i
Home range size and resource selection by the swamp wallaby, Wallabia bicolor , in a landscape modified by timber harvesting Julian Di Stefano Submitted in total fulfilment of the requirements of the degree of Doctor of Philosophy March 2007 Department of Zoology and School of Forest and Ecosystem Science The University of Melbourne ii
Abstract Timber harvesting results in patches of regenerating forest that are substantially different from surrounding unharvested stands, and provides an opportunity to investigate the effect of habitat change on forest fauna. In this thesis I used timber harvesting as an experimental treatment to investigate the effect of a changed resource base on the home range and resource selection of the swamp wallaby, Wallabia bicolor . I recorded habitat attributes at unharvested control, recently harvested (<12 months old), 5 year old and 10 year old sites. Initially, harvesting removed almost all above ground plant biomass, although the nitrogen and water content of grass on recently harvested sites was relatively high. Five years after harvesting, sites were dominated by densely regenerating 1 3 m tall Eucalyptus seedlings. Relative to unharvested sites, there was substantial lateral cover and values of a forage quality index were high. In contrast, 10 year old sites supported dense, closed stands of 3 6 m tall eucalypt regeneration, had a moderate amount of lateral cover and had low values of the forage quality index. The immediate impact of harvesting operations was to displace wallabies into adjacent unharvested forest, although in most cases, they continued to use parts of their pre harvest home range. Wallabies at recently harvested sites had similar sized home ranges to their counterparts at control locations, and selected against recently harvested patches. A survey of density based on faecal pellets showed that wallabies abandoned harvested areas for 6 8 months and then began to use these sites with increasing frequency, with relative density values increasing to five times that of unharvested control sites after 12 months. A 12 month post harvest browsing assessment showed that the impact of mammalian herbivores on regenerating Eucalyptus seedlings was insubstantial. Relative to unharvested control sites, population density at 5 year old sites was approximately 10 times greater, and diet composition was markedly different, although diet selection was only moderately changed. Negative relationships between the selection and relative availability of most food types (ferns, forbs, monocots, shrubs and trees) indicated that wallabies had a mixed feeding strategy rather than specialising on iii preferred foods. Wallabies showed strong selection for the 5 year old habitat, particularly during diurnal periods. Relative to unharvested controls, home range size was substantially (25% to 45%) smaller at 5 year old sites, and females demonstrated different patterns of diurnal and nocturnal space use. Two indices of body condition (leg length to body weight ratio and percent kidney fat) were similar for wallabies at 5 year old sites and unharvested controls. At 10 year old sites, home range size and population density were similar to values at unharvested control locations. Females selected regenerating areas during the day, and the surrounding unharvested forest at night, and relative to unharvested control sites demonstrated different patterns of diurnal and nocturnal space use. Males also selected forest over 10 year old sites at night, but showed no selection for either habitat during the day. Diurnal habitat selection of both sexes was associated with low visibility, an indicator of dense lateral cover. Nocturnal selection of females was associated with high values of a forage quality index, but this was not the case for males. Both total and diurnal home range size was strongly related to sex (males had larger ranges), and this effect was independent of body weight. Total range size had a strong negative correlation with an index of food and shelter interspersion, while diurnal range size was positively related to visibility, suggesting that smaller diurnal ranges occurred in areas containing dense vegetation. Neither total nor diurnal home range size was correlated with the forage quality index. The small home ranges observed at 5 year old sites were likely to be due to (a) relatively large contiguous patches of dense lateral cover and (b) the interspersion of shelter and adequate food. iv
Declaration This is to certify that (a) the thesis comprises only my original work towards the PhD except where indicated in the Preface (b) due acknowledgement has been made in the text to all other material used (c) the thesis is less than 100 000 words in length exclusive of tables, maps and references Signed……………………………………………………Date….…...... Julian Di Stefano v
Preface The data presented in this thesis is from a research program which, in addition to myself, had three other contributors; Jake Anson, Andrew Greenfield and Mathew Swan (two honours students and a masters student). Although it is possible to present the data separately, the four data sets are complimentary, and in combination tell a more complete story than any individual part. The contribution of each member was as follows. Jake Anson contributed to the data presented in Chapter 2. He collected the pre harvest and during harvest radio tracking data at Impact Site 2, and was involved in wallaby trapping. Andrew Greenfield contributed to the data presented in Chapters 2, 4, 5 and 6. He collected most of the post harvest radio tracking data at Impact Site 2 (referred to as a recently harvested site in Chapters 4, 5 and 6). Matthew Swan contributed to the data presented in Chapters 3, 4, 5 and 6. He collected the radio tracking data at the 10 year old sites, contributed to the collection of the forage availability and quality data in 2006, and was involved in wallaby trapping. I designed and co ordinated the research program, trapped most of the wallabies, collected the radio tracking data from animals at control sites, 5 year old sites and one of the recently harvested sites (Impact Site 1 in Chapter 2), collected forage availability and quality data in both sampling years, performed faecal pellet surveys, and conducted all laboratory work with the exception of the dry matter digestibility procedure. To date, outputs from the other three contributors are: Anson, J.A. (2005). The effects of timber harvesting on home range dynamics and habitat use in the black wallaby ( Wallabia bicolor ). Unpublished Honours Thesis, University of Melbourne. vi
Greenfield, A. (2005). The effects of coupe level timber harvesting on the home range dynamics of the swamp wallaby ( Wallabia bicolor ) in the Pyrenees State Forest, Victoria. Unpublished Research Project Report, Masters of the Environment, University of Melbourne. Swan, M. (2006). Home range and habitat selection of the swamp wallaby ( Wallabia bicolor ) in forest regenerating after timber harvesting. Unpublished Honours Thesis, University of Melbourne. vii
Acknowledgements Thanks to Graeme Coulson, Alan York and Graeme Newell for supervising this project, and to Tina Bell for some unofficial supervision. Two anonymous examiners also contributed to the clarity of the manuscript. Jake Anson, Andrew Greenfield and Matt Swan contributed to the data and to the development of ideas, and Graeme Hepworth guided the data analysis. Particular thanks to Merv Flett and Richard (Feach) Moyle who contributed substantially to Chapter 3, and to Julian Hill who conducted the analysis of dry matter digestibility. The selection of sites and access to historical site information was facilitated by Merv Flett, Bruce McTavish, Fiona Pfeil, Graeme Marshall, Graeme Schultz, Phillip Timpano, Jeremy Allen, Lachlan Spencer, Bob McPhail and Tuesday Phelan. Other assistance was given by John Di Stefano, Ann Berman, Michael Bladen, Uli Mischel, Naomi Davis, Grant Norbury, Bernadette Schmidt, Verity Bristow, Will Macaully, Gina Westhorpe, Josie Lawrence and Amanda Ashton. Funds came from the Holsworth Wildlife Research Endowment, Melbourne University’s Zoology Department and School of Forest and Ecosystem Science, the Victorian Department of Sustainability and Environment (DSE) and the Ecological Society of Australia. Peter Fagg supported my application to DSE for funds and Lynda Poke facilitated financial transactions on a number of occasions. The research was conducted in conjunction with appropriate permits (DSE Research Permit No. 10002779; Melbourne University Faculty of Science Animal Experimentation Committee approval No. 03249). viii
Table of Contents Title page …………………………………………………………………………………………………. i Abstract …………………………………………..………………………………………………………. ii
Declaration .…………………………………………………………………….……………………….. iv Preface …………………………………………………………………………………….………………v
Acknowledgements .. ...………………………………………………………………….……………. vii Table of Contents ……………………………………………………………………………………. viii List of Tables …………………………………………………………………………..……….…...…xii List of Figures ……………………………………………………………………………...…………xiv CHAPTER 1
Introduction ...... 1
Research themes ...... 3 Model species ...... 4 General objectives ...... 6 Thesis structure...... 6 Data analysis philosophy...... 8 Null hypothesis significance testing vs interval estimation...... 8 Transformations ...... 9 Multiple comparison tests ...... 10
CHAPTER 2 Interactions between timber harvesting and swamp wallabies ( Wallabia bicolor ): Space use, density and browsing impact ...... 12
Abstract...... 12 Introduction ...... 14 Methods ...... 16 Study sites...... 16 Wallaby capture and radio tracking ...... 18 ix
Wallaby density ...... 21 Browsing impact...... 22 Data analysis...... 24 Results ...... 26 Space use ...... 26 Wallaby density ...... 29 Browsing impact...... 29 Discussion...... 31 Impact of harvesting on swamp wallabies ...... 31 Potential ecosystem effects...... 34 Browsing impact...... 35 Management implications ...... 35
CHAPTER 3
Summer diet selection of swamp wallabies ( Wallabia bicolor ) in a harvested landscape: feeding strategies under conditions of changed food availability ...... 37
Abstract...... 37 Introduction ...... 39 Methods ...... 41 Study site...... 41 Collection of stomach samples...... 41 Diet analysis...... 42 Forage availability...... 43 Forage quality ...... 43 Data analysis...... 44 Results ...... 46 Discussion...... 54
x
CHAPTER 4
Effect of habitat type, sex and diel period on the space use of swamp wallabies (Wallabia bicolor )...... 58
Abstract...... 58 Introduction ...... 59 Methods ...... 60 Study site...... 60 Experimental design ...... 60 Wallaby capture and radio tracking ...... 62
Data analysis ...... 63 Results ...... 63 Discussion...... 66
CHAPTER 5
Habitat selection by the swamp wallaby ( Wallabia bicolor ) in relation to diel period, food and shelter ...... 69
Abstract...... 69 Introduction ...... 71 Methods ...... 73 Study site...... 73 Wallaby capture and radio tracking ...... 73 Habitat comparisons ...... 74 Habitat selection...... 76 Association between habitat selection and habitat quality ...... 77 Data analysis...... 77 Results ...... 78 Habitat comparisons ...... 78 Habitat selection...... 83 Association between habitat selection and habitat quality ...... 85 Discussion...... 86 xi
CHAPTER 6
The influence of resources on the home range size of the swamp wallaby ( Wallabia bicolor )...... 90
Abstract...... 90 Introduction ...... 92 Methods ...... 95 Study site...... 95 Experimental design ...... 95 Wallaby capture and radio tracking ...... 95 Factors affecting home range size...... 95 Body condition indices ...... 97 Data analysis...... 97 Results ...... 99 Treatment effects ...... 99 Factors affecting home range size...... 102 Body condition indices ...... 106
Discussion...... 106 Factors affecting home range size...... 108 Body condition...... 111 The maintenance of small home ranges ...... 111
CHAPTER 7
Synthesis ...... 113
References ...... 117
Appendix 1 ...... 141 xii
List of Tables Table 2.1. Vegetation, climate and physical geography of the study sites (pers. obs.; BOM 2006; DSE 2006; LCC 1973; LCC 1978)...... 17 Table 2.2. Percentage of wallaby locations within the area disturbed by the harvesting operation. Before = before harvest, During = between harvest onset and burning of logging debris and After = after burning. At Impact Site 2, a substantial area was burnt but not harvested so numbers in parenthesis refer to locations in either harvested or burnt areas...... 26 Table 3.1. Use, availability and selection ( B) of the five food types at unharvested forest ( n = 9) and 5 year old ( n = 10) sites. %used refers to the percentage of food types in stomach contents, %available refers to the percentage of food types derived from the field survey and B is a standardised index of diet selection (see text for details). Errors are 95% confidence intervals, or confidence intervals of the difference...... 47 Table 3.2. Linear correlations (Pearson’s correlation coefficient r) between food types and the ordination axis for both diet composition and selection. Food types with correlations < 0.70 are not shown...... 50 Table 3.3. Univariate models of frequency dependence for each of the 5 food types. For each food type, the model is log 10 B = a + β(relative abundance), where B is an index of selection. Coefficients ( β) have been standardised for comparison with other analysis. 95% Low. and 95% Upp. are the 95% lower and upper confidence limits of β. n = 22 in all cases...... 51 Table 3.4. Multivariate models for each food type comparing the effect of availability and forage quality on diet selection. For each food type the model is log 10 B = a + β(relative abundance) + β(nitrogen) + β(water) + β(digestibility) where B is an index of selection. Coefficients ( β) have been standardised. 95% Low. and 95% Upp. are the 95% lower and upper confidence limits of β...... 52 Table 3.5. Multivariate linear models for frequency dependent selection constructed using the technique of Manly (1973) – see text for details. For each model, the response variable is the selection index B and the predictors are the relative availability of each food type (the Ps). Due to the high correlation between forb and tree relative availability (r = 0.71) these two models were re run using traditional multiple regression after excluding the highly correlated predictor. 95% Low. and 95% Upp. are the 95% lower and upper confidence limits of the coefficient, and are only shown for coefficients with a non zero effect...... 53 Table 4.1. Subjectively estimated accuracy of radio tracking locations. Values are the percentage of locations (± 95% confidence interval) relating to a 1 – 5 scale of spatial accuracy; 1 = within 5 m of exact location, 2 = 5 – 25 m, 3 = 25 – 100 m, 4 = 100 – 200 m and 5 = >200 m, or when a signal could not be detected...... 62 Table 5.1. Results of the MRPP analysis for selected comparisons. Unharvested 5 yr and Unharvested 10 yr refer to unharvested forest adjacent to 5 and 10 year old sites respectively, and Unharvested All refers to pooled data from these two groups. The xiii effect size measure ( A) is a better reflection of the difference between groups than the P value as it is less affected by sample size...... 79 Table 5.2. Variables with a strong ( r ≥0.70) linear correlation with the ordination axes of Figure 5.2. FQI is a forage quality index (see text for derivation)...... 80 Table 5.3. Water content, nitrogen content and dry matter digestibility of plant functional groups at the two recently harvested sites. Ferns and shrubs are not included as they were rarely present. Values are presented for the harvested area and for an adjacent patch of unharvested forest. NP = not present...... 82 Table 5.4. Habitat selection at the population level based on faecal pellet counts. A value of 1 for the selection index ŵ is equivalent to using a habitat in proportion to its availability, while values >1 and <1 represent selection of habitats more and less than expected in relation to their availability...... 83 Table 6.1 . Model selection statistics for (A) the total home range data set and (B) the diurnal home range data set. K is the number of parameters in the model. The model with the smallest AIC c value fits the data best, but models within about 2 AIC c units of the best also have substantial support. The Akaike weights, wi, are interpreted approximately as the probability that their associated model is the best. The first four (A) and three (B) models provide an approximate 95% confidence set for the best model (∑wi ≥0.95). S = Sex; V = visibility; FQI = the forage quality index and D = the index of food and shelter interspersion. See text for further details...... 103 Table 6.2. Averaged standardised coefficients for predictors in the best model for both total and diurnal range size. 95% Lower and Upper refer to the 95% lower and upper confidence limits respectively...... 104 Table A1. 95% fixed kernel home range estimates for wallabies used in Chapter 2. . 141 Table A2 . Raw home range data used in chapters 4 and 6. The control male marked with a * was removed from all analyses in Chapter 6 (see Data analysis section). Some diurnal home ranges were not calculated due to an inadequate number of locations... 142 xiv
List of Figures Figure 1.1 . Disturbance events such as timber harvesting affect herbivorous mammals directly, and indirectly through their impact on ecosystem structure and function. Herbivorous mammals can also effect commercial timber production by damaging regenerating trees...... 3 Figure 2.1. Top: Map of Australia showing the general location of study forests within the State of Victoria (shaded). Bottom: Pyrenees State Forest showing the spatial arrangement of experimental units used in the analysis of wallaby space use...... 16 Figure 2.2. The effect of harvesting on (A) 95% range overlap, (B) 50% range overlap, (C) 95% range size and (D) 50% range size. Overlap is calculated as the percentage of the during or after range overlapping the before range. M & F refer to combined male and female data while F refers to female data only. Error bars are 95% confidence intervals...... 28 Figure 2.3. Effect of harvesting on relative wallaby density. Pre harvest data March to September 2004, post harvest data July 2005 to July 2006. Error bars are 95% confidence intervals...... 30 Figure 2.4. Relationships between adjusted stocking and (A) percent biomass removed, (B) adjusted seedling density and (C) wallaby density. Both stocking and seedling density values have been adjusted for browsing impact...... 32 Figure 3.1. Selection of plant groups by wallabies living at (A) unharvested control sites ( n = 9) and (B) 5 year old regenerating sites ( n = 10). Effects are expressed as the first named plant group minus the second named plant group, so positive values mean the first named group is selected to a greater degree. Errors are 95% confidence intervals of the difference. Mono. = Monocot, a combined grass and sedge category. 48 Figure 3.2. Ordination diagrams representing (A) the use (consumption) and (B) the selection of plant groups by wallabies. Vectors (solid lines) represent the strength and direction of the linear correlation between individual food types and the ordination axis (see Table 3.2 for coefficients). The labels 1 and 2 in diagram (B) indicate outlying points (see text for details). Diagram (A): Stress = 14.8, MRPP: A = 0.20; P <0.001. Diagram (B): Stress = 10.5, MRPP: A = 0.05, P = 0.04...... 49 Figure 4.1. The Pyrenees State Forest, western Victoria, showing the spatial arrangement of the study sites...... 61 Figure 4.2. The degree to which home range size increases with the addition of nocturnal data for each of four habitat types. Data are expressed as a percentage of diurnal range size to standardize results for animals that had markedly range sizes. * Values may be inaccurate due to small sample sizes ( n = 2 in both cases)...... 64 Figure 4.3. Pair wise comparisons between each class of harvested site and the controls. Point estimates are calculated as the value of the first listed habitat minus the control value, so a positive number represents an increase relative to the control. Error xv bars are 95% confidence intervals. * Values may be inaccurate due to small sample sizes...... 65 Figure 4.4. Relationship between home range size increase and body weight. The fitted line (solid) and its 95% confidence interval (dashed) describe the relationship for males; y = 79.2 – 3.5 weight, Adj. r2 = 0.23. The relationship does not exist for females (Adj. r2 = 0)...... 65 Figure 5.1. Pyrenees State Forest showing the spatial arrangement of sites and the polygon used to define habitat availability for the population level analysis...... 74 Figure 5.2. Ordination of habitat quality variables collected at unharvested forest adjacent to 5 year old regeneration, unharvested forest adjacent to 10 year old regeneration, 5 year old sites and 10 year old sites. Lines (vectors) represent the direction and strength of the four variables with the strongest linear correlations with the ordination axis. FQI is a food quality index. Stress = 15.8; explained variation from distance matrix = 83.1%...... 79 Figure 5.3. Values of abundance (A) nitrogen content (B) water content (C) and dry matter digestibility (D) for six plant groups in three habitats. Errors are 95% confidence intervals...... 81 Figure 5.4. Values of the forage quality index for male and female wallabies at 5 and 10 year old sites and adjacent stands of unharvested forest. Errors are 95% confidence intervals...... 82 Figure 5.5. Habitat selection of female (A) and male (B) swamp wallabies. A value >0 on the Y axis means that the first listed habitat was selected over the second listed habitat, and larger values mean stronger selection. On the X axis, 5 and 10 refer to 5 year old and 10 year old regenerating sites, and Forest refers to the unharvested forest surrounding the regenerating areas. Error bars are 95% bootstrapped confidence intervals...... 84 Figure 5.6. Measures of visibility (A, B) and forage quality (C, D) during diurnal and nocturnal periods for sites that were selected more and less than expected by chance. A habitat was selected more than expected by chance when B > B/n, where n is the number of available habitat types. Error bars are 95% confidence intervals. Numbers above the bars in graphs A and B are sample sizes, and are the same for graphs C and D...... 85 Figure 6.1. The effect of harvesting treatment on home range size for (A) total ranges and (B) diurnal ranges. Data have been back transformed for graphical presentation. Errors are 95% confidence intervals...... 100 Figure 6.2. Home range size comparisons between control sites and recently harvested, 5 year old and 10 year old sites for (A) total ranges and (B) diurnal ranges. Mean differences are calculated as the control mean minus the treatment mean, so a positive value on the X axis means that home range size at control sites is larger. Errors are 95% bootstrapped confidence intervals and a * indicates a comparison where the sample size in the treatment group was either 2 or 3...... 101 xvi
Figure 6.3. Relationship between total home range size and wallaby density for males and females. Y axis values have been back transformed for graphical presentation. 2 Males: log 10 home range size = 1.77 – 0.41 density, Adj. r = 47.8%. Females: log 10 home range size = 1.41 – 0.43 density, Adj. r2 = 79.6%. Solid and dashed lines are lines of best fit and 95% confidence intervals respectively...... 105 Figure 6.4. Relationship between total home range size and body weight for males and females. Y axis values have been back transformed for graphical presentation. Adj. r2 values are 6.8% for males and 0% for females, although the value for males also becomes 0 when the points corresponding to the two lightest wallabies are removed. Solid lines are lines of best fit and dashed lines are 95% confidence intervals. For clarity, confidence intervals are not shown for the female data...... 105 Figure 6.5. Leg length to body weight ratio for males and females at control and 5 year old sites. Errors are 95% confidence intervals...... 107 Figure 6.6. Percent kidney fat for males and females at control and 5 year old sites. The value for males at control sites is based on two wallabies, so should be treated with caution. Errors are 95% confidence intervals...... 107 1
CHAPTER 1
Introduction
Disturbance events play an important role in ecosystem regulation, and are known to influence species composition, plant succession and regeneration, the distribution and abundance of animal populations and nutrient concentration in a range of ecosystem types (Bormann and Likens 1979; Connell 1978; Hobbs and Huenneke 1992; Lugo et al. 2000; Moloney and Levin 1996; Pickett and White 1985). In forest ecosystems, natural disturbances such as wildfire and storms act as major regulating forces (Attiwill 1994a; Attiwill 1994b; Lugo 2000; Ryan 2002), and in more recent times human disturbances such as urbanisation, land clearing and timber harvesting have had a marked impact on forest extent and structure, and on the distribution and abundance of forest dwelling organisms (Abrams 2003; Dale et al. 2000; Gaston et al. 2003; Thompson et al. 2003; Wilson and Friend 1999). Timber harvesting is a major land use in many countries, and public concern over its acceptability and sustainability has increased in recent times (Chase 1995). Like any major disturbance event, native forest harvesting will have immediate and direct impacts on local animal populations. Although the direct impacts of harvesting have been infrequently studied, its effects seem to depend on the mobility of the species in question, their fidelity to pre harvest home ranges and their ability to acquire resources in the surrounding unharvested forest. For example, harvesting appeared to have little effect on the space use of mobile, non territorial female white tailed deer, Odocoileus virginianus (Campbell et al. 2004). In contrast, most hollow dependent greater gliders (Schoinobates volans ) did not disperse into adjacent forested areas and died within a week of tree fall (Tyndale Biscoe and Smith 1969). Timber harvesting may also alter the dynamics of ecological communities in complex ways by changing the structural characteristics of forest landscapes (Franklin and Forman 1987). It is argued that harvesting reduces structural complexity (Dale et al. 2000; Lindenmayer and Franklin 1997; Lindenmayer and McCarthy 2002), causing key 2 resources to become limiting for a range of forest fauna (Thompson et al. 2003). In Australia, for example, some arboreal mammals have been adversely affected by clearfelling (Lindenmayer et al. 1991; Lindenmayer and Franklin 1997) but species responses are likely to vary with harvesting method (Sullivan and Sullivan 2001), as well as the spatial and temporal extent of particular investigations. Fuller and DeStefano (2003) show that early successional communities resulting from harvesting provide abundant resources for a range of mammals, and constitute an important habitat element for these species and their predators. One group that is generally favoured by commercial timber production is medium to large generalist browsing mammals. For example, early successional patches within forest landscapes have resulted in the interspersion of shelter and food resources, contributing to high density deer populations throughout the world (Côté et al. 2004 and references therein). Elevated herbivore densities can substantially alter plant community composition (Horsley et al. 2003) and may have secondary impacts on other groups of organisms such as insects (Knops et al. 1999; Murdoch et al. 1972) and small mammals (Flowerdew and Ellwood 2001; Moser and Witmer 2000). In addition, herbivores may alter ecosystem processes through inputs of faeces and urine (Hobbs 1996), or through interactions between selective foliage consumption, litter quality, below ground plant responses and the abundance of soil micro organisms (Bardgett et al. 1998; Wardle et al. 2002). As well as incurring impacts from harvesting, some forest fauna affect commercial timber production by browsing regenerating seedlings. Throughout the world, mammalian browsers are known to have adverse effects on the survival, growth rates and form of regenerating trees in both commercial and non commercial forests (Gill 1992; Reimoser and Gossow 1996; Rooney 2001; Welch et al. 1992; Zamora et al. 2001). In southeastern Australia, mammalian browsing has also been shown to be a problem, particularly in plantation forestry (Bulinski and McArthur 1999; Bulinski and McArthur 2003), but also at native forest sites (Di Stefano 2003; Di Stefano 2005; Sebire 2001). Although more research is required, it seems likely that economic losses due to browsing are substantial in some settings (Montague 1996; Neilsen and Wilkinson 1995). 3
Harvesting may have direct and immediate impacts on forest fauna and it may have more subtle indirect effects by altering the resource base at a variety of spatial scales. Interactions between above and below ground processes have also been observed. Conversely, some species of forest fauna, notably herbivorous mammals, can have adverse effects on commercial timber production by browsing regenerating seedlings. These multi directional relationship are summarised in Figure 1.1.
Direct impacts
FOREST HERBIVOROUS ECOSYSTEM Habitat use COMMERCIAL MAMMALS Movements Indirect Food quality TIMBER PROD . Indirect Foraging behav . impacts Food quantity impacts Diet Selection (Human disturbance) Structural complexity Reproduction Nutrient cycling
Figure 1.1. Disturbance events such as timber harvesting affect herbivorous mammals directly, and indirectly through their impact on ecosystem structure and function. Herbivorous mammals can also effect commercial timber production by damaging regenerating trees. Research themes The relationships outlined in Figure 1.1 form the basis for two interrelated themes: 1. What effect does habitat alteration caused by timber harvesting have on herbivorous mammals? 2. What impact do herbivorous mammals have on commercial forestry operations when they consume regenerating seedlings? These themes form the basis of this thesis. The primary focus is on Theme one which is important from at least two perspectives. First, governments promote timber harvesting as an ecologically sustainable pursuit and acknowledge that harvested forests on public land should be managed for conservation and intrinsic as well as commercial values (e.g. Australian Government 2006; Squire et al. 1991b). To manage forests for these 4 competing values requires, amongst other things, an understanding of harvesting impacts on forest wildlife (Lindenmayer 1994). In addition, timber harvesting is a major disturbance event and thus provides an opportunity to test a related body of ecological theory. For example, harvesting results in the spatial redistribution of shelter and food resources for a number of species, and provides an opportunity to test hypotheses predicting relationships between resource availability and space use. If viewed as a landscape scale experimental treatment (Bennett and Adams 2004), harvesting provides a manipulative intervention at a spatial scale applicable to free ranging herbivores. Although there are a number of analytical challenges associated with such ‘natural experiments’ (Hobbs 2003) and they may sometimes lack elements fundamental to good experimental design (for example, random allocation of treatments to sites; Johnson 2002), using harvesting operations as experimental treatments provides a sound framework for wildlife research in managed forests (Lindenmayer 1999). Theme two is a secondary focus of this thesis and has strong management application. One aspect of sustainable timber harvesting is adequate seedling regeneration. If browsing by herbivorous mammals contributes to regeneration failure, strategies to reduce browsing impact need to be developed. Browsing by mammals has been shown to cause substantial problems for eucalypt regeneration in a number of commercially managed Victorian native forest districts (Sebire 2001) and data describing the impact browsing on regenerating seedlings, as well as the relationship between browser abundance and damage (Bulinski 2000), are required to help manage the problem effectively. Model species The swamp wallaby, Wallabia bicolor , a medium sized macropodid marsupial (Plate 1.1), was chosen as a model species to investigate the research themes and general objectives outlined above. Although it is relatively common within the harvested landscapes of southeastern Australia, the impact of harvesting on this species has not been studied in detail. In general, research on harvesting impacts in southeastern Australia have focused on arboreal marsupials (Gibbons and Lindenmayer 1996; 5
Kavanagh 2000; Lindenmayer et al. 1991; Lindenmayer and Franklin 1997; Tyndale Biscoe and Smith 1969) and birds (Abbott et al. 2003; Craig and Roberts 2005; Wardell Johnson and Williams 2000; Williams et al. 2001), and very little information exists about ground dwelling mammals. Based on past research (Floyd 1980; Hill and Phinn 1993; Lunney and O'Connell 1988; Ramsey and Wilson 1997; Troy et al. 1992), the swamp wallaby is expected to show behavioural responses to the structural changes caused by harvesting, thus making it a good candidate for study. In addition, swamp wallabies are believed to cause substantial impact to eucalypt seedlings regenerating after native forest timber harvesting in Victoria (Di Stefano 2005; Sebire 2001) and thus provide a good model to study browsing impacts of herbivorous mammals in a commercial forestry environment.
