Mechanisms Structuring Dragonfly Communities Across Canopy and Forest Cover Gradients

Mechanisms Structuring Dragonfly Communities across Canopy and Forest Cover Gradients

Sarah Kathryn French

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Department of Ecology and Evolutionary Biology University of Toronto

Mechanisms Structuring Dragonfly Communities across Canopy and Forest Cover Gradients

Sarah Kathryn French

Doctor of Philosophy

Department of Ecology and Evolutionary Biology University of Toronto

2018 Abstract

Forest regrowth, which is occurring across North America, may have large impacts on the structure of larval dragonfly assemblages. My thesis examined how habitat selection, movement behaviour, and species-sorting shape aquatic larval dragonfly assemblages in response to landscape heterogeneity, including forest cover in the landscape and canopy cover over ponds.

The mechanisms structuring aquatic communities across environmental gradients are often difficult to distinguish, so I used multiple approaches to assess the contributions of different mechanisms to these observed patterns. Firstly, I tested the role of adult habitat selection versus larval species-sorting in structuring larval dragonfly assemblages across a canopy gradient using artificial aquatic habitats (i.e. mesocosms) placed across gradients of artificial shading and of natural canopy cover. Adult dragonflies preferentially visited aquatic mesocosms situated under open canopy. I tested for species-sorting effects by placing larvae in mesocosms across a natural canopy cover gradient, but found no evidence of differential mortality or growth. Secondly, I examined the effects of forest cover in the terrestrial landscape on adult dragonfly movement.

Adults showed preferences for open field environments rather than closed forest environments, ii

moving marginally more often in fields, more often towards fields, and flying longer in fields.

However, adults’ direction of movement also depended on what environments they were released in, and their flight times were species-specific. Lastly, I investigated how forest cover in the landscape and pond canopy cover affected the distribution of larval dragonflies, as well as their reassembly in ponds following a drying event. Both local (pond canopy cover, permanence, and area) and landscape (pond connectivity based on distance and amount of forest cover between ponds) heterogeneity affected species diversity and recovery following a drying event.

My thesis work provides new insights into how aquatic communities are structured by forest and canopy cover, and how they might respond to forest regrowth.

iii

Acknowledgments

Firstly, thank you to Shannon McCauley for her endless enthusiasm and support, strong mentorship, and for fostering such a fantastic and collaborative group of colleagues. I am incredibly grateful for her patience as I navigated my way through a new field, and for taking a chance on a student who fell asleep during their first interview/car trip together to visit a field station.

The McCauley lab group has been a continuous source of support and inspiration. Thank you to Shantel Catania, for her contagious passion for aquatic invertebrates and natural history, and for her assistance in the field during my first year of experiments. Thank you to Dachin Frances, my partner in crime from the beginning, for her help in the field, expertise in the lab, and constant encouragement to the very end. Thank you to Celina Baines, Rosemary Martin, Rosalind Murray, Ilia Ferzoco, and Christopher Searcy, for their unwavering support, friendship, and feedback throughout my degree.

I am extremely grateful for the mentorship of Helen Rodd, and her guidance on my committee, and of Helene Wagner, who has helped me to cultivate a love for teaching. Thank you to Benjamin Gilbert, for his constructive feedback on my committee over the years, and to Becky Raboy and Peter Kotanen for their helpful comments on my thesis during my appraisal. A special thanks to Carolyn Moon and Stephanie do Rego for all their administrative help and for fielding an endless barrage of questions. An enormous thanks goes to Stephan Schneider for all his assistance at the Koffler Scientific Reserve, including the construction of field equipment, and for his keen knowledge of and excitement for the outdoors.

Funding for my projects was graciously provided by the departments of Ecology and Evolutionary Biology at the University of Toronto, and Biology at the University of Toronto Mississauga, the University of Toronto’s School of Graduate Studies, a Queen Elizabeth II/Pfizer Graduate Scholarship in Science and Technology, a Natural Sciences and Engineering Research Council of Canada Scholarship, and the American Museum of Natural History. Thank you to the Koffler Scientific Reserve for their research support. iv

Thank you to the entire UTM graduate community for making my time in the department so enjoyable and entertaining, and specifically Audrey Reid, Aaron LeBlanc, Connor Fitzpatrick, Krystal Nunes, Chris Wong, Ruth Rivkin, Molly Hetherington-Rauth, Mark MacDougall, and Amy Wong, for their incredible friendship throughout my PhD. I am also greatly appreciative of the ongoing support from my family and friends, particularly from the mother/mother-in-law babysitting dream team during my last stages of thesis writing.

Lastly, thank you to Shawn French for your love and encouragement, assistance in impromptu snowy field adventures, and for being my sounding board. A special thanks to Desmond, who has been a constant source of joy and who has been extremely patient with me during the writing stage, and to my thesis fetus and now-emerged little larva, Thea, who greatly improved my sleeping habits while writing, perhaps a little more so than necessary, and who thankfully waited until after my defense to make her entrance.

Table of Contents

Acknowledgments ...... iv

Table of Contents ...... vi

List of Tables ...... ix

List of Figures ...... x

List of Appendices ...... xii

Chapter 1 General Introduction ...... 1

Community Assembly ...... 1

1.1 Habitat Selection ...... 1

1.2 Forest Cover Change ...... 3

1.3 Objectives and Research Questions...... 5

1.4 Study System ...... 7

1.4.1 Study Taxon ...... 7

1.4.2 Study Location ...... 8

1.5 Goals of Chapters ...... 9

1.5.1 Chapter 2 ...... 9

1.5.2 Chapter 3 ...... 10

1.5.3 Chapter 4 ...... 11

1.6 References ...... 12

Chapter 2 Canopy Cover Affects Habitat Selection by Adult Dragonflies ...... 19

Abstract ...... 19

2.1 Introduction ...... 19

2.2 Materials and Methods ...... 23

2.2.1 Artificial Shading Experiment ...... 25

2.2.2 Natural Canopy Experiment ...... 28

2.2.3 Larval Performance Experiments ...... 30

2.3 Results ...... 32

2.3.1 Artificial Shading Experiment ...... 32

2.3.2 Natural Canopy Experiment ...... 32

2.3.3 Larval Performance Experiments ...... 35

2.4 Discussion ...... 35

2.5 Conclusion ...... 42

2.6 References ...... 43

Chapter 3 The Movement Responses of Three Libellulid Dragonfly Species to Open and Closed Landscape Cover ...... 52

Abstract ...... 52

3.1 Introduction ...... 52

3.2 Methods ...... 55

3.2.1 Statistical Analysis ...... 59

3.3 Results ...... 60

3.4 Discussion ...... 65

3.5 Conclusion ...... 71

3.6 References ...... 71

Chapter 4 Local Habitat Conditions and Landscape Effects Determine the Recovery of Larval Dragonfly Assemblages Following Pond Drying ...... 80

Abstract ...... 80

4.1 Introduction ...... 81

vii

4.2 Methods ...... 85

4.3 Results ...... 90

4.4 Discussion ...... 93

4.5 References ...... 98

Chapter 5 General Discussion ...... 106

Thesis Summary ...... 106

5.1 Conservation Implications ...... 109

5.2 Future Directions ...... 110

5.2.1 Multiple Life History Stages ...... 110

5.2.2 Community Assembly Involving Multiple Taxa ...... 111

5.2.3 Combining Local and Landscape Scales ...... 112

5.3 References ...... 113

viii

List of Tables

Table 2.1 Statistical test results for the artificial shading experiments ...... 34

Table 2.2 Statistical test results for the natural canopy experiments ...... 37

Table 2.3 Statistical test results for the larval performance experiments...... 38

Table 3.1 Morphological characteristics of the three dragonfly species used in releases ...... 66

Table 4.1 The effects of canopy cover, and land cover and distance connectivity, on diversity patterns during community reassembly of ponds following a disturbance created by pond drying ...... 94

List of Figures

Fig. 1.1 A schematic of the possible mechanisms structuring larval dragonfly assemblages at aquatic habitats ...... 6

Fig. 2.1 Species selecting habitat following dispersal ...... 21

Fig. 2.2 Map of experimental sites at Koffler Scientific Reserve...... 24

Fig. 2.3 Mean number of visits by adult dragonflies and mean abundance of larvae per mesocosm, based on a fully factorial design of no heating and heating, and shaded and open treatments ...... 33

Fig. 2.4 With an increase in natural canopy cover, the total number of adult visits per mesocosm decreased, whereas larval abundances per mesocosm were unaffected ...... 36

Fig. 3.1 Dragonflies were released at three release locations at each of 12 sites, with three replicates per species for three species at Koffler Scientific Reserve ...... 58

Fig. 3.2 Mean proportion of Sympetrum vicinum, Sympetrum hybrid, and Leucorrhinia intacta individuals that took flight when released at edges, in fields, and in forests ...... 61

Fig. 3.3 Mean number of dragonfly individuals flying towards forests and fields, when released at edges, in fields, and in forests ...... 63

Fig. 3.4 Mean flight times of those Sympetrum vicinum, Sympetrum hybrids, and Leucorrhinia intacta individuals that took flight ...... 64

Fig. 4.1 Map of the Edwin S. George Reserve showing open canopy and closed canopy landscape cover, as well as open and wooded wetlands ...... 86

Fig. 4.2 Local conditions at a site versus the Shannon diversity and richness of dragonfly larvae ...... 91

Fig. 4.3 Local conditions versus the abundance and richness of diapausing and non-diapausing dragonfly larvae ...... 92

List of Appendices

Appendix A: Supplementary Materials for Chapter 2 ...... 119

Appendix B: Supplementary Materials for Chapter 3...... 127

Appendix C: Supplementary Materials for Chapter 4...... 128

xii

Chapter 1 General Introduction Community Assembly

Local and regional processes can play important roles in the assembly of a community. At the regional scale, dispersal can connect communities and facilitate species’ presence locally. However, species may not persist at a habitat if local conditions inhibit their survival (Connell, 1985; Freestone & Inouye, 2006). Species may also be absent from or their presence limited in habitats because they simply do not arrive there: for example if they have a low dispersal ability or a limited ability to find or detect habitat during dispersal (Connell, 1985; Schooley & Wiens, 2003; Freestone & Inouye, 2006). Isolation from a source habitat may also impede an individual’s ability to reach a habitat (e.g. Urban, 2004). Metacommunity ecology has developed four paradigms to describe the assembly of communities linked by dispersal, integrating local and regional processes. These include patch dynamics, which involve a trade-off between species’ abilities to colonize and compete within and across habitats, mass effects, which involve the dynamics between source and sink habitats of varying quality, species-sorting, where species respond differently to environmental conditions such as by experiencing differential mortality across a gradient of resources and/or habitat conditions, and neutral processes, which assume functional equivalence among species (Leibold et al., 2004, Logue et al., 2011). The metacommunity perspective has increased the attention given to the role of regional processes in structuring communities but has insufficiently integrated the lessons of landscape ecology. For example, the type of environment or landscape cover between habitats may also affect dispersal and the likelihood of individuals reaching a habitat (e.g. Ricketts, 2001), depending on how individuals use and respond to these environments. My thesis bridges the fields of landscape and metacommunity ecology by explicitly considering how mechanisms at both local and landscape scales affect community structure within a given region, specifically how the presence or absence of forest cover affects processes at both scales.

1.1 Habitat Selection

Habitat selection can serve as a bridge connecting the dispersal stage and post-colonization processes (e.g. species-sorting) for actively dispersing species. Habitat selection involves the

1 2 choice and subsequent use of a habitat by an individual to survive or reproduce (Hutto, 1985; Stamps, 2009). Individuals may rely on sensory cues to identify habitat, such as auditory, visual, and olfactory cues (e.g. as in the fish Haemulon flavolineatum; Huijbers et al., 2012). Habitat selection, whether avoiding certain conditions or preferentially seeking out others, can potentially improve individual fitness through multiple mechanisms including locating higher quality habitat (for parents and/or offspring), more easily obtaining resources within sites, and a decreased predation risk (Morris, 1989; Forbes & Kaiser, 1994; Stamps et al., 2005; Zub et al., 2010; Morin, 2011). The ability to choose good quality habitats is particularly important for species with complex life histories where each life history stage uses a different environment or resource, and where one life history stage can determine the distribution of another, particularly if individuals are restricted in their dispersal at one stage. For example, the marine invertebrate Dendraster excentricus (Clypeasteroida) disperses mainly during the larval stage, with the larvae using cues from conspecific adults to inform their settlement in environments without predators (Highsmith, 1982). At the adult stage, Cassida canaliculata (Coleoptera) individuals oviposit on high quality host plants to improve the performance of their larvae, which are generally confined to one plant (Heisswolf et al., 2005).

Individuals’ habitat selection choices may, however, be evolutionary constrained, limiting their ability to evaluate differences in the quality of conditions among habitats, or their ability to detect all potentially suitable habitats (e.g. Wildermuth, 1998; Briscoe & Chittka, 2001; Bernáth et al., 2002). For example, the vision of many semi-aquatic insects relies on polarized light as a cue to detect potential habitats (Horváth, 1995). These insects may therefore only choose habitats with high levels of polarized light if they are not able to detect aquatic habitats with minimal polarized light, even if these habitats are of similar quality, or else they may be drawn to ecological traps with high levels of polarized light (e.g. gravestones, large open areas of crude oil; Horváth et al., 2007; Kriska et al., 2009). Individuals may be additionally constrained by a limited perceptual range of visual or chemical cues, which may further hamper the successful detection of suitable habitats (Lima & Zollner, 1996; Schooley & Wiens, 2003), or which may prolong the time spent searching for appropriate habitat and the associated costs of doing so (e.g. Stamps et al., 2005; Mabry & Stamps, 2008).

The effects of habitat selection on community assembly can be difficult to distinguish from those of active dispersal or species-sorting. All three mechanisms may act together or separately to

3 produce similar patterns of species distributions, and explicit tests of the role of habitat selection in shaping communities are rare (but see Foster & Tilman, 2003; Binckley & Resetarits, 2005; Vonesh et al., 2009; Knorp & Dorn, 2016). Being able to differentiate among these three drivers has important implications, not only for understanding community assembly, but also for the management and conservation of communities and ecosystems. In order to manage communities across a variety of habitats, it would be necessary to know whether immigration is bolstering species diversity at a habitat, and whether a habitat has enough resources and niches for a diversity of species (Mouillot, 2007). Management efforts to improve diversity can then focus on conserving features within a habitat, if species-sorting or mass effects are limiting species persistence based on environmental conditions within a habitat, or in the landscape surrounding habitats, if the arrival of species is being limited during dispersal or habitat selection (Mouillot, 2007).

1.2 Forest Cover Change

Forest cover, which can occur over small and large areas in the landscape, and canopy cover, or the amount of leaf and branch cover that locally affects the understory availability of sunlight, are important environmental gradients structuring terrestrial and freshwater communities. Changes to forest cover constitute a major land use change worldwide that can affect local, landscape, and regional conditions and diversity. Forest loss threatens many ecosystems and ecological communities (Pandit et al., 2007; Anderegg et al., 2013). A decline in forest cover can change the temperature, nutrient, and light regimes of terrestrial and aquatic habitats, can modify the physical structure of habitats, and can result in the displacement or extirpation of plant and animal species (Saunders et al., 1991; Sweeney et al., 2004; Richardson, 2008; García et al., 2017). However, the regrowth of forests can also impact terrestrial and aquatic communities. Forest regrowth is occurring in many regions across Europe and North America that were previously cleared for agriculture, which has since been abandoned (Foster et al., 1998; Hansen et al., 2013; Navarro & Pereira, 2015). The secondary succession of plant species following agricultural abandonment, or other disturbances, can result in very different communities than those present before or during the disturbance. Transitions in tree community composition can lead to differences in the frequency and size of gaps among trees and thus understory light availability (Clebsch & Busing, 1989) and associated air temperatures (i.e. Saunders et al., 1991).

The effects of forest regrowth on terrestrial systems are strong and well documented, however the same processes have an equal potential to affect nearby aquatic systems. Transitions in current and historical land uses can have long-standing effects on aquatic systems and communities (Harding et al., 1998; Foster et al., 2003). In lotic systems (i.e. moving waters), a decrease in riparian cover can reduce diversity, while also impairing its role as a buffer from other land uses such as logging (Davies & Nelson, 1994; Harding et al., 1998). Changes to surrounding land cover along lotic systems, such as a conversion to agriculture, can change the input of nutrients and pollutants into these environments, harming communities at the source of runoff and further downstream (Allan, 2004). In lentic systems (i.e. standing waters), changes to surrounding land cover and land use also affect aquatic communities. Gradients in tree cover surrounding lentic systems can lead to dramatic differences in nutrient levels (Kreutzweiser et al., 2008) and community composition (Werner & Glennemeier, 1999; Skelly et al., 2002; Schiesari, 2006) among habitats. Specifically, open aquatic habitats receive more light which can facilitate the growth of primary producers (Schiesari, 2006), providing more resources for higher trophic levels (Werner & Glennemeier, 1999), and are warmer environments, which can improve individual and population growth rates of some species (Skelly et al., 2002; Werner et al., 2009). Ponds with substantial canopy cover therefore have the potential to act as sink habitats for some species (Werner et al., 2009).

In addition to local impacts on terrestrial and aquatic habitats, changes in forest cover modify environments on landscape and regional scales. Forests in a landscape can either facilitate or inhibit movement among habitats. These movement responses depend on the focal taxon, and can even vary among congenerics. For example, forests can serve as movement corridors for forest-dwelling birds in fragmented landscapes (Șekercioğlu et al., 2015), and can facilitate dispersal in some salamander species that use forests to avoid desiccation while searching for aquatic habitats (Rothermel & Semlitsch, 2002). Alternatively, forests can limit the movement of organisms due to their structural complexity (e.g. of some bee species; Crall et al., 2015) and reduced understory light environment (of some butterfly species; Ross et al., 2005). Tree cover may also limit organisms’ ability to perceive potential habitats when moving through the landscape (Binckley & Resetarits, 2007). For semi-aquatic species that move through terrestrial environments to reach new habitats (e.g. dragonflies, beetles, and amphibians; Remsburg et al.,

2008; Binckley & Resetarits, 2009; Skelly et al., 2014), forests in a landscape may affect the establishment and resultant distributions of species at aquatic habitats.

1.3 Objectives and Research Questions

The objective of my thesis was to determine how forest cover in the landscape and pond canopy cover structure freshwater communities through multiple mechanisms acting at different scales, including: (1) as organisms move among habitats, (2) during habitat selection, and (3) after an individual’s colonization of a habitat (Fig. 1.1). While the effects of local environmental factors on freshwater community assembly have been explored extensively for many taxa, the effects of movement and dispersal through a heterogeneous landscape and habitat selection across environmental gradients have received far less attention, making it difficult to assess their potential contributions to structuring aquatic communities. The environmental filtering of species pools during movement, habitat selection, and post-colonization, may contribute to how well communities are connected and how communities differ in terms of species diversity or composition among habitats. Thus, examining all three mechanisms in a single system is imperative to determine their relative impact on community assembly. I used dragonflies as a study system to determine how canopy and forest cover resulted in environmental filtering at the adult terrestrial stage (i.e. during movement and habitat selection) and at the larval aquatic stage (i.e. larval growth and survival following adult colonization), and how larval dragonfly distributions might be consequently affected. Specifically, I tested the following research questions:

1. How does pond canopy cover affect adult dragonfly arrival at reproductive habitats and the performance (i.e. growth and survival) of dragonfly larvae in aquatic habitats? Does habitat selection and/or species-sorting determine the composition of larval assemblages? (Chapter 2).

2. How do open (field) versus closed (forest) matrix environments affect the small-scale movements of adult dragonflies? Are there species-specific responses in movement to forests,

Fig. 1.1 A schematic of the possible mechanisms structuring larval dragonfly assemblages in aquatic habitats. An adult dragonfly emerges from a source pond and chooses to disperse through a landscape matrix of either forest or field cover. The dragonfly then chooses an aquatic habitat at which to reproduce based on canopy cover conditions, whether due to choice or visual limitations. The resultant offspring develop in the new habitat. They may be present in open rather than closed canopy habitats because of the decisions made by the adult, or because larvae in closed habitats have not survived due to the local conditions created by the closed canopy

7 fields, and the ecotonal boundary between them? How might these behaviours scale up to patterns of adult dragonfly dispersal between pond habitats? (Chapter 3).

3. How does the interaction between conditions at the local habitat scale (i.e. pond canopy cover, hydroperiod or pond permanence, and area) and at the landscape scale (i.e. amount of forest cover and distance among ponds) affect the recovery of larval dragonfly assemblages following a disturbance from drying? (Chapter 4).

1.4 Study System

1.4.1 Study Taxon

I used dragonflies (Odonata: Anisoptera) as a focal taxon to determine the mechanisms of community assembly across a gradient of canopy cover over pond habitats and of forest cover between ponds (see Chapters 2–4). Dragonflies use different environments across their ontogeny, with larvae developing in aquatic systems, and tenerals (i.e. juveniles) and adults moving through and inhabiting terrestrial systems before returning to aquatic habitats to reproduce. Adult dragonflies are therefore instrumental in determining the developmental conditions experienced by the aquatic larvae (Remsburg & Turner, 2009). This aquatic-terrestrial linkage provides an opportunity to examine the effects of land cover on aquatic communities. Previous work in temperate regions has found that increases in pond canopy cover are associated with decreasing species richness and abundance of larval dragonflies (McCauley, 2005; McCauley et al., 2008; Werner et al., unpublished data). However, the mechanisms driving this pattern, including the effects of adult behaviour versus larval performance on species distributions, are unknown. Dragonflies from pond habitats are useful in studying metacommunity paradigms that are dependent on dispersal (i.e. Leibold et al., 2004), as adults frequently disperse to ponds distant from their natal habitat (Michiels & Dhondt, 1991; Conrad et al., 1999; Angelibert & Giani, 2003). Patterns of larval dragonfly distributions can also be well documented in natural settings, as larvae can be quick to establish at aquatic habitats across environmental gradients (e.g. McCauley et al., 2008), and larvae are easy to work with experimentally (e.g. Benke, 1970). It is therefore possible to assess the consequences of adult behaviour on larval distributions and to evaluate species-sorting mechanisms acting on larvae. In addition, dragonflies are influential in shaping community structure across their ontogeny (Wellborn et al., 1996; Knight et al., 2005), as they are predators in both terrestrial and aquatic systems (Stoks & McPeek, 2003; Olberg et

8 al., 2005). Thus, understanding the mechanisms driving dragonfly distributions may help us interpret the distribution of species at lower trophic levels.

In Chapter 2, we observed the arrival of adult dragonflies at experimental aquatic habitats (i.e. mesocosms) located within a natural landscape. Six species from two families common to Ontario, Aeshnidae and Libellulidae, were observed visiting mesocosms. Adults of 12 species were observed in the experimental area during the time period of our study, out of a regional pool of approximately 24 species that are found at our study site. Larval presence in mesocosms was limited to species in the family Libellulidae. In Chapter 4, 35 species of larvae were identified in surveys of natural habitats in Michigan, representing the families Aeshnidae, Corduliidae, Gomphidae, and Libellulidae. In Chapters 2 and 3, I used three species of dragonfly from the family Libellulidae – Leucorrhinia intacta, Sympetrum obtrusum/rubicundulum hybrids, and Sympetrum vicinum – to test for differences in larval performance and adult movement across heterogeneous pond and landscape conditions. These species were the most locally abundant during our study period and are found across a gradient of local canopy cover at ponds (i.e. 32–91% canopy cover; McCauley et al., 2008; Werner et al., unpublished data). The responses of these ‘canopy generalist’ species thus give conservative insight into how dragonflies may respond overall to forest cover.

1.4.2 Study Location

Observations from Chapters 2 and 3 were conducted at the University of Toronto’s Koffler Scientific Reserve (KSR) in King, Ontario, Canada. KSR was historically an equestrian property and estate starting in 1952, before becoming a scientific reserve in 1995. The reserve is 350 ha in size, and is now covered by a mix of fields, forests, and wetlands. There are three main ponds in the experimental area of KSR: Gazebo Pond, Barn Pond, and Dufferin Pond.

The surveys used in Chapter 4 were conducted at the University of Michigan’s Edwin S. George Reserve (ESGR) in Pinckney, Michigan, U.S.A. Before becoming a research station in 1930, the ESGR was an agricultural property with agricultural activities lasting until 1927. Secondary succession has since been allowed to occur, and trees now cover approximately 70% of the 525 ha property, in addition to fields and a number of wetlands. There are approximately 37 ponds located in the ESGR, which fall along a gradient of water permanence.

9 1.5 Goals of Chapters

To test the effects of local (i.e. canopy cover) and landscape (i.e. forest cover) conditions on the movement and distribution of dragonflies, I used a combination of mesocosm experiments, field observations, and field surveys. Through each of these techniques, I was able to assess the effects of canopy and forest cover on the movement of adult dragonflies, the arrival of adults at mesocosms, and the distribution and performance of larvae in aquatic habitats across a canopy gradient in natural and artificial settings. Specifically, I examined how a gradient of canopy cover over ponds influenced larval assemblages, and how the presence of forest versus field environments in the landscape affected the movement of dragonflies, since the type of environment between reproductive habitats may impede dragonfly movement and thus the distribution of their larvae.

1.5.1 Chapter 2

Habitat selection and species-sorting in response to habitat heterogeneity are very different mechanisms that can potentially result in similar community composition. Without experiments to differentiate their importance in a system, we may assume that species are absent from a site because they do not persist in that habitat due to environmental conditions, when in fact they may simply not have colonized that habitat in the first place. In this chapter, I tested the effects of canopy cover over experimental aquatic mesocosms on the arrival of adult dragonflies and on the survival and growth of larval dragonflies. I predicted that more adult dragonflies would arrive at open versus closed habitats, due to a greater ability to detect open habitats based on increased polarized light, and that larvae would perform better in open habitats which are more productive than closed habitats. I conducted three experiments to test these predictions. In the first experiment, twenty mesocosms were set up under open and artificially shaded treatments that simulated closed canopy cover conditions, while also artificially manipulating temperature to separately test the effects of light availability and temperature on dragonfly responses. In the second experiment, twenty mesocosms were set up under a gradient of natural canopy conditions. For these two experiments, adults were allowed to colonize the mesocosms, and the number and diversity of adult arrivals were observed. Between 2–4 weeks after the observation periods for these experiments, the number and diversity of larval offspring in mesocosms were assessed. In the third experiment, larvae were added to either 10 or 20 mesocosms under a gradient of natural canopy cover to test for differential mortality and growth among canopy

10 conditions. I found strong effects of canopy cover on the arrival of adult dragonflies to mesocosms, but no effects on the survival and growth of larval dragonflies. Habitat selection behaviour by adults may therefore play a major role in shaping larval distributions at greater scales.

This study was developed in collaboration with Shannon McCauley. I designed and performed the experiments, conducted the data analysis, and wrote the paper. Shannon McCauley co- designed the experiments and edited all versions of the manuscript. Chapter 2 has been published and has been included in this thesis with permission from Springer International Publishing. The citation is as follows:

French, S. K. & S. J. McCauley, 2018. Canopy cover affects habitat selection by adult dragonflies. Hydrobiologia 818: 129–143.

1.5.2 Chapter 3

Landscape heterogeneity can affect how organisms move among habitats. For organisms that move through environments that are compositionally dissimilar from their own habitats (i.e. matrix environments), the type of land cover between habitats can either impede or facilitate movement. In this chapter, I investigated the effects of forest versus open field matrix conditions on the movement behaviour (i.e. likelihood of movement, direction of flight, and length of time of flight) of adult dragonflies, including their responses to the boundary between fields and forests. I predicted that adult dragonflies would take flight more often in field environments than forests, would direct their movement more towards fields, and would have longer flight times in fields, as fields would be easier to manoeuvre through, and would have more light and warmer temperatures to stimulate dragonfly activity. I collected 324 adult dragonflies, representing three different species, and released them at 12 sites at KSR. Each site included release locations within a field, at a forest edge, and within a forest. I recorded whether each dragonfly took flight, and if so, recorded their first direction of movement and flight time. More individuals flew towards open field environments when released at a forest edge, but there were no differences in orientation when individuals were released in fields or forests. Individuals were marginally less likely to take flight and had shorter flight times in forests than fields, however this pattern was driven by species-specific responses to field versus forest environments. The boundary between open and closed environments elicited a strong movement response by dragonflies away from

11 forests, indicating that dragonflies likely prefer to move towards and through open environments, even though their movement abilities within each environment seem to be species-specific. Thus, maintaining a degree of openness in the landscape may facilitate the movement of adult dragonflies to reproductive habitats.

