UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Ecological niche modeling of Cryptococcus gattii in British Columbia Mak, Sunny Y. 2007

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
831-ubc_2007-0499.pdf [ 27.18MB ]
Metadata
JSON: 831-1.0093069.json
JSON-LD: 831-1.0093069-ld.json
RDF/XML (Pretty): 831-1.0093069-rdf.xml
RDF/JSON: 831-1.0093069-rdf.json
Turtle: 831-1.0093069-turtle.txt
N-Triples: 831-1.0093069-rdf-ntriples.txt
Original Record: 831-1.0093069-source.json
Full Text
831-1.0093069-fulltext.txt
Citation
831-1.0093069.ris

Full Text

ECOLOGICAL NICHE MODELING OF CRYPTOCOCCUS GATTII IN BRITISH COLUMBIA by S U N N Y Y . M A K B . A . , The University of British Columbia, 1999 Dipl .T. , British Columbia Institute of Technology, 2000 A THESIS S U B M I T T E D I N P A R T I A L F U L F I L L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F M A S T E R O F S C I E N C E in T H E F A C U L T Y O F G R A D U A T E S T U D I E S (Geography) T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A August 2007 © Sunny Y . Mak, 2007 A B S T R A C T Cryptococcus gattii unexpectedly emerged on Vancouver Island, British Columbia (BC) , Canada in 1999 causing human and animal illness. Prior to its discovery on Vancouver Island, this microscopic fungal organism was limited to tropical and sub-tropical regions of the world with eucalyptus trees as the environmental reservoir. Environmental sampling for C. gattii in southwestern B C has isolated the organism from native vegetation, soil, air and water. Since it is not possible to sample every location for the presence or absence of C. gattii on Vancouver Island or the B C mainland, ecological niche modeling using the Genetic Algorithm for Rule-set Prediction ( G A R P ) was performed to identify the optimal and potential ecological niche areas of C. gattii in B C . Human and animal surveillance and environmental sampling data were used as input data points to build and test the ecological niche models based on 15 predictor environmental data layers (topographic, climatic, biogeoclimatic, and soil). Training and testing accuracy of the C. gattii ecological niche models were 99.4% and 99.2% based on the distribution of human cases, 98.7% and 98.3% based on the distribution of animal cases, and 99.7% and 99.7% based on the distribution of positive environmental sampling locations (p-value <0.0001 for all models). Forecasted optimal C. gattii ecological niche areas in B C include the central and south eastern coast of Vancouver Island, G u l f Islands, Sunshine Coast and Vancouver Lower Mainland. They are characterized by areas of low lying elevations, daily January average temperatures above freezing, and presence within the Coastal Douglas-fir and Coastal Western Hemlock xeric maritime biogeoclimatic zones. The results of these analyses are visualized using Geographic Information Systems, and shared with public health to prioritize future C. gattii i i environmental sampling in previously unidentified areas and increase public and physician awareness of cryptococcal disease in B C . i i i TABLE OF CONTENTS Abstract i i Table of Contents iv List of Tables v i List of Figures v i i Acknowledgements x 1.0 Introduction 1 1.1 Study Objective 3 1.2 Study Area 4 1.3 Thesis Structure 5 2.0 Cryptococcus gattii 6 2.1 Biology 7 2.2 Ecology 8 2.2.1 Cryptococcus gattii in the Tropics and Subtropics 8 2.2.2 Cryptococcus gattii in Southwestern British Columbia 10 2.2.3 Physical and Biotic Characteristics of Cryptococcus gattii in the Environment... 11 2.3 Cryptococcosis 12 2.3.1 Cryptococcosis in B C 14 2.4 Laboratory Methods Used to Identify Cryptococcus gattii 15 3.0 Ecological Niche Modeling 17 3.1 The Ecological Niche 19 3.2 Selection of Model and Scale of Analysis 21 3.3 G A R P Methodology 22 3.3.1 Rule Types 23 3.3.2 Predictive Accuracy 24 3.4 Applications of G A R P Described in the Literature 25 3.5 Landscape Epidemiology 25 4.0 Data and Methods 27 4.1 Software and Hardware 27 4.2 Disease Surveillance Data 28 4.2.1 Human Cases 28 4.2.2 Animal Cases 29 4.2.3 Environmental Sampling 29 4.3 Environmental Data Layers 30 4.3.1 Topographic 31 4.3.2 Climatic 32 iv 4.3.3 Biogeoclimatic 33 4.3.4 Soi l 33 4.4 G A R P Modeling of Cryptococcus gattii 34 4.4.1 Cryptococcus gattii Data Used in the Ecological Niche Modeling 34 4.4.2 Environmental Data Layer Jackknifing 36 4.4.3 G A R P Modeling with Significant Environmental Data Layers Only 38 5.0 Results and Discussion 40 5.1 Environmental Data Layer Jackknifing 40 5.2 Ecological Niche Model Predictions 43 5.2.1 Ecological Niche Model Prediction Maps 45 5.3 Validation of the Ecological Niche Model Predictions 52 5.4 Cryptococcus gattii Ecological Niche Characterization 56 5.5 Appropriateness of the Model Properties, Methods and Data 58 5.5.1 Spatial Scale of Analysis 58 5.5.2 Cryptococcus gattii Data 59 5.5.3 Environmental Data Layers 64 5.5.4 Use of Significant Environmental Data Layers 65 5.5.5 Interpretation of the Ecological Niche Model Prediction Maps 66 6.0 Summary and Conclusion 68 6.1 Impact and Implications of the Study Findings 69 6.2 Study Strengths 71 6.3 Study Limitations 72 6.4 Proposed Possible Future Studies 74 References 77 Appendices 85 Appendix A Worldwide Distribution of Cryptococcus gattii 85 Appendix B Ethical Review Certificate of Approval 93 Appendix C Climate Station Data 94 Appendix D High Resolution Climate Modeling 98 D1.0 Introduction 98 D2.0 Data and Methods 99 D2.1 Temperature Modeling 99 D2.2 Precipitation Modeling 128 D3.0 Results and Discussion 139 Appendix E Biogeoclimatic Zones o f British Columbia 142 v L I S T O F T A B L E S Table 1. Published predictive ecological niche models 18 Table 2. Summary of environmental data layer jackknifing for C. gattii E N M based on the distribution of human cases 40 Table 3. Summary of environmental data layer jackknifing for C. gattii E N M based on the distribution of animal cases 41 Table 4. Summary of environmental data layer jackknifing for C. gattii E N M based on the distribution of positive environmental sampling locations from permanently established sites 42 Table 5. Summary of C. gattii E N M results for different data inputs and environmental data layers used to build the models 44 Table 6. Summarized environmental characteristics of the forecasted optimal C. gattii ecological niche in B C based on the distribution of human and animal cases and permanently established C. gattii sites, and environmental data layers with >95% accuracy 56 Table 7. Summarized biogeoclimatic zone characteristics of the forecasted optimal C. gattii ecological niche areas based on the distribution of human and animal cases and permanently established C. gattii sites, and environmental data layers with >95% accuracy 57 Table A l . Worldwide geography of Cryptococcus gattii 86 Table C I . Number of climate stations used in the surface interpolation models 97 Table D l . Cross-validation prediction errors of the I D W surface interpolation models 139 Table D2. Validation prediction errors of the I D W surface interpolation models 140 Table E l . Characteristics of the biogeoclimatic zones of British Columbia 143 v i L I S T O F F I G U R E S Figure 1. Geographic distribution of human and animal cryptococcosis and isolation of C. gattii from the environment in southwestern British Columbia 2 Figure 2. Worldwide distribution of exported eucalyptus trees 9 Figure 3. Chest radiograph and computed tomography scan of a patient showing three cryptococcomas in the lungs 13 Figure 4. Dark brown Cryptococcus colonies on Staib media 15 Figure 5. C G B agar turns blue for C. gattii 16 Figure 6. Restriction fragment length polymorphism profiles for selected human, animal and environmental C. gattii isolates from British Columbia, Washington and Oregon ....16 Figure 7. The fundamental and realized niches 20 Figure 8. Data reclassification of aspect to yield a simplified aspect direction raster layer 31 Figure 9. C. gattii disease surveillance and environmental sampling data used in the ecological niche modeling 35 Figure 10. Desktop G A R P settings and properties for the jackknifing o f environmental data layers for the E N M of C. gattii based on the distribution of human cases 37 Figure 11. Desktop G A R P settings and properties for the E N M of C. gattii based on the distribution of human cases and environmental data layers with training and testing accuracy >95% 39 Figure 12. C. gattii ecological niche prediction maps based on the distribution of human and animal cases and positive environmental sampling locations from permanently established sites, and environmental data layers with >95% accuracy 48 Figure 13. C. gattii ecological niche prediction maps based on the distribution of human and animal cases and positive environmental sampling locations from permanently established sites, and environmental data layers with >90% accuracy 49 Figure 14. C. gattii ecological niche prediction maps based on the distribution of human and animal cases and positive environmental sampling locations from permanently established sites, and environmental data layers with >80% accuracy 50 Figure 15. C. gattii ecological niche prediction maps based on the distribution of human arid animal cases and positive environmental sampling locations from permanently established sites, and all environmental data layers 51 Figure 16. Validation of C. gattii E N M predictions on the B C mainland based on the model input of human cases and environmental data layers with >95% accuracy 53 Figure 17. Validation of C. gattii E N M predictions on the B C mainland based on the model input of animal cases and environmental data layers with >95% accuracy 54 Figure 18. Validation of C. gattii E N M predictions on the B C mainland based on the model input of positive environmental sampling locations and environmental data layers with >95% accuracy 55 Figure C I . Data flow of climate station data used in the surface interpolation models 95 Figure C2 . Geographic distribution of climate stations used in the temperature and precipitation surface interpolation models 96 Figure D l . Histogram and QQ-plot of the adjusted January average temperature 102 v i i Figure D2. Histogram and QQ-plot of the log transformed adjusted January average temperature 102 Figure D3 . Histogram and QQ-plot of the square root transformed adjusted January average temperature 103 Figure D4. Histogram and QQ-plot of the squared transformed adjusted January average temperature 103 Figure D5 . I D W interpolation properties of the adjusted January average temperature 104 Figure D6. Validation of the adjusted January average temperature I D W interpolation 104 Figure D7. Resulting I D W surface and map of January average temperature 105 Figure D8. Elevation corrected January average temperature 106 Figure D9. Histogram and QQ-plot of the adjusted January maximum temperature 107 Figure D10. I D W interpolation properties of the adjusted January maximum temperature 107 Figure D l 1. Validation of the adjusted January maximum temperature I D W interpolation 108 Figure D12. Resulting I D W surface and map of January maximum temperature 109 Figure D13. Elevation corrected January maximum temperature 110 Figure D14. Histogram and QQ-plot of the adjusted January minimum temperature I l l Figure D15. I D W interpolation properties of the adjusted January minimum temperature I l l Figure D16. Validation of the adjusted January minimum temperature I D W interpolation 112 Figure D17. Resulting I D W surface and map of January minimum temperature 113 Figure D18. Elevation corrected January minimum temperature.... 114 Figure D19. Histogram and QQ-plot of the adjusted July average temperature 115 Figure D20. Histogram and QQ-plot of the log transformed adjusted July average temperature 115 Figure D21. Histogram and QQ-plot of the square root transformed adjusted July average temperature 116 Figure D22. Histogram and QQ-plot of the squared transformed adjusted July average temperature 116 Figure D23. I D W interpolation properties of the adjusted July average temperature 117 Figure D24. Validation of the adjusted July average temperature I D W interpolation 117 Figure D25. Resulting I D W surface and map of July average temperature 118 Figure D26. Elevation corrected July average temperature 119 Figure D27. Histogram and QQ-plot of the adjusted July maximum temperature 120 Figure D28. I D W interpolation properties of the adjusted July maximum temperature 120 Figure D29. Validation of the adjusted July maximum temperature I D W interpolation 121 Figure D30. Resulting I D W surface and map of July maximum temperature 122 Figure D31. Elevation corrected July maximum temperature 123 Figure D32. Histogram and QQ-plot of the adjusted July minimum temperature 124 Figure D33. I D W interpolation properties of the adjusted July minimum temperature 124 Figure D34. Validation of the adjusted July minimum temperature I D W interpolation 125 Figure D35. Resulting I D W surface and map of July minimum temperature 126 Figure D36. Elevation corrected July minimum temperature 127 Figure D37. Histogram and QQ-plot of total January precipitation 129 v i i i Figure D38. Histogram and QQ-plot of the log transformed total January precipitation 129 Figure D39. Histogram and QQ-plot of the square root transformed total January precipitation 130 Figure D40. Histogram and QQ-plot of climate station elevation 130 Figure D41. Histogram and QQ-plot of the log transformed climate station elevation 131 Figure D42. Histogram and QQ-plot of the square root transformed climate station elevation 131 Figure D43. Trend analysis of total January precipitation and climate station elevation ....132 Figure D44. Cokriging properties of total January precipitation surface interpolation 133 Figure D45. Resulting cokriging surface and map of total January precipitation 134 Figure D46. Comparison of original 4 km cell size P R I S M dataset and 600 m cell size re-sampled dataset 135 Figure D47. Re-sampled P R I S M January total precipitation 136 Figure D48. Re-sampled P R I S M July total precipitation 137 Figure D49. Re-sampled P R I S M annual total precipitation 138 Figure E l . System of naming and coding biogeoclimatic subzones 142 ix A C K N O W L E D G E M E N T S I am indebted to many colleagues and friends who have supported this project. First, I wish to thank my graduate studies supervisory committee for providing valuable advice, feedback and direction. Brian Klinkenberg has supported my GIS studies and work over the past decade, and I am sure that we w i l l continue to collaborate on many biomedical geography research projects in the future. I also thank Karen Bartlett for providing the environmental sampling data used in this thesis project, and for expressing her excitement in the use of GIS and spatial analysis for Cryptococcus gattii research. I would also like to acknowledge Murray Fyfe for having the vision to see the need for the use of GIS for spatial epidemiological studies at the B C Centre for Disease Control. I also acknowledge the support that I received from the B C Centre for Disease Control. I would like to thank Bonnie Henry and David Patrick for supporting my studies and career development and the Cryptococcus team for their epidemiological and laboratory support. I thank Colleen Duncan for sharing the animal case data collected from her Masters research, Sarah K i d d for providing a summary of the C. gattii molecular typing of isolates described in published studies, and Judy K w a n and Kaoru Tachiiri for providing the climatic data. Lastly and most importantly, I thank my family for their love and support of my studies, work and life. In particular, I thank my wife A n a for enduring the occasional neglect experienced at times during my studies. 1.0 I N T R O D U C T I O N On August 27th, 2001 the British Columbia Centre for Disease Control ( B C C D C ) issued a public health advisory on an outbreak of cryptococcal disease on Vancouver Island, British Columbia (BC) , Canada. The fungal organism Cryptococcus gattii was identified as the causative agent of disease. This pathogen can cause a rare, potentially lethal infection of the lungs and/or central nervous system (cryptococcosis) in humans and animals. Infection can occur simply through environmental exposure and inhalation of the microscopic-sized fungal propagules (yeast form or spores). Before its discovery on Vancouver Island, C. gattii had been typically associated with the red gum group of eucalyptus tree species in tropical and sub-tropical regions of the world such as Australia, Africa, India, Italy, Papua N e w Guinea, South America and southern California (Casadevall and Perfect, 1998; El l i s and Pfeiffer, 1990; Kwon-Chung and Bennett 1984; Sorrell, 2001). Furthermore, the worldwide distribution of cryptococcosis due to C. gattii was strongly associated with the worldwide distribution of eucalyptus trees. Environmental sampling for C. gattii has isolated it from native vegetation, soil, air and water on the south and central eastern coast of Vancouver Island, and more recently on the B C G u l f Islands, Vancouver Lower Mainland and Whatcom County o f Washington State (Kidd et al., 2004; K i d d et al., 2007a; K i d d et a l , 2007b; MacDougall et a l , 2007). To date over 170 human cases and 330 animal cases have been reported to public health in B C . This represents the highest rate of C. gattii infection and the only documented multi-species outbreak of cryptococcal disease in the world (MacDougall et al., 2007; Stephen et al., 2002). 1 The geographic distribution o f cryptococcosis in humans and animals and the isolation of C. gattii from the environment in southwestern B C are directly correlated with the Coastal Douglas-fir (CDF) and Coastal Western Hemlock xeric maritime (CWHxm) biogeoclimatic zones (Figure 1). The C D F and C W H x m biogeoclimatic zones are located in the rainshadow of the Vancouver Island and Olympic Mountain ranges and are characterized by warm, dry summers, mild, wet winters, low elevation and dry, fine-textured soils (Meidinger and Pojar, 1991). Figure 1. Geographic distribution of human and animal cryptococcosis and isolation of C. gattii from the environment in southwestern British Columbia. Cases are mapped by address o f residence and include travel-related cases to Vancouver Island. Source: K i d d et al., 2007b. © American Society for Microbiology, 2007, by permission. 2 1.1 Study Objective The purpose of this study was to help public health officials delineate the geographic areas where C. gattii is currently established and forecast areas that could support C. gattii in the future as the organism continues to spread on Vancouver Island and the B C mainland. Residents of and visitors to these areas could also be specifically informed about the public health risk, albeit low, of cryptococcal disease. Since it is not feasible to sample every location in the province for the presence or absence of C. gattii, ecological niche modeling was used to predict the spatial distribution of C. gattii by relating the environmental characteristics of field observations to the predictor variables. Geographic Information Systems (GIS) and the Genetic Algorithm for Rule-set Prediction ( G A R P ) were used to model the ecological niche of C. gattii in B C . The utility of ecological niche modeling based on human and animal C. gattii disease surveillance data was also examined. This study complements the previous work and adds to the body of knowledge o f C. gattii in B C . The epidemiology, genetic characterization, geographic distribution and environmental sampling for C. gattii in B C have been described elsewhere by Bartlett et al. (2004), Duncan et al. (2005; 2006), Fraser et al. (2003; 2005), K i d d et al. (2004; 2005; 2007a; 2007b), Lester et al. (2004), MacDougall and Fyfe (2006), MacDougal l et al. (2007), Mak et al. (2004) and Stephen et al. (2002). 3 1.2 Study A r e a British Columbia is the westernmost province of Canada. Its approximate geographic coordinate range is N48.5-60 0 latitude and W l 12-130° longitude, encompassing an area of 944,735 km 2 . The physical environment of B C is extremely diverse due to its mountainous terrain along the Pacific coast and the Continental Divide, and complex systems of rivers and valleys traversing the province. The unique geography of B C produces some of Canada's most extreme climates and landscapes. During the summer months in the B C interior, daily maximum temperatures in the Lillooet (N50°41 ' W121°56 ' ) and Lytton (N50°14 ' W121°34 ' ) areas often exceed 40°C; whereas, many alpine areas along the northern B C coast remain snow covered year round. Precipitation is equally varied with Henderson Lake (N49°06 ' W125°03 ' ) on Vancouver Island recording the greatest annual average precipitation of 6,655 mm, while the Osoyoos (N49°02 ' W l 19°27') area receives very little precipitation and is the only desert in Canada (Heidorn, 2004). Vancouver Island and the Vancouver Lower Mainland are located in the southwestern corner of the province. The 2001 population of this area was approximately 2.9 mi l l ion (700,000 on Vancouver Island and 2.2 mil l ion in the Vancouver Lower Mainland; Statistics Canada, 2003). The vast majority of the population on Vancouver Island is concentrated in urban centres distributed along the southern and central east coast, which has corresponded with the observations of human and animal cryptococcosis and isolation of C. gattii from the environment (Figure 1). 4 1.3 Thesis Structure In this thesis I provide an overview of the biology, ecology and epidemiology of C. gattii in Section 2, and a discussion of ecological niche modeling and G A R P methodology in Section 3. In Section 4 I describe the data and methods used to model the ecological niche of C. gattii in B C , and present the results and discussion in Section 5. This thesis is concluded with, in Section 6, a summary of the study findings and discussion of their implications for future C. gattii disease surveillance and research in B C . In addition, an extensive review of the C. gattii literature describing the worldwide distribution of C. gattii human and animal infections and isolation of C. gattii in the environment is presented in Appendix A . The ethical review certificate of approval granted by the University o f British Columbia Behavioural Research Ethics Board for this research is presented in Appendix B . A summary of the climate data used in the surface interpolation modeling is described in Appendix C , and the high resolution climate modeling procedure and results of the temperature and precipitation datasets used in the ecological niche modeling of C. gattii in B C are presented in Appendix D . Lastly, a brief description of the biogeoclimatic zones of B C is provided in Appendix E . 5 2.0 CRYPTOCOCCUS GATTII Cryptococcosis is one of the most prevalent life-threatening fungal diseases in the world (Meyer et al., 2003). Infection occurs when the microscopic-sized fungal propagules from the environment are inhaled into the lungs and successfully colonize there. The infection may subsequently lead to pneumonia and/or invade the central nervous system causing meningitis. The majority of cases are caused by the yeast-like organisms Cryptococcus neoformans variety grubii (serotype A ) , Cryptococcus neoformans variety neoformans (serotype D) and the hybrid variety (serotype A D ) in hosts with compromised immune systems such as H I V / A I D S patients (Casadevall and Perfect, 1998). In contrast, a third, relatively rare variety, Cryptococcus neoformans variety gattii (serotypes B and C) primarily affects immunocompetent hosts (Sorrell, 2001). C. n. var. gattii was first recognized in 1970 by Vanbreuseghem and Takashio as it differed from C. n. var. neoformans in biology, ecology, epidemiology and clinical manifestations of disease (Sorrell, 2001). Recently, C. n. var. gattii was raised to species level as Cryptococcus gattii in light of genealogical and biochemical differences between the varieties of C. neoformans (Kwon-Chung et al., 2002). Therefore, for the remainder of this thesis, the species name C. gattii w i l l be used. The biology, ecology, epidemiology and laboratory methods used to identify C. gattii are described in this section. 6 2.1 Biology C. gattii is a basidiomycetous yeast. The phylum basidiomycota consists of fungi that produce spores that are formed outside a pedestal-like structure, the basidium (Southeast Missouri State University, Department of Biology, 2006). The basidospores are approximately 1-2 jum in size, so when the spores are aerosolized and inhaled into the lungs of a mammal they can reach the deeper areas of the lungs (e.g. bronchioles) to cause infection (Casadevall and Perfect, 1998; K i d d et al., 2004). The reproduction cycle of C. gattii consists of a sexual and asexual state. Casadevall and Perfect (1998) believe that the predominant mode of reproduction is asexual budding. In the laboratory, Kwon-Chung (1976) observed sexual reproduction under conditions of nitrogen starvation. However, sexual reproduction has not been observed in nature. Molecular typing of C. gattii isolates using polymerase chain reaction (PCR), restriction fragment length polymorphism (RFLP) , amplified fragment length polymorphism ( A F L P ) , and multi-locus sequence typing ( M L S T ) fingerprinting techniques enable genetic characterization of different C. gattii strains. Four major molecular types of C. gattii have been identified: V G I , V G I I , VGII I and V G I V (Meyer et al., 2003). Molecular typing enables assessment of genetic similarity or dissimilarity among isolates - providing clues to ancestry and history of geographic dispersion. 7 2.2 Ecology C. gattii is believed to compete with a variety of microorganisms in the soil, and on the bark and in hollows of trees. The ability of C. gattii to successfully compete with other microorganisms may be a primary factor, along with a suitable climatic regime, that determines whether C. gattii can establish itself in a new environment. C. gattii has previously been introduced into new areas of the world via importation of eucalyptus trees which act as an environmental reservoir for C. gattii (Ellis and Pfeiffer, 1990). Dispersal of propagules by wind and mechanical transport by insects and birds has also been suggested (Kidd et al., 2003; K i d d et al., 2007a; Sorrell, 2001). Recently, K i d d et al. (2007b) have found evidence suggesting C. gattii dispersal via water and anthropogenic transport v ia vehicle and foot travel. 