Plate 1.1. A male swamp wallaby, Wallabia bicolor , typical of the individuals used in this study. 6
General objectives Using the themes outlined above as a conceptual framework and the swamp wallaby as a model species, the research reported in this thesis has three general objectives. 1. Quantify the impact of harvesting induced habitat change on swamp wallaby behaviour in terms of (a) space use, (b) density, (c) diet selection and (d) habitat selection. 2. Provide a test of the theory that home range size is negatively related to resource availability. 3. Quantify the impact of mammalian herbivores (predominately swamp wallabies) on Eucalyptus seedlings regenerating after harvesting. Specific hypothesis (predictions) are outlined below and described in more detail in the introduction to each chapter. Thesis structure
To facilitate a ‘chapters as papers’ structure, this thesis does not contain an initial chapter outlining the study site, study species and common methodologies. New methods and relevant aspects of the study site are described sequentially in each chapter and reference is made to previous chapters when necessary. Nevertheless, this structure results in some degree of overlap, particularly regarding the description of timber harvesting and the study species in the introduction or methods section of each chapter. In the following chapter (Chapter 2) I use Multiple Before After Control Impact (MBACI) designs to quantify the immediate impact of timber harvesting on individual wallabies (objective 1a) and the effect of harvesting on relative wallaby abundance (objective 1b). This chapter also quantifies mammalian browsing impact on regenerating areas about 12 months after harvesting (objective 3) and relates pre and post harvest swamp wallaby abundance to browsing impact. Specifically, I make two 7 predictions: 1. that the immediate impact of harvesting on swamp wallaby space use will be minor, and 2. increased food and shelter resources on regenerating areas in the 12 months post harvest will facilitate an increase in wallaby density. In Chapter 3 I quantify diet selection at sites with different relative availabilities of food types (objective 1c). I test the prediction that the relationship between the selection of a food type and its relative availability will be negative (negatively frequency dependence), and reconcile the results against two alternative feeding strategies: specialization and mixed feeding. In Chapter 4 I address objective 1a (space use) in the context of differing diel behavioural patterns. Unharvested control sites are structurally homogeneous while regenerating sites are heterogeneous, particularly with respect to lateral cover. Because swamp wallabies are expected to use lateral cover move heavily during the day than at night, I predict that diurnal and nocturnal space use will differ for individuals with access to regenerating sites but not for individuals living in the unharvested forest. Habitat selection and home range size are quantified in Chapters 5 and 6. Based on past work I expect that structural changes resulting from timber harvesting will create attractive habitats for swamp wallabies. In Chapter 5 I test this hypothesis by determining habitat selection rankings for unharvested forest, recently harvested forest and both 5 and 10 year old regenerating areas (objective 1d). I expect selection for densely vegetated habitats during diurnal periods, but do not necessarily expect this pattern to be replicated at night. I also predict that diurnal selection will be related to shelter while nocturnal selection will be related to forage quality. In Chapter 6 I test the theory that home range size is negatively correlated with resource availability (objectives 1a and 2). Resource availability within individual home ranges is quantified at unharvested sites, recently harvested sites and 5 and 10 year old regenerating areas using indicators of shelter, forage quality and the interspersion of the two. These variables, in addition to sex, are used to develop a set of competing models relating home range size to specific resources. I predict that shelter resources will be more strongly correlated with diurnal range size than with total range size, but that this pattern will be reversed with respect to food resources. In addition, I compare the body 8 condition of wallabies living at unharvested control and 5 year old sites, and consider the results in light of two alternative theoretical expectations. Chapter 7 provides a synthesis of the main findings, summarises the hypotheses generated by this work and suggests a framework within which some of these can be tested. Data analysis philosophy In this thesis my analytical techniques depart from convention on a number of occasions. Here I outline the issues and the rationale for my approach. Null hypothesis significance testing vs interval estimation
Despite the availability of alternative techniques (e.g. Altman et al. 2000; Hobbs and Hilborn 2006), null hypothesis significance testing and the use of P values to summarise statistical results has been the major analytical paradigm in ecology for many years (Anderson et al. 2000). Nevertheless, it has been argued numerous times, and from a number of different perspectives, that the practice of testing zero null hypothesis using a pre defined probability value (traditionally 0.05) to make dichotomous decisions is not the best way to derive meaning from data (e.g. Anderson et al. 2000; Berger and Sellke 1987; Cohen 1994; Ellison 2004; Johnson 1999; Rozeboom 1960; Stewart Oaten et al. 1992; Yoccoz 1991). Although some authors advocate a wholesale change in analytical philosophy (Ellison 1996; Ellison 2004; Wade 2000), others suggest that conventional statistical techniques still have much to offer, but the way analyses are presented and interpreted needs to be changed (Cumming and Finch 2005; Fidler et al. 2006). This is the view to which I subscribe. In most scientific studies, researchers should aim to measure the size of effects and estimate the associated uncertainty (Johnson 1999; Nelder 1999; Yoccoz 1991), and one way to achieve this is to use a practice known as interval estimation. The focus of interval estimation is to estimate both the magnitude of an effect and its precision, and although P values are not inconsistent with this approach, they usually do not provide 9 additional information (Johnson 1999). In this thesis, effects of interest are often the difference between two means, and in these cases P values are not used. In other cases, effect size measures are more difficult to interpret (e.g. the effect size A in a Multi response Permutation Procedure, or interaction effects in ANOVA), so P values are presented as an additional aid to interpretation. Whenever possible, I calculate 95% confidence intervals around observed effects as an estimate of their precision, although the choice of this confidence level is an arbitrary one and arguments can be made for using other values. The technical interpretation of confidence intervals is discussed elsewhere (e.g. Smithson 2003), but in colloquial terms, 95% intervals constitute a range within which the estimated parameter may plausibly lie (Hoenig and Heisey 2001). Although there are a number of problems associated with the interpretation of confidence intervals (Belia et al. 2005), these are outweighed by the benefits of their use (Cumming and Finch 2005). I use estimates of effect size and associated 95% confidence to make reasoned judgements about the importance of observed effects. In no case do I use P values to make dichotomous decisions. Transformations
The use of parametric statistical procedures often require data transformation to satisfy certain analytical assumptions (Sokal and Rohlf 1995). Transformations, however, often change the relative difference between data values and thus can lead to the loss of important biological information (Keough and Mapstone 1995). An example appears in Chapter 2 where a number of large (and biologically important) wallaby density values at impact sites become much smaller relative to control values after a log transformation. As a consequence, the large effect observed in the raw data were much smaller and appeared insubstantial when viewed in the transformed scale. Interpreting biologically important effects can also be difficult if the new measurement scale is not intuitively meaningful (Jaccard and Guilamo Ramos 2002; Stewart Oaten et al. 1992). For example, the biological relevance of an effect may be obvious using familiar units, but may be difficult to interpret if those units are transformed. For estimated means and their associated confidence intervals, this second problem can be overcome by back transforming results into the original scale of measurement, but this strategy is inappropriate for mean differences, as a confidence interval of a difference cannot be 10 back transformed. If the focus of data analysis is on the presentation of effects and their associated confidence intervals, this constitutes a substantial problem. One solution is to use bootstrapping techniques (Efron and Tibshirani 1993) to estimate mean differences and their confidence intervals. Bootstrapping requires no assumptions about the distribution of the data, and thus results can be computed in the original measurement scale. In this thesis I present bootstrapped effects and 95% confidence intervals on a number of occasions, most commonly to assess differences between means after an initial omnibus analysis. However, the validity of bootstrapped results depends on the sample data being representative of the data from the wider population, and, due to small sample sizes, this assumption is clearly not met on a number of occasions. Small sample sizes weaken inference regardless of the analytical technique, and I alert readers to this issue when it occurs. Multiple comparison tests
An issue that has received considerable attention over the years is the adjustment of P values (and in some cases confidence intervals) when multiple, related statistical tests are performed. It is argued that multiple tests within the same ‘family’ (for example, the pairwise comparison of means after an initial Analysis of Variance) require a correction so that the Type I error rate is reduced for the whole set (Ludbrook 1998), and numerous multiple comparison procedures are available to achieve this objective (reviewed by Day and Quinn 1989). Others, however, argue that multiple comparison tests are too conservative in some situations (Moran 2003; Roback and Askins 2005), or that they are not necessary at all (Nelder 1971; Perneger 1998; Rothman 1990). In reality, the decision whether or not to correct for multiple tests is a judgement rather than a statistical rule (Stewart Oaten 1995) and each researcher needs to make up their own mind. Although there are a number of mathematical, logical and practical objections to multiple comparison procedures (e.g. Moran 2003; Stewart Oaten 1995), I believe the issue is often made more complex that it needs to be. As O’Neill and Wetherill (1971) observed, there are three situations that can arise when making multiple comparisons; 11 observed effects are either large and clearly important, small and clearly unimportant, or intermediate (or large with large errors) where their importance is unclear. In the first two cases, the inference will be the same regardless of the statistical procedure used. In the third situation, the choice of procedure will make a difference to the P value (and possibly to the statistical significance of the result), but, as argued previously, P values are rarely informative. In these cases, presentation of the observed effects and their associated confidence intervals will demonstrate the uncertainty of the inference. Confidence intervals will be somewhat narrower or wider depending on the procedure used, but the general conclusion is unlikely to change. In this thesis I do not make corrections for multiple tests. I view each comparison as an independent analysis and compute 95% confidence intervals in each case. 12
CHAPTER 2
Interactions between timber harvesting and swamp wallabies (Wallabia bicolor ): Space use, density and browsing impact
Abstract I used timber harvesting in native Eucalyptus forests as an experimental treatment to study its effect on the space use and density of swamp wallabies ( Wallabia bicolor ), and on the impact of herbivorous mammals on post harvest tree regeneration. The space use and density studies used a Multiple Before After Control Impact (MBACI) design to compare changes before and after (and in some cases before and during) harvesting between unharvested control and harvested impact locations. The impact of harvesting on wallaby space use was quantified separately at two harvested sites in terms of home and core range size and overlap (95% and 50% fixed kernels), and the shift in geographic centre of location (GCL). The most obvious response to harvesting was a substantial shift in core use area and, in some cases, a large (>100%) increase in home range size. Relative to unharvested controls, GCLs shifted further at one harvested site but not at the other. Home range overlap tended to be similar between control and harvested sites indicating that harvesting had a minimal impact on the overall use of space. A year after harvesting, wallaby density was about 5 times greater at harvested sites than at control sites. This overall increase was characterised by an almost complete abandonment of harvested areas for the first 8 10 months and then a rapid influx of animals after this time. Browsing impact on 12 month old Eucalyptus seedlings (% biomass removed) ranged from 1.0 to 11.2% but was insubstantial for coppice (0.4 to 0.9%). The percentage of severely damaged seedlings ranged from 0 to 12.9%. Eucalypt regeneration success 13 was strongly or moderately related to browsing impact, seedling density and both pre and post harvest wallaby density. The reduction in stocking attributable to severe browsing ranged from 0 to 3% indicating that browsing impact had little effect on regeneration success. The results are discussed with reference to (a) the potential effect of high herbivore densities on ecosystem processes and (b) effective monitoring of browsing damage in commercially harvested native forests. 14
Introduction Sustainable management of forests used for commercial timber harvesting requires, amongst other things, an understanding of harvesting impacts on forest wildlife (Lindenmayer 1994; Simberloff 1999). An important group of forest fauna are herbivorous ground dwelling mammals who often benefit from timber harvesting in the short to medium term. Although effects can be mediated by silvicultural practices (Reimoser and Gossow 1996), harvesting generally creates patches of early successional forest adjacent to mature stands, providing high quality foraging and shelter environments for many species (Bobek et al. 1984; Fuller and Gill 2001; le Mar and McArthur 2005). Côté et al. (2004) suggested that habitat enhancement resulting from commercial timber production is one of the reasons for high density deer populations around the world. A corollary of the enhanced habitat quality afforded by harvesting is that mammalian herbivores often consume regenerating seedlings. From a commercial perspective, this can have adverse effects on the survival, growth rates and form of regenerating trees in both commercial and non commercial forests (Bulinski 1999; Bulinski and McArthur 1999; Bulinski and McArthur 2003; Di Stefano 2005; Gill 1992; Reimoser and Gossow 1996; Rooney 2001; Welch et al. 1992; Zamora et al. 2001). In addition, browsing by herbivores may alter plant community composition (Horsley et al. 2003), have secondary impacts on other groups of organisms (Flowerdew and Ellwood 2001; Moser and Witmer 2000) and can influence ecosystem processes through inputs of dung and urine (Hobbs 1996), or through interactions between selective foliage consumption, litter quality, below ground plant responses and the abundance of soil micro organisms (Bardgett et al. 1998; Wardle et al. 2002). Although retrospective studies of harvesting impacts on herbivores are common (e.g. Chubbs et al. 1993; le Mar and McArthur 2005; St Louis et al. 2000), I am aware of only one study containing pre, during and post harvest data at both control and impact locations (Campbell et al. 2004), although several others used similar designs to quantify the impact of other disturbances on herbivore behaviour (Cimino and Lovari 2003; Newell 1999). Collecting data before, during and after harvesting enables the study of immediate behavioural responses to habitat alteration, and the fate of 15 individuals to be quantified (Newell 1999). In addition, the use of Before After Control Impact (BACI, MBACI, etc.) designs provides a better basis for inferring impacts than the traditional retrospective approach (Downes et al. 2002; Keough and Mapstone 1995). Despite their strength, BACI designs are infrequently used in studies of harvesting impacts on vertebrates. In this study, my objective was to investigate interactions between commercial timber production and the swamp wallaby ( Wallabia bicolor ), a medium sized ground dwelling generalist herbivore that is widely distributed throughout the native forests of southern and eastern Australia. Past research on harvesting impacts in these forests have focused on arboreal mammals (Gibbons and Lindenmayer 1996; Kavanagh 2000; Lindenmayer et al. 1991; Lindenmayer and Franklin 1997; Tyndale Biscoe and Smith 1969), and little information on ground dwelling species is available. In addition, swamp wallabies contribute to locally severe browsing damage (Sebire 2001) and appear to favour densely regenerating areas over surrounding unharvested forest (Di Stefano 2005). Quantifying the interactions between this species and both its pre and post harvest environment may facilitate the development of browsing reduction plans (Partl et al. 2002; Reimoser and Gossow 1996). Specifically, I make two predictions related to the impact of harvesting on wallabies. Because swamp wallabies are mobile, and minor relocations of their home range are not expected to be constrained by resource availability, I expect harvesting to have little impact on space use (prediction 1), but expect that wallaby density will increase in the first year after harvesting due to increased food and shelter resources on regenerating areas (prediction 2). In addition, I relate browsing impact to the success of regenerating Eucalyptus seedlings using a local regeneration standard (stocking the proportion of plots containing a viable seedling; Dignan and Fagg 1997). Assessing browsing impact in relation to regeneration standards enables forest managers to judge if browsing impact is acceptable, or if intervention is needed (Reimoser et al. 1999). 16
Methods Study sites I collected data from the Pyrenees, Mt. Disappointment and Black Range State Forests in Victoria, Australia (Figure 2.1). All are relatively open, dry sclerophyll forests dominated by Eucalyptus spp . The Pyrenees is the driest, most open and least productive and has a seasonally abundant forb community, while Mt. Disappointment and the Black Range contain taller trees and a denser shrub layer facilitated by higher rainfall and more fertile soils. Additional details regarding the dominant plant species and physical geography of the study areas within these forests are given in Table 2.1. N Pyrenees Mt Disappointment Black Range Pyrenees State Pyrenees Forest State Forest Control Recently Harvested 5 km Figure 2.1. Top: Map of Australia showing the general location of study forests within the State of Victoria (shaded). Bottom: Pyrenees State Forest showing the spatial arrangement of experimental units used in the analysis of wallaby space use. 17
Table 2.1. Vegetation, climate and physical geography of the study sites (pers. obs.; BOM 2006; DSE 2006; LCC 1973; LCC 1978). Site characteristics Forest structure Dominant understorey plants Pyrenees State Forest, west central Victoria Rainfall: 600 700mm/year Dry, open forest Virtually absent middlestorey except for silver wattle Soils: Stony red duplex dominated by messmate (Acacia dealbata ) and cherry ballart ( Exocarpos Overstorey Ht: 15 28m (Eucalyptus cupressiformis ). Sparse shrub layer includes common Crown cover: 70 – 84% obliqua )/blue gum ( E. heath ( Epacris impressa ), gorse bitter pea ( Davisia Elevation: 500 – 700m asl globulus bicostata ) or ulicifolia ), common cassinia ( Cassinia aculeata ) and blue gum/messmate prickly wattle ( A. paradoxa ). Ground layer dominated by associations. Red austral bracken ( Pteridium esculentum ) and grasses ironbark ( E. tricarpa ), including common tussock grass ( Poa labillarderi ) and red stringybark ( E. silvertop wallaby grass ( Joycea pallida ). Supports a macrorhyncha ) yellow seasonally abundant forb community including soft crane’s box ( E. melliodora ) and bill ( Geranium potentilloides ), kidney weed ( Dichondra candlebark ( E. rubida ) repens ), creeping oxalis ( Oxalis corniculata) and bidgee occasionally present. widgee ( Acaena novae zelandiae ). Mt. Disappointment State Forest, central Victoria Rainfall: >1000mm/year Dry forest dominated by Virtually absent middlestorey except for silver wattle. Soils: Red friable earths messmate/mountain grey Moderate shrub layer includes common cassinia, prickly Overstorey Ht: 28 – 34m gum ( E. cephellocarpa ) current bush ( Coprosma quadrifida ), hop goodenia Crown cover: 70 – 84% associations. (Goodenia ovata ). Tree ferns ( Dicksonia spp.) found in Elevation: 600 – 650m asl wet gullies. Ground layer includes austral bracken, grasses and a sparse forb community. Black Range State Forest, northeast Victoria Rainfall: >1000mm/year Dry forest dominated by Middlestorey occasionally present including silver wattle, Soils: Red friable/shallow pure messmate and blanket leaf ( Bedfordia arborescens ) and hazel pomaderris stony earths messmate/peppermint ( E. (Pomaderris aspera ). Moderate shrub layer includes musk Overstorey Ht: 28 34m radiata ) associations. daisy bush ( Olearia argophylla ), common cassinia, prickly Crown cover 70 84% current bush and hop goodenia. Tree ferns found in wet Elevation: 650 – 700m asl gullies. Ground layer includes austral bracken, grasses and a sparse forb community. 18
All three forests had been subjected to selective timber harvesting throughout the nineteenth century, but since about 1970 the seed tree silvicultural system (Lutze et al. 1999) had been predominantly used. Seed tree silviculture involves the harvest of 10 30 ha patches while retaining 4 9 mature trees per hectare to provide seed for the next crop and habitat for arboreal animals. Operations generally take place between late spring and autumn (October to April) after which logging debris is burnt to prepare a seedbed and stimulate seed fall. Additional seed is added by hand or from the air if necessary, and only in exceptional circumstances are nursery grown seedlings planted. Over the years, this harvesting system has produced a matrix of differentially aged patches of regenerating forest surrounded by mature stands that show signs of historical logging operations (usually single tree selection) to a greater or lesser degree. Wallaby capture and radio tracking Wallaby movement data were collected in the Pyrenees State Forest (Figure 2.1). I collected data before, after and in some cases during harvesting and analysed it within the framework of a Multiple Before After Control Impact (MBACI) design with a single before and after sample (Downes et al. 2002). To increase the likelihood of spatial independence, unharvested control locations were defined as stands of unharvested forest at least 1.5 km from each other and from other disturbed areas. This distance was used as it was about three times the average home range length, and thus the home ranges of wallabies caught 1.5 km apart were considered unlikely to overlap. I randomly selected six control locations from a pool of 15 potential sites identified within the area used for timber harvesting (above approx. 500 m asl). Two sites originally intended for use as replicate impact location were harvested during the study period. However, the forest adjacent to one of these was subjected to a fuel reduction burn about 12 months prior to harvest and the post harvest slash burn at this site was also used to reduce fuel in another adjacent unharvested patch. These factors resulted in substantial differences between the two impact locations, so I viewed them as case studies and compared them to control sites separately. 19
Wallabies were trapped from March to October 2004 using double layered traps designed for the purpose (Di Stefano et al. 2005). I free fed with carrots up to four weeks prior to trapping, then used carrots and occasionally peanut butter as bait, setting traps in the late afternoon and checking them early the following morning. Once caught, wallabies were sedated with an intra muscular injection of Zoletil 100 (Virbac Australia. Ltd) at 0.05 mg/kg and fitted with a Sirtrack radio collar (approximately 30 g) and two Allflex ‘button’ ear tags. I glued reflective tape to both collar and tags to facilitate identification. At the point of capture I recorded the weight, crus (leg) and pes (foot) length of adults and the pes length of pouch young. I initially caught and radio collared 27 wallabies but due to death, battery failure, and collar loss, the final analysis consisted of data from 15. Seven (two male, five female) were from five control sites while the two impact sites had six (three male, three female) and two (female) wallabies respectively. Wallabies were radio tracked by homing in on foot using a hand held Yagi antenna (Telonics RA 14) and a Telonics TR2 portable receiver and positions were recorded using a Garmin 12 GPS unit, which reported an estimated error of <15 m in 96.3% of cases. I tracked in all weather and at all times of the day and night (defined as completely dark), and tracking was scheduled so that nocturnal and diurnal locations for each wallaby were spread approximately evenly throughout these periods. In order to minimize disturbance to individuals I left at least 6 hours between tracking events and obtained no more than two fixes per individual in a 24 hour period. Tracking was conducted in short bursts (field trips) with periods of relative inactivity in between, resulting in blocks of autocorrelated data that are likely to be relatively independent from one another (Katajisto and Moilanen 2006). Due to the nature of animal movement, radio tracking locations will usually be autocorrelated to a greater or lesser extent, and methods that exist to derive a statistically independent sample (Swihart and Slade 1985) may reduce biological relevance (de Solla et al. 1999; Lair 1987; Reynolds and Laundre 1990; Rooney et al. 1998). Consequently, I do not consider autocorrelation in this analysis. My objectives required the estimation of home range boundaries and size, so the focus was on acquiring a representative sample of 20 locations within a timeframe long enough for the wallabies to adequately reveal their home range (Otis and White 1999). Because I did not always see wallabies before disturbing them, I used a 1 – 5 rating system to quantify location accuracy: 1 = within 5m of exact location, 2 = 5 – 25 m, 3 = 25 – 100 m, 4 = 100 – 200 m and 5 = >200 m, or when a signal could not be detected. Due to their relative inaccuracy, a GPS reading was not taken for rating 5s. In cases where I saw or heard the wallaby (rating 1, 2 and most 3), locations were based on visual and aural cues. In cases when the wallaby moved before it was observed, (some 3 and all 4), I estimated its pre disturbed location on the basis of aural feedback from the tracking equipment in relation to the local terrain. The mean (±95% CI) percentage of locations that corresponded to each rating was: 1: 49.8% ± 11.7, 2: 19.6% ± 5.7, 3: 23.7% ± 7.3, 4: 4.7% ± 2.4 and 5: 2.2% ± 1.6. Due to their potential inaccuracy, rating 4 locations were excluded if they resulted in large range size increases, which I defined as ≥10%. In this instance, only one location was removed. Due to the time required to catch animals at multiple sites, different harvest start dates for each impact site, time delays in the harvesting process and the need to cull a number of control wallabies for another study (Chapter 3), the before and after periods at each site did not completely overlap. This introduces time as a potential confounding factor, although I do not believe its effect is likely to be large. Two control animals were tracked for extended periods and showed no temporal change in their use of space and other wallabies monitored for >12 months (Chapters 4 – 6) demonstrated strong fidelity to their home range over a number of seasons. Consequently, any effect observed in this study is most likely attributable to the harvesting treatment. The final data set contained 1379 positions with 79.7% ± 1.0 (mean ± 95% CI) collected during the day. Although this biases home range estimates towards diurnal ranges, the ratio of diurnal to nocturnal locations was similar between impact and control groups, thus validating the contrast between control and impact sites. The mean number of positions per individual within each monitoring period (before, during or after) was 32.3 ± 0.9 (max. = 34, min. = 30) and 31.2 ± 0.8 (max. = 36, min. = 26) for control and impacted locations respectively. I collected about 30 locations for each animal in each time period as 30 is considered a minimum for kernel based home range estimation 21
(Seaman et al. 1999), and similar sample sizes reduces bias from comparisons (Kenward 2001). Wallaby density Wallaby density data were obtained at 10 sites (five control, five impact) across the three study forests. Four sites (two impact, two control) were established in each of the Pyrenees and the Black Range, and two sites (one impact, one control) at Mt. Disappointment. The design conformed to a traditional MBACI with data collected at multiple control and impact sites a number of times before and after harvesting (Downes et al. 2002). Because the Black Range State Forest had been intensively harvested during the last half century, it was difficult to find potential unharvested control locations far enough away from previously harvested sites to be spatially independent of them. I therefore defined control locations in all forests as potentially harvestable stands regardless of their position relative to previously harvested areas, which differs from the spatially isolated controls used to test prediction 1. The final control locations were selected at random from a larger pool of potential sites, and were at least 1.5 km from impact locations. The population of impact sites was limited by harvesting plans and I used sites that were available within the study’s timeframe. At each of the 10 sites I defined an approximately square 10 ha area and used a randomly positioned grid to locate about 30 (min.= 25) 15 m2 permanent circular plots. Beginning in March 2004 I counted wallaby faecal pellets in these plots (Southwell 1989) every month for three months and then every two months thereafter until July 2006. I assumed that pellet numbers reflected wallaby density (Johnson and Jarman 1987), although they may also be related to changes in activity. On virtually all occasions I was able to differentiate between swamp wallaby pellets and the pellets of other macropods (mainly eastern grey kangaroos, Macropus giganteus ) on the basis of size, shape, colour and internal texture (Triggs 2004). A small number of unidentified pellets (2.6% of the sample) were excluded from subsequent analysis. 22
Data were collected on five occasions before harvesting and on seven occasions afterwards, although due to the onset of harvesting operations only three of the impacted sites contributed to all five pre harvest counts. Post harvest burning was conducted at all impacted sites during April or May 2005 and plots were re established at this time. Post harvest plots were not in exactly the same position as pre harvest plots, but were systematically spread over the same (or very similar) 10 ha area. At each site the number of pellets on each plot was converted to pellets/ha/day and then averaged to generate a measure of relative density at each site for each monitoring time. These site means were used in the analysis described below. Browsing impact A browsing assessment was conducted on all five regenerating areas during April 2006, approximately 12 months after the slash burn, and shortly after a major increase in scat numbers at these sites. Seedlings were between 5 and 50 cm tall at the time of assessment. I used the previously established scat counting plots as a sampling frame to assess browsing damage and density of regenerating Eucalyptus seedlings, although new 15 m2 plots were established in adjacent positions to remove any effect of repeated visits on regeneration success, and in one case I used smaller plots (4.37 m2) due to high seedling density. In the Pyrenees, predominant tree species included messmate ( E. obliqua) and blue gum ( E. globulus bicostata ) at site 1 and messmate, blue gum and a mix of candle bark ( E. rubida ) and red stringybark ( E. macrorhyncha ) at site 2, and were regenerating as both seedlings and coppice. Eucalypt regeneration at Mt. Disappointment and the Black Range consisted almost completely of messmate seedlings. Because both seedlings and coppice were present in the Pyrenees, I conducted assessments for both types of regeneration at these sites. I assessed between 108 and 140 seedlings per site for browsing by selecting the five (occasionally fewer in sparsely regenerating patches) seedlings closest to plot centres and estimating the amount of biomass removed to the nearest 5%. The type of damage (side leaves, growing tip, whole crown, etc) was also recorded. At the two Pyrenees sites, coppice was assessed in the same way except that the five stems were randomly selected from one or two multi stemmed coppicing stumps. Browsing impact was averaged for seedlings (or coppice stems) within plots, and then the site value expressed 23 as an average of the plot means. To assess the relationship between browsing impact and wallaby density I calculated pre harvest density as the mean of faecal pellet counts at each pre harvest monitoring time, and post harvest density as the mean of faecal pellet counts at each post harvest monitoring time prior to the browsing survey. I assumed that the majority of browsing damage was caused by swamp wallabies, although other herbivorous mammals (e.g. eastern grey kangaroos, European rabbits, Oryctolagus cuniculus , sambar and fallow deer, Cervus unicolour and Dama dama , and brushtail possums, Trichosaurus vulpecular ) were also present. Kangaroos are predominantly grazers (Sanson 1980), and thus are unlikely to eat tree seedlings, and rabbits often leave bitten off shoots lying on the ground, which was not observed. On the basis of observed faecal pellet numbers, the abundance of the other potential browsing species appeared to be very low relative to swamp wallabies. To be operationally relevant, browsing impact should be linked to regeneration standards (Reimoser et al. 1999), which in southeastern Australia are defined by stocking, the proportion of 16 m2 plots containing a viable seedling (e.g. Dignan and Fagg 1997). Although stocking surveys are usually conducted 18 to 30 months after harvesting, results as early as 12 months post harvest are still acceptable (Dignan and Fagg 1997). I related browsing impact to stocking by defining browsing impact, BI , as the difference between total stocking, ST, and stocking adjusted for browsing impact, S A:
BI = ST − SA
While BI is a measure of browsing impact, regeneration is unacceptable if SA falls below a minimum standard, which for the sites used in this study was 65%. 2 Normally, ST would be calculated as the percentage of 16 m plots containing at least one seedling, regardless of browsing damage, while SA would be the percentage of plots containing at least one undamaged or substantially undamaged seedling, where substantial damage is defined as the removal of the whole crown (Wilkinson and Neilsen 1995). Due to different sized plots (15 m2 and 4.37 m2) used in this study, 24 however, I estimated 16 m2 stocking from an h factor graph (Lutze 2003), which represents the relationship between seedling density, heterogeneity (Mount 1961) and 2 16 m stocking. ST and SA were estimated using total and adjusted seedling density respectively, where for each plot adjusted density = total density × (1 PSD), and PSD was the proportion of seedlings substantially damaged. Data analysis Conventional analysis, bootstrapping procedures and home range calculations were performed in GenStat 8, Pop Tools (Hood 2005) and Ranges VI (Anatrack Ltd.) respectively. I generated home range (95% fixed kernel) and core range (50% fixed kernel) estimates for pre harvest, post harvest and, where applicable, during harvest periods, and calculated the percentage overlap of post and during ranges on pre harvest ranges. Distances between before/during and before/after geographic centres of location (GCL) were also calculated. At control sites, range size data from before and after harvesting were converted to a single variable by calculating the before/after difference and expressing it as a percentage of the pre harvest value. At impact sites, the same procedure was followed but the before/during differences were also calculated, resulting in a before/during and a before/after contrast. Each variable (95% range size, 50% range size, 95% range overlap, 50% range overlap and distances between GCLs) was analysed separately, both for combined male and female data and for females only. The female only analysis was conducted as there is evidence that, with respect to some aspects of swamp wallaby behaviour, the sexes behave differently (see Chapters 4, 5 and 6). Inferences about harvesting impacts on the five variables were made by comparing the single value from each impact site to the mean and associated 95% confidence derived from the control sites. Due to the presence of outliers, I used 10 000 bootstrap iterations to calculate 95% confidence intervals around the mean control values. To determine an appropriate smoothing factor for the home range estimates I initially followed the recommendation of Kenward (2001) and multiplied the reference bandwidth ( href ) by 0.6, the median multiplier from the sample of Least Squares Cross Validated results. This, however, resulted in an inappropriate degree of smoothing, as 25 some home range outlines were fractured into multiple segments and biologically nonsensical. Results were reported based on a smoothing factor computed as href × 0.8, as the home range outlines generated were most consistent with the perception of space use acquired throughout the tracking regime. I used a four factor repeated measures ANOVA to assess the impact of harvesting on wallaby density. The factors were State Forest (Pyrenees, Black Range and Mt. Disappointment; used as a blocking factor), Treatment (control and impact), BA Period (before and after) and Monitoring Time nested within BA Period. Although this analysis tests multiple hypotheses, the two of primary interest are the Treatment × BA Period and Treatment × Monitoring Time (BA Period) interactions. This type of analysis is described in detail by Downes et al. (2002). I calculated the Greenhouse Geisser epsilon to assess the assumption of equal correlation between the monitoring times (G G epsilon = 0.26), and ultimately used it to adjust the outputs of the analysis. I tested assumptions of normality and homogeneity of variance with a half normal plot and a fitted value plot respectively, and deemed a log 10 (x + 0.1) transformation to be necessary. As logarithmic transformations reduce the effect of large values, they may result in the loss of biologically important patterns (Keough and Mapstone 1995). In addition, interpreting effects on the transformed scale can sometimes be difficult (Jaccard and Guilamo Ramos 2002; Stewart Oaten et al. 1992). Therefore, to examine this potential effect on the Treatment × BA Period interaction using the raw data, I generated 95% confidence intervals around the mean before/after differences at control and impact sites using 10 000 bootstrap iterations. I assessed the relationships between stocking and other variables (browsing impact, seedling density and wallaby density) using scatter plots. The use of formal statistical techniques was not warranted due to the small sample size. An issue associated with the formal analyses outlined above is the inability to randomly allocate treatments to sites, thus raising the possibility that observed effects are influenced by factors other than the experimental treatment (Johnson 2002). Although this problem is intractable, using harvesting operations as experimental treatments still 26 provides a useful framework for wildlife research in managed forests (Lindenmayer 1999). Results Space use Raw home range data are presented in Appendix 1. The percentage of pre harvest wallaby locations in areas that were subsequently harvested or disturbed by the harvesting process ranged from 3.0 to 94.4%. Wallabies with a small percentage of their pre harvest locations in subsequently disturbed areas were not attracted into them during harvesting or up to 3 months after. Wallabies that spent more time in disturbed areas before harvesting moved away when harvesting began, and used the disturbed area substantially less during and after the harvesting operation (Table 2.2). The exclusion of males from the data set had little effect on the results for either the range size or range overlap analyses, so unless otherwise specified, the following comments refer to both combined and female only data sets. Table 2.2. Percentage of wallaby locations within the area disturbed by the harvesting operation. Before = before harvest, During = between harvest onset and burning of logging debris and After = after burning. At Impact Site 2, a substantial area was burnt but not harvested so numbers in parenthesis refer to locations in either harvested or burnt areas. Impact ID % of Locations in Disturbed Area Site Before During After 1 F1 3.3 0 3.1 1 F2 9.1 13.8 0 2 F1 3.1 (6.3) 0 3.2 (3.2) 2 F2 3.0 (30.0) 0 0 (3.1) 2 F3 41.7 (94.4) 3.1 0 (30.3) 2 M1 25.8 (45.2) 12.9 11.8 (29.4) 2 M2 9.7 (34.2) 15.2 0 (10.7) 2 M3 22.6 (48.4) 3.0 0 (3.3) 27
