Macroecological Relationships Between Primary Productivity and Ecological Specialisation

Macroecological relationships between primary productivity and ecological specialisation

Hugh Munro Burley

PhD thesis

School of Biological Earth and Environmental Sciences

University of New South Wales

August 2017

Dedication

For my Grandmother Alice Burley (1917-2011), who was a geographer all her life, but did not have

the chance to attend university.

Acknowledgements

There are many people who deserve a big thank you for all their help. Firstly, I cannot thank

Shawn enough for his patience, encouragement and technical genius — to say nothing of his prompt editorials! One could not ask for a better supervisor. Similarly, I am very grateful to Karel for his invaluable conceptual and technical input. This project took much of Karel’s time, and would not have succeeded without his insight and commitment.

I am very fortunate to have been affiliated with the CSIRO macroecological modelling team.

Firstly, I am indebted to Kristen Williams for helping me apply for the CSIRO scholarship, bringing me into the team and supporting me throughout the process. Simon Ferrier has been very generous with his time, imparting much of his accumulated biogeographic wisdom. I thank him for entrusting me with these ideas, and hope this project has been useful for the team. Many thanks also to Tom

Harwood, who has provided valuable technical and conceptual insights, spatial layers, and of course the chance to play the drums for the final ‘CES who’ gig.

Many other people at CSIRO have also been very helpful. Randall Donohue provided the crucial response variable for this project — without which there would be nothing to explain (!) — and I am very grateful for his prompt and encouraging responses to all my enquiries. Helen

Murphy, Dan Metcalfe and Andrew ford provided valuable insights into the floristic datasets for the

Wet Tropics, improving the analyses substantially. Many thanks also to Dean Paini for his encouragement, and to Greg Forster, Peter Zaffina and Amar Singh for some much needed comradery whilst using the Black Mountain gym during my time in Canberra.

I have also been lucky to have fantastic mentors during the PhD process. Will Cornwell has been a supportive friend and collaborator, while Rob Kooyman has provided valuable botanical insights. I am also indebted to Andrew Thornhill for his work on the Phylogeny of the Wet Tropics flora (for future collaboration), his biogeographic expertise and some classic Australian-isms while enjoying the Mexican cuisine in Berkeley. Many thanks also to Darren Crayn for kindly hosting

Shawn and I at the Tropical Herbarium in Cairns, and to Brent Mishler for his academic guidance.

Adrian Fisher has also been a great friend during my time at UNSW, and I look forward to many more trips together to the SFS and SCG (albeit likely frustrated by the outcome!).

It has been a privilege to receive support from the Australian government, UNSW and

CSIRO. The UNSW Graduate Research School provided funding through the Australian

Postgraduate Award and the UNSW research excellence award, while CSIRO provided funding through the Office of Chief executive top up scholarship, for which I am very grateful. Many thanks must also go to all the UNSW staff who make the school of BEES work: Firoza, Jonathan, Chris and Penny, to mention but a few. Suffice to say we are so lucky to live in Australia with all this support!

Throughout life I’ve relied on my family and friends. Without the support of Jane and Neil, university wouldn’t have been possible, and I can’t thank them enough for putting our education first. Meanwhile everyone at Kanangra deserves a long holiday (perhaps to the Kimberley?). This list is woefully inadequate. But nonetheless, huge thanks to: Bill, Camille (we can, will and did),

Pat, Rowan, Chris, Greg, Kate, Vito, Delwyn — and everyone else (!) — for their friendship. Also

I’m very grateful to Pete Kostic for teaching me the drums again, and to Molly and Parker for their love and friendship during the PhD. Finally I would like to thank Cameron Tonkinwise, formerly of the University of Technology in Sydney, who first introduced the idea of sustainability in 2003 to a bunch of graphic design students, beginning a rambling adventure. Long may it continue!

vii

Abstract

A key debate in contemporary ecology concerns whether ecosystem functions such as productivity are distinctly influenced by biological diversity in natural environments. Recent work has emphasised the importance of links between ecosystem functions and measures of ecological specialisation as proxies of biodiversity. However, few studies have analysed these relationships at broad spatio-temporal scales. This research tests the empirical relationship between primary productivity and ecological specialisation at continental and bioregional scales, using two proxies of specialisation: taxonomic β-diversity and site-level environmental niche width. It also examines how the environmental niches of species vary across continental environmental gradients.

Gross primary productivity (GPP) may be influenced not only by the biological diversity at each location (α-diversity) but also by the biological turnover between locations (β-diversity).

Generalized additive models were used to test whether the magnitude or variability of GPP were distinctly influenced by either taxonomic α- or β-diversity across continental Australia, over and above environmental influences. Neither α- nor β-diversity improved the explanatory power of GPP models beyond that of environment-only models.

The realised environmental niches of species are important indicators of ecological specialisation and biogeographic history. Bivariate regression models were used to test whether species niches vary across continental environmental gradients for 1771 vascular plants from the

Australian Wet Tropics. The temperature niches of these species did not vary substantially.

However, niches were narrower in drier and less fertile environments.

The macroecological complementarity hypothesis predicts that locations with greater ecological specialisation —those with collectively narrower niches — should be more productive

viii than locations with less ecological specialisation. For pairs of environmentally similar Wet Tropics sites, linear models were used to test the pairwise relationship between differences in site GPP

(response) and differences in the median environmental niche width of all tree species present at each site (predictor). Sites with narrower temperature niche widths had higher productivity, whereas sites with narrower rainfall niche widths had lower productivity. These results will improve our understanding of the broad-scale interrelationships between ecosystem functions, environmental conditions and ecological specialisation in natural ecosystems, helping to assess utilitarian arguments for biodiversity conservation.

Table of contents

Dedication ...... v

Acknowledgements ...... vi

Abstract ...... viii

Table of contents ...... x

Publications and conference presentations (2013 – 2017) ...... xiii

1 Introduction ...... 1

1.1 Historical context and aims ...... 1

1.2 Geographic and biological scope of the thesis ...... 11

1.3 Structure of the thesis ...... 14

2 Macroecological relationships between ecosystem functions and biodiversity ...... 17

Abstract ...... 18

2.1 Introduction ...... 19

2.2 Making MB-EF relationships applicable to real ecosystems ...... 23

2.3 A broader view of MB-EF relationships in natural ecosystems ...... 26

Macroecological complementarity ...... 27

Macroecological spatio-temporal compensation ...... 30

2.4 A case study: potential MB-EF relationships for trees across an elevation gradient ...... 33

2.5 Potential avenues for testing MB—EF mechanisms in real ecosystems ...... 38

2.6 Supporting information ...... 42

3 Primary productivity is weakly related to alpha and beta diversity across Australia 44

Abstract ...... 45

3.1 Introduction ...... 46

3.2 Methods ...... 50

Overview ...... 50

3.2.1 Continental datasets ...... 51

3.2.2 Statistical Analyses ...... 58

3.3 Results ...... 61

3.4 Discussion ...... 66

3.5 Supporting information ...... 70

4 Plant niche specialisation varies along environmental gradients in the Australian

Wet Tropics ...... 73

Abstract ...... 74

4.1 Introduction ...... 75

4.2 Methods ...... 80

4.2.1 Wet Tropics datasets ...... 80

4.2.2 Statistical analysis ...... 83

4.3 Results ...... 90

4.4 Discussion ...... 96

4.5 Supporting information ...... 101

5 Primary productivity is related to niche width in the Australian Wet Tropics ...... 104

Abstract ...... 105

5.1 Introduction ...... 106

5.2 Methods ...... 111

Overview ...... 111

5.2.1 Analysis variables used to test the macroecological complementarity mechanism ..... 112

5.2.2 Statistical Analyses ...... 116

5.3 Results ...... 120

5.4 Discussion ...... 125

5.5 Supporting information ...... 131

6 Discussion ...... 132

6.1 Continental primary productivity and geographic beta diversity ...... 132

6.2 Continental variations in species niche width ...... 135

6.3 Bioregional primary productivity and community niche width ...... 138

6.4 Future implications ...... 141

6.5 Conclusions ...... 143

References ...... 145

Appendix 1: Supplementary material for Chapter 2 - Burley et al. (2016a) ...... 159

Appendix 1 references ...... 162

Appendix 2: Supplementary material for Chapter 3 - Burley et al. (2016b) ...... 163

Sensitivity analyses for chapter 3 ...... 163

Varying the radii for calculating β-diversity ...... 166

Using spatial coordinates as an alternative to subsampling ...... 167

Appendix 2 references ...... 172

Appendix 3: Supplementary material for Chapter 4 ...... 174

Appendix 4: Supplementary material for Chapter 5 ...... 189

xii

Publications and conference presentations (2013 – 2017)

Journal articles

. Mokany, K., Burley, H.M. & Paini, D.R. (2013) β Diversity contributes to ecosystem

processes more than by simply summing the parts. Proceedings of the National Academy of

Sciences, 110, E4057

. Burley, H. M., Mokany, K., Ferrier, S., Laffan, S. W., Williams, K. J. & Harwood, T. D.

(2016) Macroecological scale effects of biodiversity on ecosystem functions under

environmental change. Ecology and Evolution, 6, 2579-93.

. Burley, H. M., Mokany, K., Ferrier, S., Laffan, S. W., Williams, K. J. & Harwood, T. D.

(2016) Primary productivity is weakly related to floristic alpha and beta diversity across

Australia. Global Ecology and Biogeography, 25, 1294-1307.

. Harris, K., Burley, H., McLachlan, R., Bowman, M., Macaldowie, A., Taylor, K., Chapman,

M. & Chambers, G. M. (2016) Socio-economic disparities in access to assisted reproductive

technologies in Australia. Reproductive BioMedicine Online, 33, 575-584.

Conference presentations

. International Biogeography Society, Canberra (01/2014, 15 minute oral presentation)

Ecosystem processes and beta-diversity, a macroecological perspective

. Society for Conservation Biology, Suva (06/2014, oral presentation)

A macroecological analysis of relationships between productivity and beta-diversity

xiii

. Institute of Australian Geographers, Canberra (07/2015, oral presentation)

Does floristic niche width vary systematically across environmental space in the wet

tropics?

. Modelling & Simulation Society of AUS & NZ, Gold Coast (11/2015, oral presentation)

Effect of floristic niche width on community-level ecosystem function in the Wet tropics

. Geographic conference, Christchurch (12/2015, oral presentation)

Effect of floristic niche width on community-level ecosystem function in the Wet tropics

. Southern Connections, Puntas Arenas (01/2016, Poster session)

Effect of tree community niche width on primary productivity in the Australian Wet Tropics

. Atlas of Living Australia Symposium, Perth (06/2016, oral presentation)

Species niche widths vary along environmental gradients for the Australian wet tropics flora

. Australian Systematic Botany, Alice Springs (09/2016, oral presentation)

Quantifying floristic beta-diversity of the Australian Wet tropics using Phylogenetic

turnover

. Ecological Society of Australia, Freemantle (12/2016, oral presentation)

Primary productivity is related to niche width in the Australian Wet Tropics

xiv

1 Introduction

1.1 Historical context and aims

The aim of this thesis is to test empirical relationships between gross primary productivity and plant ecological specialisation across broad spatial, temporal and biological scales. Two measures of ecological specialisation are used: the change in the identity of plant species between locations, and the aggregated environmental distributions of all plant species occurring at a location.

Additionally, this thesis examines how individual environmental distributions of Australia’s tropical flora — their realised environmental niches — vary across continental environmental gradients, as a proxy of how specialised each plant species has become to local conditions.

The concept of ‘scale’ is central to all ecological and biogeographical theories, and has several components — notably the extent and resolution of analysis. And so it is necessary to define at the outset how scale is used in this thesis. I use the terms broad-scale and macroecological to refer to ecological analyses using data for the largest possible suite of taxa (for example the

Australian flora, thus constituting a broad biological extent), across the broadest possible geographic extents (for example continental Australia). This is distinct from the resolution component of scale, for which I also use the term fine-scale to denote the finest possible spatial resolution (for example grid cells of 1 km × 1 km).

Before defining ecological specialisation, the broader concept of biological diversity warrants consideration, given the intrinsic historical and thematic links between all concepts of biological rarity and specialisation. Initially, it is important to emphasise that the study of the multitude of organisms with which we share the planet has a long history, across all cultures. From a formal scientific perspective, the rest of life on earth came to be known as biological diversity or

biodiversity in the latter part of the 20th century (Faith, 2008), a term most directly attributed to the

American biologists E.O. Wilson (Wilson, 1988) and Tomas Lovejoy (Lovejoy, 1980). The deceptively simple fact that there is such a variety of different kinds of organisms on earth today, and that their individual and collective distributions vary with such complexity across space and time, poses a fundamental, if not rhetorical, question: what factors drive diversity? The classic ecological definition of biodiversity was devised by the American ecologist Robert Whittaker

(1960), and attempts to capture the variety of life by partitioning diversity into three components:

1). α-diversity: the count, or richness, of biological types at a particular location. For example the total number of species, genera or families recorded at an ecological survey plot of

<1ha in a tropical rainforest.

2). β-diversity: the change in biological identity, or turnover, between two or more locations within a broader region. For example the turnover of plant genera between a pair of ecological survey plots in a tract of tropical rainforest, ranging from 0 (no biological turnover, all taxa shared) to 1 (complete turnover, no taxa shared). β-diversity can thus be used to approximate the degree of biological heterogeneity within a region.

3). γ-diversity: the total number of biological types in a region. For example the total number of plant genera known to occur in a tract of tropical rainforest derived from the aggregate of many ecological survey plots.

Chapter 1 Introduction

These definitions are not absolute. Indeed they are challenged by fundamental questions about the often unexamined philosophical underpinnings of taxonomic classification itself. Despite the inherent limitations of such definitions, and regardless of the particular definition of biodiversity adopted, the core concept uniting them is that of biological variety (IUCN, 1980). This viewpoint is neatly captured by Faith (2016), who described biodiversity as the “living variation” at multiple levels of biological organisation, counted for a set of objects (for example the turnover in species, phenotypes and genotypes between forest plots). This clarifies ‘variability’ as meaning ‘variety’ with regard to the definition of biodiversity adopted by countries around the world as signatories to the Convention on Biological Diversity (CBD):

“Biological diversity means the variability among living organisms from all sources including, inter alia, terrestrial, marine and other aquatic ecosystems and the ecological complexes of which they are part; this includes diversity within species, between species and of ecosystems."

(Secretariat of the Convention on Biological Diversity, 2005), p. 4.

Within this framework, the concept of linking biological identity to place is at the core of defining biodiversity, as well as underpinning all attempts to explain the variety of life and processes that drive evolution, such as the dynamics of environmental conditions. The last 30 years have seen a subtle yet important shift of research focus in the ecological branch of the life sciences, from one of explaining the causes of biodiversity, to examining consequences. This change in focus has been motivated by another fundamental question: does diversity matter? The search for the consequences of biodiversity goes back to at least the writings of Charles Darwin, and probably much earlier. In Chapter IV of The Origin of Species (which outlines the mechanics of Natural

Chapter 1 Introduction

Selection), Darwin discusses the relationship between photosynthetic activity and plant diversity, as yet another illustration of the principle of character divergence:

“The more diversified the descendants from any one species become in structure, constitution and habits, by so much will they be better enabled to seize on many and widely diversified places in the polity of nature.

It has been experimentally proved, that if a plot of ground be sown with one species of grass, and a similar plot be sown with several distinct genera of grasses, a greater number of plants and a greater weight of dry herbage can thus be raised.” (Darwin, 1859) p. 93.

Darwin was speaking here of the taxonomic α-diversity of each respective plot — and by his use of genera, he was also invoking the phylogenetic relationships between the plants (before the genetic basis of natural selection had been revealed, of course). As Darwin hints, the ecological concept that locations in anthropogenic or natural landscapes with greater plant diversity should display greater photosynthetic capacity is longstanding, presumably from many generations of horticultural experimentation. In contemporary ecology, this concept has come to be known as the principle of biological complementarity (Grime, 1998). The core idea is that greater biological richness at a location (i.e. α-diversity) reflects a greater variety of resource acquisition strategies, which in turn should facilitate higher plant photosynthetic activity, and ultimately greater biomass production. By invoking the ecological niche, complementarity harks back to the principle of character divergence as discussed by Darwin, with important consequences for studying fluctuations in plant ecological processes across broad scales.

