Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Employing citizen-science avian data and environmental data for improved species distribution estimates… Rickbeil, Gregory James Melville 2013

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

Item Metadata

Download

Media
24-ubc_2013_fall_rickbeil_gregory.pdf [ 3.01MB ]
Metadata
JSON: 24-1.0074244.json
JSON-LD: 24-1.0074244-ld.json
RDF/XML (Pretty): 24-1.0074244-rdf.xml
RDF/JSON: 24-1.0074244-rdf.json
Turtle: 24-1.0074244-turtle.txt
N-Triples: 24-1.0074244-rdf-ntriples.txt
Original Record: 24-1.0074244-source.json
Full Text
24-1.0074244-fulltext.txt
Citation
24-1.0074244.ris

Full Text

EMPLOYING CITIZEN-SCIENCE AVIAN DATA AND ENVIRONMENTAL DATA FOR IMPROVED SPECIES DISTRIBUTION ESTIMATES AND AVIAN CONSERVATION IN BRITISH COLUMBIA by Gregory James Melville Rickbeil   A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF  THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE in The Faculty of Graduate and Postdoctoral Studies (Forestry)   The University of British Columbia (Vancouver) August 2013 ? Gregory James Melville Rickbeil 2013  ii  Abstract Previous work has shown that many populations of birds residing in British Columbia are declining in number. However, given the size and remoteness of many parts of the province, direct sampling of BC?s bird communities throughout the province is unlikely. Remote sensing has been shown to be an attractive option in these types of situations, providing environmental data in remote areas at spatial scales appropriate for a provincial level of analysis. Additionally, the spatial coverage of remotely sensed data allows for species occurrences to be estimated in un-sampled locations using models derived from areas where sampling has occurred. The overall objectives of this thesis are twofold, and were tested in two separate studies. First, the ability of remotely sensed environmental variables to predict the distribution of coastal bird species for the entire BC coast was investigated. Second, multiple environmental regionalization schemes were evaluated with regard to their ability to delineate avian Beta diversity across the province, and were compared to a regionalization built using species data directly. In Chapter 3, the distributions of 60 species of birds were estimated along the BC coast. Distribution models were built using species occupancy data linked to oceanic, terrestrial, and anthropogenic remotely sensed variables, as well as interpolated climate indices and spatial variables, in both single and ensemble models. The use of these four different types of environmental variables improved the distribution models? ability to estimate species occurrence, as did the use of ensemble modeling and the inclusion of spatial variables. Significant changes in the amount of occupied habitat by year were detected in 16 species for the eight year study period.   In Chapter 4, four environmental conservation regionalization schemes were compared using analysis of similarity (ANOSIM) tests to assess their ability to delineate Beta diversity. A new, species-based regionalization was then created to act as an ideal scenario, and was subsequently tested against each environmental regionalization scheme. These analyses demonstrated that all iii  environmental regionalization schemes delineated significant patterns in Beta diversity, with the Bird Conservation Regions scoring highest in ANOSIM testing overall and being the most similar to the species-based regionalization.   iv  Preface This thesis is the combination of two scientific papers currently in review of which I am the lead author. I was responsible for conceptual development, data analysis, writing, and editing on both manuscripts. Dr. Nicholas Coops was also involved in the development of both papers and provided editorial assistance. Dr. Mark Drever and Dr. Trisalyn Nelson were both involved in project development and provided editorial assistance with regard to Chapter 3. Dr. Meg Andrew and Douglas Bolton provided statistical advice and editorial assistance with regard to Chapter 4. Dr. Nancy Mahony and Dr. Trisalyn Nelson were both involved in project development and provided editorial assistance with regard to Chapter 4. Potential publications arising from the work completed in this thesis include: Rickbeil, G.J.M., Coops, N.C., Drever, M.C., Nelson, T.A., 2013. Improving coastal species distribution models through the integration of terrestrial, oceanic, and atmospheric data.  Rickbeil, G.J.M., Coops, N.C., Andrew, M.E., Bolton, D.K., Mahony, N., Nelson, T.A. 2013. Assessing conservation regionalization schemes: employing a Beta diversity metric to test the environmental surrogacy approach.   v  Table of Contents  Abstract ..................................................................................................................................................................... ii Preface ...................................................................................................................................................................... iv Table of Contents ................................................................................................................................................... v List of Tables ......................................................................................................................................................... vii List of Figures ...................................................................................................................................................... viii Glossary ..................................................................................................................................................................... x Acknowledgments ............................................................................................................................................... xii 1. INTRODUCTION ............................................................................................................................................. 1 1.1 Species Distribution Modeling ........................................................................................................................ 2 1.2 Avian Protected Areas Planning ..................................................................................................................... 2 1.3 Research objectives ............................................................................................................................................. 3 2. BRITISH COUMBIA AND DATA SETS USED ........................................................................................... 5 2.1 British Columbia ................................................................................................................................................... 5 2.2 Avian Species Data ............................................................................................................................................... 6 2.2.1 BC Coastal Bird Data .................................................................................................................................. 6 2.2.2 BC Breeding Bird Data .............................................................................................................................. 7 2.3 Environmental Variables .................................................................................................................................. 8 2.4 Environmental Regionalization Schemes ................................................................................................. 11 2.4.1 The North American Bird Conservation Regions ........................................................................ 11 2.4.2 The Canadian Ecoregions ...................................................................................................................... 11 2.4.3 The BC Biogeoclimatic Zones ............................................................................................................... 12 2.4.4 Remotely Sensed Unique BC Ecosystems ....................................................................................... 12 3. APPLYING LAND AND OCEAN BASED REMOTE SENSING PRODUCTS IN COASTAL SPECIES DISTRIBUTION MODELS ................................................................................................................................... 14 3.1 Introduction ......................................................................................................................................................... 14 3.2 Methods .................................................................................................................................................................. 17 3.2.1 Study Location ........................................................................................................................................... 17 3.2.2 Coastal Bird Data ...................................................................................................................................... 17 3.2.3 Environmental Data................................................................................................................................. 18 3.2.4 Species Distribution Models................................................................................................................. 19 3.2.5 Statistical Analysis ................................................................................................................................... 23 3.3 Results..................................................................................................................................................................... 25 3.3.1 Species Distribution Models................................................................................................................. 25 3.3.2 Model Prediction Variability in Space .............................................................................................. 27 3.3.3 Environmental Associations ................................................................................................................ 30 3.3.4 Conservation Applications .................................................................................................................... 31 3.4 Discussion ............................................................................................................................................................. 32 3.4.1 Single vs. Ensemble Species Distribution Models ....................................................................... 33 3.4.2 Spatial Variability of Model Predictions .......................................................................................... 35 3.4.3 Ensemble Model Environmental Variable Selection .................................................................. 36 3.4.4 Conservation Applications and Individual Species Models .................................................... 38 4. ASSESSING CONSERVATION REGIONALIZATION SCHEMES: EMPLOYING SPECIES DATA TO AUGMENT THE ENVIRONMENTAL SURROGACY APPROACH ............................................................... 42 4.1 Introduction ......................................................................................................................................................... 42 4.2 Methods .................................................................................................................................................................. 45 4.2.1 Bird Community Data ............................................................................................................................. 45 4.2.2 Environmental Regionalization Schemes ....................................................................................... 46 vi  4.2.3 Community Analyses ............................................................................................................................... 48 4.2.4 BC Avian Conservation Regions Delineation ................................................................................. 50 4.2.5 Statistical Analyses .................................................................................................................................. 52 4.3 Results..................................................................................................................................................................... 52 4.3.1 BC Breeding Bird Atlas Analyses ........................................................................................................ 52 4.3.2 ANOSIM Analyses ..................................................................................................................................... 53 4.3.3 Regional Coverage and Beta diversity delineation ..................................................................... 56 4.4 Discussion ............................................................................................................................................................. 57 4.4.1 Evaluating Environmental Regionalizations Against a Species-Based Ideal Scenario 57 4.4.2 Spatial and Environmental Structuring of Avian Beta Diversity .......................................... 58 4.4.3 Thematic Resolution ............................................................................................................................... 60 5. CONCLUSIONS .............................................................................................................................................. 62 5.1 Key Findings ......................................................................................................................................................... 62 5.2 Conservation Applications ............................................................................................................................. 64 5.3 Future Research .................................................................................................................................................. 65 REFERENCES ......................................................................................................................................................... 67 APPENDIX 1 .......................................................................................................................................................... 86 APPENDIX 2 .......................................................................................................................................................... 90    vii  List of Tables Table 2.1: The 22 spatial, environmental, and anthropogenic variables used as predictor variables in the single and ensemble Species Distribution Models. ..................................................................................... 10 Table 3.1: The frequency of modeled AUC values by model type (n=60 in all cases except n=59 for ANN). MEAN = mean ensemble model, WMEAN = weighted mean ensemble model. .............................. 25 Table 3.2: Post-hoc paired t-tests with a Bonferroni correction evaluating differences in mean transformed model AUC values (P values presented; n=60 in all cases except n=59 for ANN). MEAN = mean ensemble model, WMEAN = weighted mean ensemble model. ......................................................... 26 Table 4.1: A summary of each environmental regionalization scheme and selected remotely sensed and grid system regionalization schemes. Thematic resolution refers to the number of regions within each regionalization scheme containing >2 sampled Atlas cells. Cell number refers to the minimum or maximum number of sampled cells within a particular region for each regionalization. ....................................................................................................................................................................................................... 48   viii  List of Figures Figure 2.1: The British Columbia coast with the 2012 Coastal Waterbird Survey sites shown (survey sites vary by year depending on volunteer availability). ........................................................................................ 7 Figure 2.2 British Columbia with the BC Breeding Bird Atlas cell centroids shown (3547 pixels in total). ............................................................................................................................................................................................. 8 Figure 3.1: Untransformed AUC values by model type and spatial covariate inclusion. Error bars indicate 95% confidence intervals (n=60 in all cases except n=59 for ANN). MEAN = mean ensemble model, WMEAN = weighted mean ensemble model, sp = spatial variables. ................................................. 26 Figure 3.2: The weighted mean ensemble model AUC values grouped by (A) primary habitat type, and (B) functional feeding group. Error bars indicate 95% confidence intervals. Sample sizes are n = 9, 7, 44 respectively for habitat type, and n = 9, 7, 8, 5, 12, 19 respectively for functional feeding group. .......................................................................................................................................................................................... 27 Figure 3.3: The average standard deviation (SD) of predicted probabilities by site ? average distance to survey location (A) and minimum distance to survey location (B) model relationships. Average distance to site half-max = 113.7 km, asymptote = 0.29 SD. Minimum distance to site half-max = 8.1 km, asymptote = .28 SD. ................................................................................................................................. 28 Figure 3.4: The average standard deviation (Average SD) of predicted probabilities by site, as well as the Coastal Waterbird Survey (CWS) survey sites along the BC coast. Insert A focuses on the southern coast region, insert B on the costal fjords founds along the central coast, and insert C on the north coast and Haida Gwaii. .................................................................................................................................... 29 Figure 3.5: Most influential variable selection frequency, separated by primary habitat type (A and C; land, shore or water birds) or functional feeding group (B and D).  Graphs A and B include spatial variables (latitude and longitude), while C and D only consider environmental variables. See Appendix 1 for functional feeding group and habitat use classifications by species. ............................... 30 Figure 3.6: Mean species richness and Beta diversity for the entire study period. A missing coastal site indicates no oceanic data was available for that location for any year during the study period (coastal coverage = 89.4%). .............................................................................................................................................. 31 Figure 3.7: The species richness ? distance to road (A) and distance to cities (B) model relationships. Distance to road half-max = 10.7 km, asymptote = 41.1 species. Distance to city half-max = 458.8 km, asymptote = 65.7 species................................................................................................................. 32 Figure 4.1: Schematic diagram of the averaging kernel calculating local mean Beta diversity. ........... 50 Figure 4.2: The experimental semi-variogram of local mean Jaccard?s distance values with the fitted exponential model (range=200 km, sill=.023, RMSE= 2.56x10-3). .................................................................. 51 Figure 4.3: (A) The kriged local Beta diversity values using the modeled semi-variogram (Figure2), (B) the derived high Beta diversity (.8, .85, .9) and low Beta diversity (.45, .5, .55) isolines, and (C) the Beta diversity regions derived using the high Beta diversity isolines (shown). ................................. 52 Figure 4.4: Frequency distribution of hours of effort across the 3547 sample Atlas cells. .................... 53 ix  Figure 4.5: The Grand Overall (GO) and Grand Adjacent (GA) ANOSIM tests for (A) all environmental regionalization schemes and (B) the remotely sensed regions at varying thematic resolutions (all regionalizations delineated highly significant patterns in Beta diversity, P<.001). Also, the percentage of significant regions are shown (?=.05) for the Regional Overall (RO) and Regional Adjacent (RA) tests for both(C) the environmental regionalizations and (D) the remotely sensed regions at varying thematic resolutions. ...................................................................................................... 54 Figure 4.6: Maps of the Bird Conservation Regions (BCR) and Beta diversity derived regions (Beta) showing per-region ANOSIM R values for both the Regional Overall (RO) and Regional Adjacent (RA) tests. Diagonal lines indicate a non-significant ANOSIM R value. ........................................................... 55 Figure 4.7: (A) ANOSIM R values for both the Grand Overall (GO) and Grand Adjacent (GA) tests for all gird regionalizations (all regionalizations delineated highly significant patterns in Beta diversity, P<.001). (B) The percentage of significant regions for each grid regionalization for both the Regional Overall (RO) and Regional Adjacent (RA) tests...................................................................................... 56 Figure 4.8: Linear models of the proportion of each region sampled by atlas cells (ranging from 0 or no coverage to 1 or complete coverage) for the Bird Conservation Regions (BCR), Ecoregions (ECR), Biogeoclimatic zones (BEC), and Beta diversity derived regions (Beta) regressed against the regions Regional Overall (RO) and Regional Adjacent (RA) ANOSIM R values.  Graphs lacking lines indicate a non-significant relationship. *P<.1, **P<.01, ***P<.001. ....................................................................................... 