Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Winter use and habitat selection of moose in openings and adjacent upland forested habitats Catton, Robert Bruce 2007

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

Item Metadata

Download

Media
831-ubc_2007-0045a.pdf [ 1.83MB ]
Metadata
JSON: 831-1.0074938.json
JSON-LD: 831-1.0074938-ld.json
RDF/XML (Pretty): 831-1.0074938-rdf.xml
RDF/JSON: 831-1.0074938-rdf.json
Turtle: 831-1.0074938-turtle.txt
N-Triples: 831-1.0074938-rdf-ntriples.txt
Original Record: 831-1.0074938-source.json
Full Text
831-1.0074938-fulltext.txt
Citation
831-1.0074938.ris

Full Text

WINTER USE AND HABITAT SELECTION OF MOOSE IN OPENINGS AND ADJACENT UPLAND FORESTED HABITATS by Robert Bruce Catton A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE in The Faculty of Graduate Studies (Forestry) THE UNIVERSITY OF BRITISH COLUMBIA May 2007 © Robert Bruce Catton, 2007 Abstract Two designs were developed to model and compare the effects of scaleon resource utilization by moose in a managed lodgepole pine (Pinus contorta) forest. Statistical models, based on population and individual design levels (using microsite and moderate polygon habitat scales, respectively), were used to 1) test the hypotheses that moose track presence was increased i) in openings of increasing shrub cover, ii) in forests adjacent to openings of increasing shrub cover, 2) model moose utilization of forest cover, based on distance-from-nearest-opening, to determine appropriate leave strip widths adjacent to openings used by moose, 3) further understanding of how different opening and forest cover types influence moose presence and utilization and 4) compare these results from the different designs. Moose track occurrence and habitat attributes were recorded on 55 snow track transects to model moose presence as a function of distance-from-opening-edge and variation in vegetation cover. Over 15,000 U T M relocations were obtained from 15 GPS collared cow moose over two winters, in the same area. Based on track transect data, the model containing bog birch (Betula glandulosa), Salix species and mean shrub height was the best predictor of moose presence. Resource utilization functions indicated natural and forest management openings, up to 40 years old, were utilised more than older forest habitats. Modeling at both scales confirmed the use of early serai openings (natural or harvested) and indicated that forested distance-from-an-opening-edge does not influence resource utilization or presence of moose. Track presence was greater in wetlands with average shrub cover between 6 - 20% and > 20% than within adjacent forest cover but there was no difference between wetlands with shrub cover < 6% and adjacent forest cover. Tracks were more numerous in forest cover adjacent to wetlands with shrub cover > 20% than in forest cover adjacent to wetlands with shrub cover < 6% and 6 - 20%. The study implies that leave strips may have little immediate effect on the use of early serai openings by moose in winter although, this does not rule out that climate, predation or hunting pressure will influence habitat selection in the future. Table of Contents Abstract ii Table of Contents iv List of Tables vi List of Figures vii List of Symbols viii Acknowledgements x 1.0 Introduction 1 2.0 Methods 4 2.1 Study Area 4 2.2 Habitat attributes, moose presence and utilisation distributions 4 2.3 Field Methodology - Track Transects 5 2.4 Data Analysis - Track Transects 7 2.5 Data Analysis - GPS Relocations 9 3.0 Results 15 3.1 Track Transect Model Selection and Probability of Track Presence 15 3.2 Track Transect lrm and GEE 18 3.3 Track Transect Model Assessment and Goodness-of-Fit 18 3.4 GPS Relocational Multimodel Inference Based on AIC Weights and Selection Frequencies 21 4.0 Discussion 23 4.1 Distance from Opening Edge 23 4.2 Shrub Cover and Height 24 iv 5.0 Conclusion 29 References 31 References - Appendices 33 Appendix 1: Plot Shrub Ground Cover and Browse by Species 34 Appendix 2: Spokin Lake Weather Station3 Temperature Data 35 v List of Tables Table 1. Terrain and habitat attributes measured at each plot sampled along transect lines 6 Table 2. Terrain and habitat variables obtained from digitized forest cover maps and air photo interpretation of the Spokin Lake study area, south-central British Columbia 10 Table 3. The eleven a priori models of moose resource 11 Table 4. "Best" six logistic regression models as selected by Burnham and Anderson's (1998) AjAICc recommendations with coefficients, standard errors, and variable significance values using the Wald test for logistic regression models (lrm) and generalized estimating equations (GEE). Logistic model selection and output based on all plots whereas for GEE, plots were clustered by transect 16 Table 5. Assessment and goodness-of-fit for selected models 19 Table 6. Chi-square analysis of track presence within wetlands, and forest cover adjacent to wetlands, of differing shrub cover classes 20 Table 7. Model weights and selection probabilities based on the best models selected from 1000 bootstrap samples 21 Table 8. Single best and model averaged unstandardized winter resource utilisation functions (RUFs) coefficient estimates and standard errors based on Akaike weights (wi) and model selection frequencies from 1000 bootstraps for 15 GPS collared cow moose in south-central British Columbia 22 v i List of Figures Figure 1. Winter snow track transect layout 16 Figure 2. Probability of track presence for models BSA, TA, mBSA and CTA 17 Figure 3. Number of tracks and track presence within 25 m distance classes from wetland edge 20 List of Symbols AIC Akaike's information criterion. A method of selecting a "best approximating model(s)" for data analysis and inference. AIC = -2(log-likelihood) + 2K, where K is the number of estimable parameters, including intercept and variance, of a regression model and represents a "penalty" for adding more parameters to a model. 2K(K + \) AICC = (n — K — X)' Corrected AIC, used when sample size is small compared to the number of parameters i.e. n/K < 40, where n is sample size and K is number of model parameter. Aj = AIC, — AlCm,„ A I C A , A model with an A I C A , value of 0 is considered the best or closest to the 'true' model and 'best' fit for the data modeled. Subsequent models are ranked from lowest to highest given the difference in AIC value (A,) relative to the best model. AIC*j Model AIC from a given bootstrap. AlCbcsi Best overall model. AIC*bcst Single best model obtained from all bootstraps. AIC*mj„ Model with minimum AIC value from given bootstrap. Bm Number of times models bootstrapped; i.e. Bm = 1000 indicates 1000 bootstraps. bi Best model selected from given bootstrap. n* Number of records (data) randomly sampled for bootstrap runs; i.e. n* = 1000 indicates the data set of each bootstrap was comprised of 1000 records N Total number of GPS collared moose relocations that were pooled and sampled for model bootstrapping bi /Bm. Relative model selection frequency or probability obtained from bootstrapping. e x p ( - A , . / 2 ) _^exp(-A 12)' Akaike weights. Summing to one (1) where w, is the /=] probability that model i is the best model for the sampling data. exp(-A*/2) ^]exp(-A* 12) ' ^kaike weight for given bootstrap sample. Acknowledgements Financial support for this research was provided by Forest Renewal BC, BC Habitat Conservation Trust Fund (HCTF), the Science Council of BC and Dr. Peter Arcese. In-kind assistance was provided by the B.C. Ministry of Water, Land and Air Protection and the B.C. Ministry of Forests. I thank Dr. Peter Arcese for his academic guidance, mental nurishment and support, as well as my fellow Arcese lab mates for comradery and manuscript feedback beyond the call of duty. Further thanks to my committee members Dr. Thomas Sullivan, Dr. David Shackleton and Dr. Michael Pitt. Randy Wright, John Youds and Jim Young (from the BC Ministry of Environment), Michaela Waterhouse (BC Ministry of Forests), Claire Trethewey and my parents (Hardie and Delia Catton) deserve well founded thanks for their advice, assistance and all-round support. Additional thanks to Kristy Palmantier and Jean Williams of the Williams Lake Indian Band. And in saving one of the best for last, Shannon MacLachlan deserves many thanks, not only for her assistance with field work, but for her continued and ongoing support and friendship. 