The effect of harvesting was most clearly shown by changes in home range overlap. In most cases, 95% home range overlap was less at impact sites than controls (Figure 2.2 A), and, with the exception of the before/after contrast at Impact Site 1, this effect was accentuated when 50% overlap was measured (Figure 2.2 B). Particularly for the female data at Impact Site 2, this demonstrated an almost complete shift away from pre harvest core use areas in the during and post harvest periods. The effect of harvesting on home range size (Figure 2.2 C and D) differed markedly between the two impact sites. Relative to controls, 95% range size (Figure 2.2 C) decreased at Impact Site 1 but tended to increase at Impact Site 2, particularly for the before/after contrast. Nevertheless, the large before/after contrast values at Impact Site 2 represented the mean response of a number of individuals and hides substantial within site variation. For the combined male and female data, the mean (95% CI) change in home range size was 116% (3 to 229%) and is composed of three large range increases and two moderate range reductions. Results from the 50% range size analysis (Figure 2.2 D) were variable at both control and impact sites , and no clear patterns were evident. At control sites for male and female data combined, the mean (95% CI) distance between the before and after geographic centre of location (GCL) was 65.5m (31.9 to 124.4m). At Impact Site 1 the distance between both the before/during (100.5m) and before/after (39.0m) GCLs was within the control site confidence interval, although the latter value was close to the lower bound. At Impact Site 2 the distance between both the before/during (151.7m) and before/after (204.4m) GCLs was outside the control site confidence interval. The results were similar when males were excluded. The largest distance moved was by M3 (Table 2.2) whose post harvest GCL was 340m from his pre harvest one. Even so, his post harvest 95% home range overlapped his pre harvest home range by 20.5%. 28
○ Impact Site 1 Before/During □ Impact Site 2 Before/During ● Impact Site 1 Before/After ■ Impact Site 2 Before/After ▲ Controls 95% Kernel 50% Kernel B A Home Range Overlap Home Range Overlap 90 90 80 80 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0
95% Kernel 50% Kernel C D Home Range Change Home Range Change 140 140 120 120 100 100 80 80 60 60 40 40 20 20 0 0 20 20 40 40 Impact Controls Impact Controls Impact Controls Impact Controls M & F M & F M & F F F M & F F F Figure 2.2. The effect of harvesting on (A) 95% range overlap, (B) 50% range overlap, (C) 95% range size and (D) 50% range size. Overlap is calculated as the percentage of the during or after range overlapping the before range. M & F refer to combined male and female data while F refers to female data only. Error bars are 95% confidence intervals. 29
Wallaby density There was a clear effect of harvesting on relative wallaby density. The raw data (Figure 2.3) suggested a substantial Treatment × BA Period interaction, and although not detected statistically by the analysis using transformed data (df = 1, F = 0.02, P = 0.9), analysis of the raw data showed a substantial effect. Based on 10 000 bootstrapped samples, the mean (95% CI) before/after difference between control and impact locations was 21.5 (12.3 to 30.4) pellets/ha/day. The mean before/after difference is equivalent to the Treatment × BA Period interaction, and an effect is implied as the lower confidence bound is substantially larger than zero. In addition, the analysis of transformed data provide strong statistical evidence for the Treatment × Time (BA Period) interaction (df = 10, F = 10.97, P = <0.001), which was driven by the contrast between the first three and last four post harvest measurements (Figure 2.3). During the first three post harvest measurements, wallaby density at controls was substantially more than at impact sites (e.g. control minus impact difference (± CI of difference) at July 2005 was 6.9 ± 3.1 pellets/ha/day), but this pattern was reversed at subsequent monitoring times. At impact sites, there was a clear trend of increasing density with time after harvest, although the shape of the trend beyond July 2006 is unknown. Browsing impact Mean biomass loss from eucalyptus seedlings ranged from 1.0 to 11.2% with both sites in the Pyrenees and one in the Black Range experiencing the highest browsing impact. In contrast, biomass loss from coppice stems was 0.9 and 0.4% at the two Pyrenees sites. The percentage of seedlings with missing crowns (considered to be seriously damaged; Wilkinson and Neilsen 1995) ranged from 0 to 12.9%. In order of increasing effect, the reduction in stocking due to severe browsing, BI , was 0%, 0%, 1%, 2% and 3%, indicating that browsing impact had an insubstantial effect on stocking rates. 30
180 170 Control 160 150 Impact 140 130 120 110 100 90 80 Main harvesting 70 period 60 50 40 30 20 10 0 2004 2005 2006 Figure 2.3. Effect of harvesting on relative wallaby density. Pre harvest data March to September 2004, post harvest data July 2005 to July 2006. Error bars are 95% confidence intervals. There was a general negative relationship between adjusted stocking and biomass loss (Figure 2.4 A), although the data point in the top right hand corner of the graph demonstrates that biomass loss may not always be a good predictor of stocking. Adjusted stocking was positively related to adjusted seedling density (Figure 2.4 B) and negatively related to both pre and post harvest wallaby density (Figure 2.4 C). In these figures, both stocking and seedling density have been adjusted for browsing impact so that values relate to seedlings that had not been seriously damaged. The small sample size, in addition to the strong correlation between the four predictor variables (Pearson’s correlation coefficient ≥ 0.57), makes it difficult to infer causal relationships. 31
Discussion Impact of harvesting on swamp wallabies Consistent with prediction 1, immediate and short term (3 months post harvest) effects of harvesting on space use were minor. Although most wallabies shifted their core use area in response to harvesting and some appeared to greatly increase the size of their home range, individuals were relatively unperturbed by the harvesting process, even when logging machinery was operating. The wallabies most affected were at Impact Site 2 (e.g. F3, M1 and M3 in Table 2.2). While these individuals modified their movements slightly to avoid the harvesting operation, all continued to use parts of their pre harvest range within the during and post harvest periods. These results are generally consistent with other BACI type studies investigating the effects of harvesting on medium to large generalist herbivores. For roe deer ( Capreolus capreolus ), white tailed deer ( Odocoileus virginianus ) and elk ( Cervus elaphus ), indices of home range size and overlap differed little between individuals that occupied harvested sites and those that did not (Campbell et al. 2004; Edge et al. 1985; Linnell and Andersen 1995). Mobile animals like wallabies and deer have the ability to avoid harvesting and other disturbances while still using familiar areas. Although I did not collect these data, the observed changes in space use appeared to have little effect on the availability of important resources in the short term, although for other species this may depend on factors such as population density and territoriality. This is in contrast to the immediate and short term effect of harvesting and other similar disturbances on arboreal animals (Gibbons and Lindenmayer 1996; Kavanagh 2000; Lindenmayer et al. 1991; Lindenmayer and Franklin 1997; Newell 1999; Tyndale Biscoe and Smith 1969), which suffer a direct reduction of important resources, and may die as a result. 32
A 100 90 80 70 60 50 40 30 20 10 0 0.0 2.0 4.0 6.0 8.0 10.0 12.0 Browsing Impact (% biomass loss) B 100 90 80 70 60 50 40 30 20 10 0 0 1500 3000 4500 6000 7500 9000 Adjusted Seedling Density (s/ha) C 100 90 80 70 60 50 40 30 20 Pre harvest 10 Post harvest 0 0 5 10 15 20 25 Wallaby Density (pellets/ha/day) Figure 2.4. Relationships between adjusted stocking and (A) percent biomass removed, (B) adjusted seedling density and (C) wallaby density. Both stocking and seedling density values have been adjusted for browsing impact. 33
Reduced use of the disturbed areas in the first months after harvesting by radio tracked individuals is consistent with results from the faecal pellet survey. Relative to the pre harvest baseline, post harvest density was initially very low but, in support of prediction 2, began to increase about 8 10 months after the post harvest burn and by the end of the monitoring period was at least 5 times pre harvest levels. Although it is possible that a post harvest differential in pellet detectibility developed due to rapid vegetation growth at impact sites (e.g. Bulinski and McArthur 2000), any associated bias was likely to be minor relative to the size of the observed effect. Similar short term increases in the density of other mammalian herbivores has been found after both native forest (St Louis et al. 2000) and plantation (le Mar and McArthur 2005) harvesting, although data regarding causal factors are limited. Although it was not an objective of this study to quantify the reasons for the observed density effect, qualitative observations suggest that the presence of shelter was important. During this and a previous study (Di Stefano 2005), I observed wallabies on harvested areas only after regenerating vegetation (mainly eucalypt seedlings, eucalypt coppice and austral bracken, Pteridium esculentum ) had grown to wallaby height. On a number of occasions I disturbed individuals sheltering next to 1 2m tall coppice regrowth and pellet numbers close to re sprouting vegetation appeared relatively high. Whatever the causal factors, it is clear that swamp wallabies perceived harvested areas to be a favourable habitat less than a year after the post harvest burn, and that this effect was consistent across the landscape. Although the changes in wallaby density observed during the ca. 15 month post harvest monitoring period were almost certainly due to the movements of pre existing individuals (a functional response), it is possible that both functional and numerical mechanisms are important in the longer term. If regenerating sites do constitute high quality habitat for this species, wallabies with access to these areas may benefit from increased reproductive success, or decreased mortality, resulting in numerical population expansion. The measurement of (for example) reproductive success over time in individuals with and without access to regenerating areas is required before the mechanisms driving density increases can be quantified. 34
Potential ecosystem effects
Selection of high quality patches and subsequent increases in herbivore density may affect the forest ecosystem. Horsley et al. (2003) found that over a 10 year period, browsing by white tailed deer in harvest areas reduced plant diversity and the density of favoured dietary species, and that higher deer densities increased the effect. Changes to community composition can have secondary impacts on other groups of organisms (Flowerdew and Ellwood 2001; Moser and Witmer 2000) and in some situations, mammalian herbivores may interact with other perturbing factors (such as disturbance) to cause irreversible switches in plant species abundance (Augustine et al. 1998). Mammalian herbivores may also effect nutrient cycling in a number of different ways (Bardgett et al. 1998; Hobbs 1996; Pastor and Naiman 1992). Although inputs of faeces and urine increase the amount of nitrogen available to plants, this does not appear to compensate for nitrogen losses due to reduced litter quantity and quality (Pastor et al. 1993; Schoenecker et al. 2004), and overall nitrogen loss appears to increase with herbivore density (Feeley and Terborgh 2005; Persson et al. 2005). Changes to leaf litter quantity, as well as plant physiological responses stimulated by browsing, may also affect soil micro organisms and the processes they facilitate (Bardgett et al. 1998), although there is evidence for both positive and negative affects (Wardle et al. 2001; Wardle et al. 2002). It seems clear that factors resulting in high herbivore densities can cause feedbacks that effect ecosystem structure and function. In the context of the present study, the effect of high wallaby densities in regenerating patches of native forest is unknown and warrants further investigation. A specific prediction, for example, is that high population densities and subsequent consumption of foods high in nitrogen (Osawa 1990) will reduce the overall quality of leaf litter to the decomposer sub system. An adequate test of this prediction would require naturally disturbed areas (burnt by wildfire, for example) as a control, so that effects in harvested areas could placed in the context of normal ecosystem change. 35
Browsing impact In addition to herbivorous mammals, the success of post harvest eucalypt regeneration is influenced by factors including seed supply, seedbed condition, overwood competition and local climate (Squire et al. 1991a), and final regeneration outcomes are likely to result from some combination of these effects. However, the development of an efficient browsing reduction strategy requires the effect of browsing on regeneration success to be separated from other factors, and the index of browsing impact (BI ) presented in this study enables this to be achieved. By directly linking browsing impact to an operational measure of regeneration success, the effect of browsing on regeneration standards can be quantified. For regeneration to be deemed acceptable for the harvested areas assessed in this study, stocking must be ≥ 65%. Of the 5 monitored sites, two were well below this level and three well above it, so regeneration standards were either met or not regardless of the small browsing impact (max. BI value = 3%). In situations where total stocking levels are just above the standard, small reductions attributable to browsing could be important. Thus the importance of browsing impact not only depends on browsing severity, but on the total stocking level as well. It should be noted that BI is a useful measure only if herbivory does not completely remove seedlings from the site (e.g. by killing them or making them difficult to find by removing almost all the biomass). If this occurred, total stocking, ST, would be underestimated, resulting in the underestimation of BI . Available evidence from southeastern Australia indicates that although browsing can substantially damage seedlings, it does not kill them in the short term (Bulinski 1999; Bulinski and McArthur 1999; Di Stefano 2005) . Nevertheless, careful monitoring may be required to identify live seedlings with missing crowns and generate an accurate estimate of ST. Management implications Harvesting had little impact on the space use of swamp wallabies. It did, however, result in regenerating patches within the forested landscape that became highly attractive to this species and supported relatively large numbers of individuals. As 36 previously outlined, high concentrations of herbivores have been found to alter aspects of ecosystems including plant species composition and nutrient cycling. Although I did not measure the effects of elevated wallaby densities on the forest ecosystem, the potential impact is of management concern, particularly given the longevity of the elevated density response (Chapter 6) and its generality across the landscape. From a management perspective, browsing impact is only important if it causes measures of regeneration success to fall below established standards. Thus the variable of interest is the reduction in regeneration success attributable to browsing, defined here as BI . In the context of southeastern Australia, browsing impact that causes stocking values to fall below the minimum standard (65% for the sites used in this study) should be deemed unacceptable. Currently, operational assessment of browsing impact is made by measuring height loss and recording the amount of biomass removed (Dignan and Fagg 1997; Forestry Tasmania 1999). As demonstrated in this paper, such measures of raw impact may not reflect the degree to which browsing mammals alter regeneration success. Using BI as an index of browsing impact will provide forest managers with the information to decide whether browsing is acceptable, or a problem requiring action. Finally, it seems that in areas where the practice is acceptable, the promotion of coppice stems may reduce the effect of browsing on regeneration success, as browsing on coppice was virtually non existent. 37
CHAPTER 3
Summer diet selection of swamp wallabies ( Wallabia bicolor ) in a harvested landscape: feeding strategies under conditions of changed food availability
Abstract Mixed diets and specialisation on preferred food items are two alternative foraging strategies observed in mammalian herbivores. I used the stomach contents of swamp wallabies ( Wallabia bicolor ) foraging on five major food types (ferns, forbs, monocots, shrubs and trees) to determine if one of these alternatives was being employed. The study was conducted in a landscape where the relative availability of different food types had been substantially altered by timber harvesting. Forbs were the major dietary component at both unharvested control sites (mean ± 95% CI: 59.3 ± 11.8%) and 5 year old harvested areas (40.1 ± 9.3%), although moderate amounts of shrub (20.9 ± 10.8% and 14.5 ± 8.2%) and monocot (14.0 ± 6.9% and 12.7 ± 5.0%) were also consumed. Trees and ferns were rarely eaten at unharvested sites (trees: 2.4 ± 1.7%; ferns; 3.5 ± 3.0%) but were consumed at higher proportions in recently harvested areas (trees: 15.2 ± 4.8%; ferns: 17.5 ± 4.7%). Non metric multidimentional scaling and a multi response permutation procedure (MRPP) demonstrated a substantial difference in diet composition between wallabies occupying unharvested forest and 5 year old regenerating sites (MRPP: A = 0.20, P <0.001). However, when analysed using an index of diet selection, the difference between the two groups was substantially smaller (A = 0.05, P = 0.04). At unharvested sites the selection rank (most selected to least selected) was: shrub, forb, tree, monocot, fern. At 5 year old sites it was: forb, shrub, fern, monocot, tree. Analyses of frequency dependence indicated negative relationships between selection and relative availability on most occasions, and were generally consistent with a mixed 38 feeding strategy. The analyses also demonstrated a positive relationship between tree selection and forage quality, and showed that selection of most food types was more strongly related to the relative abundance of other types than to its own. Additional data at a finer resolution and in different seasons are required to test the generality of these conclusions at different scales. 39
Introduction Generalist mammalian herbivores consume a wide variety of food types, but often select some to a greater degree than others (e.g. Morrison et al. 2002; Sprent and McArthur 2002; Tixier et al. 1997). However, patterns of selection do not stay constant in space or time, and may be influenced by factors such as toxin and nutrient concentration (Lawler et al. 2000; Tripler et al. 2002) or forage availability (Danell and Ericson 1986; Edenius et al. 2002). The effect of forage availability on selection (known as frequency dependence) has been rarely studied for free ranging mammalian herbivores, and there are at least two inconsistent theoretical expectations. On one hand, a positive relationship between selection and availability (positive frequency dependence) at the scale of an animal’s home range is consistent with diet specialisation as predicted by simple optimal foraging models, at least for preferred food types (Hubbard et al. 1982; Pyke et al. 1977). However, mixed diets are expected if consumption is constrained by nutrient availability (partial preference hypothesis; Pulliam 1975; Westoby 1974), or if eating a range of foods reduces the negative effect of plant toxins (detoxification limitation hypothesis; Freeland and Janzen 1974; Marsh et al. 2006). In these latter cases, a negative relationship between selection and relative availability is expected (negative frequency dependence). A final possibility is that selection will be independent of forage availability (frequency independence). Tests of frequency dependence in generalist mammalian herbivores have often been conducted in controlled environments where individuals are presented with varying availabilities of two food types. These data have most often demonstrated frequency independence (Bergvall and Leimar 2005; Chevallier Redor et al. 2001; Lundberg et al. 1990; Verheyden Tixier et al. 1998), although Bergvall and Leimar (2005) also demonstrated positive frequency dependence when the two food types differed in tannin concentration. Data from field studies suggest a more complex picture. Although negative frequency dependence has been demonstrated on a number of occasions (e.g. Danell and Ericson 1986; Edenius et al. 2002), Fortin et al. (2003) demonstrated both frequency independence and positive frequency dependence, and showed that patterns of selection were influenced by forage species, season and spatial scale. Apparently, 40 there are no studies of frequency dependence in free ranging mammalian herbivores that have considered the selection of multiple food types concurrently. In this study, I used data from swamp wallabies ( Wallabia bicolor ) living in a landscape altered by timber harvesting to quantify the effect of relative availability on the selection (defined as use in relation to availability; Norbury and Sanson 1992) of five plant functional groups, hereafter referred to as food types: ferns, forbs, monocots, shrubs and trees. I used wallabies from unharvested locations and young regenerating (5 years post harvest) sites, which differed substantially in terms of vegetation abundance and lateral cover. Swamp wallabies are 10 25 kg macropodid marsupials that have been classified as browsers on the basis of dental morphology (Sanson 1980). Diet analyses (de Munk 1999; Hollis et al. 1986; Osawa 1990) show that this species consumes relatively large quantities of forbs and shrubs, although substantial consumption of grasses and ferns may also occur. The only study that has quantified diet selection found strong positive selection for shrubs and strong negative selection for grasses and ferns (Wood 2002). The diet of swamp wallabies varies substantially in space and time and may be influenced by foliar nitrogen content (Osawa 1990) or by plant toxins (Lawler and Foley 1999). There is some evidence that they are opportunists, taking advantage of high quality food resources when they become available (Osawa 1990). In this study, my first objective was to quantify both diet composition and selection for wallabies living at unharvested and 5 year old regenerating sites. My second objective was to determine the relationship between the relative availability of food types and their selection at the scale of individual home ranges, and to reconcile the results against two alternative foraging strategies: specialization and mixed feeding. If, as the available evidence suggests, swamp wallabies are mixed feeders, they should demonstrate negative frequency dependence. If the relative availability of a food item spans an adequate range, this should manifest itself as decreased selection as relative availability increases. 41
Methods Study site This study was conducted in the Pyrenees State Forest which has been described in Chapter 2. Since 1990 the harvesting regime in the Pyrenees has resulted in around 25 regenerating blocks (coupes) of various ages surrounded by unharvested forest. I used unharvested forest and 5 year old regenerating sites to define two general habitat types for study. The two habitats differed most obviously with respect to vegetation abundance and lateral cover. Five year old regenerating sites were dominated by 1 3 m tall densely regenerating Eucalyptus seedlings, and were early enough in the successional process to have a relatively high abundance of forbs and grass. Species that regenerate well after mechanical disturbance or fire mainly silver wattle (Acacia dealbata ; defined as a shrub for the purposes of this study) and austral bracken ( Pteridium esculentum ) were also relatively abundant. Unharvested sites have been described in Chapter 2. They were relatively open and supported substantial quantities of forbs, grass and occasional patches of shrub. They tended to be further from major ridge lines and on steeper slopes than harvested sites and the density and size of mature trees varied depending on local harvesting history. Collection of stomach samples Nine wallabies from unharvested sites (seven females, two males), 10 with access to 5 year old regenerating sites (five females, five males) and three with access to 10 year old sites (all female) were shot in accordance with the Code of Practice for the Humane Shooting of Kangaroos (Environment Australia 1990). On the basis of radio tracking data, none of the wallabies in unharvested forest used regenerating areas, and none of the wallabies at regenerating sites moved between age classes. Of the 10 wallabies at 5 year old sites, eight were shot on the regenerating area and two in the adjacent forest, while all three wallabies at 10 year old sites were shot in the adjacent forest. With one exception, wallabies with access to 5 and 10 year old sites used them heavily, particularly during the day (see Chapter 5). Individuals were selected haphazardly from 42 a larger pool of 32 radio collared animals, although selection was somewhat biased towards less flighty wallabies. All wallabies were shot in autumn; twelve during April 2005 and 10 during March 2006, with each site type being represented in each year. In both cases the preceding summer had been dry and autumn rain had not begun at the time of shooting. Directly after shooting, equal sized samples of stomach contents (approx. 1cm 3) were taken from the base of the oesophagus, the sacciform forestomach, the tubiform forestomach and the hindstomach, pooled, and stored in 70% ethanol. I used a pooled sample to reflect an extended consumption period that was consistent with forage availability sampled from within the whole home range (see below). Diet analysis I used the microscopic technique described by Norbury (1988) to determine diet composition. Each sample of stomach contents was thoroughly mixed, placed in 4% sodium hypochlorite for one hour, rinsed through a 125 m sieve and dyed in 1% Gentian Violet for one minute. After additional rinsing, dyed material was suspended in corn syrup, and placed on three or four microscope slides sealed with nail polish. I viewed slides under a compound microscope and attempted to identify 250 plant fragments per sample using intersections on a grid measuring 2 × 2 fields of view as a sampling frame. The plant fragment closest to each grid intersection (defined by cross hairs) was identified on the basis of microhistological features (e.g. Storr 1961) as either forb, grass, sedge, shrub, fern, tree or unknown, using a previously established reference collection of plants from the study site. Very few sedges were positively identified, so grass and sedge were collapsed into a single monocotyledon category. In almost every case, ferns were austral bracken. The number of fragments belonging to each plant group was expressed as a proportion of the total number successfully identified. Using reference slides containing fragments with the same size distribution as the stomach contents, I determined the proportion of identifiable fragments in each plant group. These slides were prepared as described above except that fragments were left in 43 the sodium hypochlorite solution for three hours. The data were used to correct for unequal rates of identification between food types (Norbury 1988). Forage availability Within two months of shooting I sampled vegetation at the intersections of a randomly positioned grid overlayed on each wallabies’ home range (see Chapter 2 for home range methodology). I set sampling intensity at 1 plot/ha with the exception of home ranges smaller than 10 ha or larger than 30 ha, where I used 10 and 30 plots respectively. When wallabies used more than one habitat type a minimum of 10 plots were established in each. Data from plots within each home range (or habitat within home range) were averaged to generate a single value for each area. At each plot I measured the abundance of food types (ferns, forbs, grasses, sedges, shrubs and trees) as percent projected cover in three 1 m2 by 1.5 m high sub plots positioned 3m apart along a linear transect. Very small cover values were recorded as 1% and values between 5% and 100% were estimated to the nearest 5%. For each plant group, cover values for the three sub plots were averaged to generate a plot mean. I defined forbs as non woody plants while shrubs were defined as all woody plants except for Eucalyptus spp . which were categorised as trees. Except for the occasional rockfern (Cheilanthes austrotenuifolia ) all ferns were austral bracken, and all sedges were the spiny headed mat rush ( Lomandra longifolia ). Grasses included common tussock grass (Poa labillardieri ) and silvertop wallaby grass ( Joycea pallida ). For consistency with the diet analysis procedure, I combined cover values for grasses and sedges into a single monocotyledon category. Due to the low abundance of sedges, cover values of the monocot category were almost totally attributable to grass. Forage quality For each plant group I used the plots described above to collect healthy foliage growing below wallaby browsing height (1.5 m). Samples were pooled from each plot, stored in plastic zip lock bags to limit evaporation, weighed in the field, and transferred to paper bags for drying. Although foliage was only sampled from broad plant groups, sampling different species in proportion to their abundance was important so that subsequent 44 analyses were not unduly weighted by rare or patchily distributed species. I achieved this for forbs by harvesting all plant material within a number (five to 20, depending on local forb density) of 10 cm 2 × 1.5 m high sub plots established along a 10 m transect. Plants from all other functional groups were collected from within or near to the 1 m2 × 1.5 m high plots used to estimate their abundance. For trees and shrubs, I made a qualitative record of the species at each plot and used this to ensure that relative proportions of different species in the pooled samples were approximately correct. For ferns and sedges sampling the correct species mix posed little problem as almost all ferns were P. esculentum and all sedges were L. longifolia . Grasses were difficult to identify to species, so I simply ensured that a variety of types present at each plot were included in the sample. In the laboratory, I dried samples at 60°C for 48 hours, weighed them, and ground the dry material through a 0.5 mm sieve using a Culatti grinder. I used the dried material to derive three variables quantifying forage quality; water content, nitrogen content and dry matter digestibility. I expressed water content as a percentage of wet weight (wet weight – dry weight / wet weight), and nitrogen content (% DM) was calculated with a LECO CHN analyser. To determine dry matter digestibility, the ground sample was weighed and incubated at 40°C with 20 mL of 0.2% pepsin in dilute hydrochloric acid for 24 hours. Then, 20 mL of cellulase solution was added to the supernatant and the sample incubated for a further 24 hours at 40°C. The residue was then dried, weighed, ignited at 540°C for four hours, cooled, weighed again, and expressed as a percentage of the initial dry weight. A detailed description of the method can be found in Dowman and Collins (1982). Data analysis The sample was too small to examine differences between males and females so I pooled sexes for all analyses. Overall diet consumption and selection at unharvested forest and 5 year old regenerating sites were analysed by subjecting data standardised by the maximum value (a range transformation) to non metric multidimentional scaling (NMDS) using the Bray Curtis distance measure. Consumption data were expressed as the proportion of each plant group found in the gut of each wallaby and selection was 45
expressed using a standardised selection index B. Bij = ( uij/a ij) / ∑uij/a ij where uij and aij are the proportion used and available respectively of food type i, for wallaby j. The values for this index range from 0 to 1 and are interpreted as the probability that a food type would be the next selected if all food types were equally available (Manly et al. 2002). I used a Multi response Permutation Procedure (MRPP; Mielke et al. 1976) to investigate overall (multivariate) differences in both consumption and selection between wallabies at unharvested and 5 year old sites. All analyses were conducted using PC ORD 4.25 (McCune et al. 2002). I used compositional analysis (Aebischer et al. 1993) to generate mean selection ranks for the five food types at both unharvested forest and 5 year old sites. When food types were available but not used, the zero use value was replaced by 0.01. In cases where food types were neither used nor available, values were treated as missing but replaced by the mean of non missing values to calculate the ranking matrix. Analysis were conducted using Compos Analysis 6.2 (Smith 2004). Analysis of frequency dependence was assessed in three stages. First, I used simple linear regression to assess the relationship between diet selection ( B) and relative abundance of each food type separately. Second, I used multivariate linear regression to quantify the effect of relative availability on selection when three forage quality variables (nitrogen content, water content and dry matter digestibility) were also entered into the model. Third, I used Manly’s (1973) linear model to assess frequency dependence in a multivariate context. The model
Selection ( B) = βPforb + βPmono + βPshrub + βPfern + βPtree was fitted separately to the Bs for each food type using the relative availability of all food types (the Ps) as predictors (Manly 1973). As relative availability values sum to 1, models for each food type were constructed by systematically excluding the abundance data for one food type at a time and using the intercept from the partial model as the coefficient for the excluded term. Assumptions of normality, homogeneity of variance and linearity were assessed using normal probability plots and plots of residuals against fitted values, and in the first two 46
analyses the response variable, B, was log 10 transformed (log 10 + 0.01 in the case of fern and tree data) to stabilise variances. In the third analysis, transformation was not conducted as the model was developed for use with raw data (B. Manly, pers. comm.). In this case, up to two outliers were removed from each data set to meet analytical assumptions, although variances in the tree selection model were still somewhat heterogeneous. There were no substantial correlations between predictors in the multivariate analyses with the exception of forb and tree relative availability in analysis three ( r = 0.71). Consequently, I re ran the forb and tree models using traditional multiple regression after excluding tree and forb relative abundance respectively. All regression analyses were conducted in Minitab 13.31. Results Regardless of the habitat, forb constituted the major portion of the wallabies’ diet, with moderate amounts of shrub and grass also consumed. The consumption of tree and fern was very low in the unharvested forest, but rose to moderate levels at 5 year old sites. Substantial differences in consumption between unharvested and 5 year old sites was observed for fern, forb and tree (Table 3.1). The mean relative availability of forb, monocot and tree also differed between unharvested and 5 year old sites, and the selection index B was substantially different for fern, forb, shrub and tree (Table 3.1). Although the effects for fern, shrub and tree were not statistically significant at the 5% level (i.e. the 95% confidence intervals around the difference between B values overlapped zero), the point estimates represent substantial mean (or median) changes within a variable data set. Simulation studies show that the population parameter is more likely to be near the centre of a confidence interval than at either extreme (Cumming and Finch 2005), so I consider cases where wide confidence intervals just overlap zero to be strong evidence for an effect. At unharvested sites compositional analysis indicated that the most highly selected food type was shrub followed by forb, tree, monocot and fern. At 5 year old sites the most highly selected type was forb followed by shrub, fern, monocot and tree. Comparisons of selection between different food types (Figure 3.1) are represented as the value of the first listed type minus the value of the second listed type, so a positive estimate represents the selection of the first over the second. Major differences between the 47 unharvested forest and 5 year old sites (Figure 3.1 A and B) are exemplified by the shrub vs forb comparison and all comparisons containing fern. In addition, all estimates are relatively precise at 5 year old sites indicating reduced variation between individuals. Table 3.1. Use, availability and selection ( B) of the five food types at unharvested forest ( n = 9) and 5 year old ( n = 10) sites. %used refers to the percentage of food types in stomach contents, %available refers to the percentage of food types derived from the field survey and B is a standardised index of diet selection (see text for details). Errors are 95% confidence intervals, or confidence intervals of the difference.