Chapter 1 Introduction

One of the most important plant ecological processes on earth is primary productivity — the rate at which plants assimilate carbon from the atmosphere through photosynthesis. This thesis focuses solely on gross primary productivity, or GPP, which can be defined as the spatio-temporal photosynthetic flux of carbon between the vegetation and the atmosphere (Donohue et al., 2014).

GPP is thus a biophysical parameter, summarising the photosynthetic activity of a suite of plants at a particular location. It will be shaped not only by the collective attributes of plants at each location, but potentially by plants in the surrounding region, depending on the spatial, temporal and biological scales under investigation.

Gross Primary Productivity (GPP) is but one of many ecosystem functions, a broader ecological concept defined by Ghilarov (2000) as “Stocks or fluxes of matter and energy derived from biological activity”. In this framework, the “dry herbage” referred to by Darwin is the stock

(i.e. the biomass), and GPP is the flux. The ecosystem function concept is more ambiguous and value laden than variables such as GPP and rainfall, because the use of the word ‘function’ implies utility, attaching a somewhat anthropogenic connotation to plant-mediated processes (for example carbon, water and nutrient cycles, or fire regimes). The key to the ecosystem function concept is that it switches biological diversity from the response to the explanatory axis, as a predictor variable used to explain ecosystem functions such as productivity. In reality, measures of productivity, environmental conditions and biological diversity are interrelated at multiple spatio-temporal scales.

These interrelationships are illustrated by the fact that different measures of biological diversity and primary productivity have been used as both response and explanatory variables over several decades of research, often amid considerable controversy [e.g. Grime (1973); Mittelbach et al.

(2001); Lavers and Field (2006); Grace et al. (2014)].

The impetus to switch measures of biological diversity from the response to the explanatory axis was born of conservation movements in Europe and North America in the latter part of the 20th

Chapter 1 Introduction century. The conservation logic behind this transposition of axes is relatively straightforward. In order to preserve the ecosystem functions at a location — for example primary productivity — it is argued that the entire suite of biological taxa occurring there must be preserved (the α-diversity), not simply the dominant species that may drive productivity (for example large trees in a tropical rainforest). The supposedly causal relationship between ecosystem functions (response variables) and biodiversity (predictor variables, referred to throughout this thesis as B-EF research) has been most directly attributed to the work of the American ecologist David Tilman and his colleagues, beginning with a series of field experiments at Cedar Creek in Minnesota (Tilman, 1982). The original B-EF research dates back to at least the early 1980s, although the first peer-reviewed publications in prominent scientific journals appeared later [e.g. Tilman and Downing (1994)]. The

B-EF concept has proliferated in the literature in the years since [e.g. see Cardinale et al. (2012) for a recent review], with thousands of publications on this general topic since 1990.

The typical B-EF experiments test the complementarity mechanism by establishing small plots, approximately 1m2 in size, which are subject to the same environmental conditions. A range of biodiversity treatments are then applied, and the ecosystem function response of each plot is measured over several years (the longest running experiment being Cedar Creek, for approximately

30 years). This format is much the same as that alluded to by Darwin. For example previous experiments have varied the number of species (Gross et al., 2014) and their genetic diversity

(Cadotte, 2013). The utilitarian argument for conserving biodiversity arising from these experiments — that destroying diversity imperils ecosystem functions on which humanity depends

(Faith et al., 2010) — gathered pace with E.O. Wilson’s popularization of the term biodiversity

(Wilson, 1988). Similarly, the momentum to support conservation initiatives generated by the Rio

Earth Summit and the Kyoto Protocol, as well as attempts to quantify the increased rates of extinction driven by human activity [e.g. Vitousek et al. (1997); Ceballos et al. (2015)], have added political weight to the overarching B-EF argument. Despite a substantial volume of literature

Chapter 1 Introduction demonstrating experimental links between different measures of ecosystem functions and biological diversity [e.g. Hooper et al. (2005); Cardinale et al. (2012)], the B-EF concept faces a pivotal test: can the relationships from plot-scale, artificial experiments be generalised to broader continental scales relevant for conservation initiatives?

The empirical evidence for B-EF relationships has thus far proven equivocal, particularly across broad spatio-temporal scales. For example Chisholm et al. (2013) found that relationships between the productivity of forest plots and α-diversity can in fact be negative at spatial resolutions that are coarser than typical B-EF experiments (for example 1 hectare). Some authors have dismissed bivariate analyses of productivity and diversity as incapable of capturing the network of interactions needed to accurately characterise the ecological dynamics in complex systems [e.g.

Grace et al. (2014); Grace et al. (2016)]. Similarly, it was recently argued that broad-scale analyses of B-EF relationships — such as regional or continental studies using observational biological data and interpolated environmental data — cannot sufficiently characterise the underlying mechanisms

(Pasari et al., 2013). These are worthy criticisms, given the innate complexity of multi-scale ecological interactions. Nevertheless, if the complementarity principle is generally applicable in nature, then it should hold across scales. Furthermore, broad-scale ecological analyses must necessarily quantify variables at much coarser spatial resolutions than manipulated experiments (for example 1km2 vs 1m2), which may alter the observed statistical relationships (Openshaw, 1983).

This is a somewhat mundane but important point, not only for this thesis, but for any broad-scale empirical analyses of the B-EF concept.

It was recently proposed that ecologists should shift their focus from the α-dimension of biodiversity (biological richness) to the β-dimension (biological turnover), in order to find broad- scale, cause and effect links between ecosystem functions and biodiversity in natural environments

(Pasari et al., 2013). This idea has intuitive appeal. It seems plausible that the spatial and temporal

Chapter 1 Introduction turnover of plants between locations within a region, and hence the turnover in their characteristics, would affect the magnitude and variability of ecosystem functions such as primary productivity, both at individual locations and across entire regions. This is because floristic turnover — in geographic and environmental space, and also through time — indirectly approximates spatio- temporal changes in the biophysical characteristics of the flora, which could influence spatio- temporal fluctuations in the photosynthetic flux of carbon (Figure 1.1).

Figure 1.1. Classic plot-scale B-EF experiments (a) suggest that ecosystem functions (e.g. productivity, biomass production and nutrient cycling) increase monotonically with biological richness [i.e. α-diversity, see Cardinale et al. (2012)]. These controlled experiments remove the biological interactions between locations that would occur in natural environments by fencing off plots [b, reproduced from Mokany et al. (2013b)], thus negating the potential effect of biological turnover on ecosystem functions (Regional EF). Compositional differences between interacting communities (c) enable new species with different physiological responses (e.g. fire response functional traits such as seeding or sprouting in the Australian context) to disperse between communities as environmental conditions change. This temporal turnover in community composition is likely to be strongly influenced by the initial spatial β-diversity of the metacommunity in which each community resides. Thus empirical analyses across broad spatio-temporal and biological scales could account for the important within (blue arrows) and between (red arrows) biological turnover between communities, which may influence ecosystem functions (EF) at both local and regional scales.

The concept of ecological specialisation provides the theoretical link between ecosystem functions and biological turnover. Regions exhibiting high biological turnover between locations, for example tropical rainforests, are characterised by greater biological heterogeneity than regions exhibiting low biological turnover, such as boreal forests. This heterogeneity means that many

Chapter 1 Introduction plants occur in narrow portions of geographic and environmental space. Indeed some plants may be known to occur only in small parts of a particular region, making them endemic to that region.

Endemic taxa can become highly specialised — physiologically, and thus ecologically — to current environmental conditions though adaptive evolutionary processes. Consequently, these narrowly distributed taxa may perform processes such as photosynthesis more optimally at the centre of their environmental distribution than taxa which are distributed across the whole landscape. Put another way, if every kind of plant occurred everywhere across a tropical rainforest, there would be no biological turnover in environmental space. All locations within the forest would then be populated by plants with generalised physiological attributes, and the landscape would be rendered biologically homogenous. Throughout this thesis I am asking the general question: would we expect the productivity at each location within a landscape to be higher or lower if the landscape was biologically heterogeneous, rather than homogenous?

In the conceptual framework outlined above, it is the degree of ecological specialisation — both at individual locations, and across regions — that can directly affect ecosystem functions, rather than the biological turnover itself (which is simply an indirect proxy of specialisation and the biogeographic processes that have formed it). For this research, I define the ecological specialisation of an individual plant species as its realised environmental niche (Austin et al., 1990)

— for example the range of rainfall and temperature conditions across all geographic locations where a tropical tree species has been recorded. I further define ecological specialisation with regard to an individual location, or site, as the median of the realised environmental niches for all species present at that site. Hence the crucial information for calculating measures of ecological specialisation is location data, derived from ecological surveys recording what plants occur where, which can be linked to ecosystem functions and environmental conditions. Recent theoretical, experimental and empirical analyses have demonstrated links between ecosystem functions and measures of biological turnover as proxies of ecological specialisation (Pasari et al., 2013; Wang &

Chapter 1 Introduction

Loreau, 2014; Van Der Plas et al., 2016). However, these relationships have not as yet been demonstrated across broad spatial, biological and environmental scales in real ecosystems.

Addressing the interrelationships between ecosystem functions, environmental conditions and species’ environmental distributions outlined above, this research tests three components of the relationship between gross primary productivity (GPP) and ecological specialisation:

1) Is GPP related to floristic β-diversity in geographic space across continental environmental gradients?

2) Does the ecological specialisation of individual species vary systematically across continental environmental gradients?

3) Is the GPP of individual sites related to the ecological specialisation of the species which occur at those sites?

There are two key analytical avenues by which these questions could be tested. Firstly, by empirical means, using measurements for GPP, environmental conditions and ecological specialisation; and secondly, by simulating these variables and their interrelationships. Here I argue that B-EF relationships can only be useful for broad-scale land management — a key justification in the literature — if links between ecosystem functions and biodiversity can be demonstrated in nature. Thus I focus on testing these questions empirically, across the broadest possible spatial, temporal and biological scales. The following Chapter (Chapter 2) outlines a broader problem space than the questions given above: that of empirically testing current and future relationships between ecosystem functions and ecological specialisation. The remainder of this Chapter describes the geographic and biological scope of the thesis, concluding with an outline of the thesis structure.

Chapter 1 Introduction

1.2 Geographic and biological scope of the thesis

The analyses in this thesis were conducted using geographic distribution records for vascular plant species which occur within the Australian continent. The geographic species records for Australian plants are among the best in the Atlas of Living Australia (www.ala.org.au), which is the Australian node of the Global Biodiversity Information Facility (GBIF, www.gbif.org). Thus

Australia represents an ideal case study for testing empirical relationships between primary productivity and ecological specialisation. Nonetheless, it is important to note that some of the taxa that occur within Australia are also distributed more widely across South East Asia and the Pacific.

Unfortunately, the spatial density of the geographic species records in other countries is lower than the records from within Australia. Given the importance of accurate ‘what and where’ data to testing the research questions, I have only analysed the environmental distributions of plants within the Australian continent.

The Australian vascular flora constitutes a highly diverse mixture of approximately 20,000 vascular plant species (Orchard, 1999). The flora are derived from two main sources: those considered to have evolved on the Gondwanan land mass (for example plants in the genera

Eucalyptus and Banksia), and those thought to have arrived in the last 33 million years (Crayn et al., 2015) via dispersal from South East Asia (for example palms in the genus Livistona).

Australia’s Gondwanan ancestry, combined with its geological isolation since separating from

Antarctica — and the subsequent evolution of its climate and soils during this period of isolation — have shaped the character of Australian plants and their geographic and environmental distributions

(Crisp & Cook, 2013).

After the Australian landmass separated from Antarctica approximately 38 million years ago, it drifted northward towards the equator, creating a drier, more seasonal climate, compared with the wetter conditions that prevailed in Australia prior to separation (White, 1994; Crayn et al.,

Chapter 1 Introduction

2015). With increasing proximity to Asia, monsoonal conditions arose in northern Australia. The prolonged seasonal drying that accompanied the monsoon fostered the adaptive radiation of sclerophyllous, fire-adapted taxa which dominate contemporary plant habitats across large areas of arid and tropical Australia [Crisp et al. (2011); Crisp and Cook (2013); Pfautsch et al. (2016),

Figure 1.2 a]. Thus most species which originated on the Australian tectonic plate have adapted to dry, fire-prone conditions over tens of millions of years. However, some species have become geographically restricted to the remaining moister environments in coastal and upland areas created by Australia’s drift into tropical latitudes.

The biogeographic forces that have shaped the Australian flora outlined above are encapsulated within the Wet Tropics region of North Queensland (Figure 1.2 c), which contains a diverse admixture of plants of Gondwanan and Asian origins (Barlow & Hyland, 1988; Metcalfe &

Ford, 2009). The Wet Tropics is characterised not only by high taxonomic α- and β-diversity, but also by a high proportion of narrowly distributed endemic species relative to the rest of Australia

(Crisp et al., 2001; Laffan & Crisp, 2003; González-Orozco et al., 2011; Crisp & Cook, 2013).

Indeed only the Southwest Australian Floristic region has higher endemism than the Wet Tropics

(Hopper & Gioia, 2004), with both regions being global biodiversity hotspots (Williams et al.,

2011). Because endemic species necessarily occupy narrow portions of geographic and environmental space, a high proportion of Wet Tropics species could be considered ecologically and physiologically specialised — making it an ideal bioregion for analysing the relationship between primary productivity and specialisation.

Chapter 1 Introduction

Figure 1.2. Broad biogeographic patterns across continental Australia form the context for the statistical relationships analysed in this thesis. Australia’s floristic biomes [a, reproduced from Crisp and Cook (2013)], and patterns of species richness in 20 km × 20 km grid cells for the Wet Tropics taxa analysed (b, c). Contemporary continental patterns of mean monthly primary productivity (d, g C m-2 month-1 2001-2012) and mean annual rainfall (e, mm, 2001-2012).

Chapter 1 Introduction

Three broad taxonomic subsets were used to test the research questions. The first question

— is GPP related to species turnover in geographic space? — was tested using geographic presence-only records for approximately 12,500 vascular plant species across continental Australia from the Australian National Heritage Assessment Tool database [ANHAT, see Williams et al.

(2012)]. The second question — does ecological specialisation vary across continental environmental gradients? — was tested using geographic presence-only records for 4292 vascular plant species known to occur in the Australian Wet Tropics Bioregion. The third question — is GPP related to ecological specialisation? — was tested using 948 tree species known to occur in the Wet

Tropics. These taxonomic subsets encompass sufficient biological, geographic and environmental space to allow plausible empirical tests of the research questions.

1.3 Structure of the thesis

This thesis addresses the three questions outlined in section 1.1 using existing data sources and a suite of analyses, which are combined, arranged and utilised in novel ways. Chapter 2

[published as Burley et al. (2016b)] is an extended review of the literature, outlining the broader problem space spanned by this thesis — that of empirically testing current and future relationships between ecosystem functions and ecological specialisation. Several reviews have summarised the findings of B-EF studies that focus on biological α-diversity, but no single article had previously outlined the conceptual and technical challenges faced when testing empirical relationships between ecosystem functions and β-diversity at broad spatio-temporal and biological scales.

Chapter 3 performs a continental-scale test of the strength and direction of relationships between gross primary productivity (GPP) and α- and β-diversity [published as Burley et al.

(2016a)]. It uses continental surfaces of GPP, rainfall and temperature, combined with empirical

Chapter 1 Introduction estimates for taxonomic floristic α- and β-diversity, to test whether GPP is influenced by biological diversity, above and beyond environmental effects.

Chapter 4 analyses how one particular measure of ecological specialisation — the realised environmental niche width of plant species — varies across continental environmental gradients for the flora of the Australian Wet tropics (in review with Journal of Biogeography). I use the environmental niche widths of the Wet Tropics flora to determine if niches are wider in particular parts of environmental space, building on the classic niche theories advanced by Gauch and

Whittaker (1972). If the realised evironmental niches of individual species can be aggregated to the community level, it will be important to control for systematic changes in niche width across environmental space. This would be necessary to isolate the effects of community niche width on community GPP from the effects of environmental conditions.

Chapter 5 tests whether the GPP of sites in the Wet Tropics is related to the degree of ecological specialisation at each site, above and beyond the effects of site environment (in preparation for Global Ecology and Biogeography). I use the median environmental niche width of all tree species occurring at each site to approximate ecological specialisation, combining the niche widths calculated in Chapter 4 with ecological plot data and continental surfaces for GPP, rainfall and temperature. The analyses are conducted for pairs of sites, holding the environmental differences between the site-pairs relatively constant while allowing the difference in community niche width between sites to vary — following the logic of plot-scale B-EF experiments.

