Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Effects of fine-scale and landscape-level habitat features on a sagebrush breeding birds of the southern… Paczek, Susan 2002

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

Item Metadata


831-ubc_2002-0197.pdf [ 4.3MB ]
JSON: 831-1.0090260.json
JSON-LD: 831-1.0090260-ld.json
RDF/XML (Pretty): 831-1.0090260-rdf.xml
RDF/JSON: 831-1.0090260-rdf.json
Turtle: 831-1.0090260-turtle.txt
N-Triples: 831-1.0090260-rdf-ntriples.txt
Original Record: 831-1.0090260-source.json
Full Text

Full Text

EFFECTS OF FINE-SCALE AND LANDSCAPE-LEVEL HABITAT FEATURES ON SAGEBRUSH BREEDING BIRDS OF THE SOUTHERN OKANAGAN AND SIMILKAMEEN VALLEYS, BRITISH COLUMBIA by Susan Paczek B.Sc, McGill University, Montreal, 1994 A THESIS S U B M I T T E D I N P A R T I A L F U L F I L L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F M A S T E R O F S C I E N C E in T H E F A C U L T Y OF G R A D U A T E S T U D I E S Facu l t y o f Fo res t ry (Department o f Forest Sc ience) We accept this^hesis as conforming to the required standard T H E U N I V E R S I T Y ^ F B R I T I S H C O L U M B I A A p r i l , 2002 © Susan Paczek, 2002 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of -Vp/g^^ *S<^ > g,^ vca-S The University of British Columbia Vancouver, Canada Date fiprl\ I'b IT-OPT-DE-6 (2/88) ABSTRACT I determined habitat associations for five species of songbirds breeding in sagebrush of the southern Okanagan and Similkameen Valleys, British Columbia. I examined the relative importance of scale for the presence and relative abundance of these species, through the measurement of vegetation floristics and structure at a local level (<100 m), and habitat context at three landscape scales (500 m, 1 km and 2 km). Vegetation and bird survey data were collected at 245 point count stations in 1998. Local-level habitat variables were derived from field surveys, while landscape-level variation was classified from a single Landsat Thematic Mapper ( TM ) image from 1996. Within sagebrush habitat the Landsat TM image was also classified at a fine-scale to determine if local-level habitat variation could be mapped by satellite data. Accuracy of the classification was assessed in 1999 by ground-truthing. Overall accuracy was 85%, and 78% for the sagebrush 'subtypes'. Local-level models from Landsat TM classified sagebrush habitat subtypes agreed with habitat associations identified from vegetation survey data, indicating that satellite data may be used as a surrogate for field data, although these associations were relatively weak. Performance of local, landscape, and local + landscape-level models was assessed from ranked Akaike's Information Criteria (AIC) scores. For all songbird species, logistic regression models showed the strongest habitat associations at a local level. Floristic variables were often more important than vegetation structure variables. Brewer's Sparrow was associated with large tufted perennials: parsnip-flowered buckwheat (Eriogonum heracleoides) and lupine (Lupinus sericeus or sulphureus). Lark Sparrow was positively associated with sand dropseed grass (Sporobolus cryptandrus), Vesper Sparrow (Pooecetes gramineus) was positively associated with lupines, and Western Meadowlark was positively associated with needle-and-thread grass (Stipa comatd). The addition of landscape-level variables usually improved the predictive ability of survey-derived local habitat association models. Habitat associations varied markedly for each species, and songbird relative abundance responded differently to the scale of measurements. Herb layer species identified as important in habitat associations were in turn correlated with rangeland management practices. Management recommendations are presented to direct conservation efforts in this highly threatened area. n TABLE OF CONTENTS ABSTRACT LIST OF TABLES LIST OF FIGURES ACKNOWLEDGEMENTS CHAPTER ONE: General Introduction CHAPTER TWO: Classification of Landsat TM data Introduction Methods Study Site Selection Classification Accuracy Assessment Assessing the utility of sagebrush subtypes 1 Results • 1 Discussion CHAPTER THREE: The effect of local and landscape-level variation on songbird habitat associations ; Introduction Methods Bird Surveys Vegetation Surveys GIS data: topography GIS data: habitat classes Statistical Analyses Results Summary of Bird Survey Data Local-level Models: Floristics Versus Structure The importance of Landscape-level variables at different scales Discussion Habitat associations of shrubsteppe birds at a local level Utility of Landsat TM data, and landscape-level associations CHAPTER FOUR: Management Recornrnendations iii REFERENCES 62 APPENDIX 1 69 APPENDIX II 70 APPENDIX III 72 APPENDIX IV 73 APPENDIX V 74 iv LIST OF TABLES TABLE 1. Names of study areas with approximate elevation 7 TABLE 2. Sagebrush Habitat Classes Used for Image Classification 10 TABLE 3. Error Matrix of classification of the 14 Landsat TM classes 14 TABLE 4: Summary of user (commission) and producer accuracy (omission) for the Landsat TM classes 15 TABLE 5. Results of individual logistic regressions 38 TABLE 6. Results of local level stepwise regressions for the focal species 41 TABLE 7. Results of stepwise logistic regression models with Landsat TM derived sagebrush subtype variables, and sagebrush subtype models with addition of topographic variables... 42 TABLE 8. Backwards elimination stepwise logistic regression model results from landscape-level variables at the 500 m, 1 km, and 2 km scale 47 TABLE 9. Significance of the step when the blocks of landscape variables from the models in Table 8 were introduced to the local-level models (100 m) from Table 6 49 TABLE 10. Ranked AICc scores and AICc weights for all models for all species 50 v LIST OF FIGURES FIGURE 1. Study area showing the distribution of study sites and point count stations 6 FIGURE 2. Mean and standard deviation of digital numbers of TM Bands 4 and 5 13 FIGURE 3. Relationship of Landsat TM classes to sagebrush abundance. 16 FIGURE 4. Relationship of Landsat TM classes to sand dropseed grass abundance 17 FIGURE 5. Relationship of Landsat TM classes to lupine abundance 18 FIGURE 6. Brewer's Sparrow relative abundance 31 FIGURE 7. Lark Sparrow relative abundance 32 FIGURE 8. Grasshopper Sparrow relative abundance 33 FIGURE 9. Western Meadowlark relative abundance 34 FIGURE 10. Vesper Sparrow relative abundance 35 vi ACKNOWLEDGEMENTS Funding for this thesis was generously provided by HCTF, NSERC, Environment Canada and Career Edge. I would like to thank my committee, Dr. Pam Krannitz, Dr. Kathy Martin, Dr. Brian Klinkenberg and Dr. Jamie Smith for their useful comments along the way. I would also like to thank Dr. Val Lemay for the considerable effort she put into editing my thesis as my external examiner for my thesis defense. The British Columbia Ministry of Forests provided the Landsat TM data, and the Ministry of Environment, Lands and Parks provided access to TRIM and TEM data. The following people and organizations graciously permitted me to conduct bird surveys on their land: Osoyoos First Nations.Band, Penticton First Nations Band, Lower Similkameen First Nations Band, the Nature Trust, Bill Clements, Wilson Clifton, Bob Ellis, Dick Francis, Lynn and Noni Kusler, Leigh McFayden, Clarence Schneider and John McGibbon. I am indebted to Kristin Dust, Laura Stepnisky, Kim Lisgo and Paul Williams for all the early mornings and long hours they put in slogging through the desert to assist with data collection. I would like to thank my family for their support since I began this project. Finally, Stephanie Melles, Clive Goodinson, Arnold Moy and numerous other members of the Centre for Applied Conservation Biology provided me with valuable, technical advice that was essential to the completion of my thesis. vn CHAPTER ONE: GENERAL INTRODUCTION In habitat association studies, ecologists often ignore the importance of sampling scale in their study design by selecting a single, anthropocentric scale on the basis of tradition (Wiens 1989). Since the perception of a landscape depends on the organism being studied, a single scale cannot be appropriate for the assessment of the relative strengths of songbird habitat associations over a heterogenous region (Wiens 1989). Choice of scale affects the efficacy of habitat selection studies, since variables that are important at one particular scale may not be important at others. For example, Red-winged Blackbirds (Agelaius phoeniceus) prefer nesting sites with a high density of cattail shoots, but cattail density is not important at the scale of blackbird territories (Pribil and Pieman 1997). Ecologists have been slow, relative to scientists in other fields, to recognize the importance of scale, perhaps in part because of a lack of available tools in this discipline to expand perception (Wiens 1989). This has changed in recent years with the increased use of Geographic Information Systems (GIS) and remote sensing data in ecological studies, which has facilitated the quantification of habitat variables at different scales. This in turn permits the identification of habitat features that may be important in habitat selection. For example, remote sensing data and GIS were used to show that the suitability of elk calving habitat increased closer to seep pits (Bian and West 1997). Measurement of landscape variables further enables researchers to evaluate whether local or landscape-level variables provide better predictors of habitat suitability (i.e. Melles 2000, Bolger etal. 1997). Within a landscape, habitat associations can be described at local scales ranging from the floristic (vegetation species-level) through to broad vegetation structure. In this thesis, I define vegetation floristics as data pertaining to the abundance of particular plant species (Rotenberry 1985), as opposed to vegetation structure. Vegetation structure, or cover, in this thesis refers to a two-dimensional measure of structure, which I define as overall grass or overall forb percent cover. At local levels, the search for broad generalizations in habitat associations has occurred at the expense of accuracy in identifying important relationships (Wiens and Rotenberry 1981). The tradition of collapsing vegetation data into summary variables such as diversity indices, coupled with the difficulty of analyzing data from individual vegetation species, could explain why vegetation structure has been used more often than floristics to study avian habitat associations (Rotenberry 1985). In this thesis, I examined habitat associations through a range of scales, from the floristic level through to the landscape level. 1 Grasslands and shrubsteppe of British Columbia contain a disproportionately high number of endangered species relative to the 3% of area that these habitat types occupy (Hooper and Pitt 1995). At the northern extent of the Great Basin, the shrubsteppe habitat of the southern Okanagan and Similkameen Valleys supports many species that are at the limit of their range and are unique in Canada. This area contains a high concentration of bird species at risk, including provincially red-listed shrubsteppe birds such as Brewer's Sparrow (Spizella breweri breweri), Grasshopper Sparrow (Ammodrammus savannarum) and Lark Sparrow (Chondestes grammacus). In addition, North American Breeding Bird Survey data compiled for the period 1966 to 1996 showed significant decreases in populations of Brewer's Sparrow, Lark Sparrow, Grasshopper Sparrows and Western Meadowlarks (Sturnella neglecta; Peterjohn and Sauer 1999). While the distributions of songbird species are well known within the South Okanagan (Cannings et al. 1987), the specific habitat associations of bird communities within sagebrush habitat of this region are not. The South Okanagan and Similkameen Valleys of British Columbia provide an ideal location to test the importance of different scales in habitat selection by breeding songbirds. The variety and structure of the sagebrush habitats in this area permits an examination of patterns of habitat association at a range of scales. Variation in topography and disturbance histories has resulted in a mosaic of shrubsteppe patches that are interspersed with other habitats including forest, riparian strips, agriculture and urban development. Within sagebrush patches, there is variation not only in shrub size and density, but also in the structure and taxonomy of the herb layer. Associations between bird species and the floristics or structure of the herb layer could explain the patchy distribution of birds that has previously gone unmeasured. In Chapter Two of this thesis I describe the methods I used to classify a Landsat Thematic Mapper (TM) image. This classification produced the landscape-level variables and sagebrush habitat classes used to model habitat associations in Chapter Three and the accuracy of this ' classification is assessed. I discuss the fine-scale sagebrush habitat description from satellite data, which appears to be a potentially useful tool for extrapolating bird-habitat associations over landscapes. In Chapter Three I describe bird-habitat associations for five shrubsteppe songbird species at a range of scales by using logistic regression to relate bird survey data to a combination of vegetation survey and remote sensing (Landsat Thematic Mapper) data. Bird survey and vegetation survey data were collected in a single breeding season in 1998. Within sagebrush habitat at a local (100 m radius) scale, I examined the relative importance of herb layer floristics 2 and the proportion of overall vegetation cover to sagebrush breeding songbirds. I tested the utility of a fine-scale Landsat Thematic Mapper (TM) classification of sagebrush habitat to determine if satellite data can be used to map local-level habitat associations. Additional habitat variables classified from Landsat TM data are used to determine the importance of the surrounding habitat, or 'context' of bird survey stations at 3 landscape scales: 500 m, 1 km and 2 km. The relative importance of landscape (500 m, 1 km, 2 km) and local-level (100 m) variation was then determined for all species using ranked Akaike's Information Criteria (AIC) scores of logistic regression models. In all cases, local-level models gave the best predictors of bird-habitat associations, although addition of landscape-level variables often improved their predictability I conclude with a discussion of management implications for the five focal songbirds I studied (chapter four). The disparity among the habitat associations of these species, is emphasized in order to encourage managers to maintain specific habitat features required by each species. 3 CHAPTER TWO: CLASSIFICATION OF LANDSAT T M DATA ABSTRACT A single classified Landsat Thematic Mapper image from 1996 was used to quantify landscape-level variation surrounding sagebrush habitat, and fine-scale variation within sagebrush habitat in the South Okanagan and Similkameen valleys, British Columbia. I used a combination of unsupervised and supervised classification methods with data derived from vegetation surveys conducted in 1998 and colour aerial photographs from 1996. Accuracy of the classification was assessed in 1999 by ground-truthing. Overall accuracy was 85%, and 78% for the five sagebrush 'subtypes'. INTRODUCTION In contrast to typical aerial photographs, satellite data detect wavelength regions of the electromagnetic spectrum that are both within and beyond human perception, and convey a large amount of information on vegetation abundance and health (Lillesand and Kiefer 1994). The pixel-based digital format of satellite data further allows for the interpretation of each unit to be automated by computer methods, thus allowing users of satellite imagery to classify landscapes over large areas relatively quickly and consistently. Spectral data can be translated into biologically meaningful variables, permitting the use of these data in inventory and monitoring of agricultural areas, including rangeland (Tueller 1992). Landsat-MSS data have been used successfully to distinguish grass species with different grazing responses in Patagonia (Paruelo and Golluscio 1994), and to monitor the impact of drought on rangelands in Australia (Graetz et al. 1988). In grasslands in the United States, the Normalized Difference Vegetation Index (NDVI), a satellite data derived index, was positively correlated with Aboveground Net Primary Production (ANPP) which was sampled in the field (Paruelo et al. 1997). Recently, the analysis of satellite spectral information has been applied to conservation biology. In particular, it has been used to quantify fine-scale patterns of vegetation communities that may relate to species distributions. Landsat TM data were used to identify attributes of old growth forest related to the habitat of Spotted Owls (Congalton et al. 1993). Satellite images in the Sahel, Africa, have been used to map bird diversity through the successful identification of wet areas that are associated with the distribution of migrating birds (Nohr and Jorgensen 1997). 4 Lauver and Whistler (1993) characterized grasslands in Kansas with satellite classification, and successfully predicted sites with rare plant species - 'high quality' grasslands were those that contained abundant forbs. In the following chapter, Landsat TM derived variables are used to assess the importance of landscape-level variation on grassland bird distribution and relative abundance. Associations are further described between songbirds and five sagebrush 'subtypes' that are the product of a fine-scale classification of sagebrush habitat. In this chapter, I describe the methods I used to classify a Landsat TM image into these different landscape classes and sagebrush subtypes, and I assess the accuracy of this classification. METHODS Study Site Selection Study sites were located in the lower South Okanagan and Similkameen valleys, British Columbia, between approximately 49° 00' 00" N and 49° 25' 00" N, and 119° 49' 00" W and 119° 27' 00" W (Figure 5). Elevations ranged between 345 m and 1200 m, within the Ponderosa Pine - Bunchgrass and Interior Douglas-fir biogeoclimatic zones (Medinger and Pojar 1991). All of the survey areas were dominated by big sagebrush, Artemisia tridentata. This eliminated any confounding effects of having sites that had high abundance of antelope bitterbrush (Purshia tridentata), a dominant shrub species of the region that is restricted to sandy benchlands. Variation in the herb understory was used to guide point count station selection. Understory types included areas dominated by sand dropseed grass, needle-and-thread grass and areas with abundant forbs. Understory types were initially defined by visual observation. Detailed vegetation sampling was then conducted to quantify differences between point count stations. The sites occurred in 15 geographically distinct areas, several of which contained more than one habitat type (Figure 5). Herb understory types were replicated at three or more of these areas. Study sites had a range of sagebrush densities, varied in surrounding landscape characteristics, and included a range of aspects. Point count stations were at least 300 m apart. Within an area, stations were placed so that each plot was in the center of an area of similar sagebrush density and homogeneous understory. The 15 areas had between 3 and 34 point count stations, depending on the size of the area, for a total of 245 survey stations (Table 4). Survey stations were mapped to +5 m with a handheld Trimble GPS unit that was corrected for selective availability error with a Global Surveyor. 