<<

University of Calgary PRISM: University of Calgary's Digital Repository

Graduate Studies The Vault: Electronic Theses and Dissertations

2015-12-16 Scaling of Density-Dependent Reproduction in Bee-Pollinated Forbs of Logged Forests

Johnson, Sarah A.

Johnson, S. A. (2015). Scaling of Density-Dependent Reproduction in Bee-Pollinated Forbs of Logged Forests (Unpublished master's thesis). University of Calgary, Calgary, AB. doi:10.11575/PRISM/27709 http://hdl.handle.net/11023/2685 master thesis

University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission. Downloaded from PRISM: https://prism.ucalgary.ca UNIVERSITY OF CALGARY

Scaling of Density-Dependent Reproduction in Bee-Pollinated Forbs of Logged Forests

by

Sarah Alexandra Johnson

A THESIS

SUBMITTED TO THE FACULTY OF GRADUATE STUDIES

IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE

DEGREE OF MASTER OF SCIENCE

GRADUATE PROGRAM IN BIOLOGICAL SCIENCES

CALGARY, ALBERTA

DECEMBER, 2015

© Sarah A. Johnson 2015

ABSTRACT

Plant reproduction can be impacted by a variety of influences at a range of spatial scales. In the face of accelerating anthropogenic habitat disturbance, it is worth understanding how communities function within highly altered landscapes. I examined how size and number varied for nine species of understory forb within logged foothills forests of southern Alberta. I examined local relationships between reproductive output and floral neighbourhood, bee abundance, and habitat variation, and how these might be modified along a gradient of landscape-scale clearcut logging. I found that local variables best explained seed production, and heterospecifics were generally more beneficial than expected. Further, logging in the landscape modified local interactions above a threshold point of approximately 50% logging in a 1.77 km2 area, predominantly for more habitat-specialized species. These results have implications for forest management, and for the importance of testing for interactions between explanatory variables, even across spatial scales.

ii ACKNOWLEDGEMENTS

This thesis only came to fruition due to the help of multiple people and funding sources, and I am forever in their debt. Thank you to Ralph Cartar for guiding me through and out the other side of many treacherous graduate school peaks and troughs; I could not have asked for a more supportive and knowledgable supervisor. Thank you to my committee members (new and old) and my external examiner for their words of wisdom and guidance along the way: Kathreen Ruckstuhl, Jana Vamosi, Paul Galpern and David Goldblum. Thank you to my parents, NSERC CANPOLIN, NSERC graduate scholarships, and ACA Biodiversity Grants for their contributions to my project and salary funds (and a bit of extra thanks to Mom and Dad). Thank you to Michelle Seifert, Evan Whitmore, Savannah Steinhilber, and Mary Fleet for help with collecting, counting, measuring, and entering data for an obscene number of . Thank you to Stevi Vanderzwan and Stephen Hausch for their comaradery and support during the final thesis- writing push. Thank you to Kyle Wilson for his unwavering emotional support and lots of help with analyses, coding, and editing. Thank you to the past and present members of the Cartar Reid lab for being there for me when I needed it, especially: Alex Farmer, Riley Waytes, Sam Robinson, Rola Kutby, Jenn Retzlaff, Megan Goulding, Leanna Lachowsky and Haydeé Peralta- Vásquez. Without all of these people (and more) believing in me, I definitely wouldn’t have made it this far, and I am eternally grateful.

iii TABLE OF CONTENTS ABSTRACT ...... ii ACKNOWLEDGEMENTS ...... iii TABLE OF CONTENTS ...... iv LIST OF TABLES ...... vi LIST OF FIGURES ...... viii LIST OF APPENDICES ...... xii CHAPTER 1: GENERAL INTRODUCTION ...... 1 CHAPTER 2: REPRODUCTION OF UNDERSTORY FOREST FORBS: THE RELATIVE EFFECTS OF FLORAL NEIGHBOURHOOD, POLLINATORS, AND HABITAT ACROSS SPATIAL SCALES ...... 8 2.1 Introduction ...... 8 2.1.1 Background ...... 8 2.1.2 Hypotheses and predictions ...... 14 2.2 Methods ...... 20 2.2.1 Field sites ...... 20 2.2.2 Study species ...... 20 2.2.3 Sampling design ...... 22 2.2.4 Sample processing ...... 25 2.2.5 Spatial scale ...... 26 2.2.6 Data analysis ...... 27 2.3 Results ...... 30 2.3.1 Spatial scale ...... 30 2.3.2 Variable effects ...... 35 2.4 Discussion ...... 46 CHAPTER 3: CROSS-SCALE INTERACTIVE EFFECTS OF LOGGING ON UNDERSTORY FORB REPRODUCTION ...... 66 3.1 Introduction ...... 66 3.1.1 Background ...... 66 3.1.2 Hypotheses and predictions ...... 70 3.2 Methods ...... 73 3.3 Results ...... 77 3.4 Discussion ...... 91 CHAPTER 4: GENERAL DISCUSSION ...... 97 LITERATURE CITED ...... 100 APPENDIX A ...... 123 APPENDIX B ...... 125

iv APPENDIX C ...... 129 APPENDIX D ...... 136

v LIST OF TABLES Table 2.1: Simplified hypothesis table outlining expected relationships between variables and seed size and number, at both the local (=smaller) and patch (=larger) scales...... 19

Table 2.2: Species of the most common bee-pollinated wildflowers sampled and analyzed for impacts on reproductive potential (seed size and number). Samples were collected in Kananaskis Country, AB from June-August 2012. The habitat counts indicate number of sites in which the species was sampled. Species are binned into three floral morphology categories, increasing in handling time for bumble bee visitors...... 21

Table 2.3: Summary statistics for seed size and number for each species analyzed. Number of samples, means, and standard errors are reported for each respective seed character. As seeds sampled for size were only a subset of all seeds counted, mean seed count for each species was calculated as a mean per pod, grouping structure, or mean “patch total seed production”...... 26

Table 2.4: Candidate models to explain seed size or seed number, selected based upon ecological hypotheses. All models also contained nuisance covariates and random effects of location to control for their potential influence on the response variable...... 29

Table 2.5: Generalized additive mixed model selection from the set of a priori candidate models (see Table 2.4), examining effects on seed size for all plant species. Only models for which ΔAIC was ≤ 10 are reported. Akaike model weights (wi) show probability that the given model is the best among the entire set of candidate models for that particular species. All models include nuisance covariates (Julian day, Spatial PC1) and nested random effects (transect location in habitat in transect in site) controlling for repeated measures within scaling locations. Species are binned into three floral morphology categories, increasing in handling time for bumble bee visitors...... 31

Table 2.6: Generalized additive mixed model selection from the set of a priori candidate models (see Table 2.4), examining effects on seed number for plant species that produced variable numbers of seeds. Only models for which ΔAIC was ≤ 10 are reported. Akaike model weights (wi) show probability that the given model is the best among the entire set of candidate models for that particular species. All models include nuisance covariates (Julian day, Spatial PC1) and nested random effects (transect location in habitat in transect in site) controlling for repeated measures within scaling locations. Species are binned into three floral morphology categories, increasing in handling time for bumble bee visitors...... 32

Table 2.7: Summary of the results of 6 studies and the present study, when examining differential effects of increasing spatial scale on and/or pollinators, and their interactions. The range of scales tested in each study is reported, along with the most important scale(s) and the processes that acted at the selected scale(s). Some studies reported alternative measurements of scale (e.g., radii), but all were converted to area in m2 for ease of comparison...... 50

Table 2.8: Summary of each focal species’ mating system and pollen limitation, from the literature. Species are ordered broadly from “most pollen-limited” to “least pollen-limited”, with no previous data found for L. ochroleucus...... 63

vi Table 3.1: Summary of each species’ habitat preferences, displayed through frequency of presence in forests and clearcuts, and mean local conspecific floral densities (not size adjusted) in each habitat type. Percentages were calculated for each to use as habitat preference metrics – presence of sites was calculated as the proportion of all sites sampled that the species was found in that particular habitat (forest or clearcut), and clearcut density preferences (CC%) were calculated scaled to species-specific floral number (i.e., meanConForest / (meanConForest + meanConClearcut)). Species are ordered by their presence in site forests, and the scale at which local densities were most important in Chapter 2 is bolded for each species...... 78

Table 3.2: Summary of each focal species’ habitat preferences, from the literature...... 79

Table 3.3: Model selection from the set of candidate landscape interactions, examining effects on seed size and number for different plant species. Only the base local model (Table 2.4) and interaction models for which ΔAIC was ≤ 10 are reported. Akaike model weights (wi) show probability that the given model is the best among the set of candidates. All models (GAMM) include nuisance covariates and nested random effects (see Tables 2.4, 2.5). Species are organized by categories of their best-selected models (logging, habitat, and no landscape interaction (= local model)), and within logging interactions, by how continuous logging was. “Type” from Table 3.1 is listed with each species (F = forest, CC = clearcut)...... 81

vii LIST OF FIGURES Figure 2.1: Density-visitation curve modified from Rathcke (1983) showing how pollinator visitation and subsequent plant reproduction are expected to respond to changes in floral density. At very low floral densities, isolated plants may suffer from Allee effects, at low floral densities, facilitation should dominate, and densities higher than maximum per- visitation should result in competitive interactions among ...... 16

Figure 2.2: A modification of Rathcke’s (1983) density-visitation curve, indicating how effects of neighbouring conspecific and heterospecific densities may differ from one another. Heterospecific facilitation (positive slope) is limited to lower floral densities, and competition (negative slope) also begins at lower floral densities for heterospecifics than conspecifics. ... 17

Figure 2.3: Locations of sixteen 176 hectare field sites along the eastern foothills slopes of Kananaskis Country, AB. Sites are categorized into 3 levels of logging intensity: low (≤30% logging; blue, n=5), medium (31%-47%; orange, n=6) and high (>47%; red, n=5). See Chapter 3 for further explanation of logging impacts. Image source: Google Earth. Image date: 2013... 21

Figure 2.4: Schematic of sites representing a) low landscape logging (25% clearcuts in pink and blue; Powderface Trail) and b) high landscape logging (60% clearcut in yellow, Sibbald Flats) with black lines indicating sampled transects with 7 per habitat type (forest and clearcut) per site. Magnified lower-right diagram displays sampling design of an example transect (100 m by 3 m, not to scale). Small circles grouped at approximately 25 m and 75 m represent samples of local floral neighbourhoods...... 23

Figure 2.5: An example schematic (not to scale) outlining the spatial scales focussed upon in this chapter – two levels of “local” indicating small-scale counts of neighbours (conspecific in red and heterospecific in blue) immediately around seed collection sites (focal individuals) located at points within a transect “patch”, and two levels of larger-scale counts of neighbours in the general area surrounding focal individuals...... 27

Figure 2.6: A visualization of the relative importance of 7 a priori models (reported in Table 2.4) across spatial scales for explaining seed size for each focal species. ΔAIC values are calculated for all models across all spatial scales per each individual species (ie. ΔAIC = AICmodel,spec1 - AICmin,spec1) and axes for each panel are inverted so that the best models are on top. Species are indicated by abbreviations in the lower right corner of each plot: a) Arnica cordifolia (A.c), b) Eurybia conspicua (E.c), c) Campanula rotundifolia (C.r), d) Castilleja miniata (C.m), e) Geranium richardsonii (G.r), f) Mertensia paniculata (M.p), g) Linnaea borealis (L.b), h) Lathyrus ochroleucus (L.o), and i) Vicia americana (V.a). Species are clustered into boxes representing categories of floral morphology: blue discs, purple pendular, and red zygomorphs. Legend in the top right corner displays models, where green models involve plant effects, orange bee effects, blue bee and plant effects combined, and red habitat effects. The solid black line across each graph is plotted at ΔAIC = 10, and any models located below that line do not have substantial evidence supporting their estimates. Models with ΔAIC > 50 are not plotted...... 33

Figure 2.7: A visualization of the relative importance of 7 a priori models (reported in Table 2.4) across spatial scales for explaining seed number for each focal species. ΔAIC values are

viii calculated for all models across all spatial scales per each individual species (ie. ΔAIC = AICmodel,spec1 - AICmin,spec1) and axes for each panel are inverted so that the best models are on top. Species are indicated by abbreviations in the middle right of each plot: a) Arnica cordifolia (A.c), b) Campanula rotundifolia (C.r), c) Mertensia paniculata (M.p), d) Eurybia conspicua (E.c), e) Geranium richardsonii (G.r), f) Castilleja miniata (C.m), and g) Vicia americana (V.a). Species are clustered into boxes representing categories of floral morphology: blue discs, purple pendular, and red zygomorphs. Legend in the top right corner displays models, where green models involve plant effects, orange bee effects, blue bee and plant effects combined, and red habitat effects. The solid black line across each graph is plotted at ΔAIC = 10, and any models located below that line do not have substantial evidence supporting their estimates. Models with ΔAIC > 80 are not plotted...... 34

Figure 2.8: Summary of all component smooth functions for additive effects. Each smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, controlling for all other variables in the model. Plots are arranged by response variable (top row: seed size; bottom row: seed number). Species are grouped into categories of floral morphology: the four panels within the purple box are pendular and the six panels within the red box are zygomorphic. Species are indicated by abbreviations in the upper left corner of each plot: Campanula rotundifolia (C.r), Castilleja miniata (C.m), Linnaea borealis (L.b), Mertensia paniculata (M.p) and Vicia americana (V.a). Labels a-j before species abbreviations are included for in-text referencing. Shaded areas surrounding each curve indicate 2 * standard error bounds. See Appendix D for complete per-species model information...... 37

Figure 2.9: Direct comparison between effects of conspecifics (blue) and heterospecifics (red) for the two species with additive impacts on seed traits. Each smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Upper panel shows seed size effects and lower two panels show effects on seed number. Effects of conspecifics are shown in blue, and heterospecifics in red, where the center line represents the estimate and the red or blue shaded regions represent 2 * standard error bounds, including the uncertainty about the overall mean...... 39

Figure 2.10: Additive effects of both local (10 m2) canopy cover and habitat type (CC = clearcut, F = forest) on ln(seed size) in M. paniculata (pendular). The y-axis is adjusted to represent the partial component contribution of each variable, controlling for all other variables in the model. Shaded curves indicate 2 * standard error bounds, and dotted lines on habitat plot indicate one standard error. See Appendix D for complete model information...... 40

Figure 2.11: Summary of the relative location of peaks in seed production (both seed size and seed number) for all species in which a model with an interaction between two or more variables was best supported. The upper left panel (a) summarizes species whose best model was Plant 3 (see Table 2.4 for variables in this model), and the lower two panels (b, c) summarize species whose best model was Bee Plant 2 (see Table 2.4). Species are colour-coded by their flower morphology groupings. See Appendix D for complete per-species model information with more detailed individual figures...... 41

ix Figure 2.12: Summary of the relative location of seed production lows (both seed size and seed number) for all species in which a model with an interaction between two or more variables was best supported. The upper left panel (a) summarizes species whose best model was Plant 3 (see Table 2.4 for variables in this model), and the lower two panels (b, c) summarize species whose best model was Bee Plant 2 (see Table 2.4). Species are colour-coded by their flower morphology groupings. See Appendix D for complete per-species model information with more detailed individual figures...... 42

Figure 3.1: Frequency histograms of percent logging in the landscape (left panel) and average age of clearcut in the landscape (right panel) across all sites...... 74

Figure 3.2: Summary of the percent logging in the landscape (1.77 km2, circular, per site) over all 16 sites sampled. The pair of dashed lines indicate where sites were split for pooling into three states of logging: low, medium, and high. The single dotted line indicated where the split was made for two-stated logging: low, and high...... 75

Figure 3.3: Summary of how percent landscape logging was split into 2- or 3-states for each species analyzed. Dashed lines separate the three states of logging (low, medium, high) and dotted lines separate two states of logging (low, high). Red indicates when a particular landscape-scale interaction best explained seed size, number, or both in a particular species, when points are red, continuous logging was the selected interactor...... 76

Figure 3.4: A comparison of local and local-landscape interactive influences on seed size in M. paniculata. Right panel depicts strictly local influence of canopy cover on ln(seed size), from Fig 2.10. Left panel contour plot depicts effects of the interaction between landscape logging intensity (continuous) and canopy cover on ln(seed size). This surface estimate was obtained from the GAMM: ln(seed size) ~ Julian + SPC1 + spline(canopy cover, logging) + random(loc in tran in hab in site). Pink contours represent the highest values of ln(seed size), and blue contours indicate smaller seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps...... 82

Figure 3.5: A comparison of local and local-landscape interactive influences on seed number in M. paniculata. Top panels show effects of the local model (presented in Fig 2.8). Lower panel contour plots depict the interactive effect of landscape logging intensity (continuous). These lower surface estimates were obtained from the GAMM: sqrt(seed count) ~ Julian + SPC1 + spline(con, logging) + spline(het, logging) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps...... 83

Figure 3.6: A comparison of local and local-landscape interactive influences on seed number in L. borealis. Top panel shows the local model effect. Lower panel component smooth plots for conspecifics depict the differential effects of landscape logging intensity (low, medium, high). This surface estimate was obtained from the GAMM: ln(seed size) ~ Julian + SPC1 + spline(con, by: 3-state logging) + random(loc in tran in hab in site)...... 84

x Figure 3.7: A comparison of local and local-landscape interactive influences on seed number in A. cordifolia. Top panel shows effects of the local model. Lower panel contour plots depict the interactive effect of 3-state landscape logging intensity (low, medium, high). These lower surface estimates were obtained from the GAMM: sqrt(seed count) ~ Julian + SPC1 + spline(con, het, by: 3-state logging) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps...... 86

Figure 3.8: A comparison of local and local-landscape interactive influences on seed size in E. conspicua. Top panel shows effects of the local model. Lower panel contour plots depict the interactive effect of landscape logging intensity (2-state). These lower surface estimates were obtained from the GAMM: ln(seed size) ~ Julian + SPC1 + spline(con, het, by: 2-state logging) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps...... 87

Figure 3.9: A comparison of local and local-landscape interactive influences on seed number in E. conspicua. Top panel shows effects of the local model. Lower panel contour plots depict the interactive effect of landscape logging intensity (2-state). These lower surface estimates were obtained from the GAMM: sqrt(seed count) ~ Julian + SPC1 + spline(con, het, by: 2-state logging) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps...... 88

Figure 3.10: A comparison of local and local-landscape interactive influences on seed number in V. americana. Top panels show effects of the local model. Lower panel contour plots depict the interactive effect of landscape logging intensity (2-state). These lower surface estimates were obtained from the GAMM: sqrt(seed count) ~ Julian + SPC1 + spline(con, bee, by: 2-state logging) + spline(het, bee, by: 2-state logging) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps...... 90

Figure 3.11: A comparison of local and local-landscape interactive influences on seed size in A. cordifolia. Top panel shows effects of the local model. Lower panel contour plots depict the interactive effect of habitat type (forest or clearcut). These lower surface estimates were obtained from the GAMM: ln(seed size) ~ Julian + SPC1 + spline(con, het, by: habitat) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps...... 92

xi LIST OF APPENDICES Appendix A: Plant species identified as bee-pollinated forbs and counted as heterospecifics for transects and local floral neighbourhoods at all sites where they were present. Flower size is reported, as all density counts were adjusted by maximum straight-line measurement (whether it be height, depth, width, length) per flower, taken as the median value of ranges reported in Moss (1983). The subset of the most common species sampled for seed size and seed number are listed first (“focal species”)...... 100

Appendix B: Species descriptions for all nine sampled focal species that were sufficiently replicated for analysis...... 102

Appendix C: General local effects of heterospecific floral morphology on all focal species that selected heterospecifics as having an additive or interactive effect on seed size or number. . 106

Appendix D: Specific model information and individualized figures for the best selected model by AIC for each species in Chapter 2. All component plots have an adjusted y-axis to represent the partial contribution of each variable, controlling for all other variables in that model. ... 113

xii CHAPTER 1: GENERAL INTRODUCTION

Theory predicts that resource-consuming individuals should distribute themselves in an environment according to the suitability of its habitat patches, a suitability that is determined by inherent characteristics (e.g., resources and predators) as well as density of competitors (Fretwell and Lucas 1969). For competing foragers, competitive ability can change with competitor density such that foraging rate is different when individuals are alone or at different local abundances (Creswell 1998). From a plant’s perspective, density-dependent interactions are constrained by immobility, and resource-consuming individuals (e.g., pollinators, herbivores) can impact plant fitness dependent upon both plant density and the density of the animals interacting with plants. Survival, growth and reproduction in plants are also influenced by density-related competitive interactions (Watkinson 1980, Freestone 2006, Gunton and Kunin 2009). Density dependence is not limited to single species, and may influence plant demography at the community level as well (Goldberg et al. 2001).

The majority of angiosperms rely on animal-mediated pollination (Ollerton et al. 2011), and most boreal forest forbs with showy flowers are dependent on bees in particular for effective sexual reproduction (Barrett and Helenurm 1987, Mitchell et al. 2009). A plant’s ability to attract pollinators has a significant impact on its fitness (Hegland et al. 2009, Sheehan et al. 2012). Pollinator attraction occurs through nutritional rewards for the visiting animal pollinator, thus forming a basis for a coevolved mutualism between pollinators and the plants they visit (Kevan and Baker 1983). Bumble bees as pollinators have formed this mutualistic relationship with understory plants of temperate forests and meadows, as bees depend on the forage that floral nectar and pollen provide (Heinrich 1979). The plant-pollinator mutualism is important for the success and stability of both plant and bee populations.

Pollination is the primary step in seed formation, and self-pollination, limited outcrossing, or outbreeding depression – affecting both the quality and quantity of pollen received – can result in fewer fertilized ovules, impaired seed development, and increased ovule abortions (Waser and Price 1994, Wilcock and Neiland 2002, Edmands 2007). Risk of pollination failure can increase for a variety of reasons: too little pollen is delivered, too much pollen is delivered, pollen is delivered too late, the wrong kind of pollen is delivered, or low- quality pollen is delivered. For plants, pollen limitation of seed production (as opposed to

1 limitation of reproduction by other resources) is common, and has been detected in the majority of natural plant populations (Burd 1994, Larson and Barrett 2000), particularly in fragmented habitats (Knight et al. 2005, Aguilar et al. 2006). However, the conclusion that pollen limitation is occurring when seed size or number increases in response to supplementary pollination is confounded by an alternative explanation: that the addition of pollen shifts the present-future life history tradeoff in reproductive effort to the present, in the case of polycarpic plants (Zimmerman and Pyke 1988, Ashman et al. 2004, Knight et al. 2005). Increased seed number or size in response to supplementary pollination is therefore consistent with more than one explanation: pollen limitation, or shifting present-future tradeoff. The level at which pollination failure occurs can vary between closely related plant species as well as within the same species across years, and it is not consistently predicted from species-level traits (Larson and Barrett 2000, Wilcock and Neiland 2002, Ashman et al. 2004, Knight et al. 2005).

Such variation in pollination failure could be due to the dependence of effective pollination on highly mutable characteristics – the number of effective visits (quantity) that successfully deposit compatible (quality) pollen strongly relies on both the spatial arrangement of individuals within plant populations, the density of their pool of pollinators, and the behaviour of their pollinators in response to such arrangements (Kunin 1993, Chittka et al. 1997, Ghazoul 2005, Mitchell et al. 2009, Ye et al. 2014). The movement of pollinators in response to variation in arrangement of plant patches will depend on the number of flowers in the area, as well as the distance between those flowers (Gunton and Kunin 2009). To add to the complexity of this pollination relationship, quality and quantity of pollen deposited can interact to influence the outcome of sexual reproduction in plants. For example, seed size or number may be impacted by the number of pollen grains deposited only at the lowest range of pollen deposition (when per- capita siring success is high due to reduced competition between individual grains). However, past a certain ‘limitation’ threshold of pollen deposition, it is likely that the quality of pollen deposited and competition between grains has a greater influence on seed set than pollen quantity (Aizen and Harder 2007).

In this thesis, I am interested in density-dependent pollination and its influence on reproduction in flowering plants. Initially proposed by Rathcke (1983) and examined with increasing frequency since, the density dependent pollination hypothesis involves foraging pollinators and how they interact with changes in patch-level floral density. Rathcke (1983)

2 suggested that at low plant densities, an increase in neighbouring flowers may lead to a positive, or facilitative, effect on visitation by pollinators for individual flowers located within such a patch. As the patch becomes more attractive due to increasing floral resources concentrated in a single area, Allee effects experienced by rare individuals are mediated (Asmussen 1979). However, positive plant-plant interactions for pollinators should transition into negative (competitive) interactions at some higher floral density, as the advantage of drawing more pollinators to a patch is overcome by a declining likelihood for pollinators to visit individual flowers. Multiple studies have examined density-dependent pollination and reproduction in plants, with dissimilar results (Levin and Anderson 1970, Silander 1978, Campbell 1985, Callaway 1995, Ghazoul 2006, Xi et al. 2015). Rathcke’s (1983) predicted relationship may also be further complicated by the addition of multiple species interactions, leading to different density-dependent relationships depending upon the proportion of a single important species with respect to others (Feinsinger et al. 1991), or the modification of density-dependence based upon distances between individuals, or differences between density dependent effects mediated by spatial scale (Fedriani et al. 2015).

Ecological interactions tend to be considered primarily at the local scale, as relationships that interest ecologists (e.g., competition, predation, mutualism) occur locally (Dayton 1971, Goh 1976), particularly for plants. However, the consideration of processes occurring over multiple scales and how interactions change depending upon the scale analyzed is highly important for the proper understanding of such interactions (Wiens 1989, Denny et al. 2004). Experiments are often designed with little regard for scale, and similar questions can result in different outcomes depending upon how they were spatially analyzed (Wiens 1989, Cadotte and Fukami 2005). What is considered a “landscape” for one species may be quite different for another species, depending upon their mobility, size, or other important aspects of their ecology (Addicott et al. 1987, Wiens 1989, Wiens and Milne 1989). There is no single best scale that all ecological phenomena should be studied at, and the resulting patterns that we observe may be driven by phenomena operating on alternate scales or across a collection of scales (Levin 1992). It is therefore highly important to consider multiple scales that could be relevant to focal individuals, as well as to be aware of hidden scaling phenomena that can lead to difficulty in the interpretation of observed dynamics (Sandel and Smith 2009).

3 A significant confounding issue for the examination of plant-pollinator relationships is that of spatial scale. When examining effects of habitat fragmentation or loss on populations, a small fragment in one study may be classified as a large fragment in another, and comparing these studies, we might see inappropriate conclusions arising from the differences in categorization (Hobbs and Yates 2003). Bumble bees often forage at great distances from their nests (Knight et al. 2005, Osbourne et al. 2008, Elliott 2009) and within a range of distances, individuals may select foraging habitat on a number of different scales. However, flowering herbaceous plants are limited by their small size and immobility, restricting direct effects to immediate local scales. Therefore, there are potentially contrasting implications of scale as it relates to plant reproduction, as influenced by pollinator mobility and habitat choice. Effects of disturbance can differ depending upon scale (Chaneton and Facelli 1991), and processes at different scales may interact or confound interpretation of results (Cash et al. 2006, Peters et al. 2007, Sandel and Smith 2009, Soranno et al. 2014).

Local effects, though important, are not constant within similar environmental conditions, and regional and historical processes also play a key role in community organization (Ricklefs 1987). Currently, human impacts on the environment are large in impact and broad in scale (McIntyre and Hobbs 1999, Chapin et al. 2000, Fontúrbel et al. 2015). These impacts include conversion of landscapes from their natural disturbance-based patterns to new, artificially altered structures with potentially degraded habitat quality (Fischer and Lindenmayer 2007, Fischer et al. 2008, Fontúrbel 2012). Forests in particular are increasingly fragmented and destroyed through human development and harvesting practices, requiring more intensive research and informed management strategies (Azevedo et al. 2010), as global monitoring of forest fragmentation currently lags (McDowell et al. 2015).

Spatial heterogeneity (i.e., differences among places) has always been an important component of ecological systems (Turner 2005). Due to large-scale extent being implicit for their definition, landscapes tend to be spatially heterogeneous. The ecological processes affecting community dynamics in heterogeneous landscapes remain poorly understood (McGarigal and Cushman 2002). Understanding interactions among species within disturbed landscapes is increasingly important as anthropogenic influences on ecosystems continue to accelerate.

4 A disturbance can be defined as either a single destructive event or an environmental fluctuation that somehow alters the structure of a community, ecosystem, or population (Pickett and White 1985). Habitat disturbances can result in comparatively open space within the original landscape type, and as their severity can be variable across a landscape, they can thus result in complex spatial variation at a variety of different scales (Turner 2005). Disturbance can have wide effects on physical environment as well as the biogeographical environment (Saunders et al. 1991). Fragmentation can lead to remnant areas of native vegetation, surrounded by a matrix of agricultural land, developed land, or a landscape that is in some other way altered from its pristine form (Saunders et al. 1991).

Fragmentation and habitat loss or conversion can influence plant-pollinator dynamics in a variety of both positive and negative ways (Rathcke and Jules 1993, Goverde et al. 2002, Tscharntke and Brandl 2004, Aguilar et al. 2006, González-Varo et al. 2009, Wagenius and Lyon 2010, Breed et al. 2013, Diaz-Forero et al. 2013, Ferreira et al. 2013). Given that the outcome of multi-scaled interactions within disturbed landscapes is so variable, the relative strengths of habitat fragmentation, conversion, or loss on plant-pollinator communities remain controversial, perhaps because of difficulties in their measurement (Ewers and Didham 2006, Hadley and Betts 2011).

Plant response to fragmentation can be trait-specific (Dauber et al. 2010), and species- specific traits such as self-incompatibility and specialization for particular pollinators can affect reproductive response (Rathcke and Jules 1993, Kearns and Inouye 1997, Kearns et al. 1998, Darvill et al. 2006, Aizen et al. 2002, Farwig et al. 2009). It has been suggested that pollen limitation (either in quantity or quality) may be the main or most proximate cause of reduced reproductive success in plant populations in fragmented habitats where reproductive success is determined by production or seed production (Aguilar et al. 2006). When testing multiple mechanisms behind demographic shifts towards extinction for a plant residing in a highly fragmented landscape, biotic interactions (i.e., pollination and seed predation) strongly influenced recruitment, and resource limitation, germination, and survivorship were not related to population collapse along edges (Jules and Rathcke 1999). However, the selection of plants examined for consequences of fragmentation has been highly skewed towards certain traits (Heinken and Weber 2013). In particular, most plant species analyzed thus far for effects of disturbance were from a few specific habitat types (such as grasslands) and had traits that may

5 have made them more vulnerable to habitat alteration (e.g., self-compatible species were underrepresented). Therefore, studies that concurrently examine a wide range of species with variation in traits that might influence response to habitat disturbance or alteration are lacking within the present literature.

Currently within the boreal forest, logging is the predominant force of change in habitat structure (Niemalä 1999, Cyr et al. 2009), and can have significant impacts on the species residing within these habitats (Drapeau et al. 2000, Schmiedinger et al. 2012). Clearcut logging results in a forest structure (i.e., species, age structure, and spatial pattern) that deviates from that expected through natural disturbances such as fire, potentially leading to an increase in the isolation and amount of old growth forest remnants (McRae et al. 2001, Rees and Juday 2002, Didion et al. 2007, Bouchard and Pothier 2011, McKechnie and Sargent 2013). Twentieth century logging practices have led to a decline in coniferous-dominated stands, a shift towards younger forests, and a destruction of the altitudinal forest type gradient (Boucher et al. 2009, Cyr et al. 2009). Contemporary disturbance has also lead to different species of trees dominating the differentially disturbed habitats (Bouchard and Pothier 2011).

Logging can have a diversity of impacts on ecological interactions, particularly plants and pollinators. Extensive forest management disturbance can lead to a decline in understory plant diversity (Schmiedinger et al. 2012), and succession back towards mature forest understory communities appears to be slow (Kreyling et al. 2008). However, due to the fact that logging disturbances tend to lead to openings in forest canopies, decreasing light limitation with positive impacts on early successional species, an increase in plant species abundance and diversity and a resultant positive effect on pollinator communities can also be typical (Romey et al. 2007, Pengelly and Cartar 2010). Low retention logging was best in comparison to clearcutting, as bees were still able to forage optimally while floral densities increased, but bees in unlogged forests were negatively affected by the presence of adjacent logging (Pengelly and Cartar 2010). Romey et al. (2007) also found a plateau in positive effects of logging, as 60% removal led to higher abundances of several pollinator species when compared to 100% tree removal.

Though logging may have a positive effect on the abundance and diversity of plants and pollinators, it may also alter their mutualistic relationship with one another. Given that habitat disturbance alters habitat structure, this can influence the way that pollinators track resources.

6 The most common observed deviation from ideal free pollinator service within fragmented landscapes (particularly in forest fragments) has been undermatching, where higher density resource patches have proportionally too few pollinators and lower density patches have proportionally too many pollinators (Cartar 2005, Pengelly and Cartar 2010, Farmer 2014). Any alteration in pollinator-resource match could potentially affect plant reproductive fitness, and interactions with other spatial or non-spatial variables could modify this result.

In a precursor study to mine, sharing my study sites, clearcuts produced higher floral abundances, leading them to be a “magnet” habitat for bees due to highly abundant forage (Farmer 2014). However, colonies within clearcuts performed poorly, suggesting the potential presence of an “ecological trap” in clearcuts with respect to bumble bees in the boreal forest (Roberston et al. 2006). An ecological trap is a habitat that is attractive for animals to settle in, but quality of the habitat is not optimal and leads to depressed success of animals in those habitats – for example, birds preferentially nested at higher densities in clearcuts, but nest success was half that of adjacent naturally disturbed habitat, due to a higher abundance of nest predators in artificially disturbed habitat (Robertson and Hutto 2007). However, whether clearcutting in the landscape leads to a positive outcome (e.g., increased diversity and abundance), a negative outcome (e.g., decreased diversity and a highly altered community structure) or a combination (e.g., an “ecological trap”) for understory plants is yet unclear.

This thesis aims to investigate a process-based question – what are the combined effects of floral density and pollinator abundance at varying spatial scales, within a landscape matrix of different logging patterns, on understory plant reproduction? This question has practical applications to forest management practices. By examining multiple common understory forest forbs within logged landscapes throughout the southern Alberta foothills forests, I was able to determine the importance of floral neighbourhood and pollinator abundance for seed production and the scale of that importance (Chapter 2), and subsequently examine how local interactions vary across a landscape-level gradient in anthropogenic disturbance (Chapter 3). Given the current rate of conversion of old growth forest to early-succession clearcuts occurring in the boreal forest, the significance of this research is in determining the impacts of this dominant form of habitat conversion and fragmentation on natural communities, and how species interactions within them are modified.