Chapter 2 is structured as a literature review. The three data Chapters (3, 4 and 5) are each structured as journal articles to successively address the research questions. Some conceptual and methodological elements therefore overlap between chapters, with each having its own introduction and discussion. Chapter 6 summarises the key findings from Chapters 3, 4 and 5, and presents a series of recommendations for future work.

2 Macroecological relationships between ecosystem functions and biodiversity

Hugh M. Burley, Karel Mokany, Simon Ferrier, Shawn W. Laffan, Kristen J. Williams and Tom D.

Harwood

Published as: Macroecological scale effects of biodiversity on ecosystem functions under environmental change (2016) Ecology and evolution, 6, 2579-93

Contributions: Hugh Burley developed the hypotheses, ran the case study analysis and wrote the article. Karel Mokany helped develop the hypotheses, wrote code to perform part of the case study and helped write the article. Simon Ferrier helped develop the hypotheses and provided advice on the publication strategy. Shawn Laffan provided supervisory advice and wrote code to create the species distributions in the transect figure. Kristen Williams provided suggestions on both the literature review and the treatment of ecosystem functions and services in the manuscript. Tom

Harwood provided suggestions on refining the hypotheses and the examples discussed in the text.

All co-authors provided editorial suggestions for the article.

This Chapter is presented as published in Ecology and Evolution, with some minor changes to the text, and the reference list is incorporated into the main reference list for the thesis

Chapter 2 Ecosystem functions and macroecological biodiversity

Abstract

Conserving different spatial and temporal dimensions of biological diversity is considered necessary for maintaining ecosystem functions under predicted global change scenarios. Recent work has shifted the focus from spatially local (α-diversity) to macroecological scales (β- and γ- diversity), emphasising links between macroecological biodiversity and ecosystem functions (MB-

EF relationships). However, before the outcomes of MB-EF analyses can be useful to real world decisions, empirical modelling needs to be developed for natural ecosystems, incorporating a broader range of data inputs, environmental change scenarios, underlying mechanisms and predictions. We outline the key conceptual and technical challenges currently faced in developing such models, and in testing and calibrating relationships assumed in these models using data from real ecosystems. These challenges are explored in relation to two potential MB-EF mechanisms:

‘macroecological complementarity’ and ‘spatio-temporal compensation’. Several regions have been sufficiently well studied over space and time to robustly test these mechanisms by combining cutting edge spatio-temporal methods with remotely sensed data, including plant community data sets in Australia, Europe and North America. Assessing empirical MB-EF relationships at broad spatio-temporal scales will be crucial in ensuring these macroecological processes can be adequately considered in the management of biodiversity and ecosystem functions under global change.

Chapter 2 Ecosystem functions and macroecological biodiversity

2.1 Introduction

The study of ecosystem functions or processes, defined as stocks and fluxes of matter and energy derived from biological activity (Ghilarov, 2000), has received considerable attention in recent decades. While environmental conditions most directly affect these stocks and fluxes, the potential influence of biodiversity on ecosystem functions (B-EF relationships) is also considered important (see Table 2.1 for definitions of all terms used in this article). Although a substantial body of research has investigated the effects of different dimensions of biodiversity (i.e. taxonomic, functional and phylogenetic) on the magnitude and stability of key ecosystem functions (Cardinale et al., 2012), the results are often equivocal and controversial (Schwartz et al., 2000). This work has focused primarily on local-scale manipulative experiments (Gross et al., 2014) or simulations

(Loreau & de Mazancourt, 2013). The effect of the biological variation at single locations on a range of ecosystem functions is typically assessed over relatively short periods (<10 years) using small numbers of species or biological types. This is essentially an aspatial, ‘-diversity’ perspective, focusing on the effect of the number of unique biological types within local communities, most commonly floristic species richness at ecological sites. Local scale B-EF research has been important in highlighting the potential consequences for ecosystem functions following expected local biodiversity losses under global change (Cardinale et al., 2012).

Chapter 2 Ecosystem functions and macroecological biodiversity

Table 2.1. Definitions for key terms used in this article.

Term Definition

Ecosystem functions Stocks and fluxes of matter and energy derived from biological activity (Ghilarov, 2000), e.g. primary productivity, evapotranspiration, decomposition (i.e. ecosystem functions and processes are synonymous). Definitions of “ecosystem services” vary, but generally constitute those “provisioning” or “regulating” ecosystem functions valued by society [e.g. food, water quality, etc. Cardinale et al. (2012)]. Here we focus solely on ecosystem functions.

B-EF (Biodiversity-Ecosystem Function) The study of relationships between different components of biological diversity as studies explanatory variables (Cardinale et al., 2012), and ecosystem functions as response variables.

α-diversity The number of biological types — taxonomic, functional or phylogenetic — occurring at a particular location (i.e. an ecological plot).

β-diversity The turnover in biological types (i.e. change in biological composition) between locations over space and or time, both across biogeographic regions, and across entire continents. β-diversity may be quantified as a single measure for a whole region (Whittaker, 1960) or as a unique value for every pair of locations (Sørensen, 1948).

γ-diversity The total number of biological types in a region (e.g. all vascular plant species in California), being a function of both α and β-diversity (Whittaker, 1960). β- and γ- diversity thus constitute the macroecological scale of biodiversity in this article, related to, yet distinct from, diversity at the local scale. Macroecological diversity is used synonymously with β- and γ-diversity in this article.

Functional traits Any aspect of an organism’s phenotype which impacts fitness indirectly via its effects on growth, reproduction and survival (Violle et al., 2007). Functional traits can both influence the effect of organisms on ecosystem functions (functional effect traits, e.g. organism size) and their response to environmental changes [response traits, e.g. fire response, Mori et al. (2013)].

MB-EF studies The study of relationships between biodiversity at macroecological scales as explanatory variables (i.e. β- and γ-diversity across biogeographic regions and continents), and ecosystem functions as response variables.

Macroecological complementarity The hypothesis that biologically heterogeneous regions with high β-diversity, populated by physiological specialists, display greater magnitudes of ecosystem functions under current conditions than regions where generalists dominate.

Macroecological spatio-temporal The hypothesis that high β- and γ-diversity will facilitate spatio-temporal compensation biological exchanges (Loreau et al., 2003) between local communities within a region when environments fluctuate, maintaining regional stability and magnitudes of ecosystem functions under environmental change (Wang & Loreau, 2014).

Chapter 2 Ecosystem functions and macroecological biodiversity

Recent work has extended the B-EF framework by highlighting the potential importance of biodiversity at broader spatial scales in influencing ecosystem functions. Gamma-diversity (γ), the total number of biological types in a biogeographic region (Table 2.1), is a function of both α and β- diversity (Whittaker, 1960). β-diversity can be defined in many ways for different purposes

(Baselga, 2010; Tuomisto, 2010b, a; Barton et al., 2013). Nonetheless, these definitions are unified by the concept of biological dissimilarity, which is inherently spatial, being derived from the proximity and connectivity between locations over space and time. The most important aspect of β- diversity for ecosystem functions is the spatial and temporal turnover in biological composition within and between locations across a biogeographic region (Table 2.1). This is because the biogeographic processes that structure turnover, which are otherwise difficult to measure across broad spatio-temporal extents, could play important roles in influencing ecosystem functions.

Experimental analyses have shown positive effects of γ- and β-diversity on multiple regional-level ecosystem functions (Pasari et al., 2013), while simulations have linked greater β-diversity to more stable regional ecosystem functions (Wang & Loreau, 2014). However, extending this initial research to achieve a broader understanding of links between “macroecological biodiversity” (β- and γ-diversity) and ecosystem functions (MB-EF relationships) in natural ecosystems, and to thereby inform real-world management decisions, will require a new focus from ecologists.

The argument has now been made that conserving biodiversity at all threes scales (α, β and

γ) could have practical, positive implications for landscape management strategies to maintain the stability of future ecosystem processes (Cardinale et al., 2012; Pasari et al., 2013; Wang & Loreau,

2014; Isbell et al., 2015a). Similarly, it has been previously suggested that rapid species turnover under changing environmental conditions could salvage the contentious prediction that α-diversity maximises the magnitude and stability of ecosystem functions (Schwartz et al., 2000). Under rapid environmental change, managers will be increasingly required to decide which actions to implement at particular locations across large jurisdictions to achieve different objectives. Local-scale B-EF

Chapter 2 Ecosystem functions and macroecological biodiversity research provides some general guidance relevant to managing ecosystem functions within individual areas, such as promoting the maintenance of functional diversity within a site (Cardinale et al., 2012). Extending this research to macroecological scales (Pasari et al., 2013; Wang &

Loreau, 2014) may provide greater potential for developing modelling approaches to make predictions across entire regions. Such approaches could account for large changes in distributions expected for some species under climate change, along with changes in the composition of communities, and the subsequent effects of these changes on ecosystem functions. Importantly, existing MB-EF studies implicitly assume that these relationships are positive, and that they generally hold true in real ecosystems. However, they have only been analysed in controlled or simulated settings (Pasari et al., 2013; Wang & Loreau, 2014).

Before management applications can even be considered, MB-EF research needs further development in several key respects. In this paper we identify the major challenges in testing and characterising MB-EF relationships under plausible bioclimatic change scenarios, using data from multiple biological dimensions — taxonomic, functional and phylogenetic. Incorporating these data sources will provide the foundation for modelling ecosystem functions across broad spatio-temporal extents (section 2.2). Development of this capability first requires describing testable hypotheses for current and future MB-EF relationships: ‘macroecological complementarity’ and ‘macroecological spatio-temporal compensation’ (section 2.3). The importance of considering ecological context when assessing these hypotheses in natural systems is illustrated here using a simple practical example of tree communities across an altitudinal transect, where macroecological diversity is hypothesised to drive broad-scale biomass (section 2.4). Finally we outline the main avenues, potential methods and example data sources for testing both MB-EF mechanisms in real ecosystems

(section 2.5).

Chapter 2 Ecosystem functions and macroecological biodiversity

2.2 Making MB-EF relationships applicable to real ecosystems

Macroecological analyses that consider likely outcomes for ecosystem functions under a plausible range of current and future bioclimatic change scenarios would provide a more robust test of the B-EF concept as a whole. Importantly, initial work on the MB-EF concept has considered only short term or random variation in environmental conditions and has simulated, or controlled for, changes in biological composition (Loreau & de Mazancourt, 2013; Pasari et al., 2013; Wang

& Loreau, 2014). This is despite strong evidence that future bioclimatic shifts are likely to be spatio-temporally directional, auto-correlated and at least partly deterministic (Barton et al., 2015;

Oliver et al., 2015). For example, observations and predictions for Australia indicate that rainfall will continue to decrease across south-western biogeographic regions (Suppiah & Hennessy, 1998;

Gallant et al., 2013), while increasing in north-western regions. These shifts will significantly impact the distributions of both individual species and ecological communities [e.g. alterations in competitive regimes between C3 and C4 grasses, Hughes (2003), and poleward shifts in avian species, Vanderwal et al. (2013)]. Similarly, direct human modifications that are focused in particular biogeographic regions or landforms, such as land clearing, have non-random impacts on macroecological biodiversity patterns and processes (Cardinale et al., 2012; Harfoot et al., 2014).

Together these deterministic environmental shifts will shape the strength and direction of any MB-

EF relationships (positive, neutral or negative) through altered patterns of taxonomic, functional and phylogenetic diversity. Existing modelling applications provide the template for including deterministic changes into MB-EF analyses (Loreau, 2010; Mokany et al., 2012). This could be achieved by integrating spatio-temporally explicit projections for future environmental conditions, macroecological diversity patterns, ecosystem functions and management strategies.

Chapter 2 Ecosystem functions and macroecological biodiversity

Following this logic, MB-EF relationships should also be analysed across multiple biological dimensions — taxonomic, functional and phylogenetic — to robustly test the relevant macroecological processes and mechanisms assumed within these models. Local-scale research suggests that mechanistic B-EF links arise primarily through the diversity and composition of functional ‘response’ and ‘effect’ traits (i.e. particular phenotypes), that influence how biota respond to environmental conditions, and influence ecosystem functions, respectively (Mori et al.,

2013). Here it must be emphasised that trait categories are not mutually exclusive, being influenced by intra-specific variation and are best characterised as overlapping continuums (e.g. Figure 2.1).

Nonetheless, current ecosystem functions are by their very nature facilitated by the spatial distribution of particular phenotypes. Indeed current functional β-diversity should reflect contemporary phylogenetic patterns (Wang et al., 2015), and will effectively shape functional α- diversity (Figure 2.1).

Thus we may expect functional β-diversity to strongly influence how macroecological biodiversity responds to various directional global change scenarios across space and time (Corlett

& Westcott, 2013). Similarly, traits should also influence the effect of macroecological biodiversity distributions on regional-scale ecosystem functions. However, initial MB-EF analyses have considered only compositional (i.e. taxonomic) β-diversity (Pasari et al., 2013; Wang & Loreau,

2014). From a macroecological perspective, we expect that the maximum heights, leaf areas, growth rates (effect traits) and fire syndromes (response traits) of dominant tree species will turn over across broad gradients of altitude, temperature and precipitation. These shifts will generate patterns of functional — and phylogenetic — β-diversity. Again, such changes could entail positive, negative or neutral impacts on the magnitude and stability of ecosystem functions. For example, if trees with unproductive phenotypes replace each other in sections of the gradient under particular environmental change scenarios, little appreciable impact on stand biomass would be expected.

Utilising the best available information on functional and phylogenetic diversity in a spatially

Chapter 2 Ecosystem functions and macroecological biodiversity explicit manner across broad extents should therefore help improve the mechanistic basis for MB-

EF analyses.

Figure 2.1. Current β-diversity patterns (green box) are shaped by environmental conditions. Ecosystem functions (brown boxes and arrows, EF) are produced by biological-environment interactions at local and regional scales. These interactions are facilitated by the distribution of particular phenotypes (i.e. functional traits) within communities, which allow species to respond to environmental conditions and influence ecosystem functions. Functional β-diversity may then influence the magnitude and stability (x̅, σ) of ecosystem functions at local and regional scales, under both current conditions and environmental change (t).

Modelling of MB-EF relationships in real ecosystems also needs to focus more explicitly on ecosystem functions directly relevant to planning and management decisions for global environmental change and biodiversity loss. The impact of directional bioclimatic changes on the future magnitude and stability of particular ecosystem functions will vary for different areas within biogeographic regions, and across entire continents. For example, altered streamflow regimes will affect clean water yields differently at particular points across forested catchments (Milly et al.,

Chapter 2 Ecosystem functions and macroecological biodiversity

2005; Schelker et al., 2014). Similarly, stand-level biomass from different forests and woodland sites across continental environmental gradients of altitude, temperature and precipitation will not be uniformly affected under deterministic bioclimatic change scenarios (Paquette & Messier, 2011;

Morin et al., 2014; Ruiz-Benito et al., 2014). At the same time, conservation strategies often focus on promoting the persistence of all native species across large regions, through targeted habitat protection, restoration and threat minimisation (Wilson et al., 2011; Pulsford et al., 2012). The most useful information for managing biodiversity and ecosystem functions under global change scenarios will therefore be spatio-temporally explicit predictions (Lindenmayer et al., 2012) encompassing fundamental MB-EF links. Several ecosystem functions can now be either measured or modelled with reasonable accuracy across broad spatio-temporal extents, such as primary productivity, evapotranspiration and fire regime (Haverd et al., 2013; Donohue et al., 2014; Fang et al., 2016). These data provide opportunities to robustly calibrate MB-EF models with consistent measurements at relatively fine resolutions. The next section describes two key MB-EF hypotheses which offer potential to better integrate the required data inputs and outputs for testing these ideas across broad spatio-temporal extents.

2.3 A broader view of MB-EF relationships in natural ecosystems

Building on the recent work of Wang and Loreau (2014), here we outline two broad mechanisms that may underpin MB-EF relationships in natural systems: 1) ‘macroecological complementarity’ and; 2) ‘macroecological spatio-temporal compensation’. These mechanisms provide testable hypotheses for how macroecological biodiversity could interact with environmental conditions to influence ecosystem functions under both current conditions and directional environmental change.

Chapter 2 Ecosystem functions and macroecological biodiversity

Macroecological complementarity

To understand how macroecological biodiversity patterns may influence current ecosystem functions across large regions, we need to consider the different biogeographic processes responsible for shaping these patterns and their contrasting implications for ecosystem functions.