2.2.1 Cryptococcus gattii in the Tropics and Subtropics El l i s and Pfeiffer (1990) first discovered C. gattii in the natural environment from decaying wood, bark and leafy Eucalyptus camaldulensis debris in Australia. Since then, C. gattii has been largely associated with the red gum group of eucalyptus trees (E. camaldulensis, E. tereticornis, E. rudis, E. blakelyi and E. gomphocephala) in the tropics and subtropics (Ellis and Pfeiffer, 1990; Sorrell, 2001). These eucalypt species have been exported commercially and established outside o f Australia and Papua N e w Guinea in southeast As ia , China, the Indian subcontinent, Africa, southern Europe, the United States (especially in California and Hawaii), Central America and South America (Figure 2; Casadevall and Perfect, 1998; 8 Penfold and Wi l l i s , 1961 and Zacharin, 1978 in El l i s and Pfeiffer, 1990; Harden, 2003; Sorrell, 2001). A small number of cold-tolerant eucalyptus species are able to grow in southeastern Vancouver Island and the Gul f Islands, Canada but not the red gum eucalypts (University of British Columbia Botanical Garden and Centre for Plant Research, 2004). Figure 2. Worldwide distribution of exported eucalyptus trees. Data from Casadevall and Perfect, 1998; El l i s and Pfeiffer, 1990; Harden, 2003; Sorrell, 2001. The ability of C. gattii to occupy non-eucalypt ecological niches was recently validated when the fungus was isolated from decaying wood of Moquilea tomentosa, Erythrina jambolanum and Guettarda acreana in Brazi l and Terminalia catappa in Columbia (Callejas et al., 1998; Fortes et al., 2001; Lazera et al., 2000). The unexpected and dramatic isolation o f C. gattii from a variety o f native tree species on Vancouver Island such as Douglas-fir (Pseudotsuga menziesii), red alder (Alnus rubra) and Garry oak (Quercus garryana), indicates that C. gattii is able to establish ecological niches in a broader array of tree species and environments than previously described (Kidd et al., 2004; K i d d et a l , 2007b). 9 The worldwide distribution of reported cryptococcosis due to C. gattii and isolation of C. gattii in the environment are described in Appendix A . Appendix A also includes an extensive review of the literature that details the molecular typing of C. gattii isolates (molecular data provided by S. K i d d , unpublished data). 2.2.2 Cryptococcus sattii in Southwestern British Columbia The environmental search for C. gattii on Vancouver Island was prompted by unusual reports of animal and human cases of cryptococcosis to public and veterinary health in September 2000. Woody debris and soil from case properties and nearby parks were initially sampled during the pilot study (Kidd et al., 2007b). A small number of eucalypts on Vancouver Island were also sampled without successful isolation o f C. gattii. Finally, after 8 months of environmental sampling, C. gattii was isolated from a mature Douglas-fir in Rathtrevor Beach Provincial Park (N49°19 ' W124°16 ' ) . Subsequent sampling by K i d d et al. (2004; 2007b) on Vancouver Island has identified C. gattii from the bark of Douglas-fir (Pseudotsuga menziesii), western red cedar (Thuja plicata), red alder (Alnus rubra), maple (Acer sp.), arbutus (Arbutus menziesii), Garry oak (Quercus garryand), pine (Pinus sp.), fir (Abies sp.), spruce (Picea sp.), dogwood (Cornus sp.), and bitter cherry (Prunus emarginata). C. gattii has also been isolated from soil, air and water samples. C. gattii has subsequently been isolated from the environment beyond Vancouver Island on Saltspring Island (N48°45 ' W123°29 ' ) and Whatcom County, Washington State (Kidd et al., 10 2007b). Positive air samples from Langley and Chil l iwack have also identified C. gattii in the Vancouver Lower Mainland but follow-up environmental sampling could not isolate it from woody debris, soil or water in these areas. K i d d et al. (2007b) suspects that C. gattii may have been recently introduced to the B C mainland via imported woody material from endemic areas on Vancouver Island. The successful colonization of C. gattii in a new ecological niche on the B C mainland remains to be seen. To date, all environmental and clinical isolates from Vancouver Island, the B C mainland and Whatcom County are C. gattii serotype B (Kidd et al., 2004; MacDougall et a l , 2007). Subtyping of these isolates found that the majority (-93%) are V G I I a and VGIIb (Kidd et al., 2004); V G I is found less often. 2.2.3 Physical and Biotic Characteristics of Cryptococcus sattii in the Environment To date there are no reports available, from B C or elsewhere, that describe the specific physical (temperature, humidity, sunlight, pH) and biotic characteristics necessary for C. gattii survival in the environment. General characterizations from previous studies in Australia (Ellis and Pfeiffer, 1990) and Columbia (Granados and Castaneda, 2005) are available, however. El l i s and Pfeiffer (1990) suggested the sudden appearance and dispersal of C. gattii coincided with the flowering of E. camaldulensis in Australia during late spring, while Granados and Castaneda (2005) reported an increase in the rate of positive C. gattii samples during the wet and humid months of A p r i l and M a y in Columbia. Granados and Castaneda (2006) also identified an association between rainy climatic conditions and 11 increased occurrence of C. gattii serotype B (the same serotype found in B C ) in trees in the temperate climate of Bogota: high precipitation and humidity, low sunshine hours and less extreme temperatures during the rainy months. In B C , K i d d et al. (2007b) did not observe a clear seasonal pattern for C. gattii positivity. A i r samples taken during the summer months, which are associated with warm, dry conditions and low relative humidity, appeared to be more likely positive though. K i d d et al. (2007b) also observed C. gattii from soils within the p H range of 4.33 - 7.48, and its presence was associated with lower moisture content and lower organic carbon content. Furthermore, C. gattii was isolated from both fresh water and seawater. Survival assays in the laboratory revealed a high survival rate of C. gattii in seawater at 4°C - broadly representing the natural seawater conditions of B C . Finally, C. gattii was also isolated from inanimate, non-organic objects such as vehicle tires, wheel wells and shoe bottoms which had traveled through the endemic C. gattii areas on Vancouver Island during the recent and not so recent past (Kidd et al., 2007b). 2.3 Cryptococcosis C. gattii can cause a rare but severe pulmonary and central nervous system disease in immunocompetent humans and animals. The infection may cause mass lesions (cryptococcomas) in the lungs (Figure 3) and brain that may manifest as a variety of symptoms including prolonged cough (lasting weeks or months), sharp chest pain, 12 unexplained shortness of breath, severe headache, fever, night sweats and weight loss ( B C C D C , 2005). Exposure to C. gattii is almost exclusively by inhalation of propagules, although direct inoculation into damaged skin has been reported (Hamann et al., 1997). It is believed that the majority of people who are exposed to C. gattii do not experience any adverse health effects as their immune systems are able to ward off infection. Risk factors for cryptococcosis due to C. gattii include genetic susceptibility and increased environmental exposure based on studies from northern Australia (Chen et al., 2000). Underlying lung conditions and oral steroid use may also increase the risk of disease acquisition ( M . Fyfe, unpublished data). Neither person to person, nor animal to animal transmission has been recorded in the literature. The geographic and temporal clustering of cryptococcal infections on Vancouver Island is the first and only one ever recorded throughout the world. Figure 3. Chest radiograph (A) and computed tomography scan (B) of a patient showing three cryptococcomas in the lungs. Source: Modified figure from Lindberg et al., 2007. © Lindberg et al., 2007, by permission. A ' I B 13 2.3.1 Cryptococcosis in B C A retrospective analysis of human cryptococcosis records suggests that the earliest evidence o f C. gattii in B C dates back to 1999 ( M . Fyfe, unpublished data). The traditional cases of cryptococcal disease in B C were due to C. n. var. neoformans infections which are typically associated with persons of compromised immune systems such as those with H I V / A I D S , tuberculosis or pulmonary disease. Prior to the identification o f C. gattii in B C , cryptococcosis was not a reportable disease in the province. Cryptococcosis due to C. gattii on Vancouver Island has plateaued at 36 cases per mi l l ion population per year during 2002-2005 (MacDougall et al., 2007). Between January 1999 and November 2004 all reported human cases of C. gattii in B C were among those l iving on or traveling to Vancouver Island during the year before symptoms appeared (MacDougall et al., 2007). However, in December 2004, the first reported human case of C. gattii on the B C mainland without travel to Vancouver Island or other C. gattii endemic areas in the world was identified. C. gattii originating from Vancouver Island is believed to have been dispersed to the B C mainland via imported woody material or mechanical transport (Kidd et al., 2007a). Subsequently, a total of 4 human, 11 animal and 7 positive environmental samples (5 air and 2 soil samples) on the B C mainland have been identified ( B C C D C and K . Bartlett, unpublished data). To date over 170 human cases have been reported to public health in B C ; the average age is 59 years (range = 2-92 years) and 56% are male. 14 2.4 Laboratory Methods Used to Identify Cryptococcus gattii The presence o f C. gattii in the environment cannot be discerned by visual observation since the fungal organism is microscopic in size (i.e., invisible to the human eye), and C. gattii does not harm the health o f its environmental reservoir, trees. Instead, specialized laboratory methods are required to identify C. gattii. Cryptococcus is commonly cultured on Staib (niger seed) media where it is incubated at 30°C for up to 10 days (Staib et al., 1987; University of British Columbia Cryptococcus Research, 2006). Cryptococcus grows as dark brown colonies on Staib media (Figure 4). If Cryptococcus is identified, Canavanine-glycine-bromothymol blue ( C G B ) agar is then commonly used to differentiate C. gattii from C. neoformans. C. gattii w i l l turn C G B agar blue (Figure 5). PCR-based molecular typing can be performed to discriminate the genetic variation within the molecular types (Figure 6). A genetic databank, GenBank (National Center for Biotechnology Information, 2007), containing molecular fingerprints of C. gattii isolates from around the world is available for comparison. Figure 4. Dark brown Cryptococcus colonies on Staib media. Source: University of British Columbia Cryptococcus Research, 2006. © University of British Columbia School of Occupational and Environmental Hygiene, 2006, by permission. 15 Figure 5. CGB agar turns blue (right) for C. gattii. Source: Mycology Online, 2006. © University of Adelaide, 2006, by permission. Figure 6. Restriction fragment length polymorphism (RFLP) profiles for selected human, animal and environmental C. gattii isolates from British Columbia, Washington and Oregon. The majority of isolates in BC are VGIIa and they share an identical genetic profile. VGIIb and VGI molecular types are also present. C. gattii control strains from other regions of the world are used for comparison of genetic similarity or dissimilarity. Source: Modified figure from MacDougall et al., 2007. © MacDougall et al., 2007, by permission. A l l . l l l . l . - 3 s fi a a P 5 o • aSSnaSS £ 3 3 555fS5f f3S$$i i ! 16 3.0 ECOLOGICAL NICHE MODELING This study used ecological niche modeling ( E N M ) to identify areas within B C that have suitable environmental conditions necessary for hosting C. gattii in the environment. The concept of the ecological niche, issues related to model selection and scale of analysis, methodology of the Genetic Algorithm for Rule-set Prediction, applications of E N M described in the literature, and premise of landscape epidemiology to identify C. gattii risk to humans and animals, are described in this section. Biogeographers and ecologists have long sought to explain the geographic distribution of plants and animals on the planet (Guisan and Thuiller, 2005). Most of the early researchers used the association between distributions of species, environment and climate to explain and predict the distribution of species based on empirical data. Over the past two decades, a number of sophisticated species distribution, or ecological niche, models have been developed to forecast the geographic distribution of plant and animal species based on predictor environmental data variables (Table 1). E N M methods that have been employed include artificial neural networks, boosted decision trees, climatic envelop, genetic algorithm, maximum entropy, multivariate distance, regression and generalized linear models (Elith et al., 2006; Guisan and Thuiller, 2005). Each model is, however, still fundamentally based on the species-environment relationship since suitable environmental conditions are required to host species habitat and populations. 17 Table 1. Published predictive ecological niche models. Source: Guisan and Thuiller, 2005. © Blackwell Publishing, 2005, by permission. Tool Reference Methods implemented URL BIOCLIM Busby (1991) CE http://www.arcscripts.esri.com ANUCLIM Sec BIOCLIM CE http://www.cres.anu.cdu.au/outputs/anuclim.php BAYES Aspinall (1992) BA ArcVicw extension available at the discretion of the author BIOMAPPER Hired et al. (2002) EN FA http://www.unil.ch/biomappcr BIOMOD Thuiller (2003) G L M , GAM, CART, ANN At the discretion of the author DIVA Hijmans et al. (2001) CE http:// www.diva-gis.org DOMAIN Carpenter et al. (1993) CE http://www.cifor.cgiar.org/dixs/_rcf/ rescarch_tools/domain/indcx.htm ECOSPAT Unpublished data G L M , GAM http://www.ccospat.unil.ch; at the discretion of the author CARP Stockwell & Peters (1999) GA (incl. CE, G L M , ANN) http://www.lifcmappcr.org/dcsktopgarp GDM Ferrier et al. (2002) . GDM Al the discretion of the author GRASP Lehmann et al. (2002) G L M , G A M http://www.cscf.ch/grasp MAXENT Phillips el al. (2005) ME At the discretion of the author SPECIES Pearson el al. (2002) ANN At the discretion of the author Coupled with ccllu lar automata Disperse Carey (1996) CE At the discretion of the author Shift Iverson et al. (1999) CART At the discretion of the author ANN, artificial neural networks; BA, Baycsian approach; CE, climatic envelop; CART, classification and regression trees; ENFA, ecological niche factor analysis; GA, genetic algorithm; G A M , generalized additive models; G D M , generalized dissimilarity modelling; GLM, generalized linear models; ME, maximum entropy. A common challenge in determining the geographic distribution and range of plant and animal species is the lack of data, especially in remote areas where it is difficult for humans to sample and report species observations. Extensive sampling is typically required, but in most cases field sampling is prohibitively expensive. Absence data are even scarcer since it is difficult to determine that a species does not exist in an area (Guisan and Thuiller, 2005). Furthermore, Stockwell and Peters (1999) point out that existing species data may not be readily available to researchers because the data have typically been stored in separate museums and among separate state and national jurisdictions. A s a result, academics, environmental groups and governments work largely with incomplete data (i.e., patchy geographical coverage), especially as it pertains to larger areas spanning multiple states or nations. 18 However, when species observation data are made available E N M can provide accurate forecasts of a species' ecological niche - even in areas where there are no data available -based on the environmental characteristics of known species occurrence locations. Recent and exciting advancements in E N M include the identification of suitable habitat for invasive species in new environments and projection of climate change impacts on ecological niches of plant and animal species. The availability of finer scale environmental data such as elevation, temperature and precipitation layers in GIS formats enable more detailed analyses at the regional and local scales. Furthermore, the application of E N M has been employed outside of its traditional role in biodiversity and conservation research. Peterson (2006), for example, recently encouraged the public health community to use E N M to forecast the distribution of vectorborne and environmental diseases based on the distribution of vectors and environmental reservoirs. 3.1 The Ecological Niche The concept of the ecological niche was proposed by Grinnell in 1917. He described the ecological niche of a species as the set of physical and biological conditions under which the species can maintain its population without immigration. For a species to maintain its population, its individuals must be able to survive and reproduce in the physical environment, obtaining energy and nutrients while avoiding predation (Pidwirny, 2006). 19 Hutchinson (1959) further defined the ecological niche concept by making distinctions between the fundamental and realized niches. The fundamental niche consists of the total potential area that meets all the physical and biological requirements of a species; whereas, the actual distribution of a species is determined by a variety of factors such as dispersal, history and physical barriers - an area described as the realized niche (Figure 7). This distinction is particularly important in the E N M of C. gattii since the present observed distribution of C. gattii in B C (i.e., realized niche) may not reflect its full potential geographic range in B C (i.e., fundamental niche) due to anthropogenic factors such as the transportation of woody material from endemic areas on Vancouver Island to other areas of the province. Figure 7. The fundamental and realized niches. Source: Pidwirny, 2006. © Pidwirny, 2006, by permission. Fundamental Niche Moisture Realized Niche Temperature 20 3.2 Selection of Model and Scale of Analysis A number of sophisticated species distribution models have been developed and are available to researchers. The criteria for selecting the most appropriate model can include model methodology, accuracy, performance, cost, hardware and software requirements, format and availability of reference data, availability o f documentation and user support, and credibility among the user community. The Genetic Algorithm for Rule-set Prediction was selected for this study because it has a proven record of accuracy and performance for predicting the ecological niches o f a variety of species (Peterson, 2001; Soberon and Peterson, 2005; Stockwell and Peters, 1999; Stockwell and Peterson, 2002), the specialized software is freely available (University of Kansas Center for Research, 2002) and is relatively easy to operate on standard desktop computers, software documentation and user support exists (University of Kansas Center for Research, 2002), and datasets in raster GIS format can easily be imported in and exported out o f the software program. Scale is a critical issue in E N M . The scale at which the modeling is performed should match the biological characteristics o f the species in question. Modeling and analysis at finer spatial resolution are more appropriate for fixed or less mobile organisms, and usually provides better predictions (Guisan and Thuiller, 2005). Often though, the scale of analysis is dictated by the spatial resolution o f the datasets available. In particular, the spatial resolution of climate data is typically coarse (4-18.5 km; Hijmans et al., 2005). Working with finer 21 resolution data has an increased "cost" in terms of increased computer processing time, hardware requirements and disk space though. 3.3 G A R P Methodology The Genetic Algorithm for Rule-set Prediction ( G A R P ) model was developed by Stockwell and Peters (1999). G A R P is an iterative, artificial intelligence based approach to E N M (Peterson, 2001). It employs a "superset" of rules to identify the ecological niche of a species based on the environmental characteristics of known occurrence locations, to search for non-random correlations between species presence, species absence and environmental parameter values (University of Kansas Center for Research, 2002). G A R P divides the species occurrence data (model input points) into training and testing datasets, and then environmental data layers relevant to the ecology of the species in question are added to construct the model. Typical environmental datasets used for E N M include topographic (elevation, aspect, slope), climatic (temperature and precipitation) and biologic (vegetation, other species distributions) data layers. Similar to the additions, deletions and mutations of genes, the rules for predicting a species' ecological niche are randomly developed and progressively employed on the training dataset. These sets of rules are essentially a collection of "if-then" relationships which are iteratively applied (Stockwell and Peters, 1999). A t each iteration of G A R P ' s processing, the predictive accuracy of the model under the employment of that rule is evaluated based on the testing dataset (Peterson et al., 2002b). The rule is accepted and incorporated into the model i f the 22 change in predictive accuracy increases from one iteration to the next; otherwise, the rule is rejected and dropped. A n one-tailed x2 test on the difference between the probability of the predicted value before and after the employment of the rule is used to determine the statistical significance of these rules (Stockwell and Peters, 1999). The model runs 1,000 iterations or until convergence is reached. 3.3.1 Rule Types G A R P employs four types of rules to identify the ecological niche of a species: atomic, range, negated range and logistic regression (Payne and Stockwell, 1996). Atomic rules are the simplest form of rule - it uses only a single value of the variable(s) in the precondition of the rule. A n example atomic rule takes the form of: i f the biogeoclimatic zone is C D F and elevation is 100 m, then C. gattii is present. Range rules use a range of values instead of a single value to determine which predictor variables are relevant for the rule. A n example range rule takes the form of: i f average January temperature is greater than 0°C and less than 10°C, then C. gattii is present. Conversely, negated range rules can be applied to values outside of the indicated range to determine which predictor variables are irrelevant for the rule. Lastly, logistic regression rules transform the logistic regression equation output into a probability to determine whether the rule should be applied. If the output probability is greater than 0.75 (default value in G A R P ) then the logistic regression is evaluated for use in the model. A n example logistic regression rule takes the form of: i f the output probability of 0.1 - (elevation * 0.2) + (average January temperature * 0.5) is > 0.75, then C. gattii is present. 23 3.3.2 Predictive Accuracy Predictive accuracy of the model is determined by the number of correct predictions and errors o f commission and omission (University of Kansas Center for Research, 2002). A commission error occurs when the model predicts a species to occur where it does not (i.e., including areas not actually inhabited); whereas, an omission error occurs when the model fails to predict a species occurrence where it does in fact occur (i.e., leaving out the true distributional area). Accuracy is calculated using the expression: (a + b) / (a + b + c + d) where a is the number of points where the model predicted presence and the input point was a presence record (i.e., the model has predicted the point successfully), b is the number of points where the model predicted absence and the input point was an absence record (i.e., the model has predicted the point successfully), c is the number of points where the model predicted absence but the input point was a presence record (i.e., an omission error or "false negative") and d is the number of points where the model predicted presence but the input point was an absence record (i.e., a commission error or "false positive"). Training accuracy is calculated using the training data points and testing accuracy is calculated using the testing data points. 24 3.4 Applications of GARP Described in the Literature G A R P has been used extensively to predict the ecological niches of a variety of plant and animal species for conservation biology and applied ecology purposes. Recent software developments have enabled G A R P to be used for invasive species and climate change modeling. For example, Peterson et al. (2003) describe the use of G A R P to identify suitable areas in a new environment for an invasive species based on the ecological characteristics of known occurrences in the species' native distribution. These forecasted areas represent where the invasive species are likely to be present but are not yet discovered, or where they may become established after dispersal (Ron, 2005). With respect to climate change modeling, examples of the use of G A R P to predict the new geographic range of species habitats due to climate change include: sand flies (Phlebotomus sp.; Peterson and Shaw, 2003), monarch butterflies (Danaus plexippus; Oberhauser and Peterson, 2003), birds (Ortalis poliocephala; Peterson et al., 2002a) and salamanders (Pseudoeurycea sp.; Parra-Olea et al., 2005). With respect to infectious diseases, G A R P has been used to infer the geographic distribution of leishmaniasis (Peterson and Shaw, 2003; Peterson et al., 2004b), Chagas disease (Peterson et al., 2002b) and filovirus disease (Peterson et al., 2004a) based on the geography of vector species. 3.5 Landscape Epidemiology E N M was used in this study to forecast the ecological niche of C. gattii to identify the "landscape" of human and animal cryptococcocal disease in B C . Landscape epidemiology, 25 which provides the basis for using E N M of C gattii to identify geographic areas of risk to C . gattii exposure, is an emerging and evolving field o f research which explores the relationship between the ecology and epidemiology of infectious diseases to identify geographical areas where disease transmission occurs (Peterson, 2006). The theory behind landscape epidemiology is that by knowing the environmental conditions (e.g. climate, geology, vegetation) necessary for the maintenance of specific pathogens in nature, one can use the landscape to identify the spatial and temporal distribution of disease risk (National Aeronautics and Space Administration, 2006). Key environmental factors such as elevation, temperature, rainfall and humidity, influence the presence, development, activity and longevity of pathogens, vectors, zoonotic reservoirs of infection, and their interactions with humans (Meade et al., 1988). Landscape epidemiology, by the very nature of the interaction of hosts and the pathogens within the environment, lends itself to the use of GIS, spatial analysis and E N M as a way to summarize the complex relationships associated with disease transmission (Glass, 2001). 