5 N A 10 15 Kilometers FIGURE 1. Study area showing the distribution of study sites and point count stations in the southern Okanagan and Similkameen Valleys for songbird surveys conducted in 1998. TABLE 1. Names of study areas with approximate elevation. Area Number Area Name #Stations Elevation (Figure 5) (m) 1 Armstrong Creek 29 920 2 Blind Creek 21 720 3 West Similkameen 20 430 4 East Similkameen 11 420 5 Chopaka 26 500 6 International Grasslands West 28 910 7 KilpoolaLake 3 840 8 International Grasslands East 23 900 9 Ecoreserve E12 13 530 10 Richter Pass 18 620 11 Manuel Flats 13 430 12 White Lake 34 540 13 Yellowlake Creek 8 1030 14 T6 Ranch 8 800 15 Matron Valley 5 910 7 Classification I classified bands 1 through 7 of a single Landsat TM scene taken on July 6, 1996. The pixel size was 30 m by 30 m. Vegetation data collected at 245 survey stations (described in chapter three) were used to guide the classification of sagebrush subtypes. The surrounding landscape within 2 km of stations was classified using colour 1:50 000 aerial photographs taken July 1996. The photos were orthorectified at a scale of 2 m by 2 m using PCI software and Terrain Resource Inventory Mapping (TRIM) road and elevation data. Orthorectification of aerial photographs means that the aerial photographs are scanned into digital format and are corrected (rectified) for differences in elevation. Orthophotos may be viewed with other spatial data, and the location of landscape features identified on these photos can be used in satellite classification. I used two methods of satellite image classification: 'unsupervised' and 'supervised' classification. Unsupervised classification .is similar to cluster analysis in that satellite data are sorted into spectrally pure classes based on differences in the digital numbers of each band of the satellite image (i.e., spectrally relevant). Unsupervised classification was used to separate shrubsteppe-dominated areas of the image. I used the Idrisi version 2.01 software module ISOCLUST, an iterative self-organizing classifier, to classify the image into 16 spectrally defined classes based on all seven bands of Landsat TM data. The resultant classes were compared to the orthophotos, and three classes were identified as being primarily shrubsteppe. ISOCLUST was applied a second time to the areas encompassed by these three classes and two more clusters that did not correspond to shrubsteppe habitat were removed. This step allowed the subsequent supervised classification of different sagebrush habitat types to be limited to shrubsteppe areas. In contrast to unsupervised classification, supervised classification is based on features that the user identifies as being important (thematically relevant). Supervised classification proceeds with the definition of training sites. Training sites are user-defined areas of thematic similarity; several are typically defined for each class. Each of the seven bands of a Landsat TM image represents a different area of the electromagnetic spectrum, from visible through thermal infrared. Each pixel in the image has an associated digital number for each of these bands. The classification software uses this spectral information to assign the remaining pixels to one of the user defined classes, based on the similarity of the digital numbers of unclassified pixels to those of the pixels in the training sites. Classes identified in supervised classification are not 8 necessarily spectrally distinct, so it is important to assess the accuracy of images that are classified with this method. I used supervised classification within sagebrush areas and the surrounding landscape. Six classes within sagebrush habitat were determined by sorting vegetation data presented in Chapter Three. These classes were based on sagebrush density, forb density, grass type and density, and the proportion of bare soil (Table 1). Classes were defined by sorting the data for these variables and then visually choosing cut-off values which corresponded with "sparse, mid-density and dense" sagebrush, as well as "dry" versus "forb" sites in the field. I used the vegetation survey data to identify training sites for these six sagebrush classes. Training sites were digitized around the study plots that best represented these site types. Vegetation data and location information from the survey stations were incorporated into a GIS layer and overlaid on the orthophotos in Idrisi as a guide to delineating training sites. This ensured that I excluded trees and other features so that the training sites were as thematically pure as possible. I used only the orthophotos to identify training sites for supervised classification of the following habitat classes in the surrounding landscape: agriculture, orchard/vineyard, meadow, deciduous forest/shrub thickets, urban, water, silt/bare soil and mixed forest/shrubsteppe. Field data was not felt to be necessary for defining these classes, as they were easily identified from the high-resolution photos. 'Agriculture' refers to row crops and dry pastures, 'meadow' is wet meadow areas including wetlands, and the 'mixed forest/sagebrush' category is the interface between coniferous forest and sagebrush habitat (10 - 50% tree cover). All training sites contained at least 70 pixels (6.3 ha), and were obtained from three or more areas of the image. I used the MAXLIKE (maximum likelihood) module in Idrisi to classify the remaining pixels based on the training sites. This is a hard classifier, meaning that it assigns pixels to a single class based on their similarity (maximum likelihood) to the signature files created from the training areas. The system defaults were used with all classes having equal prior probability, and no pixels were excluded from the classification. 9 TABLE 2. Sagebrush Habitat Classes Used for Image Classification Code Name Description DF Dense sagebrush with forb understory >30% sage cover, understory dominated by large forbs, especially lupine, < 20% bare soil. Dominant grasses include: spreading needlegrass (Stipa richardsonii), junegrass (Koeleria macranthd). MF Mid density sagebrush with forb understory >15% sage cover <25%, understory as above SF Sparse sagebrush with forb understory sage cover <10%, understory as above DD Dense sagebrush with dry understory >30% sage cover, understory dominated by bare, sandy or rocky soil (>30%), low forb abundance, Dominant grasses include: red 3-awn grass (Aristida longisetd) and sand dropseed grass MD Mid density sagebrush with dry understory >15% sage cover <25%, understory as above SD Sparse sagebrush with dry understory sage cover <10%, understory as above Accuracy Assessment Sagebrush classes were validated in 1999, between 25 June and 10 July, approximately the time of year that the image was taken. Selection of random verification sites was performed prior to field visits in Idrisi. First, the final classification was filtered with the PATTERN module. This reduced error associated with locating points from the satellite image on the ground. I used PATTERN so that a pixel was only eligible for ground-truthing if it was the center of a block of nine or more pixels of the same type. Areas of the image within the training sites were not used for verification. The result was six layers of eligible pixels, one for each class. A layer of random points was then generated and overlaid on each layer to generate a list of points in each class for verification on the ground. Field verification was mainly limited to areas between and around the point count stations since the classification was designed to describe habitat <2 km from the study sites. I visited all ground-truthed sites. Sites were located with a handheld GPS and attached global surveyor that corrected the selective availability error 10 of GPS satellite signals to + 5 m. A protocol was used to assess the accuracy of ground-truthed points, by visually determining how they compared to definitions for the habitat classes (Appendix I). Habitat classes in the landscape surrounding sagebrush were assessed visually with the orthophotos. Random sites were selected for verification using the methodology described above. To assess the accuracy of the verification sets, an error matrix was constructed by assigning each ground-truthed site to the habitat class that it best represented. There are two types of classification accuracy: user accuracy and producer accuracy (Lillesand and Kiefer 1994). User accuracy (commission error) is the probability that a pixel as coded on the map is actually what the map says it is on the ground, and indicates how useful the classified map is for that particular habitat type. Producer accuracy (omission error) is the probability that a pixel on the ground has been correctly identified on the map, and therefore indicates how well the classification scheme has performed at mapping a habitat type. Overall accuracy was further determined for sagebrush classes, and for all classes. Cohen's kappa was calculated for all classes in order to provide a statistical estimate of classification accuracy (Lillesand and Keifer 1994). Assessing the utility of sagebrush subtypes Three vegetation species (sagebrush, sand dropseed grass, and lupine) were among those that had significant associations with shrubsteppe bird distribution and abundance (Chapter Three, Table 6). If strong relationships exist between the Landsat TM categories and these species, as determined by plotting the abundance of the species against the sagebrush subtypes, then the sagebrush classification is likely to predict habitat suitability for these bird species. RESULTS Success of the supervised classification can be visualized with bivariate plots of the mean and standard deviation of digital numbers of each habitat class for the seven different TM bands. Most classes that I chose as thematically important appeared to be spectrally separate (Figure 2). In this example, Bands 4 and 5 were used. The exceptions were the 'mid-density sagebrush with forb understory' (MF) and 'dense sagebrush with forb understory' (DF) classes, which had almost identical distributions. Classes with high forb density separated out from the dry classes, as they had a higher value for Band 4. Band 4 is the near infrared band that indicates vegetative 11 biomass (Lillesand and Keifer 1994). Band 5 indicates vegetation and soil moisture (Lillesand and Keifer 1994). For the sagebrush subtypes, the value of this band increased as sagebrush density decreased (Figure 2). The 'sparse sagebrush with dry understory' category had the highest value in this band, likely due to the high reflectance of bare soil. The majority of misclassifications occurred between classes that were thematically similar (Table 3). The accuracy assessment, based on the ground-truthed points, yielded a 78% user accuracy for sagebrush categories overall (Table 4). I collapsed the 'mid-density sagebrush with forb understory' (MF) and the 'dense sagebrush with forb understory' (DF) classes into a single class (MDF) to reduce error for analyses in Chapter three (66% user accuracy, 85% producer accuracy). At the coarsest level, Landsat TM data permitted a highly accurate separation of sagebrush from other habitat classes. When all of the six sagebrush types were pooled in one 'sagebrush' category the accuracy was 215 out of 229 correctly mapped pixels for a user accuracy of 94%, and 215/220 correctly identified ground sample pixels for a producer accuracy of 98% (Table 3). A large source of error in classifying the sagebrush habitat types was found between the dense sagebrush and the mixed forest/shrub classes. User accuracy was low for the 'mid density sagebrush with forb understory' and 'dense sagebrush with forb understory' classes because many of these mapped pixels were 'mixed forest/sagebrush' habitat on the ground (Table 3). I manually corrected this error in Arc View (version 3.1) by using the orthophotos as reference to re-label misclassified pixels for all areas within 2 km of the point count stations. The corrected classification was used for the analyses in Chapter Three. 12 70 65 -\ 60 MF DF SF SD 55 A 50 A DD MD 45 65 70 75 80 85 90 95 100 105 TM Band 5 (digital number) F I G U R E 2. Mean and standard deviation of digital numbers of T M Bands 4 (near infrared) and 5 (mid-infrared) for training sites of sagebrush classes. Higher values of Band 4 indicate vegetation vigor, biomass content and soil moisture. Band 5 indicates vegetation and soil moisture, with drier sites having higher values. Separation of the sagebrush classes along these axes indicates whether or not they are spectrally distinct. D D = dense sagebrush with dry understory, D F = dense sagebrush with forb understory, M D = mid density sagebrush with dry understory, M F = mid density sagebrush with forb understory, SD = sparse sagebrush with dry understory, SF = sparse sagebrush with forb understory. 13 T 3 00 00 P © O OS H H u o Q fa to Q Q Q Q 9 H oi •< r-~ o T t — ' —< T t — ' V ) ^ ( S V I T t — ' T t V ) VO T f v© r-- T T T t T t m m as VO T t in vo T t < N f 3 e T t - H ( N U"> 90 T t 00 ve oo o -*-> 00 1-1) a 3 O fa 1/3 -3 T3 43 .•S X i 0O r H X> ID t>0 (D eo cd 00 3 -a oo a OH 1/3 II Pi 00 o 13 o 60 C O u 00 3 cd 3 -a c 3 a 3 xs | J H 3 O £ CL> O l — 1 3 2 o X i 00 <D 00 00 a> 00 3 X i oo s X) < D cd 00 C J 00 XI 00 ±2 3 u -"^ C »- oo II H •a 00 3 O 3 ' o <u X) II Q »—i CJ w Q xf 00 00 cd >, <u 3 cd XI o o >^ O oo Q 2 1 1 00 O "§ fa o X> ^ 2 X I C D d cd O 00 O (U oo -o C J X) II £5 Q X S A PH H ffl -° s s s XI oo 00 T3 3 j d o cd (U a TABLE 4: Summary of user (commission) and producer accuracy (omission) for the Landsat TM classes1. Kappa coefficient of agreement is 81% (p<0.001). USER ACCURACY PRODUCER ACCURACY Landsat TM Class #correct/ %correct #correct/ %correct map total ground total DD 25/38 66 25/25 100 DF 20/32 63 20/20 100 MD 32/36 89 32/45 71 MF 26/38 68 26/34 76 SD 41/47 87 41/54 76 SF 34/38 89 34/42 81 MIXED 56/61 92 56/79 71 DECID 42/56 75 42/42 100 CONIFER 46/53 87 46/56 82, AGRIC 54/64 84 54/80 68 MEADOW 44/45 98 44/59 75 ORCHARD 52/69 75 52/59 88 SILT 41/54 76 41/41 100 URBAN 47/50 94 47/48 98 SAGE CLASSES 178/229 78 178/220 81 SAGEBRUSH 215/229 94 215/220 98 overall accuracy 582/685 85 582/685 85 1 DD = dense sagebrush with dry understory, DF = dense sagebrush with forb understory, MD = mid density sagebrush with dry understory, MF = mid density sagebrush with forb understory, SD = sparse sagebrush with dry understory, SF = sparse sagebrush with forb understory, MIXED = 10 - 50% coniferous forest/sagebrush, DECID = deciduous forest/shrub thicket, CONIFER = coniferous forest, AGRIC = agriculture, MEADOW = wet meadow/wetlands, ORCHARD = orchard/vineyard, SILT = silt/bare soil, URBAN = urban areas, SAGE CLASSES = overall accuracy of 6 sagebrush subtypes, SAGEBRUSH = overall accuracy of 6 sagebrush subtypes treated as one class, overall accuracy = accuracy of all 14 classes. When the Agriculture, Meadow and Orchard categories were merged, the accuracy of the resulting category was 178 out of 179 correctly mapped pixels (99%), and 178/198 correct pixels on the ground (90%). Since these categories were functionally similar, they were combined for analyses in Chapter Three. To assess the utility of the sagebrush 'subtype' classes, they were plotted against the abundance of selected vegetation species (Figures 2-4). The 'sparse' categories had markedly lower abundance of sagebrush (Figure 2). Sand dropseed grass was most abundant in the 'dry' categories (Figure 3), and lupine was most abundant in the 'forb' categories (Figure 4). In the 15 next chapter I will show that certain shrubsteppe bird species exhibit strong associations with lupine and sand dropseed. Since these plant species are also described by the Landsat TM classification of sagebrush subtypes, this classification may be useful in creating maps of habitat suitability 0 A ' 1 ' 1 1 1 1 • 1 1 , 1 1 r -SD SF MD MDF DD LANDSAT C L A S S FIGURE 3. Relationship of Landsat TM classes to sagebrush abundance (% cover + standard error). Sagebrush data are derived from vegetation surveys at 245 point count stations, as described in chapter one. SD = sparse sagebrush with dry understory, SF = sparse sagebrush with forb understory, MD = mid-density sagebrush with dry understory, MDF = mid/dense sagebrush with forb understory, DD = dense sagebrush with dry understory. 16 SD SF MD MDF LANDSAT C L A S S DD F I G U R E 4. Relationship of Landsat T M classes to sand dropseed grass abundance (% cover + standard error). Sand dropseed grass data are derived from vegetation surveys at 245 point count stations, as described in chapter one. SD = sparse sagebrush with dry understory, SF = sparse sagebrush with forb understory, M D = mid-density sagebrush with dry understory, M D F = mid/dense sagebrush with forb understory, D D = dense sagebrush with dry understory. 17 4 > o o d 2 - | c 'o . 0 4 ' , ' ' , 1 , 1 , ' 1 r— SD SF MD MDF DD LANDSAT C L A S S F I G U R E 5. Relationship of Landsat T M classes to lupine abundance (% cover + standard error). Lupine data are derived from vegetation surveys at 245 point count stations, as described in chapter one. SD = sparse sagebrush with dry understory, SF = sparse sagebrush with forb understory, M D = mid-density sagebrush with dry understory, M D F = mid/dense sagebrush with forb understory, D D = dense sagebrush with dry understory. 