7 CHAPTER 2: REPRODUCTION OF UNDERSTORY FOREST FORBS: THE RELATIVE EFFECTS OF FLORAL NEIGHBOURHOOD, POLLINATORS, AND HABITAT ACROSS SPATIAL SCALES

2.1 INTRODUCTION

2.1.1 BACKGROUND

Floral density effects: facilitation of and competition for pollination

Individuals generally do not exist in isolation, and many co-flowering individuals (of the same or different species) may attract and share pollinators (Rathcke 1983, Feldman 2008). Pollinators can assess the foraging costs and rewards of these patch-level floral arrangements, and visitation can be sensitive to the density and dispersion of such resources (Rathcke 1983). Hence, a flowering individual’s floral neighbourhood affects both the quantity and the quality of pollinator service for that individual, resulting in positive, negative or neutral effects on reproductive success (Rathcke 1983, Kunin 1993, Dauber et al. 2010). Positive effects of neighbours on a focal plant species are referred to as facilitative effects, whereas negative effects of neighbours are generally understood to be competitive effects. Both types of interactions have been observed within plant communities, sometimes simultaneously (Holmgren et al. 1997, Choler et al. 2001). Though competitive interactions between individuals have historically been recognized as the dominating force structuring plant communities, facilitation may play an equally important role in individual ecology and for shaping community structure (Freestone 2006, Sargent and Ackerly 2008, Soliveres et al. 2015).

Pollination service should change with increasing floral density (Rathke 1983): there are generally more positive effects of neighbours (facilitation) at lower local floral densities, i.e. more visits per flower (Dauber et al. 2010), and weaker, neutral, or negative (competitive) effects at higher densities, i.e. fewer visits per flower (Essenberg 2012). At low floral densities, both low pollinator visitation and isolation from compatible mates can be problematic for effective reproduction. Smaller, more isolated flower patches may not be as attractive to energy- optimizing foraging pollinators that are sensitive to flight costs incurred when moving between plants while foraging; isolated patches tend to receive fewer, lower quality visits (Feldman 2008), and this decline in pollination at very low densities can lead to an Allee effect (Asmussen 1979), with subsequent reproductive declines (Kunin 1992, Groom 1998). In cases where small,

8 isolated flower patches receive pollen carried over longer distances, highly unrelated mates can lead to outbreeding depression (Waser and Price 1994, Edmands 2007). At very high floral densities, competition between individuals is stronger as the number of flowers increases past a point where pollinators are less likely to visit each individual flower within a dense patch, and pollinators have a tendency to inspect and reject more flowers when floral densities are high (Heinrich 1979).

Pollination in low density patches is impacted by a variety of processes. When small, isolated patches do receive visits, proportionally more flowers are pollinated in comparison to high density patches (Cibula and Zimmerman 1984, Klinkhamer and de Jong 1990, Van Treuren et al. 1994, Goverde et al. 2002), potentially because searching for unvisited flowers in small patches may be easier for a pollinator (Cibula and Zimmerman 1984, Goulson 2000), or because foragers are predicted to stay longer in patches (i.e., visit more flowers) when the environment is of poor quality (Charnov 1976). Thus, at low floral densities with associated ineffective pollination, any small increase in plant patch size that entices additional pollinators can lead to a strong positive effect on reproductive success (Silander 1978, Moeller 2004, Ye et al. 2014), as pollinators can more effectively transfer pollen between flowers. However, pollinators that visit proportionally more flowers in small patches can also increase geitonogamy, or inbreeding if all individuals within a patch are highly related (Bosch and Waser 1999). Inbreeding depression can result in increased ovule abortion and decreased seed quality (Wilcock and Neiland 2002). Thus, seed production in plants growing within low density floral patches reflects a tension between visitation (increasing fitness through the reduction of pollen limitation) and inbreeding depression.

Through facilitation in groups, individual plants recruit more pollinators to a patch than each could in isolation, but each individual then must compete for pollination within this mutually gained pollinator pool (Rathcke 1983). Foragers spend more time and visit more flowers in larger patches (Sih and Baltus 1987, Grindeland et al. 2005, Bennett et al. 2014, Hegland 2014) but they often visit proportionally fewer of the available flowers in these high- density areas (Sowig 1989, Klinkhamer and de Jong 1990, Robertson and Macnair 1995, Goulson 2000, Mustajarvi et al. 2001), as expected from the marginal value theorem (Charnov 1976). At some density threshold, a patch may become saturated with pollinators, leading to a dilution effect for each competing flower with further increases in floral density (Rathcke 1983).

9 Competition for pollination, in this case through a dilution mechanism, would increase with an increase in floral density beyond that threshold, given that pollinators tend to visit proportionally fewer flowers in high-density patches. There are additional proposed mechanisms behind competition for pollination: varying pollinator preference (Rathcke 1988, Chittka and Schürkens 2001) and unsuitable pollen transfer (McLernon et al. 1996, Neiland and Wilcock 1999), both of which require mixed species assemblages. These highlight the importance of neighbourhood context when examining individual reproduction in plant communities.

Facilitation and competition: community-level mechanisms

Any change in the composition of a community – reflected in altered abundances and/or distributions of different plant species – can affect pollination services, and the abundance of either conspecifics or heterospecifics should have differential effects on visitation and reproductive success of an individual. For example, at small scales, fruit set of self-incompatible plants was not affected by the density of heterospecifics, but was positively impacted by conspecific density (Feldman 2008). Individual reproduction should depend on how flowers interact for pollination within a single species, as well as how heterospecific individuals interact for pollinators, and how multiple pairwise pollination interactions play out across whole communities (Sargent and Ackerly 2008).

If pollinators are generalists, as most bumble bee species are, quality of pollination can decline in cases where interspecific pollen transfer is prevalent (Campbell 1985). In populations of rare species in particular, interspecific pollen transfer can potentially lead to increased pollen limitation and reproductive isolation (Colas et al. 2001). Heterospecific pollen mixed with conspecific pollen can lead to reproductive inhibition and failure through physical (clogging) and chemical (allelopathy) inhibition, triggering premature stigmatic closure, or altering the stigmatic environment in some way (McLernon et al. 1996, Wilcock and Neiland 2002). Impacts of heterospecific pollen receipt depend on traits of both the donor and recipient, as well as the phylogenetic relatedness of the interacting species (Schemske 1981, Ashman and Arceo-Gómez 2013).

Competitive interactions have been the historical focus of study of interactions for pollinators among heterospecific plants, and competition for pollination can dominate in mixed- species neighbourhoods (Bell et al. 2005). More recently, research has uncovered that what were

10 previously considered anomalous positive (facilitative) interactions between individuals of different species might in fact play an important role in structuring community dynamics (Soliveres et al. 2015). Such multispecies facilitative effects may arise when indirect interactions between plants of different species, through shared pollinators, lead to the joint attraction or maintenance of a larger or more diverse pollinator pool (Ghazoul 2006).

Multi-species facilitiation of pollinator attraction within a patch can occur through a diverse suite of mechanisms. For example, facilitation may arise from similar floral syndromes attracting generalist pollinators to a collective floral display that appears larger (collective attraction; Schemske 1981, Ghazoul 2006), or from divergent floral syndromes attracting an enhanced, more diverse community of pollinators to the collective display (Ghazoul 2006). Each species within a patch may produce complementary resources for pollinators, for example if one species produces nectar and another species produces pollen, and if these separate required resources are present in a single patch, it may be more attractive to pollinators than if each species were in a monospecific patch. Another postulated mechanism is that a single species in the mixture may simply have a higher level of attractiveness to pollinators compared to the others, leading to a “magnet species effect” benefit for those less attractive species (Laverty 1992, Johnson et al. 2003). In high density patches that attract large abundances of diverse pollinators, competition between pollinators for individual flowers may lead to a displacement of bees into the “competitor-free space”. This refuge provided by less attractive flowers may induce more pollination of the less attractive species, as well as maintaining a “pollinator reserve” for the most attractive species, once the dominant pollinators have moved on (Ghazoul 2006).

Indirect plant-plant interactions mediated through pollinators are potentially complex. Both competition and facilitation can occur dynamically in a system, such that a particular individual or species of plant may experience both competitive and facilitative interactions under different external conditions. Facilitation and competition may even occur simultaneously, depending upon community context and pollinator foraging responses. However, plant community interactions with the pollinator pool are not necessarily only occurring at very small scales, as pollinator foraging decisions can occur over broader spatial scales, whose extent depends on a pollinators’ mobility, a correlate of body size.

11 Scale of pollinator foraging decisions

To interpret density-dependent choice of patches by pollinators (as suggested by Ghazoul 2006), theory predicts that resource-consuming individuals should distribute themselves among patches in an environment according to the suitability of those habitat patches, a suitability that is determined by inherent characteristics (e.g., resources and predators) as well as density of competitors (Fretwell and Lucas 1969). Bumble bees often forage at great distances from their nests, typically between 0.1 to 2.7 km (Osborne et al. 1999, Knight et al. 2005, Osborne et al. 2008, Elliott 2009), with more recent estimates reaching up to 11.6 km (Rao and Strange 2012) or down to 110 m (Geib et al. 2015). Within this range of distances, individuals may select foraging habitat at a number of different scales, as recognized by hierarchy theory (O’Neill et al. 1989). Multiple scales of plant density, such as: the density of flowers within individual plants, the density of individuals within patches, the larger-scale density of those patches of plants, and the size of the overall plant population, can all impact pollinator behaviour and demography (Hadley and Betts 2011). Though the local decision-making behaviour of pollinators that leads to selection of individual flowers has been examined thoroughly (Kunin 1997, Goulson 2000, Wilcock and Neiland 2002, Hadley and Betts 2011), and small-scale impacts on plants and pollinators are most commonly found within the literature (Janovský et al. 2013, Williams and Winfree 2013), how large-bodied bees select which patches to visit and at which scale these decisions are most important is an empirical question which remains to be clarified in the literature. Decisions to visit patches may be made by bees at larger scales than have been studied to date.

The impact of community composition on foraging decisions of highly mobile pollinators may also change as the spatial scale of interest changes. For instance, higher plant densities at larger scales (i.e., more patches of individual plants in a larger area), may have stronger facilitative effects on flower visitation than smaller-scale interactions, should large patches act as a “magnet habitat” to recruit more pollinators to the general area. This effect should apply to increases of either conspecific or heterospecific density at a large scale, as the facilitative mechanisms of recruiting more pollinators to a patch through a larger collective floral display or more diverse floral display act within both monospecific and multi-species communities. Conversely, a collective large-scale facilitative effect may be comprised of a diversity of local effects.

12 Floral complexity and pollinator specialization

Display size and floral density strongly influence pollinator visitation (Hegland and Totland 2005), but other species-level traits can also mediate direct interactions between individual plants and pollinators, or indirect interactions between individuals of different species in competition for or in facilitation of pollination (Sargent and Otto 2006). In particular, traits that affect how dependent a species is on pollination for reproduction (i.e., self-compatibility), and how specialized its pollinators are, can have an impact on plant reproductive responses to external influences such as habitat fragmentation (Rathcke and Jules 1993, Kearns and Inouye 1997, Kearns et al. 1998, Darvill et al. 2006, Farwig et al. 2009). Though the majority of flowering plants are visited by more than one species of animal pollinator, pollination syndromes in plants can impact the frequency of visitors from different functional groups (Fenster et al. 2004). Floral symmetry, or level of complexity, can affect specialization for particular pollinators, which in turn can impact the reproductive response of individuals of different species competing for pollination within a community.

Flowers tend to fall into two broad categories of symmetrical complexity: bilaterally symmetrical (zygomorphic) and radially symmetrical (actinomorphic). Zygomorphic species tend to be more restrictive in how pollinators can approach and move between flowers, which may promote greater specialization or a “fixed preference” from pollinators, leading to higher visitation constancy and thus greater conspecific pollen transfer (Neal et al. 1998, Rodríguez et al. 2004, Sargent 2004). Greater floral complexity may also limit which pollinator groups are able to forage on a particular plant, as different pollinators have different learning, floral handling, and pollen deposition “quality” capabilities (Laverty 1980, Herrera 1987). Traits such as bilateral symmetry, inverted orientation or “pendular” flowers (Ushimaru and Hyodo 2005), tubular or semi-tubular morphologies (Peng et al. 2012), or any other form of complexity can preclude pollinators lacking specialized morphology or specific learned behaviours from access. Results of this increased complexity may be positive, due to increased floral constancy arising from a less diverse (and potentially less generalized) pollinator pool. However, a decrease in pollinator species breadth may not always be positive, as an increase in the diversity of pollinators or pollinator functional groups can increase seed set in many plant species (Klein et al. 2003, Hoehn et al. 2008). The impact of floral symmetry on pollinator response to different scales of groupings of flowers is not clear – when floral symmetry becomes important for a

13 discriminating pollinator may vary dependent upon the grouping of individual flowers, inflorescences, and arrangement of patches of conspecifics (Neal et al. 1998).

This chapter examines how floral density (conspecific and heterospecific flowers), pollinator density (bee abundance), and habitat quality (canopy cover) influence reproduction (as estimated by seed size and seed number) at multiple spatial scales in several common species of understory forbs flowering throughout the foothills forests of southern Alberta, . I examine these effects to detect which variables or combination of variables are most important for reproductive potential, and at which scale they have the strongest influence, across a diverse suite of plant species that vary significantly in their floral morphology, and potentially pollinator specialization. I ask: what effect or set of effects is most important in accounting for seed size and number, at what spatial scale is reproduction most strongly associated with the selected influences, and do these relationships change when looking at individual species vs. the entire co-flowering community?

2.1.2 HYPOTHESES AND PREDICTIONS

Spatial scale

I expect that characteristics of the local neighbourhood will have the biggest effect on seed size and number, because physical proximity is likely to increase opportunities for ecological interactions, and previous studies have generally found local impacts to be more important than landscape-level effects (Wright et al. 2003, Marini et al. 2007, Marini et al. 2008, Williams and Winfree 2013). However, these findings may be artifacts of data collected on a fine scale, with more variation with which to explain plant reproduction than the pooled samples that tend to arise from landscape-level metrics. Given that all interactions occur locally (particularly competition with immediate neighbours for nutrients, water, and light), but that the participants in these interactions are often the consequence of broader-scale processes, there appears to be a logical hierarchy in expectation of outcome. Ecological interactions at local scales should always be detectable, and, depending on circumstance, they might also reflect processes at broader spatial scales, with the expectation of impact decreasing as scale increases. Perhaps, due to their immobility and small size, herbaceous plants show a much stronger reliance upon their immediate surroundings. I therefore hypothesize that plant-pollinator interactions (e.g.,

14 frequency of visitation), and their outcomes (e.g., seeds set and their size), may be more determined by processes at local scales. However, since bumble bees are large-bodied pollinators that can travel long distances to forage, and the literature lacks clarity in explaining at what spatial scale they primarily make their foraging decisions, I test larger-scale patch-level effects on seed size and number as well.

Floral density

If the density of conspecifics in the vicinity of a focal individual impacts its reproductive potential, I expect the relationship to reflect one of three mechanisms of interaction between individuals: facilitation, competition, or both. For facilitation, more neighbours of the same species would attract more species-specific pollinators, allowing for more outcrossing and improved individual reproduction (Fig 2.1, region to the left of the peak). This relationship would take the shape of either a continuous increase in seed size or seed number with an increase in same-species neighbours, or a diminishing increase to a saturation point. For competition, as conspecific density increases, having more same-species neighbours may lead to dilution or intraspecific competition due to a declining likelihood for an individual pollinator to visit all conspecific flowers or individuals in a patch. This would take the shape of a monotonic decrease in seed size or number with increasing conspecific density (Fig 2.1, region to the right of the peak). However, given a sufficient range of plant densities, it is more likely that both facilitation and competition act together (Fig 2.1; Rathcke 1983). At very low densities, Allee effects should lead to smaller or fewer seeds produced, but as neighbouring density increases, facilitation should draw pollinators to a patch, up to a certain point. This transition point would be at a threshold after which more neighbours do not draw in enough additional pollinators to effectively visit all flowers in a patch, leading to dilution or competition, above which seed size or number should decrease as neighbouring individuals compete with each other for visits.

If the density of heterospecifics has impacts on reproductive potential, I expect the resultant relationship between density and seed size or number to be similar to that of conspecifics (Fig 2.1), but with slightly different outcomes. As with conspecific effects, more flowers in the neighbourhood might attract more pollinators or a higher diversity of pollinators (“collective attraction”) but in the case of heterospecifics, the higher probability of improper pollen transfer will likely limit benefits to a smaller range of densities (leading to a left-

15

Figure 2.1: Density-visitation curve modified from Rathcke (1983) showing how pollinator visitation and subsequent plant reproduction are expected to respond to changes in floral density. At very low floral densities, isolated plants may suffer from Allee effects, at low floral densities, facilitation should dominate, and densities higher than maximum per-flower visitation should result in competitive interactions among flowers. compressed Rathcke (1983) curve; Fig 2.2). However, certain plant species may react more positively (i.e., steeper slope to the left of the peak) to low densities of heterospecific individuals if they are of greater interest to pollinators (e.g., higher reward, reduced handling times or travel costs). Alternatively, more neighbours of different species may result in strictly competitive interactions for visitation, leading to more interspecific pollen transfer, a decline in seed size or number, and therefore a negative relationship between increasing numbers of neighbouring heterospecifics and seed production.

Conspecific and heterospecific densities may also interact with one another to affect seed size or seed number in plants in ways that are difficult to predict. Variation in either heterospecific or conspecific density with respect to the other might influence how either impacts seed size, where, for example, increasing conspecific density could have a different impact on pollinator behaviour when there are lots of heterospecifics in the vicinity versus when there are few, or when the community is monospecific. Thus, interactions between these two traits were tested for in all species examined.

16

Figure 2.2: A modification of Rathcke’s (1983) density-visitation curve, indicating how effects of neighbouring conspecific and heterospecific densities may differ from one another. Heterospecific facilitation (positive slope) is limited to lower floral densities, and competition (negative slope) also begins at lower floral densities for heterospecifics than conspecifics.

Floral specialization: modifying conspecific and heterospecific effects

Based on Sargent and Otto’s (2006) model of evolution of floral morphology as impacted by local species abundances, I hypothesize that if a plant is locally common (higher conspecific density), it is expected that specialization should have less of an impact on its reproductive success, as there is an increased chance that transferred pollen will be genetically compatible. If a plant is locally rare (lower conspecific density), plants that are more specialized should benefit from increasing local densities, as attracting pollinators that discriminate more effectively should reduce the rate of heterospecific pollen transfer experienced by those rare, specialized species. Because of lower rates of heterospecific pollen transfer, specialists should therefore be more resilient to negative effects of pollen competition at low conspecific densities or high heterospecific densities, while less specialized species should experience greater competition from heterospecifics.

17 Pollinator abundance

Abundance of pollinators may also influence plant reproduction. I expect that increased bee abundance should be associated with increases in seed size and number, due to the greater likelihood of pollination. However, this effect can saturate due to stigma loading, or negative interactions between pollinators (e.g., bumble bees often reject flowers they have detected as having previously visited by another bumble bee; Goulson 1999). Pollinator abundance may also interact with conspecific and/or heterospecific density – in an area saturated with bees, dilution by conspecific flowers may be weaker if there are enough pollinators to effectively visit most flowers, relative to stronger dilution effects expected for flowers in response to increases in floral densities under low pollinator densities. Conversely, with few pollinators, there may be a stronger facilitative effect of increasing neighbours at lower conspecific flower densities, as low density patches may be less likely to attract any pollinators at all. However, if total floral densities are low (at broader spatial scales), individual pollinators that tend to specialize on fewer floral species may expand their diet in face of declining resource availability (as predicted by the diet model; MacArthur and Pianka 1966), which may lead to increased interspecific pollen transfer at low conspecific densities, and thus a stronger heterospecific competitive effect (Smithson and MacNair 1997, Chittka and Schurkens 2001, Ashworth et al. 2004, Fontaine et al. 2008).

Habitat

The physical environment surrounding individual plants can impact their ability to acquire resources and might alter the way they partition those resources, which can affect pollination and its downstream effects (e.g., seed development). For this chapter, the inclusion of the test of whether habitat at multiple scales influences seed size and seed number was to control for abiotic effects; resource limitation (vs. pollen limitation) of reproduction can occur over different time scales (e.g., within a season or over the course of a lifetime) and have different strengths depending on the biotic and abiotic environment (Zimmerman and Aide 1989, Campbell and Halama 1993). Canopy cover as a measure of local light conditions, and habitat type (forest versus clearcut), can both impact the availability and flux of nutrients within a system (Saunders et al. 1991), and variation in microclimate (light, temperature and water in particular) is predictably different between harvested and old growth forests (Aussenac 2000). I

18 expect that if important at all, habitat impacts would depend upon the habitat preferences of each species.

The above hypotheses have been distilled into a table for quick reference below (Table 2.1).

Table 2.1: Simplified hypothesis table outlining expected relationships between variables and seed size and number, at both the local (=smaller) and patch (=larger) scales.

Scale Local Patch Conspecifics  Monotonic increase (facilitation, quality  Monotonic increase (at a and quantity of pollen deposition larger scale, should expect increases, more bees are attracted to higher densities to be a bigger patches so plants get more visits magnet for bees, being and potentially more genetically diverse within a higher density pollen, may plateau at high densities) “patch” is good for long-  Monotonic decrease (inbreeding distance foragers) depression at low densities, competition  Increase to a maximum, then due to dilution at high densities, more decrease (pollen limitation, likely in generalists) outbreeding depression,  Increase to a maximum then decrease dilution) (facilitation and competition, Rathcke (1983), through mechanisms listed above) Heterospecifics  Monotonic decrease (interspecific  Facilitation and competition pollen transfer, competition for with a higher threshold than pollinators) smaller scale, variety of  Facilitation restricted to lower densities resources could attract more or in species that have little reward, but pollinators to patch, support an increase in density leads to pollen more colonies competition (Rathcke (1983), peak shifted left) Bees  Monotonic increase (could plateau)  Monotonic increase (could plateau) Habitat  Shape of relationship depending on  Habitat type and canopy habitat preferences of each species cover at the patch level may (affects physical environment of plants, impact bees more strongly and therefore their ability to sequester than plants, given their resources (nutrients, light, water) or mobility across larger-scale pollinators (changes their energy habitats budgets))

19 2.2 METHODS

2.2.1 FIELD SITES

Sixteen circular 1.77 km2 sites in Kananaskis Country, Alberta were sampled over the course of July and August 2012 (Fig 2.3). Sites were a subset of those surveyed in a precursor study (Farmer 2014). Each location was separated from all others by at least 2 km to reduce spatial non-independence, and each took an average of 2 consecutive days to sample. Dominant tree species varied within and between sites, but most were coniferous: species such as lodgepole pine (Pinus contorta) and white spruce (Picea glauca) were most abundant. There were also sites containing mixtures of coniferous and the deciduous trees trembling aspen (Populus tremuloides) and balsam poplar (Populus balsamifera).

The highly mobile bumble bee (Bombus spp.) was the focal pollinator sampled in this study. Due to their large body size, bumble bees can forage widely from their colony (0.1 to 11.6 km) (Knight et al. 2005, Osbourne et al. 2008, Elliott 2009, Rao and Strange 2012). Thus, I assume that all flowers within a given site qualify as prospective forage for an individual bee.

2.2.2 STUDY SPECIES

All flowers of species that were known to be pollinated by bumble bees were counted in census surveys (see Appendix A), for a total of 43 different species from 14 different families (as classified by Desmet and Brouillet 2013). A subset of the most common species found in the study area (decided in situ) were sampled for seed size and number, and only those focal species that were present at ≥ 7 sites (n=9 species) were analyzed further (Table 2.2). All species analyzed were commonly present in both forest and clearcut locations across all sites (Table 2.2) to enable analysis of the effects of habitat type on plant reproduction, though in some cases a species may have been present in a single habitat type (i.e., forest or clearcut) within a given site. See Appendix B for detailed species descriptions for all focal species examined in this study.

For each bee pollinated species included in the survey, I categorized floral morphology (i.e., shape) into three broad categories, increasing in handling time for bumble bee visitors, as a proxy for “specialization” (Larson and Barrett 2000); categories were disc shaped and pendular

20

Figure 2.3: Locations of sixteen 176 hectare field sites along the eastern foothills slopes of Kananaskis Country, AB. Sites are categorized into 3 levels of logging intensity: low (≤30% logging; blue, n=5), medium (31%-47%; orange, n=6) and high (>47%; red, n=5). See Chapter 3 for further explanation of logging impacts. Image source: Google Earth. Image date: 2013.

Table 2.2: Species of the most common bee-pollinated wildflowers sampled and analyzed for impacts on reproductive potential (seed size and number). Samples were collected in Kananaskis Country, AB from June-August 2012. The habitat counts indicate number of sites in which the species was sampled. Species are binned into three floral morphology categories, increasing in handling time for bumble bee visitors. Species Family N Sites N Forests N Clearcuts Disc Arnica cordifolia Asteraceae 13 13 6 Eurybia conspicua Asteraceae 7 3 6 Geranium richardsonii Geraniaceae 12 10 7 Pendular Campanula rotundifolia Campanulacae 11 5 8 Linnaea borealis 12 12 6 Mertensia paniculata Boraginaceae 12 12 7 Zygomorphic Castilleja miniata Orobanchaceae 13 6 10 Lathyrus ochroleucus Fabaceae 9 7 7 Vicia americana Fabaceae 11 4 8

21 (both actinomorphic), and zygomorphic, modeled after terminology used by Lázaro et al. (2013). Flowers with disc morphology included completely open flowers (e.g., Geranium richardsonii), as well as composite flowers, where inflorescences are composed of many small, clustered tubular flowers in a single open “flower head” (e.g., Asteraceae); both of which contain rewards that are easily accessible to a broad pollinator fauna. Pendular species included plants with actinomorphic tubular and semi-tubular flowers, usually bell-shaped with inverted orientation (e.g., Mertensia paniculata), requiring specific pollinator landing behaviours in order to access rewards (Ushimaru and Hyodo 2005). Zygomorphic was the most complex floral morphology category, and included species with bilaterally symmetrical flowers, often closed and/or requiring advanced learned behaviours in order for pollinators to appropriately handle them (e.g., Fabaceae). This categorization was used to examine whether there were clear patterns of effects on my focal species within categories of floral “accessibility”; accessibility is assumed to be related to specialization and floral constancy, or how restrictive flowers are to visitation from a diverse pollinator pool. A preliminary examination of the effects of floral shape of heterospecifics in the floral neighbourhood on focal species was tabulated, but interpretation of these relationships was beyond the scope of this study (see Appendix C for methods and summary results).

2.2.3 SAMPLING DESIGN

Transect censuses

Each site contained 14 transects, 7 within each habitat type (forest and clearcut) or 8 transects within control sites containing only forest (n=3). Each transect was 100 m long and 3 m wide, and each was at least 100 m from both habitat edges and all other transects, to reduce edge effects and non-independence (Fig 2.4).

I counted bees (bumble bees and solitary bees) and focal flowers (46 total species) along all transects: this census was taken at a slow walking speed, spending approximately 15-20 minutes per transect. Bees were counted if they were observed within the bounds of a transect, whether they were seen visiting a flower, nest searching, resting on foliage, or flying. Though solitary bees were counted in addition to bumble bees, solitary bees made up less than 2% of total transect bee abundance. Other species of pollinators were observed in the field (e.g., true

22

Figure 2.4: Schematic of sites representing a) low landscape logging (25% clearcuts in pink and blue; Powderface Trail) and b) high landscape logging (60% clearcut in yellow, Sibbald Flats) with black lines indicating sampled transects with 7 per habitat type (forest and clearcut) per site. Magnified lower-right diagram displays sampling design of an example transect (100 m by 3 m, not to scale). Small circles grouped at approximately 25 m and 75 m represent samples of local floral neighbourhoods. flies, syrphid flies), but their abundances were not recorded and they made up a comparably small proportion of the pollinator pool (pers. obs.). All open, visitable flowers (i.e., not unopened buds or withered corollas) were counted and their abundances recorded. However, when a particular flower species tended to have many flowers grouped together in a larger inflorescence, inflorescences were counted and later multiplied by an average flower count per inflorescence for that species (method from Toft 1983; mean floral counts per inflorescence for the region obtained from G. R. Earle, unpublished data). For species within the family Asteraceae, with many small flowers clustered in a large inflorescence, the entire inflorescence was counted as a single flower. All bees were pooled as numbers of individual pollinators per half and full transect (150 m2, 300 m2; Fig 2.4), while all flowers were identified to species and later pooled together per half and full transect into the categories of either conspecifics or heterospecifics, according to

23 focal plant species. Flower counts were later adjusted by median flower size (maximum straight- line measurement; see Appendix A) in order to quantify display-sized count and adjust for differential effects of differently sized heterospecifics (method from Toft 1983), as display-size is known to have an effect on pollinator resource tracking and is often proportional to forage value of a flower (Goulson 1999).

Site characteristics were measured and recorded along each transect either every 25 metres (temperature, elevation, UTM coordinates, time of day, Julian date) or at every local floral neighbourhood surveyed (canopy cover or “reflecting light availability”, measured with a densiometer). Site characteristics measured multiple times at the transect level were included as covariates to control for spatial and temporal heterogeneity within and among sites, and local site characteristics were used as an abiotic control when examining impacts of local floral density.

Local floral neighbourhoods

At fixed locations (25 and 75 m) along each 100 m transect, samples of bee-pollinated plants were obtained from within 1 m2 areas, selected for their comparatively high and low conspecific local density within a 1 m2 quadrat (when available; Fig 2.4). All individual point samples were located within 20 m of their respective transect locations. The local floral neighbourhoods were quantified, identifying the density and identity of all co-flowering species of interest (both hetero- and conspecific) in the inner 1 m2 and outer 10 m2 local areas, respectively. When they were present, flowers within the two local areas that had gone to seed (but were not directly sampled) were included within “floral density” counts, but detection of seeds was much more difficult than detection of showy floral displays, so seed densities were not quantified along transects. Both local sampling areas were circular, measured using a custom- made 1 m2 hoop and a 1.78 m radius string around a central pole. The outer local area did not include the counts recorded within the central 1 m2 to reduce non-independence, so though its realized area is actually 9 m2, it will be referred to as 10 m2 for all remaining discussions.

While quantifying floral neighbourhoods, haphazard samples of up to 6 different seed pods were collected from focal individuals within each 1 m2 local area, for which size (individual seed area or weight) and abundance (per-pod seed counts) were later quantified. Each pod was collected from a different individual when possible, but individual plant identity was not recorded. I chose to sample seeds in a cross-sectional manner; using this method of sampling to

24 examine reproductive potential at different stages of each plant is similar to a static life table, in that a cross-section of the age structure of a population is examined at one point in time (Harper 1977). That is, seeds that were set in a period ~2 weeks prior to my flower counts were related to current flower and bee densities, with the assumption that conditions during their pollination were adequately captured by my current measures of flower neighbourhoods, as many species produce flowers sequentially throughout the season (Ashman et al. 2004).

While collecting seed samples from focal individuals, any bees observed within the selected transect location were recorded as representative of local-level pollinator density. As I only took point counts of bees foraging in the area while sampling at local and transect-level spatial scales, I cannot directly relate visitation to pollination. However, I have included bee counts as a measure of general bee abundance, as presence of bees in these scaled areas may affect reproductive potential through provision of pollination service.

2.2.4 SAMPLE PROCESSING

When seeds were clearly grouped, either in pods (i.e. Vicia americana, Lathyrus ochroleucus), inflorescences (i.e. Arnica cordifolia, Eurybia conspicua), or other structures (i.e. Campanula rotundifolia, Castilleja miniata), seed counts were recorded per group for a random sample of up to 10 seed groups, per 1 m2. When seeds tended to be detached from any grouping structure present during collection (i.e. Mertensia paniculata, Geranium richardsonii), seed counts were recorded per total sample of focal species within the 1 m2, representing a “patch total seed production” metric.

Seed size was either measured as weight (mg) or surface area. When individual seeds were too small to weigh individually (i.e., < 0.01 mg), overhead pictures of seeds lying on their flattest surface (using Olympus E-420 digital camera) were used to digitally measure cross- sectional area with GIMP (GNU Image Manipulation Program) 2.8.10 (GIMP Development Team 2013) using the intelligent scissors tool to contrast seeds from background. Seed area was measured in pixels and converted to mm2 using a scale inserted in each photo. Seed area was measured for a random subsample of all seeds when counts per inflorescence were high, with the goal of obtaining at least 10 replicates of seed area per point location (1 m2). Summary statistics

25 for means, standard errors, and sample size for each seed character for all plant species analysed were calculated (Table 2.3).

Table 2.3: Summary statistics for seed size and number for each species analyzed. Number of samples, means, and standard errors are reported for each respective seed character. As seeds sampled for size were only a subset of all seeds counted, mean seed count for each species was calculated as a mean per pod, grouping structure, or mean “patch total seed production”. Seed size Seed count Species N(seed) Mean(mm2) SE(mm2) N(pod) Mean(#) SE(#) Arnica cordifolia 1854 3.30 0.02 165 54.99 2.30 Eurybia conspicua 1572 0.70 0.01 210 23.70 1.42 Geranium richardsonii 790 1.06 0.05 132 3.25 0.15 Campanula rotundifolia 1525 0.21 0.003 176 217.07 28.75 Linnaea borealis 374 1.72 0.04 NA NA NA Mertensia paniculata 683 1.44 0.04 81 25.28 2.20 Castilleja miniata 2948 0.67 0.01 107 68.35 4.38 Lathyrus ochroleucus 961 0.30 0.02 NA NA NA Vicia americana 1040 6.34 0.17 165 54.99 2.30

2.2.5 SPATIAL SCALE

In this chapter, I am interested in determining the most important spatial scale at which the suites of variables affecting seed size and number operate. These scales can be grouped into two “types” of scale, which will be referred to as “local” and “patch” (Fig 2.5). As outlined above, data on floral neighbourhood (categorized into conspecifics and heterospecifics) and bee abundance were collected at different scales. Sample metrics grouped by the two areas immediately surrounding the focal plant (1 m2 and 10 m2 local neighbourhoods) are considered local-scale. In contrast, sample metrics grouped by the half transect (150 m2) and full transect (300 m2) represent the broader floral neighbourhood within which an individual is embedded, and was considered the patch-scale.

Certain variables (bee abundance, canopy cover) were measured once for the two different local scales (1 m2 and 10 m2), and hence for these variables, results and figures contain identical values for the 1 m2 and 10 m2 categories.

26

Figure 2.5: An example schematic (not to scale) outlining the spatial scales focussed upon in this chapter – two levels of “local” indicating small-scale counts of neighbours (conspecific in blue and heterospecific in red) immediately around seed collection sites (focal individuals) located at points within a transect “patch”, and two levels of larger-scale counts of neighbours in the general area surrounding focal individuals.

2.2.6 DATA ANALYSIS

I analyzed variation in reproductive success (seed size and number) across 9 different plant species, each at 4 different spatial scales (1 m2 and 10 m2 “local scales”, 150 m2 and 300 m2 “patch scales”) with generalized additive mixed models (GAMMs; package “gamm4” version 0.2-3, procedure “gamm4” in R version 3.1.0; R Development Core Team 2014), using the information theoretic model selection framework. The GAMM method of analysis allows for the fitting of smooth spline curves to nonlinear relationships (Wood 2006). Given that the hypothesis I am testing is intrinsically non-linear (predictions being typically dome-shaped or asymptotic, Fig 2.1 and Table 2.1), I chose GAMMs to more easily evaluate the nonlinear effects of conspecific flower density, heterospecific flower density, and bee abundance on seed production. Splines are less restrictive than polynomial terms within linear models, and therefore allowed me

27 to estimate potentially more complex curves with multiple inflection points. The mixed model framework of GAMMs allowed me to account for the random effects of replication at each level of sampled spatial location: site, transect, and location within transect.

