determine the actual role that a givenspecies plays in regional transmission.The absence of an association between disease incidence and broader measures of communitystructure may also result from the overly simplistic nature of these measures (i.e richness anddiversity). Favoured hosts can vary between locations, even for a single vector species, withsome species playing a larger role than their abundance would indicate (Hamer et al., 2008,2011; Komar et al., 2001; Molaei et al., 2006). Novel measures which incorporate vector feedingdata - like the force of infection - may better capture the complex interplay between aviancommunity structure and disease incidence than coarse estimates derived from point countsonly (Hamer et al., 2011; Kent et al., 2009). Avian species with a high amplification fraction12likely drive WNV transmission (McKenzie & Goulet, 2010) and are expected to be a betterpredictor for future evaluations than coarse measures of total community diversity. However,the absence of regional vector blood meal analyses in Western Canada makes the use of suchmeasures impossible, while also limiting our ability to conclusively determine if Chardriiformesor Ardeid birds are regionally important reservoirs. The inconsistent findings presented inthis thesis should be interpreted with caution as they fail to account for the full ecologicalcomplexities of avian-vector interactions.The dilution effect may also be rarer than widely believed, and only occurs under certainconditions (Ostfeld & Keesing, 2000a). The dilution effect specifically requires that the numberof infected vectors feeding per host decreases with increasing reservoir diversity. Detection of thedilution effect therefore requires controlling for vector abundance or infection rates, a criteria12Amplification fraction is the number of infectious vectors resulting from feeding on a specific host species(Hamer et al., 2009)1375.2. Associations between WNV incidence and ecological conditions in Saskatchewannot met in this thesis because of the limited vector sampling data. Randolph & Dobson (2012)do a thorough job of critiquing the dilution hypothesis and make it clear that the suspectedrarity of the dilution effect does not mean that reservoir community structure is unimportantto viral amplification - only that other aspects of community structure beyond diversity affectassociations with human disease incidence. Given the complexity of the dilution hypothesis,it is not surprising that species richness or Shannon diversity indices alone are not associatedwith transmission as they represent a simplification of reservoir populations' complex structuralmakeup. Such indices remain useful measures for processes that are truly affected by the broaddiversity of a community, yet are less relevant for analyses of WNV because of the disproportion-ate importance of select reservoir species in driving amplification (Hamer et al., 2009; McKenzie& Goulet, 2010).The BBS remains the best source of freely available, long term and wide ranging data onavian populations. However, it does have key limitations as a data source. First, the BBS onlycaptures bird counts near roads (Bart et al., 1995), which may favour species that thrive in edgehabitat (Villard, 1998). This could impact results if avian populations near roads differ fromthe broader populations, as our measures would therefore fail to reflect the reservoir potential ofthe region. Second, inter-observer reliability has well-documented issues, specifically in relationto changes in reliability over time (Sauer et al., 1994). However, this analysis is focused onexplaining variation between units, and long term changes in observer reliability are unlikelyto seriously affect its results. Finally, the BBS primarily captures information about adultbirds and may fail to capture immunologically naive fledglings. The bias toward adult birdsmay also be problematic as immunologically naive fledging and nestling birds may be uniquelysusceptible to mosquito bites because of their sedentary nature, and therefore important toarbovirus amplification. Yet the role of fledgling and nestling birds to WNV amplificationremains unclear. Immunologically naive hatch-year birds are thought to drive Cx. pipiensdriven amplification in the Chicago area (Hamer et al., 2008), yet nestling passerines were notcommonly infected with WNV in the same region (Loss et al., 2009a) and only 2% of nestlingsparrows were positive for the virus in Saskatchewan in 2006 (Millins et al., 2011). The role ofnestling and hatch year birds may therefore be context specific, depending on the primary diseasevectors, regional avian communities, and the timing of amplification relative to the presence offledglings. Despite this and the previously noted limitations, the BBS provides better estimatesof habitat specific density and avian abundance than other sources like the Christmas BirdCount (Newson et al., 2005).The unexpected associations between WNV and avian community structure may also stemfrom the residual confounding that has resulted from a failure to include landscape in avianmodels. Consequently, positive associations with non-passerine species may occur if certainlandscapes favour WNV transmission for reasons not related to reservoir community, and alsohave a greater abundance of non-passerines. Previous work has suggested that WNV incidence inSaskatchewan is greater in key ecoregions, confirming that the disease is broadly linked to specific1385.2. Associations between WNV incidence and ecological conditions in Saskatchewanhabitat types (Curry, 2004). High Cx. tarsalis abundance in some regions of Saskatchewanhas previously been attributed to the presence of wetlands (Curry, 2004), and competitiverelease of mosquitoes from predators after drought may elevate mosquito populations in andnear to wetlands (Landesman et al., 2007). Saskatchewan contains approximately 1.5 millionwetlands covering 1.7 million hectares, the majority of which are small and temporary, withless than 0.25% being larger than 50 hectares (Huel, 2000). Small wetlands are more stronglyassociated with WNV incidence than are large wetlands, especially in drought years (Skaff &Cheruvelil, 2016). Wetlands may therefore be the uncontrolled-for confounder responsible for theunexpected associations between incidence and avian community structure in Saskatchewan ifregions with small wetlands have high Cx. tarsalis abundance, as well as elevated non-passerinediversity and abundance. For such an association to occur, the role of wetlands in increasingincidence via elevated mosquito abundance must be greater than the proposed dampening effectsof elevated avian diversity (Ezenwa et al., 2007). Additional work examining the associationbetween landscape, avian diversity and vector or human infection is required to clarify thepotential confounding role of landscape in this study.Significant positive relationships with temperature were seen in all models, mirroring resultsfrom the irrigation analysis and supporting previous research (see Section 2.7.1). However,associations between other covariates and disease in the best fit avian model appear to contradictprevious research and the previously presented irrigation analyses. This was especially apparentfor associations with human population density; it is possible that the absence of this associationmay be driven by several high leverage points. Re-running the best fit model without thesepoints caused population density to become significantly negatively associated with incidence,with effect estimates similar to those seen in 2003 in the irrigation modelling. Similarly, effectestimates for DDs and precipitation are in line with the irrigation modelling results and suchrelationships are logically coherent with our understanding of disease transmission. It shouldbe noted that this does not guarantee the internal validity of the avian models. The negativeassociation with population over the age of 50 was unexpected as age is a known clinical riskfactor for more severe disease (Bode et al., 2006), and this association is maintained even afterthe exclusion of high leverage points. This and other unexpected results suggest that observedpatterns may result from small sample size, residual confounding, and/or bias (Szklo et al.,2007).5.2.3 Issues Relevant to Both AnalysesComing to grips with fluctuating and conflicting effect estimates between models is perhaps thegreatest challenge in understanding the modelling of WNV incidence is Saskatchewan. Whileecological findings related to irrigation, avian community structure, and temperature appearconsistent across models, effect estimates and p values for demographic covariates vary betweenyears in the irrigation models, and between irrigation and avian models. These discrepantfindings are likely partially driven by differences in the two data sets. The avian analyses1395.2. Associations between WNV incidence and ecological conditions in Saskatchewanwas performed on only a subset of RMs with overlapping BBS routes that are not necessarilyrepresentative of all RMs in Saskatchewan. This means that selection bias may be partiallyresponsible for the observed varying effect estimates of key predictors (Szklo et al., 2007).Small sample size is not uncommon for studies evaluating processes operating at this spatialscale, or for evaluations of the link between avian community structure and WNV incidence(McKenzie & Goulet, 2010). It does, however, limit the number of predictors and the feasiblecomplexity of resulting models. Covariates were included if they were previously identified asaffecting disease incidence, and only if high quality data was available for them. As a result,many predictors were left out of the model, potentially explaining why even best fit modelsexplained a low amount of observed variation in WNV incidence. The limited number of pre-dictors and use of coarse weather variables makes residual confounding a potential explanationfor some unexpected results. For example, observed associations between non-passerines andinfection may not be causal, but may instead mirror the distribution of landscapes that supporta high abundance of non-passerines and have additional characteristics that promote WNV am-plification (see Section 2.8). Wetlands are one uncontrolled-for landscape feature that may bea regionally important driver of disease in Saskatchewan. The ecological fallacy may also havedriven observed patterns for population density and age. By using aggregate measures at theRM level, we are not effectively measuring gradients of urbanization running from populationcentres to the surrounding countryside (Gómez et al., 2008), or true gradients of population age.In the case of population density, we therefore fail to measure variation at the scale at which itseffects are expected to occur most strongly.All models explained a limited amount of total variation in disease incidence, despite the sig-nificance of key predictors. This was not entirely unexpected. Irrigation and avian communitystructure are distal drivers in a complex causal pathway for WNV incidence (Table 4.5, Figure2.2). The models tested here are simplistic and fail to control for many factors affecting the spa-tial distribution of WNV incidence, including vector abundance. Vector data was not includedbecause of the sparsity of mosquito traps used in Saskatchewan over the study period. Moreimportantly, the urban focus of vector sampling in the region would lead to an underestimationof the true abundance of the rural Cx. tarsalis if spatial interpolation was used. The absenceof vector data is a noted limitation of previous provincial modelling and risk mapping efforts(Epp & Waldner, 2009). It should also be recognized that mosquito abundance is a proximalfactor along the same causal chain linking irrigation and disease (Bowden et al., 2011), and itsexclusion may be appropriate given our focus on the effects of irrigation. However, the samecannot be said for our analysis of avian community structure.The variability in the significance of differing predictors may also be a natural part of thecomplex ecological system that WNV amplification encompasses. Recognizing that WNV repre-sents a complex ecological system also means recognizing that such a system evolves over time,is affected by factors acting at differing spatial scales, involves non-linear responses, and includesfeedback loops. Standard modelling approaches may therefore be insufficient to accurately de-1405.2. Associations between WNV incidence and ecological conditions in Saskatchewanscribe the dynamics of the system (Parrott, 2010). Two specific ecological complexities shouldbe explicitly