26 4.0 D A T A A N D M E T H O D S The software and hardware, C. gattii disease surveillance data, environmental data layers and methods used in the ecological niche modeling of C. gattii are described in this section. 4.1 Software and Hardware ArcGIS 9.1 and A r c V i e w 3.2 (Environmental Systems Research Institute, Redlands, C A ) were used to perform the GIS mapping. The Geostatistical and Spatial Analyst extensions were used to interpolate continuous surface layers and manipulate raster datasets for use in Desktop G A R P 1.1.6 (University of Kansas Biodiversity Research Center, Lawrence, K S ) . ArcToolbox and ArcCatalog were used to process and convert GIS datasets between shapefile, grid and A S C I I formats. Desktop G A R P was used to model the ecological niche of C. gattii. Excel 2003 and Access 2003 (Microsoft Corporation, Redmond, W A ) were used to manipulate climatic data and summarize results. These software applications were run on a Pentium 4 (Intel Corporation, Santa Clara, C A ) based computer, operating at 2.0 G H z with 512 M B of R A M and 60 G B of hard disk space on the Windows X P operating system Service Pack 2 (Microsoft Corporation, Redmond, W A ) . 27 4.2 Disease Surveillance Data Human and animal surveillance and environmental sampling data were used as input data points to build and test the ecological niche models. The geographic locations of reported cryptococcal disease cases were geocoded by street address or postal code of residence using ArcGIS . Street address matching was against the CanMap Streetfiles v7.0 (dmtiSpatial, Markham, ON) and postal code geocoding was against the Canada Post Postal Code Conversion File - September 2005 (Statistics Canada, Ottawa, ON) . Environmental sampling locations (latitude and longitude coordinates) were recorded with a recreational grade Global Positioning System receiver (eTrex, Garmin, Olathe, K S ) and mapped with ArcGIS . 4.2.1 Human Cases Human cases were reported to public health by family physicians, microbiologists and respirologists. Shortly after the initial identification of cryptococcosis due to C. gattii in B C , the B C C D C developed an enhanced disease surveillance program to identify new cases of cryptococcosis. A formal case definition, detailed questionnaire and standardized diagnostic laboratory tests were created. Public health officials interviewed each case to collect demographic information, address of residence, travel history, medical history and environmental exposure. This information was recorded and managed in an Access database. In 2003, cryptococcosis due to C. gattii became a reportable disease in B C . The use of human case data in this study was reviewed and approved by the University of British Columbia Behavioral Research Ethics Board (Appendix B) . 28 4.2.2 Animal Cases Animal cases were detected and reported by veterinarians at the Central Veterinary Laboratory (Langley, B C ) , Animal Health Centre (Abbotsford, B C ) and Centre for Coastal Health (Nanaimo, B C ) . In particular, Duncan (2005) performed an extensive review of medical records from various veterinary clinics on Vancouver Island and Vancouver Lower Mainland, and developed an animal cryptococcosis database. Subsequent animal cases have been recorded and managed by K . Bartlett and S. Lester (unpublished data). Standard demographic information, address of residence, travel history, clinical diagnosis and other relevant data variables were collected where available. The animal owner's consent was required to have information released to public health. 4.2.3 Environmental Sampling Sampling for C. gattii in the environment included swab, soil, air and water sampling. Sterile medical transport swabs were brushed against the bark surface and crevasses of trees and other objects. Soi l samples were collected from the top 5 cm o f the ground and placed in sealable plastic bags. A i r sampling was conducted with two types of air samplers: R C S Plus (Biotest, Dreieich, Germany) and Andersen six-stage (Graseby Andersen, Atlanta, G A ) . Water samples (approximately 500 ml) were collected in clean screw top jars or strong sealable plastic bags (double bagged). In the laboratory, the contents were transferred to and cultured on Staib media. The concentration of C. gattii in soil, air and water were quantified in colony forming units (CFU) per gram, cubic meter and 100 milliliters, respectively. The 29 environmental sampling and laboratory methods for identifying C. gattii are described in more detail in K i d d et al. (2007b) and University of British Columbia Cryptococcus Research (2006). Environmental sampling was conducted in areas with and without an associated suspicion of C. gattii colonization, although most samples were taken in communities with human or animal cases, including properties where the cases resided and along forested walking trails and recreational parks (Kidd et al., 2007). Repeat sampling in select locations was performed to determine the degree of C. gattii colonization in an area, and whether a seasonal pattern of C. gattii presence in the environment existed. Field sampling information such as the sample identification number, date and location of collection, and species of tree or type of object sampled were recorded and managed in an Access database. For each environmental sampling location, a longitudinal review of the sampling results was performed to determine whether a location was "permanently" positive or "transiently" positive for C. gattii. 4.3 Environmenta l Data Layers Fifteen (15) environmental data layers believed to be relevant to C. gattii biogeography were used in the E N M : elevation, aspect, slope, biogeoclimatic zone, January average, maximum and minimum temperature, July average, maximum and minimum temperature, annual, January and July total precipitation, and soil drainage and development. These data themes have been routinely used to model the ecological niche of a variety of species including plants, animals, insects and fungi. Each environmental data layer had complete geographic 30 coverage across the province. The data layers were overlaid and clipped so that only the cells that were present in all layers were used in the E N M . Based on the spatial resolution of the digital elevation model used, the approximate cell size of the raster data layers was 600 m x 600 m. A total of 2,628,833 cells equating to 946,380 k m 2 spatial coverage were used in the E N M of C. gattii in B C . 4.3.1 Topographic GTOPO30 , a 30 arc second (approximately 1 km spatial resolution at the equator and 600 m in B C due to the convergence o f latitude and longitude towards the poles) digital elevation model from the United States Geological Survey (1996) was processed to produce aspect and slope raster layers. ArcGIS Spatial Analyst was used to calculate the slope and aspect from the digital elevation model. The aspect layer was reclassified to yield an aspect direction layer that summarized the raster values into an eight directional cardinal compass (Figure 8.). Figure 8. Data reclassification of aspect to yield a simplified aspect direction raster layer. For example, an original aspect value of 10° was reclassified to an aspect direction of " N " . 337.5° 22.5° 202.5° i v 157.5° 31 4.3.2 Climatic The January and July average, maximum and minimum temperature data layers were created by surface interpolation of climate station data from 1971-2000 (the "Normals"; Environment Canada, 2006; National Oceanic and Atmospheric Administration, 2006). Surface interpolation enables prediction of unmeasured values at any location based on the observed values (climate stations) of surrounding locations. Inverse distance weighted surface interpolation with elevation adjustments to account for the effect of lapse rate (decrease in air temperature of 6.5°C for every 1 k m gain in altitude) was used to model temperature. A r c G I S Geostatistical Analyst was used to perform the surface interpolation. Cokriging surface interpolation of the annual, January and July total precipitation data layers was also attempted but the measure o f error was unacceptably large using this method. Instead, the Parameter-Elevation Regressions on Independent Slopes Mode l ( P R I S M ; Daly et al., 2002) for precipitation was used. ArcGIS Spatial Analyst was used to re-sample the original 2.5 arc minute (approximately 4 k m spatial resolution) P R I S M datasets to 600 m cell size raster layers. A 7 x 7 cell low pass filter was then applied to smooth the 600 m cell size data values, thereby removing the abrupt changes in data values at the cell boundaries of the original P R I S M datasets. In Appendix C the climate station data used are presented, and in Appendix D the methodology for the surface interpolation modeling and re-sampling of the P R I S M datasets is described. 32 4.3.3 Biogeoclimatic Meidinger and Pojar (1991) have created an ecological classification system for B C that group ecologically similar environs based on vegetation, soils and climate. Fourteen (14) biogeoclimatic zones, which are generally named after the geographic region and dominant tree species of the area, have been defined for the province. Within each biogeoclimatic zone, subzones and variants may exist based on slight variations of climate and continentality. For example, the Coastal Western Hemlock biogeoclimatic zone has a xeric (very dry) maritime subzone with two variants (1 and 2). This detailed level of ecosystem classification was used for the E N M of C. gattii. The dataset was obtained from the B C Ministry of Forests (Biogeoclimatic Ecosystem Classification v6.0, Victoria, B C ) in shapefile format, and ArcGIS was used to convert it to raster format. 4.3.4 Soil The Soi l Landscapes of Canada v2.2 (Agriculture and Agri-Food Canada, Ottawa, ON) describe the major characteristics of soil and land such as surface and parent material, drainage, soil development and vegetation cover. ArcGIS was used to convert the A R C / I N F O coverages into two raster datasets: soil drainage and soil development. 33 4.4 G A R P Mode l ing of Cryptococcus gattii Desktop G A R P was used to perform the E N M of C. gattii. The 15 environmental data layers in E S R I grid format were converted to A S C I I and then R A W format using ArcToolbox and G A R P Dataset Manager. The C. gattii human and animal cases and environmental positive samples in shapefile format were used, in separate models, to train and test the respective models. G A R P is able to use a variety o f different data types in the E N M process because it employs a number of different rule types (atomic, range, negated range and logistic regression). The C. gattii, aspect direction, biogeoclimatic and soil datasets were nominal data. The elevation, slope, temperature, precipitation datasets were ratio data. 4.4.1 Cryptococcus gattii Data Used in the Ecological Niche Modeling Human and animal cases within the known endemic areas for C. gattii (i.e., along the south-central east coast of Vancouver Island from Campbell River in the north to Greater Victoria in the south, and the G u l f Islands) were used as the E N M input data points. Please note that these locations, mapped by address of residence, do not definitively represent the presence of C. gattii in the local area or where exposure to C. gattii occurred. Instead, these mapped locations were proxies for where exposure may have likely occurred since people, and especially companion animals, tend to spend most of their time and activities around their area o f residence (the social geography concepts of activity space and distance decay; Zipf, 34 1949; Carlstein and Thrift, 1978; Meentemeyer, 1989). Travel-related cases from the B C mainland and non-endemic areas on northern Vancouver Island were excluded from the E N M process. Therefore, 122 human and 135 animal cases from endemic areas on Vancouver Island and the G u l f Islands were used in the E N M of C. gattii (Figure 9). Three hundred and eighty four (384) positive environmental sampling locations from permanently established sites were also used as E N M input data points. These locations represent confirmed sites of C. gattii presence. Figure 9. C. gattii disease surveillance and environmental sampling data used in the ecological niche modeling. 0 10 20 K M 40 35 A small number of mainland cases (4 human and 11 animal cases, but only 8 of the 11 animal cases had map-able location information) without travel to Vancouver Island were identified on the B C mainland. Fifty five (55) transiently positive environmental sampling locations including 7 samples from 4 sites on the B C mainland were also identified. The locations of these mainland cases and positive environmental samples were used to verify the predictive accuracy of the C. gattii E N M since the empirical data from Vancouver Island were used to forecast the ecological niche of C. gattii on the B C mainland. 4.4.2 Environmental Data Layer Jackknifing Desktop G A R P provides functionality to identify which environmental factors are more significant or important than others for predicting the ecological niche o f the species in question. This is accomplished by "jackknifing" environmental data layers, which involves running multiple E N M tasks with only a subset of the layers (Stockwell and Peters, 1999). The jackknifing procedure w i l l determine which environmental data layers contribute positively to the model's performance; a good model has small commission and omission errors, and high training and testing accuracy. The settings and properties for the jackknifing of environmental data layers for the E N M of C. gattii based on the distribution of human cases are presented in Figure 10. The human cases were divided into training (70%) and testing (30%) sub-datasets. A l l 15 environmental data layers were jackknifed individually through 30 model runs, for a total of 450 model runs Each model run consisted of a maximum of 1000 iterations or until the convergence limit of 36 0.01 was reached; the term convergence in computer science and mathematics denotes the approach towards a definite value. Figure 10. Desktop G A R P settings and properties for the jackknifing of environmental data layers for the E N M of C. gattii based on the distribution of human cases. The human cases dataset (122 cases) was divided into training (70% of cases) and testing (remaining 30% of cases) sub-datasets. A l l 15 environmental data layers were jackknifed individually through 30 model runs. Each model run consisted of a maximum of 1000 iterations or until the convergence limit of 0.01 was reached. A l l rule types were employed: atomic, range, negated range, logistic regression. Desktop Garp - humanl 22_notravel_vionly_jacknife._xl File Datasets Model Results Help Species Data Points Species List: (1 selected) Hi Upload Data Points human notravel vionly (122) Options: U s e l ? " % for training r At least (20 training points Optimization Parameters (30 Runs 01 Convergence limit |1000 Max iterations Rule types: Atomic fv Range fv Negated Range Logistic Regression (Logit) f " All combinations of the selected rules (1 rule comb.) (30 total runs) Best Subset Selection Parameters r Active sion measure: Extrinsic C Intrinsic Omi sion threashold: <• r |10 * omission Total models under hard omission threshold Max models per spp. Commission threshold: n % of distribution Projection Layers Available datasets Current datasets for projection: (besides the training dataset) Environmental Layers Dataset: |bcenmcgattii Layers to be used: * bcelevation • bcaspectdir C* bcslope • bcbgczones6 *• bcjanavetemp v bcjanmaxtemp >/ bcjanmintemp </ bcjulavetemp v bcjulmaxtemp • bcjulmintemp Background sample: |—Random-How layer will be used: r All selected layers C All combinations of the selected layers « All combinations of size pj (13 comb.) Output Maps as: P Bitmaps T ASCII Grids T ARC/INFO Grids Output directory: v 3 Models: <? All models C Bestsubse' C:\Temp J 37 Environmental data layer jackknifing was also performed for the E N M of C. gattii based on the distribution of animal cases and positive environmental sampling locations from permanently established sites (figures not shown). 4.4.3 G A R P Modeling with Significant Environmental Data Layers Only E N M of C. gattii was then performed using only the significant environmental data layers identified in the jackknifing procedure (results are described in Section 5). Model accuracy is optimized when only the significant environmental data layers are used in the E N M process. The input C. gattii datasets were divided into training (70%) and testing (30%) sub-datasets, and 20 model runs were performed (Figure 11). A separate grid data layer was created for each model run. The model output grid layers simply reported cell values of 1 to indicate locations with appropriate ecological conditions to support C. gattii or 0 to indicate unsuitable locations. ArcGIS Spatial Analyst was used to summate the cell values of all 20 model runs. Since the threshold value for determining what constitutes a significant environmental data layers is not universally defined, several scenarios o f the G A R P modeling procedure were performed: environmental data layers with training and testing accuracy >95%, >90%, >80% and all environmental data layers. 38 Figure 11 . Desktop G A R P settings and properties for the E N M of C. gattii based on the distribution of human cases and environmental data layers with training and testing accuracy >95%. The elevation, biogeoclimatic and January average, maximum and minimum temperature data layers were used as the predictor variables to predict the ecological niche of C. gattii in B C based on the distribution of human cases as the model input dataset. Desktop Garp - human_notravel_vionly_enm.gxl File Datasets Model Results Help Species Data Points Species List: (1 selected) Upload Data Points human notravel_vionly (122) Optimization Parameters \2Q Runs [01 Convergence limit |l 000 Max iterations Rule types: Atomic Iv Range I* Negated Range I* Logistic Regression (Logit) [~ All combinations of the selected rules (1 rule comb.) (20 total tuns) Projection Layers Available datasets: Current datasets for projection: (besides the training dataset) Options: Use \7Q % for training r At least [20 training points Best Subset Selection Parameters r~ Active Omission measure: Extrinsic C Intrinsic On- <? Hard C Soft Total models under hard omission threshold commission threshold: I % of distribution w Environmental Layers Dataset: |bcenmcgattii Layers to be used: <* bcelevation bcaspectdir bcslope • bcbgczones6 * bcjanavetemp • bcjanmaxtemp • bcjanmintemp bcjulavetemp bcjulmaxtemp bcjulmintemp Background sample: |—Random— w \ How layer will be used: [• All selected layers C All combinations of the selected layers <~ All combinations of size |~~ (i comb.) Output Maps as: r Bitmaps r ASCII Grids p ARC/INFO Grids Output directory: Models: All mt C Best'. C:\Temp 39 5.0 RESULTS AND DISCUSSION A series of C. gattii ecological niche models were created based on the different data inputs (i.e., human cases, animal cases and positive environmental sampling locations) and significant environmental data layers (i.e., those with training and testing accuracy >95%, >90%, and >80%, and all environmental data layers) used to build the models. The results o f the E N M , including the interpretation of the C. gattii prediction maps, are discussed here. 5.1 Environmental Data Layer Jackknifing The average commission errors, omission errors, training accuracy, testing accuracy and p-value results based on the 30 model runs per environmental data layer provided quantitative measures for identifying which environmental data layers were important for accurately predicting the ecological niche o f C. gattii in B C (Table 2 for the human cases scenario; Table 3 for the animal cases scenario; Table 4 for the environmental positives scenario). Table 2. Summary of environmental data layer jackknifing for C. gattii E N M based on the distribution of human cases. Training and testing accuracy are expressed as percentages (%). Environmental Layer Commission Errors Omission Errors Training Accuracy Testing Accuracy P-value Elevation 4.5 0.4 97.8 97.6 <0.0001 Aspect Direction 27.5 72.1 43.2 42.5 0.0524 Slope 35.9 9.9 73.6 73.3 <0.0001 Biogeoclimatic Zone 2.4 3.6 97.4 96.8 O.0001 January Average Temp. 4.1 0.6 98.0 97.6 <0.0001 40 Environmental Layer Commission Errors Omission Errors Training Accuracy Testing Accuracy P-value January Maximum Temp. 4.0 0.6 98.0 97.7 O.0001 January Minimum Temp. 4.8 0.4 97.5 97.4 O.0001 July Average Temp. 13.2 12.6 88.1 86.6 <0.0001 July Maximum Temp. 25.5 27.5 68.7 66.9 O.0001 July Minimum Temp. 6.9 4.5 94.8 94.2 O.0001 Annual Total Precipitation 26.3 15.2 78.0 75.8 O.0001 January Total Precipitation 23.7 4.0 85.7 85.1 <0.0001 July Total Precipitation 6.2 5.8 91.8 91.8 <0.0001 Soil Drainage 14.0 61.4 57.4 56.6 O.0001 Soil Development 21.3 44.5 78.0 78.2 O.0001 Table 3. Summary of environmental data layer jackknifing for C. gattii E N M based on the distribution of animal cases. Training and testing accuracy are expressed as percentages (%). Environmental Layer Commission Errors Omission Errors Training Accuracy Testing Accuracy P-value Elevation 4.6 1.1 97.6 97.0 <0.0001 Aspect Direction 12.7 95.9 35.1 33.8 0.0857 Slope 36.7 14.8 71.3 71.4 O.0001 Biogeoclimatic Zone 1.4 1.3 98.7 98.8 <0.0001 January Average Temp. 4.0 0.8 98.0 97.5 <0.0001 January Maximum Temp. 3.8 1.3 98.0 97.4 <0.0001 January Minimum Temp. 4.5 0.3 97.7 97.5 O.0001 July Average Temp. 15.5 0.2 92.3 92.1 O.0001 July Maximum Temp. 29.0 19.7 72.3 71.6 <0.0001 July Minimum Temp. 6.0 3.4 95.9 95.2 <0.0001 Annual Total Precipitation 19.7 22.1 76.3 74.1 <0.0001 January Total Precipitation 19.1 5.8 88.0 86.7 <0.0001 July Total Precipitation 3.3 3.3 96.5 96.0 <0.0001 41 Environmental Layer Commission Errors Omission Errors Training Accuracy Testing Accuracy P-value Soil Drainage 10.5 84.2 51.9 51.7 <0.0001 Soil Development 19.9 36.7 80.8 81.1 <0.0001 Table 4. Summary of environmental data layer jackknifing for C. gattii E N M based on the distribution of positive environmental sampling locations from permanently established sites. Training and testing accuracy are expressed as percentages (%). Environmental Layer Commission Errors Omission Errors Training Accuracy Testing Accuracy P-value Elevation 4.6 0.3 97.7 97.3 O.0001 Aspect Direction 24.1 61.1 51.5 51.2 0.0016 Slope 37.7 8.6 74.8 74.5 <0.0001 Biogeoclimatic Zone •1.4 0.0 99.3 99.3 O.0001 January Average Temp. 4.3 0.0 97.8 • 97.7 <0.0001 January Maximum Temp. 4.2 0.1 97.9 97.7 <0.0001 January Minimum Temp. 4.8 0.2 97.6 97.4 <0.0001 July Average Temp. 10.1 6.7 87.6 87.9 <0.0001 July Maximum Temp. 15.9 12.0 81.2 81.0 O.0001 July Minimum Temp. 8.2 1.7 94.0 94.0 <0.0001 Annual Total Precipitation 31.0 5.0 79.4 78.8 <0.0001 January Total Precipitation 27.4 6.9 83.5 83.3 O.0001 July Total Precipitation 7.6 5.1 91.3 90.7 O.0001 Soil Drainage 13.0 86.9 44.9 42.8 0.0070 Soil Development 22.2 48.0 74.9 76.1 <0.0001 The elevation, biogeoclimatic and January average, maximum and minimum temperature data layers had training and testing accuracy >95% for the jackknifing of environmental data layers based on the distribution of human and animal cases and positive environmental 42 sampling locations from permanently established sites; July minimum temperature and July total precipitation also had training and testing accuracy >95% for the jackknifing o f environmental data layers based on the distribution of animal cases. These environmental data layers had small commission and omission errors as wel l , making them good predictor variables for modeling the ecological niche of C. gattii in B C . Conversely, aspect, slope, July maximum temperature, annual total precipitation, soil drainage and soil development were poor predictor variables for modeling the ecological niche of C. gattii in B C since their commission and omission errors were large, and training and testing accuracy were low. A full description of the various C. gattii E N M scenario results based on the distribution of human cases, animal cases and permanently positive environmental sampling locations using environmental data layers with >95% accuracy, >90% accuracy, >80% accuracy, and all environmental data layers follows in section 5.2. 5.2 Ecological Niche Model Predictions The E N M of C. gattii using environmental data layers with >95% training and testing accuracy produced the best models (Table 5). Training and testing accuracy were 99.4% and 99.2% for the C. gattii E N M based on the distribution of human cases, 98.7% and 98.3% for the C. gattii E N M based on the distribution of animal cases, and 99.7% and 99.7% for the C. gattii E N M based on the distribution of positive environmental sampling locations from permanently established sites. Commission errors (incorrectly predicting an area that was not actually occupied by C. gattii) and omission errors (incorrectly leaving out an area that was 43 actually occupied by C. gattii) were very small: only 1.2 commission errors and 0.4 commission errors per model run for the C. gattii E N M based on the distribution of human cases, only 1.2 commission errors and 0.5 commission errors per model run for the C. gattii E N M based on the distribution of animal cases, and only 0.7 commission errors and no commission errors per model run for the C. gattii E N M based on the distribution o f positive environmental sampling locations from permanently established sites. A s a rule of thumb, a good model has fewer than 5 commission and omission errors, and training and testing accuracy greater than 95% (Peterson et al., 2003). The probability o f these modeling results to have occurred by random chance was less than 1 in 10,000 (p-value <0.0001). Table 5. Summary of C. gattii E N M results for different data inputs and environmental data layers used to build the models. Training and testing accuracy are expressed as percentages (%). The number of environmental data layers used to build the models is also indicated. C. gattii ENM Scenario Commission Errors Omission Errors Training Accuracy Testing Accuracy P-value Human (>95% - 5 layers) 1.2 0.4 99.4 99.2 O.0001 Animal (>95% - 7 layers) 1.2 0.5 98.7 98.3 <0.0001 Env'tal (>95% - 5 layers) 0.7 0.0 99.7 99.7 <0.0001 Human (>90% - 7 layers) 1.7 0.0 98.2 98.2 O.0001 Animal (>90% - 8 layers) 1.2 0.1 88.5 88.3 <0.0001 Env'tal (>90% - 7 layers) I- 2 0.0 99.1 99.0 <0.0001 Human (>80% - 9 layers) 1.6 0.0 73.8 73.6 <0.0001 Animal (>80% - 10 layers) 1.3 0.4 83.4 83.0 O.0001 Env'tal (>80% -10 layers) 0.9 0.1 86.0 85.7 <0.0001 Human (Al l 15 layers) 1.1 0.1 52.2 52.2 <0.0001 Animal (Al l 15 layers) 0.9 0.6 82.9 82.4 O.OOOT Env'tal (Al l 15 layers) 0.9 0.2 71.4 71.0 <0.0001 44 E N M of C. gattii based on the distribution of human cases and positive environmental sampling locations from permanently established sites using environmental data layers with >90% training and testing accuracy (elevation, biogeoclimatic zones, January average, maximum and minimum temperature, July minimum temperature, and July total precipitation) also produced good model results (Table 5). However, the C. gattii E N M based on the distribution o f animal cases had a considerable decrease in training and testing accuracy (88.5% and 88.3%, respectively) - a model accuracy decrease of 10% when compared to the E N M scenario using environmental data layers with >95% training and testing accuracy. This may be due to the inclusion of the July average temperature data layer i n the C. gattii E N M scenario based on the distribution o f animal cases (i.e., July average temperature is not a good predictor variable for modeling the ecological niche of C. gattii in B C since it decreases the overall model accuracy). E N M o f C. gattii using environmental data layers with >80% training and testing accuracy, and using all environmental data layers produced less than optimal model results (Table 5). This highlights the importance in identifying the significant environmental data layers (predictor variables) for accurately predicting the ecological niche of C. gattii in B C . 