This study was developed in collaboration with Shannon McCauley. I designed and executed the study, conducted the data analysis, and wrote the paper. Shannon McCauley co-designed the study and edited all versions of the manuscript.

1.5.3 Chapter 4

Local and landscape conditions can interact to determine species distributions. Heterogeneity at habitat and landscape scales may jointly affect how individuals move through the landscape and select habitats, whereas habitat heterogeneity alone may affect post-colonization processes. The recovery of a community in habitats experiencing infrequent disturbances may depend on how dispersing and colonizing individuals respond to local and landscape conditions. In Chapter 4, I analyzed a 15-year dataset that documented the larval dragonfly assemblages found in 37 pond and wetland habitats. I used this dataset to examine the distribution of larval dragonflies in ponds in response to local conditions (i.e. pond canopy cover, hydroperiod, and area), and landscape connectivity (i.e. amount of forest cover and distance between ponds). I then tested how local and landscape factors influenced the recovery of dragonfly assemblages following a drying event. I predicted that an increase in pond canopy cover, and a decrease in pond hydroperiod/permanence and area, as well as decreased connectivity to nearby ponds, would result in lower diversity at a focal pond. Similarly, I expected that following a disturbance, ponds with more canopy cover that were less connected in the landscape would take longer to return to pre-disturbance levels of diversity and species composition. Local conditions, specifically increased canopy cover and decreased hydroperiod and pond area, resulted in decreased diversity at ponds. Local canopy cover had strong effects on larval dragonfly distributions and the recovery of larval diversity following a drying event, however the interaction between local and landscape factors also influenced diversity. These results indicate that conditions at both local and landscape scales should be considered to better understand the drivers of community assembly, and community reassembly following disturbances.

This study was developed in collaboration with Michael Benard, Jason Hoverman, Rick Relyea, David Skelly, Earl Werner, Kerry Yurewicz, and Shannon McCauley. I developed the study, conducted the data analysis, and wrote the paper. Shannon McCauley co-developed the study and edited all versions of the manuscript. Shannon McCauley and the remaining authors developed the sampling regime at the ESGR, conducted the sampling, processed and identified all larval dragonflies, and measured all of the environmental parameters at the ESGR.

1.6 References

Allan, J. D., 2004. Landscapes and riverscapes: the influence of land use on stream ecosystems. Annual Review of Ecology, Evolution, and Systematics 35: 257–284.

Anderegg, W. R. L., J. M. Kane & L. D. L. Anderegg, 2013. Consequences of widespread tree mortality triggered by drought and temperature stress. Nature Climate Change 3: 30–36.

Angelibert, S. & N. Giani, 2003. Dispersal characteristics of three odonate species in a patchy habitat. Ecography 26: 13–20.

Benke, A. C., 1970. A method for comparing individual growth rates of aquatic insects with special reference to the Odonata. Ecology 51: 328–331.

Bernáth, B., G. Szedenics, H. Wildermuth & G. Horváth, 2002. How can dragonflies discern bright and dark waters from a distance? The degree of polarisation of reflected light as a possible cue for dragonfly habitat selection. Freshwater Biology 47: 1707–1719.

Binckley, C. A. & W. J. Resetarits Jr., 2005. Habitat selection determines abundance, richness and species composition of beetles in aquatic communities. Biology Letters 1: 370–374.

Binckley, C. A. & W. J. Resetarits Jr., 2007. Effects of forest canopy on habitat selection in treefrogs and aquatic insects: implications for communities and metacommunities. Oecologia 153: 951–958.

Binckley, C. A. & W. J. Resetarits Jr., 2009. Spatial and temporal dynamics of habitat selection across canopy gradients generates patterns of species richness and composition in aquatic beetles. Ecological Entomology 34: 457–465.

Briscoe, A. D. & L. Chittka, 2001. The evolution of color vision in insects. Annual Review of Entomology 46: 471–510.

Clebsch, E. E. C. & R. T. Busing, 1989. Secondary succession, gap dynamics, and community structure in a southern Appalachian cove forest. Ecology 70: 728–735.

Connell, J. H., 1985. The consequences of variation in initial settlement vs. post-settlement mortality in rocky intertidal communities. Journal of Experimental Marine Biology and Ecology 93: 11–45.

Conrad, K. F., K. H. Willson, I. F. Harvey, C. J. Thomas & T. N. Sherratt, 1999. Dispersal characteristics of seven odonate species in an agricultural landscape. Ecography 22: 524– 531.

Crall, J. D., S. Ravi, A. M. Mountcastle & S. A. Combes, 2015. Bumblebee flight performance in cluttered environments: effects of obstacle orientation, body size and acceleration. Journal of Experimental Biology 218: 2728–2737.

Davies, P. E. & M. Nelson, 1994. Relationships between riparian buffer widths and the effects of logging on stream habitat, invertebrate community composition and fish abundance. Australian Journal of Marine and Freshwater Research 45: 1289–1305.

Forbes, L. S. & G. W. Kaiser, 1994. Habitat choice in breeding seabirds: when to cross the information barrier. Oikos 70: 377–384.

Foster, B. L. & D. Tilman, 2003. Seed limitation and the regulation of community structure in oak savanna grassland. Journal of Ecology 91: 999–1007.

Foster, D. R., G. Motzkin & B. Slater, 1998. Land-use history as long-term broad-scale disturbance: regional forest dynamics in central New England. Ecosystems 1: 96–119.

Foster, D., F. Swanson, J. Aber, I. Burke, N. Brokaw, D. Tilman & A. Knapp, 2003. The importance of land-use legacies to ecology and conservation. BioScience 53: 77–88.

Freestone, A. L. & B. D. Inouye, 2006. Dispersal limitation and environmental heterogeneity shape scale-dependent diversity patterns in plant communities. Ecology 87: 2425–2432.

García, L., W. F. Cross, I. Pardo & J. S. Richardson, 2017. Effects of landuse intensification on stream basal resources and invertebrate communities. Freshwater Science 36: 609–625.

Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L. Chini, C. O. Justice & J. R. G. Townshend. 2013. High-resolution global maps of 21st-century forest cover change. Science 342: 850–853.

Harding, J. S., E. F. Benfield, P. V. Bolstad, G. S. Helfman & E. B. D. Jones III, 1998. Stream biodiversity: the ghost of land use past. Proceedings of the National Academy of Sciences 95: 14843–14847.

Heisswolf, A., E. Obermaier & H. J. Poethke, 2005. Selection of large host plants for oviposition by a monophagous leaf beetle: nutritional quality or enemy-free space? Ecological Entomology 30: 299–306.

Highsmith, R. C., 1982. Induced settlement and metamorphosis of sand dollar (Dendraster ecentricus) larvae in predator-free sites: adult sand dollar beds. Ecology 63: 329–337.

Horváth, G., 1995. Reflection-polarization patterns at flat water surfaces and their relevance for insect polarization vision. Journal of Theoretical Biology 175: 27–37.

Horváth, G., P. Malik, G. Kriska & H. Wildermuth, 2007. Ecological traps for dragonflies in a cemetery: the attraction of Sympetrum species (Odonata: Libellulidae) by horizontally polarizing black gravestones. Freshwater Biology 52: 1700–1709.

Huijbers, C. M., I. Nagelkerken, P. A. C. Lössbroek, I. E. Schulten, A. Siegenthaler, M. W. Holderied & S. D. Simpson, 2012. A test of the senses: fish select novel habitats by responding to multiple cues. Ecology 93: 46–55.

Hutto, R. L., 1985. Habitat selection by nonbreeding, migratory land birds. In Cody, M. L. (ed), Habitat selection in birds. Academic Press, Orlando FL: 455–476.

Knight, T. M., M. W. McCoy, J. M. Chase, K. A. McCoy & R. D. Holt, 2005. Trophic cascades across ecosystems. Nature 437: 880–883.

Knorp, N. E. & N. J. Dorn, 2016. Mosquitofish predation and aquatic vegetation determine emergence patterns of dragonfly assemblages. Freshwater Science 35: 114–125.

Kreutzweiser, D. P., P. W. Hazlett & J. M. Gunn, 2008. Logging impacts on the biogeochemistry of boreal forest soils and nutrient export to aquatic systems: a review. Environmental Reviews 16: 157–179.

Kriska, G., B. Bernáth, R. Farkas & G. Horváth, 2009. Degrees of polarization of reflected light eliciting polarotaxis in dragonflies (Odonata), mayflies (Ephemeroptera) and tabanid flies (Tabanidae). Journal of Insect Physiology 55: 1167–1173.

Leibold, M. A., M. Holyoak, N. Mouquet, P. Amarasekare, J. M. Chase, M. F. Hoopes, R. D. Holt, J. B. Shurin, R. Law, D. Tilman, M. Loreau & A. Gonzalez, 2004. The metacommunity concept: a framework for multi-scale community ecology. Ecology Letters 7: 601–613.

Lima, S. L. & P. A. Zollner, 1996. Towards a behavioral ecology of ecological landscapes. Trends in Ecology & Evolution 11: 131–135.

Logue, J. B., N. Mouquet, H. Peter, H. Hillebrand, P. Declerck, A. Flohre, S. Gantner, N. Gülzow, P. Hörtnagl, S. Meier & B. Pecceu, 2011. Empirical approaches to metacommunities: a review and comparison with theory. Trends in Ecology & Evolution 26: 482–491.

Mabry, K. E. & J. A. Stamps, 2008. Dispersing brush mice prefer habitat like home. Proceedings of the Royal Society B 275: 543–548.

McCauley, S. J., 2005. Species distributions in anisopteran odonates: effects of local and regional processes. PhD thesis. University of Michigan, Ann Arbor MI.

McCauley, S. J., C. J. Davis, R. A. Relyea, K. L. Yurewicz, D. K. Skelly & E. E. Werner, 2008. Metacommunity patterns in larval odonates. Oecologia 158: 329–342.

Michiels, N. K. & A. A. Dhondt, 1991. Characteristics of dispersal in sexually mature dragonflies. Ecological Entomology 16: 449–459.

Morin, P. J., 2011. Community ecology, 2nd ed. Wiley-Blackwell, Chichester UK.

Morris, D. W., 1989. Density-dependent habitat selection: testing the theory with fitness data. Evolutionary Ecology 3: 80–94.

Mouillot, D., 2007. Niche-assembly vs. dispersal-assembly rules in coastal fish metacommunities: implications for management of biodiversity in brackish lagoons. Journal of Applied Ecology 44: 760–767.

Navarro, L. M. & H. M. Pereira, 2015. Rewilding abandoned landscapes in Europe. In Pereira, H. M. & L. M. Navarro (eds), Rewilding European landscapes. Springer, New York NY: 3–23.

Olberg, R. M., A. H. Worthington, J. L. Fox, C. E. Bessette & M. P. Loosemore, 2005. Prey size selection and distance estimation in foraging adult dragonflies. Journal of Comparative Physiology A 191: 791–797.

Pandit, M. K., N. S. Sodhi, L. P. Koh, A. Bhaskar & B. W. Brook, 2007. Unreported yet massive deforestation driving loss of endemic biodiversity in Indian Himalaya. Biodiversity and Conservation 16: 153–163.

Remsburg, A. J., A. C. Olson & M. J. Samways, 2008. Shade alone reduces adult dragonfly (Odonata: Libellulidae) abundance. Journal of Insect Behavior 21: 460–468.

Remsburg, A. J. & M. G. Turner, 2009. Aquatic and terrestrial drivers of dragonfly (Odonata) assemblages within and among north-temperate lakes. Journal of the North American Benthological Society 28: 44–56.

Richardson, J. S., 2008. Aquatic arthropods and forestry: effects of large-scale land use on aquatic systems in Nearctic temperate regions. Canadian Entomologist 140: 495–509.

Ricketts, T. H., 2001. The matrix matters: effective isolation in fragmented landscapes. The American Naturalist 158: 87–99.

Ross, J. A., S. F. Matter & J. Roland, 2005. Edge avoidance and movement of the butterfly Parnassius smintheus in matrix and non-matrix habitat. Landscape Ecology 20: 127–135.

Rothermel, B. B. & R. D. Semlitsch, 2002. An experimental investigation of landscape resistance of forest versus old-field habitats to emigrating juvenile amphibians. Conservation Biology 16: 1324–1332.

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

Schiesari, L., 2006. Pond canopy cover: a resource gradient for anuran larvae. Freshwater Biology 51: 412–423.

Schooley, R. L. & J. A. Wiens, 2003. Finding habitat patches and directional connectivity. Oikos 102: 559–570.

Șekercioğlu, Ç. H., S. R. Loarie, F. Oviedo-Brenes, C. D. Mendenhall, G. C. Daily & P. R. Ehrlich, 2015. Tropical countryside riparian corridors provide critical habitat and connectivity for seed-dispersing forest birds in a fragmented landscape. Journal of Ornithology 156: 343–353.

Skelly, D. K., S. R. Bolden & L. K. Freidenburg, 2014. Experimental canopy removal enhances diversity of vernal pond amphibians. Ecological Applications 24: 340–345.

Skelly, D. K., L. K. Freidenburg & J. M. Kiesecker, 2002. Forest canopy and the performance of larval amphibians. Ecology 83: 983–992.

Stamps, J. 2009. Habitat selection. In Levin, S. A. (ed), The Princeton guide to ecology. Princeton University Press, Princeton NJ: 38–44.

Stamps, J. A., V. V. Krishnan & M. L. Reid, 2005. Search costs and habitat selection by dispersers. Ecology 86: 510–518.

Stoks, R. & M. A. McPeek, 2003. Predators and life histories shape Lestes damselfly assemblages along a freshwater habitat gradient. Ecology 84: 1576–1587.

Sweeney, B. W., T. L. Bott, J. K. Jackson, L. A. Kaplan, J. D. Newbold, L. J. Standley, W. C. Hession & R. J. Horwitz, 2004. Riparian deforestation, stream narrowing, and loss of stream ecosystem services. Proceedings of the National Academy of Sciences 101: 14132–14137.

Urban, M. C., 2004. Disturbance heterogeneity determines freshwater metacommunity structure. Ecology 85: 2971–2978.

Vonesh, J. R., J. M. Kraus, J. S. Rosenberg & J. M. Chase, 2009. Predator effects on aquatic community assembly: disentangling the roles of habitat selection and post-colonization processes. Oikos 118: 1219–1229.

Wellborn, G. A., D. K. Skelly & E. E. Werner, 1996. Mechanisms creating community structure across a freshwater habitat gradient. Annual Review of Ecology and Systematics 27: 337–363.

Werner, E. E. & K. S. Glennemeier, 1999. Influence of forest canopy cover on the breeding pond distributions of several amphibian species. Copeia 1999: 1–12.

Werner, E. E., R. A. Relyea, K. L. Yurewicz, D. K. Skelly & C. J. Davis, 2009. Comparative landscape dynamics of two anuran species: climate-driven interaction of local and regional processes. Ecological Monographs 79: 503–521.

Wildermuth, H., 1998. Dragonflies recognize the water of rendezvous and oviposition sites by horizontally polarized light: a behavioural field test. Naturwissenschaften 85: 297–302.

Zub, K., E. Pugacewicz, B. Jędrzejewska & W. Jędrzejewski, 2010. Factors affecting habitat selection by breeding Lesser Spotted Eagles Aquila pomarina in northeastern Poland. Acta Ornithologica 45: 105–114.

Chapter 2 Canopy Cover Affects Habitat Selection by Adult Dragonflies

This paper is published as: French, S. K. & S. J. McCauley, 2018. Canopy cover affects habitat selection by adult dragonflies. Hydrobiologia 818: 129–143.

Abstract

The mechanisms structuring aquatic communities across environmental gradients are often difficult to distinguish from one another and can produce similar patterns of species distributions. In freshwater systems, the amount of canopy cover from surrounding trees is often associated with transitions in local community structure. These community changes could be driven by habitat selection prior to colonization of the aquatic habitat and/or species-sorting post- colonization. To assess the contributions of pre- versus post-colonization processes in structuring larval dragonfly assemblages, we tested the impact of artificial and natural canopy cover on the selection of experimental aquatic mesocosms by adult dragonflies, and monitored the performance (i.e. growth and survival) of larval dragonflies that were placed in mesocosms under a gradient of natural canopy cover. We found that greater levels of canopy cover resulted in fewer adult visits to mesocosms, and more natural canopy cover decreased the species richness of visitors. There were no effects of canopy cover on the growth and survival of larvae added to the mesocosms. Our results suggest that adult habitat selection plays a dominant role in structuring larval dragonfly assemblages across a canopy cover gradient, and that canopy cover can be an important environmental filter on species distributions.

2.1 Introduction

Heterogeneous landscapes can affect species’ establishment within habitats and their resultant distributions across environmental gradients by influencing dispersal, individuals’ choice of habitat, and post-colonization processes (Lima & Zollner, 1996; Leibold et al., 2004; Keller et al., 2012). Animals dispersing to new patches may select habitats across gradients of available resources, refuges, or other environmental conditions (e.g. Boyce et al., 2003; Rieger et al., 2004). Alternatively, animals may have a limited ability to detect particular habitats (Schooley & Wiens, 2003; Schtickzelle et al., 2007). Following the colonization of a habitat, colonist or

19 20 offspring persistence can depend on differential performance (e.g. variable mortality) along environmental gradients, affecting community assembly and structure in the process (i.e. species- sorting; Leibold et al., 2004; Logue et al., 2011).

Studies on community assembly across heterogeneous habitats generally focus on two mechanisms, dispersal and species-sorting, while overlooking habitat selection as a critical link between the two (i.e. Fig. 2.1; Binckley & Resetarits, 2005; Resetarits, 2005). Individuals can choose and establish at habitats that they perceive to be of high quality (Binckley & Resetarits, 2005, 2007). This individual-level behavioural strategy of selecting habitats potentially minimizes the risks posed by colonizing low-quality habitats, and decreases the extent to which post-colonization species-sorting acts on community structure. Habitat selection to avoid predators occurs in a number of groups (e.g. Binckley & Resetarits, 2003, 2005; Brodin et al., 2006). For example, some Chaoborus spp. (Diptera: Chaoboridae) use chemical cues to avoid reproductive habitats with fish predators, where their offspring would experience greater mortality (Berendonk, 1999; Garcia & Mittelbach, 2008). However, not all animals can detect predators within the juvenile habitat (e.g. adult odonates; McPeek, 1989; McCauley, 2005a; Knorp & Dorn, 2016), and consequently may be limited in their ability to use habitat selection to improve post-colonization performance. The extent to which colonists acquire and use information about habitat quality, whether based on predators or other factors, is an important determinant of whether pre- or post-colonization processes dominate in shaping species distributions across environmental gradients.

In aquatic species with a dispersing terrestrial stage, a common life-history for many aquatic insects, the conditions in the terrestrial environment at both local and regional scales are important for adult dispersal and colonization, whereas local aquatic conditions determine larval performance (Rothermel & Semlitsch, 2002; Remsburg & Turner, 2009). The focal taxa in our study, dragonflies (i.e. infraorder Anisoptera), have an aquatic larval stage and a terrestrial adult stage, with habitat choices by adults potentially leading to reproduction across a gradient of habitat quality (Switzer, 2002). Adult dragonflies disperse and may select aquatic habitats in response to visual cues from both aquatic and terrestrial environments (Bernáth et al., 2002; Corbet, 2004; Remsburg et al., 2008; Knorp & Dorn, 2016). Habitat selection by adults can

Fig. 2.1 Species selecting habitat following dispersal. Adults of species A, B, and C, disperse from a source habitat and select between two reproductive habitats (X and Y) of varying quality. Adults of A select X, adults of C select Y, and adults of B select both habitats. The offspring of A and B survive in X, and those of C survive in Y. Unsuitable conditions in Y lead to mortality in the offspring of B due to species-sorting. Without observations of both adult habitat selection and offspring species-sorting, we could not determine the mechanism generating patterns of offspring distributions. For example, we might infer that offspring of A and B cannot persist in Y. However, adults are simply not arriving at Y, and we have no indication of whether their offspring could survive in Y

22 occur across a hierarchy of spatial scales: among ponds at the landscape scale, to among sites within a single pond (Buchwald, 1994 in Corbet, 2004; Wildermuth, 1994 in Corbet, 2004).

Forest canopy gradients provide an important context in which to examine the effects of adult habitat selection versus larval performance on community structure in aquatic systems. Forest cover has dramatically changed in many regions of North America, including widespread deforestation in some regions (Hansen et al., 2013) and significant forest regrowth in others (Foster et al., 1998). Forest regrowth following the abandonment of agricultural land in parts of eastern North America has led to the highest levels of forest cover in 200 years (Foster et al., 1998). Reforestation in this region is particularly likely to have large effects on small freshwater habitats such as ponds, as these habitats can be more fully shaded than lakes. Canopy cover over aquatic habitats can decrease light availability and water temperature, while increasing organic input, all of which can impact community structure within these habitats (Richardson, 2008). In temperate regions, tree cover near aquatic habitats can also restrict visitation and colonization rates at ponds by amphibians (Skelly et al., 2014), adult dragonflies (Remsburg et al., 2008), and beetles (Binckley & Resetarits, 2009). Canopy cover may decrease the amount of polarized light reflected from aquatic surfaces, a visual cue commonly used by dragonflies to detect oviposition habitats (Wildermuth, 1998), as well as the amount of sunlight available for adult thermoregulation (Horváth et al., 2007; Remsburg et al., 2008; De Marco et al., 2015). Low light availability at these habitats may affect the composition of a dragonfly assemblage by constraining adults’ detection of habitat or limiting their persistence in the area, thus reducing colonization success (e.g. Remsburg et al., 2008; De Marco et al., 2015). In highly shaded ponds, eggs and larvae may develop more slowly due to lower temperatures (Pritchard et al., 1996; Corbet, 2004; Frances et al., 2017), potentially decreasing survival in these often desiccation- prone habitats.

Greater levels of canopy cover are also associated with changes in larval dragonfly assemblages. In temperate regions in the northern hemisphere, where more sunlight is reflected off surfaces than in tropical regions, dragonfly species that are not generally forest associated experience reduced larval abundances and diversity with increasing canopy cover (McCauley, 2005a; McCauley et al., 2008; Werner et al., unpublished data). However, it is difficult to discern whether reduced dragonfly diversity in closed habitats is driven by adult habitat selection or by

23 poorer performance (i.e. growth and survival) and species-sorting imposed on larvae by within- pond habitat conditions. Indeed, empirical studies aimed at teasing apart these two mechanisms in freshwater systems have been limited (e.g. Binckley & Resetarits, 2005; Vonesh et al., 2009; Knorp & Dorn, 2016). Our study assessed the extent to which larval dragonfly distributions were related to adult habitat selection behaviour versus differences in larval performance across varying canopy conditions. Because polarized light is important in the visual detection of aquatic habitats by dragonflies (Wildermuth, 1998; Briscoe & Chittka, 2001; Bernáth et al., 2002), we predicted that more adults would visit open versus closed canopy habitats, resulting in more larvae in open canopy habitats. We also expected that larvae would have greater survival and growth rates in open canopy habitats because they are warmer and more productive (Schiesari, 2006). Adult arrivals (abundance and diversity) at a habitat were therefore expected to be strongly related to canopy cover, with subsequent larval distributions being principally shaped by the arrival of adult colonists, but reinforced by improved larval performance in open environments.

2.2 Materials and Methods

To determine how adult dragonflies respond to canopy cover while selecting aquatic habitats, field experiments using aquatic mesocosms were performed at the University of Toronto’s Koffler Scientific Reserve (KSR; King City, ON, Canada, 44.03 N, 79.53 W). Approximately 24 species of dragonflies are found at KSR (French & McCauley, unpublished data). Three separate field experiments tested whether shading (i.e. artificial cover) and canopy cover (i.e. natural tree cover) reduced the abundance and diversity of adult and larval dragonflies at artificial ponds. Each experiment used twenty mesocosms (100-gallon cattle tanks) as artificial ponds for potential dragonfly visitors during each experiment. Mesocosms were filled with filtered pond water, and in the case of the shading experiment, covered with netting until observations began to prevent aquatic insects from colonizing prior to starting the experiment. After approximately 25 days for the shading experiments and 1 day for the natural canopy experiments, all mesocosms were inoculated once with zooplankton collected from Gazebo Pond (Fig. 2.2), including cladocerans and copepods, to serve as a food source, particularly for early insect colonists. Four pieces of rabbit food were also added to serve as a nutrient base for algal growth for both experiments, and two handfuls of leaves collected from the terrestrial environment were

Fig. 2.2 Map of experimental sites at Koffler Scientific Reserve, with the arrangement of mesocosms used for the artificial shading, natural canopy, and larval performance experiments. Landscape cover around the experiments is shown, as are the potential source ponds (Gazebo Pond and Barn Pond)

25 added to the mesocosms of the shading experiments to serve as a nutrient base and to add structural complexity (these are standard conditions for colonization arrays; e.g. McCauley, 2006). Other aquatic invertebrates were allowed to naturally colonize the mesocosms, which also provided a food source for larval dragonflies. An artificial perch (a 1.2-m-long square dowel weighted at the bottom and extending past the water surface) for adult dragonflies to land on was added to each mesocosm (McCauley, 2006). Artificial vegetation was added to provide structural complexity, and consisted of two frayed rope bundles approximately 60 cm in length that were weighted to the mesocosm bottom and extended to the water surface, which larval dragonflies have been shown to use (McCauley, 2006, 2008). Observations started approximately 20 days after the zooplankton inoculation for shading experiments and five days for the natural canopy experiments. All observations of adult dragonflies, both at mesocosms and ponds, were conducted by sight and using binoculars, and the number of individuals and their species and sex were identified when possible.

Mesocosms were located near two ponds that contain fish (Gazebo Pond and Barn Pond; Fig. 2.2), which were potential source habitats. One fishless pond is also present at KSR (Dufferin Pond, 509 m from nearest fish pond), but was a less likely source of visitors due to isolation by distance. There was an open field landscape between Gazebo/Barn Ponds and the mesocosm locations, making access to mesocosms under varying levels of canopy cover more standardized. Only the nearest pond for each experiment was surveyed for dragonflies, since identifying the regional pool of potential visitors was prioritized over knowing their exact source, and visitors may have stopped over at the nearest pond before visiting the mesocosms. Additionally, the larval and adult dragonfly assemblages at these two ponds with fish appeared to be identical (French & McCauley, personal observation).