considered when interpreting the results of this study: the changing dynamics ofthe system in outbreak and non-outbreak years, and the relevance of the spatial scale at whichpredictors are quantified.The evaluation of irrigation used only data from years with WNV outbreaks in order tominimize the effects of an overabundance of zeroes and inflated incidence estimates that couldoccur in non-outbreak years. However, these outbreak years may be anomalous in the long-term epidemiology of the disease. Evaluating the drivers of disease incidence in both outbreakand non-outbreak years is important because time-varying ecological factors like avian herdimmunity affect incidence (Kwan et al., 2012). Although key demographic associations areexpected to remain constant with varying transmission intensity, some associations may onlyoccur, or be detectable, above key incidence thresholds. To my knowledge there have beenno explicit comparisons of the relationships between incidence and demographic and landscapefactors during outbreak and non-outbreak years.Ecological theory also suggests that abiotic processes like weather are particularly importantin determining patterns at large spatial scales, while biotic factors like inter-species competitionexert their effects at smaller spatial scales (Levin, 1992; McGill, 2010). However, the majority ofstudies evaluating drivers of VBD incidence, including this one, measure environmental factors ata single spatial scale determined by political boundaries (e.g. census tract, RM). The importanceof considering scale for WNV was recently shown as species richness and community structurewere most predictive of incidence when measured at smaller spatial scales, while weather oranthropogenic landscape changes were more strongly associated with patterns of disease whenmeasured over larger spatial scales (Cohen et al., 2016). This study's failure to consider scalemeans that key climate and ecological predictors may be quantified at a scale other than that atwhich they exert their effects. This may explain the unexpected associations between incidenceand both precipitation and reservoir diversity.The identification of distal determinants can provide a glimpse into the ecological conditionsthat affect WNV incidence, while also improving our understanding of the spatial distributionof risk. The inconsistency of the results presented in this thesis, however, make it difficultto translate findings regarding the primary predictor of WNV incidence into actionable publichealth interventions. The study results should not be used to support reductions in irriga-tion or modifications to avian communities. Agriculture is too important an economic driver,and any attempts to modify avian communities would undoubtedly have unintended ecologicalconsequences. Despite the inconsistent findings for key predictors, the consistent impact of tem-perature in the model and the cumulative 14-day 109 DD threshold in the descriptive statisticsdo suggest that the latter may be a useful predictor for large outbreaks in the region.In summary, study results demonstrate that heat is a dominant driver of WNV in Saskatchewan,while irrigation and avian community structure are associated with disease in inconsistent andunexpected ways. However, these study findings must be interpreted with caution despite their1415.3. WNV Decision Support in BCstatistical significance. Effect estimates for both irrigation and avian community structureshould be viewed as imprecise until additional data is acquired, or until we can expand the geo-graphic scope of the analyses. Based on study findings, only cumulative heat showed consistentand statistically significant associations sufficient to recommend its inclusion in decision supporttools. Studies comparing vector and avian infection rates in irrigated and non-irrigated land-scape may clarify the role of irrigation in facilitating transmission, while blood meal analyses forCx. tarsalis in non-urban landscapes could help identify non-passerine species that may driveamplification in these prairie regions. Work of this kind would improve our understanding ofhow landscape and reservoir community structure drive regional amplification, however, evalu-ating such features independently will only advance the field so much. Our understanding of theinteractions and interdependencies of ecological and social variables remains limited (Stephenet al., 2015), and these interdependencies can be difficult to capture using aggregate statisticalmodel (Petersen et al., 2012a). Novel approaches that attempt to deal with these interactions,and with the system as a whole, may be required. Work of this kind would also be strengthenedby a renewed emphasis on primary data collection designed around specific research questions.Doing so would address the biases and limitations that arise when creating analytic datasetsfrom data collected for other purposes.5.3 WNV Decision Support in BCThe results of the previous analyses show that our understanding of the distal drivers of diseaseremains incomplete, in part because of the importance of the contextual factors that driveamplification, and in part due to limitations of using aggregate statistical models to evaluatecomplex systems. Accurate prediction may be infeasible in BC, and instead regional publichealth should focus on improving regional situational awareness and preparedness for the rareoccasions when WNV spillover does occur (Stephen et al., 2015). Decision-making tools thatcan coarsely track the current stage of WNV amplification and can differentiate spillover fromenzootic transmission are likely sufficient for public health decision-making and planning in BC.The simplicity and ease of such tools is important because regions with sporadic activity maylack dedicated staff trained in VBD or WNV ecology.This decision support tool proposed in this research represents its practical summation andlinks key surveillance and weather inputs with a simple amplification based hazard classificationscaled to historic activity in the province. This general approach to decision support and riskestimation for WNV is not novel, as similar tools have been used in California (CaliforniaDepartment of Public Health, 2017) and Florida (Day et al., 2015). Further, decision supporttools with a similar structure have also been used previously by BC governmental decision-makers (BC Ministry of Environment, 2016). The decision tool presented here was not createdin response to any perceived failings of these other tools. Rather, the tool recognizes that aregion's immunological history and ecology play an important role in driving WNV transmission,1425.3. WNV Decision Support in BCwhich necessitates that decision-making tools be regionally focused and regionally parameterizedin order to have public health relevance (Ruiz et al., 2010). The tool as currently structuredrequires continued refinement, but can support future decision-making should WNV expand itsprovincial range beyond the Okanagan Valley.Although the development of this tool required a holistic understanding of the WNV trans-mission pathway, the selected surveillance inputs represent only a subset of factors that areimportant in determining patterns of WNV disease. Many drivers of disease, like irrigation andavian community structure, were excluded. This exclusion was done for several reasons. First,some drivers of disease were impractical for hazard monitoring because of data limitations, pri-marily limitations related to data timeliness. Second, the aim of the tool was to evaluate changesto hazard through the amplification season in a single location. Those surveillance inputs thatwere associated with spatial patterns of disease, but failed to track changing amplification overtime - like landscape - were therefore not included. Such features are better suited to approacheslike ecological niche modelling (Mak et al., 2010a,b; Wimberly et al., 2008) or like approachespreviously previously taken in Saskatchewan (Epp & Waldner, 2009). Population density andage were not included for similar reasons despite being included in statistical models. While thedemographic makeup of the region may change over years, they are unlikely to drastically shiftwithin season.Identifying appropriate thresholds for surveillance inputs proved challenging for several rea-sons. First, the rarity of WNV in BC and Washington State meant that all available datawas used to identify tentative regional thresholds. Evaluating the ability of the tool to retro-spectively classify within season hazard was therefore not possible. In addition, the contextualnature of WNV amplification means that care must be taken when applying thresholds iden-tified elsewhere. Three of the identified thresholds represented lab confirmed presence-absencethresholds that confirm the current stage of amplification (high certainty) and do not requireadditional validation. Temperature and precipitation thresholds (medium certainty) were ex-pected to hold in BC, although the temperature threshold is expected to show limited regionalvariation. The third group of thresholds, which contains WNV surveillance inputs from Wash-ington State and VIR estimates, is the most uncertain because these threshold values are basedsolely on descriptive comparisons. Despite this uncertainty, the thresholds for these surveillanceinputs were included as placeholders to ensure their consideration in future decision support.Tools like this should be constantly refined, however, threshold uncertainty should not preventthe tool from being used and being useful. When used in combination with expert opinion, thetool can continue to provide an estimate of seasonal WNV hazard that can inform preparednesseven as the tool and decision process evolve over time.Uncertainty also remains about how to best incorporate information on the intensity of theprevious year's amplification and preceding weather events, both of which can affect the currentseasons' transmission risk via regional immunity and community level ecological processes likecompetitive release (Kwan et al., 2012; Landesman et al., 2007; Shaman et al., 2005). Although1435.3. WNV Decision Support in BCantecedent conditions are important, it is difficult to capture them with a decision support toolbecause they can be negated by inclement weather in the current season. Antecedent conditionswere included using the classification of environmental predisposition, but no actionable publichealth activities were associated with this hazard classification and the information is includedonly to provide a context or situational awareness for users. Antecedent conditions were notincluded in any of the survey scenarios, making it impossible to determine if respondents equallyvalued the inclusion of such conditions for decision support.Although surveillance input-based decision support tools have been used elsewhere, thechosen set of inputs, as well as the input thresholds used in this thesis, require additionalrefinement. While traditional sensitivity- and specificity-based validation is impossible becauseof the limited regional activity, the user acceptance survey evaluated the: 1) link between riskperception and stage of amplification, 2) appropriateness of stage-specific prevention measures,and 3) ecological literacy of potential tool users.In general, the survey results confirmed that those involved with provincial and regionalWNV surveillance had a nuanced understanding of WNV disease ecology. Furthermore, respon-dents recognized the inherent link between the stages of the amplification cycle and hazard,validating the use of the proposed hazard categorization. The higher perceived risk for theearly-vs-late season detection of enzootic transmission showed that respondents had an under-standing of the higher amplification potential represented by intense early season transmissionand supported the inclusion of seasonality in the decision tool. Early season enzootic transmis-sion was perceived as having a risk similar to that of confirmed equine spillover or the presence ofasymptomatic human cases, showing a recognition of both the importance of non-human mam-mals as indicators of spillover and the assumption that early season enzootic activity will lead tolate season spillover. This assumption is likely true in the absence of subsequent unfavourableweather conditions that halt transmission.The greatest variability in respondent risk perception occurred for the scenarios that de-scribed a single asymptomatic case detected via blood donation and the scenario that describedthe early detection of enzootic transmission. Variations in the former were likely driven by prob-lems