Food Type Unharvested 5 year old Difference forest forest (Unharv. – 5 yr) % Used Fern 3.5 ± 3.0 17.5 ± 4.7 14.0 ± 5.3 Forb 59.3 ± 11.8 40.1 ± 9.3 19.2 ± 13.7 Monocot 14.0 ± 6.9 12.7 ± 5.0 1.3 ± 7.8 Shrub 20.9 ± 10.8 14.5 ± 8.2 6.4 ± 12.4 Tree 2.4 ± 1.7 15.2 ± 4.8 12.8 ±4.9 1 % Available Fern 8.6 ± 6.0 13.6 ± 8.4 5.0 ± 9.8 Forb 33.2 ± 13.9 11.0 ± 6.4 22.3 ± 14.6 1 Monocot 47.4 ± 8.0 28.5 ± 7.0 18.9 ± 9.8 Shrub 5.5 ± 3.0 5.9 ± 3.1 0.3 ±4.0 Tree 5.2 ± 5.2 41.1 ± 14.7 35.9 ± 15.1 1 Selection (B) 2 Fern 3 0.07 (0.00, 0.18) 0.16 (0.12, 0.26) 0.09 ( 0.19, 0.03) Forb 0.24 ±0.14 0.43 ± 0.12 0.19 ± 0.17 Monocot 0.03 ±0.02 0.06 ± 0.04 0.03 ± 0.04 Shrub 0.43 ± 0.16 0.28 ± 0.14 0.15 ± 0.20 Tree 3 0.09 (0.01, 0.22) 0.03 (0.01, 0.08) 0.06 ( 0.03, 0.18) 1 Confidence interval calculated under the assumption of unequal variances. 2 Random feeding (i.e., foods selected in proportion to their availability) corresponds to 0.20 (1/ n where n is the number of food types). 3 Due to the presence of outliers, medians were used as a measure of central tendency, and confidence intervals calculated using 10 000 bootstrapped iterations. 48
COMPARISON (A) Unharvested forest (B) 5 Years old
Shrub vs Forb Shrub vs Forb
Shrub vs Tree Shrub vs Tree
Shrub vs Mono Shrub vs Mono
Shrub vs Fern Shrub vs Fern
Forb vs Tree Forb vs Tree
Forb vs Mono Forb vs Mono
Forb vs Fern Forb vs Fern
Tree vs Mono Tree vs Mono
Tree vs Fern Tree vs Fern
Mono vs Fern Mono vs Fern 3 1 1 3 5 7 3 1 1 3 5 7 Log ratio Difference Log ratio Difference Figure 3.1. Selection of plant groups by wallabies living at (A) unharvested control sites ( n = 9) and (B) 5 year old regenerating sites ( n = 10). Effects are expressed as the first named plant group minus the second named plant group, so positive values mean the first named group is selected to a greater degree. Errors are 95% confidence intervals of the difference. Mono. = Monocot, a combined grass and sedge category. Overall diet composition differed substantially between unharvested forest and 5 year old sites (Figure 3.2 A, MRPP: A = 0.20, P <0.001). Relative to wallabies living at 5 year old sites, individuals at unharvested sites ate more forb and less fern and tree (Table 3.1), and this was reflected by the strong linear correlations between these food types and ordination Axis 2 (Table 3.2). Unharvested forest and 5 year old sites were more similar in terms of overall selection, although a statistical difference was still observed (Figure 3.2 B, MRPP: A = 0.05, P = 0.04). The effect size index A showed the observed difference in diet selection was about four times smaller than the difference in diet composition. On the basis of linear trends, the difference was associated with shrub and forb selection, with wallabies in the unharvested forest selecting shrub more strongly and forb less strongly (Table 3.2). A bubble plot (not shown) indicated that fern selection had a strong non linear association with the difference between groups, with wallabies at 5 year old sites selected fern more strongly. 49
Unharvested forest 5 yr old sites A
Tree Fern
Axis 2 Shrub
Forb
Axis 1 B
2
Forb Axis 2 Tree
Shrub
1
Axis 1 Figure 3.2. Ordination diagrams representing (A) the use (consumption) and (B) the selection of plant groups by wallabies. Vectors (solid lines) represent the strength and direction of the linear correlation between individual food types and the ordination axis (see Table 3.2 for coefficients). The labels 1 and 2 in diagram (B) indicate outlying points (see text for details). Diagram (A): Stress = 14.8, MRPP: A = 0.20; P <0.001. Diagram (B): Stress = 10.5, MRPP: A = 0.05, P = 0.04. 50
Table 3.2. Linear correlations (Pearson’s correlation coefficient r) between food types and the ordination axis for both diet composition and selection. Food types with correlations < 0.70 are not shown.
Plant Group Correlation Correlation with Axis 1 ( r) with Axis 2 ( r) Diet composition Tree 0.19 0.95 Fern 0.00 0.89 Forb 0.77 0.72 Shrub 0.85 0.19 Diet selection Shrub 0.11 0.78 Forb 0.72 0.70 Tree 0.89 0.31 Wallabies at 5 year old sites (filled circles) formed two distinct clusters along Axis 2 in Figure 3.2 B. The group of four lower on Axis 2 selected shrub to a greater extent and forb to a lesser extent than their counterparts. The labelled points in Figure 3.2 B represent uncharacteristically strong selection for tree (point 1) and fern (point 2), and are influenced by the sensitivity of selection indices to very low availability values (Lechowicz 1982). The removal of point 1 had little effect on the analysis, but the removal of point 2 resulted in a approximate 40% increase in the difference between unharvested and 5 year old sites ( A = 0.07 after removal). In Stage 1 of the analysis of frequency dependence, standardised simple linear regression coefficients and their associated 95% confidence intervals (Table 3.3) indicated negative but imprecise relationships between selection and relative availability for ferns, forbs and trees, while shrub selection was clearly independent of relative abundance. In the case of monocots, there was a moderate but imprecise negative relationship and the inference remains uncertain. In Stage 2 of the analysis (Table 3.4), relative availability was a substantial negative predictor of selection for every food type except for shrub, while in most cases the three forage quality variables were poor predictors. On a number of occasions (e.g. monocot model, shrub model) one or more of the forage quality variables had a moderate 51 negative influence on selection, but the a priori expectation was that correlations would be positive if a causal relationship existed. In the case of the tree model, however, both nitrogen and water content showed moderate positive correlations with selection, indicating that along with relative availability, these variables may have influenced the selection of tree foliage to some degree. The coefficients in Table 3.4 are standardised and are directly comparable to those in Table 3.3. Table 3.3. Univariate models of frequency dependence for each of the 5 food types. For each food type, the model is log 10 B = a + β(relative abundance), where B is an index of selection. Coefficients ( β) have been standardised for comparison with other analysis. 95% Low. and 95% Upp. are the 95% lower and upper confidence limits of β. n = 22 in all cases. Food Type r2 (Adj. r2) βββ 95% Low. 95% Upp. Fern 23.3 (19.5) 0.483 0.882 0.084 Forb 31.3 (27.9) 0.560 0.946 0.174 Monocot 8.1 (3.5) 0.284 0.730 0.162 Shrub 1.8 (0.0) 0.133 0.596 0.330 Tree 26.9 (23.2) 0.519 0.917 0.120 Stage 3 of the analysis (Table 3.5) presents multivariate models relating diet selection of each food type to the relative availability of all foods. The original models constructed using Manly’s (1973) technique showed, in every case, selection of a food was independent of its own relative availability but positively correlated with the relative availability of at least one other type. In the case of the forb and tree models, however, the strong correlation between forb and tree relative availability make the coefficients associated with these food types difficult to interpret. After removing the tree data from the forb model and the forb data from the tree model, selection of both food types was negatively correlated with its own relative availability. 52
Table 3.4. Multivariate models for each food type comparing the effect of availability and forage quality on diet selection. For each food type the model is log 10 B = a + β(relative abundance) + β(nitrogen) + β(water) + β(digestibility) where B is an index of selection. Coefficients ( β) have been standardised. 95% Low. and 95% Upp. are the 95% lower and upper confidence limits of β. Predictors βββ 95% Low. 95% Upp. Partial r2 Fern model. r 2 =30.1, Adj. r 2 = 12.6, n = 21 Rel. availability 0.537 1.009 0.065 24.7 Nitrogen 0.060 0.469 0.589 0.3 Water 0.244 0.781 0.293 3.9 Digestibility 0.126 0.351 0.603 1.3 Forb model. r 2 =50.2, Adj. r 2 = 37.8, n = 21 Rel. availability 0.517 0.915 0.119 22.9 Nitrogen 0.234 0.638 0.170 4.6 Water 0.272 0.651 0.107 7.0 Digestibility 0.199 0.207 0.605 3.2 Monocot model. r 2 =38.3, Adj. r 2 = 22.8, n = 21 Rel. availability 0.542 1.019 0.065 21.8 Nitrogen 0.315 0.727 0.097 9.9 Water 0.416 0.872 0.040 17.2 Digestibility 0.190 0.254 0.634 3.1 Shrub model. r 2 =26.9, Adj. r 2 = 0.3, n = 16 Rel. availability 0.255 0.414 0.924 4.5 Nitrogen 0.046 0.593 0.501 0.2 Water 0.355 0.936 0.226 11.4 Digestibility 0.398 0.909 0.113 18.5 Tree model. r 2 =66.5, Adj. r 2 = 56.9, n = 19 Rel. availability 0.576 0.944 0.208 26.1 Nitrogen 0.425 0.076 0.774 15.8 Water 0.394 0.037 0.750 13.0 Digestibility 0.042 0.289 0.373 0.2 53
Table 3.5. Multivariate linear models for frequency dependent selection constructed using the technique of Manly (1973) – see text for details. For each model, the response variable is the selection index B and the predictors are the relative availability of each food type (the Ps). Due to the high correlation between forb and tree relative availability ( r = 0.71) these two models were re run using traditional multiple regression after excluding the highly correlated predictor. 95% Low. and 95% Upp. are the 95% lower and upper confidence limits of the coefficient, and are only shown for coefficients with a non zero effect. Predictors Coefficient 95% Low. 95% Upp. Fern model, n = 21 B = 0.200 Pforb + 0.090 Pmono 0.359 Pshrub 0.155 Pfern +0.389 Ptree Tree availability 0.389 0.266 0.513 Forb model, n = 21 B = 0.077 Pforb – 0.092 Pmono + 3.966 Pshrub + 0.330 Pfern + 0.379 Ptree Shrub availability 3.966 2.940 4.992 Tree availability 0.379 0.218 0.540 Fern availability 0.330 0.041 0.619 Forb model excluding tree B = 0.379 – 0.003 P forb – 0.005 Pmono + 0.036 Pshrub – 0.001 Pfern Shrub availability 0.036 0.025 0.047 Monocot availability 0.005 0.008 0.002 Forb availability 0.003 0.006 0.000 Monocot model, n = 21 B = 0.061 Pforb – 0.053 Pmono + 0.452 Pshrub + 0.125 Pfern +0.015 Ptree Shrub availability 0.452 0.209 0.695 Fern availability 0.125 0.046 0.205 Shrub model, n = 22 B = 0.199 Pforb + 0.671 Pmono – 0.461 Pshrub + 0.248 Pfern +0.215 Ptree Monocot availability 0.671 0.112 1.230 Tree model, n = 20 B = 0.172 Pforb + 0.093 Pmono – 0.023 Pshrub + 0.018 Pfern 0.019 Ptree Forb availability 0.172 0.009 1 0.353 Tree model excluding forb B = 0.172 – 0.001 Pmono – 0.002 Pshrub – 0.002 Pfern – 0.002 Ptree Tree availability 0.002 0.004 0.000 1 Interpreted as strong evidence for a positive correlation between tree selection and forb relative abundance as the lower confidence limit only just overlaps 0. 54
Discussion In this study, I used vegetation change resulting from timber harvesting to investigate diet composition and selection in a wild population of swamp wallabies. Diet composition differed substantially between unharvested forest and 5 year old sites, and the results confirm the status of the swamp wallaby as a generalist herbivore that consumes a wide variety of foods, including low quality forage such as austral bracken and Eucalyptus foliage (Edwards and Ealey 1975; Hollis et al. 1986). The ability to process low quality foods is consistent with the species’ common status and wide geographic distribution (Menkhorst 1995), and the observed westward expansion of its range into southwestern Victoria (Bird 1992). In general, the findings were similar to the only other study of diet selection for this species (Wood 2002), in which shrub and forb was also identified as important dietary components. Wood (2002) used faecal pellets to derive selection indices for plant functional groups within two vegetation communities during two seasons, and observed differences in selection associated with both factors. Although no formal analysis was conducted, an increase in forb selection during summer was associated with reduced availability, and this is consistent with the negative relationship between forb selection and availability demonstrated here. A marked change in overall diet composition between unharvested and 5 year old sites was mirrored by a smaller but distinct difference in selection. The latter effect was primarily driven by changed selection for forb, shrub, tree and fern, and the selection ranking of tree and fern changed substantially between sites. Nevertheless, at both unharvested and 5 year old sites, wallabies consistently selected forb and shrub over monocot and tree, demonstrating that generalist mixed feeders may still exhibit distinctive foraging choices (Tixier et al. 1997). It is important to note, however, that different degree to which shrub was selected at unharvested and 5 year old sites may have been influenced by the presence of different species. While most unharvested sites had a mix of shrubs including silver wattle, prickly wattle, common heath, gorse bitter pea and common cassinia, the shrub flora at 5 year old sites was dominated by silver wattle, and this difference confounds the selection results to some degree. 55
My initial prediction, that wallabies would be mixed feeders and demonstrate negative frequency dependence, was strongly supported by univariate models for fern, forb and tree. In the case of shrub, the absence of any relationship may have been due to the small range (≤ 20.5%) of sampled relative availability values, as shrubs were never a dominant vegetation type. The multivariate models that included forage quality variables suggested similar patterns, although in this case monocot also demonstrated negative frequency dependence. Recently, Bergvall and Leimar (2005) showed the effect of relative availability on selection may be mediated by forage quality, but, with the exception of tree, the forage quality indicators used in this study did not appear to influence selection, strengthening the inference regarding relative availability. However, the positive correlation between the selection of tree foliage and both its nitrogen and water content suggested the observed pattern of tree selection was influenced by both relative availability and forage quality, although relative availability appeared to have a stronger effect. Manly’s (1973) method for the analysis of frequency dependence indicated that selection of all food types except for tree was positively related to the relative availability of other types. Evidence for negative frequency dependence existed for tree and forb, but not for the other food types. This result confirms the intuitive expectation that diet selection is a complex process that depends, amongst other factors, on the availability and quality of alternative forage, although I do not discount the possibility of spurious results given the relatively large ratio of predictors to experimental units. In the context of commercial forestry, Welch et al. (1991) found a negative correlation between the cover of ericoid plants and browsing damage to tree seedlings while Codron et al. (2006) argue that contrasting amounts of grass in the diet of two groups of African elephants ( Loxodonta africana ) was directly related to the availability of alternative forage species. Selection of a food type may be influenced by the nutrient status of other foods (Moser et al. 2006) and, presumably, by interactions between nutrients, toxins and availability. Swamp wallabies have been found to consume substantial amounts of grass in spring when, relative to other available food types, its nitrogen concentration was high (Osawa 1990). Data across seasons incorporating selection, availability and information about forage quality (e.g. Forsyth et al. 2005) are required to investigate these potential effects further. 56
The three analyses of frequency dependence present an increasingly complex picture of the factors influencing diet selection. Overall, the data provide substantial support for negative frequency dependence and a mixed feeding strategy, which may be driven by a need to consume a variety of nutrients (Pulliam 1975; Westoby 1974) or minimise the detrimental effects of plant toxins. In no case did the data support diet specialisation (positive frequency dependence), and on this basis it appears that the foraging strategy of swamp wallabies does not conform to the predictions of optimal foraging theory. The general lack of support for the diet specialisation is not particularly surprising, particularly for a mobile herbivore like the swamp wallaby. Theories predicting specialisation on preferred foods assume that the time taken to search for and handle food items are mutually exclusive (Pyke et al. 1977) but in most cases large, mobile herbivores can search for the next prey item while handling the previous one (Spalinger and Hobbs 1992). In addition, searching costs are likely to be very low for animals that consume large quantities of abundant food (Westoby 1974) and thus may have little effect on food choice. These and other factors (e.g. Westoby 1978, Perry and Pianka 1997, Illius et al. 2002) may render predictions of diet specialisation inappropriate on many occasions. Finally, the results of this study relate to the consumption of broad plant groups in autumn within areas defined by home range boundaries and should not be extended to other times or scales of measurement. Presumably wallabies also selected species within plant groups and patches within home ranges, thus data at a finer resolution may reveal different patterns. Collecting data at the scale at which behaviour takes place will facilitate accurate inference (Wiens 1989), but it is not always obvious what that scale is, or indeed if multiple scales are important. Studies at a variety of scales, and consistency of results between them, can strengthen the generality of findings from any one investigation (Ward and Saltz 1994, Bowyer and Kie 2006). In addition, theories that predict foraging behaviour are not defined in terms of scale, and different scales of investigation can generate inconsistent results. For example, Moser et al. (2006), found that foraging by roe deer within 1 × 20 m patches to be consistent with optimal foraging models while the same species foraging on individual branches did not behave optimally (Illius et al. 2002). Fortin et al. (2003) also demonstrated that the foraging behaviour of bison ( Bison bison ) was more consistent with optimal foraging theory at a small scale than at a larger scale. In contrast, Ward and Saltz (1994) found that dorcas 57 gazelles ( Gazella dorcas ) foraged optimally, and did so at both larger and smaller spatial scales. Patterns of selection observed for swamp wallabies may also change with season (Wood 2002), particularly in response to temporal changes in nutrient status of available forage (Osawa 1990). Seasonal changes in selection have been observed for agile wallabies (Macropus agilis ) in the monsoon tropics of northern Australia, a species that selected grasses and forbs in the wet season when their quality was high but substantially broadened their diet in the dry season as the quality of these preferred foods decreased (Stirrat 2002). It is also possible that generalist herbivores may switch their foraging strategy from mixed feeders to diet specialists on the basis of seasonal changes in forage quality (Lundberg et al. 1990). For swamp wallabies, it will be instructive to see if the current support for models predicting mixed feeding is also found in different seasons and when data are collected at a finer resolution. 58
CHAPTER 4
Effect of habitat type, sex and diel period on the space use of swamp wallabies ( Wallabia bicolor )
Abstract For mobile animals, determining movement patterns at different spatial and temporal scales, and among population classes defined by sex, age or status, can help reveal behavioural complexities and refine tests of ecological theory. In this study I used timber harvesting operations as experimental treatments to investigate the effect of habitat type, sex and time of day on the space use of swamp wallabies ( Wallabia bicolor ). I radio tracked 42 wallabies (24 females and 18 males) at 18 sites that conformed to one of four habitat types: unharvested forest (controls), recently harvested forest, and young and old regeneration (5 and 10 years post harvest respectively). The difference between diurnal and nocturnal space use was quantified by calculating the change in home range size attributable to night locations. I used Restricted Maximum Likelihood (REML) to model the effect of treatment on males and females separately, using body weight as a covariate for males. Relative to wallabies living in unharvested forest, the addition of nocturnal locations to diurnal data increased range size by around 30% for female wallabies at both young and old regenerating sites. Relative to controls, male range size (adjusted for body weight) changed little at recently harvested and young regenerating sites, and appeared to decrease by around 20% at old regenerating sites. Male body weight was negatively correlated with range size increase which explained 23% of the variance in data. No relationship existed for females. The results suggest that reinterpretation of findings from previous studies of swamp wallaby space use may be warranted, and have implications for the design of future research for this and other species with similar diel behavioural patterns. 59
Introduction Mobile animals often use multiple habitat patches at a variety of spatial and temporal scales to access resources (Johnson 1980; Law and Dickman 1998; Orians and Wittenberger 1991). Using different spaces during diurnal and nocturnal periods is one example of temporal differences in resource use, a pattern that has been identified in numerous terrestrial species, including birds (e.g. Shepherd and Lank 2004), bats (e.g. Law 1993), ungulates(e.g. Ager et al. 2003), carnivores (e.g. Comiskey et al. 2004) and macropodid marsupials (e.g. Fisher 2000; Johnson 1980; le Mar and McArthur 2005). Differentiating between diurnal and nocturnal space use can help to reveal behavioural complexities (Ager et al. 2003), and can refine tests of ecological theory (Fisher 2000), yet nocturnal data are lacking for many species. The swamp wallaby ( Wallabia bicolor ), for example, is a medium sized macropodid marsupial that occur throughout eastern and south eastern Australia. To date, descriptions of their movements and habitat use have been based on daytime observations (Kaufmann 1974), counts of faecal pellets (de Munk 1999; Floyd 1980; Hill and Phinn 1993; Lunney and O'Connell 1988; Ramsey and Wilson 1997) and radio location data (Edwards and Ealey 1975; Troy and Coulson 1993; Troy et al. 1992; Wood 2002). On the basis of these data, the species has been reported to have relatively small home ranges (around 15 to 30 ha) and to prefer dense vegetation. However, faecal pellet studies cannot discriminate between nocturnal and diurnal behaviour and the radio tracking studies had few if any nocturnal locations, so whether the space use of swamp wallabies differs between diurnal and nocturnal periods is unknown. In this study, I use both nocturnal and diurnal radio tracking data to quantify the effect of nocturnal locations on home range size for wallabies with access to 5 and 10 year old regenerating eucalypt forest, recently harvested sites and unharvested control locations. On the basis of data from other macropods (Arnold et al. 1992; Fisher 2000; Johnson 1980; le Mar and McArthur 2005; Stirrat 2003b; Vernes et al. 1995), and my expectation that swamp wallabies will seek resources in open and closed habitats during different parts of the 24 hour cycle, I predict that diurnal and nocturnal space use will differ for individuals with access to both dense and open habitats, but not for those with access to only one habitat type. 60
Methods Study site The data were collected from the Pyrenees State Forest in west central Victoria, described previously in Chapter 2. I used unharvested forest and three age classes of regenerating forest (recently harvested, about 5 years post harvest and about 10 years post harvest) to define four habitat types that differed most obviously with respect to vegetation structure and lateral cover. Except for retained seed trees and some remaining ground vegetation, the recently harvested areas were initially devoid of live plant biomass, although some growth occurred during the monitoring period. The 5 year old sites were dominated by 1 3 m tall, densely regenerating eucalyptus seedlings, but also had substantial quantities of silver wattle, grass and forbs. Ten year old sites supported dense, closed stands of 3 6 m tall eucalypt regeneration and consequently had low levels of forb, grass and shrub cover. The cover of bracken, however, remained relatively high. Unharvested sites were relatively open and thus supported substantial quantities of forbs, grass and occasional patches of shrub. They tended to be further from major ridge lines and on steeper slopes than harvested sites and the density and size of mature trees varied depending on local harvesting history. Experimental design I used timber harvesting operations as an experimental treatment to investigate the effect of habitat type, sex and time of day on wallaby movements. The location of unharvested control sites was limited by road access and proximity to each other and non control areas (see Chapter 2). Initially I identified around 15 potential control locations within the area used for timber harvesting (above approx. 500 m elevation) and attempted to trap wallabies at 14 of these. Five sites were subsequently rejected due to unsuccessful trapping regimes (three), and wallabies using other (non control) areas (two). Wallabies caught at the remaining nine control sites were used in the analysis. I used all available recently harvested sites ( n = 2) and 5 year old sites ( n = 5), and selected three 10 year old sites randomly from six that we deemed appropriate. A 61 number of 10 year old sites were rejected a priori due to their close proximity to sites where wallabies had already been captured. The spatial arrangement of sites is shown in Figure 4.1. With two exceptions, sites were at least 1.5 km apart to increase the likelihood of spatial independence. In one case, a 5 year old and 10 year old site were separated by about 500 metres and in another, two 5 year old sites were 300 metres apart at their closest point. Of the 13 wallabies tracked at these adjacent sites, two moved between them. On two additional occasions, minor home range overlap (<10%) occurred between animals at other adjacent sites. At each site I obtained location data (see next section) for between one and four wallabies, and sampled both sexes whenever possible. In all, the data represented 42 wallabies (24 females and 18 males) from 18 sites.