5.2.1 Ecological Niche Model Prediction Maps Twenty (20) model output grid data layers (1 for each model run) reporting the presence or absence o f suitable ecological conditions to support C. gattii in B C were created for each E N M scenario. Due to the random and iterative nature of the G A R P algorithm, individual 45 model runs may vary considerably. Therefore, the 20 model output grid data layers for each E N M scenario were overlaid and the cell values (either 1 indicating suitable ecological conditions to support C. gattii, or 0 indicating unsuitable ecological conditions to support C. gattii) were summated to produce a composite, "best subset" C. gattii E N M data layer. This method was suggested by Peterson (2001) to ensure construction of robust ecological niche model results. "Optimal" C. gattii ecological niche areas had 11-20 model output grid data layer agreement and "potential" C. gattii ecological niche areas had 1-10 model output grid data layer agreement. C. gattii ecological niche prediction maps were created for each E N M scenario: human cases, animal cases and positive environmental sampling locations from permanently established sites using environmental data layers with >95% accuracy (Figure 12), >90% accuracy (Figure 13), >80% accuracy (Figure 14), and all environmental data layers (Figure 15) to build the models. Similarities among the C. gattii ecological niche prediction map scenarios - using different environmental data layers (based on training and testing accuracy determined by the jackknifing procedure) to build the models - were the presence of optimal ecological niches along the central and south eastern coast of Vancouver Island, G u l f Islands, Sunshine Coast and Vancouver Lower Mainland. C. gattii ecological niche predictions based on the environmental data layers with >95% accuracy produced conservative results. For example, a large area o f the B C interior was identified as potential C. gattii ecological niche in the E N M scenario based on the distribution of human cases and environmental data layers with >80% accuracy (Figure 14A); whereas, the E N M scenario based on the distribution of human cases 46 and environmental data layers with >95% accuracy does not identify any suitable C. gattii ecological niche areas in the B C interior (Figure 12A). The total forecasted ecological niche area for C. gattii was greatest for the models based on the environmental data layers with >80% accuracy (mean of 53,322 k m 2 based on the composite human, animal and environmental sampling models), followed by the models based on the environmental data layers with >90% accuracy (41,590 k m ), > 95% accuracy (23,222 km ), and all layers (20,999 km 2 ) . In general, inclusion of greater numbers of environmental data layers for E N M produced larger forecasted ecological niche areas. The exception is when al l environmental data layers were used to build the models, which resulted in the smallest forecasted C. gattii ecological niche area, since the inclusion of non-significant environmental datasets produced many rule rejections and less than optimal modeling results. When comparisons are made between the C. gattii ecological niche prediction map scenarios based on the distribution of human cases, animal cases and positive environmental sampling locations from permanently established sites model scenarios, the C. gattii ecological niche predictions based on the environmental sampling locations produced the most conservative results. For example, potential C. gattii ecological niche areas on the Queen Charlotte Islands and central B C coast are practically absent in the positive environmental sampling locations from permanently established sites model scenarios (Figure 12C, Figure 14C, Figure 15C); whereas, the C. gattii E N M scenarios based on the distribution of human and animal cases identify these potential ecological niche areas quite prominently (Figure 12AB, Figure 14AB, Figure 15AB). 47 Figure 12. C. gattii ecological niche prediction maps based on the distribution of human (A) and animal cases (B ) and positive environmental sampling locations from permanently established sites (C), and environmental data layers with >95% accuracy. Atlin Ft Nel. Forecasted ecological niche (based on environmental sampling) | Optimal | Potential 1 I Not suitable PR. P.O. Kami kehv Cran o • o Pt Hardy 48 Figure 13. C. gattii ecological niche prediction maps based on the distribution of human (A) and animal cases (B) and positive environmental sampling locations from permanently established sites (C), and environmental data layers with >90% accuracy. 49 Figure 14. C. gattii ecological niche prediction maps based on the distribution of human (A) and animal cases (B) and positive environmental sampling locations from permanently established sites (C), and environmental data layers with >80% accuracy. 50 Figure 15. C. gattii ecological niche prediction maps based on the distribution of human (A) and animal cases (B) and positive environmental sampling locations from permanently established sites (C), and all environmental data layers. Forecasted ecological niche (based on human case distribution) FtNel. 3 Forecasted ecological niche (based on animal case distribution) Atlin FtNel , , Forecasted ecological niche (based on environmental sampling) Atlin FtNel , , 51 5.3 Validation of the Ecological Niche Model Predictions The non-travel (i.e., locally acquired) human and animal C. gattii cases and positive environmental samples on the B C mainland were used to validate the ecological niche model predictions since the models were based on Vancouver Island and G u l f Islands C. gattii observations (current endemic areas for C. gattii in B C ) . The 4 human (Figure 16) and 8 animal (Figure 17) C. gattii cases, and 4 positive environmental sampling sites (Figure 18) on the B C mainland were overlaid with the forecasted C. gattii ecological niche models. Note that the remaining 3 animal cases were not mapped due to missing address information, and 2 o f 4 positive environmental sampling sites were located in the northern Sunshine Coast area which is outside of the inset maps' extent and are therefore not shown in the figures. A l l 4 human cases (Burnaby, Surrey, Abbotsford and Sechelt), 5 of 8 animal cases (Delta, Maple Ridge, Abbotsford (2) and Chill iwack), and 3 of 4 positive environmental sampling sites (Langley and Powell River) on the B C mainland were located in the forecasted C. gattii ecological niche areas based on the E N M scenario using the distribution of human C. gattii cases to construct the model (Figure 16). The remaining 3 animal cases (Mission (2) and Abbotsford) and 1 positive environmental sampling site (Abbotsford) were located within 2.5 k m distance from the nearest forecasted C. gattii ecological niche areas. 52 Figure 16. Validation of C. gattii E N M predictions on the B C mainland based on the model input of human cases and environmental data layers with >95% accuracy. The distribution of mainland human (A) and animal cases (B), and positive environmental samples (C) are displayed in the inset maps. With respect to the E N M scenario using the distribution of animal C. gattii cases to construct the model, all 4 human cases, all 8 animal cases, and all 4 positive environmental sampling sites on the B C mainland were located in the forecasted C. gattii ecological niche areas (Figure 17). 53 Figure 17. Validation o f C.gattii E N M prediction on the B C mainland based on the model input of animal cases and environmental data layers with >95% accuracy. The distribution of mainland human (A) and animal cases (B), and positive environmental samples (C) are displayed in the inset maps. Finally, with respect to the E N M scenario using the distribution of positive C. gattii environmental sampling locations from permanently established sites to construct the model, 3 of 4 human cases (Burnaby, Abbotsford and Sechelt), 4 of 8 animal cases (Delta, Abbotsford (2) and Chill iwack), and 3 of 4 positive environmental sampling sites (Langley and Powell River) on the B C mainland were located in the forecasted C. gattii ecological niche areas (Figure 18). The remaining 4 animal cases (Maple Ridge, Miss ion (2) and Abbotsford) and 1 positive environmental sampling site (Abbotsford) were located within 2.5 km distance from the nearest forecasted C. gattii ecological niche areas. 54 Figure 18. Validation of C.gattii E N M prediction on the B C mainland based on the model input of positive environmental sampling locations and environmental data layers with >95% accuracy. The distribution of mainland human (A) and animal cases (B), and positive environmental samples (C) are displayed in the inset maps. Based on the results of this validation procedure, and coupled with the high training and testing accuracy measures, low number of commission and omission errors, and small p-values, the use of animal case data to construct the ecological niche model of C. gattii produced the best modeling results and highest predictive power for C. gattii presence on the B C mainland. 55 5.4 Cryptococcus gattii E c o l o g i c a l N i c h e C h a r a c t e r i z a t i o n Environmental characteristics of the forecasted ecological niche of C. gattii based on E N M using environmental data layers with >95% accuracy to construct the models are described in Table 6. L o w lying elevations (below 770 m and averaging 100 m above sea level), daily January average temperatures above 0°C, and presence within the C D F and C W H x m biogeoclimatic zones (Table 7) are strongly correlated with the ecological niche of C. gattii. The association of daily January average temperatures above 0°C with the forecasted ecological niche areas o f C. gattii, and the general absence (or transiently positive nature) o f C. gattii in geographic areas where the daily January average temperatures are below 0°C suggests that C. gattii is not able to survive and/or thrive in areas commonly experiencing freezing temperatures. Table 6. Summarized environmental characteristics of the forecasted optimal C. gattii ecological niche in B C based on the distribution of human and animal cases and permanently established C. gattii sites, and environmental data layers with >95% accuracy. Human Animal Environmental Data Layer Ave. Max. Min. Ave. Max. Min. Ave. Max. Min. Elevation (m) 120 762 1 106 762 1 94.8 448 1 Jan.Ave.Temp (°C) 2.8 4.7 0.9 2.9 5.6 1.1 2.9 5.6 0.7 Jan.Max.Temp (°C) 5.4 7.8 3.3 5.5 7.8 3.5 5.6 7.8 3.2 Jan.Min.Temp (°C) 0.1 3.9 -2.0 0.2 3.9 -2.0 0.1 3.9 -2.2 July Min.Temp (°C) ~ ~ ~ 11.5 15.2 8.1 ~ ~ ~ July Precip. (mm) ~ — ~ 41.0 111.3 15.5 ~ ~ — 56 Table 7. Summarized biogeoclimatic zone characteristics of the forecasted optimal C. gattii ecological niche areas based on the distribution of human and animal cases and permanently established C. gattii sites, and environmental data layers with >95% accuracy. Descriptions and characteristics of the biogeoclimatic zones of B C are described in Appendix E . Human Animal Environmental BGC Zone Number of Cells Area (Km2) Number of Cells Area (Km2) Number of Cells Area (Km2) CDFmm 6620 2383 6605 2378 6625 2385 CWHxm 1 10994 3958 10558 3801 11258 4053 CWHxm2 6653 2395 5310 1912 — ~ CWHmml 620 223 308 111 ~ ~ CWHmm2 54 19 41 15 ~ CWHdm — ~ 201 72 — ~ CWHdsl ~ ~ 20 7 ~ ~ CWHwh ~ ~ 83 30 ~ ~ CWHvml — - 6 2 — ~ EDFww ~ -- 40 14 ~ ~ E N M Total 24941 8979 23172 8342 17883 6438 The E N M of C. gattii based on the distribution of animal cases to construct the model also identifies mi ld daily July minimum temperatures (11.5°C) and moderate July precipitation (41 mm) in areas associated with C. gattii in the environment. Maritime weather conditions experienced in coastal southwestern B C , which is where the forecasted ecological niche of C. gattii is largely confined to, exhibit mi ld daily July minimum temperatures and moderate July precipitation. 57 5.5 A p p r o p r i a t e n e s s o f the M o d e l P r o p e r t i e s , M e t h o d s a n d D a t a The va l id i ty and appropriateness o f any mode l are determined by its propert ies, methods and data. C r i t i ca l issues and components o f the E N M procedure such as the spat ial scale o f analys is , C. gattii input and predictor env i ronmenta l data layers used to bu i l d the mode ls , and interpretation o f the mode l i ng results are d iscussed i n this sub-sect ion. 5.5.1 Spat ia l Sca le o f A n a l y s i s T h e 600 m ce l l s ize and spatial scale o f analys is used i n the E N M o f C. gattii i n B C was appropriate and advantageous for this study because C. gattii, i n the b io log i ca l context, can be considered a f i xed or less m o b i l e o rgan ism. G u i s a n and Thu i l l e r (2005) suggested E N M and analys is o f f i xed or less mob i l e organisms at h igh spatial reso lu t ion to op t im ize mode l i ng results. R o n (2005) conducted h is E N M o f the amph ib ian fungal pathogen Batrachochytrium dendrobatidis at a spatial resolut ion o f 2.5 arc minutes (4 k m ) , and it p roduced mode ls w i t h h igh predic t ive accuracy and power . The E N M o f C. gattii at 600 m spatial reso lu t ion descr ibed i n this study was approx imate ly 7 t imes more detai led than R o n ' s (2005) study. The 600 m ce l l s ize and spatial scale o f analys is used i n the E N M o f C. gattii i n B C was determined b y a number o f factors. F i rs t , the G T O P O 3 0 d ig i ta l e levat ion mode l was selected for this study because it conta ined complete geographic coverage o f the p rov ince , and the dataset was f reely avai lab le. Second , the 600 m ce l l s ize was appropriate for most o f the env i ronmenta l data layers, and espec ia l ly for the c l imat i c datasets because this spatial scale 58 was small enough to show distinctive local climatic phenomena yet preserved the overall regional climate patterns. Third, computer performance and data management for the E N M process was manageable for all the datasets at this spatial scale. Associated with the spatial scale of analysis is data uncertainty. Spatial and attribute uncertainty are inherent in raster GIS data models since information at the individual cell unit is stored as a homogenous entity. The difference between the location and value of a real-world feature and its location and value in a raster GIS model is dependent on the spatial resolution, attribute accuracy and type of data. For example, a coastline may appear "pixilated" when represented in the raster GIS data model since the entire cell is used to represent a land or sea feature. Wi th respect to attribute accuracy, the multiple characteristics o f a geographical area may be reduced to a single attribute value where two or more zones converge. Lastly, data uncertainty may differ depending on the type of data and how it was modeled. For example, the uncertainty of the aspect and slope data at 600 m spatial resolution is likely far less meaningful than that of the temperature and precipitation data since aspect and slope were derived from a detailed and accurate digital elevation model; whereas, the climatic data were derived from the spatial interpolation of geographically-biased and low density weather station observations. 5.5.2 Cryptococcus sattii Data The validity of any ecological niche model is dependent on the modeling methods ( G A R P was used in this study), quality and completeness of the species occurrence data, and quality 59 and completeness of the predictor environmental data layers used in the model (Ron, 2005). Therefore, it is important to note that the cryptococcal disease surveillance data used in this study do not definitively represent where C. gattii is found in the environment. Instead, cases were mapped by their address of residence, so these case locations were proxies to where exposure to C. gattii may have had occurred. Unfortunately, from an ecological niche modeling and epidemiological perspective, the high mobility of people in developed countries such as Canada, and the duration of time between exposure to C. gattii and onset of symptoms make identifying the actual place of exposure difficult to ascertain. However, the geographic concepts of activity space and distance decay provide basis for the use of human surveillance data as appropriate proxies for C. gattii presence in an area (Zipf, 1949; Carlstein and Thrift, 1978; Meentemeyer, 1989). E N M of C. gattii based on the distribution of human cases was therefore appropriate, but caveats pertaining to the limitations of the data, and possible resulting model errors due to the "uncertainty" of these data (whether each case truly represents the presence of C. gattii in the area), should be noted and understood. Animal cryptococcal disease surveillance data, especially the feline cases, may be more appropriate for identifying the true presence o f C. gattii in an area because companion animals are generally more sedentary, and their contact with soil and woody material (environs where C. gattii are most likely to be found) are more direct than humans - thereby providing more reliable location data for the E N M of C. gattii than human case data. Currently, the number o f reported animal cases is approximately 2 times greater than the number of reported human cases. N e w reports of cryptococcal disease in previously non-endemic C. gattii geographic areas in B C are typically identified first in animal cases, and 60 then in human cases. However, although a greater number of animal records have been reported to veterinary and public health, the detail of disease surveillance information relating to travel history and other relevant information is generally less complete than those of the human cases. A l so , a rigorous system for reporting and tracking animal C. gattii cases is not in place at the current time. Wi th respect to the use of C. gattii environmental sampling data to construct the ecological niche models, pre-selection bias in the selection o f locations sampled near previously identified human and animal cases and in environs frequented by humans may also introduce modeling errors. Reese et al. (2005) recently evaluated the effects of different sampling strategies on E N M and found that random and stratified sampling (e.g. a linear transect) produced the best overall modeling results. However, these sampling strategies are more difficult and expensive to carry out, and the success rate in finding positive results is considerably lower. The purposeful sampling strategy for C. gattii in B C can be considered a quasi-adaptive sampling design where adjacent areas to a positive sample were subsequently sampled. Reese (2005) found that although the adaptive sampling strategy produced the largest number of commission errors among the different survey designs tested, it also produced the fewest omission errors, especially for models with large sample sizes. The average number o f commission errors for the E N M of C. gattii based on environmental data layers with >95% accuracy was 0.7 (Table 5) despite the use of C. gattii environmental sampling data which was primarily collected by purposeful sampling. This was fewer than the number of errors committed in the E N M of C. gattii based on the distribution of human and animal cases - an average of 1.2 commission errors per model run. 61 Furthermore, some random sampling for C. gattii was performed in B C , especially during the initial stages of the environmental sampling strategy when cases of cryptococcosis were first identified and C. gattii was yet to be isolated from the environment. For example, random samples have been taken from the west coast of Vancouver Island, G u l f Islands, Sunshine Coast and Vancouver Lower Mainland (Figure 1). Also , a linear transect along a highway corridor from Parksville (N49°20 ' W124°19 ' ) to M a c M i l l a n Park (N49°17 ' W124°40 ' ) was systematically sampled (Kidd et al., 2007a). These sampling points were part of the overall environmental sampling dataset used in the E N M of C. gattii in B C . In addition to the quality and representativeness of the input data for E N M , predictive accuracy is related to the sample size of input data points used to train and test the ecological niche model. Stockwell and Peterson (2002) noted that predictive accuracy of E N M was related to the sample size of the input data points. In particular, accuracy on the training dataset decreased with larger sample sizes while accuracy on the test dataset increased as sample sizes increased. They concluded that G A R P required on average only ten data points to achieve 90% of maximum accuracy and was near maximal at 50 data points. These results were also observed by Hernandez et al. (2006). The E N M of C. gattii in B C described in this study used 122 human cases, 135 animal cases and 384 positive environmental sampling locations from permanently established sites which were randomly divided into 70% training and 30%) testing sub-datasets. Despite using more than 50 input data points (the optimal number according to Stockwell and Peterson, 2002) for each E N M scenario, the predictive accuracy of the C. gattii ecological niche models using environmental data layers with >95% accuracy was extremely high: training and testing accuracy >98%, and <2 commission and 62 omission errors per model run for the E N M of C. gattii based on the distribution of human and animal cases and positive environmental sampling locations from permanently established sites. Guisan and Thuiller (2005) emphasize that modeling the ecological niche of a species should be defined from empirical observations of individuals that reproduce successfully, and thus support a positive growth rate for the entire population. Therefore, the E N M of C. gattii in B C used only the permanently positive environmental sampling locations; the transiently positive environmental sampling locations were not included as input points for the ecological niche model because evidence of long term presence and establishment o f C. gattii in the environment has not been proven at these transiently positive locations to date. Spatial overlay o f the forecasted C. gattii ecological niche models (based on the environmental data layers with >95% accuracy) and transiently positive locations found that 47 of 55 transiently positive locations were located within the forecasted optimal C. gattii ecological niche based on the distribution of human cases, 44 of 55 transiently positive locations were located within the forecasted optimal C. gattii ecological niche based on the distribution of animal cases, and 41 of 55 transiently positive locations were located within the forecasted optimal C. gattii ecological niche based on the distribution of permanently positive environmental sampling locations. Transiently positive locations outside of the forecasted C. gattii ecological niche were primarily found along the highway corridor between Parksville and M a c M i l l a n Park. Inclusion of these input data points in the E N M of C. gattii would have increased the forecasted ecological niche area, but also negatively increased the number of commission errors (false presence) and decreased the overall model accuracy. 63 Guisan and Thuiller (2005) also warn that creating a model based on the observed distribution too closely (e.g. choosing an overly conservative selection of input environmental data layers) may lead to underestimating the true potential range of the species. This may explain the conservative prediction of the ecological niche model of C. gattii based on the input of positive environmental sampling locations from permanently established sites and environmental data layers with >95% training and testing accuracy (Figure 12C). The forecasted C. gattii ecological niche was considerably larger when environmental data layers with >90% accuracy were used to build the model (Figure 13C). 5.5.3 Environmental Data Layers The environmental data themes (i.e., topography, climate, biogeoclimatic zones-vegetation, and soil types) used here to predict the ecological niche of C. gattii were recommended and routinely used by other E N M researchers for various plant, animal, insect and fungi species (Guisan and Thuiller, 2005; Oberhauser and Peterson, 2003; Parra-Olea et al., 2005; Peterson, 2001; Ron, 2005; Soberon and Peterson, 2005; Stockwell and Peters, 1999). However, concerns relating to the "mechanistic approach" of E N M - determining the ecological niche of a species based on direct measures of physiographical variables and ignoring biotic interactions such as mutualistic relationships or competition and predation - have been raised by others (Soberon and Peterson, 2005). This is the situation faced here in the E N M of C. gattii in B C since all the predictor environmental data layers, with the exception of the biogeoclimatic zones data layer which is a composite of vegetation, physical geography and climate variables, are abiotic in nature. The use of these predictor environmental data layers 64 for the E N M of C. gattii in B C is not believed to be a limitation of this study however, since C. gattii has been isolated in high concentrations from gravel parking lots with little or no organic soil content (Kidd et al., 2007b). 