57   x  Glossary ANN: Artificial Neural Networks. A machine learning algorithm for estimating probability of occurrence. ANOSIM: Analysis of Similarity. A non-parametric test for assessing differences across groupings using distance metrics. AUC: Area under the receiver-operator characteristic curve. A statistic measuring the goodness of fit of a presence/absence model. BCR: Bird Conservation Regions. Developed by the North American Bird Conservation Initiative as a regionalization for North America primarily for avian conservation. BEC: Biogeoclimatic Ecosystem Classification. A classification system for British Columbia developed for forestry purposes employing abiotic and biotic information. Beta diversity: The amount of species turnover (shared vs. unique species) experienced between two locations. Community: The assemblage of species at a particular location. CWS: Coastal Waterbird Survey. A coastal bird survey conducted by Bird Studies Canada at various locations along the British Columbia coast. CTA: Classification Tree Analysis: A recursive binary splitting algorithm used to estimate probability of occurrence. Ensemble model: An algorithm which uses a combination of individual model outputs to estimate probability of occurrence. GA: Grand Adjacent. An ANOSIM test using all within region distances and only between regions distances where regions are adjacent in space. GAM: Generalized Additive Model. A semi-parametric model used for estimating probability of occurrence. GBM: Boosted Regression Trees. A machine learning algorithm employing regression and decision trees which estimates probability of occurrence. GLM: Generalized Linear Model. A parametric linear model used for estimating probability of occurrence. GO: Grand Overall. An ANOSIM test using all within region distances and all between regions distances. Jaccard?s distance: A distance metric which measures the amount of shared and unique species between two pairwise sites. One indicates no species are shared, while zero indicates all species are shared. xi  Kriging: A method which uses a measurement of the spatial autocorrelation of a metric to estimate the metric at un-sampled locations. MARS: Multivariate Adaptive Regression Splines. A regression method which fits multiple linear segments to a non-linear curve to estimate probability of occurrence. Population: The individuals of a particular species occupying a particular region at one point in time. Probability of occurrence: The likelihood that a species will occupy a particular location, ranging from zero (zero likelihood) to one (absolute likelihood). RA: Regional Adjacent: An ANOSIM test which uses one particular region?s within distances and the between distances of all adjacent regions. RF: Random Forest. A binary recursive splitting algorithm used for estimating probability of occurrence. RO: Regional Overall. An ANOSIM test which uses one particular region?s within distances and all between regions distances. SDM: Species Distribution Model. An estimate of a species? presence and absence in space and/or time. Thematic resolution: The number of regions within a regionalization.   xii  Acknowledgments This project was made possible thanks to the thousands of hours volunteers spent collecting breeding bird and coastal bird data throughout British Columbia for the BC Breeding Bird Atlas and Coastal Waterbird Survey. Bird Studies Canada is responsible for running the BC Breeding Bird Atlas and Coastal Waterbird Survey and for provided both data sets, and this work would not have been possible without their efforts. Jessica Fitterer generously provided the remotely sensed regionalizations, as well as advice on how best to use them. Funding for both projects was provided by the Canadian Space Agency?s ?BioSpace: Biodiversity monitoring with Earth Observation Data? project, as well as by the National Science and Engineering Council of Canada, Environment Canada, and the British Columbia Innovations Council.  I was fortunate to receive extensive help and guidance during the completion of this thesis. My committee members, Dr. Peter Arcese and Dr. Nancy Mahony, both provided invaluable feedback throughout my MSc program, as well as for this Thesis. Dr. Trisalyn Nelson was also involved throughout my project and helped improve both manuscripts. The members of IRSS all contributed ideas at different stages of this research, and they are all thanked for their contributions.  Lastly, this research would not have happened without the huge amount time and energy put in by Dr. Nicholas Coops, and my thanks will never be able to repay him for the role he played throughout my MSc.   xiii  For my loving parents, Jim and Doreen. This is as much a product of your work as it is mine, Thank You. 1  1. INTRODUCTION Globally, species are being lost at an increasing rate (Sala et al., 2000; Dirzo and Raven, 2003). Land conversion - which is projected to increase as the global human population grows - is currently the largest cause of species extirpation (Foley, 2005). However, climate change is now recognized as a significant threat to species worldwide, and is projected to increase throughout the 21st century potentially surpassing land conversion as the major threat to our planet?s species assemblages (Gaston et al., 2003; Dawson et al., 2011). Avian communities, the subject of this thesis, appear to be following a similar pattern of decline (Gaston et al., 2003; Norris and Harper, 2004). Land use change has been implicated as the major driver of the bird species declines (Gaston et al., 2003; Norris and Harper, 2004; Drever and Martin, 2010). However, climate change (Thomas et al., 2003) and invasive species (Clavero et al., 2009), as well as more specific threats such as mountain pine beetle in British Columbia (Drever et al., 2008), have also been identified as potential threats. In response to these threats, the North American Bird Conservation Initiative (NABCI) was established in 1998 (North American Bird Conservation Initiative, 1999). The primary goal of the NABCI is to coordinate the conservation of North American bird species across their ranges irrespective of political boundaries. An initial step in conserving avian communities is answering a simple question ? what species of birds occur where? Having an understanding of the distribution of bird species across a land base is critical for implementing conservation plans (Guisan and Thuiller, 2005; Ara?jo and Guisan, 2006), assessing habitat availability (Betts et al., 2006), and estimating how range shifts may occur under future climate scenarios (Pearson and Dawson, 2003; Thomas et al., 2004; Hamman and Wang, 2006). Once species distributions are relatively well known, conservation strategies can be implemented, with the goal of protecting populations across a land base (Margules and Pressey, 2000). Both concepts, species distributions and conservation, are tested here as the subject of this thesis.   2  1.1 Species Distribution Modeling Many areas do not have reasonable estimates regarding species distributions (Ara?jo and Guisan, 2006). This can be due to the remoteness of an area or because an area is difficult to access. A lack of resources available to researchers can also limit the amount of sampling conducted across a land base. Because of these constraints, species distribution models are becoming more common in biogeography and ecology when direct sampling is prohibitive (Guisan and Zimmerman, 2000; Elith and Leathwick, 2009). Species distribution models estimate the probability of a species occurring at a given location based on their affinity for certain environmental characteristics (Guisan and Thuiller, 2005). These probabilities can then be converted into binary presence/absence values by employing a rule which states that above ?X? probability, the species is predicted to be present.  The British Columbia coast offers a good example of an area where modeling can be an effective tool for estimating species distributions. Many sections of the coast are extremely remote and have no road access (Cannings and Cannings, 1996), meaning that direct sampling must be by boat or aircraft, both of which are expensive options. By estimating species distributions through modeling, knowledge can be gained regarding what species occur where along the coast. While these models do not offer truth, with model accuracy assessments and post-hoc evaluation of the agreement of individual species models with known species distributions, they can be an effective initial prediction from which further work can proceed.  1.2 Avian Protected Areas Planning  Biodiversity protection has been used as a major driver of conservation planning (Rands et al., 2010). One of the most common methods of protecting biodiversity has been establishing reserve systems to provide habitat and separate species from threats present elsewhere (Margules and Pressey, 2000; Rodrigues and Brooks, 2007). For a reserve system to be successful it should meet two criteria: (1) it should represent a region?s species assemblages and (2) it should ensure that the 3  species it protects persist through time (Margules and Pressey, 2000). Because a full data set for all species within an area never truly exists, biodiversity surrogates must be employed to represent a region?s communities during the planning phase of a reserve network (Margules et al., 2002). Environmental surrogates, such as habitat type, habitat heterogeneity and climate have been employed with varying levels of success to delineate conservation regions within regionalization schemes. The goal of such regionalizations is to delineate homogenous groups of species within each region, which can then be conserved within a reserve system. The North American Bird Conservation Initiative was developed as a plan for conserving avian populations across the continent regardless of political boundaries. One of the cornerstones of the plan was the delineation of the North American Bird Conservation Regions (BCR), which are, ideally, spatially contiguous regions which divide the continent into relatively homogenous communities of birds (North American Bird Conservation Initiative, 1999). These regions would then act as planning units for avian community conservation at the continental scale. The BCRs have been used in both Canada (NABCI Canada, 2012) and the United States (NABCI U.S. Committee, 2009) as a basis for planning bird conservation strategies. 1.3 Research objectives The overall objective of this thesis was to evaluate how two existing citizen-science avian data sets can be used to improve our knowledge regarding bird species distributions and conservation throughout BC.  Two research questions were posed to meet this objective, one pertaining to species distributions and another addressing avian conservation: 1. Can remotely sensed environmental indicators be used to estimate coastal bird distributions along the BC coast? 2. Which environmental regionalization schemes are most useful for avian conservation? 4  Chapter 2 describes the study region, including the entire province of BC as well as its coastal areas, and both avian community data sets in detail. It also describes the remotely sensed environmental products used in Chapter 3, and the environmental regionalization schemes used in Chapter 4. Chapter 3 addresses Research Question 1, where coastal bird presence/absence data is used in multiple species distribution models with remotely sensed (and interpolated) environmental variables to evaluate and predict species distributions along the BC coast. Model type, variable type and the inclusion of spatial covariates are all tested in a single model framework. Additionally, I demonstrate how the distribution models can be used to generate products which can be used in conservation planning scenarios. Chapter 4 addresses Research Question 2, where multiple environmental regionalization schemes are tested to determine their ability to delineate avian Beta diversity across BC. Additionally, I outline how species data can be used to define regions within a regionalization directly, and use this species-based regionalization as an ideal scenario for comparison with the environmental surrogacy approach. Lastly, Chapter 5 summarizes the conclusions arrived at in this thesis, the potential conservation applications of the work detailed here, and provides suggestions for future research.   5  2. BRITISH COUMBIA AND DATA SETS USED 2.1 British Columbia The questions addressed in this thesis will be focused within the Province of British Columbia (BC), Canada. BC lies on the west coast of Canada and as such, is heavily influenced by the Pacific Ocean (Cannings and Cannings, 1996). The province generally experiences warm, dry summers and wet, cold winters; however, climate is quite variable across the province?s landscape, with the coast generally being warmer and wetter and the interior being drier and cooler (Pojar and Meidinger, 1991). Topographically, the province is more variable than any other province in Canada, containing two major mountain ranges, vast plateaus and extensive plains (Pojar and Meidinger, 1991). Plant communities vary from high alpine meadows to grasslands to desert shrub-steppe (Parish et al., 1996); however, the majority of the province is forested as either coniferous or mixed coniferous and broadleaf forests (Pojar and Meidinger, 1991). This diversity of ecosystem types results in a wide variety of species occurring within the province ? to date the BC Breeding Bird Atlas has documented 316 breeding species of birds in the province.  BC contains the entire Canadian Pacific coast, which is over 25 000 km long, making up approximately 10% of Canada?s coast, and represents some of the province?s most remote locations. The south coast has relatively high levels of anthropogenic activity due to its proximity to Vancouver and Victoria (the two largest coastal cities in BC); the central and north coasts have lower levels of human influence. The province?s coast is considered to be a temperate marine climate due to its moderate temperatures, high levels of precipitation, and tight coupling with Pacific conditions (Klock and Mullock, 2001). The majority of the BC coast is highly productive coniferous forest, with some oak woodlands occurring in the southern portions (Pojar and Meidinger, 1991). 6  2.2 Avian Species Data 2.2.1 BC Coastal Bird Data The Coastal Waterbird Survey (CWS) data set, used in Chapter 3, is a citizen-science data set collected by Bird Studies Canada (for a more detailed survey description, see Crewe et al., 2012). Survey design involves a volunteer visiting a coastal site once monthly and recording all bird species encountered as well as their abundance. Surveyors record all detections out to 500 m offshore, as well as any birds using the onshore environment. The survey focuses on the September to April period, with some surveys being conducted year-round. The survey began in 1999 and continues today, with the number of locations per year ranging from approximately 100 sites to over 300 (Figure 2.1). The CWS data set is freely available to researchers. 7   Figure 2.1: The British Columbia coast with the 2012 Coastal Waterbird Survey sites shown (survey sites vary by year depending on volunteer availability). 2.2.2 BC Breeding Bird Data The BC Breeding Bird Atlas data set (http://www.birdatlas.bc.ca/), used in Chapter 4, is a five year project undertaken by Bird Studies Canada which is due for completion in 2012, and involves surveying 100 km2 cells throughout the province for evidence of breeding birds. To date, over 200 000 records document all breeding evidence, highest breeding evidence, rare and unique species, and raw point count data, as well as information on survey effort, measured as hours spent surveying each cell. Species data are recorded as presence/absence for the 316 breeding species detected across 3547 sampled cells (Figure 2.2). This data set is also freely available to researchers. 8   Figure 2.2 British Columbia with the BC Breeding Bird Atlas cell centroids shown (3547 cells in total). 2.3 Environmental Variables Five categories of data were used in the study described in Chapter 3: oceanic, terrestrial, atmospheric, anthropogenic and spatial. Oceanic and atmospheric data were gathered by year and averaged for each winter at each survey site. Terrestrial and anthropogenic data were assumed to be more static in nature, and therefore one mean value per site was used to represent every year in the data set. Due to the highly mobile nature of coastal birds, any data occurring within three km of a survey site was used during the averaging of the oceanic variables, while any terrestrial data within two km of a survey site was used (Guisan and Thuiller, 2005). Atmospheric data was referenced from site centroids.  9  The oceanic data, selected as direct estimates of sea surface temperature (Lobb and Buckley, 2012), oceanic sediments (Esaias et al., 1998), and productivity (Moses et al., 2009) were Level 2 variables derived from the MODIS sensors. All data from day of year 335 to 91 were averaged by survey polygon, matching the time frame for the avian data. The high temporal resolution of MODIS allowed for up to 90 scenes to be composited by winter for the length of the BC coast. The atmospheric variables were interpolated using the ClimateWNA version 7.0 program developed by Wang et al. (2006), which provided estimates of mean winter temperatures and precipitation conditions by site. The terrestrial variables estimated site productivity, broken up into overall productivity, minimum productivity, and seasonality measured using the Dynamic Habitat Index (DHI) data set (Coops et al., 2008), and land cover richness using the Earth Observation for Sustainable Development of Forests (EOSD) data set (Wulder et al.,  2007; Wulder et al.,  2008). The anthropogenic variables estimated the distance of each site from human influences (cities and roads) using the nighttime lights data and Canadian road atlas data (Andrew et al., 2012). Lastly, the spatial variables used were a site?s latitude and longitude measured at its centroid.    10  Table 2.1: The 22 spatial, environmental, and anthropogenic variables used as predictor variables in the single and ensemble Species Distribution Models. Category Variable Description Time Frame Citation Spatial Latitude Longitudinal coordinate of site centroid NA NA  Longitude Latitudinal coordinate of site centroid NA NA Terrestrial Mean DHI Mean yearly vegetation productivity (fPAR) 2000-2005 Coops et al. (2008)  Min DHI Minimum yearly vegetation productivity (fPAR) 2000-2005 Coops et al. (2008)  Cov DHI Seasonality of productivity (fPAR) 2000-2005 Coops et al. (2008)  Landcover Richness Number of different Land Cover Classes (EOSD) 2000 Wulder et al. (2007, 2009) Anthropogenic Distance to Road Distance to the nearest road from pixel centroid 2008 Andrew et al. (2012)  Distance to City Distance to nearest light source from pixel centroid 2000 Andrew et al. (2012) Atmospheric tmax Maximum Winter Temperature By year Wang et al. (2006)  tmin Minimum winter Temperature By year Wang et al. (2006)  tave Mean winter temperature By year Wang et al. (2006)  ppt Total winter precipitation By year Wang et al. (2006)  pas Total winter precipitation as snow By year Wang et al. (2006) Oceanic pic Calcite concentration By year Esaias et al. (1998)  poc Particulate organic carbon concentration By year Esaias et al. (1998)  cdom index Colored dissolved organic matter By year Esaias et al. (1998)  ipar Instantaneous Photosynthetically active radiation  By year Esaias et al. (1998)  par Photosynthetically active radiation  By year Esaias et al. (1998)  nflh Normalized fluorescence activity By year Esaias et al. (1998)  chlor a Chlorophyll-a concentration By year Moses et al. (2009)  Kd 490 Diffuse attenuation coefficient (water turbidity) By year Esaias et al. (1998)  sst Mean winter sea surface temperature By year Lobb and Buckley (2012) 11  2.4 Environmental Regionalization Schemes Four different environmental regionalizations (Figure 2.3), described below, were tested in the study described in Chapter 4. Each regionalization incorporates information from at least one of the three themes identified by Mackey et al. (2008) as useful for creating biogeographical regionalizations. 2.4.1 The North American Bird Conservation Regions The North American Bird Conservation Initiative (http://www.bsc-eoc.org/nabci.html/) was created in 1998 as a partnership between governments, industry and environmental organizations in North America to improve avian conservation (North American Bird Conservation Initiative, 1999). One of the major developments from this initiative was the North American Bird Conservation Regions (BCR), which aimed to create biologically meaningful conservation regions containing distinct bird communities around which tailored conservation strategies could be developed, without consideration of political boundaries. The delineation of these regions was based on existing land classifications as well as expert opinion. As such, the BCRs incorporate information from Mackey et al.?s (2008) species composition and ecosystem drivers themes. 2.4.2 The Canadian Ecoregions Ecozones/Ecoregions are contiguous parcels of land which are expected to contain similar biotic and abiotic conditions, with Ecoregions being the finer of the two thematically (Bailey et al., 1985; Olsen et al., 2001). The classification is based on mapped environmental, geological and climate variables, as well as expert opinion. Ecoregions have been used to identify representation targets for conservation (Olson and Dinerstein, 1998; Wiersma and Urban, 2005), and are also used as the primary regionalization scheme for conservation organizations such as The Nature Conservancy (Groves et al., 2000). The Ecoregions of Canada were delineated by the National Ecological Framework for Canada (Ecological Stratification Working Group, 1995) and are subdivisions of the 12  Canadian Ecozones. Ecoregions are characterized by distinct assemblages of landforms, climate, soil and human uses; therefore, the information used to delineate the Ecoregions falls under the ecosystem drivers theme (Mackey et al., 2008). 2.4.3 The BC Biogeoclimatic Zones The British Columbia Biogeoclimatic zones (BEC) were developed in the 1960s (Pojar et al., 1987) and are under a constant state of refinement. The BEC system uses local soil, vegetation, topography, and climate data to delineate distinct ecosystems, with vegetation being most influential (British Columbia Forest Service, 2012). Similar to Ecozones, the classification is hierarchical, with BEC zones being the broadest; however, BEC zones are not required to be contiguous, and consequently, unlike the Ecoregions, are able to reflect the zonation of environmental conditions along elevation gradients. The version of BEC used in this study defines 14 zones for BC. The BEC system incorporates information from both the species composition and ecosystem drivers themes (Mackey et al., 2008). 