with the scenario description as only two thirds of respondents believed they were givensufficient information to make a recommendation. In addition, a greater number of respondentsrequested additional information for the scenario describing asymptomatic cases than for thescenarios describing symptomatic cases, even though an equivalent amount of detail was pro-vided for both descriptions. Requests for additional information regarding the asymptomaticinfection may have stemmed from a failure to clearly identify whether the case represented re-cent or historic infection. The variation in risk perception for early season enzootic detectionwas not unwarranted, as this scenario presents the greatest range of potential outcomes; intensespillover would occur if early enzootic detection was followed by warm temperatures, but alter-natively be halted by sub-threshold temperatures. Spillover in nearby Washington State withoutconcurrent local transmission in BC received a higher perceived risk score than scenarios with1445.3. WNV Decision Support in BClocal enzootic transmission, validating the inclusion of Washington State surveillance data inthe decision support tool.The survey results not only validated the hazard-risk linkage that is the backbone of thedecision tool, but also helped to clarify appropriate stage-specific public health prevention mea-sures. Public health messaging was the most acceptable prevention approach in all enzootic orspillover scenarios, and is ideal for low incidence settings because of its low cost and wide reach.Research on public perception in Colorado indicated that messaging should clearly state that:even mild forms of the disease cause illness, young people can become ill, and infection canoccur even with low mosquito abundance (Zielinski-Gutierrez & Hayden, 2006). Despite surveyrecommendations to the contrary, messaging promoting the reporting of dead birds should beimplemented once conditions are deemed suitable for transmission, rather than after confirmedenzootic transmission. The inherent value of dead bird surveillance is debatable (Kwan et al.,2012; Petersen et al., 2012a), especially given known provincial biases in corvid reporting (Davidet al., 2007), but its true value as a surveillance input lies in its role as the first indication ofenzootic transmission. Waiting to initiate dead-bird messaging until after enzootic transmissionhas been confirmed therefore minimizes its value.The similarity of the intervention recommendations for confirmed enzootic transmission andspillover, which were made despite respondents' clearly differing perceptions of risk, was unex-pected. Respondents likely recognized the impracticality of certain prevention approaches inlight of the regional history of WNV in BC. Both adulticiding and public health recommenda-tions to reduce outdoor activity in response to spillover were firmly rejected. As respondentswere generally experienced in the field of public health, they likely understood both the bureau-cratic requirements required to implement these type of interventions (British Columbia Centrefor Disease Control., 2010a) and the public's unfavourable perception of insecticide use (Pe-tersen et al., 2012b; Zielinski-Gutierrez & Hayden, 2006). These prevention approaches may beappropriate in response to severe outbreaks, but are not seen as necessary for limited enzootictransmission or spillover. Furthermore, adulticiding would have limited value in the Okana-gan even in response to a large outbreak, because of 1) the short amplification period (Lysyk,2010), and 2) the low population density in the region (Elnaiem et al., 2008). Adulticiding andwidespread restricting outdoor activity were removed from recommended prevention approachesin response to user feedback.Respondents approved modifications to current surveillance and prevention approaches inresponse to confirmation of circulating virus, as well as the implementation of single unit testingof the blood supply. Increasing the intensity of vector surveillance will improve the detectionprobability of low-levels of circulating virus (Gu et al., 2008), while trapping twice per-weekreduces the impact of stochastic weather events on weekly trap catch. However, both approachesrequire increased surveillance and lab costs. In contrast, transitioning to a trap-specific analysisof vector data has no additional cost beyond increasing the required staff time and would helprefine estimates of the spatial distribution of hazard. In scenarios with significant enzootic1455.3. WNV Decision Support in BCtransmission, consideration could be given to implementing regional calculations of VI (Nasci &Biggerstaff, 2005). Finally, single unit testing of the blood supply can be cost effective duringsevere outbreaks (Busch et al., 2006) and can improve the sensitivity of WNV surveillance(O'Brien et al., 2010a), but it should be implemented only after the early season detection ofspillover.One limitation of this tool remains its incomplete validation. However, tools of this kindare unlikely to ever be fully validated, and will require continued refinement in response to theevolution of the virus (Moudy et al., 2007; Snapinn et al., 2007), the disease ecology (LaDeauet al., 2008; Moudy et al., 2007), and regional WNV epidemiology. The incomplete validationwould be problematic if the tool aimed to predict future risk or differentiate between levels ofpost-spillover risk. Instead, the tool presented here aims to guide decision support and improvepublic health readiness: an objective that can be met in the absence of certainty. Althoughthe final form of the tool was never officially used to track within-season WNV hazard, it hasprovided an ecological framework with which to view and understand regional hazard. Theframework and hazard categorization schema that informed the tool's development were usedby the provincial WNV committee to guide decision-making when provincial vector surveillancewas spatially restricted. Furthermore, the framework also guided the decision-making processthat led to the reduction in provincial surveillance to only human and horse reporting. While thisdoes not validate the choice of input thresholds, it does support both the hazard categorizationschema as well as the relevance of surveillance inputs selected to represent WNV hazard andthe utility of the tool in general.The general decision support framework described in this thesis is applicable to other MBDs,given a sufficient understanding of the specific ecology of the disease in question. In fact, theapplication of this framework to a regionally novel arbovirus may be its primary public healthvalue given the cessation of provincial WNV surveillance. However, the tool in its current formwould only be generalizable to VBDs in regions with low sporadic incidence, and the degreeof disease-specific parameterization would depend on the similarity between WNV and theMBD in question. For example, the tool could potentially be applied to SLE because it sharesvectors and reservoirs with WNV (Day & Stark, 2000; Reisen et al., 2008a), but would not beapplicable to diseases with human reservoirs like dengue, malaria or zika virus. This is becausethe absence of non-human reservoirs decreases the importance of climate factors relative to socialconditions like urbanization, poverty, and social interaction, none of which are included in thetool (Eisenberg et al., 2007). Seasonality and historic transmission should also be consideredif the tool is applied to another MBD, as the tool proposed in this research was structured fortemperate regions with pronounced seasonality. Regions with year-round transmission would bebetter served by modifying the approaches used in California (California Department of PublicHealth, 2017) or Florida (Day et al., 2015).The key research topics that should be focused on in order to improve regional decisionsupport in BC are as follows. First, refining our understanding the ecological requirements for1465.3. WNV Decision Support in BCWNV transmission would improve future public health decision-making. We are gaining anunderstanding of the climatic drivers of disease, yet there is much to learn about the role ofbroad ecological, landscape and geological conditions as drivers of disease. National, or cross-provincial studies, using GIS and ecological niche modelling could be specifically informative.Second, effective regional decision-making hinges on clarifying whether WNV overwinters or issporadically re-introduced into BC from nearby jurisdictions. If the virus overwinters in BC, thensynchronicity with Washington State would result from shared environmental conditions andnon-BC surveillance inputs could be ignored. Cross-border mark-release studies of key vectorscould help clarify the hypothesized role of regional migration, although such studies wouldbe challenging. Third, avian infection surveys, coupled with blood meal analyses of regionalvectors, would help identify regionally important reservoirs. Calculating regionally-relevantmeasures of avian community structure could be used to better delineate spatial patterns ofWNV hazard. This calculation could also identify regional vector feeding shifts that may refineour understanding of high-risk periods. Finally, gaining a better understanding of the effects ofclimate change on regional WNV transmission is needed to clarify future risk. MBDs vectorsmay expand northward if climate warming occurs (Mills et al., 2010), and the length of thetransmission season may also increase (Brown et al., 2015). Understanding the multifactorialimpacts of climate change, including the role of social and behavioural changes, is requiredbefore planning can mitigate its potential impacts (Kilpatrick & Randolph, 2012; Reiter, 2001,2008).This work has focused on the development of a useful regional decision support tool, whileclarifying the unique challenges facing decision support for emerging VBDs. The framework'sstrength includes its simplicity, the direct linking of hazard to the ecology of WNV transmis-sion, and the explicit consideration of contextual factors. At the same time, the tool's simplicitycomes with a cost: reducing the complexities of the WNV transmission system to a series ofcheck-boxes only generates coarse estimates of the current stage of amplification. The tool musttherefore be viewed only as a guide to help with planning and preparedness, and should beclearly differentiated from risk prediction. There is no guarantee that the tool will correctlyclassify hazard in all years. However, the consequence of misclassification is believed to beminimal for several reasons. First, the rarity of the WNV in BC led us to intentionally designthe tool to produce low-risk, conservative public health recommendations (i.e. messaging, al-ternative analyses). Hence, incorrectly raising the hazard level does not result in expensive andcontroversial interventions like aerial adulticiding. Secondly, failing to raise the hazard levelwhen true regional hazard has increased will also have limited consequences because the proba-bility of explosive WNV transmission remains low in BC. WNV is expected to remain a low riskpublic health issue in BC for the foreseeable future, however, new MBD are emerging and thetool provides a modifiable framework to support rapid response to a new MBD or to an unusualsevere WNV event should it occur. It is important to emphasize that WNV is an evolvingdisease and that sequential years of favourable conditions could lead to a regional accumulation1475.3. WNV Decision Support in BCof the virus and localized outbreaks. The tool has already served as a useful regional guide fordecision making, and remains a valuable regional tool should WNV incidence increase. At thesame time, it should not replace the nuanced group decision-making that balances the risks,costs and benefits of implementing any public health action.148Chapter 6ConclusionsThe results of this thesis demonstrate the challenges facing public health agencies when attempt-ing to make decisions related to WNV in settings with regional knowledge gaps and uncertainty.Climate and ecological conditions are associated with WNV disease through a complex web ofinterdependent predictors (Figure 2.2). While broad classes, or types, of disease drivers maybe universal (i.e. heat, mosquito breeding sites, competent avian reservoirs), the specific setof relevant predictors and thresholds will differ between regions with different ecologies. Thedynamics of transmission are therefore context specific, and decision support tools thereforemust also be regionally specific.The study's findings showed that temperature was a consistent driver of transmission