N Pyrenees State Forest
Control Recent Harvest 5 yr Regen 10 yr Regen 5 km
Figure 4.1. The Pyrenees State Forest, western Victoria, showing the spatial arrangement of the study sites. 62
Wallaby capture and radio tracking I trapped wallabies from March 2004 to October 2005 using the methods described in Chapter 2. The total trapping effort constituted 897 trap nights for 102 captures (11.4% trap success), which included 94 individuals with eight recaptures. The methods for radio tracking are also described in Chapter 2. Most wallabies were tracked for between 5 and 10 months (minimum = 3, maximum = 15) and the majority of locations were spread relatively evenly over the Spring, Summer and Autumn (September to May) of either 2004/2005 or 2005/2006. The exception were wallabies at the recently harvested sites who were tracked from May to December 2005. The final data set contained 2163 positions, 78.9% ± 0.9 (mean ± 95% CI) collected during the day and 21.1% ± 0.9 collected at night. The mean number of positions per individual was 51.5 ± 3.9, 40.6 ± 3.1 of these representing diurnal locations. The accuracy rating system described in Chapter 2 was used, and the percentage of locations given each rating in both diurnal and nocturnal periods is shown in Table 4.1. I removed a single rating 4 location from 5 individuals as these locations resulted in range size increases of ≥10%. Table 4.1. Subjectively estimated accuracy of radio tracking locations. Values are the percentage of locations (± 95% confidence interval) relating to a 1 – 5 scale of spatial accuracy; 1 = within 5 m of exact location, 2 = 5 – 25 m, 3 = 25 – 100 m, 4 = 100 – 200 m and 5 = >200 m, or when a signal could not be detected. Time Period % of Locations Given Each Accuracy Rating 1 2 3 4 5 Diurnal 42.2 ± 7.8 25.2 ± 4.1 26.4 ± 5.3 4.9 ± 1.5 1.3 ± 0.7 Nocturnal 34.7 ± 7.3 27.7 ± 4.9 30.5 ± 5.4 4.4 ± 2.0 2.7 ± 1.8 63
To differentiate between diurnal and nocturnal space use I calculated the change in home range size attributable to night locations using the following formula: Percentage
Change in Home Range Size = ( HR T – HR D) / HR D × 100 where HR T is total range size and HR D is diurnal range size. Diurnal range size was used as the denominator in the equation to standardize results for animals that had markedly different range sizes. Data analysis
I used Restricted Maximum Likelihood (REML) to model the effect of treatment on males and females separately. Separate analyses were necessary as sex and weight were confounded and could not be entered into the same model. A consequence of this approach is the absence of a formal interaction test for treatment and sex, so the interaction was assessed qualitatively using graphs of the data. Initially, site was entered into each model as a random factor but subsequently excluded due to its insubstantial effect (variance component equal to zero). Body weight was tested as a covariate for both sexes, and was included in the analysis of male data. .I tested assumptions of normality and homogeneity of variance with a half normal plot and a fitted value plot respectively, and deemed data transformation to be unnecessary. Additional assumptions relating to the covariance analysis were met after removing an outlier. I also investigated sample size as an additional covariate, but it did not explain a substantial component of the variance. Results Raw home range data are presented in Appendix 1. With the addition of nocturnal data, the home range size of female wallabies increased by around 40% at 5 year old and 10 year old sites, while the change in male range size, adjusted for body weight, was generally much smaller (Figure 4.2). Differences between male and female responses at 10 year old sites appears to be driving a treatment by sex interaction, but this may be an artefact of inadequate sample size ( n = 2 for males at 10 year old sites). 64
70 Females 60 Males 50 40 30
20 * 10 0 10 * 20 30 Control Recently 5 yr Regen. 10 yr Regen. Harvested Figure 4.2. The degree to which home range size increases with the addition of nocturnal data for each of four habitat types. Data are expressed as a percentage of diurnal range size to standardize results for animals that had markedly range sizes. * Values may be inaccurate due to small sample sizes ( n = 2 in both cases). The effect of habitat type is shown most clearly by comparing each of the harvested site classes with the unharvested control (Figure 4.3). In this figure, the point estimates for each comparison are calculated as the value of the first listed habitat minus the control value, so a positive estimate represents an increase relative to controls. Relative to wallabies living in unharvested forest, the addition of night locations to diurnal data sets increased female range size by around 30% for wallabies at both 5 year old and 10 year old sites. Relative to controls, the range size of males at 10 year old sites appeared to decrease by about 20%, but this may have been influenced by the small sample size noted above. There was a negative linear relationship (Figure 4.4) between body weight and home range size increase for males (Adj. r2 = 0.23), indicating that range sizes increased more for lighter individuals. There was no relationship for females (Adj. r2 = 0). 65
COMPARISON Change in Home Range Size Recent Harv vs Cont * Males 5 yr vs Cont Females 10 yr vs Cont * Recent Harv vs Cont 5 yr vs Cont 10 yr vs Cont 50 40 30 20 10 0 10 20 30 40 50 60 Mean Difference (%) Figure 4.3. Pair wise comparisons between each class of harvested site and the controls. Point estimates are calculated as the value of the first listed habitat minus the control value, so a positive number represents an increase relative to the control. Error bars are 95% confidence intervals. * Values may be inaccurate due to small sample sizes. 80 Male s 70 Females 60 50 40 30 20 10 0 10 20 10 12 14 16 18 20 22 24 Body Weight (kg) Figure 4.4. Relationship between home range size increase and body weight. The fitted line (solid) and its 95% confidence interval (dashed) describe the relationship for males; y = 79.2 – 3.5 weight, Adj. r2 = 0.23. The relationship does not exist for females (Adj. r2 = 0). 66
Discussion My initial prediction, that diurnal and nocturnal space use would differ for individuals with access to both dense and open habitats, was supported by data from females but not from males. For females there were substantial differences between diurnal and nocturnal movements when wallabies had access to both dense post harvest regeneration and adjacent unharvested areas. This pattern was not evident for males, and may be explained by the greater influence of reproductive efforts on male space use (Clutton Brock 1989; Fisher and Lara 1999). A fuller discussion regarding the mechanisms that may result in differential use of space by male and female mammals is included in Chapters 5 and 6. Because intra specific ranging behaviour is affected by access to resources (Gompper and Gittleman 1991; McLoughlin and Ferguson 2000), the range size of swamp wallabies may differ between sites and populations. In addition, the data presented here show that at a particular site, space use can differ substantially between diurnal and nocturnal periods. This is consistent with previous findings for macropodid marsupials (Arnold et al. 1992; Fisher 2000; Johnson 1980; le Mar and McArthur 2005) and other species (Ager et al. 2003; Beyer and Haufler 1994), but, as demonstrated by the control animals, may not be the case in every situation (Gaines and Lyons 2003; Perelberg et al. 2003). Relative to controls, range size increased up to 30% when nocturnal data (around 20% of the sample) was added to diurnal positions. I suspect that the increase would have been even greater if more nocturnal data were available. Published data describing home ranges for swamp wallabies are limited to two studies (Edwards and Ealey 1975; Troy and Coulson 1993), both conducted at the same site, Coranderrk Bushland, in southern Victoria. Both studies included some nocturnal positions, but given the heterogeneous nature of that study site, it seems likely that the reported range sizes were underestimates of the total space used. This may also be the case for the diurnal home ranges presented in the unpublished studies by Wood (2002) and Ben Ami (2005) in other parts of the species’ geographic range. 67
The use of different spaces during the day and night has important implications for experimental design (Beyer and Haufler 1994). If there is an expectation that diurnal and nocturnal differences exist, then a representative sample of both night and day locations should be collected (Aebischer et al. 1993). If an objective is to examine the differences between day and night behaviour, then samples sizes must be adequate to calculate both diurnal and nocturnal indices. If theoretical predictions about range size or habitat use are being tested, consideration should be given to whether diurnal, nocturnal or pooled data provide the best test. Consider, for example, testing the hypothesis that intra specific range size will be negatively related to resource availability. This hypothesis might be tested by examining correlations between home range size and measures of resource availability for a sample of individuals (Fisher 2000; Tufto et al. 1996). An important consideration, however, is how perceptions of resource availability differ between diurnal and nocturnal periods. Shelter and food are important resources for most animals, but their relative importance may differ throughout the 24 hour cycle. For many species, including macropodid marsupials, shelter is an important daytime resource but, relative to food, becomes less important at night (Arnold et al. 1992; Evans 1996; Johnson 1980; le Mar and McArthur 2005; Wahungu et al. 2001). In such cases, the resources affecting range size are likely to differ during diurnal and nocturnal periods, so the use of total ranges (pooled diurnal and nocturnal data) to quantify relationships may obscure important biological effects. This was demonstrated by Fisher (2000), who found that the total range size of bridled nailtail wallabies (Onychogalea fraenata ) was negatively correlated with food availability, but this relationship disappeared when only diurnal ranges were used. Diurnal range size, however, was negatively associated with vegetation density, presumably because shelter was more important during the day. I would expect the relationship between range size and food to have been strongest for nocturnal ranges, but these data were not presented. Although regression techniques have often been used to determine relationships between home range size and various predictor variables (Dahle and Swenson 2003; Fisher 2000; Gompper and Gittleman 1991; Grigione et al. 2002; Tufto et al. 1996), manipulative experiments are a better way to infer causal relationships and examine interactions (Johnson 2002). In studies of space and habitat use it would be instructive 68 to manipulate food and shelter resources and examine their interaction with factors such as sex and time of day. While manipulation of food availability is relatively common (Boutin 1990 and references therein; Cooper et al. 2006; Jonsson et al. 2002) shelter is more difficult to control, particularly at scales relevant to free ranging mammals. Timber harvesting operations, however, provide an opportunity to manipulate shelter at large spatial scales, as they result in patches of dense regeneration surrounded by more open unharvested forest. Manipulating food resources within or adjacent to regenerating patches could unravel complexities associated with ranging behaviour and habitat use of free ranging mammals. There is considerable scope for the development of such experiments in the future. 69
CHAPTER 5
Habitat selection by the swamp wallaby ( Wallabia bicolor ) in relation to diel period, food and shelter
Abstract Timber harvesting results in patches of regenerating forest that are substantially different from surrounding unharvested stands, and provides an opportunity to investigate the effect of habitat change on forest fauna. In this study, I used recently harvested sites (0 8 months post harvest), 5 and 10 year old sites and adjacent unharvested stands in a southeastern Australian Eucalyptus forest to investigate the habitat selection of swamp wallabies ( Wallabia bicolor ), a medium sized ground dwelling mammal. Non metric multidimentional scaling indicated that habitats differed most markedly with respect to visibility (an indicator of lateral cover) and a forage quality index, providing the basis for non random habitat selection. At the population level, wallabies selected 5 year old sites to a much stronger degree than other habitat, although both 5 and 10 year old sites were selected more than expected on the basis of their availability. At the scale of individual wallabies, a pooled male and female sample demonstrated selection for forest over recently harvested sites during both diurnal and nocturnal periods. A separate analysis comparing patterns of selection between 5 and 10 year old sites and the surrounding forest found that selection was influenced by sex and diel period. In general, females demonstrated more distinct choices than males. Both sexes selected 5 year old sites over the surrounding forest during the day, but only females did so at night. Females selected 10 year old sites over surrounding forest during the day, but selected the forest at night. Males also selected forest over 10 year old sites at night, but showed no selection for either habitat during the day. Both sexes selected 5 over 10 year old sites during both diurnal and nocturnal periods. 70
Diurnal habitat selection by both sexes was associated with low visibility. Nocturnal selection of females was associated with high values of a forage quality index, but this was not the case for males. For female wallabies in particular, habitat selection appeared to be influenced by predator avoidance behaviour and food acquisition, although these factors were not always in conflict. 71
Introduction Disturbance events have a major effect on ecosystems by creating a mosaic of patches at different successional stages (Moloney and Levin 1996). The resulting heterogeneous and temporally changing environment provides animal populations with complex choices with respect to habitat selection. Selection of high quality habitat is often a response to improved food resources (Geffen et al. 1992), shelter resources (McCorquodale 2003) or both (Tufto et al. 1996). Nevertheless, selection may be mediated by other factors such as predation and competition (Hughes et al. 1994), or sex and reproductive status (Johnson and Bayliss 1981). Because the factors influencing habitat selection are often spatially or temporally hierarchical (Rettie and Messier 2000), assessments at multiple scales may be required (Bowyer and Kie 2006). Although infrequently quantified (but see Nilsen et al. 2004) it is often assumed that selection of higher quality habitats is directly related to reproductive success, and thus selection is viewed as a surrogate for Darwinian fitness (Krebs and Davies 1993). Timber harvesting is a disturbance event that resets the successional process and results in patches of differentially aged regenerating forest scattered throughout the landscape. At the stand scale, the effects of harvesting on habitat selection are potentially complex, involving the interaction between a number of factors. For example, black bears ( Ursus americanus ) avoided recently harvested stands due to the absence of den trees, even though they contained more food and shelter than other habitats (Mitchell and Powell 2003). In addition, two sympatric macropodid marsupials, red bellied pademelons (Thylogale billardierii ) and red necked wallabies ( Macropus rufogriseus rufogriseus ), both selected a newly planted Eucalyptus plantation for night time foraging but selected unharvested forest and older plantation respectively as day time resting areas, possibly due to differences in predator avoidance strategies (le Mar and McArthur 2005). At the landscape scale, harvesting may affect selection by either diminishing or increasing the supply of important resources. For example, some forms of harvesting may reduce the number of trees with hollows (Gibbons and Lindenmayer 1996), and thus influence habitat choices of hollow dependent fauna. On the other hand, early successional forests associated with harvesting operations or plantation establishment provide abundant food resources for many ungulates and are a contributing factor to high density deer populations throughout the world (Côté et al. 2004). 72
In the native forests of southeastern Australia, common harvesting methods (Lutze et al. 1999) create 10 to 30 ha patches of regenerating vegetation within much larger unharvested stands. Sites are generally covered with regenerating Eucalyptus seedlings and other native vegetation three to five years after harvesting, but by ten years post harvest the abundance of early successional species has diminished and stands are dominated by close growing Eucalyptus saplings and silver wattle ( Acacia dealbata ), often with dense patches of austral bracken ( Pteridium esculentum ). These temporal changes provide an opportunity to study the effects of altered habitat quality on the habitat selection of forest fauna. In this study, I used timber harvesting as an experimental treatment to investigate the effect of vegetation structure and forage quality on the habitat selection of the swamp wallaby ( Wallabia bicolor ). These medium sized macropodid marsupials are widely distributed within the forests of southern and eastern Australia and, on the basis of limited diurnal data, have been shown to select densely vegetated areas over open ones (Troy et al. 1992; Wood 2002), and appear to use sites that contain access to both shelter and food (Hill and Phinn 1993). In the context of native forest timber harvesting, the species responds positively to densely vegetated one to two year old regenerating areas (Di Stefano 2005), but the effect has never been quantified or examined in relation to regenerating sites of other ages. My first objective was to evaluate differences in habitat quality between native unharvested forest and patches that had been harvested recently (about three months prior to data collection) and five and 10 years previously. My second objective was to quantify habitat selection (defined as the analysis of use in relation to availability) of swamp wallabies living in a harvested landscape at two spatial scales. Selection was assessed at the population level using faecal pellet counts and at the level of individual wallabies using radio telemetry. The telemetry data were also used to quantify diurnal and nocturnal differences in selection patterns. My third objective was to relate patterns of habitat selection to habitat quality, using variables describing habitat structure, forage abundance and forage quality. On the basis of past studies of habitat use (Floyd 1980; Hill and Phinn 1993; Lunney and O'Connell 1988; Ramsey and Wilson 1997; Troy et al. 1992; Wood 2002) I expected selection for densely vegetated habitats during diurnal periods, but did not necessarily expect this pattern to be replicated at night. I also 73 predicted that diurnal selection would be related to shelter while nocturnal selection would be related to forage quality. Methods Study site This study was conducted in the Pyrenees State Forest which has been described in Chapter 2. The design and protocol for site selection are described in Chapters 2 and 4, although I only used the recently harvested ( n = 2), 5 year old ( n = 5) and 10 year old ( n = 3) sites for this study (Figure 5.1). These sites defined three habitat types and the unharvested forest surrounding them defined a fourth. The unharvested forest included a number of small selectively harvested patches and some areas burnt as part of past fuel reduction operations, but these were retained as I considered them to be a component of the heterogeneity within unharvested stands. Two patches of recently burnt forest totalling about 70 ha were removed from the analysis as they were only used by four wallabies. Wallaby capture and radio tracking A total of 31 wallabies (between two and five at each site; 18 females and 13 males) were caught, radio tracked and their home ranges established using the methods described in Chapter 2. The length and timing of the tracking period is described in Chapter 4. Each wallaby was treated as an experimental unit and its locations used to define habitat use during both the day and night, with night being defined as completely dark. Although there was some error associated with location accuracy (see description of accuracy rating system in Chapter 2), approaching wallabies along habitat edges facilitated accurate identification of the occupied habitat type in the vast majority of cases. The proportion of radio tracking locations given each accuracy rating was almost exactly the same as presented in Table 4.1. 74
N Pyrenees State Forest
Recently Harvested 5 yr Regen 10 yr Regen 5 km
Figure 5.1. Pyrenees State Forest showing the spatial arrangement of sites and the polygon used to define habitat availability for the population level analysis. The final data set contained 1495 locations, 78.1% ± 1.0 (mean ± 95% CI) of which were collected during the day. The mean number of locations per individual was 48.2 ± 3.1 of which 37.6 ± 2.4 were diurnal and 10.6 ± 0.9 nocturnal. I removed a total of 58 locations from the analysis; 50 were in the excluded patches of recently burnt forest, in 4 cases the habitat could not be determined, and in a further 4 cases wallabies were located either on roads or in a campsite. Habitat comparisons I quantified the difference between the four habitat types (recently harvested forest, 5 and 10 year old regenerating sites and unharvested forest – see Chapters 2, 3 and 4 for 75 general descriptions) by sampling variables describing lateral cover, forage abundance and forage quality from within previously established home range areas (see Chapter 3 for detailed methodology). The exception was at recently harvested sites where sampling occurred within square 1 ha (harvested areas) or 4 ha (forest) blocks using nine plots established at the intersections of a randomly positioned grid. In addition to abundance and quality variables, lateral cover (hereafter refered to as visibility) was measured by moving away from a 3 cm 3 orange cube positioned 60 cm above the ground at bearings of 0 °, 120 ° and 240 °. The distance at which the block disappeared or became indistinguishable from background vegetation was recorded and averaged for the three bearings, and data from each plot was averaged to generate a single value for each home range, or for each habitat within a home range when more than one was used. At 5 and 10 year old regenerating sites and the adjacent unharvested forest, these data were collected during March and April of 2005 and February and March of 2006, before autumn rain and after a dry summer. At recently harvested sites, data were collected during the winter (August) of 2005, approximately three months after the post harvest burn. I used the forage abundance and quality variables to derive a forage quality index, FQI , for 5 and 10 year old sites and the forest adjacent to them. The index, modified from Moser et al. (2006), combined cover and three measures of quality (nitrogen content, water content and dry matter digestibility) for five plant groups (fern, forb, monocot, shrub and tree) and was calculated as:
n ∑cij × qij × Bi FQI = i=1 n where c and q are the cover and quality respectively of plant group i in habitat j, B is the average selection index for plant group i derived from a study of diet selection at the same study site (Chapter 3), and n is the number of available plant groups. Forage quality was defined as the average combination of nitrogen content, water content and dry matter digestibility and was calculated as: 76
nitrogen water digestibility qij + qij + qij qij = 3 Sedges and grasses were collapsed into a single monocot category because sedges were almost never eaten (see Chapter 3). Because monocot abundance was almost entirely attributable to grass, I used the nitrogen, water and digestibility values associated with this food type to calculate the index. The raw values of some variables differed markedly, so the index was computed using ranks. Habitat selection The study site was digitized in ArcView 3.2 using a 1:4000 aerial photograph. At the population scale, availability of the four habitat types was defined by drawing a polygon around the outermost regenerating sites (Figure 5.1), then calculating the proportion of the area within the polygon belonging to each habitat. The degree to which each habitat was used was estimated by counting the number of whole, clearly identifiable faecal pellets in either 1 × 15m (forest) or 1 × 10m (regenerating sites) plots, established within home ranges as described in Chapter 3. The raw pellet counts were converted to pellets/ha to account for the different sampling intensity between sites. At a smaller scale, habitat availability was defined separately for each wallaby as the proportion of each habitat within a circular area positioned around their geographic centre of location. The circles were 134 ha and 63 ha for males and females respectively, and were based on the average maximum distance from geographic centres to furthest locations in a separate sample of wallabies that lived entirely in unharvested forest (11 females and seven males). I chose not to use home range areas to define habitat availability as, a priori , I expected home range size and shape to be influenced by available habitat (Bowyer and Kie 2006). Habitat use was defined for each wallaby as the proportion of times they were located in each habitat in both diurnal and nocturnal periods. The spatial separation of sites and my definition of availability meant that none of the wallabies had access to all four habitat types. Although this was a biological reality, it precluded a single analysis due to the large number of missing cells in the data matrix. 77
Consequently, analyses were conducted separately for (a) wallabies with access to only recently harvested sites and the surrounding forest ( n = eight; five females and three males) and (b) wallabies with access to either 5 or 10 year old sites (or both) and the surrounding forest ( n = 23; 13 females and 10 males). In the second set of analyses, selection was assessed separately for males and females in both diurnal and nocturnal periods, but sexes were pooled in the first analyses due to the small sample size. Association between habitat selection and habitat quality For each of the 23 wallabies with access to either 5 or 10 year old sites (or both) and the surrounding forest I quantified habitat quality within each accessed habitat ( n = 43) using (a) visibility and (b) the forage quality index. I calculated the selection index B (Manly et al. 2002) for each of the 43 areas, and categorised them as either selected for or selected against during both diurnal and nocturnal periods (a habitat was selected for when B > B/n, where n is the number of available habitat types). Each data point (constituting a measure of selection and values for both visibility and the forage quality index) was then grouped into one of four categories defined by time of day and sex (females day, females night, males day, males night). In each category, values of visibility and the forage quality index were graphed for habitats that were (a) selected for and (b) selected against. Due to interdependencies between the values of B, formal statistical procedures were not used. Data analysis To assess the difference between 5 and 10 year old sites and the forest surrounding them I subjected the variables describing vegetation cover, forage abundance, forage quality and visibility to non metric multidimentional scaling (NMDS) using the Bray Curtis distance measure. Recently harvested sites were not included in the ordination as the large difference between them and the other habitats generated results that did not reflect reality. Missing values were replaced by the mean of non missing values in the same group and the final matrix was subjected to a range standardisation to remove the effect of variables measured on different scales. Pearson’s correlation coefficients were used to assess the strength of the linear relationship between the original variables and the ordination axes and I used a Multi response Permutation Procedure (MRPP; Mielke 78 et al. 1976) to investigate overall (multivariate) differences between habitats. All analyses were conducted using PC ORD 4.25 (McCune et al. 2002). I estimated habitat selection at the population level using a selection index calculated as
ŵi = (ui/u+) / ( Ai/A+) where ui is the number of used resource units in habitat i, u+ is the total number of used resource units, Ai is the number of available resource units in habitat i and A+ is the total number of available resource units (Manly et al. 2002). Values of the index are interpreted in relation to 1, where values >1 and <1 represent selection of habitats more and less than expected in relation to their availability. Habitat selection at the stand scale was quantified using compositional analysis (Aebischer et al. 1993). When habitat types were available but not used, the zero used value was replaced by 0.001. In cases where habitats were neither used nor available, values were treated as missing but replaced by the mean of non missing values to calculate the ranking matrix. The results were presented graphically and a bootstrapping procedure (n =10 000) was used to calculate 95% confidence intervals around the mean log ratio differences between habitat types. Analyses were conducted using Compos Analysis 6.2 (Smith 2004) and PopTools 2.6.7 (Hood 2005). Results
Habitat comparisons The NMDS ordination produced a moderate stress (15.8) two dimensional solution that showed distinct multivariate differences between some habitat types (Figure 5.2). Habitats separated most clearly along Axis 2 with 5 year old sites forming a distinct group at the bottom of the diagram. Selected pairwise comparisons (Table 5.1) demonstrated relatively large differences between unharvested forest and 5 year old sites and between 5 and 10 year old sites, whereas unharvested forest and 10 year old sites were more similar, as were patches of unharvested forest adjacent to 5 and 10 year old sites. 79
5 yr forest 10 yr forest 5 yr regen. 10 yr regen.
Figure 5.2. Ordination of habitat quality variables collected at unharvested forest adjacent to 5 year old regeneration, unharvested forest adjacent to 10 year old regeneration, 5 year old sites and 10 year old sites. Lines (vectors) represent the direction and strength of the four variables with the strongest linear correlations with the ordination axis. FQI is a food quality index. Stress = 15.8; explained variation from distance matrix = 83.1%. Table 5.1. Results of the MRPP analysis for selected comparisons. Unharvested 5 yr and Unharvested 10 yr refer to unharvested forest adjacent to 5 and 10 year old sites respectively, and Unharvested All refers to pooled data from these two groups. The effect size measure ( A) is a better reflection of the difference between groups than the P value as it is less affected by sample size. Comparison Effect size ( A) P value Unharvested 5 yr vs Unharvested 10 yr 0.03 0.07 Unharvested All vs 5yr 0.09 <0.001 Unharvested All vs 10yr 0.02 0.07 5 yr vs 10 yr 0.06 0.001 80
Four variables had strong ( r ≥0.70) linear relationships with the ordination axes (Table 5.2) and two of these, visibility and the forage quality index, were related to the separation of habitats along Axis 2. Relative to other habitats, 5 year old sites had low visibility, but high values of the forage quality index. A number of non linear relationships were also apparent, with substantial differences between the habitats observed for grass and tree abundance and to a lesser extent fern, forb and shrub abundance (Figure 5.3 A). There was little difference between the habitats with respect to nitrogen content, water content or dry matter digestibility (Figure 5.3 B, C and D). Forb and sedge cover were also strongly correlated with the ordination, but the abundance of these plant groups varied within habitats, rather than between them. At the two recently harvested sites, visibility was much greater than in the surrounding forest (46.9 and 46.0 m versus 21.0 and 19.7 m), or in any of the other habitats. In addition, the forage quality index was, on average, 41% lower, due mainly to the relative absence of plant biomass. In contrast to the 5 and 10 year old sites (Figure 5.3 B and C), vegetation at recently harvested sites often had a higher water and nitrogen content than vegetation in the surrounding forest (Table 5.3). Table 5.2. Variables with a strong ( r ≥0.70) linear correlation with the ordination axes of Figure 5.2. FQI is a forage quality index (see text for derivation). Variable Correlation ( r) Correlation ( r) with Axis 1 with Axis 2 Visibility 0.04 0.91 FQI 0.55 0.77 Forb cover 0.81 0.39 Sedge cover 0.73 0.53 81
Unharvested forest 5 year sites 10 year sites
A Abundance B Nitrogen 10 3 9 8 7 2 6 5 4 3 1 2 1 0 0 Fern Forb Grass Sedge Shrub Tree Fern Forb Grass Sedge Shrub Tree
C Water D Digestibility 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 Fern Forb Grass Sedge Shrub Tree Fern Forb Grass Sedge Shrub Tree Figure 5.3. Values of abundance (A) nitrogen content (B) water content (C) and dry matter digestibility (D) for six plant groups in three habitats. Errors are 95% confidence intervals. For both visibility and the forage quality index, additional differences between the 5 and 10 year old sites and the surrounding forest were apparent when the areas that males and females used were considered separately. In the forest surrounding 5 year old sites, visibility in areas used by males (20.9 m) was substantially higher than the visibility in areas used by females (16.8 m; mean difference ± 95% CI = 4.1 ± 3.1m). Differences relating to the forage quality index are shown in Figure 5.4. For females, the value of the index was much lower at 10 year old sites than it was in the surrounding forest (mean difference ± 95% CI: 94.7 ± 91.0). For males, the value of the forage quality index was substantially higher in both 5 and 10 year old sites than it was in the surrounding forest (mean difference at 5 year sites: 109.2 ± 77.7; mean difference at 10 year sites 130.4 ± 131.3). The last contrast was derived from a small sample of wallabies, and should be treated with caution. 82
Table 5.3. Water content, nitrogen content and dry matter digestibility of plant functional groups at the two recently harvested sites. Ferns and shrubs are not included as they were rarely present. Values are presented for the harvested area and for an adjacent patch of unharvested forest. NP = not present. Plant Group Site 1 Site 2 Harvested Forest Harvested Forest Water (% of wet weight) Forb 82.9 79.8 78.3 88.5 Grass 78.3 68.5 74.8 65.1 Tree NP 58.1 62.8 48.5 Sedge NP 51.7 60.1 55.3 Nitrogen (% dry matter) Forb 2.9 2.9 3.1 3.1 Grass 4.6 1.4 3.7 1.4 Tree NP 2.1 1.9 0.9 Sedge NP 1.1 2.2 1.3 Digestibility (% dry matter) Forb 33.6 51.1 38.7 50.7 Grass 30.8 27.0 1 42.2 85.7 1 Tree NP 48.7 44.5 53.5 Sedge NP 36.7 38.0 32.3 1 These values are less similar than expected, and may indicate errors in the analytical process used to determine dry matter digestibility.
Females 350 Males
300
250
200
150
100
50
0 5yr site 5yr forest 10yr site 10yr forest Habitats Figure 5.4. Values of the forage quality index for male and female wallabies at 5 and 10 year old sites and adjacent stands of unharvested forest. Errors are 95% confidence intervals. 83
Habitat selection At the population level wallabies selected both 5 and 10 year old sites more than expected in relation to their availability, although 5 year old sites were selected to a much greater degree. Unharvested forest was selected less than expected, and wallabies avoided recently harvested sites completely (Table 5.4). At the scale of individuals, the pooled male and female sample that had access to recently harvested sites and the surrounding forest selected the forest over the harvested area during both the day (mean log ratio difference; lower 95% CL to upper 95% CL: 4.1; 2.9 to 4.9) and the night (2.7; 1.1 to 4.1). In these calculations, the mean log ratio difference represents the effect size, thus selection for forest over the harvested area was substantially reduced at night. Although there were insufficient data to generate credible results for males and females separately, both sexes showed generally similar patterns. Table 5.4. Habitat selection at the population level based on faecal pellet counts. A value of 1 for the selection index ŵ is equivalent to using a habitat in proportion to its availability, while values >1 and <1 represent selection of habitats more and less than expected in relation to their availability. Habitat Pellets/ha Available ŵ 95% CI for ŵ ± 95% CI area (ha) Forest 1089 ± 354 2169 0.10 0.08, 0.12 Recently harvested 0 27 0.00 0.00, 0.52 5 year old sites 10 597 ± 2513 128 15.65 15.42, 15.88 10 year old sites 1511 ± 630 167 1.71 1.51, 1.91 84
Habitat selection for the sample of wallabies with access to 5 and 10 year old sites and the surrounding unharvested forest was affected both by sex and time of day (Figure 5.5). The Y axis of Figure 5.5 is the log ratio difference between pairs of habitats (the output of the compositional analysis), and a positive value means that the first listed habitat was selected over the second listed habitat. A value close to zero indicates that neither habitat in the pair was selected over the other. Females selected 5 year old sites over the surrounding forest during both diurnal and nocturnal periods. They also selected 10 year old sites over the forest during the day, but selected forest over 10 year old sites at night. Males also selected 5 year old sites over the surrounding forest during the day, but to a lesser extent than females. At night, no selection was shown for either habitat, a result repeated for the diurnal choice between 10 year old sites and forest. At night, however, male behaviour was similar to females as they selected forest over 10 year old sites. For both sexes, 5 year old sites were selected over 10 year old sites during both diurnal and nocturnal periods.
6 A Females 5 4 3 2 1 0 1 2 Day 3 Night 4 5 5 vs Forest 10 vs Forest 5 vs 10
6 B Males 5 4 3 2 1 0 1 2 3 4 5 5 vs Forest 10 vs Forest 5 vs 10 Figure 5.5. Habitat selection of female (A) and male (B) swamp wallabies. A value >0 on the Y axis means that the first listed habitat was selected over the second listed habitat, and larger values mean stronger selection. On the X axis, 5 and 10 refer to 5 year old and 10 year old regenerating sites, and Forest refers to the unharvested forest surrounding the regenerating areas. Error bars are 95% bootstrapped confidence intervals. 85
Association between habitat selection and habitat quality As with habitat selection, the association between selection and the two habitat quality variables (visibility and the forage quality index) depended on sex and time of day (Figure 5.6). In this figure, higher values for visibility relate to more open areas with less lateral cover while higher values of the forage quality index relate to areas with an assumed higher composite value of available forage. During the day, areas that females selected had substantially lower visibility (i.e. more lateral cover) than areas they selected against, while at night visibility did not appear to be associated with the choice of habitat. The pattern for males was similar. Diurnal habitat selection by females was not related to the forage quality index, but at night, selected areas had substantially higher index values than areas selected against. This contrast was driven primarily by the 5 females at 10 year old sites which selected areas with low index values during the day and high index values at night. Males, however, showed the opposite pattern with no relationship apparent at night, but substantially higher index values in selected areas during the day. 25 A Day 25 B Night Females 20 9 20 Males 11 10 11 15 15 13 9 13 10 10 10
5 5
0 0
300 C Day 300 D Night
250 250
200 200 150 150
100 100 50 50
0 0 Selected More Selected Less Selected More Selected Less Than Expected Than Expected Than Expected Than Expected Figure 5.6. Measures of visibility (A, B) and forage quality (C, D) during diurnal and nocturnal periods for sites that were selected more and less than expected by chance. A habitat was selected more than expected by chance when B > B/n, where n is the number of available habitat types. Error bars are 95% confidence intervals. Numbers above the bars in graphs A and B are sample sizes, and are the same for graphs C and D. 86
Discussion The large and obvious contrast between recently harvested forest patches and surrounding unharvested stands changes through time as plants re establish and grow. As plant succession progresses, the characteristics of harvested patches change markedly, and thus harvested stands of different ages, in conjunction with surrounding unharvested forest, provide different resources for forest fauna. In the Pyrenees State Forest, 5 and 10 year old harvested sites and surrounding unharvested stands differed most markedly with respect to visibility (a surrogate measure for lateral cover), the forage quality index and the abundance of a number of plant functional groups. At recently harvested sites, visibility was very high and plant abundance was very low, but the nitrogen and water content of some plants, particularly grass, was high relative to the surrounding forest. From the perspective of a ground dwelling macropodid marsupial like the swamp wallaby, these differences represent choices with respect to lateral cover and food abundance, and in some cases food quality. Because both food and cover are important resources for this species (e.g. Hill and Phinn 1993), non random habitat selection in this harvested landscape was expected. At the scale of the population, strong selection for 5 year old sites and, to a lesser extent, 10 year old sites was consistent with past studies suggesting selection of densely vegetated areas, both for swamp wallabies (Troy et al. 1992; Wood 2002) and other medium sized macropods (le Mar and McArthur 2005). Other medium sized mammalian herbivores, such as the roe deer ( Capreolus capreolus ), are also attracted to densely vegetated habitats (e.g. Tufto et al. 1996). The population level analysis, however, could not differentiate between sexes or diel period, factors that influence habitat selection in some circumstances (Ager et al. 2003; McCorquodale 2003; Mitchell and Powell 2003). The eight individuals (five females and three males) that had access to recently harvested sites and the surrounding unharvested forest selected the forest over the harvested area during both diurnal and nocturnal periods. Presumably, this was due to the almost complete absence of cover on the harvested areas, an explanation consistent 87 with the wallabies’ somewhat higher use of these areas at night. Although there is evidence that swamp wallabies select foods with a high nitrogen content (Osawa 1990), the relatively high levels of nitrogen in grass growing on recently harvested areas did not attract wallabies onto these sites during the 8 months of monitoring after the post harvest burn. Patterns of selection for wallabies with access to 5 and 10 year old sites and the forest surrounding them was affected both by sex and diel period. In general, females demonstrated distinct choices, showing strong selection during both diurnal and nocturnal periods. Choosing between available habitats appeared to be less important for males, who demonstrated no selection on a number of occasions. Although not always the case (e.g. Mosnier et al. 2003), habitat selection in mammals has been shown to differ between males and females (Ager et al. 2003; McCorquodale 2003; Mitchell and Powell 2003), and between diurnal and nocturnal periods (Beyer and Haufler 1994; le Mar and McArthur 2005). Differences between the sexes may be due to the relatively greater influence of reproduction on male behaviour (Clutton Brock 1989). For example, males often range further than females (e.g. Dahle and Swenson 2003), a behaviour that has been observed for swamp wallabies (Chapter 6) and linked to reproductive success in another macropod species (Fisher and Lara 1999). Differences between diurnal and nocturnal habitat selection were more pronounced for females than for males, but diel patterns for both sexes were generally consistent with the behaviour of a number of other macropod species who shelter during the day and forage in more open areas at night (Fisher 2000; Johnson 1980; le Mar and McArthur 2005). It is likely that this behaviour evolved as a strategy to deal with the conflicting demands of food acquisition and predator avoidance (Lima and Dill 1990), and behavioural decisions related to these two factors have been shown to influence habitat selection on a number of occasions (Cowlishaw 1997; Ferguson et al. 1988). Although predation pressure was not quantified, potential predators at the study site included foxes ( Vulpes vulpes ) and wedge tailed eagles ( Aquila audax ). Foxes may have a substantial impact on macropod populations through predation on juveniles (Banks et al. 2000), and eagles are known to hunt smaller macropods under the forest canopy (N. Mooney, pers. comm.). 88
If predator avoidance and food acquisition are in conflict and are a primary influence on behaviour, cryptic species such as swamp wallabies should select habitats with good protective cover at times when predation is a threat. Female swamp wallabies in particular appeared to behave in this way, selecting areas of relatively low visibility during the day and areas with a relatively high value of the forage quality index at night. Although data for males and females were pooled, and cover and forage quality were not quantified, red bellied pademelons and red necked wallabies with access to differentially aged Eucalyptus plantations and the surrounding forest appeared to be selecting habitats on the basis of similar imperatives (le Mar and McArthur 2005). Ager et al. (2003) also observed similar behaviour in elk ( Cervus elaphus ), which remained close to protective cover during the day, but moved into areas with relatively high forage biomass at night. Although female wallabies with access to 5 and 10 year old sites demonstrated the same general behavioural patterns with respect to visibility and overall forage value, they selected different nocturnal habitats to achieve it. As previously mentioned, trade offs between food acquisition and predator avoidance are common, and for females with access to 10 year old sites these competing factors appeared to be responsible for the difference between diurnal and nocturnal selection. At 5 year old sites, however, there was no trade off; visibility was low and the value of the forage quality index was high, and thus selection for 5 year old regenerating areas occurred in both diurnal and nocturnal periods. A similar situation was observed by Pierce et al. (2004) who found that mule deer (Odocoileus hemionus ) were less likely to be preyed upon in habitats containing the best food resources. The selection of 5 year old sites by swamp wallabies is analogous to the selection of edge habitats by other mammalian herbivores (e.g. Dussault et al. 2005; Tufto et al. 1996) due to the interspersion of multiple resources in these areas. However, as harvested stands mature and the canopy closes, habitat quality diminishes, particularly with respect to food resources. Although 10 year old sites still provided substantial shelter, the forage quality index was low relative to the surrounding forest. Unlike 5 year old sites, a conflict between predator avoidance and food acquisition did exist, and, at least for females, patterns of selection appeared to be driven by these opposing forces. 89
Finally, the preceding discussion assumes that visibility and the forage quality index are indicators of predation risk and overall food value respectively. Swamp wallabies are solitary (Kaufmann 1974), and thus do not use aggregation for predator avoidance (Jarman and Coulson 1989). In addition, they almost certainly perceive cover as protective, as do other smaller macropods such as the tammar wallaby, Macropus eugenii (Blumstein and Daniel 2002). Consequently the association between visibility and habitat selection was probably a reasonable representation of the behavioural response to the perceived risk of predation. The forage quality index, on the other hand, was a coarse indicator of overall food value based on a multiple (but incomplete) variable set. Although similar indices appear in the literature (e.g. Mitchell and Powell 2003; Moser et al. 2006), it is somewhat difficult to determine ecologically important effects, as measurement scales are arbitrary and different for each index. In addition, I did not measure all variables that may have influence forage consumption. There is direct evidence, for example, that plant toxins influence the foraging choices of mammalian herbivores in Australian ecosystems (Lawler et al. 2000; Loney et al. 2006; Wallis et al. 2002), and indirect evidence that toxins may influence foraging behaviour in swamp wallabies (Lawler and Foley 1999). As a consequence, results relating to the forage quality index should be interpreted with care, and the limitations of the index recognised. Given that an association between habitat selection and the forage quality index was detected at the home range scale, it would be profitable to determine if the inclusion of other variables in the index altered the results, and to investigate effects at different spatial and temporal scales. 90
CHAPTER 6
The influence of resources on the home range size of the swamp wallaby ( Wallabia bicolor ).