Two separate response variables were examined for plant reproduction: seed size and seed abundance (when available). Seed size was natural log transformed for all species, and seed number was square root transformed (except for Vicia americana) before models were fit with the Gaussian family distribution assuming normal residuals, using an identity link function. Use of a normal error structure combined with these transformations facilitated convergence by decreasing computational intensity. Candidate GAMMs (Table 2.4) were run individually for each plant species at each spatial scale. Assumptions of normality and homoscedasticity of residuals were visually assessed via residual plots and were adequately met for all selected models. This study provides a wealth of explanatory variables. These include: site, transect, and transect location (all random effects, of little general interest), habitat type, canopy cover, floral densities, pollinator abundance, nuisance covariates, and key interactions. All continuous main effects were included as spline smoothed terms in each model, and continuous interactive effects formed spline smoothed response surfaces. To parse this large number of variables of interest, for each species at each of 4 scale levels, 7 a priori models were assessed (Table 2.4) and compared to each other using Akaike’s Information criterion (AIC) and Akaike model weights

(wi) to determine which model best fit each species’ seed data across scales (Akaike 1973, Burnham and Anderson 2002). Candidate models can be broadly characterized into 3 groups: flower density effects (first 3 models in Table 2.4), habitat type effects (4th model) and effects including bee abundances, either alone (5th model), or in combination with flower densities (models 6-7). All models also contained the two nuisance covariates found to have the most influence on seed size and number during preliminary analyses: Julian day and the first principal component of elevation and UTM easting and northing (accounting for 63.78% of the variation in all 3 variables), which I used to control for spatial non-independence of sample locations. The spatial PC1 was always included as a linear term and not a smoothed term, and in special cases Julian day was also included as a linear term to facilitate convergence (but usually seasonal effects show a “humped” form, facilitating the use of a smoothed function for Julian day). My use of PC1 and Julian day was to remove any potential nuisance effects of spatial and temporal

28 Table 2.4: Candidate models to explain seed size or seed number, selected based upon ecological hypotheses. All models also contained nuisance covariates and random effects of location to control for their potential influence on the response variable. Model Variables Plant 1 Conspecific density Plant 2 Conspecific density + Heterospecific density Plant 3 Conspecific density * Heterospecific density Habitat Habitat type + Densiometer Bee Bee abundance Bee Plant 1 Bee abundance + Conspecific density + Heterospecific density Bee Plant 2 (Bee abundance * Conspecific density) + (Bee abundance * Heterospecific density)

heterogeneity on seed size and seed number. All candidate models also included the nested random effects of transect location within transect within habitat within site, to account for repeated measures within spatial scales.

For each response variable, all models were run once per scale and compared within a species; the aim being to identify the importance of each suite of variables for explaining reproductive potential (i.e., plant, habitat, bee, or some combination; Table 2.4), the changing importance of scale within each suite of variables, as well as the most important spatial scale across both measures of reproduction. These analyses allow general inferences about the importance of different spatial scales on plant reproduction, and specific inferences about how particular ecological variables affect seed size or number across scales.

After generating figures for the effects of each variable on seed size and/or number within each selected best model across species (i.e., lowest AIC), it was evident that in order to interpret general impacts of conspecifics, heterospecifics, and bees, species would have to be pooled as replicates and simplified effects summarized. Interactive effects (i.e., Plant 3 and Bee Plant 2 in Table 2.4) were exceptionally complex, so to streamline interpretation, “peaks” and “troughs” were identified from interaction surfaces for all species that selected an interaction as best explaining reproductive output. Each peak and trough was then classified as being located generally within “low”, “intermediate”, or “high” regions of conspecific density, heterospecific density, or bee abundance, depending on the two components of the selected interaction. Two figures were then generated to display the collective combinations (across species) of the interacting variables that led to maximal and minimal seed production.

29 2.3 RESULTS

2.3.1 SPATIAL SCALE

Across species, wildflower seed traits were most strongly influenced at the local scale. When comparing ΔAIC and Akaike weights for candidate models with empirical support (ΔAIC < 10) across spatial scales, for all species, seed size and seed number were best explained by variables at the 1 m2 or 10 m2 local scales. This was consistent across six of eight species sampled for seed size (Table 2.5, Fig 2.6) and all six species for seed number (Table 2.6, Fig 2.7). Only two species deviated from this pattern and showed a small potential influence of patch-scale dynamics (M. paniculata with 2 models wi < 0.05, and L. borealis with 1 model wi < 0.02), but this effect was only observed for seed size (Table 2.5).

For G. richardsonii, surprisingly, most of the models had similar nonexistent explanatory power: the majority of models across both model categories and spatial scales were within 10 ΔAIC units of the best model, and all models tested were within 30 ΔAIC units. Hence, variation in Geranium seed size and seed number remains unexplained in my study system (Fig 2.6e, Fig 2.7e). G. richardsonii was therefore excluded from all further analyses.

Reproductive potential in this system was most commonly best explained not by the most immediate floral neighbourhood (1 m2), but by the larger of the two most local spatial scales (10 m2). This trend held for both seed size and seed number, but with no strong patterns within floral morphology types. Seed size was best explained primarily by the 10 m2 scale (four of eight species) followed by the 1 m2 scale (three of eight species; Table 2.5, Fig 2.6). Seed number was best explained primarily by the 10 m2 scale (four of six species) followed by the 1 m2 scale (two of six species; Table 2.6, Fig 2.7). For a single species (M. paniculata), the best model (habitat + canopy) shared influence with both the 1 m2 and 10 m2 scales (Fig 2.6f), because canopy cover was only measured a single time for both small scales. In several species, when the top- performing model was for one of the local scales (1 m2 or 10 m2), there were instances of support for models of the alternative local scale (ΔAIC < 10; Fig 2.6, Fig 2.7). C. miniata was an especially strong case of this, where seed size was best explained at the 10 m2 scale, but the 2nd best model was at the 1 m2 scale (Table 2.5).

30 Table 2.5: Generalized additive mixed model selection from the set of a priori candidate models (see Table 2.4), examining effects on seed size for all plant species. Only models for which ΔAIC was ≤ 10 are reported. Akaike model weights (wi) show probability that the given model is the best among the entire set of candidate models for that particular species. All models include nuisance covariates (Julian day, Spatial PC1) and nested random effects (transect location in habitat in transect in site) controlling for repeated measures within scaling locations. Species are binned into three floral morphology categories, increasing in handling time for bumble bee visitors. Species Model df AIC ΔAIC wi Disc Arnica cordifolia Con * Het (10 m2) 11 1051.34 0 0.88 Con * Het (1 m2) 11 1055.61 4.27 0.10 Con + Het (10 m2) 12 1059.70 8.35 0.01 Con + Het (1 m2) 12 1061.78 10.44 0 Eurybia conspicua Con * Het (10 m2) 11 2296.99 0 1 Pendular Campanula rotundifolia Con (10 m2) 10 3391.00 0 0.50 Con * Het (1 m2) 11 3392.15 1.15 0.28 Con + Het (10 m2) 12 3393.75 2.75 0.13 Con * Het (10 m2) 11 3395.43 4.43 0.05 Bee * Con, Bee * Het (10 m2) 13 3395.98 4.98 0.04 Bee + Con + Het (10 m2) 14 3400.43 9.43 0.01 Linnaea borealis Con (1 m2) 10 377.16 0 0.77 Habitat + Canopy (10 m2) 11 380.96 3.80 0.11 Con * Het (1 m2) 11 382.70 5.54 0.05 Habitat + Canopy (150 m2) 11 384.64 7.49 0.02 Habitat + Canopy (300 m2) 11 385.01 7.85 0.02 Con (10 m2) 10 385.47 8.32 0.01 Con + Het (1 m2) 12 386.02 8.86 0.01 Mertensia paniculata Habitat + Canopy (10 m2) 11 1119.10 0 0.88 Habitat + Canopy (300 m2) 11 1124.99 5.88 0.05 Habitat + Canopy (150 m2) 11 1125.35 6.25 0.04 Bee * Con, Bee * Het (1 m2) 13 1126.74 7.64 0.02 Con * Het (10 m2) 11 1127.59 8.48 0.01 Zygomorphic Castilleja miniata Con + Het (1 m2) 12 7817.76 0 0.67 Bee * Con, Bee * Het (1 m2) 13 7819.52 1.76 0.28 Bee + Con + Het (1 m2) 14 7823.89 6.13 0.03 Con * Het (1 m2) 11 7826.33 8.56 0.01 Con + Het (10 m2) 12 7827.15 9.38 0.01 Lathyrus ochroleucus Con * Het (1 m2) 11 1918.96 0 0.77 Con + Het (1 m2) 12 1921.80 2.84 0.19 Bee * Con, Bee * Het (1 m2) 13 1925.76 6.80 0.03 Bee + Con + Het (1 m2) 14 1927.42 8.46 0.01 Notes: ΔAIC = AICmodel - AICmin; wi = exp(-0.5* ΔAICi)/Σ exp(-0.5* ΔAICi)

31 Species Model df AIC ΔAIC wi Zygomorphic Vicia americana Con (10 m2) 10 895.96 0 0.72 Con + Het (10 m2) 12 898.75 2.78 0.18 Con * Het (10 m2) 11 901.02 5.06 0.06 Bee * Con, Bee * Het (10 m2) 13 902.79 6.82 0.02 Bee + Con + Het (10 m2) 13 903.01 7.05 0.02 Notes: ΔAIC = AICmodel - AICmin; wi = exp(-0.5* ΔAICi)/Σ exp(-0.5* ΔAICi)

Table 2.6: Generalized additive mixed model selection from the set of a priori candidate models (see Table 2.4), examining effects on seed number for plant species that produced variable numbers of seeds. Only models for which ΔAIC was ≤ 10 are reported. Akaike model weights (wi) show probability that the given model is the best among the entire set of candidate models for that particular species. All models include nuisance covariates (Julian day, Spatial PC1) and nested random effects (transect location in habitat in transect in site) controlling for repeated measures within scaling locations. Species are binned into three floral morphology categories, increasing in handling time for bumble bee visitors. Species Model df AIC ΔAIC wi Disc Arnica cordifolia Con * Het (1 m2) 11 6105.50 0 1 Eurybia conspicua Con * Het (10 m2) 11 5484.20 0 1 Pendular Campanula rotundifolia Bee * Con, Bee * Het (10 m2) 13 10808.99 0 1 Mertensia paniculata Con + Het (10 m2) 12 1526.72 0 0.74 Bee + Con + Het (10 m2) 14 1528.88 2.16 0.25 Habitat + Canopy (10 m2) 11 1535.38 8.66 0.01 Zygomorphic Castilleja miniata Bee + Con + Het (1 m2) 13 12358.06 0 0.89 Con + Het (1 m2) 11 12362.31 4.25 0.11 Vicia americana Bee * Con, Bee * Het (10 m2) 12 1766.15 0 0.85 Con + Het (1 m2) 12 1771.43 5.29 0.06 Con + Het (10 m2) 12 1771.81 5.66 0.05 Con (10 m2) 10 1772.58 6.43 0.03 Notes: ΔAIC = AICmodel - AICmin; wi = exp(-0.5* ΔAICi)/Σ exp(-0.5* ΔAICi)

32 Figure 2.6: A visualization of the relative importance of 7 a priori models (reported in Table 2.4) across spatial scales for explaining seed size for each focal species. ΔAIC values are calculated for all models across all spatial scales per each individual species (ie. ΔAIC = AICmodel,spec1 - AICmin,spec1) and axes for each panel are inverted so that the best models are on top. Species are indicated by abbreviations in the lower right corner of each plot: a) Arnica cordifolia (A.c), b) Eurybia conspicua (E.c), c) Campanula rotundifolia (C.r), d) Castilleja miniata (C.m), e) Geranium richardsonii (G.r), f) Mertensia paniculata (M.p), g) Linnaea borealis (L.b), h) Lathyrus ochroleucus (L.o), and i) Vicia americana (V.a). Species are clustered into boxes representing categories of floral morphology: blue discs, purple pendular, and red zygomorphs. Legend in the top right corner displays models, where green models involve plant effects, orange bee effects, blue bee and plant effects combined, and red habitat effects. The solid black line across each graph is plotted at ΔAIC = 10, and any models located below that line do not have substantial evidence supporting their estimates. Models with ΔAIC > 50 are not plotted.

33 Figure 2.7: A visualization of the relative importance of 7 a priori models (reported in Table 2.4) across spatial scales for explaining seed number for each focal species. ΔAIC values are calculated for all models across all spatial scales per each individual species (ie. ΔAIC = AICmodel,spec1 - AICmin,spec1) and axes for each panel are inverted so that the best models are on top. Species are indicated by abbreviations in the middle right of each plot: a) Arnica cordifolia (A.c), b) Campanula rotundifolia (C.r), c) Mertensia paniculata (M.p), d) Eurybia conspicua (E.c), e) Geranium richardsonii (G.r), f) Castilleja miniata (C.m), and g) Vicia americana (V.a). Species are clustered into boxes representing categories of floral morphology: blue discs, purple pendular, and red zygomorphs. Legend in the top right corner displays models, where green models involve plant effects, orange bee effects, blue bee and plant effects combined, and red habitat effects. The solid black line across each graph is plotted at ΔAIC = 10, and any models located below that line do not have substantial evidence supporting their estimates. Models with ΔAIC > 80 are not plotted.

34 2.3.2 VARIABLE EFFECTS

The most important variables explaining seed size and number differed among flower species, but generalities were also apparent. Model weights for the best model tended to be 0.70 or higher (except for C. rotundifolia and C. miniata, which still exceeded 0.50; Table 2.5), so I examined the effects of all individual variables only within that single best model, for each seed attribute. With one exception, local conspecific density was always present in the best-selected model for each species, whether alone or in combination with other variables.

Seed size

Seed size for four out of eight species was best explained by the local density of neighbouring flowers, both conspecifics and heterospecifics. The top selected model contained a conspecific and heterospecific interaction in three cases and their additive effects in a single species. Seed size for three species was explained only by local conspecific flower density. Bee abundance weakly explained seed size for 4 of 8 species (wi < 0.05; Table 2.5). In the case of C. nd miniata, bee abundance may have had a stronger influence on seed size, as the 2 best model (wi = 0.28) included local bee abundance.

In contrast to biotic conditions, abiotic habitat conditions (habitat, canopy cover) did not have a strong influence on seed size. Only a single species’ (M. paniculata) best model showed an influence of local habitat conditions, with weak support for models incorporating floral neighbourhood (Table 2.5). L. borealis was the only other species to show a small, but weakly supported, effect of habitat (wi < 0.15), though its seed size was best explained by local conspecific density alone (Table 2.5).

Seed number

The density of both conspecifics and heterospecifics best explained seed number in three out of the six species – interactively in two species and additively in one species (Table 2.6). For the other three species, bee abundance also explained seed number – two of six species’ seed counts were best explained by the two-way interactions of local bee abundance with conspecific and with heterospecific densities, and, for a single species, their additive effects were most important (Table 2.6, Fig 2.7). Bee abundance was only ever included in top models for explaining seed number (Table 2.6, Fig 2.7).

35 Explanations of seed size were less consistent (i.e., more candidate models within 10 ΔAIC units) than were explanations of seed number. For size, all species except for E. conspicua had at least four models within 10 ΔAIC units of the best model (Table 2.5, Fig 2.6), while the maximum number of models with some support when explaining seed number was four (Table 2.6, Fig 2.7).

Trends in additive effects

Several hypotheses (Table 2.1) were supported through the trends shown in most species, both for conspecific and heterospecific densities (Fig 2.8), however, due to occasionally large confidence intervals, one must be cautious about over-interpreting these trends (e.g., Fig 2.8e). In particular, we see support for the expectation that increasing density (both conspecific and heterospecific) should increase reproductive output (facilitation) to a peak, with further increases in density decreasing reproductive output (competition). There was a positive effect of increasing density on seed size and number at low to medium conspecific densities (facilitation; Fig 2.8a- c,e,h) in four species (C. rotundifolia, L. borealis, C. miniata and V. americana). Of these four species, three showed negative effects of increasing density (competition) past a threshold point (C. rotundifolia, C. miniata and V. americana; Fig 2.8a,c,e) and one showed weak competition or dilution at higher densities (L. borealis; Fig 2.8b). Conspecific facilitation at low densities was also seen for seed number in C. miniata, with a weak competitive effect past a high-density threshold (~50 conspecific neighbours; Fig 2.8h). Of the species with conspecific facilitative trends, at very low conspecific densities (< 20 neighbouring flowers), two showed evidence of negative impacts (i.e., Allee effects) on seed size and number (C. rotundifolia and C. miniata; Fig 2.8a,c,h).

In two species, heterospecifics also had facilitative effects at low to medium densities (C. miniata and M. paniculata; Fig 2.8g,i), as seed number increased to a maximum (i.e., a competitive threshold). Above this optimum level of heterospecific density, effects of increasing heterospecifics were either negative or neutral. Competitive effects on seed size were evident at all levels of heterospecific density for C. miniata (Fig 2.8d). When comparing additive effects of conspecifics and heterospecifics, I expected Rathcke’s (1983) curve to be shifted to the left for heterospecifics – that is, peaks for positive effects of heterospecific density should be reached at

36

Figure 2.8: Summary of all component smooth functions for additive effects. Each smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Plots are arranged by response variable (top row: seed size; bottom row: seed number). Species are grouped into categories of floral morphology: the four panels within the purple box are tubular and the six panels within the red box are zygomorphic. Species are indicated by abbreviations in the upper right corner of each plot: Campanula rotundifolia (C.r), Castilleja miniata (C.m), Linnaea borealis (L.b), Mertensia paniculata (M.p) and Vicia americana (V.a). Labels a-j before species abbreviations are included for in-text referencing. Shaded areas surrounding each curve indicate 2 * standard error bounds, including the uncertainty about the overall mean. See Appendix D for complete per-species model information.

37 lower densities than those of conspecifics, due to stronger competitive effects arising from incompatible pollen deposition (Table 2.1, Fig 2.2).

For the species with additive heterospecific effects (C. miniata size and number, and M. paniculata number), this expectation was partially supported. For C. miniata, heterospecifics were either competitive at all densities (upper panel of Fig 2.9) or reached their most beneficial peak at a lower density than conspecifics (~30 size-weighted individuals vs. ~50; lower left panel of Fig 2.9). The opposite was the case for M. paniculata, as conspecific individuals were mostly neutral or competitive, and maximum seed number was reached at a higher density of heterospecifics than conspecifics (lower right panel of Fig 2.9). For seed number in both species, however, the optimal density of heterospecifics produced more seeds than the optimal density of conspecifics, such that heterospecifics had a stronger positive impact on seed number than did conspecifics (lower panels of Fig 2.9).

In summary, 8 of the 10 trends in Fig 2.8 partially or fully support the visitation hypothesis initially proposed by Rathcke (1983), integrating the effects of facilitation at low floral densities with competition at high densities, and a stronger competitive effect of heterospecifics was only partially supported.

While the effects of floral density dominated explanations of seed size and number, both bee abundance and habitat had detectable effects, as well. In the only species where bee abundance had an additive effect (C. miniata), more bees led to a linear increase in seed number (Fig 2.8j). In the only species where habitat conditions were important (M. paniculata), seeds were largest at very low canopy cover, and seed size decreased as cover increased (left panel of Fig 2.10). At the highest levels of canopy cover, a weak increase in seed size was observed, which may have contributed to the lack of a significant effect of the categorical habitat type variable on seed size (right panel of Fig 2.10).

Trends in interactive effects

Seed size and number were not just affected additively, as described above. In the case of 3 of the 9 species for seed size, and 4 of the 6 species for seed number, top models included covariate interactions. These patterns are more difficult to visualize, and I have simplified them by summarizing the locations of both the peaks (Fig 2.11) and troughs (Fig 2.12) in seed size or

38

Figure 2.9: Direct comparison between effects of conspecifics (blue) and heterospecifics (red) for the two species with additive impacts on seed traits. Each smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Upper panel shows seed size effects and lower two panels show effects on seed number. Effects of conspecifics are shown in blue, and heterospecifics in red, where the center line represents the estimate and the red or blue shaded regions represent 2 * standard error bounds, including the uncertainty about the overall mean.

39

Figure 2.10: Additive effects of both local (10 m2) canopy cover and habitat type (CC = clearcut, F = forest) on ln(seed size) in M. paniculata (pendular). The y-axis is adjusted to represent the partial component contribution of each variable, controlling for all other variables in the model, centered on zero. Partial residuals are plotted. Shaded curves indicate 2 * standard error bounds, including the uncertainty about the overall mean, and dashed lines on habitat plot indicate one standard error. See Appendix D for complete model information.

40

Figure 2.11: Summary of the relative location of peaks in seed production (both seed size and seed number) for all species in which a model with an interaction between two or more variables was best supported. The upper left panel (a) summarizes species whose best model was Plant 3 (see Table 2.4 for variables in this model), and the lower two panels (b, c) summarize species whose best model was Bee Plant 2 (see Table 2.4). Species are colour-coded by their flower morphology groupings. See Appendix D for complete per-species model information with more detailed individual figures.

41

Figure 2.12: Summary of the relative location of seed production lows (both seed size and seed number) for all species in which a model with an interaction between two or more variables was best supported. The upper left panel (a) summarizes species whose best model was Plant 3 (see Table 2.4 for variables in this model), and the lower two panels (b, c) summarize species whose best model was Bee Plant 2 (see Table 2.4). Species are colour-coded by their flower morphology groupings. See Appendix D for complete per-species model information with more detailed individual figures. number on the interaction surface provided in the model fit. Several general patterns emerge from this among-species comparison.

Patterns attributable to floral morphology were clearest when the best selected model included the interaction between conspecific and heterospecific floral density: seed size and number of both species with a disc morphology were most strongly impacted by this interaction. Within this interaction, density of conspecifics and heterospecifics tended to be of dissimilar positive effect. Optimal seed production (largest size and/or highest count) most commonly

42 occurred at low conspecific densities and intermediate heterospecific densities (3 species; Fig 2.11a). Peaks were never seen when local conspecific and heterospecific densities were similar (minor diagonal; Fig 2.11a), though certain mixtures were beneficial: specifically, when conspecific densities were slightly higher than heterospecific densities (intermediate conspecifics + low heterospecifics and high conspecifics + intermediate heterospecifics; Fig 2.11a). Contrasting with the general trends, for seed number in A. cordifolia, peaks were also seen as expected if there were tradeoffs between the beneficial effects of heterospecific and conspecific density – seeds were larger at high heterospecific densities with correspondingly low conspecifics, and at high conspecific densities with correspondingly low heterospecifics (Fig 2.11a). However, this tradeoff was most commonly associated with minimal seed production: combinations of conspecifics and heterospecifics where at least one variable was at low densities tended to be detrimental, with minima across species clustered mostly at low densities of both con- and heterospecifics, as well as at low conspecific and high heterospecific densities, and high conspecific and low heterospecific densities (Fig 2.12a).

When conspecific and heterospecific density had an interactive effect on seed size or number, the additive-only hypothesis for the difference between their influences (see Fig 2.2) was partially rejected. Instead of negative (competitive) effects beginning at lower densities for heterospecifics in comparison to conspecifics, the opposite was the most common result: the largest peak (seen in the most species and seed traits) was located at intermediate levels of heterospecifics and low levels of conspecifics. Though there was a peak in seed number for two species at low heterospecifics and intermediate conspecifics, heterospecifics more typically had positive effects at higher (intermediate) densities (Fig 2.11a).

The patterns of interaction involving bee abundance imply tradeoffs between bee abundance and local flower abundance. For two species of different floral morphologies (V. americana zygomorphic and C. rotundifolia pendular), floral densities (conspecific and heterospecific) interacting with local bee abundance best explained seed number (Fig 2.11b,c and Fig 2.12b,c). These species shared peaks along the major diagonal of the interaction between conspecific density and bee abundance, such that the most seeds were produced when conspecific density and bee abundance traded off (i.e. low densities of conspecifics require high bee abundance to be effectively pollinated; Fig 2.11b). Low densities of either conspecifics or heterospecifics combined with low bee abundance produced the fewest seeds (i.e., low densities

43 are even worse for attracting pollinators when pollinators are few; Fig 2.12b,c). Conspecific densities of intermediate or high levels produced a peak in seed production for one or both species, no matter the bee abundance (Fig 2.11b). Curiously, the parameter space in between intermediate and high conspecific abundance produced the lowest seed abundance, for both species at intermediate bee densities (Fig 2.12b). That is, there were multiple steep changes in the fitness surface over a relatively short range of conspecific abundances (Fig 2.11b, Fig 2.12b).

High heterospecific density combined with any level of bee presence was never beneficial for seed abundance (Fig 2.11c). Low to intermediate bee abundance was especially detrimental at high levels of heterospecifics (Fig 2.12c), implying that when bees are scarce, they favoured visiting flowers of high-density heterospecifics over those of conspecifics (i.e., heterospecifics were competitors). Intermediate to high abundances of bees combined with intermediate to low densities of heterospecifics were optimal for producing the greatest number of seeds (Fig 2.11c), implying that heterospecifics, when they were sufficiently uncommon, enhanced pollination (i.e., heterospecifics were facilitators).

Trends in effects within morphological categories

Trends within flower shape categories (i.e., disc, pendular, and zygmorphic) were not immediately clear upon initial examination of the data, due to the variation between species in variables selected and the relationships between variables within selected models. However, upon closer examination, some species within the same floral morphology category share some influences.

For scale, morphology does not seem to matter – seed size in discs was best explained by the 10 m2 scale (2/2 species), seed size in pendular flowers was explained by a mixture of 1 m2 (1/3 species) and 10 m2 (2/3 species), and seed size in zygomorphic flowers was explained by a mixture of the two local scales, but with more species influenced by the smaller of the two (1 m2, 2/3 species; Table 2.5, Fig 2.6). This trend suggests a potential decrease in the scale of importance as complexity increases, though not a strong trend given the small sample sizes. However, this decline was not reflected in the scales selected to explain seed number, which were a mixture of the two local scales for discs and zygomorphs, and strictly 10 m2 for pendular species (Table 2.6, Fig 2.7).

44 Flower shape seemed to be associated with traits explaining plant reproduction, but only for disc morphologies. For discs, the interaction between conspecific and heterospecific density was always most important, for both seed size and seed number (Tables 2.4, 2.5). Within both of the more complex floral categories, explanatory variables were not consistent, often even varying between seed characters within a species (e.g., M. paniculata (pendular) and V. americana (zygomorphic); Tables 2.4, 2.5). However, since some models shared variables (e.g., conspecific density was a variable within 5/7 best models; see Table 2.4), it was possible to compare effects of these variables within morphological categories.

For trends in individual variable effects within morphological categories, again the species with disc morphologies produced the clearest shared effects. A. cordifolia and E. conspicua generally shared a peak in seed production at low levels of conspecifics and intermediate levels of heterospecifics (Fig 2.11a) and both tended to produce minimal seeds (small size and/or low number) when conspecific and heterospecific density were simultaneously low (Fig 2.12a). Pendular flowers all shared additive effects of conspecifics on seed size, but the shape of those relationships varied – C. rotundifolia and M. paniculata shared similarities, but the shape of the conspecific density curve in L. borealis was quite different from the other two pendulars (Fig 2.8a,b,f). Seed number only varied for two of the species within the pendular morphology category (C. rotundifolia and M. paniculata), and those two species selected vastly different models explaining seed number (i.e., Bee Plant 2 vs. Habitat; Table 2.6). For the zygomorphs, conspecific density was facilitative at low densities and competitive at high densities for all additive effects (Fig 2.8c,e,h). Heterospecifics had a range of effects in zygomorphic species: strictly competitive (Fig 2.8d), Rathcke’s (1983) curve (Fig 2.8i), or the same curve restricted to low conspecific densities (L. ochroleucus; Figs 2.9a, 2.10a) or modified slightly by bee abundance (Figs 2.9c, 2.10c).

Though in some cases a subset of species within the more complex morphology categories (pendular and zygomorphic) had comparable relationships between selected variables and seed size or number, there were no strong patterns across all three species with these morphologies. Assuming that pendular and zygomorphic species are more specialized for particular pollinators, the expectation that they should therefore experience stronger facilitative effects of heterospecifics can be rejected. There were species within all floral morphologies that were strongly facilitated by intermediate densities of heterospecifics, often with optimal

45 abundances of heterospecifics in the floral neighbourhood leading to greater seed production than optimal abundances of conspecifics.

2.4 DISCUSSION

From this large-scale observational study examining the relationship between reproduction of wildflowers and the densities of neighbouring plants and pollinators at different spatial scales, my analyses have shown both overarching patterns and interspecific variation among plant responses. For all eight species analyzed, local scales, not broader “patch” scales, were most important in determining seed size and seed number. Of the local scales examined, the larger, 10 m2 level most commonly drove patterns in seed traits. At these local scales, conspecific density was the most consistently important variable. However, heterospecific density and bee abundance had both additive and interactive effects on seed size and number for a large proportion of species examined, though bee abundances only ever impacted seed number. For additive effects, I found strong support for Rathcke’s (1983) predictions involving density- dependent pollination. For both conspecifics and heterospecifics, low densities were typically facilitative and high densities were typically competitive. In my system, heterospecifics had a stronger positive effect on seed production than expected when considering the negative impacts of interspecific pollen transfer: when heterospecifics were facilitative for the focal species, at optimal densities more seeds were produced (peak was higher) than at optimal conspecific densities. For interactive effects, patterns were not as clear, for the inherent complexity of interactions that arise from indirect effects often impedes interpretation. When plant species were split into floral morphology categories ordered by expected pollinator handling time, the only clear pattern within groups was that all disc species (most generalized or easily accessible) were impacted by the interaction of conspecifics and heterospecifics for both seed size and number. For both pendular and zygomorphic morphologies, selected effects of conspecifics, heterospecifics and bees on reproduction varied both across and within species (seed size vs. seed number). The floral morphology of competing heterospecifics in the floral neighbourhood may also impact visitation and subsequent seed production, but interpretation of these impacts was beyond the scope of this study (see Appendix C for a summary of general per-species heterospecific effects).

46 Scale

Of the four spatial scales I examined, the larger of the two local scales best explained variation in seed traits across species. Patch-level densities of conspecifics and heterospecifics, or bee abundances, almost never had empirical support for affecting seed traits within this observational study system, but the most local of scales also showed less empirical support. Physical proximity is likely to increase the opportunities for ecological interactions, especially in non-mobile plants, and previous studies have generally found local impacts to be more important than effects at larger scales (Roll et al. 1997, Wright et al. 2003, Marini et al. 2007, Marini et al. 2008, Williams and Winfree 2013). However, there is evidence that animals select foraging habitat at multiple scales (Oatway and Morris 2007, Schmid et al. 2015) as recognized by hierarchy theory (O’Neill et al. 1989). Bumble bees can forage at a large range of different distances from their nests (Rao and Strange 2012, Geib et al. 2015), and resource availability, or flower abundance, tends to drive pollinator foraging patterns (Goulson 1999, 2000, Essenberg 2013a, 2013b, Bennett et al. 2014). It is therefore interesting that an intermediate scale and neither the most physically proximate scale, nor the broader patch scales were most important for seed size and number across wildflower species.

When investigating foraging ranges and site constancy of bumble bees, Osborne et al. (1999) found that bees typically fly close to the nest (~275 m) but this short travel might be an artifact of carrying a heavy transponder. Most researchers find greater travel distances (Westpal et al. 2006, Goulson and Osborne 2009). Osborne and Williams (2001) saw highest return visits of marked bees within patch groups of 729 m2, but individual foragers were also constant to smaller patches of 81 m2, when they weren’t connected to one another. All of these studies demonstrate foraging site constancy of bumble bees, but at a larger scale than was most important for reproduction in the plants that I studied. Other studies have examined alternative foraging strategies, and bees have been shown to use a “traplining” technique (i.e., visiting the same set of plants in a predictable, repeated sequence) at the plant level within larger scale arrays (~100 m2; Thomson et al. 1997, Williams and Thomson 1998). However, bees appeared to restrict their repeated foraging bouts to plants within a smaller subsection of the large arrays (Thomson et al. 1997). Bees also develop traplining strategies over time in stable arrays of small flower patches (6 flowers; Saleh and Chittka 2007), and remember the positions of individual rewarding plants within a much broader home range (Cartar 2004). Given that natural plant

47 populations tend to occur in a clustered, patchy distribution, that pollinators are more site- constant when patches are not connected (Osborne and Williams 2001), and that pollinators reduce flight costs by moving between adjacent flowers in patches (Saleh and Chittka 2007), it is reasonable to expect that bumble bees in my study system may be repeatedly visiting rewarding patches at the 10 m2 scale. They may be responding to this scale as a foraging “patch”, from whose selection occurs within a larger foraging range.

Choosing a scale or range of scales to examine within an ecological study has long been an important yet difficult task (Greig-Smith 1952, Wiens 1989). If an inappropriately broad scale is selected for the process that is being investigated, averaging fine-scale processes within can mask important patterns. Elliot and Irwin (2009) found evidence of nonlinear visitation with respect to density at the 100 m2 scale only observationally, and not within their manipulated experiment, and suggested that expanding their investigation to larger spatial scales may have altered the result. However, density was manipulated at a plant-level, and variables were measured on a per-flower basis but averaged over 100 m2 plot, possibly losing grain. If visitation varied significantly within smaller-scale patches such as the 10 m2 that I examined, a patchy distribution within their 100 m2 patches may have masked their relationship. This issue has been pointed out previously, as a collective, large-scale effect can have a diversity of local effects (i.e., competitive or facilitative), depending on local context; Janovský et al. (2013) found heterogeneity in pollinator community assemblages at the scale of tens of meters that would have been undetected had data been averaged within larger plots.

In the case of my study, selection of the local scale may also be a consequence of grain: fine-scale variation in seed size and seed number may be best explained by fine-grain variables such as local flower densities simply due to the nature of that variation, as many individual seed samples share the same coarse larger-scale flower density (i.e. transect-level measurements at 150 m2 or 300 m2), requiring variation in each seed character measured to be low between plants and between point samples of plants, or there to be very little impact of fine-scale densities, neither of which was predicted. This is a general problem associated with scaling by pooling: as scale increases, the same larger-scale metrics are shared by many differing constituents contained within the pool.

48 Examining the strength of scale effects comparatively is not always the motivation behind looking at multiple scales, and some previous studies have investigated how the types of relationships (and not their strength) differ by scale. Spigler and Chang (2008) found a negative relationship between densities of plants at scales less than 1 m and fitness in a single species, reflecting competitive interactions involving the partitioning of abiotic resources. At scales larger than 1 m, density impacts were positive and related to facilitation in pollinator visitation. Bartkowska and Johnston (2014) found a similar relationship, where at their smallest scale, increasing densities negatively impacted seed production (number of seeds) due to resource competition, and increasing densities at their larger scales were facilitative for pollination and seed number. However, Hegland (2014) found the opposite – facilitative interactions were measured at the smallest scale, and competitive interactions occurred as density increased at larger scales, though the smallest scale examined was larger in comparison to the finest scale of most other studies. In my study, when examining models across multiple scales, in the majority of species I analyzed, models at a single scale or the two smallest scales examined vastly outweighed the empirical contributions of other scales, and thus it was not of interest for me to examine the different effects of the far weaker alternative scales for this study.