1.0 Introduction Despite a large literature on moose (Alces alces), few studies have asked whether moose show preferences for particular wetland or riparian areas. In British Columbia (BC), wetlands are assumed to be key components of moose habitat1'2'3'4'5. In spring and summer, wetlands provide high quality forage and relief from insects and high temperatures6 (Peek 1998). Riparian willow and spruce/sedge meadow habitat types are important calving areas in the spring, providing access to water, forage and security cover6 (Langley and Pletscher 1994). The importance provided by these features has lead to forested buffer zones, or 'leave strips', being left adjacent 1 Schwab, F.E. 1985. Moose habitat selection in relation to forest cutting practices in northcentral British Columbia. Ph.D. thesis, University of British Columbia, Vancouver. 2 Baker, B.G. 1990. Winter habitat selection and use by moose in the west-Chilcotin region of British Columbia. MSc. Thesis, The University of British Columbia Vancouver. 3 Hatter, 1. K. Child and H. Langin. 1990. Provincial moose statement, 1990-1995. Draft. Wildlife Branch, BC Min. Environment. 54pp. 4 Young, J.A. 1991. Habitat use and population characteristics of moose in the Cariboo river area of southcentral British Columbia. Habitat Protection Section, BC Min. of Environment. Williams Lake, BC. 73pp. 5 Lemke, S.L. 1998. Upper Deadman River moose habitat study. British Columbia Conservation Foundation, # 200A-1383 McGill Road Southern Interior, BC 1 to these habitats for conservation purposes in areas where forest management activities occur. Despite reserving leave strips for resource protection, current understanding on the use of these reserves is limited (Hannon et al. 2002). Habitat selection by moose is based primarily on forage abundance and quality (Peek 1998). As wetlands are used year round and provide important foraging areas within the Southern Interior Forest Region of British Columbia, adjacent forest cover must often provide all the seasonal habitat cover attributes6 sought by moose within a local area. Baker's study4 (1990), on the Chilcotin Plateau east of Williams Lake, BC, indicated moose used wetland and forest cover types on a relatively equal basis, and that observations during winter months of moose within coniferous forest types often occurred within 100 m, but rarely > 200 m, from a wetland edge. Based on their literature review7, Sopuck et al. (1997) indicated "...there is uncertainty about the effective width and structure of forested buffer zones around wetlands that would provide adequate protection for moose...". Lemke6 (1998) also stated the need for buffers adjacent to important habitat types, but did not indicate appropriate buffer widths and forest cover attributes adjacent to a given habitat type. The need for buffers adjacent to [wetland and] aquatic forage areas is also controversial - due primarily to the large amounts of timber that would otherwise be available for harvest (Thompson and Stewart 1998). 6 Sopuck, L., K. Ovaska and R. Jakimchuk. 1997. Literature Review and problem analysis of moose/forestry interactions in the Cariboo Forest Region. Final Report for British Columbia Ministry of Forests, Central Cariboo Forest District, Williams Lake. 2 Baker4 (1990) suggested that forest harvesting should not occur within 400 m of wetlands with >15% shrub cover. The untested recommendations of Sopuck et al. (1997) were also for 400 m forested buffers around key wetlands, wetland complexes and riparian areas, 200 m buffers around wetlands > 5 ha, and 50 m around wetlands 1 - 5 ha in size. Brusnyk and Gilbert (1983) reported that moose used uncut shoreline reserves next to clearcuts more than shoreline areas with no buffers during winter, but did not recommend a reserve width. The need for forested buffers adjacent to wetland/aquatic and riparian systems seems apparent. The width, size, location and physical attributes required of these buffers is less apparent. Timmermann and Racey (1989) suggested shoreline buffers might be required for security cover only when adjacent to actual foraging areas. The width and location of these buffers would then be dependent upon the type, size and importance of a given opening type within the range of a given moose population. My objectives were to 1) test the hypotheses that moose track presence increases i) in openings of increasing shrub cover (i.e. 0 - 100% ground cover), ii) in forests adjacent to openings of increasing shrub cover, 2) model moose presence and utilization of forest cover, based on distance-from-nearest-opening, to determine appropriate leave strip widths adjacent to serai openings used by moose, 3) further the understanding of how different opening and forest cover types influence moose presence and utilization and 4) compare these results based on population and individual level based study designs. 3 2.0 Methods 2.1 Study area The study area was located in the Sub-Boreal Pine-Spruce moist-cool (SBPSmk) biogeoclimatic subzone (Steen and Coupe 1997), approximately 25 km east of Williams Lake, BC (52° 00' -52° 12' North 121° 24' - 121° 48' West). Mature (80+ years old) lodgepole pine (Pinus contorta) forests dominate and 14% of the area has been harvested and reforested to pine over the last 20 years. Approximately 10% of the area consists of wetlands and sedge (Carex spp.) meadows and 5% is used for pasture and hay cultivation. The forest undergrowth consists of pine grass (Calamagrostis rubescens), various herbs, scattered shrubs and a forest floor blanket of moss and lichens. The terrain is level to gently rolling with elevation ranging from 900 to 1300 m. Mean annual precipitation is 506 mm with up to 50%> of this being in the form of snow (Steen and Coupe 1997). The study area is important for both First Nations as well as a timber supply area. A forest harvesting deferral was in effect during the years the study was conducted, although some selection removal of mountain pine beetle (Dendroctonus ponderosae) infested lodgepole pine was permitted. 2.2 Habitat attributes, moose presence and utilisation distributions 7 The study involves third-order habitat selection analysis in order to model utilisation of different 1) wetland and opening types and 2) forest cover types. An experimental unit consists of an opening and upland habitats associated with moose use of the given opening. First-order is selection of the physical or geographical range of a species, second-order is home range selection and third-order is importance of habitat components within an animal's home range (Johnson 1980, Thomas and Taylor 1990, Samuel and Fuller 1994). 4 Two study designs, based on two different scales of habitat utilization by moose were used. Design I is based on a population level assessment of moose habitat use based on track presence / absence data obtained from the establishment of snow track transects located within differing wetland types and their adjacent forest cover. Design II is used to determine habitat utilization based on the identification of individual GPS collared animals and the relocational sampling data being overlaid onto digitized forest cover maps. 2.3 Field Methodology - Track Transects I randomly sampled 55 wetlands and adjacent forest cover with 2 m wide track transects during January to mid-March of 2001 and 2002. Tracks were recorded if intersecting any portion of the transect. Tracks observed to follow a transect line were only recorded once. Transects were an average length of 600 m, on a random bearing, with the constraints that the length of a transect was at least 300 m from any clearcuts, wetlands or active roads (Figure 1). Habitat attributes were sampled in 50 m 2 plots at 50 m intervals along each transect with approximately 1/3 of plots falling within a wetland and the remainder in the adjacent forest cover. Local terrain attributes and vegetative characteristics within plots were measured (Table 1). Shrub cover (%) and mean shrub height were based on the portion of shrub exposed above snow level and therefore available for browsing. A total of 218 wetland and 457 forest plots were sampled. 