2.2.1 Artificial Shading Experiment

To test the effects of artificial shading, water temperature, and their interaction, on the arrival of dragonfly visitors and subsequent colonization, we manipulated both in a fully factorial design. This setup allowed us to isolate the effects of shading (i.e. light availability and temperature) on dragonfly distributions from other water quality parameters influenced by canopy (e.g. nutrient input). There were four mesocosm treatments consisting of shaded, heated, shaded plus heated, and no shade or heat (i.e. control), and each treatment was replicated five times. Four rows, with

26 five mesocosms per row, were set up parallel to and moving away from a boundary of nearby trees (approximately 7.1 m away from the nearest row). Four replicates of each treatment were randomly assigned to a mesocosm in each row, such that at least one of every treatment was found in every row. A fifth replicate of each treatment was randomly assigned to the remaining mesocosms, independent of row. The trees did not shade the mesocosms during peak hours, and only shaded approximately a third of each mesocosm in the final row by 16h00. These conditions limited the potential for edge effects but nonetheless we wanted to control for any effects of this extra shading. Mesocosms in the shaded treatment had light blocked by dark green shade cloth, which reduces light penetration by 50% (Fig. S2.1 in Appendix A: Supplementary Materials for Chapter 2; note that this photo was taken prior to starting the experiment when mesocosms were still covered by netting). Mesocosms in the open canopy treatment had structure above them similar to that of the light-blocking cloth in the shade treatment but made of an open weave netting that allowed most incoming sunlight to reach the water’s surface (Fig. S2.1). The overhang from both covers was approximately 1.8 m off the ground. The shade cloth shaded an entire mesocosm during hours of peak dragonfly activity and allowed adults to fly underneath, as flight paths of < 2 m above the ground have been observed in the field (Remsburg et al., 2008). Adults flew beneath both open and closed canopy cover, and were not observed to fly through the open netting; thus the weave density of the shade cloth likely did not deter dragonflies from arriving at the mesocosms. In addition to the potential naturally occurring effects of shading on water temperature, temperature was artificially manipulated with aquarium heaters (300 W, EHEIM JÄGER thermostatic heater, Deizisau, Germany) to test whether the warmer temperatures associated with open canopy conditions were preferred by adults, potentially because of positive effects of temperature on larval performance (e.g. faster growth rates, accelerated development; Flenner et al., 2010), and to identify any confounding influence of shading on pond temperature.

The mesocosms were observed by researchers for visiting adult dragonflies. A dragonfly was considered to be visiting a mesocosm if it was observed perching, hovering, or dipping at a mesocosm. Adults flying through the study area without stopping at a mesocosm were counted separately. We were conservative in our estimates of females by considering all adults independent of sex, because the arrivals were generally male-biased, a pattern which has been observed in other studies using mesocosms as artificial ponds (McCauley, 2006), and many

27 females could not be definitively identified because in many species they resemble juvenile males. However, the abundance of males at a site can be positively related to the number of visiting ovipositing females and mating pairs (Van Buskirk, 1986; Corbet, 1962 in Horváth et al., 1998), and in a previous colonization experiment using mesocosms, male-biased adult arrivals predicted larval diversity (McCauley, 2006). In some instances, males arrive at aquatic habitats before females (Jacobs, 1955; Switzer & Walters, 1999) and remain for longer (Foster & Soluk, 2006). Males may also perch near oviposition sites and intercept approaching females (Jacobs, 1955; Switzer & Walters, 1999), or escort females to an oviposition site (Michiels & Dhondt, 1990). In addition, males and females likely use similar polarized light cues to find pond habitats where they encounter each other (e.g. Horváth et al., 1998; Wildermuth, 1998).

Observations were conducted twice a week for 6 weeks from July to August 2013, during hours of peak dragonfly activity (11h30–15h00), and when weather was suitable for dragonfly flight (no rain, zero to light cloud cover, and low wind). An observation period lasted an hour and involved the same two observers (SKF and SVLC) monitoring each of the four rows for 15 min. Two observers were situated at opposite ends of the rows, with each starting at a random row at the beginning of the period and continuing in one of two randomized directions. Two observation periods took place per day, for a total of 48 person-h of observations over 12 days. In between observation periods, we conducted a 40 person-min survey of adult dragonflies at the nearest source pond (Gazebo Pond, 39 m from nearest mesocosm) to determine the regional species pool, as even 20–40-min weekly surveys of adults can capture 87–96% of the representative species pool (Bried et al., 2012). The mesocosms were sampled for live dragonfly larvae and other colonizing insects (i.e. potential prey) after 10 weeks, with standard collection techniques across mesocosms. Prey density was calculated based on the total number of individuals from the most abundant prey families in a subsample (i.e. Chaoboridae, Chironomidae, and Baetidae). Water temperatures were measured in between observation periods, within the hours of 12h30–13h40, using an OAKTON pH/CON 10 meter (Vernon Hills, Illinois, USA). Water temperatures (average of measurements taken from the bottom and top of

2 mesocosms) were different among treatments (Kruskal–Wallis rank sum test, c 3 = 16, p = 0.001). A Nemenyi post hoc test (PMCMR package; Pohlert, 2014) indicated that shaded treatments (22.7 ± 0.2°C) had lower temperatures than heated treatments (26.6 ± 0.3°C; p = 0.002) and shaded plus heated treatments (26.0 ± 0.3°C; p = 0.03), but did not differ from control

28 treatments (23.3 ± 0.1°C; p = 0.7). Heated treatments were marginally warmer than control treatments (no shade or heat; p = 0.05). Shaded plus heated treatments did not differ from control (p = 0.3) or heated treatments (p = 0.9). When treatments were pooled based on temperature manipulation, heated treatments (26.3 ± 0.2°C) had greater temperatures than non-heated treatments (23.0 ± 0.1°C; Wilcoxon rank-sum test, W = 100, p < 0.001). Heating treatments were therefore deemed effective.

We tested the effects of the presence and/or absence of heating and shading on the number and species richness of adult dragonfly visits to the mesocosms, and on the number, richness, and diversity (using the Shannon–Wiener index) of larvae within mesocosms. Separate generalized linear mixed-effect models (GLMMs; lme4 package; Bates et al., 2015) were run with adult abundance and species richness using a Poisson distribution, and with larval diversity using a Gaussian distribution. We also tested the effects of our treatments on larval abundance and richness using a GLMM with a negative binomial distribution due to overdispersion (glmmADMB package; Fournier et al., 2012; Skaug et al., 2012). Row was included as a random factor in each test. GLMM models with and without an interaction term were compared using likelihood ratio tests and the simplest model was selected. We separately tested the effect of prey density on larval abundance using a GLMM with a negative binomial distribution. All analyses were conducted in R (version 3.4.0; R Core Team, 2017).

2.2.2 Natural Canopy Experiment

To test the effects of a natural canopy cover gradient on the arrival of dragonfly visitors and subsequent colonization, mesocosms were placed under a range of canopy cover levels. These levels were based on natural ranges (32–91% cover) measured at the Edwin S. George Reserve (ESGR) in southeast Michigan, where earlier long-term surveys of larval dragonfly assemblages found decreased abundances and diversity with increasing forest canopy cover (McCauley, 2005a; McCauley et al., 2008; Werner et al., unpublished data). This experiment allowed us to consider several components of canopy cover that our previous shading experiments did not capture (e.g. nutrient input). Twenty mesocosms were placed across five sites. At each site, one mesocosm was situated under each level of 30–40, 55–60, 70–75, and 90–100% canopy cover. Canopy cover above each mesocosm was estimated using a spherical densiometer (Model-C, Forest Densiometers, Rapid City, South Dakota, USA). Each mesocosm was observed for

29 visiting adult dragonflies for 10 min by a single observer (SKF) approximately two days a week, depending on weather. Observations were repeated for a total of 10 weeks between July and September 2014, resulting in 70 h of observations over 21 days across all mesocosms. The nearest source pond (Barn Pond, 112 m from nearest mesocosm) was also surveyed for adults halfway through each observation day, for a period of 20 min (timing based on Bried et al., 2012). Again, visits were male-biased, but were assumed to predict the number of ovipositing females arriving (Van Buskirk, 1986; Corbet, 1962 in Horváth et al., 1998) and larval diversity (McCauley, 2006). The mesocosms were sampled for live dragonfly larvae and other colonizing insects (i.e. potential prey) after 12 weeks, with standard collection techniques across mesocosms. Prey density was calculated based on the total number of individuals from the most abundant prey families in a subsample (i.e. Culicidae, Chironomidae, and Baetidae).

I-Buttons (DS1921G-F50 Thermochron, Maxim Integrated, San Jose, California, USA) were placed in two mesocosms per site to measure water temperature, resulting in 2–3 i-Buttons per level of canopy cover, between July and September 2014. The i-Button in one mesocosm malfunctioned and the data could not be recovered. The i-Buttons collected data every 2 h between 10h15 and 18h50, resulting in approximately four measurements per day. The temperatures from a single canopy level treatment were averaged across all sites and days for subsequent analyses. Water chemistry was measured in July–September 2014 using an Exo Sonde Probe (EXO2, YSI Inc., Yellow Springs, Ohio, USA): chlorophyll was averaged over 3 days, turbidity over 2 days, and dissolved oxygen, pH, and conductivity were measured once.

GLMMs tested the effects of natural canopy on the number and species richness of adult dragonfly visits to the mesocosms using a Poisson distribution, and on the presence/absence of larvae within mesocosms using a binomial distribution. Separate GLMMs tested the effects of canopy and prey density on larval abundance using a negative binomial distribution due to overdispersion (glmmADMB package). Site was included as a random factor. GLMM models were compared using likelihood ratio tests and the simplest model was selected.

A principal component analysis (PCA) was conducted using the princomp function to assess variation among mesocosms based on dissolved oxygen, pH, conductivity, chlorophyll, and water temperature, after scaling the variables. The first principal component axis explained 75%

30 of the variance among mesocosms, with moderate to strong loadings (|0.36–0.5|; Table S2.1), and was the only axis retained based on a scree plot of the components. The PCA grouped 30–40 and 55–60% canopy cover separately from 70–75 and 90–100% canopy cover based on water chemistry (Fig. S2.2), indicating that the extreme levels of canopy were different. These results were confirmed using a one-way ANOVA on the PC1 component scores (F3 = 17, p < 0.001; see Table S2.2 for all pairwise comparisons using Tukey’s test); however, in this case, 55–60 and 70–75% canopy cover were marginally similar (p = 0.09), as were 70–75 and 90–100% canopy cover (p = 0.07).

2.2.3 Larval Performance Experiments

To determine how larval performance was affected by conditions within aquatic habitats that vary with canopy cover, we conducted experiments testing performance among mesocosms distributed across the same gradient of canopy cover as in the natural canopy experiment described above. These experiments allowed us to better assess the effects of post-colonization processes (i.e. growth and survival) on larval presence and abundance. Three species within the family Libellulidae, that occur across the shaded/unshaded gradient of canopy cover (McCauley et al., 2008), were placed in cages made of insect screening within mesocosms and monitored for survival and growth (measured as head width, a standard metric of larval odonate size). Growth was calculated as ln((final–initial head width) + 1) divided by the number of days of each trial (e.g. McCauley, 2005b).

For the first trial, larval Leucorrhinia intacta were collected in September 2014 from Dufferin Pond at KSR and caged in the 20 mesocosms from the natural canopy experiment (under levels of 31–96% natural canopy cover). Across this performance trial, the mesocosms under 30–40 and 55–60% canopy cover had warmer water temperatures (10.5 ± 0.1 and 10.1 ± 0.1°C, respectively) than the mesocosms under 70–75 and 90–100% canopy cover (8.9 ± 0.1 and 9.0 ±

2 0.1°C, respectively; Kruskal–Wallis rank sum test, c 3 = 144, p < 0.001; p < 0.001 for all significant pairwise comparisons based on a Nemenyi post hoc test). Three cages containing L. intacta were placed in each mesocosm. One cage, which was approximately 89 cm ´ 75 cm in size with a plastic colander and dowels providing structural support within the cage, and which contained two pieces of artificial vegetation ranging between 10 and 20 cm long, held five

31 larvae. The remaining two cages, which were approximately 25 cm ´ 25 cm in size with support provided by PVC piping with screening on one end, and which contained one piece of artificial vegetation, contained a single larva each. Larvae were checked biweekly for survival and photographed over a 5-week period from September to November 2014. Afterwards, head widths were measured from these photographs in ImageJ (version 1.46r; Schneider et al., 2012). For the cages containing five individuals, average growth was calculated. This value was then taken with the growth of individuals to determine average growth for each mesocosm. Growth across canopy cover levels was tested using a one-sided Wilcoxon signed-rank test.

A second experiment on larval performance was conducted in the summer of 2015 when water temperatures were warmer, to test for differences in survival and growth at the extremes of natural canopy cover (31–37 and 90–96%), which were shown to have different water chemistry conditions (Fig. S2.2). Across these performance trials, the five mesocosms under open canopy had warmer average water temperatures (26.1 ± 0.3°C) than the five closed canopy mesocosms (18.0 ± 0.3°C; Wilcoxon rank-sum test, W = 2500, p < 0.001), a pattern also observed among natural ponds that vary in canopy cover levels (Werner & Glennemeier, 1999; Skelly et al., 2002). Sympetrum hybrids (i.e. combinations of Sympetrum obtrusum, Sympetrum rubicundulum, and possibly Sympetrum internum, which can all hybridize; Catling, 2007), which are found in more temporary ponds (McCauley et al., 2008), were collected in May–June 2015 from Dufferin Pond and placed in mesocosms. Larvae were checked weekly for survival and their head widths measured using digital calipers over a 4-week period from June to July 2015. Sympetrum vicinum, which occur across a wide range of pond permanence levels (McCauley et al., 2008), were collected in July 2015 from Gazebo Pond and placed in mesocosms. Larvae were checked weekly for survival and their head widths measured over a 4-week period from July to August 2015. In these two trials, three 25 cm ´ 25 cm cages, containing one larva each, were placed in a total of 10 mesocosms. Water temperatures were measured at the mesocosm bottoms within the hours of 15h00–17h30. The number of days between the beginning of the experiment and the date when larvae began their eclosion process (i.e. emergence time) was also recorded. We tested for differences in growth using Wilcoxon rank-sum tests, and differences in emergence times using two sample t tests, between open (31–37% canopy) and closed (90–96% canopy)

32 mesocosms for each species. Growth across canopy cover levels was tested using one-sided Wilcoxon signed-rank tests.

2.3 Results

Across experiments, adult dragonflies showed strong habitat selection responses to canopy cover and shading. However, these conditions had no effect on larval presence, abundance, or performance.

2.3.1 Artificial Shading Experiment

In the artificial shading experiment, the nearest source pond had 11 species present over the course of the experiment, with all 11 observed in the experimental area, but only six species visiting the mesocosms (Table S2.3). Mesocosm conditions are summarized in Table S2.4. Of the individuals visiting the mesocosms (N = 23), 11 were male, 4 were female, 6 were female or teneral (juveniles in which sex cannot be definitively determined, as females and male tenerals share similar colouring), and the sex of 2 could not be identified. Five individuals dipped, 16 perched, and 2 individuals did both at mesocosms. Under artificial shading (50% shade cloth), the mean number of adult visits per mesocosm was 28% lower than at open canopy mesocosms (p = 0.008; Fig. 2.3a, Table 2.1). However, shading only marginally lowered the species richness of visitors (p = 0.08; Table 2.1). Heating did not affect visitor abundance or species richness (Table 2.1).

Only three species of larvae (Libellulidae) were found within the mesocosms compared to six species of adults visiting the mesocosms. Libellula pulchella larvae were found in all 20 mesocosms, Libellula luctuosa larvae were found in 17 mesocosms, and Tramea lacerata were found in 2 mesocosms. Heating and shading treatments did not affect larval abundance (Fig. 2.3b), richness, or diversity (p > 0.1 in all cases; Table 2.1). Prey density was not related to larval abundance (p = 0.08; Table 2.1).

2.3.2 Natural Canopy Experiment

In the natural canopy experiment, we observed eight species at the nearest source pond over the course of the experiment, 10 species in the experimental area where mesocosms were located, and six species visiting the mesocosms (Table S2.3). Mesocosm conditions are summarized in

Fig. 2.3 (a) Mean number of visits (± 1 SE) by adult dragonflies and (b) mean abundance of larvae per mesocosm, based on a fully factorial design of no heating (grey circle) and heating (black circle), and shaded and open treatments. Open treatments had more adult visits than shaded treatments (p = 0.008), but no differences were found between temperature treatments or in larval responses

Table 2.1 Statistical test results for the artificial shading experiments. Each test was conducted using a GLMM, with 20 observations per test

Response Variables Mean Estimate Standard error z value p value Present | Absent

Table S2.5. Of the individuals visiting the mesocosms (N = 39), 13 were male, 2 were female, 2 were female or teneral, and 4 were mating pairs (where 1 pair = 1 visit). The sex of 18 Aeshna individuals could not be determined. Three individuals dipped, 23 hovered, 11 perched, and 2 individuals both hovered and perched at mesocosms. In this experiment, an increase in natural canopy cover was associated with a significant decrease in the number of adult visits per mesocosm (p = 0.001; Fig. 2.4a, Table 2.2) and a decrease in the species richness of visitors (p = 0.03; Table 2.2). Libellula pulchella larvae were found in 6 of the 20 experimental mesocosms, across a range of 34.5–89.9% canopy cover (Table S2.5). No other species of larvae were found in the mesocosms. Natural canopy cover was not associated with larval presence (p = 0.4) or abundance (p = 0.8; Fig. 2.4b, Table 2.2). Prey density was not significantly related to larval abundance (p = 0.3; Table 2.2).

2.3.3 Larval Performance Experiments

There were no effects of canopy cover on larval survival in any species. There was no mortality of caged L. intacta larvae (138/139 survived: one individual was accidentally crushed by the PVC structure). There was no growth observed in the larvae in these trials (p = 1; Table 2.3), likely due to the colder seasonal temperatures. In the S. vicinum trials, all individuals survived the larval period (N = 30), whereas in the Sympetrum hybrid trials (N = 29), 27 individuals survived the larval period, and 2 individuals died (in mesocosms with 91.9% and 94.5% canopy cover). Both Sympetrum hybrids and S. vicinum exhibited growth across treatments (p < 0.001 and p = 0.01, respectively; Table 2.3), but no differential growth between open and closed canopy conditions (p > 0.4 in both cases; Table 2.3). The Sympetrum hybrids reached emergence faster in open mesocosms (p = 0.03; Table 2.3); however, there was no difference among treatments for S. vicinum individuals (p = 0.8; Table 2.3).

2.4 Discussion

Across all experiments, shading and natural canopy cover were associated with sharp decreases in the number of visits to mesocosms by adult dragonflies. In our first study, when light hitting the mesocosms was blocked by artificial shade (50% shade cloth), there were fewer adult visits to mesocosms than when mesocosms were exposed to full sun and covered with an open weave canopy (Fig. 3a). In the natural canopy experiment, both the number (Fig. 2.4a) and species

Fig. 2.4 With an increase in natural canopy cover, (a) the total number of adult visits per mesocosm decreased (p = 0.001), whereas (b) larval abundances per mesocosm were unaffected

Table 2.2 Statistical test results for the natural canopy experiments. Each test was conducted using a GLMM, with 20 observations per test

Variables Response Estimate Standard error z value p value

Canopy Adult abundance -0.03 0.008 -3 0.001 Canopy Adult richness -0.02 0.01 -2 0.03 Canopy Larval abundance -0.01 0.04 -0.3 0.8

Prey density Larval abundance -12 11 -1 0.3

Canopy Larval presence -0.03 0.03 -0.9 0.4

Table 2.3 Statistical test results for the larval performance experiments, using Wilcoxon signed- rank tests (V), rank-sum tests (W), and two sample t tests. Means were taken across cages within each mesocosm. For Leucorrhinia individuals, responses were tested across four levels of canopy cover. For Sympetrum individuals, responses were tested under open (30–40%) and closed (90–100%) canopy cover

Mean ± standard error Test Species N Response p value open/closed statistic

Leucorrhinia intacta 20 Growth -3 ± 2 × 10-3 mm day-1 V = 48 1

Sympetrum hybrids 10 Growth 8 ± 2 × 10-3 mm day-1 V = 55 p < 0.001 Differential growth 10 ± 3 × 10-3 mm day-1 / W = 17 0.4 6 ± 2 × 10-3 mm day-1 Time to emergence 17 ± 2 days/21 ± 1 day t = -3 0.03 Sympetrum vicinum 10 Growth 0.3 ± 0.1 × 10-3 mm day-1 V = 28 0.01 Differential growth 0.4 ± 0.3 × 10-3 mm day-1/ W = 14 0.8 0.2 ± 0.1 × 10-3 mm day-1 Time to emergence 18 ± 2 days/19 ± 3 days t = -0.3 0.8

39 richness of adult visitors decreased with increasing canopy cover. Canopy cover was not consistently a strong predictor of patterns of larval abundance or diversity, nor did the number of adult visits seem to predict larval patterns. The small size of our mesocosms, which was necessary to accommodate mesocosms in our heating arrays and under higher canopy cover levels, may have limited reproduction at the mesocosms and/or the larval carrying capacity of our mesocosms, restricting our ability to detect the effects of these treatments on larval assemblages. However, small natural pond habitats (0–1 ha) can be very common across terrestrial landscapes (Semlitsch & Bodie, 1998), and the results from our mesocosms are likely to be predictive of patterns in these habitats which have lower abundances and species richness with higher levels of shading (McCauley, 2005a; McCauley et al., 2008; Werner et al., unpublished data). Additionally, our experiments ran for one field season each and it is possible that larval patterns take longer to emerge. In a previous 2-year study with mesocosms situated under open and closed canopy conditions (22 and 7 mesocosms, respectively), adult dragonfly arrivals predicted larval dragonfly assemblages for open canopy mesocosms (McCauley, 2006, 2007). In that experiment, adults were not monitored at closed canopy mesocosms. However, larval insects following adult colonization were monitored the same way in both open and closed canopy mesocosms, and the probability of a closed canopy mesocosm having larvae was significantly lower than for mesocosms under open canopy (i.e. across 2 years, no larvae were found in closed canopy mesocosms versus the 11,270 dragonfly larvae collected from open canopy mesocosms; McCauley, 2005a). These results suggest that our own results based on adult visitors would likely translate into effects on larval assemblages over longer time periods. Indeed, adult visits are a pre-requisite for larval presence in ponds and occur before environmental filters affecting the persistence of colonists and/or species-sorting can act to structure larval assemblages (e.g. Fig. 2.1). But the sampling of both adult visitors and larval presence is imperative to understand dragonfly distributions, as the sampling of only one of these communities may not accurately reflect the presence or absence of the other (e.g. Luke et al., 2017). The observed patterns of adult habitat selection at these small experimental habitats likely play a role in structuring larval dragonfly assemblages at a larger scale (see Fig. S2.3).

Our performance trials, which assessed the effects of canopy cover on the larvae of 3 species, suggest that closed canopy habitats have conditions suitable for larval dragonflies. We found that larval survival and growth were equivalent across the canopy gradient in these species,

40 suggesting that post-colonization processes are less important for determining previously observed patterns of decreased abundance and diversity with increasing canopy cover (McCauley, 2005a; McCauley et al., 2008; Werner et al., unpublished data). One caveat here is that we explored these effects during summer and fall, which are critical periods for larval dragonfly survival and growth, but are not the only time periods during which environmental conditions could act as a filter to decrease abundances and diversity within larval assemblages. In particular, winter conditions may differ across the canopy gradient and result in differential survival of larvae in open versus closed canopy ponds.

Taken together, these results suggest that adult attraction to closed canopy habitats is limited, but if these habitats are located, adults can and do successfully breed, and their larvae perform well, in these environments. This provides support for the hypothesis that previously observed patterns of lower diversity and abundance of larval dragonflies in closed canopy ponds – specifically those in temperate regions which experience greater solar incidence and reflection than tropical regions – are related at least in part to limited attraction of adults to these environments. Canopy cover could obscure ponds and decrease their detection by the highly visual adults: directly, and by changing the availability of light cues from pond surfaces. Adult dragonflies can detect horizontally polarized light (Meyer & Labhart, 1993; Horváth, 1995), which they use as a visual cue to detect oviposition sites (Wildermuth, 1998; Bernáth et al., 2002; Kriska et al., 2009). Sunlight reflected from a water surface becomes horizontally polarized, but a lower incidence of sunlight reaching aquatic habitats under a closed canopy would result in less polarized light, potentially decreasing the detection of closed canopy habitats by adult dragonflies.

Adult habitat selection has been demonstrated in other taxa. For example, 8 species of migratory birds (Passeriformes) select nesting sites with no predators in order to improve offspring survival (Fontaine & Martin, 2006). Adult habitat selection decisions, however, do not always clearly relate to offspring performance. In the pine processionary moth, Thaumetopoea pityocampa (Thaumetopoeidae), female adults that are expanding their ranges will oviposit on plants that do not improve larval performance (i.e. growth and survival), possibly due to an evolved specialization for a specific host plant in its native range (Stastny et al., 2006). Previous work in freshwater systems has found evidence of habitat selection across a canopy cover gradient in other taxa. At least one species of aquatic beetle selects habitat based on canopy cover, laying eggs in open but not closed canopy mesocosms (Binckley & Resetarits, 2009). However, other

41 beetle species were either segregated across a canopy cover gradient, only colonizing mesocosms in open canopy or closed canopy conditions, or were generalists, occupying both (Binckley & Resetarits, 2009). This suggests that canopy cover can impact several members of aquatic communities. These choices can vary, however, among species, and without information on offspring performance, it is unclear how adult habitat selection drives larval distributions (e.g. beetles; Binckley & Resetarits, 2009).

In our study, adult habitat selection appeared to be a better predictor of previously documented natural patterns of larval dragonfly distributions across a canopy gradient (McCauley, 2005a; McCauley et al., 2008; Werner et al., unpublished data) than species-sorting mechanisms, as larval performance remained consistent across increasing levels of canopy in our experiments. Adult habitat selection across the canopy gradient appeared to be unrelated to the quality of these habitats for larvae, at least within the levels of canopy cover examined. Instead, adults may avoid closed canopy habitats based on visual limitations (i.e. minimal amount of reflected polarized light; Wildermuth, 1998). Dragonflies detect surfaces that reflect horizontally polarized light through photoreceptors in their compound eyes (Meyer & Labhart, 1993; Horváth, 1995). Many taxa use polarized light as a visual cue to orient themselves towards potential habitats, as is the case for many insects with aquatic larval stages (Kriska et al., 2009), and possibly some amphibian and reptile species (Vitt & Caldwell, 2014). Different dragonfly species may also be attracted to aquatic habitats with varying levels of polarized light (Bernáth et al., 2002; Horváth et al., 2007). For our study, we generated rough estimates of reflected polarized light intensity off the mesocosms, which were collinear with canopy cover and thus not included in our analysis but which provided similar results (French & McCauley, unpublished data). Variation in the polarized light reflected from the surface of a pond may therefore be a key mechanism in structuring larval dragonfly assemblages, as species may colonize habitats across a gradient of polarized light, generated by differences in canopy cover. The selection of habitats by dragonflies may also be hierarchical in terms of choosing habitats in a landscape based on their perceptibility, and from making these large-scale decisions based on general polarized light cues to making site-level evaluations and microhabitat choices.

On a global scale, dragonflies have demonstrated species-specific responses to the presence of canopy cover over aquatic habitats. In tropical regions, the adults of some dragonfly species are very shade-tolerant (e.g. in Mayotte; Samways, 2003), whereas others thrive when [invasive] tree

42 species are removed (e.g. in South Africa; Smith et al., 2007; Samways & Sharratt, 2010) or in urbanized habitats with little canopy cover (e.g. in Brazil; Monteiro-Júnior et al., 2014). Similarly, larval and adult dragonflies in Borneo have demonstrated species-specific distributions in response to gradients of forest, oil palm, and riparian cover (Dolný et al., 2012; Luke et al., 2017). In addition, lentic (standing waters) and lotic (moving waters) attract different assemblages of dragonflies, which may be influenced differently by surrounding tree cover (Balzan, 2012; Dolný et al., 2012; Ball-Damerow et al., 2014). As such, conservation initiatives centred around dragonfly and other invertebrate species should take careful consideration of species-specific tolerances to canopy cover across habitats and regions.