Depending upon the range of scales examined for effects on plants and pollinators, different studies produced different results (Roll et al. 1997, Osborne and Williams 2001, Johnson et al. 2003, Gunton and Kunin 2009, Jakobsson et al. 2009, Hegland 2014; Table 2.7), illustrating the importance of including a broad range of potentially biologically relevant scales when examining the impacts of density-dependent processes on plant reproduction and plant- pollinator interactions. As the common plant species I chose to examine tended to all be located within heterogeneous, clumped patches across varying scales, I examined a range of local and patch-level scales which I believed may be important for interactions involving plant sexual reproduction and pollinator foraging. I found that when spanning several spatial scales that were relevant for plants in previous studies (Table 2.7), the most important scale for most species in my study system was the 10 m2 scale, potentially related to the scale of bumble bee selective foraging (Thomson et al. 1997, Cartar 2004, Saleh and Chittka 2007), and loss of fine-scale variation when averaging across larger scales (Janovský et al. 2013).

49 Table 2.7: Summary of the results of 6 studies and the present study, when examining differential effects of increasing spatial scale on plants and/or pollinators, and their interactions. The range of scales tested in each study is reported, along with the most important scale(s) and the processes that acted at the selected scale(s). Some studies reported alternative measurements of scale (e.g., radii), but all were converted to area in m2 for ease of comparison. Study Scale range Important scale(s) Important variable(s) Roll et al. 1997 3-28 m2 3 m2 Plant reproductive success affected by conspecific density Osborne and Williams 2001 81 m2, 729 m2, all Bee foraging site constancy affected by patch 40,000 m2 size Johnson et al. 2003 1 m2, 100 m2 100 m2 Pollination success in orchids affected by sympatric nectar plant density Gunton and Kunin 2009 0.1-71,000 m2 4.5 m2, 11 m2, 45 m2, Plant reproductive output affected by 531 m2, 15,394 m2 male/female conspecific flower densities Jakobsson et al. 2009 1-28 m2 1 m2, 13 m2, 28 m2 Conspecific and heterospecific pollen deposition affected by con- and heterospecific densities Hegland 2014 20-1964 m2 all Visitation rate affected by conspecific (smaller scales) and heterospecific (all scales) densities Present study 1 m2, 10 m2, 150 10 m2, and sometimes 1 m2 Seed set affect by conspecifics, m2, 300 m2 heterospecifics, bee abundance, and canopy cover

50 Effects of variables: complexity

Both density-dependent competition (Campbell 1985, Gunton and Kunin 2009) and facilitation (Callaway 1995, Knight 2003) are commonly reported in studies of flowering plants, with fewer studies producing both effects dependent on density (Callaway and Walker 1997, Choler et al. 2001) as originally proposed by Rathcke (1983). However, the concept of density- dependent pollination has been predicted and observed multiple times within the pollination literature (Sih and Baltus 1987, Ghazoul 2006, Elliot and Irwin 2009, Essenberg 2012, Hegland 2014, Xi et al. 2015). Rathcke’s (1983) prediction produces a complex curve representing the relationship between patch floral density and pollinator visitation with two inflection points, requiring more complex analysis techniques such as higher order polynomial terms (e.g., Nattero et al. 2011) or breakpoint regression (e.g., Ghazoul 2006) for estimation. Previous studies may show contrasting evidence for this nonlinear relationship between density and pollinator visitation or seed set in plants due to limited power in detecting it: polynomial terms in linear models are more restrictive than the nonparametric spline estimates of more recently developed additive analyses (GAM, GAMM; Wood 2006), and nonlinear effects are often simply not tested as alternative hypotheses (Underwood et al. 2014), even though threshold relationships are becoming increasingly more important within the ecological literature (Lennartsson 2002, Turner 2005).

Interactive effects between independent variables are also rarely fully analyzed, perhaps given their inherent added complexity. However, traditional ecological models aiming for simplification can often instead end up complicating understanding of the underlying biological mechanisms (Neill 1974). Without testing for biologically relevant interactions, more intricate underlying relationships due to modification by other species or conditions may be masked. These non-additive modifications can act synergistically or antagonistically (or even a combination of the two, when nonlinear interactions are allowed for) to change the ecological outcomes of pairwise interactions in a way that fundamentally differs from predictions (Strauss and Irwin 2004). If the outcome of a pairwise species interaction may be modified by the presence of additional species, results are not necessarily easily interpreted through only pairwise hypotheses, and without full knowledge of all potential species combinations, specific a priori predictions can be impossible (Wootton 1993). Certain interactions, such as those between competition and predation, have clear expected results (less competition in the presence of higher

51 predation) and have thus been relatively well studied (reviewed by Gurevitch et al. 2000). However, less straightforward interactions lacking clear a priori predictions are rarely tested, and as a result, we may be getting a simpler but less complete picture of the combined importance of multiple variables of interest on any one response variable. Diffuse community-level interactions across multiple species are expected in natural systems (Morris et al. 2007), and interactions are often non-additive and inconsistent in direction (Strauss and Irwin 2004). Within my system, I examined an array of these potential effects across plant species: conspecific, heterospecific, and pollinator presence, additive and non-additive, and the complexity of effects varied across species. This emphasizes the inherent complexity and diversity of plant-plant and plant-animal interactions, even within a single season, thus underlining the importance of testing for biologically motivated interspecific interactions that may be difficult to predict based on pairwise hypothesized relationships alone.

Conspecifics

For (simpler) additive effects of conspecifics, low to medium densities were typically facilitative and high densities were typically competitive, following my expectation that pollination service should change with increasing floral density (Rathke 1983): positive effects of neighbours (facilitation) at lower local floral densities, i.e. more visits per flower (Kunin 1992, Dauber et al. 2010), and weaker, neutral, or negative (competitive) effects at higher densities, i.e. fewer visits per flower (Essenberg 2012). This relationship typically takes the form of a hump- shape, but weak competitive interactions at higher densities can result in an asymptotic curve (Knight 2003). As previously introduced, this pattern has many different proposed mechanisms, and these mechanisms can depend on different effects of varying plant patch densities on pollinator attraction to a patch and their ensuing behaviour in the patch, as well as interactions among individuals (Sih and Baltus 1987, Holmgren et al. 1997, Hegland 2014). The hump- shaped relationship was expected to hold for both seed size and seed number, and patterns were similar across these seed traits, as quality and quantity of visits to a patch should increase and decrease in similar ways with respect to density-mediated visitation. Unfortunately, my study design did not allow for the detection of potential seed size-number tradeoffs within individual plants.

52 Reproduction of flowers in low density patches tends to be limited by visitation frequency (Kunin 1993, Groom 1998, Knight 2003). However, when low density patches do receive visits, proportionally more flowers are pollinated in comparison to high density patches (Klinkhamer and de Jong 1990, Goverde et al. 2002), likely because identifying unvisited flowers in small patches is simpler for a pollinator (Cibula and Zimmerman 1984, Goulson 2000), and it is optimal to forage for longer in lower quality patches (Charnov 1976). Foragers spend more time and visit more flowers in larger patches (Bosch and Waser 2001, Mustajarvi et al. 2001, Bennett et al. 2014), but they often visit proportionally fewer available flowers in these larger patches (Beattie 1976, Heinrich 1979, Sih and Baltus 1987, Sowig 1989, Goulson 2000). In my system, the strength of dilution or competition at high densities varies across species. In several cases, seed size or number at high densities was lower than at medium densities, but still greater than at very low densities. This suggests that although competitive interactions are present at high densities (whether they be for pollinators or alternative resources), the benefits of facilitative attraction of more pollinators to denser patches masks suboptimal (for a plant) departure decisions in individual pollinators, when compared to the reproductive success of isolated individuals. There was a single species, however, in which that was not the case (C. rotundifolia), where high levels of conspecific density produced smaller seeds than individuals with few conspecific neighbours.

At low densities, any increase in plant patch size can lead to a positive effect on reproductive success (Silander 1978, Moeller 2004, Ye et al. 2014), as more pollinators are recruited to the patch, and they more effectively transfer pollen between plants in smaller, more easily assessable patches. However, this energy conservation strategy employed by pollinators in small patches can also lead to an increase in self-pollination, or pollination limited within a genetically similar group of individuals if all conspecific individuals within a patch are highly related (Bosch and Waser 1999). Either (or both) of these mechanisms may be acting at very low densities in my system, as neutral, negative and positive effects were seen on seed size and number at densities lower than where the more obvious trend of facilitative interactions at medium-low densities of both conspecifics and heterospecifics took hold.

53 Heterospecifics

As I adjusted all of my count densities by the average flower size of each particular heterospecific, the effect of individual heterospecific floral display size was removed, and thus all of my densities are a proxy for combined display size and not individual inflorescence or floral count densities. Display size is often associated with a flower’s “availability” (Toft 1983), and pollinators tend to be visual foragers attracted to larger flowers and larger display sizes (Goulson 1999, Hegland and Totland 2005). As such, my adjustment allowed heterospecific species with larger (on average) flowers to contribute higher “densities”, in comparison to smaller heterospecifics. This may have an impact on how closely my results match those of other studies that did not make this adjustment, but allows me to examine the effect of heterospecific density across many pooled species of heterospecifics without allowing for the potential modifying effect of individual floral display size on effects of density on seed size and number (Grindeland et al. 2005).

Results of previous work on impacts of heterospecifics and their density on focal species has varied considerably: sometimes heterospecifics are strictly facilitative (Liao et al. 2011) or strictly competitive (Caruso 1999), and several studies have found low density facilitative and high density competitive effects (Ghazoul 2006, Muñoz and Cavieres 2008), or stronger competition at higher densities (Flanagan et al. 2010, Yang et al. 2011). In my study, additive effects of heterospecifics came in two types: strictly competitive (linear decline in seed production) or a combination of facilitative and competitive (hump-shaped). In the case where heterospecifics were increasingly competitive at higher densities, a mixture of interspecific competitive mechanisms may be contributing to negative impacts across all densities. Quality of pollination declines in cases where interspecific pollen transfer is prevalent (Waser 1978, Campbell 1985, Brown et al. 2002), and many pollinators within my system are generalists and thus have the capacity to transfer pollen from multiple species. Competition can also act through displacement of pollinators from the focal species, should heterospecific species be more attractive in some way (Chiitka and Schürkens 2001, Brown et al. 2002, Muñoz and Cavieres 2008). Heterospecific facilitation may arise due to a variety of previously outlined mechanisms (Schemske 1981, Laverty 1992, Johnson et al. 2003, Ghazoul 2006, Molina-Montenegro et al. 2008), and is typically restricted to low densities. However, in the cases where heterospecifics were facilitative in this study system, they had a stronger positive effect on seed number than

54 conspecifics did – that is, the optimal density of heterospecifics for a focal species led to the production of more seeds than the optimal density of conspecifics. This rejects my hypothesis that heterospecifics should be more competitive than conspecifics with increasing densities, due to incompatible pollen transfer. A possible mechanism behind this result may be that the focal species are less attractive to pollinators on their own (due to having limited reward, being less showy, etc.), and the presence of heterospecifics acts as a stronger “magnet” for a more enhanced, diverse pollinator pool than increased densities of the focal species (Laverty 1992, Johnson et al. 2003).

Con*Het interaction

Studies examining the combined impacts of varying density of conspecifics and heterospecifics on visitation and plant reproductive response are rare (Caruso 1999, Feinsinger et al. 1991), and often conspecific and heterospecific densities are not manipulated simultaneously (e.g., Campbell 1985, Bell et al. 2005, Muñoz and Cavieres 2008, Flanagan et al. 2010). From the current literature, impacts of heterospecifics are commonly mediated by the density and proportion of only a single heterospecific species, and heterospecific facilitation tends to decline into competition when density or proportion increases past a certain threshold (Ghazoul 2006, Muñoz and Cavieres 2008, Dietzsch et al. 2011). To my knowledge, there are no previous studies that have adequately tested for potential interactive (as opposed to simple additive, e.g. Jakobsson et al. 2009) effects of conspecific and heterospecific neighbours on plant reproductive potential; Caruso (1999) and Liao et al. (2011) found significant or near-significant interactive effects within their highly restrictive experimental frameworks, but failed to address the implications of these interactive effects.

For my study, due to the expected complexity of interactions between individual flowers, their neighbours of many different species, and their pollinators, I predicted there might be differential effects of conspecifics and heterospecifics dependent on community context. For additive effects, I expected that heterospecific densities would be more competitive at higher densities than would conspecifics, mainly because heterospecifics appear different (and may therefore be of reduced value in attracting the correct species-specific attention of a foraging bee), and provide incompatible pollen that might clog stigmas. However, I also expected that this might be altered when examining conspecific and heterospecific densities interactively, as the

55 proportion of the mixture of conspecifics vs. heterospecifics may mediate the impacts of one or the other. I found significant interactive effects of conspecifics and heterospecifics in several species, with evidence of tradeoffs between the two demonstrated through dissimilar effects of each respective density. These results do not support the prediction that heterospecifics have a restricted facilitative effect in comparison to conspecifics. Nor do they align with results of previous studies that manipulated conspecific or heterospecific density, and analyzed their impacts on visitation or seed traits without testing for interactions. It is typical for interactive effects to differ from the hypothesized (and measured) separate effects of each individual variable in either a positive or negative direction (reviewed by Strauss and Irwin 2006).

Interestingly, I found that several focal species had areas of optimal seed production clustered at low levels of conspecifics and intermediate levels of heterospecifics. Averaging across all species with interactive conspecific and heterospecific effects, this impact of heterospecifics restricted to low conspecific densities mirrors Rathcke’s (1983) hump-shape, as low seed production clusters at high and low heterospecific densities. It is curious that this strong cross-species effect is restricted to low conspecific densities, as previous studies have found that positive effects of heterospecifics tend to only occur in situations where conspecifics still make up a reasonable proportion of the mixture (Feinsinger et al. 1991, Ghazoul 2006). Since the studies that have asked similar types of questions to mine did not examine a similarly broad range of both heterospecific and conspecific density effects, and were not motivated by looking for nonlinear impacts of density, they may have missed the more complex patterns that I have observed, or simply picked up isolated components of them due to fewer treatment levels.

Factorial experiments are classically used when testing for nonadditive effects of interactions, and though experiments are preferred for isolating the cause and effect of individual variables of interest on one another, they are also more restrictive – these experiments typically lack continuous treatments across the broad ranges of situations occurring in nature (e.g. Caruso 1999, Liao et al. 2011). For example, Feinsinger et al. (1991) manipulated both conspecific and heterospecific densities in conjunction, but did not cover the entirety of the combined conspecific and heterospecific densities that I observed, and only examined the effect of a single heterospecific species. Though observational studies like mine lack power in terms of isolating causation and identifying specific mechanisms behind patterns observed, they allow for the examination of larger variation typical within natural systems.

56 Bees

In my system, bee abundance was only selected for best explaining seed number and never for seed size. If bee abundance is a proxy for increased visitation and increased pollen deposition, from these results, increased visitation does not have a strong effect on the quality (size) of seeds, but increased pollen receipt enhances seed quantity. This mirrors historical assumptions that the impacts of pollen receipt on seed production are restricted to the number of seeds produced, as many previous studies addressing pollination limitation have solely focused on variation in seed or fruit number (Zimmerman and Pyke 1988 and references therein). Though both quality and quantity of pollination are expected to interact in their impacts on seed production (Aizen and Harder 2007), it may be the case here that number of seeds is strongly limited by number of visits and number of pollen grains deposited as mediated by pollinator abundance, while size of seeds is controlled through competition between pollen grains of different quality deposited on stigmas, as mediated by compatible floral density. Since the competitive ability of pollen grains is influenced by their genetic compatibility, it follows logically that “quality” of pollen should be influenced more strongly by mate availability (e.g., pollen donor floral density) than visitation frequency (e.g., pollinator abundance). Wagenius and Lyon (2010) suggested this mechanism to explain the fact that pollen limitation on seed set was still occurring in their focal species, even though visitation frequency increased at low plant densities.

Extraordinarily, snapshot bee abundance had effects on seed number in several species of plant. Due to the temporal limitations imposed upon my study – arising from its observational framework and goal of answering multiple broad questions in a restricted time frame – how I was able to measure bee abundance is intrinsically stochastic. I censused bees present at local and patch levels for short (10 minutes, on average) surveys, during a single day sampling period when seed samples were collected. Thus, the bees that I counted would not have directly affected the developed seeds obtained on that day, that were instead a result of bee visitation ~2 weeks prior. Though it may seem incredible that such a short glimpse of bee presence in an area was actually strongly related to seed number across multiple species, bumble bees are remarkably site-constant when foraging, as long as suitable densities of flowers are continuously present in an area (Osborne et al. 1999, Osborne and Williams 2001). Thus, if higher densities of bees were

57 observed foraging in a particular sampling area, it is likely that there were higher densities of bees in that area during the flowering pollination period of the seeds collected there.

I found a single additive effect of bee abundance on seed number – as predicted, there was a linear increase in amount of seeds produced as bee abundance in the area increased. I assume that this is because as bee abundance increases, the potential for an increase in visitation frequency allowed for more pollen deposition and more seeds successfully sired in C. miniata.

Bee interaction

In two species, seed number was impacted by the interaction of bee abundance and conspecific and heterospecific densities, potentially due to a combination of visitation frequency (quantity) and mate availability (quality) being important for reproduction in these species. Though seemingly intuitive, there are few studies that have investigated the changes in effects of community context on plant reproduction as pollinator abundance varies (Lázaro and Totland 2010, Lázaro et al. 2013, Ye et al. 2014).

When plant densities and bee abundance interacted I found tradeoffs, with caveats. High bee abundance had a positive impact on seed production in almost all cases. Low bee abundance produced more numerous seeds only if conspecific densities were higher, and low conspecific densities were only positive when propped up by high bee abundances. Curiously, a region of conspecific densities between intermediate and high was detrimental for seeds, no matter the bee abundance. These results contrast with previous findings that increases in conspecific density are only facilitative when pollinator abundance is high, and at low pollinator abundance, there is no effect of community context on visitation or reproduction (Lázaro et al. 2013, Ye et al. 2014). These differences may be due to the stochastic and lagged nature of my measures of bee abundance, or the fact that both previous studies involved only two broad scale site-level treatments, high and low pollinator abundance.

High levels of heterospecifics were never optimal for seeds no matter the bee abundance, and the lowest levels of seed production were concentrated at combinations of high levels of heterospecifics and few bees. This is intuitive; higher densities of heterospecifics tend to lead to competitive effects due to increased heterospecific pollen transfer (Flanagan et al. 2010) and lower likelihood of visitation as the proportion of conspecifics declines in the mixture (Ghazoul 2006). However, heterospecifics were facilitative at intermediate and low densities, if enough

58 bees were present – this mirrors previous findings, presumably because in areas where pollinators are abundant, plants compete less with one another for pollinator attraction (Lázaro and Totland 2010, Lázaro et al. 2013, Ye et al. 2014), allowing for facilitative effects of diversity of floral displays (Ghazoul 2006).

Habitat

Habitat conditions can impact a plant’s ability to acquire resources and may alter the way it partitions resources, which can affect pollination and its downstream effects (e.g., seed development). Therefore, the habitat model was included to control for these potential abiotic effects – canopy cover is a measure of local light conditions, habitat type (forest versus clearcut) can impact the availability and flux of nutrients within a system (Saunders et al. 1991), and variation in microclimate (light, temperature and water in particular) is predictably different between harvested and old-growth forests (Aussenac 2000). I expected that if important at all, habitat impacts would depend upon the habitat preferences of each respective species.

In my system, this coarse representation of physical conditions was only important for explaining seed size in a single species – Mertensia paniculata – who produced the largest seeds in very open and very closed canopies. This likely mirrored and overshadowed the effect of habitat, as clearcuts tended to be largely open and forests were more closed. However, variation in seed size was larger within the forest habitat, probably representing the higher diversity of microclimates due to the different types of forest (coniferous and mixed wood) surveyed. M. paniculata is shade-tolerant and typically found in damp areas, but can also be successful during early post-fire regeneration (Reeves 2006), and flowering typically occurs in sunny locations (Morris 1996). It has shown a strong positive response to increased nutrient availability (Turkington et al. 1998, Arii and Turkington 2002), and commonly experiences costly nectar robbing by bumble bees (Morris 1996, Morris et al. 2010). Abiotic resource availability is typically expected to have a more direct impact on seed quality (as opposed to pollen limitation on seed number) in plants (Zimmerman and Pyke 1988), and M. paniculata is sensitive to nutrient fertilization as well as experiencing both costs and benefits to attracting pollinators – these combined factors may explain why seed size was best explained by local habitat conditions while variation in seed number was still driven by conspecific and heterospecific densities.

59 Flower morphology and specialization

Floral shape (morphology) can impact what types of pollinators are able to effectively visit a particular plant species, and the floral constancy of those pollinators (Herrera 1987, Laverty 1994, Fenster et al. 2004, Sargent 2004). Typically, simpler or more open floral morphologies tend to be generalized (i.e., can be visited by a variety of different pollinator species) and more complex flowers, with rewards that may be more difficult to access, tend to be specialized (i.e., they can only be visited by specific pollinators with better floral handling and learning skills). I divided my species into categories of increasing “complexity” or pollinator specialization (disc, pendular, and zygomorphic) and I expected that species with more specialized flowers would respond more positively to increases in heterospecific densities, due to reaping the benefits of a larger display size (Ghazoul 2006) but avoiding the detrimental impacts associated with incompatible pollen transfer from pollinators switching between species, due to increased floral constancy within specialized plants. That is, heterospecifics may act as a form of unvisited “bait”. However, my results did not support this hypothesis – select species within all categories of floral morphology experienced heterospecific facilitation that was often stronger than conspecific facilitation, and the only species that experienced strong competition from heterospecifics belonged to the most specialized category (C. miniata, zygomorphic). In contrast, the most generalized species examined (disc morphology) showed the strongest patterns, but for both discs, intermediate levels of heterospecifics and low levels of conspecifics tended to produce the largest and most numerous seeds, conflicting with the idea that rare (i.e., low density) generalists should experience strong competition from neighbouring heterospecifics. Therefore, in my study system, other factors may be contributing to the reproductive differences between species.

Specialization of both plants and pollinators depend on community context. For instance, floral size and floral density can be completely responsible for all variation in pollination, where flowers with similar traits (form and symmetry) don’t necessarily have the same visitation frequencies, leaving the classical idea of “pollination syndromes” with little support (Hegland and Totland 2005). Overall pollinator abundance could impact this relationship: Lázaro et al. (2013) found that flower traits mattered more for seed set and visitation when there was a low pollinator-to-flower ratio, whereas when pollinators competed more with one another for resources, community context (conspecific and heterospecific density) had a stronger effect. This

60 might be based on the expectation that the generalization of pollinators depends on the community of available forage: pollinators can adjust the breadth of species that they visit depending on plant patch densities, the relative densities of competing flowers, plant population densities, and levels of competition (Kunin 1993, Kunin and Iwasa 1996, Chittka et al. 1997). It is also possible that the specialization of a focal species is only important when compared to the specialization of its competitors: plants of a particular morphology may be most strongly affected by competing plants of that same morphology, if pollinators learn a particular floral syndrome and tend to be constant within that type (Rogríguez et al. 2004), or generalist heterospecifics may be exceptionally competitive if they induce more frequent switching between species in a community. It is also possible that interactions between species are only important when the competing species are similar: pollinators are more prone to switch between species when they are of similar morphology or of similar colour, leading to an increased risk of heterospecific pollen transfer (Chittka et al. 1997, Smithson and MacNair 1997, Hegland et al. 2009, Liao et al. 2011, Hegland and Totland 2012). I did not examine the effects of flower colour, but potential effects of heterospecific floral morphology are summarized in Appendix C, as a formal analysis of these impacts was beyond the scope of this study.

Species-specific considerations

Species-specific characteristics may mediate effects of floral neighbourhood, pollinators, and habitat. I examine two species here. C. miniata is considered a weak competitor, as it is parasitic, tapping into roots of heterospecifics (mostly grasses) for additional nutrients (Matthies 1997). As it is also nearly self-incompatible, it should rely more heavily on pollination for seed production than species that often self-fertilize, and this may explain its linear decrease in seed size as heterospecifics increase in density, but hump-shaped relationship with heterospecifics for seed number. Its mostly outcrossing breeding system may also explain why it was the only species that was impacted additively by bee abundance. G. richardsonii was the only focal species whose seed production could not be explained. This may be a function of its lack of variation in seed number, making seed size a very plastic trait (see Ebert 1994). G. richardsonii is also gynodioecious; individual plants produce all hermaphroditic or all female flowers (Williams et al. 2000), and I did not discern between the types of plants, and thus may have masked density effects arising from changing plant sex ratios (Gunton and Kunin 2009, Bartkowska and Johnston 2014).

61 Constraints on mechanistic conclusions

I based my hypotheses and predictions upon mechanisms of plant-pollinator interaction, but only measured their putative outcome (i.e., variation in seed size and number). Due to the multi-species nature of my project, as well as the spatial and temporal limitations imposed upon large-scale observational studies, especially when constrained within the flowering season at higher latitudes, I was not able to directly measure several potentially key components of the plant-pollinator relationships that I inferred. I assumed that all of my species were potentially pollen limited (at least for the facultative responses to changes in the densities of competitor flowers), given that pollen limitation is estimated to occur in the majority of natural plant populations (Burd 1994, Larson and Barrett 2000), and that changes in seed size and seed number were driven by changes in pollinator visitation made in response to differences in neighbourhood plant densities, not by abiotic resource availability. There is evidence from the literature supporting this assumption for almost all of the forb species that I examined, though there may be a gradient of reproductive reliance on pollination among species (Table 2.8). However, this qualitative gradient in reliance on pollination does not appear to be associated with any of my results – bee abundance was linked to improved seed production in one species from each of the four general pollen limitation categories, and there were no clear trends in density effects that can be definitively linked to differences in pollen limitation.

Relating seed size and number to the quality and quantity of pollen delivered by pollinators could not be achieved in my study. Most small-scale experiments examining plant- pollinator interactions also perform supplemental pollination and/or direct observation of pollinators at focal flowers, and degree of pollen limitation can differ between species, within the same species in different environments, and even within the same species at different times or points in the season (Larson and Barrett 2000, Ashman et al. 2004, Knight et al. 2005). Pollen limitation for each of my species may vary according to any one of these characteristics, and I was not able to quantify how that may be impacting reproductive responses. Degree of outcrossing required and realized can also vary between species, between individuals, and through time, and breeding systems in plants can range from completely self-compatible (i.e., autogamy results in no reduction in seed set) to completely self-incompatible (i.e., strictly outcrossed pollen required for seed development), with a usually unmeasured level of outbreeding depression. Degree of self-compatibility can affect plant reproductive responses

62 Table 2.8: Summary of each focal species’ mating system and pollen limitation, from the literature. Species are ordered broadly from “most pollen-limited” to “least pollen-limited”, with no previous data found for L. ochroleucus.

Species Mating system Pollen limitation Reference(s) Require outcrossing Linnaea borealis Self-incompatible Seed production, self, bagged < open Wilcock and Jennings 1999, pollination, outcrossed Scobie and Wilcock 2009 Eurybia conspicua Obligate outcrossing Low germination with self-pollen, open Jones 1978, Allen et al. 1983 pollinated plants >80% germination Castilleja miniata Nearly self- Marginally pollen-limited, outcrossing Lertzman 1981, Cariveau et al. incompatible leads to greater seed production 2004, Hersch and Roy 2007 Some evidence of pollen limitation Arnica cordifolia Sexual and asexual Seed production, bagged < unbagged Kao 2007 reproduction plants Campanula rotundifolia Self-compatible % fruit set and seeds per capsule, no Nuortila et al. 2004 pollination < self < outcrossed Mertensia paniculata Self-compatible Seeds per flower increase with Morris et al. 2010 visitation Limited evidence of pollen limitation Vicia americana Self-compatible Selfing more common than outcrossing Gunn 1965 Geranium richardsonii Self-compatible, Female reproductive success not pollen Williams et al. 2000 gynodioecious limited, autogamy and geitonogamy may lead to inbreeding depression No data Lathyrus ochroleucus N/A N/A N/A

63 to changes in the floral neighbourhood (Feinsinger et al. 1991). Though almost all of my species are to some degree self-compatible (Table 2.8), the strength of negative impacts of self- pollination for the majority has not been well characterized, so contributions of this important factor are not easily elucidated within my study system. Certain species may also show different competitive responses to increases in heterospecific densities, dependent upon varying effects of incompatible pollen deposition: for a pair of species in one study, when pollen loads deposited on stigmas contained as low as 10% heterospecific pollen, fruit set declined by more than 50% in both competitors (Tokuda et al. 2015).

Though it is possible that plant densities are associated with differences in unmeasured abiotic conditions (e.g., soil moisture, nutrients), and these abiotic differences are driving variation in seed size and number, I expect that is not entirely the case in my system. Abiotic microclimate can be heterogeneous at scales smaller than my smallest local scale (1 m2; Lechowicz and Bell 1991), and densities at the 10 m2 scale tended to best explain reproduction across species, highlighting the importance of plant densities separated from focal flowers by at least 0.5 m. Though I attempted to control for any potential effects of abiotic heterogeneity at a larger scale (~20 m) with the random effect of transect location, and any specific spatial correlation with my spatial covariate, it is still possible that some variation in seed size and number was driven by abiotic factors, which may contribute to the complexity of my results.

In my study, there was temporal limitation in my examination of the effects of floral neighbourhood on seed formation. I collected seeds at the same time as floral densities were measured, so the floral densities present at the time of seed formation were likely somewhat different from floral densities used in my analyses. Seed densities were only recorded for the portion of the season that they were present, and only at the local (1 m2 and 10 m2) scales due to detection difficulty for patch-level measurement, which may contribute to the selection of the local scale for all species. The floral densities that directly impacted visitation of flowers that produced collected seeds would have existed approximately 2 weeks prior to collection. An assumption of my study is that those densities should remain relatively constant over the time period of seed formation. Across years, at least, population floral density remains consistent (Knight 2003), and sequential flower maturation on individual plants can maintain relatively constant within-plant densities (Ashman et al. 2004), but present and past floral densities were most likely not identical.

64 Conclusion

Despite the associations that are inherently more complex in the field when all potential covariate effects are not controlled for, and the strictly observational nature of this study, I have been able to provide novel insight into patterns that may occur arising from this complexity, as opposed to more simple, directed experimental questions that have clear mechanistic answers within their more restricted framework. For instance, examining these types of questions as manipulative field experiments, it is very common to only use a single heterospecific species as a facilitator or a competitor (e.g., Bell et al. 2005, Flanagan et al. 2010, Liao et al. 2011), a single focal species (e.g., Caruso 1999, Ghazoul 2006, Liao et al. 2011), few treatments (e.g., Muñoz and Cavieres 2008, Yang et al. 2011), adjust overall patch density but not relative density of conspecifics and heterospecifics (e.g., Muñoz and Cavieres 2008, Yang et al. 2011), and an assortment of other restrictions in order to better determine direct mechanisms involved within relationships. Instead, I have examined more than one focal species, many different varieties of heterospecifics, a broad range of natural conspecific and heterospecific densities, and nonlinear interactive effects of a variety of independent variables, I have provided a window into the true complexity of plant-plant and plant-pollinator interactions in the field. In particular, my study has highlighted the importance of interactions between explanatory variables, and suggests that other studies, including experimental ones, neglect this common component at their peril.

In my system, a variety of variables were important for reproductive output across species, but only at local scales. Nonlinear effects of floral neighbours and pollinators dominated, and additive linear relationships only appeared for a single species. Conspecific density was ubiquitous in its importance, appearing in all but one model for explaining seed size and number. Conspecific effects typically were hump shaped as predicted by Rathcke (1983), although the shape of the relationship was rarely as simple as the Rathcke curve. Heterospecific densities were surprisingly facilitative, and their optimal densities additively produced more seeds than optimal conspecific densities. When interacting with conspecific densities, heterospecifics were also more strongly facilitative, producing peaks in seed production at intermediate levels when restricted to low densities of conspecifics, contrary to expectations. The impacts of floral morphology on reproduction in my plants were not clear, and further examination of the role of the diverse interacting morphologies in natural communities on reproduction is required in order to begin to answer the complex associated questions.

65 CHAPTER 3: CROSS-SCALE INTERACTIVE EFFECTS OF LOGGING ON UNDERSTORY FORB REPRODUCTION

3.1 INTRODUCTION

3.1.1 BACKGROUND

Scale of ecological interactions: local and landscape effects

The effects of landscape modification on small-scale ecological interactions are increasingly becoming a focus of conservation research, with detection of negative influences of large-scale habitat change on a multitude of species having a diversity of life history characteristics (Wu and Hobbs 2002). As interactions at local scales are often mediated through processes happening at broad spatial scales, the field of landscape ecology continues to receive increasing attention (Turner 2005). Spatial heterogeneity (i.e., differences among places) has always been an important component of ecological systems (Turner 2005). Heterogeneity can be defined in alternative ways: at very small spatial scales, within a region of interest, or across “landscapes” (Wiens and Milne 1989, Turner 2005). Landscapes are, often by virtue of their extent, spatially heterogeneous. As anthropogenic impacts on habitat complexity continue to increase, research into the effects of non-natural habitat disturbance or alteration has moved to the forefront of conservation research.

It is important to investigate how altered landscapes affect the distribution, abundance, and persistence of species living within them (Turner 2005). Severity of landscape alteration has been correlated with declines in species abundance and diversity, but the precise mechanisms of these declines are thus far unclear (Aizen and Feinsinger 1994, Wilson et al. 1999, Lennartsson 2002). In a review of fragmentation studies, McGarigal and Cushman (2002) concluded that the ecological processes affecting community dynamics in heterogeneous landscapes remain poorly understood. Understanding interactions among species within changing landscapes clearly bears further study.

One growing focus within landscape ecology has been the effect of landscape alteration on plant-pollinator dynamics – made more compelling by mounting evidence of accelerating pollinator decline worldwide, along with declines in their mutualistic plant counterparts (Potts et al. 2010). Fragmentation and habitat loss can alter native plant assemblages (Saunders et al.

66 1991), reduce plant abundance (Rathcke and Jules 1993, Lienert 2004), and reduce the abundance and diversity of pollinators (Winfree et al. 2009, Kennedy et al. 2013). Effects of fragmentation and habitat loss can also be positive, at least at habitat edges where increase in species diversity follows from a mixing of adjoining communities (Ozinga et al. 2004, Marini et al. 2008). However, the relative strength of these impacts of habitat fragmentation or loss on communities remains controversial and difficult to predict (Ewers and Didham 2006, Hadley and Betts 2011).

Given that all interactions occur locally, and that these interactions are often the consequence of broader-scale processes, there appears to be a logical hierarchy in expectation of outcome. Ecological interactions at local scales should always be detectable, and, depending on circumstance, they might also reflect processes at broader spatial scales, with a typical expectation of impact decreasing as scale increases. For plants in particular, perhaps due to their immobility and small size, local interactions tend to be stronger when explaining variables such as species richness and diversity (Wright et al. 2003, Marini et al. 2007, Marini et al. 2008). It is reasonable to expect that plant-pollinator interactions (e.g., frequency of visitation), and their outcomes (e.g., seed set or rate of pollen collection), may be more determined by local processes (e.g., shading, density and identity of interactors) than landscape processes (e.g., habitat fragmentation or loss). However, many studies of the impacts of fragmentation on pollinators and plant reproduction have found significant negative responses (Rathcke and Jules 1993, Murcia 1996, Kearns et al. 1998, Aguilar et al. 2006, Ferreira et al. 2013). Both local and landscape-level factors can have strong implications for the outcome of plant-pollinator interactions (Schüepp et al. 2014), but interactive effects across spatial scales have not been thoroughly investigated.