5 Figure 1. Winter snow track transect layout. Transects were randomly located in 55 randomly selected wetlands with the constraint that the sampled wetland and any portion of the transect line was a minimum of 300 m from other openings (natural or from land management). Vegetation plots were located every 50 m along the transect line with approximately 1/3 of all plots being located within the wetland. clearcut v clearcut ', clearcut \ Table 1. Terrain and habitat attributes measured at each plot sampled along transect lines Attribute (code) Description Distance from wetland (m) Total shrub ground cover (T) Habitat type (H) Bog birch ground cover (B) Willow species ground cover (S) Other shrub ground cover by species % browse by species Average shrub height (A) Crown closure (C) Snow depth (Sn) Slope Slope position Aspect Plot centre distance from nearest wetland/forest edge (in) Total shrub ground cover (%) for all species within plot Wetland or forest categorical variable Ocular (%) estimate of bog birch ground cover Ocular (%) estimate of Salix species ground cover Ocular (%) estimate of ground cover for given species Ocular (%) estimate of amount browsed by moose (based on presence of tracks) Average height (m) of all shrub species within plot Forest overstory densiometer measurement (%) from plot centre Average of 3 measurements taken within plol (cm) % slope Mesoslope position Direction of slope as a bearing 6 2.4 Data Analysis - Track Transects I constructed logistic regression models based on variables expected to influence the prediction of moose track presence. Slope, slope position and aspect were not modeled given only 5% of plots had a slope > 10%. Bog birch (Betula glandulosa) and willow {Salix spp.) were the only shrub species modeled given the low ground coverage and negligible browsing observed on the twelve other shrub species observed within the study area (Appendix 1). Total shrub cover was correlated with bog birch and willow ground cover, and therefore was not included within the same models. Percent cover variables were arcsine transformed in order to approximate a normal distribution. Burnham and Anderson (1998) recommended the number of parameters used in a model as n/10. For modeling purposes, a maximum of seven parameters were included in any model, including second order interactions. I selected models based on a corrected Akaike's Information Criterion (AICC) value of < 4. Burnham and Anderson (2002) indicated that models with a Aj AICc < 2 substantially support the sample data whereas models with a Aj AICc > 4 indicate considerably less support. In order to incorporate distance-from-wetland-edge as a variable, 1 assumed all plots were independent. There are precedents to this assumption; having an independent data set is less important than having collected a sufficient sample size for data analysis (Andersen and Rongstad 1989, Reynolds and Laundre 1990, McNay et al. 1994, Swihart and Slade 1997 and Otis and White 1999). Selected logistic model coefficients were transformed to probabilities and subsequently the probability of track presence was plotted for individual model variables. Although correlated 7 estimators are generally unbiased, they may lead to inconsistent (underestimated) variances (Fitzmaurice 1995; Schabenberger & Pierce 2002), subsequently leading to "false" variable significance from p-values being too low. Generalized estimating equations (GEE), using the logit link function and an auto-regressive correlation structure, were used in consideration of plot autocorrelation within transects (Liang & Zegar 1986; Horton and Lipsutz 1999; Hosmer and Lemeshow 2000) to test the significance of the selected model variables. Variable significance for both logistic regressions and GEE were determined using the Wald test and compared. The C index and Somers' D were used to assess the predictive ability of the selected models. The C index, or area under the Receiver Operating Characteristic (ROC), measures a model's ability to discriminate between the outcome values of the response variable (Hosmer and Lemeshow 2000). C index values > 0.70 are considered an acceptable level of discrimination (Hosmer and Lemeshow 2000). Somers' D assesses the fit of a model by indicating the rank correlation between the observed and predicted model response values (Harrell Jr. et al. 1996). It ranges between -1.0 and 1.0 with larger numbers indicating a higher negative or positive correlation. Nagelkerke's pseudo R2 (Nagelkerke 1991) and the Le Cessie-van Houwelingen (Le Cessie and van Houwelingen 1991) goodness-of-fit test were used to assess the ability of the selected models to fit the data. Welch's t-test was used to test if there was a significant difference in the amount of observed browse of bog birch and willow. Pearson's Chi-squared test with Yates' continuity correction analysis was used to test for differences in track presence within wetlands and within adjacent forests based on wetland shrub cover classes. Transects were placed within a given class by averaging the total shrub cover (%) of wetland plots for a given transect. For analysis purposes, plots were assumed independent. Statistical analysis was performed using the program R, version 1.8.1 (R Development Core Team 2003). All tests used a significance level of 5% (a = 0.05). 2.5 Data Analysis - GPS Relocations The second component of the study involved modeling relative habitat utilisation of 15 GPS collared cow moose using Resource Utilization Functions (Marzluff et al. 2004, Hepinstall et al. 2003) . Resource Utilization Functions (RUF) are linear regression models that, unlike ordinary least square regressions, account for autocorrelation present when modeling GIS grid data (Marzluff et al. 2004, Hepinstall et al. 2003). The RUF regression coefficients are estimated using maximum likelihood estimation (MLE). Resource importance is indicated by the estimated RUF coefficients which indicate the change in utilisation distribution (UD) when a given resource value is changed while the remaining resource values stay fixed (Marzluff et al. 2004) . Positive (+) or negative (-) coefficient values indicates use increases or decreases with the quantitative increase of a given resource (Marzluff et al. 2004). RUF variants used to model habitat utilisation of the collared moose are listed in Table 2. Akaike's Information Criteria is particularly well suited for model selection given RUF model coefficients are estimated using MLE . Model selection using AIC permits testing of specific a priori models and the subsequent ranking and comparison of these models. The model with an A I C A , value of 0 is considered the best or closest to the 'true' model and 'best' fit for the data model with subsequent models ranked from lowest to highest value given the difference in AIC value (Ai) relative to the best model (Burnham and Andersen 2002). My methodology follows the procedure initially described in Marzluff et al. (2004) and Hepinstall et al. 2003. These steps 9 involve 1) the construction of utilisation distributions (using fixed-kernels) for study animals, 2) density estimate determination of individual UD grid cells, 3) relation of habitat resources to each grid cell and 4) model resource use by relating UD grid cell density estimates to grid cell habitat resources. I subsequently added a fifth step, incorporating the model selection and unconditional averaging procedures advocated by Burnham and Anderson (2002) and suggested by Marzluff et al. (2004). I developed 11 a priori models (Table 3) which were fit to each of 1000 bootstrap runs using unstandardized regression coefficients. Use of unstandardized Table 2. Terrain and habitat variables obtained from digitized forest cover maps and air photo interpretation of the Spokin Lake study area, south-central British Columbia Attribute (code) Description Distance from wetland (m) Plot centre distance from nearest wetland/forest edge (m) Habitat description Wetland or forest categorical variable hdesForested Non harvested forest hdesHarvestedO 1 Harvested forest age class 0 — 20 years hdes Harvested2+ . Harvest forest age class 21 - 40 years hdesHE Herb dominated wetland / meadow with < 5% shrub ground cover1 hdesL Waterbody classification obtained from forest cover maps hdesSH Shrub wetland / meadow with > 5% shrub ground cover* Forest species composition2 Sx Obtained from forest