Although the causes of specific macroecological biodiversity patterns remain controversial (Kraft et al., 2011), they are clearly influenced by interactions between contemporary and past environments, the dispersal abilities of organisms and the relative strength of biogeographic barriers (Jackson &

Sax, 2010; Fernandez-Going et al., 2013). Thus it is important to consider biogeographic history when re-framing macroecological diversity as an explanatory, rather than response, variable (Table

2.1). For example, high regional β- and γ-diversity could be produced by adaptive processes such as niche specialisation (Chase & Myers, 2011) through strong competition within stable environments over long evolutionary timescales (Figure 2.2 a). Under these circumstances, we expect to observe strong relationships between species’ genotypes and phenotypes, and their ability to persist and thrive in particular environments. Adaptively assembled macroecological biodiversity patterns could then influence current ecosystem functions through environmental niche specialisation at the metacommunity scale, or ‘macroecological complementarity’.

Chapter 2 Ecosystem functions and macroecological biodiversity

Figure 2.2. Conceptual depiction of the proposed MB-EF mechanisms. Under ‘macroecological complementarity’, regions with high β-diversity resulting from the evolution of species with strong physiological specialisation and performance in particular environments (i.e. ‘deterministic β-diversity’, a) have high local ecosystem function (e.g. primary productivity) under current environmental conditions (current, dark grey lines in top ecosystem function panels). Narrow coloured niches and symbols in the central panels denote species, and black rectangular boxes denote communities. Conversely, lower β-diversity regions where more generalist species dominate (broader coloured niches and symbols, c, d) may have relatively lower current ecosystem function (dark grey lines in top ecosystem function panels). Under ‘spatio-temporal compensation’, the maintenance of ecosystem function across broad scales of space and time depends on interactions between the degree and nature of phenotypic and niche specialisation within the region, and changing environmental conditions (future, lighter dashed grey lines in top and central plots of each panel). These interactions determine the capacity of suitably adapted species to replace less well adapted species under directional environmental change (biological replacement, denoted by dashed black arrows between communities j and i). Regions where β-diversity has formed through physiological specialisation may retain higher ecosystem function because species replacement occurs (dashed grey light lines for future in top panel, a), but could experience a greater decline in ecosystem function where biological loss occurs (‘stochastic β- diversity’, b).

Chapter 2 Ecosystem functions and macroecological biodiversity

The ‘niche complementarity’ hypothesis was originally framed at the local spatial scale

(Table 2.1), predicting that greater taxonomic α-diversity approximates more diverse resource use strategies, thereby enhancing the efficiency of ecosystem functions (Loreau, 1998; Petchey, 2003;

Ruiz-Benito et al., 2014). We can extend complementarity to the spatial dimension by considering the effects of β- and γ-diversity. To hypothesise mechanistic relationships between ecosystem functions and macroecological biodiversity, we must assume that β-diversity patterns have formed through deterministic evolutionary processes (Chase & Myers, 2011). If regional β-diversity has been deterministically structured, it will reflect the degree of phenotypic (i.e. physiological) optimisation to current conditions through niche partitioning of environmental space (Devictor et al., 2010; Zuppinger-Dingley et al., 2014). In such regions, collections of narrower environmental niches (higher β-diversity, Figure 2.2 a, b) could result in more efficient performance of ecosystem functions at any point in environmental space (Baltzer et al., 2007) than regions where broader niches dominate (low β-diversity, Figure 2.2 c, d).

High macroecological biodiversity (i.e. β- and γ-diversity) can also result from more neutral processes operating within fluctuating environments over shorter evolutionary timescales (Chase &

Myers, 2011). Under these conditions, current environmental gradients may be fluctuating or weak, reducing the potential for niche specialisation. However, events such as the formation of biogeographic barriers may generate high spatial β- and γ-diversity simply through reduced dispersal (i.e. vicariance), without necessarily leading to strong phenotypic adaptation to particular niches. In reality the forces shaping macroecological biodiversity patterns — niche specialisation and vicariance — are unlikely to have operated independently, instead interacting across space and time to generate the different levels of current β-diversity.

Because the macroecological complementarity mechanism is underpinned by phenotypic optimisation, MB-EF links should arise from patterns of functional “effect” traits that influence

Chapter 2 Ecosystem functions and macroecological biodiversity ecosystem functions, such as plant height and leaf size. These patterns reflect the evolutionary processes that generate phenotypic variation among organisms (Cadotte et al., 2011; Siefert et al.,

2013), allowing them to exploit different niches and influence ecosystem functions. For example the evolution of diverse plant leaf shapes affected growth and reproduction (Nicotra et al., 2011), facilitating the colonisation and partitioning of new environments. The strength of relationships between functional β-diversity and current ecosystem functions should then help to reveal the relative influences of adaptive and neutral processes in shaping contemporary β-diversity patterns.

Assuming primarily adaptive processes, we hypothesize that more biologically heterogeneous biogeographic regions, populated by physiological specialists, will display greater magnitudes of ecosystem functions under current conditions than homogenous regions where generalists dominate

(Figure 2.2). The macroecological complementarity of species’ physiological functioning across metacommunities, as approximated by - and γ-diversity, represents the first mechanism through which contemporary macroecological biodiversity patterns could influence current ecosystem functions.

Macroecological spatio-temporal compensation

Macroecological biodiversity could also influence ecosystem functions in a more dynamic manner through the modulating effects of biological heterogeneity across space and time as environmental conditions change. The “spatial insurance” hypothesis (Loreau et al., 2003) lays the foundation for understanding dynamic MB-EF relationships, and was recently advanced by a statistical model for “ecosystem stability in space” (Wang & Loreau, 2014). This new framework partitions the stability of ecosystem functions into local, spatial and regional components akin to the partitioning of compositional diversity (Whittaker, 1960), but does not empirically quantify relationships between macroecological biodiversity and ecosystem functions. By implication, high

Chapter 2 Ecosystem functions and macroecological biodiversity regional β- and γ-diversity is thought to facilitate spatio-temporal biological exchanges (Loreau et al., 2003) between local communities within a region when environments change, promoting regional stability of ecosystem functions (Wang & Loreau, 2014). However, this prediction may only hold when dispersal is non-limiting, given its influence on the diversity and composition of local communities over time (Matthiessen & Hillebrand, 2006).

The new spatial ecosystem-stability framework proposed by Wang and Loreau (2014) provides a valuable foundation for advancing MB-EF research. However, it needs to be extended in several respects to allow plausible, empirical testing across large regions. One key extension is to consider the effects of β-diversity on the magnitude of ecosystem functions, in addition to the focus on stability adopted by Wang and Loreau (2014). Both stability and the absolute magnitude of ecosystem functions are important in natural resource management, particularly for deriving ecosystem services from particular functions (Table 2.1). For example the taxonomic, functional and phylogenetic β-diversity in different forest systems across broad environmental gradients may provide the same variability of primary production under climate change simulations, but different overall magnitudes (section 2). Similarly, a broader dynamic MB-EF mechanism should explicitly consider the effects of spatio-temporal compositional differences between interacting communities on ecosystem functions. These compositional differences facilitate the dispersal between communities of new species with different physiological responses as environmental conditions change [Figure 2.2 a, c, Mokany et al. (2013b); Mori et al. (2013)]. Aggregated regional estimates of MB-EF relationships (Pasari et al., 2013; Wang & Loreau, 2014) omit the effect of these interactions on the magnitude and stability of ecosystem functions at individual locations within regions. This is an important distinction given the inherently spatio-temporal nature of conservation planning for global change, whereby managers must allocate increasingly scarce resources between and within regions (Pressey et al., 2007; Kujala et al., 2013).

Chapter 2 Ecosystem functions and macroecological biodiversity

Another important extension of the spatial stability framework will be to consider the effects of directional spatio-temporal environmental change [Oliver et al. (2015), section 2], alongside the relatively stochastic changes already considered (Wang & Loreau, 2014). Directional bioclimatic shifts are already occurring and are likely to strengthen. Thus we suggest extending the spatial stability framework to consider a more comprehensive mechanism of macroecological spatio- temporal compensation (Figure 2.2), incorporating directional change in both environmental conditions and biodiversity distributions into predicted outcomes. Under directional change, MB-

EF links could be positive, negative or neutral depending on the ecological context, particularly the taxa, spatio-temporal extent and resolution considered. Moreover, demonstrating macroecological complementarity is not a prerequisite for testing the spatio-temporal compensation mechanism, given the strong likelihood that significant shifts in macroecological diversity patterns will affect ecosystem functions under rapid environmental change.

Dynamic MB-EF relationships will also depend on how the biological composition of a region is impacted by environmental change, including human modifications. Functional β-diversity should thus be fundamental to dynamic MB-EF links, reflecting inherent spatio-temporal trade-offs

(Mori et al., 2013; Oliver et al., 2015) in species physiological responses (e.g. fire and drought tolerance) and their effects on key ecosystem functions (e.g. gross primary productivity, evapotranspiration, fire intensity). Because gradients of human modification disproportionately impact particular environments, phenotypes and genotypes (Laliberté et al., 2010), they will interact with climatic changes to shape future MB-EF relationships. For example rapid, directional shifts in environmental conditions could have strong impacts on the biodiversity of non-contiguous rainforest metacommunities in the Australian Wet Tropics, due to inhibited dispersal resulting from land clearing (Williams et al., 2009). The finely adapted functional traits and genes endemic to regions with high β-diversity could fail to migrate or adapt by virtue of their specialisation and isolation (Feeley & Rehm, 2012). The temporal dynamics of β-diversity would then be expressed

Chapter 2 Ecosystem functions and macroecological biodiversity through loss of local functional α-diversity. High β-diversity therefore also has the potential to reduce the magnitude and stability of future ecosystem functions (Figure 2.2), if niche differentiation becomes disadvantageous under rapid environmental change.

2.4 A case study: potential MB-EF relationships for trees across an elevation gradient

To illustrate the importance of ecological context to the proposed macroecological complementarity and spatio-temporal compensation mechanisms, we consider a simple practical example of tree communities across an altitudinal transect in south eastern Australia. We first downloaded georeferenced occurrence records from the Atlas of Living Australia (ALA, http://www.ala.org.au/) for the 30 most common tree species known to occur along this transect in south eastern Australia (see Appendix 1, Table S1.1. Shrubs and herbs were not included because biomass estimates are not available for these taxa — thus biomass will be underestimated). Our coastal-inland transect was 1km × 500km, centred on latitude -36.48 to encompass a broad range in elevation, temperature and precipitation. This transect crosses six of Australia’s 85 continental

Interim Biogeographic Regions [IBRAs, see http://www.environment.gov.au/land/nrs/science/ibra,

Thackway and Cresswell (1997)], and 18 of Australia’s 23 major vegetation groups (see http://www.environment.gov.au/land/native-vegetation/national-vegetation-information-system,

Appendix 1, Figure S1.1). This study system primarily comprises Eucalypt-dominated forests and woodlands — specifically Eucalypt Open Forest and Woodland vegetation groups — where taxonomic diversity can be calculated with reasonable accuracy in the Australian context. Species records were restricted to those observations with spatial errors <2km, occurring within native vegetation and recorded since 1970 (e.g. http://biocache.ala.org.au/occurrences/search?q=Angophora%20floribunda&fq).

Chapter 2 Ecosystem functions and macroecological biodiversity

The spatial species records were then used to fit simple convex hulls for each species as a function of the current (1990) mean annual temperature (°C) and precipitation (mm) at the record locations. These values were derived from 1 km resolution interpolated climate surfaces for the

Australian continent (see http://www.emast.org.au/). Convex hulls were fitted with the alphahull R package, version 1.0 [R version 3.1.2, R Core Team (2015)]. This method was used because the analysis outcome was the presence or absence of tree species in each transect cell. Alternatively, the species' total occurrences in Australia could be used to predict their presence in the transect, which could improve predictions for those species only marginally occurring in the transect. However, using all Australian records would likely over-predict occurrences. Thus convex hulls were thus useful for making simple predictions of the current and future distribution of each species along the

500km transect using the presence-only ALA data. Dispersal limitations were incorporated into future distribution predictions by limiting the 2100 distributions to within 5km (i.e. 5 grid cells) of the current occurrence records. The current (2015) and future occurrences (2100) of each tree species in each cell are plotted in the central panel of Figure 2.3 in beige and red, respectively (see

“tree species” panel, Figure 2.3).

The predicted occurrences of all 30 tree species in each grid cell of the 500km transect were then combined to create α-diversity estimates for each cell by summing the species occurrences columns for each row. Community above-ground biomass estimates (t ha-1) were predicted by averaging the mean stand biomass for all tree species either potentially present (2015), or predicted to occur (2100), in each transect cell. See Appendix 1, Table S1.1 for biomass estimates for each species (Grierson et al., 1992; Turner et al., 1999; Keith et al., 2000; Raison et al., 2003). Biomass estimates will therefore be affected by changes in the stand structure and successional status of the particular Eucalypt forest or woodland that each cell occurs in. We then used the predicted species composition for each transect cell to create a site × site matrix of Sørensen dissimilarity values

(Sørensen, 1948) for pairs of sites along the transect using the betapart R package, [version 1.3,

Chapter 2 Ecosystem functions and macroecological biodiversity

Baselga (2010), R version 3.1.2, R Core Team (2015)]. These pairwise dissimilarity values were used to calculate the mean pairwise β-diversity within a five km radius surrounding each transect cell. This distance was chosen to match the maximum dispersal distance specified in the species distribution shifts.

Figure 2.3. A simple case study illustrating the importance of ecological context to the macroecological complementarity and spatio- temporal compensation mechanisms, using 30 key tree species and tree stand biomass (tonnes per ha-1, bottom panel). The top map in grey shows the 1km extant vegetation mask for all of Australia. The grey strip depicts the remaining vegetation across our case study system: a 500 km × 1 km altitudinal transect in south eastern Australia (red line on map, where ‘sites’ are 1km x 1km grid cells along the transect, see text). Current (2015, black lines) and future (2100, red lines) temperature (°C) and precipitation (mm) are plotted at each point along this transect. Temperature increases by 5.39 °C on average across the transect, with the greatest difference near 100km (+5.6 °C) and the smallest difference near 500km (+4.96 °C). Conversely, rainfall decreases by 327.5 mm on average across the

Chapter 2 Ecosystem functions and macroecological biodiversity transect, with the greatest difference near 350km (-839.2 mm) and the smallest difference at 500km (110.4 mm). Variations in elevation, temperature and precipitation from east to west drive changes in current environmental conditions, and subsequent variations in species distributions (light blue, green and darker blue arrows connecting environmental conditions, species distributions, potential biodiversity values and estimated biomass). The beige lines in the central panel (“tree species”) represent the current (2015) occurrences of each trees species in each transect cell according to their convex hulls, and the red lines represent the predicted future occurrences (2100). Current (2015, black lines) and future (2100, red lines) potential species richness (pt α-diversity) and potential species turnover (pt β- diversity) are also plotted at each transect point, estimated from the occurrence records for all 30 tree species.

Changes in altitude and increasing aridity from east to west across the 500km transect drove spatial turnover in current environmental conditions (indicated by light grey arrows connecting top

3 panels in Figure 2.3). Current environmental turnover then generates changes in the occurrences of tree species between transect cells, determining the potential current α-diversity of each cell and potential β-diversity surrounding each cell (Figure 2.3, pt α-diversity and pt β-diversity). The estimated stand biomass of each cell is thus a by-product of these macroecological patterns.

However, we must emphasize that the methods outline above are not intended to be realistic or necessarily accurate indicators of future B-EF relationships. Rather they provide a useful model to highlight important methodological limitations of the techniques traditionally used in the B-EF literature, while pointing to potential avenues for advancement.

Importantly, the realised climatic niche of each species was unrealistically assumed to remain constant (i.e. stationary) through time and under future climate conditions [with similar assumptions made in both plot scale B-EF experiments and simulations, e.g. see Wang and Loreau

(2014)]. Indeed species distribution models are known to be very poor at predicting future distributions for this reason, given that regression parameters will vary over space and time [i.e. display spatio-temporal non-stationarity, Brunsdon et al. (1998)]. The stationarity assumption undoubtedly shaped our biomass estimates under future climates, producing artefactual results. For example productivity is predicted to increase in a 50km block at around 400km (Figure 2.3), derived from the prediction that only one highly productive tree species will occur there in 2100.