2.4.4 Remotely Sensed Unique BC Ecosystems Fitterer et al. (2012) created a regionalization scheme for BC by modeling 10 remotely sensed physical (elevation and soil wetness), available energy (solar insulation and snow melt) and vegetation productivity (fraction of photosynthetically active radiation) variables in a two-step hierarchical clustering framework, creating a unique ecosystem classification for the province. This regionalization scheme incorporates information from both the ecosystem drivers and ecosystem responses themes (Mackey et al., 2008). The resulting regionalization scheme can be produced at variable thematic resolutions based solely on environmental variables without any spatial continuity requirements.  13   Figure 2.3: The four environmental regionalization schemes tested in Chapter 4.    14  3. APPLYING LAND AND OCEAN BASED REMOTE SENSING PRODUCTS IN COASTAL SPECIES DISTRIBUTION MODELS 3.1 Introduction Coastal ecosystems are complex, occurring at the interface between land and water, be it oceans or large freshwater bodies (Wilkinson et al., 1997). This complexity is reflected in coastal biotic communities, with many species using both terrestrial and aquatic resources to fulfill different portions of their life history requirements. Coastal bird communities are an excellent example of this, with many species exploiting both terrestrial resources, such as nesting or roosting sites (Burger et al., 2010), and aquatic resources such as access to prey (Janssen et al., 2011). With the additional effect of microclimate conditions affecting thermoregulatory capabilities (Carrascal and Diaz, 2006; Gutierrez et al., 2012), it becomes apparent that many different environmental variables may govern coastal bird community composition. Many of the world?s coastlines are extremely remote, compounding the problem of understanding coastal environmental conditions and their effects on species composition by making direct sampling infeasible, and as a result, indirect estimates of communities may be the only option available to researchers (Guisan and Thuiller, 2005). The British Columbia (BC) coastline is spatially extensive, rugged, and remote; to date, no community of terrestrial animals has been modeled for the length of its coast. Coastal birds, in particular, have been identified as a community which has experienced significant declines in many species along a portion of the BC coast (Bower, 2009; Crewe et al., 2012), and have been suggested as a candidate community to assess coastal ecosystem health (Bower, 2009; Lopez et al., 2010; Wiegand et al., 2010).  Species distribution models (SDMs), which predict a species? occurrence based on environmental and spatial correlates, are becoming more common in many fields of natural sciences (Guisan and Zimmerman, 2000; Guisan and Thuiller, 2005; Ara?jo and Guisan 2006; Thuiller, 2007; Elith and Leathwick, 2009). The growth in the use of SDMs can be attributed to multiple factors ? the 15  increase in available, spatially expansive environmental and species data (Guisan and Zimmerman, 2000), and the development of increasingly sophisticated modeling techniques allowing for more accurate distribution models (Marmion et al., 2009). The effectiveness of different modeling techniques has been the subject of much evaluation without clear consensus (Austin, 2002; Ara?jo and New, 2007; Elith and Graham, 2009; Marmion et al., 2009). The evaluation of different SDM techniques is of paramount importance, as effective distribution models can inform situations where species data are unavailable, historical records are lacking, or when estimating species occurrences under future scenarios (Pearson and Dawson, 2003; Elith and Leathwick, 2009; but see Fitzpatrick and Hargrove, 2009). A more recent development is that of ensemble modeling, which employs multiple different model types under the assumption that different techniques have different strengths, and a combination of models may be more able to model a species across its range than any one single technique alone (Ara?jo and New, 2007; Marmion et al., 2009).  A large focus of current SDM development has been on future projections under varying climate scenarios; however, SDMs show promise for modeling spatially difficult areas as well as different temporal states (Randin et al., 2006; Elith and Leathwick, 2009). Since SDMs are correlative models where relationships between a species occurrence and environmental conditions are estimated, they require environmental data with full coverage of a given study area, as well as species data. Many of the world?s coastlines are inaccessible to traditional forms of environmental data collection; however, remotely sensed products can provide environmental data, irrespective of spatial location, for SDMs ranging from terrestrial productivity (Coops et al., 2008) to oceanic sea surface temperature (Haines et al., 2007). Remote sensing, due to its ability to provide terrestrial, oceanic, and anthropogenic data in remote locations, may be uniquely positioned to provide this wide range of environmental information required for estimating species occurrence in complex 16  environments, with the added advantage of most often being freely available to researchers. The synthesis and testing of these four environmental categories has not, to our knowledge, been evaluated. The inclusion of spatial variables in distribution models has also received attention, with some suggesting that their inclusion delineates the modeling of realized niches from fundamental niches due to the use of observational species data (Guisan and Zimmerman, 2000; Ara?jo and Guisan, 2006), while others have argued that the niche concept has little applicability in species distribution modeling (McInerny and Etienne, 2012).  Scientific questions surrounding range size and species richness (Jetz and Rahbek 2002), habitat loss and fragmentation (Betts et al., 2006), and the effects of different climate change scenarios (Pearson and Dawson, 2003; Thomas et al., 2004; Hamman and Wang, 2006) have all been addressed using SDM techniques. Equally, SDMs are becoming more common in applied conservation situations, guiding conservation policy, reserve design, and protected areas planning (Wilson et al., 2005; Rodriguez et al., 2007; Schuster and Arcese, 2012). SDMs lend themselves well to estimating common indices used in systematic conservation planning, such as species richness and Beta diversity (Tuomisto, 2010a, b) due to their binomial species occurrence outputs. Additionally, species distribution models can be used as guides to assess species representation in a protected areas network, again due to the presence/absence estimates generated by the models.  The goals of this Chapter are to examine the feasibility of using freely available remotely sensed and interpolated environmental data to model the distribution of coastal bird species, while also examining which SDM techniques are most appropriate in this setting. Our research questions are:  1) Can current SDM techniques ? employing terrestrial, aquatic, anthropogenic, and climatic data ? accurately model coastal species? distributions, and if so which model types are most effective? 2) How does model performance vary across species groupings such as functional feeding groups or primary habitat type? 3) How do model predictions vary spatially? 4) Which categories of 17  environmental data are most important for estimating coastal bird species? and 5) How might these models be used to inform systematic conservation planning? 3.2 Methods 3.2.1 Study Location The scope of this study was the entire BC coast (for a detailed location description see Section 2.1; Figure 2.1). 3.2.2 Coastal Bird Data The bird species data used here was derived from the Coastal Waterbird Survey data set (for a detailed data description see Section 2.2.1). Species abundances were recorded in three locations, on-shore (on land), near-shore (0-100 m from the shoreline), and offshore (100-500 m from the shoreline). Abundance data with multiple visits within each winter was converted into presence/absence by species by considering a species present if it was sampled in any of the four surveys conducted by winter in any number. Absences were modeled as true absences given the four repeat visits per winter, resulting in 970 survey sites across eight years. To avoid issues with randomly selecting zero presences in the model verification data set, species were modeled only if they were present in a minimum of 5% samples within the entire study period, which lead to 60 species being modeled.  Species were classified by functional feeding group using published literature where available (Bower, 2009; Conover, 1983; de Graaf et al., 1985; Greenhalgh, 1952; Pierotti and Annet, 1991; Sydeman et al., 2001; see Appendix 1 for functional feeding group assignments by species), or the Cornell Lab of Ornithology Online Bird Guide (http://birds.cornell.edu/onlineguide/), to assess variable importance and model fit (model AUC) across functional feeding groups. Species were also coarsely classified by primary habitat use (land, shore, or water; Appendix 1) and by primary 18  detection location (on-shore, near-shore, or off-shore) to test whether model performance varied due to these factors.  The 60 modeled species represent a diverse group, with open ocean foragers such as grebes (Family: Podicipedidae) and cormorants (Phalacrocorax spp.), coastal birds such as sandpipers (Family: Scolopacidae), to more diverse families such as gulls (Family: Laridae). The time frame used in this study began in December 2002 and continued to March of 2010. The survey window of December 1st to March 31st coincided with the time when the majority of overwintering coastal birds used the BC coast, but before most migratory species using the pacific flyway might be present in survey sites. Environmental data availability dictated the start and end study years (oceanic data was not available until 2002, and atmospheric data ended in 2010). 3.2.3 Environmental Data As discussed earlier, different species of coastal birds use coastal resources in various ways. As such, the environmental variables employed here were selected as broad scale measures of environmental conditions governing nutrition, habitat, thermoregulation, human influences, and/or spatial effects. Environmental data (described in Section 2.3; Table 2.1) were linked spatially (and temporally for the oceanic and atmospheric data) to the CWS sites as well as the entire BC coast, which was divided up into sections varying from two to five km in length, matching the size distribution of the survey sites. Due to the difficult nature of remote sensing coastlines for oceanic variables, not all survey polygons or coastal polygons were able to be linked to oceanic variables, meaning certain areas could not be modeled each year. The yearly coastal coverage ranged from 63.3% in 2002 to 75.0% in 2008. 19  3.2.4 Species Distribution Models Species distribution models were developed using the Biomod2 package (Thuiller et al., 2009) for R statistical software version 2.15.1 (R Core Development Team, http://www.r-project.org/). The data were partitioned first into training and verification data using an 80/20 split (see Marmion et al. (2009) for a more detailed outline of data partitioning for ensemble models). I acknowledge here that the verification data set does not represent an independent model validation data set as recommended by Ara?jo and Guisan, (2006); however, alternatives were not available given the uniqueness of the data set. Seven individual models were built: Generalized Additive Models (GAM), Generalized Linear Models (GLM), Boosted Regression Trees (GBM), Multivariate Adaptive Regression Splines (MARS), Classification Tree Analysis (CTA), Artificial Neural Networks (ANN) and Random Forests (RF).  3.2.4.1 Generalized Additive Models (GAM) GAM is an extension of GLM (see Section 3.2.4.2) that is similar in approach, but which is classified as non-parametric (Yee and Mitchell, 1991). GAM uses a link function to describe the relationship between predictor and response variables; however, the link is determined by the data and not given beforehand (Guisan et al., 2002). A smoothing spline was used here with the response data defined as binomial (Hastie, 1992). Forward stepwise AIC was used as the model selection criteria to achieve parsimony. 3.2.4.2 Generalized Linear Models (GLM) GLM has been used extensively in the past to analyze and estimate ecological relationships. It is a parametric, linear model, similar to that of linear regression. However, GLM can incorporate non-normal relationships between predictor and explanatory variables, non-constant variance structures, and transformations to represent non-linear relationships (Guisan et al., 2002; Venables et al., 2002). This is done by specifying a link function and a data distribution. GLM then uses a 20  maximum likelihood procedure to converge on parameters estimates used in the final model (Venables et al, 2002). A binomial distribution and a logit link function were employed here due to the presence/absence data, with maximum iterations set at 200. As was the case for GAMs, forward stepwise AIC was used as the model selection criteria to achieve parsimony. 3.2.4.3 Boosted Regression Trees (GBM) GBM is a relatively new method for estimating species distributions and is a machine learning approach combining regression techniques and classifications trees in one model framework (Elith et al., 2008). GBM can handle many data types, missing data, and correlated response variables. GBM uses two algorithms: regression trees and boosting. Regression trees involve recursive binary splitting of the data, fitting classification trees which are insensitive to outliers and differing scales of predictor variables. However, classification trees have difficulty modeling linear responses, and are typical out performed by GAM and GLM (Elith et al., 2008). Boosting addresses this issue, by building multiple trees, each which is selected in a forward stagewise procedure (stagewise implying that the previous models are unchanged) as the tree which reduces model residual error most (Leathwick et al., 2006). Similar to GLM, a link function can be defined a priori to accommodate many response types, with binomial being used here. A learning rate, which determines the importance of each new model to the overall model, of 0.001 and a bag fraction of 0.5, which controls the amount of data to be used in each iteration, were set for all GBM models (Thuiller et al., 2009). 3.2.4.4 Multivariate Adaptive Regression Splines (MARS) MARS is another relatively new model type for fitting distribution models. It involves fitting multiple regression segments allowing for complex, non-linear relationships to be estimated (Friedman, 1991; Leathwick et al., 2005). Breaks are identified which are optimized using stepwise 21  cross-validation, which also determines which variables are retained in the final model. A binomial distribution was used here to describe the presence/absence data (Thuiller et al., 2009).   3.2.4.5 Classification Tree Analysis (CTA) CTA involves recursive binary splitting which aims to separate the response data into two groups with maximum internal homogeneity (Thuiller et al., 2003). Trees are then pruned backwards by eliminating splits which are deemed less important by their relative contribution to variance explained. Like other classification tree approaches, CTA does not require a prior knowledge regarding distribution types, and can handle multiple forms of data.  3.2.4.6 Artificial Neural Networks (ANN) ANN is a machine learning algorithm which can incorporate non-normal, non-linear and noisy data in a single model framework (Lek and Guegan, 1999). ANN constructs models based on predictor variables with known results (Lek et al., 1996). Predictor variables are fed through a number of hidden layers in a uni-directional manner (Lek et al., 1996). The nodes, or connections between predictor and hidden layers, are analyzed, and if the final output provides the wrong result, the network weights are adjusted until the prediction of the model is more likely to be correct (Lek and Guegan, 1999). A single hidden layer was used here (Thuiller et al., 2009). 3.2.4.7 Random Forests (RF) Random forests is a classification tree approach which has shown to be highly effective at modeling ecological relationships. It builds on traditional CTA by creating many trees where at each node, only a subset of all predictor variables are available for use for binary partitioning (Cutler et al., 2007). The final model can include thousands of trees, which are then averaged to provide one overall model (Prasad et al., 2006). A maximum of 1000 trees was used here.  22  3.2.4.8 Ensemble Models The individual models were then combined into ensemble models using the two techniques found to be most promising by Marmion et al. (2009) - mean and weighted mean. The mean ensemble algorithm uses a simple mean of estimated occurrence probabilities; however, the weighted mean algorithm requires a fit statistic to be calculated to assign weights to the best models. To achieve this, the build data were randomly divided using a second 80/20 split with the larger set used to build individual models and the smaller set as individual model verification. This procedure was repeated ten times for each single model resulting in 70 models built for each species from which ensemble models were calculated. This second, iterative split also acted as a form of sub-sampling within the data, helping to address any potential spatial biases which existed by randomly selecting build data during each model run. Predicted probabilities generated by each model were converted into binary/presence absence predictions (or kept for calculating the standard deviation of prediction probabilities by site, see below) using a threshold which maximized model AUC scores. The area under the receiver-operator characteristic curve (AUC) was used in all evaluations of model fit (Marmion et al., 2009; O?Hanley, 2009; Seo et al., 2009; but see Lobo et al., 2008). Models were built with and without latitude and longitude covariates to examine the effects of their inclusion. Both ensemble models were projected onto the entire coast for each year of the study period where full environmental data were available. The standard deviation (SD) was calculated by site using the seven individual model?s predicted probabilities for each of their ten runs, resulting in an n = 70 for each site. This produced estimates of the variability of model predictions by site, with a low SD indicating agreement between individual models regarding the predicted probability of occurrence for a given species at each site. An overall SD for each site was calculated (due to multiple years of predictions) as the mean SD of all models across all species, resulting in one overall estimate of prediction variability by site.  23  3.2.5 Statistical Analysis All statistical analyses were carried out in R statistical software. Model performance (AUC values) was analyzed using a two-way, fixed effects ANOVA with interaction, with model type and the inclusion of spatial covariates as the fixed effects. Data was transformed using a Blom rank-based inverse normal transformation (Blom, 1958) to meet assumptions of ANOVA. Multiple t-tests with a Bonferroni correction were employed post-hoc to test for differences where appropriate. The weighted mean ensemble model performance (untransformed model AUC) was also analyzed by functional feeding group, primary habitat type, and primary detection location using one-way ANOVA and post-hoc t-tests. Assessing individual variable importance in ensemble models is complicated due to their reliance on multiple single models, which may or may not include all, some, or none of the same environmental and/or spatial variables. As such, variable importance was determined for each species? ensemble models using a ?leave-one variable out? approach where models were built without each variable (repeated 10x per model) and evaluated against the ensemble model using a Pearson?s correlation coefficient (r) (Thuiller, 2009). The assumption is that the more different a model is due to the removal of a particular variable (measured by correlating the full model with the model lacking variable ?x?) the more influential that particular variable is. Variables were ranked using a 1 ? r calculation, so that the most important variables would score highest. If a model did not change due to the removal of a variable, its r would equal 1, and it would score 0 for variable importance.  The SD of predicted occurrence probabilities was modeled against average and minimum distance to survey site to assess whether model prediction variability changed with regard to its proximity to survey locations. Non-linear least squares ?half-max? asymptotic equations were used to represent these relationships, with model fit measured using root mean-squared error (RMSE). 24  Simple linear regression was used to analyze change in available habitat through time by species, with available habitat defined as the amount of habitat predicted to be occupied by the weighted mean ensemble model for a given species. The proportion of predicted occupied habitat, expressed as a percentage of total habitat measured for a given year, was used to compensate for unequal spatial coverage between years (as discussed earlier, coastal coverage ranged from 63.3%-75.0%).  Species richness, calculated as the sum of all predicted presences for a given location, was evaluated against both anthropogenic variables again using non-linear least squares ?half-max? asymptotic equations. A natural logarithm transformation was then applied to species richness and both anthropogenic variables to meet assumptions of linear regression, and both anthropogenic variables were employed as predictors of species richness in single and multiple linear regression models. Local Beta diversity, which is an estimate of site uniqueness (Whittaker, 1972), was estimated using Jaccard?s distance (Andrew et al., 2012), with pairwise comparisons being calculated between a site and any other site occurring within 20 km measured from the site centroid. Species richness and Beta diversity were averaged through time by site, and compared against one another using simple linear regression.  