5.5.4 Use of Significant Environmental Data Layers Model accuracy is also largely determined by the selection of environmental data layers as predictor variables to forecast the ecological niche of a species. The quantitative threshold value that determines whether a predictor environmental data layers is included or excluded is somewhat arbitrary. For this reason, a number of E N M scenarios were performed and the model results were examined to determine the best modeling procedures (indicated by small commission and omission errors, and high training and testing accuracy). E N M o f C. gattii based on the use of environmental data layers with >95% accuracy (elevation, biogeoclimatic zone and January average, maximum and minimum temperature for the human cases and positive environmental sampling locations from permanently established sites scenarios; elevation, biogeoclimatic zone, January average, maximum and minimum temperature, July minimum temperature, and July total precipitation for the animal cases scenario) produced the best modeling results. Wi th respect to determining the predictive accuracy of E N M , Peterson and Cohoon (1999) found that predictive accuracy was related to the number of predictor environmental data layers. They performed their E N M of 3 bird species with 8 original predictor environmental data layers, and jackknife analysis identified 5 environmental data layers which contributed 65 to high predictive accuracy of the models. Similarly, the E N M of C. gattii in B C described in this study used 15 original predictor environmental data layers, and jackknife analysis identified 5 (human cases and positive environmental sampling locations from permanently established sites scenarios) and 7 (animal cases scenario) environmental data layers, respectively, which contributed to high predictive accuracy of the models. 5.5.5 Interpretation of the Ecological Niche Model Prediction Maps A composite, "best subset" C. gattii E N M data layer was created for each scenario (human cases, animal cases and positive environmental sampling locations from permanently established sites) following the methods described by Peterson (2001). "Optimal" and "potential" C. gattii ecological niche areas were identified in B C (Figure 12). These are areas within the province that have suitable environmental conditions to support C. gattii. Specifically, the east coast of Vancouver Island from Campbell River (N50°01 ' W125°15 ' ) to Sooke (N48°22 ' W123°44 ' ) , Port Alberni Val ley (N49°15 ' W124°48 ' ) , Nimpkish Val ley (N50°24 ' W126°58 ' ) , Sunshine Coast (N50°00 ' W123°45 ' ) , and Vancouver Lower Mainland are forecasted optimal ecological niche areas for C. gattii in B C . In addition, small, isolated geographical areas on the Queen Charlotte Islands, B C central coast, west coast of Vancouver Island and southern B C interior are forecasted potential ecological niche areas for C. gattii in B C . Only a small percentage of the province (<2%) is geographically located within the forecasted ecological niche area of C. gattii. However, approximately 2/3 of B C ' s population 66 resides in this area, primarily in the Vancouver Lower Mainland and central to southern east coast of Vancouver Island. To date, it does not appear that C. gattii has been permanently established on the B C mainland. A small number of positive environmental samples have been isolated on the B C mainland in the past 2 years but subsequent, repeat environmental sampling has not isolated it in the same locations. Furthermore, the incidence and rate of human illness on the B C mainland is still disproportionately low as compared to the rates observed on Vancouver Island. The E N M of C. gattii performed in this study identifies, however, that the Vancouver Lower Mainland has optimal ecological niches that can support the establishment of C. gattii in the local environment. If C. gattii does become established in the Vancouver Lower Mainland in concentrations similar to the endemic areas on Vancouver Island, and i f similar rates of infection experienced in the endemic areas on Vancouver Island are also experienced in the Vancouver Lower Mainland, we would expect to see 79 cases of cryptococcosis per year (36 cases per mi l l ion population on Vancouver Island per year during 2002-2005 multiplied by 2.2 mil l ion population in the Vancouver Lower Mainland) in the Vancouver Lower Mainland alone! E N M provides the ability to forecast and visualize the ecological niche of a species in geographic space. Although it is interesting to visualize the similarities and differences between model outputs, it is not possible to evaluate the accuracy of the E N M based on visual analysis. Instead, the quantitative measures of commission and omission errors, training and testing accuracy, and statistical significance should be examined to determine the predictive accuracy of the models (Table 5). 67 6.0 SUMMARY AND CONCLUSION In summary, this study found that the "optimal" ecological niche areas of C. gattii in B C are limited to the central and south eastern coast of Vancouver Island, the G u l f Islands, Sunshine Coast and Vancouver Lower Mainland. Small, isolated geographical areas on the Queen Charlotte Islands, B C central coast, west coast of Vancouver Island and southern B C interior have environmental conditions that could potentially support the establishment of C. gattii in these areas. These forecasted ecological niche areas of C. gattii in B C are characterized by low lying elevations (below 770 m and averaging 100 m above sea level), above freezing daily January average temperatures, and presence within the C D F and C W H x m biogeoclimatic zones. In particular, the association of daily January average temperatures above 0°C with the forecasted ecological niche areas of C. gattii, and the general absence (or transiently positive nature) of C. gattii in geographic areas where the daily January average temperatures are below 0°C suggests that C. gattii is not able to survive and/or thrive in areas commonly experiencing freezing temperatures. In other words, C. gattii does not appear to be tolerant to freezing temperatures. This ecological observation - freezing temperatures as a limiting factor of C. gattii survival in the environment - can be tested in the laboratory. To the author's knowledge, experiments to determine the minimum thermal limit for C. gattii survival has not been conducted or reported since C. gattii has traditionally been associated with tropical and sub-tropical climates (i.e., above freezing temperatures). 68 A number of human (8) and animal (11) C. gattii cases without travel to Vancouver Island, and transiently positive C. gattii environmental samples (from 4 unique geographic sites) have been identified on the B C mainland (Sunshine Coast and Vancouver Lower Mainland). A l l o f these observations were located within or directly adjacent (within 2.5 k m proximity) to the forecasted C. gattii ecological niche areas on the B C mainland, thereby validating the usefulness and predictive power of E N M employed in this study. E N M proved useful and effective in predicting the ecological niche of C. gattii in B C based on the distribution of human and animal cases and positive environmental sampling locations from permanently established sites. The predictive accuracy of the C. gattii ecological niche models using environmental data layers with >95% accuracy was extremely high: training and testing accuracy was >98%, and <2 commission and omission errors were committed per model run. 6.1 Impact and Implications of the Study Findings These study findings are consistent with previous observations indicating the association of C. gattii distribution within the C D F and C W H x m biogeoclimatic zones (Kidd et al., 2004; K i d d et al., 2007b; MacDougall et al., 2007). They also provide an explanation as to why the distribution of C. gattii, especially in permanently established sites, is geographically restricted to the central and south eastern coast of Vancouver Island and G u l f Islands in B C despite the anthropogenic transport of woody material (Kidd et al., 2007a), and wind and mechanical dispersal of propagules to other areas of the province - freezing temperatures 69 experienced in areas throughout the rest of B C outside of the C D F and C W H x m biogeoclimatic zones appear to be the primary environmental limiting factor for C. gattii. However, geographical areas along the Sunshine Coast and Vancouver Lower Mainland, which are identified as optimal ecological niche areas for C. gattii, are forecasted to become endemic C. gattii areas in the near future i f the anthropogenic and mechanic dispersion of C. gattii into these areas continue, since these areas have suitable environmental conditions to support C. gattii. Furthermore, the inadvertent anthropogenic dispersion of C. gattii v ia transport of woody material such as lumber and landscaping products has the potential for public health and economic trade implications, especially i f these products are exported internationally. The use of human and animal cryptococcosis surveillance data to model the ecological niche of C. gattii was also found effective. The forecasted ecological niche of C. gattii based on the distribution of human and animal cases corresponded well to the forecasted ecological niche of C. gattii based on the distribution of positive environmental sampling locations from permanently established sites. Similarly, the use of animal surveillance data and C. gattii environmental sampling data were useful and effective in forecasting the geographic distribution of human cryptococcosis. In particular, animal surveillance data proved to be good indicators for C. gattii presence in an area. This suggests that surveillance for animal cryptococcosis may be useful as an early human disease alert system in B C because new reports of cryptococcosis in previously non-endemic C. gattii geographical areas have been identified in animals before humans and the number of reported animal cases is approximately 2 times greater than the number of reported human cases. The creation of a 70 comprehensive, province wide animal cryptococcosis surveillance system to track laboratory confirmed and clinically diagnosed cases of cryptococcosis due to C. gattii is recommended. 6.2 Study Strengths This study used existing human and animal cryptococcosis surveillance data, and environmental sampling data, to identify and model the ecological niche of C. gattii in B C using GIS and G A R P . It is the first attempt to describe the ecological niche and forecasted geographic range of C. gattii in B C . A review of the current E N M literature indicates that the scale of analysis at which the E N M of C. gattii in B C was performed (600 m cell size) is the most detailed of any species distribution model described to date. A finer spatial scale of analysis allows for greater identification and discernment of C. gattii ecological niche areas. The biophysical characteristics of C. gattii (i.e., the environmental limiting factors) were also inferred in this study. The investigation into the detection of C. gattii in the environment and identification of human and animal infections has been a model of cooperation between public health, veterinary health and academia (Galanis et al., 2006). This study highlights the value and strength of using a multi-disciplinary, landscape epidemiology approach to communicable disease surveillance and research since the health of human and animal populations are in large part determined by their interaction with the environment around them. 71 6.3 Study Limitations A common criticism of E N M is its tendency to identify the realized niche o f a species rather than the fundamental niche due to the use of empirical data to build the model (Guisan and Thuiller, 2005). The fundamental niche consists of the total potential area that meets all the physical and biological requirements of a species; whereas, the realized niche describes the actual distribution of a species which is influenced by factors such as dispersal, history and physical barriers. A s a result, E N M has a tendency to produce conservative predictions. A conservative prediction is beneficial for increasing the likelihood of isolating C. gattii in previously unsampled locations within the forecasted ecological niche, but it may exclude geographical areas of the C. gattii fundamental niche. Human populations in communities within these excluded geographical areas of the C. gattii fundamental niche may not receive the targeted public health notification and awareness of C. gattii in their community. Therefore, public health notification and awareness of C. gattii campaigns should be focused at the regional level (e.g. Fraser Val ley or Fraser East Health Service Delivery Area) rather than targeting specific communities (e.g. Abbotsford and Mission). A s discussed previously in Section 5.5, potential sources for errors in the E N M procedure were failure to include variables important for defining the ecological niche of C. gattii, and poor representativeness of the C. gattii observation data used to build the models. The human and animal cryptococcosis surveillance data used to build the ecological niche models do not definitively represent the presence of C. gattii in the local environment. Instead, they were proxy locations to where exposure to C. gattii may have occurred. Pre-selection bias of 72 sampling locations near previously identified human and animal cases and in environs frequented by humans may also introduce modeling errors. Furthermore, only occurrence data was used in the E N M of C. gattii because the geographic density and temporal frequency of environmental sampling was not adequate to definitively indicate C. gattii absence in an area. Unfortunately, the detection of C. gattii in the environment cannot be made with visual observation since the fungal organism is microscopic in size and C. gattii does not harm the health of its environmental host, trees. The climatic data used in the E N M of C. gattii were the 1971-2000 Normals. These long term average climate data provide more stable observations but the recent effects of global warming in B C should not be dismissed. The weather conditions experienced in southwestern B C over the past decade have generally been warmer and wetter in the winter months, and drier in the summer months, than normal (British Columbia Ministry of Water, Land and A i r Protection, 2002). This may be particularly important since C. gattii appears to have emerged in B C around 1999 based on the first identification of animal and human cryptococcosis in B C (i.e., are recent changes in climatic conditions responsible for the emergence of C. gattii in B C ? ) . Unfortunately, reliable climate data (in terms of quality control, completeness and wide geographic coverage) collected from 2000 onwards are not readily available or appropriate for high resolution climate modeling and E N M of C. gattii. Land use data were not used in the E N M of C. gattii. Deforestation and soil disturbance are hypothesized to increase dispersion, through aerosolization, of C. gattii (Kidd et al., 2007a). Unfortunately, recent land use data with provincial wide coverage is not readily available. 73 Access to and processing of this information is also prohibitively expensive and technically challenging. Performing this work at a local scale may be feasible, though. Despite these possible sources of error, the ecological niche models for C. gattii showed a high predictive power for identifying localities on Vancouver Island, G u l f Islands, Sunshine Coast and Vancouver Lower Mainland where C. gattii has been predicted and identified to date. 6.4 Proposed Possible Future Studies The results of this study provide opportunities for possible future studies. The role of climate change on the future geographic expansion of C. gattii in B C has been suggested by K i d d et al. (2004) and Bartlett et al. (2004). A warming trend w i l l produce larger geographical areas with suitable ecological habitat for C. gattii in B C (e.g. more areas with above freezing daily January average temperatures), and increase the potential for human illness as the landscape of C. gattii in the environment expands. Ecological niche modeling of C. gattii under climate change scenarios can be explored - the G A R P modeling software already provides this functionality. Climate change modeling data are also available; however, the spatial resolution of this data may currently be too coarse for this analysis though. Finer spatial resolution climate change data may become available in the future. For example, researchers at the University of Victoria (2007) have installed a number of new weather stations across Vancouver Island to collect real-time measurements of temperature, humidity, wind speed and direction, precipitation, solar and U V radiation, and atmospheric pressure to augment the system of weather stations operated by Environment Canada. It is hoped that with this 74 additional information, finer scale modeling of climate and climate change can be achieved in the near future. Environmental sampling for C. gattii in forecasted "optimal" and "potential" ecological niches should be performed to identify whether C. gattii is present in these areas to further validate the model, and to warn residents of the presence and potential health risks to C. gattii in the local environment. O f the areas identified as optimal C. gattii ecological niches -the east coast of Vancouver Island from Campbell River to Sooke, Port Alberni Valley, Nimpkish Valley, Sunshine Coast, and Vancouver Lower Mainland - only the Nimpkish Val ley area (located south of Port Hardy in northern Vancouver Island) lacks any environmental sampling data. Therefore, future environmental sampling for C. gattii should include the Nimpkish Valley, as well as areas in the forecasted potential ecological niche areas for C. gattii in B C : the Queen Charlotte Islands, B C central coast, west coast o f Vancouver Island and southern B C interior. Furthermore, environmental sampling for C. gattii in the Vancouver Lower Mainland and Sunshine Coast should continue so that the extent of C. gattii presence and establishment on the B C mainland can be determined. The ancestral source of C. gattii in B C has not been determined. K i d d et al. (2004) hypothesized two scenarios: that C. gattii was recently introduced from the tropics or subtropics and has now been established in specific ecological niches in B C , or that C. gattii has been present in the Pacific Northwest of North America for quite some time but recent changes in climate or disturbance o f forested landscapes has enabled C. gattii to thrive here. Environmental sampling for C. gattii in remote forecasted ecological niche areas where C. 75 gattii was not likely to be introduced by anthropogenic activity could be performed to determine whether C. gattii is present there. Isolation of C. gattii in remote areas would support the long time presence of C. gattii in B C and further challenge the traditional understanding of a tropical and sub-tropical ecology of C. gattii. 76 R E F E R E N C E S Babish G . 2006. Geostatistics without tears: a practical guide to surface interpolation, geostatistics, variograms and kriging. Environment Canada, Regina, S K . Bartlett K H , Fyfe M , MacDougall L , Duncan C, Mak S, K i d d S. 2004. Cryptococcus gattii, a tropical pathogen emerging in a temperate climate zone. American Meteorological Society 16 t h Conference on Biometeorology and Aerobiology. Vancouver, B C . 23-27 August 2004. Accessed 9 M a y 2006. http://ams.confex.com/ams/pdfpapers/80027.pdf [ B C C D C ] British Columbia Centre for Disease Control. 2005. Cryptococcal disease. Accessed 23 October 2006. http://www.bccdc.org/topic.php7iteiiiM09 British Columbia Ministry of Water, Land and A i r Protection. 2002. Indicators of climate change for British Columbia, 2002. Queens' Printer, Victoria, B C . Brown D P , Comrie A C . 2002. Spatial modeling of winter temperature and precipitation in Arizona and N e w Mexico, U S A . Climate Research. 22(2): 115-28. Callejas A , Ordonez N , Rodriguez M C , Castaneda E . 1998. First isolation of Cryptococcus neoformans var. gattii, serotype C , from the environment in Colombia. Medical Mycology. Oct;36(5):341-4. Carlstein T, Thrift N . 1978. Afterward: towards a time-space structured approach to society and environment. In Human Activi ty and Time Geography, pp. 225-263. Edited by Carlstein T, Parkes D , Thrift N . John Wiley and Sons, N e w York, N Y . Casadevall A , Perfect JR. 1998. Ecology of Cryptococcus neoformans. In "Cryptococcus neoformans". American Society for Microbiology, Washington, D . C . pp 41-70. Chen S, Sorrell T, Nimmo G , Speed B , Currie B , El l i s D , Marriott D , Pfeiffer T, Parr D , Byth K . 2000. Epidemiology and host- and variety-dependent characteristics of infection due to Cryptococcus neoformans in Australia and N e w Zealand. Australasian Cryptococcal Study Group. Clinical Infectious Diseases. Aug;31(2):499-508. Daly C , Gibson W P , Taylor G H , Johnson G L , Pasteris P. 2002. A knowledge-based approach to the statistical mapping of climate. Climate Research. 22(2):99-l 13. Duncan C . 2005. M . S c . thesis. University o f Saskatchewan, Saskatoon, Canada. Duncan C , Schwantje H , Stephen C, Campbell J, Bartlett K . 2006. Cryptococcus gattii in wildlife of Vancouver Island, British Columbia, Canada. Journal of Wildlife Diseases. Jan;42(l): 175-8. 77 Duncan C, Stephen C, Lester S, Bartlett K H . 2005. Sub-clinical infection and asymptomatic carriage of Cryptococcus gattii in dogs and cats during an outbreak of cryptococcosis. Medical Mycology. Sep;43(6):511-6. Dunkley R. 2004. Personal communication. Environment Canada, Atmospheric Environment Service. Vancouver, B C . Occurred 7 June 2004. El i th J, Graham C H , Anderson R P , Dudik M , Ferrier S, Guisan A , Hijmans R J , Huettmann F, Leathwick JR, Lehmann A , L i J, Lohmann L G , Loiselle B A , Manion G , Mori tz C , Nakamura M , Nakazawa Y , Overton J M , Peterson A T , Phillips SJ, Richardson K , Scachetti-Pereira R, Schapire R E , Soberon J, Will iams S, Wisz M S , Zimmermann N E . 2006. Novel methods improve prediction of species' distributions from occurrence data. Ecography. 29(2): 129-51. El l i s D H , Pfeiffer TJ . 1990. Natural habitat of Cryptococcus neoformans var. gattii. Journal of Clinical Microbiology. Jul;28(7): 1642-4. Environment Canada. 2006. Canadian climate normals or averages 1971-2000. Accessed 1 A p r i l 2006. http://www.climate.weatheroffice.ec.gc.ca/climate normals/index e.html Fortes ST, Lazera M S , Nishikawa M M , Macedo R C , Wanke B . 2001. First isolation of Cryptococcus neoformans var. gattii from a native jungle tree in the Brazilian Amazon rainforest. Mycoses. 44(5): 137-40. Fraser J A , Giles SS, Wenink E C , Geunes-Boyer S G , Wright JR, Diezmann S, A l l e n A , Stajich JE , Dietrich FS, Perfect JR, Heitman J. 2005. Same-sex mating and the origin of the Vancouver Island Cryptococcus gattii outbreak. Nature. Oct 27;437(7063): 1360-4. Fraser J A , Subaran R L , Nichols C B , Heitman J. 2003. Recapitulation of the sexual cycle of the primary fungal pathogen Cryptococcus neoformans var. gattii: implications for a outbreak on Vancouver Island, Canada. Eukaryotic Cel l . Oct;2(5): 1036-45. Galanis E , MacDougall L , K i d d S, Duncan C , Mithani S, M a k S, Stephen C, Bartlett K . 2006. Investigation of the emergence and spread of Cryptococcus gattii in B C : a collaboration between public health and academic professionals. Canadian Public Health Association 97* Annual Conference. M a y 28-31,2006. Vancouver, B C . Accessed 23 October 2006. http://www.conference.cpha.ca/english/documents/06_FINAL-program_e2.pdf http://www.conference.cpha.ca/english/documents/Abstracts Mondayl030.doc Glass G E . 2001. Geographic information systems. In Nelson K E , Will iams C M , Graham N M H (eds). Infectious disease epidemiology: theory and practice. Aspen Publishers, N e w York, N Y . 78 Granados D P , Castaneda E . 2005. Isolation and characterization of Cryptococcus neoformans varieties recovered from natural sources in Bogota, Colombia, and study of ecological conditions in the area. Microbial Ecology. Feb;49(2):282-90. Granados D P , Castaneda E . 2006. Influence of climatic conditions on the isolation of members of the Cryptococcus neoformans species complex from trees in Colombia from 1992-2004. F E M S Yeast Research. Jun;6(4):636-44. Grinnell J. 1917. Field tests of theories concerning distributional control. The American Naturalist. 51:115-28. Guisan A , Thuiller W . 2005, Predicting species distributions: offering more than simple habitat models. Ecology Letters. 8(9):993-1009. Hamann ID, Gillespie R J , Ferguson J K . 1997. Primary cryptococcal cellulites caused by Cryptococcus neoformans var. gattii in an immunocompetent host. Australian Journal of Dermatology. 1997 Feb;38(l):29-32. Harden G . 2003. Eucalypts - evolution and distribution. Royal Botanic Gardens, Sydney. Accessed 24 October 2006. http://www.amonline.net.au/factSheets/eucalypts.htm Heidorn K C . 2004. The weather doctor's weather almanac of British Columbia: land of a bi l l ion climates. Accessed 28 February 2007. http://www.islandnet.com/~see/weather/almanac/arc2004/alm04nov.htm Hernandez P A , Graham C H , Master L L , Albert D L . 2006. The effect of sample size and species characteristics on performance o f different species distribution modeling methods. Ecography. 29(5):773-85. Hijmans R J , Cameron SE , Parra J L , Jones P G , Jarvis A . 2005. Very high resolution interpolated climate surfaces for global land areas. International Journal of Climatology. 25(15): 1965-78. Hutchinson G E . 1959. Homage to Santa Rosalia, or why are there so many kinds of animals? The American Naturalist. 93:145-59. Johnston K , Ver Hoef J M , Krivoruchko K , Lucas N . 2001. Using ArcGIS geostatistical analyst. E S R I Press, Redlands, C A . K i d d SE, Bach PJ , Hingston A O , Mak S, Chow Y , MacDougall L , Kronstad JW, Bartlett K H . 2007a. Cryptococcus gattii dispersal mechanisms, British Columbia, Canada. Emerging Infectious Diseases. Jan;13(l):51-7. 79 K i d d SE , Chow Y , M a k S, Bach PJ , Chen H , Hingston A O , Kronstad JW, Bartlett K H . 