Chapter 2 Ecosystem functions and macroecological biodiversity

Similarly, the use of published productivity estimates (Appendix 1, Table S1.1) to represent what each species will do in under future climates — and in a different species mix — assumes no adaptation to local environmental change (with the B-EF literature again making similar assumptions). Although accounting for stationarity and adaptation will be difficult across macroecological scales, this case study highlights the impact these assumptions can have on B-EF relationships (see Figure 2.4, current potential β-diversity vs. biomass change).

Figure 2.4. Plot of change in stand biomass for all 500 cells in the altitudinal transect (t ha-1 as estimated from the species occurrences) against current potential α- and β-diversity. α-diversity values are counts of species, and β-diversity values are the Sørensen dissimilarity (between 0-1). Deviance explained values (%) for generalised additive models of each plot using four knots are given in the left half of the panel (orange lines are the fitted spline regressions).

Chapter 2 Ecosystem functions and macroecological biodiversity

2.5 Potential avenues for testing MB-EF mechanisms in real ecosystems

To thoroughly account for ecological context in MB-EF mechanisms across sufficiently broad scales, empirical analyses must overcome a longstanding challenge in ecology: the effect of spatio-temporal non-stationarity on ecological data. Non-stationarity is not a new concept in geography (Tobler, 1970; Fortin & Dale, 2005; O'Sullivan & Unwin, 2010), but needs greater consideration in ecosystem function research. The inherent context dependence of MB-EF links — arising from biogeographic history and directional environmental change — means that relationships will vary spatially, temporally and with scale, forming complex emergent patterns

(e.g. Figure 2.3, Figure 2.4). For example geographically structured environmental variation should be the primary influence on ecosystem functions relating to the carbon cycle (Kanniah et al., 2013).

However, the strength of environmental influences on biological variables will vary strongly across space and time [e.g. Brunsdon et al. (1998); Miller et al. (2007); Osborne et al. (2007)]. Because environmental and biological influences on ecosystem functions are interrelated, spatio-temporal non-stationarity will affect the magnitude, accuracy, meaning and interpretation of MB-EF predictions, at both individual locations and across biogeographic regions. Although analysing nonstationary macroecological relationships represents significant challenges, these problems are not insurmountable and can be partially addressed with existing methods and data.

Chapter 2 Ecosystem functions and macroecological biodiversity

There are three main avenues for plausibly testing MB-EF mechanisms in natural ecosystems.

Firstly, macroecological complementarity could be tested in geographic space by quantifying the unique contributions of α, β and γ-diversity to ecosystem functions over and above environmental conditions at continental scales. This could be achieved by combining either empirical biodiversity data, or models of α, β and γ-diversity predicted by environmental turnover across continental scales (Ferrier et al., 2007), with fine resolution surfaces for environmental conditions and ecosystem functions (e.g. remotely sensed gross primary productivity, Table 2.2). Secondly, macroecological complementarity could be tested in environmental space at the regional scale. This could be achieved by quantifying bivariate relationships between species and community-level niche widths across key environmental gradients and functional traits approximating physiological performance (e.g. plant height, leaf area, seed size, Table 2.2). Quantifying environmental niche widths at multiple biological dimensions (i.e. phenotypic, genotypic) could better approximate the key mechanism of ecological specialisation (Devictor et al., 2010) hypothesised to underpin MB-

EF mechanisms. If niche widths are strongly related to ecological performance, multivariate relationships between ecosystem functions, environmental conditions and community-level niche widths could then be quantified using techniques such as causal networks and spatio-temporally explicit statistical modelling [e.g. Grace et al. (2014); Lamb et al. (2014); Fotheringham et al.

(2015)].

Chapter 2 Ecosystem functions and macroecological biodiversity

Table 2.2. The main avenues, potential methods and example Australian data sources for testing both MB-EF mechanisms in real ecosystems. EF denotes datasets quantifying ecosystem functions, ENV denotes environmental datasets, and BIO denotes biodiversity datasets.

Avenue Methods Example data sources and spatial extents

Test macroecological  Apply spatially interpolated models of α,  EF: Monthly continental remotely- complementarity in geographic β, and γ-diversity (Ferrier et al., 2007) to sensed gross primary productivity space test their unique contributions to EF, layers at 250m resolution [GPP, over and above the contribution of Donohue et al. (2014)]. Annual environmental conditions, at potential evapotranspiration at continental scales. 1km resolution (http://www.emast.org.au/).  ENV: Monthly continental climate surfaces at 1km resolution.  BIO: Vascular plant occurrence records at 1km resolution across a continent (http://www.ala.org.au/).

Test macroecological  Quantify relationships between  Proxies (e.g. functional traits) of complementarity in environmental niche widths (ENW) for physiological performance. environmental space individual species and ecological  EF: GPP layers downscaled to performance, e.g. niche width along soil 250m moisture gradient vs. plant growth  ENV: Monthly continental climate  Quantify multivariate relationships surfaces downscaled to 250m. Soil between EF, environment and attribute layers at 90m resolution community level ENW (cENW) using (Viscarra Rossel et al., 2014). causal networks and structural equation modelling [SEM, Lamb et al. (2014)].  BIO: Vascular plant occurrence records and community survey

plots (Mokany et al., 2014). Trait and phylogenetic databases (Kattge et al., 2011).

Test spatio-temporal  Develop and apply spatio-temporal  Same data as above, but must compensation in geographic models integrating biodiversity consider how to project complex and environmental space composition and ecosystem function relationships under environmental under rapid environmental (cENW, EF and taxonomic, functional and change scenarios (e.g. combining change scenarios phylogenetic α, β and γ-diversity). climate surfaces for 2100 with new Continue developing MB-EF simulations, biodiversity models and parameterised using macroecological simulations). datasets.  Long-term ecological monitoring  Quantify multivariate relationships sites, e.g. http://www.tern- between EF, environment, cENW, supersites.net.au/supersites/fnqr functional traits and phylogeny under various environmental change scenarios.

Chapter 2 Ecosystem functions and macroecological biodiversity

Third, the capacity for spatio-temporal compensation and its impact on ecosystem functions under rapid environmental change could be assessed by developing and applying new simulation and modelling approaches. These methods could build on existing applications (Loreau, 2010;

Mokany et al., 2012), combining mechanistic projections for both biodiversity and ecosystem functions at regional to continental scales (Table 2.2). Such information could potentially add capability to existing management tools for resource allocation (Wilson et al., 2011). Under rapid environmental change, adherence to a single management strategy (e.g. maximising α, β and γ- diversity, or connectivity) is unlikely to deliver optimal outcomes in all locations (Mokany et al.,

2013a). A number of regions around the world have been sufficiently well studied over space and time to robustly test MB-EF relationships at appropriate spatio-temporal resolutions, including plant community data sets in Australia (Metcalfe & Ford, 2009; Mokany et al., 2014) , North America

(Potter & Woodall, 2014) and Europe (Schaminée et al., 2009). Combining spatially extensive vegetation survey networks such as these with functional trait databases (Kattge et al., 2011) and phylogenies, provides an ideal test-bed for MB-EF hypotheses. This not only includes investigating contemporary MB-EF links, but also assessing the spatio-temporal compensation mechanism in both geographic and environmental space (Table 2.2). Explicitly spatio-temporal, neighbourhood- based analyses that do not assume stationarity [e.g. Sengupta and Cressie (2013); Mellin et al.

(2014)] would thus be important to accurately quantify local variations in MB-EF relationships.

In conclusion, the effect of macroecological biodiversity on ecosystem functions could be positive, negative or neutral, depending on ecological context. Thus the argument that biodiversity must be preserved at multiple spatio-temporal dimensions in order to maintain ecosystem functions under environmental change can only be tested comprehensively across broad scales. We have outlined the key conceptual and technical challenges requiring further investigation for conducting plausible, empirical tests of relationships between macroecological-scale biodiversity and ecosystem functions. The likelihood that significant shifts in macroecological patterns of

Chapter 2 Ecosystem functions and macroecological biodiversity biodiversity will affect future ecosystem functions highlights the importance of investigating the underlying mechanisms with the most appropriate methods and data. Then we will be better able to assess the potential role of biological diversity in maintaining broad-scale ecosystem functions during rapid environmental change, with possible implications for more effective management strategies.

2.6 Supporting information

Additional supporting information published online for Ecology and Evolution is included in the thesis as:

 Appendix 1: Supplementary material —for Chapter 2 — additional table and figures for

case-study of MB-EF relationships for tree species across an elevation gradient

3 Primary productivity is weakly related to alpha and beta diversity across Australia

Hugh M. Burley, Karel Mokany, Simon Ferrier, Shawn W. Laffan, Kristen J. Williams and Tom D.

Harwood

(2016) Global Ecology and biogeography, 25, 1294-1307

Contributions: Hugh Burley developed the hypotheses, wrote the R code to perform the analyses and wrote the article. Karel Mokany helped develop the hypotheses, wrote code to create the biodiversity data, helped with code development and helped write the article. Simon Ferrier helped develop the hypotheses and refine the analyses for publication. Shawn Laffan provided supervisory advice and helped with code development. Kristen Williams provided suggestions on the literature regarding productivity-diversity relationships, and Tom Harwood helped refine the hypothesis and methods. All co-authors provided editorial suggestions for the article.

This Chapter is presented as published in Global Ecology and Biogeography, with some minor changes to the text, and the reference list is incorporated into the main list for the thesis

Chapter 3 Continental productivity and macroecological biodiversity

Abstract

Aim Ecosystem functions, such as productivity, may be influenced not only by the biological diversity at each location (α-diversity), but also by the biological turnover between locations (β- diversity). We perform a continental-scale test of the strength and direction of relationships between gross primary productivity (GPP) and both α- and β-diversity.

Location Continental Australia.

Methods Species occurrence records were used to quantify the taxonomic α-diversity of vascular plants at ~11,000 1 km × 1 km grid cells across Australia, and to calculate the mean β-diversity within a 10 km radius around each cell. The magnitude and variability of monthly, MODIS-derived remotely sensed GPP (2001-2012) were summarised for continental Australia, as were rainfall and temperature over the same period. Generalised additive models were then used to test whether GPP magnitude or variability were distinctly influenced by either biodiversity measure, over and above the influence of environmental conditions.

Results Precipitation and temperature explained large proportions of deviance in the magnitude

(75.6 %) and variability (38.3 %) of GPP across the Australian continent. GPP was marginally more strongly related to species richness than it was to species turnover. However, neither diversity measure provided substantial increases in the explanatory power of GPP models, over and above that of environment-only models (always <1 %).

Main conclusions The relationship between primary productivity and taxonomic α- and β-diversity was weak for the Australian flora. Our findings question the generality of key assumptions, predictions and results in the literature regarding the strength of empirical relationships between

Chapter 3 Continental productivity and macroecological biodiversity productivity and biodiversity across multiple biological levels (α-, β- and γ-diversity) at macroecological scales.

3.1 Introduction

The study of relationships between ecosystem functions as response variables, and biological diversity as explanatory variables (B-EF research), is considered important in the context of global environmental change. B-EF research may be especially relevant for global conservation strategies, given the potential spatiotemporal congruence between key ecosystem functions, such as primary productivity, and global biodiversity distributions (Strassburg et al., 2010). Although the

B-EF perspective emerged over 30 years ago, it is intrinsically related to the substantial literature examining relationships between biodiversity measures as response variables, and productivity measures as explanatory variables (Grime, 1973; Waide et al., 1999; Mittelbach et al., 2001; Adler et al., 2011; Grace et al., 2014). When measured at fine spatial scales, biological richness is thought to be greatest at intermediate levels of productivity (Fraser et al., 2015), while broad-scale richness may increase monotonically with productivity, due to positive correlations between biological heterogeneity and productivity (Lavers & Field, 2006).

Within this historical context, most B-EF research has focused on the influence of experimentally controlled or simulated biological richness or variation at single locations (i.e. taxonomic, phenotypic and genotypic α-diversity) on ecosystem functions (Loreau, 2000; Hooper et al., 2005; Cardinale et al., 2012; Cadotte, 2013; Gross et al., 2014; Tilman et al., 2014; Fraser et al.,

2015; Isbell et al., 2015b; Poorter et al., 2015). These spatially ‘local’ approaches have typically constituted plot-scale analyses of the order of square metres. Thus they do not provide spatially explicit predictions at broad spatio-temporal scales (e.g. for different locations within biogeographic

Chapter 3 Continental productivity and macroecological biodiversity regions and across continents), limiting their applicability to global conservation assessment.

Furthermore, spatially local analyses ignore the broad-scale shifts in both environmental conditions and the distributions of entire biological communities which are already underway, and are likely to accelerate this century (Garcia et al., 2014). In order to usefully inform conservation planning under likely global change scenarios, B-EF analyses should therefore incorporate the effects of changes in biodiversity distributions in real systems across broad spatio-temporal extents. A key step in understanding how broad-scale B-EF relationships may affect ecosystem functions in coming decades is to quantify current relationships.

Recent work has highlighted the potential importance of biodiversity at macroecological scales as a distinct influence on ecosystem functions. Experimental analyses and theoretical simulations have predicted that the biological turnover between locations (β-diversity) and the total biological variation within a region (γ-diversity) could increase the magnitude and decrease the variability of a range of individual and collective ecosystem functions, such as primary productivity

(Pasari et al., 2013; Wang & Loreau, 2014; Wang & Loreau, 2016). Macroecological biodiversity could influence ecosystem functions (MB-EF relationships) through an extension of the

“complementarity” effect of biodiversity (Ruiz-Benito et al., 2014). Where causality is assumed to flow from biodiversity to primary productivity, complementarity is the key mechanism proposed to underlie B-EF relationships. Greater diversity at a location (i.e. more unique species, phenotypes or genotypes) is thought to drive higher overall resource use through complementary resource acquisition strategies (Ruiz-Benito et al., 2014; Poorter et al., 2015). Thus higher α-diversity is predicted to increase magnitudes of ecosystem functions such as primary productivity, up to a point.

Previous research suggests that relationships between alpha diversity and the magnitudes of ecosystem functions are generally positive curvilinear (Hooper et al., 2005; Cardinale et al., 2012;

Pasari et al., 2013; Ruiz-Benito et al., 2014), and negative curvilinear for the variability of ecosystem functions (Loreau et al., 2003). These relationships are thought to arise from increasing

Chapter 3 Continental productivity and macroecological biodiversity redundancy of functional effect traits that influence ecosystem processes (Mori et al., 2013) as α- diversity increases.

Local complementarity can be extended to a broader ‘macroecological complementarity’ mechanism by incorporating the inherently spatial β-diversity dimension (Pasari et al., 2013; Wang

& Loreau, 2014). Contemporary patterns of spatial biological turnover (β-diversity) are the net result of regional biogeographic history, shaped by evolutionary processes such as deterministic and stochastic community assembly. For example, competitive assembly, allopatric speciation and more recent biological colonisation have all influenced current β-diversity patterns (Chase & Myers,

2011; Flores-Moreno & Moles, 2013). In regions where the biota has undergone niche specialisation through competition over long evolutionary periods of low environmental variability, the fine partitioning of species across environmental space could reflect greater physiological optimisation to prevailing conditions. In these regions, the mean niche width in environmental space should be narrower and the mean level of β-diversity higher, reflecting a greater degree of ecological specialisation. Under macroecological complementarity, biologically heterogeneous

‘neighbourhoods’ with greater β-diversity, populated by physiological specialists, could display higher magnitudes and lower variabilities of ecosystem functions than biologically homogeneous neighbourhoods (Pasari et al., 2013; Wang & Loreau, 2016).

Importantly, high levels of macroecological biodiversity can also result from more stochastic evolutionary processes operating over shorter timescales (Chase & Myers, 2011). For example, plant populations may become isolated when biogeographic barriers emerge that prevent dispersal. These events could foster speciation – thus increasing β- and γ-diversity – without the phenotypes of these new species necessarily being adaptively specialised to particular ecological niches (Burley et al., 2016b). Similarly, more recent biological colonisations could boost magnitudes of particular ecosystem functions simply through the local dominance of widespread

Chapter 3 Continental productivity and macroecological biodiversity yet highly productive species, despite overall regional reductions in β-diversity (Vellend et al.,

2013; Savage & Vellend, 2015). Due to a lack of suitable data over broad extents in natural ecosystems, we are currently unable to determine whether existing patterns of higher β-diversity reflect greater physiological optimisation through competitive assembly, stochastic biogeographic forces (e.g. vicariance events), or simply result from regional environmental heterogeneity. While we cannot yet directly test a continent-wide macroecological complementarity mechanism in environmental space, we can use spatial patterns of β-diversity to approximate the effects of undifferentiated processes (optimisation, stochasticity and heterogeneity) on ecosystem functions.