18 DISCUSSION Sagebrush habitat was classified into five subtypes with a user accuracy of 78%. The sagebrush subtypes distinguished two moisture levels, 'dry' versus 'forb' sites. Knick et al. (1997) distinguished shrubsteppe habitats from grassland habitats with 80% accuracy. However, their classification of shrubsteppe and grassland subcategories was less successful, at 64% accuracy. This was likely due to the use of plant species to define categories, and also to the fact that many of their training sites were located in heterogeneous habitats (Knick et al. 1997). In this study, my training sites were defined from homogeneous areas, and the classification relates directly to moisture (i.e., total forbs, and per cent bare soil cover). Hair et al. (2000) also used a classification scheme based on moisture to divide grassland into five types and this was linked to the distribution of seven species of grassland birds in Colorado. Landsat TM data are more useful for discerning variables based on physical attributes, rather than plant species that may be similar in structure and have similar reflectances. My results are consistent with those of Lauver and Whistler (1993) who found that their 'high quality' grassland sites were characterized by a high abundance of living plant biomass (including high forb abundance), and moisture. Their classified image assisted in the discovery of populations of threatened plants in eastern Kansas (Lauver and Whistler 1993). Satellite data maps of fine-scale vegetation features can inform management decisions for the conservation of wildlife. For example, Homer et al. (1993) used satellite imagery to generate accurate models of winter habitat for sage grouse. If habitat associations can be linked to satellite data, managers suddenly have access to habitat suitability information over vast, unsampled areas, and the potential rewards of using these maps are great. Clearly, the successful use of such classifications will be scale dependent. For example, the classification in this thesis could be to distinguish sagebrush habitat from other habitats, and to assert that these habitats will contain sagebrush-breeding birds. However, this would be of lesser conservation value compared to a classification that could quantify variation within sagebrush habitat, and link these classes to the precise distributions of sagebrush bird species. The sagebrush habitat types that were defined are related to vegetation sample data, which are also related to songbird presence and relative abundance. Thus, this classification can be used to describe bird-habitat associations. However, as satellite images are used to examine smaller scales, errors increase due to limitations imposed by minimum pixel size (approximately 30 m by 30 m in my study). 19 Habitat suitability for threatened species in the South Okanagan has been mapped using Terrestrial Ecosystem Mapping (TEM) data (Warman 2001). TEM mapping is based on the interpretation of digitized aerial photographs. Unlike the TEM data available for my study area, Landsat TM classification can be used at a scale that is likely to be relevant to shrubsteppe birds. TEM polygons are often many times larger than the expected home range of some of the small shrubsteppe songbirds in my study. Furthermore, the TEM polygons lacked information on sage density and other fine-scale habitat features, and each polygon contains as many as three habitat types. In contrast, each pixel of the classified Landsat TM image has a single habitat type associated with it, and therefore these data may be easier to use in analyses. When hard classification methods are used, such as those used in this thesis, it is important to remember that there is still error associated with the classification. The classified habitats are not discrete but composed of patches with borders that do not overlap the pixels in the image perfectly. Unfortunately, it is difficult to assess the impact of this error in further analyses. An alternative to using a hard classification method would be to use a fuzzy classification method. Unlike hard classifiers, fuzzy classifiers produce a probability of class membership instead of assigning a single class to a pixel. Fuzzy classifiers may therefore give a more accurate representation of variable landscapes with indistinct edges (Foody 1996). Accuracy assessment of this method is difficult, but fuzzy classification should be considered for future analyses. It is important for users of classified data to recognize that although error has been minimized, it still exists and can affect the strength of relationships in habitat association studies. My results in this chapter indicate that satellite data are useful for creating a classification of sagebrush habitat that is relevant to shrubsteppe bird distribution. An additional application of these data may be to detect habitat loss of sagebrush as it is converted to urban and agricultural areas (Knick and Rotenberry 2000). Landsat TM classification could be an efficient way to update the TEM polygon coverage of the South Okangan as shrub-steppe habitat is lost to development. The classification in this analysis was merely a first pass at exploring the use of Landsat TM imagery for describing sagebrush breeding bird distributions. It is important to note that the graphs used to display the utility of the classification (Figures 3-5) were created from the same vegetation data that were used to define training sites. Thus, the correspondence is somewhat circular, and a more accurate assessment of classification performance should be made with independent data. 20 The utility of a classification needs to be balanced against the time and expense of such an undertaking. The remnants of the South Okanagan shrubsteppe that I studied were small and easily accessible, but in larger areas of similar habitat, thorough field surveys would require more time and staff. A s satellite data and the techniques to classify it improve, greater levels of refinement are likely to become possible, thus increasing the economic advantages of perfecting classification techniques. 21 CHAPTER THREE: THE EFFECT OF LOCAL AND LANDSCAPE-LEVEL VARIATION ON SONGBIRD HABITAT ASSOCIATIONS ABSTRACT Habitat associations were determined for five species of songbirds breeding in the sagebrush habitat of the South Okanagan and Similkameen Valleys, British Columbia. The effect of scale on the relative abundance of these species was examined through the measurement of vegetation floristics and vegetation cover at a local level (<100 m), and habitat context at three landscape scales (500 m, 1 km and 2 km) used to determine at which level habitat associations were strongest. Local-level habitat variables were derived from field surveys, while landscape-level variation was quantified from a classified Landsat TM image. For all songbird species, logistic regression models showed the strongest habitat associations at a local level. Floristic variables were often more important than vegetation structure variables. Brewer's Sparrows were associated with large tufted perennials: parsnip-flowered buckwheat (Eriogonum heracleoides) and lupine (Lupinus sericeus). Lark Sparrows were positively associated with sand dropseed grass (Sporobolus cryptandrus), Vesper Sparrows (Pooecetes gramineus) were positively associated with lupines, and Western Meadowlarks were positively associated with needle-and-thread grass (Stipa comatd). Local-level models from Landsat TM classified sagebrush habitat subtypes agreed with habitat associations identified from vegetation survey data. The addition of landscape-level variables usually improved the predictability of survey-derived local habitat association models. Habitat associations varied markedly for each species, and songbird relative abundance responded differently to the scale of measurements. This information can be used to direct management efforts in this highly threatened area. INTRODUCTION Habitat associations in songbirds have been described at a variety of scales from the floristic level (Rotenberry 1985, Plentovich et al. 1995) through to the landscape level (Bolger et al. 1997, Knick and Rotenberry 1995). While some recent studies have compared associations at landscape and local levels (e.g. Bolger et al. 1997, Vander Haegen et al. 2000, Melles 2001), fewer studies have considered the importance of the context, or surrounding habitat type, of areas occupied by songbirds. The context of shrubsteppe patches within a landscape varies because of 22 both anthropogenic and natural disturbances. Humans change the context of shrubsteppe habitat directly through development, or indirectly through processes such as fire suppression which can lead to increased forest cover due to tree encroachment (Turner and Krannitz 2000). The importance of the landscape varies among different bird communities. Landscape-level variables explained more of the variation in bird distribution in urban landscapes than local-level variables, (Bolger 1997, Melles 2000) and were also superior in describing the distribution of birds in cottonwood riparian areas of Idaho (Saab 1999). In contrast, local-level variation was a better predictor of pairing success of Ovenbirds than landscape-level variation in forested landscapes of Pennsylvania (Rodewald and Yahner 2000). Landscape-level variation may be of greater importance for birds living in areas that have been fragmented due to alteration by humans rather than by natural disturbance (Berry and Bock 1998). The distribution of shrubsteppe birds breeding in Washington state (Vander Haegen et al. 2000) and Idaho (Knick and Rotenberry 1995) was best explained by a combination of local and landscape-level variables. Assessment of the relative importance of local versus landscape-level effects is necessary to ensure that species are managed at the correct scale. In the southern Okanagan and Similkameen Valleys of British Columbia, the remaining shrubsteppe habitat consists of fragments of various shapes and sizes, embedded in a matrix of forest, riparian, agricultural, and urban areas. It is possible that these different habitat contexts could affect the presence and abundance of songbirds occupying the shrubsteppe areas that they surround. In this thesis, classified Landsat TM satellite data was used to quantify the landscape-level variation surrounding songbird point count stations at different scales. This ensured that a plot in the centre of a vast shrubsteppe area was not treated on par with one that was surrounded by forest or farmland. By looking beyond the sample plot to quantify its context, local-level habitat associations can be described more accurately. Description of bird-habitat associations at a local level have often been made with measures of vegetation structure and indices of vegetation variables, such as richness and diversity, based •on tradition and convenience (Rotenberry 1985). Mac Arthur and Mac Arthur (1961) first argued that bird species should be more highly correlated with vegetation structure than floristics. However, breeding songbirds in temperate regions are likely to be insectivorous migrants that are adapted to certain insect prey, and often have better success in the type of vegetation with which abundance of these insects is correlated (Cody 1981). Host specialization is widespread among herbivorous insects (Minckley et al. 2000) and therefore floristics could indirectly affect habitat selection of insectivorous songbirds. For example, grasshopper species composition is directly 23 related to vegetation species composition in some grasslands (Quinn and Walgenbach 1990), and in North American deserts, certain bee species specialize on the creosote bush (Larrea tridentata; Minckley et al. 2000). In other cases, plant species may reflect environmental variables such as moisture that are directly linked to insect productivity (Dunning and Brown 1982). The measurement of floristics is of particular value in grazed habitats as grazing impacts both composition and structure of vegetation. While species richness of grassland birds declines in areas of extreme grazing pressure (Bradford et al. 1998), bird species have variable responses to grazing intensities in the middle of this spectrum (Saab et al. 1995). Grazing affects birds indirectly by altering their habitat. For example, litter and cryptogramic crust cover decrease while bare soil increases with grazing (Laycock 1967). Certain plant species are less palatable and/or more robust to livestock grazing and will increase under grazing pressure, while preferred forage or sensitive plant species will decline (Ryder 1980). Insectivorous songbirds can be further affected by grazing, since changes in plant community structure impacts herbivorous insects. Grazing has been positively correlated with increased grasshopper abundance (Smith 1940, Nemey 1958), but has a negative impact on the abundance of small and relatively sedentary insects and arachnids (Dennis et al. 1998). It is difficult to adequately assess grazing effects because of the need for temporal and spatial replication, but the effect of grazing on wildlife can be inferred indirectly from correlations between vegetation and wildlife variables (Hooper and Pitt 1995). Research of this type can produce reliable grazing management prescriptions. Shrubsteppe songbird communities are often dominated by one or two bird species that may exhibit a large amount of variation in their abundance (Wiens and Rotenberry 1981). Habitat associations of shrubsteppe birds have been studied to explain the patchiness of their distributions (Rotenberry and Wiens 1980, Wiens and Rotenberry 1981). At a continental scale, distinct avifauna communities were found within different shrubsteppe and grassland habitat types (Rotenberry and Wiens 1980). In a study of grassland birds within shrubsteppe habitat, the strongest habitat associations were observed at a shrub species level, while no significant relationships were found between bird abundance and herb layer structure (Wiens and Rotenberry 1981). Wiens and Rotenberry (1981) therefore suggested that floristics could provide more information on habitat associations. More detailed habitat associations were described for songbirds breeding in sagebrush habitat of Washington state (Vander Haegen et al. 2000). Forbs and grasses were identified as 'annual' 24 or 'perennial'. Annual grasses were mainly cheatgrass (Bromus tectorum), an exotic weedy species, while perennial species were associated with native rangeland, therefore these classes corresponded to range quality. Vesper Sparrows were negatively associated with annual grasses, while Grasshopper Sparrows were positively associated with perennial grasses. Significant associations were also found between songbirds and additional variables such as latitude, range quality and soil type. For example, Brewer's Sparrow presence was more likely in areas with deep soil. Vander Haegen et al. (2000) suggested that birds might require the vegetation communities that these soils support. My analyses in this chapter focus on five of the most abundant sagebrush breeding songbird species: Brewer's Sparrow, Lark Sparrow, Grasshopper Sparrow, Western Meadowlark and Vesper Sparrow. Habitat associations of these species can be found in the literature, although they have not been examined in detail in the southern Okanagan and Similkameen Valleys. The Brewer's Sparrow is typically the most common bird species in big sagebrush habitats of the United States, although it is declining throughout its range due to habitat loss (Rotenberry et al. 1999). In the southern Okanagan and Similkameen Valleys, this species is at the northern extent of its range and is patchily distributed. Lark Sparrows are commonly associated with edge, recently burned or otherwise disturbed habitats and are often overlooked in community studies (Martin and Parrish 2000). Consequently, the habitat associations of this species are poorly understood throughout its range (Martin and Parrish 2000). It has been suggested that Grasshopper Sparrows require large tracts of contiguous grassland habitat (Vickery 1996). This species forages exclusively on the ground and has been found to select sites with exposed soil (Vickery 1996). Western Meadowlark is a common species of open grasslands and pastures. Meadowlarks avoid forest edges, which suggests that they require large tracts of native grassland habitat (Lanyon 1994). Vesper Sparrows are abundant in the grassland and open ponderosa pine habitats of the Okanagan, where they nest on the ground, typically beneath grass or sagebrush (Cannings et al. 1987). A hierarchical approach was used to model bird-habitat associations for these five songbird species at three scales ranging from vegetation floristics, through vegetation structure (both within 100 m), through to a landscape level (500 m, 1 km, 2 km). The specific objectives of these analyses are: Objective 1: To examine whether bird presence and relative abundance is better predicted by vegetation floristics or by two-dimensional vegetation structure (cover). 