Forest regrowth is a valuable landscape change but can dramatically reshape communities. Secondary forests, which can dominate a landscape following agricultural abandonment (Bellemare et al., 2002), support structurally dissimilar terrestrial communities in comparison to primary forests (Foster et al., 1998), which could differentially impact aquatic communities. In the context of more temperate regions, less canopy cover at lentic habitats may be more favourable for the management or restoration of macroinvertebrate communities (e.g. Batzer et al., 2004). However, the complete removal of canopy cover in these areas is not necessarily warranted, as adult dragonflies still arrive and larvae still persist at habitats with low levels of canopy (e.g. McCauley, 2005a; Remsburg et al., 2008), and other taxa use shaded ponds (e.g. Skelly et al., 2002; Earl & Semlitsch, 2013). Instead, a mix of open and closed canopy cover over habitats in restored and/or managed landscapes may improve regional diversity. Understanding the impacts of both open and closed canopy conditions on communities is essential to inform management decisions about whether to maintain open or closed canopy at a habitat, including considerations of the type of habitat (e.g. lotic or lentic systems) and the history of local habitat conditions (e.g. persisting effects of historical land use; Bellemare et al., 2002). Other regional environmental factors, such as landscape type between habitats, may also play a role in maintaining connectivity among aquatic habitats, and therefore must be considered in conjunction with local factors.

2.5 Conclusion

Our results help to distinguish the mechanisms driving dragonfly distributions across an important environmental gradient, canopy cover. Habitat selection and species-sorting are

43 difficult to separate, especially across aquatic and terrestrial environments (Vanschoenwinkel et al., 2010), and are generally inferred from patterns of species distributions (Binckley & Resetarits, 2005). Widespread forest regrowth across mid-latitudinal North America following agricultural abandonment is modifying both aquatic and terrestrial systems (Foster et al., 1998). Since forests may impact habitat colonization, understanding how habitat selection behaviour interacts with dispersal and spatial connectivity in patchy habitats can provide insight into how these processes determine community structure across aquatic habitats.

2.6 References

Ball-Damerow, J. E., L. K. M’Gonigle & V. H. Resh, 2014. Local and regional factors influencing assemblages of dragonflies and damselflies (Odonata) in California and Nevada. Journal of Insect Conservation 18: 1027–1036.

Balzan, M. V., 2012. Associations of dragonflies (Odonata) to habitat variables within the Maltese Islands: a spatio-temporal approach. Journal of Insect Science 12: 1–18.

Bates, D., M. Mächler, B. M. Bolker & S. C. Walker, 2015. Fitting linear mixed-effects models using lme4. Journal of Statistical Software 67: 1–48.

Batzer, D. P., B. J. Palik & R. Buech, 2004. Relationships between environmental characteristics and macroinvertebrate communities in seasonal woodland ponds of Minnesota. Journal of the North American Benthological Society 23: 50–68.

Bellemare, J., G. Motzkin & D. R. Foster, 2002. Legacies of the agricultural past in the forested present: an assessment of historical land-use effects on rich mesic forests. Journal of Biogeography 29: 1401–1420.

Berendonk, T. U., 1999. Influence of fish kairomones on the ovipositing behavior of Chaoborus imagines. Limnology and Oceanography 44: 454–458.

Binckley, C. A. & W. J. Resetarits Jr., 2003. Functional equivalence of non-lethal effects: generalized fish avoidance determines distribution of gray treefrog, Hyla chrysoscelis, larvae. Oikos 102: 623–629.

Binckley, C. A. & W. J. Resetarits Jr., 2005. Habitat selection determines abundance, richness and species composition of beetles in aquatic communities. Biology Letters 1: 370–374.

Boyce, M. S., J. S. Mao, E. H. Merrill, D. Fortin, M. G. Turner, J. Fryxell & P. Turchin, 2003. Scale and heterogeneity in habitat selection by elk in Yellowstone National Park. Écoscience 10: 421–431.

Bried, J. T., B. J. Hager, P. D. Hunt, J. N. Fox, H. J. Jensen & K. M. Vowels, 2012. Bias of reduced-effort community surveys for adult Odonata of lentic waters. Insect Conservation and Diversity 5: 213–222.

Briscoe, A. D. & L. Chittka, 2001. The evolution of color vision in insects. Annual Review of Entomology 46: 471–510.

Brodin, T., F. Johansson & J. Bergsten, 2006. Predator related oviposition site selection of aquatic beetles (Hydroporus spp.) and effects on offspring life-history. Freshwater Biology 51: 1277–1285.

Buchwald, R., 1994. Experimentelle untersuchungen zu habitatselektion und biotopbindung bei Ceriagrion tenellum De Villers, 1789 (Coenagrionidae, Odonata). Zoologische Jahrbücher Abteilung für Systematik, Ökologie Und Geographie Der Tiere 121: 71–98.

Catling, P. M., 2007. Variation of hind-wing colour and length in Sympetrum internum (Odonata: Libellulidae) from the Canadian prairie provinces. Canadian Entomologist 139: 872–880.

Corbet, P. S., 1962. A biology of dragonflies. Witherby, Warwick UK.

Corbet, P. S., 2004. Dragonflies: behaviour and ecology of Odonata, Revised ed. Harley Books, Colchester UK.

De Marco Jr., P., J. D. Batista & H. S. R. Cabette, 2015. Community assembly of adult odonates in tropical streams: an ecophysiological hypothesis. PLoS ONE 10: e0123023.

Dolný, A., F. Harabiš, D. Bárta, S. Lhota & P. Drozd, 2012. Aquatic insects indicate terrestrial habitat degradation: changes in taxonomical structure and functional diversity of dragonflies in tropical rainforest of East Kalimantan. Tropical Zoology 25: 141–157.

Earl, J. E. & R. D. Semlitsch, 2013. Spatial subsidies, trophic state, and community structure: examining the effects of leaf litter input on ponds. Ecosystems 16: 639–651.

Flenner, I., O. Richter & F. Suhling, 2010. Rising temperature and development in dragonfly populations at different latitudes. Freshwater Biology 55: 397–410.

Fontaine, J. J. & T. E. Martin, 2006. Habitat selection responses of parents to offspring predation risk: an experimental test. The American Naturalist 168: 811–818.

Foster, D. R., G. Motzkin & B. Slater, 1998. Land-use history as long-term broad-scale disturbance: regional forest dynamics in central New England. Ecosystems 1: 96–119.

Foster, S. E. & D. A. Soluk, 2006. Protecting more than the wetland: the importance of biased sex ratios and habitat segregation for conservation of the Hine’s emerald dragonfly, Somatochlora hineana Williamson. Biological Conservation 127: 158–166.

Fournier, D. A., H. J. Skaug, J. Ancheta, J. Ianelli, A. Magnusson, M. N. Maunder, A. Nielsen & J. Sibert, 2012. AD Model Builder: using automatic differentiation for statistical inference of highly parameterized complex nonlinear models. Optimization Methods and Software 27: 233–249.

Frances, D. N., J. Y. Moon & S. J. McCauley, 2017. Effects of environmental warming during early life history on libellulid odonates. Canadian Journal of Zoology 95: 373–382.

Garcia, E. A. & G. G. Mittelbach, 2008. Regional coexistence and local dominance in Chaoborus: species sorting along a predation gradient. Ecology 89: 1703–1713.

Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L. Chini, C. O. Justice & J. R. G. Townshend, 2013. High-resolution global maps of 21st-century forest cover change. Science 342: 850–853.

Horváth, G., 1995. Reflection-polarization patterns at flat water surfaces and their relevance for insect polarization vision. Journal of Theoretical Biology 175: 27–37.

Horváth, G., B. Bernáth & G. Molnár, 1998. Dragonflies find crude oil visually more attractive than water: multiple-choice experiments on dragonfly polarotaxis. Naturwissenschaften 85: 292–297.

Jacobs, M. E., 1955. Studies on territorialism and sexual selection in dragonflies. Ecology 36: 566–586.

Keller, D., M. J. van Strien & R. Holderegger, 2012. Do landscape barriers affect functional connectivity of populations of an endangered damselfly? Freshwater Biology 57: 1373– 1384.

Knorp, N. E. & N. J. Dorn, 2016. Mosquitofish predation and aquatic vegetation determine emergence patterns of dragonfly assemblages. Freshwater Science 35: 114–125.

Lima, S. L. & P. A. Zollner, 1996. Towards a behavioral ecology of ecological landscapes. Trends in Ecology & Evolution 11: 131–135.

Luke, S. H., R. A. Dow, S. Butler, C. V. Khen, D. C. Aldridge, W. A. Foster & E. C. Turner, 2017. The impacts of habitat disturbance on adult and larval dragonflies (Odonata) in rainforest streams in Sabah, Malaysian Borneo. Freshwater Biology 62: 491–506.

McCauley, S. J., 2005a. Species distributions in anisopteran odonates: effects of local and regional processes. PhD thesis. University of Michigan, Ann Arbor MI.

McCauley, S. J., 2005b. Relationship between habitat distribution, growth rate, and plasticity in congeneric larval dragonflies. Canadian Journal of Zoology 83: 1128–1133.

McCauley, S. J., 2006. The effects of dispersal and recruitment limitation on community structure of odonates in artificial ponds. Ecography 29: 585–595.

McCauley, S. J., 2007. The role of local and regional processes in structuring larval dragonfly distributions across habitat gradients. Oikos 116: 121–133.

McCauley, S. J., 2008. Slow, fast and in between: habitat distribution and behaviour of larvae in nine species of libellulid dragonfly. Freshwater Biology 53: 253–263.

McCauley, S. J., C. J. Davis, R. A. Relyea, K. L. Yurewicz, D. K. Skelly & E. E. Werner, 2008. Metacommunity patterns in larval odonates. Oecologia 158: 329–342.

McPeek, M. A., 1989. Differential dispersal tendencies among Enallagma damselflies (Odonata) inhabiting different habitats. Oikos 56: 187–195.

Meyer, E. P. & T. Labhart, 1993. Morphological specializations of dorsal rim ommatidia in the compound eye of dragonflies and damselflies (Odonata). Cell Tissue Research 272: 17– 22.

Michiels, N. K. & A. A. Dhondt, 1990. Costs and benefits associated with oviposition site selection in the dragonfly Sympetrum danae (Odonata: Libellulidae). Animal Behaviour 40: 668–678.

Monteiro-Júnior, C. S., L. Juen & N. Hamada, 2014. Effects of urbanization on stream habitats and associated adult dragonfly and damselfly communities in central Brazilian Amazonia. Landscape and Urban Planning 127: 28–40.

Pohlert, T., 2014. The pairwise multiple comparison of mean ranks package (PMCMR). R package. http://CRAN.R-project.org/package=PMCMR.

Pritchard, G., L. D. Harder & R. A. Mutch, 1996. Development of aquatic insect eggs in relation to temperature and strategies for dealing with different thermal environments. Biological Journal of the Linnean Society 58: 221–244.

R Core Team, 2017. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.

Remsburg, A. J., A. C. Olson & M. J. Samways, 2008. Shade alone reduces adult dragonfly (Odonata: Libellulidae) abundance. Journal of Insect Behavior 21: 460–468.

Resetarits Jr., W. J., 2005. Habitat selection behaviour links local and regional scales in aquatic systems. Ecology Letters 8: 480–486.

Richardson, J. S., 2008. Aquatic arthropods and forestry: effects of large-scale land use on aquatic systems in Nearctic temperate regions. Canadian Entomologist 140: 495–509.

Rieger, J. F., C. A. Binckley & W. J. Resetarits Jr., 2004. Larval performance and oviposition site preference along a predation gradient. Ecology 85: 2094–2099.

Samways, M. J., 2003. Threats to the tropical island dragonfly fauna (Odonata) of Mayotte, Comoro archipelago. Biodiversity and Conservation 12: 1785–1792.

Samways, M. J. & N. J. Sharratt, 2010. Recovery of endemic dragonflies after removal of invasive alien trees. Conservation Biology 24: 267–277.

Schiesari, L., 2006. Pond canopy cover: a resource gradient for anuran larvae. Freshwater Biology 51: 412–423.

Schneider, C. A., W. S. Rasband & K. W. Eliceiri, 2012. NIH Image to ImageJ: 25 years of image analysis. Nature Methods 9: 671–675.

Schooley, R. L. & J. A. Wiens, 2003. Finding habitat patches and directional connectivity. Oikos 102: 559–570.

Schtickzelle, N., A. Joiris, H. Van Dyck & M. Baguette, 2007. Quantitative analysis of changes in movement behaviour within and outside habitat in a specialist butterfly. BMC Evolutionary Biology 7: 1–15.

Semlitsch, R. D. & J. R. Bodie, 1998. Are small, isolated wetlands expendable? Conservation Biology 12: 1129–1133.

Skaug, H. J., D. A. Fournier, A. Nielsen, A. Magnusson & B. M. Bolker, 2012. Generalized linear mixed models using AD Model Builder. R package version 0.7.2.12.

Skelly, D. K., S. R. Bolden & L. K. Freidenburg, 2014. Experimental canopy removal enhances diversity of vernal pond amphibians. Ecological Applications 24: 340–345.

Skelly, D. K., L. K. Freidenburg & J. M. Kiesecker, 2002. Forest canopy and the performance of larval amphibians. Ecology 83: 983–992.

Smith, J., M. J. Samways & S. Taylor, 2007. Assessing riparian quality using two complementary sets of bioindicators. Biodiversity and Conservation 16: 2695–2713.

Stastny, M., A. Battisti, E. Petrucco-Toffolo, F. Schlyter & S. Larsson, 2006. Host-plant use in the range expansion of the pine processionary moth, Thaumetopoea pityocampa. Ecological Entomology 31: 481–490.

Switzer, P. V., 2002. Territory quality, habitat selection, and competition in the amberwing dragonfly, Perithemis tenera (Say) (Odonata: Libellulidae): population patterns as a consequence of individual behaviour. Journal of the Kansas Entomological Society 75: 145–157.

Switzer, P. V. & W. Walters, 1999. Choice of lookout posts by territorial amberwing dragonflies, Perithemis tenera (Anisoptera: Libellulidae). Journal of Insect Behavior 12: 385–398.

Van Buskirk, J., 1986. Establishment and organization of territories in the dragonfly Sympetrum rubicundulum (Odonata: Libellulidae). Animal Behaviour 34: 1781–1790.

Vanschoenwinkel, B., A. Waterkeyn, M. Jocqué, L. Boven, M. Seaman & L. Brendonck, 2010. Species sorting in space and time – the impact of disturbance regime on community assembly in a temporary pool metacommunity. Journal of the North American Benthological Society 29: 1267–1278.

Vitt, L. J. & J. P. Caldwell, 2014. Herpetology: an introductory biology of amphibians and reptiles, 4th ed. Academic Press, Waltham MA.

Werner, E. E. & K. S. Glennemeier, 1999. Influence of forest canopy cover on the breeding pond distributions of several amphibian species. Copeia 1999: 1–12.

Wildermuth, H., 1994. Habitatselektion bei Libellen. Advances in Odonatology 6: 223–257.

Wildermuth, H., 1998. Dragonflies recognize the water of rendezvous and oviposition sites by horizontally polarized light: a behavioural field test. Naturwissenschaften 85: 297–302.

Chapter 3 The Movement Responses of Three Libellulid Dragonfly Species to Open and Closed Landscape Cover

This paper will be submitted for publication as: French, S. K. & S. J. McCauley. The movement responses of three libellulid dragonfly species to open and closed landscape cover.

Abstract

The land cover between habitats (i.e. matrix environment) can affect connectivity by impacting organismal movement. However, many active dispersers have preferences for various matrix environments, which can affect their movement through the landscape. We examined how different terrestrial matrix environments impacted the fine-scale movement of adult dragonflies. Based on previous studies of dragonfly dispersal patterns, we hypothesized that dragonflies would prefer to enter fields rather than forests, and that forests would be a barrier to dragonfly movement, due to forests’ structural complexity and low understory light availability and air temperatures. To test how dragonflies responded to various terrestrial environments, we released 324 adults at field-forest ecotones and assessed their preferences for fields and forests. Individual behavioural responses were recorded, including their probability of taking flight, their direction of movement with respect to the two matrix types, and flight time. Adult dragonflies took flight marginally more often in fields than forests, but these responses were species-specific. Adults also moved more frequently towards fields than forests when released at a forest edge. Individuals released within forests had shorter flight times, but again these responses were species-specific. The presence of an open matrix (field or meadow) is likely important for facilitating movement in dragonflies, however forests do not hinder movement in all dragonfly species. Integrating assays of matrix and habitat preferences can provide insight into how landscape connectivity can be maintained.

3.1 Introduction

The physical characteristics of a landscape can affect habitat connectivity – the potential for organisms to move among habitats (Calabrese & Fagan, 2004). In heterogeneous landscapes, patches of habitat where a focal species lives, are surrounded by compositionally similar and/or dissimilar (i.e. matrix) environments (Forman & Godron, 1981, 1986; Kracker, 1999). When

52 53 moving away from their source habitat, both passively and actively dispersing individuals pass through and respond to these environments before settling in a new habitat (Baguette & Van Dyck, 2007). The type of landscape between habitats can affect habitat connectivity by promoting or restricting organismal movement (Ricketts, 2001; Murphy & Lovett-Doust, 2004; Prugh et al., 2008). In particular, the choices of active dispersers moving through a matrix environment can affect their success in dispersing to a new habitat, as their decisions in the matrix may shape their direction of travel and ultimate settlement decisions, and are thus important when considering the drivers of species distributions.

The behavioural responses of individuals to a matrix environment can provide critical insight into how species move through, use, and distribute themselves across heterogeneous landscapes. Few studies have been able to connect individual-based behaviour to processes seen at the community and landscape scale (Lima & Zollner, 1996; Bélisle, 2005; Gordon, 2011). Yet, the observation of these behaviours at a local scale can be a useful tool in predicting patterns of regional dispersal (Haddad, 1999). Indeed, behavioural movements are an important mechanism in driving the evolution of species dispersal, particularly across changing landscapes (Van Dyck & Baguette, 2005; Baguette & Van Dyck, 2007). However, without information on how actively dispersing animals respond behaviourally to different elements in the landscape, our estimates of habitat connectivity for these animals will likely be flawed. Heterogeneous matrix environments may provide differential costs and benefits to individuals moving through the landscape (Chardon et al., 2003). This landscape of costs and benefits may be complex and drive non- intuitive patterns of movement. For example, matrix environments that provide resources or refuge may be preferred to ones that are easy to move through, despite the costs imposed by the difficulty of moving through these environments (Haynes & Cronin, 2006; Prugh et al., 2008; Kuefler et al., 2010). Choices made during dispersal can depend on a number of other factors, including the time it takes for individuals to find a new habitat depending on matrix type (Stamps et al., 2005), and intra- and inter-specific variation (Bélisle, 2005).

The suitability of, or preference for, specific types of matrix environments by species and individuals can vary depending on the potential risks and gains organisms experience while moving through these environments, whether during rare and long-distance movements from a source habitat to a new habitat (i.e. dispersal) or during short-distance movements that occur frequently (e.g. daily foraging). One type of matrix in terrestrial landscapes that has dramatic

54 effects on the movement of organisms is forest cover (e.g. Ricketts, 2001; Driscoll et al., 2013; Damschen et al., 2014). For example, green hermit hummingbirds (Phaethornis guy, Apodiformes) prefer to move through forested corridors, possibly due to increased resource availability (Volpe et al., 2016). Some species of the Ambystoma salamander (Caudata) preferentially move through forests rather than fields in order to avoid desiccation (Rothermel & Semlitsch, 2002; Rittenhouse & Semlitsch, 2006). Yet, forests do not always facilitate connectivity and recruitment to new habitats – in some cases forests can increase the risk of mortality for dispersing individuals (e.g. Phengaris spp., Lepidoptera; Nowicki et al., 2014) or restrict their movements (e.g. Leucorrhinia hudsonica, Odonata; Chin & Taylor, 2009). Forests can also be energetically costly for organisms to manoeuvre through due to their structural complexity (e.g. for Bombus impatiens, Hymenoptera; Crall et al., 2015), and in the case of ectotherms, due to the reduced solar radiation available for thermoregulation (e.g. large-bodied dragonfly species; De Marco et al., 2015). Forest cover is also rapidly changing worldwide, both in terms of forest regrowth and loss (Foster et al., 1998; Hansen et al., 2013), which may lead to changes in how organisms move in the landscape.

Terrestrial matrix conditions are similarly important in determining connectivity among isolated aquatic habitats, particularly for semi-aquatic insects, many of which have an aquatic larval stage and a dispersing terrestrial life history stage. In these taxa, their post-colonization distribution depends at least in part on the permeability of the matrix (Joly et al., 2001; Semlitsch, 2008), however for taxa that use multiple environments across their ontogeny, measurements of functional connectivity are more difficult to obtain (Kadoya, 2009). Our study taxon, dragonflies (specifically three species in the family Libellulidae), disperse through terrestrial systems during their adult stage and lay eggs at aquatic habitats (Switzer, 2002). The linkage between aquatic and terrestrial systems is therefore an important factor in shaping distributions of aquatic larval dragonflies (McCauley, 2006, 2007) and potentially other aquatic insects. Dragonflies are a good model taxon for understanding active dispersers’ responses to changes in the landscape, as adults frequently disperse to and establish at new aquatic habitats (Oertli, 2008), and may thus exhibit differential responses to habitat heterogeneity. For example, forests can obstruct adult dragonfly dispersal to aquatic reproductive habitats at distances ≥ 500 m (Chin & Taylor, 2009), and can restrict their arrival at aquatic habitats (French & McCauley, 2018). At shorter distances (e.g. 125 m) however, forests do not seem to impede movement to aquatic habitats (Chin & Taylor,

2009). The behavioural mechanisms behind this pattern of increased connectivity at small scales among dragonfly breeding habitats with an intervening forest matrix remain unclear. Forests may somehow facilitate dragonfly movement to aquatic habitats (Chin & Taylor, 2009), however it is also possible that once dragonflies enter a forested area, their movement may be effectively confined to this area if they perceive it to be costly to search for new aquatic habitats (e.g. Stamps et al., 2005). Adult dragonflies may also exhibit species-specific movement responses to land cover, which may explain previously observed patterns of changes in the composition of larval dragonfly assemblages across a gradient of pond canopy cover and surrounding forest cover (McCauley et al., 2008).

We therefore sought to determine whether a behavioural preference by dragonflies for forest or field environments at small scales might explain their previously observed patterns of dispersal. We tested whether dragonflies preferentially moved in and into field and/or forest matrix environments, and observed dragonflies’ behavioural responses to forest edges. We hypothesized that dragonflies would prefer to move and enter field environments and that forests would be a barrier to movement, as dragonflies may face difficulties in navigating the structural complexity of forests during flight (e.g. Crall et al., 2015), and the lowered light and air temperatures conditions imposed by the forest canopy (Chen et al., 1999) would reduce dragonfly flight activity (e.g. Csabai et al., 2006; De Marco et al., 2015). Alternatively, the open and less structurally complex field habitats, with higher air temperatures and light intensity, were expected to be preferred because they facilitate movement (e.g. Csabai et al., 2006; Crall et al., 2015). Specifically, we predicted that across species, dragonflies would have a higher probability of flying towards fields versus forests, and that they would move more and for longer time periods in fields versus forests. We tested these predictions by releasing three species of adult dragonflies, which may have generalist responses to forest cover as their larvae inhabit ponds across a gradient of surrounding canopy cover, near boundaries between fields and forests, and observing their likelihood of taking flight, and their direction and time of flight.

3.2 Methods

To test how adult dragonflies moved in response to open versus closed environments, we observed flight patterns of dragonflies released in field and forest matrix environments, and at forest edges. Our work was conducted at the University of Toronto’s Koffler Scientific Reserve

(KSR; King City, ON, Canada, 44.03 N, 79.53 W). In the summer of 2015, we collected 108 adult males of each of three species, Leucorrhinia intacta, Sympetrum obtrusum/rubicundulum hybrids, and Sympetrum vicinum, for a total of 324 individuals, from local ponds at KSR. These species are all members of the family Libellulidae (Odonata, Anisoptera), their larvae are common in lentic systems in this region, and adults of these species were abundant at KSR. The larvae of all three species have been observed across a gradient of pond canopy cover and surrounding landscape forest cover (McCauley et al., 2008), indicating that the larvae and/or adults of these species may have generalist responses to canopy and forest cover. Males were used as they were more readily available than female dragonflies. Male dragonflies tend to appear first at aquatic reproductive habitats, and have been observed to bring females to these habitats (Jacobs, 1955; Michiels & Dhondt, 1990; Switzer & Walters, 1999). Little is known about whether the dispersal of dragonflies is sex-biased, however previous studies have found male-biased arrivals to aquatic habitats (McCauley, 2006), and limited female dispersal among aquatic habitats for some species (Conrad et al., 1999; Chin & Taylor, 2009; McCauley et al., 2010). Adults were used over tenerals (i.e. juveniles) as dragonflies may more readily disperse at the adult stage (Angelibert & Giani, 2003), and adults are less easily damaged during collection and are more identifiable to sex and species than tenerals.

Mature male dragonflies were collected using insect nets from Gazebo Pond and Dufferin Pond at KSR while they were either perching or flying. Individuals were placed in covered plastic cups in a cooler and chilled using ice packs to anesthetize them, in order to measure and mark them effectively. Using digital calipers (to 0.01 mm), we measured individuals’ head width as a proxy for body size/mass to test for effects on their ability to take flight (e.g. McCauley, 2005a; Serrano-Meneses et al., 2007), and forewing length, as a proxy for their flight ability (e.g. McCauley et al., 2014). Both head width and wing length are positively related to body length, another common measure of body size (Serrano-Meneses et al., 2007). The right forewing and hindwing of each individual were numbered using permanent black ink (Sharpie™) to allow for any necessary identification after release. Individuals were then stored in hanging mesh nets to allow them to warm up and become alert before translocations occurred. Collections and releases took place between June and October 2015, during the period when individuals of each species were most abundant (L. intacta: June–July, S. obtrusum/rubicundulum: July–August, S. vicinum: August–October).

Individuals were translocated to sites distributed across KSR in which a field was adjacent to a forest edge. A forest area was defined as any continuous line of tree cover where the other side of the forest could not be seen from the edge. A release area (i.e. where sites were close to one another) contained 2–4 sites for a total of 12 sites across KSR (Fig. 3.1). To account for any tendency for dragonflies to orient themselves in particular directions, such as in response to the sun (Corbet, 2004; Chahl & Mizutani, 2012), sites were set up across KSR such that three sites had forest edge boundaries to the north, three to the south, three to the west, and three to the east. One individual was released at one of three points (i.e. release locations) at each site (Fig 3.1): at 10 m into the field, directly at the forest/field boundary, and 10 m into the forest (e.g. Ross et al., 2005; Schultz et al., 2012). After adults of a single species were collected for one ‘release replicate’ (i.e. 36 individuals), they were released at 36 different locations (forest, edge, and field locations, at each of 12 sites). For each release replicate, the order of release at sites for each release area (i.e. general location across the reserve) was randomized, followed by the order of release locations at that site. Only one release replicate was performed over a given time period, and the three replicates for a single species were performed before conducting the three replicates of the next species. If individuals were collected from different source ponds during a ‘release replicate’, they were pooled together and released randomly across sites and release locations. At each release location, translocated dragonflies were placed on grass or another available forb, approximately three feet off the ground and facing parallel to the forest edge. The dragonflies were then observed for up to five minutes after release for whether they took flight or not. If they took flight, we recorded their initial flight orientation: towards a field or forest matrix, parallel to a forest edge, or up. We also recorded the angle of this flight direction (e.g. Rothermel & Semlitsch, 2002) using a compass, and the length of time of their flight. If the landing of the dragonfly was not observed, their flight time was the length of time observed before visual contact was lost. Individuals that flew up were pooled with those that flew parallel to the forest edge, because we assumed both responses were indicative of not having a preference for forest or field environments.