These effects are otherwise difficult to measure across broad extents.

Several authors have also argued that different measures of biological diversity should influence ecosystem functions just as strongly as environmental conditions (Paquette & Messier,

2011; Maestre et al., 2012; Gaitán et al., 2014; Ruiz-Benito et al., 2014). While this assertion may find statistical support in particular ecological contexts, it relies on somewhat circular reasoning and requires further investigation across a broad range of natural ecosystems. For example, recent analyses suggest that at broad scales, geographically structured environmental variation at multiple spatio-temporal scales are the primary drivers of both ecosystem functions – for example primary productivity (Kanniah et al., 2013; Poorter et al., 2015) – and biodiversity distributions. These unavoidable interrelationships between variables are particularly relevant in biomes where environmental conditions associated with greater magnitudes of ecosystem functions, such as productivity, also strongly affect biodiversity patterns (e.g. tropical rainforests). Therefore to comprehensively analyse the generality of real-world MB-EF relationships, we need to control for relationships between ecosystem functions and environmental conditions across the broadest possible spatio-temporal extents.

Chapter 3 Continental productivity and macroecological biodiversity

Here we assess the strength and direction of relationships between a key ecosystem function, gross primary productivity (GPP), and local α- and regional β-diversity of plant communities across continental Australia. We test whether the magnitude and variability of GPP are significantly related to the α- and β-diversity of vascular plants, over and above the influence of environmental conditions. Given the existing evidence for local complementarity and the proposed macroecological complementarity mechanism, we make two continental-scale predictions for GPP

~ diversity relationships. Firstly, that the relationships between GPP magnitude and both α- and β- diversity should be positive curvilinear. Secondly, that the relationships between GPP variability and α- and β- diversity should be negative curvilinear. If the predicted saturating complementarity relationship is indeed driven by increasing functional redundancy as α-diversity increases, we would expect this relationship to also hold for β-diversity, as a proxy of ecological specialisation.

We assess the strength and nature of these relationships for each diversity measure at the continental scale.

3.2 Methods

Overview

To determine the strength of relationships between GPP and α- and β-diversity at relatively fine spatial resolution across continental Australia, we combined observational species records with continuous spatial layers for GPP and environmental conditions. Firstly, we derived the mean

(magnitude, GPP mean) and coefficient of variation (variability, GPP CV) of GPP, rainfall and temperature using existing monthly surfaces from January 2001 to December 2012 at 1 km × 1 km spatial resolution. We then extracted GPP and environmental values for approximately 27,000 1 km

× 1 km grid cells where the species records provided reasonable α-diversity estimates. We then

Chapter 3 Continental productivity and macroecological biodiversity further restricted the data to approximately 11,000 cells with the highest-quality species records, and calculated empirical β-diversity values. From these refined cells, we subsampled approximately

1000 cells within 20 km 3 20 km blocks, repeating the subsampling 1000 times as described below.

Generalized additive models were then fitted to all 1000 subsamples to test relationships between

GPP and biodiversity.

3.2.1 Continental datasets

Gross primary productivity

Our response variable, GPP, was defined as the gross photosynthetic uptake of atmospheric carbon. We employed monthly, 250 m × 250 m spatial resolution estimates of GPP from January

2001 to December 2012 [GPP, gC m-2 month-1, Donohue et al. (2014)]. The surfaces created by

Donohue et al. (2014) were used to quantify total GPP, the estimated monthly photosynthetic flux from three plant functional types: trees, C3 and C4 grasses. These surfaces employ a standard light use efficiency model (Monteith, 1972) modified to account for the effects of diffuse radiation

(Donohue et al., 2014), and incorporate MODIS derived fPAR as an input (the fraction of photosynthetically active radiation absorbed by plants). We up-scaled these surfaces from 250 m ×

250 m spatial resolution to 1 km × 1 km, calculating the magnitude of GPP as the mean of each 1 km × 1 km grid cell across all 144 months (Figure 3.1 a, GPP mean). GPP variability was calculated as the coefficient of variation for 1 km × 1 km grid cell across all months (Figure 3.1 b, GPP CV).

Remotely sensed measures such as GPP CV will be affected by climatic extremes and various sources of error between locations (De Keersmaecker et al., 2014). However, the diffuse light model GPP surfaces provide considerable accuracy improvements for Australia compared with

Chapter 3 Continental productivity and macroecological biodiversity standard global MODIS products (Donohue et al., 2014). Similarly, the coefficient of variation provides a consistent measure of variability in productivity relevant to existing B-EF hypotheses.

Figure 3.1. Current Primary productivity and environmental conditions for continental Australia at 1 km × 1 km spatial resolution. Maps for the mean and coefficient of variation (CV) surfaces of gross primary productivity [GPP; gC m-2 month-1 in brown, a, b, Donohue et al. (2014)], rainfall (in blue, c, d) and temperature (in orange, e, f) are shown in the left panels. Plots of each variable across the time series (January 2001-December 2012) are shown in the right panels for two points in eastern Australia (black and grey triangles in the bottom right panel, 152°11'4.356" E 29°15'26.352" S, and 149°14'33.193" E 30°21'23.725" S, respectively). Inset panels on the right show a detail of the CV surface for each variable across the time series (GPP CV in brown, rainfall CV in blue and temperature CV in orange, which are indicated on the main maps by red and black inset panels).

Chapter 3 Continental productivity and macroecological biodiversity

Environmental conditions

Spatio-temporal fluctuations in GPP are primarily driven by climatic conditions at continental scales, and particularly by rainfall in the Australian context. Key environmental conditions were defined as the total monthly rainfall (mm month-1) and the mean monthly maximum temperature (°C) from January 2001 to December 2012. We used spatially interpolated climate surfaces from the Australian Water Availability Project based on meteorological stations

(http://www.bom.gov.au/jsp/awap) to quantify rainfall and temperature (Jones et al., 2009). The temperature and rainfall surfaces were then resampled from 5 km × 5 km spatial resolution to 1 km

× 1 km, using nearest neighbour resampling. This algorithm determines the value of each cell in the

1km output raster using the value of the nearest cell in the 5km input raster (ESRI, 2015). Although nearest neighbour resampling is associated with error (Ebdon, 1985), it is appropriate for resampling climate surfaces because the original values are retained rather than smoothed. We again calculated the mean (magnitude) and coefficient of variation (variability) for rainfall and temperature across all 144 months (Figure 3.1 c, d, e and f).

Alpha and beta diversity of vascular plants across Australia

The original biological data consisted of approximately 2.2 million species occurrence records from the Australian National Heritage Assessment Tool database [ANHAT, Slatyer et al.

(2007)], comprising ~12,500 vascular plant species and 83 families (Williams et al., 2012). These records were aggregated to 1 km × 1 km grid cells, creating species count observations for approximately 400,000 locations. Unfortunately, the paucity of occurrence records for many grid cells resulted in erroneous estimates of α- and β-diversity, limiting the potential for testing empirical relationships between productivity and biodiversity in a spatially complete manner. Thus

Chapter 3 Continental productivity and macroecological biodiversity we limited our analyses to those locations where plant species have been sampled with sufficient quality and density to allow reasonable biodiversity estimates.

We took the number of species recorded in each grid cell as a proxy for sampling intensity.

For each of the original cells (Figure 3.2 a), we determined the maximum number of species recorded in any cell within a 20 km radius. We then took the mean of this grid surface within a 50- km radius surrounding each cell. The resulting values, Smax, of the 50-km smoothed surface, approximated the greatest number of species we expected to be recorded in each of the original cells. We then omitted those cells with species counts that were <25% of Smax (which we call

Smin), as well as any remaining cells with fewer than 10 species recorded (Figure 3.2). Of the remaining approximately 27,000 cells, we further omitted those cells containing fewer than 10 retained locations within a 10-km radius, which equated to cells with fewer than 45 site pairs. This additional thresholding reduced the risk of overestimating β-diversity through insufficient site pairs, and left approximately 11,000 grid cells with high quality compositional observations for further analysis (Figure 3.2).

Given that the diverse Australian vascular flora will be under sampled, particularly within the arid and semi-arid biomes, it is unlikely that any 1 km × 1 km grid cell truly contains fewer than

10 species. Thus using grid cells with fewer than 10 records would systematically underestimate α- diversity and overestimate β-diversity, undermining our analyses. Consequently, the data quality thresholds should be considered lenient for continental Australia. The α-diversity of each retained grid cell was then assumed to be the number of species recorded. To quantify the β-diversity in a 10 km radius surrounding each retained grid cell, we first calculated the Simpson’s dissimilarity (βsim) for each pair of grid cells in that area (equation 1):

Chapter 3 Continental productivity and macroecological biodiversity

min(푏,푐) β = (1) 푠푖푚 푖,푗 푎 + min (푏,푐)

Where a is the number of species present in both sites i and j, b is the number of species present in the first site but not in the second, and c is the number of species present in the second site but not in the first. The Simpson dissimilarity was used to quantify differences in composition whilst minimising the effect of differences in species richness (Baselga, 2010). We then determined the weighted mean Simpson dissimilarity in the 10 km radius around each focal grid cell by weighting the contribution of the dissimilarity of each site pair to the average [Wij = (wi + wj) /2] based on the distance of each site (di) in the pair from the focal cell [wi = 1 – (di / dmax); where dmax

= 10 km, Figure 3.2 f]. For each of the retained grid cells (Figure 3.3 a, b), we used the values for

GPP and environmental conditions from each spatial grid for subsequent analysis (Figure 3.3 c).

Chapter 3 Continental productivity and macroecological biodiversity

Figure 3.2. Flow diagram depicting the quantification of α- and β-diversity values. The original species occurrences (a) were aggregated into approximately 400,000 1 km × 1 km grid cells (b). We then calculated the maximum α-diversity in any cell within a 20 km radius (c). This 20km surface was then averaged within a 50 km radius of each cell, creating Smax: the mean of the average maximum α-diversity within a 20 km radius (d). Cells were excluded if α-diversity was either <10 or <25% of the Smax, and also if fewer than 10 locations were retained within a 10 km radius. For the remaining grid cells (e), β-diversity was quantified using a linear weighting scheme to calculate the pairwise Simpson dissimilarity within a 10 km radius (i.e. dmax = 10km). The focal cell (in red) received a weight of 1, and the peripheral cells a weight of 0. The average β-diversity for the focal cell (f) was then calculated as the mean weighted average of all pairwise comparisons [i.e. Wij = (Wi + Wj )/2)].

Chapter 3 Continental productivity and macroecological biodiversity

Figure 3.3. (a) Mean species richness of the approximately 11,000 candidate 1 km × 1 km sites where the number of site pairs within 10 km was >45, averaged within 20 km × 20 km blocks (‘α’ in panels, dark blue indicates higher α). b). Mean Simpson dissimilarity within a 10 km radius of all sites (‘β’ in panels, dark red indicates higher β). A zoom of the Darwin region is shown for both diversity measures (the colour scales for α and β match between the maps and the scatterplots). c). Scatterplots of the response (GPP mean and CV) and predictor variables for all candidate sites (top half of scatterplot, note the greater range of axes values) and a subsample of 1107 sites (bottom half). GPP is the mean monthly photosynthetic flux of carbon (gC m-2 month-1, January 2001-December 2012/12), GPP CV is the standard deviation of GPP divided by the mean. ‘Rain’ is the average total monthly precipitation (mm month-1), temp is the mean monthly maximum temperature (°C).

Chapter 3 Continental productivity and macroecological biodiversity

3.2.2 Statistical Analyses

Continental subsampling

Within the set of approximately 11,000 retained grid cells where α- and β-diversity were quantified (Figure 3.3 a, b), there remained substantial spatial bias and marked variation in the reliability of the biodiversity estimates. We therefore undertook our analyses on 1000 random subsamples of these retained grid cells. For each subsample, we selected a single grid cell within the approximately 5000 20 km × 20 km spatial blocks across Australia that contained retained grid cells, based on the joint probability that both α (Pα) and β-diversity (Pβ) were reliably estimated.

The probability that the species richness of a grid cell i (Si) was estimated reliably [Pα, Mokany et al. (2012)] was determined as:

1 푃 = 1 − (2) 훼 1−푏푐.푆푟푒푙

Where b = 2, c = 6.63, and the relative species richness (Srel) calculated as:

푆 +푆 푆 − 푚푖푛 푚푎푥 푖 2 푆푟푒푙 = 푆 +푆 (3) 푆 − 푚푖푛 푚푎푥 푚푎푥 2

The probability that the β-diversity for a focal grid cell was reliably estimated (Pβ) was based on the number of site pairs within a 10 km radius (np), using the complement of a negative exponential function (with z = 0.2):

−푛푝푧 푃훽 = 1 − 푒 (4)

Chapter 3 Continental productivity and macroecological biodiversity

Of the spatial blocks that contained retained grid cells, approximately 1000 had data of the highest quality according to the joint probability of Pα and Pβ. Each random subsample therefore contained approximately 1000 points, and we repeated the subsampling 1000 times. These final

1000 continental subsamples of GPP, environmental conditions and biodiversity measures were used for the statistical analyses.

Statistical modelling of the GPP response variable

To test the strength of relationships between GPP variables and biodiversity measures, over and above the influence of environmental conditions, we compared results from models with and without these two biodiversity measures. Models were run for both GPP magnitude and variability, fitting the models to each of the 1000 random sub-samples and averaging the results. Preliminary analyses and correlation plots illustrated that GPP was non-linearly related to rainfall and temperature (Figure 3.4). Moreover, the true functional nature of macroecological relationships will vary considerably across Australia, due to strong spatio-temporal dependencies. Thus we adopted a flexible approach using generalised additive models (GAMs). Three model comparisons were run:

(1) Environment (rainfall + temperature) + α vs. environment

(2) Environment + β vs. environment

(3) Environment + α + β (i.e. full model) vs. environment + α

Chapter 3 Continental productivity and macroecological biodiversity

GAMs were employed with 5 knots specified for the spline of each covariate to balance over-parameterisation with model fit, using the Tweedie distribution with a logistic link function

[mgcv R package, version 1.8.4, R Core Team (2015)]. Preliminary analyses demonstrated that increasing the knots caused overfitting, without substantially changing the results. Similarly, the

Tweedie distribution is commonly used to model positive, continuous responses such as biomass

(Dunstan et al., 2013), and provided a better fit to both GPP responses than the gamma distribution.

Including a spline term for the spatial location of each grid cell also did not change the results (see

Appendix 2, Table S2.3, Table S2.4), thus we omitted them from the main models.

For each comparison between the nested models, we tested the contributions of α- and β- diversity using four diagnostic measures. The additional percentage of deviance explained (ΔDE) was used to indicate the explanatory power induced by adding each diversity measure (i.e. the strength of net relationships). The change in the Bayesian information criterion values between each model minus the environmental model (ΔBIC) was used to quantify differences in model fit [BIC values being less prone to overfitting with large datasets than the Akaike information criterion, Mac

Nally (2000)]. Initial testing indicated that relationships between GPP variables and biodiversity values were weak. Rather than explicitly testing for the curvilinear relationships predicted by the local and macroecological complementarity hypotheses (see introduction), we visually assessed predicted GPP values for each model when holding rainfall and temperature at their mean. The overall directions of relationships were quantified by comparing the GPP values predicted by the

GAMs at the highest and lowest values of α- and β-diversity while holding environmental variables at their mean values (ΔGPP). Approximate tests of whether the reduction in total deviance induced by adding each biodiversity measure to the environmental models was significant (P) were performed using the anova.gam function in the mgcv R package. Importantly, anova.gam p-values for the smoothed components of nested GAMs are often too low (Wood, 2013), particularly for large datasets. Thus we considered all four diagnostic measures in concert.