All statistical tests employed an alpha level of 0.05, save linear regression assessing change in occupied habitat through time, which employed an alpha value of 0.1 due to a lack of data availability (the 8 year study yield only 1 data point per year for occupied habitat estimates, for a total of 8 estimates).   25  3.3 Results 3.3.1 Species Distribution Models Table 3.1: The frequency of modeled AUC values by model type (n=60 in all cases except n=59 for ANN). MEAN = mean ensemble model, WMEAN = weighted mean ensemble model. AUC Value GLM GBM GAM CTA ANN MARS RF MEAN WMEAN >0.9 1 4 1 1 0 1 1 4 4 >0.8 7 30 15 11 7 21 7 31 32 >0.7 27 21 20 18 13 28 27 22 21 >0.6 21 5 22 18 28 7 19 3 3 >0.5 4 0 2 11 11 3 5 0 0  Species distribution models were run as single predictive models and ensemble models, both with and without spatial covariates, to test the effects of single vs. ensemble modeling as well as the inclusion of spatial data (Figure 3.1; see Appendix 1 for individual species scores). Weighted mean was the most successful model type when considering AUC score ranges, with four models having an AUC above 0.9, and another 32 models above 0.8 (Table 3.1), and was therefore used in all model projections. When using the industry standard for acceptable models (AUC scores above 0.7; Marmion et al., 2009) both ensemble models produced 57 (out of 60 models, or 95%) species distribution models which would be classified as acceptable.  Both model type and the inclusion of spatial covariates were significant determinants of transformed mean AUC values (P<.001 in both cases). The interaction between spatial covariates and model type was found to be non-significant (P=.11). The inclusion of spatial covariates significantly improved model AUC values (P<.001). The two ensemble model types (mean and weighted mean) were found to have significantly higher transformed AUC values than all single models save boosted regression trees (Table 3.2) with weighted mean also producing the highest mean AUC value. Weighted mean ensemble model AUC values ranged from 0.645 for herring gull (Larus smithsonianus) to 0.976 for sanderling (Calidris alba).  26   Figure 3.1: Untransformed AUC values by model type and spatial covariate inclusion. Error bars indicate 95% confidence intervals (n=60 in all cases except n=59 for ANN). MEAN = mean ensemble model, WMEAN = weighted mean ensemble model, sp = spatial variables. Table 3.2: Post-hoc paired t-tests with a Bonferroni correction evaluating differences in mean transformed model AUC values (P values presented; n=60 in all cases except n=59 for ANN). MEAN = mean ensemble model, WMEAN = weighted mean ensemble model. Model Type ANN CTA GAM GBM GLM MARS MEAN RF CTA 1 - - - - - - - GAM <0.001 <0.001 - - - - - - GBM <0.001 <0.001 <0.001 - - - - - GLM 0.001 0.015 1 <.001 - - - - MARS <0.001 <0.001 1 0.028 0.453 - - - MEAN <0.001 <0.001 <0.001 1 <0.001 <0.001 - - RF 1 1 0.083 <0.001 0.985 <0.001 <0.001 - WMEAN <0.001 <0.001 <0.001 1 <0.001 <0.001 1 <0.001  The weighted mean ensemble model?s mean AUC values (used because it was found to have the highest overall AUC values) did not differ significantly based on primary detection location (P = 0.246) or primary habitat type (P = 0.067, Figure 3.2 A), but was shown to have significant differences across functional feeding groups (P = 0.007, Figure 3.2 B). Post-hoc t-tests revealed a 27  significant difference in mean AUC between insectivores and piscivores (P = 0.016) and near-significant difference between insectivores and carnivores (P = 0.076). No other functional feeding groups were found to differ.  Figure 3.2: The weighted mean ensemble model AUC values grouped by (A) primary habitat type, and (B) functional feeding group. Error bars indicate 95% confidence intervals. Sample sizes are n = 9, 7, 44 respectively for habitat type, and n = 9, 7, 8, 5, 12, 19 respectively for functional feeding group. 3.3.2 Model Prediction Variability in Space Model prediction variability at a particular site was found to be related to the distance of that site from CWS survey locations (Figure 3.3). Average site SD was more strongly related to average distance to survey (RMSE = 0.016 SD) than minimum distance to survey (RMSE = 0.023 SD). Figure 3.3 (A) predicts a 47% increase in the SD of predicted site probabilities when average distance to survey site increased from 100 to 300 Km, while Figure 3.3 (B) predicted a 63% increase in SD when the minimum distance to a survey site increased from 0 to 50 Km. Both predict an asymptote in this relationship between 0.28 and 0.29 SD. 28   Figure 3.3: The average standard deviation (SD) of predicted probabilities by site ? average distance to survey location (A) and minimum distance to survey location (B) model relationships. Average distance to site half-max = 113.7 km, asymptote = 0.29 SD. Minimum distance to site half-max = 8.1 km, asymptote = .28 SD. Model prediction variability was found to be lowest in the southern portion of the BC coast (Figure 3.4 A) and highest in the central and north coast regions (Figure 3.4 C). Model prediction variability was also found to increase towards the terminal end of many coast fjords (Figure 3.4 B, C). Minimum SD was found to be 0.13, while maximum SD was 0.31. 29   Figure 3.4: The average standard deviation (Average SD) of predicted probabilities by site, as well as the Coastal Waterbird Survey (CWS) survey sites along the BC coast. Insert A focuses on the southern coast region, insert B on the costal fjords founds along the central coast, and insert C on the north coast and Haida Gwaii. 30  3.3.3 Environmental Associations  Figure 3.5: Most influential variable selection frequency, separated by primary habitat type (A and C; land, shore or water birds) or functional feeding group (B and D).  Graphs A and B include spatial variables (latitude and longitude), while C and D only consider environmental variables. See Appendix 1 for functional feeding group and habitat use classifications by species. The spatial covariates were the most important variables in every functional feeding group, except insectivores, where anthropogenic variables were also selected as most important (Figure 3.5; see Appendix 2 for variable loadings by species). The most important environmental variable type after spatial was anthropogenic in all functional feeding groups except carnivores. Each environmental variable type was, however, selected as most important in at least two species distribution models. Similarly, spatial variables were most often selected as most important across primary habitat type. Interestingly, and as would be expected, oceanic variables were never selected as most important for species classified as land or shore dwelling species.  31  3.3.4 Conservation Applications  Figure 3.6: Mean species richness and Beta diversity for the entire study period. A missing coastal site indicates no oceanic data was available for that location for any year during the study period (coastal coverage = 89.4%). Of the 60 species examined, there were 16 detected changes in occupied habitat throughout the eight year study period (see Appendix 1 for each species? rate of change). Predicted occupied habitat of 11 species significantly declined, while occupied habitat for five species increased. Mean species richness varied from 6.75 to 47.75 species, while mean Beta diversity varied from 0.04 to 0.80 (Figure 3.6). Mean species richness and mean Beta diversity were strongly negatively associated (P<.001, R2=.61; Beta=0.75-0.015*richness).  Species richness was modeled against both anthropogenic variables to examine the effects of human activities. Richness tended to increase in a non-linear fashion as human influence decreased; however, the influence of road networks on coastal bird richness became less pronounced after 32  approximately 30 km, while the effects of cities was more constant even in more remote locations (Figure 3.7). Richness was estimated to be lowest in areas proximal to human development in the south portion of the coast, and highest in the remote central coast region. Log transformations on species richness and both anthropogenic variables produced highly significant correlations in both cases (P < .001), with (log) distance to roads explaining (r2) 14% of variation in (log) species richness, and (log) distance to cities explaining (r2) 13% of variation in (log) species richness. Collectively, the log transformed anthropogenic variables explained (R2) 34% of the variation in (log) species richness in multiple linear regression.  Figure 3.7: The species richness ? distance to road (A) and distance to cities (B) model relationships. Distance to road half-max = 10.7 km, asymptote = 41.1 species. Distance to city half-max = 458.8 km, asymptote = 65.7 species. 3.4 Discussion The effects of modeling species distributions using single and ensemble models, as well as the inclusion of spatial variables was tested to determine how to best estimate coastal bird 33  distributions. Information regarding model variability, coastal bird species distributions, available habitat, species richness and Beta diversity was inferred from these models. 3.4.1 Single vs. Ensemble Species Distribution Models  Both model type and the inclusion of spatial covariates were tested to determine their effects on coastal bird distribution model AUC values. Species distribution model AUC values were found to vary significantly depending on which model was used. The two ensemble models found to be most effective by Marmion et al., (2009) produced significantly better AUC values than any single model, except boosted regression trees, which agrees with the multiple studies that have also demonstrated how ensemble algorithms can improve model fit (Ara?jo et al., 2005; Pearson et al., 2006; Ara?jo and New, 2007). Boosted regression trees have also been found to be highly effective at fitting distribution models (Elith et al., 2006, 2008; Elith and Graham, 2009; Capelle et al., 2010) making its inclusion amongst the top performing model types not entirely surprising and supports its future use as a single model. More unusual, however, was the poor performance of the Random Forest model, which is also commonly among the most powerful single model types employed in SDMs (Breiman, 2001; Marmion et al., 2009). Both classification models (RF and CTA) were among the least effective models, being outperformed by all other model types save ANN. An important difference between classification techniques and approaches involving regression in some form (i.e. GLM, GAM, GBM, and MARS) is that regression may be able to estimate a species response curve more accurately beyond the range of data than classification techniques (Thuiller et al., 2004).  The inclusion of spatial covariates also significantly improved model AUC values. SDMs lacking spatial explicit factors have been described as potential habitat models (or fundamental niche models) due to their lack of consideration for dispersal limitation, competition and other biotic factors, while models including spatial variables have been referred to as potential geographical distributions (or realized niche models) since these models include spatially explicit factors (Guisan 34  and Zimmerman, 2000; Ara?jo and Guisan, 2006). Since SDMs based on observation data model realized niches (Pearson and Dawson, 2003; Pearson et al., 2006), models which incorporate spatial covariates may have an advantage due to their inclusion of spatial variables that address potential dispersal limitation (Guisan and Thuiller, 2005), as was the case here. Commonly, a spatial variable such as latitude may also act as surrogate for an un-described environmental variable such as temperature or productivity. The importance of spatial variables in this study may, at least in part, owe itself to the environmental variables not fully describing the environmental conditions at each site (Cottenie, 2005). No difference was detected for the weighted mean ensemble model?s AUC values across major habitat type or detection location of a species. However, mean AUC values were found to vary owing to functional feeding group, with insectivores having a significantly larger mean AUC than piscivores, and potentially carnivores although this test was non-significant. The insectivorous birds in this study are all considered shorebirds, which typically have very strong affinities for certain types of coastal habitats, in particular coastal mudflats. These habitats are fairly rare along the BC coast, and it could be that the models were able to identify these habitats well using spatial variables such as latitude, longitude, and the anthropogenic metrics, allowing for higher mean AUC results. Additionally, the ease of detection of these species, owing to their large flock sizes and onshore habitat usage may have increased model accuracy by providing better detection of presences and absences versus some of the more offshore species. Shorebird were found to have the highest mean AUC by primary habitat type, while land birds were found to have the lowest mean AUC. This may reflect the generalist nature of multiple species included in the land bird category (bald eagles (Haliaeetus leucocephalus), common ravens (Corvus corax), and northwestern crows (Corvus caurinus) for example) making detecting strong environmental associations, which tends to improve correlative models, difficult. 35  3.4.2 Spatial Variability of Model Predictions The variability of model predictions was found to be highly related to the distance of a particular site from survey locations. This may be due, in part, to environmental conditions occurring on the relatively un-sampled north and central coast which were outside the range represented by the sampled locations on the south coast. Attempting to project models built mainly on data at sites which do not reflect the full range of variability for a study region can lead to more variable, and erroneous, model predictions (Elith and Graham, 2009; Elith and Leathwick, 2009). Randin et al. (2006) examined SDM transferability in space and found that model predictive ability (using GLM and GAM single models) was significantly reduced when transferring models to new regions, most likely due to overfitting of the models to their training data. Both the results described here, as well as those of Randin et al. (2006), underscore the need for careful analysis when projecting models into un-sampled locations. A strength of the approach used here is the large number of models considered as well as the multiple iterations of each model. Although model predictions became more variable with increasing distance from survey locations, with multiple (70) model predictions, the mean of these predictions (the resulting probability calculated for both ensemble models) provides a reasonable estimate of a site?s actual probability of occurrence, provided the predicted probabilities are not biased in one direction. If users of SDMs are particularly concerned with model variability at its effects on the predictive abilities of ensemble models, a more conventional method of assigning presence or absence may be to use the confidence intervals (which can be derived from the SD or directly using BIOMOD2) of ensemble predictions rather than the mean. As an example, a species could be assigned as present if the entire range of the 95% confidence interval falls above the given cut-off value, or absent if the entire 95% confidence interval is below. The limitation of this approach would be some ambiguous predictions where the 95% confidence interval spans the cut-off value. 36  The model variability-distance relationship can also be used to empirically determine to what distance model projection can be considered accurate. Users could set a percentage increase in model variability that they are willing to tolerate (as an example a 50% increase in SD from the model intercept SD occurring at the minimum distance value) and allow projections out to this distance. In this case, that distance would be any site within 300 Km, on average, of all other sites (using the average distance relationship). This approach would require, however, an arbitrary decision regarding the increase in model variability users are willing to accept. Ultimately, model accuracy is the desired metric to assess when projecting spatially, which requires a second data set with occurrence records in certain un-sampled locations. Interestingly, model predictions became more variable towards the terminal end of multiple coastal fjords. This is most likely due to the difficulty of obtaining oceanic environmental estimates along complex and narrow sections of coast (discussed in section 3.4.3). In many cases only one to five data points were obtained for coastal sites located towards the end of fjords, contrasted against 30 -90 data points for more open ocean sites, resulting in less accurate estimates of mean oceanic conditions at those sites and increased model prediction variability. 3.4.3 Ensemble Model Environmental Variable Selection Environmental variables were selected to represent a wide range of environmental conditions due to the diverse community of species being considered, with each variable being selected as a direct estimate of local environmental conditions dictating habitat or nutritional availability, or thermoregulatory constraints. Each category of environmental variables was selected as most important in multiple models, supporting our hypothesis that a wide range of environmental correlates should be employed when modeling a community such as coastal birds. Further, I argue that distribution models should consider and incorporate this new, freely available, suite of 37  remotely sensed variables which cover additional dimensions of the environment beyond terrestrial vegetation productivity (such as the Normalised Difference Vegetation Index (NDVI)). The Dynamic Habitat Index (derived from the MODIS constellation) and EOSD Land Cover (derived from Landsat 7 data) have both been shown to be useful predictors of both single species distributions and species richness (Coops et al., 2009a; Coops et al., 2009b), and both were selected as important covariates in multiple models, as was the ClimateNWA data, derived from recent global circulation and PRISM models (Daly et al., 2002; Wang et al., 2006). Lippitt et al. (2008) argued that SDMs must incorporate anthropogenic effects due to their ability to influence species distributions, and our findings agree, as anthropogenic variables were selected as the most important environmental category more often than any other. The high temporal repeat of the oceanic and atmospheric data, which is another advantage when considering remotely sensed data, also allowed for model improvement. The near daily repeat of MODIS provided a large quantity of oceanic data by winter which was linked to species data along with the interpolated climate data. This was a unique application of MODIS aquatic environmental estimates and their inclusion as important variables (Figure 3.2) in multiple species models supports their use in future coastal SDM scenarios. The SDMs built here do share some of the common shortcomings of many distributional studies. We acknowledge that our models incorporate verification data as opposed to validation data (Ara?jo and Guisan, 2006). Given the data available, the data split, coupled with the iterative splitting of the training data did provide high quality model verification. The CWS survey locations also presented a challenge, being clustered along certain sections of coast (Figure 2.1). Again, the iterative splitting of the training data acted as a form of subsampling, helping address spatial bias (Ara?jo and Guisan, 2006). The largest single challenge for producing accurate distribution predictions here was the difficulty obtaining oceanic environmental products in a coastal setting. As described earlier, at 38  most only 75% of the coast was able to be modeled during any given year, restricting our ability to get full assessments of any species entire coastal BC distribution. This issue may be more pronounced among species which select for sheltered inlets, fjords, and other sections of coast which form complex shapes, as these are the most difficult areas for oceanic remote sensing to gather data. Presence of land in a pixel, shallow water, excessive wave action, and sun glint can all affect the MODIS sensor?s ability to accurately estimate oceanic variables, leading to pixels being masked when one of these conditions occur. This difficulty leads to a loss of useable data, particularly when considering water which is in close proximity to a coastline. An interpolation method such as kriging might offer a solution in situations such as is presented here, but was not tested.  