inWestern Canada. Although the findings from BC are not conclusive, they suggest that temper-ature and landscape combined to limit amplification along the coast, in the heavily populatedGeorgia Basin, and in northern BC. This limitation effectively restricted transmission to thehot valleys of the provincial interior. Temperature was the key driver of inter-year variabil-ity in WNV incidence in BC, even in regions with usual warm temperatures, because periodsof cool temperatures interrupt the amplification cycle and reduce the probability of spillover.In Saskatchewan, temperature was significantly associated with within-season patterns of inci-dence, and more importantly, appeared to determine when outbreaks occurred. In contrast, theeffects of precipitation appeared to be context-specific and difficult to quantify because mech-anistically derived summary measures have not yet been identified. Highlighting the factorslimiting WNV activity in BC helped refine our understanding of the environmental range inwhich transmission is possible. In addition, identifying climate correlates for WNV activityimproved our understanding of the regional distribution of risk in BC.Landscape and avian community structure were also statistically associated with WNV in-cidence, although variation between years and between models leaves the exact nature of theseassociations unclear. In BC, mountainous landscapes likely acted as a barrier to viral introduc-tion and likely continue to limit the amount of suitable landscape for WNV transmission. Thelimited suitable habitat for WNV amplification in BC may make the region uniquely suscep-tible to stochastic weather events. Irrigated landscape was associated with WNV incidence inSaskatchewan in some years, especially when precipitation was limited. The positive associa-tions with irrigation, and the mediating effects of precipitation, further support an evidence baselinking WNV incidence to landscape changes that alter regional hydrology. In addition, thisis one of the first studies that explicitly showed how differing irrigation methods deferentially149Chapter 6. Conclusionsaffect WNV incidence, which is a previously identified research gap (Eisen et al., 2010). How-ever, the relationship between incidence and irrigation was present in only one of two outbreaksexamined. Further work is needed to determine if the association observed in 2003 was true, orwas in fact driven by an association with other unmeasured ecological factors.How avian community structure relates to WNV incidence remains unclear. Positive asso-ciations between both non-passerine diversity and WNV incidence in Saskatchewan, and non-passerine abundance and disease incidence, did not support the dilution hypothesis proposedby others (see Section 2.5). The limited sample size and potential for residual confoundingmake it difficult to accept study findings that contradict the generally accepted understandingof associations between avian community structure and WNV risk. At the same time, thesecontrasting findings may also result from the lack of nuance that resulted from a hypothesis andmethodological approach that focused on simplistic measures of avian community structure.Previously documented associations between avian diversity and WNV incidence were likelynot driven by biodiversity per se, but instead by the abundance of key reservoir species thatdisproportionately drive amplification (Randolph & Dobson, 2012; Salkeld et al., 2013). A spe-cific focus on regionally relevant reservoir species may have led to more consistent associations.However, regional vector feeding studies are needed before such associations can be confirmed.Residual confounding resulting from unmeasured landscape features and differences in vectorcommunity structure or other ecological and environmental factors continue to offer a very realexplanation for observed patterns.The climate and ecological associations presented in this thesis should be interpreted care-fully. Despite the expectation that WNV would establish and cause consistent morbidity in BC,it has only occurred sporadically since 2009. The near absence of WNV detected in humans inBC has limited both the scope of feasible analytical approaches as well as strong, statisticallysupported conclusions. Transmission is clearly driven by the abundance and behaviour of vec-tors and reservoirs, but its measured outcomes (eg. human disease, vector infection rates) arefiltered through a complex public health reporting structure. This results in several forms ofpotential bias. First, anything that affects how we detect spatial or temporal patterns of diseasewill bias results. The high percentage of asymptomatic cases and the low negative predictivevalue of low intensity surveillance limited our ability to differentiate between low levels of WNVactivity and a true absence of activity. The consequent misclassification may be significantin low incidence settings when case reporting is passive. This is particularly relevant for thebetween-year comparisons in the BC analysis. Other surveillance inputs, like vector abundance,are also strongly affected by the sampling regime and the spatial distribution of surveillance in-puts. Analyses of spatial and temporal patterns of WNV incidence were potentially confoundedby undocumented behavioural changes like repellent use, changes in outdoor activity (Gujralet al., 2007; Reisen, 2013), or changes in the intensity of prevention measures like larviciding.In addition to these traditional limitations, we must also recognize that WNV amplificationand transmission are ecological processes. The study results suggest that associations exist150Chapter 6. Conclusionsbetween WNV incidence and climate or ecological conditions, yet the inconsistency of the find-ings hints at the limitations of using relatively simplistic epidemiological methods to capturethe complexity of the WNV transmission system. Complex ecological systems are character-ized by a hierarchical structure, emergent properties, scaling, self-organization, local interaction,multiple stable states, chaotic attractors, and unpredictability (Parrott, 2002, 2010). Recogniz-ing that WNV transmission occurs within a complex ecological system requires accepting thatsuch systems are subject to the "general indeterminacy from specific predictability" that haschallenged field ecologists for decades (Parrott, 2010). Model calibration is extremely difficultand generalizing findings to other locations can be almost impossible (Proulx, 2007). Further,models parameterized at a single point in time will not accurately capture the non-equilibriumdynamics and multiple stable states of most ecosystems (Liu et al., 2007). Linking ecological andenvironmental predictors to human WNV incidence requires two unique and complex systemsbe considered; that of WNV amplification and that of human society (i.e. health-seeking be-haviour, socioeconomic conditions, WNV prevention measures, and urbanization trends). Cou-pled societal-ecological systems reveal levels of complexity that are not seen in the componentsystems (Liu et al., 2007; Stephen et al., 2015), which further challenges modelling.The results presented here demonstrate that novel approaches may be needed to capturethe complexity of the WNV system. Viewing WNV as a complex system also has importantimplications for risk prediction. The complexity of these systems, coupled with the importance ofcontextual region-specific factors, make precise risk prediction nearly impossible. A combinationof methods, including modeling (both statistical and mathematical), surveillance and expertopinion, are likely required to guide public health decision making until we are better able todeal with complex systems and their associated uncertainty (Plowright et al., 2008). Publichealth decision support should therefore focus on simpler surveillance input-based approachesthat coarsely track the current stage of the amplification cycle. Thinking of WNV decisionmaking through a lens of situational awareness may be useful. Adopting this approach reflects aconscious recognition of ecological complexity, an acknowledgement of the stochastic nature ofthe system, and an acceptance of the need for new methods to augment current risk predictionapproaches.The decision support tool was designed with this recognition in mind, and is targeted towardpublic health agencies that must make decisions with incomplete information. BCCDC-ledWNV surveillance was halted before this tool could be fully implemented for within-seasonhazard tracking. At the same time, the tool has already shown public health value by providingthe groundwork for the decision-making process that led to the cessation of all non-humanor non-horse surveillance in the province in 2015. This decision was also supported by studyfindings suggesting that WNV will remain sporadic in BC and limited to the less populatedmountain valleys of the southern BC interior. However, the number of WNV cases in BC couldincrease if the virus expands into the Fraser Valley, where Cx. pipiens could facilitate urbantransmission. The decision support tool therefore remains a useful contingency should WNV151Chapter 6. Conclusionsincidence increase in BC, or should another zoonotic VBD affect the health of people in theprovince. The tool is currently parameterized to BC, but remains modifiable and applicableto other settings and other MBDs. At the same time, current threshold values are based onlimited data and are not fully validated; additional refinement of the tool may be required asnew information on WNV is discovered. The evaluation framework and lessons learned duringthe creation of the tool may also serve others attempting to create similar regional tools.The research presented in this thesis focused primarily on non-modifiable distal determinantsof disease, and as such, does not identify novel risk factors that are amenable to public healthintervention. Any negative consequences of agricultural landscape modification are outweighedby their positive benefits. Although the study results provided limited evidence that transi-tioning from surface-to-sprinkler based irrigation may reduce WNV incidence, this evidence isinsufficient to be the basis for public health interventions. Similarly, interventions based onavian community structure are not recommended given the inconsistency of the findings andthe controversy regarding the true nature of the association between community structure andWNV incidence (Randolph & Dobson, 2012; Salkeld et al., 2013). However, this controversyis, at its core, not a debate about whether reservoir populations affect amplification. Instead,it focuses on exactly what characteristics of avian community structure are most strongly as-sociated with human incidence. Once these associations are clarified and effective predictorsare identified, information on avian community structure may be combined with vector data toidentify areas that are ecologically primed for WNV amplification. Furthermore, this work hasimproved our understanding of the regional drivers of WNV incidence and strengthened regionalpreparedness for WNV and other MBDs.Although WNV has now circulated in North America for nearly 18-years, WNV researchremains rare in Washington State and BC. Additional research in this geographic area wouldimprove our understanding of WNV dynamics along the extremes of its habitat gradient. Re-gional research needs to: 1) clarify whether WNV overwinters in BC or is introduced annually,2) identify regionally important reservoirs, 3) advance a better understanding of the drivers ofinter-annual variation in disease activity, and 4) refine our understanding of the potential im-pacts of climate change on WNV transmission (see Section 5.3). Although traditional statisticalmodelling may be insufficient to capture the dynamics of the WNV system, it is likely that pub-lic health will continue to rely on such approaches until new analytical methods are developedto better deal with complex systems. The next steps for the statistical modelling done hereinclude: 1) incorporating other landscape factors to control for ecological conditions like breed-ing locations and, 2) re-running models with additional case data, ideally from other regions ofCanada, and including interaction terms in these models, and 3) explicitly considering spatialautocorrelation to remove the impact of uncontrolled-for spatial processes. More broadly, theglobal WNV research community needs to work toward gaining a coherent understanding of therelative importance of drivers of disease, and of how conditions interact to affect transmissionand spillover. In essence, WNV researchers have