Abstract For many mammals, home range size is inversely related to the availability or quality of important resources. In this study, I used habitat changes resulting from timber harvesting to investigate the factors influencing home range size in swamp wallabies (Wallabia bicolor ), a medium sized (10 – 25 kg) macropodid marsupial distributed widely throughout eastern Australia. I used harvested sites that were <12 months, 5 years and 10 years old as experimental treatments to quantify the impact of habitat type and sex on both total and diurnal home range size. I modelled home range size as a function of sex and three indicators of resource quality: visibility (an indicator of lateral cover), a forage quality index, and an index of food and shelter interspersion, and used Akaike’s Information Criteria to differentiate between a pre determined set of explanatory models. Relationships between (a) home range size and population density and (b) home range size and body weight were analysed separately. I also compared the body condition of wallabies living at control and 5 year old sites using two indices of body condition: (a) percent kidney fat and (b) leg length:body weight ratio. For both total and diurnal data sets, male ranges were about twice as large as female ranges in all habitats, and this difference was attributed to sex rather than body weight. Relative to wallabies living at unharvested control sites, the home range of wallabies at 5 year old sites was reduced by 40 45% for females and 25 30% for males. Range size at recently harvested and 10 year old sites were similar to controls. For total and diurnal home range size respectively, simple models including two predictors explained about 70% and 65% of the variance in the data. Sex was the most 91 important predictor of home range size for both data sets. The index of food and shelter interspersion was an important negative predictor of total range size, and visibility was an important positive predictor of diurnal ranges size. Home range size was negatively related to population density, and body condition was similar for wallabies at unharvested control and 5 year old sites. I suggest that two factors were responsible for the maintenance of small total home ranges observed at 5 year old sites: (a) relatively large contiguous patches of dense lateral cover and (b) the interspersion of shelter and adequate food. 92
Introduction
Studies of inter specific variation in the home range size of eutherian mammals (e.g. Damuth 1981; Harestad and Bunnell 1979; Kelt and Van Vuren 2001; Lindstedt et al. 1986; Mace and Harvey 1983; McLoughlin and Ferguson 2000; McNab 1963) have found that range size is strongly related to energetic requirements. Energy demands are primarily driven by body size (McNab 1963), but may also be related to factors such as trophic status, habitat productivity (Harestad and Bunnell 1979; Kelt and Van Vuren 2001) and sociality (Damuth 1981). Processes that influence intra specific variation in home range size may often be similar but are likely to operate at shorter temporal and smaller spatial scales (McLoughlin and Ferguson 2000). Within species, home range size is affected by resource availability (Tufto et al. 1996), resource dispersion (Macdonald 1983), population density (Kjellander et al. 2004), predation pressure (Desy et al. 1990) and sex (Dahle and Swenson 2003). Short term habitat change resulting from disturbance events or seasonal weather conditions can also effect the ranging behaviour of individual animals. For example, Stirrat (2003b) found that home ranges of both male and female agile wallabies ( Macropus agilis ) were substantially smaller in the wet season than in the dry season, and attributed this finding to seasonal differences in food availability. The effect of resources on intra specific home range size has often been considered in terms of forage abundance, or, more generally, habitat productivity (e.g. Gompper and Gittleman 1991; Relyea et al. 2000), but other resources, such as shelter, are also important. This is particularly true for arboreal species that are obligate users of tree hollows (e.g. Martin 2006), or species that hide from predators, such as many herbivorous mammals (e.g. le Mar and McArthur 2005; Tufto et al. 1996). Herbivorous mammals may minimise the trade off between anti predator behaviour and food acquisition by using areas containing both food and shelter (Pierce et al. 2004; Chapter 5), so both the availability of these resources and their proximity to one another may influence movement patterns and home range size. 93
In the native forests of southeastern Australia, timber harvesting creates 10 to 30 ha patches of regenerating vegetation within much larger unharvested stands. Sites are densely covered with regenerating Eucalyptus seedlings and other native vegetation three to five years after harvesting, but by ten years post harvest the abundance of early successional species has diminished and stands are dominated by close growing Eucalyptus saplings and species such as silver wattle ( Acacia dealbata ) and austral bracken ( Pteridium esculentum ) that regenerate well after mechanical disturbance or fire. These temporal changes, and the contrast between harvested sites and the surrounding forest, provide an opportunity to study the effects of altered habitat quality on forest fauna. In this study, I used timber harvesting as an experimental treatment to quantify the effect of food and shelter resources on the home range size of swamp wallabies (Wallabia bicolor ). Home ranges were defined at unharvested control locations and at sites harvested <12 months, 5 years and 10 years previously, and these areas differed markedly with respect to lateral cover, vegetation abundance, and wallaby density. Relative to other sites, 5 year old locations has substantially more lateral cover, vegetation and wallabies than other sites, and greater vegetation abundance resulted in larger (better) values of a forage quality index (Chapter 5). Swamp wallabies are medium sized (10 to 25 kg) solitary macropodid marsupials, and are widely distributed throughout the forests of southern and eastern Australia. They have been classified as browsers on the basis of dental morphology (Sanson 1978) and diet (e.g. Hollis et al. 1986; Chapter 3), and select densely vegetated habitats during the day (Troy et al. 1992) but may move into open habitats to forage at night (Chapter 5; Swan et al. unpublished data). Swamp wallabies have relatively small (15 – 40 ha), temporally stable diurnal home ranges (Troy and Coulson 1993; Wood 2002), but their total ranges may be substantially larger (Chapter 4). Factors affecting home range size have never been quantified. My first objective was to quantify the effect of habitat change precipitated by timber harvesting on both total (combined diurnal and nocturnal data) and diurnal home range size. Specifically, I aimed to test the hypothesis that animals living in relatively resource rich environments will have smaller home ranges. Because 5 year old 94 regenerating sites appeared to have greater resources in terms of both food and shelter (Chapter 5), I expected home range sizes to be reduced in these areas. My second objective was to quantify the relationship between home range size and the availability of specific resources. Because both shelter and food are important for swamp wallabies, and their density has been observed to increase in areas that contain both resources in close proximity (Floyd 1980; Hill and Phinn 1993; Lunney and O'Connell 1988), I used indicators of (a) shelter, (b) forage quality and (c) the spatial aggregation of the two (an index of food and shelter interspersion) as predictors of home range size. Sex was also used as a variable in this analysis as the relationship between male home range size and resources such as food and shelter may be influenced by the need to search for mates (Clutton Brock 1989; Fisher and Lara 1999). On the basis of previous work where different resources appear to be more important to swamp wallabies at different times of the diel cycle (Chapter 5), I expected the relative importance of the predictor variables to depend on whether total or diurnal range size was used as the response variable. A specific prediction was that shelter resources will be more strongly correlated with diurnal range size than with total range size, but that this pattern would be reversed with respect to food resources. My third objective was to quantify the body condition of wallabies living in unharvested forest and 5 year old sites, and consider the results in light of two alternative expectations. First, if higher values of the food quality index recorded at 5 year old sites (Chapter 5) actually indicated an ecologically important increase in food quality, wallabies living in these areas were expected to be in better condition. An alternative is that wallaby populations conform to the Ideal Free Distribution model (Fretwell and Lucas 1970) which predicts that population density varies between habitats of differing suitability but fitness does not. If body condition was assumed to be an indicator of fitness, the condition of individuals living at unharvested control and 5 year old sites was also expected to be similar. 95
Methods Study site This study was conducted in the Pyrenees State Forest, which has been described in previous chapters. Experimental design The general design for this study is described in Chapters 2 and 4. The data set for the analysis of total home ranges consisted of 48 wallabies (28 female and 20 male) at 18 sites. It contained 2358 positions, with 78.7% ± 0.9 (mean ± 95% CI) collected during the day, and thus total home ranges were somewhat biased towards diurnal space use. The mean number of positions per individual was 49.1 ± 3.9 (diurnal 38.6 ± 3.1; nocturnal 10.4 ± 0.9). On average, I located females 8.8 ± 7.5 more times than males. The data set for the analysis of diurnal home ranges used fewer wallabies (42; 24 females and 18 males) as six individuals were excluded due to an inadequate (<30) number of positions. The final diurnal data set consisted of 1706 positions with an average of 40.6 ± 3.1 per individual. Females had 7.8 ± 5.2 more locations than males. Wallaby capture and radio tracking Methods of wallaby capture and radio tracking were outlined in Chapter 2. For both total and diurnal data sets, the proportion of radio tracking locations given each accuracy rating is almost exactly the same as presented in Table 4.1. Factors affecting home range size Both food and shelter resources are important to swamp wallabies, and it has been suggested that habitat quality is enhanced when these resources occur together (e.g. Floyd 1980; Hill and Phinn 1993; Lunney and O'Connell 1988). In terms of the predicted relationship between ranging behaviour and resource quality (McLoughlin and Ferguson 2000), I expected that home range size would be negatively correlated with either food, shelter, or some combination of the two, but that relationships may 96 vary depending on sex. From this basis, I developed an a priori set of 10 candidate linear models relating home range size to sex and three indicators of resource quality: (a) visibility (a surrogate for lateral cover), (b) a forage quality index and (c) an index of forage and shelter interspersion. The set of models comprised each of the variables as individual predictors (four models), as well as the combination of sex with each of the other three predictors both with (three models) and without (three models) an interaction term. Visibility and the forage quality index represented shelter and food resources at the scale of the whole home range, and their derivation has been described in Chapters 3 and 5. The index of forage and shelter interspersion (hereafter referred to as the interspersion index) represented the degree to which food and shelter occurred together at the same location, and was calculated as
n ∑()Vi + FQI i Interspersion = i=1 n where Vi and FQI i are visibility and the forage quality index respectively at plot i, and n is the number of plots (see Chapter 3 for sampling methodology). So that each variable had equal weight prior to summation, visibility was represented as 1/visibility and then both variables were transformed by dividing by their maximum values so that they ranged between 0 and 1. Relationships between (a) home range size and body weight and (b) home range size and wallaby density were assessed in separate analysis. I used the standing crop of faecal pellets within each home range as an indicator of relative density in analysis (b) (see Chapter 5 for detailed methodology). Each of the analysis described above was applied to both total and diurnal ranges. Due to the timing of the harvesting operation, data for wallabies at recently harvested sites were collected in winter and spring (see Chapter 5), and thus were not included in the analysis due to the potential confounding effect of season. Data for a number of other wallabies was also missing or excluded due to an insufficient (<30) number of radio 97 tracking locations, thus sample sizes for the analysis of total range and diurnal range were 36 and 32 respectively. Body condition indices I used two indices of body condition: (a) leg length (crus) to body weight ratio and (b) percentage kidney fat. Both indices have been used previously as indicators of body condition in macropods (e.g. Moss and Croft 1999; Shepherd 1987; Stirrat 2003a). The indices were compared between control and 5 year old sites as the kidney fat data were not available from recently harvested or 10 year old sites. Because body weight is likely to be influenced by environmental conditions, I categorised conditions at the time of wallaby capture and weighing as either high stress (March to October inclusive) or low stress (November to February). The high and low stress categorisation was based on observed patterns of vegetation growth at the study site and incorporated a one month time lag, as time intervals between vegetation growth and changes to body condition have been noted for other macropods (e.g. Moss and Croft 1999). Both kidneys were extracted from each wallaby shot for a study of diet analysis (Chapter 3) and stored in 70% ethanol for later processing. The ureter was severed 10cm below the kidney, and all fat and associated mesentery above this point was removed, weighted, and expressed as a percentage of the kidney weight. Values from the two kidneys were averaged to generate a kidney fat index for each individual. Because the presence of young can affect the energy use of female macropods (Louden 1987), I used the presence of young at foot and teat size to categorise female reproductive status. Females with known young at foot or late lactational teats were categorised as high energy users, and these data were used as a blocking factor in subsequent analysis. Data analysis 98
I used the Restricted Maximum Likelihood (REML) routine in GenStat 8 to model the effect of sex, harvesting treatment (recently harvested, 5 year old, 10 year old and unharvested control), and their interaction on both total and diurnal home range size, using site as a random factor. The assumptions of normality and homogeneity of variance were tested with a half normal plot and a fitted value plot respectively, and home range data were log 10 transformed to stabilise the variance. A single data point was excluded from all subsequent analysis; it belonged to a male control wallaby who had an uncharacteristically small home range (total range = 16.6 ha) and appeared to be restricted in his movements. I investigated the use of body weight and the number of radio tracking locations as covariates but neither of these variables was substantially related to home range size. For each analysis, planned contrasts between control sites and the three treatments were made for both males and females. In order to represent the effects in the original units, the contrasts and associated 95% confidence intervals were generated using 10 000 bootstrapped iterations. REML was also used to assess the effect of treatment (unharvested control and 5 year old sites), sex and their interaction on the two body condition indices, with site used as a random factor. In the leg length/body weight analysis I initially used stress (high or low) as a blocking factor. Although it altered the body condition index by about 10%, condition was shown to be better at times of high stress, a counter intuitive result. In addition, preliminary analysis on the data for the group of culled animals demonstrated, for both males and females, insubstantial differences in body condition at capture (all seasons) and death (early autumn only) (analyses not shown). Consequently, stress was not used as a factor in the final analysis. In the kidney fat analysis I first analysed female data only using reproductive status as a blocking factor. Although females in a high energy state had less kidney fat than others (mean difference ± 95%CI = 9.6% ± 28.1), the estimate was imprecise and the removal of this factor had little effect on treatment means. Consequently, the analysis of the total data set was conducted without taking female reproductive status into account. The assumptions of these analyses were checked using the graphical techniques described above, and data were not transformed. One individual was removed from the kidney fat analysis as her fat percentage was 149% (next largest = 53%) and the inclusion of this data point greatly influenced the results. 99
I used the small sample size construction of Akaike’s Information Criteria (AIC c) and
Akaike weights ( wi) to differentiate between the linear models relating the four predictor variables (sex, visibility, the forage quality index and the interspersion index) to home range size (Burnham and Anderson 1998). The model with the smallest AIC c value was considered to fit the data best, and other models within about two AIC c units of the best also had substantial support (Burnham and Anderson 1998). The importance of each variable xj was assessed by summing values of wi for each model in the set containing that variable to obtain the predictor weight w+(j). The variable with the largest predictor weight is estimated to be the most important. I used a model averaging approach to calculate coefficients and associated 95% confidence intervals for the best models. Assumptions of normality, homogeneity of variance and linearity were assessed using normal probability plots and plots of residuals against fitted values. To better meet these assumptions, home range size and visibility were log 10 transformed, while the index of food and shelter interspersion was given an exponential ( ex) transformation. I used simple linear regression to quantify the relationship between (a) home range size and body weight and (b) home range size and relative wallaby density. As before, home range data were log 10 transformed. Results Treatment effects Back transformed means and 95% confidence intervals from the REML analysis demonstrated a substantial treatment effect for both males and females when both total (Figure 6.1 A) and diurnal (Figure 6.1 B) ranges were used. In all cases, home range size was substantially smaller at 5 year old sites than anywhere else, although data for males at recently harvested and 10 year old sites were based on only two and three wallabies respectively. Four of the eight females sampled at 5 year old sites had some 100
110 A 100 Females 90 Males 80
70 60 50 40 30 20 10 0 110 B 100 90
80 70 60 50 40 30 20 10 0 Control Recently 5 yr Regen. 10 yr Regen. Harvested
Figure 6.1. The effect of harvesting treatment on home range size for (A) total ranges and (B) diurnal ranges. Data have been back transformed for graphical presentation. Errors are 95% confidence intervals. 101 of the smallest total home ranges ever recorded for this species (3.4, 4.8, 6.0 and 9.8 ha). The pattern of change across treatments was similar for both sexes, indicating no sex by treatment interaction. Raw home range data for all wallabies are presented in Appendix 1. Contrasts between home range size at unharvested control and treatment sites for both total (Figure 6.2 A) and diurnal (Figure 6.2 B) ranges demonstrated that the average range size at control sites was substantially larger than the average range size at 5 year old sites for both males and females. Relative to control sites, female ranges at 5 year old sites were 41.8% and 46.1% smaller for total and diurnal data sets respectively. The corresponding values for males were 23.9% and 32.7%. For females, differences in home range size between control sites and both recently harvested and 10 year old sites were minor. For males, ranges at recently harvested sites and 10 year old sites appeared to be somewhat larger than ranges at control sites, but the low precision of the estimates (due to small sample sizes) makes the inference unreliable. COMPARISON * A Cont vs Recent Harv Males Cont vs 5 yr Females * Cont vs 10 yr Cont vs Recent Harv Cont vs 5 yr Cont vs 10 yr 50 40 30 20 10 0 10 20 30 40 Mean Difference in Total Home Range Size (ha) B Cont vs Recent Harv * Cont vs 5 yr Cont vs 10 yr * Cont vs Recent Harv Cont vs 5 yr Cont vs 10 yr 50 40 30 20 10 0 10 20 30 40 Mean Difference in Diurnal Home Range Size (ha) Figure 6.2. Home range size comparisons between control sites and recently harvested, 5 year old and 10 year old sites for (A) total ranges and (B) diurnal ranges. Mean differences are calculated as the control mean minus the treatment mean, so a positive value on the X axis means that home range size at control sites is larger. Errors are 95% bootstrapped confidence intervals and a * indicates a comparison where the sample size in the treatment group was either 2 or 3. 102
Factors affecting home range size The model including the main effects of sex and the interspersion index best described 2 the variation in total home range size ( wi = 0.44; Adj. r = 70.3), although two other models also had substantial support (Table 6.1 A). The fact that the largest Akaike weight was only 0.44 indicates substantial uncertainty associated with the best model, and the first four models provide an approximate 95% confidence set within which the best model is likely to be found (∑wi ≥0.95). The evidence ratio (wi model 1 / wi model 2) was 1.69, indicating that the first model was about 1.7 times more likely than the second. The effect of each predictor variable averaged over the set of candidate models (the predictor weights) were 1.0, 0.65, 0.34 and 0.00 for sex, the interspersion index, visibility and the forage quality index respectively, indicating that after sex the interspersion index was the most important variable, and the forage quality index had no effect. The model including the main effects of sex and visibility best described the variation in 2 diurnal home range size ( wi = 0.50; Adj. r = 65.8), although the model containing the sex by visibility interaction had a similar degree of support (Table 6.1 B). This is confirmed by the evidence ratio of model 1 to model 2 (1.19), indicating that model 1 is only slightly more likely to be the best model. Nevertheless, the standardised coefficient (± 95% CI) for the interaction term in model 2 was 0.066 ± 0.089 suggesting that the interaction effect was insubstantial. As above, the maximum wi value of 0.50 indicates substantial model selection uncertainty, and in this case the first three models provide an approximate 95% confidence set within which the best model is likely to be found. The predictor weights were 1.00, 0.92, 0.09 and 0.00 for sex, visibility, the interspersion index and the forage quality index respectively, indicating that sex and visibility were both important variables. Averaged standardised coefficients and associated unconditional 95% confidence intervals for the predictors in the best models are shown in Table 6.2. The average coefficient and confidence interval for each predictor was derived from the set of models within which it occurred, thus incorporating model selection uncertainty into the estimates (Burnham and Anderson 1998). In the model for total range size the coefficients show that that males had larger ranges than females and range size tended 103 to decrease with increasing spatial aggregation of food and shelter. The model for diurnal range size demonstrates a similar sex effect, and ranges tended to increase with visibility, or, in other words, ranges were larger in areas with less lateral cover. Table 6.1. Model selection statistics for (A) the total home range data set and (B) the diurnal home range data set. K is the number of parameters in the model. The model with the smallest AIC c value fits the data best, but models within about 2 AIC c units of the best also have substantial support. The Akaike weights, wi, are interpreted approximately as the probability that their associated model is the best. The first four (A) and three (B) models provide an approximate 95% confidence set for the best model (∑wi ≥0.95). S = Sex; V = visibility; FQI = the forage quality index and D = the index of food and shelter interspersion. See text for further details. A 2 Model K AIC c AIC c wi Adj. r S + D 4 122.05 0.00 0.44 70.3 S + V 4 121.02 1.03 0.26 69.4 S+D+S×D 5 120.55 1.50 0.21 70.4 S+V+S×V 5 118.73 3.32 0.08 68.8 D 3 101.32 20.73 0.00 45.0 V 3 100.42 21.63 0.00 43.6 S+FQI 4 98.72 23.33 0.00 43.2 S+FQI+S×FQI 5 97.05 25.00 0.00 43.1 S 3 92.54 29.51 0.00 38.0 FQI 3 84.40 37.65 0.00 12.0 B 2 Model K AIC c AIC c wi Adj. r S+V 4 96.04 0.00 0.50 65.8 S+V+S×V 5 95.72 0.32 0.42 67.3 S+D 4 91.28 4.77 0.05 60.3 S+D+S×D 5 91.90 5.15 0.04 61.9 S+FQI+S×FQI 5 79.45 16.59 0.00 45.5 S+FQI 3 78.69 17.36 0.00 41.2 V 3 77.93 18.12 0.00 36.8 D 3 75.22 20.83 0.00 31.2 S 3 73.53 22.51 0.00 36.7 FQI 3 65.21 30.84 0.00 5.9 104
Table 6.2. Averaged standardised coefficients for predictors in the best model for both total and diurnal range size. 95% Lower and Upper refer to the 95% lower and upper confidence limits respectively. Predictor βββ 95% Lower 95% Upper Model for total range size Sex 0.157 0.096 0.219 Interspersion index 0.192 0.270 0.115 Model for diurnal range size Sex 0.185 0.109 0.262 Visibility 0.193 0.111 0.275
For both males and females, strong linear relationships were observed between log 10 home range size and relative wallaby density, measured as pellets/ha (Figure 6.3). The analysis is only shown for total range data, as results for diurnal ranges were very similar. The regression equations for males and females respectively are: log 10 home 2 range size = 1.77 – 0.41 density, Adj. r = 47.8%; log 10 home range size = 1.41 – 0.43 density, Adj. r2 = 79.6%. The strong trend observed in the female data was influenced by a number of data points with high leverage.
In contrast, there was no substantial relationship between log 10 home range size and body weight for either males or females (Figure 6.4). Again, the analysis is only shown for the total range data, as results for diurnal data were similar. The relationship for females was non existent (Adj. r2 = 0) and although the relationship for males was somewhat positive (Adj. r2 = 6.8) it was strongly influenced by two data points corresponding to the two lightest wallabies. When these points were removed, Adj. r2 = 0. The lack of relationship within each sex suggests that the larger male ranges observed in Figure 6.1 were due to behavioural differences related to gender rather than differences in body weight. 105
90 Male 80 80 Female 70 60 50 40 30 20 10 0 0 5000 10000 15000 20000 25000 Wallaby Density (pellets/ha) Figure 6.3. Relationship between total home range size and wallaby density for males and females. Y axis values have been back transformed for graphical presentation. Males: log 10 2 home range size = 1.77 – 0.41 density, Adj. r = 47.8%. Females: log 10 home range size = 1.41 – 0.43 density, Adj. r2 = 79.6%. Solid and dashed lines are lines of best fit and 95% confidence intervals respectively. 90 Male 80 Female 70 60 50 40 30 20 10 0 0 5 10 15 20 25 Body Weight (kg) Figure 6.4. Relationship between total home range size and body weight for males and females. Y axis values have been back transformed for graphical presentation. Adj. r2 values are 6.8% for males and 0% for females, although the value for males also becomes 0 when the points corresponding to the two lightest wallabies are removed. Solid lines are lines of best fit and dashed lines are 95% confidence intervals. For clarity, confidence intervals are not shown for the female data. 106
Body condition indices Both body condition indices (leg length to body weight: Figure 6.5; kidney fat: Figure 6.6) provided no substantive evidence that body condition differed between wallabies living in unharvested controls or 5 year old sites. Taking factors such as female reproductive condition and environmental stress into account had little effect on the results. Although there was a substantial difference between the kidney fat values for males (Figure 6.6), sample size for control values was small ( n = 2) and thus does not provide a basis for strong inference.
Discussion
Habitat change precipitated by timber harvesting had a marked effect on home range size. Relative to unharvested control areas both male and female wallabies living at 5 year old regenerating sites had substantially smaller home ranges, and patterns of change were similar when the analysis was conducted using total and diurnal data sets. Reduced home ranges were expected at 5 year old sites, as both food and shelter resources appeared to be more abundant in these areas. The observed effect is consistent with hypotheses predicting an inverse relationship between resource availability and home range size (see Harestad and Bunnell 1979 for a discussion in an inter specific context) and has been observed in many species (e.g. Beckmann and Berger 2003; Fisher 2000; Gompper and Gittleman 1991; Tufto et al. 1996). The fact that range size at recently harvested and 10 year old sites were similar to control values is also likely to be related to resource availability. Although wallabies did occasionally visit them, recently harvested sites provided few resources, and wallabies using these areas prior to harvesting simply moved into the adjacent forest where they established home ranges of similar sizes (Chapter 2). Although there was substantial lateral cover at 10 year old sites, food resources were limited and apparently insufficient to sustain small home ranges. Female wallabies in particular showed a 107
30 Control
5 yr 25
20
15
10
5
0 Female Male
Figure 6.5. Leg length to body weight ratio for males and females at control and 5 year old sites. Errors are 95% confidence intervals.