Cross-scale interactions: landscape configuration indirectly affects local relationships

It is difficult to predict, understand and reliably detect impacts of habitat fragmentation and loss across large scales when individuals interact locally, but it does not follow that no impacts are occurring (Ewers and Didham 2006, Hadley and Betts 2011). When investigating for direct large-scale impacts, logistical issues can drive a stronger limitation on sampling efforts, representing a challenge for gathering enough broadly scaled data (Soranno et al. 2014). However, it is increasingly apparent that ecological processes are often influenced by a broad

67 array of components that are multi-scaled both spatially and temporally, and the importance of examining these complex relationships has recently come into broader focus (Soranno et al. 2014, Heffernan et al. 2014).

Cross-scale or cross-level interactions are processes at a single scale (spatial, in this case) that interact with processes at a different scale, often leading to nonlinear effects that are difficult to predict and interpret (Cash et al. 2006, Peters et al. 2007, Soranno et al. 2014). Macrosystems ecology, an emerging subdiscipline of ecology, is focused on processes occurring at the continental scale and how these broad, diverse ecological phenomena may impact and interact with processes occurring at smaller scales (Heffernan et al. 2014). Within ecology, the identification of broad-scale drivers that interact with fine-scale processes to regulate system dynamics is increasingly important, and when left undetected this can result in errors in the generalization of specific results across regions (Peters et al. 2007, Soranno et al. 2014). Problems of ignorance (the failure to recognize the importance of cross-scale interactions altogether), assumed additivity (the failure to consider non-additive effects of landscape), and plurality (the failure to recognize that there isn’t always a single best scale that will apply to the system as a whole, or for all actors in the system) can lead to issues with management and conservation. For example, the failure to identify dynamic linkages across levels can result in inaccurate assessment of problems and inability to implement ecologically sustainable solutions (Cash et al. 2006).

Recent work that has tested for cross-scale interactions has found integrated impacts of multiple scales on bird habitat selection (Vergara and Armesto 2009), small mammal successional habitat selection (Schweiger et al. 2000), and seed dispersal by frugivores (Vergara et al. 2010). These effects are driven by the fact that – similar to pollinators – birds, mammals, and frugivores can select habitats at different scales, and landscape-level habitat context can moderate spatial heterogeneity in resources. When large-scale spatial context is altered, this can alter the ways that highly mobile individuals select habitat at the fine scale. For instance, high degrees of fragmentation led to birds underusing high quality local habitat patches (Vergera and Armesto 2009) – thus, conservation efforts involving the addition of small-scale high quality habitat patches would not be sustainable in a highly fragmented environment for this system.

68 Mechanisms behind cross-scale effects

Given the inherent complexity and difficulty of predicting the effects of habitat fragmentation and loss on plant-pollinator dynamics (Hadley and Betts 2011), this important ecological interaction may be driven by more complex cross-scale dynamics; plant-plant interactions for pollinators occur on a small scale, but pollinators may be selecting their foraging habitat at a variety of different scales (Essenberg 2013b). Wildflower reproduction in fragmented habitats is pollen-limited, and the local abundances of pollinators causing this limitation show no impact of landscape-scale effects (Williams and Winfree 2013). However, local habitat characteristics did explain pollinator visitation, with a higher abundance of pollinators recorded in brighter or larger patches, depending on focal plant species. Given that landscape-scale fragmentation and habitat loss can moderate the level of sunlight penetration in forests (Geiger 1965), which can impact insect foraging activity (Kevan and Baker 1983, Herrera 1995), and that habitat alterations can change the density and distribution of floral resources within landscapes (Saunders et al. 1991), large-scale processes may have more of an indirect effect on plant- pollinator interactions. It is important to investigate whether the large-scale context these interactions occur within has predictable effects on plant-pollinator relationships (Maron et al. 2014).

Spatial arrangement of habitat at a variety of scales affects plant reproduction. Large- scale habitat modifications can affect the availability and distribution of abiotic resources important for plants and pollinators, potentially leading to cascading effects in local communities. For example, nectar production may be altered by changes in water flux (Zimmerman 1983), which may influence pollinator attraction. Small-scale fragmentation changes bumble bee foraging behaviour (Goverde et al. 2002), and potentially indirectly impacts plant reproduction. Changes in seed production were strongly impacted by changes in pollinator behaviour with respect to changing floral densities, but only above a threshold of landscape disturbance (Ghazoul et al. 1998). At a more local scale, visitation might also respond to the context in which the floral patches are situated, which is affected by the habitat matrix – when floral displays were surrounded by bare ground as opposed to a community of grasses, visitation and reproduction was positively affected (Diekötter et al. 2007). Local habitat configuration and plant community can be jointly important for pollinator visitation, as floral densities were most important for forager patch use, but distance to forest edge was a factor as well, in that

69 communities near forest edges were visited more often, and forest cover reduced visitation at the 10 m scale and increased visitation at the 20 m scale (Bennett et al. 2014).

Bertness and Callaway (1994) proposed the stress-gradient hypothesis (SGH), which suggests that since biotic interactions (e.g., plant-plant interactions and plant-pollinator interactions) are mediated by environmental conditions, the nature of the biotic interactions should vary along an environmental stress gradient: facilitative interactions between individuals or between species tend to be more common in stressful environments. There is evidence of this phenomenon of changing interactions along an environmental gradient (e.g., Choler et al. 2001, Pugnaire and Luque 2001), but its generalizability has been controversial (Maestre et al. 2005, Lortie and Callaway 2006, Maestre et al. 2006), and its application tends to focus primarily on plant-plant interactions (e.g., Wheeler et al. 2015). In plant-pollinator systems, it is even possible that the SGH might take the form of stronger positive impacts of increasing floral densities at high levels of habitat fragmentation. Namely, if fragmentation induces stress in pollinators, making if difficult for them to find appropriate forage, there may be stronger benefits of mutual attraction for individual flowers. However, increased floral abundance at lower levels of fragmentation can result in more pollinators in agricultural landscapes with more flowers and more semi-natural land (Nayak et al. 2014), though this didn’t have consequences for crop seed set. Facilitation can also fail to vary across stress gradients (García et al. 2015).

3.1.2 HYPOTHESES AND PREDICTIONS

Within fragmented habitats, local conditions could be more important in additive effects than landscape conditions for explaining the broad structure of both plant and pollinator communities (Wright et al. 2003, Kennedy et al. 2013, Williams and Winfree 2013). In contrast, I propose that landscape-scale changes can also alter small-scale (i.e., microhabitat) relationships between individuals. That is, landscape conditions should interact with local variables to explain plant reproduction. As previously noted, fragmentation can affect a variety of abiotic environmental characteristics of microhabitats. This abiotic heterogeneity can then influence floral displays (e.g., number of inflorescences, inflorescence size, nectar production, nectar quality), degree of isolation, level of connectivity and the local community. Habitat alterations leading to changes in floral quality may have different impacts depending on the plant species, as

70 certain wildflowers may be better adapted to disturbed habitats, while others may be more suited to remnant habitat conditions in highly fragmented landscapes. Microhabitat alterations can also impact pollinator behaviour (Thomson et al. 1987), and consequently visitation and plant reproductive success. Different pollinator species may react differently to local environment changes, further complicating interpretation of the effects of larger-scale habitat conversion (Steffan-Dewenter et al. 2002, Weiner et al. 2014).

For the plant-pollinator relationship in particular, spatial scales drive interactions between individuals differently in plants and animals: plants are limited by their immobility, restricting plant-plant interactions to local scales, but pollinators can be highly mobile and, depending on their body size, may choose foraging habitat at larger scales. Pollinators may be strongly impacted by fragmentation and loss of habitat (Rathcke and Jules 1993, Murcia 1996, Kearns et al. 1998, Aguilar et al. 2006, Ferreira et al. 2013) and in animal pollinated plants the most common reason for pollination failure is changes in pollinator communities (Wilcock and Neiland 2002). It is therefore reasonable to hypothesize that impacts of habitat change at the landscape scale on plant reproduction may be driven by landscape-altered plant-pollinator relationships at the small scale, particularly if pollinators adjust how they visit different local patches based on area and/or fragmentation of habitats at the landscape level.

Currently within the boreal forest, logging is the predominant force of change in habitat structure (Niemalä 1999, Cyr et al. 2009), and it can have significant impacts on ecological functioning (Drapeau et al. 2000, Boucher et al. 2009, Schmiedinger et al. 2012). Logging also strongly distorts the distribution of forest stand ages (Cyr et al. 2009) and impacts plant communities differentially relative to the natural disturbance within the boreal forest: fire (McRae et al. 2001). Within logged northern forests, a plant’s reproductive success could be altered dependent upon either: the degree of alteration from forest to clearcut in the surrounding landscape, or whether a plant is located within a forest or clearcut habitat, irrespective of the large-scale proportion of that habitat. Clearcuts and old growth forests are very different habitat types for both plants and pollinators, and may impose different environmental stressors on their interactions. Conspecific and heterospecific competition for pollinators may increase as the proportion of logged habitat increases in the landscape, since clearcuts can support higher densities of early-successional plant species adapted to gap conditions and increased light availability (Suding and Goldberg 2001). In areas of high logging, high densities of flowers in

71 clearcuts combined with lower densities in nearby close-canopied forest fragments may lead to depressed reproduction in forests, as pollinators are drawn to forage in clearcuts with abundant floral resources, sometimes disproportionately so (Cartar 2005). Meanwhile, in forests, where increasing densities should be facilitative in landscapes with lower logging, plants may instead suffer from Allee effects due to pollinator inattention (hypothesized by Farmer 2014). Given that different plant species are better suited to growth in old growth forests or open clearcuts, and understory communities are altered with increasing logging in the boreal landscape (Cartar 2005), the identity and strength of the facilitative or competitive impacts of increasing densities with respect to increasing clearcutting may be different dependent on the habitat specialization of the plant species. Thus, early-successional species that thrive in clearcuts may be positively or neutrally affected by increased logging in the landscape, where forest specialist species are more likely to be negatively impacted by high levels of logging.

Habitat type alone (independent of its proportion) may be an important landscape component. The effects of local variables such as conspecific and heterospecific floral densities and bee abundance on seed set may therefore differ between these different habitat types. Clearcutting dramatically changes the level of sunlight penetration in forests (Geiger 1965, Aussenac 2000), which may impact insect foraging activity (Kevan and Baker 1983, Herrera 1995) and competition between plants for light availability (Suding and Goldberg 2001). Conspecific and heterospecific facilitation could be stronger in clearcuts, as declines in tree cover at edges can lead to improved visibility of patches for pollinators (e.g., Bennett et al. 2014).

Habitat fragmentation and loss can disrupt plant-pollinator interactions and threaten local persistence of plant and pollinator populations (Rathcke and Jules 1993), but whether clearcut logging poses this threat in Canada’s forests is generally unclear. In Chapter 2, I found strong local controls on seed set. Impacts of landscape alteration on plant reproduction and plant- pollinator interactions are of increasing concern for conservation, and the effects of clearcut logging on understory plants motivated my site selection within the foothills forests of southern Alberta. For the purposes of this study, landscape is constrained to large-scale variation in habitat type within a preselected study site – large scale for small understory plants, but within the typical foraging range of their bumble bee pollinators. I therefore test for landscape-local interactions, in attempt to disentangle how large-scale influences may alter the outcomes of local

72 interactions, as measured by plant reproductive output, and detailed in Chapter 2. By examining the variables that best explained seed size and seed number at the local scale interacting with a set of larger-scale variables that may modify local plant-pollinator interactions, I assess whether local effects vary across the environmental gradient proportion of landscape logging, and how they vary, particularly with respect to plant species habitat preferences.

3.2 METHODS

All data used for the analyses interpreted in this chapter are from the same field season as Chapter 2, sampled between July and August 2012. See Chapter 2 for detailed field methods.

Landscape-level logging

The sixteen 1.77 km2 sites sampled (Fig 2.3) varied in level of landscape disturbance from 0% to 75% clearcut, to allow examination of the effects of logging intensity on plant reproduction (Fig 3.1). To balance sampling of landscape-level logging throughout the season, each consecutively sampled site was randomly chosen from within strata of the lower third (≤30% logging), the middle third (31%-47%) and the upper third (>47%) of the range of disturbance in rotating sequence (lower, middle, upper). Clearcut age also varied across sites, though all cuts were 35 years old or younger to reduce effects of succession (Fig 3.1).

Data analysis

The analyses for this chapter are sequentially based off of those that were ran in Chapter 2. After selecting the model (of seven a priori types, see Table 2.4) with the lowest AIC across local and patch scales for each species, I then built off of that best-selected model for each interactive landscape-scale model. To test for large-scale interactions, I interacted each variable within the best-selected model with previously determined logging variables. Four a priori logging interactors were assessed for their influence on local effects: percent logging in the landscape (as a continuous variable, a 3-state variable and a 2-state variable), and habitat type (forest or clearcut). Logging categories were selected so that the number of sites within each category was balanced (see Fig 3.2). Since the estimation of effects of logging is quite coarse in this study due to limited site sample size (n=16), certain species were not found at all sites and were thus restricted in the range of percent logging that they experienced (Fig 3.3). Logging was

73

Figure 3.1: Frequency histograms of percent logging in the landscape (left panel) and average age of clearcut in the landscape (right panel) across all sites.

74

Figure 3.2: Summary of the percent logging in the landscape (1.76 km2, circular, per site) over all 16 sites sampled. The pair of dashed lines indicate where sites were split for pooling into three states of logging: low, medium, and high. The single dotted line indicated where the split was made for two-stated logging: low, and high.

75

Figure 3.3: Summary of how percent landscape logging was split into 2- or 3-states for each species analyzed. Dashed lines separate the three states of logging (low, medium, high) and dotted lines separate two states of logging (low, high). Red indicates when a particular landscape-scale interaction best explained seed size, number, or both in a particular species, when points are red, continuous logging was the selected interactor.

76 tested in both continuous and categorical forms to attempt to adjust for non-continuous sampling in site blocks, to simplify interpretation, and to increase the likelihood of successful model fit (convergence), due to the complex nature of some interacting local variables (multiple two-way interactions, e.g., bee*con + bee*het).

Species habitat preferences

Three species were always present in the forested habitat of each site they were sampled from, but achieved higher local floral densities in clearcuts (Table 3.1). L. borealis grew at slightly higher floral densities in clearcuts. A. cordifolia produced comparatively depressed floral densities in the forest, and M. paniculata produced slightly higher floral densities at the smaller local scales in clearcuts and dramatically higher densities at the 10 m2 scale in clearcuts. L. ochroleucus was equally likely to be found in either habitat type, with similar densities in forests and clearcuts as well. Four species were found in the majority of clearcuts within the sites they were sampled at (> 70%; Table 3.1), but with a range of floral densities depending on habitat type. Floral densities of C. miniata in forests and clearcuts were similar. V. americana grew at slightly higher densities in clearcuts. C. rotundifolia was present at higher densities in clearcuts, and E. conspicua was found at much higher densities in the more disturbed habitat type. Though there appears to be a range of habitat preferences across species, there are no clear “forest specialists” within this group, as almost all were found at greater floral densities in clearcuts. This is reflected by literature on each species, as most have a variety of habitat preferences and have been found recolonizing disturbed areas (Table 3.2). Several are also known to only flower, or flower more extensively, in canopy gaps in forests (e.g., A. cordifolia, E. conspicua, M. paniculata; Table 3.2). Competition may be stronger in more stable habitats like forests, than in relatively disturbed habitats like clearcuts, such that plants may be larger in stable habitats, but produce more flowers in disturbed ones (e.g., Solbrig 1971).

3.3 RESULTS

Logging has a significant influence on the effects of local factors on seed characters across wildflower species. Landscape-scale interactions were species-dependent, but at least one seed trait in the majority of species examined (five of eight) showed improved AIC values with

77 Table 3.1: Summary of each species’ habitat preferences, displayed through frequency of presence in forests and clearcuts, and mean local conspecific floral densities in each habitat type. Percentages were calculated for each to use as habitat preference metrics – presence of sites was calculated as the proportion of all sites sampled that the species was found in that particular habitat (forest or clearcut), and clearcut density preferences (CC%) were calculated scaled to species-specific floral number (i.e., meanConForest / (meanConForest + meanConClearcut)). The scale at which local densities were most important in Chapter 2 is bolded for each species. Species are separated into forest and clearcut “types” by the percentage of sites they were located in each habitat type (F = forest, CC = clearcut, B = both), and ordered by their locally important densities within those “types” (i.e., forest types are ordered from lowest CC% to highest CC% and clearcut types are ordered from highest CC% to lowest CC%). Type Species Presence (# of sites) Mean 1 m2 density Mean 10 m2 density Forest % Clearcut % Forest Clearcut CC% Forest Clearcut CC% F Linnaea borealis 12/12 100 6/12 50 39 59 60 39 49 56 Mertensia paniculata 12/12 100 7/12 58 25 27 52 11 22 66 Arnica cordifolia 13/13 100 6/13 46 4 13 76 2 13 87 B Lathyrus ochroleucus 7/9 78 7/9 78 20 18 45 12 15 56 CC Eurybia conspicua 3/7 43 6/7 86 6 20 77 3 14 82 Campanula rotundifolia 5/11 45 8/11 73 5 14 74 4 10 71 Castilleja miniata 6/13 46 10/13 77 13 15 54 9 10 53 Vicia americana 4/11 36 8/11 73 17 27 61 12 13 52

78 Table 3.2: Summary of each focal species’ habitat preferences, from the literature. Species Habitat preference Reference(s) Arnica cordifolia Moist, shady woods to drier, more exposed locales, Weber 1972, Moss 1983, Young 1983 coniferous forests to subalpine meadows, open habitats better for growth and seed production than shaded Eurybia conspicua Woodlands and clearings, moist to dry meadows, forest Moss 1983, Breaitung 1988, Boufford openings, thickets, and clearings, tolerant to forestry 1997, Harper and Macdonald 2002 harvest, flower extensively in canopy openings Campanula rotundifolia Moist to dry hillsides, meadows and open woods, rocky Moss 1983, Pahl and Smreciu 1999, sites, alpine outcrops, responds quickly to changes in land Lindborg and Eriksson 2004, Lindborg et usage, spreads quickly through open sites al. 2005 Linnaea borealis Woodlands, important component of boreal and mixed Howard 1993, Tannas 1997 wood forest understory, key ground cover component in climax communities but also found in disturbed areas e.g., cutblocks Mertensia paniculata Shade-tolerant, damp places such as woodlands, thickets, Beckingham and Archibald 1996, Morris meadows, stream banks, but present in early regeneration 1996, Tannas 1997, Turkington et al. after fire since flowering occurs in sunny locations, prefers 1998, Reeves 2006 nutrient-rich soils Castilleja miniata Wide variety of habitat types such as open woods, wet to Moss 1983, Tannas 2004, Klinkenberg dry meadows, fens, grassy slopes, roadsides, early 2014 successional Lathyrus ochroleucus Moist woods and clearings, can grow naturally in disturbed Moss 1983, Harper and Macdonald 2002 areas e.g., forest edges after clearcutting Vicia americana Typically found in open woods and meadows but can Moss 1983, Coladonato 1993, Gerling et colonize disturbed and agricultural land e.g., coal mines, al. 1996, Pahl and Smreciu 1999 road sides, adapted to early successional conditions

79 the addition of a large-scale interaction (Table 3.3). Of the interacting variables, logging amount was the most commonly selected modifier of local effects: logging affected at least one seed trait of all species where landscape interactions mattered, but the variant of logging (continuous, 2- state, 3-state) that mattered most differed between species. Habitat type was only selected once as an interactor with local variables, for its influence on seed size in A. cordifolia.

Model weights varied more than they did in Chapter 2: most top models had weights exceeding 0.65, but the best model weight for several species was less than 0.50 (C. rotundifolia, M. paniculata, and V. americana). However, to maintain consistency and clarity, I examine only interactive effects within the single best model for each species. Thus, two species for which the best model was the basic local model had strong secondary contributions of landscape-scale interactions, but were excluded from further discussion (C. rotundifolia continuous logging with wi = 0.36 and V. americana habitat type with wi = 0.26; Table 3.3).

Two species (C. rotundifolia and C. miniata) are not present in Table 3.3 for effects on seed number because none of the large-scale interaction models could be fit successfully. For several other species, convergence of a subset of the candidate landscape interaction models failed, and this happened more frequently when trying to explain seed number (five models across two species) than seed size (three models across three species, two of which did not select landscape interaction models). Model failure was attributed to either non-convergence (model could not be optimized successfully, and “timed out”) or lack of variation in the data (insufficient sampling within all levels of the interacting landscape variable).

Logging interactions

Though presence of an interaction of logging in the landscape with locally chosen important variables was widespread across species, the nature of the interaction was not wholly consistent. However, there were some patterns among these landscape-altered effects. In all cases, the large-scale interaction altered local-only dynamics in such a way that the original local interactions were unrecognizable in some landscapes.

M. paniculata was the only species for which a local impact of habitat quality on seed size was found, and closed canopies and open canopies both had positive effects on seed size (left panel of Fig 3.4). The addition of a continuous effect of logging altered this relationship, as open canopies (low canopy cover) were associated with larger seeds only in more highly

80 Table 3.3: Model selection from the set of candidate landscape interactions, examining effects on seed size and number for different plant species. Only the base local model (Table 2.4) and interaction models for which ΔAIC was ≤ 10 are reported. Akaike model weights (wi) show probability that the given model is the best among the set of candidates. All models (GAMM) include nuisance covariates and nested random effects (see Tables 2.4, 2.5). Species are organized by categories of their best-selected models (logging, habitat, and no landscape interaction (= local model)), and within logging interactions, by how continuous logging was. “Type” from Table 3.1 is listed with each species (F = forest, CC = clearcut). Species Model df AIC ΔAIC wi Logging interaction (F) Mertensia Size Logging (continuous) 12 1118.66 0 0.43 paniculata Local model 11 1119.10 0.44 0.35 Logging (2-state) 13 1120.10 1.44 0.21 Logging (3-state) 15 1127.35 8.69 0.01 Number Logging (continuous) 13 986.63 0 1 Local model 12 1526.72 540.10 0 (F) Linnaea Size Logging (3-state) 14 363.27 0 0.67 borealis Logging (2-state) 12 364.79 1.52 0.31 Logging (continuous) 11 370.28 7.00 0.02 Local model 10 377.16 13.88 0 (F) Arnica Number Logging (3-state) 17 6045.39 0 1 cordifolia Local model 11 6105.50 60.11 0 (CC) Eurybia Size Logging (2-state) 14 2226.88 0 0.77 conspicua Logging (3-state) 17 2229.32 2.44 0.23 Local model 11 2296.99 70.10 0 Number Logging (2-state) 14 5383.04 0 0.91 Logging (3-state) 17 5387.73 4.68 0.09 Local model 11 5484.20 101.13 0 (CC) Vicia Number Logging (2-state) 17 1760.54 0 0.94 americana Local model 12 1766.15 5.61 0.06 Habitat-type interaction (F) Arnica Size Habitat type 14 1043.69 0 0.96 cordifolia Local model 11 1051.34 7.66 0.02 Logging (2-state) 14 1052.71 9.02 0.01 Logging (3-state) 17 1053.78 10.10 0.01 Local model only (CC) Campanula Size Local model 10 3391.00 0 0.41 rotundifolia Logging (continuous) 11 3391.29 0.29 0.36 Habitat type 12 3392.69 1.69 0.18 Logging (2-state) 12 3395.18 4.19 0.05 (CC) Castilleja Size Local model 12 7817.76 0 0.86 miniata Habitat type 16 7822.54 4.77 0.08 Logging (continuous) 13 7823.27 5.51 0.06 Notes: ΔAIC = AICmodel - AICmin; wi = exp(-0.5* ΔAICi)/Σ exp(-0.5* ΔAICi)

81 Species Model df AIC ΔAIC wi Local model only (B) Lathyrus Size Local model 11 1918.96 0 0.85 ochroleucus Habitat type 14 1922.66 3.70 0.13 Logging (2-State) 14 1927.09 8.13 0.01 (CC) Vicia Size Local model 10 895.96 0 0.44 americana Habitat type 12 897.07 1.11 0.26 Logging (2-State) 12 898.33 2.36 0.14 Logging (continuous) 11 898.49 2.53 0.13 Logging (3-State) 14 900.84 4.87 0.04 Notes: ΔAIC = AICmodel - AICmin; wi = exp(-0.5* ΔAICi)/Σ exp(-0.5* ΔAICi)

Figure 3.4: A comparison of local and local-landscape interactive influences on seed size in M. paniculata. Left panel depicts strictly local influence of canopy cover on ln(seed size), from Fig 2.8. Right panel contour plot depicts effects of the interaction between landscape logging intensity (continuous) and canopy cover on ln(seed size). This surface estimate was obtained from the GAMM: ln(seed size) ~ Julian + SPC1 + spline(canopy cover, logging) + random(loc in tran in hab in site). Pink contours represent the highest values of ln(seed size), and blue contours indicate smaller seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps. logged landscapes, and closed canopies (high canopy cover) were only associated with larger seeds in sites with low levels of logging (right panel of Fig 3.4). M. paniculata also had a unique response to logging with respect to seed number – locally, conspecific and heterospecific densities had an additive effect, where higher densities of conspecifics were competitive and higher densities of heterospecifics were facilitative (top panels of Fig 3.5), but the interactive effect of logging mostly overwhelmed density impacts. In M. paniculata, intermediate levels of logging always produced the most numerous seeds, with a slightly facilitative effect of

82

Figure 3.5: A comparison of local and local-landscape interactive influences on seed number in M. paniculata. Top panels show effects of the local model (presented in Fig 2.8). Lower panel contour plots depict the interactive effect of landscape logging intensity (continuous). These lower surface estimates were obtained from the GAMM: sqrt(seed count) ~ Julian + SPC1 + spline(con, logging) + spline(het, logging) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

conspecifics at high densities in low and high levels of logging, and a slightly competitive effect of heterospecifics at high densities for all levels of logging except the highest levels – mostly opposite of the original local impacts (Fig 3.5).

Seed size in L. borealis at the local scale was best explained by a facilitative effect of increasing conspecific density, but fit was improved by the addition of a 3-state interaction with logging (Fig 3.6). The local pattern of facilitation persisted at low and medium levels of logging, while at high levels of logging the slope was steeper – in highly logged landscapes, low

83

Figure 3.6: A comparison of local and local-landscape interactive influences on seed number in L. borealis. Top panel shows the local model effect (presented in Fig 2.8). Lower panel component smooth plots for conspecifics depict the differential effects of landscape logging intensity (low, medium, high). This surface estimate was obtained from the GAMM: ln(seed size) ~ Julian + SPC1 + spline(con, by: 3-state logging) + random(loc in tran in hab in site).

84 conspecific densities produced much smaller seeds, but there was a strong facilitative effect of increasing densities above that (Fig 3.6).

As with L. borealis, seed number in A. cordifolia was best explained by the 3-state interaction of logging with local effects. A. cordifolia was the only species for which a different landscape interaction was selected for seed size vs. seed number: logging best explained seed number and habitat type best explained seed size (Table 3.3). The best local model for seed number was the interaction between conspecific and heterospecific density, and A. cordifolia produced the most seeds at low levels of heterospecifics and intermediate levels of conspecifics (top panel of Fig 3.7). However, taking into account the effect of landscape-level logging altered the solely local effects of conspecifics and heterospecifics at all levels of logging. Similar to L. borealis, high levels of logging altered the local interaction surface most drastically in A. cordifolia – at both low and medium levels of logging, high levels of conspecifics and low levels of heterospecifics were associated with the highest numbers of seeds (bottom left panels of Fig 3.7). At high levels of logging, only low levels of conspecifics were associated with higher seed production, and increasing densities of conspecifics only produced more seeds at high densities of heterospecifics.

Models fitted for E. conspicua were similar to models for A. cordifolia for both seed size and number, with the best model featuring the interaction between local conspecific, heterospecific density and a 2-state logging variable. The trends for both seed size and number were similar in E. conspicua, both local-only and large-scale interactive (Fig 3.8, 3.9). Locally, intermediate levels of heterospecifics combined with low or high densities of conspecifics was optimal for seed production, and at low levels of logging intermediate heterospecifics and intermediate to high conspecifics produced the largest, most numerous seeds (Fig 3.8, 3.9). At high levels of logging, however, all densities of conspecifics were facilitative, but fewer conspecifics were only facilitative when there were also fewer heterospecifics in the local community (bottom right panels of Fig 3.8, 3.9).

Seed number in V. americana was locally influenced by a complex relationship between bee abundance and conspecific and heterospecific density. This was the only species with such local relationships in which models with landscape-scale interactions were successfully fit (Table 3.3). At the 1 m2 scale, seed number at low conspecific densities was improved by the presence

85

Figure 3.7: A comparison of local and local-landscape interactive influences on seed number in A. cordifolia. Top panel shows effects of the local model. Lower panel contour plots depict the interactive effect of 3-state landscape logging intensity (low, medium, high). These lower surface estimates were obtained from the GAMM: sqrt(seed count) ~ Julian + SPC1 + spline(con, het, by: 3-state logging) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

86

Figure 3.8: A comparison of local and local-landscape interactive influences on seed size in E. conspicua. Top panel shows effects of the local model. Lower panel contour plots depict the interactive effect of landscape logging intensity (2-state). These lower surface estimates were obtained from the GAMM: ln(seed size) ~ Julian + SPC1 + spline(con, het, by: 2-state logging) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

87

Figure 3.9: A comparison of local and local-landscape interactive influences on seed number in E. conspicua. Top panel shows effects of the local model. Lower panel contour plots depict the interactive effect of landscape logging intensity (2-state). These lower surface estimates were obtained from the GAMM: sqrt(seed count) ~ Julian + SPC1 + spline(con, het, by: 2-state logging) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

88 of more bees, but medium and high levels of conspecifics were facilitative no matter what the bee abundance, and a band of intermediate densities between medium and high conspecifics led to a decline in seed number, no matter the bee abundance (top left panel of Fig 3.10). More seeds were produced at intermediate densities of heterospecifics, and the same number of seeds could be produced either through fewer heterospecifics or more locally abundant bees (top right panel of Fig 3.10). More complex patterns arose with the addition of interaction involving landscape- level logging. At sites with low levels of logging, the conspecific and bee interaction was similar to that seen locally, and heterospecifics were associated with more seeds only at low levels, combined with high bee abundance (bottom left panels of Fig 3.10). However, at high levels of logging, all densities of conspecifics were facilitative and only high densities of heterospecifics were facilitative, but these positive impacts were restricted to low, and not high, local bee abundances (bottom right panels of Fig 3.10).

Though landscape alterations of local impacts on seed size varied across species, there was a clear trend supported or partially supported by the majority: at high levels of logging in the landscape, overall seed production was lower. That is, regardless of the shape of the interaction surface, the largest seeds and the smallest seeds within highly fragmented landscapes were smaller than their equivalents at lower levels of landscape disturbance, or highly logged landscapes produced fewer seeds overall (L. borealis, Fig 3.6; A. cordifolia, Fig 3.7; E. conspicua, Fig 3.8, 3.9; and partially supported by V. americana, Fig 3.10). The second major trend was that high levels of logging tended to produce the most altered form of local relationships across species: low and moderate levels of landscape logging were usually similar in their effects, and more closely mimicked the local relationships with seed size and number than did high levels of logging (L borealis, Fig 3.6; A. cordifolia, Fig 3.7; V. americana, Fig 3.10). Species were arranged by how similar the effects of logging on their reproduction were, using coarse habitat preference groupings (see Table 3.1). Species most often found in forests tended to have more continuous logging interactions than clearcut species, and species of similar habitat types shared certain components of each interaction (e.g., interacting variables, or how many states of logging were selected for; Table 3.3). It was also more likely for species that had shared habitat preferences or similar densities in forests and clearcuts (see Table 3.1) to select the local model with no logging interaction (e.g., L. ochroleucus, C. miniata, size in V. americana; Table 3.3).

89

Figure 3.10: A comparison of local and local-landscape interactive influences on seed number in V. americana. Top panels show effects of the local model. Lower panel contour plots depict the interactive effect of landscape logging intensity (2-state). These lower surface estimates were obtained from the GAMM: sqrt(seed count) ~ Julian + SPC1 + spline(con, bee, by: 2-state logging) + spline(het, bee, by: 2-state logging) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

90 Interaction with habitat type

A habitat interaction (differential local effects dependent upon whether the plant was growing in a forest or a clearcut) was only detected for a single seed trait for a single species in this study system. Seed size in A. cordifolia was best explained by habitat’s influence on the local interaction between conspecific and heterospecific densities (Table 3.3). Locally, large seeds resulted from low densities of conspecifics combined with intermediate or high densities of heterospecifics, or lots of conspecifics and few heterospecifics (top panel of Fig 3.11). In clearcuts, that local interaction was mirrored, but forests showed a substantially different relationship – low levels of conspecifics and low to intermediate levels of heterospecifics were associated with larger seeds (bottom panels of Fig 3.11). However, forest seeds were smaller, on average, than in clearcuts. In A. cordifolia, since habitat type had an influence on seed size, whereas level of logging in the landscape had an influence on seed number (Fig 3.7), overall quality of reproduction (a combination of seed size and number) may be affected in complex ways, by both where an individual is growing (forest or clearcut) and the context of its level of landscape disturbance.

3.4 DISCUSSION

Reproduction in the majority of understory forbs studied was better explained by a landscape-scale interaction with local influences, than simply local influences alone. For five of eight species analyzed, an interaction with logging or habitat type improved the local explanation of seed size and/or seed number, with logging being the predominantly selected interactor. The influence of habitat specialization of individual plants on response to logging was not completely clear, possibly because all species were originally chosen for analysis based upon how common they were within the entire study area (both forest and clearcut habitats), and though some species were found more frequently in forested areas, almost all species grew at higher densities in clearcuts (see Table 3.1). However, species that were more equally abundant across habitat types tended to be best explained only by local conditions, and not a landscape interaction. Responses to landscape-level variables could be generally linked to species-specific traits, and might to be related to their relative competitive ability in disturbed landscapes. Stronger

91

Figure 3.11: A comparison of local and local-landscape interactive influences on seed size in A. cordifolia. Top panel shows effects of the local model. Lower panel contour plots depict the interactive effect of habitat type (forest or clearcut). These lower surface estimates were obtained from the GAMM: ln(seed size) ~ Julian + SPC1 + spline(con, het, by: habitat) + random(loc in tran in hab in site). Pink is associated with higher seed counts, and blue contours represent fewer seeds. Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps. competition may be occurring in highly logged sites, due to higher floral densities in clearcuts. For almost all species analyzed, high levels of logging in the landscape (when compared to medium or low levels) had the strongest effect on local interactions, suggesting a possible threshold of disturbance, and high levels of logging almost always led to depressed seed production, potentially arising from the aforementioned increased competition for pollinators or

92 resources. Across species, when logging or habitat type had an interactive influence on seed production, local relationships were sometimes substantially different from the landscape-free model. This highlights the importance of examining ecologically relevant cross-scale interactions within fragmented landscapes.