discovered a variety of puzzle pieces, each152Chapter 6. Conclusionsof which impacts transmission and disease in isolation. Now, they need to determine how thepieces fit together. A unified picture is starting to come together thanks to fascinating researchon vector feeding preferences and regionally important reservoirs (Hamer et al., 2009; Janouseket al., 2014; Kilpatrick et al., 2006; Levine et al., 2016); the interacting effects of temperatureand precipitation (Shand et al., 2016); the interaction between landscape and weather conditions(Skaff & Cheruvelil, 2016), and; how time-varying immunity in avian populations drives cyclicalpatterns of incidence (Kwan et al., 2012). Finally, additional research is needed to understandhow true landscape processes affect disease transmission (Crowder et al., 2013; Despommieret al., 2006; Lambin et al., 2010; Pradier et al., 2008; Skaff & Cheruvelil, 2016). This does notmean only summarizing the amount of specific landscape classes, but also explicitly consideringthe role of connectivity, edge effects and total habitat when attempting to understand distaldrivers of WNV incidence.This program of research was undertaken under the guidance of a university, yet wouldhave been impossible without access to provincial WNV surveillance data. The multi-speciestransmission cycle and potential for WNV transmission to occur over large areas makes it difficultfor a university-based research groups to study the disease without collaborating with publichealth institutions (Mills et al., 2010). Only stable government funded programs can providedata with the spatial extent and multi-year duration required to understand the epidemiologyof WNV, especially in low incidence regions. Recent reductions to WNV surveillance in the USand Canada (Hadler et al., 2015) will likely restrict future research to locations retaining strongsurveillance or research programs like California and Chicago, reducing our ability to validate keyresearch findings cross-regionally. In BC, the elimination of the vector surveillance program hasreduced our capacity to respond to future VBD threats. This decision is warranted as ecologicaland climate conditions appear to limit regional VBD activity; public health dollars previouslyspent on WNV surveillance should therefore be applied to more pressing public health issues.Public health experts' ten years of experience with coordinated vector surveillance has giventhe provincial public health community the knowledge necessary to re-initiate the program ifnecessary. Continued tracking of weather conditions throughout the WNV season may, however,continue to guide public health messaging in the key amplification periods.In summary, the aim of this thesis was to better understand the factors that drive, or pre-vent, WNV activity in BC and Saskatchewan, and to use the resulting information to createa practical WNV decision support tool for provincial and regional decision-making in BC. An-alyzing epidemiological, environmental, and ecological data from BC and Saskatchewan hasimproved our understanding of both the regional determinants of WNV disease in BC and thecause of the prolonged absence of the virus in the province. Doing so has also improved thesituational awareness of public health agencies in the province, leaving them better prepared forfuture arbovirus threats. However, it must be recognized that these improvements did not onlyresult from the identification of precise effect estimates or verified causal pathways, but from aholistic evaluation of WNV across locations in North America. Research on emerging diseases153Chapter 6. Conclusionsoften suffers from sparse data that can limit the complexity of available models and opens thedoor for uncontrolled confounding. This is certainly the case for this thesis. Waiting for theideal dataset is not feasible when delayed action can adversely affect public health. The specificobjectives of this thesis were met despite the data limitations and methodological shortcomings.This work has provided evidence to support the role of irrigation along the northern range ofthe WNV, while also providing additional support for the importance of temperature as a pri-mary determinant of both inter-year and within-year variability in WNV incidence. However,the inconsistency of findings cannot be ignored and point towards the challenges of using tra-ditional statistical approaches to understand complex ecological systems. Finally, the decisionsupport framework and tool - although similar to approaches used elsewhere - provides a novelapplication with proven utility to a northern region with sporadic WNV activity. Althoughmuch remains unknown about the drivers of WNV in BC and Saskatchewan, this work fits afew pieces into the overall puzzle of WNV transmission in Western Canada. Those puzzle piecesmust be clarified and refined over time through study replication and verification.154BibliographyAgriculture and Agri-Food Canada (2016). Lessons learned from the canadian drought years 2001 and 2002: Syn-thesis report. Date Retrieved: 2017-04-23. URL http://www.agr.gc.ca/eng/programs-and-services/list-of-programs-and-services/drought-watch/managing-agroclimate-risk/lessons-learned-from-the-canadian-drought-years-2001-and-2002/?id=1463593613430.Agriculture Knowledge Centre (2011). http://www.agriculture.gov.sk.ca/AKC. URL http://www.agriculture.gov.sk.ca/AKC.Aiken, L. (2003). Health Canada "nearly blindsided" by West Nile virus incidence. Canadian Medical AssociationJournal, 168, 756756.Akaike, H. (1974). A new look at the statistical model identification. Automatic Control, IEEE Transactionson, 19, 716723.Alberta Health (2016). West Nile Virus - Surveillance in Alberta. Date Retrieved: 2017-01-05. URL http://www.health.alberta.ca/health-info/WNv-evidence.html.Allan, B., Langerhans, R., Ryberg, W., Landesman, W., Griffin, N., Katz, R., Oberle, B., Schutzenhofer, M.,Smyth, K., de StMaurice, A. et al. (2008). Ecological correlates of risk and incidence of West Nile virus in theUnited States. Oecologia.Allen, J. (1976). A modified sine wave method for calculating degree days. Environmental Entomology, 5,388396. ISSN 0046-225X.Anderson, A. (2011). The Encyclopedia of Saskatchewan: Population Trends. Date Retrieved: 2017-01-20. URLhttp://esask.uregina.ca/entry/population_trends.html.Anderson, J., Ferrandino, F., Dingman, D., Main, A., Andreadis, T. & Becnel, J. (2011). Control of mosquitoesin catch basins in connecticut with bacillus thuringiensis israelensis, bacillus sphaericus, and spinosad. Journalof the American Mosquito Control Association, 27, 4555.Anderson, J. F., Andreadis, T. G., Main, A. J. & Kline, D. L. (2004). Prevalence of West Nile Virus in TreeCanopy Inhabiting Culex Pipiens and Associated Mosquitoes. The American Society of Tropical Medicine andHygiene, 71, 112119.Anderson, S. (2002). An evaluation of spatial interpolation methods in air temperature in Phoenix, Az. Tech.rep., Department of Geography, Arizona State University.Andreadis, T. G. (2012). The Contribution of Culex pipiens Complex Mosquitoes to Transmission and Persistenceof West Nile Virus in North America. Journal of American Mosquito Control Association, 28, 137151. ISSN8756-971X.Anonymous (1999). Update: West Nile-Like Viral Encephalitis New York, 1999. MMWR, 48, 890892.Apperson, C., Harrison, B., Unnasch, T., Hassan, H., Irby, W., Savage, H., Aspen, S., Watson, D., Rueda,L., Engber, B. et al. (2002). Host-feeding habits of Culex and other mosquitoes (Diptera: Culicidae) in theBorough of Queens in New York City, with characters and techniques for identification of Culex mosquitoes.Journal of Medical Entomology, 39, 777785.Arias, J. F. (2006). What is an epidemiological week and why do we use them? The Skeeter: Newsletter of theVirginia Mosquito Control Association, 66, 7.155BibliographyArtsob, H. (2000). Arthropod-borne disease in canada: A clinician's perspective from the `cold zone'. Paediatrics& child health, 5, 206.Artsob, H., Gubler, D., Enria, D., Morales, M., Pupo, M., Bunning, M. & Dudley, J. (2009). West Nile virus inthe New World: trends in the spread and proliferation of West Nile virus in the Western Hemisphere. Zoonosesand Public Health, 56, 357369. ISSN 1863-2378.Bailey, S., Hoffmann, B. & Eliason, D. (1965). Flight and Dispersal of the Mosquito Culex Tarsalis Coquillett inthe Sacramento Valley of California. Univ. of Calif.Baker, S. L. & Yan, N. D. (2010). Accumulated organic debris in catch basins improves the efficacy of S-methoprene against mosquitoes in Toronto, Ontario, Canada. Journal of the American Mosquito ControlAssociation, 26, 172182.Banerjee, S. N., Banerjee, M., Fernando, K., Burgdorfer, W. & Schwan, T. G. (1998). Tick-borne relapsingfever in British Columbia, Canada: First isolation of Borrelia hermsii. Journal of Clinical Microbiology, 36,35053508. ISSN 00951137.Barber, L. M., Schleier III, J. J. & Peterson, R. K. D. (2010). Economic Cost Analysis of West Nile VirusOutbreak, Sacramento County, California, USA, 2005. Emerging Infectious Diseases, 16, 480486. ISSN1080-6040.Barker, C., Bolling, B., Black, I., William, C., Moore, C. & Eisen, L. (2009). Mosquitoes and West Nile virusalong a river corridor from prairie to montane habitats in eastern Colorado. Journal of Vector Ecology, 34,276293.Barker, C., Eldridge, B. & Reisen, W. (2010a). Seasonal Abundance of Culex tarsalis and Culex pipiens ComplexMosquitoes (Diptera: Culicidae) in California. Journal of Medical Entomology, 47, 759.Barker, C. M., Johnson, W. O., Eldridge, B. F., Park, B. K., Melton, F. & Reisen, W. K. (2010b). Temporalconnections between Culex tarsalis abundance and transmission of Western equine encephalomyelitis virus inCalifornia. American Journal of Tropical Medicine and Hygiene, 82, 11851193. ISSN 00029637.Barker, C. M., Reisen, W. K. & Kramer, V. L. (2003). California state mosquito-borne virus surveillance andresponse plan: a retrospective evaluation using conditional simulations. American Journal of Tropical Medicineand Hygeine, 68, 50818.Bart, J., Hofschen, M. & Peterjohn, B. G. (1995). Reliability of the Breeding Bird Survey: Effects of RestrictingSurveys to Roads. The Auk, 112, 758761. ISSN 0004-8038.Barzon, L., Pacenti, M., Franchin, E., Squarzon, L., Lavezzo, E., Cattai, M., Cusinato, R. & Palù, G. (2013).The complex epidemiological scenario of West Nile virus in Italy. International Journal of EnvironmentalResearch and Public Health, 10, 46694689. ISSN 16617827.Bates, D., Mächler, M., Bolker, B. & Walker, S. (2015). Fitting linear mixed-effects models using lme4. Journalof Statistical Software, 67, 148.BC Ministry of Environment (2016). British Columbia Drought Response Plan. Date Retrieved: 2017-04-23. URL http://www2.gov.bc.ca/assets/gov/environment/air-land-water/water/drought-info/drought-response-plan-update-june-2015.pdf.Becker, N. (2008). Influence of climate change on mosquito development and mosquito-borne diseases in Europe.Parasitology Research, 103, 1928.Begon, M. (2009). Ecological epidemiology. The Princeton Guide to Ecology, 220.Bickel, R. (2007). Multilevel Analysis for Applied Research: It's Just Regression! The Guilford Press.Biggerstaff, B. (2017). PooledInfRate, version 4.0. Division of Vector-Borne Infectious Diseases. Centers forDisease Control, Fort Collins, CO.Blanford, J. I., Blanford, S. & Murdock, C. C. (2013). Temperature variation makes ectotherms more sensitiveto climate change. Global Change Biology, 23732380.156BibliographyBlitvich, B. (2008). Transmission dynamics and changing epidemiology of West Nile virus. Animal health researchreviews / Conference of Research Workers in Animal Diseases, 116.Bode, A., Sejvar, J., Pape, W., Campbell, G. & Marfin, A. (2006). West Nile virus disease: a descriptive study of228 patients hospitalized in a 4-county region of Colorado in 2003. Clinical Infectious Diseases, 42, 12341240.ISSN 1058-4838.Bødker, R., Akida, J., Shayo, D., Kisinza, W., Msangeni, H., Pedersen, E. & Lindsay, S. (2003). Relationshipbetween altitude and intensity of malaria transmission in the Usambara Mountains, Tanzania. Journal ofMedical Entomology, 40, 706717.Boere, G. & Stroud, D. (2006). The flyway concept: what it is and what it isn't. In. G.C. Boere, C.A. Galbraith& D.A. Stroud. Waterbirds around the world, 4047.Bolker, B., Brooks, M., Clark, C., Geange, S., Poulsen, J., Stevens, M. & White, J. (2009). Generalized linearmixed models: a practical guide for ecology and evolution. Trends in Ecology and Evolution, 24, 127135.Bolling, B., Barker, C., Moore, C., Pape, W. & Eisen, L. (2009). Seasonal patterns for entomological measuresof risk for exposure to Culex vectors and West Nile virus in relation to human disease cases in northeasternColorado. Journal of Medical Entomology, 46, 15191531.Bouden, M., Moulin, B. & Gosselin, P. (2008). The geosimulation of West Nile virus propagation: a multi-agentand climate sensitive tool for risk management in public health. International Journal of Health Geographics,7, 35.Bowden, S. E., Magori, K. & Drake, J. M. (2011). Regional Differences in the Association Between Land Coverand West Nile Virus Disease Incidence in Humans in the United States. The American Journal of TropicalMedicine and Hygiene, 84, 234238.Boyce, W. M., Lawler, S. P., Schultz, J. M., McCauley, S. J., Kimsey, L. S., Niemela, M. K., Nielsen, C. F.