25 Objective 2: To assess the utility of a Landsat TM classification of sagebrush subtypes in predicting shrubsteppe songbird presence. Objective 3: To determine if the measurement of landscape-level variables quantified at three spatial scales (500 m, 1 km, 2 km) will increase the predictive power of habitat association models that contain only local-level variables. METHODS Bird Surveys Point count data were collected by three experienced observers between 12 May and 3 July, 1998. To ensure consistency among observers, practice surveys were conducted simultaneously at point count stations by all observers from 2 May to 11 May, and the observations compared. Flags were placed at several stations at 50 and 100 m to assist with estimating distances to singing birds. Stations were surveyed on a 10 day rotation. On each day, three separate routes were surveyed, one per observer. Each observer surveyed each route at least once, to reduce observer bias. Most stations were surveyed four times. Surveys were conducted within three hours of sunrise. All birds seen or heard within five minutes were recorded to species. Observers noted the direction and distance of vocalizations or sightings on circles within radii of 50 m, 100 m, 200 m and unlimited radius. A radius of 100m was used for analyses to give an index of abundance and to eliminate overlap between stations. In a comparison of various point count radii in grassland habitat, 100m yielded almost as many detections as an unlimited radius for most bird species (Savard and Hooper 1995). Since my objective was to compare the relative abundance within a focal bird species across the landscape, I did not determine the absolute densities of each species. An analysis of the relative probability of detections was not performed because it was felt that each of the five focal songbird species could be heard within this radius. Of the five focal species, Grasshopper Sparrow has the softest call, but could still be detected at distances greater than 100 m. Although the probability of detection is bound to increase at shorter distances, a 100 m radii was used for convenience. For analyses, bird counts were averaged across survey dates. 26 Vegetation Surveys Habitat measurements were collected at all point count stations between 18 May and 8 August 1998. Two 50 m tapes were laid out to intersect at each station: one tape in a random direction (chosen by the random number function on a calculator), and the second tape at 90 degrees to the first tape. The line intercept method was used (Brower et al. 1989) to measure percent linear cover of sagebrush and other shrubs. Twenty Daubenmire (1959) plots (0.5 m x 0.2 m) were placed at 5 m intervals along the tapes to measure ground cover. Percent cover of grasses and forbs were recorded at the species level. Percent cover of bare soil, rocky soil, crust, litter, rock, dead wood and cattle droppings were also recorded. Percent cover was averaged for the 20 plots per station. GIS data: topography Terrain Resource Information Management (TRIM) maps for my study area were provided by the British Columbia Ministry of Environment, Lands and Parks (MoELP), Penticton office. A Digital Elevation Model (DEM) at 25 m resolution was constructed from the TRIM elevation data in Arc/Info (version 7.1.2) and the elevation in metres was extracted for each point count station. The DEM was also used to generate slope and aspect maps in ArcView. Percent slope was recorded for each point count station. A buffer with radius of 100 m was created around each of the point count stations and the median aspect was recorded within this area. The median aspect was then classed as north (315° - 44°), east (45 0 - 134 °), south (135 0 - 224 °) or west (225° - 3 1 4 ° ) . GIS data: habitat classes I classified a single Landsat TM scene from July 1996, provided by British Columbia Ministry of Forests, (MoF) Penticton, in order to derive landscape-level variables. Within sagebrush habitat, I identified five subtypes: mid density sagebrush with forb understory, dense sagebrush with dry understory, mid-density sagebrush with dry understory, sparse sagebrush with forb understory, and sparse sagebrush with dry understory. Habitat classes used for the remaining landscape were coniferous forest, deciduous forest and agriculture (orchards, vineyards, pasture and wet meadows). Urban areas, silt and water were excluded from analyses as these habitat types were less abundant within two km of point count stations, and were not evenly distributed throughout the study areas. Where each of these variables were used in 27 further analyses, they were not treated as class variables, but rather each variable represented the proportion of area within each radii belonging to that category. The methods and accuracy assessment of the Landsat classification were presented in Chapter Two of this thesis. Statistical Analyses A l l statistical analyses were performed in SPSS (version 9). Data were plotted and tested for normality by examining frequency histograms and using the Kolmogorov-Smirnov test (alpha level = 0.05). Most of the habitat and bird variables were not normally distributed; this could not be remedied with transformations. A l l habitat associations were modeled using logistic regression because it is robust with non-normal data, and it can utilize a number of both continuous and categorical independent variables (Menard 1995). This method allows variables to be introduced in blocks, so that the relative importance of local versus; landscape variables could be assessed. I used stepwise logistic regression with backward elimination to reduce the risk of excluding variables that have a significant relationship to the dependent variable (Menard 1995). The maximum likelihood ratio was used to decide which variables to retain. Alpha levels used in all models were p<0.05 for entry, and p>0.10 for removal of variables. A higher p value (such as p>0.10) is recommended for removal in order to prevent the elimination of important variables (Menard 1995). Local-level Models Vegetation data were analyzed at species level, but some of the less abundant forb species were grouped to genus or family (i.e. pussytoes, Antennaria). Data were sorted by frequency and abundance. Plant species occurring at fewer than 50 stations were excluded. To examine the importance of floristics relative to overall structure, all forbs and all grasses were combined into two additional variables: total forbs and total grasses. These two variables were created to represent vegetation structure of the herb layer. Significance of either total forbs or total grasses rather than the significance of individual species would imply that vegetation structure was more important than floristics. B y assessing the importance of floristics, I was able to ensure, for example, that native perennials were not treated on par with exotic weeds. Elevation, aspect, and moisture were included in these models, as I anticipated that they would be strongly associated with floristic patterns. For all species, bird presence versus bird absence was used as the dependent variable for logistic regression models. For the three most abundant species, Vesper Sparrow, Western 28 Meadowlark, and Brewer's Sparrow, additional models were created to explore the value of different habitats (Begg and Gray 1984, Bender and Grouven 1998). The presence data were further divided into low versus high abundance. As there was no biologically meaningful reason to choose a particular cut off between low and high abundance, cut-offs were calculated within the presence sites for all species that would give an equal number of stations in the low and high classes. Since Vesper Sparrow was highly abundant, few point count stations were surveyed where they were absent. For this species, presence refers to sites with an average of more than 1.33 birds, and absence sites have fewer than 1.33 birds. To screen data for the selection of variables for stepwise models, individual regressions were performed for each bird species with each of the most common plant species and other independent variables (Hosmer and Lemeshow 1989). This step was used to prioritize variables, since sample size was not sufficient for all plant species to be entered into a single stepwise regression. Variables were retained if the individual model was significant (p<0.05). Data were log transformed where this improved the fit of these regressions (Menard 1995). Where this was the case, the log-transformed variables were included in the models. Landsat TM Classification Assessment Backward elimination stepwise logistic regression was used to predict bird species presence within 100 m. For these models, only the proportion of each of the five sagebrush subtypes were used to see how different species associate with the different subtypes. To assess the performance of models created with only the remote sensing data, backward elimination stepwise logistic regression models were repeated with three additional variables: elevation, aspect, and moisture index. The resulting models were examined to determine whether sagebrush subtypes from the Landsat TM classification still contributed to significant models. Landscape-level Models To examine the impact of sagebrush context, the proportion of each habitat type within three circular radius plots of 500 m, 1 km, and 2 km was determined for each point count station. For each class, the buffer excluded the first 100 m radius, so that the predictive ability of these variables would be a function of the larger landscape scale, and not due to strong correlations within the first 100 m. Each buffer width (i.e. 100 m - 500 m, 100 m - 1 km, 100 m - 2 km) was analyzed separately to reduce correlations among independent variables. 29 Landscape-level variables used in analyses were the proportion of surrounding agriculture, coniferous forest, deciduous forest, and each of the five sagebrush subtypes, for a total of eight independent variables. Stepwise logistic regression with backward elimination was used to examine the significance of landscape variation at each scale. To determine if the addition of the landscape-level models improved local-level models, landscape variables were introduced to the local-level variables as a second block in logistic regression models. Significance of the chi-squared value at this step indicates that the inclusion of landscape-level variables improved the predictability of local-level models. Akaike's Information Criteria with a second order correction (AICc), and AICc weights were calculated (Bumham and Anderson 1998). Within a set of models, a low score of AICc indicates a model with better fit. AICc weight is the easiest AICc metric to interpret as it gives an indication of the relative plausibility of the models (Burnham and Anderson 1998). For example, a model with an AICc weight of 0.8 explains the data twice as well as a model with an AICc weight of 0.4. AICc scores and weights were ranked to compare predictability between the three different scales of landscape models. AICc scores and weights were also used to rank these models with Landsat TM and Landsat TM + topography models at 100 m radius, local models, and landscape + local models. RESULTS Summary of Bird Survey Data Within 100 m of point count stations, 67 species of birds were recorded (Appendix II). An additional 32 species were recorded outside of this radius (Appendix III). The five focal species were summarized by study area (Appendix IV). The number of detections of the five focal species was highly variable both among areas as well as within areas (Appendix IV, Figures 6 -10). 30 BRSP Relative Abundance 0 0 A >0-1 H >1 - 2.75 • >2.75 - 4.5 / V ' Roads Water N A 10 15 Kilometers FIGURE 6. Brewer's Sparrow (BRSP) relative abundance in the southern Okanagan and Similkameen Valleys, based on 245 point count stations surveyed May - June 1998. Abundance ranged from 0 to 4.5 and was the average number of birds detected over repeated surveys at each station. 31 LASP Relative Abundance o 0 A >0 - 0.75 • >0.75-1.25 • >1.25-2.2 A /Roads 1 Water N A 10 15 Kilometers FIGURE 7. Lark Sparrow (LASP) relative abundance in the southern Okanagan and Similkameen Valleys, based on 245 point count stations surveyed May - June 1998. Abundance ranged from 0 to 2.2 and was the average number of birds detected over repeated surveys at each station. 32 GRSP Relative Abundance o 0 A >0 - 0.5 • >0.5-1.5 • >1.5-2.5 A / ' Roads mm Water N A 10 15 Kilometers FIGURE 8. Grasshopper Sparrow (GRSP) relative abundance in the southern Okanagan and Similkameen Valleys, based on 245 point count stations surveyed May - June 1998. Abundance ranged from 0 to 2.5 and was the average number of birds detected over repeated surveys at each station. 33 WEME Relative Abundance 0 0 A >0 - 0.5 • >0.5-1.5 • >1.5-3 A / ' Roads WH Water N A 10 15 Kilometers FIGURE 9. Western Meadowlark (WEME) relative abundance in the southern Okanagan and Similkameen Valleys, based on 245 point count stations surveyed May - June 1998. Abundance ranged from 0 to 6 and was the average number of birds detected over repeated surveys at each station. 34 N A 10 15 Kilometers FIGURE 10. Vesper Sparrow (VESP) relative abundance in the southern Okanagan and Similkameen Valleys, based on 245 point count stations surveyed May - June 1998. Abundance ranged from 0 to 6 and was e average number of birds detected over repeated surveys at each station. 35 Local-level Models: Floristics Versus Structure In initial logistic regression analyses, all dependent variables were significantly associated, either positively or negatively, with six or more independent variables (Table 5). In most cases, significant associations occurred between bird species and specific vegetation types (Table 5). Variables listed as significant were used in the stepwise models. The results of these regressions (Table 5) were used to verify that the direction of association of the independent variables did not change when entered in stepwise models. The direction of an association was only interpreted if it was supported by regression results from individual models (Table 5), since the direction of these variables could change when they are used in a group. All significant variables in subsequent models had the same relationship as in these single regressions. Scientific names of all plant species are listed in Appendix V. Probability of Brewer's Sparrow presence was higher at sites having higher elevations, and at sites with more parsnip-flowered buckwheat, junegrass (Koeleria macrantha), litter and crust (Table 6, 'presence vs. absence'). Brewer's Sparrow seldom occurred at sites with a north aspect, cactus (Opuntia fragilis) and sand dropseed grass. Although 'total forb' was positively associated with Brewer's Sparrow at the individual model-level (Table 5), this variable was not significant in either the presence vs. absence or low vs. high stepwise models, while particular forb species were (Table 6). Within occupied sites, high relative abundance of Brewer's Sparrow was positively associated with parsnip-flowered buckwheat, lupine, sagebrush and litter (Table 6, low versus high), as the direction of these associations was supported by Table 5. Elevation was not significant, meaning that for occupied sites at a given elevation, Brewer's Sparrow relative abundance was predicted by parsnip-flowered buckwheat, lupine, sagebrush and litter. Lark Sparrow presence was positively associated with bare soil and sand dropseed grass, and negatively associated with total forbs and elevation (Table 6). As for all the other species, the direction of these associations in stepwise models consistent with the direction of association in individual regressions in Table 5. The difference in distribution of Lark Sparrow and Brewer's Sparrow is reflected by the marked contrast in both individual (Table 5) and logistic regressions (Table 6) for these species, at the local level. While Lark Sparrow was positively associated with sand dropseed grass and negatively associated with forbs, Brewer's Sparrow had opposite associations with these variables. Grasshopper Sparrow was positively associated with cheatgrass and pasture sage (Artemisia frigida), and negatively associated with spreading needlegrass (Stipa richardsonii), sagebrush 36 and north aspects (Table 6, direction of associations consistent with Table 5). 'Total grass' was not important to this species in either individual (Table 5) or stepwise regressions (Table 6). Needle-and-thread grass was positively associated with Grasshopper Sparrow in individual regressions (Table 5), but dropped out of the stepwise regression (Table 6), possibly because it was confounded with low sagebrush density in the study areas. Many of the point count stations with sparse sage cover were dominated by needle-and-thread grass. Western Meadowlark presence was positively associated with pasture sage, total grasses and knapweed (Centaurea diffusa), and negatively associated with moisture, elevation and balsamroot (Balsamorhiza sagittata) (Table 6, direction of association of variables consistent with Table 5). Within occupied sites, higher abundance was more likely at drier sites with needle-and-thread grass and three-tip sage (Artemisia tripartita) (Table 6). Like Grasshopper Sparrow, Western Meadowlark was more abundant in sites with low sagebrush density (Table 6). In this case, while the measurement of total grasses could better predict the presence of meadowlarks, high abundance was associated with the presence of needle-and-thread grass in particular. Vesper Sparrow presence was positively associated with lupine and pasture sage, and negatively associated with pussytoes, dead wood and the combination of rock and Selaginella spp (Table 6 direction of association of variables consistent with Table 5). Total forbs was not a significant predictor in either individual (Table 5) or stepwise models (Table 6). Within occupied sites, higher abundance of Vesper Sparrow was positively associated with yarrow (Achillea millefolium) and pasture sage, and negatively associated with sand dropseed, sagebrush and moisture (Table 6). Utility of Landsat Classification At the 100 m radius (Table 7), associations between birds and the sagebrush subtypes mirrored the results of the local-level models, suggesting that the five sagebrush subtypes classified from Landsat TM data were relevant to songbird habitat selection. Brewer's Sparrow was more often found within the mid density sagebrush with forb understory and the sparse sagebrush with forb understory subtypes, which in turn were associated with high lupine abundance (Chapter Two, Figure 4). Lark Sparrow was positively associated with the 'mid density sagebrush with dry understory' class, which was in turn positively associated with sand dropseed (chapter two, Figure 3), a significant variable in Lark Sparrow local models (Table 6). 