50 Control
5 yr 40
30
20
10
0 Female Male
Figure 6.6. Percent kidney fat for males and females at control and 5 year old sites. The value for males at control sites is based on two wallabies, so should be treated with caution. Errors are 95% confidence intervals. 108 strong cyclic pattern of diurnal use of the regenerating area and nocturnal use of the adjacent forest (Chapters 4 and 5), a behaviour that resulted in ranges of a similar size to their control site counterparts. Factors affecting home range size
The set of linear models relating home range size to sex, and three measures of resource quality (visibility, forage quality and food and shelter interspersion) enabled a more detailed assessment of the mechanisms that may have regulated home range size in this system. The most important predictor of both total and diurnal home range size was sex. Males had substantially larger home ranges than females, an observation that is consistent with one previous study of range size in swamp wallabies (Wood 2002) but not with another (Troy and Coulson 1993), although sample size in both was small. The home range of male mammals, including macropodid marsupials, are often larger than females (e.g. Dahle and Swenson 2003; Stirrat 2003b), although it has sometimes been difficult to differentiate between the effect of sex and body weight. For example, Grigone et al. (2002) found that both sex and body weight of mountain lions ( Puma concolor ) was correlated with home range size, but that these two variables were themselves highly correlated. The present analysis demonstrated that body weight was unimportant, and that the observed effect was primarily due to sex. Swamp wallabies are solitary, non territorial and polygynous (Croft 1989), so males are expected to gain fitness benefits by searching widely for mates (Clutton Brock 1989). In a study of another wallaby (bridled nailtail wallaby, Onychogalea fraenata ) with a similar social and mating system, Fisher and Lara (1999) found that male home range size was positively related to reproductive success. Thus the larger male home ranges observed in this study were probably due to the different imperatives of males and females. Although the range size of both sexes is affected by access to resources, females are an important resource for males, and maximising their access to females is expected to result in the relative expansion of male ranges. Consistent with my initial expectation, the relative importance of the interspersion index and visibility as predictors of home range size changed depending on whether total or 109 diurnal range size was used as the response variable. This result supports the argument (Chapter 4) that space use may be influenced by different factors at different times of the day. Also in line with initial expectations, visibility was a more important predictor of diurnal than total home range size, a finding consistent with results from a study of another medium sized macropodid, the bridled nailtail wallaby (Fisher 2000). Many herbivorous mammals use dense vegetation as a diurnal refuge from predators, but it often becomes less important relative to food at night (Ager et al. 2003; Johnson 1980; le Mar and McArthur 2005; Chapter 5). For species that behave in this way, access to dense lateral cover, especially when it occurs in large contiguous patches, renders excessive diurnal movement unnecessary, and thus may result in relatively small diurnal ranges. Overall space use was negatively related to the interspersion index, an indicator of the spatial aggregation of food and shelter resources within each home range. In a similar analysis, Tufto et al. (1996) found no statistically significant effect of the presence of both food and shelter on the home range size of roe deer ( Capreolus capreolus ). These authors, however, calculated the mean value of a food and shelter variable for each home range and used the interaction between the two to indicate food and shelter interspersion. While this variable represents the average combination of food and shelter resources, it does not measure the degree to which food and shelter occur at the same location – the interaction could take on a relatively high value due to the presence of spatially distinct patches of high quality food and shelter within the same home range. My findings, however, suggest that spatially distinct patches of food and shelter may act to increase total home range size due to the diel pattern of movement between the two resources. The index of food and shelter interspersion used in the present study provided a more meaningful measure of the local mix of food and shelter resources by recalculating the index at multiple points within each home range. Contrary to expectations the forage quality index was an equally poor predictor of both total and diurnal range size, although this may have been influenced by the relatively small proportion of nocturnal locations in the total data set. If nocturnal data were insufficient to fully establish nocturnal extensions to diurnal ranges, the relationship between total range size and the forage quality index may have been suppressed by the diurnal component. Alternatively, the forage quality index may be more closely related 110 to nocturnal range size, rather than total or diurnal ranges, although the size of nocturnal data sets was not sufficient to test this hypothesis. Further, as discussed in Chapter 5, the index was based on an incomplete variable set which may have compromised its relevance. In the present study, density was identified as an important negative predictor of home range size. This is contrary to the theoretical expectation, as for solitary species that do not defend territories, higher conspecific density is expected to result in larger home ranges due to increased competition for resources (Kjellander et al. 2004). In reality, the causal relationship between density and home range size is often difficult to define. For example, the reduction of white tailed deer ( Odocoileus virginianus ) populations has resulted in both increased (Henderson et al. 2000) and decreased (Kilpatrick et al. 2001) home range size, and Dahle and Swenson (2003) suggested that the true effect of density on home range size is often confounded by the uncontrolled influence of food resources. It is possible that the observed changes in swamp wallaby density were due to concomitant changes in resource quality, and were not causally related to home range size at all. Experimental manipulation of food resources and density within a chipmunk (Tamias striatus ) population showed that better resources can cause both reductions in range size and increased density (Mares et al. 1982), and a number of observational studies have also suggested the causal influence of resource availability on both home range size and population density (e.g. Glessner and Britt 2005; Jorge and Peres 2005). In addition, previous work on swamp wallabies (Chapter 2) has shown that changes to plant communities resulting from timber harvesting was causally related to a substantial increase in wallaby density at harvested sites. On this basis I suggest that increased lateral cover and forage availability resulting from post harvest plant regeneration caused both the density increase and smaller home ranges observed at 5 year old sites, and that increased density itself had little effect on space use. 111
Body condition
The body condition of wallabies living at unharvested control locations and 5 year old regenerating sites was similar, indicating that both populations had access to adequate food resources. Even if food resources per unit area were better at 5 year old sites, it appeared that wallabies in the unharvested forest accounted for this by increasing the size of their foraging area, and that the energetic cost associated with this behaviour did not effect body condition. In combination, the body condition and relative density data are consistent with the Ideal Free Distribution model (Fretwell and Lucas 1970) which predicts that population density should change with habitat suitability, but fitness should not. An inherent assumption is that body condition is an accurate fitness indicator which of course may not be the case. A better test of Ideal Free model predictions would be to measure longevity and reproductive success, with the expectation that it would be similar for individuals living at unharvested control and 5 year old sites. These data remain to be collected. The maintenance of small home ranges
Based on the results of this study I suggest that two factors were critical to the maintenance of the small total home range areas observed at 5 year old regenerating sites: (a) the continuity of lateral cover and (b) interspersion of shelter with adequate food. When it is available, swamp wallabies often hide in dense vegetation during the day, presumably in response to their perceived threat of predation (Floyd 1980; Troy et al. 1992; Chapter 5). The 5 and 10 year old sites used in this study constituted large (>15 ha) continuous patches of densely regenerating Eucalyptus trees that provided high quality shelter for resident wallabies. In contrast, the distribution of shelter at unharvested sites (e.g. clumps of austral bracken, trunks of large trees, steep gullies) was relatively patchy. Because home ranges are predicted to be smaller when important resources are more evenly distributed (Clutton Brock and Harvey 1978; Macdonald 1983), the continuous nature of the shelter resource at regenerating sites would be expected to facilitate smaller home ranges. 112
Large patches of dense shelter alone, however, are unlikely to result in substantial reductions in home range size, as food resources must also be adequate to maintain small home ranges. For swamp wallabies, this is consistent with the observed negative relationship between the interspersion index and total home range size discussed above. At 5 year old sites, food and shelter occurred together and home ranges were small. In contrast, 10 year old sites contained good shelter resources, but, relative to the adjacent forest, had little food. This spatial partitioning of food and shelter resulted in wallabies moving between the two resources and maintaining relatively large home ranges. The interspersion of food and shelter is important for many species of herbivorous mammal and their co occurrence has been linked to browsing damage (Reimoser and Gossow 1996) and the selection of edge habitats (e.g. Fisher 2000; Tufto et al. 1996). However, it is possible that while resource interspersion was primarily responsible for smaller female ranges, male range size was also influenced by the distribution and abundance of females (Clutton Brock 1989). Males often have larger ranges than females because searching for mates increases reproductive success (Fisher and Lara 1999). Because female range size was reduced and density increased at 5 year old sites, males in these areas would not need to search as widely to encounter the same number of females as their counterparts at other sites. It seems plausible that this factor was as much responsible for the observed reduction in male range size as increased access to interspersed food and shelter resources. 113
CHAPTER 7
Synthesis
Native forest timber harvesting in southeastern Australia can be viewed as a disturbance event that resets the process of vegetation succession and results in 10 to 30 hectare patches of differentially aged regenerating forest scattered throughout the harvested landscape. These patches provide alternative habitat for forest fauna, the suitability (quality) of which changes with time since harvesting. I used patches of differentially aged native forest regenerating after timber harvesting as a template to study the effect of habitat change on the space use, population density, habitat selection and diet of swamp wallabies, Wallabia bicolor . In some components of this project, harvesting operations were viewed as an experimental treatment and used to formally test the effect of habitat change on aspects of wallaby behaviour. I used swamp wallabies as a model species because, on the basis of past research, their behaviour was expected to change in response to the alternative habitats generated by the harvesting process. I recorded habitat attributes at unharvested control, recently harvested (<12 months old), 5 year old and 10 year old sites to establish the temporal pattern of change occurring after the initial harvesting operation. I attempted to quantify the different habitats in a way meaningful to the study species by recording variables believed to represent important resources, such as lateral cover, forage abundance and forage quality. Initially the harvesting process removed almost all above ground plant biomass with the exception of a few mature trees, but within 12 months substantial regeneration of grasses, forbs, austral bracken ( Pteridium esculentum ), silver wattle ( Acacia dealbata ) and eucalypt seedlings (mainly Eucalyptus obliqua and E. globulus ) had occurred. Three months after the post harvest burn, the water and nitrogen content of grass growing on regenerating areas was high relative to the adjacent unharvested forest. Five years after harvesting sites were dominated by densely regenerating 1 3 m tall Eucalyptus seedlings interspersed with silver wattle and dense patches of austral bracken. Relative to unharvested sites, visibility was low, indicating an abundance of 114 lateral cover, and values of a forage quality index were high. Despite the developing canopy, 5 year old sites were early enough in the successional process to sustain relatively abundant grass and forb communities. In contrast, 10 year old sites supported dense, closed stands of 3 6 m tall Eucalyptus regeneration and consequently had lower levels of forb, grass and shrub cover than either 5 year old sites or the unharvested forest. The cover of bracken, however, remained relatively high. Visibility values were intermediate between unharvested stands and 5 year old sites, and values of the forage quality index were low relative to the surrounding forest. The immediate impact of harvesting operations was to displace wallabies into adjacent unharvested forest. They did not move far, and, in most cases, continued to use parts of their pre harvest home range. Although some moderate changes in home range characteristics were observed, the forced relocation appeared to have no detrimental effects. In the 8 months after the post harvest burn, these wallabies had similar sized home ranges to their counterparts at control locations, and selected against the recently harvested area. On the basis of faecal pellet surveys, wallabies abandoned harvested areas for 8 10 months and then began to use the sites with increasing frequency. Twelve months after the post harvest burn, population density was at least 5 times higher than on comparable unharvested sites. An assessment of the impact that herbivores (probably mainly wallabies) were having on regenerating eucalypt seedlings at this time showed that in terms of stocking (a measure of eucalypt regeneration success), their impact was minimal. The resources available at 5 year old sites made them highly suitable for swamp wallabies and, relative to unharvested locations, stimulated major changes in behaviour. Population density at 5 year old sites was approximately 10 times greater than at unharvested sites. Diet composition was markedly different, although diet selection was only moderately changed. The diet data enabled a comparison between two alternative foraging strategies (mixed feeding and diet specialisation), and theories predicting mixed feeding were generally supported. Relative to the surrounding unharvested forest, wallabies showed strong selection for the 5 year old habitat, particularly during diurnal periods, and females demonstrated different patterns of diurnal and nocturnal space use. In addition, home range size was 115 substantially (25% to 45%) smaller, which was consistent with the theoretical prediction that individuals with access to better quality resources will require less space to meet their energy needs. Total range size had a strong negative correlation with an index food and shelter interspersion, while diurnal range size was positively related to visibility, an indicator of lateral cover, suggesting that smaller diurnal ranges occurred in areas containing dense vegetation. Neither total nor diurnal home range size was correlated with the forage quality index. Small home ranges at 5 year old sites were attributed to the presence of relatively large, contiguous patches of dense lateral cover interspersed with adequate food resources. At 10 year old sites, home range size and population density were both similar to values at unharvested control locations. Female wallabies demonstrated a distinct diel movement pattern, which manifested itself as selection for the regenerating areas during the day, but for the surrounding forest at night. This was probably caused by the spatial separation of food and shelter resources, which was in contrast to the interspersion of food and shelter at 5 year old sites. Overall, the later successional status of 10 year old sites, particularly the effect that canopy closure had on food resources, made them less suitable habitat than 5 year old sites. Male and female wallabies behaved somewhat differently. Males had substantially larger home ranges than females, and relative to unharvested control sites, the addition of nocturnal data to diurnal home ranges at 5 and 10 year old sites resulted in a range expansion for females but not for males. Females demonstrated more distinct habitat choices, and the habitat selection data demonstrated an interaction between sex and time of day at 10 year old sites. In addition, habitat selection by females appeared to be influenced by predator avoidance behaviour and food acquisition to a greater extent than males. Throughout the course of this work I proposed a number of hypotheses that remain untested. In Chapter 2 I suggested that increased wallaby density may effect ecosystem processes. One specific prediction is that consumption of nitrogen rich forage will reduce the quality of leaf litter available to the decomposer sub system. This assumes that wallabies select plants high in nitrogen, a hypothesis that has been suggested elsewhere (Osawa 1990) and is generally consistent with the data presented in Chapter 116
3. In several places (Chapters 4, 5 and 6) I suggested that reproduction had a greater effect on male than female behaviour, and this explained behavioural differences between the sexes. Although there is a strong theoretical basis for this hypothesis (Clutton Brock 1989), the relevant data remain to be collected. In Chapter 6 I proposed a number of hypotheses related to space use. I predicted that improved habitat quality caused both smaller home ranges and increased population density, and that forage quality would be more closely related to nocturnal than to diurnal or total range size. I also suggested that because females are an important resource for males, male space use will be affected by the distribution and abundance of females. Finally, I predicted that two factors caused the observed range size reduction at 5 year old sites; (a) relatively large, contiguous patches of dense lateral cover, and (b) the interspersion of cover and adequate food. Formal tests of some of these predictions are difficult, as they would require large scale experimental manipulations of resources in forest ecosystems. Manipulations are possible, however, if planned, large scale management actions are used as experimental treatments. In this study, differentially aged harvested patches were used as an experimental treatment to test the hypothesis that wallabies with access to better resources would have smaller home ranges. Although the data supported the prediction, this simple experiment was insufficient to determine specific causal factors. Nevertheless, there is potential to extend the basic design to determine the influence of specific factors on wallaby behaviour. Harvesting, for example, can be viewed as a manipulation of lateral cover, and the individual effects of food and shelter could be determined by manipulating food resources within and adjacent to harvested stands. More generally, using harvesting (and other planned large scale disturbances) as an experimental overlay provides an opportunity to study the impact of disturbance and resource redistribution on mobile forest dwelling mammals, and to test related bodies of ecological theory. Although such experiments are no panacea, and close collaboration with land management agencies will be required to address some of the technical issues associated with large scale management experiments (e.g. Hobbs 2003), I believe there is scope for the development of such designs in the future. 117
References
Abbott I, Mellican A, Craig MD, Williams M, Liddelow G, Wheeler I (2003). Short term logging and burning impacts on species richness, abundance and community structure of birds in open eucalypt forest in Western Australia. Wildlife Research 30:321 329.
Abrams MD (2003). Where has all the white oak gone? Bioscience 53:927 939.
Aebischer NJ, Robertson PA, Kenward RE (1993). Compositional analysis of habitat use from animal radio tracking data. Ecology 74:1313 1325.
Ager AA, Johnson BK, Kern JW, Kie JG (2003). Daily and seasonal movements and habitat use by female rocky mountain elk and mule deer. Journal of Mammalogy 84:1076 1088.
Altman DG, Machin D, Bryant TN, Gardner MJ (2000). Statistics with Confidence, 2nd edn. British Medical Journal Books, London, 240p.
Anderson DR, Burnham KP, Thompson WL (2000). Null hypothesis testing: Problems, prevalence, and an alternative. Journal of Wildlife Management 64:912 923.
Arnold GW, Steven DE, Grassia A, Weeldenburg J (1992). Home range size and fidelity of western gray kangaroos ( Macropus fuliginosus ) living in remnants of wandoo woodland and adjacent farmland. Wildlife Research 19:137 143.
Attiwill PM (1994a). The disturbance of forest ecosystems the ecological basis for conservative management. Forest Ecology and Management 63:247 300.
Attiwill PM (1994b). Ecological disturbance and the conservative management of eucalypt forests in Australia. Forest Ecology and Management 63:301 346.
Augustine DJ, Frelich LE, Jordan PA (1998). Evidence for two alternate stable states in an ungulate grazing system. Ecological Applications 8:1260 1269. 118
Australian Government (2006). About Sustainable Forest Management in Australia. Department of Agriculture, Forestry and Fisheries website: http://www.affa.gov .au/content/output.cfm?ObjectID=F8063123 711F 4E8E 98148FDA13C16A88.
Banks PB, Newsome AE, Dickman CR (2000). Predation by red foxes limits recruitment in populations of eastern grey kangaroos. Austral Ecology 25:283 291.
Bardgett RD, Wardle DA, Yeates GW (1998). Linking above ground and below ground interactions: How plant responses to foliar herbivory influence soil organisms. Soil Biology and Biochemistry 30:1867 1878.
Beckmann JP, Berger J (2003). Using black bears to test ideal free distribution models experimentally. Journal of Mammalogy 84:594 606.
Belia S, Fidler F, Williams J, Cumming G (2005). Researchers misunderstand confidence intervals and standard error bars. Psychological Methods 10:389 396.
Bennett LT, Adams MA (2004). Assessment of ecological effects due to forest harvesting: approaches and statistical issues. Journal of Applied Ecology 41:585 598.
Berger JO, Sellke T (1987). Testing a point null hypothesis the irreconcilability of P values and evidence. Journal of the American Statistical Association 82:112 122.
Bergvall UA, Leimar O (2005). Plant secondary compounds and the frequency of food types affect food choice by mammalian herbivores. Ecology 86:2450 2460.
Beyer DE, Haufler JB (1994). Diurnal versus 24 hour sampling of habitat use. Journal of Wildlife Management 58:178 180.
Bird PR (1992). Expansion of the range of the black wallaby in western Victoria. Victorian Naturalist 109:89 91.
Blumstein DT, Daniel JC (2002). Isolation from mammalian predators differentially affects two congeners. Behavioral Ecology 13:657 663.
Bobek B, Boyce MS, Kosobucka M (1984). Factors affecting red deer ( Cervus elaphus ) population density in southeastern Poland. Journal of Applied Ecology 21:881 890.
BOM (2006). Bureau of Meteorology (Australia) web site: http://www.bom.gov.au/ . 119
Bormann FH, Likens GE (1979). Pattern and Process in a Forested Ecosystem: Disturbance, Development, and the Steady State Based on the Hubbard Brook Ecosystem Study. Springer Verlag, New York, 253p.
Boutin S (1990). Food supplementation experiments with terrestrial vertebrates: patterns, problems, and the future. Canadian Journal of Zoology 68:203 220.
Bowyer RT, Kie JG (2006). Effects of scale on interpreting life history characteristics of ungulates and carnivores. Diversity and Distributions 12:244 257.
Bulinski J (1999). A survey of mammalian browsing damage in Tasmanian eucalypt plantations. Australian Forestry 62:59 65.
Bulinski J (2000). Relationships between herbivore abundance and browsing damage in Tasmanian eucalypt plantations. Australian Forestry 63:181 187.
Bulinski J, McArthur C (1999). An experimental field study of the effects of mammalian herbivore damage on Eucalyptus nitens seedlings. Forest Ecology and Management 113:241 249.
Bulinski J, McArthur C (2000). Observer error in counts of macropod scats. Wildlife Research 27:277 282.
Bulinski J, McArthur C (2003). Identifying factors related to the severity of mammalian browsing damage in eucalypt plantations. Forest Ecology and Management 183:239 247.
Burnham KP, Anderson DR (1998). Model Selection and Inference: a Practical Information Theoretic Approach. Springer, New York, 353p.
Campbell TA, Laseter BR, Ford WM, Miller KV (2004). Movements of female white tailed deer ( Odocoileus virginianus ) in relation to timber harvests in the central Appalachians. Forest Ecology and Management 199:371 378.
Chase A (1995). In a Dark Wood. Houghton Mifflin, Boston, 535p. 120
Chevallier Redor N, Verheyden Tixier H, Verdier M, Dumont B (2001). Foraging behaviour of red deer Cervus elaphus as a function of the relative availability of two tree species. Animal Research 50:57 65.
Chubbs TE, Keith LB, Mahoney SP, McGrath MJ (1993). Responses of woodland caribou ( Rangifer tarandus caribou ) to clear cutting in east central Newfoundland. Canadian Journal of Zoology 71:487 493.
Cimino L, Lovari S (2003). The effects of food or cover removal on spacing patterns and habitat use in roe deer ( Capreolus capreolus ). Journal of Zoology 261:299 305.
Clutton Brock TH (1989). Mammalian mating systems. Proceedings of the Royal Society of London Series B Biological Sciences 236:339 372.
Clutton Brock TH, Harvey PH (1978). Mammals, resources and reproductive strategies. Nature 273:191 195.
Codron J, Lee Thorp JA, Sponheinier M, Codron D, Grant RC, De Ruiter DJ (2006). Elephant ( Loxodonta Africana ) diets in Kruger National Park, South Africa: Spatial and landscape differences. Journal of Mammalogy 87:27 34.
Cohen J (1994). The earth is round ( p<.05). American Psychologist 49:997 1003.
Comiskey EJ, Eller AC, Perkins DW (2004). Evaluating impacts to Florida panther habitat: How porous is the umbrella? Southeastern Naturalist 3:51 74.
Connell JH (1978). Diversity in tropical rain forests and coral reefs high diversity of trees and corals is maintained only in a non equilibrium state. Science 199:1302 1310.
Cooper SM, Owens MK, Cooper RM, Ginnett TF (2006). Effect of supplemental feeding on spatial distribution and browse utilization by white tailed deer in semi arid rangeland. Journal of Arid Environments 66:716 726.
Côté SD, Rooney TP, Tremblay JP, Dussault C, Waller DM (2004). Ecological impacts of deer overabundance. Annual Review of Ecology Evolution and Systematics 35:113 147. 121
Cowlishaw G (1997). Trade offs between foraging and predation risk determine habitat use in a desert baboon population. Animal Behaviour 53:667 686.
Craig MD, Roberts JD (2005). The short term impacts of logging on the jarrah forest avifauna in south west Western Australia: implications for the design and analysis of logging experiments. Biological Conservation 124:177 188.
Croft DB (1989). Social organization in the Macropodoidea. In: Grigg G, Jarman P, Hume I (eds) Kangaroos, Wallabies and Rat Kangaroos. Surrey Beatty, Sydney.
Cumming G, Finch S (2005). Inference by eye: confidence intervals and how to read pictures of data. American Psychologist 60:170 180.
Dahle B, Swenson JE (2003). Home ranges in adult Scandinavian brown bears ( Ursus arctos ): effect of mass, sex, reproductive category, population density and habitat type. Journal of Zoology 260:329 335.
Dale VH, Brown S, Haeuber RA, Hobbs NT, Huntly N, Naiman RJ, Riebsame WE, Turner MG, Valone TJ (2000). Ecological principles and guidelines for managing the use of land. Ecological Applications 10:639 670.
Damuth J (1981). Home range, home range overlap, and species energy use among herbivorous mammals. Biological Journal of the Linnean Society 15:185 193.
Danell K, Ericson L (1986). Foraging by moose on 2 species of birch when these occur in different proportions. Holarctic Ecology 9:79 83.
Day RW, Quinn GP (1989). Comparisons of treatments after an analysis of variance in ecology. Ecological Monographs 59:433 463. de Munk FG (1999). Resource use by the eastern grey kangaroo and the black wallaby in a managed remnant woodland community. PhD Thesis, Deakin University, Geelong, 252p. de Solla SR, Bonduriansky R, Brooks RJ (1999). Eliminating autocorrelation reduces biological relevance of home range estimates. Journal of Animal Ecology 68:221 234. 122
Desy EA, Batzli GO, Liu J (1990). Effects of food and predation on behavior of prairie voles: a field experiment. Oikos 58:159 168.
Di Stefano J (2003). Mammalian browsing in the Mt Cole State Forest: defining a critical browsing level and assessing the effect of multiple browsing events. Australian Forestry 66:287 293.
Di Stefano J (2005). Mammalian browsing damage in the Mt. Cole State forest, southeastern Australia: analysis of browsing patterns, spatial relationships and browse selection. New Forests 29:43 61.
Di Stefano J, Moyle R, Coulson G (2005). A soft walled, double layered trap for capture of swamp wallabies Wallabia bicolor . Australian Mammalogy 27:235 238.
Dignan P, Fagg PC (1997). Eucalypt Stocking Surveys. Native Forest Silviculture Guideline No. 10. Department of Natural Resources and Environment, Melbourne, 84p.
Dowman MG, Collins FC (1982). The use of enzymes to predict the digestibility of animal feeds. Journal of the Science of Food and Agriculture 33:689 696.
Downes BJ, Barmuta LA, Fairweather PG, Faith DP, Keough MJ, Lake PS, Mapstone BD, Quinn GP (2002). Monitoring Ecological Impacts: Concepts and Practice in Flowing Waters. Cambridge University Press, Cambridge, 434p.
DSE (2006). Forest Explorer Map. Web site. http://www.dse.vic.gov.au/dse/dsencor. nsf/LinkView/836EE128E54D861FCA256DA200208B945FD09CE028D6AA58CA25 6DAC0029FA1A.
Dussault C, Ouellet JP, Courtois R, Huot J, Breton L, Jolicoeur H (2005). Linking moose habitat selection to limiting factors. Ecography 28:619 628.
Edenius L, Ericsson G, Naslund P (2002). Selectivity by moose vs the spatial distribution of aspen: a natural experiment. Ecography 25:289 294.
Edge WD, Marcum CL, Olson SL (1985). Effects of logging activities on home range fidelity of elk. Journal of Wildlife Management 49:741 744. 123
Edwards GP, Ealey EHM (1975). Aspects of the ecology of the swamp wallaby Wallabia bicolor (Marsupialia: Macropodidae). Australian Mammalogy 1:307 317.
Efron B, Tibshirani RJ (1993). An introduction to the bootstrap. Chapman and Hall, New York, 436p.
Ellison AM (1996). An introduction to Bayesian inference for ecological research and environmental decision making. Ecological Applications 6:1036 1046.
Ellison AM (2004). Bayesian inference in ecology. Ecology Letters 7:509 520.
Environment Australia (1990). Code of Practice for the Humane Shooting of Kangaroos. Environment Australia, Canberra, 5p.
Evans M (1996). Home ranges and movement schedules of sympatric bridled nailtail and black striped wallabies. Wildlife Research 23:547 556.
Feeley KJ, Terborgh JW (2005). The effects of herbivore density on soil nutrients and tree growth in tropical forest fragments. Ecology 86:116 124.
Ferguson SH, Bergerud AT, Ferguson R (1988). Predation risk and habitat selection in the persistence of a remnant caribou population. Oecologia 76:236 245.
Fidler F, Burgman MA, Cumming G, Buttrose R, Thomason N (2006). Impact of criticism of null hypothesis significance testing on statistical reporting practices in conservation biology. Conservation Biology 20:1539 1544.
Fisher DO (2000). Effects of vegetation structure, food and shelter on the home range and habitat use of an endangered wallaby. Journal of Applied Ecology 37:660 671.
Fisher DO, Lara MC (1999). Effects of body size and home range on access to mates and paternity in male bridled nailtail wallabies. Animal Behaviour 58:121 130.
Flowerdew JR, Ellwood SA (2001). Impacts of woodland deer on small mammal ecology. Forestry 74:277 287.
Floyd RB (1980). Density of Wallabia bicolor (Desmarest) (Marsupialia: Macropodidae) in eucalypt plantations of different ages. Australian Wildlife Research 7:333 337. 124
Forestry Tasmania (1999). Monitoring and protecting eucalypt regeneration. Native Forest Silviculture Bulletin No. 12. Forestry Tasmania, Hobart, 40p.
Forsyth DM, Richardson SJ, Menchenton K (2005). Foliar fibre predicts diet selection by invasive Red Deer Cervus elaphus scoticus in a temperate New Zealand forest. Functional Ecology 19:495 504.
Fortin D, Fryxell JM, O'Brodovich L, Frandsen D (2003). Foraging ecology of bison at the landscape and plant community levels: the applicability of energy maximization principles. Oecologia 134:219 227.
Franklin JF, Forman RT (1987). Creating landscape patterns by forest cutting: ecological consequences and principles. Landscape Ecology 1:5 18.
Freeland WJ, Janzen DH (1974). Strategies in herbivory by mammals: the role of plant secondary compounds. American Naturalist 108:269 289.
Fretwell SD, Lucas HL (1970). On territorial behaviour and other factors influencing habitat distribution in birds. Acat Biotheoretica 19:16 36.
Fuller RJ, Gill RMA (2001). Ecological impacts of increasing numbers of deer in British woodland. Forestry 74:193 199.
Fuller TK, DeStefano S (2003). Relative importance of early successional forests and shrubland habitats to mammals in the northeastern United States. Forest Ecology and Management 185:75 79.
Gaines WL, Lyons AL (2003). Crepuscular and nocturnal activity patterns of black bears in the North Cascades of Washington. Northwest Science 77:140 146.
Gaston KJ, Blackburn TM, Goldewijk KK (2003). Habitat conversion and global avian biodiversity loss. Proceedings of the Royal Society of London Series B 270:1293 1300.
Geffen E, Hefner R, Macdonald DW, Ucko M (1992). Habitat selection and home range in the Blanford Fox, Vulpes cana : Compatibility with the resource dispersion hypothesis. Oecologia 91:75 81. 125
Gibbons P, Lindenmayer DB (1996). Issues associated with the retention of hollow bearing trees within eucalypt forests managed for wood production. Forest Ecology and Management 83:245 279.
Gill RMA (1992). A review of damage by mammals in north temperate forests. 3. Impact on trees and forests. Forestry 65:363 388.
Glessner KDG, Britt A (2005). Population density and home range size of Indri indri in a protected low altitude rain forest. International Journal of Primatology 26:855 872.
Gompper ME, Gittleman JL (1991). Home range scaling: intraspecific and comparative trends. Oecologia 87:343 348.
Grigione MM, Beier P, Hopkins RA, Neal D, Padley WD, Schonewald CM, Johnson ML (2002). Ecological and allometric determinants of home range size for mountain lions ( Puma concolor ). Animal Conservation 5:317 324.
Harestad AS, Bunnell FL (1979). Home range and body weight a reevaluation. Ecology 60:389 402.
Henderson DW, Warren RJ, Cromwell JA, Hamilton RJ (2000). Responses of urban deer to a 50% reduction in local herd density. Wildlife Society Bulletin 28:902 910.
Hill GJE, Phinn SR (1993). Revegetated sand mining areas, swamp wallabies and remote sensing: North Stradbroke Island, Queensland. Australian Geographical Studies 31:3 13.
Hobbs NT (1996). Modification of ecosystems by ungulates. Journal of Wildlife Management 60:695 713.
Hobbs NT (2003). Challenges and opportunities in integrating ecological knowledge across scales. Forest Ecology and Management 181:223 238.
Hobbs NT, Hilborn R (2006). Alternatives to statistical hypothesis testing in ecology: A guide to self teaching. Ecological Applications 16:5 19.
Hobbs RJ, Huenneke LF (1992). Disturbance, diversity, and invasion Implications for conservation. Conservation Biology 6:324 337. 126
Hoenig JM, Heisey DM (2001). The abuse of power: the pervasive fallacy of power calculations for data analysis. The American Statistician 55:19 24.
Hollis CJ, Robertshaw JD, Harden RH (1986). Ecology of the swamp wallaby (Wallabia bicolor ) in north eastern NSW. I. Diet. Australian Wildlife Research 13:355 361.
Hood GM (2005). Pop Tools v. 2.6.7. Available at: http://www.cse.csiro.au/poptools .
Horsley SB, Stout SL, DeCalesta DS (2003). White tailed deer impact on the vegetation dynamics of a northern hardwood forest. Ecological Applications 13:98 118.
Hubbard SF, Cook RM, Glover JG, Greenwood JJD (1982). Apostatic selection as an optimal foraging strategy. Journal of Animal Ecology 51:625 633.