3.4.4 Conservation Applications and Individual Species Models The distribution models produced in this study have similar use to those described elsewhere, such as: protected areas planning (Hernandez et al., 2006), species at risk assessments (Engler et al., 2004), and examination of range shifts (Pearson and Dawson, 2003; Pearson et al., 2006), with one important addition. Because habitat occupancy was able to be estimated by year, trend assessments through time were possible. These % changes in occupied habitat through time analyses were done more to demonstrate the power of this approach for monitoring rather than actual trend detection, but the results are notable. Significant changes in occupied habitat were detected in 16 out of 60 species, with 11 species percent habitat decreasing and five increasing. These findings agree with those of Crewe et al., (2012) in their assessment of coastal bird populations in the Georgia Basin. It should be noted that our analysis was for the entire coast however, and habitat and population trends are not always linked. Also, an analysis of overall habitat should include an assessment of habitat quality, which was not done here.  39  A comparison of trends between change in occupied habitat estimated here and Bower?s (2009) estimates of changes in waterbird abundance in the Pacific northwest yielded 13 species where both studies detected significant changes (% occupied habitat in this study vs. abundance in Bower (2009)) in the same direction, or no significant changes, and 20 species where findings differed. Significant declines detected in both studies were for: black scoter (Melanitta Americana), bufflehead (Bucephala albeola), and lesser scaup (Aythya affinis). Disagreement between models in most cases (14 out of 20) involved similar estimates of the direction of change (+ vs. -); however, significance was found in one circumstance and not the other. The short time frame of this study may have limited its ability to detect long term changes in habitat when compared with Bower (2009) which analyzed changes from 1979 to 2005. Although species richness and Beta diversity have commonly been found to be positively related (Gering and Crist, 2002) the negative relationship found here between mean species richness and mean Beta diversity is expected when considering a small group of species (60). As richness increased in an area along the coast, the likelihood that sites within that area shared species increased, and since rarities were not considered here, once species richness reached a maximum, sites must have shared all species, meaning Beta diversity was equal to zero. Jost (2007) described the same phenomenon when using data sets with relatively low levels of species richness, and the inevitability of having similar species present in two plots when richness counts are both high compared to the rest of the data set. Despite this, the mean species richness and Beta diversity products as well as their relationship can provide useful conservation information, as sites falling above the fitted linear relationship ? indicating higher estimated Beta diversity than expected given that sites? species richness ? can be identified for further investigation or conservation. Additionally, the averaging through time for each of these products allowed for better coastal coverage than in any one year (89% compared with a maximum of 75% in 2008). 40  SDMs are only useful after careful scrutiny of their ecological relevance, usually involving an examination of the variables selected for model building and their consistency with known species ecology and life history requirements (Elith and Leathwick, 2009). Here, we acknowledge that certain models may be more applicable than others. The western grebe (Aechmophorus occidentalis), which is a species that has experienced significant declines over the past 50 years (Bower, 2009), produced a model where the three most important variables were particulate organic carbon, average winter temperature, and average winter precipitation respectively. Western grebes are piscivores, and the importance of particulate organic carbon in the model could be viewed as providing a food source for their prey species. The model accuracy for the western grebe was poor however (AUC=.66). The associations found in this model are plausible given the species ecology. Despite the attractiveness of an SDM in this circumstance, the usefulness of this particular model is questionable given its poor fit.  Dunlin (Calidris alpina), which are a species of ecological significance in southern BC coast (Butler and Vermeer, 1994), were modeled with a relatively high degree of accuracy (AUC=0.84). Dunlin tend to arrive on the BC coast in mid-October, with approximately 30 000 - 60 000 individuals overwintering in the Fraser river delta in southern BC (Shepherd, 1997). The distribution estimates provided here reflects their known distribution fairly well, with the majority of sites where dunlin were estimated to be present occurring in southern BC on or near the Fraser River Delta, with a second grouping of presence sites occurring in Haida Gwaii where dunlin are also known to occur. The most important variable in the dunlin model was longitude which most likely represents their selection for mud flats along the BC coast, which are few. Distance to cities was the second most important variable, which again may act as a proxy measure for distance to mud flats as most are concentrated in the Fraser river delta which is also close to multiple cities. It is unlikely that this species actually selects for habitat near cities directly, therefore, while the model does a fair job at 41  representing dunlin?s coastal distribution, the ecological knowledge which can be gained from the model is limited.  A good example of a species where model fit and environmental associations both seem to indicate the model is both accurate and represents real ecological relationships is the northern shoveler (Anas clyptea) model. The ensemble models both produced an AUC of .872, while the three most important variables were particulate inorganic carbon, distance to city, and mean productivity. The northern shoveler is an omnivorous waterbird, supporting its association with particulate concentrations in water, and is known to associate with cities. Other species SDMs, such as the rhinoceros auklet (Cerorhinca monocerata, AUC=.79), horned grebe (Podiceps auritus, AUC=.82), pacific loon (Gavia pacifica, AUC=.82) , glaucous-winged gull (Larus glaucescens, AUC=.89), and barrow?s goldeneye (Bucephala islandica, AUC=.91) ? all of which experienced significant declines in abundance throughout the Georgia basin over the same time frame as this study (Crewe et al., 2012) ? may also be more appropriate for conservation applications, but again, only after careful scrutiny of the environmental associations in each model.    42  4. ASSESSING CONSERVATION REGIONALIZATION SCHEMES: EMPLOYING SPECIES DATA TO AUGMENT THE ENVIRONMENTAL SURROGACY APPROACH 4.1 Introduction The decline of global biodiversity has been well documented (IPCC, 2007; Pereira et al., 2010) and as a result biodiversity has become a major driver of conservation planning (Knight et al., 2007; Rands et al., 2010). Generally, the method most often used for protecting biodiversity has been establishing reserve systems, which provide habitat for target species and separate them from threats present elsewhere (Margules and Pressey, 2000; Rodrigues and Brooks, 2007). For a reserve system to be successful, it should meet two criteria: it should (1) represent a region?s biodiversity and (2) ensure that the biodiversity it protects persists through time (Margules and Pressey, 2000). To address the first criteria, representation, a conservation strategy will often begin by sub-dividing the land base into conservation units/regions/land cover types which are expected to delineate environmental conditions, and thus distinct communities of target species (Leathwick et al., 2003; Oliver et al., 2004; Olson et al., 2001). Since species data are typically not available with adequate spatial coverage at the local scale most relevant for conservation planning, environmental proxies are commonly employed with the goal of representing distinct species assemblages (Margules et al., 2002). Once a conservation regionalization scheme has been developed, a reserve system can be designed with specific regional site prioritization (Belbin, 1993). This strategy employs the concept of complementarity ? that multiple reserves located within different regions of a planning area will be able to fully represent a large portion of that area?s biodiversity (Howard et al., 1998; Pressey et al., 1993; Rodrigues and Brooks, 2007). If the regionalization scheme is effective at delineating distinct communities of target species, then the resulting reserve system should conserve the full suite of 43  species without large amounts of redundancy occurring (Reyers et al., 2002). However, if the regionalization is ineffective at delineating distinct faunal communities, the reserve system risks missing important species and/or requiring more land than is necessary for full species representation (Fox and Beckley, 2005). Considerations when developing a conservation regionalization should also include thematic resolution (i.e., the number of regions within a regionalization) (Pearson and Carroll, 1999; Pharo and Beattie, 2001). With too fine a thematic resolution, the area requirements for a reserve system may be excessively large with high levels of redundancy between reserves (Andrew et al., 2011). Conversely, a regionalization scheme with an excessively coarse thematic resolution risks missing species protected in a reserve system due to difficulties representing the full variability of large, heterogeneous regions (Pressey and Logan, 1994; Wiens, 1989). A range of methods have been proposed for developing conservation regionalization schemes, each of which employs their own suite of environmental variables and rules for delineation, leading to inconsistent partitioning of the landscape (Bailey et al., 1985; Leathwick et al., 2003; Mackey et al., 2008; Pressey and Logan, 1994; Oliver et al., 2004; Olson et al., 2001). Complicating issues further, the effectiveness of these schemes to act as a surrogate for delineating distinct communities of target species has been called into question due to problems surrounding community spatial autocorrelation and regionalizations not capturing environmental variables which are meaningful to community composition (Bonn and Gaston, 2005; Kirkpatrick and Brown, 1994; Lombard et al., 2003; but see Trakhtenbrot and Kadmon, 2005).  As the goal of a regionalization is to delineate distinct ecological communities (Margules et al., 2002), it follows that the core areas of each region should have low levels of species turnover, or Beta diversity (Whittaker, 1972), while the border areas should experience higher levels of Beta diversity, since these should be areas of transition between communities. For this reason, Beta 44  diversity metrics such as Jaccard?s and Euclidean distance have been used in multiple studies to evaluate different methods for delineating land bases into distinct groups of species (Anderson et al., 2011; Andrew et al., 2011; Tuomisto, 2010b).  An alternative approach to developing environmental regionalization schemes, which is becoming more feasible with the increase in available, spatially expansive community data, is to delineate regions within a regionalization scheme based on where community Beta diversity values are highest. A species-based approach allows for the delineation of regions of distinct communities using observed species data without relying on environmental surrogates. Such a species-based regionalization scheme could act as a best case scenario for comparison with environmental regionalization schemes. Data sets such as bird and butterfly atlases, which focus on a particular taxon for one period in time and one region (i.e., a state or province), and breeding bird surveys are allowing different approaches for developing conservation regionalization schemes to be examined against one another in an empirical manner. As such, the primary goals of this paper are: first, to analyze the performance of different types of environmental regionalization schemes, and second, to develop a regionalization scheme built using Beta diversity gradients for comparison with the previously tested environmental regionalizations. The following questions were asked with regard to conservation regionalization schemes: (1) from a selection of regionalization schemes employing environmental surrogates, which most effectively delineates Beta diversity? (2) Which environmental regionalization scheme is most similar to a regionalization scheme built using species data directly? (3) Which thematic resolution optimizes Beta diversity delineation within the study region? These questions were addressed within British Columbia, Canada, using data collected for the BC Breeding Bird Atlas project. 45  4.2 Methods 4.2.1 Bird Community Data 4.2.1.1 The BC Breeding Bird Atlas The bird species data used here was obtained from the BC Breeding Bird Atlas data set (for a detailed data description see Section 2.2.2; Figure 2.2). Highest breeding evidence data from 2008-2011 were used for this study (2012 data are not yet available) as they summarize the most conclusive evidence of breeding from the all breeding evidence data set, producing one presence/absence value for each species within each cell. 4.2.1.2 Bird Community Data Analysis Citizen science data sets are known to contain biases due to uneven sampling effort (Andrew et al., 2011; Moerman and Estabrook, 2006), sampling intensity (Schulman et al., 2007) and surveyor skill (Ahrends et al., 2011). The confounding effects of variable sampling effort (measured as hours spent surveying each cell) and intensity (the number of adjacently sampled cells ? including diagonals) on observations of bird community composition were tested using partial Canonical Correspondence Analysis (pCCA) (Borchard et al., 1992; Palmer, 1993; Ter Braak, 1986) to evaluate the amount of variation in Atlas data explained by different regionalizations which was also explained by sampling covariates. Survey effort recordings were missing from 827 cells, which were therefore excluded from these analyses, leaving 2720 cells. Effort and intensity were partitioned out individually and collectively during the pCCA analyses to test their ability to explain community composition, as well as shared explanatory power with each regionalization scheme tested. Additionally, simple linear regression was used to evaluate if per-region performance (see section 2.3.1) was influenced by sampling coverage by region. 46  4.2.2 Environmental Regionalization Schemes Four environmental regionalization schemes (Figure 2.3) and one spatial regionalization scheme were tested, each of which represents a different method for partitioning a land base using environmental, spatial and abiotic variables at varying thematic resolutions (number of regions within a regionalization) and with differing spatial constraints (contiguous regions vs. non-contiguous regions; Table 1). 4.2.2.1 The North American Bird Conservation Regions The BCR system is described in Section 2.4.1. There are five spatially contiguous BCRs contained within BC, largely corresponding to the Ecozones put forward by the National Ecological Framework for Canada (Ecological Stratification Working Group, 1995) ? described in more detail in the following section. The BCRs have been used in planning exercises by Environment Canada (NABCI Canada, 2012) and the United States Fish and Wildlife Service (NABCI U.S. Committee, 2009), among others. To our knowledge, the BCR system has not been assessed to determine how effective it is at delineating distinct North American bird communities, which is surprising, given that this was the original purpose of the project.  4.2.2.2 The Canadian Ecoregions The Canadian Ecoregions are described in Section 2.4.2. There are 44 spatially contiguous Ecoregions in British Columbia. Tests of the effectiveness of Ecoregions for delineating distinct communities of organisms have been mixed. McDonald et al. (2005) found little evidence that Ecoregions are effective at delineating distinct bird, mammal or tree species. Supporting evidence, however, has been found for bumblebees (Williams, 1996), birds (Williams et al., 1999) and butterflies (Andrew et al., 2011). 47  4.2.2.3 The BC Biogeoclimatic Zones The BEC system is described in Section 2.4.3. To our knowledge, BEC has not been tested for its ability to delineate distinct communities of organisms, although similar types of regionalizations have been developed recently (see Mackey et al., 2008).  4.2.2.4 Remotely Sensed Unique BC Ecosystems The unique BC ecosystems regionalization is described in Section 2.4.4. The thematic resolutions tested were: 5, 7, 10, 14 and 17 classes produced at a 10 km spatial resolution to match the avian community data.  4.2.2.5 Spatial Grids Proximity in space is known to be a significant determinant of community composition (Nekola and White, 1999) and as such communities typically are less similar as the distance between them increases (Andrew et al., 2012). It is common for species composition to be spatially autocorrelated due to autocorrelation in abiotic conditions and/or dispersal limitation (Legendre, 1993; Nekola and White, 1999). In fact, it has been argued that community spatial autocorrelation is necessary for community functioning and should be considered a positive, rather than a statistical nuisance (Legendre, 1993). To examine the effects of space on community structure, eight square grids ranging from 100 to 800 km resolution were developed and used as regionalization schemes that did not explicitly consider environment variables (Andrew et al., 2011). This allowed for a repeatable comparison of the effects of thematic resolution on community spatial structure without any consideration of environmental effects. 48  Table 4.1: A summary of each environmental regionalization scheme and selected remotely sensed and grid system regionalization schemes. Thematic resolution refers to the number of regions within each regionalization scheme containing >2 sampled Atlas cells. Cell number refers to the minimum or maximum number of sampled cells within a particular region for each regionalization. Regionalization Scheme Thematic Resolution Mean Region Size (Km2) Min. Pixel Number Max. Pixel Number Contiguous Regions BCR 5 188947 242 1566 Yes Ecoregion 44 21471 6 392 Yes BEC 14 67481 29 647 No Beta 12 78728 112 556 Yes RS 5 5 188947 52 1877 No RS 17 17 55573 7 635 No 100 Km Grid 5 188947 2 82 Yes 800 Km Grid 121 7808 24 1319 Yes 4.2.3 Community Analyses Each conservation regionalization scheme was tested on its ability to delineate Beta diversity using the Analysis of Similarity test (ANOSIM) developed by Clark (1993), while Jaccard?s distance (identified as a suitable dissimilarity measure by Anderson et al., (2011); Tuomisto, 2010; see formula below), which is a measure of species not shared between sites in a pairwise comparison and does not consider joint absences, was used to represent overall Beta diversity.                     (   )  (   )(   ) ANOSIM tests whether values (Jaccard?s distance in this case) contained within regions are significantly different than those from between regions. This is done in a non-parametric fashion: first, by rank ordering all Jaccard?s distances from both within and between regions, and second, by taking the difference of mean within region rank and mean between region rank, standardized by sample size (See formula below). The ANOSIM test produces an R statistic ranging from -1 to 1. A value of 1 indicates that all within region distances were ranked lower than all between region distances, while -1 indicates the opposite. A value of 0 indicates no difference in mean rank for within or between region distances.  49                      (                                            ) (   )   The ANOSIM test is desirable as it makes no assumptions regarding the distribution of distances and provides an R statistic which allows for regional comparisons beyond significant/non-significant results. Significance was tested by permuting the data 1000 times and evaluating the calculated R statistic against the randomly generated distribution of R statistics (?=.05; see formula below).   (     (                        ) (               )  Four different ANOSIM tests were performed based on Andrew et al.?s (2011) analysis of regionalization schemes using butterfly community data. The first two, termed ?Grand? ANOSIMs, analyzed the ability of each regionalization scheme as a whole to partition British Columbia into regions of distinct avian communities. Grand Overall (GO) tested all regional within and between distances, providing one ANOSIM R statistic for each regionalization. Similarly, Grand Adjacent (GA) produced a single R statistic for each regionalization using all within region distances, but restricted between region distances to regional pairs that are adjacent in physical or environmental space.  Not all regions within a regionalization are expected to perform equally, however, which prompted the two additional sets of tests, termed ?Regional? ANOSIMs, of individual regions within each regionalization scheme. The Regional Overall (RO) test evaluated a given region?s within distances against all between region distances to analyze each region?s contribution to overall Beta diversity delineation. Lastly, the Regional Adjacent (RA) test evaluated a region?s within distances against the between distances shared with its neighboring regions in physical or environmental space, assessing regional redundancy when considering only the regions expected to be most similar to it. 50  4.2.4 BC Avian Conservation Regions Delineation  4.2.4.1 Beta Diversity Spatial Modeling Since Jaccard?s distance is a pairwise comparison of sites located at two different points in space, a single Beta diversity value calculated between sites cannot be located spatially. Modeling Beta diversity at a point in space requires averaging pairwise Beta diversity values between that point and multiple other points. This study used the average pairwise Jaccard?s distances between each cell and its direct neighbors ? including diagonals ? within a 3x3 moving kernel (Figure 4.1). Beta diversity was only estimated for cells with at least two neighbors (yielding a total of 3377 cells used to map average Beta diversity).  Figure 4.1: Schematic diagram of the averaging kernel calculating local mean Beta diversity. Due to gaps in the spatial coverage of the Atlas data, interpolation was required prior to delineation of the land base along Beta diversity gradients. The map of local average Beta diversity was tested for spatial autocorrelation using a semi-variogram, which estimates the similarity of points 51  (semivariance) within a given distance. Eight different model fits were iteratively tested, with the lowest Root Mean Squared Error model selected for (model = exponential, range = 200 km, sill = .023, RMSE = 2.56x10-3; see Figure 4.2). Universal kriging was then carried out to estimate Beta diversity of all non-sampled cells (6070 non-sampled cells were estimated) (Olea, 1974; Pearson and Carroll, 1999). Beta diversity contours were identified using the isolines tool in ArcGIS 10.0 (ESRI, 2011).  Figure 4.2: The experimental semi-variogram of local mean Jaccard?s distance values with the fitted exponential model (range=200 km, sill=.023, RMSE= 2.56x10-3). 