A wedge prism (BAF 2, Cruise Master Prisms Inc., Universal Field Supplies) was used to estimate tree stand density (i.e. basal area) at each release location (i.e. forest, field, or forest edge), to test the effects of structural complexity on flight patterns. A spherical densiometer (Model-C, Forest Densiometers, Rapid City, South Dakota, USA) was used to estimate canopy

Fig. 3.1 Dragonflies were released at three release locations (white circles in inset; forest, field, and forest edge) at each of 12 sites, with three replicates per species for three species at Koffler Scientific Reserve. Local source ponds are also indicated (Gazebo, Barn, and Dufferin Ponds)

59 cover at each release location to examine the effects of light on flight behaviour. Field release locations had an average of 13 ± 2% (standard error) canopy cover and a stand density of 0.3 ± 0.2 m2 ha-1, forests had 95 ± 1% canopy cover and a stand density of 32 ± 3 m2 ha-1, and forest edges had 81 ± 3% canopy cover and a stand density of 15 ± 2 m2 ha-1. Wind speed and direction were recorded during releases to account for any influence on flight patterns. An anemometer (Vantage VUE 6250, Davis Instruments, Hayward, California, USA) was used for part of the study to measure wind speed and direction at a central location among nearby sites in a release area, and a portable weather meter was used for the remainder of the study at the release points at each site. The anemometer provided cardinal, ordinal, and secondary intercardinal directions, whereas the portable meter provided 360° compass directions. Each wind direction was converted to a wind bearing (i.e. the direction toward which the wind was moving) to facilitate comparisons with the orientation of the dragonflies’ flight. The anemometer measured wind speed and direction at 15-minute intervals, so for each release the direction was determined from the closest time point to the time of release. Across our releases, wind speeds ranged between 0– 8 m s-1, with 87% of releases taking place with wind speeds of 0–2.2 m s-1.

3.2.1 Statistical Analysis

We tested the probability of dragonflies taking flight in response to release location, species identity, head width, and the interaction among the three, using a generalized linear model (GLM) with binomial errors. Differences in the flight time of those individuals that took flight were tested against release location, species identity, and wing length, using a GLM with gamma errors and a log link. We removed interaction terms and main effects sequentially and compared models using the ‘anova’ function in R with a chi-squared test, and used least-square means (package: emmeans; Lenth, 2018) to compare the effects of specific release locations and species identities on movement and flight times.

To test whether dragonflies preferentially flew towards one of the matrix environments (field or forest), we compared their orientations using a saturated GLM with species identity, release location (i.e. forest, field, and forest edge), and orientation (i.e. towards field or forest) as predictor variables, and count data as our response variable with a Poisson error distribution. We removed interaction terms sequentially and compared models using the ‘anova’ function with a chi-squared test. Least-square means were used to compare the effect of specific release

60 locations on dragonfly orientation. We assumed that individuals moving parallel to the forest edge or flying up after release had no preference for a matrix environment.

To see if canopy cover affected dragonfly activity within forested release locations, we tested the effect of canopy cover on the probability of individuals taking flight using a GLM with quasibinomial errors. We also tested the effect of tree density within forested release locations on flight time using a GLM with gamma errors and a log link, to see if increased structural complexity affected the dragonflies' flight capabilities. Kruskal-Wallis rank sum tests with post hoc Nemenyi-tests (PMCMR package; Pohlert, 2014) were used to compare head widths and wing lengths separately among species.

A Moore’s test for paired angles (Pewsey et al., 2013) was used to test whether the compass direction of dragonfly flight corresponded to wind direction, separately for wind directions measured by the anemometer and the handheld meter. We considered the absolute value of the differences in angles, as we were interested in the magnitude of the differences rather than the direction of the differences between wind and dragonfly flight directions. The probability of individuals taking flight in response to wind speed and time of day was tested using a GLM with binomial errors. Wind speed was taken as the average between anemometer and handheld meter measurements when available, whereas time of day was calculated as decimal time using the TIMEVALUE function in Excel. All analyses were performed in R (version 3.4.0; R Core Team, 2017).

3.3 Results

Of the 324 individuals released, across species, 169 flew towards a certain matrix environment, 77 did not fly towards either matrix environment, 61 did not take flight, and 17 disappeared immediately after taking flight. Overall, there were mixed effects of release location and species identity on the flight responses of dragonflies to forest and field matrix environments.

Release location, species identity, and the interaction between the two affected movement (i.e. the proportion of individuals taking flight) in adult dragonflies (p = 0.02, p < 0.001, and p = 0.04, respectively; Fig. 3.2). When movement was averaged across species identity, adult dragonflies were marginally less likely to take flight in forests versus fields (z = 2, p = 0.07). When movement was averaged across release location, L. intacta individuals were less likely to move

Fig. 3.2 Mean proportion (± SE) of Sympetrum vicinum, Sympetrum hybrid, and Leucorrhinia intacta individuals that took flight when released at edges, in fields, and in forests. Across species, adult dragonflies were marginally less likely to take flight in forests than in fields (p = 0.07). Leucorrhinia intacta individuals moved less than Sympetrum hybrids when released at edges (p = 0.01) and marginally less than both Sympetrum species when released in forests (p = 0.07). Within a single species, individuals were equally likely to take flight across release locations

62 than Sympetrum hybrids (z = -4, p < 0.001). Within a species, there were no differences in movement among release locations. However, L. intacta individuals were less likely to move than Sympetrum hybrids at forest edges (p = 0.01), and marginally less likely to move than both Sympetrum spp. in forests (z = -2, p = 0.07). Head width and the interaction between head width and species identity did not significantly affect movement (p = 0.2 and p = 0.1, respectively), and were not retained in the best model. For those individuals released in the forest, increased canopy cover did not affect the probability of taking flight (z107 = -0.09, p = 0.9).

Across all species and of the adult dragonflies that took flight towards a certain matrix environment (169/324 individuals), individuals oriented themselves more towards fields than forests (z17 = -4, p < 0.001; Fig. 3.3). However, the release location of individuals (i.e. forest, field, or at forest edge) also interacted with the orientation of dragonflies (p = 0.003). When released at the edge, individuals flew more often towards the field than the forest (z = 4, p < 0.001), however, there were only marginal differences in orientation when individuals were released in the field (z = 2, p = 0.06), and no differences when released in the forest (z = 0.7, p = 0.5). Species identity did not affect the number of dragonflies moving towards fields or forests (p = 0.9), and was dropped from the model as it was not an important parameter in comparison to the best model.

Of the individuals that took flight, release location, species identity, and the interaction between the two had significant effects on flight time (p < 0.001 in all cases; Fig. 3.4). Across species, individuals released in forests had significantly shorter flight times than those released at forest edges and in fields (p < 0.003 in both cases). The relationship between release location and flight time seemed to be driven by the responses of S. vicinum individuals, who had shorter flight times in forests than at forest edges or in fields (p < 0.001 in both cases). In fields and at forest edges, S. vicinum individuals also had longer flight times than L. intacta and Sympetrum hybrids (p < 0.001 for all pairwise comparisons), whereas in the forest, all three species had similar flight times (p > 0.5 for all pairwise comparisons). Wing length did not affect the flight time of dragonflies (t231 = -1, p = 1) and only had marginal interactions with release location and species identity on flight time (p = 0.09 and p = 0.09, respectively). Wing length was dropped from the model as it was not an important parameter in comparison to the best model. Of those individuals that took flight in forests, increased tree density (i.e. a more complex and cluttered environment) marginally decreased flight time (t67 = -2, p = 0.07). Overall, L. intacta had greater head widths

Fig. 3.3 Mean number of dragonfly individuals (± SE) flying towards forests and fields, when released at edges, in fields, and in forests. Overall, more individuals flew towards fields than forests (p < 0.001; indicated by an asterisk), however individuals responded differently based on their release location. A dagger indicates a marginally significant difference in orientation from a single release location (i.e. p = 0.06) and “ns” indicates no significance (i.e. p = 0.5)

Fig. 3.4 Mean flight times (s; ± SE) of those Sympetrum vicinum, Sympetrum hybrids, and Leucorrhinia intacta individuals that took flight. Across species, individuals had longer flight times when released at forest edges and in the field than in the forest (p < 0.003), however this was likely due to species differences. At forest edges and in fields, S. vicinum individuals had the longest flight times (p < 0.001), however all species had similar flight times when released in forests (p > 0.6)

2 2 (χ 2 = 218, p < 0.001) and wing lengths (χ 2 = 134, p < 0.001) than both Sympetrum species (Table 3.1), however whether differences in taking flight and subsequent flight times were due to body size or differences in monthly air temperatures between species’ release dates could not be discerned.

Wind direction did not correspond with the flight direction of dragonflies (i.e. they were significantly dissimilar), either using the cup anemometer (R = 2, p < 0.001; Fig. S3.1a in Appendix B: Supplementary Materials for Chapter 3) or the handheld wind meter (R = 1, p =

0.005; S3.1b). Individuals tended to take flight less often with increasing wind speed (z272 = -3, p

< 0.001). Time of day also affected the probability of individuals taking flight (z272 = -2, p = 0.03), with individuals being less likely to move into the late afternoon and early evening.

3.4 Discussion

The dragonfly species in our study showed preferences for open field environments compared to forest environments, however their movements were affected by where individuals were released. Dragonflies flew more often towards fields than forests when released at the boundary between these two environments, the forest edge, when both land cover types were visible and readily accessible (Fig. 3.3). This difference, however, was marginal when individuals were released in fields and disappeared when they were released in forests. The responses when released in fields and forests may be due to a lack of perception of the ecotonal boundary (e.g. Ross et al., 2005), or its associated risks. This may be especially likely for dragonflies released in forests as the cluttered conditions of these environments would reduce or eliminate their ability to perceive open fields outside the forest. Species showed markedly different movement and flight behaviours: in some cases these were general species differences while in other cases, species exhibited differential responses to matrix conditions. Leucorrhinia intacta individuals were less likely to move than Sympetrum individuals at forest edges and in forests (Fig 3.2), and only S. vicinum individuals showed a reduction in flight times in forests (Fig. 3.4). These results support previous observational studies that suggested that dragonflies prefer to fly through fields and other open environments to reach aquatic habitats (Chin & Taylor, 2009) but also indicate the importance of understanding species-specific movement behaviour.

Table 3.1 Morphological characteristics of the three dragonfly species used during releases. Shared letters indicate similarity in characteristics (p > 0.05)

Species Head width (mm) ± SE Wing length (mm) ± SE

Sympetrum hybrida 4.91 ± 0.01 23.63 ± 0.08 Sympetrum vicinuma 4.97 ± 0.01 23.87 ± 0.08 Leucorrhinia intactab 5.59 ± 0.02 25.12 ± 0.07

At the small scales used in our study, adult dragonflies may be more familiar with their landscape, as they make frequent (i.e. daily) movements of over 50 m for activities such as foraging and moving between roosting and reproductive habitats (Eason & Switzer, 2006). The individuals used in our study were likely familiar with at least some of the locations they were released at, having possibly encountered them after leaving a pond as tenerals, and moving to and from ponds as mature adults. Dragonflies moving during larger scale dispersal, where they would encounter new landscapes and habitats, may differ in their responses to matrix conditions compared with the animals we tested, moving in more familiar conditions. Even so, previous studies that have involved translocating individuals to unfamiliar landscapes have also resulted in movements that differ from expected dispersal movements: in some cases individuals have not moved as great a distance as they would during dispersal, and in other cases, they move faster than expected during dispersal (Knowlton & Graham, 2010). Despite these potential differences between movement following translocations and long-distance dispersal, our results do match patterns of dragonfly movement at larger scales (i.e. Chin & Taylor, 2009), suggesting that some species of adult dragonfly move best in open environments across scales. At larger scales (i.e. ³ 500 m), forests decrease dispersal to reproductive habitats (Chin & Taylor, 2009). At smaller scales (e.g. 125 m), forests do not seem to impact movement to nearby aquatic habitats, either as a matrix en route to an aquatic habitat or when dragonflies are released in forests (Chin & Taylor, 2009; French & McCauley, unpublished data), although canopy cover surrounding reproductive habitats can restrict arrival by dragonfly colonists (Remsburg et al., 2008; French & McCauley, 2018). At these small scales, where reproductive habitats may be closer and less costly to reach, dragonflies may need to spend less time searching for new habitats (e.g. Stamps et al., 2005), or else may be arriving at aquatic habitats while also using surrounding forests for daily activities such as roosting (Corbet, 2004; Timofeev, 2016). Travelling through open environments such as fields may improve the chances of detecting an aquatic habitat when individuals choose to undergo larger scale dispersal, or quicken the detection of these habitats. Forests may therefore act as a partial boundary for local movements rather than a complete structural boundary, providing refuge to dragonflies in between foraging attempts or overnight (Corbet, 2004) at small scales, but deterring large-scale dispersal to reproductive habitats (Chin & Taylor, 2009). It is not known whether forests have other effects on dragonfly movement and survival. Fields and forests may support different predator communities and may vary in predation risk, which may also vary seasonally, potentially leading to differential risks to

68 dragonflies temporally, and spatially across a heterogeneous landscape, and thus changes in dragonflies’ movement behaviour (Brown, 1999; Haynes & Cronin, 2006).

A prominent forest matrix in a landscape may therefore reduce functional connectivity when distances between pond habitats are greater (e.g. Stevens et al., 2004). Scale-dependent connectivity has also been observed in damselflies, where small-scale dispersal depended on stream connectivity and open landscapes facilitated large-scale dispersal (Keller & Holderegger, 2013). Given the behavioural preferences for open matrix environments that we observed here and results from other studies (i.e. Chin & Taylor, 2009; French & McCauley, unpublished data), it is likely that the presence of open matrix environments such as fields improves the chances of a dispersing dragonfly finding a pond habitat. Although dragonflies may fly over forests to avoid such a boundary, the height of flight of many dragonfly species has not been documented, and they may still prefer to fly over open environments to improve their chances of detecting reproductive habitats during large-scale dispersal. We observed 15% of individuals flying directly up after release, and a number of individuals that didn’t fly up immediately gained altitude while flying: both behaviours potentially giving them a better vantage point for further movement. Although larger bodied dragonfly taxa (e.g. Aeshnidae) have been observed to fly higher than the tree canopy (French & McCauley, personal observation), many smaller bodied dragonfly species, such as those used in our study, likely fly lower than tree height, possibly to avoid predation.

Our results showed that adults of dragonfly species that are relative generalists with respect to pond canopy cover at the larval stage (i.e. Leucorrhinia intacta, Sympetrum obtrusum/rubicundulum hybrids, and Sympetrum vicinum; McCauley, 2005b; McCauley et al., 2008; French & McCauley, 2018; Werner et al., unpublished data), have clear preferences for orienting towards open environments but show differences in movement abilities between open and closed environments. We used ‘canopy-generalist’ species to provide conservative estimates of overall dragonfly responses to forest cover; we would expect that species that are specialized to open environments might show stronger preferences for open matrix environments. Dragonflies may differ in how they respond to canopy or forest cover depending on species- specific traits. For example, species with a shorter adult stage may prefer more open environments, whereas those with a longer adult stage may prefer more closed environments (Kadoya et al., 2008). Previous studies, each using a single dragonfly species to test for

69 landscape effects on movement, have found that adults may move more readily through open environments (Libellula depressa in Angelibert & Giani, 2003; Leucorrhinia hudsonica in Chin & Taylor, 2009). However, adult dragonflies may also show plasticity in their behavioural responses to forest boundaries. Such behaviour has been observed in the Fender’s blue butterfly (Icaricia icarioides fenderi, Lepidoptera), which generally prefers to move through open environments but will cross a forest boundary when the forest contains the host plant for its larvae (Schultz et al., 2012). The behavioural responses of adult dragonflies to an ecotonal boundary may be species-specific however, as larval dragonflies can have species-specific distributions across a gradient of pond canopy cover and landscape forest cover (McCauley et al., 2008). Differential responses among species have been observed in two congeneric Nymphalidae spp. (Lepidoptera), where their likelihood of entering a forest may depend on a species-specific perceptual range or ability to evaluate the risk of crossing the boundary (Stasek et al., 2008).

The landscapes across KSR have proportionally even amounts of open and closed land cover, with approximately 49% fields and 51% forests on the western side of the reserve. The reserve was previously an equestrian property on which forest regrowth has occurred relatively recently. In agricultural landscapes that have had longer time periods to experience forest regrowth, such as the Edwin S. George Reserve (ESGR) in Michigan with approximately 70% forested land cover (based on maps by Howard, 1998), larval dragonfly assemblages are found in ponds across the reserve (see McCauley et al., 2008) despite potential forest boundaries. These patterns suggest that within this range of distances between ponds (approximately 10–2700 m; Werner et al., unpublished data), adult dragonflies still colonize aquatic habitats that are surrounded by forests (e.g. McCauley, 2006). However, there is little information on how adult dragonflies in more forested regions respond to open versus closed land cover types while moving among aquatic habitats. In a previous study with more variable landscape cover between a release point and ponds (i.e. either >40% forest cover or >40% open cover), dragonflies appeared to disperse through open landscapes at greater scales (500–1400 m) but traversed forested landscapes at shorter distances (125 m; Chin & Taylor, 2009). Additionally, it is difficult to untangle the historical influences on these patterns of species distributions. The ESGR was historically much less forested, like much of the upper Midwestern and northeastern United States (Foster et al., 1998), and its current levels of forest cover reflect secondary succession following agricultural abandonment. The rates of extinction and colonization in the ponds of the ESGR found that 6 of

11 species had higher rates of extinction than colonization across a period of 13 years (McCauley et al., 2014). Current assemblages of larval dragonflies may therefore be unstable and shifting (McCauley et al., 2014), potentially reflecting patterns of colonization that occurred earlier in this successional sequence, and possibly leading to decreased rates of habitat colonization due to a changing matrix environment.

Our study demonstrates the effects of different matrix types on animal movements. One species in our study, S. vicinum, had shorter flight times in forests than fields, and across species, individuals had marginally shorter flight times with increasing stand complexity (i.e. tree density) in forests. The structural simplicity of a field may therefore play an important role in promoting movement in the landscape for dragonflies. Structurally complex or cluttered environments can have large impacts on movement, particularly for flying organisms. For example, complex environments can slow down flight in a bumblebee species (Bombus impatiens, Hymenoptera), especially for larger bodied individuals (Crall et al., 2015). More complex environments can also affect foraging in larger bat species, as they are less able to manoeuvre or detect prey in denser environments (Patriquin & Barclay, 2003; Schnitzler et al., 2003) and may be injured by running into these structures (Stockwell, 2001). However, environmental complexity can also provide benefits to many species. For some species moving through terrestrial habitats, structural complexity can help to minimize risk for individuals by limiting their exposure to adverse conditions or predators (e.g. Rothermel & Semlitsch, 2002), while also providing access to resources (McElhinny et al., 2005).

On a landscape level, the presence of ecotones or other less prominent features in a landscape may affect how individuals move. Some species of butterfly (e.g. Lycaena helle; Fischer et al., 1999) and bumblebee (e.g. Bombus spp.; Cranmer et al., 2012) use ecotonal edges as a linear feature to direct movement, and dragonflies may also use landmarks in the landscape to direct their flight (Eason & Switzer, 2006). The extent to which dragonflies use linear features to navigate, and how this affects their risk of movement through terrestrial environments, is largely unexplored. Dragonflies may also use forest edges for critical resources including a safe place to roost during their inactive overnight period (Corbet, 2004; Timofeev, 2016). Future work on these topics may provide insight into how features in the landscape affect connectivity for dragonflies and other flying insects, and how a mosaic of forests and fields in the landscape might be important for adult dragonflies and other semi-aquatic taxa. In addition, when assessing

71 habitat connectivity for restoration or conservation purposes, incorporating the responses of multiple species to the landscape may be ideal (e.g. Ricketts, 2001), although more difficult to accomplish.

3.5 Conclusion

Forest regrowth is occurring in some regions around the world (Hansen et al., 2013), particularly across mid-latitudinal North America following the abandonment of agricultural fields (Foster et al., 1998; Foster et al., 2003). Our results, taken with those of previous studies, suggest that some dragonfly species show a marked preference for moving in open terrestrial environments. Consequently, increased forest cover may negatively affect connectivity among aquatic environments for dragonflies, potentially decreasing the movement of animals among ponds, and as a result reducing genetic, species, and functional diversity among populations, communities, and ecosystems (Crooks & Sanjayan, 2006; Dixo et al., 2009; Staddon et al., 2010). However, this is not a general response across taxa, and our results should not, therefore, be interpreted to mean that forest removal is an effective tool to facilitate connectivity between ponds for all dragonflies, particularly since dragonflies use forest edges for roosting (Corbet, 2004; Timofeev, 2016), or for other pond-dwelling species. Understanding individuals’ responses to ecotonal boundaries is vital to predicting their success in moving across changing landscapes to colonize habitats.

3.6 References

Angelibert, S. & N. Giani, 2003. Dispersal characteristics of three odonate species in a patchy habitat. Ecography 26: 13–20.

Baguette, M. & H. Van Dyck, 2007. Landscape connectivity and animal behavior: functional grain as a key determinant for dispersal. Landscape Ecology 22: 1117–1129.

Bélisle, M., 2005. Measuring landscape connectivity: the challenge of behavioral landscape ecology. Ecology 86: 1988–1995.

Brown, J. S., 1999. Vigilance, patch use and habitat selection: foraging under predation risk. Evolutionary Ecology Research 1: 49–71.

Calabrese, J. M. & W. F. Fagan, 2004. A comparison-shopper’s guide to connectivity metrics. Frontiers in Ecology and the Environment 2: 529–536.

Chahl, J. & A. Mizutani, 2012. Biomimetic attitude and orientation sensors. IEEE Sensors Journal 12: 289–297.

Chardon, J. P., F. Adriaensen & E. Matthysen, 2003. Incorporating landscape elements into a connectivity measure: a case study for the Speckled wood butterfly (Pararge aegeria L.). Landscape Ecology 18: 561–573.

Chen, J., S. C. Saunders, T. R. Crow, R. J. Naiman, K. D. Brosofske, G. D. Mroz, B. L. Brookshire & J. F. Franklin, 1999. Microclimate in forest ecosystem and landscape ecology. BioScience 49: 288–297.

Chin, K. S. & P. D. Taylor, 2009. Interactive effects of distance and matrix on the movements of a peatland dragonfly. Ecography 32: 715–722.

Conrad, K. F., K. H. Willson, I. F. Harvey, C. J. Thomas & T. N. Sherratt, 1999. Dispersal characteristics of seven odonate species in an agricultural landscape. Ecography 22: 524– 531.

Corbet, P. S., 2004. Dragonflies: behaviour and ecology of Odonata, Revised ed. Harley Books, Colchester UK.

Cranmer, L., D. McCollin & J. Ollerton, 2012. Landscape structure influences pollinator movements and directly affects plant reproductive success. Oikos 121: 562–568.

Crooks, K. R. & M. Sanjayan, 2006. Connectivity conservation: maintaining connections for nature. In: Crooks, K. R. & M. Sanjayan (eds), Connectivity conservation. Cambridge University Press, Cambridge UK: 1–19.

Csabai, Z., P. Boda, B. Bernáth, G. Kriska & G. Horváth, 2006. A ‘polarisation sun-dial’ dictates the optimal time of day for dispersal by flying aquatic insects. Freshwater Biology 51: 1341–1350.

Damschen, E. I., D. V. Baker, G. Bohrer, R. Nathan, J. L. Orrock, J. R. Turner, L. A. Brudvig, N. M. Haddad, D. J. Levey & J. J. Tewksbury, 2014. How fragmentation and corridors affect wind dynamics and seed dispersal in open habitats. Proceedings of the National Academy of Sciences 111: 3484–3489.

De Marco, P. Jr, J. D. Batista & H. S. R. Cabette, 2015. Community assembly of adult odonates in tropical streams: an ecophysiological hypothesis. PLoS ONE 10: e0123023.

Dixo, M., J. P. Metzger, J. S. Morgante & K. R. Zamudio, 2009. Habitat fragmentation reduces genetic diversity and connectivity among toad populations in the Brazilian Atlantic Coastal Forest. Biological Conservation 142: 1560–1569.

Driscoll, D. A., S. C. Banks, P. S. Barton, D. B. Lindenmayer & A. L. Smith, 2013. Conceptual domain of the matrix in fragmented landscapes. Trends in Ecology & Evolution 28: 605– 613.

Eason, P. K. & P. V. Switzer, 2006. Spatial learning in dragonflies. International Journal of Comparative Psychology 19: 268–281.

Fischer, K., B. Beinlich & H. Plachter, 1999. Population structure, mobility and habitat preferences of the violet copper Lycaena helle (Lepidoptera: Lycaenidae) in Western Germany: implications for conservation. Journal of Insect Conservation 3: 43–52.

Forman, R. T. T. & M. Godron, 1981. Patches and structural components for a landscape ecology. BioScience 31: 733–740.

Forman, R. T. T. & M. Godron, 1986. Landscape ecology. John Wiley & Sons, New York NY.

Foster, D. R., G. Motzkin & B. Slater, 1998. Land-use history as long-term broad-scale disturbance: regional forest dynamics in central New England. Ecosystems 1: 96–119.

Foster, D., F. Swanson, J. Aber, I. Burke, N. Brokaw, D. Tilman & A. Knapp, 2003. The importance of land-use legacies to ecology and conservation. BioScience 53: 77–88.

French, S. K. & S. J. McCauley, 2018. Canopy cover affects habitat selection by adult dragonflies. Hydrobiologia 818: 129–143.

Gordon, D. M., 2011. The fusion of behavioral ecology and ecology. Behavioral Ecology 22: 225–230.

Haddad, N. M., 1999. Corridor use predicted from behaviors at habitat boundaries. The American Naturalist 153: 215–227.

Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L. Chini, C. O. Justice & J. R. G. Townshend, 2013. High-resolution global maps of 21st-century forest cover change. Science 342: 850–853.

Haynes, K. J. & J. T. Cronin, 2006. Interpatch movement and edge effects: the role of behavioral responses to the landscape matrix. Oikos 113: 43–54.

Howard, T. G., 1998. Edwin S. George Reserve GIS database and maps. Created for the University of Michigan Museum of Zoology and the Edwin S. George Reserve, Ann Arbor MI, using ARC/INFO version 7.1.

Jacobs, M. E., 1955. Studies on territorialism and sexual selection in dragonflies. Ecology 36: 566–586.

Joly, P., C. Miaud, A. Lehmann & O. Grolet, 2001. Habitat matrix effects on pond occupancy in newts. Conservation Biology 15: 239–248.

Kadoya, T., 2009. Assessing functional connectivity using empirical data. Population Ecology 51: 5–15.

Kadoya, T., S.-i. Suda, Y. Tsubaki & I. Washitani, 2008. The sensitivity of dragonflies to landscape structure differs between life-history groups. Landscape Ecology 23: 149–158.

Keller, D. & R. Holderegger, 2013. Damselflies use different movement strategies for short- and long-distance dispersal. Insect Conservation and Diversity 6: 590–597.

Knowlton, J. L. & C. H. Graham, 2010. Using behavioral landscape ecology to predict species’ responses to land-use and climate change. Biological Conservation 143: 1342–1354.

Kracker, L. M., 1999. The geography of fish: the use of remote sensing and spatial analysis tools in fisheries research. The Professional Geographer 51: 440–450.

Kuefler, D., B. Hudgens, N. M. Haddad, W. F. Morris & N. Thurgate, 2010. The conflicting role of matrix habitats as conduits and barriers for dispersal. Ecology 91: 944–950.

Lenth, R., 2018. Emmeans: estimated marginal means, aka least-squares means. R package. https://CRAN.R-project.org/package=emmeans.

Lima, S. L. & P. A. Zollner, 1996. Towards a behavioral ecology of ecological landscapes. Trends in Ecology & Evolution 11: 131–135.

McCauley, S., 2005a. Differential dispersal propensities between individuals in male Leucorrhinia intacta (Odonata: Libellulidae). International Journal of Odonatology 8: 223–232.

McCauley, S. J., 2005b. Species distributions in anisopteran odonates: effects of local and regional processes. PhD thesis. University of Michigan, Ann Arbor MI.

McCauley, S. J., 2006. The effects of dispersal and recruitment limitation on community structure of odonates in artificial ponds. Ecography 29: 585–595.

McCauley, S. J., 2007. The role of local and regional processes in structuring larval dragonfly distributions across habitat gradients. Oikos 116: 121–133.

McCauley, S. J., T. Brodin & J. Hammond, 2010. Foraging rates of larval dragonfly colonists are positively related to habitat isolation: results from a landscape-level experiment. The American Naturalist 175: E66–E73.