Hughes JJ, Ward D, Perrin MR (1994). Predation risk and competition affect habitat selection and activity of Namib desert gerbils. Ecology 75:1397 1405.
Illius AW, Duncan P, Richard C, Mesochina P (2002). Mechanisms of functional response and resource exploitation in browsing roe deer. Journal of Animal Ecology 71:723 734.
Jaccard J, Guilamo Ramos V (2002). Analysis of variance frameworks in clinical child and adolescent psychology: Advanced issues and recommendations. Journal of Clinical Child and Adolescent Psychology 31:278 294.
Jarman PJ, Coulson G (1989). Dynamics and adaptiveness of grouping in macropods. In: Grigg G, Jarman P, Hume I (eds) Kangaroos, Wallabies and Rat kangaroos. Surrey Beatty, Sydney, pp 527 547.
Johnson CN, Bayliss PG (1981). Habitat selection by sex, age and reproductive class in the red kangaroo, Macropus rufus , in western New South Wales. Australian Wildlife Research 8:465 474.
Johnson CN, Jarman PJ (1987). Macropod studies at Wallaby Creek. VI. A validation of the use of dung pellet counts for measuring absolute densities of populations of macropodids. Australian Wildlife Research 14:139 146. 127
Johnson DH (1999). The insignificance of statistical significance testing. Journal of Wildlife Management 63:763 772.
Johnson DH (2002). The importance of replication in wildlife research. Journal of Wildlife Management 66:919 932.
Johnson KA (1980). Spatial and temporal use of habitat by the red necked pademelon, Thylogale thetis (Marsupialia: Macropodidae). Australian Wildlife Research 7:157 166.
Jonsson P, Hartikainen T, Koskela E, Mappes T (2002). Determinants of reproductive success in voles: space use in relation to food and litter size manipulation. Evolutionary Ecology 16:455 467.
Jorge MSP, Peres CA (2005). Population density and home range size of red rumped agoutis ( Dasyprocta leporina ) within and outside a natural Brazil nut stand in southeastern Amazonia. Biotropica 37:317 321.
Katajisto J, Moilanen A (2006). Kernel based home range method for data with irregular sampling intervals. Ecological Modelling 194:405 413.
Kaufmann JH (1974). Habitat use and social organization of nine sympatric species of macropodid marsupials. Journal of Mammalogy 55:66 80.
Kavanagh RP (2000). Effects of variable intensity logging and the influence of habitat variables on the distribution of the Greater Glider Petauroides volans in montane forest, southeastern New South Wales. Pacific Conservation Biology 6:18 30.
Kelt DA, Van Vuren DH (2001). The ecology and macroecology of mammalian home range area. American Naturalist 157:637 645.
Kenward RE (2001). A manual for wildlife radio tagging. Academic Press, London, 286p.
Keough MJ, Mapstone BD (1995). Protocols for Designing Marine Ecological Monitoring Programs Associated With BEK Mills. CSIRO, Canberra, 177p. 128
Kilpatrick HJ, Spohr SM, Lima KK (2001). Effects of population reduction on home ranges of female white tailed deer at high densities. Canadian Journal of Zoology 79:949 954.
Kjellander P, Hewison AJM, Liberg O, Angibault JM, Bideau E, Cargnelutti B (2004). Experimental evidence for density dependence of home range size in roe deer (Capreolus capreolus L.): a comparison of two long term studies. Oecologia 139:478 485.
Knops JMH, Tilman D, Haddad NM, Naeem S, Mitchell CE, Haarstad J, Ritchie ME, Howe KM, Reich PB, Siemann E, Groth J (1999). Effects of plant species richness on invasion dynamics, disease outbreaks, insect abundances and diversity. Ecology Letters 2:286 293.
Krebs JR, Davies NB (1993). An Introduction to Behavioural Ecology, Third edn. Blackwell, Oxford, 420p.
Lair H (1987). Estimating the location of the focal center in red squirrel home ranges. Ecology 68:1092 1101.
Law BS (1993). Roosting and foraging ecology of the Queensland Blossom Bat (Syconycteris australis ) in north eastern New South Wales flexibility in response to seasonal variation. Wildlife Research 20:419 431.
Law BS, Dickman CR (1998). The use of habitat mosaics by terrestrial vertebrate fauna: implications for conservation and management. Biodiversity and Conservation 7:323 333.
Lawler IR, Foley WJ (1999). Swamp wallabies and Tasmanian pademelons show intraspecific preferences for foliage. Australian Forestry 62:17 20.
Lawler IR, Foley WJ, Eschler BM (2000). Foliar concentration of a single toxin creates habitat patchiness for a marsupial folivore. Ecology 81:1327 1338.
LCC (1973). Report on the Melbourne Study Area. In. Land Conservation Council, Melbourne, 444p. 129
LCC (1978). Report on the North Central Study Area. In. Land Conservation Council, Melbourne, 252p. le Mar K, McArthur C (2005). Comparison of habitat selection by two sympatric macropods, Thylogale billardierii and Macropus rufogriseus rufogriseus , in a patchy eucalypt forestry environment. Austral Ecology 30:674 683.
Lechowicz MJ (1982). The sampling characteristics of electivity indices. Oecologia 52:22 30.
Lima SL, Dill LM (1990). Behavioral decisions made under the risk of predation A review and prospectus. Canadian Journal of Zoology 68:619 640.
Lindenmayer DB (1994). Timber harvesting impacts on wildlife: Implications for ecologically sustainable forest use. Australian Journal of Environmental Management 1:56 68.
Lindenmayer DB (1999). Future directions for biodiversity conservation in managed forests: Indicator species, impact studies and monitoring programs. Forest Ecology and Management 115:277 287.
Lindenmayer DB, Cunningham RB, Tanton MT, Nix HA, Smith AP (1991). The conservation of arboreal marsupials in the montane ash forests of the Central Highlands of Victoria, south east Australia. 3. The habitat requirements of Leadbeaters Possum Gymnobelideus leadbeateri and models of the diversity and abundance of arboreal marsupials. Biological Conservation 56:295 315.
Lindenmayer DB, Franklin JF (1997). Managing stand structure as part of ecologically sustainable forest management in Australian mountain ash forests. Conservation Biology 11:1053 1068.
Lindenmayer DB, McCarthy MA (2002). Congruence between natural and human forest disturbance: a case study from Australian montane ash forests. Forest Ecology and Management 155:319 335.
Lindstedt SL, Miller BJ, Buskirk SW (1986). Home range, time, and body size in mammals. Ecology 67:413 418. 130
Linnell JDC, Andersen R (1995). Site tenacity in roe deer: short term effects of logging. Wildlife Society Bulletin 23:31 35.
Loney PE, McArthur C, Sanson GD, Davies NW, Close DC, Jordan GJ (2006). How do soil nutrients affect within plant patterns of herbivory in seedlings of Eucalyptus nitens ? Oecologia 150:409 420.
Louden ASI (1987). The reproductive energetics of lactation in a seasonal macropodid marsupial: comparison of marsupial and eutherian herbivores. In: Louden ASI, Race P (eds) Reproductive Energetics in Mammals. Clarendon Press, Oxford, pp 127 147.
Ludbrook J (1998). Multiple comparison procedures updated. Clinical and Experimental Pharmacology and Physiology 25:1032 1037.
Lugo AE (2000). Effects and outcomes of Caribbean hurricanes in a climate change scenario. Science of the Total Environment 262:243 251.
Lugo AE, Rogers C, Nixon S (2000). Hurricanes, coral reefs and rainforests: Resistance, ruin and recovery in the Caribbean. Ambio 29:106 114.
Lundberg P, Astrom M, Danell K (1990). An experimental test of frequency dependent food selection: winter browsing by moose. Holarctic Ecology 13:177 182.
Lunney D, O'Connell M (1988). Habitat selection by the swamp wallaby, Wallabia bicolor , the red necked wallaby, Macropus rufogriseus , and the common wombat, Vombatus ursinus , in logged, burnt forest near Bega, New South Wales. Australian Wildlife Research 15:695 706.
Lutze M (2003). Standardized measures of regeneration success for sustainable management of Australian native forests. Forest and Wood Products Research and Development Corporation Project No. PN99.810. Available at: http://www.fwprdc. org.au/menu.asp?id=37&start=1&order=.
Lutze MT, Campbell RG, Fagg PC (1999). Development of silviculture in the native State forests of Victoria. Australian Forestry 62:236 244.
Macdonald DW (1983). The ecology of carnivore social behavior. Nature 301:379 384. 131
Mace GM, Harvey PH (1983). Energetic constraints on home range size. American Naturalist 121:120 132.
Manly BFJ (1973). A linear model for frequency dependent selection by predators. Researches on Population Ecology 14:137 150.
Manly BFJ, McDonald LL, Thomas DL, McDonald TL, Erickson WP (2002). Resource Selection by Animals: Statistical Design and Analysis for Field Studies. Kluwer Academic Publishers, Dordrecht, 221p.
Mares MA, Lacher TE, Willig MR, Bitar NA, Adams R, Klinger A, Tazik D (1982). An experimental analysis of social spacing in Tamias striatus . Ecology 63:267 273.
Marsh KJ, Wallis IR, Andrew RL, Foley WJ (2006). The detoxification limitation hypothesis: Where did it come from and where is it going? Journal of Chemical Ecology 32:1247 1266.
Martin JK (2006). Den use and home range characteristics of bobucks, Trichosurus cunninghami , resident in a forest patch. Australian Journal of Zoology 54:225 234.
McCorquodale SM (2003). Sex specific movements and habitat use by elk in the Cascade Range of Washington. Journal of Wildlife Management 67:729 741.
McCune B, Grace JB, Urban DL (2002). Analysis of Ecological Communities. MjM Software Design, Gleneden Beach, 300p.
McLoughlin PD, Ferguson SH (2000). A hierarchical pattern of limiting factors helps explain variation in home range size. Ecoscience 7:123 130.
McNab BK (1963). Bioenergetics and determination of home range size. American Naturalist 97:133 140.
Menkhorst PW (1995). Black wallaby. In: Menkhorst PW (ed) Mammals of Victoria Distribution, Ecology and Conservation. Oxford University Press, Melbourne, pp 153 154.
Mielke PW, Berry KJ, Johnson ES (1976). Multiresponse permutation procedures for a priori classifications. Communications in Statistics A5:1409 1424. 132
Mitchell MS, Powell RA (2003). Response of black bears to forest management in the southern Appalachian mountains. Journal of Wildlife Management 67:692 705.
Moloney KA, Levin SA (1996). The effects of disturbance architecture on landscape level population dynamics. Ecology 77:375 394.
Montague TL (1996). The extent, timing and economics of browsing damage in eucalypt and pine plantations of Gippsland, Victoria. Australian Forestry 59:120 129.
Moran MD (2003). Arguments for rejecting the sequential Bonferroni in ecological studies. Oikos 100:403 405.
Morrison SF, Forbes GJ, Young SJ (2002). Browse occurrence, biomass, and use by white tailed deer in a northern New Brunswick deer yard. Canadian Journal of Forest Research 32:1518 1524.
Moser B, Schutz M, Hindenlang KE (2006). Importance of alternative food resources for browsing by roe deer on deciduous trees: The role of food availability and species quality. Forest Ecology and Management 226:248 255.
Moser BW, Witmer GW (2000). The effects of elk and cattle foraging on the vegetation, birds, and small mammals of the Bridge Creek Wildlife Area, Oregon. International Biodeterioration and Biodegradation 45:151 157.
Mosnier A, Ouellet JP, Sirois L, Fournier N (2003). Habitat selection and home range dynamics of the Gaspé caribou: a hierarchical analysis. Canadian Journal of Zoology 81:1174 1184.
Moss GL, Croft DB (1999). Body condition of the red kangaroo (Macropus rufus) in arid Australia: The effect of environmental condition, sex and reproduction. Australian Journal of Ecology 24:97 109.
Mount AB (1961). Regeneration surveys for cut over areas of ash type eucalypt forests. Appita 15:77 86.
Murdoch WW, Peterson CH, Evans FC (1972). Diversity and pattern in plants and insects. Ecology 53:819 829. 133
Neilsen WA, Wilkinson GR (1995). Browsing damage, site quality and species selection a case study of their effects on the economic viability of eucalypt plantations in north eastern Tasmania. Proceedings of the 10th Australian Vertebrate Pest Control Conference, Hobart, pp 161 170.
Nelder JA (1971). Contribution to the discussion of R. O'Neill and G.B Wetherill 1971 The present state of multiple comparison methods. Journal of the Royal Statistical Society Series B 33:218 250.
Nelder JA (1999). From statistics to statistical science. Journal of the Royal Statistical Society Series D 48:257 267.
Newell GR (1999). Responses of Lumholtz's tree kangaroo ( Dendrolagus lumholtzi ) to loss of habitat within a tropical rainforest fragment. Biological Conservation 91:181 189.
Nilsen EB, Linnell JDC, Andersen R (2004). Individual access to preferred habitat affects fitness components in female roe deer Capreolus capreolus . Journal of Animal Ecology 73:44 50.
Norbury GL (1988). Microscopic analysis of herbivore diets a problem and a solution. Australian Wildlife Research 15:51 57.
Norbury GL, Sanson GD (1992). Problems with measuring diet selection of terrestrial, mammalian herbivores. Australian Journal of Ecology 17:1 7.
O'Neill R, Wetherill GB (1971). The present state of multiple comparison methods. Journal of the Royal Statistical Society Series B 33:218 250.
Orians GH, Wittenberger JF (1991). Spatial and temporal scales in habitat selection. American Naturalist 137:S29 S49.
Osawa R (1990). Feeding strategies of the swamp wallaby, Wallabia bicolor , on North Stradbroke Island, Queensland. 1. Composition of diets. Australian Wildlife Research 17:615 621.
Otis DL, White GC (1999). Autocorrelation of location estimates and the analysis of radiotracking data. Journal of Wildlife Management 63:1039 1044. 134
Partl E, Szinovata V, Reimoser F, Schweiger Adler J (2002). Forest restoration and browsing impact by roe deer. Forest Ecology and Management 159:87 100.
Pastor J, Dewey B, Naiman RJ, McInnes PF, Cohen Y (1993). Moose browsing and soil fertility in the boreal forests of Isle Royale national park. Ecology 74:467 480.
Pastor J, Naiman RJ (1992). Selective foraging and ecosystem processes in boreal forests. American Naturalist 139:690 705.
Perelberg A, Saltz D, Bar David S, Dolev A, Yom Tov Y (2003). Seasonal and circadian changes in the home ranges of reintroduced Persian fallow deer. Journal of Wildlife Management 67:485 492.
Perneger TV (1998). What's wrong with Bonferroni adjustments. British Medical Journal 316:1236 1238.
Persson IL, Pastor J, Danell K, Bergstrom R (2005). Impact of moose population density on the production and composition of litter in boreal forests. Oikos 108:297 306.
Pickett STA, White PS (1985). The Ecology of Natural Disturbance and Patch Dynamics. Acedemic Press, Orlando, Florida, 472p.
Pierce BM, Bowyer RT, Bleich VC (2004). Habitat selection by mule deer: Forage benefits or risk of predation? Journal of Wildlife Management 68:533 541.
Pulliam HR (1975). Diet optimization with nutrient constraints. American Naturalist 109:765 768.
Pyke GH, Pulliam HR, Charnov EL (1977). Optimal foraging: a selective review of theory and tests. Quarterly Review of Biology 52:137 154.
Ramsey DSL, Wilson JC (1997). The impact of grazing by macropods on coastal foredune vegetation in southeast Queensland. Australian Journal of Ecology 22:288 297.
Reimoser F, Armstrong H, Suchant R (1999). Measuring forest damage of ungulates: what should be considered. Forest Ecology and Management 120:47 58. 135
Reimoser F, Gossow H (1996). Impact of ungulates on forest vegetation and its dependence on the silvicultural system. Forest Ecology and Management 88:107 119.
Relyea RA, Lawrence RK, Demarais S (2000). Home range of desert mule deer: Testing the body size and habitat productivity hypotheses. Journal of Wildlife Management 64:146 153.
Rettie WJ, Messier F (2000). Hierarchical habitat selection by woodland caribou: its relationship to limiting factors. Ecography 23:466 478.
Reynolds TD, Laundre JW (1990). Time intervals for estimating pronghorn and coyote home ranges and daily movements. Journal of Wildlife Management 54:316 322.
Roback PJ, Askins RA (2005). Judicious use of multiple hypothesis tests. Conservation Biology 19:261 267.
Rooney SM, Wolfe A, Hayden TJ (1998). Autocorrelated data in telemetry studies: time to independence and the problem of behavioural effects. Mammal Review 28:89 98.
Rooney TP (2001). Deer impacts on forest ecosystems: a North American perspective. Forestry 74:201 208.
Rothman KJ (1990). No adjustments are needed for multiple comparisons. Epidemiology 1:43 46.
Rozeboom WW (1960). The fallacy of the null hypothesis significance test. Psychological Bulletin 57:416 428.
Ryan KC (2002). Dynamic interactions between forest structure and fire behavior in boreal ecosystems. Silva Fennica 36:13 39.
Sanson GD (1978). The evolution and significance of mastication in the macropodidae. Australian Mammalogy 2:23 28.
Sanson GD (1980). The morphology and occlusion of the molariform cheek teeth in some macropodinae (Marsupialia, Macropodidae). Australian Journal of Zoology 28:341 365. 136
Schoenecker KA, Singer FJ, Zeigenfuss LC, Binkley D, Menezes RSC (2004). Effects of elk herbivory on vegetation and nitrogen processes. Journal of Wildlife Management 68:837 849.
Seaman DE, Millspaugh JJ, Kernohan BJ, Brundige GC, Raedeke KJ, Gitzen RA (1999). Effects of sample size on kernel home range estimates. Journal of Wildlife Management 63:739 747.
Sebire I (2001). Browsing in Victorian Native State Forests Results of the 1998 Statewide Survey. Department of Natural Resources and Environment, Melbourne, 29p.
Shepherd N (1987). Condition and recruitment of kangaroos. In: Caughley G, Shepherd N, Short J (eds) Kangaroos Their Ecology and Management in the Sheep Rangelands of Australia. Cambridge University Press, Cambridge, pp 135 158.
Shepherd PCF, Lank DB (2004). Marine and agricultural habitat preferences of dunlin wintering in British Columbia. Journal of Wildlife Management 68:61 73.
Simberloff D (1999). The role of science in the preservation of forest biodiversity. Forest Ecology and Management 115:101 111.
Smith PG (2004). Automated log ratio analysis of compositional data: software suited to analysis of habitat preference from radiotracking data. Bat Research News 45:16.
Smithson M (2003). Confidence Intervals. Quantitative Applications in the Social Sciences Series, No. 140. Sage, Thousand Oaks, 104p.
Sokal RR, Rohlf FJ (1995). Biometry, Third edn. W.H. Freeman and Company, New York, 887p.
Southwell C (1989). Techniques for monitoring the abundance of kangaroo and wallaby populations. In: Grigg G, Jarman P, Hume I (eds) Kangaroos, Wallabies and Rat Kangaroos. Surrey Beatty, Sydney, pp 659 693.
Spalinger DE, Hobbs NT (1992). Mechanisms of foraging in mammalian herbivores: new models of functional response. American Naturalist 140:325 348. 137
Sprent JA, McArthur C (2002). Diet and diet selection of two species in the macropodid browser grazer continuum: do they eat what they 'should'? Australian Journal of Zoology 50:183 192.
Squire RO, Dexter BD, Eddy AR, Fagg PC, Campbell RG (1991a). Regeneration Silviculture for Victoria's Eucalypt Forests. SSP Technical Report No. 6. Department of Conservation and Environment, Melbourne, 38p.
Squire RO, Flinn DW, Campbell RG (1991b). Silvicultural research for sustained wood production and biosphere conservation in the pine plantations and native eucalypt forests of south eastern Australia. Australian Forestry 54:120 133.
St Louis A, Ouellet JP, Crete M, Maltais J, Huot J (2000). Effects of partial cutting in winter on white tailed deer. Canadian Journal of Forest Research 30:655 661.
Stewart Oaten A (1995). Rules and judgments in statistics: Three examples. Ecology 76:2001 2009.
Stewart Oaten A, Bence JR, Osenberg CW (1992). Assessing effects of unreplicated perturbations: No simple solutions. Ecology 73:1396 1404.
Stirrat SC (2002). Foraging ecology of the agile wallaby ( Macropus agilis ) in the wet dry tropics. Wildlife Research 29:347 361.
Stirrat SC (2003a). Body condition and blood chemistry of agile wallabies (Macropus agilis) in the wet dry tropics. Wildlife Research 30:59 67.
Stirrat SC (2003b). Seasonal changes in home range area and habitat use by the agile wallaby ( Macropus agilis ). Wildlife Research 30:593 600.
Storr GM (1961). Microscopic analysis of faeces, a technique for ascertaining the diet of herbivorous mammals. Australian Journal of Biological Sciences 14:157 165.
Sullivan TP, Sullivan DS (2001). Influence of variable retention harvests on forest ecosystems. II. Diversity and population dynamics of small mammals. Journal of Applied Ecology 38:1234 1252. 138
Swihart RK, Slade NA (1985). Testing for independence of observations in animal movements. Ecology 66:1176 1184.
Thompson ID, Baker JA, Ter Mikaelian M (2003). A review of the long term effects of post harvest silviculture on vertebrate wildlife, and predictive models, with an emphasis on boreal forests in Ontario, Canada. Forest Ecology and Management 177:441 469.
Tixier H, Duncan P, Scehovic J, Yani A, Gleizes M, Lila M (1997). Food selection by European roe deer ( Capreolus capreolus ): Effects of plant chemistry, and consequences for the nutritional value of their diets. Journal of Zoology 242:229 245.
Triggs B (2004). Tracks, Scats and Other Traces: A Field Guide to Australian Mammals. Oxford University Press, Melbourne, 340p.
Tripler CE, Canham CD, Inouye RS, Schnurr JL (2002). Soil nitrogen availability, plant luxury consumption, and herbivory by white tailed deer. Oecologia 133:517 524.
Troy S, Coulson G (1993). Home range of the swamp wallaby, Wallabia bicolor . Wildlife Research 20:571 577.
Troy S, Coulson G, Middleton D (1992). A comparison of radio tracking and line transect techniques to determine habitat preferences in the swamp wallaby ( Wallabia bicolor ) in south eastern Australia. In: Priede IG, Swift SM (eds) Wildlife Telemetry: Remote Monitoring and Tracking of Animals. Ellis Horwood, Chichester, pp 651 660.
Tufto J, Andersen R, Linnell J (1996). Habitat use and ecological correlates of home range size in a small cervid: The roe deer. Journal of Animal Ecology 65:715 724.
Tyndale Biscoe CH, Smith RFC (1969). Studies on the marsupial glider, Schoinobates volans (Kerr). III. Response to habitat destruction. Journal of Animal Ecology 38:651 659.
Verheyden Tixier H, Duncan P, Ballon P, Guillon N, Guillon N (1998). Selection of hardwood saplings by European roe deer: Effects of variation in the availability of palatable species and of understory vegetation. Revue D Ecologie La Terre Et La Vie 53:245 253. 139
Vernes K, Marsh H, Winter J (1995). Home range characteristics and movement patterns of the red legged pademelon ( Thylogale stigmatica ) in a fragmented tropical rainforest. Wildlife Research 22:699 708.
Wade PR (2000). Bayesian methods in conservation biology. Conservation Biology 14:1308 1316.
Wahungu GM, Catterall CP, Olsen MF (2001). Predator avoidance, feeding and habitat use in the red necked pademelon, Thylogale thetis , at rainforest edges. Australian Journal of Zoology 49:45 58.
Wallis IR, Watson ML, Foley WJ (2002). Secondary metabolites in Eucalyptus melliodora : Field distribution and laboratory feeding choices by a generalist herbivore, the common brushtail possum. Australian Journal of Zoology 50:507 519.
Ward D, Saltz D (1994). Foraging at different spatial scales: dorcas gazelles foraging for lilies in the Negev desert. Ecology 75:48 58.
Wardell Johnson G, Williams M (2000). Edges and gaps in mature karri forest, south western Australia: Logging effects on bird species abundance and diversity. Forest Ecology and Management 131:1 21.
Wardle DA, Barker GM, Yeates GW, Bonner KI, Ghani A (2001). Introduced browsing mammals in New Zealand natural forests: Aboveground and belowground consequences. Ecological Monographs 71:587 614.
Wardle DA, Bonner KI, Barker GM (2002). Linkages between plant litter decomposition, litter quality, and vegetation responses to herbivores. Functional Ecology 16:585 595.
Welch D, Staines BW, Scott D, French DD (1992). Leader browsing by red and roe deer on young sitka spruce trees in western Scotland. 2. Effects on growth and tree form. Forestry 65:309 330.
Welch D, Staines BW, Scott D, French DD, Catt DC (1991). Leader browsing by red and roe deer on young sitka spruce trees in western Scotland. 1. Damage rates and the influence of habitat factors. Forestry 64:61 82. 140
Westoby M (1974). Analysis of diet selection by large generalist herbivores. American Naturalist 108:290 304.
Wiens JA (1989). Spatial scaling in ecology. Functional Ecology 3:385 397.
Wilkinson GR, Neilsen WA (1995). Implications of early browsing damage on the long term productivity of eucalypt forests. Forest Ecology and Management 74:117 124.
Williams MR, Abbott I, Liddelow GL, Vellios C, Wheeler IB, Mellican AE (2001). Recovery of bird populations after clearfelling of tall open eucalypt forest in Western Australia. Journal of Applied Ecology 38:910 920.
Wilson BA, Friend GR (1999). Responses of Australian mammals to disturbance: A review. Australian Mammalogy 21:87 105.
Wood MS (2002). Resource Partitioning in Sympatric Populations of Red necked and Black Wallabies. PhD Thesis, Deakin University, Geelong, 288p.
Yoccoz NG (1991). Use, overuse, and misuse of significance tests in evolutionary biology and ecology. Bulletin of the Ecological Society of America 72:106 111.
Zamora R, Gomez JM, Hodar JA, Castro J, Garcia D (2001). Effect of browsing by ungulates on sapling growth of Scots pine in a Mediterranean environment: consequences for forest regeneration. Forest Ecology and Management 144:33 42. 141
Appendix 1. Raw home range data.
Table A1. 95% fixed kernel home range estimates for wallabies used in Chapter 2. Site Sex ID Home Range Estimate n Before n During n After Impact 1 F F1 30 19.4 26 14.6 32 14.6 Impact 1 F F2 33 24.9 29 18.4 31 19.3 Impact 2 F F1 33 19.2 32 12.0 32 15.4 Impact 2 F F2 32 9.7 32 21.0 31 20.3 Impact 2 F F3 36 8.8 32 15.3 30 33.2 Impact 2 M M1 31 57.9 31 46.5 na na Impact 2 M M2 31 19.0 33 13.5 28 64.2 Impact 2 M M3 31 39.3 33 31.5 30 28.3 Control 1 F na 33 14.0 na na 30 11.4 Control 2 F na 32 19.0 na na 30 26.8 Control 3 F na 33 13.0 na na 34 15.2 Control 4 M na 32 25.1 na na 30 29.9 Control 4 M na 32 15.7 na na 32 6.3 Control 5 F na 32 20.1 na na 32 36.8 Control 5 F na 34 29.1 na na 34 19.7 142
Table A2 . Raw home range data used in chapters 4 and 6. The control male marked with a * was removed from all analyses in Chapter 6 (see Data analysis section). Some diurnal home ranges were not calculated due to an inadequate number of locations. Total 95% home Diurnal 95% Site Type Sex Weight (kg) Total n range (ha) Diurnal n home range (ha) Recent harvest F 14.0 56 24.29 43 16.78 Recent harvest F 12.0 54 16.77 43 14.90 Recent harvest F 15.5 38 35.08 31 33.26 Recent harvest F 15.0 55 22.00 44 16.66 Recent harvest F 13.0 55 25.83 43 20.81 Recent harvest M 20.0 41 35.01 33 27.84 Recent harvest M 24.0 53 77.44 42 85.93 5 yr old F 14.0 56 20.04 44 14.48 5 yr old F 14.0 62 6.03 48 3.36 5 yr old F 15.0 37 3.39 30 2.80 5 yr old F 14.0 49 9.78 37 6.17 5 yr old F 14.0 47 24.95 34 16.16 5 yr old F 14.0 64 14.27 50 12.58 5 yr old F 13.5 59 19.76 48 17.85 5 yr old F 13.0 33 4.76 na na 5 yr old M 21.5 46 42.77 35 42.78 5 yr old M 17.5 44 57.07 33 53.89 5 yr old M 14.5 37 16.04 30 11.95 5 yr old M 19.0 39 23.86 32 17.24 5 yr old M 22.0 47 30.18 35 29.45 5 yr old M 18.0 48 29.47 36 24.71 5 yr old M 22.5 41 40.91 33 36.30 10 yr old F 15.0 47 32.79 37 24.37 10 yr old F 13.5 47 29.03 36 18.00 10 yr old F 13.5 58 20.30 43 14.83 10 yr old F 10.5 54 18.18 41 10.70 10 yr old F 14.0 59 18.37 45 16.38 10 yr old M 17.0 50 82.57 39 74.75 10 yr old M 17.5 47 71.28 36 77.90 10 yr old M 11.0 42 33.75 34 18.78 control F 12.5 63 13.48 52 10.76 control F 17.0 47 18.09 39 15.81 control F 14.0 33 24.92 na na control F 12.0 30 19.42 na na control F 11.0 63 22.26 49 21.37 control F 14.5 67 18.36 55 17.95 control F 9.0 35 22.39 na na control F 13.5 38 15.29 32 16.62 control F 15.0 87 29.23 68 23.14 control F 12.0 64 37.53 54 30.58 control M 21.0 32 46.93 26 50.70 control M 23.0 52 78.11 41 82.13 control M 17.0 30 38.91 na na control M 20.0 36 33.80 29 30.80 control M 18.5 33 46.45 26 46.00 control M 24.0 32 43.64 na na control M 18.0 62 27.86 49 19.97 control* M 21.5 89 16.59 71 15.12
Minerva Access is the Institutional Repository of The University of Melbourne
Author/s: DI STEFANO, JULIAN
Title: Home range size and resource selection by the swamp wallaby, Wallabia bicolor, in a landscape modified by timber harvesting
Date: 2007
Citation: Di Stefano, J. (2007). Home range size and resource selection by the swamp wallaby, Wallabia bicolor, in a landscape modified by timber harvesting, PhD thesis, Zoology, University of Melbourne.
Publication Status: Unpublished
Persistent Link: http://hdl.handle.net/11343/39323
File Description: Home range size and resource selection by the swamp wallaby, Wallabia bicolor, in a landscape modified by timber harvesting
Terms and Conditions: Terms and Conditions: Copyright in works deposited in Minerva Access is retained by the copyright owner. The work may not be altered without permission from the copyright owner. Readers may only download, print and save electronic copies of whole works for their own personal non-commercial use. Any use that exceeds these limits requires permission from the copyright owner. Attribution is essential when quoting or paraphrasing from these works.