“Forest” species trends

The seed size of M. paniculata individuals may be under disruptive selection, where larger seeds were found in high logging in the landscape (i.e., in open locations when the landscape was composed of more open locations, being highly logged), and in shaded areas in sites that were primarily forested (i.e., closed canopies). If logging disturbance selects against shade-adapted individuals, rapid evolution can occur on contemporary timescales, such as within decades (Stockwell et al. 2003), and the majority of clearcuts within this study system were 10- 35 years old (Fig 3.1). It is possible that M. paniculata has two locally adapted ecotypes, dependent upon landscape disturbance levels (Hufford and Mazer 2003). Seed number of M. paniculata was highest at intermediate levels of logging, with minimal impacts of local conspecific and heterospecific densities once logging was taken into account. For M. paniculata, optimal seed production (the largest, most numerous seeds) occurs within intermediate levels of landscape logging in more open habitats. Since M. paniculata is highly nutrient-dependent (Beckingham and Archibald 1996, Turkington et al. 1998, Arii and Turkington 2002) and flowers densely in open areas (Morris 1996), the reproductive output of M. paniculata may have been more strongly influenced by abiotic factors (canopy cover, nutrient fluxes and microclimate) altered by forest harvest (McRae et al. 2001, Heithecker and Halpern 2006) than by pollinators.

Both L. borealis and A. cordifolia grew in forests in all of the sites they were sampled from, but both produced higher floral densities in clearcuts than forests, though floral densities of A. cordifolia were more highly depressed in forested habitats than L. borealis (Table 3.1). In both species, there appeared to be a threshold effect of high levels of logging in the landscape, where low and medium levels of logging produced effects similar to the non-interactive local impacts, and high levels of logging altered the relationship between locally measured variables. For both species, higher densities of conspecifics were facilitative at lower levels of logging, but at high levels of logging, facilitation of seed size (A. cordifolia) or number (L. borealis) occurred

93 only at low conspecific densities. These effects may have occurred because of higher levels of plant-plant competition for pollinators in highly logged landscapes: clearcuts support higher overall (i.e., combining all species) floral densities. In low density patches within a densely populated landscape, seed production may be depressed since pollinators have the freedom to visit more highly productive patches, and individuals have a difficult time getting noticed unless they are growing at higher densities. Perhaps for A. cordifolia in highly disturbed sites, patches only attract pollinators when they contain a diversity of resources (i.e., more heterospecifics; Ghazoul 2006), whereas L. borealis tends to always grow in monospecific patches (S. Johnson and R. Cartar, pers. obs.).

A. cordifolia was also the only species for which a habitat-type interaction best explained seed size. In clearcuts, low densities of conspecifics combined with intermediate or high levels of heterospecifics produced the largest seeds, but only lower densities of both conspecifics and heterospecifics produced larger seeds in the forest. Seeds produced by forest-dwelling plants tended to be smaller than clearcut seeds. Combining logging effects on seed number and habitat- type effects on seed size in A. cordifolia may lead to unexpected impacts on overall demography of this species. In landscapes with a high proportion of forests, high conspecific densities and low heterospecific densities lead to more seeds, but forest plants produced smaller seeds at high densities of conspecifics. In landscapes with abundant clearcuts, low levels of conspecifics and high densities of heterospecifics lead to higher numbers of seeds, and clearcut plants produced larger seeds in those same conditions. Therefore, A. cordifolia may fare best demographically in highly logged landscapes, when surrounded by abundant heterospecifics.

“Clearcut” species trends

E. conspicua is known to be tolerant to disturbance (Harper and Macdonald 2002) and to flower extensively in canopy openings (Breitung 1988). This was reflected in its apparent habitat preferences (Table 3.1) and its response to high levels of logging in the landscape. In highly disturbed areas, E. conspicua produced similar numbers of similarly sized seeds over its entire range of conspecific densities, though primarily at low levels of heterospecifics, and competitive interactions (inferred by small seed size) only occurred when densities of E. conspicua were proportionally low with respect to competing flowers. In forested landscapes (i.e., low levels of logging), optimal seed production only occurred at intermediate to high levels of conspecifics

94 and intermediate densities of heterospecifics. If E. conspicua is one of the competitively dominant species in clearcuts, pollinators may prefer it, as there is a tendency for bees in particular to specialize on the most common floral variety when foraging (Smithson and Macnair 1996, Smithson and Macnair 1997). Thus when growing in a landscape composed of mostly clearcuts (i.e., high logging), all densities of E. conspicua could be facilitative, as long as heterospecifics were not in substantially larger proportions (Feinsinger et al. 1991, Chittka et al. 1997). However, in landscapes composed mostly of forests (i.e., low logging), patches may need to be sufficiently attractive (i.e., high densities of conspecifics combined with intermediate densities of heterospecifics) to induce visitation to E. conspicua when it is rare.

V. americana can colonize disturbed sites (Coladonato 1993, Gerling et al. 1996, Pahl and Smreciu 1999) and is highly attractive to bees (Inouye 1980), but in my study system it appeared to only be successful in highly disturbed areas when bee abundance was low. When visitors are few, flower species compete with one another for visitation, and greater seed production in V. americana during these conditions might be related to bumble bee preference (Inouye 1980) and floral traits (e.g., floral shape) being more important (Lázaro et al. 2013). In highly logged landscapes with locally abundant bees, V. americana seemed unable to attract pollinators, even when flowering at high densities. This may be because of the increased importance of community context when the bee-to-flower ratio is high (Lázaro et al. 2013), and the fact that V. americana had a preference for growing in clearcuts, but presents its flowers at relatively low density, and usually low in the vegetation, with concomitant barriers between flowers. In clearcuts, all other species surveyed were comparatively much denser, therefore potentially drawing pollinator attention away in cases of stronger pollinator competition.

Overarching trends

For almost all species, high levels of logging in the landscape led to depressed seed production – seeds were smaller and fewer across all ranges of local plant and bee abundances in highly disturbed sites. I expect that this may reflect higher levels of competition between plants in highly logged landscapes, because clearcuts tend to support higher diversity and densities of floral resources overall (Cartar 2005, Pengelly and Cartar 2010, Farmer 2014). However, clearcuts also attract bumble bees, often disproportionately so with respect to their additional floral resources (Cartar 2005, Farmer 2014), which should lead to a higher per-flower visitation

95 rate in clearcuts. Therefore, altered pollinator foraging behaviour in dense patches (e.g., visiting fewer flowers per patch or modifying visitation to large inflorescences in dense patches; Mustajarvi et al. 2001, Grindeland et al. 2005), increased competition between pollinators leading to diet breadth expansion and thus greater risk for interspecific pollination (Fontaine et al. 2008), or competition for alternative resources such as water or nutrients may be limiting seed production for these densely flowering forbs. Though it appears that open canopies in clearcuts foster increased flower production (see Table 3.1), increased competition for resources or altered pollinator foraging behaviour may have led to lower reproductive output for these more highly abundant plants.

High levels of logging also appeared to impose a stronger modification of local effects than low or medium levels of logging in the landscape, suggesting that clearcut logging disturbances may have a threshold-like effect on the reproductive performance of understory plants. Disturbance thresholds are a common thread within landscape ecology, and exist when the effects of fragmentation or disturbance are undetectable below a particularly rapid transition point in ecosystem quality (Turner 1989, Metzger and Décamps 1997, Groffman et al. 2006, Peters et al. 2007). In my study system, this transition point appears to be at roughly 50% logging in the landscape. Given this threshold, forestry management should be adjusted with respect to the clustering of clearcutting within a “landscape”, and block cuts planned at densities of less than half of a ~2 km2 region.

96 CHAPTER 4: GENERAL DISCUSSION

My results offer insight into the complexity of interactions within pollination systems, highlighting the importance of local floral neighbourhood as well as landscape context for seed production in understory forest forbs within the fragmented southern Albertan foothills forests. In Chapter 2, using a range of species differing in their phylogeny and floral traits, I found that immediate local scale was more important than broader patch scale for seed size and seed number, with the most important scale being the 10 m2 area surrounding focal individuals. Conspecific densities were a constant influence on reproduction across species, but heterospecific densities and bee abundances often additively or interactively affected reproductive output, sometimes producing a stronger effect on seed production than conspecific densities alone. In Chapter 3, I found that for a majority of species, especially those that showed differential presence or abundance between forested and clearcut habitat types, large-scale interactions involving logging within the site landscape altered the local interactions found in Chapter 2. High levels of logging in the landscape had the strongest impact on local relationships, leading to the largest alterations of local effects, and depressed seed production in most species.

At the local scale, variation between species was substantial both in the models selected and the relationships between variables within those models, and that variation remains predominantly unexplained, particularly for effects of heterospecifics. Surprisingly, heterospecifics were important for seed production in a variety of species, and most commonly were more beneficial for seed production at their optimal densities than were conspecifics. Few studies examine effects of multiple heterospecifics on reproduction in species across the broad ranges of densities that occur in nature (especially in disturbed landscapes), and community-level analyses are important for guiding management decisions, since flowering plant species occur in diverse communities and strictly pairwise interactions between species in the field are rare (Strauss and Irwin 2004). Future work might focus on selecting multiple focal species along a gradient of known species-specific floral characteristics such as compatibility, complexity, or specialization, and attempt to control for other non-focal characteristics such as colour. It would also be beneficial to manipulate potentially facilitating heterospecific densities in a way that allows testing of mechanistic hypotheses. To pinpoint the mechanisms of increased

97 heterospecific facilitation in this system, more care must be paid to species of heterospecifics, their traits, their densities, and their interactions.

Amount of logging in the landscape was important for driving the reproductive outcomes of local plant-plant and plant-pollinator interactions, but effects were indirect. Cross-scale interactions are notoriously difficult to predict and interpret (Cash et al. 2006, Peters et al. 2007, Soranno et al. 2014), but may be particularly important for understanding ecological interactions in an increasingly fragmented world. In Chapter 3, I found that plants that were equally likely to be present in either habitat type, or produced comparable densities in both habitat types, were less likely to be impacted by a landscape-level interaction with local variables. When a plant species was affected by an interaction in my study system, high levels of logging in the landscape most likely led to stronger competition between individuals for pollination or resources as population-level plant densities increased, and effects were strongest within 1.77 km2 sites that consisted of more than half clearcuts. In all cases where I found landscape interactions, local influences were dramatically altered at varying levels of landscape change, highlighting the importance of paying attention to and testing for potentially confounding “lurking” multi-scaled interactions in ecological systems (Sandel and Smith 2009).

As timber harvesting within the Canadian boreal forest continues to increase its footprint, forestry planners managing for biodiversity could benefit from attention to my results, as well as through the integration of similar studies into adaptive management practices. Given my detection of a threshold impact of logging above approximately 50% removal within my measured landscapes (1.77 km2), it may be beneficial to restrict clearcutting to proportions below this threshold in order to mitigate alterations to local processes within natural communities, particularly for understory plants and their pollinators examined in this thesis. Since intermediate levels of logging did not appear to be highly detrimental in my system, and for some species were even associated with positive effects on reproduction, clearcutting does not necessarily need to be eliminated, but decisions with respect to the scale and arrangement of clearcuts should be informed by ecological data, and sites with intensive logging should be continuously monitored for impacts on natural communities.

Within such a large-scale study of so many plant species, it was difficult to isolate the mechanisms behind landscape-mediated changes in local interactions, and future work should

98 focus on attempting to test clearer hypotheses and predictions that may be guided by habitat preferences of each plant species. In order to determine whether changes in pollination services or abiotic resource sharing are causing the effects of increased logging on reproduction in these understory species, direct observation of visitation within patches of different densities in different habitat types, or supplemental pollination experiments may be useful. For this to be logistically practical, sampling intensity would have to be diminished in some way, whether it be through examining fewer focal species, examining fewer sites but restricting them to the high and low ends of the logging gradient, or restricting other variables measured in some way. My results offer insight into the complexity of ecological interactions embedded within altered landscapes, and the importance of examining that complexity within a natural observational framework, even though specific mechanisms behind certain relationships were difficult to identify.

This study provides a window into the complexity of interactions influencing reproductive output in understory forbs, and the importance of examining natural processes on a broad scale. Through the quest to answer the multiple questions about the effects of density dependent pollination, spatial scale, and landscape disturbance on reproduction in understory forest forbs that motivated of this project, I have both shed light on natural complexity, and its resolution in the reproduction of forest plants, and unearthed additional questions. Large scale observational studies are important for examining multispecies interactions in the field, and are by definition of general relevance. They also raise unexpected results that are not easily replicated in restrictive experimental scenarios. However, when examining so many different interacting processes, it can be difficult to determine causation. Given the results of my study, future work can be guided in such a way that individual questions (i.e., how do different species of heterospecifics in the community drive reproduction of a focal species, or how does logging in the landscape alter local effects on plant reproduction?) are examined over sufficient levels of naturally occurring variation, with manipulations of the study system performed in the field when necessary, to elucidate mechanisms affecting seed production in fragmented landscapes.

99 LITERATURE CITED

Addicott, J. F., J. M. Aho, M. F. Antolin, D. K. Padilla, S. John, D. A. Soluk, and J. S. Richardson. 1987. Ecological neighborhoods: scaling environmental patterns. Oikos 49:340–346.

Aguilar, R., L. Ashworth, L. Galetto, and M. A. Aizen. 2006. Plant reproductive susceptibility to habitat fragmentation: review and synthesis through a meta-analysis. Ecology Letters 9:968–980.

Aizen, M. A., and P. Feinsinger. 1994. Forest fragmentation, pollination, and plant reproduction in a chaco dry forest, Argentina. Ecology 75:330–351.

Aizen, M. A., and L. D. Harder. 2007. Expanding the limits of the pollen-limitation concept: effects of pollen quantity and quality. Ecology 88:271–281. Akaike, H. 1973. Information theory and an extension of the maximum likelihood principle. Pages 267–281 in B. N. Petrov and S. Caski, editors. Proceedings of the Second International Symposium on Information Theory. Akademiai Kaido, Budapest, Hungary.

Allen, G. A., M. L. Dean, and K. L. Chambers. 1983. Hybridization studies in the Aster occidentalis (Asteraceae) polyploid complex of western . Brittonia 35:353– 361.

Arii, K., and R. Turkington. 2002. Do nutrient availability and competition limit plant growth of herbaceous species in the boreal forest understory? Arctic, Antarctic and Alpine Research 34:251–261.

Ashman, T. L., and G. Arceo-Gómez. 2013. Toward a predictive understanding of the fitness costs of heterospecific pollen receipt and its importance in co-flowering communities. American Journal of Botany 100:1061–1070.

Ashman, T.-L., T. M. Knight, J. A. Steets, P. Amarasekare, M. Burd, D. R. Campbell, M. R. Dudash, M. O. Johnston, S. J. Mazer, R. J. Mitchell, M. T. Morgan, and W. G. Wilson. 2004. Pollen limitation of plant reproduction: ecological and evolutionary causes and consequences. Ecology 85:2408–2421.

Asmussen, M. A. 1979. Density-dependent selection II. The Allee effect. The American Naturalist 114:796–809.

Aussenac, G. 2000. Interactions between forest stands and microclimate: Ecophysiological aspects and consequences for silviculture. Annals of Forest Science 57:287–301.

Azevedo, J. C., A. H. Perera, and M. A. Pinto. 2010. Forest landscapes and global change: challenges for research and management. Springer, New York.

100 Barrett, S. C. H., and K. Helenurm. 1987. The reproductive biology of boreal forest herbs. I. Breeding systems and pollination. Canadian Journal of Botany 65:2036–2046.

Bartkowska, M. P., and M. O. Johnston. 2014. The sexual neighborhood through time: competition and facilitation for pollination in Lobelia cardinalis. Ecology 95:910–919.

Beattie, A. J. 1976. Plant dispersion, pollination and gene flow in Viola. Oecologia 25:291–300. Beckingham, J. D., and J. H. Archibald. 1996. Field guide to ecosites of Northern Alberta. Special Report 5. Natural Resources Canada, Canadian Forest Service, Northern Forestry Centre, Edmonton, Alberta, Canada.

Bell, J. M., J. D. Karron, and R. J. Mitchell. 2005. Interspecific competition for pollination lowers seed production and outcrossing in Mimulus ringens. Ecology 86:762–771.

Bennett, J. A., G. C. Gensler, and J. F. J. Cahill. 2014. Small-scale bee patch use is affected equally by flower availability and local habitat configuration. Basic and Applied Ecology 15:260–268.

Bertness, M. D., and R. Callaway. 1994. Positive interactions in communities. Trends in Ecology and Evolution 9:191–193.

Bingham, R. A., and A. R. Orthner. 1998. Efficient pollination of alpine plants. Nature 391:238.

Bosch, M., and N. M. Waser. 1999. Effects of local density on pollination and reproduction in Delphinium nuttallianum and Aconitum columbianum (Ranunculaceae). American Journal of Botany 86:871–879.

Bouchard, M., and D. Pothier. 2011. Long-term influence of fire and harvesting on boreal forest age structure and forest composition in eastern Québec. Forest Ecology and Management 261:811–820.

Boucher, Y., D. Arseneault, L. Sirois, and L. Blais. 2009. Logging pattern and landscape changes over the last century at the boreal and deciduous forest transition in Eastern Canada. Landscape Ecology 24:171–184. Boufford, D. E. 1997. Eurybia. Page 365 in Flora of North America Editorial Committee. Flora of North America North of . New York and Oxford, New York, New York, USA.

Breed, M. F., K. M. Ottewell, M. G. Gardner, M. H. K. Marklund, E. E. Dormontt, and A. J. Lowe. 2013. Mating patterns and pollinator mobility are critical traits in forest fragmentation genetics. Heredity:1–7. Breitung, A. J. 1988. Distribution of the showy Aster, Aster conspicuus. Canadian Field Naturalist 102:523-526.

101 Brown, B. J., R. J. Mitchell, and S. A. Graham. 2002. Competition for pollination between an invasive species (purple loosestrife) and a native congener. Ecology 83:2328–2336. Burd, M. 1994. Bateman’s principle and plant reproduction: the role of pollen limitation in fruit and seed set. Botanical Review 60:83–139. Burnham, K. P. and D. R. Anderson. 2002. Model selection and multimodel inference. a practical information-theoretic approach. Springer, New York, New York, USA. Burton, C.M. and P. Burton. 2003. A manual for growing and using seed from herbaceous plants native to the interior of northern British Columbia. Symbios Research and Restoration, Smithers, British Columbia, Canada.

Cadotte, M. W., and T. Fukami. 2005. Dispersal, spatial scale, and species diversity in a hierarchically structured experimental landscape. Ecology Letters 8:548–57.

Callaway, R. M. 1995. Positive interactions among plants. Botanical Review 61:306–349.

Callaway, R. M., and L. R. Walker. 1997. Competition and facilitation: a synthetic approach to interactions in plant communities. Ecology 78:1958–1965.

Campbell, D. R. 1985. Pollinator sharing and seed set of Stellaria pubera: competition for pollination. Ecology 66:544–553.

Campbell, D. R., and K. J. Halama. 1993. Resource and pollen limitations to lifetime seed production in a natural plant population. Ecology 74:1043–1051. Caplow, F., 2004. Reintroduction plan for golden paintbrush (Castilleja levisecta). US Fish and Wildlife Service, Western Washington Fish and Wildlife Office, Lacey, Washington, USA. Cariveau, D., R. E. Irwin, A. K. Brody, L. S. Garcia-Mayeya, and A. Von Der Ohe. 2004. Direct and indirect effects of pollinators and seed predators to selection on plant and floral traits. Oikos 104:15–26.

Cartar, R. V. 2004. Resource tracking by bumble bees: responses to plant-level differences in quality. Ecology 85:2764–2771.

Cartar, R. V. 2005. Short-term effects of experimental boreal forest logging disturbance on bumble bees, bumble bee-pollinated flowers and the bee-flower match. Biodiversity and Conservation 14:1895–1907.

Caruso, C. M. 1999. Pollination of Ipomopsis aggregata (Polemoniaceae): effects of intra- vs. interspecific competition. American Journal of Botany 86:663–668.

Cash, D. W., W. N. Adger, F. Berkes, P. Garden, L. Lebel, P. Olsson, L. Pritchard, and O. Young. 2006. Scale and cross-scale dynamics: governance and information in a multilevel world. Ecology and Society 11:8.

102 Chaneton, E. J., and J. M. Facelli. 1991. Disturbance effects on plant community diversity: spatial scales and dominance hierarchies. Vegetatio 93:143–155.

Chapin, F. S., E. S. Zavaleta, V. T. Eviner, R. L. Naylor, P. M. Vitousek, H. L. Reynolds, D. U. Hooper, S. Lavorel, O. E. Sala, S. E. Hobbie, M. C. Mack, and S. Díaz. 2000. Consequences of changing biodiversity. Nature 405:234–242.

Charnov, E. L. 1976. Optimal foraging, the marginal value theorem. Theoretical Population Biology 9:129–136.

Chittka, L., A. Gumbert, and J. Kunze. 1997. Foraging dynamics of bumble bees: correlates of movements within and between plant species. Behavioral Ecology 8:239–249.

Chittka, L., and S. Schürkens. 2001. Successful invasion of a floral market. Nature 411:653.

Choler, P., R. Michalet, and R. M. Callaway. 2001. Facilitation and competition on gradients in alpine plant communities. Ecology 82:3295–3308.

Cibula, D. A., and M. Zimmerman. 1984. The effect the of plant density on departure decisions: testing the marginal value theorem using bumblebees and Delphinium nelsonii. Oikos 43:154–158. Coladonato, M., 1993. Vicia americana in The fire effects information system. Department of Agriculture, Forest Service, Intermountain Research Station, Intermountain Fire Sciences Laboratory, Missoula, Montana, USA.

Colas, B., I. Olivieri, and M. Riba. 2001. Spatio-temporal variation of reproductive success and conservation of the narrow-endemic Centaurea corymbosa (Asteraceae). Biological Conservation 99:375–386.

Cresswell, W. 1998. Relative competitive ability changes with competitor density: evidence from feeding blackbirds. Animal Behaviour 56:1367–1373. Crouch, G. L. 1985. Effects of clearcutting a subalpine forest in central Colorado on wildlife habitat. RM-258. U.S. Department of Agriculture, Forest Service, Rocky Mountain Forest and Range Experiment Station, Fort Collins, Colorado, USA.

Cyr, D., S. Gauthier, Y. Bergeron, and C. Carcaillet. 2009. Forest management is driving the eastern North American boreal forest outside its natural range of variability. Frontiers in Ecology and the Environment 7:519–524.

Darvill, B., J. S. Ellis, G. C. Lye, and D. Goulson. 2006. Population structure and inbreeding in a rare and declining bumblebee, Bombus muscorum (Hymenoptera: Apidae). Molecular Ecology 15:601–611.

Dauber, J., J. C. Biesmeijer, D. Gabriel, W. E. Kunin, E. Lamborn, B. Meyer, A. Nielsen, S. G. Potts, S. P. M. Roberts, V. Sõber, J. Settele, I. Steffan-Dewenter, J. C. Stout, T. Teder, T.

103 Tscheulin, D. Vivarelli, and T. Petanidou. 2010. Effects of patch size and density on flower visitation and seed set of wild plants: a pan-European approach. Journal of Ecology 98:188–196.

Dayton, P. K. 1971. Competition, disturbance, and community organization: the provision and subsequent utilization of space in a rocky intertidal community. Ecological Monographs 41:351–389.

Denny, M. W., B. Helmuth, G. H. Leonard, C. D. G. Harley, L. J. H. Hunt, and E. K. Nelson. 2004. Quantifying scale in ecology: lessons from a wave-swept shore. Ecological Monographs 74:513–532.

Desmet, P., and L. Brouillet. 2013. Database of Vascular Plants of Canada (VASCAN): a community contributed taxonomic checklist of all vascular plants of Canada, Saint Pierre and Miquelon, and Greenland. PhytoKeys 25:55–67.

Diaz-Forero, I., V. Kuusemets, M. Mänd, A. Liivamägi, T. Kaart, and J. Luig. 2013. Influence of local and landscape factors on bumblebees in semi-natural meadows: a multiple-scale study in a forested landscape. Journal of Insect Conservation 17:113–125.

Didion, M., M.-J. Fortin, and A. Fall. 2007. Forest age structure as indicator of boreal forest sustainability under alternative management and fire regimes: a landscape level sensitivity analysis. Ecological Modelling 200:45–58.

Diekötter, T., K. J. Haynes, D. Mazeffa, and T. O. Crist. 2007. Direct and indirect effects of habitat area and matrix composition on species interactions among flower-visiting insects. Oikos 116:1588–1598.

Dietzsch, A. C., D. A. Stanley, and J. C. Stout. 2011. Relative abundance of an invasive alien plant affects native pollination processes. Oecologia 167:469–479.

Drapeau, P., A. Leduc, J. F. Giroux, J. P. L. Savard, Y. Bergeron, and W. L. Vickery. 2000. Landscape-scale disturbance and changes in bird communities of boreal mixed-wood forests. Ecological Monographs 70:423–444.

Ebert, D. 1994. Fractional resource allocation into few eggs: Daphnia as an example. Ecology 75:568–571.

Edmands, S. 2007. Between a rock and a hard place: evaluating the relative risks of inbreeding and outbreeding for conservation and management. Molecular Ecology 16:463–475.

Elliott, S. E. 2009. Subalpine bumble bee foraging distances and densities in relation to flower availability. Environmental Entomology 38:748–756.

104 Elliott, S. E., and R. E. Irwin. 2009. Effects of flowering plant density on pollinator visitation, pollen receipt, and seed production in Delphinium barbeyi (Ranunculaceae). American Journal of Botany 96:912–919.

Essenberg, C. J. 2012. Explaining variation in the effect of floral density on pollinator visitation. The American Naturalist 180:153–166.

Essenberg, C. J. 2013a. Explaining the effects of floral density on flower visitor species composition. The American Naturalist 181:1–21.

Essenberg, C. J. 2013b. Scale-dependent shifts in the species composition of flower visitors with changing floral density. Oecologia 171:187–196.

Ewers, R. M., and R. K. Didham. 2006. Confounding factors in the detection of species responses to habitat fragmentation. Biological Reviews of the Cambridge Philosophical Society 81:117–42. Farmer, A. 2014. Flower visitation and colony success of bumble bees in logged landscapes. MSc Thesis, University of Calgary, Calgary, Alberta, Canada.

Farwig, N., D. Bailey, E. Bochud, J. D. Herrmann, E. Kindler, N. Reusser, C. Schüepp, and M. H. Schmidt-Entling. 2009. Isolation from forest reduces pollination, seed predation and insect scavenging in Swiss farmland. Landscape Ecology 24:919–927.

Fedriani, J. M., T. Wiegand, G. Calvo, A. Suárez-Esteban, M. Jácome, M. Żywiec, and M. Delibes. 2015. Unraveling conflicting density- and distance-dependent effects on plant reproduction using a spatially-explicit approach. Journal of Ecology 103:1344–1353.

Feinsinger, P., H. M. I. Tiebout, and B. E. Young. 1991. Do tropical bird-pollinated plants exhibit density-dependent interactions? Field experiments. Ecology 72:1953–1963.

Feldman, T. S. 2008. The plot thickens: Does low density affect visitation and reproductive success in a perennial herb, and are these effects altered in the presence of a co-flowering species? Oecologia 156:807–817.

Fenster, C. B., W. S. Armbruster, P. Wilson, M. R. Dudash, and J. D. Thomson. 2004. Pollination syndromes and floral specilization. Annual Review of Ecology, Evolution, and Systematics 35:375–403.

Ferreira, P. A., D. Boscolo, and B. F. Viana. 2013. What do we know about the effects of landscape changes on plant–pollinator interaction networks? Ecological Indicators 31:35– 40.

Fischer, J., and D. B. Lindenmayer. 2007. Landscape modification and habitat fragmentation: a synthesis. Global Ecology and Biogeography 16:265–280.

105 Fischer, J., D. B. Lindenmayer, and R. Montague-Drake. 2008. The role of landscape texture in conservation biogeography: a case study on birds in south-eastern Australia. Diversity and Distributions 14:38–46.

Flanagan, R. J., R. J. Mitchell, and J. D. Karron. 2010. Increased relative abundance of an invasive competitor for pollination, Lythrum salicaria, reduces seed number in Mimulus ringens. Oecologia 164:445–454.

Fontaine, C., C. L. Collin, and I. Dajoz. 2008. Generalist foraging of pollinators: diet expansion at high density. Journal of Ecology 96:1002–1010.

Fontúrbel, F. E. 2012. Does habitat degradation cause changes in the composition of arboreal small mammals? A small-scale assessment in Patagonian temperate rainforest fragments. Revista Latinoamericana de Conservación 2:68–72.

Fontúrbel, F. E., A. B. Candia, J. Malebrán, D. A. Salazar, C. González-Browne, and R. Medel. 2015. Meta-analysis of anthropogenic habitat disturbance effects on animal-mediated seed dispersal. Global Change Biology. doi: 10.1111/gcb.13025

Freestone, A. L. 2006. Facilitation drives local abundance and regional distribution of a rare plant in a harsh environment. Ecology 87:2728–2735.

Fretwell, S. D., and H. L. J. Lucas. 1969. On territorial behavior and other factors influencing habitat distribution in birds. Acta Biotheoretica 19:16–36.

García, M. C., M. Y. Bader, and L. A. Cavieres. 2015. Facilitation consequences for reproduction of the benefactor cushion plant Laretia acaulis along an elevational gradient: costs or benefits? Oikos 000:001–009.

Geib, J. C., J. P. Strange, and C. Galen. 2015. Bumble bee nest abundance, foraging distance, and host-plant reproduction: implications for management and conservation. Ecological Applications 25:768–778. Geiger, R. 1965. The climate near the ground. Harvard University Press, Cambridge, Massachusetts, USA. Gerling, H.S., M.G. Willoughby, A. Schoepf, K.E. Tannas and C.A Tannas, 1996. A guide to using native plants on disturbed lands. Alberta Agriculture, Food and Rural Development and Alberta Environmental Protection, Edmonton, Alberta, Canada.

Ghazoul, J. 2005. Pollen and seed dispersal among dispersed plants. Biological Reviews of the Cambridge Philosophical Society 80:413–443.

Ghazoul, J. 2006. Floral diversity and the facilitation of pollination. Journal of Ecology 94:295– 304.

106 Ghazoul, J., K. A. Liston, and T. J. B. Boyle. 1998. Disturbance-induced density-dependent seed set in Shorea siamensis (Dipterocarpaceae), a tropical forest tree. Journal of Ecology 86:462–473. GIMP Development Team. 2013. GNU Image Manipulation Program. University of , Berkeley, California, USA.

Goh, B. S. 1976. Global stability in two species interactions. Journal of Mathematical Biology 3:313–318.

Goldberg, D. E., R. Turkington, L. Olsvig-Whittaker, and A. R. Dyer. 2001. Density dependence in an annual plant community: variation among life history stages. Ecological Monographs 71:423–446.

González-Varo, J. P., J. Arroyo, and A. Aparicio. 2009. Effects of fragmentation on pollinator assemblage, pollen limitation and seed production of Mediterranean myrtle (Myrtus communis). Biological Conservation 142:1058–1065.

Goulson, D. 1999. Foraging strategies of insects for gathering nectar and pollen, and implications for plant ecology and evolution. Perspectives in Plant Ecology, Evolution and Systematics 2:185–209.

Goulson, D. 2000. Why do pollinators visit proportionally fewer flowers in large patches? Oikos 91:485–492.

Goulson, D., and J. L. Osborne. 2009. Foraging range and the spatial distribution of worker bumble bees. Pages 97–111 Food Exploitation By Social Insects: Ecological, Behavioral, and Theoretical Approaches. Taylor & Francis Group, LLC.

Goverde, M., K. Schweizer, B. Baur, and A. Erhardt. 2002. Small-scale habitat fragmentation effects on pollinator behaviour: experimental evidence from the bumblebee Bombus veteranus on calcareous grasslands. Biological Conservation 104:293–299. Green, B. B. 1978. Comparative ecology of Geranium richardsonii and Geranium nervosum. Bulletin of the Torrey Botanical Club 105:108-113.

Greig-Smith, P. 1952. The use of random and contiguous quadrats in the study of the structure of plant communities. Annals of Botany 16:293–316.

Groffman, P. M., J. S. Baron, T. Blett, A. J. Gold, I. Goodman, L. H. Gunderson, B. M. Levinson, M. A. Palmer, H. W. Paerl, G. D. Peterson, N. L. Poff, D. W. Rejeski, J. F. Reynolds, M. G. Turner, K. C. Weathers, and J. Wiens. 2006. Ecological thresholds: the key to successful environmental management or an important concept with no practical application. Ecosystems 9:1–13.

Groom, M. J. 1998. Allee effects limit population viability of an annual plant. The American Naturalist 151:487–496.

107 Gunn, C. R. 1965. The Vicia americana complex (Leguminosae). PhD Thesis, Iowa State University, Ames, Iowa, USA.

Gunton, R. M., and W. E. Kunin. 2009. Density-dependence at multiple scales in experimental and natural plant populations. Journal of Ecology 97:567–580.

Gurevitch, J., J. A. Morrison, and L. V. Hedges. 2000. The interaction between competition and predation: a meta-analysis of field experiments. The American Naturalist 155:435–453.

Hadley, A. S., and M. G. Betts. 2011. The effects of landscape fragmentation on pollination dynamics: absence of evidence not evidence of absence. Biological Reviews 87:526–544. Harper, J.L. 1977. Population biology of plants. Academic Press, New York, New York, USA.

Harper, K. A., and S. E. Macdonald. 2002. Structure and composition of edges next to regenerating clear-cuts in mixed-wood boreal forest. Journal of Vegetation Science 13:535–546.

Heffernan, J. B., P. A. Soranno, M. J. Angilletta, L. B. Buckley, D. S. Gruner, T. H. Keitt, J. R. Kellner, J. S. Kominoski, A. V. Rocha, J. Xiao, T. K. Harms, S. J. Goring, L. E. Koenig, W. H. McDowell, H. Powell, A. D. Richardson, C. A. Stow, R. Vargas, and K. C. Weathers. 2014. Macrosystems ecology: Understanding ecological patterns and processes at continental scales. Frontiers in Ecology and the Environment 12:5–14.

Hegland, S. J. 2014. Floral neighbourhood effects on pollination success in red clover are scale- dependent. Functional Ecology 28:561–568.

Hegland, S. J., J.-A. Grytnes, and Ø. Totland. 2009. The relative importance of positive and negative interactions for pollinator attraction in a plant community. Ecological Research 24:929–936.

Hegland, S. J., and Ø. Totland. 2005. Relationships between species’ floral traits and pollinator visitation in a temperate grassland. Oecologia 145:586–594.

Heinken, T., and E. Weber. 2013. Consequences of habitat fragmentation for plant species: Do we know enough? Perspectives in Plant Ecology, Evolution and Systematics 15:205–216.

Heinrich, B. 1979. Resource heterogeneity and patterns of movement in foraging bumblebees. Oecologia 40:235–245.