McCauley, S. J., C. J. Davis, R. A. Relyea, K. L. Yurewicz, D. K. Skelly & E. E. Werner, 2008. Metacommunity patterns in larval odonates. Oecologia 158: 329–342.

McCauley, S. J., C. J. Davis, E. E. Werner & M. S. Robeson II, 2014. Dispersal, niche breadth and population extinction: colonization ratios predict range size in North American dragonflies. Journal of Animal Ecology 83: 858–865.

McElhinny, C., P. Gibbons, C. Brack & J. Bauhus, 2005. Forest and woodland stand structural complexity: its definition and measurement. Forest Ecology and Management 218: 1–24.

Michiels, N. K. & A. A. Dhondt, 1990. Costs and benefits associated with oviposition site selection in the dragonfly Sympetrum danae (Odonata: Libellulidae). Animal Behaviour 40: 668–678.

Murphy, H. T. & J. Lovett-Doust, 2004. Context and connectivity in plant metapopulations and landscape mosaics: does the matrix matter? Oikos 105: 3–14.

Nowicki, P., V. Vrabec, B. Binzenhöfer, J. Feil, B. Zakšek, T. Hovestadt & J. Settele, 2014. Butterfly dispersal in inhospitable matrix: rare, risky, but long-distance. Landscape Ecology 29: 401–412.

Oertli, B., 2008. The use of dragonflies in the assessment and monitoring of aquatic habitats. In Córdoba-Aguilar, A. (ed), Dragonflies and damselflies: model organisms for ecological and evolutionary research. Oxford University Press, Oxford UK: 79–96.

Patriquin, K. J. & R. M. R. Barclay, 2003. Foraging by bats in cleared, thinned and unharvested boreal forest. Journal of Applied Ecology 40: 646–657.

Pewsey, A., M. Neuhäuser & G. D. Ruxton, 2013. Circular statistics in R. Oxford University Press, Oxford UK.

Pohlert, T., 2014. The pairwise multiple comparison of mean ranks package (PMCMR). R package. http://CRAN.R-project.org/package=PMCMR.

Prugh, L. R., K. E. Hodges, A. R. E. Sinclair & J. S. Brashares, 2008. Effect of habitat area and isolation on fragmented animal populations. Proceedings of the National Academy of Sciences 105: 20770–20775.

R Core Team, 2017. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. http://www.R-project.org/.

Remsburg, A. J., A. C. Olson & M. J. Samways, 2008. Shade alone reduces adult dragonfly (Odonata: Libellulidae) abundance. Journal of Insect Behavior 21: 460–468.

Ricketts, T. H., 2001. The matrix matters: effective isolation in fragmented landscapes. The American Naturalist 158: 87–99.

Rittenhouse, T. A. G. & R. D. Semlitsch, 2006. Grasslands as movement barriers for a forest- associated salamander: migration behavior of adult and juvenile salamanders at a distinct habitat edge. Biological Conservation 131: 14–22.

Ross, J. A., S. F. Matter & J. Roland, 2005. Edge avoidance and movement of the butterfly Parnassius smintheus in matrix and non-matrix habitat. Landscape Ecology 20: 127–135.

Schnitzler, H.-U., C. F. Moss & A. Denzinger, 2003. From spatial orientation to food acquisition in echolocating bats. Trends in Ecology & Evolution 18: 386–394.

Schultz, C. B., A. M. A. Franco & E. E. Crone, 2012. Response of butterflies to structural and resource boundaries. Journal of Animal Ecology 81: 724–734.

Semlitsch, R. D., 2008. Differentiating migration and dispersal processes for pond-breeding amphibians. Journal of Wildlife Management 72: 260–267.

Serrano-Meneses, M. A., A. Córdoba-Aguilar, V. Méndez, S. J. Layen & T. Székely, 2007. Sexual size dimorphism in the American rubyspot: male body size predicts male competition and mating success. Animal Behaviour 73: 987–997.

Staddon, P., Z. Lindo, P. D. Crittenden, F. Gilbert & A. Gonzalez, 2010. Connectivity, non- random extinction and ecosystem function in experimental metacommunities. Ecology Letters 13: 543–552.

Stamps, J. A., V. V. Krishnan & M. L. Reid, 2005. Search costs and habitat selection by dispersers. Ecology 86: 510–518.

Stasek, D. J., C. Bean & T. O. Crist, 2008. Butterfly abundance and movements among prairie patches: the roles of habitat quality, edge, and forest matrix permeability. Environmental Entomology 37: 897–906.

Stevens, V. M., E. Polus, R. A. Wesselingh, N. Schtickzelle & M. Baguette, 2004. Quantifying functional connectivity: experimental evidence for patch-specific resistance in the Natterjack toad (Bufo calamita). Landscape Ecology 19: 829–842.

Stockwell, E. F., 2001. Morphology and flight manoeuvrability in New World leaf-nosed bats (Chiroptera: Phyllostomidae). Journal of Zoology 254: 505–514.

Switzer, P. V. & W. Walters, 1999. Choice of lookout posts by territorial amberwing dragonflies, Perithemis tenera (Anisoptera: Libellulidae). Journal of Insect Behavior 12: 385–398.

Timofeev, A. N., 2016. The ecology and behavioral features of dragonflies (Insecta, Odonata) in the forest–steppe zone of Central Chernozem region. Russian Journal of Ecology 47: 501–507.

Van Dyck, H. & M. Baguette, 2005. Dispersal behaviour in fragmented landscapes: routine or special movements? Basic and Applied Ecology 6: 535–545.

Volpe, N. L., W. D. Robinson, S. J. K. Frey, A. S. Hadley & M. G. Betts, 2016. Tropical forest fragmentation limits movements, but not occurrence of a generalist pollinator species. PLoS ONE 11: e0167513.

Chapter 4 Local Habitat Conditions and Landscape Effects Determine the Recovery of Larval Dragonfly Assemblages Following Pond Drying

This paper will be submitted for publication as: French, S. K., M. F. Benard, J. T. Hoverman, R. A. Relyea, D. K. Skelly, E. E. Werner, K. L. Yurewicz & S. J. McCauley. Local habitat conditions and landscape effects determine the recovery of larval dragonfly assemblages following pond drying.

Abstract

Local conditions at a habitat as well as the composition of the landscape surrounding habitats can have strong effects on species distributions. Following a disturbance, community recovery at a habitat may depend on how well dispersers recolonize habitats, in part driven by their responses to the heterogeneous landscape surrounding habitats. To assess the impacts of local conditions and landscape type on community reassembly following a disturbance – pond drying – we surveyed larval dragonfly communities over 15 years, in ponds situated in a landscape containing field and forest environments. We predicted that local conditions including an increase in pond canopy cover, and a decrease in pond permanence and area, as well as landscape variables, specifically increased forest cover and distance between ponds, would interact to limit diversity and slow the reassembly of larval assemblages following drying, by limiting adult dispersal and colonization. Local variables (i.e. more canopy cover, and decreased pond permanence and area) reduced the diversity of larval dragonflies across years (p < 0.03). Landscape variables alone did not predict larval distributions, but did interact with local variables to affect larval diversity. Species generally colonized ponds within the first year following a drying event, independent of local or landscape conditions (p > 0.2 in all cases). During recovery of pond assemblages following drying, greater levels of pond canopy cover resulted in an increase in relative diversity and more similar assemblages (p < 0.02 in both cases) compared to pre-drying diversity, although open ponds had greater overall diversity and abundances than closed ponds. Landscape connectivity also affected diversity after drying, particularly when interacting with canopy cover. Local habitat conditions therefore strongly affect the distribution of larval dragonflies, however

80 81 the interaction between local and landscape variables may better predict how semi-aquatic species detect and colonize aquatic habitats following a disturbance.

4.1 Introduction

Local habitat conditions and processes are instrumental in shaping communities (Ricklefs, 1987; Leibold et al., 2004). For communities linked by dispersal, local processes based on the quality of a habitat patch (e.g. species-sorting and mass effects) and competition (e.g. patch dynamics) have strong effects on species distributions and the assembly of communities (Leibold et al., 2004; Logue et al., 2011). However, the landscape context in which habitat patches are situated, including the type of land cover in between or surrounding habitat patches, can be an important determinant of species distributions in heterogeneous landscapes. These conditions can exert effects during dispersal when landscape cover can facilitate or limit movement (Baguette & Van Dyck, 2007) or as individuals select among habitats (Resetarits, 2005). The type of landscape surrounding a suitable habitat for a population or community, both immediately at a habitat boundary and at greater spatial scales, can affect how individuals of a species move and disperse (Haynes & Cronin, 2006), detect habitat (Lima & Zollner, 1996; Fahrig, 2007), and ultimately whether species arrive or establish at habitat patches (Ricketts, 2001; Donald & Evans, 2006; Herse et al., 2017). The ability of organisms to move through a landscape to reach a new habitat, or the functional connectivity among habitats, will therefore depend heavily on how dispersing individuals respond to landscape cover. Heterogeneity in the landscape may result in differential movement costs and benefits for dispersing individuals, potentially affecting the distance or time an organism travels before finding a suitable habitat (e.g. Chardon et al., 2003; Stamps et al., 2005; Wood et al., 2018).

For habitats affected by disturbances, landscape context may affect how quickly species are able to re-colonize these habitats. Community reassembly generally follows a disturbance, with some members of the original community potentially persisting through the disturbance (Chase, 2003; Symons & Arnott, 2014). Local habitat conditions can affect the quality and perceptibility of the target habitat (e.g. Binckley & Resetarits, 2007; Fahrig, 2007), and thus whether organisms establish or persist at the habitat following a disturbance. For example, canopy cover over aquatic habitats can lower water temperatures and productivity, reducing growth rates in frog species that are not specialized to these conditions (Skelly et al., 2002). Even when primary

82 productivity is standardized between open and closed canopy conditions, some frog and aquatic beetle species still lay eggs in open aquatic habitats, perhaps because they do not perceive closed canopy habitats (Binckley & Resetarits, 2007). However, community recovery can also be delayed by landscape factors. Disturbed habitats that are more isolated from a source habitat may take longer to achieve pre-disturbance levels of diversity (e.g. Damschen & Brudvig, 2012; Flinn & Vellend, 2005). Higher connectivity between habitats can accelerate community recovery by facilitating the movement of organisms. The presence of corridors, which are compositionally similar to the habitat where a focal organism lives, can improve movement among open habitats (e.g. for plant species; Damschen & Brudvig, 2012), as can shorter distances between source and disturbed habitats (e.g. for tree species; Flinn & Vellend, 2005). For habitats that are not connected by compositionally similar corridors, the land cover type in between two habitats (i.e. matrix environments, which organisms can move through but do not reside in) can either facilitate or restrict movement, potentially affecting the arrival of organisms at new habitats. For example, many butterfly species prefer to move through open environments such as meadows and prairies; when these habitats are surrounded by a forest matrix, individuals choose to move through more open and less dense forest environments (Ricketts, 2001; Ross et al., 2005; Schultz et al., 2012). In contrast, some amphibian species prefer to more through forested areas as opposed to open environments when moving among aquatic reproductive habitats (Rothermel & Semlitsch, 2002). However, heterogeneity in landscape cover directly at and in between habitats are not always considered in tandem in studies of community assembly or reassembly following disturbance (Biswas & Wagner, 2012).

Understanding this interaction between local and landscape processes can be crucial in determining how or whether communities assemble or reassemble following disturbances. Dispersal has direct effects on local community assembly (Ricklefs, 1987; Chase, 2003; Leibold et al., 2004). Additionally, landscape factors may contribute to the composition of local communities, including how habitat patches are arranged in a landscape, the presence of physical barriers to movement, and large-scale differences in landscape cover (Biswas & Wagner, 2012; Ryberg & Fitzgerald, 2015; Zarnetske et al., 2017). However, more local landscape conditions such as the type of land cover immediately surrounding a habitat can also affect community assembly (Biswas & Wagner, 2012; Le Gall et al., 2018). The consideration of both local and landscape effects on species distributions can give us better insight into how populations and

83 communities are structured, and can improve our ability to predict how well communities can recover from either natural or anthropogenic disturbance and/or what scales to target for restoration, especially if the landscape constrains local conditions or species’ return to the regional species pool (e.g. Ngai et al., 2008; Matthews et al., 2009).

In freshwater systems, habitat permanence is an important local driver of community assembly (Wellborn et al., 1996; Urban, 2004; McCauley et al., 2008). Temporary habitats are typically occupied by organisms that have life-history adaptations which allow them to tolerate frequent and/or predictable drying events. These include having the capacity to persist (e.g. as eggs or other drought-resistant forms), shorter developmental periods, and/or the ability to re-colonize quickly following a drying event (Wellborn et al., 1996; Jocqué et al., 2007; Vanschoenwinkel et al., 2010). Aquatic habitats that experience intermittent drying may, however, serve as sinks for species without these adaptations. These species may become locally extinct, reducing diversity in these habitats (e.g. McCauley et al., 2008; Amundrud & Srivastava, 2015; Gleason & Rooney, 2018). Taxa that are semi-aquatic and actively disperse have the added difficulty of transiting through the terrestrial landscape from a source habitat to reach and colonize new habitats (e.g. dragonflies and frogs; Skelly et al., 1999; McCauley et al., 2008; Werner et al., 2009).

In terrestrial landscapes, forest cover has a major influence on species movement and distributions. The presence of forests between habitats can restrict dispersal (e.g. butterflies; Schultz et al., 2012) and species movement (e.g. some toad species; Stevens et al., 2004). Tree cover at a habitat can also impact species’ distributions, even in aquatic systems. At pond habitats, surrounding or overhanging canopy cover can result in lowered primary production (Schiesari, 2006), which may result in reduced population growth rates and abundances in taxa of higher trophic levels (e.g. some frog species; Skelly et al., 2002; Werner et al., 2009). Canopy cover may also limit some species’ ability to detect and colonize aquatic habitats (Binckley & Resetarits, 2007; French & McCauley, 2018).

The type of landscape cover in the matrix therefore has the potential to affect community reassembly processes following a disturbance: by altering connectivity between source and sink habitats, and by shaping local conditions, which can affect the re-colonization process following a disturbance. We analyzed the effects of forest cover in the landscape and pond canopy cover on larval dragonfly distributions in ponds, and on community assembly in a set of semi-permanent

84 ponds following natural drying events. Semi-permanent habitats were defined here as those ponds which dried intermittently, as opposed to temporary ponds, which usually dry every year (McCauley, 2007). We used dragonflies as our study taxon, as they have differential responses to landscape cover types during their terrestrial adult stage (Chin & Taylor, 2009; Chapter 3) and thus may be affected by landscape heterogeneity when dispersing to aquatic habitats, and because flying insects are a common active colonizer of aquatic habitats following drying (Vanschoenwinkel et al., 2007, 2010). Firstly, we assessed the effects of local conditions (i.e. pond canopy cover, hydroperiod, and area) and landscape conditions (i.e. connectivity to nearby ponds based on distance and landscape cover) on the diversity of larval dragonflies across a 15- year period, to establish the drivers of larval dragonfly assembly in ponds. We expected that an increase in pond canopy cover and decrease in hydroperiod and area would result in decreased larval diversity as has been observed in previous studies of dragonfly diversity at the ESGR (see McCauley et al., 2008). We also expected that isolation generated both by distance and by landscape permeability would affect pond diversity. Specifically, we predicted that focal ponds at increasing isolation from other ponds (e.g. McCauley, 2006) including more forest cover in the landscape between these ponds would result in decreased diversity at a focal pond. Secondly, we tested the degree to which larval dragonfly communities in ponds recover following a drying event, and how their recovery is shaped by local habitat conditions, distance to source habitats, and the composition of land cover in the matrix. Recovery was evaluated as differences in pre- and post-drying diversity, including how compositionally similar assemblages were following a drying event, and the time it took for non-diapausing species to return to a pond, as they would not be able to persist locally through a drying period. We predicted that forest cover and distance between ponds, along with greater levels of canopy cover at ponds, would slow the reassembly of larval communities following drying. These conditions would then result in lower diversity, by limiting adult dispersal and colonization. We tested these predictions using a dataset from a long-term survey that collected dragonflies and other organisms in a standardized way. The surveyed aquatic habitats were located within a single biological reserve but differed in their surrounding terrestrial conditions and in the matrix conditions between the habitats and their potential source ponds.

85 4.2 Methods

To assess the impacts of landscape type on community reassembly following natural pond drying, we used data from a survey that collected aquatic insects and amphibians, including larval dragonflies, from 37 ponds situated among fields and forests over 15 years. The study was conducted between 1996–2010 at the University of Michigan’s Edwin S. George Reserve (ESGR; Michigan, USA, UTM Zone 17, 253360.6 E, 4705953.4 N). The reserve, which is approximately 5.25 km2 in area, consists primarily of fields and forests, with a high density of ponds and both open and forested wetlands (Fig. 4.1). Over the 15-year period, 10 of the 37 ponds were semi-permanent (i.e. remained wet for at least 3 consecutive years following a focal drying event), 23 ponds were temporary (dried most years or remained wet for £ 2 consecutive years following a drying event), and 4 ponds did not dry at all.

Ponds were sampled in May and July of every year between 1996–2010 by a consistent team of researchers. Sampling was conducted using pipe sampling, dipnetting, and seining. In several of these years (1996 and 2003–2010), an additional survey was conducted in June using dipnets only. Data from these 2–3 sampling periods within a year were pooled. More detailed information on the sampling techniques used can be found in Werner et al. (2007). Most larval dragonflies collected across ponds could be identified to species (96% of 14,355 larvae), and we used these individuals in our analyses. The following larvae could not be identified to species, but were nonetheless included as a single species each: Sympetrum internum individuals which were combined with Sympetrum danae/Sympetrum internum individuals that could not be differentiated, Epitheca cynosura/spinigera individuals that could not be differentiated, and Sympetrum obtrusum/rubicundulum individuals, as these species are known to hybridize (Catling, 2007). Although Tramea individuals could not be identified to species, they were retained as a single species so as not to lose the genus in further analyses, and because they likely represent a single species based on observations of adults at the ESGR (Tramea lacerata; McCauley, 2006; McCauley, personal observation). We excluded the remaining larvae that could not be identified to species (4% of individuals) from our analyses. In this study, the ESGR likely supported a single population for most species, with only the adult population of the migratory species Anax junius likely having a greater ecological neighbourhood than the extent of the ESGR. During the larval period, populations are subdivided into discreet subpopulations in a

Fig. 4.1 Map of the Edwin S. George Reserve showing open/field (yellow) and closed/forest (green) landscape cover, as well as open and wooded wetlands (blue). Thirty-seven ponds were sampled every year between 1996–2010 for dragonfly larvae

87 single pond, which do not interact with the subpopulations in other ponds. We therefore consider each pond to have a distinct subpopulation during this larval stage. Canopy cover over all ponds on the ESGR was measured once in the summer of 2001 using a spherical densiometer (Model- C, Forest Densiometers, Rapid City, South Dakota, USA), and was assumed to not change across the 15-year study. Four measurements were taken at water level in each cardinal direction at a single site within a pond, and their average was calculated. This was repeated at 5 different sites per pond (north, south, east, west, and centre of pond), and the average across sites was then taken, resulting in a single measurement of canopy cover per pond. The amount of open and closed forest cover in between pond habitats was quantified based on a GIS database of landscape cover at the ESGR. Mean hydroperiod for each site was calculated based on the proportion of days the pond was wet in a year (i.e. when the pond did not dry up completely), and taking the average of these values across years. Water depth was measured from depth gauges located in each pond and used to determine whether the pond held water or had dried. The area of the ponds was delineated based on the GIS database and a map from Werner et al. (2007). Pond characteristics are summarized in Table S4.1.

Data were analyzed based on two sets of data: one for all 37 ponds where factors related to larval dragonfly diversity in these sites were examined, and one for a subset of the surveyed ponds (including 4 permanent and 10 semi-permanent ponds) where we examined the recovery of larval dragonfly communities following a drying event. Ponds in the subset were selected based on level of pond permanence: all 4 permanent and 10 semi-permanent ponds were included in the subset, whereas 23 temporary ponds that dried frequently were excluded, as our focus was on unpredictable drying events where species did not necessarily have adaptations to drying.

We examined the effects of local and landscape factors on the diversity of larval dragonflies, specifically the mean richness and Shannon diversity per pond averaged across all years, in ponds at the ESGR. Local conditions included pond canopy cover, hydroperiod, and area. Local conditions were combined into a single principal component axis (PC1), as these three variables are highly correlated (McCauley et al., 2008), using the princomp function in R. PC1 described 74% of the variation among ponds. Landscape conditions included the connectivity of a single pond to all other ponds using Euclidean distances (hereafter referred to as ‘distance connectivity’), and using least-cost paths based on open or closed landscape cover (hereafter referred to as ‘land cover connectivity’). Spatial autocorrelation (Moran’s I) in species richness

88 and Shannon diversity was also tested among all ponds using the correlog function in R (ncf package; Bjornstad, 2018) with an increment of 0.1 for distance classes and 99 permutations, to see if ponds that were closer together were similar in diversity across years. A Euclidean distance matrix was then calculated among all 37 ponds based on the centroid of each pond using qgis (v. 2.18.9). Accumulated costs and least cost paths were calculated for each of the 37 ponds as a destination with SAGA (v. 2.2.3) in qgis, using rasterized data of the original ESGR vector data, with a cell size of 10. For the subset of ponds (i.e. 4 permanent and 10 semi-permanent ponds), least cost paths were measured to each semi-permanent pond from all the available source ponds in a given year (i.e. permanent ponds and other semi-permanent ponds that were not dry in the focal time period). For open land cover types (i.e. open water bodies and fields) that were assumed to be easy for dragonflies to move through (Chin & Taylor, 2009; Chapter 3), a resistance value of 1 was assigned, whereas for closed land cover types (i.e. wooded wetlands and forested regions) that were assumed to be difficult to move through, a resistance value of 30 was assigned. These values are based on common levels of differential resistance used in the literature (e.g. forests versus pastures for mammal movement in Gurrutxaga et al., 2011). The choice of the values themselves were somewhat arbitrary (e.g. Chardon et al., 2003), but reflected the difference in resistance between open and closed land cover for various dragonfly species.

Separate calculations of connectivity were conducted to represent distance and land cover connectivity among ponds for all ponds, and for the subset of ponds. Connectivity was calculated using Hanski’s index of connectivity (1999):

!" = $ exp (−,-"()/( ()"

where a is the reciprocal of the dispersal distance of a species (1/dispersal distance in metres), dij is the distance (centroid to centroid) between a pond (i) and a source pond (j), and Aj is the area of the source pond (McCauley, 2006). We used a value of a = 0.001 (indicating an ability to disperse 1000 m; McCauley, 2006) as a representation of adult dragonflies’ general dispersal abilities across species (adults of some species at the ESGR have been observed to disperse 218– 1886 m; McCauley, unpublished data; the maximum distance between ponds at the ESGR is 2745 m). Varying a-values between short and long dispersal distances has not been shown to

89 affect the relationship between diversity and connectivity for dragonflies (McCauley, 2006). This metric of connectivity represents how well a single pond is connected to others in the landscape (i.e. multiple source ponds; greater value indicates greater connectivity).

Generalized linear models (GLMs) were used to test the effects of local (PC1) and landscape conditions on species richness and Shannon diversity (with an added constant of one), separately, using Gamma errors with a log link, as the data were positively skewed. We also tested whether species with diapausing and non-diapausing life history strategies separated across a gradient of local conditions. We did not test for a response to landscape conditions based on life history strategy, as dispersal abilities through the landscape can vary considerably within each group (McCauley, 2007; McCauley, unpublished data). All Aeshna and Sympetrum spp. were included as diapausing species, and all other species were considered to be non-diapausing (see Table S4.2; Walker & Corbet, 1975; Needham et al., 2000). For each pond, the total number of species present in each life history group and the abundance of individuals in each group was determined for every year, and the mean abundance and species richness across years for that pond was calculated. We used a GLM with Gamma errors and a log link to test the interaction between PC1 and life history trait on species richness (plus an added constant of 1) and Shannon diversity (plus 1).

Richness and Shannon diversity were calculated for the year preceding and following a drying event, and the relative change between the two was determined ((final-initial)/(initial+final); e.g. Carreño-Rocabado et al., 2012). Bray-Curtis and Sørensen indices were calculated between the year preceding and following a drying event, using vegdist in R (vegan package; Oksanen et al., 2018). The Bray-Curtis index was used to examine patterns of species abundances and the Sørensen index was used to examine patterns of species presence or absence, to assess how similar communities were before and after a drying index, and whether this changed based on the index used. Generalized linear models were used to test the effects of canopy cover and distance and land cover connectivity on Bray-Curtis and Sørensen values, separately, using quasibinomial errors. We also used GLMs with Poisson errors to test the effects of these three variables on the number of years it took for non-diapausing species to return to a pond following a drying event, in comparison to diapausing species which could persist through a drying event. These statistical analyses were conducted in R (version 3.4.4; R Core Team, 2017).

90 4.3 Results

Across all ponds at the ESGR, local pond conditions including canopy cover, hydroperiod, and area had strong effects on species diversity, whereas landscape conditions had mixed effects on diversity. Along the PC1 axis summarizing local conditions, all three variables loaded strongly, with canopy cover loading positively (0.59), and hydroperiod and area loading negatively (-0.55 and -0.60, respectively) on this axis. An increase in PC1 scores therefore represented an increase in canopy cover, and a decrease in hydroperiod and area. Greater PC1 scores were associated with decreased Shannon diversity (t36 = -2, p = 0.03; Fig. 4.2a), however there was no effect of land cover connectivity (t36 = -0.5, p = 0.6) or distance connectivity (t36 = 0.05, p = 1) on

Shannon diversity. The interaction between PC1 and land cover connectivity was significant (t36

= -3, p = 0.002), but that between PC1 and distance connectivity was not (t36 = 2, p = 0.08). Similar patterns were seen with species richness. Greater PC1 scores resulted in decreased species richness (t36 = -4, p = 0.001; Fig. 4.2b), however there was no effect of land cover connectivity (t36 = -0.4, p = 0.7) or distance connectivity (t36 = 0.2, p = 0.8). The separate interactions between PC1 and each connectivity variable were significant (land cover: t36 = -4, p

< 0.001; distance: t36 = 3, p = 0.006). Interactions among all three variables and between the two landscape variables were not retained in the best model. Across years, ponds showed minimal spatial autocorrelation for both species richness and Shannon diversity, indicating that neighbouring ponds in the landscape did not appear to influence the diversity of one another.

When considering the effects of local conditions on diapausing and non-diapausing species separately, the abundance of dragonfly larvae (Fig. 4.3a) and the number of species (Fig. 4.3b) within each group that were present at a pond decreased with an increase in PC1 (more canopy cover, shorter hydroperiod, and smaller area), but to different degrees (life history strategy ´

PC1: abundance: t73= -6, p < 0.001; richness: z73 = -7, p < 0.001). The slope of the relationship between PC1 and abundance/richness for diapausing species was more gradual than that for diapausing species (Figs. 4.3a and b), suggesting that the canopy/hydroperiod/area gradient had less of an effect on the richness and abundance of species capable of diapause.