37 00 <D OD CH £ > 2 00 oo 0- > .2° 00 W oo 5 8 PH > V\ oo O o . c s oo CM > C/3 oo ^ a> oo ^ fc- X i oo > PH on ^ _ , M oo PH > t/2 oo CQ a R * * * 00 i - ^ * o o o o o o c d o o o o o o o o , — s * * S3 S3 00 S3 * * * * 00 # 00 00 , ., a c c + 00 00 2— 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 S 3 S 3 ^ S 3 S 3 S 3 S 3 S 3 S 3 S 3 C S 3 00 00 00 00 c c a c # * Cd 00 00 00 a c c a * * * 00 00 # 00 00 oo cd S3 a S3 S3 S3 (3 00 00 00 00 a a a c # 00 /—v S3 + •X1 j f . j f . j f . -Jf * r„ * * * * * * oo * * o o o o o o * oo oo cd oo oo oo # * * * 00 * * jf. S3 T"-r^>^ 00 c * * 00 * 00 + * s * -00 00 00 00 00 c c c c c 00 * 00 00 00 00 00 S3 s3 c i3 a * 00 00 * 00 00 00 00 13 (3 c e * * 00 00 00 !3 S3 S3 00 00 00 00 00 00 00 S3 S3 S3 S3 S3 S3 S3 00 / v 13 + 00 00 00 00 00 S3 S3 £3 S3 S3 00 00 00 00 00 00 00 S3 S3 S3 S3 S3 S3 C * * * * * + + 00 00 00 00 S3 S3 S3 S3 « — ° g 5 w> O 41 3 3 R fc- O R o o 3 £ § o <W 00 OH ft 00 Q Q ~ o CX) a CO U •S > o l a & >-60 O a, o -4—» a cd CD CJ 0 o CD CD > o ts kH 00 a _o 'cn co C D fcH M ID 1-O '•*-> cn '5b . o =3 O -O O 'I | a -d . a 13 =3 > to CD CD — i & 13 CD W OH oo CQ 13 £.8 WO oo Xi es oo w ^ > a ox 00 > © oo eS 00 > 00 C J LH a PH > C/3 w C U 00 00 PH • J a es 00 P H ps5 |e .Sf" CQ 5 £ 00 P H > cn t« 00 00 00 CO a a c e + ^ ca n + * on * a T CO CO CO C C C 00 CO C C * oo * a " + CO CO C C * * 00 * CO * 00 * a ~" * + + * C C + # * * CO ,—^ * •a CD CD a o 00 CD o > co co C C CO CO * c c CO CO CO CO C C C C CO CO CO C C C C C C C C * 4^  CO CO CO C C C C C CO CO CO CO CO C C C C C * CO CO * C C T * # CD tS ?? ~V CD 03 JH <D . a 43 cd ? a £ -° >> £ £ ^ £ -a P o CO CO CO CO C C C C CO CO CO CO C C C C # # CO CO CO CO C C C C C + CO CO CO CO C C C C * # -a o r <u  -a ^ ^ ^ ^ & S a f a o C o 2 u « S ^ u ' 5 S -CO ^ 00 2 00 CO CO C C CO CO C C * * * * * * CO CO C C C C T3 cd CD M CO a + CO CO C C CO CO CO C C C CO CO CO C C C CO CO C C CO CO CO C C C * CD a CJ CO -a c 2 « 8 2 - 3 $ C D ciii oo ^ OH o fcn CD C3 — H CD -a C O 00 O 3 2 cd > X J OH cd UH 60 O OH O -4—» -a c cd UH 3 o £3 O CJ CJ > o UH 00 C o C J UH 60 C J UH o *-UH 00 '5b . o 3 o - a o o ^ 00 , i X! oj 3 > 00 C J <u - H & 13 o O CJ cd 3 B ^ '•3 °° S3 CJ v j aj W OH m - a CQ 13 <£ t> H £ P H > OX) 00 © P H C Z ) W > oo « 00 >• oo fc* a X i ox) oo > O PH C Z ) oi o P H C Z ! |2 P H C Z ) P3 oo X i et oo > oo CJ u a oo > oo CJ oo I- X i a . cs oo > oo CD oo fc< X i O H CS oo 5 2 P H C Z ) PQ # * # * * CO * * +^  # * * * * oo S3 * # * * * # oo * 1=1 T + 00 S3 oo * > 00 * •X-oo S3 CJ 00 fc. X i a . CS XI oo 2 X ) oo S3 * * * 00 * 3 + , 00 S3 00 00 00 S3 S3 S3 oo % S3 + * _ * 00 jt S3 * - s + 00 S3 00 00 00 S3 S3 S3 + 00 00 S3 S3 00 00 00 S3 S3 S3 00 00 S3 S3 # 00 # S3 -T * * * * 00 S3 * * 00 00 S3 S3 * * + # * C J ft 60 cd 00 • ^ 6 0 « r 5 $ K OX) oo JH H ? o 3 S u « CQ H S C ^ P H C ^ U £ < W OO o „ m II m — 1 ^ S W m > 2 CJ H—» OO II m w o fc ca OH 00 UH C J OH OH o XI oo oo cd UH o ca _ C J .2 3 i d cd S3 -C • - H cd ^ w cd in o o -c o ° OH + J oo g O J U td O o X j oo 00 00 S3 cd u _> *H-H cd 60 CJ S3 cd T 3 CJ H—» £3 CJ oo C J UH OH C J O C J OH 00 cd C J 00 S3 o ft oo C J S3 r v X ) < OH 00 PH* O & o fc cd ft 00 UH cd h-l P H 00 o fc cd ft 00 S3 oo • H cj cd > "3 C J o O co S3 2 00 <+H 4J O ft S3 >- .9 .O J3 a . S 'oo oo oo • - " C J ^-v i-1 - J 6 0 > «-> UH — 0 <=> ? ? O OH oo "5 (_, C J 4J O oo S3 « r 00 "—' H J UH I—I P H O O ft « 00 ^ aj * S3 cd o 'S 60 X ! o cd o T i -ro 00 o C J •n C J & 00 cd C J C J o T t I o in td 2 13 w 1 - 1 i-H 00 X H H-> • 2 w O ft 00 OO r. Cd o ^ X I ^ t; m O ^ S3 I X ! ft * H X ! OO 00 C J o o T t T t cd O T t CQ <D cd CQ cd CO H 2 'S 00 CU • •4 cu CU O l C Z ) o o \ cTi oo oo (N O O © •—; ' - J o ©° © © © © i/~> C O C N C O O — ' C N C N o © o © cu 00 a 00 CU oo T3 cd cu 0 ^ « S O M ^ A ^ g I ^  j J J 3 S J T3 cd CU | d cd CU' CD 4 3 00 2 4 2 CD 00 cd 00 & 52 ' 2 4 2 CD OO cd 00 CD CD (-1 4 3 t—1 CU CU • • CJ 00 s d oo \ S eS " oo >• cu cu e cu 00 CU 1 -Q. T3 CU J h J h J * I O +3 "es eu o os r- r- oo r - H O O © —J O O © © ' © OH OH 00 CU 00 S -o cd oo . 5 O w <D (50 O 3 -S ^ $ "2 cd 3 =3 i 2 w & H h-l O H O Q J J J P i J O 5"? "S d C cu <D 00 co cu •£> cd OH I * s * -a £ 5 £ o S cu cu S3 £ s o -2 Ui es u . es £ a > CZ) eu g, s > a C3 O \S ^ »-i - " fc £3 § O « 91 o 55 CD cN Cd OH — ; p O O p p O O O O O O O O O O c d CD 4 3 C J =5 43 d .& .2 g h i S h i U h i J - < h i T3 CD CD CO CO OH 00 O cd ±i i-i -O co o d cj y 3 cd CD ^ •=, 00 O H U CO co o o\ —1 <-> o o 0 0 0 0 cd CD 4 3 CJ 33 42 OH c^f C— 0\ O —; p p —; O © © ' © CD • S S3 OH £ =3 cd hJ P H 4 3 co 2 43 CD 00 cd 00 Hi hJ J hJ o 00 CD cd PQ -a CD CD 00 OH o -a . c cd 00 co -8 2 13 o H h J W h J o M fc! .2 « 3 2 cu "* Q. «J tZ)| ^ *Ui CU & CU >-CM cu • • CJ CO G d cu O 00 42 O es ^ > eu eu C eu 00 cu u OH H ^ o CD O o OH 0 S -i^r co ^ 0 0 ^ V. . . •S d d eu CD co co cu 42 C cd OH C N 0 0 0 0 O ^ p — ; C N O O O O O 4 3 CD I 8 « 3 "S o > CD 00 i g M cu cd ^3 S co 5 00 d . " cd co • c d > H P H -3 CQ 00 J J l J J o es - d p., oc cz) 2 la • CU 00 > 5 o r~ 1 d - 1 ^ . M l ° 4 3 O > 0 C N C O O N C N ^ - — ; ^ C N O O O O O CD £ -d CD CD CO 00 CO ^ 2 ,cd 00 00 XI co cd S3 <D " 0 0 ^ ^ « -2 8 H - S M w « o S 001 td 43 CD OH • « CH U OHOO CQ J J < J J O CO CO CO row 13 C S3 row O - ^H H—» O C2> pa O ian O -3 0 La. U 0 > -d « T J T II d CZ) O •»—1 T3 a. CZ) 00 high ct pre II d 5, n=" parn ct pre 94, n C N loppe ct pre 0 0 C N CU vs. rre C N lO CZ) CD - H C d ent iS 00 00 rre d eu cu 0 4 3 u O cu CO es 0 cu CO CJ 2? es 0 4 3 J ^ CO C U u 0 GO cu PQ 0 S 0 4 2 cd kH OH 0 4 2 cd i-H OH TABLE 7. Results of stepwise logistic regression models with Landsat TM derived sagebrush subtype variables', and sagebrush subtype models with addition of topographic variables2. All models are based on bird surveys and digital data within 100 m of point count stations. Only significant variables (p<0.10) are listed. All models are significant (p<0.05) unless otherwise indicated, "r" is the partial correlation of variables within models. Species: 100 m Landsat T M only Variable r Species: 100 m Landsat T M + topography Variable r Brewer's Sparrow MDF 0.21 Brewer's Sparrow ELEV 0.24 presence vs. absence SF 0.08 presence vs. absence MDF 0.08 %correct predictions: DD -0.07 %correct predictions: ASPECT -0.14 absent: 70 absent: 77 present: 63 present: 61 Brewer's Sparrow no model Brewer's Sparrow MOIST 0.11 low vs. high low vs. high model p>0.05 %correct predictions: low: 77 high: 38 Lark Sparrow MD 0.08 Lark Sparrow SF -0.14 %correct predictions: MDF -0.07 %correct predictions: ELEV -0.21 absent: 94 absent: 91 present: 17 present: 35 Grasshopper Sparrow SF 0.31 Grasshopper Sparrow SF 0.29 %correct predictions: SD 0.27 %correct predictions: SD 0.26 absent: 100 DD 0.11 absent: 100 DD 0.10 present: 17 present: 28 ASPECT -0.15 Western Meadowlark SF 0.18 Western Meadowlark SF 0.22 presence vs. absence SD 0.18 presence vs. absence SD 0.13 %correct predictions: MD 0.12 %correct predictions: MDF 0.11 absent: 0 absent: 45 MOIST -0.12 present: 100 present: 95 ELEV -0.21 Western Meadowlark SD 0.28 Western Meadowlark SF 0.21 low vs. high SF 0.25 low vs. high ELEV -0.11 %correct predictions: SF 0.23 %correct predictions: MOIST -0.29 low: 70 SD 0.23 low: 69 high: 69 MD 0.22 high: 77 42 TABLE 7. (Continued). Results of stepwise logistic regression models with Landsat TM derived sagebrush subtype variables1, and sagebrush subtype models with addition of topographic variables2. All models are based on bird surveys and digital data within 100 m of point count stations. Only significant variables (p<0.10) are listed. All models are significant (p<0.05) unless otherwise indicated, "r" is the partial correlation of variables within models. Species: 100 m Landsat T M only Variable r Species: 100 m Landsat T M + topography Variable r Vesper Sparrow SD 0.15 Vesper Sparrow SF 0.15 presence vs. absence SF 0.13 presence vs. absence SD 0.13 %correct predictions: MDF 0.10 %correct predictions: MDF 0.10 absent: 20 MD 0.09 absent: 20 MD 0.09 present: 93 DD 0.05 present: 93 DD 0.05 Vesper Sparrow SD 0.23 Vesper Sparrow ASPECT 0.16 low vs. high SF 0.23 low vs. high SF 0.15 %correct predictions: MDF 0.15 %correct predictions: ELEV 0.14 low: 60 MD 0.12 low: 66 SD 0.13 high: 75 high: 75 MDF 0.10 MD 0.09 DD 0.05 MOIST -0.15 1 MDF = mid/dense sagebrush with forb understory, DD = dense sagebrush with dry understory , MD = mid density sagebrush with dry understory, SF = sparse sagebrush with forb understory, SD = sparse sagebrush with dry understory,2 ELEV = elevation, MOIST = moisture. 43 Grasshopper Sparrow, Western Meadowlark and Vesper Sparrow were negatively associated with sagebrush (Table 6). These species were more associated with the sparse classes, which were associated with less sagebrush (Chapter Two, Figure 2). The addition of some combination of other digitally derived variables - elevation, aspect or moisture - improved Landsat TM models in every case except for Vesper Sparrow presence versus absence (Table 7). For all improved models except Brewer's Sparrow low versus high, sagebrush subtypes remained as important predictors of songbird abundance and distribution at the 100 m scale (Table 7), thus highlighting the utility of a fine-scale classification of Landsat TM data. The combination of this classification with other digitally derived data such as elevation, moisture and aspect, can improve the ability of users to predict bird habitat suitability in the absence of field surveys. The importance of Landscape-level variables at different scales Significant landscape-level models could be created for all bird species except for Brewer's Sparrow low vs. high at the 500 m and 1 km scale (Table 8). At the 500 m scale, Brewer's Sparrow avoided sites with surrounding agriculture, and were more likely to occur in sites with sagebrush subtypes that indicated forb abundance (Table 8). Brewer's Sparrow selected sites that had more coniferous forest at both the 1 km and 2 km scales (Table 8). They were also positively correlated with most sagebrush subtypes at these scales, although their strongest relationship was with the forb subtypes (Table 8). Lark Sparrow was negatiyely associated with surrounding coniferous forest at the 500 m and 1 km scale, and also negatively associated with all sagebrush subtypes except for mid-density sagebrush with dry understory (Table 8). At the 2 km scale, this species was positively associated with agriculture, and negatively associated with forb subtypes (Table 8). Again, habitat association models for this species contrast those for Brewer's Sparrow. Grasshopper Sparrow consistently avoided sites with surrounding coniferous and deciduous forests at all scales (Table 8), although the negative association with conifers at the 500 m scale was quite weak. This species did not exhibit positive associations with any of the sagebrush subtypes, but avoided sites that were surrounded by the mid-density 44 sagebrush with dry understory and mid/dense sagebrush with forb understory subtypes (Table 8). Western Meadowlark avoided deciduous forest, and selected habitats with sparse subtypes, near agriculture, at the 500 m and 1 km scale (Table 8). At the 2 km scale, this species was negatively associated with all habitat classes except for the sparse sagebrush with dry understory subtype. At all scales, Western Meadowlark had lower relative abundance in areas with coniferous forest in the surrounding landscape (Table 8, Western Meadowlark, low vs. high). Vesper Sparrow was more likely to occur at sites dominated by the sparse subtypes at the 500 m and 1 km scale (Table 8). At the 2 km scale, this species occurred more in sites with sparse sagebrush with forb understory while avoiding mid-density sagebrush with forb understory (Table 8). Vesper Sparrow abundance (Table 5, Vesper Sparrow low vs. high) was positively associated with all sagebrush subtypes at the 500 m and 1 km scale, with the sparse subtypes having the highest values of r (partial correlation). The importance of quantifying landscape-level variation can be determined by the significance of the addition of these variables to the local models derived from vegetation survey data (Table 9). The Brewer's Sparrow local-level models were only improved for presence versus absence, at the 1 km and 2 km scale and not at 500 m, while the Lark Sparrow model was only significantly improved at the 1 km scale (Table 9). The Grasshopper Sparrow model improved significantly at the 500 m scale. Western Meadowlark and Vesper Sparrow local models were consistently improved by the addition of landscape-level variables (Table 9). The value of the Landsat TM classification relative to field data is best examined by comparing the AICc scores and weights for the different models. The AICc scores and weights for models constructed from Landsat TM and Landsat TM + topography variables within 100 m were always considerably lower than for local-level models (Table 10). For all species, local-level variables provided better models than landscape variables alone (Table 10). However, some combination of both of these sets of variables provided the strongest model, except in the case of Brewer's Sparrow low versus high abundance. The most useful scale measuring landscape-level variation was species dependent, as illustrated by AICc rankings. Presence versus absence was best predicted 45 at the 500 m scale for Grasshopper Sparrow, at the 1 km scale for Lark and Brewer's Sparrow, and at the 2 km scale for Vesper Sparrow and Western Meadowlark. B y referring to the landscape variables that are pertinent at these scales (Table 8), one can make inferences about habitat selection. For example, Grasshopper Sparrow preferred habitats that did not have trees, 'mid density sagebrush with forb understory', or 'mid-density sagebrush with dry understory' within a 500 m radius. Lark Sparrow preferred habitats with no conifer cover within 1 km, while Brewer's Sparrow was found at sites with conifer cover within 1 km. Western Meadowlark avoided tree cover and agriculture at the 2km scale, although they did not avoid agricultural context at the 500 m and 1 km scales. This could mean that while meadowlarks may select large tracts of shrubsteppe that are relatively less developed, they might not necessarily avoid agricultural edges within these areas. Vesper Sparrow preferred sparse sagebrush habitats at the 2 km scale and were not affected by tree cover or agriculture. 46 -O cd . CO 0 f o cu co 5 td cd ^3 .a ("I oo 1 £ >^  co o > oo _CO CO o <~~* 00 IT) -O O S3 r-i •2 V s a O 4 - . f<-i S3 ^ cd oo o oo S3 CO 00 CO co T3 fa O cd £ - u .2 o oo oo S3 CO & ' < ui •a « 52 to 5b x: O co * — l Ul aj cd oo , — v * 2 co o J2 | ftl O oo c3 •^J co S3 2 3 S •S 2 .2 .5 cd 5 ~ > £ 00 CO C O _ 00 "5 V. « "o 1 « .2 'S cd o . M > cd 'j^ H—i CO S3 o W O '+2 P J O P cu 'I ^CU "3 cu l oo 1 cu 'I oo *s cu ft| 1 ^ cu 'I u l > 00 oo O t-- ro in T f r--N N H H O © © © o © fa I—I , fa o 9 9 Q CU 2 -es cs ft • oo g 00 S3 O oo -a CO cu co c s cu cu n r -oo S3 g CO oo O O X <D = x cd U ft >o m h h « CN CN —i >— —i O © © © © © © fa n * ~ fa Q 9 O Q cu cu £ a © *> r oo es es O-00 S3 .2 T3 CO ts ^ v! £ S g O CO oo o 55 co 00 "fc. OU CU U fc- 2 U ft O 2^ CN 1 . d © 9 S 00 < a cu cu 00 c S3 cu . O 00 *H—» X i CO es T3 00 CO Ui > ft ^ H CN cu H ^ oo cu o resen 00 m ocorre absent resent ft m absent ft S3 cd CO " C O « S 'oo O fc-fc-cs a oo fc-cu cu fc-pa in JS © OX o 2 A , . ftl 00 — > — CO •a. S3 "O 2 CN S 0 a "CO W 1 -2P SS OO © fc. fc-« ft 00 00 "fc-CU CU fc. pp Xi OX) oo o © © "co O a © a a a "co r3 "« a § .2? C 'oo © fc-fc-cs ft 00 oo *fc-cu ^ cu fc-pp in X J q 00 o 3 A , 00 (-1 ,—i > a o |§ ° is in © © © © © © O 2 fa o 9 fa < 2 00 S3 o o fc-fc-cs ft oo S3 fc- H ed M h J CN g O CO oo o s; <u 5^ cd ft CTl ^ h r H O r H rH r H (N © ' © ' © ' © © fa M P Q Q O P H Q oo S U oo oo S3 O © fc-fc. 03 ft 00 ^ S3 fc- H -a CO ts ^ n C S3 co O O o i co 55 co X> fa cd ft r~- co © . - i © © © © © fa H H Q oo U 2 oo 00 S3 O O fc. fc. es ft 00 co 1 ^ ON C N C O ^ S C S3 co U _ O co oo S © O C/3 CO « © ^ X) fa © T t C3N —i CN CN © © © fa Q 5 n u O 9 W U S Q 00 S3 o fc. cu ft © ^ Xi o 00 IM 00 s_ « S3 fc- a CO S ON ? ! C S3 g O CO oo o co o X> Ui O 00 C N ^ 0 \ H I T ) H © — —< C N © ' © ' © ' © Q fa • hH fa o z Q w o 9 S Q o S 00 S3 O fc. CU ft © ^ £i O oo U v> fc- £2 « es R fc. Q H - ^ 00 o -a co t3 0 0 . . g O CO oo o 55 co \ 0 X) " o-- cd C N I > C O C O © © — - C N © ' © ' © " © fa Q U Q S S3 o fc. cu a © ^ X i o oo u S3 oo u S3 ^ ft © -a co I 15 g O CO oo o 55 co o X> u O 00 i n cd EU T t C J 4 3 „ GO sr 2 u 4 3 t i -4* cd o cd ,3 S3 00 CJ 3 cd •c cd > 00 CJ 4 3 CJ O In 0 0 j u 00 I CJ 2 3 _2 ^ I ? S 3 s ~ * s » s 00 c 2 .5? 00 . 2 « O cd B ™ S3 ,2 . 2 ^ cn fl 00 C cu _ —I —3 0 0 < < CJ 13 \E3 cj 0 0 4 3 49 cj l-H cj cd co v 2 CJ O to V a 3 2 to +2 CJ cd -s w s '5 o 3 3 51 4 3 cd .2 .-S 1 l i s >. cd CO -a CQ t o § c u o •5 3 c S3 cj O O co • U fl 3 w 5 3 • fc O O ( N Q pq T3 cj « S - l __ § * h ^ a 00 eu '3 cu =4 c/) CO =4 CZ) CZ) c n 0 0 (N r t - H 4-1 CN CN CM CN CN c n 0 0 o <N 0 0 0 0 0 0 0 0 0 0 U 2 9 ^ o 2 0 0 < 4 H . M cu cu e cu CO es CO >• cu cu .fl eu _, eu _b C- CN S3 O T 3 CJ P H 1—1 & fc O O P S U P OH 4 H !C z : as fl S3 cj CJ co _ CO CJ O 43 —< N cd OH o CJ CO 0 O v o g © © © 0 Q 0 0 es © cd eu eu 2 < 0 0 U W Q eu cu e eu CO 43 es CO > eu cu C eu co CU I- ' a • CO fl O •+-> o -a CJ >-H _ . _ OH ^ " 0 cj fc 1=5 d G cj CJ So 0 -„ -CJ CO CD o 42 >H cd OH a \ 10 as cn —1 ,-1 0 —; O O O O Q 0 0 0 0 P H 3 u O W < Q _< u es o T3 es eu US e eu CO eu eu eu fl eu CO 42 es CO > cu eu c fl eu fl .T eu O u o Q. un co S3 *H—» 0 3 CJ CJ S3 fc o o _ N ? 42 e>* cd C . cj CJ co co cu P & O fc ^ w o 9 o o S es o •a es CU c o e 43 _< CU 40 -to > E eu > O vo 0 0 O O O o o o •a CJ , 10 42 3 o Q es o es eu 2 Q m Q Q 6* fl J= u . eu co « t B fc o u CO S3 o T 3 CJ l-H OH +-» ^ ° 3 r - H O —I CN o o o Q & ^ u 5 Q W O Q Q U u es o TS es eu CO S3 O 43 OX) fl 43 !_ eu oo ** > CO eu k H S2H & ° O 0 — un T 3 cu C-> v^. O V© CN CN O O o *o 0 0 (N 4 H o o O O O O P H Q P H Q Q h - H U Q W 00 P eu cu fl ° -s es es S? a tZ) eu >- fl S3 O •a cu > O . C N O O P H Q 0 0 0 0 eu eu fl ^ s 0 "2 s- es -1 es «> a ^ CZ) eu cu fl a , eu co cu « h • > a . - H 0 0 s <^  g 3 o cu p 4 2 cd o es -3 0 0 cz) 3 -£H" fcn • S3 eu ^ k> _3 CN so m CN CN CN O o o P H 0 0 S3 o - O cu 8 CN Q P H 0 0 OO P H P 2-CO S3 _o *H-H O 3 eu SH OH c n H—< O CN rre S3 0 cu CJ CO ab O CN - H o o P H P 0 0 0 0 eu eu fl 0 -2 u es u . es j« a > CZ! eu eu t £ (> O. un P H P 0 0 0 0 co c 0 *H—» CJ 3 cu B fc 1 r-CN cU • • 2 3 O cj CJ CO p - ° N cd SD o o s-•* f es — as a- 6* v „ • S3 eu g 2 S ^ . O H > _2 • t— * 0 M CN CN CN o o o C N o o P 9 P s co fl _o o 3 CJ 4 . C H -4-» 8 CN fc ^ 2 ^ ^ _2 r~~ un o d -a, 4 3 P H P O -< ^ 4< es — ^ 0 , 0 X 1 0 0 cz) 2 LH • S3 eu =« fl fe- o fc _4 l § OH » > — un p p p CO 0 0 CJ CJ 4 . O , -4* CJ CJ SO fc un CO un -4 II P P M 4 ? o ^ CJ S3 • ° 4 3 3 ^ * -S •d 2 fc 2. 