2007b. Characterization of environmental sources of Cryptococcus gattii in British Columbia, Canada, and the Pacific Northwest. Appl ied and Environmental Microbiology. Mar;73(5):1433-43. K i d d SE , Guo H , Bartlett K H , X u J, Kronstad JW. 2005. Comparative gene genealogies indicate that two clonal lineages of Cryptococcus gattii in British Columbia resemble strains from other geographical areas. Eukaryotic Ce l l . Oct;4( 10): 1629-38. K i d d SE , Hagen F, Tscharke R L , Huynh M , Bartlett K H , Fyfe M , MacDougal l L , Boekhout T, Kwon-Chung K J , Meyer W . 2004. A rare genotype of Cryptococcus gattii caused the cryptococcosis outbreak on Vancouver Island (British Columbia, Canada). Proceedings of the National Academy of Sciences of the United States of America. Dec;101(49):17258-63. Kwon-Chung K J . 1976. Morphogenesis of Filobasidiella neoformans, the sexual state of Cryptococcus neoformans. Mycologia. Jul-Aug;68(4):821-33. Kwon-Chung K J , Bennett JE. 1984. Epidemiologic differences between the two varieties of Cryptococcus neoformans. American Journal of Epidemiology. Jul;120(l):123-30. Kwon-Chung K J , Boekhout T, Fel l JW, Diaz M . 2002. Proposal to conserve the name Cryptococcus gattii against C. hondurianus and C. bacillisporus (Basidiomycota, Hymenomycetes, Tremellomycetidae). Taxon. Nov;51(4):804-6. Kolivras K N , Comrie A C . 2003. Modeling valley fever (coccidioidomycosis) incidence on the basis of climate conditions. International Journal of Biometeorology. Mar;47(2): 87-101. Lazera M S , Salmito Cavalcanti M A , Londero A T , Trilles L , Nishikawa M M , Wanke B . 2000. Possible primary ecological niche of Cryptococcus neoformans. Medical Mycology. Oct;38(5):379-83. Lester SJ, Kowalewich N J , Bartlett K H , Krockenberger M B , Fairfax T M , M a l i k R. 2004. Clinicopathologic features of an unusual outbreak of cryptococcosis in dogs, cats, ferrets, and a bird: 38 cases (January to July 2003). Journal of the American Veterinary Medical Association. Dec 1;225(11):1716-22. Lindberg J, Hagen F , Laursen A , Stenderup J, Boekhout T. 2007. Cryptococcus gattii risk for tourists visiting Vancouver Island, Canada. Emerging Infectious Diseases. Jan;13(l): 178-9. MacDougall L , Fyfe M . 2006. Emergence of Cryptococcus gattii in a novel environment provides clues to its incubation period. Journal of Clinical Microbiology. May;44(5): 1851-2. 80 MacDougall L , K i d d SE , Galanis E , M a k S, Leslie M J , Cieslak PR, Kronstad J W , Morshed M , Bartlett K H . 2007. Spread of Cryptococcus gattii in British Columbia, Canada and detection in the Pacific Northwest, U S A . Emerging Infectious Diseases. Jan;13(l):42-50. M a k S, Duncan C , Bartlett K , Stephen C , Fyfe M , MacDougall L , Raverty S. 2004. Using GIS to track cryptococcosis in B C . GisVef04. University of Guelph, O N . July 21-25, 2004. Accessed 19 October 2006. http://www.gisvet.org/Documents/GisVet04/Poster/Mak.pdf Meade M S , Florin JW, Gesler W M . 1988. Medical geography. The Guilford Press, N e w York, N Y . Meentemeyer V . 1989. Geographic perspectives of space, time, and scale. Landscape Ecology. 3(3/4): 163-73. Meidinger D , Pojar D . 1991. Special report series 6: ecosystems of British Columbia. British Columbia Ministry of Forests. Accessed 19 October 2006. http://www.for.gov.bc.ca/hfrJ/pubs/docs/Srs/SRseries.htm http://www.for.gov.bc.ca/hre/becweb/resources/maps/gisproducts.html Meyer W , Castaneda A , Jackson S, Huynh M , Castaneda E ; IberoAmerican Cryptococcal Study Group. 2003. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerging Infectious Diseases. Feb;9(2): 189-95. Mycology Online, University of Adelaide. 2006. Cryptococcus gattii. Accessed 9 October 2006. http://www.mycologv.adelaide.edu.au/Fungal Descriptions/Yeasts/ Cryptococcus/Cgattii .html National Aeronautics and Space Administration, Center for Health Applications of Aerospace Related Technologies. 2006. Landscape epidemiology and RS/GIS . Accessed 5 January 2007. http://geo.arc.nasa.gov/sge/health/landepi.html National Center for Biotechnology Information. 2006. GenBank overview. Accessed 28 October 2006. http://www.ncbi.nlm.nih.gov/Genbank/ National Oceanic and Atmospheric Administration, Satellite and Information Service. 2006. National climatic data center. Accessed 27 March 2006. http://www5.ncdc.noaa.gov/cgi-bin/climatenormals/climatenormals.pl Ninyerola M , Pons X , Roure J M . 2000. A methodological approach of climatological modelling of air temperature and precipitation through GIS techniques. International Journal of Climatology. 20(14): 1823-41. 81 Oberhauser K , Peterson A T . 2003. Modeling current and future potential wintering distributions of eastern North American monarch butterflies. Proceedings of the National Academy of Sciences of the United States of America. Nov; 100(24): 14063-8. Parra-Olea G , Martinez-Meyer E , Perez-Ponce de Leon G . 2005. Forecasting climate change effects on salamander distribution in the highlands of central Mexico . Biotropica. Jun;37(2):202-8. Payne K , Stockwell D R B . 1996. G A R P modelling system user's guide and technical reference. Accessed 24 June 2007. http://landshape.org/enm/garp-modelling-system-users-guide-and-technical-reference/ Penfold A R , Wi l l i s JL . 1961. The eucalypts. Leonard H i l l , London, U K . Peterson A T . 2001. Predicting species' geographic distributions based on ecological niche modeling. The Condor. Feb;103(3):599-605. Peterson A T . 2006. Ecologic niche modeling and spatial patterns of disease transmission. Emerging Infectious Diseases. Dec;12(12):1822-6. Peterson A T , Bauer JT, M i l l s J N . 2004a. Ecologic and geographic distribution of filovirus disease. Emerging Infectious Diseases. Jan;10(l): 40-7. Peterson A T , Cohoon K P . 1999. Sensitivity of distributional prediction algorithms to geographic data completeness. Ecological Modell ing. 117(1):159-164. Peterson A T , Ortega-Huerta M A , Bartley J, Sanchez-Cordero V , Soberon J, Buddemeier R H , Stockwell D R . 2002a. Future projections for Mexican faunas under global climate change scenarios. Nature. Apr;416(6881):626-9. Peterson A T , Papes M , K l u z a D A . 2003. Predicting the potential invasive distributions of four alien plant species in North America. Weed Science. 51:863-8. Peterson A T , Pereira R S , Neves V F . 2004b. Using epidemiological survey data to infer geographic distributions of leishmaniasis vector species. Revista da Sociedade Brasileira de Medicina Tropical. Jan-Feb; 17(1): 10-4. Peterson A T , Sanchez-Cordero V , Beard C B , Ramsey J M . 2002b. Ecologic niche modeling and potential reservoirs for Chagas disease, Mexico. Emerging Infectious Diseases. Jul;8(7):662-7. Peterson A T , Shaw J. 2003. Lutzomyia vectors for cutaneous leishmaniasis in Southern Brazi l : ecological niche models, predicted geographic distributions, and climate change effects. International Journal of Parasitology. 33:919-31. 82 Pidwirny M . 2006. Chapter 9: Introduction to the biosphere (g). Concept of ecological niche. Accessed 23 January 2007. http://www.physicalgeography.net7fiindamentals/9g.html Reese G C , Wilson K R , Hoeting J A , Flather C H . 2005. Factors affecting species distribution predictions: a simulation modeling experiment. Ecological Applications. 15(2):554-64. Ron SR. 2005. Predicting the distribution of the amphibian pathogen Batrachochytrium dendrobatidis in the new world. Biotropica. Jun:37(2);209-21. Soberon J, Peterson A T . 2005. Interpretation of models of fundamental ecological niches and species' distributional areas. Biodiversity Informatics. 2:1-10. Sorrell T C . 2001. Cryptococcus neoformans variety gattii. Medical Mycology. Apr;39(2): 155-68. Southeast Missouri State University, Department of Biology. 2006. Basidiomycetes. Accessed 9 October 2006. http://www.basidiomycetes.org/ Staib F , Seibold M , Antweiler E , Frohlich B , Weber S, Blisse A . 1987. The brown colour effect ( B C E ) of Cryptococcus neoformans in the diagnosis, control and epidemiology of C. neoformans infections in A I D S patients. Zentralblatt fur Bakteriologie, Mikrobiologie und Hygiene. Series A , Medical Microbiology, Infectious Diseases, Virology, Parasitology. Aug;266(l-2): 167-77. Statistics Canada. 2003. Lower mainland and southern Vancouver Island. Accessed 1 March 2007. http://geodepot3b.statcan.ca/Diss/Highlights/Page9/Page9c e.cfm http://geodepot.statcan.ca/Diss/Maps/ThematicMaps/vancouver e.cfm Stephen C , Lester S, Black W , Fyfe M , Raverty S. 2002. Multispecies outbreak of cryptococcosis on southern Vancouver Island, British Columbia. Canadian Veterinary Journal. Oct;43(10):792-4. Stockwell D , Peters D . 1999. The G A R P modelling system: problems and solutions to automated spatial prediction. International Journal of Geographic Information Science. Mar;13(2):143-58. Stockwell D R B , Peterson A . 2002. Effects of sample size on accuracy of species distribution models. Ecological Modell ing. 148(1):1-13. Tachiiri K , Klinkenberg B , Mak S, Kazmi J. 2006. Predicting outbreaks: a spatial risk assessment of West Ni l e virus in British Columbia. International Journal of Health Geographies. M a y 16; 5:21. United States Geological Survey, Earth Resources Observation and Science. 1996. GTOPO30 . Accessed 26 M a y 2006. http://edc.usgs.gov/products/elevation/gtopo30/gtopo30.html 83 University of British Columbia Botanical Garden and Centre for Plant Research. 2004. Hardy Eucalyptus varieties. Accessed 30 June 2007. http://www.ubcbotanicalgarden.org/ forums/showthread.php?s=e 1 f55c4a7c69f50678fa469al 71407fb&t=233 5 University of British Columbia Cryptococcus Research. 2006. Laboratory methods. Accessed 2 November 2006. http://www.cher.ubc.ca/crvptococcus/new/methods.htm University of Kansas Center for Research. 2002. Desktop G A R P : users manual. Accessed 12 September 2006. http://nhm.ku.edu/desktopgarp/index.html University of Victoria. 2007. School weather stations expand across Vancouver Island. The Ring. 33(1). Accessed 9 M a y 2007. http://ring.uvic.ca/07ian05/weather-stations.html Vanbreuseghem R, Takashio M . 1970. A n atypical strain of Cryptococcus neoformans (San Felice) Vui l lemin 1894. II. Cryptococcus neoformans var. gattii var. nov. Annales de la Societe Beige de Medecine Tropicale. 50(6):695-702. Zacharin R F . 1978. Emigrant eucalypts: gum trees as exotics. Melbourne University Press, Melbourne, Australia. Z i p f G K . 1949. Human behavior and the principle of least effort. Addison-Wesley Press, Cambridge, United Kingdom. 84 APPENDICES Appendix A. Worldwide Distribution of Cryptococcus gattii A n extensive review of the literature was performed to describe the worldwide distribution of C. gattii human and animal infections, and C. gattii isolated in the environment. Molecular typing information was collated by S. K i d d (2006, unpublished data). Table A l describes the worldwide geography of C. gattii and cites the reference(s) for each observation. 85 Table A l . Worldwide geography of Cryptococcus gattii Country Human Animal Environment Reference Cited Notes Argentina VGII, VGIII Bava AJ, Negroni R. The epidemiological characteristics of 105 cases of cryptococcosis diagnosed in the Republic of Argentina between 1981-1990. Rev Inst Med Trop Sao Paulo. 1992 Jul-Aug;34(4):335-40.; Meyer W, et al. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerg Infect Dis. 2003 Feb;9(2):189-95. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2):155-68. Argentina VGI Meyer W, et al. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerg Infect Dis. 2003 Feb;9(2):189-95. Aruba VGII Boekhout T, et al. Hybrid genotypes in the pathogenic yeast Cryptococcus neoformans. Microbiology. 2001 Apr;147(Pt 4): 891-907. Kidd SE, et al. A rare genotype of Cryptococcus gattii caused the cryptococcosis outbreak on Vancouver Island (British Columbia, Canada). Proc Natl Acad Sci USA. 2004 Dec 7;101(49):17258-63. Animal: goat Australia VGI, VGII, VGIII Kwon-Chung KJ, Bennett JE. Epidemiologic differences between the two varieties of Cryptococcus neoformans. Am J Epidemiol. 1984 Jul;120(l): 123-30. Australia VGI, VGII Malik R, et al. Cryptococcosis in cats: clinical and mycological assessment of 29 cases and evaluation of treatment using orally administered fluconazole. JMed VetMycol. 1992;30(2):133-44. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68. Animal: cat, dog, sheep, horse, koala, ferret, native echidna Australia VGI, VGII, VGIII Ellis DH, Pfeiffer TJ. Natural habitat of Cryptococcus neoformans var. gattii. J Clin Microbiol. 1990 Jul;28(7):1642-4.; Sorrell TC, et al. Concordance of clinical and environmental isolates of Cryptococcus neoformans var. gattii by random amplification of polymorphic DNA analysis and PCR fingerprinting. J Clin Microbiol. 1996 May;34(5):1253-60. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68.; Kidd SE, et al. Isolation of two molecular types of Cryptococcus neoformans var. gattii from insect frass. Med Mycol. 2003 41(2):171-6.; Kidd SE, PhD Thesis 2003. Environment: Eucalyptus camaldulensis, E. blakelyi, E. tereticornis, E. rudis, E. gomphocephala Country Human Animal Environment Reference Cited Notes Brazil VGII Meyer W, et al. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerg Infect Dis. 2003 Feb;9(2): 189-95. Brazil VGII Lazera MS, et al. Isolation of both varieties of Cryptococcus neoformans from saprophytic soruces in the City of Rio de Janeiro, Brazil. J Med VetMycol. 1993 31:449-454. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2):155-68.; Meyer W, et al. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerg Infect Dis. 2003 Feb;9(2): 189-95. Environment: bat guano, Moquilea tomentosa Cambodia VG Ellis DH, Pfeiffer TJ. Natural habitat of Cryptococcus neoformans var. gattii. J Clin Microbiol. 1990 Jul;28(7): 1642-4. Canada, British Columbia VGI, VGIIa, VGIIb Kidd SE, et al. A rare genotype of Cryptococcus gattii caused the cryptococcosis outbreak on Vancouver Island (British Columbia, Canada). Proc Natl Acad Sci USA. 2004 Dec 7;101(49):17258-63.; MacDougall L, et al. Spread of Cryptococcus gattii in British Columbia, Canada and detection in the Pacific Northwest, USA. Emerg Infect Dis. 2007 Jan;13(l):42-50. Canada, British Columbia VGI, VGIIa, VGIIb Stephen C, et al. Multispecies outbreak of cryptococcosis on southern Vancouver Island, British Columbia. Can Vet J. 2002 Oct;43(10):792 4.; Duncan C, et al. Sub-clinical infection and asymptomatic carriage of Cryptococcus gattii in dogs and cats during an outbreak of cryptococcosis. Med Mycol. 2005 Sep;43(6):511-6. Kidd SE, et al. A rare genotype of Cryptococcus gattii caused the cryptococcosis outbreak on Vancouver Island (British Columbia, Canada). Proc Natl Acad Sci USA. 2004 Dec 7;101(49):17258-63. Animal species: cat, dog, ferret, porpoise, llama, bird, horse Canada, British Columbia VGI, VGIIa, VGIIb Kidd SE, et al. Comparative gene genealogies indicate that two clonal lineages of Cryptococcus gattii in British Columbia resemble strains from other geographical areas. Eukaryot Cell. 2005 Oct;4(10):1629-38.; Kidd SE, et al. Characterization of environmental sources of Cryptococcus gattii in British Columbia, Canada, and the Pacific Northwest. Appl Environ Microbiol. 2006 Mar;73(5): 1433-43 Environment: Douglas-fir, red alder, maple, Garry oak, red cedar, pine, soil, water, air Country Human Animal Environment Reference Cited Notes Central Africa VG Ellis DH, Pfeiffer TJ. Natural habitat of Cryptococcus neoformans var. gattii. J Clin Microbiol. 1990 Jul;28(7): 1642-4. Chile VGII Meyer W, et al. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerg Infect Dis. 2003 Feb;9(2):189-95. China VGI Li A, et al. The isolation of Cryptococcus neoformans from pigeon droppings and serotyping of naturally and clinically sourced isolates in China. Mycopathologia. 1993 124:1-5. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68.; Boekhout T, et al. Hybrid genotypes in the pathogenic yeast Cryptococcus neoformans. Microbiology. 2001 Apr;147(Pt 4):891-907. China VG Li A, et al. The isolation of Cryptococcus neoformans from pigeon droppings and serotyping of naturally and clinically sourced isolates in China. Mycopathologia. 1993 124:1-5. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68. Columbia VGI, VGII, VGIII, VGIV Meyer W, et al. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerg Infect Dis. 2003 Feb;9(2):189-95. Kidd SE, et al. Comparative gene genealogies indicate that two clonal lineages of Cryptococcus gattii in British Columbia resemble strains from other geographical areas. Eukaryot Cell. 2005 Oct;4(10):1629-38. Columbia VGIII, VGIV Callejas A, et al. First isolation of Cryptococcus neoformans var. gattii serotype C from the environment in Columbia. Med Mycol. 1998 36:341-344. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68.; Boekhout T, et al. Hybrid genotypes in the pathogenic yeast Cryptococcus neoformans. Microbiology. 2001 Apr;147(Pt 4):891-907. Environment: Terminalia catappa Guatemala VGIII Meyer W, et al. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerg Infect Dis. 2003 Feb;9(2):189-95. Honduras VGI Boekhout T, et al. Hybrid genotypes in the pathogenic yeast Cryptococcus neoformans. Microbiology. 2001 Apr;147(Pt 4): 891-907. Country Human Animal Environment Reference Cited Notes India VGI, VGII, VGIII, VGIV Padhye AA, et al. Cryptococcus neoformans var. gattii in India. J Med Vet Mycol. 1993;31(2): 165-8.; Boekhout T, et al. Hybrid genotypes in the pathogenic yeast Cryptococcus neoformans. Microbiology. 2001 Apr;147(Pt4):891-907. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68.; Boekhout T, et al. Hybrid genotypes in the pathogenic yeast Cryptococcus neoformans. Microbiology. 2001 Apr;147(Pt 4):891-907.; Kidd SE. PhD Thesis 2003. Italy VGI Montagna MT, et al. Cryptococcus neoformans var. gattii en Italic Note I. Premier cas autochtone de meningite a serotype B chez un sujet VIH positif. J Mycol Med. 1997 7:90-92. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2):155-68.; Kidd SE. PhD Thesis 2003. Malaysia VG Keah KC, et al. Serotypes of clinical isolates of Cryptococcus neoformans in Malaysia. Trop Biomed. 1994 11:205-207. Sorrell TC. Cryptococcus neoformans variety gattii:. Med Mycol. 2001 Apr;39(2):155-68. Mexico VGI, VGIII, VGIV Castanon-Olivares LR, et al. Frequency of Cryptococcus species and varieties in Mexico and their comparison with some Latin American countries. Rev Latinoam Microbiol. 2000 Jan-Mar;42(l):35-40. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68.; Meyer W, et al. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerg Infect Dis. 2003 Feb;9(2):189-95. Mexico VG Licea BA, et al. Aislamiento de Cryptococcus neoformans var. gattii de Eucalyptus tereticornis. RevIberMicol. 1996 13:27-28. Casadevall A, Perfect JR. Cryptococcus neoformans. Washington: American Society for Microbiology Press. 1998. Environment: Eucalyptus tereticornis Nepal VG Ellis DH, Pfeiffer TJ. Natural habitat of Cryptococcus neoformans var. gattii. J Clin Microbiol. 1990 Jul;28(7): 1642-4. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2):155-68. New Zealand VGIII Kidd SE. PhD Thesis 2003. New Zealand VGI Hill FI, et al. Cryptococcosis in a North Island brown kiwi (Apteryx australis mantelli) in New Zealand. J Med Vet Mycol. 1995 Sep-Oct;33(5):305-9. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68.; Kidd SE. PhD Thesis 2003. Animal: kiwi Papau New Guinea VGI Currie BJ, et al. Cryptococcus neoformans var. gattii. Lancet. 1990 Dec 8;336(8728):1442. So rell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68. Paraguay VG Kwon-Chung KJ, Bennett JE. Epidemiologic differences between the two varieties of Cryptococcus neoformans. Am J Epidemiol. 1984 Jul;120(l):123-30. Country Human Animal Environment Reference Cited Notes Peru VGI Bustamante Rufino B, Swinne D. Cryptococcus neoformans var. gattii isolates from two Peruvian patients. Rev Iberoam Micol. 1998 15:22-24. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68.; Meyer W, et al. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerg Infect Dis. 2003 Feb;9(2):189-95. Rwanda VG Bogaerts J, et al. AIDS-associated cryptococcal meningitis in Rwanda (1983-1992): epidemiologic and diagnostic features. J Infect. 1999 Jul;39(l):32 7. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68. South Africa VGI, VGIII, VGIV Crewe-Brown H, et al. A 10 year study of cryptococcal meningitis. In Program and Abstracts of the 4th International Conference on Cryptococcus and Cryptococcosis. London, UK. September 12-16, 1999. The Royal Society. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2):155-68.; Kidd SE. PhD Thesis 2003. South Africa VGIV Bolton LA, et al. Cryptococcosis in captive cheetah (Acinonyx jubatus): two cases. J S Afr Vet Assoc. 1999 70:35-39. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68. Animal: king cheetah Spain VG Colom MF, et al. First case of human cryptococcosis due to Cryptococcus neoformans var. gattii in Spain. J Clin Microbiol. 2005 Jul;43(7):3548-50. Spain VGI Baro T, et al. First identification of autochthonous Cryptococcus neoformans var. gattii isolated from goats with predominantly severe pulmonary disease in Spain. J Clin Microbiol. 1998 36:458-461. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68.; Boekhout T, et al. Hybrid genotypes in the pathogenic yeast Cryptococcus neoformans. Microbiology. 2001 Apr;147(Pt 4):891-907. Animal: goat Taiwan VG Hsu MM, et al. Serotypes and mating types of clinical strains of Cryptococcus neoformans isolated in Taiwan. Mycopathologia. 1994 125:77-81. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2): 155-68. Thailand VGIIb Ellis DH, Pfeiffer TJ. Natural habitat of Cryptococcus neoformans var. gattii. J Clin Microbiol. 1990 Jul;28(7): 1642-4. Country Human Animal Environment Reference Cited Notes Uruguay VGII Gezuele E, et al. Isolation of Cryptococcus neoformans var. gattii from nest of the wasp Polybi occidentalis. Rev Iberoam Micol. 1993 10:5-6. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2):155-68.; Fraser JA, et al. Same-sex mating and the origin of the Vancouver Island Cryptococcus gattii outbreak. Nature. 2005 Oct 27;437(7063):1360-4. Environment: wasp nest USA, California VGII, VGIII Kwon-Chung KJ, Bennett JE. Epidemiologic differences between the two varieties of Cryptococcus neoformans. Am J Epidemiol. 1984 Jul;120(l):123-30. USA, California VGI Miller WG, et al. Cryptococcosis in a bottlenose dolphin (Tursiops truncatus ) caused by Cryptococcus neoformans var. gattii. J Clin Microbiol. 2002 Feb;40(2):721-4. Animal: dolphin, parrot USA, California VGI, VGIIa, VGIII Diaz MR, et al. Molecular sequence analyses of the intergenic spacer (IGS) associated with rDNA of the two varieties of the pathogenic yeast, Cryptococcus neoformans. Syst Appl Microbiol. 2000 Dec;23(4):535-45. Kidd SE, et al. Comparative gene genealogies indicate that two clonal lineages of Cryptococcus gattii in British Columbia resemble strains from other geographical areas. Eukaryot Cell. 2005 Oct;4(10): 1629-38.; Fraser JA, et al. Same-sex mating and the origin of the Vancouver Island Cryptococcus gattii outbreak. Nature. 2005 Oct 27;437(7063):1360-4. Environment: Eucalyptus camaldulensis USA, Hawaii V G Ellis DH, Pfeiffer TJ. Natural habitat of Cryptococcus neoformans var. gattii. J Clin Microbiol. 1990 Jul;28(7): 1642-4. USA, Oklahoma V G Muchmore HG, Scott EN, Felton FG, Fromtling RA. Cryptococcus neoformans serotype groups encountered in Oklahoma. Am J Epidemiol. 1980 Jul;112(l):32-8. Casadevall A, Perfect JR. Cryptococcus neoformans. Washington: American Society for Microbiology Press. 1998. USA, Oregon VGII-different genotypes to those in BC MacDougall L, et al. Spread of Cryptococcus gattii in British Columbia, Canada and detection in the Pacific Northwest, USA. Emerg Infect Dis. 2007 Jan;13(l):42-50. Country Human Animal Environment Reference Cited Notes USA, Washington VGIIa Boekhout T, et al. Hybrid genotypes in the pathogenic yeast Cryptococcus neoformans. Microbiology. 2001 Apr;147(Pt 4): 891-907. Kidd SE, et al. Comparative gene genealogies indicate that two clonal lineages of Cryptococcus gattii in British Columbia resemble strains from other geographical areas. Eukaryot Cell. 2005 Oct;4(10):1629-38. USA, Washington VGIIa MacDougall L, et al. Spread of Cryptococcus gattii in British Columbia, Canada and detection in the Pacific Northwest, USA. Emerg Infect Dis. 2007 Jan;13(l):42-50. Animal: cats USA, Washington VGIIa MacDougall L, et al. Spread of Cryptococcus gattii in British Columbia, Canada and detection in the Pacific Northwest, USA. Emerg Infect Dis. 2007 Jan;13(l):42-50.; Kidd SE, et al. Characterization of environmental sources of Cryptococcus gattii in British Columbia, Canada, and the Pacific Northwest. Appl Environ Microbiol. 2006 Mar;73(5):1433-43. Environment: soil Venezuela VGII, VGIII Kwon-Chung KJ, Bennett JE. Cryptococcosis. In Kwon-Chung KJ, Bennett JE, eds. Medical Mycology. Philadelphia: Lea& Febiger. 1992: 397-445. Sorrell TC. Cryptococcus neoformans variety gattii. Med Mycol. 2001 Apr;39(2):155-68.; Meyer W, et al. Molecular typing of IberoAmerican Cryptococcus neoformans isolates. Emerg Infect Dis. 2003 Feb;9(2): 189-95. Vietnam VG Ellis DH, Pfeiffer TJ. Natural habitat of Cryptococcus neoformans var. gattii. J Clin Microbiol. 1990 Jul;28(7): 1642-4. Zaire VGI Boekhout T, et al. Hybrid genotypes in the pathogenic yeast Cryptococcus neoformans. Microbiology. 2001 Apr;147(Pt 4): 891-907. Appendix B. Ethical Review Certificate of Approval UBC The University of British Columbia Office of Research Services Behavioural Research Ethics Board Suite 102, 6190 Agronomy Road, Vancouver, B.C. V6T 1Z3 CERTIFICATE OF APPROVAL - MINIMAL RISK PRINCIPAL INVESTIGATOR: ,Brian Klinkenberg INSTITUTION / DEPARTMENT: UBC/Arts/Geography UBC BREB NUMBER: IH07-01235 NSTITUTION(S) WHERE RESEARCH WILL BE CARRIED OUT: Institution Site BC Centre for Disease Control Other locations where the research will be conducted: N/A BC Centre for Disease Control CO-INVESTIGATOR(S): Karen H. Bartlett Sunny Mak SPONSORING AGENCIES: N/A PROJECT TITLE: Ecological Niche Modeling of Cryptococcus gattii in British Columbia CERTIFICATE EXPIRY DATE: June 4, 2008 DOCUMENTS INCLUDED IN THIS APPROVAL: DATE APPROVED: June 4, 2007 N/A [The application for ethical review and the document(s) listed above have been reviewed and the procedures were found to be acceptable on ethical grounds for research involving human subjects. Approval Is issued on behalf of the Behavioural Research Ethics Board and signed electronically by one of the following: Dr. Peter Suedfeld, Chair Dr. Jim Rupert, Associate Chair Dr. Arminee Kazanjian, Associate Chair Dr. M. Judith Lynam, Associate Chair Dr. Laurie Ford, Associate Chair 93 Appendix C. Climate Station Data The 30 year average (1971-2000 Normals) monthly climate station data from Environment Canada and the National Climatic Data Center (National Oceanic and Atmospheric Administration) were used for the temperature and precipitation surface interpolation models (Figure C I ) . B C climate station data were extracted from the Environment Canada Pacific and Yukon Region (Vancouver, B C ) reference database. Climate station data from adjacent jurisdictions ( A B , Y T , N T , A K , W A , ID, M T ) within 200 k m of the B C border were also obtained from the Environment Canada (2006) and National Oceanic and Atmospheric Administration (2006) internet portals to minimize edge effects in the surface interpolation procedures. Data values from the American climate stations were converted to metric units, and then merged with the Canadian climate stations in an Access database. Climate stations with at least 15 years of data for temperature and/or precipitation were queried from the Access database, mapped in ArcGIS and used for the surface interpolation modeling (Figure C2). A total of 634 climate stations were used (Table C I ) . The climate station dataset was divided into training (80%) and testing (20%) sub-datasets for the temperature and precipitation surface interpolation modeling procedure. 