Fig. 4.2 Local conditions (PC1) at a site versus the Shannon diversity and richness of dragonfly larvae. Scores of the PC1 axis represented 74% of the variation in local conditions among ponds. An increase in PC1 indicates an increase in mean percent canopy cover, but a decrease in the hydroperiod averaged across 15 years and pond area. Greater PC1 scores were negatively related to the (a) mean Shannon diversity and (b) mean richness of larval dragonfly species per site, averaged across all years

Fig. 4.3 Local conditions (PC1) versus the abundance and richness of diapausing and non- diapausing dragonfly larvae. Scores of the PC1 axis represented 74% of the variation in local conditions among ponds. An increase in PC1 indicates an increase in mean percent canopy cover, but a decrease in the hydroperiod averaged across 15 years and pond area. Greater PC1 scores were negatively related to the (a) mean abundance of dragonfly larvae across years and (b) mean richness of larvae across years. Non-diapausing species (black circles and solid lines) were more likely to be found across PC1 (had a more gradual slope), whereas diapausing species (white circles and dashed lines) were more prominent in open, large habitats with long hydroperiods. A single pond is represented twice for each PC1 score

For the subset of semi-permanent ponds that experienced a drying disturbance in a given year, upon examining the recovery of their communities after drying, canopy cover had strong effects on diversity in the year following the drying event, whereas connectivity had mixed effects. For semi-permanent ponds, relative species richness and Shannon diversity of ponds were greater in the year following a drying event in comparison with the year before the drying event in response to greater levels of local canopy cover (p = 0.01 and p = 0.02, respectively; Table 4.1). Connectivity to source ponds, both in terms of land cover and distance, had mixed effects on diversity, as did interactions with canopy (see Table 4.1). Shannon diversity was greater following a drying event with greater land cover connectivity (p = 0.03) and with the interaction between local and landscape conditions (Table 4.1), but was not affected by isolation by distance alone (p = 0.1). Species richness was greater following a drying event when considering the interaction between local canopy cover and connectivity (canopy ´ land cover: p = 0.02; canopy ´ distance: p = 0.02), but did not depend on connectivity measures alone (land cover: p = 0.3; distance: p = 0.6). Canopy cover affected Bray-Curtis turnover, a measure of community composition based on species abundances, with ponds with greater canopy cover being more similar before and after a drying event (p = 0.03; Table 4.1). Connectivity also had strong effects on Bray-Curtis turnover with more connected sites having more similar community composition before and after a drying event (land cover: p = 0.02; distance: p = 0.01; Table 4.1). However, when comparing community composition before and after drying based on presence/absence- based community comparisons (Sørensen index), we did not find an effect of local or landscape conditions (p > 0.1 in all cases; Table 4.1).

Across ponds, larvae of diapausing species were all found within one year following a drying event. Many larvae of non-diapausing species also returned to ponds within one year following a drying event (i.e. 7 out of the 10 semi-permanent ponds), with abundances of these individuals representing between 17–87% of the individuals found at those ponds. However, neither canopy cover (z9 = 1, p = 0.2), distance connectivity (z9 = -0.4, p = 0.7), or land cover connectivity (z9 = -0.3, p = 0.7) affected the timing of non-diapausing species’ return to ponds.

4.4 Discussion

Canopy cover, hydroperiod, and area were important drivers of local diversity of larval dragonflies in pond communities at the ESGR. An increase in canopy cover at ponds was

Table 4.1 The effects of canopy cover, and land cover (least-cost) and distance (Euclidean) connectivity, on diversity patterns during community reassembly of ponds following a disturbance created by natural pond drying. Connectivity indices assumed a dispersal distance of 1000 m. The relative species richness and Shannon diversity, and the Bray-Curtis and Sørensen indices, were calculated between the year before and the year following a drying event. Significant effects are in bold and marginal effects are in italics

Response Predictor t value p value

Relative richness Canopy -10 0.01 Land cover connectivity -2 0.3 Distance connectivity -0.6 0.6 Canopy ´ land cover connectivity 6 0.02 Canopy ´ distance connectivity 7 0.02 Land cover ´ distance connectivity -0.2 0.9 Canopy ´ land cover ´ distance connectivity -3 0.08 Relative Shannon Canopy -5 0.02 diversity Land cover connectivity 4 0.03 Distance connectivity 2 0.1 Canopy ´ land cover connectivity 4 0.03 Canopy ´ distance connectivity 4 0.02 Land cover ´ distance connectivity -5 0.02 Bray-Curtis Canopy -3 0.03 Land cover connectivity 4 0.02 Distance connectivity -5 0.01 Canopy ´ land cover connectivity -2 0.08 Canopy ´ distance connectivity 3 0.03 Sørensen Canopy -2 0.1 Land cover connectivity 2 0.1 Distance connectivity -0.5 0.6

95 associated with lower diversity, whereas longer hydroperiods and greater pond area were positively related to local diversity (i.e. Shannon diversity and species richness; Figs. 4.2a and b). Connectivity, both in terms of distance among ponds and the least cost path among ponds based on forest cover did not influence the overall pattern of larval diversity on their own. However, the interaction between local effects (PC1) and these landscape terms had significant effects on local diversity: specifically land cover connectivity affected Shannon diversity and both land cover and distance connectivity affected species richness. These results indicate that connectivity in the landscape may play a role in determining larval dragonfly distributions when interacting with local conditions.

In general, species were quick to re-colonize ponds, often within the year following a drying event across diapausing and non-diapausing life history strategies. During the recovery of communities following drying, an increase in canopy cover resulted in greater Shannon diversity and species richness, as well as an increase in Bray-Curtis similarity, but did not affect the Sørensen index (Table 4.1). These unexpected increases in diversity in more closed ponds may have been a result of a number of factors. More open canopy ponds tended to be completely wet in the year(s) preceding a drying event, which may have led to more drastic decreases in absolute abundances and diversity at these sites. Closed canopy ponds on the other hand tended to experience frequent drying events in the years leading up to the focal drying event so already had reduced abundances and diversity. Open ponds also tended to have more individuals than closed canopy ponds in the years before and after a drying event, but closed ponds had a greater relative increase in individuals following a drying event. Open ponds also had more species than closed ponds before a drying event, but the two were relatively similar after drying. Connectivity also affected diversity: land cover connectivity directly increased Shannon diversity, whereas the interaction between canopy and land cover/distance connectivity affected Shannon diversity and species richness. Greater connectivity also increased community similarity between the year before and after a drying event. These results suggest that factors at both the local (i.e. canopy cover) and landscape (land cover and distance) scale influence diversity and community recovery following a disturbance, including compositional changes (i.e. Bray-Curtis index) before and after a drying event. There were no effects of life history strategy (i.e. diapausing or non- diapausing) on time to re-colonize a pond, suggesting that non-diapausing species are quick to re-colonize ponds following a drying event, and their abundances can be restored rather quickly,

96 while diapausing species are able to persist and also re-establish their numbers soon after a drying event.

Our results support previous work that suggests that canopy cover is a major influence on larval dragonfly distributions, but that forest cover in the landscape may have mixed effects on adult dispersal to and colonization of ponds. Larvae may not be as abundant at covered ponds due to the negative effects of canopy cover on the arrival of adults, potentially due to the adults’ limited attraction to shaded ponds (French & McCauley, 2018). Canopy cover over ponds does not appear to affect growth or mortality of dragonfly larvae at later instars (French & McCauley, 2018). It is therefore unlikely that the larval distributions observed in our study are a result of larval species-sorting in response to temperature or other abiotic conditions following the colonization of ponds by adults. However, closed canopy habitats generally have lower primary productivity than open habitats (e.g. Schiesari, 2006), which may reduce populations of zooplankton and other small aquatic herbivores thus reducing food availability for earlier stages of dragonfly larvae. Across all ponds and years in our study, local conditions played a stronger role in shaping larval distributions than landscape factors, however there appeared to be important effects of the interaction between local and landscape factors on these distributions. These patterns are in congruence with previous observational studies that suggest that although field environments facilitate dragonfly movement, the movement of adult dragonflies is not completely impeded by landscape forest cover, at least up to distances of 400–1400 m (Chin & Taylor, 2009; French & McCauley, unpublished data).

The ESGR has a high density of forest cover: as of 2001 approximately 60% of the reserve was covered in forests, with 17% open cover, 13% open wetlands, and 10% wooded wetlands. In contrast, previous studies of dragonfly responses to landscape conditions have occurred in areas with lower proportions of forest matrix. These studies examined behaviour of translocated dragonflies and the relationship between matrix conditions and the straight-line distances between a release point and ponds (e.g. a mix of >40% forest and >40% open matrix depending on the site, Chin & Taylor, 2009; 51% forest matrix and 49% open matrix, Chapter 3). Because of the high level of forest cover on the ESGR, an adult dragonfly flying to a destination pond at the ESGR would always encounter forest cover along its route. Thus, at least in this system, canopy cover over ponds appears to play a more important role in determining larval dragonfly distributions than forest cover in between ponds. This is likely due in part to the large amount of

97 forest cover at the ESGR resulting in relatively similar forest matrix conditions surrounding ponds, and therefore little variation in how these matrix conditions affect diversity among ponds. In systems with less overall forest cover, landscape effects may play a stronger role in determining community composition (e.g. Chin & Taylor, 2009). Teneral (i.e. juvenile) and adult dragonflies may also use different parts of the landscape before moving to find reproductive habitats, and/or may use alternative paths en route to aquatic habitats. For example, the ESGR has large patches of field environments that may serve as habitat for dragonflies before and after they start to use reproductive habitats (Corbet, 2004; Eason & Switzer, 2006; Foster & Soluk, 2006). The proximity of field environments to aquatic habitats may therefore be important to consider when examining the connectivity of a destination pond. The ESGR also has a comprehensive network of roads situated under both open and closed landscape cover, which could increase connectivity among ponds if dragonflies can maintain flight along covered roads (e.g. Soluk et al., 2011; Timofeev, 2016). In addition, historical land use, such as agricultural cover in the case of the ESGR, may have persistent effects on aquatic systems (Foster et al., 2003; Surasinghe & Baldwin, 2014), and may better predict species distributions than current land cover.

The dragonfly species surveyed in our study varied in their distributions across a canopy cover gradient. Previous studies have shown species-specific tolerances to canopy cover. Among aquatic beetles (Coleoptera), some species are found solely in open habitats, solely in closed habitats, or in both (Binckley & Resetarits, 2007, 2009). The distribution of these beetle species relies heavily on the behavioural choices of the adults, which can move readily in and out of ponds before reproducing, however the mechanism behind these choices is not clear (Binckley & Resetarits, 2007, 2009). The ponds with heavy canopy in our study (i.e. >79% canopy cover) were dominated by Sympetrum spp., as were several ponds with less canopy (i.e. 32–78% canopy cover). The flexibility of Sympetrum spp. to persist in both open and closed environments may allow them to exploit habitats with fewer competitors and/or predators (e.g. Anderson et al., 2015). This may be particularly important as Sympetrum eggs hatch in spring, when larvae of non-diapausing species that hatched the summer before will be relatively large, potentially imposing risks as intraguild predators. Sympetrum individuals may also be more exploratory, as they tend to have longer teneral periods (Walker & Corbet, 1975; Needham et al., 2000). Adults of these species have also been found at greater distances from aquatic habitats (e.g. in

98 graveyards; Horváth et al., 2007), indicating that these Sympetrum species may have the opportunity to explore more during daily flights. Sympetrum females also oviposit in a number of wet and dry substrates (Michiels & Dhondt, 1990), indicating that they may not be as constrained in where they reproduce as other species.

The results of our study show the importance of understanding how habitat conditions and landscape cover interact to affect community recovery after a disturbance. In this case, the local conditions of a habitat may influence the habitat selection behaviour of adult dragonflies (French & McCauley, 2018) more so than landscape heterogeneity, and thus these local conditions may have the largest role in determining larval community structure. While this interaction is of particular importance for semi-aquatic invertebrates that use the terrestrial landscape during dispersal, it has wide-reaching implications. Changes to the terrestrial matrix within a landscape may reduce habitat connectivity for certain species, inhibiting their dispersal and thus reducing local diversity (Prugh et al., 2008; Koh & Ghazoul, 2010; Ruffell et al., 2017), but must be considered alongside local conditions.

Furthermore, for aquatic habitats, drying events, which are becoming more frequent with climate change due to increased evapotranspiration (Johnson et al., 2010), may interact with habitat fragmentation to reduce habitat connectivity, jeopardizing the persistence of species and the recovery of communities. For example, the loss of wetlands due to land use change may inhibit community recovery in response to drying events, by creating greater distances among aquatic habitats (Gibbs, 1993). However, if the landscape between habitats is also changing due to deforestation or forest regrowth, species may interact differently with their environment, thus affecting the dispersal and resultant distribution of organisms (Foster et al., 1998; Gurevitch & Padilla, 2004; Lagrue et al., 2011). Further studies are therefore needed to determine how communities recover from disturbances in different landscape and regional contexts.

4.5 References

Amundrud, S. L. & D. S. Srivastava, 2015. Drought sensitivity predicts habitat size sensitivity in an aquatic ecosystem. Ecology 96: 1957–1965.

Anderson, T. L., B. H. Ousterhout, W. E. Peterman, D. L. Drake & R. D. Semlitsch, 2015. Life history differences influence the impacts of drought on two pond-breeding salamanders. Ecological Applications 25: 1896–1910.

Baguette, M. & H. Van Dyck, 2007. Landscape connectivity and animal behavior: functional grain as a key determinant for dispersal. Landscape Ecology 22: 1117–1129.

Biswas, S. R. & H. H. Wagner, 2012. Landscape contrast: a solution to hidden assumptions in the metacommunity concept? Landscape Ecology 27: 621–631.

Bjornstad, O. N., 2018. Spatial nonparametric covariance functions. R package. https://CRAN.R- project.org/package=ncf.

Carreño-Rocabado, G., M. Peña-Claros, F. Bongers, A. Alarcón, J.-C. Licona & L. Poorter, 2012. Effects of disturbance intensity on species and functional diversity in a tropical forest. Journal of Ecology 100: 1453–1463.

Catling, P. M., 2007. Variation of hind-wing colour and length in Sympetrum internum (Odonata: Libellulidae) from the Canadian prairie provinces. Canadian Entomologist 139: 872–880.

Chase, J. M., 2003. Community assembly: when should history matter? Oecologia 136: 489–498.

Chin, K. S. & P. D. Taylor, 2009. Interactive effects of distance and matrix on the movements of a peatland dragonfly. Ecography 32: 715–722.

100

Corbet, P. S., 2004. Dragonflies: behaviour and ecology of Odonata, Revised ed. Harley Books, Colchester UK.

Damschen, E. I. & L. A. Brudvig, 2012. Landscape connectivity strengthens local–regional richness relationships in successional plant communities. Ecology 93: 704–710.

Donald, P. F. & A. D. Evans, 2006. Habitat connectivity and matrix restoration: the wider implications of agri-environment schemes. Journal of Applied Ecology 43: 209–218.

Eason, P. K. & P. V. Switzer, 2006. Spatial learning in dragonflies. International Journal of Comparative Psychology 19: 268–281.

Fahrig, L., 2007. Non-optimal animal movement in human-altered landscapes. Functional Ecology 21: 1003–1015.

Flinn, K. M. & M. Vellend, 2005. Recovery of forest plant communities in post-agricultural landscapes. Frontiers in Ecology and the Environment 3: 243–250.

Foster, D. R., G. Motzkin & B. Slater, 1998. Land-use history as long-term broad-scale disturbance: regional forest dynamics in central New England. Ecosystems 1: 96–119.

Foster, D., F. Swanson, J. Aber, I. Burke, N. Brokaw, D. Tilman & A. Knapp, 2003. The importance of land-use legacies to ecology and conservation. BioScience 53: 77–88.

French, S. K. & S. J. McCauley, 2018. Canopy cover affects habitat selection by adult dragonflies. Hydrobiologia 818: 129–143.

Gibbs, J. P., 1993. Importance of small wetlands for the persistence of local populations of wetland-associated animals. Wetlands 13: 25–31.

Gleason, J. E. & R. C. Rooney, 2018. Pond permanence is a key determinant of aquatic macroinvertebrate community structure in wetlands. Freshwater Biology 63: 264–277.

101

Gurevitch, J. & D. K. Padilla, 2004. Are invasive species a major cause of extinctions? Trends in Ecology & Evolution 19: 470–474.

Gurrutxaga, M., L. Rubio & S. Saura, 2011. Key connectors in protected forest area networks and the impact of highways: a transnational case study from the Cantabrian Range to the Western Alps (SW Europe). Landscape and Urban Planning 101: 310–320.

Hanski, I., 1999. Metapopulation ecology. Oxford University Press, Oxford UK.

Haynes, K. J. & J. T. Cronin, 2006. Interpatch movement and edge effects: the role of behavioral responses to the landscape matrix. Oikos 113: 43–54.

Herse, M. R., M. E. Estey, P. J. Moore, B. K. Sandercock & W. A. Boyle, 2017. Landscape context drives breeding habitat selection by an enigmatic grassland songbird. Landscape Ecology 32: 2351–2364.

Jocqué, M., B. J. Riddoch & L. Brendonck, 2007. Successional phases and species replacements in freshwater rock pools: towards a biological definition of ephemeral systems. Freshwater Biology 52: 1734–1744.

Johnson, W. C., B. Werner, G. R. Guntenspergen, R. A. Voldseth, B. Millett, D. E. Naugle, M. Tulbure, R. W. H. Carroll, J. Tracy & C. Olawsky, 2010. Prairie wetland complexes as landscape functional units in a changing climate. BioScience 60: 128–140.

Koh, L. P. & J. Ghazoul, 2010. A matrix-calibrated species-area model for predicting biodiversity losses due to land-use change. Conservation Biology 24: 994–1001.

Lagrue, C., J. S. Kominoski, M. Danger, J.-M. Baudoin, S. Lamothe, D. Lambrigot & A. Lecerf, 2011. Experimental shading alters leaf litter breakdown in streams of contrasting riparian canopy cover. Freshwater Biology 56: 2059–2069.

102

Le Gall, M., M. Fournier, A. Chaput-Bardy & A. Husté, 2018. Determinant landscape-scale factors on pond odonate assemblages. Freshwater Biology 63: 306–317.

Lima, S. L. & P. A. Zollner, 1996. Towards a behavioral ecology of ecological landscapes. Trends in Ecology & Evolution 11: 131–135.

Matthews, J. W., A. L. Peralta, D. N. Flanagan, P. M. Baldwin, A. Soni, A. D. Kent & A. G. Endress, 2009. Relative influence of landscape vs. local factors on plant community assembly in restored wetlands. Ecological Applications 19: 2108–2123.

McCauley, S. J., 2006. The effects of dispersal and recruitment limitation on community structure of odonates in artificial ponds. Ecography 29: 585–595.

McCauley, S. J., 2007. The role of local and regional processes in structuring larval dragonfly distributions across habitat gradients. Oikos 116: 121–133.

McCauley, S. J., C. J. Davis, R. A. Relyea, K. L. Yurewicz, D. K. Skelly & E. E. Werner, 2008. Metacommunity patterns in larval odonates. Oecologia 158: 329–342.

Michiels, N. K. & A. A. Dhondt, 1990. Costs and benefits associated with oviposition site selection in the dragonfly Sympetrum danae (Odonata: Libellulidae). Animal Behaviour 40: 668–678.

Needham, J. G., M. J. Westfall Jr. & M. L. May, 2000. Dragonflies of North America, Revised ed. Scientific Publishers, Gainesville FL.

103

Ngai, J. T., K. R. Kirby, B. Gilbert, B. M. Starzomski, A. J. D. Pelletier & J. C. R. Conner, 2008. The impact of land-use change on larval insect communities: testing the role of habitat elements in conservation. Écoscience 15: 160–168.

Oksanen, J., F. G. Blanchet, M. Friendly, R. Kindt, P. Legendre, D. McGlinn, P. R. Minchin, R. B. O’Hara, G. L. Simpson, P. Solymos, M. H. H. Stevens, E. Szoecs & H. Wagner, 2018. Vegan: community ecology package. R package. https://CRAN.R- project.org/package=vegan.

R Core Team, 2017. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.

Resetarits, W. J. Jr, 2005. Habitat selection behaviour links local and regional scales in aquatic systems. Ecology Letters 8: 480–486.

Ricketts, T. H., 2001. The matrix matters: effective isolation in fragmented landscapes. The American Naturalist 158: 87–99.

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

Ross, J. A., S. F. Matter & J. Roland, 2005. Edge avoidance and movement of the butterfly Parnassius smintheus in matrix and non-matrix habitat. Landscape Ecology 20: 127–135.

Ruffell, J., M. N. Clout & R. K. Didham, 2017. The matrix matters, but how should we manage it? Estimating the amount of high-quality matrix required to maintain biodiversity in fragmented landscapes. Ecography 40: 171–178.

104

Ryberg, W. A. & L. A. Fitzgerald, 2015. Landscape composition, not connectivity, determines metacommunity structure across multiple scales. Ecography 38: 001–010.

Schiesari, L., 2006. Pond canopy cover: a resource gradient for anuran larvae. Freshwater Biology 51: 412–423.

Schultz, C. B., A. M. A. Franco & E. E. Crone, 2012. Response of butterflies to structural and resource boundaries. Journal of Animal Ecology 81: 724–734.

Skelly, D. K., L. K. Freidenburg & J. M. Kiesecker, 2002. Forest canopy and the performance of larval amphibians. Ecology 83: 983–992.

Skelly, D. K., E. E. Werner & S. A. Cortwright, 1999. Long-term distributional dynamics of a Michigan amphibian assemblage. Ecology 80: 2326–2337.

Soluk, D. A., D. S. Zercher & A. M. Worthington, 2011. Influence of roadways on patterns of mortality and flight behavior of adult dragonflies near wetland areas. Biological Conservation 144: 1638–1643.

Stamps, J. A., V. V. Krishnan & M. L. Reid, 2005. Search costs and habitat selection by dispersers. Ecology 86: 510–518.

Surasinghe, T. & R. F. Baldwin, 2014. Ghost of land-use past in the context of current land cover: evidence from salamander communities in streams of Blue Ridge and Piedmont ecoregions. Canadian Journal of Zoology 92: 527–536.

Symons, C. C. & S. E. Arnott, 2014. Timing is everything: priority effects alter community invasibility after disturbance. Ecology and Evolution 4: 397–407.

Timofeev, A. N., 2016. The ecology and behavioral features of dragonflies (Insecta, Odonata) in the forest–steppe zone of Central Chernozem region. Russian Journal of Ecology 47: 501–507.

105

Urban, M. C., 2004. Disturbance heterogeneity determines freshwater metacommunity structure. Ecology 85: 2971–2978.

Vanschoenwinkel, B., C. De Vries, M. Seaman & L. Brendonck, 2007. The role of metacommunity processes in shaping invertebrate rock pool communities along a dispersal gradient. Oikos 116: 1255–1266.

Walker, E. M. & P. S. Corbet, 1975. The Odonata of Canada and Alaska. Volume 3, Part III: the Anisoptera–three families. University of Toronto Press, Toronto ON.

Wellborn, G. A., D. K. Skelly & E. E. Werner, 1996. Mechanisms creating community structure across a freshwater habitat gradient. Annual Review of Ecology and Systematics 27: 337–363.

Werner, E. E., D. K. Skelly, R. A. Relyea & K. L. Yurewicz, 2007. Amphibian species richness across environmental gradients. Oikos 116: 1697–1712.

Wood, C. W., E. W. Wice, J. del Sol, S. Paul, B. J. Sanderson & E. D. Brodie III, 2018. Constraints imposed by a natural landscape override offspring fitness effects to shape oviposition decisions in wild forked fungus beetles. The American Naturalist 191: 524– 538.

Zarnetske, P. L., B. Baiser, A. Strecker, S. Record, J. Belmaker & M.-N. Tuanmu, 2017. The interplay between landscape structure and biotic interactions. Current Landscape Ecology Reports 2: 12–29.

Chapter 5 General Discussion Thesis Summary

In my thesis, I tested the effects of local and landscape heterogeneity on the distribution of larval dragonflies. Specifically, I assessed 1) how canopy cover at a pond habitat affected adult arrival at reproductive aquatic habitats and subsequent larval performance, 2) how forest cover in the terrestrial landscape between ponds affected the small-scale movement of adult dragonflies in response to field and forest land cover, and 3) how pond canopy cover and forest cover in the landscape interacted to influence the distribution of larval dragonflies, and the recovery of larval communities following a drying event in response to these factors. Dragonflies have an aquatic larval stage and assemblages of larvae are connected among aquatic habitats via dispersal during the terrestrial adult stage. Habitat heterogeneity, a key driver of differences in structure among communities (Leibold et al., 2004; Logue et al., 2011), may therefore influence the larval and/or adult stage of dragonflies. Differential mortality following habitat colonization (i.e. species- sorting) in response to habitat heterogeneity is an important mechanism structuring aquatic communities, and has been shown to affect larval dragonfly assemblages through predation (Knorp & Dorn, 2016) and hydroperiod (McCauley et al., 2008). Environmental conditions that affect dragonfly movement and reproduction at aquatic habitats are also likely to play a role (e.g. McCauley, 2006; Chin & Taylor, 2009). Because adult dragonflies do not appear to detect aquatic predators (McCauley, 2005; Knorp & Dorn, 2016), other environmental gradients are likely influential in shaping larval dragonfly assemblages. Specifically, pond canopy cover and forest cover surrounding habitats can limit the arrival of adult dragonflies at these habitats, and their dispersal to reproductive habitats, respectively (Remsburg et al., 2008; Chin & Taylor, 2009).

My results from Chapter 2 demonstrated that increased canopy cover over aquatic mesocosms resulted in a decrease in the arrival of adult dragonflies. Canopy cover did not however affect the performance of larvae of ‘canopy generalist’ species placed in mesocosms. The species used in this study have been previously observed to be the most abundant species in ponds across a gradient of canopy cover (McCauley et al., 2008; Werner et al., unpublished data). It is possible that species that are more specialized to open aquatic habitats may be more affected by closed

106 107 conditions than those observed and used in these experiments, as might eggs or earlier instar larvae. These results suggest that habitat selection by adult dragonflies plays a formative role in determining larval distributions, and that the habitat conditions created by canopy cover (e.g. lower water temperatures and increased nutrient input) do not detrimentally affect larval performance, at least for canopy generalist species.

In Chapter 3, I established that adult dragonflies, when released at an ecotonal boundary (i.e. forest edge) preferred to direct themselves more towards open field environments than closed forest environments. Individuals had longer flight times and took flight more often in fields than forests, however this response was species-dependent. When released in forests, the amount of canopy cover did not affect the likelihood of individuals taking flight, whereas the density of trees resulted in marginally decreased flight times of individuals. Adult dragonflies therefore make small-scale movements that tend to avoid forested areas, but are capable of manoeuvring through and around forests if necessary to return to open environments at these scales. These responses likely translate to the patterns seen at greater distances in previous studies, where dragonflies are more likely to disperse through open environments to reach aquatic habitats (Chin & Taylor, 2009). Forests however do not seem to be a complete barrier to dragonfly movements, particularly at small scales (Chin & Taylor, 2009).

In Chapter 4, I determined that local conditions (i.e. pond canopy cover, hydroperiod, and area) played a major role in determining larval dragonfly distributions, as did the interaction between local conditions and landscape connectivity. Both local canopy cover and landscape connectivity (amount of forest and distance between ponds) affected species diversity and community composition following a drying event. The consideration of both local and landscape mechanisms may therefore be useful in understanding species distributions, particularly for active dispersers that respond strongly to environmental conditions.

Together, the results of these chapters indicate that both local canopy cover over ponds and forest cover in the landscape are major environmental gradients structuring larval dragonfly assemblages. The interaction between the two types of cover on the movement and habitat selection of adult dragonflies may be key in predicting larval distributions. My results indicate that canopy cover severely reduces the number of dragonfly species and the abundance of individuals visiting aquatic habitats, possibly due to limited visual attraction to covered ponds.

108

However, if adult dragonflies are able to detect covered ponds, by chance or due to a more generalist species-specific response to canopy cover, their larvae can persist under closed conditions (Chapters 2 and 4). Forest cover in the landscape may in general reduce the movement and flight abilities of adult dragonflies (Chin & Taylor, 2009; Chapter 3), however these responses can be species-specific. In landscapes that are dominated by forest cover, adult dragonflies can still colonize pond habitats, as their larvae have been found across a gradient of pond canopy cover and surrounding forest cover (Chapter 4), however their behaviour in these landscapes is yet undocumented. The presence of ecotonal boundaries (i.e. forest edges) may repel adult dragonflies (Chapter 3), but may also provide them with roosting habitat (Corbet, 2004; Timofeev, 2016). It is therefore possible that a mix of cover types in the landscape facilitates the dispersal of dragonflies to aquatic habitats and their persistence in an area. At greater distances to an aquatic habitat, open environments may be easiest to move through, and may offer better vantage points to detect these habitats. However, at small scales, forest cover in the vicinity of an aquatic habitat may provide refuge for adults.