42 0 0 cj cd 0 0 M cd cu co 4< cd OH CO P H 0 0 C? o -4* CO = 3 4? co cj VH c 2 co „ 3 —1 ."S § * 3 -d . 2 2 11 " 1 1 0 0 P S H H u ^ m .t3 P S3 ~ CJ CO ^ CJ T3 c 2 " § § P )H Q s o CJ 3 3 u 2 o 4 3 o 2 3 ^ $ * CO M § g S3 cd co 3? rt - 4 C S 4> U OO a o o IT ) a , o d =3 ^ ^ OS CM O d CM < ^ O •3 C M O fc Cd ft 00 CJ CD l-i CQ OS CN d os CN d in d ca o (3 co oo X ) cd 00 > co o S3 CO 00 CO ft x; XI 00 o fc ca ft oo CO & CO I-CQ in cn o d T t T t CN —I in o CO CO S3 CO on X ) ca > co o c co c/3 CO u. ft o fc ca ft oo M v-cd O ro in ro O CO o S3 co oo Id oo > CJ o S3 (U 00 CO O fc cd ft 00 l-l CO ft ft o X i oo oo cd * o o oo in ro O O ro CS o o os in d T t —i cs co CJ S3 co 00 X> cd 00 > CO o S3 co 00 CO cd O T3 cd CJ CJ -*-» 00 CO * O o OS ro * o o d cs T t * o o CN in in xl 60 oo > O cd o -a cd CO a E CJ -4-» O0 CO o o CN CN CS cs * cs o d ro 0O * cs o ro oo * o o o £ £ CO o S3 co 00 X ) cd 00 > CO CJ S3 CO 00 CO U ft o fc cd ft 00 u, CO ft oo CJ > O T t * o o d CN © T t * o o d cs in os ro X ! 00 oo > o fc cd ft 00 s-l CJ ft oo CJ > OS T t m o d V ft * CD —I o 0 0 c S3 CD cj CO cd .S3 CD cd CD cd O o CO o U CD o CD ' o CD OH CO cd xi CD O CD cd S-H .2 CJ - 4 * CJ 4"J 6 0 S3 .5°. 43 '53 P 3 cd cj s3 T3 S3 cd CO CJ .2P 'S O u 9 S M 43 o 6 0 S CO Is 2 cd o rV cj ^ co 2 u W < I—I -H CQ « H -2 it 1-_ 03 IH it o u -. - J - H u M it o 1-u w. O l o I. In S3 C J £ 9 ,2? w cj ft °^ r-" o —< CN CN so 00 00 o O O o o o o o W W W W W W W W 00 r~ T - H o oo 00 CN p iq vq 00 r~; XT — < cn — < •n CN oo oo p oo CN rn rn oo 4 H CN CN CN CN •3- cn i n ^ ' H C S t N t N ^ - ' + ' ^ i O i n o o cn C 13 + _3 ° S -2 6 T3 5 X I + -2 + ^ n o cj cj -^i O o + H E o o •H - H ^ H C N C N C N c n c n m ^ o p p p p p p p p C J O O ^ O ' ^ - C N O - H — i — i J 00 O0 - ' O O O icf CN —i m' —'—'(-' j^-' —.'—.'—'^d v© oo os o —> V") U") u-i \c (N (N CN CN CN s E E M .y o o CN m T3 T3 T3 lan lan lan nd + + + ,2 13 13 c3 13 E cj cj O o _o _o _o —c oo oo -rr vo vo vo r~ CN CN CN CN O o si H -2 E E o o o o —' cn O O O O O m Os Os CN CN O O o o pq m w w w w w CA 0\ O M N p i n r-; CN CN CN CN CN oo i n ^ T(-os o\ o\ o\ so m >n >n oo oo in n in MX) cc oo 00 00 OS OS OS OS OS S 5 CN •o -o -o ~\ ^ c H c £ c e Q o o o o o a o E + CN O c3 E ^ 2 § S 9 so pq '53 ^ 9 —1 M (S m in « H O O O O O O O — 1 W W W W W t i H p q w m r f m c N O O c N O ^ O r ^ p p p i n p c N c n cn—<—<rn—<—-CN^O • ^ i n i n ^ i n ^ i n ^ ; m ^ o ^ o r ^ r ^ o o o o o CNCNCNCNCNCNCNrn —. CN T3 O O cn •O c C3 1 1 O o 13 + + 13 13 o o o 2 P P CN - H CJ CJ C3 O CO CO c -c3 o m g "2 o > o o =H1 c/l u a> a (/) o o u C/3 — 1 — i f N C N i n i n i n v o r ^ ^ o o o o o o o o o C J C N — 1 — i v o o v m o o o J r O « C N ' * C 7 \ 0 \ ^ i ^ - H K - H ' « T t O \ C \ ( N O \ - H o—<—1 o o\ d \ i n « i n \ 3 o o a \ 0 \ o o o o o — • ^ - H - ^ - ^ - ^ t N I C N I t N l f N J C S ) E E " § CN —' m o D H o S - 2 - 2 - 2 - 2 - 2 i S i S 2 2 <u 00 ^ OS and and E g g 1—1 0 M H H + + 0 •n -1 CN E E 13 13 "ca loc loc loc lan lan lan 0 O m cn 00 0 O 0 O 0 0 0 0 — ' — | W W w W w w w w < • — H CN CN CN CN p O cn CN CN CN —• — ' — ' • * ' — < 00 cn rn m cn cn 0O CN CN O 0 vd 00 OS 0 0 00 00 0 0 CN CN CN CN CN CN m m J E o M O <N —1 cn 13 C C C C3 C3 w i l l "5 "ca P CJ CJ CJ g o o o op pq 53 os ^ 00 m m o o I I W W o\ c^ CN O CN o o CN CN CN CN E E E M 0 0 CN —^H m cc! •a T3 T3 O c C C _o -2 .2 JS m •n 0 0 0 p W w W W m — H o\ p — ^ rn — < CN* o\ <o •^ f rn OS r~" CN CN CN m CN CN CN CN o C C o H E o o 00 00 o o (N CN E 0 0 km km m —' CN T3 T3 e C C .2 _co -2 + + + 13 "c3 13 CJ O CJ 0 O 0 o a o + in g - g CN — i c N c N i n v o r ^ r ^ o o m _ 2 o o o p p p p p - H c j r ~ o o v o o \ — 1 h m 00 00 ^ t-~; —< p CN rn 00 o d r ^ i n ' c N c N ^ c N r n c N o vq *o CN CJ> Tt i n CT\ Tfr CN y r n ' c > d c ^ ^ C T \ C N r n r - ' - ^ > i O ^ « C O 0 0 C M ^ O \ N T3 C -2 13 + 2 -2 o o cn •a c C3 CN i n a e cu ca o C C o CJ o £3 cj CO CD — H OH 2 s «s § cj --3 3 CO co cd CD CO OH > 35 2 o ^ s 43 O ' > 2 "CD + s fl § I-§ > co « CD CD 43 6 0 43 OH cd o OH O H—» + H H-» cd CO -0 S3 cd —1 CO 33 CO CJ > CJ CJ c cd 13 S3 3 •8 CJ _> *H-» g o CO tyi 1-*—1 d 43 cd ™ 1—1 CO - > CJ OH p ' o 8 •§ g S5 co cd ^ " — 1 cd cd o O o in DISCUSSION Habitat associations of shrubsteppe birds at a local level In this chapter, I was able to make inferences about the relationship between bird relative abundance and vegetation features by considering broadly distributed occupied and unoccupied survey stations within one habitat type: big sagebrush. While songbirds were significantly associated with variation in the surrounding landscape, the strongest relationships were measured at a local level. The habitat associations I described are directly linked to range management practices and can be used to guide conservation planning. However, it is important to note that these relationships are strictly correlational rather than causal (i.e. they are associations rather than requirements), and are based on a single year of study, which rules out an exploration of other processes such as site tenacity and weather events. My results confirm the suggestion made by Rotenberry (1985) that floristics play an important role in habitat associations. All five focal songbird species in my study had significant associations with particular understory plant species. These associations may reflect dietary preference of adult birds, habitat for insect prey, or structural requirements for nesting. The inclusion of these floristic variables improved the description of habitat associations, as compared to associations identified by Wiens and Rotenberry (1981), whose study included all of my focal bird species except for Grasshopper Sparrow. For most of the bird species in my study, associations with certain plants were stronger than associations between birds and 'total grasses' or 'total forbs'. It is possible that grass and forb species may be selected for the seed component of the breeding adult diet (Goebel and Berry 1976), as well as insect habitat and nesting cover. Gut content analysis of breeding adult songbirds collected in June in Washington State have showed that mustard (Brassicaceae) and grass (Poaceae) seeds made up 11.2% of Brewer's Sparrow diet and 36.6% of Vesper Sparrow diet (Stephens 1985). Adult Brewer's Sparrows have been seen eating native grass seeds at my study areas (Mahony, University of British Columbia, pers. comm). Seeds accounted for 5% of Western Meadowlark, 33% of Grasshopper Sparrow, and 61% of Lark Sparrow gut contents in adult birds collected in Nebraska during the breeding season, while nestlings of these species were fed almost entirely on arthropods (Kaspari and Joern 1993). Previous studies have shown that Brewer's Sparrow was positively associated with shrub cover (Knick and Rotenberry 1995). I found that while sagebrush density was useful in distinguishing between Brewer's Sparrow presence and absence, it was not a significant 51 predictor of high versus low relative bird abundance, while large tufted perennials such as parsnip flowered buckwheat and lupine were. The forb-rich areas that Brewer's Sparrow select may have correspondingly high insect abundance and species richness (Krannitz unpublished data). Lepidopteran larvae make up 72% of nestling Brewer's Sparrow diet (Petersen and Best 1986), and adult Brewer's Sparrow have been observed feeding their chicks large larvae from areas of my study sites where lupines are abundant (Mahony pers. comm.). Lepidopterans often have association with specific 'host plants', including plants for larval food (Swengel and Swengel 1999). Large perennial forbs have also been used as nesting substrate in burned areas (Mahony, University of British Columbia, unpublished data). In eastern Washington, Brewer's Sparrow was more abundant at northern latitudes, and in sites with loamy soils (Vander Haegen et al. 2000). While my study did not include latitude or soil type, I found that Brewer's Sparrow was more likely to occur at higher elevations, which are likely to have deeper soil (Florinsky and Kuryakova 1995). Similarly, Vander Haegen et al. (2000) reported that, at southern latitudes, Brewer's Sparrow and Vesper Sparrow were more abundant at higher sites where rainfall and plant productivity were more similar to northern latitudes, and these species were less abundant at sandy sites at low elevations. Lark Sparrow was strongly associated with sand dropseed, a native grass species that is commonly associated with disturbance and overgrazing (Archer 1953). Local-scale plant variables were similarly important to Lark Sparrow in Colorado, where they preferred shale plant communities (Haire et al. 2000). As in Arizona (Bock and Bock 1992), Lark Sparrow at my sites was positively associated with bare soil. Bare soil is positively associated with grazing (Milchunas et al. 1989), which could further indicate an association between Lark Sparrow and disturbed areas. Other studies have shown that Grasshopper Sparrow is associated with 'total perennial grasses' (Vander Hagen et al. 2000). I did not observe a strong relationship between Grasshopper Sparrow occurrence and native grasses, although individual logistic regression models showed a strong relationship between this bird species and needle-and-thread grass. Grass cover is important for concealing nests (Vickery 1996), although Grasshopper Sparrow also select areas with bare soil, presumably to aid in foraging on the ground for insects (Whitmore 1981, Vickery 1996). Grasshopper Sparrow was more abundant in areas with abundant Eurasian weeds relative to native plants, in Manitoba, Illinois and Colorado (Wilson and Belcher 1989, Walk and Warner 2000, Haire et al. 2000). Therefore, it is possible that Grasshopper Sparrow responds more to plant structure rather than to floristics (Walk and Warner 52 2000). The positive association I found for Grasshopper Sparrow with cheatgrass and pasture sage, both weedy species, may be because these plants are widespread in the dry, open areas of my study sites which this species preferred. However, grasshopper abundance, an important component of Grasshopper Sparrow diet (Vickery 1996), tends to increase on overgrazed rangeland invaded by exotic species (Nerney 1958, Smith 1940). A positive association of Western Meadowlark presence with total grass abundance has been found in previous studies (Knick and Rotenberry 1995, Wiens and Rotenberry 1981). However, within occupied sites in my study, Western Meadowlark abundance was positively associated with abundance of needle-and-thread grass, rather than total grass cover. In other studies, Western Meadowlark was positively correlated with native grass species, and negatively associated with Eurasian weeds (Wilson and Belcher 1989, Haire et al. 2000). Seeds are a minor part of Western Meadowlark diet during breeding season (Kaspari and Joern 1993), but grass is important in nest cover (Lanyon 1994), and native grasses could provide habitat for insect prey. As in my study, Western Meadowlark in the Chilcotin-Cariboo grasslands of British Columbia were also more common at low elevations (Hooper and Pitt 1995). Birds in the Chilcotin-Cariboo were found to be more often associated with topographic features than with vegetation structure (Hooper and Pitt 1995). As the importance of floristics was not tested in Hooper and Pitt's (1995) study, it is possible that the correlation between topography and habitat suitability is a consequence of preferred herb layer species growing at lower elevations (Florinsky and Kuryakova 1995). Similar to Brewer's Sparrow, I found that Vesper Sparrow was more likely to occur in areas with abundant lupines, and were less abundant in areas with sand dropseed grass. Vesper Sparrow was more abundant in dry open sites with pasture sage and sparser cover of big sagebrush, which agreed with the findings of Kantrud and Kogoloski (1983, in Dechant et al. 2000). Despite the suggestion that the measurement of the abundance of particular plant species may provide insights into habitat selection, it is only recently that the importance of floristics to bird habitat associations has been addressed. More variation in habitat selection by grassland birds was attributable to plant floristics rather than to physiognomy in a survey of bird communities within a range of grassland types (Rotenberry 1985). Plentovich et al. (1999) found that two plant species indicative of moisture, pitcher plants (Sarracenia spp.) and the grass Panicum verrucosum, were superior to structural variables in explaining the presence of wintering Henslow's Sparrows (Ammodramus henslowii). Floristics was important in distinguishing 53 brood habitat of two grouse species in Wyoming (Klott and Lindzey 1990). Sharp-tailed Grouse (Tympanuchus phasianellus) broods were associated with oniongrass (Melica spp.), and sulphur buckwheat (Eriogonum umbellatum), while sites used by Sage Grouse (Centrocercus urophasianus) contained needle-and-thread grass (Stipa comatd) and desert alyssum (Alyssum desertorum) (Klott and Lindzey 1990). Although there is evidence that Western Meadowlarks, Lark Sparrows and Vesper Sparrows in Washington State were selective in their choice of seeds during breeding season (Goebel and Berry 1976), I have not found any other studies that have examined habitat associations of these birds with individual grass and forb species. Utility of Landsat TM data, and landscape-level associations • All species were modeled successfully with the Landsat TM data at a 100 m.scale, with the exception of Brewer's Sparrow low versus high abundance. Fine-scale classification of sagebrush habitats using Landsat TM data, as into the five subtypes used in this thesis, could therefore be used as a surrogate for detailed, local-level vegetation surveys. These methods have the advantage of allowing a user to map habitat suitability over large areas beyond the sample plot level. The addition of digitally-derived topographic variables improved most models, increasing the ability to predict bird distributions in the absence of survey data. However, all Landsat TM models were relatively weak compared to models constructed using local habitat variables measured from field data. For all species the landscape models had an AICc score at least two points higher than the local models, so these models should be used with caution. While variables derived from Landsat TM never matched local vegetation models in their ability to predict bird presence and abundance, they could be used to quantify the surrounding landscape, or context, of point count stations. This would be difficult to measure without remote sensing data. For all species, local-level models were superior to models of landscape-level features derived from Landsat TM, and this was scale dependent. Despite the relative weakness of landscape models compared to local models, I found that the addition of landscape-level variables at one or more scales improved my ability to predict the presence and relative abundance of songbirds. Ribic and Sample (2001) found that individual species of grassland birds in Wisconsin had variable responses to landscape measures to three different scales: 200, 400, and 800 m. They found that Grasshopper Sparrow was best modeled with landscape cover variables within 400 m (Ribic and Sample 2001), which was similar to my study where this species was best modeled at the 500 m scale. In contrast to my study, Berry and Bock (1998) did 54 not find that measures of landscape context improved local-level models for a foothills scrub area. They suggested that because their study areas were naturally fragmented, the songbird species they studied were adapted to this habitat and therefore not affected by the interspersion of forest (Berry and Bock 1998). I expected all species to exhibit negative associations with neighbouring trees and agricultural land because proximity to these habitats could elevate rates of corvid and mammalian predation, and susceptibility to parasitism by Brown-headed Cowbirds (Molothrus ater; Johnson and Temple 1990). Grasshopper Sparrow occurrence was negatively associated with both coniferous forest and deciduous forest at all three spatial scales, similar to the negative association of Lark Sparrow and Western Meadowlark with coniferous or deciduous forest at one or more scale. In contrast, Brewer's Sparrow was positively associated with conifers at a 1 km and 2 km scale. Conifer cover increased with elevation, as did forb abundance, so this result may not represent a preference for tree cover, but rather illustrate that preferred habitats are likely to be within 2 km of conifers. The loamy soils favoured by Brewer's Sparrow in Washington (Vander Haegen et al. 2000) are probably more prevalent at high elevations in my study areas. It was possible that shrubsteppe within 1 km of forest edge in my study was of such high quality (i.e. contains high forb density) that Brewer's Sparrow occurred in these habitats despite potential added risk. Dispersing Brewer's Sparrow juveniles have been seen in aspen stands adjacent to sagebrush habitat (Yu 2001), which indicates that there might be some advantage in selecting habitat near tree cover. It is also possible that tree cover within 1 km does not increase the threat of predation or parasitism to Brewer's Sparrow. The effect of surrounding tree cover on Brewer's Sparrow nest predation in the Okanagan is currently under investigation (Welstead, unpublished thesis). Positive associations of Western Meadowlark and Vesper Sparrow with agriculture were also not expected, but understandable given that they were the most common species, and both are known to occur in pastures (Lanyon 1994, Campbell et al. 2001). Lark Sparrow was also positively associated with agriculture at the 2 km scale. The sandy, disturbed sites preferred by Lark Sparrow in my study are often in valley bottoms close to agricultural land. Habitat associations identified in this chapter highlight the importance of scale, in examining bird-habitat relationships. A multi-scale approach to the identification of habitat associations was also superior to the use of a single-scale in the study of urban landbirds in Vancouver, although in this study birds were more strongly associated with landscape-level influences (Melles 2000). Multiple scales can also be examined within a local level. In a study of forest 55 songbird response to thinning, habitat selection was influenced by local vegetation, predictive