94 Figure C I . Data flow o f climate station data used in the surface interpolation models. E C = Environment Canada. P Y R = Pacific and Yukon Region. N O A A = National Oceanic and Atmospheric Administration. EC PYR Reference Database EC Climate Normals N O A A Climate Normals data extract I data download AB, YT, NT merge datasets query for climate stations wi th >= 15 years o f temperature and/or precipitation data f Access Database process data for A r c G I S mapping map climate stations wi th latitude/longitude coordinates ArcGIS Shapefile I A K , WA, ID, M T include stations within 200 k m o f B C J convert to metric values Analysis Dataset 95 Figure C2. Geographic distribution of climate stations used in the temperature and precipitation surface interpolation models. Y T N T L E G E N D Climate station data used in the surface interpolation modeling Precipitation and temperature O Precipitation only o Temperature only 200 km buffer around BC 0 100 200 KM 400 96 Table C I . Number of climate stations used in the surface interpolation models. Prov = province or state. Temp = temperature. Prec = precipitation. Province Prov Total January July Annual Prec Temp Prec Temp Prec British Columbia 412 309 398 310 391 301 Alberta 55 47 55 47 55 55 Yukon Territory 9 9 9 9 9 9 Northwest Territories 2 2 2 2 2 2 Alaska 17 15 17 15 17 17 Washington 97 89 97 89 97 97 Idaho 11 11 11 11 11 11 Montana 31 30 31 30 31 31 Total 634 512 620 513 613 523 97 Appendix D. High Resolution Climate Modeling D1.0 Introduction Continuous surface interpolation of climate station data was performed to create high resolution (600 m cell size) temperature and precipitation data layers to be used for the E N M of C. gattii because high resolution climate data in spatial format are not readily available for B C . The lack of high resolution climate data in spatial format is largely due to the inherent difficulty in modeling B C ' s temperature and precipitation because of the influence of both oceanic and continental weather systems and the complex topography of the province (R. Dunkley, 2004, personal communication). P R I S M (Daly etal. , 2001) is an "expert" climate modeling system with complete geographic coverage for B C but its spatial resolution of 2.5 arc minute (approximately 4 km) did not match the high resolution (600 m cell size) of the other environmental data layers used for the E N M of C. gattii. High resolution precipitation modeling was attempted, but unfortunately it was subsequently abandoned due to large measures of error in the resulting precipitation models. Surface interpolation enables prediction of unmeasured values at any location within the study area based on the observed values (climate stations) of surrounding locations. Several different surface interpolation methods are available. Inverse distance weighted (IDW) and kriging are the more popular methods used to model natural phenomena such as climate (Babish, 2006; Johnston et al., 2001). Regression based methods for modeling climate are also commonly used but they require a high degree of expertise (Brown and Comrie, 2002; 98 Daly et al., 2002; Ninyerola et a l , 2000). Therefore, I D W and kriging (cokriging specifically) methods were used in this study. D2.0 Data and Methods Climate station data used for the surface interpolation modeling were previously described in Appendix C . ArcGIS Geostatistical Analyst was used to interpolate continuous surface layers of January and July average, minimum and maximum temperature using I D W . Surface interpolation of annual, January and July total precipitation using cokriging was also attempted; however, the model performed poorly (large measure of error). Instead, the 4 k m P R I S M precipitation datasets were re-sampled to 600 m cell size. D2.1 Temperature Modeling Temperature values for each climate station were adjusted by the standard lapse rate (decrease o f 6.5°C for every 1 k m gain in altitude) to account for the effect of elevation prior to the surface interpolation. The following function was used: [ T E M P A D J U S T ] = [ E L E V ] * 6.5 /1000 where [ E L E V ] is the climate station elevation (based on the G T O P O 3 0 digital elevation model) and [ T E M P A D J U S T ] is the temperature correction factor to be applied to the average, minimum and maximum temperature values. 99 Next, the structure (frequency distribution) of the temperature data was explored using histograms, QQ-plots and descriptive statistics. Continuous surface interpolation techniques, kriging in particular, performs best when the input data are normally distributed. Since the 1/2 2 data distributions were all non-normal, log (ln(x)), square root (x ) and squared (x ) transformations were performed in an attempt to yield a normal distribution; the January temperature data were converted to Ke lv in units to allow transformation of negative data values. Unfortunately, the data remain skewed. Therefore, I D W was selected as the method for surface interpolation of temperature because I D W is not as sensitive to data skewness as compared to kriging which requires a normal distribution for best results. A minimum of 16 neighbouring points (at least two from each of the cardinal and ordinal directions) within 200 k m radius were used for the I D W surface interpolation. This distance reflects the influence of continentality and latitude on temperature (Tachiiri et al., 2006). Cross validation and validation procedures were performed to evaluate the accuracy of the models as the interpolated surfaces were created. The interpolated temperature surfaces were converted back to the real-world (elevation corrected) values using the following function: [ T E M P E R A T U R E ] = [ I D W S U R F A C E ] - ( [DEM] * 6.5 / 1000 ) where [ I D W S U R F A C E ] is the I D W temperature surface, [DEM] is the GTOPO30 digital elevation model and [ T E M P E R A T U R E ] is the resulting temperature data layer in °C. Figures D l , D2 , D3 , D4, D9, D14, D19, D20, D21, D22, D27 and D32 describe the frequency distributions of January and July average, minimum and maximum temperature using histograms, QQ-plots and descriptive statistics. The I D W interpolation properties and 100 validation results of the models are described in Figures D5 , D6, D10, D 1 1 , D 1 5 , D 1 6 , D23, D24, D28, D29, D33 and D34. The interpolated surfaces and finalized, elevation corrected temperature datasets are illustrated in Figures D7, D8, D12, D13, D17, D18, D25, D26, D30, D 3 1 , D 3 5 a n d D 3 6 . 101 u o i >-CD a. £ u u 60 03 ( H U | —. U ca u o CQ o IS, a a -a o -+-» • i—i X 1 I 2C J I i | s I CD 03 s-<u a. S CD CU e CD 1—11 -a u S3 O 60 o s I a a i s ea — O c/i X Q g bp • IN go m * LT) tr> iri uXIIui w _______ [ s i l l Kiri iri uiTS-] 1— * 102 Figure D3. Histogram (A) and QQ-plot (B) of the square root transformed adjusted January average temperature (K). Tip: Click 01 drag over bars to select Add to Layout Bars: [20 ^ 15 Statistics / Transformation Transformation: I E*ametei: :'6.5 / Data Source Layer. _ |adLian_512 Count Mm Max : 512 : 29.534 : 31.472 : 31,952 S kevmess : -1.7262 Kurlnsit : 7.35 Is * Quart** r 30.942 31.1 31.245 3 05 Data-10"' •*! JJANAVEKELV B -3.1 -2 48 -1.24 -062 0 0.62 Standard Normal Value Tip: Dick or drag over points to select Add to Layout • Transformation Transforrnation / Data Sot Layer: j£j Earameter [&5~ ~3 JJANAVEKELV 1.86 2.48 •3 Figure D4. Histogram (A) and QQ-plot (B) of the squared transformed adjusted January average temperature (K). Frequency A 30.9 31.74 32.59 33.42 34.26 Tip: Pick or drag over bats to select Add to Layout Bars: J20 1] tv Statistics / Tran starvation Transformation: / Data Source Layer: 512 30909 3922? 37310 1231.3 Skewness Kuttosit ] 1 sl -1.6009 G 6969 36801 37510 138170 3534 36.76 37.62 38.46 B fit* . -0.62 0 0.62 Standard Normal Value 1.86 2.48 Tip: Click or drag over points to select Add to Layout / Transformation Transformation. | / Data Source Layer n*] Parameter [2 JJANAVEKELV -Figure D5. I D W interpolation properties of the adjusted January average temperature (°C). Geostatistical Wizard IDW Interpolation: Step 1 of 2 Set Parameter- Geoslatistical Wizard IDW Interpolation: Step 2 of 2 Cti Optimize Power Value Power: 22674 <Back Ne*t > | Finish j Cancel Chart Predated Error j ,. 0.71 £ 0.18 j -0.34 1 -D 8 B £ -1 39 -1.91 -9 44 | • B S C ] • ^ * r-1 -2 44 -1.91 -1 39 -0.86 -0.34 0.18 0 71 Measued. 10-1 Regiession (unction: 0.845" K + 0.177 Prediction Errors Mean-Root-Mean-Square: Samples: 512of 512 0.05577 1.617 Included | x Iy | Measured | Predicted Ves 231660 1689000 -33 •8.3007 Yes 301790 1884700 -16.6 •10378 Yes 376650 1516400 •0.3 -33896 Yes 402100 1830000 •10.8 -13388 Yes 413140 1650400 •8.3 -6.9 Yes 413140 1650400 •6.9 •8.3 Yes 424840 1536300 •2.2 •3.4035 v < Ml Stave Doss Validation. Figure D6. Validation of the adjusted January average temperature (°C) I D W interpolation. d IDW Interpolation: Step 1 of 3 Set Paramete - IDW Interpolation: Step 3 of 3 - Validation Optimize Power Value Power: 1.2087 Symbol Size: J3 Method: [Neighborhood " 3 MMV4 Review type: | Neighbors -Neighbors to Include: Iv Include at Least: Shape Type: Shape Angle: Major Semiaxis: Minor Semiaxis: Anisotropy Factor: Test Location X: [1174868.6 Neighbors 28 Estimated = 3.5E o|©|®|F (200000 200000 Y: 451496.78 0 18 0.71 MeasiRed. 10-1 Regiession function 0.674" K 10.323 Predicton tiiors Mean: 0.2125 Roor-Mean-9quare: 23 Samples: 409 ol 409 Included | x Iy Measured Predicted ~ Yes 231660 1683000 -3.3 •6.8785 Yes 301790 1684700 •166 •9.5521 Yes 375850 1516400 -0.3 •3.7664 Yes 413140 1650400 -83 -6.9 Yes 462000 1623000 •35 33884 Yes 47679D 1867500 -164 -94319 Yes 495320 1519500 -29 -44066 I *, . . . m Save Validation. Figure D7. Resulting I D W surface (A) and map (B) of January average temperature (°C). Legend UI w < h I j jn i a/e_stirf Prediction Map [adjJan_512].[ADJJAHAVE] Filled Contours •24.4000DD --13.4651 DO •13.485100 --6 J866376 -6J866376 --2.884340 -2J884340 - -0.481361 M M -0.481361 -0.968732 H 0J968 732- 1.843798 1J843 798 - 2.371862 2 371 862 • 3.246926 | B 3246926 - 4.697019 ^ • 4697019-7.100000 150 J I 600 I * * • • * ** • $* v * • • # * • • 3 * A • i / r B S Legend id wall_adj jan a/e_surf Pre<liction Map [adj jan_512].[AD J JAM AVE ] Filled Contours -24.400000 --13.465100 •13.465100 --6 J866376 -6J866376 --2.884340 -2884340- -0.481361 • 1 -0.481361 -0.968732 0368732- 1.843798 1J843798- 2.371862 2371 862- 3.246926 H 3 246926- 4.697019 ^ • 4697019- 7.100000 150 300 l _ I in 105 Figure D8. Elevation corrected January average temperature (°C). 107 Figure D l 1. Validation of the adjusted January maximum temperature (°C) I D W interpolation. Geostatistical Wizard - IDW Interpolation: Step 1 of 3 - Set Parameters D Optimize Power Value ] Power: jl.5672 Symbol Size |3 ±j Method: JNwghbofhood Neighbors to Include: \* Include at Least Shape Type: Shape Angle: Major Semiaxis: Minor Semiaws: Antsotropy Factor: t Location |1174968.8 Neighbors 28 Estimated = 6.2777 Preview type: | Neighbors Geostatistical Wizard - IDW Interpolation: Step 2 of 3 - Cross Validation Chart Predicted j Eror _ 09 S 0.S8 J 0.26 1 "° 0 6 fi -0.37 -0 69 -1 01 , • , . - . , . ****** — /% J^ i • i -1.01 -0 69 -0.37 -0.06 0.26 0.58 0 90 Measiued, 10-1 Regression functjon: 0.621 "x + 1.825 Prediction Errors Mean: R oot-M ean-Square: Samples: 103 of 103 0.03642 2.153 Save Cross Validation.. Included Ix IY Measured Predicted "» Yet 402100 1830000 -6 -4 7943 Yes 413140 1650400 -3.8 -1.6653 Yes 424840 1536X0 0.3 -22994 Yet 457980 1621600 42 •26655 Yes 473400 1643900 -2.3 •1.3973 Yes 613160 931820 5.6 2711 Yes 625610 1704900 -10.1 -1.6021 » < Mi < Back | Next) Finish Cancel | Geostatistical Wizard IDW Interpolation: Step 3 of 3 - Validation 0.51 1.01 Measuted, 10-1 0.688* x + 1.421 Prediction Erors Mean: Pi oot-M ean-Square: Samples: 409 ot 409 Save Validation.. 0.1787 2.045 Included 1 X I Y I Measured | Predicted * Yes 231660 1689000 0.2 -15417 Yes 301790 1884700 •102 •55222 Yes 375850 1516400 1.3 -tt 78628 Yes 413140 1650400 -47 -3.8 Yet 462000 1623000 -13 -04173 Yes 476790 1867500 •10.2 -55799 Yet 495320 1519500 -0.6 -1.3435 •! < t8l Figure D12. Resulting I D W surface (A) and map (B) of January maximum temperature (°C). s Legend id wall_adj jan max_airf Prediction Map [adjjan_512].[AD JJAll MAX] Filled Contours •19.799999 --9.420524 •9.420524--3.156972 •3.156972 -0.622801 0J622801 - 2.903724 2 J9D3 724-4.280162 — 4280162-5.110780 _ _ _ 5.110780 -5JS12021 _ _ _ 5J612D21 - 6.442638 _ _ _ 6.442638 - 7.819075 ^•7J819075- 10.100000 Io0 xo .1 l _ I III 600 J I 1 1 *« • t * * * * *• ... /He, B Legend i < I w.Jll_a I j ja 1111,ix_si II f Prediction Map [adj Jan_512].[AD JJAll MAX] Filled Contours -19.799999 --9.420524 -9.420524--3.156972 -3.156972 -0.622801 0J622801 - 2.903724 _____ 2J903724 - 4.280162 _ _ _ 4280162-5.110780 _ _ _ 5.110780 -5B12021 _ _ _ 5J612021 - 6.442638 _ _ _ 6.442638 - 7.819075 • 1 7819075- 10.100000 300 I in •500 I 109 Figure D13. Elevation corrected January maximum temperature (°C). Figure D14. Histogram (A) and QQ-plot (B) of the adjusted January minimum temperature (°C). Data •10"' Tip: Ckck or drag over bar* to select Add to Layout Bare |20 ^ Siafatict / T ranstorrnation T rarulonnabon | None / Data Source Layer: At_rbute: Data's Quantte-10'' 0.56 -3.1 -2 48 -1.86 -124 -0.62 0 0B2 124 186 248 3.1 Stanford Normal Value Tip: Click or drag over points lo select Add to Layout / T ransfr*matron T ransfomvobon: j None /Data Source Layer Attribute: __U_n_512 "\ ^^^^MmWBmWmWmWmjmWKL ^ Figure D15. I D W interpolation properties of the adjusted January minimum temperature (°C). rd IDW Interpolat ion: Step 1 of 2 Set Parameter ; Geostat ist ical W i z a r d IDW Interpolation: Step I ol 1 Cross Val idat ion Optimize Power Value Power 2.1371 G} Q \5 • Preview type: JNeighrbort p a , . -/«W_____\ >"•*'-" -' t^3BBoP^ "'• Symbol Size: J3 -jrj Method |Nekjhtorhood" Neighbors lo Include: ^ Include at Least Shape Type: Shape ~ 3 |16 _ _ oj©|®)|® Angle: Mapr Semraxis: Minor Semiaas: Anisotiooy Factor Test Location j 200000 X |1174868.B Neighbors 97 Estimated - 1.5926 Predicted Error ] _ 0 47 S -0 09 ?' -0.65 • ______ *_M S -1.21 • t | -1.77 -2.33 ______ • 7 89 I 1 1 -2 89 -2.33 -1 77 -1.21 -0.65 -0.09 0.47 Mea«iii«d. 10- 1 Regression function: 0 826 • x » 0.374 Prediction Errors Mean: Root-Mean-Square: Samples: 512 of 512 Save Doss Validation-. 0.09512 1 912 Included fx Iv Measured Predicted Yes 231660 1689000 _s •12.792 Ye* 301790 1864700 23 •14.905 Yet 375850 1516400 •2.1 61399 Yet 402100 1830000 •15.5 •18.29 Yet 413140 1650400 -11.9 9.9 Yet 413140 16504m •9.9 •11 9 Yet 424840 1536300 •4.7 6.0555 > Figure D16. Validation of the adjusted January minimum temperature (°C) I D W interpolation. Geoslatistical Wizard IDW Interpolation: Step 1 of 3 - Set Par, Optimize Power Value ~j Power: J2.0043 Symbol Size: Method: fr Neighbofs to Include: IV Include at Least: Shape Type: Shape Angle: Major Semi axis: Minor Semiaxis: Antsotropy Factor: Test Location X: |16 |2~ "3 "3 o|__®j[® [oo~^_ 1174868.7 Neighbois E stimated Y: |451496.78 86 1.5748 Preview type | Neighbors < Back j Next> Finish Cancel | Geostatistical Wizard - IDW Interpolation: Step 3 of 3 Validation Chart Predicted j Error _ 0.38 ? -0.01 _ -0.39 -0.78 -1.17 .2 — S. -1 55 -1.94 I L ._**** , rJ i . - • i i | • I I i -1 94 -1 .55 -1 .17 Regression function: 0.780 ' x • -0.283 -0.01 0.38 Meosuied. 10 1 Predicbon Errors Mean: R oot-M ean-S quare: Samples: 103 of 103 Save Validation... 0.113 1.897 Included Ix I Y Measured I Predicted *s Yes 402100 1830000 -15.5 -18.57 Yet 413140 1650400 •as •11.9 Yet 424840 1536300 +7 •5.9163 Yes 457980 1621600 -5.7 -5.8121 Yet 473400 1643900 7.4 -8.5391 Yes El 3160 931820 0.7 1.256 Yes E25610 1704900 -19.4 -17.196 m • H i Geostatistical Wi, Chart Predicted I Error IDW Interpolation: Step ? of 3 - Cross Validation 0.09 0.47 Me.isuied. 10 1 R egression function: 0.834 "x* -0.423 Predicbon Errors Mean; Root-Mean-Square: Samples: 409 of 409 0.0691 1.983 Save Cross Validation... Included I x 1 Y Measured Predicted *t Yet 231660 168SO00 •69 •12533 m ' Yes 301790 1884700 -23 •14.952 Yes 375850 1516400 21 •7.9452 Yes 413140 1650400 •11.9 -9.62 Yes 462000 1623000 •5.7 •12076 Yes 476790 1867500 225 -17.941 Yes 495320 1519500 •5 2 -6.3697 mi M < Back | Next> M l Cancel Figure D17. Resulting I D W surface (A) and map (B) of January minimum temperature (°C). 7*5 Legend i< I w , I j jaii 11 i i i_s< i if Prediction Map [ad]Jan_S12].[AD J JAM Mill ] Filled Contours -28000000 --18388731 -18388731 --11.657312 •11.657312 • -7346508 ___ -7346508 - -4.585868 S_ i -4585868 - -2.817950 __H-2317950 --1.68577 7 __|-1J68 5777 --0.96073 2 ____-0960732 • 0.171441 ___ 0.171441- 1.939359 __• 1S39 359- 4.700000 150 300 I In 600 •» • . • * * __ *$ * <* v • • • * , • • _ _ _ _ _ _ _ * • J T . • • • • • «* V* ••>.•> :•. •••• • * B Legend idwall_acljjaiiniin_sinf Prediction Map [adj Jan_512].[AD JJAll Mil ] Filled Contours -28fl00000 - -18388731 •18388731 --11.657312 •11.657312 - -7346508 -7346508--4.585868 -4585868--2.817950 -2317950 --1.685777 -158 5777 - -0.960732 -0960732 - 0.171441 0.171441 - 1.939359 1939359- 4.700000 ISC J I X O I in I I i _ aw 113 Figure D19. Histogram (A) and QQ-plot (B) of the adjusted July average temperature (°C). Data's Quantfe 275 B -0 62 0 0 62 Standard Normal Value Tp: Clck or drag over points to select Add to Layout / T raraformatjon Trarrsfcfmatiori [None /Data Source Layer: ^3 2.48 3.1 Figure D20. Histogram (A) and QQ-plot (B) of the log transformed adjusted July average temperature (°C). 3.07 3.16 325 334 Bars: |2t3 _ _ M Statutes / Trarisformation T rans tor mat ion: {_|___________| ^  / Data Source Layer: Atnibute ~ _ _ |ADJJULAVE Normal QQPlot am Data's Quantife'ID B 31.54 ^^^^ p O ° 0 29.76 2798 26.2 •3.1 -248 -1 86 -1 24 -0 62 0 0 62 Standard Normal Value 124 1 86 2 48 3 1 Tip: Cfck or drag over points to select Add t 3 Layout f uansrormarion T ransf or mat ion rj___j ••rid / Data Source Layer: Attribute: adLML51 3 __j JADJJULAVE Figure D21. Histogram (A) and QQ-plot (B) of the square root transformed adjusted July average temperature (°C). rXurlom 1 sl Qua. tile Median 3rd Quartlre -0.40725 2.401 6.4143 47 62 51 56 55.3 59.04 62.78 Tip: rjfck or drag over bats to sctoct Add to Layout Bars: [20 3 P Statistic* / Transformation Tramformation / Data Source Layer |adjju1_513 66 52 70J26 Data'10 wi Parameter Q5 "3 JAOJJULAVE Data's Quar.tJe-10 85 62 B -3 1 -2 46 -1.24 -0.62 Standard NoffrofVakK Tip: Click or drag ovei points to select Add to Layout / T ransf ormatron Ttamlrxmaton: j /Data Source Layer |Tj Eararneter: |0 5 !ad_ju_513 JADJJUUVE "3 Figure D22. Histogram (A) and QQ-plot (B) of the squared transformed adjusted July average temperature (°C). Count 513 Skewnest 0.023305 Win M a i Mean Std Dev 65.625 377.63 207.24 65.464 Kurtoti* 1 st Quartile Median 3 r d Quartile 2.0296 156.15 203.52 264 2 23 2.55 Data'10"* Tip: Cick or drag over bars to select Bar* |_o" ^ P Statistics / Transformation Transformation /Data Source Layer: T] parameter [_T "3] |ADJJULAVE Data's Quantrle'10 384 -3.1 -2.46 B | --0.62 0 0 62 Standard NormalValue Tip: Cbck or drag over points to select Add to Layout / Transformation Trarriformation: [0EIE!l__B. * Parameter 2 / Data Source Layer •. ~3 JADJJULAVE "3 I Figure D23. I D W interpolation properties of the adjusted July average temperature (°C). Geostatistical Wizard IDW Interpolation: Step 1 of 2 • Set Paramete • phmtze Power Value I Power: J2.794 B M 3 J Preview type: JNe_trbot« Method: | Neighborhood Neighbors to Indude: V Include at Leas! Shape Type: Shape Angle: Major Semiaxis Minor Semiaws: Anisotropy Factor: . 3 o|e|®||i§ X |1174868.8 Y: |451496.78 Neighbors 97 Estimated = 17.704 <Back Ne*t> j Finish j Cancel | d IDW Interpolation: Step 2 of 2 Cn - 2.75 . 2.48 t 2.22 1 1 9 5 £ 1.68 1.42 1.15 1.15 1.42 1.6 Regresson tuncbon: 0.932" x • 1.463 i I i-^.i "*• I - - ' • — r w. I 2.46 2.75 MeasnerL 10-1 Reaction Errors Mean: Rcot-MearvSauare: Samples: 513o(513 0. 03408 1. D41 ! Measured I Predicted Y e i Yes Y e . Yes Yes Yes Yes < 231660 301790 375950 402100 413140 413140 424840 1683000 12.1 16.224 1884700 182 17566 1516400 11.9 13.899 1830000 187 18093 1650400 157 15.9 1650400 159 15.7 153S300 13.3 13.B34 Save Cross Validation.. Figure D24. Validation of the adjusted July average temperature (°C) I D W interpolation. rd - IDW Interpolation: Step 1 of 3 - Set Paramete Optimize Power Value | Power: J26455 5 Symbol See: j3 ^ | Method: Ne Neighbors to Include: tv Include at Least: Shape Type: Shape Angle: Major Semiaxis: Minor Semiaxis: Arrtsotiopy Factor: T est Loceirjion X: |1174868.0 Neighbors 85 Estimated = 17 6 3 . 13 oje)®ll_. b.o ; : Geostatistical Wizard IDW Interpolation: Step 3 of 3 - Validation Chart Predicted | Error | C 2.62 r . 2.38. | 2.14 . _ 191 _ 167 . 1.43 . 1.19 E i • • . i • - I i 119 Regression function: 2.38 2.62 Measiaed. 10-1 Prediction Errors Mean: R cot -M earv-S quare: Samples: 103 of 103 Save Validation. 0.3544 1.161 Included Ix IY 1 Measured ] Predicted Yes 375850 1516400 11.9 15.459 Yes 402100 1830000 18.7 18.104 Yes 424B40 1536300 13.3 15.329 Yes 457980 1621600 166 15.501 Yes 498740 1515400 138 14.602 Yes 526420 1508300 122 14.853 Yes 538040 1650800 17.5 17.312 V _______ > Figure D25. Resulting I D W surface (A) and map (B) of July average temperature (°C). Legend idw<ill_adjjiilave_siirf Prediction Map [adjJul_513. [ADJJULAVE] Filled Contours 11.500000 - 14221989 14.221969 - 16 368105 16368105 - 18 J.60226 18060226 - 19 394381 • 1 19394381 -20.446293 M l 20.446293 - 21 275675 21275675 -22 327587 • i 22327587 -23 661741 | B 23 681741 - 25 3538 63 _ _ • 25 353863 - 27 5000 00 ISO 300 I In L_ 600 B 3 ? Legend idw.ill_.ixliiiil«we_suif Prediction Map [adj Jul_513. [AD J JU LAVE ] Filled Contours 11.500000- 14221969 14221969-16 368105 16368105 - 18 060226 18060226-19 394381 19394381 -20.446293 tgf, 20.448293-21 275675 fgm 21275675 -22 327587 • I 22327587 -23 661741 K 23861741 - 25 353863 ^ • 25353863 -27 .00000 ISO I 300 L_ lln L_ 800 I 118 Figure D27. Histogram (A) and QQ-plot (B) o f the adjusted July maximum temperature (°C). Figure D28. I D W interpolation properties o f the adjusted July maximum temperature (°C). 31 Geostatistical Wizard - IDW Interpolation: Step 1 of 2 • Set Parameter; Optimize Power Value Power: 3.1202 -!- <3 G} \} A Sunbol Site pp_ H __j .5* H " 3 1200000 Neighbors to Include: J1E tv Include at Least Shape Type: Shape Angle: Major Semianis: Minor Sen-axis: Amsotropy Factor: Test Location X: 1174868 Neighbors Estimated |2 _| o | e j ® | | t » |0.0 _) (200000 1 97 227 < Back No»t> 1 Finish | Cancel | Geostatistical Wizard IDW Interpolation: Step 2 of 2 - Cr _ 3 58 S 3.22 | 2.86 tZ 2 . 1 3 1.77 1.41 i kjrVSf^• i •• • I s i 1 41 1.77 2 1 3 2.49 2 . 8 6 3 .22 3 .58 Me . I S I Med. 10-1 Recession rurKAon: 0921 -K + Z 1 D 3 Prediction Erors Mean: Root-Mean-Square: Samples: 513o*513 •0.01333 1.37 Included | x IY i Measured I Predicted A Yet 231660 1689000 157 21.759 Yet 301790 1884700 24.7 23684 Yet 375850 1516400 14.1 18.594 Yet 402100 1830000 24.7 24.999 Yet 413140 1650400 21.5 21.9 Yet 413140 1650400 21.9 21.5 Yet 424840 1536300 182 17.261 I < 111 • Save Doss Validation. Figure D29. Validation of the adjusted July maximum temperature (°C) I D W interpolation. Geostatistical Wizard - IDW Interpolation: Step 1 of 3 - Set Parameters 0 ptimize Power Value Power 12.8661 Symbol Size: |3 - | J Method: |Neighborhood J: ••••• „' Preview type: | Neighbors Neighbors to Include: KV Include at Least Shape Type: Shape Angle: Major Semiaxis: Minor Semiaxis: Amsohopy Factor |1B o | e | ® | ( ® |200000 Neighbors Estimated 1174888 8 85 - 22.695 Y: 1451496.78 (Back Next > FWsh Cancel Geostatistical Wizard - IDW Interpolation: Step 2 of 3 - Cross Validation Chart Predicted ] Eror _ 3.58 £ 3.23 Y 2.87 2 52 2 17 1.81 1 46 1.46 Regression function: I I ^ T P -r* •- j i i i 1 81 0.906-2.17 .2515 323 358 Me.ismed. 10-1 Prediction Erors Mean: R oot-M ean-S quare: Samples: 410 of 410 0.007748 1.418 Included Ix Iv Measure i 1 Predcted •» Yet 231660 1689000 15.7 22 88 Yes 301790 1884700 24.7 22764 Yes 413140 1650400 21.5 21.9 Yes 413140 1650400 21.9 21.5 Yet 462000 1623000 19.6 20506 Yes 473400 1643900 20.2 20 018 Yes 476790 1867500 25.7 25.597 » < all Save Cioss Vaidation I Wi/,ti rl IDW Interpolation: Step 3 of 3 Valid-tic Chart Predicted j Eror | _. 343 3 09 2 76 2 42 2 08 1 75 1.41 1.41 „ I _ —t"*—. 2 08 3.09 3.43 MeastNed. 10-1 Regression function: 0.895 " x + 2.895 Prediction Erors Mean: Root-Mean-S quare: Samples: 103 of 103 0.2765 1.56 Included Ix Iv Measured Piedcled " Yet 375850 1516400 14.1 20 407 Yes 402100 1830000 24 7 24 963 Yes 424840 1536300 182 20.069 Yet 457980 1621600 20.7 19.607 Yet 498740 1515400 17 9 19.001 Yet 526420 1508300 16.2 19.262 Yet 538040 1650800 24 3 22.803 HI Mi—1 Mi Figure D30. Resulting I D W surface (A) and map (B) of July maximum temperature (°C). Legend idwall_adjjiilmax_siirf Prediction Map [adj Jul_5131 [AD JJU LMAX ] Filled Contours 14.100000-17508094 17B08094-20 J910732 23506775 25J678940 | 27.49 6435 | 291)17172 | 30289604 | 31810339 I33J627838 20J910732 23 506775 25 J678940 27.496435 29D17172 30 289604 31J810339 33 J627838 35.799999 150 300 L_ aso i B Legend idwall_adjjiilmax_siirf Prediction Ma|> [adj Jul_51 JJ. [AD JJU LMAX ] Filled Contours 14.100000 -17 J808094 17SD8094-20910732 23506775 25J678940 | 27.496435 | 29017172 | 30289604 | 31 B10339 I33J627838 201910732 23 506775 25 5789 40 27 .496435 29 £117172 30 289604 31 810339 33 J627838 35 .799999 0 150 X0 Un 600 I I I I I I 1 1 1 122 Figure D31. Elevation corrected July maximum temperature (°C). Figure D32. Histogram (A) and QQ-plot (B) of the adjusted July minimum temperature (°C). Count 513 Skewnen a32543 Mm Has Mean Std. Dev. 8.1 198 13 487 2.21 E Kurtosts 1 St Quaihlc Medun 3 id QiMMhte 2.5578 11.8 13.3 15 Date-10"1 Tip; Cick or drag over bars to select Add to Layout Bars: [20 __ W Statistics / Transformation Transform at en JNone / Data Source Ion ,ii qnpiot Data's QiMntte'tO'1 1 77 B 1 53 1 29 1 OS >^ riRic -3.1 -2.48 -1 86 -124 -0.62 0 062 1 24 1.86 Standard Normal Value 248 3.1 Tip Click or drag over points to select Add to Layout / i ransronnBoon Trar^ rxTTiatron: | None __J s Data Source Layer: AMrbute |ad_jul_513 Figure D33. I D W interpolation properties of the adjusted July minimum temperature (°C). rd IDW Interpolation: Slep 1 of 2 Set Parai Dptrnize Power Value - : - <•}<=} ^ # Pow 11.8392 Symbol S Preview type: |Nergnbc« " Method: Nagtoorhood Ne_hbors tolridude. fl? v Include at Least: Shape Type: Shape Angle: Major Semaws: Minor Semiaws: Antsotropy Factor: Test Lccation " 3 "3 |j a Oj©|®|[® 200000 1200000 X: |1174868.8 Neighbors Estimated Y: 451496 78 ~3 IDW Interpolation: Step 2 of 2 Cr< r. 1 £ 1.78 , 1.59 | 1 39 £ 1.2 1.01 0.81 0.81 Regression function; i • 1 i • _ • * " i J M *• *i — v 1.01 0 708-1 20 .3.944 1.39 178 198 Measmed. 