Chapter 3 Continental productivity and macroecological biodiversity

3.3 Results

Continental analysis of the relationship between GPP and α-diversity

Our relatively simple additive models of GPP mean that included α-diversity (Table 3.1, rainfall + temperature + α-diversity) explained a substantial amount of deviance (76.7%), with most of the deviance explained by rainfall (64.02%). This is unsurprising in the Australian context, given that water availability, rather than temperature, is the key limiting factor for plant growth and reproduction across the majority of the continent. The average explanatory power for the environment + α-diversity models of continental GPP CV was considerably weaker (Table 3.2,

39.34%). Although both continental GPP mean and GPP CV were more strongly related to α- diversity than to β-diversity, the results clearly demonstrate that α-diversity alone explained only a small amount of deviance in primary productivity (e.g. Table 3.1, 2.48 % for GPP mean, and Table

3.2, 5.87 % for GPP CV). While the environmental models (rainfall + temperature) had higher BIC values than the full models for GPP mean (Table 3.1, rainfall + temperature + α-diversity + β- diversity ΔBIC), α-diversity did not contribute meaningful information to environmental models of

GPP CV (Table 2, ΔBIC).

Chapter 3 Continental productivity and macroecological biodiversity

Table 3.1. Total percentage (%) of deviance explained (DE) and Bayesian information criterion values (BIC, ΔBIC) for the analyses of mean gross primary productivity (GPP) as a function of environmental conditions and biodiversity measures. DE and BIC values are shown for each predictor variable by itself and the four model combinations used in the continental analyses. ΔBIC indicates the average difference between the BIC of each model and that of the full model (Rain + Temp + α + β). Rain denotes total monthly rainfall (mm month-1 2001-2012), and Temp denotes mean monthly maximum temperature (°C).

Predictors GPP Rain Temp α-diversity β-diversity DE (%) BIC ΔBIC 76.701 11033.51 0 76.45 11034.65 1.14 76.05 11036.65 3.14 75.70 11045.23 11.71 64.02 11464.52 431.01 15.61 12452.07 1418.55 2.48 12617.49 1583.98 0.4497 12639.65 1606.13

The strength of relationships between GPP and α-diversity over and above that explained by environmental conditions was slightly greater than for β-diversity (Table 3.1, Table 3.2). However, the additional amount of deviance explained for both GPP mean and GPP CV by adding α-diversity was small relative to the deviance explained by environmental conditions (Table 3.3, ΔDE = 0.76% for GPP mean and 0.95% for GPP CV). The differences in BIC values between environmental models of GPP with and without α-diversity indicate that α was associated with greater improvements in model fit for GPP magnitude than for GPP variability (Table 3.3, ΔBIC = -10.76 for GPP mean and -0.19 for GPP CV). The modelled relationship between GPP mean and α- diversity was negligible once relationships between productivity and climate were accounted for

(Figure 3.4), yet stronger than for β-diversity (Table 3.3, ΔGPP). The spline functions for GPP magnitude suggest an increase, then plateau, over low to moderate richness, but no clear overall trend for GPP CV vs. α-diversity (Figure 3.4). Nonetheless, the reduction in total deviance of GPP mean and CV induced by including α-diversity was only moderately significant and should be

Chapter 3 Continental productivity and macroecological biodiversity interpreted with caution, given the large sample sizes and low additional explanatory power (Table

3.3, P <0.002 for both responses).

Table 3.2. Total percentage (%) of deviance explained (DE) and Bayesian information criterion values (BIC) for the analyses of the coefficient of variation in GPP (GPP CV) as a function of environmental conditions and biodiversity measures. DE and BIC values are shown for each predictor variable by itself and the four model combinations used in the continental analyses. ΔBIC indicates the average difference between the BIC of each model and that of the full model (Rain + Temp + α + β). CV Rain denotes the coefficient of variation in total monthly rainfall, and CV Temp denotes the coefficient of variation in mean monthly maximum temperature (2001-2012).

Predictors GPP CV CV rain CV temp α-diversity β-diversity DE (%) BIC ΔBIC 39.34 -1501.55 -12.54 38.39 -1501.26 -12.26 38.71 -1489.31 -0.30 39.69 -1489.01 0 25.98 -1321.45 167.56 8.56 -1080.40 408.61 5.87 -1052.97 436.04 1.02 -996.70 492.31

Continental analysis of the relationship between GPP and β-diversity

The weakest individual predictor of continental GPP mean or GPP CV was β-diversity

(Table 3.1, Table 3.2). This was confirmed by the comparative models, where adding β-diversity to the full models provided no substantial increase in the additional percentage of deviance explained for either GPP response (i.e. Table 3.3, environment + α + β vs. environment + α ΔDE = 0.002% for GPP mean and 0.003% for GPP CV). Similarly, adding β-diversity to the full models of both

GPP responses was associated with either only minor BIC reductions or, in some cases, increases

(Table 3.3, ΔBIC = -1.1 for GPP mean and ΔBIC = 12.69 for GPP CV). This was highlighted by the considerable variation in β-diversity values across the entire range of both GPP responses (Figure

3.4). The spline functions show that once environmental conditions were accounted for, the net

Chapter 3 Continental productivity and macroecological biodiversity relationship between both GPP measures and β-diversity was flat, rather than positive or negative curvilinear (see introduction). Strengthening this interpretation, adding β-diversity to the full models was mildly significant for GPP mean (Table 3.3, environment + α + β vs. environment + α

P = 0.012) and non-significant for GPP CV (P = 0.099). In concert, these results demonstrated that the relationship between Australian continental primary productivity and taxonomic floristic β- diversity was weak.

Table 3.3. Results for the generalised additive models of gross primary productivity (GPP), averaged across 1000 random subsamples. For each model comparison (i.e. base model vs. base model + added variable), four model diagnostics are given (env denotes total monthly rainfall + mean monthly maximum temperature, 2001-2012). ΔDE represents the average additional percentage of deviance explained by adding that variable. ΔBIC indicates the average BIC values for each model (i.e. base model + added variable) minus the environmental model. ΔGPP denotes the average difference in predicted gross primary productivity values (GPP) at the highest and lowest values of the added variable, while holding all other variables at their mean. For GPP mean, ΔGPP units are in gC m-2 month-1. For GPP CV models, ΔGPP is the coefficient of variation.

Response Base model Added variable ΔDE (%) ΔBIC ΔGPP P GPP mean env α 0.756 -10.76 -57.10 <0.001 env β 0.346 -8.61 -37.58 0.001 env+α β 0.002 -1.11 -31.59 0.012 GPP CV env α 0.945 -0.19 -0.05 0.002 env β 0.312 12.07 0.03 0.118 env+α β 0.003 12.69 0.00 0.099

Chapter 3 Continental productivity and macroecological biodiversity

Figure 3.4. Plots of one of the continental generalised additive models, showing raw observations for one random subsample (blue points). Predicted gross primary productivity values (GPP and GPP CV) for models using only that predictor [e.g. GPP = f (rain) for top row of panel] are plotted in green. Predicted GPP values for models holding all other variables at their mean [e.g. GPP = f (rain + mean temp + mean α + mean β) for top row of panel] are also plotted in orange. GPP is the mean monthly photosynthetic flux of carbon (gC m-2 month-1, January 2001-December 2012, see section 3.2.1). GPP CV is the ratio of the standard deviation in GPP to the mean.

Chapter 3 Continental productivity and macroecological biodiversity

3.4 Discussion

Weak relationships between continental GPP and taxonomic floristic α- and β-

diversity

The magnitude and variability of primary productivity across Australia were weakly related to the taxonomic α- and β-diversity of vascular plants, even before accounting for environmental conditions. Although both GPP variables were more strongly related to α-diversity than β-diversity, the difference was minor. Specifically, neither diversity measure was associated with substantial increases in deviance explained, improvements in model fit or consistent direction of relationships, after accounting for environmental conditions (Table 3.3, Figure 3.4). This result is particularly surprising for α-diversity, given the predictions of strong relationships for the complementarity mechanism in the literature (Cardinale et al., 2012). Recent studies have used theoretical analyses

(Wang & Loreau, 2014) and simulations (Pasari et al., 2013) to argue that even if the effect of α- diversity on ecosystem functions is weak (Vellend et al., 2013), the loss of β- and γ-diversity could still increase the variability of productivity in regional ecosystems. However, our results do not support this reasoning: the variability of primary productivity was not meaningfully related to spatial variation in taxonomic β-diversity.

Assuming biological heterogeneity has formed through adaptive evolutionary processes, the macroecological complementarity hypothesis states that greater physiological specialisation across environmental space should increase the magnitude and reduce the variability of primary productivity (Burley et al., 2016b). Contrary to this expectation, and counter to those of Pasari et al.

(2013), we demonstrated that incorporating larger, more realistic gradients of community structure and composition did not yield increased, distinct contributions of α- and β-diversity to primary productivity. This contrasts with analyses of taxonomic and functional α-diversity in northern hemisphere forest ecosystems (Paquette & Messier, 2011; Morin et al., 2014; Ruiz-Benito et al.,

Chapter 3 Continental productivity and macroecological biodiversity

2014), drylands (Maestre et al., 2012; Gaitán et al., 2014) and tropical forests (Poorter et al., 2015), which found significant, net relationships between productivity and biological richness. However, it is noteworthy that Pasari et al. (2013) reported only small contributions for the β-diversity of experimental landscapes to above ground productivity, when considered as an individual ecosystem function.

We found no support for any particular functional form of the relationship between both productivity variables and either diversity measure, contradicting both the local and macroecological complementarity hypotheses (Figure 3.4). The hypothesised relationships between the variability of productivity and both biological richness and heterogeneity are thought to arise from asynchronous responses of species with different functional traits (Morin et al., 2014).

Unfortunately, asynchronous species responses cannot feasibly be quantified using continental, presence-only taxonomic datasets. It is also possible that our continental communities (i.e. grid cells) are relatively saturated with species, whereby they are already at the diversity level where productivity asymptotes with richness. Indeed this saturation has been observed by macroecological analyses in simpler ecosystems (Ruiz-Benito et al., 2014). Furthermore, the macroecological complementarity mechanism may only manifest in locations with low levels of ecological specialisation. In this case, most of our grid cells would have high enough levels of taxonomic turnover, such that changes in β-diversity would not appreciably affect productivity.

We have only considered relationships between one ecosystem function (GPP), and measures of taxonomic α- and β-diversity. It has been suggested that positive effects for β-diversity will emerge only when multiple ecosystem functions are considered simultaneously, and over longer time-frames (Jones et al., 2009; Pasari et al., 2013; Lefcheck et al., 2015). Here it could be argued that multi-functionality indices sacrifice interpretability for the sake of statistical significance (Bradford et al., 2014). The effects of β-diversity could indeed be stronger when

Chapter 3 Continental productivity and macroecological biodiversity considering the causal networks between multiple environmental conditions and individual ecosystem functions which have more mechanistic relationships with biological turnover (Grace et al., 2014). Unfortunately, many of these ecosystem functions are not as easily measured across broad spatio-temporal extents as remotely sensed productivity (e.g. nutrient cycling, fire regime, resistance to biological invasion). Furthermore, our use of simple geographic measures of taxonomic β-diversity does not directly test the macroecological complementarity mechanism

(Burley et al., 2016b). Quantifying the mean environmental niche width of all species, phenotypes and genotypes at a location could better isolate the effects of ecological specialisation on ecosystem functions from other factors that drive geographic biological turnover (e.g. vicariance processes and spatial environmental configuration).

Context-dependence of B-EF relationships

The apparent contradiction between our findings and much previous research highlights that

B-EF relationships are context dependent and difficult to generalise across ecosystems, spatio- temporal and biological scales (Bengtsson, 1998; Schwartz et al., 2000; Vellend et al., 2013). We must then consider how our methods could have affected the observed relationships between productivity and biodiversity. Our empirical approach comes with its own biases, driven by decreasing sampling intensity further from the primarily coastal Australian population centres.

Rainfall, productivity and α-diversity also decay strongly with distance from the Australian coastline. As noted by (Lavers & Field, 2006), this decline is likely driven by simultaneously increasing energy inputs, and decreasing water inputs, habitat volumes and gradient distances with increasing aridity. Choosing the best sampled sites will therefore have biased the analyses towards the wettest, most productive and most species rich areas. However, supplementary analyses that sampled more extensively in all of Australia’s biomes still supported the main results (Appendix 2,

Chapter 3 Continental productivity and macroecological biodiversity

Table S2.1). Although β-diversity could be calculated in many ways [e.g. as a site contribution,

Legendre and De Cáceres (2013), or as a site-pair comparison between productivity and turnover differences], the biological turnover between sites was most appropriate to our test. Furthermore,

GPP was weakly related to β-diversity, regardless of the scale at which it was calculated (i.e. even when calculated across larger radii, Appendix 2, Table S2.2).

The concept of scale, both spatial and biological, is central to all ecological hypotheses. We must therefore acknowledge the disparity in spatial resolution between ‘local scale’ B-EF experiments (i.e. 1 m × 1 m), and our macroecological analyses (1 km × 1 km). Our spatial resolution cannot be varied below 1 km, the scale at which the vascular plant records were captured.

Substantial net relationships between ecosystem functions and α- and β-diversity could conceivably manifest at finer spatial resolutions, where increasing biological richness and turnover may not be functionally redundant (Poorter et al., 2015). Interestingly, the strength of relationships between the productivity of forest plots and α-diversity may decrease, and even become negative, with coarser spatial resolution (Chisholm et al., 2013). This apparent scale-dependence of B-EF relationships could result from biological ‘saturation’ at macroecological scales across real ecosystems, whereby additional biological richness and heterogeneity become redundant. Overall, our substantial geographic and biological coverage suggests we would have detected stronger relationships between GPP and biodiversity if they existed. Thus we cannot infer from these datasets that continental GPP has a meaningful relationship with geographic measures of taxonomic α- and β- diversity in the Australian context.

Chapter 3 Continental productivity and macroecological biodiversity

Conclusions and future directions

The analysis of broad-scale relationships between productivity and biodiversity is not new.

However, empirical, continental-scale tests of relationships between ecosystem functions as response variables, and biological richness and heterogeneity as explanatory variables, are in their infancy. We found that adding taxonomic α- and β-diversity to simple models of the magnitude and variability of continental primary productivity contributed no meaningful explanatory power, with a striking lack of support for longstanding assumptions in the literature. It is reasonable to ask whether these results question the logic of using simple measures of biological diversity – however they are quantified – as explanatory variables for broad-scale productivity measures. Indeed our results may indicate that both measures of productivity and simple diversity indices are more appropriately placed on the y-axis, responding to the same environmental conditions (Lavers &

Field, 2006), at least across such broad spatio-temporal and biological scales. Nevertheless, more targeted measures of ecological specialisation – such as site-level environmental niche widths – may potentially have distinct relationships with certain ecosystem functions, particularly under environmental change scenarios.

3.5 Supporting information

Additional supporting information published online for Global Ecology and Biogeography is included here as:

 Appendix 2: Supplementary material for Chapter 3 — sensitivity analyses for

continental relationships between GPP and alpha and beta diversity.

Chapter 3 Continental productivity and macroecological biodiversity

Acknowledgements

We thank R. Donohue for granting access to the gross primary productivity dataset. Dr

Chris Ware helped improve earlier versions of the manuscript through the CSIRO internal review process for which we are grateful. UNSW and CSIRO provided scholarship funding through the

Australian Postgraduate Award (APA) and the Office of Chief Executive (OCE) respectively.

4 Plant niche specialisation varies along environmental gradients in the Australian Wet Tropics

Hugh M. Burley, Karel Mokany, Shawn W. Laffan, Will K. Cornwell, Robert M. Kooyman,

Andrew H. Thornhill, Kristen J. Williams, Tom D. Harwood and Simon Ferrier

This Chapter represents an article submitted the Journal of Biogeography (in review).

Contributions: Hugh Burley developed the hypotheses, wrote the R code for the analyses and wrote the article. Karel Mokany helped develop the hypotheses, wrote code to randomise the regressions of species niche width and helped write the article. Shawn Laffan wrote code to summarise species niches and randomise the regressions of species niche width, provided supervisory advice and helped with code development. Will Cornwell helped develop the hypotheses and wrote code to subset the angiosperm phylogeny for the phylogenetically corrected regressions. Rob Kooyman and Andrew Thornhill advised on biogeographic interpretations of the results. Kristen Williams provided suggestions on additional ecological factors affecting niche width for discussion, such as biotic competition and ecological opportunity. Tom Harwood advised on the niche vs. trait analyses and the niche volume analyses. Simon Ferrier helped develop the hypotheses and advised on the publication strategy. All co-authors provided editorial suggestions.

Chapter 4 Continental species niches

Abstract

Aim The range of environmental conditions that species are known to occupy — their realised environmental niche widths — are important indicators of physiological specialisation and biogeographic history. Here we test whether the realised continental environmental niches of vascular plants occurring in the Australian Wet Tropics (WT) vary across important environmental gradients, and link the results to biogeographic interpretations of Australia’s flora.