power was greater at the plot level than at the nest-site level (Easton and Martin 1998). B y examining response from the floristic through to the landscape level, I determined that songbirds were strongly correlated with individual plant species. However, it is still possible that these associations represent a dependence on other factors that I did not measure. Site fidelity, intra or interspecific competition, vegetation productivity, prey abundance and predation risk are all examples of processes that could interact to further affect habitat selection of songbirds in this area. Despite my effort to study a range of scales, my results only represent a fragment of the total picture of how songbirds respond to their environment, and because reproductive success was not measured, habitat quality can only be inferred from relative abundance. In planning the management of these species, the potential influence of other processes should be considered by examining concurrent songbird research in the study area (i. e. Y u 2001 and Mahony and Welstead, University of British Columbia, unpublished theses). Management implications from this chapter are discussed further in Chapter Four. 56 CHAPTER FOUR: MANAGEMENT RECOMMENDATIONS The association of songbird species with particular plant species provides a means by which wildlife managers can communicate with range managers in order to ensure the protection of songbird habitat, both on public and private land. The importance of considering the ecological requirements of individual songbird species in the southern Okanagan and Similkameen Valleys was clearly illustrated by the variation in habitat use of each of my five focal species. Single species models for the five focal species showed variation among their habitat preferences, so it may important to understand the habitat associations of each. B y addressing the habitat associations of individual species, rather than sagebrush breeding birds as a group, it w i l l be necessary to manage for a diversity of sagebrush habitat types that benefit all species. For example, by focussing conservation efforts solely on Brewer's Sparrow, one would risk the inadequate management of Lark Sparrow, a species that occurs in areas of low Brewer's Sparrow abundance. The variation in habitat quality for different species is further illustrated by examining the variation in songbird relative abundance within the different study areas (Appendix IV) . Brewer's Sparrow is red-listed in the province of British Columbia ( M E L P 1998) and, of the focal species I examined, it is the most restricted to sagebrush habitat. M y results have shown that the patchy distribution of this species can be attributed to herb layer variation, with higher Brewer's Sparrow abundance coinciding with lupine and buckwheat. Managers should therefore ensure that adequate areas of this specific sagebrush habitat type are maintained for Brewer's Sparrow populations. Some plant species that are associated with Brewer's Sparrow abundance are directly affected by grazing and burning. Lupines, for example, are reported to increase after fires (Grigore and Tramer 1996). Although not all of the burn histories of my point count stations are known, some of the areas of greatest forb productivity had been recently burned. Lupines are also toxic to cattle and thought to increase under grazing pressure, but in overgrazed areas they may still be used as forage by cows. Cows often eat lupines in the International Grasslands (Mahony, University of British Columbia, pers. comm.) and lupines are highly abundant in a grazing exclosure in this area, relative to the surrounding grazed pasture. Brewer's Sparrow nests have also been lost to cattle trampling (Mahony, pers. comm), further suggesting that this species is negatively affected by heavy grazing. The year I collected my data, 1998, was the warmest in the past decade, and the apparent preference of Brewer's Sparrow for high and moist 57 sites could be atypical. However, surveys at 12 regions in the South Okanagan in 1991 also indicated low Brewer's Sparrow density at Chopaka, which was the lowest elevation site that was counted (Harvey 1992). Lark Sparrow is red-listed in British Columbia (MELP 1998), although they are more widespread throughout the province than Grasshopper Sparrow and Brewer's Sparrow, and is found in antelope bitterbrush grasslands as well as sagebrush (Krannitz and Rohner 2000, Campbell et al. 2001). My results indicate that Lark Sparrow is unique among the other sagebrush breeding birds in my study in that they occur mainly at low elevations and in sandy habitats, which would probably not be able to support the vegetation communities preferred by Brewer's Sparrow and other species. Areas preferred by Lark Sparrow are perhaps the most susceptible to development and degradation as they are close to existing settlement in the valley bottoms. The strong association of Lark Sparrow with the native grass sand dropseed suggests that this species would benefit from management strategies that curb or reverse invasion by exotic species. Maintenance of some of these sandy, low elevation areas for Lark Sparrow habitat is crucial for the conservation of this species. While El 2 and Chopaka are conservation reserves, most remaining Lark Sparrow habitat is on First Nations reserve land or other private property. The preference of this species for what may appear to be degraded habitat indicates that it is preferable that private owners maintain their property as range, rather than sell it for development. The discovery of Grasshopper Sparrow at sites that are not usually accessed by birders is of conservation value, as this species is red-listed in British Columbia due to low population numbers (MELP 1998). Models for Grasshopper Sparrow were based on a relatively low number of detections, and therefore not necessarily as robust as for the other species. However, it was clear from my research (Paczek, unpublished data) that Grasshopper Sparrow is an area-sensitive species in the southern Okanagan and Similkameen Valleys, requiring large tracts of land away from forest edges. More importantly, they require a sparse cover of sagebrush, and native herb species cover did not seem to be important to this species. In North Dakota, both Grasshopper Sparrow and Western Meadowlark were absent from grassland sites that had been subject to long-term (>15 year) fire-suppression, suggesting that fire regimes that mimic natural disturbance levels are beneficial to both species (Madden et al. 1999). Prescribed burns in winter in discrete blocks of Grasshopper Sparrow range could be an efficient way to maintain the sparse cover of sage they require, and this would also benefit Western Meadowlark. Grasshopper Sparrow was also an effective indicator of butterfly species that specialize in prairie habitat 58 throughout the Midwestern USA (Swengel and Swengel 1999), and it is possible that other taxa of conservation concern would be protected through the management of this species. Western Meadowlark breeds in grasslands from northeastern to central and southern British Columbia (Campbell et al. 2001). While this species has expanded its range towards the north in British Columbia in the past 50 years, it has also been extirpated from grasslands on Vancouver Island due to urbanization (Campbell et al. 2001). Western Meadowlark does not have listed status in British Columbia, but according to North American Breeding Bird Survey Data they are among the most rapidly declining grassland bird species in North America (Herkert 1995). Maintenance of native rangeland (i.e. needle-and-thread grass) is important for this species (Wilson and Belcher 1989). Therefore, the spread of exotic plant species should be controlled, and grazing practices adjusted accordingly on crown land and conservation reserves. As with Grasshopper Sparrow, large tracts of sparse shrub land maintained by burning would benefit Western Meadowlark populations. It is possible that extirpation of Western Meadowlark from Vancouver Island may partly be due to fire suppression, as this species was absent from unburned grassland (Madden et al. 1999). Fire suppression may have already altered the burn properties of shrubsteppe in my study areas due to fuel accumulation, which could result in relatively hot burns (Madden et al. 1999). As grassland and shrubsteppe ecosystems are maintained by disturbance, it would be valuable to use crown land to experiment with prescribed burns in the fall to early spring. Effects of range burning on bird populations and spread of exotic plant species could then be monitored. Burning 50 ha parcels in rotation would lead to a mosaic of different shrub and densities (Madden et al. 1999) and this would help to fulfill the habitat requirements of all five species. Vesper Sparrow was the most abundant species at my sites. This species is not listed in British Columbia and breeds throughout the central southwestern region of the province (Campbell 2001). While it is adapted to more general habitats than the other focal species, it most commonly nests in sagebrush dominated shrubsteppe in BC (Campbell 2001). Vesper Sparrow had habitat associations in common with Grasshopper Sparrow, Brewer's Sparrow and Western Meadowlark, preferring lupine, but also sparse sagebrush. Management practices that conserve adequate habitat for listed bird species will likely fulfill requirements for the maintenance of Vesper Sparrow populations. Like Brewer's Sparrow, this species is not associated with the sand dropseed habitat preferred by the Lark Sparrow. The strong, significant negative association of Vesper Sparrow with dead wood (Table 6) suggests that this species may also benefit from prescribed burning of shrubsteppe in the fall to early spring. 59 Perhaps the most crucial issue for shrubsteppe bird management is the ownership of rangeland. Despite recent acquisition of crown land for parks in the southern Okanagan, between 60-70% of potential habitat for the red-listed birds in my study is owned privately or by First Nation reserves (MELP 1998). It is essential for wildlife managers to communicate with these private landowners, since many people may be unaware of the ecological importance of sagebrush habitat (Hooper and Pitt 1995). The First Nation Band members and private landowners who allowed me on their land for my surveys are clearly interested in aiding in conservation efforts. Fostering a good relationship between managers and cattle ranchers could be beneficial for both parties, since management strategies designed to enhance shrubsteppe bird habitat do not necessarily conflict with the maintenance of healthy rangeland. For example, in Idaho, fall grazing improved rangeland in poor condition, with forage lost by spring deferment offset by increased grazing rates in the fall (Laycock 1967). Replacement of season-long grazing with rotational grazing systems can be less damaging to native grasslands (Stoddart et al. 1975). Prescribed burning also decreases shrub cover, and promotes the growth of grasses (Madden et al. 1999). In a successful program in the Netherlands, dairy farmers were given financial incentives for each clutch of declining meadowbird species found on their land (Musters et al. 2001). Farmers participated in nest searching, and altered farming practices in ways that have led to an increase in the productivity of declining meadowbirds. Although grassland bird nests are much more difficult to find, a similar scheme of financial incentives for providing bird habitat in the South Okanagan and Similkameen Valleys may increase the involvement of private landowners in conservation practices. A common rangeland management practice of private landowners that could be discouraged is the re-seeding native grasslands with exotic forage species such as crested wheatgrass (Agropyron cristatum). Few point count stations in my study were dominated by crested wheatgrass, but where it did occur, at Blind Creek, Kilpoola Lake, and Ecoreserve El2, the areas with abundant crested wheatgrass supported fewer songbirds than neighbouring sagebrush with a native understory. Spot mapping is needed to confirm this impression. The greatest threat to shrubsteppe bird populations is habitat loss due to irreversible conversion to urban or agricultural property. Across the United States and Canada, declines in native prairie since European settlement vary, but are as high as 99.9% in Manitoba and some mid-western states (Samson and Knopf 1994). Over 90% of land in the Okanagan and lower Similkameen valleys has already been altered from its original state (Redpath 1990, in Cannings 1995). It is crucial to the conservation of shrubsteppe bird populations that the maintenance of 60 remaining native grasslands be encouraged. The use of specific habitat associations for these threatened songbird species may be used to prioritize land for conservation. The translation of these habitat variables into Landsat T M classifications may be a useful tool to monitor the amount and quality of habitat as it changes over time. 61 REFERENCES Archer, S. G. 1953. The American Grass book; a manual of pasture and range practices. University of Oklahoma press, Norman, OK. Begg, C. B. and R. Gray. 1984. Calculation of polychotomous logistic regression parameters using individualized regressions. Biometrika 71: 11-18. Bender, R. and U. Grouven. 1998. Using binary logistic regression models for ordinal data with non-proportional odds. Journal of Clinical Epidemiology 51: 809 - 816. Berry, M. E. and C. E. Bock. 1998. Effects of habitat and landscape characteristics on avian breeding distributions in Colorado foothills shrub. Southwestern Naturalist 43: 453 -461. Bian, L. and E. West. 1997. GIS modeling of elk calving habitat in a prairie environment with statistics. Photogrammetric Engineering & Remote Sensing 63: 161 - 167. Bock, C. E., and J. H. Bock. 1992. Response of birds to wildfire in native versus exotic Arizona grassland. Southwestern Naturalist 37: 73 - 81. Bolger, D. T., T. A. Scott and J. T. Rotenberry. 1997. Breeding bird abundance in an urbanizing landscape in coastal southern California. Conservation Biology 11: 406 -All. Bradford, D. F., Franson, S. E., Neale, A. C , Heggem, D. T., G. R. Miller and G. E. Canterbury. 1998. Bird species assemblages as indicators of biological integrity in Great Basin rangeland. Environmental Monitoring and Assessment 49: 1 - 22. Brower, J. E., J. H. Zar and C. N. von Ende. 1989. Field and laboratory methods for general ecology. 3rd ed. Wm C. Brown Publishers. Burnham, K. P. and D. R. Anderson. 1998. Model Selection and Inference. A Practical Information-Theoretic Approach. Springer-Verlag Inc, NY. Campbell, R.W., N.K. Dawe, I. McTaggart-Cowan, J.M. Cooper, G.W. Kaiser, A.C. Stewart and M.C.E. McNall. 2001. The birds of British Columbia, Volume 4: Passerines, wood-warblers through Old World sparrows. BC Ministry of Environment, Lands and Parks and Canadian Wildlife Service. Cannings, R.J. 1995. Status report on the grasshopper sparrow, Ammodramus savannarum, in British Columbia. B.C. Ministry of Environment, Lands and Parks, Victoria, BC. Wildlife Bulletin #B-77. Cannings, R.J., R.G. Cannings and S. Cannings. 1987. Birds of the Okanagan Valley, British Columbia. Royal British Columbia Museum. Victoria BC. Cody, M. L. 1981. The roles of vegetation structure, competitors, and productivity. Bioscience 31: 107-113. 62 Congalton, R. G., K. Green and J. Teply. 1993. Mapping old growth forests on national forest and park lands in the Pacific Northwest from remote sensing data. Photogrammetric Engineering and Remote Sensing 59: 529 - 535. Daubenmire, R. 1959. A canopy-coverage method of vegetational analysis. Northwest Science 33:43 -64. Dechant, J. A., M. F. Dinkins, D. H. Johnson, L. D. Igl, C. M. Goldade, and B. R. Euliss. 2000. Effects of management practices on grassland birds: Vesper Sparrow. Northern Prairie Wildlife Research Center, Jamestown, ND. Jamestown, ND: Northern Prairie Wildlife Research Center. Dennis, P., M. R. Young and I. J. Gordon. 1998. Distribution and abundance of small insects and arachnids in relation to structural heterogeneity of grazed, indigenous grasslands. Ecological Entomology 23: 253 - 264. Dunning, J. B and J. H. Brown. 1982. Summer rainfall and winter sparrow densities: a test of the food limitation hypothesis. Auk 99: 123 - 129. Easton, W. E. and K. Martin. 1998. The effect of vegetation management on breeding bird communities in British Columbia. Ecological Applications, 8: 1092 - 1103. Florinksy, I. V. and G. A. Kuryakova. 1995. Influence of topography on some vegetation cover properties. Cantena 27: 123- 141. Foody, G. M. 1996. Approaches for the production and evaluation of fuzzy land cover classifications from remotely-sensed data. International Journal of Remote Sensing 17: 1317- 1340. Goebel, C. J. and G. Berry. 1976. Selectivity of range grass seeds by local birds. Journal of Range Management 29: 393 - 395. Graetz, R. D., R. P. Pech and A. W. Davis. 1988. The assessment and monitoring of sparsely vegetated rangelands using callibrated Landsat data. International Journal of Remote Sensing 9: 1201 - 1222. Grigore, M. T. and E. J. Tramer. 1996. The short-term effect of fire on Lupinus perennis (L). Natural Areas Journal 16: 41 - 48. Haire, S. L., C E . Bock, B. S. Cade and B. S. Bennett. 2000. The role of landscape and habitat characteristics in limiting abundance of grassland nesting songbirds in an urban open space. Landscape and Urban Planning 48: 65 - 82. Harvey, D. W. 1992. The distribution, density and habitat of Brewer's Sparrows Spizella breweri in the South Okanagan Valley of British Columbia. Unpubl. report. Simon Fraser University, Burnaby BC. 63 Herkert, J. R. 1995. An analysis of Midwestern breeding bird population trends: 1966-1993. American Midland Naturalist 134: 41 - 50. Homer, C. G., Edwards, T. C., Ramsey, R. D., and K. P. Price. 1993. Use of remote sensing methods in modelling sage grouse winter habitat. Journal of Wildlife Management 57: 78 - 84. Hooper, T. D. and M. D. Pitt. 1995. Problem analysis for Chilcotin-Cariboo grassland biodiversity. B. C. Min. Environ., Lands and Parks, Widl. Branch, Wildl. Bull. No. b-82, Victoria, BC. Hosmer, D. W. and S. Lemeshow. 1989. Applied Logistic Regression. John Wiley and Sons, New York, NY. Johnson, R. G. and S. A. Temple. 