& Reisen, W. K. (2007). Nontarget effects of the mosquito adulticide pyrethrin applied aerially during aWest Nile virus outbreak in an urban California environment. Journal of the American Mosquito ControlAssociation, 23, 335339.Bradley, C. & Altizer, S. (2007). Urbanization and the ecology of wildlife diseases. Trends in Ecology andEvolution, 22, 95102.Bradley, C., Gibbs, S. E. J. & Altizer, S. (2008). Urban Land Use Predicts West Nile Virus Exposure in Songbirds.Ecological Applications.Branch, I. (2011). Statistics by R.M. and Irrigation Type. Tech. rep., Saskatchwan Government.Brault, A., Langevin, S., Bowen, R., Panella, N., Biggerstaff, B., Miller, B. & Komar, N. (2004). DifferentialVirulence of West Nile Strains for American Crows. Emerging infectious diseases, 10, 2161.Briese, T. & Bernard, K. (2005). West Nile virusan old virus learning new tricks? Journal of Neurovirology,11, 46975.Brighton, B., Bhandari, M., Tornetta, P., Felson, D. T. et al. (2003). Hierarchy of evidence: from case reportsto randomized controlled trials. Clinical Orthopaedics and Related Research, 413, 1924.British Columbia Centre for Disease (2015). Communicable Disease Control Reportable Zoonoses Guidelines.Tech. Rep. September, British Columbia Centre for Disease Control. URL http://www.bccdc.ca/resource-gallery/Documents/GuidelinesandForms/GuidelinesandManuals/Epid/CDManual/Chapter1-CDC/CompleteReportableZoonosesGuidelineFinalVers.pdf.British Columbia Centre for Disease Control (2005). West Nile Virus Surveillance Manual for Field Staff, 2005.British Columbia Centre for Disease Control. (2007). West Nile Virus Activity in British Columbia: 2007Surveillance Program Results. Date Retrieved: 03-07-2017. URL http://www.bccdc.ca/resource-gallery/Documents/Statistics%20and%20Research/Statistics%20and%20Reports/Epid/Vector-bourne/WNV/2007/WNV_2007SurveillanceReport.pdf.157BibliographyBritish Columbia Centre for Disease Control. (2008). West Nile Virus Activity in British Columbia: 2008Surveillance Program Results. Date Retrieved: 03-07-2017. URL http://www.bccdc.ca/resource-gallery/Documents/Statistics%20and%20Research/Statistics%20and%20Reports/Epid/Vector-bourne/WNV/2008/WNV_2008SurveillanceReport.pdf.British Columbia Centre for Disease Control. (2009). West Nile Virus Activity in British Columbia: 2009Surveillance Program Results. Date Retrieved: 03-07-2017. URL http://www.bccdc.ca/resource-gallery/Documents/Statistics%20and%20Research/Statistics%20and%20Reports/Epid/Vector-bourne/WNV/2009/WNVSurveillanceresults2009v2.pdf.British Columbia Centre for Disease Control. (2010a). Arbovirus Surveillance and Response Guidelines for BritishColumbia.British Columbia Centre for Disease Control. (2010b). West Nile Virus Activity in British Columbia: 2010Surveillance Program Results. Date Retrieved: 03-07-2017. URL http://www.bccdc.ca/resource-gallery/Documents/Statistics%20and%20Research/Statistics%20and%20Reports/Epid/Vector-bourne/WNV/2010/WNVSurveillanceResults20101231v3_Jan6.pdf.British Columbia Centre for Disease Control. (2011). West Nile Virus Activity in British Columbia: 2011Surveillance Program Results. Date Retrieved: 03-07-2017. URL {http://www.bccdc.ca/resource-gallery/Documents/Statistics%20and%20Research/Statistics%20and%20Reports/Epid/Vector-bourne/WNV/2011/WNVSurveillanceResults2011v2.pdf}.British Columbia Centre for Disease Control. (2012). West Nile Virus Activity in British Columbia: 2012Surveillance Program Results. Date Retrieved: 03-07-2017. URL http://www.bccdc.ca/resource-gallery/Documents/Statistics%20and%20Research/Statistics%20and%20Reports/Epid/Vector-bourne/WNV/2012/WNVAnnualSurveillanceReport2012.pdf.British Columbia Centre for Disease Control. (2013). West Nile Virus Activity in British Columbia: 2013Surveillance Program Results. Date Retrieved: 03-07-2017. URL http://www.bccdc.ca/resource-gallery/Documents/Statistics%20and%20Research/Statistics%20and%20Reports/Epid/Vector-bourne/WNV/2013/WNV2013.pdf.British Columbia Centre for Disease Control. (2014). West Nile Virus Activity in British Columbia: 2014Surveillance Program Results. Date Retrieved: 03-07-2017. URL http://www.bccdc.ca/resource-gallery/Documents/Statistics%20and%20Research/Statistics%20and%20Reports/Epid/Vector-bourne/WNV/2015/2014WNVSurveillanceReport_Jan82015_final.pdf.British Columbia Government (2012). Living water smart: Drought information. Date Retrieved: 2017-09-12. URL http://www2.gov.bc.ca/gov/content/environment/air-land-water/water/drought-flooding-dikes-dams/drought-information.British Columbia Government (2014). Animal Health Act. Date Retrieved: 2017-08-23. URL http://www.bclaws.ca/civix/document/id/complete/statreg/14016.British Columbia Government (2017). B.C. Equine Disease Surveillance and Report. Date Retrieved: 2017-02-03.URL http://www2.gov.bc.ca/gov/content/industry/agriculture-seafood/animals-and-crops/animal-health/reportable-notifiable-diseases/equine-disease-report.Brown, H., Childs, J., Diuk-Wasser, M. & Fish, D. (2008). Ecologic Factors Associated with West Nile VirusTransmission, Northeastern United States. Emerging Infectious Diseases, 14, 15391545.Brown, H. E., Young, A., Lega, J., Andreadis, T. G., Schurich, J. & Comrie, A. (2015). Projection of climatechange influences on U.S. West Nile Virus vectors. Earth Interactions, 19. ISSN 10873562.Brownstein, J., Holford, T. & Fish, D. (2004). Enhancing west nile virus surveillance, united states. EmergingInfectious Diseases, 10, 1129.Burnham, K. & Anderson, D. (2002). Model Selection and Multimodel Inference: a Practical Information-Theoretic Approach. Springer Verlag. ISBN 0387953647.158BibliographyBusch, M., Wright, D., Custer, B., Tobler, L., Stramer, S., Kleinman, S., Prince, H., Bianco, C., Foster, G.,Petersen, L. et al. (2006). West Nile virus infections projected from blood donor screening data, United States,2003. Emerging Infectious Diseases, 12, 395402.Bustamante, D. M. & Lord, C. C. (2010). Sources of Error in the Estimation of Mosquito Infection Rates Usedto Assess Risk of Arbovirus Transmission. The American Journal of Tropical Medicine and Hygiene, 82,11721184.Buth, J., Brust, R., Ellis, R. et al. (1990). Development time, oviposition activity and onset of diapause in Culextarsalis, Culex restuans and Culiseta inornata in southern Manitoba. Journal of the American MosquitoControl Association, 6, 5563.Butler, M., Lebrun, R. A., Ginsberg, H. S. & Gettman, A. D. (2006). Efficacy of methoprene for mosquito controlin storm water catch basins. Journal of the American Mosquito Control Association, 22, 333338.California Department of Public Health (2017). California State mosquito-borne virus surveillance and responseplan. Tech. rep., California Department of Public Health. URL https://www.cdph.ca.gov/Programs/CID/DCDC/CDPH%20Document%20Library/2017CAMBVirusSurveillanceResponsePlan.pdf.Calisher, C. (1994). Medically important arboviruses of the United States and Canada. Clinical MicrobiologyReviews, 7, 89.Calisher, C. H. (2000). West Nile virus in the new world: Appearance, persistence, and adaptation to a newEconiche - An opportunity taken. Viral Immunology, 13, 411414. ISSN 08828245.Campbell, G., Marfin, A., Lanciotti, R. & Gubler, D. (2002). West Nile virus. The Lancet Infectious Diseases,2, 51929.Campbell, R. & Branch, B. (1990). The birds of British Columbia. Univ of British Columbia Pr.Canada, E. (2003). Environment Canada's Top Ten Weather Stories for 2003.Canada, E. (2007). Summer 2007 Was Warmer than Normal Climate Trends and Variations Bulletin for Sum-mery 2007. Date Retrieved: 2009-04-23. URL http://www.newswire.ca/news-releases/summer-2007-was-warmer-than-normal-534309191.html.Canadian Wildlife Health Cooperative (2017). Canadian Wildlife Health Cooperative. Last Checkecd: July 23,2017. URL http://www.cwhc-rcsf.ca/index.php.Cardenas, V. M., Jaime, J., Ford, P. B., Gonzalez, F. J., Carrillo, I., Gallegos, J. E. & Watts, D. M. (2011).Yard Flooding by Irrigation Canals Increased the Risk of West Nile Disease in El Paso, Texas. Annals ofEpidemiology, 21, 922929. ISSN 10472797.Carney, R. M., Ahearn, S. C., Mcconchie, A., Glaser, C., Jean, C., Barker, C., Park, B., Padgett, K., Parker, E.,Aquino, E. & Kramer, V. (2011). Early Warning System for West Nile Virus Risk Areas, California, USA.Emerging Infectious Diseases, 17, 14451454. ISSN 10806059.Carney, R. M., Husted, S., Jean, C., Glaser, C. & Kramer, V. L. (2008). Efficacy of aerial spraying of mosquitoadulticide in reducing incidence of West Nile virus, California, 2005. Emerging Infectious Diseases, 14, 747754.Carson, P. J., Borchardt, S. M., Custer, B., Prince, H. E., Dunn-Williams, J., Winkelman, V., Tobler, L.,Biggerstaff, B. J., Lanciotti, R., Petersen, L. R. & Busch, M. P. (2012). Neuroinvasive disease and West NileVirus infection, North Dakota, USA, 1999-2008. Emerging Infectious Diseases, 18, 684686. ISSN 10806040.Centers for Disease Control and Prevention (2012). Developing an Effective Evaluation Plan, 1115.Centers for Disease Control and Prevention (2013). West Nile Virus in the United States: Guidelines forSurveillance, Prevention, and Control. Tech. rep., US Center for Disease Control. URL http://www.cdc.gov/westnile/resources/pdfs/wnvGuidelines.pdf.Centers for Disease Control and Prevention (2016). CDC: West Nile Virus - West Nile Virus Data and Maps.http://www.cdc.gov/ncidod/dvbid/westnile/. URL http://www.cdc.gov/ncidod/dvbid/westnile/.159BibliographyChancey, C., Grinev, A., Volkova, E. & Rios, M. (2015). The Global Ecology and Epidemiology of West NileVirus. BioMed Research International, 2015. ISSN 23146141.Chase, J. M. & Knight, T. (2003). Drought-induced mosquito outbreaks in wetlands. Ecol Lett, 6, 10171024.Chase, J. M. & Shulman, R. S. (2009). Wetland isolation facilitates larval mosquito density through the reductionof predators. Ecological Entomology, 34, 741747. ISSN 03076946.Chen, S., Blanford, J. I., Fleischer, S. J., Hutchinson, M., Saunders, M. C. & Thomas, M. B. (2013). EstimatingWest Nile virus transmission period in Pennsylvania using an optimized degree-day model. Vector-Borne andZoonotic Diseases, 13, 48997. ISSN 1557-7759.Chevalier, V., Tran, A. & Durand, B. (2013). Predictive modeling of West Nile Virus transmission risk in theMediterranean Basin: How far from landing? International Journal of Environmental Research and PublicHealth, 11, 6790. ISSN 16604601.Childs, C. (2004). Interpolating Surfaces in ArcGIS Spatial Analyst. ArcUser, July-September, 3235.Childs, J. E. (2007). Pre-spillover prevention of emerging zoonotic diseases: What are the targets and whatare the tools? Wildlife And Emerging Zoonotic Diseases: The Biology, Circumstances And Consequences OfCross-Species Transmission, 315, 389443.Childs, J. E. & Gordon, E. R. (2009). Surveillance and Control of Zoonotic Agents Prior to Disease Detectionin Humans. Mount Sinai Journal Of Medicine, 76, 421428.Chivian, E. & Bernstein, A. (2004). Embedded in Nature: Human Health and Biodiversity. EnvironmentalHealth Perspectives, 112, A12A13.Chuang, T. W., Hildreth, M. B., Vanroekel, D. L. & Wimberly, M. C. (2011). Weather and land cover influenceson mosquito populations in Sioux Falls, South Dakota. Journal of medical entomology, 48, 66979. ISSN0022-2585.Chuang, T. W. & Wimberly, M. C. (2012). Remote Sensing of Climatic Anomalies and West Nile Virus Incidencein the Northern Great Plains of the United States. PLoS ONE, 7, 110. ISSN 19326203.Ciota, A. T., Drummond, C. L., Drobnack, J., Ruby, M. a., Kramer, L. D. & Ebel, G. D. (2011). Emergenceof Culex pipiens from overwintering hibernacula. Journal of the American Mosquito Control Association, 27,2129. ISSN 8756-971X.Ciota, A. T. & Kramer, L. D. (2013). Vector-virus interactions and transmission dynamics of West Nile virus.Viruses, 5, 30213047. ISSN 19994915.Ciota, A. T., Matacchiero, A. C., Kilpatrick, A. M. & Kramer, L. D. (2014). The effect of temperature on lifehistory traits of Culex mosquitoes. Journal of Medical Entomology, 51, 5562. ISSN 0022-2585.Clements, J., Schulenberg, T., Lliff, M., Sullivan, B., Wood, C. & Roberson, D. (2012). The ebird/clementschecklist of birds of the world. version 6.7. Tech. rep., The Cornell Lab of Ornithology. URL http://www.birds.cornell.edu/clementschecklist/downloadable-clements-checklist.Cleveland, W. S. & Devlin, S. J. (1988). Locally Weighted Regression: An Approach to Regression Analysis byLocal Fitting. Journal of the American Statistical Association, 83, 596. ISSN 01621459.Cohen, J. M., Civitello, D. J., Brace, A. J., Feichtinger, E. M., Ortega, C. N., Richardson, J. C., Sauer, E. L.,Liu, X. & Rohr, J. R. (2016). Spatial scale modulates the strength of ecological processes driving diseasedistributions. Proceedings of the National Academy of Sciences, 113, E3359E3364. ISSN 0027-8424. URLhttp://www.pnas.org/content/113/24/E3359.abstract.Colborn, J. M., Smith, K. A., Townsend, J., Damian, D., Nasci, R. S. & Mutebi, J.