10-1 Piediction Eirots Mean Root-Mean-Squere: Samples: 51301 513 Save Cioss Validation 0.06078 1.306 Included |x IY Measured I Predicted 1 Yes 231660 1689000 B5 10 763 Yes 301790 1884700 11.7 11.201 Yes 375650 1516400 9.7 9.5328 Yes 402100 1830000 127 11.169 Yes 413140 1650400 3.5 104 Yes 413140 1650400 104 85 Yes 424840 153S3O0 84 10202 V • • m tktti < Back j 1 Finish Figure D34. Validation of the adjusted July minimum temperature (°C) IDW interpolation. Geostatistical Wizard IDW interpolation: Step 1 of 3 - Set Parameters mm Optimize Power Value j -!- O * Power: 1.7895 Symbol Size: J3 ±j Method: j Neighborhood 3 Neighbors to Include: |1E IV Include at Least: [2 jFj Shape Type: Q | (&\ <g)|[jj Shape Angle: Major Semiaxis: Minor Semiaxis: Anisotropy Factor: 1 Test Location X: {1174868.8 Y: J451496.78 Neighbors 85 Estimated = 12.426 Preview type: JNerc^ bors ~3 Geostatistical Wizard - IDW Interpolation: Step 2 of 3 - Cn Chart Predicted 1 Error Prediction Errors Mean: R oot-M ean-S quare: Samples: 410 of 410 0.006486 1.312 I E j Measured I Predicted Yes Yes Yes Yes Yes Yes Yes • • 231660 301790 413140 413140 462000 473400 476790 1689000 1884700 1650400 1650400 1623000 1643900 1867500 B.5 11.7 9.5 104 11.4 10.5 11.1 1089 10221 104 9.5 10557 11.152 11.61 Saw Cross Validation... Geostatistical Wizard IDW Interpolation: Step 3 of 3 Validation Chart Predicted ! Error Regression function: 0.673 • x + 4.609 Prediction Errors Mean: 0.4124 Root-Mean-Square: 1.457 Included Ix I Y : Measured Predicted M Yes 375850 1516400 97 10.573 Yes 402100 1830000 12.7 11.213 Samples: 103oM03 Yes 424840 1536300 8.4 10597 Yes 457980 1621600 125 11.343 Yes 498740 1515400 9.6 10222 Yes 526420 1508300 8.1 10.621 Yes 538040 1650B0O 10.7 11.583 m I M > Save Validation. < Back J 1 Finish Cancel Figure D35. Resulting I D W surface (A) and map (B) of July minimum temperature (°C). Legend idw<ill_<vljj(iliniii_stirf Prediction Map [adj Jul_51 J J. [AD JJU LMIH] Filled Contours 8.1 DO ODD- 9.708084 9.798084-11.022768 11.906027 12543047 13 DO 2475 | 13J639494 | 14522754 | 15.747437 I 17.445520 111)22788 - 11.906027 - 12 5 43047 - 13D02475 - 13J639494 -14522754 -15.747437 -17.445520 - 19.799999 130 I 300 L_ I In l _ EDO I o o B Legend i<lwall_acijjMlmin_surf Prediction Map [adj jul_5131 [AD JJU LMIH] Filled Contours 8.100000- 9.798084 9.798084-11.022768 11.906027 12543047 | 13 DO 2475 | 13J63 9494 | 14522754 I 15.747437 111)22768 - 11.906027 - 12 5 43047 -13 D02475 -13 539494 -14522754 -15 .747437 -17.445520 17.445520 - 19.799999 500 _l_ I In l_ SM 126 Figure D36. Elevation corrected July minimum temperature (°C). D2.2 Precipitation Modeling Surface interpolation of annual, January and July precipitation was performed by cokriging of precipitation and elevation values from the climate station dataset. The effect of elevation has the greatest influence on precipitation - precipitation increases as altitude increases (Brown and Comrie, 2002; Daly et al., 2002; Ninyerola et al., 2001). Figures D37, D38, D39, D40, D41, D42, D43, D44 and D45 illustrate the cokriging modeling process and properties. Unfortunately, the climate station precipitation and elevation data distributions were non-normal and large measures of error resulted in the models. A s a result, high resolution modeling of precipitation was subsequently abandoned. Instead, the 2.5 arc minute (approximately 4 k m spatial resolution) P R I S M precipitation datasets were used. ArcGIS Spatial Analyst was used to re-sample the P R I S M datasets to 600 m cell size raster layers. A 7 x 7 cell low pass filter was then applied to smooth the 600 m cell size data values, thereby removing the abrupt data values at the cell boundaries of the original P R I S M datasets. Re-sampling, or downscaling, of climate data in raster data format for E N M has been described by Parra-Olea et al. (2005). Figure D46 illustrates the comparison of the original and re-sampled P R I S M datasets. Figures D47, D48 and D49 illustrate the 600 m cell size re-sampled P R I S M raster layers for annual, January and July precipitation. 128 Figure D37. Histogram (A) and QQ-plot (B) of total January precipitation. [Ftequency'lO"' 1.18 1 75 Tip: Click or drag over bars to select 2.32 Add to Layout Bore: [20 j § j P Statistics / T ranslormation Transformation: .None / Data Source Layer: Count : 609 S kewness 1.1671 Mm : 9.6 tCurtosis 3.7927 Max : 571.3 1-*t Quartile 39.625 Mean : 136.48 Median 98 Std. Dev. : 118.68 3-id Quartile 203.9 3.46 4.03 T 3 I Data's Quantle-10"' B j i -3.1 S -2.52 Tip: Click or drag over ports to . / Trensfomralion -1.26 Add to Layout -0.63 0 0.B3 Standard Normal Value Transformation: JNone / Data Source Layer: 2 5 2 3.16 SO Figure D38. Histogram (A) and QQ-plot (B) of the log transformed total January precipitation. Count : 609 Min , 2 2618 \ 6.3479 I 4.5059 Std. Dev. I 0.95214 Skewnets Kurt otit 1 it Quartile Median 3 rdQuaitile Tip: Click or drag over bars to select | Bars: [20 ^ jv Statistics / T i ans formation s Data Source I Ldve> 3 6795 4.585 5.3176 ~3 JJANTDTPREC r>ormal QQPlot O S Data s Quantile 5 54 B I S 4.72 3 9 LjS^i 3 OS ocrf 1 7f£ -3.15 - 2 5 2 •1 B9 - 1 3 6 -0.63 0 0JS3 Standard Normal Value 1 26 1.89 2.52 3.15 Tp: Click or drag over points to select Add to Layout / i ransiormation T rans f orma (i on: [IBH M i d / Data Source Layer Attribute: | jan_pr ec_G13 j ^ J JJANTOTPREC Figure D39. Histogram (A) and QQ-plot (B) of the square root transformed total January precipitation. Count - EOS S k e w n e s s 0 .53177 M i n : 4 1368 Kurt ox i i 2 .255 M a , : 45 .B04 1 s t Qua i t i l e 1 0 . 5 9 M e a n : 1 9 . 2 3 M e d i a n 17 .799 Slot D e v . : 3 . 7656 3-fd Quai t i le 26 .559 0.38 1.24 2.53 2.96 Data •TO"1 Tip: Click or drag over bars to select Bare J2D 3 & Statistics / T \ansformation Transformation: [J^ ffl / Data Source Layer: T] Parameter: jb.5 ~3 I J A N T O T P R E C Normal QQPIor Data's Quantrle'10"1 B 3.74 2.89 2.04 Q o 0 0 occar*ff ln.- -n ° ° . .._: -3.15 -252 -1 89 -1.26 -063 0 0.63 Stand ardNormalVarue 1 26 1.89 2.52 3.1 S Tip. CJck or drag over points to select Add to Layout / T rensformation Transformation; ^S3JEBB_ZLI Ea'ameter: W$ / Data Source Layer; Attribute: |jan_prec_E13 JANT0TPREC Figure D40. Histogram (A) and QQ-plot (B) of climate station elevation. Count M i n 2 1 7 5 516 08 471.51 S k e w n e u : 0.69061 Kur tos i t : 2 .B617 1 s t Qua i t i l e : 6 0 M e d i a n : 4 4 3 3- id Quartrle : 858 .5 Bars: J20 ]|] S" Statistics / Transformation Transtorrnation: [None 3 / Data Source Layer: ] jon_prec_613 Normal QQPlot b I Data's 22.16 17.73 QuanWe>10"a B (ft 13.3 4.44 -315 -2.52 -1.89 -126 -0.63 0 0.63 Standard Normal Value 126 1.89 2.52 3. 5 Tip: Click or drag over points to select Add to Layout / T lansrormatron / Data b ounce Layer: Attribute: j ian_prec_613 ( D E M E L E V E Figure D41. Histogram (A) and QQ-plot (B) of the log transformed climate station elevation. B \ 0 a^^^mmmm^^a ——I -32133 -257.06 -192 79 -128.52 -64.25 0.02 64.29 1 2856 192.83 2571 321 37 Standard Normal Varus • 10! Tip: Crick or drag over points to select Add to Layout / Transformation Transformation ^SSIHBHrl * • / Data Source Layer Attribute: • ! IDEMELEVE Figure D42. Histogram (A) and QQ-plot (B) of the square root transformed climate station elevation. : 809 Skewneu -0.015644 0 Kurtosis 1 7328 : 91.274 1st Quart lie 13.492 : 36 831 Median 40.095 : 23.609 3 id Quaitile 56.6 T ip: Click or drag over bars to select Add to Layout Ban; [20 HO ! * Statistics / Transfomiation Transformation: il^ESSBH Pa'ametei: !0.5 / Data Source Layer: Data's Quantie'10 915 B n i> in ' 1 -3.15 -1.26 -0.63 0 0.63 Standard Normal Value Tip: Cick or drag over points to select Add to Layout / T ransformation Transformation: j Bon-Cox - \ Parameter; |u\5 .-"Data Source Layer j D E M E L £ V E Figure D43. Trend analysis of total January precipitation (A) and climate station elevation (B). Trend Analysis 17 Legend Rotation Angles Location: 0° — 3D Graph — Horizontal: 129° Vertical: -29° Rotate: | Graph \1\ n J J M J Perspective: [< Add to Layout LI I? Projected Data 17 Trend on Projections 17 Sticks 17 Axes (7 Input Data Points / Data Source Number of Grid Lines X: [T~±3 Y: [ T I E Grid Line Width: |1 ± | Attribute: | jan_prec_613 • ra r^aV y jKJ | | jANiuTPREr Trend Analysis B B P Legend B Rotation Angles Location: 0° — 3D Graph — Horizontal: 129° Vertical: -29° Rotate: Graph Perspective: | < . ] Ills I Tip. oi drag ov Add to Layout / Graph Options 171 17 Projected Data 17 Trend on Projections 17 Sticks 17 Axes 17 Input Data Points / Data Source Layer: I jan_prec_613 Number of Grid Lines X: |i~~]jf] Y: ( T H Z: Grid Line Width: " 3 Attribute: E 132 Figure D44. Cokriging properties of total January precipitation surface interpolation. Geostatistical Wizard: Step 2 of 5 - Detrending (Standard Options) Dataset Setection: j Dataset 1 Advanced Opbons » pow«: r Neighborhood Ne*> Frah Cancel Geostatistical Wizard: Step 3 of 5 - Sei View Semiverir^iam/Covariarice Surface P Stow Search Direction Sernivariogtam/Covarianees: |V»1 t W I 3 Models V M o d e l l j Model2 ] T Model 3 | Major Range Q T etrespherical Pentasphericel Eicponential Gaussian Rational Quadratic Hots Elfecl K-Bessel J-Bessel Stable |l 44570 P Anisotropy Partial S i 0.19921 ly Nugget B 0 10.074832 T Shifts B l / J -X. [None" Lag 015921 -Spherical! 44570KI 074832-Nuggat Size: 122020 Back | Mem > f~ Error Modeling Y: |None Numboj . , of Lag* fl2 ]0 Finish Cancel Geostatislical Wizard: Step 4 of 5 Searching Neighborhood Dataset Selection: i Dataset 1 Symbol Size: |3 ± • s M '• • * ! * •" • Preview type: | NeigHxis A -Neighbors to Include: 16 P Include at Least |5 Shape Type: r Shape Angle: Major Semiaas Minor Semiaws: Anisotropy Factor: Test Location X: ~3 |1174868.8 Neighbors 103 <Back | Nent> Geostatistical Wizard: Step 5 of 5 Cross Validation Chart Predicted | Error j Standardced Error | QQProt j <vr 5.71 S 4.76 I 3.81 £ 286 | 19 0 95 0 0 Regression function: • •* • • • • • >8rW? •• • fa - • |. 1 0 95 1.90 0.939 •« +10.601 476 5.71 Me«iSile<J. 10 2 Prediction Enors Mean: Root-Mean-Square: Average Standard Error Mean Standardized 0.578 46.29 4.416e39 003255 Root-Mean Square Standairjzed: 0.7077 Samples: 607 ot 609 Save Cross Valdation... Measured | Predicted Yes Yes Yes Yes Yes Yes Yes < 402100 405560 413140 424840 431320 457360 462000 1830000 1689000 1850400 153G300 1533500 1621600 1623000 17 61 176.4 165.6 125.2 138 4 1247 0 13195 10119 115.56 176.31 187.9 118.3 106.96 mi Cancel Figure D45. Resulting cokriging surface (A) and map (B) of total January precipitation. A , i aEj , s Legend coki i <j oil janpi ec_surf Prediction Map Danj>rec_613].[JAHTOTPRECJ Filled Contours 9 £00000 - 23.70822 D 23.708220 -33.585407 33585407 - 47.693626 47 J693626 -67.845291 67845291 -96.629196 • H 96J629196 - 137.743073 • I 137.743073 - 196.468628 196.468628-280350067 2803 50067 - 400.163239 mmW 400.163239 - 571299988 150 B O L _ Mn B Legend cokrigall janprec_siirf Prediction Map Dan_prec_613].[JAN TOTPRE C] Filled Contours 9 BOO 000- 23.708220 23.708220-33 585407 33585407 - 47 693626 47593626-67 845291 67845291 -96 529196 i 96529196 - 137.743073 | 137.743073- 196.468628 | 196.468628 -280350067 | 280350067-400.163239 I 400.163239-571299988 300 I Hi •500 I 134 Figure D46. Comparison of original 4 km cell size P R I S M dataset (A) and 600 m cell size re-sampled dataset (B). 135 Figure D47. Re-sampled (600 m cell size) P R I S M January total precipitation. Figure D48. Re-sampled (600 m cell size) P R I S M July total precipitation. Figure D49. Re-sampled (600 m cell size) P R I S M annual total precipitation. L E G E N D w oc A n n u a l Tota l Prec ip i ta t ion High: 10426.9 mm Low: 226.7 nun I i i i I i i • I 0 (50 120 K M 240 D3.0 Results and Discussion I D W surface interpolation of January and July average, minimum and maximum temperature produced good modeling results. Tables D l and D2 summarize the prediction errors of the models via cross-validation and validation procedures. Cross-validation uses all of the data, both training and testing sub-datasets, to build and evaluate the surface interpolation model by iteratively removing one data location to predict the associated data value (Babish, 2006). Validation on the other hand uses the training sub-dataset to build the model and the testing sub-dataset to evaluate the "appropriateness" of the model decisions. A s a rule of thumb, a model that provides accurate predictions has a mean error close to zero and a root-mean-square error close to one (Johnston et al., 2001). Table D l . Cross-validation prediction errors of the I D W surface interpolation models. IDW Modeled Dataset Mean Error Root-Mean-Squared Error January Average Temperature 0.05577 1.617 January Maximum Temperature 0.01969 1.518 January Min imum Temperature 0.09512 1.912 July Average Temperature 0.03408 1.041 July Maximum Temperature -0.01339 1.370 July Min imum Temperature 0.06078 1.306 139 Table D2. Validation prediction errors of the I D W surface interpolation models. IDW Modeled Dataset Mean Error Root-Mean-Squared Error January Average Temperature 0.2125 2.300 January Maximum Temperature 0.1787 2.045 January Min imum Temperature 0.1130 1.897 July Average Temperature 0.3544 1.161 July Maximum Temperature 0.2765 1.560 July Min imum Temperature 0.4124 1.457 The root-mean-squared error of the January temperature modeling (1.518-1.912 for the cross-validation and 1.897-2.300 for the validation) was greater than that of the July temperature modeling (1.041-1.370 for the cross-validation and 1.161-1.457 for the validation). This may be due to greater temperature variability experienced across the province during January than compared to July. For example, during January the coastal areas of B C are largely above freezing temperatures; whereas, the interior and northern regions experience very cold temperatures. In addition to examining the quantitative measures of model accuracy, the interpolated temperature datasets were visually inspected to check that the models made intuitive sense -Daly et al., (2002) emphasize the importance and value of "human-expert" methods to infer climate patterns. For example, as expected the coastal areas had the mildest temperatures during January and moderate temperatures during July which reflect the moderating oceanic effect, valley regions in the province's southern interior had the hottest temperatures during July, and mountainous areas had the coldest temperatures year round which reflect the effect of altitude. 140 The accuracy of the 600 m cell size re-sampled P R I S M precipitation datasets could not be discerned by quantitative methods because the only readily available precipitation data that could be used to evaluate the model are the same climate station data (1971-2000 Normals from Environment Canada and the National Oceanic and Atmospheric Administration) that were used to build the original P R I S M datasets. Instead, the 600 m cell size re-sampled P R I S M datasets were visually inspected to check that artifactual patterns of the precipitation data were not created as a result of the re-sampling procedure. The 7 x 7 low pass filter successfully removed the abrupt changes in the data values at the cell boundaries of the original P R I S M datasets while preserve the general pattern of the original data. The datasets also made intuitive sense. A s expected, high precipitation values were predicted along the B C coastline and mountainous regions (Coast Mountains and Rocky Mountains), especially on windward locations; whereas, valley areas and leeward locations experienced much less precipitation. A seasonal difference is also seen in the pattern of precipitation in B C : convective precipitation in the B C interior and Peace-Liard plateau during July, but absent in January, as opposed to orographic precipitation along the coast and mountainous regions year round. Micro-climate patterns of temperature and precipitation in particular, do exist in B C due to the complex topography of the province and influence of both oceanic and continental weather systems. The modeling of micro-climate is an active area of research but it is outside the scope of this study. The 600 m cell size temperature and precipitation data layers created for the E N M of C. gattii are appropriate for this study because these datasets provide a high degree of spatial resolution and display distinctive local and regional climatic patterns. 141 Appendix E . Biogeoclimatic Zones of British Columbia Meidinger and Pojar (1991) have created an ecological classification system for B C that group ecologically similar environs based on vegetation, soils and climate. Fourteen biogeoclimatic zones, which are generally named after the geographic region and dominant tree species of the area, have been defined for the province. The ecological characteristics of these 14 zones are briefly described in Table E l . Within each biogeoclimatic zone, subzones and variants may exist based on slight variations of climate and continentality (Figure E l ) . For example, the Coastal Western Hemlock biogeoclimatic zone has a xeric (very dry) maritime subzone with two variants based on temperature and precipitation: C W H x m 2 where C W H is the zone, x m the subzone and 2 the variant. Figure E l . System of naming and coding biogeoclimatic subzones. Source: Meidinger and Pojar, 1991. © Government of British Columbia Ministry of Forests, 1991, by permission. SUBZONE CODE 1st tetter 2nd l e t t e r interior zones temperature regime coastal zones continentality h = hot w = warm m = mild k = cool (kool) c <= cold v = very cold h = hypormaritime m <= maritime s = submaritime 142 Table E l . Characteristics of the biogeoclimatic zones of Brit ish Columbia. Biogeoclimatic Zone Location Climate Elevation Soil Vegetation Coastal Douglas-fir Eastern coast of south Vancouver Island in the rainshadow of the Vancouver Island and Olympic Mountains, Gulf Islands, Sunshine Coast, southwest edge of the Fraser River Delta Warm, dry summers; mild wet winters; monthly average of daily minimum temperatures is above 0°C; 650-1250 mm of annual precipitation with only 5% falling as snow Mostly below 150 meters Brunisols, Humo-Ferric Podzols Coastal Douglas-fir, western red cedar, grand fir, arbutus, Garry Oak, red alder IS « Coastal >, Western * H e m l o c k Western coast of British Columbia from Washington to Alaska border, Vancouver Island, Queen Charlotte Islands Cool summers; mild winters; mean daily temperature of coolest month is 4.7°C in the south and -6.6 °C in the north; 1000-4400 mm of annual precipitation Sea level to 1050 meters Ferro-Humic Podzols, Humo-Ferric Podzols, Brunisols, Folisols Western hemlock, amabilis fir, yellow cedar, Douglas-fir, lodgepole pine, red alder, Sitka spruce, black cottonwood Biogeoclimatic Zone Mountain Hemlock Location Climate Elevation Soil Vegetation Sub-alpine elevations above the Coastal Western Hemlock zone on the Coast Mountains along the west coast of British Columbia, and Insular Mountains of Vancouver Island and the Queen Charlotte Islands Short, cool summers; long, cool, wet winters; average temperature remains below 0°C for 1-5 months, and above 10°C for 1-3 months; mean annual precipitation is 1700-5000 mm with heavy snow cover for several months 400-1000 meters in the north, 900-1800 meters in the south Podzols, Folisols Mountain hemlock, amabilis fir, yellow cedar, western hemlock, western red cedar, Douglas-fir, white pine, Sitka spruce, sub-alpine fir, whitebark pine Okanagan Valley, Similkameen River Valley, Thompson River Valley, Nicola River Valley, middle Fraser Valley, lower Chilcotin River Valley Hot, dry summers; moderately cold winters; precipitation is bimodal with the majority of it falling in December-January and June Valley bottoms up to 1000 meters Brown, Dark Brown, Black and Dark Gray Chernozems Bunchgrasses: wheatgrass, sagebrush, cheatgrass, junegrass, antelope brush Biogeoclimatic Zone Location Climate Elevation Soil Vegetation Ponderosa Pine Low elevations along the very dry valleys of southern British Columbia: Lytton-Lillooet area, lower Thompson and Nicola Rivers, Similkameen River, lower Kettle River, Okanagan Lake, southern Rocky Mountain Trench Hot, dry summers; cool winters with light snowfall; 280-500 mm of annual precipitation 335-900 meters Dark Brown Chernozems, Orthic or Eluviated Eutric Brunisols Ponderosa pine, wheatgrass, interior Douglas-fir, water birch, paper birch, black cottonwood, sagebrush, bluegrass, cheatgrass, bulrushes Interior Douglas-fir Low to mid elevation landscapes of south-central interior British Columbia: southern Interior Plateau, southern Rocky Mountain Trench, leeward side of the Coast Mountains Warm, dry summers; cool winters; mean annual precipitation is 350-1000 mm with 20-50% falling as snow; average temperature is below 0°C for 2-5 months, and above 1 0 ° C f o r 3 - 5 months 350-1450 meters Orthic or Dark Gray Luvisols, Eutric or Dystric Brunisols Interior Douglas-fir, lodgepole pine, Ponderosa pine, white spruce, grand fir, paper birch, western larch, Rocky Mountain juniper, wheatgrass, needlegrass, junegrass Biogeoclimatic Zone Interior Cedar -Hemlock Location Climate Elevation Soil Vegetation Low to middle elevations in southeastern British Columbia: lower Columbia Mountains, windward slopes of the Rocky Mountains, Shuswap and Quesnel Highlands; Nass Basin, Hazelton, Iskut and Stikine Mountains Warm, dry summers; cool, wet winters; mean annual precipitation is 500-1200 mm with 25-50% falling as snow; average temperature is below 0°C for 2-5 months, and above 1 0 ° C f o r 3 - 5 months 100-1000 meters in the south, 400-1500 meters in the north Humo-Ferric or Ferro-Humic Podzols, Brunisolic or Orthic Gray Luvisols, Brunisols, Gleysols Western red cedar, western hemlock, grand | fir, white spruce, Engelmann spruce, sub-alpine fir, Sitka | spruce, Roche spruce, Ponderosa pine, white pine, black cottonwood, lodgepole pine, paper birch Middle elevations in the Fraser Plateau, Southern Interior Plateau, leeward side of the Coast and Cascade Mountains, southern Rocky Mountains and Rocky Mountain Trench Moderately short, warm summers; cold winters; mean annual precipitation is 380-900 mm; average temperature is below 0°C for 5 months, and above 10°Cfor 2-4 months 1150-1700 meters Brunisolic or Othic Gray Luvisols, Eutric Brunisols, Humo-Ferric Podzols, Dystric Brunisols White spruce, sub-alpine fir, black huckleberry, Interior Douglas-fir, Utah honeysuckle, grouseberry, lodgepole pine Biogeoclimatic Zone Location Climate Elevation Soil Vegetation Su b-Boreal Pine-Spruce High plateau areas (in the rainshadow of the Coast Mountains) in the west central interior of British Columbia known as the Chilcotin Cool, dry summers; cold, dry winters; mean annual precipitation is 335-580 mm with 30-50% falling as snow; average temperature is below 0°C for 4-5 months, and above 10°Cfor 1-3 months 850-1300 meters in the north, 1100-1500 meters in the south Brunisolic Gray Luvisols, Orthic Dystric Brunisols, Gleysols Lodgepole pine, white spruce, trembling aspen, interior Douglas-fir, sub-alpine fir, black spruce, black cottonwood Sub-Boreal Spruce Montane zone of central interior British Columbia: Nechako and Fraser plateaus, Fraser basin Relatively warm, moist, short summers; severe, snowy winters; mean annual precip. is 440-900 mm with 25-50% falling as snow; average temp, is below 0°C for 4-5 months, and above 10°C for 2-5 months 1100-1300 meters Podzols, Brunisolic and Orthic Gray Luvisols White spruce, sub-alpine fir, lodgepole pine, trembling aspen, paper birch, interior Douglas-fir, black spruce Biogeoclimatic Zone Location Climate Elevation Soil Vegetation \ ^ Engelmann Lies below the alpine tundra in the Rocky Mountains, leeward side of the Coast, Skeena and Omineca Mountains; highest elevations of the Interior Plateau Cool, short summers; cold, long winters; mean annual precipitation is 400-500 mm in drier areas and up to 2200 mm in wetter areas with 50-70% falling as snow; average temp, is below 0°C for 5-7 months, and above 10°C for 0-2 months 900-1700 meters in the north, 1200-2100 meters in the southwest, 1500-2300 meters in the southeast Humo-Ferric Podzols Engelmann spruce, sub-alpine fir, lodgepole pine, whitebark pine, limber pine, alpine larch N 9kj W P * ^ Great Plains (Alberta Plateau), Nelson Lowland, Peace, Stikine, Liard, Fort Nelson, and Petitot Valleys Short, warm summers; long, very cold winters; mean annual precip. is 330-570 mm; average temp, is below 0°C for 5-7 months, and above 10°C for 2-4 months 230-1300 meters Gray Luvisols, Cumulic Regosols, Chernozems, Brunisols White spruce, trembling aspen, lodgepole pine, black spruce, balsam poplar, tamarack, sub-alpine fir, paper birch Biogeoclimatic Zone Location Middle elevations of the northern Rocky Mountains, northern-most Skeena, Omineca and Cassiar Mountains, Stikine, Yukon and Liard plateaus On high mountains throughout the province Climate Short, cool summers; long, cold winters; mean annual precipitation is 460-700 mm with 35-60% falling as snow; average temperature is above 10°C for 1-3 months Harsh alpine climate that is cold, windy and snowy; mean annual precipitation is 700-3000 mm with 70-80% falling as snow; average temperature is below 0°C for 7-11 months Elevation 1000-1700 meters in the southern portions of this zone, 900-1500 meters in the northern portions Above 2250 meters in the south east, above 1650 meters in the south west, above 1400 meters in the north east, above 1000 meters in the north west Soil Humo-Ferric Podzols, Brunisols Orthic and Humic Regosols, Brunisols, Cryosols Vegetation White spruce, sub-alpine fir, black spruce, lodgepole pine, trembling aspen, Shrubs (arctic willow, grey-leaved willow), | scrub birch, herbs, bryophytes, lichens 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0093069/manifest

Comment

Related Items