cover maps Px Obtained from forest cover maps Fd Obtained from forest cover maps At Obtained from forest cover maps Ac Obtained from forest cover maps conAll Total conifer cover (%) obtained from forest cover maps Decid Total deciduous cover (%) obtained from forest cover maps Ocular (%) estimate of shrub ground cover from air photos. Forest cover species occurring in < 5% of the 15,000 utilisation points were not considered (BI, Ep, Cw, Lw) in the analysis. Forest crown closure was considered as a variable but dropped given the majority of unharvested forested ranged between 45 to 65% . Harvested polygons < 40 years were considered as these early serai stages tend to have greater shrub production (hence browse availability) than older forest serai stages. coefficients allows expected resource use to be predicted and averaged by pooling relocational data from the GPS collared moose in this component of the study (Marzluff et al. 2004). A I C for all models were calculated ( A I C * , ) , then ranked based on the AICA / calculated for each bootstrap (B). The single best model (AIC*b e s t ) from all bootstraps was then obtained. The best model 10 from each bootstrap (bi) was selected and averaged using both Akaike weights and model selection frequencies. Table 3. The eleven a priori models of moose resource utilisation evaluated using Akaike's Information Criteria (AIC) Model number Model 1) use ~ dis + hdes + Sx+ Px + Fd + At + Ac 2) dis + hdes + Sx + At 3) dis + hdes + conAll + Decid 4) dis + hdes + Sx + Decid 5) dis + hdes + Decid 6) dis + hdes V) dis 8) hdes 9) Sx 10) At H) Decid Prior to bootstrapping, 300 GPS relocations were randomly selected from the 3D (differentially corrected) winter relocational data collected for each of 15 cow moose. The January to April 30 time period was considered to represent winter conditions given the majority of lakes within the study area were still frozen and browse species had not yet flushed. Relocation UTM coordinates were overlaid onto digitized forest cover maps using GIS. Kernel utilization distribution (UD) home ranges were estimated for each animal using the Arc View Animal Movement extension (Hooge and Eichenlaub 2000) based on a 25 x 25 m grid cell resolution. The UD was overlaid on the forest cover maps, permitting relative use values to be assigned to each 25 m grid cell within mapsheet polygons. Utilisation distributions give a quantitative view of the relative resource use of animals within an area by considering the distribution of all relocations together versus a single location at a time (Marzluff et al. 2001). The 25m resolution was deemed a practical scale upon which to determine leave strip/buffer widths adjacent to natural openings. This resolution should also incorporate any forest cover boundary mapping and GPS relocational 11 error. Sampling points were created in the centre of each grid cell and distance to the nearest unforested or harvested polygon (boundary) was determined using the Arc View 3.x Focal Patch extension (Hurvitz 2002). Relative moose utilisation and forest cover attributes for each grid cell were extracted for statistical modeling. Approximately 89,000 utilisation records were derived for the 15 moose (range of 1005 - 9255 grid cells per home range). One thousand utilisation points were randomly selected from the home range of each collared cow moose and pooled (total of 15,000 points) for analysis. One thousand points were used to ensure the equal weighting of each moose given differences in the number of utilization points for each animal (due to differences in kernel home range sizes). Forest cover species occurring in < 5% of the 15,000 points were not considered in the analysis. Forest crown closure was considered as a variable but dropped given the majority of unharvested forested ranged between 45 to 65%. Harvested polygons < 40 years were considered as these early serai stages tend to have greater shrub production (hence browse availability) than older forest serai stages. From the 15,000 (TV) pooled points, 1000 points («*) were randomly sampled, with replacement, and each model was run for the given bootstrap sample; this was repeated 1000 times (B,„ — 1000). Bootstrapped data sets were limited to 1000 points for computer efficiency purposes. I used the bootstrap output to estimate model selection uncertainty using two methods. Akaike weights (w,) and relative model selection frequencies (7ij), were calculated to quantify model selection uncertainty and permit model averaging of the best models selected from bootstrap sampling. Relative model selection frequencies for each of the models were determined based on the frequency a given model was selected 'best' (AIC*mjn) model from each of the bootstraps i.e. model with minimum AIC from each bootstrap. Akaike weights indicate the weight of evidence, 12 or probability, of a given model being the best of those models considered or in other words indicate the relative support of the data for the models (Burnham and Anderson 2002). Selection probabilities (7tj) quantify model selection uncertainty by providing a measurement of the amount of sample variation occurring in the selection of the best model (Burnham and Anderson 2002). The Akaike weights for each bootstrap (w,*) were calculated and coefficients averaged for each model over the 1000 bootstraps (Burnham and Anderson 2002). Similarly, relative model selection frequencies were calculated and model coefficients averaged. Model-averaged estimates were derived by averaging the estimated parameter values from the best model selected from each bootstrap (£>,) based on 1) Akaike weights (w,*) and 2) bootstrap selection frequencies (7 1) (Burnham and Anderson 2002). The w,*'s from all (the best) models including a given parameter estimate (®) were summed and then renormalized so the sum of all A w,-*'s was equal to 1 (E w,* = 1). The original estimated model parameter estimates (Q) were multiplied by the renormalized weights and summed to give the model-averaged parameter A estimates (Q ). Similarly, model averaged parameter estimates based on bootstrap selection frequencies (n) were calculated. The estimated coefficients from 1) the overall best model, 2) Akaike weights and 3) relative model selection frequencies from the best model selected from each of 1000 (B) bootstraps were then compared. The overall best model was the single best model (i.e. A1CA, = 0) of the 1000 bootstraps. Eleven a priori models (Table 3) were developed with the objective of determining the "best approximating" model of moose resource utilisation, as evaluated using Akaike's Information Criteria (AIC). 13 Moose utilisation data derived from the kernel home range estimates and the forest cover data were heavily skewed and could not be transformed into a normalized distribution. The subsequent resource utilisation function (RUF) analysis relies on the robustness of the 'normal' model and its distribution. The model averaged RUFs assume collared moose home ranges were independent of each other. The R environment for statistical computing and graphics (www.r-project.org) using Handcock's (2003) code was used for estimation of Resource Utilization Functions. 14 3.0 Results Willow occurred in 43% and bog birch in 59% of wetland track transect plots sampled. Both of these species were used as food in the wetlands where they were found, but the intensity of browsing on each did not differ (t45.5 = -1.43,^ -value = 0.16). Seven other shrub species were recorded, but none occurred in > 4% of the wetland plots and none had been browsed (Appendix 1). Signs of browsing were also generally absent from 14 shrub species observed within forest cover adjacent to the sampled wetlands (Appendix 1). Where present, willow was the most heavily browsed forest shrub, although the majority of this was older browse and could not be attributed directly to moose. Shrub cover was sparse, being < 1 % total ground cover in 80% of the forested plots. Average snow depth for years 2001 and 2002 was 36 cm and 55 cm in wetlands and 29 cm and 32 cm for adjacent forested areas, respectively. No observations of snow cratering for forage, or indications of graminoids or other vegetation being grazed were recorded within either wetland or adjacent forested areas. 