-p. (2013). West NileVirus Outbreak in Phoenix , Arizona  2010 : Entomological Observations and Epidemiological Correlations.Journal of the American Mosquito Control Association, 29, 123132.Colon-Gonza, Lez, F. J., Fezzi, C., Lake, I. R. & Hunter, P. R. (2013). The Effects of Weather and ClimateChange on Dengue. PLoS Neglected Tropical Diseases, 7, 19. ISSN 19352727.160BibliographyConly, J. & Johnston, B. (2007). Why the west in west nile virus infections? The Canadian Journal of InfectiousDiseases & Medical Microbiology, 18, 285.Cook, D. R. (1977). Detection of Influential Observation in Linear Regression. Technometrics, 19, 1518. ISSN1361259X.Cook, R. L., Xu, X., Yablonsky, E. J., Sakata, N., Tripp, J. H., Hess, R., Piazza, P. & Rinaldo, C. R. (2010).Demographic and clinical factors associated with persistent symptoms after West Nile virus infection. TheAmerican Journal of Tropical Medicine and Hygiene, 83, 11336. ISSN 1476-1645.Cooke, W., Grala, K. & Wallis, R. (2006). Avian GIS models signal human risk for West Nile virus in Mississippi.International Journal of Health Geographics, 5.Costanza, R., Wainger, L., Folke, C. & Mäler, K.-G. (1993). Modeling complex ecological economic systems.American Institute of Biological Sciences, 43, 545555.Cote, M. (2011). The Encyclopedia of Saskatchewan: Climate. http://esask.uregina.ca/entry/climate.html.Crowder, D. W., Dykstra, E. A., Brauner, J. M., Duffy, A., Reed, C., Martin, E., Peterson, W., Carrière, Y.,Dutilleul, P. & Owen, J. P. (2013). West Nile Virus Prevalence across Landscapes Is Mediated by Local Effectsof Agriculture on Vector and Host Communities. PLoS ONE, 8, e55006. ISSN 19326203.Curriero, F., Heiner, K., Samet, J., Zeger, S., Strug, L. & Patz, J. (2002). Temperature and mortality in 11 citiesof the eastern United States. American Journal of Epidemiology, 155, 8087.Curry, P. (2004). Saskatchewan Mosquitoes and West Nile Virus. Blue Jay, 62, 104111.David, S. T., Mak, S., MacDougall, L. & Fyfe, M. (2007). A bird's eye view: using geographic analysis to evaluatethe representativeness of corvid indicators for West Nile virus surveillance. International Journal of HealthGeography, 6, 3.Davis, A. J., Jenkinson, L. S., Lawton, J. H., Shorrocks, B. & Wood, S. (1998). Making mistakes when predictingshifts in species range in response to global warming. Nature, 391, 7836. ISSN 0028-0836.Davis, R. S. & Peterson, R. K. D. (2008). Effects of single and multiple applications of mosquito insecticides onnontarget arthropods. Journal of the American Mosquito Control Association, 24, 270280.Day, J. (2001). Predicting St. Louis Encephalitis Virus Epidemics: Lessons from Recent, and Not So Recent,Outbreaks. Annual Reviews in Entomology, 46, 111138.Day, J. (2005). Host-seeking strategies of mosquito disease vectors. Journal of the American Mosquito ControlAssociation, 21, 1722.Day, J. F. & Stark, L. M. (2000). Frequency of Saint Louis encephalitis virus in humans from Florida, USA:1990-1999. Journal of medical entomology, 37, 626633. ISSN 0022-2585.Day, J. F., Tabachnick, W. J. & Smartt, C. T. (2015). Factors that Influence the Transmission of West NileVirus in Florida. Journal of Medical Entomology, 52, 743754. ISSN 00222585.DeGroote, J., Sugumaran, R., Brend, S., Tucker, B. & Bartholomay, L. (2008). Landscape, demographic,entomological, and climatic associations with human disease incidence of West Nile virus in the state of Iowa,USA. International Journal of Health Geographics, 7, 19.Demarchi, D. (2011). The British Columbia Ecoregion Classification. Tech. rep., Ecosystem Information Section,Minestry of the Environment. URL http://www.env.gov.bc.ca/ecology/ecoregions/.Despommier, D., Ellis, B. & Wilcox, B. (2006). The role of ecotones in emerging infectious diseases. EcoHealth,3, 281289.DiMenna, M., Bueno, R., Parmenter, R., Norris, D., Sheyka, J., Molina, J., LaBeau, E., Hatton, E. & Glass, G.(2006). Emergence of West Nile virus in mosquito (Diptera: Culicidae) communities of the New Mexico RioGrande Valley. Journal of Medical Entomology, 43, 5949.161BibliographyDinu, S., Cotar, A. I., Panculescu-Gatej, I. R., E., F., Prioteasa, F. L., Sirbu, A., Oprisan, G., Badescu, D.,Reiter, P. & Ceianu, C. S. (2015). West Nile virus circulation in South-Eastern Romania, 2011 to 2013. EuroSurveillance : European Communicable Disease Bulletin, 20. ISSN 15607917.Diuk-Wasser, M., Brown, H., Andreadis, T. & Fish, D. (2006). Modeling the Spatial Distribution of MosquitoVectors for West Nile Virus in Connecticut, USA. Vector-Borne and Zoonotic Diseases, 6, 283295.Dobson, A. (2004). Population Dynamics of Pathogens with Multiple Host Species. Source: The AmericanNaturalist, 164, 6478. ISSN 1537-5323.Dodson, B. L., Kramer, L. D. & Rasgon, J. L. (2012). Effects of larval rearing temperature on immaturedevelopment and West Nile virus vector competence of Culex tarsalis. Parasites & Vectors, 5, 199. ISSN1756-3305.Dohm, D., O'Guinn, M. & Turell, M. (2002). Effect of Environmental Temperature on the Ability of Culexpipiens (Diptera: Culicidae) to Transmit West Nile Virus. Journal of Medical Entomology, 39, 221225.Dohm, D. & Turell, M. (2001). Effect of Incubation at Overwintering Temperatures on the Replication of WestNile Virus in New York Culex pipiens (Diptera: Culicidae). Journal of Medical Entomology, 38, 462464.Drummond, C., Drobnack, J., Backenson, P., Ebel, G. & Kramer, L. (2006). Impact of trap elevation on estimatesof abundance, parity rates, and body size of Culex pipiens and Culex restuans (Diptera: Culicidae). Journalof Medical Entomology, 43, 177184.Dusek, R., Mclean, R. G., Kramer, L., Ubico, S., Dupuis, A., Ebel, G. & Guptill, S. (2009). Prevalence of WestNile virus in migratory birds during spring and fall migration. The American Journal of Tropical Medicineand Hygiene, 81, 1151.Dzodzomenyo, M. & Simonsen, P. (1999). Bancroftian filariasis in an irrigation project community in southernGhana. Tropical Medicine & International Health, 4, 1318.Ecological Stratification Working Group (1995). A National Ecological Framework for Canada. Tech. rep., Centrefor Land and Biological Resources Research. URL http://sis.agr.gc.ca/cansis/publications/ecostrat/index.html.Eidson, M., Kramer, L., Stone, W., Hagiwara, Y. & Schmit, K. (2001). Dead bird surveillance as an earlywarning system for West Nile virus. Emerging Infectious Diseases, 7, 631635. ISSN 10806040.Eisen, L., Barker, C. M., Moore, C. G., Pape, W. J., Winters, A. M. & Cheronis, N. (2010). Irrigated agricultureis an important risk factor for West Nile virus disease in the hyperendemic Larimer-Boulder-Weld area ofnorth central Colorado. Journal of Medical Entomology, 47, 93951. ISSN 0022-2585.Eisen, L., Bolling, B. G., Blair, C. D., Beaty, B. J. & Moore, C. G. (2008). Mosquito species richness, composition,and abundance along habitat-climate-elevation gradients in the northern Colorado Front Range. Journal ofMedical Entomology, 45, 800811.Eisenberg, J. N. S., Desai, M. A., Levy, K., Bates, S. J., Liang, S., Naumoff, K. & Scott, J. C. (2007). Envi-ronmental determinants of infectious disease: a framework for tracking causal links and guiding public healthresearch. Environmental Health Perspectives, 115, 121623.Eisler, D., McNabb, A., Jorgensen, D. & Isaac-Renton, J. (2004). Use of an internal positive control in a multiplexreverse transcription-PCR to detect West Nile virus RNA in mosquito pools. Journal of Clinical Microbiology,42, 841. ISSN 0095-1137.El Adlouni, S., Beaulieu, C., Ouarda, T., Gosselin, P. & Saint-Hilaire, A. (2007). Effects of climate on West NileVirus transmission risk used for public health decision-making in Quebec. International Journal of HealthGeographics, 6, 17. ISSN 1476-072X.Eldridge, B. (1987). Strategies for surveillance, prevention, and control of arbovirus diseases in western northamerica. The American Journal of Tropical Medicine and Hygiene, 37, 77S.Ellis, R. (2005). Vector Indices as an Indicator of Risk of West NIle Virus Outbreak. Tech. rep., Prairie PestManagement.162BibliographyElnaiem, D., Kelley, K., Wright, S., Laffey, R., Yoshimura, G., Reed, M., Goodman, G., Thiemann, T., Reimer,L., Reisen, W. et al. (2008). Impact of aerial spraying of pyrethrin insecticide on Culex pipiens and Culextarsalis (Diptera: Culicidae) abundance and West Nile virus infection rates in an urban/suburban area ofSacramento County, California. Journal of Medical Entomology, 45, 751757.Endsley, M. R. (1995). Toward a Theory of Situation Awareness in Dynamic Systems. Human Factors: TheJournal of the Human Factors and Ergonomics Society, 37, 3264. ISSN 0018-7208.Endy, T. P., Weaver, S. C. & Hanley, K. A. (2010). Dengue virus: past, present and future. In: Frontiers inDengue Virus Research. Horizon Scientific Press, pp. 39.Environment Canada (2012). Breeding Bird Survey - General Information: Maps and Coverage.http://ec.gc.ca/reom-mbs/default.asp?lang=En&n=02CB949F-1.Environment Canada (2016). Canada's National Climate Archive. Date Retrieved: 2017-01-04. URL http://climate.weather.gc.ca/.Epp, T. & Waldner, C. (2009). Predicting Geographical Human Risk of West Nile Virus-Saskatchewan, 2003and 2007. Canadian Journal of Public Health.Epp, T., Waldner, C., Corrigan, R. & Curry, P. (2008). Public health use of surveillance for West Nile virus inhorses: Saskatchewan, 2003-2005. Transboundary and Emerging Diseases, 55, 411416. ISSN 18651674.Epstein, P. R. & Defilippo, C. (2001). West Nile virus and drought. Global Change and Human Health, 2,19992001.ERSI (2011). ArcGIS Desktop: Release 10. Environmental Systems Research Institute.Ezenwa, V., Godsey, M., King, R. & Guptill, S. (2006). Avian diversity and West Nile virus: testing associationsbetween biodiversity and infectious disease risk. Proceedings of the Royal Society B - Biological Sciences, 273,109117.Ezenwa, V., Milheim, L., Coffey, M., Godsey, M., King, R. & Guptill, S. (2007). Land Cover Variation and WestNile Virus Prevalence: Patterns, Processes, and Implications for Disease Control. Vector-Borne and ZoonoticDiseases, 7, 173180.Farley, A. (1979). Atlas of British Columbia. The University of British Columbia Press.Fauci, A. S. & Morens, D. M. (2016). Zika Virus in the Americas - Yet Another Arbovirus Threat. The NewEngland Journal of Medicine, 374, 601603. ISSN 00416193.Florida Department of Health (2014). Surveillance and control of selected mosquito-borne diseases in Florida.Fluid Surveys (2014). Fluid Surveys: Canadian Survey Software. Date Retrieved: 2017-09-01. URL http://fluidsurveys.com/popular-topics/canada/.Fonseca, D., Keyghobadi, N., Malcolm, C., Mehmet, C., Schaffner, F., Mogi, M., Fleischer, R. & Wilkerson, R.