Heithecker, T. D., and C. B. Halpern. 2006. Variation in microclimate associated with dispersed- retention harvests in coniferous forests of western Washington. Forest Ecology and Management 226:60–71.

Herrera, C. M. 1987. Components of pollinator “quality”: comparative analysis of a diverse insect assemblage. Oikos 50:79–90.

108 Hersch, E. I., and B. A. Roy. 2007. Context-dependent pollinator behavior: an explanation for patterns of hybridization among three species of Indian paintbrush. Evolution 61:111–124.

Hobbs, R. J., and C. J. Yates. 2003. Impacts of ecosystem fragmentation on plant populations: generalising the idiosyncratic. Australian Journal of Botany 51:471–488.

Hoehn, P., T. Tscharntke, J. M. Tylianakis, and I. Steffan-Dewenter. 2008. Functional group diversity of bee pollinators increases crop yield. Proceedings of the Royal Society B 275:2283–2291.

Holmgren, M., M. Scheffer, and M. A. Huston. 1997. The interplay of facilitation and competition in plant communities. Ecology 78:1966–1975. Howard, J.L. 1993. Linnaea borealis in The fire effects information system. United States Department of Agriculture, Forest Service, Intermountain Research Station, Intermountain Fire Sciences Laboratory, Missoula, Montana, USA.

Inouye, D. W. 1980. The effect of proboscis and corolla tube lengths on patterns and rates of flower visitation by bumblebees. Oecologia 45:197–201.

Jakobsson, A., A. Lázaro, and Ø. Totland. 2009. Relationships between the floral neighborhood and individual pollen limitation in two self-incompatible herbs. Oecologia 160:707–719.

Janovský, Z., M. Mikát, J. Hadrava, E. Horčičková, K. Kmecová, D. Požárová, J. Smyčka, and T. Herben. 2013. Conspecific and heterospecific plant densities at small-scale can drive plant-pollinator interactions. PloS One 8:1–11.

Johnson, S. D., C. I. Peter, L. A. Nilsson, and J. Ågren. 2003. Pollination success in a deceptive orchid is enhanced by co-occurring rewarding magnet plants. Ecology 84:2919–2927. Jones, G. N. and F. F. Jones. 1943. A revision of the perennial species of Geranium of the United States and Canada. Rhodora. 45:5-26.

Jones, A. G. 1978. Observations on reproduction and phenology in some perennial asters. The American Midland Naturalist 99:184–197.

Jules, E. S., and B. J. Rathcke. 1999. Mechanisms of reduced Trillium recruitment along edges of old-growth forest fragments. Conservation Biology 13:784–793.

Kao, R. H. 2007. Asexuality and the coexistence of cytotypes. New Phytologist 175:764–772.

Kao, R. H. 2008. Origins and widespread distribution of co-existing polyploids in Arnica cordifolia (Asteraceae). Annals of Botany 101:145–152. Kearns, C. A. 1990. The role of fly pollination in montane habitats. Ph.D. dissertation, University of Maryland, College Park, Maryland, USA.

109 Kearns, C. A., and D. W. Inouye. 1997. Pollinators, flowering plants, and conservation biology: Much remains to be learned about pollinators and plants. BioScience 47:297–307.

Kearns, C. A., D. W. Inouye, and N. M. Waser. 1998. Endangered mutualisms: the conservation of plant-pollinator interactions. Annual Review of Ecology and Systematics 29:83–112.

Kennedy, C. M., E. Lonsdorf, M. C. Neel, N. M. Williams, T. H. Ricketts, R. Winfree, R. Bommarco, C. Brittain, A. L. Burley, D. Cariveau, L. G. Carvalheiro, N. P. Chacoff, S. A. Cunningham, B. N. Danforth, J.-H. Dudenhöffer, E. Elle, H. R. Gaines, L. A. Garibaldi, C. Gratton, A. Holzschuh, R. Isaacs, S. K. Javorek, S. Jha, A. M. Klein, K. Krewenka, Y. Mandelik, M. M. Mayfield, L. Morandin, L. A. Neame, M. Otieno, M. Park, S. G. Potts, M. Rundlöf, A. Saez, I. Steffan-Dewenter, H. Taki, B. F. Viana, C. Westphal, J. K. Wilson, S. S. Greenleaf, and C. Kremen. 2013. A global quantitative synthesis of local and landscape effects on wild bee pollinators in agroecosystems. Ecology Letters 16:584–599.

Kevan, P. G., and H. G. Baker. 1983. Insects as flower visitors and pollinators. Annual Review of Entomology 28:407–453.

Klein, A.-M., I. Steffan-Dewenter, and T. Tscharntke. 2003. Fruit set of highland coffee increases with the diversity of pollinating bees. Proceedings Of The Royal Society B 270:955–961.

Klinkhamer, P. G. L., and T. J. de Jong. 1990. Effects of plant size, plant density and sex differential nectar reward on pollinator visitation in the protandrous Echium vulgare (Boraginaceae). Oikos 57:399–405.

Knapp, A. K., W. K. Smith, and D. R. Young. 1989. Importance of intermittent shade to the of subalpine herbs ecophysiology. Functional Ecology 3:753–758.

Knight, T. M. 2003. Floral density, pollen limitation, and reproductive success in Trillium grandiflorum. Oecologia 137:557–563.

Knight, T. M., J. A. Steets, J. C. Vamosi, S. J. Mazer, M. Burd, D. R. Campbell, M. R. Dudash, M. O. Johnston, R. J. Mitchell, and T.-L. Ashman. 2005. Pollen limitation of plant reproduction: pattern and process. Annual Review of Ecology, Evolution, and Systematics 36:467–497.

Kreyling, J., A. Schmiedinger, E. Macdonald, and C. Beierkuhnlein. 2008. Slow understory redevelopment after clearcutting in high mountain forests. Biodiversity and Conservation 17:2339–2355.

Kunin, W. E. 1992. Density and reproductive success in wild populations of Diplotaxis erucoides (Brassicaceae). Oecologia 91:129–133.

Kunin, W. E. 1993. Sex and the single mustard: population density and pollinator behavior effects on seed-set. Ecology 74:2145–2160.

110 Kunin, W. E. 1997. Population size and density effects in pollination: pollinator foraging and plant reproductive success in experimental arrays of Brassica kaber. Journal of Ecology 85:225–234.

Kunin, W., and Y. Iwasa. 1996. Pollinator foraging strategies in mixed floral arrays: density effects and floral constancy. Theoretical Population Biology 49:232–263.

Larson, B. M. H., and S. C. H. Barrett. 2000. A comparative analysis of pollen limitation in flowering plants. Biological Journal of the Linnean Society 69:503–520.

Laverty, T. M. 1980. The flower-visiting behaviour of bumble bees: floral complexity and learning. Canadian Journal of Zoology 58:1324–1335.

Laverty, T. M. 1992. Plant interactions for pollinator visits: a test of the magnet species effect. Oecologia 89:502–508.

Lázaro, A., A. Jakobsson, and O. Totland. 2013. How do pollinator visitation rate and seed set relate to species’ floral traits and community context? Oecologia. doi: 10.1007/s00442- 013-2652-5

Lázaro, A., and Ø. Totland. 2010. Population dependence in the interactions with neighbors for pollination: A field experiment with Taraxacum officinale. American Journal of Botany 97:760–769.

Lechowicz, M. J., and G. Bell. 1991. The ecology and genetics of fitness in forest plants. II. Microspatial heterogeneity of the edaphic environment. Journal of Ecology 79:687–696.

Lennartsson, T. 2002. Extinction thresholds and disrupted plant-pollinator interactions in fragmented plant populations. Ecology 83:3060–3072. Farmer, A. 2014. Flower visitation and colony success of bumble bees in logged landscapes. MSc Thesis, University of Calgary, Calgary, Alberta, Canada.

Lertzman, K. P. 1981. Pollen transfer: processes and consequences. MSc Thesis, University of British Columbia, Vancouver, British Columbia, Canada.

Levin, D. A., and W. W. Anderson. 1970. Competition for pollinators between simulataneously flowering species. The American Naturalist 104:455–467.

Levin, S. A. 1992. The problem of pattern and scale in ecology: the Robert H. MacArthur Award lecture. Ecology 73:1943–1967.

Liao, K., R. W. Gituru, Y. H. Guo, and Q. F. Wang. 2011. The presence of co-flowering species facilitates reproductive success of Pedicularis monbeigiana (Orobanchaceae) through variation in bumble-bee foraging behaviour. Annals of Botany 108:877–884.

111 Lienert, J. 2004. Habitat fragmentation effects on fitness of plant populations – a review. Journal for Nature Conservation 12:53–72.

Lindborg, R., S. A. O. Cousins, and O. Eriksson. 2005. Plant species response to land use change - Campanula rotundifolia, Primula veris and Rhinanthus minor. Ecography 28:29–36.

Lindborg, R., and O. Eriksson. 2004. Effects of restoration on plant species richness and composition in Scandinavian semi-natural grasslands. Restoration Ecology 12:318–326.

Lortie, C. J., and R. M. Callaway. 2006. Re-analysis of meta-analysis: support for the stress- gradient hypothesis. Journal of Ecology 94:7–16.

MacArthur, R. H., and E. R. Pianka. 1966. On optimal use of a patchy environment. American Society of Naturalists 100:603–609.

Maestre, F. T., F. Valladares, and J. F. Reynolds. 2005. Is the change of plant-plant interactions with abiotic stress predictable? A meta-analysis of field results in arid environments. Journal of Ecology 93:748–757.

Maestre, F. T., F. Valladares, and J. F. Reynolds. 2006. The stress-gradient hypothesis does not fit all relationships between plant-plant interactions and abiotic stress: further insights from arid environments. Journal of Ecology 94:17–22.

Marini, L., M. Scotton, S. Klimek, J. Isselstein, and A. Pecile. 2007. Effects of local factors on plant species richness and composition of Alpine meadows. Agriculture, Ecosystems and Environment 119:281–288.

Marini, L., M. Scotton, S. Klimek, and A. Pecile. 2008. Patterns of plant species richness in Alpine hay meadows: Local vs. landscape controls. Basic and Applied Ecology 9:365–372.

Maron, J. L., K. C. Baer, and A. L. Angert. 2014. Disentangling the drivers of context-dependent plant-animal interactions. Journal of Ecology. 102:1485-1496.

Matthies, D. 1997. Parasite-host interactions in Castilleja and Orthocarpus. Canadian Journal of Botany 75:1252–1260.

McDowell, N. G., N. C. Coops, P. S. A. Beck, J. Q. Chambers, C. Gangodagamage, J. A. Hicke, C. Huang, R. Kennedy, D. J. Krofcheck, M. Litvak, A. J. H. Meddens, J. Muss, R. Negrón- Juarez, C. Peng, A. M. Schwantes, J. J. Swenson, L. J. Vernon, A. P. Williams, C. Xu, M. Zhao, S. W. Running, and C. D. Allen. 2015. Global satellite monitoring of climate- induced vegetation disturbances. Trends in Plant Science 20:114–123.

McGarigal, K., and S. A. Cushman. 2002. Comparative evaluation of experimental approaches to the study of habitat fragmentation effects. Ecological Applications 12:335–345.

112 McIntyre, S., and R. Hobbs. 1999. A framework for conceptualizing human effects on landscapes and its relevance for management and research models. Conservation Biology 13:1282–1292.

McKechnie, I. M., and R. D. Sargent. 2013. Do plant traits influence a species’ response to habitat disturbance? A meta-analysis. Biological Conservation 168:69–77.

McLernon, S. M., S. D. Murphy, and L. W. Aarssen. 1996. Heterospecific pollen transfer between sympatric species in a midsuccessional old-field community. American Journal of Botany 83:1168–1174.

McRae, D. J., L. C. Duchesne, B. Freedman, T. J. Lynham, and S. Woodley. 2001. Comparisons between wildfire and forest harvesting and their implications in forest management. Environmental Reviews 9:223–260.

Metzger, J.-P., and H. Décamps. 1997. The structural connectivity threshold: an hypothesis in conservation biology at the landscape scale. Acta Oecologia 18:1-12.

Mitchell, R. J., R. J. Flanagan, B. J. Brown, N. M. Waser, and J. D. Karron. 2009. New frontiers in competition for pollination. Annals of Botany 103:1403–13.

Moeller, D. A. 2004. Facilitative interactions among plants via shared pollinators. Ecology 85:3289–3301.

Molina-Montenegro, M. A., E. I. Badano, and L. A. Cavieres. 2008. Positive interactions among plant species for pollinator service: assessing the “magnet species” concept with invasive species. Oikos 117:1833–1839.

Morris, W. F. 1996. Mutualism denied? Nectar-robbing bumble bees do not reduce female or male success of bluebells. Ecology 77:1451–1462.

Morris, W. F., R. A. Hufbauer, A. A. Agrawal, J. D. Bever, V. A. Borowicz, G. S. Gilbert, J. L. Maron, C. E. Mitchell, I. M. Parker, A. G. Power, M. E. Torchin, and D. P. Vaźquez. 2007. Direct and interactive effects of enemies and mutualists on plant performance: A meta- analysis. Ecology 88:1021–1029.

Morris, W. F., D. P. Vazquez, and N. P. Chacoff. 2010. Benefit and cost curves for typical pollination mutualisms. Ecology 91:1276–1285. Moss, E. H. 1983. Flora of Alberta. University of Toronto Press, Toronto, Ontario, Canada.

Muñoz, A. A., and L. A. Cavieres. 2008. The presence of a showy invasive plant disrupts pollinator service and reproductive output in native alpine species only at high densities. Journal of Ecology 96:459–467. Murcia, C. 1996. Forest fragmentation and the pollination of neotropical plants. Pages 19-33 in Forest Patches in Tropical Landscapes. Island Press, London, England, UK.

113 Mustajarvi, K., P. Siikamaki, S. Rytkonen, and A. Lammi. 2001. Consequences of plant population size and density for plant–pollinator interactions and plant performance. Journal of Ecology 89:80–87.

Nattero, J., R. Malerba, R. Medel, and A. Cocucci. 2011. Factors affecting pollinator movement and plant fitness in a specialized pollination system. Plant Systematics and Evolution 296:77–85.

Nayak, G. K., S. P. M. Roberts, M. Garratt, T. D. Breeze, T. Tscheulin, J. Harrison-Cripps, I. N. Vogiatzakis, M. T. Stirpe, and S. G. Potts. 2014. Interactive effect of floral abundance and semi-natural habitats on pollinators in field beans (Vicia faba). Agriculture, Ecosystems & Environment 199:58–66.

Neal, P. R., A. Dafni, and M. Giurfa. 1998. Floral symmetry and its role in plant-pollinator systems: terminology, distribution, and hypotheses. Annual Review of Ecology and Systematics 29:345–373.

Neiland, M. R. M., and C. C. Wilcock. 1999. The presence of heterospecific pollen on stigmas of nectariferous and nectarless orchids and its consequences for their reproductive success. Protoplasma 208:65–75.

Neill, W. E. 1974. The community matrix and interdependence of the competition coefficients. The American Naturalist 108:399–408.

Niemelä, J. 1999. Management in relation to disturbance in the boreal forest. Forest Ecology and Management 115:127–134.

Nuortila, C., M.-M. Kytoviita, and J. Tuomi. 2004. Mycorrhizal symbiosis has contrasting effects on fitness components in Campanula rotundifolia. New Phytologist 164:543–553.

Nyman, Y. 1992. Pollination mechanisms in six Campanula species (Campanulaceae). Plant Systematics and Evolution 181:97–108.

O’Neill, R. V., A. R. Johnson, and A. W. King. 1989. A hierarchical framework for the analysis of scale. Landscape Ecology 3:193–205.

Oatway, M. L., and D. W. Morris. 2007. Do animals select habitat at small or large scales? An experiment with meadow voles (Microtus pennsylvanicus). Canadian Journal of Zoology 85:479–487.

Ollerton, J., R. Winfree, and S. Tarrant. 2011. How many flowering plants are pollinated by animals? Oikos 120:321–326.

Osborne, J. L., S. J. Clark, R. J. Morris, I. H. Williams, J. R. Riley, A. D. Smith, D. R. Reynolds, and A. S. Edwards. 1999. A landscape-scale study of bumble bee foraging range and constancy, using harmonic radar. Journal of Applied Ecology 36:519–533.

114 Osborne, J. L., A. P. Martin, N. L. Carreck, J. L. Swain, M. E. Knight, D. Goulson, R. J. Hale, and R. A. Sanderson. 2008. Bumblebee flight distances in relation to the forage landscape. Journal of Animal Ecology 77:406–415.

Osborne, J. L., and I. H. Williams. 2001. Site constancy of bumble bees in an experimentally patchy habitat. Agriculture, Ecosystems and Environment 83:129–141.

Ozinga, W. A., R. M. Bekker, J. H. J. Schaminee, and J. M. Van Groenendael. 2004. Dispersal potential in plant communities depends on environmental conditions. Journal of Ecology 92:767–777. Pahl, M. and Smreciu, A. 1999. Growing native plants of western Canada: common grasses and wildflowers. Alberta Agriculture, Food and Rural Development and Alberta Research Council, Edmonton, Alberta, Canada.

Peng, D.-L., Z.-Q. Zhang, B. Xu, Z.-M. Li, and H. Sun. 2012. Patterns of flower morphology and sexual systems in the subnival belt of the Hengduan Mountains, SW China. Alpine Botany 122:65–73.

Pengelly, C. J., and R. V. Cartar. 2010. Effects of variable retention logging in the boreal forest on the bumble bee-influenced pollination community, evaluated 8–9 years post-logging. Forest Ecology and Management 260:994–1002.

Peters, D. P. C., B. T. Bestelmeyer, and M. G. Turner. 2007. Cross-scale interactions and changing pattern-process relationships: Consequences for system dynamics. Ecosystems 10:790–796. Pickett, S. T. A., and P. S. White. 1985. The ecology of natural disturbance as patch dynamics. Academic Press, New York, New York, USA.

Pugnaire, F. I., and M. T. Luque. 2001. Changes in plant interactions along a gradient of environmental stress. Oikos 93:42–49. R Development Core Team. 2014. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

Rao, S., and J. P. Strange. 2012. Bumble bee (Hymenoptera: Apidae) foraging distance and colony density associated with a late-season mass flowering crop. Environmental Entomology 41:905–915.

Rathcke, B. 1983. Competition and facilitation among plants for pollination. Pages 305–329 in Pollination Biology. Academic Press, Inc., Ann Arbor, Michigan.

Rathcke, B. 1988. Interactions for pollination among coflowering . Ecology 69:446–457.

Rathcke, B. J., and E. S. Jules. 1993. Habitat fragmentation and plant-pollinator interactions. Current Science 65:273–277.

115 Rees, D. C., and G. P. Juday. 2002. Plant species diversity on logged versus burned sites in central . Forest Ecology and Management 155:291–302. Reeves, S.L. 2006. Mertensia paniculata in The fire effects information system. United States Department of Agriculture, Forest Service, Intermountain Research Station, Intermountain Fire Sciences Laboratory, Missoula, Montana, USA.

Ricklefs, R. E. 1987. Community diversity: relative roles of local and regional processes. Science 235:167–171.

Robertson, A. W., and M. R. MacNair. 1995. The effects of floral display size on pollinator service to individual flowers of Myosotis and Mimulus. Oikos 72:106–114.

Robertson, B. A., and R. L. Hutto. 2007. Is selectively harvested forest an ecological trap for Olive-sided Flycatchers? The Condor 109:109–121.

Robertson, B. A., R. L. Hutto, and L. Hutto. 2006. A framework for understanding ecological traps and an evaluation of existing evidence. Ecology 87:1075–1085.

Rodríguez, I., A. Gumbert, N. H. De Ibarra, J. Kunze, and M. Giurfa. 2004. Symmetry is in the eye of the “beeholder”: innate preference for bilateral symmetry in flower-naïve bumblebees. Naturwissenschaften 91:374–377.

Roll, J., R. J. Mitchell, R. J. Cabin, and D. L. Marshall. 1997. Reproductive success increases with local density of conspecifics in a desert mustard (Lesquerella fendleri). Conservation Biology 11:738–746.

Romey, W. L., J. S. Ascher, D. A. Powell, and M. Yanek. 2007. Impacts of logging on midummer diversity of native bees (Apoidea) in a northern hardwood forest. Journal of the Kansas Entomological Society 80:327–338. Royer, F. and R. Dickinson, 1996. Harebell Campanula rotundifolia L. Page 43 in Wild Flowers of Edmonton and Central Alberta. The University of Alberta Press, Edmonton, Alberta, Canada.

Saleh, N., and L. Chittka. 2007. Traplining in bumblebees (Bombus impatiens): A foraging strategy’s ontogeny and the importance of spatial reference memory in short-range foraging. Oecologia 151:719–730.

Sandel, B., and A. B. Smith. 2009. Scale as a lurking factor: incorporating scale-dependence in experimental ecology. Oikos 118:1284–1291.

Sargent, R. D. 2004. Floral symmetry affects speciation rates in angiosperms. Proceedings of the Royal Society B 271:603–608.

Sargent, R. D., and D. D. Ackerly. 2008. Plant-pollinator interactions and the assembly of plant communities. Trends in Ecology and Evolution 23:123–130.

116 Sargent, R. D., and S. P. Otto. 2006. The role of local species abundance in the evolution of pollinator attraction in flowering plants. The American Naturalist 167:67–80.

Saunders, D. A., R. J. Hobbs, and C. R. Margules. 1991. Biological consequences of ecosystem fragmentation: A review. Conservation Biology 5:18–32.

Schellenberg, M. P., and M. R. Banerjee. 2002. The potential of legume- mixtures for optimum forage production in southwestern Saskatchewan: A greenhouse study. Canadian Journal of Plant Science 82:357–363.

Schemske, D. W. 1981. Floral convergence and pollinator sharing in two bee-pollinated tropical herbs. Ecology 62:946–954.

Schmid, B., H. Nottebrock, K. J. Esler, J. Pagel, A. Pauw, K. Bohning-Gaese, F. M. Schurr, and M. Schleuning. 2015. Responses of nectar-feeding birds to floral resources at multiple spatial scales. Ecography 38:001-011.

Schmiedinger, A., J. Kreyling, M. J. Steinbauer, S. E. Macdonald, A. Jentsch, and C. Beierkuhnlein. 2012. A continental comparison indicates long-term effects of forest management on understory diversity in coniferous forests. Canadian Journal of Forest Research 42:1239–1252.

Schüepp, C., F. Herzog, and M. H. Entling. 2014. Disentangling multiple drivers of pollination in a landscape-scale experiment. Proceedings Of The Royal Society B 281:1–8.

Schweiger, E. W., J. E. Diffendorfer, R. D. Holt, R. Pierotti, and M. S. Gaines. 2000. The interaction of habitat fragmentation, plant, and small mammal succession in an old field. Ecological Monographs 70:383–400.

Scobie, A. R., and C. C. Wilcock. 2009. Limited mate availability decreases reproductive success of fragmented populations of Linnaea borealis, a rare, clonal self-incompatible plant. Annals of Botany 103:835–846.

Sheehan, H., K. Hermann, and C. Kuhlemeier. 2012. Color and scent: How single genes influence pollinator attraction. Cold Spring Harbor Symposia on Quantitative Biology 77:117–133. Shelter, S.G. and N.R. Morin. 1986. Seed morphology in North American Campanulaceae. Annals of the Missouri Botanical Garden 73:653- 688.

Shepherd, T. D., and M. K. Litvak. 2004. Density-dependent habitat selection and the ideal free distribution in marine fish spatial dynamics: considerations and cautions. Fish and Fisheries 5:141–152.

Sih, A., and M.-S. Baltus. 1987. Patch size, pollinator behavior, and pollinator limitation in catnip. Ecology 68:1679–1690.

117 Silander, J. A. J. 1978. Density-dependent control of reproductive success in Cassia biflora. Biotropica 10:292–296.

Smithson, A., and M. R. Macnair. 1996. Frequency-dependent selection by pollinators: mechanisms and consequences with regard to behaviour of bumblebees Bombus terrestris (L.) (Hymenoptera: Apidae). Journal of Evolutionary Biology 9:571–588.

Smithson, A., and M. R. Macnair. 1997. Density-dependent and frequency-dependent selection by bumblebees Bombus terrestris (L.) (Hymenoptera: Apidae). Biological Journal of the Linnean Society 60:401–417.

Solbrig, O. T. 1971. The population biology of dandelions: these common weeds provide experimental evidence for a new model to explain the distribution of plants. American Scientist 59:686–694.

Soliveres, S., C. Smit, and F. T. Maestre. 2015. Moving forward on facilitation research: Response to changing environments and effects on the diversity, functioning and evolution of plant communities. Biological Reviews 90:297–313.

Soranno, P. A., K. S. Cheruvelil, E. G. Bissell, M. T. Bremigan, J. A. Downing, C. E. Fergus, C. T. Filstrup, E. N. Henry, N. R. Lottig, E. H. Stanley, C. A. Stow, P.-N. Tan, T. Wagner, and K. E. Webster. 2014. Cross-scale interactions: quantifying multi-scaled cause–effect relationships in macrosystems. Frontiers in Ecology and the Environment 12:65–73.

Sowig, P. 1989. Effects of flowering plant’s patch size on species composition of pollinator communities, foraging strategies, and resource partitioning in bumblebees (Hymenoptera: Apidae). Oecologia 78:550–558.

Spigler, R. B., and S. M. Chang. 2008. Effects of plant abundance on reproductive success in the biennial Sabatia angularis (Gentianaceae): spatial scale matters. Journal of Ecology 96:323–333.

Steffan-Dewenter, I., U. Munzenberg, C. Burger, C. Thies, and T. Tscharntke. 2002. Scale- dependent effects of landscape context on three pollinator guilds. Ecology 83:1421–1432.

Stockwell, C. A., A. P. Hendry, and M. T. Kinnison. 2003. Contemporary evolution meets conservation biology. Trends in Ecology and Evolution 18:94–101.

Strauss, S. Y., and R. E. Irwin. 2004. Ecological and evolutionary consequences of multispecies plant-animal interactions. Annual Review of Ecology, Evolution, and Systematics 35:435– 466.

Suding, K. N., and D. Goldberg. 2001. Do disturbances alter competitive hierarchies? Mechanisms of change following gap creation. Ecology 82:2133–2149. Tannas, K., 1997. Grasses, grass-like species, trees and shrubs. Volume 1 in Common plants of the western rangelands. Lethbridge Community College, Lethbridge, Alberta, Canada.

118 Tannas, K., 2004. Forbs. Volume 3 in Common plants of the western rangelands. Olds College, Olds, Alberta and Alberta Agriculture, Food and Rural Development, Edmonton, Alberta, Canada.

Thomson, J. D., S. C. Peterson, and L. D. Harder. 1987. Response of traplining bumble bees to competition experiments: shifts in feeding location and efficiency. Oecologia 71:295–300.

Thomson, J. D., M. Slatkin, and B. A. Thomson. 1997. Trapline foraging by bumble bees: II. Definition and detection from sequence data. Behavioral Ecology 8:199–210.

Toft, C. A. 1983. Community patterns of nectivorous adult parasitoids (Diptera, Bombyliidae) on their resources. Oecologia 57:200–215.

Tokuda, N., M. Hattori, K. Abe, Y. Shinohara, Y. Nagano, and T. Itino. 2015. Demonstration of pollinator-mediated competition between two native Impatiens species, Impatiens noli- tangere and I. textori (Balsaminaceae). Ecology and Evolution 5:1271–1277.

Van Treuren, R., R. Bijlsma, N. J. Ouborg, and M. M. Kwak. 1994. Relationships between plant density, outcrossing rates and seed set in natural and experimental populations of Scabiosa columbaria. Journal of Evolutionary Biology 7:287–302.

Tscharntke, T., and R. Brandl. 2004. Plant-insect interactions in fragmented landscapes. Annual Review of Entomology 49:405–430.

Turkington, R., E. John, C. J. Krebs, M. R. T. Dale, V. O. Nams, R. Boonstra, S. Boutin, K. Martin, A. R. E. Sinclair, and J. N. M. Smith. 1998. The effects of NPK fertilisation for nine years on boreal forest vegetation in northwestern Canada. Journal of Vegetation Science 9:333–346.

Turner, M. G. 1989. Landscape ecology: the effect of pattern on process. Annual Review of Ecology and Systematics 20:171–197.

Turner, M. G. 2005. Landscape ecology: what is the state of the science? Annual Review of Ecology, Evolution, and Systematics 36:319–344.

Underwood, N., B. D. Inouye, and P. A. Hambäck. 2014. A conceptual framework for associational effects: when do neighbours matter and how would we know? The Quarterly Review of Biology 89:1–19.

Ushimaru, A., and F. Hyodo. 2005. Why do bilaterally symmetrical flowers orient vertically? Flower orientation influences pollinator landing behaviour. Evolutionary Ecology Research 7:151–160.

Vergara, P. M., and J. J. Armesto. 2009. Responses of Chilean forest birds to anthropogenic habitat fragmentation across spatial scales. Landscape Ecology 24:25–38.

119 Vergara, P. M., C. Smith, C. A. Delpiano, I. Orellana, D. Gho, and I. Vazquez. 2010. Frugivory on Persea lingue in temperate Chilean forests: interactions between fruit availability and habitat fragmentation across multiple spatial scales. Oecologia 164:981–991.

Wagenius, S., and S. P. Lyon. 2010. Reproduction of Echinacea angustifolia in fragmented prairie is pollen-limited but not pollinator-limited. Ecology 91:733–742.

Waser, N. M. 1978. Interspecific pollen transfer and competition between co-occurring plant species. Oecologia 36:223–236.

Waser, N. M., and M. V Price. 1994. Crossing-distance effects in Delphinium nelsonii: Outbreeding and inbreeding depression in progeny fitness. Evolution 48:842–852.

Watkinson, A. R. 1980. Density-dependence in single-species populations of plants. Journal of Theoretical Biology 83:345–357. Weber, W. A. 1972. Rocky mountain flora. Colorado Associated University Press, Boulder. Colorado, USA.

Weiner, C. N., M. Werner, K. E. Linsenmair, and N. Blüthgen. 2014. Land-use impacts on plant- pollinator networks: interaction strength and specialization predict pollinator declines. Ecology 95:466–474.

Westphal, C., I. Steffan-Dewenter, and T. Tscharntke. 2006. Bumblebees experience landscapes at different spatial scales: possible implications for coexistence. Oecologia 149:289–300.

Wheeler, J. A., F. Schnider, J. Sedlacek, A. J. Cortés, S. Wipf, G. Hoch, and C. Rixen. 2015. With a little help from my friends: community facilitation increases performance in the dwarf shrub Salix herbacea. Basic and Applied Ecology 16:202–209.

Wiens, J. A. 1989. Spatial scaling in ecology. Functional Ecology 3:385–397.

Wiens, J. A., and B. T. Milne. 1989. Scaling of “landscapes” in landscape ecology, or, landscape ecology from a beetle’s perspective. Landscape Ecology 3:87–96.

Wilcock, C. C., and S. B. Jennings. 1999. Partner limitation and restoration of sexual reproduction in the clonal dwarf shrub Linnaea borealis L. (Caprifoliaceae). Protoplasma 208:76–86.

Wilcock, C., and R. Neiland. 2002. Pollination failure in plants: why it happens and when it matters. Trends in Plant Science 7:270–275.

Williams, C. F., M. A. Kuchenreuther, and A. Drew. 2000. Floral dimorphism, pollination, and self-fertilization in gynodioecious Geranium richardsonii (Geranieacae). American Journal of Botany 87:661–669.

120 Williams, N. M., and J. D. Thomson. 1998. Trapline foraging by bumble bees: III. Temporal patterns of visitation and foraging success at single plants. Behavioral Ecology 9:612–621.

Williams, N. M., and R. Winfree. 2013. Local habitat characteristics but not landscape urbanization drive pollinator visitation and native plant pollination in forest remnants. Biological Conservation 160:10–18.

Wilson, J. D., A. J. Morris, B. E. Arroyo, S. C. Clark, and R. B. Bradbury. 1999. A review of the abundance and diversity of invertebrate and plant foods of granivorous birds in northern Europe in relation to agricultural change. Agriculture, Ecosystems & Environment 75:13– 30.

Winfree, R., R. Aguilar, D. P. Vazquez, G. LeBuhn, and M. A. Aizen. 2009. A meta-analysis of bees’ responses to anthropogenic disturbance. Ecology 90:2068–2076. Wood, S. N. 2006. Generalized additive models: An introduction with R. Chapman and Hall/CRC, Boca Raton, Florida, USA.

Wootton, J. T. 1993. Indirect effects and habitat use in an intertidal community: interaction chains and interaction modifications. The American Naturalist 141:71–89.

Wright, J. P., A. S. Flecker, and C. G. Jones. 2003. Local vs. landscape controls on plant species richness in beaver meadows. Ecology 84:3162–3173.

Wu, J., and R. Hobbs. 2002. Key issues and research priorities in landscape ecology: An idiosyncratic synthesis. Landscape Ecology 17:355–365.

Xi, X., J. Mu, Y. Peng, N. Eisenhauer, and S. Sun. 2015. Capitulum density-dependent effects generate peak seed yield at an intermediate density of a Tibetan lotus. Journal of Plant Ecology:1–7.

Yang, S., M. J. Ferrari, and K. Shea. 2011. Pollinator behavior mediates negative interactions between two congeneric invasive plant species. The American Naturalist 177:110–118.

Ye, Z.-M., W.-K. Dai, X.-F. Jin, R. W. Gituru, Q.-F. Wang, and C.-F. Yang. 2014. Competition and facilitation among plants for pollination: can pollinator abundance shift the plant–plant interactions? Plant Ecology 215:3–13.

Young, D. R. 1983. Comparison of intraspecific variations in the reproduction and photosynthesis of an understory herb, Arnica cordifolia. American Journal of Botany 70:728–734.

Zimmerman, J. K., and T. M. Aide. 1989. Patterns of fruit production in a neotropical orchid: pollinator vs. resource limitation. American Journal of Botany 76:67–73.

Zimmerman, M. 1983. Plant reproduction and optimal foraging: experimental nectar manipulations in Delphinium nelsonii. Oikos 41:57–63.

121 Zimmerman, M., and G. H. Pyke. 1988. Reproduction in Polemonium: assessing the factors limiting seed set. The American Naturalist 131:723–738.

122 APPENDIX A Plant species identified as bee-pollinated forbs and counted as heterospecifics for transects and local floral neighbourhoods at all sites where they were present. Flower size is reported, as all density counts were adjusted by maximum straight-line measurement (whether it be height, depth, width, length) per flower, taken as the median value of ranges reported in Moss (1983). The subset of the most common species sampled for seed size and seed number are listed first (“focal species”).