3.1 T rack Transect M o d e l Selection and Probabi l i ty of T r a c k Presence Track presence was supported by six models, including one second order interaction, based on an Aj AICc < 4 (Table 4). All six models included average shrub height. The best model (BSA) included average shrub height and the transformed bog birch and willow ground cover variables, while the very similar second and third best models were the second order interaction (T*A) and main effect (TA) models of total shrub ground cover and average shrub height (Table 4). These three best models had a Aj AICc < 2. The fourth, fifth and sixth models had a Aj AICc < 4 and 15 included distance from wetland edge and forest crown closure. Snow depth was not considered to influence probability of track presence given models in which it was included had a Aj AICc > 5. Table 4. "Best" six logistic regression models as selected by Burnham and Anderson's (1998) AjAICc recommendations with coefficients, standard errors, and variable significance values using the Wald test for logistic regression models (Inn) and generalized estimating equations (GEE). Logistic model selection and output based on all plots whereas for GEE, plots were clustered by transect model" K Log(L) AICc AiAlC c variable coefficient Inn std. error p-value coefficient GEE sld. error /;-value BSA 4 -378.40 765.60 0 intercept -1.394 0.123 0.000 -1.150 0.194 0.000 B 2.908 0.609 0.000 0.043 0.011 0.000 s 1.848 0.825 0.025 0.023 0.012 0.049 A 0.721 0.207 0.001 0.327 0.143 0.022 T*A* 4 -378.79 766.38 0.78 intercept -1.650 0.160 0.000 -1.413 0.221 0.000 T 4.080 0.919 0.000 0.064 0.015 0.000 A 0.984 0.289 0.001 0.590 0.213 0.006 T*A -1.314 0.773 0.089 -0.023 0.011 0.035 TA 3 -380.07 766.61 1.01 intercept -1.501 0.128 0.000 -1.219 0.198 0.000 T 2.974 0.619 0.000 0.039 0.009 0.000 A 0.678 0.216 0.002 0.319 0.152 0.036 mBSA 5 -378.38 767.99 2.38 intercept -1.416 0.167 0.000 -1.033 0.240 0.000 m 0.000 0.001 0.839 -0.001 0.001 0.275 B 2.968 0.679 0.000 0.039 0.011 0.001 S 1.871 0.834 0.025 0.021 0.012 0.084 A 0.724 0.208 0.001 0.320 0.144 0.027 mTA 4 -379.998 768.80 3.19 intercept -1.457 0.172 0.000 -1.059 0.243 0.000 m 0.000 0.001 0.706 -0.001 0.001 0.156 T 2.887 0.659 0.000 0.036 0.009 0.000 A 0.670 0.216 0.002 0.307 0.152 0.043 CTA 4 -380.05 768.90 3.30 intercept -1.541 0.237 0.000 -0.861 0.297 0.004 C 0.052 0.256 0.841 -0.009 0.006 0.144 T 3.030 0.681 0.000 0.032 0.009 0.001 A 0.678 0.216 0.002 0.292 0.151 0.053 model variable codes: B - bog birch, S - willow species, A - average shrub height, T - total shrub ground cover (%), m -distance (m) from wetland edge, C - forest crown closure (%)). second order interaction. The probabilities of track presence based on individual model variables, while holding all others constant, of models BSA, TA, mBSA and CTA are illustrated in Figure 2. The probability plots 16 illustrate a minimum probability of track presence of 20%, regardless of habitat variable considered. Probability of track presence is similar for given variables amongst the different models. The plotted model variables illustrate the positive influence of bog birch, willow, total Figure 2. Probability of track presence, with 95% confidence intervals, based on variable coefficients from models BSA (AICC = 0), TA (A1CC = 1.02), mBSA (AICC = 2.37.) and CTA (AICC = 3.30). o Q. 1.0 0.8 0.6 0.4 0.2 0.0 Model: BSA 0 40 80 1.0 0.8 0.6 0.4 0.2 0.0 0 40 80 1.0 0.8 0.6 0.4 0.2 0.0 0 CD O 1.0 0.8 0.6 0.4 0.2 0.0 bog birch (%) Model. 0 40 80 w illow (%) 1.0 0.8 0.6 0.4 0.2 0.0 shrub height (m) total shrub (%) 1.0 0.8 0.6 0.4 0.2 0.0 Model: mBSA shrub height (m) 1.0 0.8 0.6 0.4 0.2 0.0 -400 0 400 0 40 80 1.0 0.8 0.6 0.4 0.2 0.0 0 40 80 1.0 0.8 0.6 0.4 0.2 0.0 X I CO .a o 1.0 0.8 0.6 0.4 0.2 0.0 distance (m) Model: CTA 0 40 80 bog birch (%) 1.0 0.8 0.6 0.4 0.2 0.0 0 40 80 w illow (%) 1.0 0.8 0.6 0.4 0.2 0.0 0 shrub height (m) crow n closure (%) total shrub (%) shrub height (m) 17 shrub ground cover and average shrub height and the lack of influence distance from wetland edge and forest crown closure have on probability of track presence. Logistic regression model and GEE Wald significance tests indicated the non-significance of the distance and crown closure variables (Table 4). Probability of track presence based on bog birch (and total shrub) ground cover was somewhat greater than willow. Based on 20% ground cover, the probability of track presence was 36% for bog birch and 31 % for willow. Probability of track presence based on total shrub cover was similar to that of bog birch. 3.2 Track Transect Irm and GEE Post hoc review of models selected via A1CC indicated the significance of ground cover variables when modeled using Inn and GEE, with the exception of willow ground cover in GEE model mBSA (Table 4). Distance from wetland edge and crown closure were not significant for selected models with either Inn or GEE; this is similarly indicated by probability of track presence plots for models mBSA and CTA in Figure 2. 3.3 Track Transect Model Assessment and Goodness-of-Fit The assessment and goodness-of-fit for the selected models are listed in Table 5. All selected models were significantly different from the null model (p < 0.01). Similar to testing a models ability to correctly predict track presence, the C index indicated the probability of all selected models to correctly predict track presence 71% of the time. Fit of the selected models as measured by Somers' D were all approximately 0.42. Nagelkerke's pseudo R (Nagelkerke 1991) indicated all selected models explained approximately 20% of the variation of probability of track presence. The lack of significance (p > 0.28) indicated by the Le Cessie-van Houwelingen (Le Cessie and van Houwelingen 1991) goodness-of-fit tests indicated there is no reason to reject any of the selected models ability to fit the data. Track presence and numbers 18 increased within wetlands with increasing total shrub cover (Figure 3). Chi-square tests indicated there was no difference in observed track presence between wetlands with plot averaged total shrub coverages of < 6% and adjacent forest cover (X2 = \.79, df= 1,p-value = 0.18). Table 5. Assessment and goodness-of-fit for selected models Model C D X Y goodness-of-fit'' (n = 675) (p-value) BSA 0.71 0.421 0.202 0.28 TA 0.707 0.414 0.197 0.44 TA, 0.709 0.417 0.201 0.36 mBSA 0.71 0.419 0.202 0.31 mTA 0.712 0.423 0.197 0.59 CTA 0.708 0.416 0.197 0.35 " C index (area under ROC curve); where track presence is indicated by P(y=l) > 0.5 and track absence is indicated by P(y=i) < 0.5 * Somers' Dxy Nagelkerke's R 2 index goodness-of-fit - Le Cessie-van Houwelingen Track presence was significantly greater in wetlands with plot averaged total shrub coverages of 6 - 20% ( r = 23.66, df= 1, p-value = < 0.001) and > 20 - 56% {%2= 30.79, df= 1,p-value = < 0.001) than within their respective adjacent forest cover. Chi-square tests indicated there was a significant difference in observed track presence between wetlands of differing plot averaged total shrub coverages (Table 6). Chi-square tests indicated no difference in observed track presence within forest cover adjacent to wetlands with plot averaged total shrub coverages of < 6% and > 6 - 20% (Table 6). There was significantly greater track presence within forest cover adjacent to wetlands with plot averaged total shrub coverages > 20 - 56% when compared to both wetlands with plot averaged total shrub coverages of < 6% and > 6 - 20% (Table 6). 