4.2.4.2 Species-Based Conservation Regions Delineation A new, species-based, spatially contiguous conservation regionalization scheme was developed using the Beta diversity map (Figure 4.3) and the thematic resolution indicated as optimal in the testing of Fitterer et al.?s (2012) environmental regionalization schemes and the spatial grid regionalization schemes (12 regions were delineated; see section 4.3.2). Regions were delineated manually along zones of high turnover, identified using Beta diversity isolines (Figure 4.3). The performance of the species-based regionalization scheme was then compared to the environmental regionalizations using all four ANOSIM models.  52   Figure 4.3: (A) The kriged local Beta diversity values using the modeled semi-variogram (Figure 4.2), (B) the derived high Beta diversity (.8, .85, .9) and low Beta diversity (.45, .5, .55) isolines, and (C) the Beta diversity regions derived using the high Beta diversity isolines (shown). 4.2.5 Statistical Analyses All ANOSIM tests as well as the averaging kernel were written by the authors in R version 2.15.1 (R Core Development Team, http://www.r-project.org/). Functions in the ?vegan? add-on package were used to calculate Jaccard?s distances and to perform pCCA (Oksanen et al., 2008). Universal kriging was carried out using functions in the add-on package ?gstat? (Pebesma, 2004). All statistical tests, except where indicated as different, employed an alpha level of 0.05. 4.3 Results  4.3.1 BC Breeding Bird Atlas Analyses Survey effort (Figure 4.4) and intensity (mean = 5.3 neighbouring sites, SD = 1.3) explained low yet significant variation in the BC breeding bird atlas data through pCCA (1.49% variation explained, pCCA P < 0.005; data not shown). Effort (hours surveyed) explained 0.78% of community variation, while intensity (n neighboring sites), 0.91%. On average, the estimated ability of environmental 53  regionalizations to represent avian communities was 88.34% (SD = 7.21%) uncontaminated by the effects of sampling effort and intensity. Given the low explanatory power of effort and intensity and lack of shared explanatory power between these covariates and the regionalization schemes, all survey points, regardless of recorded effort in the ANOSIM analyses were used during further testing.    Figure 4.4: Frequency distribution of hours of effort across the 3547 sample Atlas cells. 4.3.2 ANOSIM Analyses All regionalizations that incorporated measures of ecological variation (i.e., excluding the purely spatial grid regions) delineated highly significant patterns in Beta diversity (P < .001) (Figure 4.5). When considering all within and between distances (GO), the Ecoregions were the most effective at delineating Beta diversity (R = .27), followed closely by the regions constructed using Beta diversity values (R = .26). However, when restricting between distances to neighboring regions (GA), the regions constructed using Beta diversity values were most effective at delineating patterns in Beta diversity (R = .20), followed by the Bird Conservation Regions (R = .18) and the remotely sensed regions at a thematic resolution of 14 (R = .17). A peak in ANOSIM R values was evident for the 54  remotely sensed regionalizations ? consistent between GO and GA tests ? occurring between 10 and 14 regions, indicating an optimal environmental thematic resolution in this range.  Figure 4.5: The Grand Overall (GO) and Grand Adjacent (GA) ANOSIM tests for (A) all environmental regionalization schemes and (B) the remotely sensed regions at varying thematic resolutions (all regionalizations delineated highly significant patterns in Beta diversity, P<.001). Also, the percentage of significant regions are shown (?=.05) for the Regional Overall (RO) and Regional Adjacent (RA) tests for both(C) the environmental regionalizations and (D) the remotely sensed regions at varying thematic resolutions. The relative performance of regionalizations when considering the proportion of statistically significant regions within each regionalization scheme (RO/RA) was similar to the R value results of the overall tests (GO/GA) in some cases, but not all (Figure 4.5). In three out of four cases for the environmental regionalizations, the RA test produced more statistically significant regions than the 55  RO test, which contrasts with the GO and GA findings where the GO values were larger in all four cases. A similar pattern to the GO/GA test was produced in the remotely sensed RO and RA tests; however, there were two peaks in ANOSIM R values at thematic resolutions of 7 and 14. The BEC zones produced the highest proportion of statistically significant regions in the RO test (93%), while all regions (100%) were significant for both the BEC and Bird Conservation Regions in the RA test. The per-region ANOSIM R values varied widely in both the RO and RA tests, from slightly negative (R > -0.1) to strongly positive (R > 0.6) (Figure 4.6).   Figure 4.6: Maps of the Bird Conservation Regions (BCR) and Beta diversity derived regions (Beta) showing per-region ANOSIM R values for both the Regional Overall (RO) and Regional Adjacent (RA) tests. Diagonal lines indicate a non-significant ANOSIM R value. When comparing the purely spatial schemes, the 100 km grid regionalization scheme had the highest recorded R statistic of any regionalization for the GO test (R = .34), with declining R statistic values for the GO test as thematic resolution became coarser (Figure 4.7). When considering 56  neighbors only (GA), the pattern was more variable with a slight increase in R statistic values as the thematic resolution coarsened. There was a large increase in R statistic values for both the GO (37%) and GA (48%) tests at 500 km (Figure 4.7 A) which corresponded to a thematic resolution of 12 regions.  Figure 4.7: (A) ANOSIM R values for both the Grand Overall (GO) and Grand Adjacent (GA) tests for all gird regionalizations (all regionalizations delineated highly significant patterns in Beta diversity, P<.001). (B) The percentage of significant regions for each grid regionalization for both the Regional Overall (RO) and Regional Adjacent (RA) tests. In contrast to the Grand tests for the grid system, the per-region tests (RO and RA) showed an increase in the number of statistically significant regions with decreasing thematic resolution for both the RO and RA tests (Figure 4.7 B). The RA test produced a larger increase from 100 to 800 km (40% to 100%), while the overall test was more consistent (63% to 79%) across different thematic resolutions. 4.3.3 Regional Coverage and Beta diversity delineation To test whether spatial coverage of a region had an effect on a region?s ability to delineate Beta diversity, the proportion of each region sampled by atlas cells was compared with per-region ANOSIM R values (RO/RA) for all environmental regionalization schemes, as well as the species-based regionalization scheme, using simple linear regression (Figure 4.8). Six of the eight 57  regionalization schemes revealed positive relationships between sampling coverage and ANOSIM R values, which was strongest for the BEC regionalization.   Figure 4.8: Linear models of the proportion of each region sampled by atlas cells (ranging from 0 or no coverage to 1 or complete coverage) for the Bird Conservation Regions (BCR), Ecoregions (ECR), Biogeoclimatic zones (BEC), and Beta diversity derived regions (Beta) regressed against the regions Regional Overall (RO) and Regional Adjacent (RA) ANOSIM R values.  Graphs lacking lines indicate a non-significant relationship. *P<.1, **P<.01, ***P<.001. 4.4 Discussion Our results indicate that patterns of Beta diversity within bird communities are structured both environmentally and spatially. All environmental and spatial regionalization schemes were able to capture significant patterns in Beta diversity, with the BCR system being most effective overall. 4.4.1 Evaluating Environmental Regionalizations Against a Species-Based Ideal Scenario The results of our ANOSIM analyses support the use of environmental regionalization schemes in conservation planning. The Ecoregions performed virtually identically to the species-based regionalization in the GO test; however, the species-based regionalization had an ANOSIM R value more than twice the Ecoregions or BEC zones in the GA tests. The regionalization that performed most similarly to the species-based regionalization across all tests was the BCR system, showing a 58  similar, moderate decline which was actually less than the species-based regionalization from GO to GA tests versus the more dramatic declines in the Ecoregions and BEC zones. This is encouraging, considering the amount of avian conservation planning currently underway in North America which employs the BCR system, and relates to Ferrier?s (2002) third strategy of testing regionalization schemes in data rich environments for improved regional conservation planning. While the assessment of the species-based regionalization may seem circular, in that data used to produce the Beta diversity map were then used during its assessment, the number of pairwise comparisons considered during the creation of the species-based regionalization scheme (17 958 pairwise comparisons) made up 0.30 % of the total pairwise comparisons used in the ?Grand? ANOSIM tests (over 6.2 million pairwise comparisons). Essentially, the ANOSIM tests considered regional dissimilarities as a whole, while the species-based delineation identified areas of abrupt species turnover in physically adjacent sites (i.e., regional contents vs. regional boundaries).  Caution should be applied, however, when interpreting the ANOSIM results on a per-region basis, as proportional coverage was shown to be significantly related to regional ANOSIM R values for most regionalizations. It is possible that volunteers preferentially sampled more desirable birding areas that harbor more distinct species assemblages, in which case the observed patterns reflect true spatial variation. If this relationship does represent a bias in the data, one option could be to adjust conservation targets based on deviations from expected ANOSIM RO and RA values (i.e. residual values by region in the models shown in Figure 4.7) for a given level of regional sampling coverage. 4.4.2 Spatial and Environmental Structuring of Avian Beta Diversity All environmental regionalizations delineated highly significant patterns in Beta diversity, with the spatially contiguous environmental regionalizations performing best when considering the overall tests. This agrees with the findings of Andrew et al. (2011) that the Ecozones/Ecoregions system 59  was most efficient at delineating butterfly Beta diversity at the national scale, and contrasts with McDonald et al. (2005) who found that Ecoregions did not delineate Beta diversity effectively. The Ecoregions/BCR approach to regionalizations and their incorporation of multiple environmental variables agrees with several studies which suggest that multiple environmental inputs best represents diversity as a whole (Bonn and Gaston, 2005; Sarkar et al., 2005). However, the BEC system, which does not require contiguity within each region and produced the lowest overall ANOSIM R values, actually produced the highest percentage of non-redundant regions. The environmental findings of this study agree with the numerous studies which indicate that avian communities are structured to some degree by environmental factors which can be linked to plant communities, including: species richness and vegetation productivity (Bailey et al., 2004; Coops et al., 2009; Hawkins, 2004; Luck, 2007) and to plant structural complexity (Kissling et al., 2008).  The testing of the remotely sensed regionalization was able to address one of the questions poised in Andrew et al. (2011) ? whether increases in environmental variables were needed for improved remotely sensed regionalizations. Andrew et al. (2011) found that a regionalization scheme constructed along five remotely sensed variables (Coops et al., 2008) resulted in slightly poorer performance in the Grand ANOSIM tests than the remotely sensed regionalization of the present study, which was defined along twice as many variables but at the same thematic resolution. However, there are differences in study scale and taxa between the studies which may influence these patterns. Results from our other comparisons indicated that an increase in environmental variables does not necessarily improve the ability of remote sensing to produce regionalization schemes. Instead, our results indicate that the spatial structuring of faunal communities plays a significant part in community delineation, leading to the increased ANOSIM R values resulting from the spatially contiguous environmental regionalizations (in agreement with Andrew et al. (2011)). A remotely sensed regionalization scheme that explicitly incorporates spatial covariates may address this issue and warrants further study.  60  While the spatial dependence of ecological communities is a well described phenomenon, its role in planning conservation regionalization schemes has not been investigated in much depth (but see Andrew et al., 2011; Oliver et al., 2004). Ferrier et al. (2007) clearly describe the decay of similarity between sites owing to spatial separation irrespective of ecological dissimilarity. Our results support this concept and the importance of spatial structuring to conservation, with the 100 km grid regionalization scheme being most able to delineate Beta diversity of any scheme in the GO test, while all grid sizes delineated highly significant patterns of Beta diversity for both the GO and GA tests. This clearly indicates that avian communities are structured spatially, and I therefore argue that regionalization schemes that do not explicitly incorporate space are lacking a potentially critical variable in their creation, likely decreasing their ability to represent distinct groups of species.  4.4.3 Thematic Resolution Thematic resolution and its effects within regionalization schemes involve a trade-off between regional distinctiveness and internal heterogeneity (Andrew et al., 2011). As thematic resolution becomes coarser, internal heterogeneity tends to increase (Wiens, 1989) while regional redundancy will decrease. Striking a balance becomes difficult; however, our results indicate that estimating an optimal thematic resolution is possible for a given study area. By testing thematic resolution in two ways, considering only environment and no space (the remotely sensed unique ecosystems at varying thematic resolutions) and space without environmental considerations (the grid system at varying thematic resolutions), I identified the thematic resolution at which ANOSIM R values were maximized for both the GO and GA tests. The grid ANOSIM R values had a marked increase at 500 km, which corresponds to a thematic resolution of 12 regions, while the RS regions were maximized between 10 and 14 regions for these tests. The agreement between the RO and RA tests was less pronounced, with the grid regionalization showing a distinct increase in the 61  percentage of significant regions as thematic resolution coarsened, while the remotely sensed regionalizations were more variable.  A notable result of this study is the consistency found between the bird communities analyzed here and the butterfly communities tested in Andrew et al. (2011) when considering the shape of the curves for the GO and GA grid tests. In both studies a decreasing pattern in GO ANOSIM R values was observed, with a distinct increase at a particular grid size, corresponding to an increase in GA values as well (Figure 6, and see Figure 4 in Andrew et al., 2011). This increase was at 400 km for butterflies, and 500 km for birds, and I suggest that this scale difference is due to the increased vagility of bird species over butterflies, and could explain why these patterns fall out over larger scales. Additionally, this difference between birds and butterflies highlights the specificity of thematic resolution with regard to taxa, and evaluations of thematic resolution should be conducted with this limitation in mind.   62  5. CONCLUSIONS 5.1 Key Findings The work here outlines how citizen-science data sets can be used to explore ecological relationships and improve conservation planning. Two research questions were addressed using these freely available avian data sets. Chapter 3 outlined how remote sensing, coupled with interpolated climate indices and spatial variables, can effectively estimate species distributions in a complex coastal environment. Chapter 4 assessed different environmental regionalization schemes for their ability to delineate Beta diversity, and outlined how a second type of regionalization, built using species data directly, can also be used to assess the environmental surrogacy approach. In Chapter 3, ensemble models were found to be most effective at fitting SDMs against all single model types except boosted regression trees, which supports their continued use as a method for improving species distribution estimates when compared with traditional, single model approaches. The incorporation of spatial covariates significantly improved model fit, indicating that their inclusion is an important consideration when building species distribution models, and supports the concept of realized niche modeling in distribution models. No difference in model performance was detected across primary habitat type or primary detection location; however, mean AUC was found to vary across functional feeding groups. Model prediction variability was found to be highly dependent on average distance from survey sites, supporting the use of ensemble models with multiple iterations to estimate mean probability of occurrence. Model variability was also found to increase towards the ends of coastal fjords, highlighting the challenges of using oceanic remote sensing in coastal locations.  Each environmental predictor was selected as most important in, at minimum, two separate species models. Therefore, it can be argued that when modeling a large and diverse group of species, a wide range of environmental variables is necessary to describe the diverse habitat requirements of the 63  community in question. When working at the provincial scale, remotely sensed data was shown to provide effective estimates of many of the environmental variables necessary to build accurate species distribution models. Additionally, the use of remote sensing, with many repeat observations of an area per season, allowed for species data to be linked temporally to environmental data, which enabled species distributions and the amount of occupied habitat by species to be estimated by year. This approach identified 16 species (out of 60) which were estimated to have experienced significant changes in occupied habitat, with 11 estimated declines and five estimated increases in available habitat. In summary: ? Ensemble models offer improved distribution estimates when compared with most single model types ? The inclusion of spatial variables improves model estimates across all model types ? Ensemble model performance varied by functional feeding group, indicating that this approach may be more appropriate for species with certain life-history characteristics (insectivorous shorebirds) than others ? Model predictions become more variable away from survey site locations ? A wide range of environmental variables are necessary to model multiple species ? Remotely sensed environmental variables offer an effective means to build species distribution models at the BC coast scale ? By linking environmental and species data by year, occupied habitat for each species can be estimated allowing for inferences to be made with regard to habitat changes through time In Chapter 4, all four of the environmental regionalization schemes tested (BCRs, Ecoregions, BEC zones and remotely sensed regions) delineated highly significant patterns in Beta diversity, supporting their use in conservation planning, with the BCR system performing most similarly to 64  the species-based system when considering the GO and GA test results. When using the species-based regionalization as a benchmark, the BCR system would have to be considered the most effective environmental regionalization due to this similarity. Spatially contiguous regionalizations were superior to non-contiguous regionalizations in ANOSIM testing, indicating that space should be incorporated into a regionalization scheme explicitly by requiring regions to be contiguous parcels of land. This was supported by the spatial grid system, which also delineated significant patterns in Beta diversity. Additionally, the spatial grid system was used to test various thematic resolutions for delineation of Beta diversity values alongside remotely sensed regionalizations, allowing for an optimized resolution to be determined for British Columbia. Common spatial and environmental patterns were observed between avian and butterfly communities, reinforcing the findings of Andrew et al. (2011) and emphasizing the need for spatial consideration during the conservation regionalization scheme development process.  In summary: ? All environmental regionalization schemes effectively delineate Beta diversity, with the Bird Conservation Regions performing most similarly to the species-based regionalization ? The thematic resolution of a regionalization is an important consideration when during development, and as was outlined, can be tested empirically ? Regions within a regionalization should be spatially contiguous ? There are potential common patterns between birds and butterflies which warrant further investigation 5.2 Conservation Applications The research outlined in this thesis has the potential to be used in multiple avian conservation situations. Chapter 3 outlined how the species distribution models can be applied in systematic conservation planning, using both the species richness and Beta diversity indices. The integration of 65  oceanic and terrestrial reserve networks is increasingly being explored, and the models developed here could be integrated into systematic conservation planning along the BC coast. Coastal waterbirds may be uniquely positioned to act as a community, for which data exists, that requires both oceanic and terrestrial habitats. As such, their use in a reserve system which aims to conserve both oceanic and terrestrial habitat seems logical.  