1990. Nest predation and brood parasitism of tallgrass prairie birds. Journal of Wildlife Management 54: 106 - 111. Kaspari, M. and Joern, A. 1993. Prey choice by 3 insectivorous grassland birds: re-evaluating opportunism. Oikos 68: 414 - 430. Klott, J. H. and F. G. Lindzey. 1990. Brood habitats of sympatric sage grouse and Columbian Sharp Tailed Grouse in Wyoming USA. Journal of Wildlife Management 54: 84 - 88. Knick, S. T., and J. T. Rotenberry. 2000. Ghosts of habitats past: contribution of landscape change to current habitats used by shrubland birds. Ecology 81: 220 - 227. Knick, S. T. and J. T. Rotenberry. 1995. Landscape characteristics of fragmented shrubsteppe habitats and breeding passerine birds. Conservation Biology 9: 1059 - 1071. Knick, S. T., Rotenberry, J. T., and T. J. Zarriello. 1997. Supervised classification of Landsat Thematic Mapper imagery in a semi-arid rangeland by nonparametric discriminant analysis. Photogrammetric Engineering and Remote Sensing 63: 79 - 86. Krannitz, P.G. and C. Rohner. 2000. Habitat relationships of endangered grassland birds in the south Okanagan. Pp. 823-829 In: L. M. Darling (Ed.) Proceedings of a conference on the biology and management of species and habitats at risk, Kamloops,BC. 15-19 Feb, 1999. Volume One, BC Ministry of Environment, Lands and Parks, Victoria,BC, and University College of the Cariboo, Kamloops. Lanyon, W. E. 1994. Western Meadowlark (Sturnella neglecta). In The Birds of North America, No. 104 (A. Poole and F. Gill, eds.). The Birds of North America, Inc., Philadelphia, PA. Lauver, C. L. and J. L. Whistler. 1993. A hierarchical classification of Landsat TM Imagery to identify natural grassland areas and rare species habitat. Photogrammetric Engineering and Remote Sensing, 59: 627 - 634. Laycock, W. A. 1967. How heavy grazing and protection affect sagebrush-grass ranges. Journal of Wildlife Management: 20: 206-213. 64 Lillesand, T. M. and R. W. Keifer. 1994. Remote Sensing and Image Interpretation. Third Edition, John Wiley and Sons, New York, NY. MacArthur, R. H. and J. W. MacArthur. 1961. On bird species diversity. Ecology 42: 594 -598. Madden, E. M., A. J. Hansen, and R. K. Murphy. 1999. Influence of prescribed fire history on habitat and abundance of passerine birds in northern mixed-grass prairie. The Canadian Field-Naturalist 113: 627 - 640. Martin, J. W. and J. R. Parrish. 2000. Lark Sparrow (Chondestes grammacus). In The Birds of North America, No. 488 (A. Poole and F. Gill, eds.). The Birds of North America, Inc., Philadelphia, PA. Medinger, D. V. and J. P. Pojar. 1991. Ecosystems of British Columbia. Special report series, British Columbia Ministry of Forests, no. 6, Melles, S. J. 2000. Effects of landscape and local habitat features on bird communities: A study of an urban gradient in Greater Vancouver. MSc. Thesis. University of British Columbia, Vancouver BC. MELP, British Columbia Ministry of Environment, Lands and Parks. 1998. Habitat atlas for wildlife at risk: South Okanagan & Lower Similkameen, Victoria BC. Menard, 2. 1995. Applied logistic regression analysis. SAGE Publications, Thousand Oaks, CA. Milchunas, D. G., Lauenroth, W. K., Chapman, P. L and M. K. Kazempour. 1989. Effects of grazing, topography, and precipitation on the structure of a semiarid grassland. Vegetatio 80: 11 -23. Minckley, R. L., J. H. Cane, and L. Kervin. 2000. Origins and ecological consequences of pollen specialization among desert bees. Proceedings of the Royal Society of London, Series B 267: 265-271. Musters, C. J. M., M. Kruk, H. J. De Graaf, and W. J. Ter Keurs. 2001. Breeding birds as a farm product. Conservation Biology 15: 363 - 369. Nerney, N. J. 1958. Grasshopper infestations in relation to range condition. Journal of Range Management 11: 247. Nohr, H. and A. F. Jorgensen. 1997. Mapping of biological diversity in Sahel by means of satellite image analyses and ornithological surveys. Biodiverstiy and Conservation 6: 545-566. Paruelo, J. M., Epstein, H. E., Lauenroth, W. K., and I. C. Burke. 1997. ANPP estimates from NDVI for the central grassland region of the United States. Ecology 78: 953 - 958. 65 Paruelo, J. M. and R. A. Golluscio. 1994. Range assessment using remote sensing in Northwest Patagonia (Argentina). Journal of Range Management 47: 498 - 502. Peterjohn, B. G. and J. R. Sauer. 1999. Population status of North American grassland birds from the North American Breeding Bird Survey, 1966-1996. Studies in Avian Biology 19:27-44. Petersen , K. L. and L. B. Best. 1986. Diets of nestling sage sparrows and Brewer's Sparrow in an Idaho sagebrush community. Journal of Field Ornithology 57: 283 - 294. Plentovich, S., Holler, N. R. and G. E. Hill. 1999. Habitat requirements of Henslow's Sparrows wintering in silvicultural lands of the Gulf Coastal Plain. Auk 116: 109-115. Pribil, S. and J. Pieman. 1997. The importance of using the proper methodology and spatial scale in the study of habitat selection by birds. Canadian Journal of Zoology 75: 1835 -1844. . Quinn, M. A and D. D. Walgenbach. 1990. Influence of grazing history on the community structure of grasshoppers of a mixed-grass prairie. Environmental Entomology 19: 1756 - 1766. Ribic, C. A. and D. W. Sample. 2001. Associations of grassland birds with landscape factors in Southern Wisconsin. American Midland Naturalist 146: 105- 121. Rodewald, A. D., and R. H. Yahner. 2000. Influence of landscape and habitat characteristics on Ovenbird pairing success. Wilson Bulletin 112: 238 - 242. Rotenberry, J. T. 1985. The role of habitat in avian community composition: physiognomy or floristics? Oceologia 67: 213 - 217. Rotenberry, J. T. and J. A. Wiens. 1980. Habitat structure, patchiness and avian communities in North American steppe vegetation: a multivariate analysis. Ecology 61: 1228 - 1250. Rotenberry, J. T., Patten, M. A. and K. L. Preston. 1999. Brewer's Sparrow (Spizella breweri). In The Birds of North America, No. 390 (A. Poole and F. Gill, eds.). The Birds of North America, Inc., Philadelphia, PA. Ryder, R. A. 1980. Effects of grazing on bird habitats. In Management of Western Forests and Grasslands for Nongame Birds. USDA - Forest Service General Technical Report. INT - 86, Ogden, UT. Saab, V. 1999. Importance of spatial scale to habitat use by breeding birds in riparian forests: a hierarchical analysis. Ecological Applications 9: 135-151. Saab, V. A., Bock, C. E., Rich, T. D and S. Dobkin. 1995. Livestock grazing effects in western North America. Pp: 311 - 353. In: Ecology and Management of Neotropical Migratory Birds. Martin, T. E. and D. M. Finch (eds). Oxford University Press, NY. Samson, F. and F. Knopf. 1994. Prairie conservation in North America. Bioscience 44: 418 -421. 66 Savard, J-P. L and T. D. Hooper. 1995. Influence of survey length and radius size on grassland bird surveys by point counts at Williams Lake, British Columbia. USDA Forest Service Gen. Tech. Rep. PSW-GTR-149. Smith, C. C. 1940. The effect of overgrazing and erosion upon the biota of the mixed-grass prairie of Oklahoma. Ecology 21: 381 - 397. Stephens, D. A. 1985. Foraging ecology of shrubsteppe birds in central Washington. MSc Thesis. Central Washington University, WA. Stoddart, L. A., A. D. Smith and T. W. Box. 1975. Range management. 3rd ed., McGraw-Hill Book Co., New York, NY. Swengel, S. R. and A. B. Swengel. 1999. Correlations in abundance of grassland songbirds and prairie butterflies. Biological Conservation 90: 1-11. Tueller, P. T. 1992. Overview of remote sensing for range management. Geocarto international 1:5 — 10. Turner, J. S. and P. G. Krannitz. 2001. Conifer density increases in semi-desert habitats of British Columbia in the absence of fire. Northwest Science 75: 176 - 182. Vander Haegen, W. M., F. C. Dobler and D. J. Pierce. 2000. Shrubsteppe bird response to habitat and landscape variables in Eastern Washington, U.S.A. Conservation Biology 14: 1145- 1160. Vickery, P. D. 1996. Grasshopper Sparrow (Ammodramus savannarum). In The Birds of North America, No. 239 (A. Poole and F. Gill, eds.). The Birds of North America, Inc., Philadelphia, PA. Walk, J. E. and R. E. Warner. 2000. Grassland management for the conservation of songbirds in the Midwestern U.S.A. Biological Conservation 94: 165- 172. Warman, L. 2001. Identifying priority conservation areas using systematic reserve selection and GIS at a fine spatial scale: a test case using threatened vertebrate species in the Okanagan, British Columbia. MSc. Thesis. University of British Columbia, Vancouver, BC. Whitmore, R. C. 1981. Structural characteristics of Grasshopper Sparrow habitat. Journal of Wildlife Management 45: 811 - 814. Wiens, J. A. 1989. Spatial scaling in ecology. Functional Ecology 3: 385 - 397. Wiens, J. A. and J. T. Rotenberry. 1981. Habitat associations and community structure of birds in shrubsteppe environments. Ecological Monographs 51: 21 -41. Wilson, S. D. and J. W. Belcher. 1989. Plant and bird communities of native prairie and introduced Eurasian vegetation in Manitoba, Canada. Conservation Biology 3: 39 - 44. 67 , J. 2001. Postfledging habitat use and movements of Brewer's Sparrows (Spizella breweri breweri) in the South Okanagan region. M S c Thesis. University of British Columbia, Vancouver, B C . 68 APPENDIX I. Protocol used to ground-truth sagebrush classes. Real-time corrected GPS data were available in the field for ground-truthing purposes. Points were located using field maps and GPS within 5 m, and verified by myself. Where sagebrush density was not obvious: sparse: <5%, mid density: 5 - 30%, dense: 30%+, a tape was extended for 30 m at a random direction through the point and sagebrush density was measured as the proportion of sagebrush that intercepted the tape. TABLE 1.1. Criteria used to assess 'dry' versus 'forb' sites. DRY Less than 75% vegetated Presence of rocky or sandy soil or scree No forbs or few forbs indicative of dry sites such as phlox and mariposa lily Presence of cactus, grass species inci. red 3-awn, sand dropseed FORB more than 75% vegetated Presence of moist soil High density of forbs inci. lupine, buckwheat, balsamroot, aster sp Grass species inci. June grass, Kentucky bluegrass Sites were rated as agreeing with the classification as: not at all, poor, fair, and excellent. The classes of field verification sites were sometimes ambiguous. For example a site with exactly 30% sage density would be noted as high or mid. A site with bare sandy soil and a lot of lupines would be recorded as 'fair: dry or forb'. TABLE I. 2. Results rigorous analysis of sage accuracy data for the six sage classes, counting only those with an 'excellent' score as correct. USER ACCURACY PRODUCER ACCURACY Habitat Class #correct/ %correct #correct/ %correct map total ground total DD 22/38 58 22/25 88 DF 18/32 56 18/20 90 MD 32/36 89 32/45 71 MF 23/38 61 23/34 68 SD 39/47 87 39/54 72 SF 25/38 65 25/42 59 SAGE CLASSES 159/229 69 159/220 72 overall accuracy 563/685 82 563/685 82 69 APPENDIX II. Bird species surveyed in the southern Okanagan and Similkameen Valleys, British Columbia, 1998. Species are listed in decreasing order of the number of survey stations at which they occurred within a radius of 100 m. 'Average' is the number of birds per survey over all 245 stations. Common Name Taxonomic Name Number Average of number Stations birds Vesper Sparrow Pooecetes gramineus 233 1.875 Western Meadowlark Sturnella neglecta 196 0.916 Brewer's Sparrow Spizella breweri 142 0.753 Chipping Sparrow Spizella passerina 109 0.305 Lark Sparrow Chondestes grammacus 69 0.128 Spotted Towhee Pipilo maculatus 47 0.087 American Robin Turdus migratorius 46 0.068 Brown Headed Cowbird Molothrus ater 32 0.147 Grasshopper Sparrow Ammodramus savannarum 29 0.088 Northern Flicker Colaptes auratus . 24 0.030 Dark-eyed Junco Junco hyemalis 24 0.031 Clay-colored Sparrow Spizella pallida 24 0.031 Brewer's Blackbird Euphagus cyanocephalus 23 0.118 Black billed Magpie Pica pica 22 0.042 Lazuli Bunting Passerina amoena 20 0.047 Western Kingbird Tyrannus verticalis 15 0.020 Eastern Kingbird Tyrannus tyrannus 11 0.013 Common Nighthawk Chordeiles minor 11 0.017 American Goldfinch Carduelis tristis 11 0.014 Mountain Bluebird Sialia currucoides 9 0.018 Long-billed Curlew Numenius americanus 8 0.020 American Crow Corvus brachyrhynchos 8 0.013 Red-winged Blackbird Agelaius phoeniceus 7 0.011 Killdeer Charadrius vociferus 7 0.021 Black-capped Chickadee Parus atricapillus 7 0.010 California Quail Callipepla californica 6 0.008 Yellow-rumped Warbler Dendroica coronata . 5 0.006 Tree Swallow Tachycineta bicolor 5 0.009 Mountain Chickadee Parus gambeli 5 0.006 Lewis' Woodpecker Melanerpes lewis 5 0.007 Western Wood-peewee Contopus sordidulus 4 0.004 European Starling Sturnus vulgaris • 4 0.009 Pine Siskin Carduelis pinus 4 0.030 House Wren Troglodytes aedon 4 0.005 Clark's Nutcracker Nucifraga columbiana 4 0.005 Bullock's Oriole Ictera galbula 4 0.004 Western Bluebird Sialia mexicana 3 0.004 Warbling Vireo Vireo gilvus 3 0.003 70 A P P E N D I X II. (Continued) Bird species surveyed in the southern Okanagan and Similkameen Valleys, British Columbia, 1998. Species are listed in decreasing order of the number of survey stations at which they occurred within a radius of 100 m. 'Average' is the number of birds per survey over all 245 stations. Common Name Taxonomic Name Number of Stations Average number birds Say's Phoebe Sayornis saya 3 0.003 Dusky Flycatcher Empidonax oberholseri 3 0.004 American Kestrel Falco sparverius 3 0.004 Willow Flycatcher Empidonax traillii 2 0.002 Swainson's Thrush Cathaus ustulatus 2 0.002 Sage Thrasher Oreoscoptes montanus 2 0.002 Savannah Sparrow Passerculus sandwichensis 2 0.002 Red tailed Hawk Buteo jamaicensis 2 0.003 . . Cedar Waxwing Bombycilla cedrorum 2 0.004 Calliope Hummingbird Stellula calliope 2 0.004 Winter Wren Troglodytes troglodytes 1 0.002 Wilson's Phalarope Phalaropus tricolor 1 0.009 Western Tanager Piranga ludoviciana 1 0.001 Short eared Owl Asio flammeus 1 0.001 Red Crossbill Loxia curvirostra 1 0.001 Ruffed Grouse Bonasa umbellus 1 0.001 Rock Wren Salpinctes obsoletus 1 0.001 Ring-necked Pheasant Phasianus colchicus 1 0.002 Red-eyed Vireo Vireo olivaceus 1 0.001 Olive-sided Flycatcher Contopus borealis 1 0.001 Orange-crowned Warbler Vermivora celata 1 0.001 Loggerhead Shrike Lanius ludovicianus 1 0.001 House Finch Carpodacus mexicanus 1 0.002 Gull Larus sp. 1 0.001 Gray Jay Perisoreus canadensis 1 0.001 • Common Raven Corvus corax 1 0.005 Canyon Wren Catherpes mexicanus 1 0.001 Blue Grouse Dendragapus obscurus 1 0.001 Bam Swallow Hirundo rustica 1 0.001 71 APPENDIX III. Bird species surveyed in the southern Okanagan and Similkameen Valleys, British Columbia, 1998. These species occurred beyond the radius of 100 m, and are listed in descending order according to number of stations where they were seen or heard, within an unlimited radius. Common Name Taxonomic Name Number of stations Mourning Dove Zenaida macroura 69 Common Snipe Gallinago gallinago 40 Chukar Alectoris chukar 28 Canada Goose Branta cadensis 11 Yellow-headed Blackbird Xanthocephalus xanthocephalus 11 Yellow Warbler Dendroica petechia 11 Red-breasted Nuthatch Sitta canadensis 10 Violet-green Swallow Tachycineta thalassina 2 White-throated Sparrow Zonotrichia albicollis 4 White-breasted Nuthatch Sitta carolinensis , . , . 3 Red-necked Phalarope Phalaropus lobatus "'.2 Red-naped Sapsucker Sphyrapicus ruber 2 Sora Porzana Carolina 2 Veery Catharas fuscescens 2 Alder Flycatcher Empidonax alnorum 1 Cassin's Finch Carpodacus cassinii 1 Downy Woodpecker Picoides pubescens 1 Gray Catbird Dumetella carolinensis 1 Northern Harrier Circus cyaneus 1 Mallard Anas platyrhynchos 1 Marsh Wren Cistothorus palustris 1 Northern Waterthrush Seiurus noveboracensis 1 Pileated Woodpecker Dryocopus pileatus 1 Redhead Athya americana 1 Rough-legged Hawk Buteo lagopus 1 Rock Dove Columba livia 1 Steller's Jay Cyanocitta stelleri 1 Townsend's Warbler Dendroica townsendi 1 White-crowned Sparrow Zonotrichia leucophrys 1 White-throated Swift Aeronautes saxatalis 1 Golden-crowned Kinglet Regulus satrapa 1 Solitary Vireo Vireo solitarius 1 72 00 fa r OS s o tN CN Tt so © d o o ^ ° oo © OS CN Tj- i n 3 S >^ • , - ; © ro CN O oo r-~ vq p d ro f~ o CN OO . _• r< O i n C O d i n E E > W c/j SO CN O © O VO Tt i n SO os m >n co CN SO ~ O — d d ° ° Tf —; © o SO Tf CN CO d d £ £ " ! ^. ^ o <N r - <N oo m O 00 d —I SO Os 0 M OO " ro —c oo >n o o f- f~- o ,-; © £ w S w T? 3 P n ^ © ro r~~ o OS CN o i/1 d Tf" CN Os o o ro oo so uo >n oo r~ oo °- 2 © P _ ; O n o o o o o o <N _ ; - ro co i-! P ^ „• o m S ^ q o o _ < ° C O CN so o o o © S - P co d -H <=> <^ d d ° ~ Os ro 0 ro so P O O ° £ ^ O O s E Ca c*> CM o o o o o o O ro ro oo O O d d ° ° SO Tf ro ro o 2 P § 2 q S q « ,_: o o O -H o o © © S 2 O m r-s O © © 2 o 00 r-, d © OS _ Tf SO P © © ° OS l O — i CN P d © ° ° © m d d Tf Tt - - , CN E E J -(, c/i PM § § q q ,_; © © 3 _C P °° o o o o © © © © £ - p ^ § § q q r-s © © TJ P <N " O —i © © © o o o 00 00 © o © © CN so © o © © © © Q © © p d o 0 0 o o © © © © © © — i n © © © ro © © o o o o © © £ E O Pi CZ) PH 2 M ca ^ .s u 7 * i > H X> CN ca o «.s • O ° fi-ll £ g o Y\ t £ S .2 fa ca o n > - Cfi A > T3 r o A P P E N D I X V . Plant species used in statistical analyses in this thesis. Plants were identified in the South Okanagan/Similkameen Valleys, British Columbia, 1998. Common Name Taxonomic Name Grasses Bluebunch wheatgrass Agropyron spicatum Cheatgrass Bromus tectorum Junegrass Koeleria macrantha Kentucky bluegrass Poa pratensis Needle-and-thread grass Stipa comata Sand dropseed Sporobolus cryptandrus Sandberg bluegrass Poa secunda Spreading needlegrass Stipa richardsonii Six weeks fescue Vulpia octoflora Forbs Balsamroot Balsamorhiza sagittata Cactus Opuntia fragilis Dandelion Taraxacum officinale Desert-parsley Lomatium spp. Fleabane (daisies) Erigeron spp. Knapweed Centaurea diffusa Knotweed Polygonum majus Lemonweed Lithospermum ruderale Lupine Lupinus sericeus or sulphureus Milk-vetches Astralagus spp. Mulle in Verbascum thapsus Parsnip-flowered buckwheat Eriogonum heracleoides Phlox Phlox longifolia Pussytoes Antennaria spp. Snow buckwheat Eriogonum niveum Woolly plantain Plantago patagonica Yarrow Achillea millefolium Yel low salsify Tragopogon dubius Shrubs B i g sagebrush Artemisia tridentata Pasture sage Artemisia frigida Rabbit-brush Chrysothamnus nauseosus Snowberry Sumphoricarpos albus Three-tip sagebrush Artemisia tripartita 74 


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items