(2004). Emerging vectors in the Culex pipiens complex. Science, 303, 1535.Gates, M. & Boston, R. (2009). Irrigation linked to a greater incidence of human and veterinary West Nile viruscases in the United States from 2004 to 2006. Preventive Veterinary Medicine.George, T. L., Harrigan, R. J., LaManna, J. A., DeSante, D. F., Saracco, J. F. & Smith, T. B. (2015). Persistentimpacts of West Nile virus on North American bird populations. Proceedings of the National Academy ofSciences, 1507747112-1507747112. ISSN 0027-8424.Gibbs, S., Wimberly, M., Madden, M., Masour, J., Yabsley, M. & Stallknecht, D. (2006). Factors Affecting theGeographic Distribution of West Nile Virus in Georgia, USA: 2002-2004. Vector-Borne and Zoonotic Diseases,6, 7382.Giladi, M., Metzkor-Cotter, E., Martin, D. A., Siegman-Igra, Y., Korczyn, A. D., Rosso, R., Berger, S. A.,Campbell, G. L. & Lanciotti, R. S. (2001). West Nile encephalitis in Israel, 1999: The New York connection.Emerging Infectious Diseases, 7, 659661. ISSN 10806040.163BibliographyGinsberg, H., Rochlin, I. & Campbell, S. (2010). The use of early summer mosquito surveillance to predict latesummer west nile virus activity. Journal of Vector Ecology, 35, 3542.Goddard, L., Roth, A., Reisen, W. & Scott, T. (2003). Vertical transmission of West Nile virus by three CaliforniaCulex (Diptera: Culicidae) species. Journal of Medical Entomology, 40, 743746.Godsey, M. S., King, R. J., Burkhalter, K., Delorey, M., Colton, L., Charnetzky, D., Sutherland, G., Ezenwa,V. O., Wilson, L. A., Coffey, M., Milheim, L. E., Taylor, V. G., Palmisano, C., Wesson, D. M. & Guptill,S. C. (2013). Ecology of potential West Nile virus vectors in Southeastern Louisiana: Enzootic transmissionin the relative absence of Culex quinquefasciatus. American Journal of Tropical Medicine and Hygiene, 88,986996. ISSN 00029637.Gómez, A., Kilpatrick, A., Kramer, L., II, A., Maffei, J., Goetz, S., Marra, P., Daszak, P. & Aguirre, A. (2008).Land Use and West Nile Virus Seroprevalence in Wild Mammals. Emerging Infectious Diseases, 14, 962965.Government of Saskatchewan (2010). Covered population (saskatchewan health). Date Retrieved: 2017-03-09.URL http://www.publications.gov.sk.ca/details.cfm?p=29128.Government of Saskatchewan (2012). Bureau of Statistics. Date Retrieved: February 20, 2011. URL http://www.stats.gov.sk.ca/.Gratto-Trevor, C. (2011). The Encyclopedia of Saskatchewan: Shorebirds. Date Retrieved: 2017-01-16. URLhttp://esask.uregina.ca/entry/shorebirds.html.Gratz, N. (1988). The impact of rice production on vector-borne disease problems in developing countries. In:Vector-borne disease control in humans through rice agro-ecosystem management. Proceedings of the workshopon research and training needs in the field of integrated vector-borne disease control in riceland agro-ecosystemsof developing countries.Gray, T. J. & Webb, C. E. (2014). A review of the epidemiological and clinical aspects of West Nile virus.International Journal of General Medicine, 7, 193203. ISSN 11787074.Gu, W., Lampman, R. & Novak, R. J. (2003). Problems in estimating mosquito infection rates using minimuminfection rate. Journal of medical entomology, 40, 595596. ISSN 0022-2585.Gu, W., Unnasch, T., Katholi, C., Lampman, R. & Novak, R. (2008). Fundamental issues in mosquito surveillancefor arboviral transmission. Transactions of the Royal Society of Tropical Medicine and Hygiene, 102, 817822.Gubler, D. (1998). Resurgent Vector-Borne Diseases as a Global Health Problem. Emerging Infectious Diseases,4, 442450.Gubler, D. (2007). The continuing spread of West Nile virus in the western hemisphere. Clinical InfectiousDiseases, 45, 103946.Gubler, D. J. (2002). The global emergence/resurgence of arboviral diseases as public health problems. Archivesof medical research, 33, 330342.Gubler, D. J. (2009). Vector-borne diseases. Revue Scientifique et Technique-Office International des Epizooties,28, 583588.Gubler, D. J., Reiter, P., Ebi, K. L., Yap, W., Nasci, R. & Patz, J. A. (2001). Climate variability and change inthe United States: potential impacts on vector-and rodent-borne diseases. Environmental Health Perspectives,109, 223.Gujral, I., Zielinski-Gutiérrez, E. C., LeBailly, A. & Nasci, R. S. (2007). Behavioral Risks for West Nile VirusDisease, Northern Colorado, 2003. Emerging Infectious Diseases, 13, 419425.Guptill, S. C., Julian, K. G., Campbell, G. L., Price, S. D. & Marfin, A. A. (2003). Early-season avian deaths fromWest Nile virus as warnings of human infection. Emerging Infectious Diseases, 9, 483484. ISSN 10806040.Hadler, J. L., Patel, D., Nasci, R. S., Petersen, L. R., Hughes, J. M., Bradley, K., Etkind, P., Kan, L. & Engel,J. (2015). Assessment of arbovirus surveillance 13 years after introduction of West Nile virus, United States.Emerging Infectious Diseases, 21, 11591166. ISSN 10806059.164BibliographyHahn, M. B., Monaghan, A. J., Hayden, M. H., Eisen, R. J., Delorey, M. J., Lindsey, N. P., Nasci, R. S. &Fischer, M. (2015). Meteorological Conditions Associated with Increased Incidence of West Nile Virus Diseasein the United States, 2004-2012. The American Journal of Tropical Medicine and Hygiene, 92, 10131022.ISSN 1476-1645.Haley, R. W. (2012). Controlling urban epidemics of West Nile virus infection. JAMA: The Journal of theAmerican Medical Association, 308, 13251326.Halpin, K., Hyatt, A. D., Plowright, R. K., Epstein, J. H., Daszak, P., Field, H. E., Wang, L. & Daniels, P. W.(2007). Emerging viruses: coming in on a wrinkled wing and a prayer. Clinical Infectious Diseases, 44,711717. ISSN 1537-6591.Hamer, G., Kitron, U., Brawn, J., Loss, S., Ruiz, M., Goldberg, T. & Walker, E. (2008). Culex pipiens (Diptera: Culicidae): A bridge vector of West Nile virus to humans. Journal of Medical Entomology, 45, 125128.Hamer, G. L., Chaves, L. F., Anderson, T. K., Kitron, U. D., Brawn, J. D., Ruiz, M. O., Loss, S. R., Walker,E. D. & Goldberg, T. L. (2011). Fine-scale variation in vector host use and force of infection drive localizedpatterns of West Nile virus transmission. PLoS ONE, 6, e23767.Hamer, G. L., Kitron, U. D., Goldberg, T. L., Brawn, J. D., Loss, S. R., Ruiz, M. O., Hayes, D. B. & Walker,E. D. (2009). Host Selection by Culex pipiens Mosquitoes and West Nile Virus Amplification. The AmericanJournal of Tropical Medicine and Hygiene, 80, 268278.Han, L., Popovici, F., Alexander, J., Laurentia, V., Tengelsen, L., Cernescu, C., Gary, H., Ion-Nedelcu, N.,Campbell, G. & Tsai, T. (1999). Risk factors for West Nile virus infection and meningoencephalitis, Romania,1996. The Journal of Infectious Diseases, 179, 2303.Hanson, R. P. (1988). Koch is dead. J Wildl Dis, 24, 193200. ISSN 00903558. URL http://www.ncbi.nlm.nih.gov/pubmed/3286904.Harb, M., Faris, R., Gad, A., Hafez, O., Ramzy, R. & Buck, A. (1993). The resurgence of lymphatic filariasis inthe Nile delta. Bulletin-World Health Organization, 71, 4949.Harrigan, R. J., Thomassen, H. A., Buermann, W. & Smith, T. B. (2014). A continental risk assessment of WestNile virus under climate change. Global Change Biology, 20, 24172425. ISSN 13652486.Harrus, S. & Baneth, G. (2005). Drivers for the emergence and re-emergence of vector-borne protozoal andbacterial diseases. International Journal for Parasitology, 35, 13091318.Hartley, D. M., Barker, C. M., Le Menach, A., Niu, T., Gaff, H. D. & Reisen, W. K. (2012). Effects of temperatureon emergence and seasonality of West Nile virus in California. American Journal of Tropical Medicine andHygiene, 86, 884894. ISSN 00029637.Hayes, C. (2001). West Nile virus: Uganda, 1937, to New York City, 1999. Annals of the New York Academy ofSciences, 951, 2537.Hayes, E., Komar, N., Nasci, R., Montgomery, S., O'Leary, D. & Campbell, G. (2005). Epidemiology andtransmission dynamics of West Nile virus disease. Emerging Infectious Diseases, 11, 116773.Healy, J. M., Reisen, W. K., Kramer, V. L., Fischer, M., Lindsey, N., Nasci, R. S., Macedo, P. A., White, G.,Takahashi, R., Khang, L. & Barker, C. M. (2014). Comparison of the efficiency and costs of west nile virussurveillance methods in california. American Journal of Tropical Medicine and Hygiene, 254. ISSN 0002-9637.Hill, A. (1965). The environment and disease: association or causation? Proceedings of the Royal Society ofMedicine, 58, 295.Hobbs, N. T. (2003). Challenges and opportunities in integrating ecological knnowledge across scales. ForestEcology and Management, 181, 223238.Hochachka, W., Dhondt, A., McGowan, K. & Kramer, L. (2004). Impact of West Nile Virus on American Crowsin the Northeastern United States, and Its Relevance to Existing Monitoring Programs. EcoHealth, 1, 6068.ISSN 1612-9202.165BibliographyHubalek, Z. (2000). Measures of species diversity in ecology: An evaluation. Folia Zoologica, 49, 241260. ISSN01397893.Huel, D. (2000). Managing Saskatchewan Wetlands. A landowner's guide. 1st edn. Saskatchewan WetlandConservation Corporation, Houghton Boston, Saskatoon, Saskatchewan. ISBN 1-896793-26-6.Huhn, G. D., Sejvar, J. J., Montgomery, S. P. & Dworkin, M. S. (2003). West Nile virus in the United States:An update on an emerging infectious disease. American Family Physician, 68, 653660+671. ISSN 0002838X.Hurlbert, S. H. (1984). Pseudoreplication and the design of ecological field experiments. Ecological Monographs,54, 187212. ISSN 0012-9615.Ivan, M., Schopflocher, D. P., Svenson, L. W., Tilley, P. & Keays, G. (2005). Estimating the infection rate ofWest Nile Virus in Alberta. Tech. rep., Alberta Health and Wellness.Jackson, L. E., Kurtz, J. & Fisher, W. S. (2000). Evaluation guidelines for ecological indicators. Tech. rep.,Environmental Protection Agency, Office of Research and Development.Janousek, W. M., Marra, P. P. & Kilpatrick, a. M. (2014). Avian roosting behavior influences vector-hostinteractions for West Nile virus hosts. Parasites & vectors, 7, 399. ISSN 1756-3305.Jean, C., Honarmand, S., Louie, J. & Glaser, C. (2007). Risk factors for West Nile virus neuroinvasive disease,California, 2005. Emerging Infectious Diseases, 13, 1918.Johnson, B. J., Munafo, K., Shappell, L., Tsipoura, N., Robson, M., Ehrenfeld, J. & Sukhdeo, M. V. K. (2012).The roles of mosquito and bird communities on the prevalence of West Nile virus in urban wetland andresidential habitats. Urban Ecosystems, 15, 513531. ISSN 10838155.Johnson, B. J. & Sukhdeo, M. V. K. (2013). Drought-induced amplification of local and regional West Nile virusinfection rates in New Jersey. Journal of Medical Entomology, 50, 195204. ISSN 0022-2585.Jones, C., Lounibos, L., Marra, P. & Kilpatrick, A. (2012). Rainfall influences survival of culex pipiens (diptera:Culicidae) in a residential neighborhood in the mid-atlantic united states. Journal of Medical Entomology, 49,467473.Karesh, W. B., Dobson, A., Lloyd-Smith, J. O., Lubroth, J., Dixon, M. A., Bennett, M., Aldrich, S., Harrington,T., Formenty, P., Loh, E. H., MacHalaba, C. C., Thomas, M. J. & Heymann, D. L. (2012). Ecology ofzoonoses: Natural and unnatural histories. The Lancet, 380, 19361945. ISSN 01406736.Keesing, F., Holt, R. D. & Ostfeld, R. S. (2006). Effects of species diversity on disease risk. Ecology Letters, 9,48598.Keiser, J., Maltese, M., Erlanger, T., Bos, R., Tanner, M., Singer, B. & Utzinger, J. (2005). Effect of irrigated riceagriculture on Japanese encephalitis, including challenges and opportunities for integrated vector management.Acta Tropica, 95, 4057.Keller, C. & Scallan, J. (1999). Potential roadside biases due to habitat changes along breeding bird surveyroutes. Condor, 101, 5057. ISSN 0010-5422.Kent, R., Juliusson, L., Weissmann, M., Evans, S. & Komar, N. (2009). Seasonal Blood-Feeding Behavior ofCulex tarsalis (Diptera: Culicidae) in Weld County, Colorado, 2007. Journal of Medical Entomology, 46,380390.Kessler, S. & Guerin, P. M. (2008). Responses of Anopheles gambiae, Anopheles stephensi, Aedes aegypti, andCulex pipiens mosquitoes (Diptera: Culicidae) to cool and humid refugium conditions. Journal of VectorEcology, 33, 145149. ISSN 1081-1710.Kettyls, G., Verrall, V., Wilton, L., Clapp, J., Clarke, D. & Rublee, J. (1972). Arbovirus infections in man inbritish columbia. Canadian Medical Association Journal, 106, 1175.Kilpatrick, A., Daszak, P., Jones, M., Marra, P. & et al. (2006). Host heterogeneity dominates West Nile virustransmission. Proceedings of the Royal Society B - Biological Sciences.166BibliographyKilpatrick, A., Kramer, L., Campbell, S., Alleyne, E., Dobson, A. & Daszak, P. (2005). West Nile virus riskassessment and the bridge vector paradigm. Emerging Infectious Diseases, 11, 4259.Kilpatrick, A. M. (2011). Globalization, land use and the invasion of West Nile virus. Science, 21, 323327. ISSN15378276.Kilpatrick, A. M., LaDeau, S. L. & Marra, P. P. (2007). Ecology of West Nile virus transmission and its impacton birds in the western hemisphere. Auk, 124, 11211136.Kilpatrick, A. M., Meola, M. A., Moudy, R. M., Kramer, L. D. & Buchmeier, M. J. (2008). Temperature, ViralGenetics, and the Transmission of West Nile Virus by Culex pipiens Mosquitoes. PLoS Pathog, 4, e1000092.Kilpatrick, A. M. & Pape, W. J. (2013). Predicting human West Nile Virus infections with mosquito surveillancedata. American Journal of Epidemiology, 178, 829835. ISSN 00029262.Kilpatrick, A. M. & Randolph, S. E. (2012). Drivers, dynamics, and control of emerging vector-borne zoonoticdiseases. Lancet, 380, 19461955. ISSN 15378276.Klee, A. L., Maidin, B., Edwin, B., Poshni, I., Mostashari, F., Fine, A., Layton, M. & Nash, D. (2004). Long-termprognosis for clinical West Nile virus infection. Emerging infectious diseases, 10, 140511. ISSN 1080-6040.Komar, N., Burns, J., Dean, C., Panella, N., Dusza, S. & Cherry, B. (2001). Serologic evidence for West Nilevirus infection in birds in Staten Island, New York, after an outbreak in 2000. Vector-Borne and ZoonoticDiseases, 1, 191196.Komar, N., Langevin, S., Hinten, S., Nemeth, N., Edwards, E., Hettler, D., Davis, B., Bowen, R. & Bunning, M.(2003). Experimental Infection of North American birds with the New York 1999 S