19 Figure 3. Number of tracks and track presence based on number of transect plots falling within 25 m distance classes from wetland edge. Total shrub ground cover for a transect's wetland plots were averaged and associated plots were placed into given shrub cover class. There are 17, 21 and 17 transects in the < 6, > 6 - 20, and > 20 - 56% shrub cover classes, respectively. I 1 No. of plots in dislance class No. of plots with hack presence • • • • No. of Iracks in dislance class Table 6. Chi-square analysis of track presence within wetlands, and forest cover adjacent to wetlands, of differing shrub cover classes wetland forest Wetland shrub cover ~> r df p-value r df p-value < 6% vs. > 6 - 20% 14.19 1 < 0.01 3.61 1 0.057 > 6 - 20% vs. > 20 - 56% 22.06 1 < 0.01 19.19 1 < 0.01 < 6% vs. > 20 - 56% 58.88 1 < 0.01 33.05 1 < 0.01 20 3.4 GPS Relocational Multimodel Inference Based on AIC Weights and Selection Frequencies No single model was strongly supported based on the AIC weights and selection frequencies derived from bootstrapping (Table 7); indicating model selection uncertainty was considerable. Model 8 (hdes) was the single best model with the lowest overall AIC value (AIC^/) and was also selected as best model based on the weighted average (w, = 0.44) and selection frequencies (6j= 0.22); estimated coefficients for model 8, including the averaged model coefficients, are located in Table 8. Models 6 and 5 followed as second and third best based on model averaging. Table 7. Model weights and selection probabilities based on the best models selected from 1000 bootstrap samples Model number Wi1 W*i i ^i (as selection probability) 1 0.003 0.056 52 0.052 2 0.0463 0.089 45 0.045 3 0.029 0.076 19 0.019 4 0.035 0.082 25 0.025 5 0.075 0.142 130 0.130 6 • 0.202 0.188 208 0.208 7 0.047 0.055 47 0.047 8 t 0.470 0.225 438 0.438 9 0.030 0.026 8 0.008 10 0.030 0.030 12 0.012 11 0.033 0.030 16 0.016 sum I 1 1000 1 weights of models from the bootstrap run (b,) producing the single best model of 1000 bootstraps 2 weighted average of given model based on 1000 bootstraps of all models 3 estimator of relative best model selection frequency based on 1000 bootstraps of all models t hdes - best model (hdes: HarvestedOJ, Harvested2+, HERB, L A K E , SHRUB) • distance + hdes - 2 n d best model (dis: distance from wetland edge) x distance + hdes + deciduous - third best model Eight of the 11 models (8, 6, 5, 2, 4, 3, 1, 7) were required to obtain an 91% confidence set of models recommended by Burnham and Anderson (1998) based on averaged Akaike weights. Six 21 models (8, 6, 5, 1, 7, 2) were required for a 92% confidence set based on the selection frequencies. All five categorical opening habitat descriptors (hdes) positively influenced use, with magnitudes of about 0.5 to 2.6, depending on the variable and method of calculation; the exception being the negative "hdesHarvested2+" coefficient value derived from the single best model (Table 8). All other forest cover components and distance from opening edge had a negligible influence on use, with magnitudes from 0.1 to 0.001. Table 8. Single best and model averaged unstandardized winter resource utilisation functions (RUFs) coefficient estimates and standard errors based on Akaike weights (wi) and model selection frequencies from 1000 bootstraps for 15 GPS collared cow moose in south-central British Columbia Parameter 0' Standard Standard Error Standard Error 02(byw*) 6 3 (by ) Error Intercept 8.02 0.120 9.97 10.071 9.63 9.727 distance 0.004 — -0.003 0.002 -0.003 0.001 hdesHarvestedO 1 0.704 0.060 0.96 0.879 0.55 0.609 hdesHarvested2+ -1.42 0.1 15 0.45 0.490 0.22 0.346 hdesHE 2.61 0.082 2.00 1.806 1.23 0.336 hdesL 0.95 0.181 1.10 1.098 0.71 0.884 hdesSH 1.83 0.063 1.78 1.596 1.12 0.077 Sx 0.02 0.006 0.01 0.0002 Px -0.008 0.001 -0.01 0.0001 Fd -0.004 0.0003 -0.004 0.0001 At 0.030 0.006 0.02 0.0004 Ac 0.270 0.015 0.27 0.003 conAll -0.020 0.002 0.02 0.0000 Decid 0.040 0.012 0.05 0.004 Single best model from 1000 bootstraps; standard error is conditional (i.e. based on the single model output). " Model averaged parameter estimates derived using Akaike weights. * Model averaged parameter estimates derived using model relative selection frequencies. Although magnitudes of use were similar for the habitat descriptor variables (hdes), there is still considerable variation as to the effect on use depending on the method of model averaging (wi or bi) used. Model Averaged coefficients derived using Akaike weights, whether the single best or 22 bootstrap derived, were quite similar relative to the coefficients based on model selection frequencies derived using the bootstrap. 4.0 Discussion The ability to model provides a way of understanding the structure of a system and predicting its condition in future (Burnham and Anderson 1998). Habitat models based on different study designs provided me with a method to identify how wetland and forest features influence moose track presence and resource utilisation at local and landscape scales. Similar to other studies, serai stage wetland, natural openings (meadows) and forest habitat types < 40 years of age, dictated resource utilisation of the GPS collared moose within the study area. The relatively high minimum probability (20%) of moose track presence indicated the study area was well utilized by moose. This high minimum probability of track presence is likely a function of browse availability not only within wetlands but within recent (< 40 years) local forest harvested stands as well. Modeling of GPS relocational data indicated increased utilisation of serai harvested areas and considerably less in forest stands greater than 40 years old. Although not sampled, I observed moose browsing shrubs and trembling aspen (Populus tremuloides) saplings within these early serai stage cutblocks when accessing track transect lines. Trembling aspen was scattered and provided minimal forage within the forested areas sampled, although I often observed moose sign such as the heavy browsing of aspen saplings and cambium chewing of older aspen trees. 4.1 Distance from opening edge Unlike Baker's4 (1990) telemetry study, both design types of this study indicated that moose track presence and utilisation within forest cover adjacent to openings were not influenced by distance from opening edge. In Baker4 (1990), moose locations were equally divided between wetland and adjacent forest cover, while my track transect data indicated a disproportionately greater track presence within wetlands versus adjacent forest cover when wetland shrub cover 24 exceeded 5%. Although moose track presence was greater within and adjacent to wetlands of increasing shrub cover, distance from wetland edge did not influence the probability of track presence whereas in Baker's4 (1990) study, moose were primarily located within 100 m of the edge. Riparian zones adjacent to wetland openings within my study area were mostly absent or only several metres wide and did not affect distance from wetland edge. 4.2 Shrub Cover and Height Track count data and modeling of track presence demonstrated increased moose presence in shrubby versus non-shrubby openings or forest habitat, with browsing almost exclusively on bog birch and willow. This is similar to Baker3 (1990), although he also noted the browsing of lodgepole pine which was not observed in my study. Modeling also indicated the probability of moose presence increased with shrub height, with a levelling trend commencing at approximately 4 m. With the average shoulder height of moose at 180 cm (Banfield 1974), shrub heights > 4 m not only exceed the reach of moose (though may be bent over for browsing if palatable) but offer no further gain in security cover. Unlike other regions where moose crater for vegetation6 (Renecker and Schwartz 1998), no cratering was observed to indicate the foraging on sedges (Carex spp.), other graminoids or vegetation buried beneath the snow. Probability of track presence was less and confidence intervals greater for willow as compared to bog birch cover. Bog birch was found primarily within wetlands whereas willow was found within wetland and forested areas. Willow was observed in 44% of forested plots. Ninety-two percent of these forested plots with willow had < 1 % ground cover and < 4% of these plots exhibited direct evidence of moose browse. The majority of willow stems had been heavily browsed in the past, leaving little "current" growth available for browse. Given the occurrence, but minimal ground coverage of willow within forested areas, would located the limited amounts of willow either by specifically wandering throughout forested areas in search of browse or randomly locating it 25 while moving to another area. Shrub cover was so minimal within the forested areas that wetlands, or early serai stage forest openings, with greater shrub cover and height may offer more security cover. Relocational modeling did not differentiate between natural meadows or fields cultivated for hay and straw production or used for pasture. Factoring out these components may lead to a different result as the cultivated fields may offer more palatable winter forage than of the natural meadows. The natural meadows sampled appeared