The ability to estimate available habitat by year and the continued supply of remotely sensed environmental variables could allow for a monitoring system to be developed which tracks available habitat by year for each species included in the modeling process. Caution should be exercised here however, as certain models were found to better estimate distributions than others (see dulin vs. western grebe), and certain species-environment associations were more plausible than others (see northern shoveler vs. dunlin). No differences in model performance were detected by functional feeding group, primary habitat type, and primary detection location, indicating the models are applicable across a wide range of species. The model variability-distance to survey sites locations allows for model users to assess the models ability to project in space, and could be used to set a limit on how far models can be projected away from survey locations. In Chapter 4, the BCR system performed most similarly to the species-based regionalization, supporting its future use in avian conservation. However, the optimal resolution for BC was found to be finer than that of the BCRs, and further refinement of this regionalization system should incorporate an analysis of thematic resolution. Ultimately, all regionalizations delineated highly significant patterns in Beta diversity, and could be considered candidates for acting as conservation regionalization scheme in an avian conservation network. 5.3 Future Research As more broad scale species data sets become freely available to researchers, the opportunity to answer questions at large scales will increase. Remote sensing offers a unique platform to gather 66  data at these large scales, with the added advantage of being repeated in time. As remote sensing archives increase in duration, the opportunity to develop time series analyses of both environmental conditions and species distributions will increase. Ultimately, models that monitor species distributions through time may allow for large scale monitoring of habitat availability where direct sampling is prohibitive. The research outlined in Chapter 3 was largely related to the process of modeling species distributions at the coastal scale; however, many potential ecological questions remain. How consistent are the patterns in occupancy across time? Are certain species predicted to exhibit high site fidelity while others are more variable in their predicted site selection? How consistent is model fit along the BC coast/ does model fit vary in space? Model validation with a second, independent data set and/or by experts in coastal bird distributions along the BC coast could answer many of these questions and should be further investigated. The continued development of the Coastal Waterbird Survey data set would allow for model refinement, with a focus on obtaining data in un-sampled areas along the central and north coast. Unfortunately, this may prove difficult given the nature of volunteer surveyors residing largely in, or near, urban areas.  The research described in Chapter 4 identified an interesting pattern in ANOSIM R values decreasing due to decreased thematic resolution until a spike at a particular value, which was shared by the avian community tested here and the butterfly community examined by Andrew et al., (2012). While this peak occurred at larger spatial scales for birds when compared with butterflies, the shape of the response was very similar. Testing this in other taxa, and/or in other environments may shed light on whether this represents an ecological pattern, or is simply coincidental. Ultimately, the continued use of citizen science data sets coupled with remotely sensed and spatial data will allow for more refined scientific questions to be addressed at large spatial scales. 67  REFERENCES Ahrends A., Rahbek C., Bulling M.T., Burgess N.D., Platts P.J., Lovett J.C., Kindemba V.W., Owen N., Sallu A.N., Marshall A.R., Mhoro B.E., Fanning E. & Marchant R. (2011) Conservation and the botanist effect. Biological Conservation, 144, 131-140. Anderson, M.J., Crist, T.O., Chase, J.M., Vellend, M., Inouye, B.D., Freestone, A.L., Sanders, N.J., Cornell, H.V., Comita, L.S., Davies, K.F., Harrison, S.P., Kraft, N.J.B., Stegen, J.C. & Swenson, N. G. (2011) Navigating the multiple meanings of ? diversity: a roadmap for the practicing ecologist. Ecology Letters, 14, 19-28. Andrew M.E., Wulder M.A. & Coops N.C. (2012) Identification of de facto protected areas in boreal Canada. Biological Conservation, 146, 97?107.  Andrew M.E., Wulder M.A. & Coops N.C. (2011) How do butterflies define ecosystems? A comparison of ecological regionalization schemes. Biological Conservation, 144, 1409-1418. Andrew M.E., Wulder M.A., Coops N.C. & Baillargeon G. (2012) Beta-diversity gradients of butterflies along productivity axes. Global Ecology and Biogeography, 21, 352-364. Ara?jo M.B. & Guisan A. (2006) Five (or so) challenges for species distribution modelling. Journal of Biogeography, 33, 1677?1688.  Ara?jo M.B. & New M. (2007) Ensemble forecasting of species distributions. Trends in Ecology & Evolution, 22, 42?7.  Ara?jo M.B., Whittaker R.J., Ladle R.J. & Erhard M. (2005) Reducing uncertainty in projections of extinction risk from climate change. Global Ecology and Biogeography, 14, 529-538. Austin M. (2002) Spatial prediction of species distribution: an interface between ecological theory and statistical modelling. Ecological Modelling, 157, 101?118.  68  Bailey R.G., Zoltai S.C. & Wiken E.B. (1985) Ecological regionalization in Canada and the United States. Geoforum, 16, 265-275. Bailey S.-A., Horner-Devine M.C., Luck G., Moore L.A., Carney K.M., Anderson S., Betrus C. & Fleishman, E. (2004) Primary productivity and species richness: relationships among functional guilds, residency groups and vagility classes at multiple scales. Ecography, 27, 207-217. Belbin L. (1993) Environmental representativeness: regional partitioning and reserve selection. Biological Conservation, 66, 223-230. Betts M.G., Forbes G.J., Diamond A.W. & Taylor P.D. (2006) Independent effects of fragmentation on forest songbirds: an organism-based approach. Ecological Applications, 16, 1076?89.  Blom G. (1958) Statistical estimates and transformed beta-variables. Wiley, New York.  Bonn A. & Gaston K.J. (2005) Capturing biodiversity: selecting priority areas for conservation using different criteria. Biodiversity and Conservation, 14, 1083-1100. Borcard D., Legendre P. & Drapeau P. (1992) Partialling out the spatial component of ecological variation. Ecology, 73, 1045?1055. Bower J.L. (2009) Changes in marine abundance?in the Salish Sea: 1975 to 2007. Marine Ornithology, 17, 9?17.  Breiman L. (2001) Random Forests. Machine Learning, 45, 5?32.  British Columbia Forest Service (2012) Biogeoclimatic Ecosystem Classification Program. <http://www.for.gov.bc.ca/hre/becweb/index.html>. 69  Buchanan J.B., Lyons J.E., Salzer L.J., Carmona R., Arce N., Wiles G.J., Brady K., Hayes G.E., Desimone S.M., Schirato G. & Michaelis W. (2012) Among-year site fidelity of Red Knots during migration in Washington. Journal of Field Ornithology, 83, 282?289.  Burger A.E., Ronconi R.A., Silvergieter M.P., Conroy C., Bahn V., Manley I.A., Cober A. & Lank D.B. (2010) Factors affecting the availability of thick epiphyte mats and other potential nest platforms for Marbled Murrelets in British Columbia. Canadian Journal of Forest Research, 40, 727?746.  Butler R.W., & Vermeer K. (1994) The abundance and distribution of estuarine birds in the Strait of Georgia, British Columbia. Canadian Wildlife Service Technical Report #83. Cannings R. & Cannings S. (1996) The Pacific Province. In: British Columbia: a Natural History. Greystone Books, Vancouver. Cappelle J., Girard O., Fofana B., Gaidet N. & Gilbert M. (2010) Ecological modeling of the spatial distribution of wild waterbirds to identify the main areas where avian influenza viruses are circulating in the Inner Niger Delta, Mali. EcoHealth, 7, 283?93.  Carrascal L.M. & Diaz L. (2006) Winter bird distribution in abiotic and habitat structural gradients: a case study with Mediterranean montane oakwoods. Ecoscience, 13, 100-110. Chapin III F.S., Zavaleta E.S., Eviner V.T., Naylor R.L., Vitousek P.M., Reynolds H.L., Hooper D.U., Lavorel S., Sala O.E., Hobbie S.E., Mack M.C. & Diaz S. (2000) Consequences of changing biodiversity. Nature, 405, 234-242.  Clark K.R. (1993) Non-parametric multivariate analyses of changes in community structure. Australian Journal of Ecology, 18, 117-143. 70  Clavero M., Brotons L., Pons P. & Sol D. (2009) Prominent role of invasive species in avian biodiversity loss. Biological Conservation, 142, 2043-2049. Conover M.R. (1983) Recent changes in ring-billed and California gull populations in the western United States. The Wilson Bulletin, 95, 362-383. Coops N.C., Waring R.H., Wulder M.A., Pidgeon A.M. & Radeloff V.C. (2009a) Bird diversity: a predictable function of satellite?derived estimates of seasonal variation in canopy light absorbance across the United States. Journal of Biogeography, 36, 905-918. Coops N.C., Wulder M.A., Duro D.C., Han T., & Berry S. (2008) The development of a Canadian dynamic habitat index using multi-temporal satellite estimates of canopy light absorbance. Ecological Indicators, 8, 754?766.  Coops N.C., Wulder M.A. & Iwanicka D. (2009) Exploring the relative importance of satellite-derived descriptors of production, topography and land cover for predicting breeding bird species richness over Ontario, Canada. Remote Sensing of Environment, 113, 668-679. Cottenie K. (2005) Integrating environmental and spatial processes in ecological community dynamics. Ecology letters, 8, 1175?82.  Crewe T., Barry K., Davidson P. & Lepage D. (2012) Coastal waterbird population trends in the Strait of Georgia 1999-2011: results from the first 12 years of the British Columbia Coastal Waterbird Survey. British Columbia Birds, 22, 8?35.  Cutler D.R., Edwards T.C., Beard K.H., Cutler A., Hess K.T., Gibson J. & Lawler J.J. (2007) Random forests for classification in ecology. Ecology, 88, 2783-2792. 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, 99-113 71  Dawson T.P., Jackson S.T., House J.I., Prentice I.C. & Mace G.M. (2011) Beyond predictions: biodiversity conservation in a changing climate. Science, 332, 53-58.  de Graaf R.M., Tilghman N.G. & Anderson S.H. (1985) Foraging Guilds of North American Birds. Environmental Management, 9, 493?536.  D?az S., Fargione J., Chapin F.S. & Tilman, D. (2006) Biodiversity loss threatens human well-being. PLoS Biology, 4, e277.  Dirzo R. & Raven P.H. (2003) Global state of biodiversity and loss. Annual Review of Environment and Resources, 28, 137-167. Drever M.C. & Martin. K. (2010) Response of woodpeckers to changes in forest health and harvest: implication for conservation of avian biodiversity. Forest Ecology and Management, 259, 958-966. Drever M.C., Aitken K.E.H., Norris A.R. & Martin K. (2008) Woodpeckers as reliable indicators of bird richness, forest health and harvest. Biological Conservation, 141, 624-634. Ecological Stratification Working Group (1995) A National Ecological Framework for Canada. Agriculture and Agri-Food Canada, Research Branch, Center for Land and Biological Resources Research and Environment Canada, State of the Environment Directorate, Ecozone Analysis Branch, Ottawa/Hull. Elith J. & Graham C.H. (2009) Do they? How do they? WHY do they differ? On finding reasons for differing performances of species distribution models. Ecography, 32, 66?77.  Elith J. & Leathwick J.R. (2009) Species distribution models: ecological explanation and prediction across space and time. Annual Review of Ecology, Evolution, and Systematics, 40, 677?697.  72  Elith J., Leathwick J.R. & Hastie T. (2008) A working guide to boosted regression trees. The Journal of Animal Ecology, 77, 802?13.  Engler R., Guisan A., & Rechsteiner L. (2004) An improved approach for predicting the distribution of rare and endangered species from occurrence and pseudo-absence data. Journal of Applied Ecology, 41, 263?274.  Esaias W.E., Abbott M.R., Barton I., Brown O.B., Campbell J.W., Carder K.L., Clark D.K., Evans R.H., Hoge F.E., Gordon H.R., Balch W.M., Letelier R. & Minnett P.J. (1998) An overview of MODIS capabilities for ocean science observations. IEEE Transactions on Geoscience and Remote Sensing, 36, 1250-1265. Ferrier S. (2002) Mapping spatial pattern in biodiversity for regional conservation planning: where to from here? Systematic Biology, 51, 331-363. Fitterer J. L., Nelson T. A., Coops N. C. & Wulder M. A. (2012) Modeling the ecosystem indicators of British Columbia using Earth observation data and terrain indices. Ecological Indicators, 20, 151-162. Fitzpatrick M.C. & Hargrove W.W. (2009) The projection of species distribution models and the problem of non-analog climate. Biodiversity and Conservation, 18, 2255?2261.  Foley J.A., DeFries R., Asner G.P., Barford C., Bonan G., Carpenter S.R., Chapin F.S., Coe M.T., Daily G.C., Gibbs H.K., Helkowski J.H., Holloway T., Howard E.A., Kucharik C.J., Monfreda C., Patz J.A., Prentice I.C., Ramankutty N. & Snyder P. K. (2005). Global consequences of land use. Science, 309, 570-574.  73  Fox N.J. & Beckley L.E. (2005) Priority areas for conservation of Western Australian coastal fishes: a comparison of hotspot, biogeographical and complementarity approaches. Biological Conservation, 125, 399-410. Friedman J.H. (1991) Multivariate Adaptive Regression Splines. The Annals of Statistics, 19, 1?67.  Gaston K.J., Blackburn T.M. & Goldewijk K.K. (2003) Habitat conversion and global avian biodiversity loss. Proceedings of the Royal Society of London. Series B: Biological Sciences, 270, 1293-1300.  Gering J.C. & Crist T.O. (2002) The alpha-beta-regional relationship: providing new insights into local-regional patterns of species richness and scale dependence of diversity components. Ecology Letters, 5, 433-444. Gibbs J.P. (2000) Demography versus habitat fragmentation as determinants of genetic variation in wild populations. Biological Conservation, 100, 15-20. Greenhalgh C.M. (1952) Food habits of the California gull in Utah. The Condor, 54, 302-308. Groves C., Valutis L., Vosick D., Neely B., Wheaton K., Touval J. & Runnels B. (2000) Designing a Geography of Hope: A Practitioners Handbook for Ecoregional Conservation Planning. The Nature Conservancy, Arlington. Guisan A. & Thuiller W. (2005) Predicting species distribution: offering more than simple habitat models. Ecology Letters, 8, 993?1009.  Guisan A. & Zimmermann N.E. (2000) Predictive habitat distribution models in ecology. Ecological Modelling, 135, 147?186.  Guisan A., Edwards Jr. T.C., & Hastie T. (2002) Generalized linear and generalized additive models in species distribution: setting the scene. Ecological Modelling, 157, 89-100  74  Guti?rrez J.S., Dietz M.W., Masero J. a., Gill Jr R.E., Dekinga A., Battley P.F., S?nchez-Guzm?n J.M. & Piersma T. (2012) Functional ecology of saltglands in shorebirds: flexible responses to variable environmental conditions. Functional Ecology, 26, 236?244.  Haines S.L., Jedlovec G.J. & Lazarus S.M. (2007) A MODIS Sea Surface Temperature Composite for Regional Applications. IEEE Transactions on Geoscience and Remote Sensing, 45, 2919?2927.  Hamman A. & Wang T. (2006) Potential effects of climate change on ecosystem and tree species distribution in British Columbia. Ecology, 87, 2773?2786.  Hastie T. J. (1992) Generalized additive models. In: Statistical Models in S (eds: Chambers J.M. & Hastie T.J.). Chapman and Hall, London. Hawkins B.A. (2004) Summer vegetation, deglaciation and the anomalous bird diversity gradient in eastern North America. Global Ecology and Biogeography, 13, 321-325. Hernandez P.A., Graham C.H., Master L.L., Albert D.L. & The A.D.L. (2006) The effect of sample size and species characteristics on performance of different species distribution modeling methods. Ecography, 5, 773?785.  Howard P.C., Viskanic P., Davenport T.R.B., Kigenyi F.W., Baltzer M., Dickinson C.J., Lwanga J.S., Matthews R.A. & Balmford A. (1998) Complementarity and the use of indicator groups for reserve selection in Uganda. Nature, 394, 472-475. IPCC (2007) Climate change 2007: impacts, adaptations and vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge. Janssen M.H., Arcese P., Kyser T.K., Bertram D.F. & Norris D.R. (2011) Stable isotopes reveal strategic allocation of resources during juvenile development in a cryptic and threatened 75  seabird , the Marbled Murrelet (Brachyramphus marmoratus). Canadian Journal of Zoology, 868, 859?868.  Jetz W. & Rahbek C. (2002) Geographic range size and determinants of avian species richness. Science, 297, 1548?51.  Jost L. (2007) Partitioning diversity into dependant alpha and beta components. Ecology, 88, 2427-2439. Kirkpatrick J.B. & Brown M.J. (1994) A comparison of direct and environmental domain approaches to planning reservation of forest higher plant communities and species in Tasmania. Conservation Biology, 8, 217-224. Kissling W.D., Field R. & Bohning-Gaese K. (2008) Spatial patterns of woody plant and bird diversity: functional relationships or environmental effects. Global Ecology and Biogeography, 17, 327-339. Klock R. & Mullock, J. (2001) Weather patterns of British Columbia. In: The weather of British Columbia. NAV CANADA, Government of Canada. Knight A.T., Smith R.J., Cowling R.M., Desmet P.G., Faith D.P., Ferrier S., Gelderblom C.M., Grantham H., Lombard A.T., Maze K., Nel J.L., Parrish J.D., Pence G.Q.K., Possingham H.P., Reyers B., Rouget M., Roux D. & Wilson K.A. (2007) Improving the key biodiversity areas approach for effective conservation planning. Bioscience, 57, 256-261. Leathwick J., Elith J., Francis M., Hastie T. & Taylor P. (2006) Variation in demersal fish species richness in the oceans surrounding New Zealand: an analysis using boosted regression trees. Marine Ecology Progress Series, 321, 267?281.  76  Leathwick J.R., Rowe D., Richardson J., Elith J. & Hastie T. (2005) Using multivariate adaptive regression splines to predict the distributions of New Zealand?s freshwater diadromous fish. Freshwater Biology, 50, 2034?2052.  Leathwick J.R., Overton J.M. & McLeod M. (2003) An environmental domain classification of New Zealand and its use as a tool for biodiversity management. Conservation Biology, 17, 1612-1623. Leathwick J.R., Rowe D., Richardson J., Elith J., & Hastie T. (2005) Using multivariate adaptive regression splines to predict the distributions of New Zealand?s freshwater diadromous fish. Freshwater Biology, 50, 2034-2052. Legendre P. (1993) Spatial autocorrelation: trouble or new paradigm? Ecology, 74, 1659-1673. Lek S., Delacoste M., Baran P., Dimopoulos I., Lauga J., Aulagnier S. (1996) Application of neural networks to modelling nonlinear relationships in ecology. Ecological Modelling, 90, 39-52. Lek S. & Guegan J.F. (1999) Artificial neural networks as a tool in ecological modelling, an introduction. Ecological Modelling, 120, 65-73. Lippitt C.D., Rogan J., Toledano J., Sangermano F., Eastman J.R., Mastro V. & Sawyer A. (2008) Incorporating anthropogenic variables into a species distribution model to map gypsy moth risk. Ecological Modelling, 210, 339-350. Lobb M.G. & Buckley J.R. (2012) Does MODIS Sea Surface Temperature accurately represent the temperature of the dynamically significant surface layer of the ocean? IEEE Geoscience and Remote Sensing Symposium, 2621?2624.  Lobo J.M., Jim?nez-Valverde A. & Real R. (2008) AUC: a misleading measure of the performance of predictive distribution models. Global Ecology and Biogeography, 17, 145?151.  77  Lombard A.T., Cowling R.M., Pressey R.L. & Rebelo A.G. (2002) Effectiveness of land classes as surrogates for species in conservation planning for the Cape Floristic Region. Biological Conservation, 112, 45-62. L?pez E., Aguilera P.A, Schmitz M.F., Castro H. & Pineda F.D. (2010) Selection of ecological indicators for the conservation, management and monitoring of Mediterranean coastal salinas. Environmental Monitoring and Assessment, 166, 241?56.  Luck G.W. (2007) The relationships between net primary productivity, human population density and species conservation. Journal of Biogeography, 34, 201-212. Mackey B.G., Berry S.L. & Brown T. (2008) Reconciling approaches to biogeographical regionalizations: a systematic and generic framework examined with a case study of the Australian continent. Journal of Biogeography, 35, 213-229. Margules C.R. & Pressey R.L. (2000) Systematic conservation planning. Nature 405: 243-253. Margules C.R., Pressey R.L. & Williams P.H. (2002) Representing biodiversity: data and procedures for identifying areas for conservation. Journal of Biosciences, 27, 309-326. Marmion M., Parviainen M., Luoto M., Heikkinen R.K. & Thuiller W. (2009) Evaluation of consensus methods in predictive species distribution modelling. Diversity and Distributions, 15, 59?69.  McDonald R., McKnight M., Weiss D., Selig E., O?Connor M., Violin C. & Moody A. (2005) Species compositional similarity and Ecoregions: do Ecoregion boundaries represent zones of high species turnover? Biological Conservation, 126, 24-40. McInerny G.J. & Etienne R.S. (2012) Ditch the niche ? is the niche a useful concept in ecology or species distribution modeling? Journal of Biogeography, 39, 2096-2102. 78  Mittelhauser G.H., Tudor L. & Connery B. (2012) Within-year movements and site fidelity of Purple Sandpipers during the nonbreeding season. Journal of Field Ornithology, 83, 32?40.  Moerman D.E. & Estabrook G.F. (2006) The botanist effect: counties with maximal species richness tend to be home to universities and botanists. Journal of biogeography, 33, 1969-1974. Moses W.J., Gitelson A.A., Berdnikov S. & Povazhnyy V. (2009) Estimation of chlorophyll- a concentration in case II waters using MODIS and MERIS data?successes and challenges. Environmental Research Letters, 4, 1-8.  NABCI Canada (2012) The state of Canada?s birds, 2012. Environment Canada. Ottawa, Canada. NABCI U.S. Committee (2009) The state of the birds, United States of America, 2009. U.S. Department of Interior. Washington, D.C. Nekola J.C. & White P.S. (1999) The distance decay of similarity in biogeography and ecology. Journal of Biogeography, 26, 867-878. Norris K. & Harper N. (2004) Extinction processes in hot spots of avian biodiversity and the targeting of pre-emptive conservation action. Proceedings of the Royal Society of London. Series B: Biological Sciences, 271, 123-130. North American Bird Conservation Initiative (1999) Strategy and Action Plan.  <http://www.bsc-eoc.org/nabcstrategy.html>. O?Hanley J.R. (2009) NeuralEnsembles: a neural network based ensemble forecasting program for habitat and bioclimatic suitability analysis. Ecography, 32, 89?93.  Oksanen J., Kindt R., Legendre P., O?Hara B., Simpson G.L., Solymos P., Stevens M.H.H. & Wagner H. (2008) Vegan: Community Ecology Package. <http://cran.r-project.org/>. 79  Olea R.A. (1974) Optimal contour mapping using universal kriging. Journal of Geophysical Research, 79, 695-702. Oliver I., Holmes A., Dangerfield J.M., Gillings M., Pik A.J., Britton D.R., Holley M., Montgomery M.E., Raison M., Logan V., Pressey R.L. & Beattie A.J. (2004) Land systems as surrogates for biodiversity in conservation planning. Ecological Applications, 14, 485-503. Olsen D.M. & Dinerstein E. (1998) The global 200: a representative approach to conserving the Earth?s most biologically valuable Ecoregions. Conservation Biology, 12, 502-515. Olsen D.M., Dinerstein E., Wikramanayake E.D., Burgess N.D., Powell G.V.N., Underwood E.C., D?amico J.A., Itoua I., Strand H.E., Morrison J.C., Loucks C.J., Allnutt T.F., Ricketts T.H., Kura Y., Lamoreux J.F., Wettengel W.W., Hedao P. & Kassem K.R. (2001) Terrestrial Ecoregions of the world: a new map of life on Earth. BioScience, 51, 933-938. Palmer M.W. (1993) Putting things in even better order: the advantages of canonical correspondence analysis. Ecology, 74, 2215-2230. Parish R., Coupe R. & Lloyd D. (1996) Plants of Southern British Columbia. Lone Pine Publishing, Vancouver.  Pearson R.G. & Dawson T.P. (2003) Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Global Ecology and Biogeography, 12, 361?371.  Pearson R.G., Thuiller W., Ara?jo M.B., Martinez-Meyer E., Brotons L., McClean C., Miles L., Segurado P., Dawson T.P. & Lees D.C. (2006) Model-based uncertainty in species range prediction. Journal of Biogeography, 33, 1704?1711.  80  Pearson D.L. & Carroll S.S. (1999) The influence of spatial scale on cross-taxon congruence patterns and prediction accuracy of species richness. Journal of Biogeography, 26, 1079-1090. Pebesma E.J. (2004) Multivariate geostatistics in S: the gstat package. Computers & Geosciences, 30, 683-691. Pereira H.M., Leadley P.W., Proenca V., Alkemade R., Scharlemann J.P.W., Fernandez-Manjarres J.F., Ara?jo M.B., Balvanera P., Biggs R., Cheung W.W.L., Chini L., Cooper H.D., Gilman E.L., Guenette S., Hurtt G.C., Huntington H.P., Mace G.M., Oberdorff T., Revenga C., Rodrigues P., Scholes R.J., Sumaila U.R. & Walpole M. (2010) Scenarios for global biodiversity in the 21st century. Science, 330, 1496-1501. Pharo E.J. & Beattie A.J. (2001) Management forest types as a surrogate for vascular plant, bryophyte and lichen diversity. Australian Journal of Botany, 49, 23-30. Pierotti R. & Annett C.A. (1991) Diet choices in the herring gull: constraints imposed by reproductive and ecological factors. Ecology, 72, 319-328. Pojar J., Klinka K. & Meidinger D.V. (1987) Biogeoclimatic ecosystem classification in British Columbia. Forest Ecology and Management, 22, 119-154. Pojar J. & Meidinger D.V. (1991). British Columbia: the environmental setting. In: Ecosystems of British Columbia. Research Branch, Ministry of Forests, Victoria. Prasad A.M., Iverson L.R. & Liaw A. (2006) Newer classification and regression tree techniques: bagging and random forests for ecological prediction. Ecosystems, 9, 181-199. Pressey R.L. & Logan V.S. (1994) Level of geographic subdivision and its effects on assessments of reserve coverage: a review of regional studies. Conservation Biology, 8, 1037-1046. 81  Pressey R.L., Humphries C.J., Margules C.R., Vane-Wright R.I. & Williams P.H. (1993) Beyond opportunism: key principles for systematic reserve selection. Trends in Ecology and Evolution, 8, 124-128. Randin C.F., Dirnb?ck T., Dullinger S., Zimmermann N.E., Zappa M. & Guisan A. (2006) Are niche-based species distribution models transferable in space? Journal of Biogeography, 33, 1689?1703.  Rands M.R.W., Adams W.M., Bennun L., Butchart S.H.M., Clements A., Coomes D., Entwistle A., Hodge I., Kapos V., Scharlemann J.P.W., Sutherland W.J. & Vira B. (2010) Biodiversity conservation: challenges beyond 2010. Science, 329, 1298-1303. Reyers B., Wessels K.J. & van Jaarsveld A.S. (2002) An assessment of biodiversity surrogacy options in the Limpopo Province of South Africa. African Zoology, 37, 185-195. Rodrigues A.S.L. & Brooks T.M. (2007) Shortcuts for biodiversity conservation planning: the effectiveness of surrogates. Annual Review of Ecology, Evolution and Systematics, 38, 713-737. Rodr?guez J.P., Brotons L., Bustamante J. & Seoane J. (2007) The application of predictive modelling of species distribution to biodiversity conservation. Diversity and Distributions, 13, 243?251.  Sala O.E., Chapin F.S., Armesto J.J., Berlow E., Bloomfield J., Dirzo R., Huber-Sanwald E., Huenneke L.F., Jackson R.B., Kinzig A., Leemans R., Lodge D.M., Mooney H.A., Oesterheld M., Poff L., Sykes M.T., Walker B.H., Walker M. & Wall D.H. (2000). Global biodiversity scenarios for the year 2100. Science, 287, 1770-1774. 82  Sarkar S., Justus J., Fuller T., Kelley C., Garson J. & Mayfield M. (2005) Effectiveness of environmental surrogates for the selection of conservation area networks. Conservation Biology, 19, 815-825. Schulman L., Toivonen T. & Ruokolainen K. (2007) Analyzing botanical collecting effort in Amazonia and correcting for it in species range estimation. Journal of Biogeography, 34, 1388-1399. Schuster R. & Arcese P. (2012) Using bird species community occurrence to prioritize forests for old growth restoration. Ecography, in publication.  Schwartz M.W., Brigham C.A., Hoeksema J.D., Lyons K.G., Mills M.H. & van Mantgem P.J. (2000) Linking biodiversity to ecosystem function: implications for conservation ecology. Oecologia, 122, 297-305.  Seo C., Thorne J.H., Hannah L. & Thuiller W. (2009) Scale effects in species distribution models: implications for conservation planning under climate change. Biology letters, 5, 39?43.  Shepherd P. (1997) The Winter Ecology of Dunlin (Calidris alpina pacifica) in the Fraser River Delta. Environment Canada Technical Report (DOE FRAP 1997-38).  Sydeman W.J., Hester M.H., Thayer J.A., Gress F., Martin P. & Buffa J. (2001) Climate change, reproductive performance and diet composition of marine birds in the southern California Current system, 1969-1997. Progress in Oceanography, 49, 309-329. ter Braak C.J.F. (1986) Canonical correspondence analysis: a new eigenvector technique for multivariate direct gradient analysis. Ecology, 67, 1167-1179. Thomas C.D., Cameron A., Green R.E., Bakkenes M., Beaumont L.J., Collingham Y.C., Erasmus B.F.N., De Siqueira M.F., Grainger A., Hannah L., Hughes L., Huntley B., Van Jaarsveld A.S., Midgley 83  G.F., Miles L., Ortega-Huerta M.A., Peterson  A.T., Phillips O.L. & Williams S.E. (2004) Extinction risk from climate change. Nature, 427, 145?148.  Thuiller W. (2003) BIOMOD - optimizing predictions of species distributions and projecting potential future shifts under global change. Global Change Biology, 9, 1353?1362.  Thuiller W. (2004) Patterns and uncertainties of species? range shifts under climate change. Global Change Biology, 10, 2020?2027.  Thuiller W. (2007) Climate change and the ecologist. Nature, 448, 1?3.  Thuiller W., Lafourcade B., Engler R. & Ara?jo M.B. (2009) BIOMOD - a platform for ensemble forecasting of species distributions. Ecography, 32, 369?373.  Thuiller W., Ara?jo M.B. & Lavorel S. (2003) Generalized models vs. classification tree analysis: predicting spatial distributions of plant species at different scales. Journal of Vegetation Science, 14, 669-680. Trakhtenbrot A. & Kadmon R. (2005) Environmental cluster analysis as a tool for selecting complementary networks of conservation sites. Ecological Applications, 15, 335-345. Tuomisto H. (2010a) A diversity of beta diversities: straightening up a concept gone awry. Part 1. Defining beta diversity as a function of alpha and gamma diversity. Ecography, 33, 2-22. Tuomisto H. (2010b) A diversity of beta diversities: straightening up a concept gone awry. Part 2. Quantifying beta diversity and related phenomena. Ecography, 33, 23-45. Venables W.N. & Ripley B.D. (2002) Generalized linear models. In: Modern Applied Statistics with S. Springer, New York. 84  Wang T., Hamann A., Spittlehouse D.L. & Aitken S.N. (2006) Development of scale-free climate data for Western Canada for use in resource management. International Journal of Climatology, 26, 383?397.  Warnock S.E. & Takekawa J.Y. (1992) Wintering site fidelity and movement patterns of Western Sandpipers Calidris maun in the San Francisco Bay estuary. Ibis, 138, 160?167.  Whittaker R.H. (1972) Evolution and measurement of species diversity. Taxon, 21, 213-251. Wiegand J., Raffaelli D., Smart J.C.R. & White P.C.L. (2010) Assessment of temporal trends in ecosystem health using an holistic indicator. Journal of Environmental Management, 91, 1446?55.  Wiens J.A. (1989) Spatial scaling in ecology. Functional Ecology, 3, 385-397. Wilkinson W.B., Leeksb G.J.L., Morrisc A. & Wallingd D.E. (1997) Rivers and coastal research in the Land Ocean Interaction Study. The Science of the Total Environment, 195, 5?14.  Williams P.H. (1996) Mapping variations in the strength and breadth of biogeographic transition zones using species turnover. Proceedings of the Royal Society of London. Series B: Biological Sciences, 263, 579-588. Williams P.H., de Klerk H.M. & Crowe T.M. (1999) Interpreting biogeographical boundaries among afrotropical birds: spatial patterns in richness gradients and species replacement. Journal of Biogeography, 26, 459-474. Wilson K.A., Westphal M.I., Possingham H.P. & Elith J. (2005) Sensitivity of conservation planning to different approaches to using predicted species distribution data. Biological Conservation, 122, 99?112.  85  Wulder M.A., White J.C., Cranny M., Hall R.J., Luther J.E., Beaudoin A., Goodenough D.G. & Dechka J.A. (2009) Monitoring Canada?s forests . Part 1?: Completion of the EOSD land cover project. Canadian Journal of Remote Sensing, 34, 549?562. Wulder M.A., White J.C., Magnussen S. & McDonald S. (2007) Validation of a large area land cover product using purpose-acquired airborne video. Remote Sensing of Environment, 106, 480?491.  Yee T.W. & Mitchell N.D. (1991) Generalized additive models in plant ecology. Journal of Vegetation Science, 2, 587-602.   86  APPENDIX 1 Species distribution AUC values for models including spatial covariates, by species, as well as each species % change in habitat P value and slope as well as habitat designation. Species Habitat Designation GLM AUC GBM AUC GAM AUC CTA AUC ANN AUC MARS AUC RF AUC MEAN AUC WMEAN AUC %Habitat Change P Value %Habitat Change Slope Aechmophorus occidentalis water 0.55 0.63 0.61 0.52 0.55 0.55 0.60 0.66 0.66 0.18 0.00 Anas acuta water 0.76 0.85 0.85 0.81 0.67 0.85 0.79 0.87 0.87 0.44 0.01 Anas americana water 0.68 0.80 0.83 0.76 0.64 0.78 0.75 0.83 0.83 0.06 0.01 Anas carolinensis water 0.79 0.81 0.80 0.75 0.62 0.81 0.74 0.84 0.84 0.29 0.00 Anas clypeata water 0.83 0.88 0.59 0.54 0.77 0.91 0.77 0.87 0.87 0.49 0.00 Anas penelope water 0.77 0.82 0.76 0.53 0.62 0.80 0.74 0.83 0.83 0.84 0.00 Anas platyrhynchos water 0.63 0.73 0.64 0.68 0.61 0.66 0.73 0.73 0.74 0.13 0.01 Anas strepera water 0.77 0.72 0.62 0.57 0.70 0.56 0.64 0.77 0.77 0.52 -0.01 Aphriza virgata shore 0.74 0.73 0.56 0.55 0.64 0.75 0.60 0.79 0.79 0.44 0.00 Ardea herodias land 0.62 0.72 0.70 0.64 0.60 0.74 0.60 0.72 0.72 0.02 -0.01 Arenaria melanocephala land 0.78 0.82 0.79 0.77 0.78 0.78 0.76 0.83 0.83 0.27 0.01 Aythya affinis water 0.55 0.76 0.75 0.53 0.61 0.75 0.64 0.82 0.82 0.01 -0.01 Aythya marila water 0.74 0.83 0.80 0.73 0.63 0.78 0.76 0.83 0.83 0.34 0.00 Brachyramphus marmoratus water 0.69 0.82 0.70 0.67 0.60 0.75 0.66 0.80 0.80 0.26 0.00 Branta bernicla water 0.74 0.85 0.81 0.80 0.60 0.86 0.77 0.87 0.87 0.02 -0.01 Branta canadensis water 0.70 0.77 0.69 0.76 0.51 0.74 0.74 0.77 0.77 0.72 0.00 Bucephala albeola water 0.81 0.84 0.70 0.54 0.81 0.84 0.64 0.85 0.85 0.09 -0.01 Bucephala clangula water 0.67 0.75 0.69 0.68 0.66 0.75 0.66 0.76 0.76 0.58 0.00 Bucephala islandica water 0.84 0.91 0.88 0.88 0.80 0.88 0.85 0.91 0.91 0.71 0.00 Buteo jamaicensis land 0.69 0.77 0.69 0.62 0.73 0.72 0.65 0.82 0.82 0.51 -0.01 Calidris alba shore 0.77 0.99 0.93 0.65 0.75 0.89 0.65 0.97 0.98 0.04 -0.02 Calidris alpina shore 0.75 0.85 0.84 0.80 0.57 0.81 0.75 0.85 0.85 0.51 0.00 Ceoohus columba water 0.71 0.84 0.79 0.75 0.66 0.78 0.78 0.84 0.84 0.13 0.00 Cerorhinca monocerata water 0.71 0.80 0.65 0.65 0.75 0.75 0.52 0.79 0.79 0.31 0.00 Charadrius vociferus shore 0.63 0.77 0.79 0.54 0.65 0.77 0.69 0.78 0.78 0.24 -0.01 Chroicocephalus philadelphia water 0.70 0.66 0.61 0.68 0.70 0.72 0.57 0.68 0.69 0.59 0.00 Circus cyaneus land 0.92 0.88 0.83 0.90 0.87 0.90 0.92 0.88 0.88 0.96 0.00 87  Clangula hyemalis water 0.69 0.85 0.78 0.78 0.58 0.78 0.80 0.83 0.83 0.18 0.02 Corvus caurinus land 0.61 0.76 0.69 0.64 0.62 0.74 0.78 0.77 0.77 0.07 0.03 Covus corax land 0.70 0.76 0.77 0.67 0.69 0.77 0.69 0.79 0.79 0.00 0.01 Cygnis buccinator water 0.83 0.86 0.88 0.62 0.70 0.83 0.63 0.88 0.88 0.01 0.01 Gavia immer water 0.78 0.85 0.80 0.78 0.54 0.82 0.78 0.85 0.85 0.80 0.00 Gavia pacifica water 0.71 0.81 0.78 0.75 0.72 0.74 0.80 0.82 0.82 0.38 0.00 Gavia stellata land 0.79 0.85 0.77 0.70 0.71 0.81 0.70 0.83 0.84 0.11 0.01 Haematopus bachmani shore 0.69 0.83 0.79 0.80 0.71 0.84 0.76 0.83 0.83 0.08 0.00 Haliaeetus leucocephalus land 0.63 0.70 0.67 0.59 0.56 0.67 0.61 0.71 0.71 0.50 0.00 Histrionicus histrionicus water 0.80 0.88 0.89 0.84 0.76 0.89 0.85 0.91 0.91 0.12 -0.01 Larus califronicus water 0.59 0.68 0.67 0.64 0.68 0.66 0.67 0.71 0.71 0.47 0.00 Larus canus water 0.67 0.80 0.77 0.78 0.65 0.77 0.70 0.80 0.80 0.66 0.00 Larus delawarensis water 0.87 0.90 0.82 0.83 0.80 0.86 0.77 0.89 0.89 0.13 -0.01 Larus glaucescens water 0.88 0.91 0.65 0.81 0.81 0.81 0.75 0.89 0.89 0.95 0.00 Larus occidentalis water 0.62 0.79 0.66 0.49 0.64 0.56 0.49 0.77 0.77 0.15 0.01 Larus smithsonianus water 0.61 0.63 0.62 0.65 0.66 0.61 0.65 0.65 0.65 0.05 -0.02 Larus thayeri water 0.77 0.84 0.79 0.81 0.73 0.80 0.76 0.84 0.84 0.11 0.00 Lophodytes cucullatus water 0.71 0.78 0.71 0.75 0.70 0.75 0.75 0.78 0.78 0.79 0.00 Megaceryle alcyon land 0.68 0.73 0.72 0.65 0.53 0.71 0.66 0.75 0.75 0.35 0.00 Melanitta americana water 0.73 0.85 0.83 0.76 0.63 0.81 0.80 0.86 0.86 0.08 0.00 Melanitta fusca water 0.70 0.86 0.83 0.80 0.68 0.82 0.82 0.86 0.86 0.16 0.00 Melanitta perspicillata water 0.76 0.84 0.83 0.77 0.81 0.79 0.81 0.86 0.86 0.01 -0.02 Mergus merganser water 0.62 0.72 0.69 0.72 0.67 0.69 0.63 0.71 0.71 0.07 -0.01 Mergus serrator water 0.66 0.69 0.70 0.67 0.65 0.66 0.64 0.71 0.72 0.26 0.00 Phalacrocorax auritus water 0.72 0.79 0.84 0.69 0.64 0.77 0.79 0.83 0.84 0.48 0.00 Phalacrocorax pelagicus water 0.70 0.82 0.74 0.79 0.67 0.84 0.81 0.83 0.83 0.35 0.00 Phalacrocorax penicillatus water 0.59 0.73 0.69 0.57 0.62 0.74 0.62 0.76 0.76 0.76 0.00 Pluvialis squatarola shore 0.77 0.88 0.85 0.81 0.72 0.86 0.71 0.90 0.90 0.01 0.01 Podiceps auritus water 0.68 0.81 0.76 0.82 0.61 0.74 0.79 0.82 0.82 0.29 0.01 Podiceps grisegena water 0.73 0.82 0.82 0.73 0.77 0.79 0.81 0.85 0.86 0.26 0.00 Podilymbus podiceps water 0.73 0.75 0.63 0.69 0.55 0.73 0.52 0.80 0.80 0.03 -0.02 Tringa melanoleuca shore 0.82 0.91 0.73 0.89 0.81 0.89 0.86 0.92 0.93 0.98 0.00 88  Uria aalge water 0.61 0.76 0.66 0.60 0.50 0.67 0.54 0.75 0.75 0.56 0.00  Species Distribution AUC values for models not including spatial covariates, by species, as well as species Functional Feeding Group designation. Species Functional Feeding Group GLM AUC GBM AUC GAM AUC CTA AUC ANN AUC MARS AUC RF AUC MEAN AUC WMEAN AUC Aechmophorus occidentalis piscivore 0.79 0.77 0.50 0.57 0.72 0.83 0.58 0.82 0.82 Anas acuta herbivore 0.81 0.84 0.78 0.53 0.69 0.84 0.69 0.87 0.87 Anas americana herbivore 0.66 0.72 0.71 0.69 0.64 0.67 0.72 0.75 0.75 Anas carolinensis herbivore 0.74 0.81 0.77 0.74 0.75 0.77 0.83 0.84 0.84 Anas clypeata omnivore 0.67 0.71 0.70 0.56 0.68 0.53 0.57 0.70 0.70 Anas penelope herbivore 0.68 0.66 0.72 0.59 0.67 0.62 0.52 0.72 0.72 Anas platyrhynchos herbivore 0.69 0.65 0.63 0.56 0.56 0.62 0.59 0.74 0.75 Anas strepera omnivore 0.64 0.67 0.71 0.58 0.55 0.58 0.52 0.70 0.70 Aphriza virgata carnivore 0.84 0.90 0.88 0.66 0.72 0.84 0.73 0.91 0.91 Ardea herodias piscivore 0.70 0.82 0.77 0.73 0.68 0.78 0.76 0.83 0.83 Arenaria melanocephala piscivore 0.70 0.76 0.68 0.67 0.72 0.74 0.69 0.75 0.76 Aythya affinis omnivore 0.84 0.82 0.86 0.64 0.67 0.79 0.73 0.86 0.86 Aythya marila omnivore 0.77 0.81 0.74 0.69 0.80 0.75 0.79 0.83 0.83 Brachyramphus marmoratus piscivore 0.73 0.88 0.76 0.66 0.60 0.71 0.65 0.89 0.89 Branta bernicla herbivore 0.65 0.73 0.67 0.58 0.67 0.71 0.65 0.73 0.73 Branta canadensis herbivore 0.73 0.77 0.81 0.71 0.78 0.77 0.74 0.80 0.80 Bucephala albeola benthivore 0.84 0.80 0.49 0.81 0.61 0.68 0.72 0.82 0.82 Bucephala clangula benthivore 0.58 0.69 0.64 0.67 0.68 0.67 0.63 0.72 0.72 Bucephala islandica benthivore 0.73 0.79 0.82 0.82 0.59 0.72 0.75 0.83 0.83 Buteo jamaicensis carnivore 0.75 0.77 0.80 0.66 0.71 0.81 0.69 0.80 0.80 Calidris alba insectivore 0.74 0.81 0.68 0.69 0.70 0.76 0.80 0.80 0.80 Calidris alpina insectivore 0.70 0.79 0.77 0.79 0.61 0.72 0.78 0.82 0.82 Ceoohus columba piscivore 0.63 0.59 0.56 0.52 0.63 0.65 0.50 0.69 0.69 Cerorhinca monocerata piscivore 0.70 0.59 0.62 0.60 0.55 0.57 0.49 0.66 0.67 Charadrius vociferus insectivore 0.77 0.82 0.78 0.74 0.78 0.72 0.78 0.84 0.84 89  Chroicocephalus philadelphia piscivore 0.68 0.73 0.59 0.61 0.64 0.80 0.68 0.73 0.73 Circus cyaneus carnivore 0.58 0.61 0.58 0.59 0.58 0.58 0.62 0.64 0.64 Clangula hyemalis benthivore 0.67 0.84 0.77 0.73 0.74 0.81 0.81 0.84 0.85 Corvus caurinus omnivore 0.69 0.81 0.74 0.75 0.73 0.71 0.69 0.82 0.82 Covus corax omnivore 0.79 0.78 0.62 0.78 0.76 0.78 0.69 0.78 0.79 Cygnis buccinator herbivore 0.70 0.75 0.75 0.69 0.68 0.71 0.65 0.76 0.76 Gavia immer piscivore 0.58 0.61 0.64 0.52 0.50 0.61 0.57 0.67 0.67 Gavia pacifica piscivore 0.72 0.70 0.73 0.60 0.57 0.65 0.52 0.73 0.73 Gavia stellata piscivore 0.71 0.73 0.72 0.55 0.69 0.73 0.60 0.76 0.77 Haematopus bachmani carnivore 0.66 0.75 0.70 0.61 0.64 0.77 0.66 0.74 0.74 Haliaeetus leucocephalus carnivore 0.73 0.81 0.76 0.72 0.67 0.81 0.82 0.83 0.83 Histrionicus histrionicus benthivore 0.71 0.75 0.74 0.66 0.60 0.72 0.71 0.79 0.79 Larus califronicus omnivore 0.70 0.77 0.75 0.64 0.69 0.72 0.70 0.78 0.78 Larus canus omnivore 0.73 0.76 0.75 0.76 0.67 0.72 0.73 0.78 0.78 Larus delawarensis omnivore 0.67 0.67 0.69 0.54 0.61 0.62 0.62 0.72 0.72 Larus glaucescens omnivore 0.66 0.79 0.70 0.66 0.73 0.78 0.67 0.79 0.79 Larus occidentalis omnivore 0.64 0.76 0.61 0.60 0.49 0.68 0.58 0.76 0.76 Larus smithsonianus carnivore 0.58 0.60 0.60 0.52 0.58 0.53 0.57 0.66 0.66 Larus thayeri omnivore 0.81 0.73 0.74 0.53 0.69 0.85 0.61 0.82 0.82 Lophodytes cucullatus benthivore 0.77 0.81 0.81 0.70 0.73 0.85 0.82 0.85 0.85 Megaceryle alcyon piscivore 0.68 0.74 0.63 0.63 0.66 0.66 0.74 0.75 0.75 Melanitta americana benthivore 0.75 0.82 0.76 0.73 0.76 0.75 0.74 0.83 0.83 Melanitta fusca benthivore 0.80 0.78 0.79 0.69 0.75 0.69 0.62 0.79 0.80 Melanitta perspicillata benthivore 0.76 0.81 0.68 0.52 0.70 0.75 0.62 0.82 0.82 Mergus merganser piscivore 0.71 0.79 0.74 0.68 0.66 0.69 0.62 0.78 0.78 Mergus serrator piscivore 0.77 0.77 0.75 0.53 0.70 0.68 0.65 0.80 0.80 Phalacrocorax auritus piscivore 0.82 0.89 0.88 0.78 0.80 0.86 0.83 0.90 0.90 Phalacrocorax pelagicus piscivore 0.68 0.68 0.71 0.49 0.75 0.59 0.50 0.78 0.79 Phalacrocorax penicillatus piscivore 0.63 0.63 0.63 0.60 0.51 0.59 0.65 0.66 0.66 Pluvialis squatarola insectivore 0.72 0.81 0.47 NA 0.60 0.94 0.66 0.84 0.85 Podiceps auritus piscivore 0.74 0.77 0.77 0.69 0.76 0.71 0.72 0.80 0.80 Podiceps grisegena piscivore 0.68 0.75 0.59 0.54 0.50 0.58 0.63 0.76 0.76 90  Podilymbus podiceps carnivore 0.87 0.86 0.77 0.61 0.89 0.81 0.75 0.87 0.87 Tringa melanoleuca insectivore 0.87 0.88 0.83 0.84 0.74 0.81 0.66 0.88 0.88 Uria aalge piscivore 0.58 0.67 0.57 0.58 0.55 0.67 0.71 0.70 0.71  APPENDIX 2 Model variable loadings by species, showing the 3 most important variables for the weighted-mean ensemble model determined through Pearson?s correlation and a leave-one out model approach. Species Variable 1 Variable 2 Variable 3 Aechmophorus occidentalis poc tave ppt Anas acuta lon d2ct poc Anas americana d2ct lon dhi_cov Anas carolinensis lon d2ct lat Anas clypeata pic d2ct dhi_mean Anas penelope lon poc Kd_490 Anas platyrhynchos d2ct lon tmin Anas strepera pic poc chlor_a Aphriza virgata lat tmin lon Ardea herodias d2ct lon ipar Arenaria melanocephala lon d2ct tmin Aythya affinis d2ct lon lat Aythya marila lon d2ct tmin Brachyramphus marmoratus lat dhi_mean d2ct Branta bernicla lon poc d2ct Branta canadensis lon d2ct m_dhi_cov Bucephala albeola dhi_cov dhi_mean d2rd Bucephala clangula d2ct lon lat Bucephala islandica lon lat dhi_mean Buteo jamaicensis dhi_min lon lat Calidris alba lon d2ct dhi_mean Calidris alpina lon d2ct tmin Ceoohus columba lat d2ct lon Cerorhinca monocerata lat chlor_a ppt Charadrius vociferus d2ct lon lat Chroicocephalus philadelphia dhi_mean dhi_min dhi_cov Circus cyaneus lon tave tmax Clangula hyemalis lon lat d2ct Corvus caurinus lat m_d2ct lon Covus corax eosd_rich chlor_a d2ct Cygnis buccinator lat lon tmin 91  Gavia immer lon d2ct tmin Gavia pacifica lon d2ct dhi_min Gavia stellata lon dhi_min ppt Haematopus bachmani lon lat dhi_min Haliaeetus leucocephalus lon lat dhi_mean Histrionicus histrionicus lat lon d2ct Larus califronicus lon d2ct tmin Larus canus d2ct dhi_cov dhi_min Larus delawarensis lon lat poc Larus glaucescens d2ct dhi_mean pas Larus occidentalis poc lon lat Larus smithsonianus lon dhi_mean d2ct Larus thayeri lon d2ct lat Lophodytes cucullatus lat lon tave Megaceryle alcyon lon d2ct dhi_mean Melanitta americana lon d2ct dhi_cov Melanitta fusca lon d2ct tmin Melanitta perspicillata d2ct lon tmin Mergus merganser lon dhi_mean m_dhi_min Mergus serrator lon dhi_cov d2ct Phalacrocorax auritus d2ct lat lon Phalacrocorax pelagicus d2ct lon lat Phalacrocorax penicillatus lat lon d2rd Pluvialis squatarola lon d2ct lat Podiceps auritus lon d2ct dhi_min Podiceps grisegena lon d2ct lat Podilymbus podiceps d2ct ppt poc Tringa melanoleuca d2ct lon lat Uria aalge lat d2ct tave  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0074244/manifest

Comment

Related Items