Species Maximum floral measurement Flower size adjustment Focal species Arnica cordifolia 2-7 cm across 5 Eurybia conspicua 3-4 cm wide 3.5 Campanula rotundifolia 1.5-2.5 cm deep 2 Castilleja miniata 2-3.5 cm deep 2.75 Geranium richardsonii 2-4 cm across 3 Lathyrus ochroleucus 1.2-1.5 cm long 1.35 Linnaea borealis 0.8-1.5 cm deep 1.15 Mertensia paniculata 0.8-1.4 cm deep 1.1 Vicia americana 1-2 cm long 1.5 Heterospecifics Agoseris glauca 2-5 cm across 3.5 Allium cernuum 0.4-0.6 cm deep 0.5 Aquilegia flavescens 2-4 cm long 3 Astragalus vexilliflexus 0.6-0.8 cm deep 0.7 Astragulus americanus 1.3-1.5 cm deep 1.4 Delphinium glaucum 2-2.5 cm long 2.25 Chamerion angustifolium 2-4 cm across 3 Erigeron glabellus 1.8-3.5 cm across 2.65 Gentianella amarella 1-2 cm deep 1.5 Geranium viscosissimum 3-4 cm across 3.5 Geum rivale 1.5-2 cm across, 0.7-1 cm deep 1.75 Hedysarum alpinum 1.2-1.8 cm deep 1.5 Hedysarum sulphurescens 1.5-1.8 cm deep 1.65 Hieracium umbellatum 2-2.5 cm across 2.25 Lonicera involucrata 1-1.8 cm long 1.4 Orthilia secunda 0.5-1 cm across 0.75 Oxytropis monticola 1.2-1.7 cm deep 1.45 Oxytropis splendens 1-1.5 cm deep 1.25 Pedicularis bracteosa 1.5-2 cm deep 1.75 Potentilla fruticosa 2-3 cm 2.5 Pyrola asarifolia 0.8-1.4 cm across 1.1

123 Species Maximum floral measurement Flower size adjustment Pyrola chlorantha 0.8-1.5 cm across 1.15 Rhinanthus minor 1-2 cm deep 1.5 Rosa acicularis 4-6 cm across 5 Rubus idaeus 1-1.2 cm across 1.1 Rubus pubescens 1.2-2 cm across 1.6 Senecio spp. 0.5-2.5 cm across 1.5 Solidago spathulata 0.5-0.8 cm across 0.7 Taraxacum officinale 2-5 cm across 3.5 Trifolium hybridum 1-2 cm across 1.5 Trifolium pratense 2-5 cm across 3.5 Vaccinium caespitosum 0.4-0.5 cm deep 0.45 Vaccinium vitis-idaea 0.5 cm deep 0.5 Viola adunca 0.8-2 cm deep 1.4 Viola canadensis 1-2.5 cm across 1.75

124 APPENDIX B Species descriptions for all nine sampled focal species that were sufficiently replicated for analysis.

Arnica cordifolia (heartleaf arnica) is a rhizomatous perennial herb that lives in a broad variety habitat types: from moist, shady woods to drier, more exposed locales, ranging from coniferous forests to subalpine meadows (Weber 1972, Moss 1983). Its floral morphology was classified as disc for the purposes of this study, and flowers are yellow in colour. Plants can reproduce both sexually and asexually, and the number of sets of chromosomes determines the primary mode of reproduction for an individual (diploid vs. polyploidy; Kao 2008). Asexual reproduction via rhizomes can lead to the common occurrence of discrete, monospecific patches (Young 1983), but A. cordifolia can also produce asexual seeds (Kao 2008). However, plants under open pollination produced more seeds than bagged plants, suggesting some role of pollen limitation for seed set in this species (Kao 2007). Seeds are dispersed by wind. Individual plants from open microhabitats tend to grow larger and produce more seeds than those from shaded locations, and single-species patches are denser and have greater reproductive effort in open habitats (Young 1983). However, ideal germination conditions are in shade. Thus, A. cordifolia is best adapted to a successional forest habitat type (Knapp et al. 1989). In my study area, A. cordifolia was predominantly found in forests (Table 2.2).

Campanula rotundifolia (harebell) is a circumpolar perennial forb and can be found in a wide variety of habitats including: moist to dry hillsides, meadows and open woods, rocky sites, and alpine outcrops (Moss 1983). It is native to and common throughout Alberta (Royer and Dickinson 2007). Flowers are blue, and morphology falls within the pendular classification for the purposes of this study, as it is bell-shaped and tends to have an inverted orientation. C. rotundifolia can persist in low nutrient sites and responds quickly to changes in land usage (Lindborg and Eriksson 2004, Lindborg et al. 2005). It spreads quickly through later seral stages in open sites, especially in well-drained gravel or shallow soils (Pahl and Smreciu 1999). In my study area, C. rotundifolia was slightly more common in clearcuts than in forested habitats (Fig 2.2). This species is protandrous, so pollen is released from anthers before that particular flower’s stigma is receptive to pollen (Nyman 1992). C. rotundifolia is pollinated primarily by bumble bees and solitary bees (Bingham and Orthner 1998). Though self-compatible, cross-

125 pollination results in more seeds than self-pollination (Nuortila et al. 2004). Seeds may be wind or fluvial dispersed (Shelter and Morin 1986).

Castilleja miniata (indian paintbrush) is a widely distributed common perennial herb and can be found in a variety of habitat types: open woods, wet to dry meadows, fens, grassy slopes and roadsides (Moss 1983). In my study area, C. miniata was typically found growing in clearcuts (Fig 2.2). C. miniata is an early successional species with few germination requirements, but is rarely an aggressive competitor (Tannas 2004). C. miniata is a root hemiparasite, and relies heavily on neighbouring individuals; taproots grow until they contact heterospecific plant roots, frequently grasses and legumes, and then penetrate the roots of these plants to access nutrients (Matthies 1997). Castilleja are nearly self-incompatible, benefitting from outcrossed pollen (Lertzman 1981, Cariveau et al. 2004), and are pollinated by bumble bees, flies, hover flies and hummingbirds (Hersch and Roy 2007). Flowers consist of green tubular corollas surrounded by showy ranging from red to yellowish-white in colour (Moss 1983), and due to the complexity of corollas, C. miniata is classified as morphologically zygomorphic. Seeds can be dispersed over short distances by wind, but are usually just dropped at the base of the plant (Caplow 2004).

Eurybia conspicua (showy aster) is an herbaceous perennial from the composite family, primarily located in woodlands (spruce-fir, pine, aspen-conifer and aspen) and clearings and can be found throughout western North America (Moss 1983, Boufford 1997). Flowers are purple and morphologically classified as disc for the purposes of this study, as floral heads consist of many small tubular flowers arranged in clusters in open inflorescences (i.e., Asteraceae; as classified in Lázaro et al. 2013). It is insect pollinated by a variety of species including butterflies, moths, bees and flies and is self compatible, but most ovules require outcrossed pollen for development into seeds (Jones 1978, Allen et al. 1983). E. conspicua is commonly found in moist to dry meadows, forest openings, thickets, and clearings at low to middle elevations. These plants are able to expand vegetatively under closed forest canopies, and then flower extensively when the canopy opens (Breitung 1988). E. conspicua is tolerant to habitat disturbances including both fire (Boufford 1997) and forestry harvest (Harper and Macdonald 2002). In my study area, E. conspicua was predominantly found in clearcuts (Table 2.2).

126 Geranium richardsonii (Richardson’s geranium) is a long-lived perennial of moist thickets and open woods and is the most widespread native geranium in North America (Jones and Jones 1943, Moss 1983). Its floral morphology was classified as disc-shaped within this study system as it produced open, easily accessible flowers that are white in colour. It tends to do best in moderately disturbed habitats – for example, understory cover of Richardson's geranium decreased in the four years following logging, but increased in year 5 (Crouch 1985). In my study area, G. richardsonii was commonly found in both habitats (Table 2.2). G. richardsonii is gynodioecious; individual plants produce all hermaphroditic or all female flowers, and it is self compatible and can suffer from inbreeding depression via geitonogamy (Williams et al. 2000). Flowers are visited by a wide array of potential pollinators, the most common being bees, beetles and flies (Kearns 1990). G. richardsonii is an active seed disperser, and seeds can be thrown some distance from the parent plant through a catapult-like mechanism. However, seeds do not remain viable for more than one winter (Green 1978).

Lathyrus ochroleucus (creamy peavine) is a climbing legume with a habitat preference for moist woods and clearings (Moss 1983). It prefers loam to sandy loam soil and can grow naturally in disturbed areas. In my study area, L. ochroleucus was located in both forests and clearcuts across sites (Fig 2.2). Peavine is not considered a good competitor even though it can tolerate saline soils, as it is sensitive to deviations from neutral pH (Burton and Burton 2003). However, L. ochroleucus can tolerate some level of habitat disturbance, as densities increased at the forest edge within a year after a clearcut logging event (Harper and Macdonald 2002). It is insect pollinated, primarily by bees and butterflies, but currently there have been no studies into its level of self-compatibility. It produces white to creamy coloured flowers and falls into the zygomorphic morphology category due to high floral complexity.

Linnaea borealis (twinflower) is a circumpolar creeping woodland plant, an important component of the boreal and mixed wood forest understory throughout Alberta (Tannas 1997). In my study area, L. borealis was more commonly located in the forest understory (Fig 2.2). It is a key species for providing ground cover in climax forest communities, but has been found in disturbed areas such as cut blocks as well (Howard 1993). It is pollinated by insects, is generally self-incompatible (Scobie and Wilcock 2009), but has been reported to self-fertilize on rare occasions (Howard 1993). Flowers are pink/white and actinomorphic but tubular, leading to pendular classification for the purposes of this study. The seeds come in pairs (from paired sets

127 of flowers) and are barbed, catching on fur of mammals for dispersal (Scobie and Wilcock 2009).

Mertensia paniculata (northern bluebell) is a shade-tolerant perennial herb typically found in damp places such as lush woodlands, willow thickets, depressions, meadows and stream banks (Tannas 1997). It prefers moist, nutrient rich soils (Beckingham and Archibald 1996, Turkington et al. 1998, Arii and Turkington 2002) and can be found among the regenerating post-fire community in early succession (Reeves 2006), as flowering typically occurs in sunny locations (Morris 1996). In my study area, M. paniculata was predominantly a plant of forested habitats (Fig 2.2). It is self-compatible but is insect pollinated, primarily by bumble bees, and seeds are wind dispersed (Reeves 2006, Morris et al. 2010). It also suffers from nectar robbing by bumble bees (Morris 1996, Morris et al. 2010). Flowers are tubular and typically inverted, falling under the pendular morphology classification, and colour changes from pink to blue as they mature and open (Morris 1996).

Vicia americana (purple vetch) is an insect-pollinated climbing legume typically found in open woods and meadows, but it can colonize disturbed and agricultural land (Moss 1983, Pahl and Smreciu 1999). For example, it has been found flourishing within disturbed alpine rangelands, re-vegetated coal mines and along road sides (Coladonato 1993, Gerling et al. 1996, Pahl and Smreciu 1999). In my study area, V. americana was more frequently located in clearcuts (Fig 2.2). The vine is common throughout most of Alberta, as it is adapted to early seral conditions and is highly drought tolerant. Flowers are purple in colour and floral morphology is zygomorphic, as flowers are closed, complex and tubular. As with all legumes, it may aid in the success of neighbouring plants through improvement of local soil nutrient conditions via nitrogen fixation (Schellenberg and Banerjee 2002). Bumble bees have a strong preference for V. americana over a Lathyrus species (Inouye 1980). V. americana is self-compatible once stigmas are made receptive, but this requires visitation by bees (Gunn 1965). Seeds are usually dispersed when pods dehisce, spraying seeds a short distance from the plant, though there is some animal dispersal.

128 APPENDIX C General local effects of heterospecific floral morphology on all focal species that selected heterospecifics as having an additive or interactive effect on seed size or number.

C.1 METHODS

In Chapter 2, for all species that selected one of the candidate models (Table 2.4) that contained heterospecific density as an explanatory variable for either seed size or seed number, I investigated whether specific species or specific floral morphologies dominated at heterospecific densities that were highly positive or highly negative in their impacts on seed traits. To do this, I examined species-specific figures (see Appendix D) for peaks and troughs in seed size or abundance for both additive and interactive effects, but always treating the effect of heterospecifics as if it were additive. From these peaks and troughs I determined the range(s) of heterospecific densities that were the best and worst for seeds for each species, and subset the original dataset into point locations (of either 1 m2 or 10 m2, depending on the best selected model) that contained heterospecific densities within those ranges. I then calculated the average proportion of each individual heterospecific across all subsampled locations (for beneficial and detrimental subsets; Table C.1). Then, to generate average proportions of each floral morphology type (see Section 2.2.2), I took the sum of all mean heterospecific proportions as they fit within each shape category (Table C.2, C.3). I also calculated the average proportion of each heterospecific species/morphology type across all sampled locations for valid comparative interpretation between average proportional heterospecifics (heterospecifics that tend to naturally grow around a particular focal species) and highly beneficial or highly detrimental heterospecifics.

129 Table C.1: A summary identifying the species of heterospecifics present (3 most common) for all situations in which heterospecifics were selected as a component of the best selected model for that focal species. Focal species are ordered from least restrictive to most restrictive morphology group (disc to zygomorphic). Most common heterospecifics were the species most often found surrounding the focal species at the selected scale across all locations sampled, most beneficial heterospecifics were the species most often found in the subset of locations that produced the largest or most numerous seeds, and most detrimental were the species most often found in locations that produced the smallest or fewest seeds. The most common beneficial and detrimental heterospecific for each seed character is bolded if it was a different species from the most common across all locations for that focal species.

Focal species Scale Most common heterospecifics Most beneficial heterospecifics Most detrimental heterospecifics Arnica 10 m2 Mertensia paniculata (22.4%) Size Eurybia conspicua (49.9%) Eurybia conspicua (48.0%) cordifolia Epilobium angustifolium (20.4%) Epilobium angustifolium (38.2%) Epilobium angustifolium (26.1%) (Disc) Linnaea borealis (14.5%) Oxytropis monticola (10.8%) Linnaea borealis (24.0%) 1 m2 Mertensia paniculata (18.9%) Number Mertensia paniculata (25.0%) Linnaea borealis (23.3%) Linnaea borealis (17.9%) Linnaea borealis (20.2%) Mertensia paniculata (16.5%) Epilobium angustifolium (16.3%) Eurybia conspicua (8.2%) Epilobium angustifolium (13.0%) Eurybia 10 m2 Epilobium angustifolium (18.5%) Size Epilobium angustifolium (39.2%) Campanula rotundifolia (22.3%) conspicua Trifolium hybridum (15.7%) Trifolium hybridum (27.3%) Trifolium hybridum (20.3%) (Disc) Campanula rotundifolia (14.4%) Astragulus americanus (19.5%) Epilobium angustifolium (14.2%) Number Trifolium hybridum (21.0%) Campanula rotundifolia (22.3%) Campanula rotundifolia (20.5%) Trifolium hybridum (20.3%) Epilobium angustifolium (19.7%) Epilobium angustifolium (14.2%) Campanula 10 m2 Rhinanthus minor (16.4%) Number Eurybia conspicua (17.7%) Solidago spathulata (48.8%) rotundifolia Eurybia conspicua (14.2%) Epilobium angustifolium (13.6%) Rhinanthus minor (32.9%) (Pendular) Epilobium angustifolium (11.4%) Oxytropis monticola (8.0%) Epilobium angustifolium (8.0%) Mertensia 10 m2 Rubus pubescens (22.2%) Number Geranium richardsonii (25.5%) Rubus pubescens (27.5%) paniculata Rubus idaeus (17%) Rubus pubescens (21.5%) Rubus idaeus (17.8%) (Pendular) Arnica cordifolia (15.9%) Vicia americana (9.7%) Arnica cordifolia (15.4%)

130 Focal species Scale Most common heterospecifics Most beneficial heterospecifics Most detrimental heterospecifics Castilleja 1 m2 Campanula rotundifolia (17.5%) Size Campanula rotundifolia (29.3%) Geranium richardsonii (48.4%) miniata Lathyrus ochroleucus (15%) Lathyrus ochroleucus (25.1%) Eurybia conspicua (29.1%) (Zygomorphic) Hedysarum sulphurescens (13.2%) Senecio pauperculus (10.0%) Astragalus americanus (12.3%) Number Lathyrus ochroleucus (53.3%) Campanula rotundifolia (25.0%) Mertensia paniculata (12.5%) Lathyrus ochroleucus (16.6%) Geranium richardsonii (12.5%) Rosa acicularis (10.0%) Lathyrus 1 m2 Vicia americana (19.5%) Size Hedysarum suphurescens (50%) Vicia americana (26.2%) ochroleucus Hedysarum sulphurescens (14.8%) Epilobium angustifolium (34.5%) Linnaea borealis (15.4%) (Zygomorphic) Linnaea borealis (12.9%) Rubus idaeus (6.1%) Geranium richardsonii (11.7%) Vicia 10 m2 Trifolium hybridum (19.6%) Number Trifolium hybridum (18.5%) Trifolium hybridum (18.3%) americana Potentilla fruticosa (11.9%) Potentilla fruticosa (11.9%) Potentilla fruticosa (17.8%) (Zygomorphic) Geranium richardsonii (9.5%) Geranium richardsonii (9.7%) Rosa acicularis (12.9%)

131 Table C.2: A summary identifying the groups of heterospecific morphology (disc, tubular, and zygomorphic) present for all situations in which heterospecifics were selected as a component of the best selected model for that focal species. Focal species are ordered from least restrictive to most restrictive morphology group (disc to zygomorphic, colour coded). The most average heterospecific column contains the mean proportion of the 3 morphology groups surrounding the focal species at the selected scale across all locations sampled, the most beneficial column contains the mean proportion found in the subset of locations that produced the largest or most numerous seeds, and the detrimental column contains the mean proportion at the locations that produced the smallest or fewest seeds.

Focal species Scale Average proportion of Seed Most beneficial Most detrimental heterospecific groups character group proportions group proportions Arnica cordifolia 10 m2 45.2% tubular Size 88.14% disc 74.6% disc (Disc) 40.8% disc 10.8% zygomorphic 25.4% tubular 13.9% zygomorphic 1.0% tubular 0.0% zygomorphic 1 m2 42.7% tubular Number 52.8% tubular 47.7% tubular 41.2% disc 25.8% disc 34.4% disc 16.1% zygomorphic 21.4% zygomorphic 17.9% zygomorphic Eurybia conspicua 10 m2 42.5% disc Size 53.6% zygomorphic 44.4% disc (Disc) 31.2% zygomorphic 41.0% disc 32.8% tubular 26.3% tubular 5.4% tubular 22.8% zygomorphic Number 43.2% disc 44.4% disc 28.6% tubular 32.8% tubular 7.3% zygomorphic 22.8% zygomorphic Campanula 10 m2 47.9% disc Number 55.5% disc 65.7% disc rotundifolia 42.4% zygomorphic 33.1% zygomorphic 34.2% zygomorphic (Tubular) 9.5% tubular 11.4% tubular 0.1% tubular Mertensia paniculata 10 m2 78.0% disc Number 71.9% disc 78.5% disc (Tubular) 16.3% zygomorphic 21.2% zygomorphic 15.3% zygomorphic 5.7% tubular 6.9% tubular 6.3% tubular

132

Focal species Scale Average proportion of Seed Most beneficial Most detrimental heterospecific groups character group proportions group proportions Castilleja miniata 1 m2 47.0% zygomorphic Size 39.6% tubular 79.9% disc (Zygomorphic) 27.0% disc 36.7% zygomorphic 13.6% zygomorphic 26.0% tubular 23.7% disc 6.5% tubular Number 65.9% zygomorphic 46.0% zygomorphic 21.6% disc 33.8% tubular 12.5% tubular 20.1% disc Lathyrus ochroleucus 1 m2 51.7% zygomorphic Size 54.8% zygomorphic 51.6% zygomorphic (Zygomorphic) 33.2% tubular 45.2% disc 35.2% tubular 15.1% disc 0.0% tubular 13.2% disc Vicia americana 10 m2 57.0% zygomorphic Number 57.6% zygomorphic 50.3% zygomorphic (Zygomorphic) 39.1% disc 38.4% disc 46.6% disc 3.9% tubular 4.0% tubular 3.1% tubular

133 Table C.3: A summary identifying the groups of heterospecific morphology (disc, pendular, and zygomorphic) present for all situations in which heterospecifics were selected as a component of the best selected model for that focal species. Focal species are ordered from least restrictive to most restrictive morphology group (disc to zygomorphic, colour coded). Changes in the proportional presence of each morphology type are presented for densities of heterospecifics that were the most beneficial and most detrimental (see Appendix A for species-specific positive and negative heterospecific density ranges). The change (positive or negative proportional difference) represents whether morphology types were more or less abundant for highly positive or highly negative seed production, in comparison to their overall average abundance across all sampling locations. When presence of a particular morphology type was more than 10% different than its presence on average, it is highlighted with its respective morphology colour.

Focal species Scale Seed Change from average group Change from average group character proportions, beneficial proportions, detrimental Arnica cordifolia 10 m2 Size +47.3% disc +33.8% disc (Disc) -44.2% pendular -19.8% pendular -3.1% zygomorphic -13.9% zygomorphic 1 m2 Number -15.4% disc -6.8% disc +10.1% pendular +5.0% pendular +5.3% zygomorphic -1.8% zygomorphic Eurybia conspicua 10 m2 Size +22.4% zygomorphic -8.4% zygomorphic (Disc) -20.9% pendular +6.5% pendular -1.5% disc +1.9% disc Number -23.9% zygomorphic -8.4% zygomorphic +2.3% pendular +6.5% pendular +0.7% disc +1.9% disc Campanula rotundifolia 10 m2 Number -9.3% zygomorphic +17.8% disc (Pendular) +7.6% disc -9.4% pendular +1.9% pendular -8.2% zygomorphic Mertensia paniculata 10 m2 Number -6.1% disc -1.0% zygomorphic (Pendular) +4.9% zygomorphic +0.6% pendular +1.2% pendular +0.5% disc

134

Focal species Scale Seed Change from average group Change from average group character proportions, beneficial proportions, detrimental Castilleja miniata 1 m2 Size +13.6% pendular +52.9% disc (Zygomorphic) -10.3% zygomorphic -33.4% zygomorphic -3.3% disc -19.5% pendular Number +18.9% zygomorphic +7.8% pendular -13.5% pendular -6.9% disc -5.4% disc -1.0% zygomorphic Lathyrus ochroleucus 1 m2 Size -33.2% pendular +2.0% pendular (Zygomorphic) +30.1% disc -1.9% disc +3.1% zygomorphic -0.1% zygomorphic Vicia americana 10 m2 Number -0.7% disc +7.5% disc (Zygomorphic) +0.6% zygomorphic -6.7% zygomorphic +0.1% pendular -0.8% pendular

135 APPENDIX D Specific model information and individualized figures for the best selected model by AIC for each species in Chapter 2. All component plots have an adjusted y-axis to represent the partial contribution of each variable, controlling for all other variables in that model.

Species-specific seed size effects

Table D.1: AIC-selected GAMM for A. cordifolia best predicting seed size (ln[mm2]) at the 10 m2 scale, from covariates and the interaction between conspecific and heterospecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day F (smooth) 8.740 <0.001 Spatial PC1 t (linear) 0.969 0.333 Conspecific * Heterospecific F (smooth) 6.585 <0.001

136

Figure D.1: Contour plot depicting the paired influence of conspecific and heterospecific density within 10 m2 on ln(seed size) in A. cordifolia. Pink peaks represent combinations of conspecific and heterospecific density that resulted in the highest values of ln(seed size), and blue troughs indicate smaller seeds. This surface estimate was obtained from the GAMM: ln(seed size) ~ spline(Julian) + SPC1 + spline(het, con) + random(loc in tran in hab in site). Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

Table D.2: AIC-selected GAMM for E. conspicua best predicting seed size (ln[mm2]) at the 10 m2 scale, from covariates and the interaction between conspecific and heterospecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day F (smooth) 0.002 0.969 Spatial PC1 t (linear) 1.235 0.217 Conspecific * Heterospecific F (smooth) 10.238 <0.001

137

Figure D.2: Contour plot depicting the paired influence of conspecific and heterospecific density within 10 m2 on ln(seed size) in E. conspicua. Pink peaks represent combinations of conspecific and heterospecific density that resulted in the highest values of ln(seed size), and blue troughs indicate smaller seeds. This surface estimate was obtained from the GAMM: ln(seed size) ~ spline(Julian) + SPC1 + spline(het, con) + random(loc in tran in hab in site). Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

Table D.3: AIC-selected GAMM for C. rotundifolia best predicting seed size (ln[mm2]) at the 10 m2 scale, from covariates and conspecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day F (smooth) 11.10 <0.001 Spatial PC1 t (linear) 0.437 0.662 Conspecific density F (smooth) 16.67 <0.001

138

Figure D.3: Component smooth function plot of the additive effect of conspecific density within 10 m2 on ln(seed size) in C. rotundifolia. This spline estimate was obtained from the GAMM: ln(seed size) ~ spline(Julian) + SPC1 + spline(con) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

Table D.4: AIC-selected GAMM for L. borealis best predicting seed size (ln[mm2]) at the 1 m2 scale, from covariates and conspecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Marginally non-significant effects are italicized. Predictor term Test-statistic Value p-value Julian day F (smooth) 0.022 0.8823 Spatial PC1 t (linear) 1.882 0.0606 Conspecific density F (smooth) 6.609 <0.001

139

Figure D.4: Component smooth function plot of the additive effect of conspecific density within 1 m2 on ln(seed size) in L. borealis. This spline estimate was obtained from the GAMM: ln(seed size) ~ spline(Julian) + SPC1 + spline(con) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

Table D.5: AIC-selected GAMM for M. paniculata best predicting seed size (ln[mg]) at the local scale, from covariates, habitat type and percent canopy cover. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day F (smooth) 20.107 <0.001 Spatial PC1 t (linear) -3.476 <0.001 Habitat type t (factor) -0.251 0.8018 % canopy cover F (smooth) 3.461 0.0165

140

Figure D.5: Categorical plot of the additive effect of habitat type (CC = clearcut, F = forest) on ln(seed size) in M. paniculata. This estimate of the influence of the categorical variable habitat type was obtained from the GAMM: ln(seed size) ~ spline(Julian) + SPC1 + habitat type + spline(canopy cover) + random(loc in tran in hab in site). Dashed lines indicate one standard error.

141

Figure D.6: Component smooth function plot of the additive effect of local canopy cover on ln(seed size) in M. paniculata. This spline estimate was obtained from the GAMM: ln(seed size) ~ spline(Julian) + SPC1 + habitat type + spline(canopy cover) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

Table D.6: AIC-selected GAMM for C. miniata best predicting seed size (ln[mm2]) at the 1 m2 scale, from covariates and the additive effects of conspecific and heterospecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Marginally non-significant terms are italicized. Predictor term Test-statistic Value p-value Julian day F (smooth) 2.597 0.0771 Spatial PC1 t (linear) -1.964 0.0496 Conspecific F (smooth) 12.438 <0.001 Heterospecific F (smooth) 43.966 <0.001

142

Figure D.7: Component smooth function plot of the additive effect of conspecific density within 1 m2 on ln(seed size) in C. miniata. This spline estimate was obtained from the GAMM: ln(seed size) ~ spline(Julian) + SPC1 + spline(con) + spline(het) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

143

Figure D.8: Component smooth function plot of the additive effect of heterospecific density within 1 m2 on ln(seed size) in C. miniata. This spline estimate was obtained from the GAMM: ln(seed size) ~ spline(Julian) + SPC1 + spline(con) + spline(het) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

Table D.7: AIC-selected GAMM for L. ochroleucus best predicting seed size (ln[mm2]) at the 1 m2 scale, from covariates and the interaction between conspecific and heterospecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day F (smooth) 0.369 0.544 Spatial PC1 t (linear) 1.483 0.138 Conspecific * Heterospecific F (smooth) 6.728 <0.001

144

Figure D.9: Contour plot depicting the paired influence of conspecific and heterospecific density within 1 m2 on ln(seed size) in L. ochroleucus. Pink peaks represent combinations of conspecific and heterospecific density that resulted in the highest values of ln(seed size), and blue troughs indicate smaller seeds. This surface estimate was obtained from the GAMM: ln(seed size) ~ spline(Julian) + SPC1 + spline(het, con) + random(loc in tran in hab in site). Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

Table D.8: AIC-selected GAMM for V. americana best predicting seed size (ln[mm2]) at the 10 m2 scale, from covariates and conspecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day F (smooth) 1.433 0.232 Spatial PC1 t (linear) 0.685 0.494 Conspecific density F (smooth) 9.624 <0.001

145

Figure D.10: Component smooth function plot of the additive effect of conspecific density within 10 m2 on ln(seed size) in V. americana. This spline estimate was obtained from the GAMM: ln(seed size) ~ spline(Julian) + SPC1 + spline(con) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

146 Species-specific seed number effects

Table D.9: AIC-selected GAMM for A. cordifolia best predicting seed number (sqrt[#]) at the 1 m2 scale, from covariates and the interaction between conspecific and heterospecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day F (smooth) 1.869 0.146 Spatial PC1 t (linear) 0.398 0.691 Conspecific * Heterospecific F (smooth) 19.802 <0.001

Figure D.11: Contour plot depicting the paired influence of conspecific and heterospecific density within 1 m2 on sqrt(seed count) in A. cordifolia. Pink peaks represent combinations of conspecific and heterospecific density that resulted in the highest values of ln(seed size), and blue troughs indicate smaller seeds. This surface estimate was obtained from the GAMM: sqrt(seed count) ~ spline(Julian) + SPC1 + spline(het, con) + random(loc in tran in hab in site). Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

147 Table D.10: AIC-selected GAMM for E. conspicua best predicting seed number (sqrt[#]) at the 10 m2 scale, from covariates and the interaction between conspecific and heterospecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day F (smooth) 3.414 0.0319 Spatial PC1 t (linear) 2.274 0.0231 Conspecific * Heterospecific F (smooth) 16.426 <0.001

Figure D.12: Contour plot depicting the paired influence of conspecific and heterospecific density within 10 m2 on sqrt(seed count) in E. conspicua. Pink peaks represent combinations of conspecific and heterospecific density that resulted in the highest values of ln(seed size), and blue troughs indicate smaller seeds. This surface estimate was obtained from the GAMM: sqrt(seed count) ~ spline(Julian) + SPC1 + spline(het, con) + random(loc in tran in hab in site). Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

148 Table D.11: AIC-selected GAMM for C. rotundifolia best predicting seed number (sqrt[#]) at the 10 m2 scale, from covariates and the interactions between bee density and both conspecific and heterospecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Marginally non-significant terms are italicized. Predictor term Test-statistic Value p-value Julian day F (smooth) 3.519 0.0253 Spatial PC1 t (linear) 0.773 0.44 Bee * Conspecific F (smooth) 8.989 <0.001 Bee * Heterospecific F (smooth) 2.315 0.0793

Figure D.13: Contour plot depicting the paired influence of conspecific floral density and bee abundance within 10 m2 on sqrt(seed count) in C. rotundifolia. Pink peaks represent combinations of conspecific and heterospecific density that resulted in the highest values of ln(seed size), and blue troughs indicate smaller seeds. This surface estimate was obtained from the GAMM: sqrt(seed count) ~ spline(Julian) + SPC1 + spline(con, bee) + spline(het, bee) + random(loc in tran in hab in site). Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

149

Figure D.14: Contour plot depicting the paired influence (p=0.08) of heterospecific floral density and bee abundance within 10 m2 on sqrt(seed count) in C. rotundifolia. Pink peaks represent combinations of conspecific and heterospecific density that resulted in the highest values of ln(seed size), and blue troughs indicate smaller seeds. This surface estimate was obtained from the GAMM: sqrt(seed count) ~ spline(Julian) + SPC1 + spline(con, bee) + spline(het, bee) + random(loc in tran in hab in site). Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

Table D.12: AIC-selected GAMM for M. paniculata best predicting seed number (sqrt[#]) at the 10 m2 scale, from covariates and the additive effects of conspecific and heterospecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day F (smooth) 0.668 0.511 Spatial PC1 t (linear) -2.198 0.0283 Conspecific F (smooth) 16.464 <0.001 Heterospecific F (smooth) 79.114 <0.001

150

Figure D.15: Component smooth function plot of the additive effect of conspecific density within 10 m2 on sqrt(seed number) in M. paniculata. This spline estimate was obtained from the GAMM: sqrt(seed number) ~ spline(Julian) + SPC1 + spline(con) + spline(het) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

151

Figure D.16: Component smooth function plot of the additive effect of heterospecific density within 10 m2 on sqrt(seed number) in M. paniculata. This spline estimate was obtained from the GAMM: sqrt(seed number) ~ spline(Julian) + SPC1 + spline(con) + spline(het) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

Table D.13: AIC-selected GAMM for C. miniata best predicting seed number (sqrt[#]) at the 1 m2 scale, from covariates and the additive effects of bee abundance and conspecific and heterospecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day t (linear) 1.585 0.113 Spatial PC1 t (linear) -0.465 0.642 Bee F (smooth) 10.57 0.00116 Conspecific F (smooth) 56.64 <0.001 Heterospecific F (smooth) 18.87 <0.001

152

Figure D.17: Component smooth function plot of the additive effect of conspecific density within 1 m2 on sqrt(seed number) in C. miniata. This spline estimate was obtained from the GAMM: sqrt(seed number) ~ Julian + SPC1 + spline(con) + spline(het) + spline(bee) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

153

Figure D.18: Component smooth function plot of the additive effect of heterospecific density within 1 m2 on sqrt(seed number) in C. miniata. This spline estimate was obtained from the GAMM: sqrt(seed number) ~ Julian + SPC1 + spline(con) + spline(het) + spline(bee) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

154

Figure D.19: Component smooth function plot of the additive effect of bee abundance within 1 m2 on sqrt(seed number) in C. miniata. This spline estimate was obtained from the GAMM: sqrt(seed number) ~ Julian + SPC1 + spline(con) + spline(het) + spline(bee) + random(loc in tran in hab in site). The smoothed line represents the deviation of the fitted model from the mean Y value for changes in each covariate, centered on zero, controlling for all other variables in the model. Partial residuals are plotted. Shaded curves indicate 2 standard error bounds, including the uncertainty about the overall mean.

Table D.14: AIC-selected GAMM for V. americana best predicting seed number (sqrt[#]) at the 10 m2 scale, from covariates and the interactions between bee density and both conspecific and heterospecific floral density. Random effects are transect location nested within transect, within habitat, within site. Significant terms (approximate) are bolded. Predictor term Test-statistic Value p-value Julian day t (linear) -0.002 0.999 Spatial PC1 t (linear) 0.047 0.962 Bee * Conspecific F (smooth) 14.727 <0.001 Bee * Heterospecific F (smooth) 4.444 0.00252

155

Figure D.20: Contour plot depicting the paired influence of conspecific floral density and bee abundance within 10 m2 on sqrt(seed count) in V. americana. Pink peaks represent combinations of conspecific and heterospecific density that resulted in the highest values of ln(seed size), and blue troughs indicate smaller seeds. This surface estimate was obtained from the GAMM: sqrt(seed count) ~ Julian + SPC1 + spline(con, bee) + spline(het, bee) + random(loc in tran in hab in site). Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

156

Figure D.21: Contour plot depicting the paired influence of heterospecific floral density and bee abundance within 10 m2 on sqrt(seed count) in V. americana. Pink peaks represent combinations of conspecific and heterospecific density that resulted in the highest values of ln(seed size), and blue troughs indicate smaller seeds. This surface estimate was obtained from the GAMM: sqrt(seed count) ~ Julian + SPC1 + spline(con, bee) + spline(het, bee) + random(loc in tran in hab in site). Breaks in the surface indicate large gaps in data, as spline estimates are unreliable within such gaps.

157