to be dominated by Carex species, although confirmation of the graminoid components of these meadows needed to be done during the summer growing season and was not a component of this study. Natural meadows sampled using track transects indicated no track presence in these openings at all, while cultivated fields were not sampled. During winter reconnaissance flights, moose were often observed in cultivated fields versus natural openings; this may have been a result of visibility given the large size of the cultivated fields in relation to the smaller size of the natural meadows. Utilisation of forest cover > 40 years of age, whether coniferous, deciduous or categorized by species, appeared to be of minimal value to moose based on the GPS relocational study component. Similarly, snow depth and crown closure had no influence on track presence based on transect data. Given reduced utilisation and track presence in older forest cover as compared to wetland and stands < 40 years old, one might suggest that mature forest was of less value to moose. However, mature forest does provide thermal, security and snow interception cover during periods of extreme cold and deep snow. Thermal cover is particularly important during winter as moose are prone to thermal stress when temperatures are > -5°C or < -30°C (Renecker et al. 1978; Renecker and Hudson 1986). Moose may have experienced heat stress in 2001 and 26 2002, and possibly cold stress in 2002, based on temperatures recorded within the study area (Appendix 2). Schwab and Pitt (1991) suggested heat stress influenced winter habitat selection and can compromise foraging efficiency (Renecker and Schwartz 1998). Although not analyzed, the majority of moose beds observed during track transect surveys were located within shrubbier wetlands (i.e. not meadows) and along shrubby wetland / mature forest edges, than within mature forest itself. Shading, and security cover, by tall shrubs within openings used for browsing, or along opening edges by adjacent forest cover, may provide adequate resting or bedding sites during higher daytime temperatures. The use of forest cover, not directly on wetland/opening edges, for thermal regulation may not be required under these conditions. Sufficient forest structure is necessary to provide thermal buffers during periods of extreme cold and wind chill conditions as well as adequate security cover. The mature lodgepole pine forest dominating the study area provides reduced wind cover compared to shade tolerant conifer forest types of similar height and/or age class (e.g. spruce). Snow depth may also influence habitat selection, although it did not influence the probability of track presence during the two years of this study. Peek (1998) indicated moose will seek alternative habitat when snow depths of > 70 cm occur. Given Timmermann and Racey's (1989) suggestion that buffers may only be required adjacent to feeding sites, buffering requirements adjacent to non-shrubby meadows or openings may be less • than openings offering greater browsing opportunities because of increased shrub availability. However, this assumes such sites are not found to be utilised outside the winter season. The lack of track presence or evidence of snow cratering for grazing purposes within openings with < 6% shrub cover, based on track transects, indicated the non-utilisation of such openings during winter months. Within the study area there was little need for security cover from predators with the exception of human hunters. Two GPS collared moose were shot by poachers and a third 8 Bog birch and Salix spp; Alnus spp were not observed to be browsed. 27 carcass was found with evidence of scavenging by coyotes. Over the two winters of conducting snow track transects, fewer than a dozen predator tracks were observed9. Security cover from hunters is an issue not only of visual cover but also of hunter accessibility. Security coverage may be better addressed by incorporating habitat structure/forest stand density, adjacent to the serai habitat of importance, with distance as a variable of interaction rather than distance alone, which was addressed in this study. Insight from modeling this interaction variable would perhaps lead to a more biological approach for developing buffer/leave strips adjacent to well utilised habitats than defining a set distance based for an opening alone. This approach may also aid in defining the amount of 'leave' required to provide adequate thermal cover during periods of low air temperatures and/or high wind chill factors. A buffer/leave strip of 1-1/2 - 2 times the "secure" distance from the habitat of value, providing an adequate winter coniferous crown closure, would provide horizontal cover from extreme wind chill factors. It would also provide vertical cover providing snow interception and/or shade from solar radiation during warmer periods when heat stress is an issue. A greater distance is probably not necessary from a security point of view as it is unlikely moose predators, or hunters, would be in active pursuit at temperatures low enough to stress moose. Measures to prevent windthrow and the reduction of cover effectiveness should be addressed within forest management plans. Given the importance of natural shrubby openings to moose, limiting land conversion and use for agricultural purposes (i.e. hay and straw production, grazing from livestock) and determination of buffer widths appropriate for maintaining the hydrological integrity of these openings would further aid in maintaining these components of moose habitat. Continued clearcut harvesting in the area would also provide shrub and early serai aspen browse within the study area. 9 Predator tracks observed were cougar, wolf and coyote. 28 The degree in which moose perceive their surrounding environment may have been overestimated in this study. Track transect plots were analyzed based on track presence and ground cover within 25 m, either direction, along transect lines from plot centre while resource utilisation of relocational data was derived from a 25 m grid. Use of coarser resolutions may reveal different resource use patterns of moose at a stand level versus the sub-stand level of use considered in my study though may not have been adequate to designate buffer/leave strip widths, had distance from wetland edge, crown closure and (indirectly) stand age been factors in predicting probability of moose presence or resource utilisation. These finer scale resolutions could be converted to coarser scales, compared and then used to further estimate or refine the extent to which moose perceive the local environment within their home range from the available data. 29 5.0 Conclusion I compared low and high-tech study designs using two different scales of assessment and then predicted habitat use by moose. Analyses of track transect data indicated that shrub cover, particularly bog birch and willow, and average shrub height had a positive effect on moose use, whereas distance from an opening edge and forest crown closure adjacent to openings had no detectable effect. My GPS relocational study design also indicated that distance-from-an-opening-edge and forest composition by species had no influence on moose use of habitat, but that openings and young forest, often associated with increased shrub (browse) production, had a positive influence on habitat utilisation. Thus, the different designs lead to similar conclusions: that moose favoured areas of high browse production but showed no response to distance from the edge of openings in adjacent forested habitats. These results imply that leave strips/buffers may have little immediate effect on the use of early serai opening by moose in winter. It is possible that resources and environmental conditions during my study, based on population size, predation, hunting, forest management activities, did not lead moose to seek snow, security or thermal cover in forests adjacent to openings. However, this possibility does not rule out that climate, predation or hunting pressure will influence habitat selection in future. Although my track transect design met and fulfilled the main objectives of my study, data from the GPS study design provided opportunities to explore a broader range of habitats at the landscape level. Additional analysis of relocational data with utilisation distributions and movement paths may provide further insight as to how and what "course-of-action" moose take when moving through areas of greater versus lower utilisation. Further relocational data analysis 30 m a y identi fy forest resource uti l isat ions dur ing seasons other than w i n t e r or w i t h i n narrower, or specif ic t ime frames ( M a z l u f f et a l . 