Environmental filtering during dispersal and habitat selection therefore seem to be important mechanisms driving larval dragonfly distributions, whereas post-colonization species-sorting does not, at least for the developmental stage examined (i.e. mid- to late-instar larvae). Studies on community assembly across environmental gradients generally focus on differential post- colonization responses (Leibold et al., 2004; Logue et al., 2011). However, my thesis demonstrates that for some taxa, it is important to consider organisms’ movement and habitat selection behaviours, particularly in response to the landscapes surrounding habitats, in order to better understand species assembly. Although predation is an important gradient that drives differences in organismal behaviour and mortality (e.g. Wellborn et al., 1996; Binckley & Resetarits, 2005), my results show the influence of other environmental gradients, such as canopy and forest cover, on species’ distributions. When the drivers of species’ distributions in response to environmental gradients are not known, studies conducted at multiple life history stages across an individual’s ontogeny can be particularly beneficial: in this case during a dragonfly’s adult stage when dispersing and selecting habitats, and during the post-colonization larval stage.

109 5.1 Conservation Implications

Forest regrowth, which is occurring in many regions across Europe and North America following agricultural abandonment (Foster et al., 1998; Hansen et al., 2013; Navarro & Pereira, 2015), has the potential to influence the interactions between species and their environment, thereby affecting their dispersal and distribution (Foster et al., 1998; Gurevitch & Padilla, 2004; Lagrue et al., 2011). In the studies I conducted for my thesis, canopy and forest cover affected how dragonflies moved through the landscape and colonized aquatic habitats. Forest cover at larger scales than examined here has also impacted the dispersal of dragonflies to aquatic habitats (Chin & Taylor, 2009). Canopy and forest cover therefore have the potential to affect functional connectivity among habitat patches and dragonfly communities. Although I did not detect any local effects of canopy cover on the performance of larval dragonflies, it is possible that certain species may be more tolerant to open conditions and may experience reduced performance under closed conditions, where habitats can have decreased temperatures and light availability, and increased organic input (Richardson, 2008).

Land use and land cover change can lead to habitat fragmentation, which for many species can impede dispersal among populations and among communities by isolating species in remote patches of habitat and decreasing the likelihood of individuals finding appropriate or high quality habitat (Fahrig, 2003; McCoy et al., 2013). For species that prefer to move through open or uncluttered environments, such as dragonflies, even forest regrowth may result in a loss of connectivity among aquatic habitats, especially if the landscape was historically open. In this case, fragmentation that creates more open landscapes may be beneficial to some taxa, as has been shown for some butterfly species using logged corridors to move between open habitats (Haddad & Tewksbury, 2005).

Habitat fragmentation is a common occurrence in anthropogenically modified terrestrial systems (Haddad et al., 2015), but can also occur among aquatic habitats. The loss of aquatic habitats, whether permanent due to land use change (Wood et al., 2003) or temporary due to drying events (e.g. Boulton, 2003; Johnson et al., 2010), can result in greater distances among prospective habitats. Semi-permanent ponds, the focal habitats in Chapter 4, are particularly vulnerable to drying with climate change (Johnson et al., 2010), and can result in species losses if species are not adapted to drying events (Wellborn et al., 1996). If individuals can disperse, the type of cover

110 in the landscape may influence their ability to find and establish at new habitats (Chapter 4). Changes in the land cover surrounding aquatic habitats also have the potential to interact with the loss of aquatic habitat to affect the success of species establishing at new habitats. For example, in the face of wetland loss, forest cover in between aquatic habitats may facilitate the persistence of Blanding’s and musk turtles in an area (Quesnelle et al., 2013). Habitats may become more difficult to detect or reach however, especially if species rely on perceptual ranges developed for historical landscapes (Lima & Zollner, 1996), or if species cannot adjust their dispersal behaviour to changes in the landscape (Desrochers & Hannon, 1997; McCoy et al., 2013).

5.2 Future Directions

For communities connected by dispersal, a consideration of the effects of heterogeneity on processes preceding colonization is imperative. Spatial heterogeneity has the potential to facilitate or impede dispersal, as well as the detection and arrival at a habitat in the case of active dispersers, all of which can affect connectivity among habitats. A growing body of literature also highlights the importance of habitat selection by individuals in determining species distributions. Depending on the taxa, the integration of observations of individual-based behaviour with population- and/or community-based patterns may better elucidate how species are distributed.

5.2.1 Multiple Life History Stages

An increasing number of studies are focussing on how the life history stages of a species affect one another (e.g. carry-over effects; Benard & McCauley, 2008; Smith et al., 2009). For species with life history stages that span different environments (e.g. semi-aquatic insects), conducting studies across their ontogeny may give a more complete understanding as to their habitat requirements (Kolding, 1981; Orians & Wittenberger, 1991) and their distributions. The documentation of patterns at one stage, such as the patterns of reduced larval dragonfly diversity and abundance with greater levels of pond canopy cover, smaller pond area, and shorter hydroperiods (McCauley et al., 2008; Chapter 4), may give an incomplete picture of the mechanisms driving species distributions at this stage. Although larval dragonflies experience differential mortality along a gradient of pond permanence (Wellborn et al., 1996), they may be less sensitive to changes in pond canopy cover (Chapter 2). Instead, the selection of open aquatic habitats by adult dragonflies may determine the distributions of larval dragonflies across this environmental gradient (Chapter 2). Thus, considering species’ responses to environmental

111 conditions across their life history may better inform our understanding of where species are found.

For actively dispersing taxa, movement between habitats may occur during a single life history stage, but can affect individuals across their ontogeny. During their early life history stages, individuals may select habitats that satisfy resource requirements across life history stages (e.g. polychaete worm larvae; Thiyagarajan et al., 2006), or may face reduced fitness at later life history stages due to the selection of suboptimal habitat (Stamps et al., 2005). Choosing habitats may also occur during later life history stages, particularly if individuals are spatially confined during earlier stages, as is the case for dragonflies (Chapter 2). Adult individuals can minimize risks to fitness by reproducing in high quality habitats, such as beetle species laying eggs on high quality host plants (Heisswolf et al., 2005), or by investing in multiple habitats during reproduction. For example, the fire salamander (Salamandra infraimmaculata) can deposit eggs across a number of temporary pools at risk of drying up, in order to maximize their potential number of offspring through risk-spreading (Segev, 2011).

Knowledge of habitat selection behaviour and habitat requirements at multiple life history stages would allow for comprehensive habitat conservation and/or restoration plans, enabling species conservation efforts to target the most vulnerable life history stage of a species, and/or to develop conservation plans specific to each life history stage (Smith et al., 2009). Similarly, knowledge of these species-specific behaviours and requirements can allow for insight into the mechanisms of dispersal by invasive species, in order to better control or prevent the spread of the species (Aubry et al., 2006). Overall, knowing how and why organisms select habitat can help to determine how the connectivity of habitats changes across life history stages of organisms, and how changes in habitat conditions can impact populations and communities.

5.2.2 Community Assembly Involving Multiple Taxa

My thesis primarily focussed on assemblages of dragonflies, the species of which are functionally similar. However, many taxa are found in habitats with dragonflies, such as mayflies, midges, and mosquitoes (see Chapter 2). Changes in the behaviour or dispersal patterns of one functional group in response to land use change such as forest regrowth may affect species at other trophic levels. Dispersal can determine the assembly of communities and can affect species interactions, such as local predator-prey dynamics and competition (Palmer et al.,

112

1996; Verreydt et al., 2012). Trophic interactions and community composition in particular can be affected by the arrival of a top predator, especially when dispersal rates into a system are high (Verreydt et al., 2012). Species that attempt to colonize a habitat may then be influenced by the presence of predators or competitors, such as mosquitoes preferring to breed in pools without predatory notonectids (Eitam & Blaustein, 2004). Larval dragonflies are predators in aquatic systems, and therefore have the potential to shape aquatic communities (Stoks & McPeek, 2003). However, actively dispersing taxa at lower trophic levels, which colonize similar aquatic habitats to dragonflies, may respond similarly to dragonflies in response to a change in the landscape such as forest regrowth. For example, some midge and mayfly species may also prefer to move through open environments (Delettre & Morvan, 2000; Petersen et al., 2004), and thus could continue to share similar habitats with dragonflies. More holistic approaches to community assembly, involving multiple taxa, would therefore provide a more complete picture of how entire communities and food webs respond to spatial heterogeneity and habitat fragmentation. For example, future studies could conduct network analyses for a range of semi-aquatic taxa while considering species-specific dispersal abilities and responses to spatial heterogeneity, to determine the connectivity of aquatic systems and communities based on landscape variables (e.g. Saura et al., 2011; Bishop-Taylor et al., 2015; Albert et al., 2017).

5.2.3 Combining Local and Landscape Scales

Studies spanning multiple spatial scales are time- and resource-intensive but can provide valuable ecological information. Not only can the movement choices by individuals occur across multiple scales (e.g. Hutto, 1985), but the choices made by individuals at local scales may also inform patterns observed at landscape scales (e.g. Haddad, 1999). The consideration of multiple scales to address questions in ecology can also result in the integration of information from several ecological levels, including behavioural, population, community, landscape, and ecosystem ecology, to develop more integrated and holistic approaches to explaining ecological patterns (Schoener, 1986; Lima & Zollner, 1996; Levin et al., 2000). This information can be particularly valuable in predicting the outcomes of large-scale environmental perturbations, such as climate change, where species loss or extirpation from an area may be best explained by a combination of individual-level and community responses to local changes (Woodward et al., 2010).

113 5.3 References

Albert, C. H., B. Rayfield, M. Dumitru & A. Gonzalez, 2017. Applying network theory to prioritize multispecies habitat networks that are robust to climate and land-use change. Conservation Biology 31: 1383–1396.

Aubry, S., C. Labaune, F. Magnin, P. Roche & L. Kiss, 2006. Active and passive dispersal of an invading land snail in Mediterranean France. Journal of Animal Ecology 75: 802–813.

Benard, M. F. & S. J. McCauley, 2008. Integrating across life-history stages: consequences of natal habitat effects on dispersal. The American Naturalist 171: 553–567.

Binckley, C. A. & W. J. Resetarits Jr., 2005. Habitat selection determines abundance, richness and species composition of beetles in aquatic communities. Biology Letters 1: 370–374.

Bishop-Taylor, R., M. G. Tulbure & M. Broich, 2015. Surface water network structure, landscape resistance to movement and flooding vital for maintaining ecological connectivity across Australia’s largest river basin. Landscape Ecology 30: 2045–2065.

Boulton, A. J., 2003. Parallels and contrasts in the effects of drought on stream macroinvertebrate assemblages. Freshwater Biology 48: 1173–1185.

Chin, K. S. & P. D. Taylor, 2009. Interactive effects of distance and matrix on the movements of a peatland dragonfly. Ecography 32: 715–722.

Corbet, P. S., 2004. Dragonflies: behaviour and ecology of Odonata, Revised ed. Harley Books, Colchester UK.

Delettre, Y. R. & N. Morvan, 2000. Dispersal of adult aquatic Chironomidae (Diptera) in agricultural landscapes. Freshwater Biology 44: 399–411.

Desrochers, A. & S. J. Hannon, 1997. Gap crossing decisions by forest songbirds during the post-fledging period. Conservation Biology 11: 1204–1210.

Eitam, A. & L. Blaustein, 2004. Oviposition habitat selection by mosquitoes in response to predator (Notonecta maculata) density. Physiological Entomology 29: 188–191.

114

Fahrig, L., 2003. Effects of habitat fragmentation on biodiversity. Annual Review of Ecology, Evolution, and Systematics 34: 487–515.

Foster, D. R., G. Motzkin & B. Slater, 1998. Land-use history as long-term broad-scale disturbance: regional forest dynamics in central New England. Ecosystems 1: 96–119.

Gurevitch, J. & D. K. Padilla, 2004. Are invasive species a major cause of extinctions? Trends in Ecology & Evolution 19: 470–474.

Haddad, N. M., 1999. Corridor use predicted from behaviors at habitat boundaries. The American Naturalist 153: 215–227.

Haddad, N. M., L. A. Brudvig, J. Clobert, K. F. Davies, A. Gonzalez, R. D. Holt, T. E. Lovejoy, J. O. Sexton, M. P. Austin, C. D. Collins, W. M. Cook, E. I. Damschen, R. M. Ewers, B. L. Foster, C. N. Jenkins, A. J. King, W. F. Laurance, D. J. Levey, C. R. Margules, B. A. Melbourne, A. O. Nicholls, J. L. Orrock, D.-X. Song & J. R. Townshend, 2015. Habitat fragmentation and its lasting impact on Earth’s ecosystems. Science Advances 1: e1500052.

Haddad, N. M. & J. J. Tewksbury, 2005. Low-quality habitat corridors as movement conduits for two butterfly species. Ecological Applications 15: 250–257.

Hutto, R. L., 1985. Habitat selection by nonbreeding, migratory land birds. In Cody, M. L. (ed), Habitat selection in birds. Academic Press, Orlando FL: 455–476.

115

Knorp, N. E. & N. J. Dorn, 2016. Mosquitofish predation and aquatic vegetation determine emergence patterns of dragonfly assemblages. Freshwater Science 35: 114–125.

Kolding, S., 1981. Habitat selection and life cycle characteristics of five species of the amphipod genus Gammarus in the Baltic. Oikos 37: 173–178.

Levin, P. S., N. Tolimieri, M. Nicklin & P. F. Sale, 2000. Integrating individual behavior and population ecology: the potential for habitat-dependent population regulation in a reef fish. Behavioral Ecology 11: 565–571.

Lima, S. L. & P. A. Zollner, 1996. Towards a behavioral ecology of ecological landscapes. Trends in Ecology & Evolution 11: 131–135.

McCauley, S. J., 2005. Species distributions in anisopteran odonates: effects of local and regional processes. PhD thesis. University of Michigan, Ann Arbor MI.

116

McCauley, S. J., 2006. The effects of dispersal and recruitment limitation on community structure of odonates in artificial ponds. Ecography 29: 585–595.

McCauley, S. J., C. J. Davis, R. A. Relyea, K. L. Yurewicz, D. K. Skelly & E. E. Werner, 2008. Metacommunity patterns in larval odonates. Oecologia 158: 329–342.

McCoy, E. D., K. A. Basiotis, K. M. Connor & H. R. Mushinsky, 2013. Habitat selection increases the isolating effect of habitat fragmentation on the gopher tortoise. Behavioral Ecology and Sociobiology 67: 815–821.

Navarro, L. M. & H. M. Pereira, 2015. Rewilding abandoned landscapes in Europe. In Pereira, H. M. & L. M. Navarro (eds), Rewilding European landscapes. Springer, New York NY: 3–23.

Orians, G. H. & J. F. Wittenberger, 1991. Spatial and temporal scales in habitat selection. The American Naturalist 137: S29–S49.

Palmer, M. A., J. D. Allan & C. A. Butman, 1996. Dispersal as a regional process affecting the local dynamics of marine and stream benthic invertebrates. Trends in Ecology & Evolution 11: 322–236.

Petersen, I., Z. Masters, A. G. Hildrew & S. J. Ormerod, 2004. Dispersal of adult aquatic insects in catchments of differing land use. Journal of Applied Ecology 41: 934–950.

Quesnelle, P. E., L. Fahrig & K. E. Lindsay, 2013. Effects of habitat loss, habitat configuration and matrix composition on declining wetland species. Biological Conservation 160: 200– 208.

Remsburg, A. J., A. C. Olson & M. J. Samways, 2008. Shade alone reduces adult dragonfly (Odonata: Libellulidae) abundance. Journal of Insect Behavior 21: 460–468.

Richardson, J. S., 2008. Aquatic arthropods and forestry: effects of large-scale land use on aquatic systems in Nearctic temperate regions. Canadian Entomologist 140: 495–509.

117

Saura, S., C. Estreguil, C. Mouton & M. Rodríguez-Freire, 2011. Network analysis to assess landscape connectivity trends: application to European forests (1990–2000). Ecological Indicators 11: 407–416.

Schoener, T. W., 1986. Mechanistic approaches to community ecology: a new reductionism? American Zoologist 26: 81–106.

Segev, O., M. Mangel, N. Wolf, A. Sadeh, A. Kershenbaum & L. Blaustein, 2011. Spatiotemporal reproductive strategies in the fire salamander: a model and empirical test. Behavioral Ecology 22: 670–678.

Smith, R. F., L. C. Alexander & W. O. Lamp, 2009. Dispersal by terrestrial stages of stream insects in urban watersheds: a synthesis of current knowledge. Journal of the North American Benthological Society 28: 1022–1037.

Stamps, J. A., V. V. Krishnan & M. L. Reid, 2005. Search costs and habitat selection by dispersers. Ecology 86: 510–518.

Stoks, R. & M. A. McPeek, 2003. Predators and life histories shape Lestes damselfly assemblages along a freshwater habitat gradient. Ecology 84: 1576–1587.

Thiyagarajan, V., L. Soo, P. K. S. Shin & P. Y. Qian, 2006. Spatio-temporal variation in sediment biochemistry alters larval habitat selection and juvenile performance in the polychaete Capitella sp. I. Marine Ecology Progress Series 327: 207–222.

Timofeev, A. N., 2016. The ecology and behavioral features of dragonflies (Insecta, Odonata) in the forest–steppe zone of Central Chernozem region. Russian Journal of Ecology 47: 501–507.

Verreydt, D., L. De Meester, E. Decaestecker, M.-J. Villena, K. Van Der Gucht, P. Vannormelingen, W. Vyverman & S. A. J. Declerck, 2012. Dispersal-mediated trophic interactions can generate apparent patterns of dispersal limitation in aquatic metacommunities. Ecology Letters 15: 218–226.

118

Wellborn, G. A., D. K. Skelly & E. E. Werner, 1996. Mechanisms creating community structure across a freshwater habitat gradient. Annual Review of Ecology and Systematics 27: 337–363.

Wood, P. J., M. T. Greenwood & M. D. Agnew, 2003. Pond biodiversity and habitat loss in the UK. Area 35: 206–216.

Woodward, G., D. M. Perkins & L. E. Brown, 2010. Climate change and freshwater ecosystems: impacts across multiple levels of organization. Philosophical Transactions of the Royal Society B 365: 2093–2106.

Appendix A: Supplementary Materials for Chapter 2

Fig. S2.1 Mesocosms with closed and open canopy structure. The solid green material generated 50% canopy cover, whereas the more open green material was an open canopy structure, creating no shade on the mesocosms. The open canopy structures created similar structural barriers for flying dragonflies as the shade structures, and served as a sham treatment to control for the effects of the structures without the effects of shading. The mesocosms were covered with white netting pre-experiment to prevent colonization by dragonflies or other aquatic animals. This netting was removed when the experiment started

119 120

Fig. S2.2 Component scores of 20 mesocosms from a principal component analysis of water chemistry variables (dissolved oxygen, pH, conductivity, chlorophyll, and water temperature), grouped by canopy cover level (A: 30–40, B: 55–60, C: 70–75, and D: 90–100%). Only PC1 was retained for further analysis, however both PC1 and PC2 are shown for better visualization

121

Fig. S2.3 Patterns of expected and observed larval and adult abundances, at the Edwin S. George Reserve and in our experiments at the Koffler Scientific Reserve. We see natural patterns of decreasing larval abundance at ESGR with an increase in percent canopy cover (Natural larvae; McCauley, 2005a; McCauley et al., 2008; Werner et al., unpublished data). If this decrease were due to species-sorting, we would expect a consistent number of adult visits across canopy conditions (Adult null if larval SS). Instead, we observed a decrease in adult visits to mesocosms across the canopy gradient (Adult experiments), and did not witness any changes in larval abundance or survival across this gradient (Larval experiments). Therefore, we expect that habitat selection is a strong driver of larval dragonfly distributions

122

Table S2.1 Loadings of the first principal component axis that explained 75% of the variance among mesocosms for the natural canopy experiments

Variable Component 1

Dissolved oxygen -0.491 pH -0.500 Conductivity 0.436 Chlorophyll -0.356 Temperature -0.438

123

Table S2.2 Tukey’s test results for all pairwise comparisons from a one-way ANOVA comparing water chemistry variables to canopy cover. The ANOVA was conducted on differences in PC1 scores of water chemistry variables (dissolved oxygen, pH, conductivity, chlorophyll, and water temperature) from a principal component analysis, among groupings of canopy cover (30–40, 55–60, 70–75, and 90–100%). The standard error in all cases was 0.7

Canopy cover grouping (%) Canopy cover grouping (%) Estimate t value p value

30–40 55–60 0.9 1 0.5 70–75 3 4 0.006 90–100 4 7 p < 0.001 55–60 70–75 2 3 0.09 90–100 4 5 p < 0.001 70–75 90–100 2 3 0.07

124

Table S2.3 Species of adult dragonflies that visited the source pond, experimental area, and mesocosms during the colonizing experiments. Author information was obtained from the Michigan Odonata Survey (michodonata.org)

Experiment Source pond Experimental area Mesocosms

Artificial shading Aeshna sp. (Fabricius, 1775) Aeshna sp. L. luctuosa Anax junius (Drury, 1773) A. junius L. pulchella Celethemis elisa (Hagen, 1861) C. elisa P. longipennis Erythemis simplicicollis (Say, 1839) E. simplicicollis P. lydia Libellula luctuosa (Burmeister, 1839) L. luctuosa S. obtrusum Libellula pulchella (Drury, 1773) L. pulchella S. vicinum Pachydiplax longipennis (Burmeister, 1839) P. longipennis Plathemis lydia (Drury, 1773) P. lydia Sympetrum obtrusum (Hagen, 1867) S. obtrusum Sympetrum vicinum (Hagen, 1861) S. vicinum Tramea lacerata (Hagen, 1861) T. lacerata Natural canopy Aeshna sp. Aeshna sp. Aeshna sp. Anax junius A. junius L. luctuosa Celethemis elisa Leucorrhinia intacta (Hagen, 1861) L. pulchella Libellula luctuosa L. luctuosa P. lydia Libellula pulchella L. pulchella S. obtrusum Pachydiplax longipennis P. longipennis S. vicinum Sympetrum obtrusum P. lydia Sympetrum vicinum S. obtrusum S. vicinum Tramea lacerata or Pantala sp. (Hagen, 1861)

125

Table S2.4 Mesocosm conditions during the artificial shading experiment. For the control treatment, shading and heating were not artificially manipulated. Values indicate means (± 1 SE) for each treatment

Mesocosm Turbidity Prey density Polarized light Treatment temperature (°C) (% transmission) (individuals mL-1) (µmol m-2 s-1)

Control 23.3 ± 0.2 99.7 ± 0.9 0.3 ± 0.1 1.3 ± 0.2 Heated 26.6 ± 0.1 96.7 ± 0.9 0.3 ± 0.1 1.3 ± 0.2 Shaded 22.7 ± 0.2 99.2 ± 0.6 0.2 ± 0.1 0.9 ± 0.1 Shaded plus heated 26.0 ± 0.1 97.7 ± 0.7 0.3 ± 0 0.9 ± 0.1

126

Table S2.5 Mesocosm conditions during the natural canopy experiment. Values indicate means (± 1 SE) for each treatment

Canopy Mesocosm Prey density Turbidity Chlorophyll Distance Elevation cover temperature (individuals -1 (NTU) (µg L ) (m) (m) (%) (°C) mL-1)

34 ± 0.9 21.5 ± 0.1 1.1 ± 0.3 0.09 ± 0.03 20 ± 7 133 ± 7 281 ± 3 57 ± 0.6 20.5 ± 0.8 1 ± 0 0.06 ± 0.03 13 ± 2 135 ± 8 282 ± 2 72 ± 0.6 16.9 ± 0.4 0.9 ± 0.1 0.2 ± 0.1 17 ± 4 138 ± 7 283 ± 2 93 ± 1 15.9 ± 0.5 0.4 ± 0.1 0.3 ± 0.3 6 ± 3 137 ± 8 283 ± 2

Canopy Dissolved Polarized light Conductivity Adult Larval cover pH oxygen (µmol m-2 s-1) (µS cm-1) abundance abundance (%) (mg L-1)

34 ± 0.9 8.4 ± 0.1 28 ± 1 265 ± 12 9.8 ± 0.3 3 ± 0.7 10 ± 8 57 ± 0.6 8.2 ± 0 23 ± 4 270 ± 10 9.2 ± 0.1 2 ± 0.6 20 ± 14 72 ± 0.6 7.9 ± 0.1 10 ± 3 275 ± 13 7.2 ± 0.8 2 ± 1.4 6 ± 6 93 ± 1 7.7 ± 0 3 ± 0.6 329 ± 12 6.3 ± 0.3 0.2 ± 0.2 7 ± 7

Appendix B: Supplementary Materials for Chapter 3

a b l Dragonfly flight direction l Dragonfly flight direction l Wind direction l Wind direction

Fig. S3.1 Flight direction of adult dragonflies (black circles) and wind direction (grey circles) measured using (a) an anemometer and (b) a portable weather meter. Wind direction did not correspond to the flight direction of dragonflies (p < 0.005 in both cases)

127

Appendix C: Supplementary Materials for Chapter 4

Table S4.1 Characteristics of each of the permanent and semi-permanent ponds. Ponds with fish present had minnow species

Year of focal Canopy Potential Fish Permanence class Pond drying event cover (%) source present

Permanent Burt Pond n/a 61 Yes No Crane Pond n/a 42 Yes Yes East Marsh n/a 34 Yes Yes George Pond n/a 63 Yes No Semi-permanent Cattail Marsh 2006 77 No No Crescent Pond 2006 55 No No Dreadful Hollow 2006 87 No No Fishhook Marsh 2002 49 Yes Yes Ilex Pond 2006 75 No No Southwest Swamp 1999 32 Yes Yes Southwest Woods Pond 2005 88 Yes No Spring Pond North 2006 91 No No Star Pond 2006 59 No No West Marsh Dam Pond 2005 34 Yes Yes

128 129

Table S4.2 List of species of larval dragonflies found across the surveyed ponds

Species Diapausing Species author*

Aeshna canadensis Y Walker (1908) Aeshna clepsydra Y Say (1839) Aeshna constricta Y Say (1839) Aeshna (Rhionaeschna) mutata Y Hagen (1861) Aeshna tuberculifera Y Walker (1908) Aeshna umbrosa Y Walker (1908) Aeshna verticalis Y Hagen (1861) Anax junius N Drury (1773) Arigomphus furcifer N Hagen (in Selys, 1878) Basiaeshna janata N Say (1839) Cordulia shurtleffi N Scudder (1866) Dorocordulia libera N Selys (1871) Epiaeshna heros N Fabricus (1798) Epitheca canis N McLachlan (1886) Epitheca cynosura/spinigera N Say (1839) / Selys (1871) Erythemis simplicicollis N Say (1839) Ladona julia N Uhler (1857) Leucorrhinia frigida N Hagen (1890) Leucorrhinia intacta N Hagen (1861) Leucorrhinia proxima N Calvert (1890) Libellula luctuosa N Burmeister (1839) Libellula pulchella N Drury (1773) Libellula quadrimaculata N Linnaeus (1758) Pachydiplax longipennis N Burmeister (1839) Perithemis tenera N Say (1839) Plathemis lydia N Drury (1773) Somatochlora elongata N Scudder (1866) Somatochlora kennedyi N Walker (1918) Somatochlora williamsoni N Walker (1907) Sympetrum ambiguum Y Rambur (1842) Sympetrum danae/internum Y Sulzer (1776) / Montgomery (1943) Sympetrum obtrusum/rubicundulum Y Hagen (1867) / Say (1839) Sympetrum semicinctum Y Say (1839) Sympetrum vicinum Y Hagen (1861) Tramea spp. N Hagen (1861)

* Author information gathered from the Michigan Odonata Survey (michodonata.org)