2001). U t i l i z a t i o n distributions c o u l d be created to determine c ircadian habitat ut i l i sa t ion patterns, response to specif ic environmental condit ions such as the temperature extremes considered stressful to moose, spr ing m o v e m e n t and habitat u t i l i z a t i o n patterns dur ing c a l v i n g season (i.e. site f idel i ty) and s i m i l a r l y , m o v e m e n t and habitat use i n relat ion to fa l l breeding and/or hunt ing season. T h e shape o f these ut i l i sa t ion distr ibutions w i l l further our perception o f the space and scale ( M a r z l u f f et a l . 2001) i n w h i c h m o o s e perceive and ut i l i se the avai labi l i ty o f different habitats. 31 References Banfield, A.W.F. 1974. The mammals of Canada. University of Toronto Press. 438pp. Brusnyk, L.M. and Gilbert, F.F. 1983. Use of shoreline timber reserves by moose. Journal of Wildlife Management. 47:673-685. Burnham, K.P. and Anderson, D.R. 1998. Model selection and inference: a practical information-theoretic approach. Springer-Verlag, New York. Burnham, K.P. and Anderson, D.R. 2002. Model selection and multimodel: a practical information-theoretic approach. 2nd Ed. Springer-Verlag, New York. 489pp. Hepinstall, J.A., J.M. Marzluff, M.S. Handcock and P. Hurvitz. 2003. Incorporating resource utilization distributions into the study of resource selection: dealing with spatial autocorrelation In Proceedings of the first international conference of resource selection by animals, University Of Wyoming, Departments Of Statistics, Zoology and Physiology and West, Inc. Omnipress. Le Cessie, S. and van Houwelingen, J.C. 1991. A goodness-of-fit test for binary regression models, based on smoothing methods. Biometrics. 47:1267-1282. Fitzmaurice, G.M. 1995. A caveat concerning independence estimating equations with multivariate binary data. Biometrics. 51:309-317. Hannon, Susan J. Cynthia A. Paszkowski, Stan Boutin, Jordan DeGroot, S. Ellen Macdonald, Matt Wheatley, and Brian R. Eaton. 2002. Abundance and species composition of amphibians, small mammals, and songbirds in riparian forest buffer strips of varying widths in the boreal mixedwood of Alberta. Can. J. For. Res. 32(10): 1784-1800 (2002) Harrell Jr., F.E., Lee, K.L., and Mark, D.B. 1996. Multivariate prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Statistics in Medicine. 15:361-387. Johnson, D.H. 1980. The comparison of usage and availability measurements for evaluating resource preference. Ecology. 61:65-71. Langley, M.A. and Pletscher, D.H. 1994. Calving areas of moose in northwestern Montana and southeastern British Columbia. Alces. 30:127-135. Nagelkerke, N.J.D. 1991. Miscellanea: A note on a general definition of the coefficient of determination. Biometrika. 78(3): 691-692. Marzluff, J.M., S.T. Knick and J.J. Millspaugh. 2001. High-tech behavioural ecology: modeling the distribution of animal activities to better understand wildlife space use and resource selection. In Millspaugh, J.J. And J.M. Marzluff. 2001. Radio tracking and animal populations. Academic Press. San Diego and London. Chap 12., pp. 309-326. 32 Marzluff, J.M., J.J. Millspaugh, P. Hurvitz and M.S. Handcock. 2004. Relating resources to a probabilistic measure of space use: forest fragments and steller's jays. Ecology, 85(5): 1411-1427. Peek, J.M. 1998. Habitat relationships. In Ecology and management of the North American moose. Franzmann, A.W. and Schwartz, C.C. (eds). Smithsonian Institution Press, Washington and London, Chap. 11, pp. 351-376. R Development Core Team (2003). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-00-3, URL http://www.R-project.org. Renecker, L.A. and Hudson, R.J. 1986. Seasonal energy expenditure and thermoregulatory response of moose. Canadian Journal of Zoology. 64:322-327. Renecker, L.A and Schwartz, C.C. 1998. Food habits and feeding behaviour. In Ecology and management of the North American moose. Franzmann, A.W. and Schwartz, C.C. (eds). Smithsonian Institution Press, Washington and London, Chap. 13, pp. 403-439. Renecker, L.A., Hudson, R.J., Christophersen, M.K. and Arelis, C. 1978. Effect of posture, feeding, low temperature and wind on energy expenditure of moose calves. In Procedures of North American Moose Conference and Workshop, Number 14, April 1978, Halifax, Nova Scotia, pp. 126-140. Samuel, M.D. and M.R. Fuller. 1994. Wildlife telemetry. Pages 317-418 in T.A Bookhout, ed. Research and management techniques for wildlife and habitat. Fifth ed. The Wildlife Society, Bethesda, Md. Steen, O.A. and Coupe, R.A. 1997. A field guide to forest site identification and interpretation for the Cariboo Forest Region. B.C. Ministry of Forests, Land Management Handbook 39. Crown Publications Inc., Victoria, B.C. Schwab, F.E. and Pitt, M.D. 1991. Moose selection of canopy cover types related to operative temperature, forage, and snow depth. Canadian Journal of Zoology. 69:3071-3077. Thomas, D.L. and E.J. Taylor. 1990. Study designs and test for comparing resource use and availability. J. Wildl. Manage. 54:322-330. Thompson, I.D. and Stewart, R.W. 1998. Management of moose habitat. In Ecology and management of the North American moose. Franzmann, A.W. and Schwartz, C.C. (eds). Smithsonian Institution Press, Washington and London, Chap. 12, pp. 377-401. Timmermann, H.R. and Racey, G.D. 1989. Moose access routes to an aquatic feeding site. Alces. 25:104-111. 33 References - Appendices Environment Canada. 2004. Monthly Data Report for 2001 and 2002: Spokin Lake 4E. Available from http://www.climate.weatheroffice.ec.gc.ca/climateData/monthlydata_e.html [cited 17 January 2005]. Environment Canada. 2004. Canadian Climate Normals or Averages 1971 - 2000: Spokin Lake 4E. Available from http://www.climate.weatheroffice.ec.gc.ca/climate_normals/ results_e.html [cited 17 January 2005]. 34 Appendix 1: Plot shrub ground cover and browse by species. Wetland plots (n = 218) Forest plots (n = 457) Shrub species #(%) # (%) # (%) average #(%) # (%) # (%) average and code plots plots plots with amount plots plots plots with amount with where direct browsed with where direct browsed species present evidence (%) species present evidence (%) present with < browse where present with < browse where 1% observed" browse 1% observed" browse ground present ground present cover cover Alnus incana - - - - 14 (3) 8 (57) 1 (7) 10 (ALIN) Amelanchier 1 (<]) 1 (100) - - 83 (18) 82 (99) 1 (1) 5 alnifolia ( A M A L ) Betula 136 (62) 19 (14) 25 (18) 10.2 25 (5) 18 (72) - -glandulosa (BEGL) Corn us 1 (<1) 1 (100) - - 6 (1) 6 (100) - -slolonifera (COST) Ledum - - - - 2 (<1) 1 (50) - -groenlandicum (LEGR) Lonicera 7 (3) 6 (86) 1 (14) 1 115 (25) 106 (92) 3 (3) 8.7 in volucrata (LOIN) Ribes spp. 1 (<1) 1 (100) - - 4 (1) 4 (100) - -(Ribes) Rosa spp. 6 (3) 6 (100) 1 (17) 1 284 (62) 272 (96) 2 (<1) 55.5 (Rosa) Salix spp. 75 (34) 28 (37) 22 (29) 18 202 (44) 186 (92) 7 (3) 25 (Salix) Shepherdia candensis (SHCA) Spiraea belulifolia (SPBE) Symphoricarpos albas (SYAL) Vaccinium myrtilloides ( V A M Y ) Viburnum edule (VIED) 121 (26) 96 (79) 1 (<1) 1 (<1) I (100) 28 (6) 23 (82) 32 (7) 32 (100) 42 (9) 42 (100) 40 " where species present 35 Appendix 2: Spokin Lake weather station3 temperature data (Environment Canada 2004) Year Month Mean maximum temperature ( °Q Mean temperature (°C) Mean minimum temperature (°C) Extreme maximum temperature (°C) Extreme minimum temperature (°Q 2001 January 1.1 -4.5 -10.0 6.0 -20.0 February -1.4 -8.8 -16.2 7.0 -31.0 March 5.2 -1.2 -7.6 14.0 -17.5 2002 January- -1.9 -6.5 -11.2 7.0 -34.5 February 0.8 -5.5 -11.8 5.5 -31.0 March -0.2 -7.1 -13.9 11.0 -33.5 Climate Normals (Averages)6 January -2.8 -8.1 -13.4 9.0 -43.0 February 0.7 -5.4 -11.5 11.0 -43.0 March 5.4 -1.0 -7.4 17 -35.0 " Latitude: 52 10'N Longitude: 121 4 1 ' W Elevation: 1040 m b based on average temperature readings from 1984 to 2000. 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0074938/manifest

Comment

Related Items