Location Continental Australia

Methods We obtained herbarium records spanning continental Australia for 4292 vascular plant species that have been recorded within the WT. For each species we calculated the realised continental niche width (95th -5th percentile) and niche median for annual rainfall (mm), maximum temperature (°C) and soil nitrogen (%) using interpolated surfaces. Public-access databases were used to quantify maximum recorded plant height (m) and estimate leaf area (cm2). We then tested whether niche width was related to niche median and traits, using spatially constrained randomisations of linear and generalised additive models for all species (n = 1771) and tree species only (n = 647). Finally, we used a structural equation model to test whether the combined elliptical niche volume of trees (volume = π/6 × rainfall niche width × temperature niche width × nitrogen niche width) was related to leaf area and height, while controlling for the effect of rainfall, temperature and nitrogen niche median.

Results For all species and tree species only, realised temperature niche width was independent of niche median (adjusted R2 = 0.085). However, realised rainfall and nitrogen niche widths were wider in wetter and more fertile environments (deviance explained = 0.387 and 0.513, respectively). Similarly, tree species with niche widths centred in wetter environments had larger

Chapter 4 Continental species niches leaves (adjusted R2 = 0.162). A moderate amount of variation in niche volume was explained by realised niche medians and traits (R2 = 0.123).

Main conclusions The WT flora is not strongly specialised across the continental temperature gradient, while displaying greater specialisation towards drier and less fertile conditions. Although these patterns likely reflect the biogeographic history of the WT flora, further studies are required to investigate the role of adaptation and competition in influencing continental environmental distributions.

4.1 Introduction

The physiological specialisation of biological taxa to environmental conditions is an important concept, and can be defined and quantified in myriad ways. Central to the current interest in physiological specialisation is the hypothesized relationship between rarity (environmental, geographic or otherwise) and extinction rates under global environmental change scenarios [e.g.

Feeley and Silman (2010); Gallagher (2016)]. In addition to assessing extinction risk, analysing variation in physiological specialisation holds great interest for quantifying the relative importance and generality of different biogeographic factors such as dispersal, adaptive radiation and vicariance in explaining extant distribution patterns (Crisp & Cook, 2013; Crayn et al., 2015; Schluter, 2015).

A key element of environmental specialisation is the species niche concept, which comprises factors such as habitat, resources, environmental requirements and physiological attributes, that vary among taxa, and can be quantified in different ways (Roughgarden, 1972; Rabinowitz, 1981;

Slatyer et al., 2013).

Given the multiple abiotic and biotic factors that influence species distributions across macroecological scales, quantifying the environmental niches of individual species is problematic

Chapter 4 Continental species niches

(Devictor et al., 2010; Boulangeat et al., 2012; Slatyer et al., 2013). Previous analyses have used geographic records to classify species as specialists or generalists based on co-occurrence patterns

(Gaston & Fuller, 2009; Boulangeat et al., 2012; Slatyer et al., 2013; Vimal & Devictor, 2015).

Other studies have attempted to incorporate the multivariate nature of the niche by calculating niche volumes (i.e. multi-dimensional measures of the niche space, based on geometry or probability) using either empirical or simulated data (Broennimann et al., 2012; Swanson et al., 2015).

Regardless of the methods used, complex interactions between niche overlap, niche width and biotic factors such as competition mean that no single measure can conclusively summarise the niche (Bar-Massada, 2015).

A potentially useful approach to analysing macroecological variation in niches is to quantify how different niche parameters vary across environmental space, using simple empirical measures for environmental gradients of known physiological importance to particular taxa. Here we consider the realised environmental niche width of an individual species: for example the continental range of rainfall, temperature and soil nitrogen conditions across which a tropical plant has been recorded

(Austin et al., 1990). Such measures cannot quantify species’ fundamental physiological tolerances.

Nonetheless, in the same way that patterns of biological diversity can be formed by divergent processes (Chase & Myers, 2011), different physiological and biogeographic mechanisms have shaped realised environmental distributions along individual gradients (Figure 4.1).

Chapter 4 Continental species niches

Figure 4.1. Graphical representation of how the realised environmental niche width of different plant species (coloured lines) could vary along an environmental gradient in relation to physiological performance (P). If the environmental gradient (for example rainfall) was important for plant physiology, and its effect on physiology varied along the gradient, a systematic relationship between realised niche width and niche median (i.e. niche centre) would manifest (a). However, this pattern could also result from biogeographic processes, being formed by the long term specialisation of species to conditions at the dry end, then rapid evolution of species to newly available conditions at the wet end (see discussion). Alternatively, if the environmental condition was important, but its effect on physiology was constant across the gradient, a range of realised niche widths could occur at any point (b). However, this non-systematic pattern could also emerge if the particular environmental gradient had no effect on plant physiology. If the pattern in panel a) holds, we would also expect the niche centre to be positively related to functional traits (bottom left panel). For example plants with niches centred in dried areas may have smaller leaves.

For example, assuming that rainfall, temperature and soil nitrogen are important for the physiology of a tropical flora, and that the effect of each environmental condition on plant physiology varies along continental environmental gradients, a systematic relationship between niche width and niche centre should manifest (Figure 4.1 a). Alternatively, if each environmental condition was important, but their physiological effects were constant across the gradient, wide or

Chapter 4 Continental species niches narrow niches could occur at any point along the gradient (Figure 4.1 b). Following a similar vein of logic, Gauch and Whittaker (1972, 1976) hypothesized that despite strong species replacement with environmental turnover, such a neutral relationship could manifest in diverse ecosystems such as tropical forests. Importantly, environmental niches are species-specific properties. A particular environmental gradient might be important to none, few, many or all of the taxa comprising a flora.

Thus, analysing the variation in niche width for an entire flora could blend the two patterns depicted in Figure 4.1, diluting the effect of the gradient on physiological specialization. Nonetheless, as a logical starting point, we hypothesize that the realised environmental niche width of vascular plants is positively related to the niche median: thus niche width should systematically increase across rainfall, temperature and nitrogen space.

Species’ extant distributions provide the means to define ecological parameters such as niche width, which can then be compared to variation in functional traits that more directly reflect plant physiology. The link between environmental niches and traits may arise from shifts in the ecological competitiveness of different trait values along environmental gradients (Westoby &

Wright, 2006). However, strong relationships between environmental distributions and functional traits may have several causes (Reich et al., 2003), including ancient molecular divergence, genotypic pre-adaptation to new environments and recent adaptive divergence.

Considering the physiological manifestation of these factors across a continental rainfall gradient, for example, we expect that plants with environmental niches centred at the drier end of the gradient should be physiologically specialised to dry conditions. Thus these plants should have smaller, harder, and thicker leaves, greater stomatal control, and generally lower stature [e.g.

Körner and Cochrane (1985); Fonseca et al. (2000); McDonald et al. (2003), Figure 4.1, bottom left panel]. Conversely, plants with niches centred at the wetter end of the continental gradient should be physiologically specialised to wet conditions, and have larger, softer, and thinner leaves, less

Chapter 4 Continental species niches stomatal control, and taller stature. Given these physiological expectations, we pose a second hypothesis: that the realised niche median in rainfall, temperature and nitrogen space is positively related to both the leaf area and maximum height of vascular plants. Similarly, we can extend these bivariate niche concepts to multivariate space, and hypothesize that combined niche volume is positively related to leaf area, maximum height and the niche median in rainfall, temperature and nitrogen space for vascular plants.

Few empirical studies have analysed variation in the environmental distributions and functional traits of diverse tropical floras [e.g. Feeley and Silman (2010, 2011); Petter et al. (2016)].

In this context, the vascular flora of the Australian Wet Tropics is among the best studied globally

(Metcalfe & Ford, 2009), with extensive geographic records available to quantify the environmental distributions of a highly diverse, endemic admixture of congeneric, confamilial and more distantly related taxa (Costion et al., 2015; Crayn et al., 2015; Thornhill et al., 2016). We test whether the environmental niches and traits of the Wet Tropics flora vary systematically across the continental range of rainfall, temperature and soil nitrogen conditions in Australia.

Chapter 4 Continental species niches

4.2 Methods

4.2.1 Wet Tropics datasets

Biological data: species occurrence records and functional trait values

Our study focuses on species present in the Wet Tropics, analysing their occurrences across the entire range of climates found in Australia. First, biological data were obtained for species occurring within the Wet Tropics bioregion of Australia (Thackway & Cresswell, 1995) and the surrounding area within a 100 km buffer [approximately 113,003 km2, Figure 4.2, Mokany et al.

(2014)]. Geographic records were sourced from the Atlas of Living Australia (www.ala.org.au/ , accessed 23 June 2015) using a list of 4292 vascular plants that are known to occur in the Wet

Tropics based on Queensland Herbarium surveys (Mokany et al., 2014). The ALA4R R package

(github.com/AtlasOfLivingAustralia/ALA4R) was used to download geographic records for the

Wet Tropics species that had a spatial error ≤1000 m, across mainland Australia and Tasmania. This yielded a total of 1,951,173 species records with a median of 73 records per species.

Unfortunately, the spatial density of the geographic species records in Southeast Asia and the Pacific is lower than the records from within Australia. Thus we have only quantified environmental niches within Australia, where the spatial accuracy of each record can be better quantified with denser geographic sampling. Values for maximum plant height (m), leaf width

(mm) and leaf length (mm) were obtained for each species using publicly available databases, including the Australian Tropical Rain forest Plants identification system

(www.anbg.gov.au/cpbr/cd-keys/rfk/) and the Flora of Australia Online

(www.anbg.gov.au/abrs/online-resources/flora/). Estimated leaf area was calculated in cm2 using the leaf length and width from these data sources, with the simplifying assumption that all species have elliptically shaped leaves. Given the rarity of many Wet Tropics species, detailed information

Chapter 4 Continental species niches on leaf shape is unavailable for most species. However, future analyses could use digital herbarium records to build large databases of leaf area, while also varying the assumption of elliptically shaped leaves for the subset of species with leaf shape data.

Environmental conditions

Species’ realised continental environmental niche widths were estimated using average annual values from 1976-2005 of total annual rainfall (mm) and maximum temperature of the warmest period (°C) at 250 m × 250 m spatial resolution. We used the 9 arc-second Australian digital elevation model in ANUCLIM [version 6.1, Xu and Hutchinson (2013)] to generate separate interpolation surfaces for both Australia and the Wet Tropics only, then merged the two surfaces to produce a single surface. This was done to better incorporate the effects of slope and aspect on climate in the Wet Tropics (Figure 4.2). Surfaces (250 m × 250 m) for the mass fraction of total nitrogen by weight (%) in the top three soil profiles (0-5, 5-15 and 15-30 cm) were obtained from the Soil and Landscape Grid of Australia (www.clw.csiro.au/aclep/soilandlandscapegrid/). These surfaces represent the best available data and are based on geostatistical interpolations of historical soil data and soil spectra (Viscarra Rossel et al., 2015; Viscarra Rossel & Bui, 2016). We averaged the values for the top three surfaces to derive a single 0-30cm soil nitrogen surface.

Chapter 4 Continental species niches

Figure 4.2. Maps of the biological and environmental data. The records of the vascular plants (a) used to analyse species environmental niches (366,111 geographic records of 1771 species plotted in blue. This is the subset of all 4292 species with niche medians within the Wet Tropics environmental bounds, as defined in section 4.2.2. The broader Wet Tropics study region is shown in red). The records for the tree species (b) with measured maximum height and leaf dimensions (98,337 geographic records of 647 species plotted in blue. This is the subset of all 1351 tree species with niche medians within the Wet Tropics environmental bounds, as defined in section 4.2.2). As an example, the geographic points for the purple Laurel tree, Cryptocarya hypospodia, are plotted in pink over the major Australian ecoregions (c): desert/arid (grey), montane (cyan), Mediterranean (tan), Savanna (khaki), tropical forest (dark green), temperate forest (light green, shapefile of global ecoregions: http://www.worldwildlife.org/publications/terrestrial-ecoregions-of-the-world). 250 m × 250 m surfaces for annual rainfall (blue, d), maximum temperature of the warmest period (orange, e) and mass fraction of total soil nitrogen in the top 30cm (purple, f, in panels d-f the bioregions has been enlarged). Boxplots for the environmental values of all records of Cryptocarya hypospodia [g, using points on map c) showing the rain (RW), temperature (TW) and nitrogen niche width (NW, coloured dashes on y-axes)]. In panels d- f, the bioregion is enlarged, and the legends below the Australian maps are scaled to the entire continent, whereas the legends on the right of the Wet Tropics refer to the Wet Tropics only.

Chapter 4 Continental species niches

4.2.2 Statistical analysis

Quantifying continental niche widths for the Wet Tropics vascular flora

We extracted the environmental values for all the geographic records of each species, under the assumption that the species data represent a random sample across their continental environmental range. Realised niche width was defined as the 95th percentile minus the 5th percentile for all continental environmental values of each species. Similarly, realised niche median was defined as the median for all continental environmental values of each species (hereafter all references to niche parameters are realised niches).

Some species are widespread across several biomes (see Figure 4.2), being relatively rare in tropical ecosystems yet abundant in other landscapes (for example Australia’s most widely distributed tree, the River Red Gum, Eucalyptus camaldulensis). These environmentally widespread species would bias the analysis if their realised environmental niches were truncated to only the

Wet Tropics, creating edge effects [e.g. Feeley and Silman (2010)]. Therefore we excluded 2521 species with niche medians either less than the 5th percentile of rainfall for just the Wet Tropics

(580.59 mm), or that did not fall within the middle 90% of temperature values (29.59-35.43°C,

Figure 4.2). Hereafter, these rainfall and temperature limits are referred to as the ‘Wet Tropics environmental bounds’. This left 1771 species for statistical analysis, or 41% of the original 4292 species that occur in the Wet Tropics.

By excluding the most environmentally widespread species, we are assuming this will not bias the analyses towards species with narrower environmental niches. Similarly, the use of the niche median to quantify the niche centre may bias the analyses if species are abundant in sub- optimal conditions due to confounding factors, such as increased competition in more optimal locations or sampling bias (e.g. proximity from roads, research institutions and botanical expertise).

Chapter 4 Continental species niches

However, testing showed that analysis results were consistent regardless of the data subset used or the measure of niche centre chosen (see Appendix 3, Table S3.1, Table S3.2).

Bivariate regression models

To test whether realised niche width was positively related to niche median, we ran separate bivariate regressions between realised niche width (response) and niche median (predictor) across each environmental gradient. We ran the analyses on species with ≥20 records per species to balance the creation of robust summaries of species’ environmental distributions against the arbitrary exclusion of genuinely rare Wet Tropics species. Here we note that a similar threshold of

>30 was used by Feeley and Silman (2010), and the pattern of relationships between niche width and niche median was similar when using both lower and higher thresholds (see, Appendix 3 Table

S3.1).

These regressions were run on two data subsets: 1) All vascular plant species that have their rainfall and temperature medians within the Wet Tropics environmental bounds, as defined in the previous section (n = 1771), and 2) All tree species within the Wet Tropics environmental bounds with empirical measurements for their maximum height, leaf width and length (n = 647, hereafter we simply refer to ‘all trees’ for this subset). Although 2989 of 4292 species have empirical measurements for their functional traits, we analysed only trees with measured functional traits (n =

647) to make the analyses of realised niches and functional traits directly comparable, given that statistical relationships hold across life forms (Appendix 3, Table S3.3, Figure S3.1, Figure S3.2).

The relationship between niche width and niche median was non-linear and strongly skewed across the rainfall and nitrogen gradients (Figure 4.4). However, the relationship between temperature niche width and niche median was linear. Therefore we used generalised additive

Chapter 4 Continental species niches models (GAMs) of the natural log of rainfall and nitrogen niche width (response) vs. the natural log of niche median (predictor). Three knots were specified for the splines of the niche median to capture overall trends without over-parameterising the models [mgcv R package, v.1.8.4; R Core

Team (2015)]. Preliminary analyses demonstrated that running GAMs on the natural log of rainfall and nitrogen niche width (response) and niche median (predictor) provided substantial improvements in the model fit and the residuals, compared with regressions without spline smoothers (e.g. second order polynomials).

Because ecological specialization can be reflected in both narrower environmental niches, and less variable functional traits, we treat both environmental distributions and traits as different components of the realised physiological specialization for each species. To test the hypothesis that niche median was positively related to both the leaf area and maximum height, we ran separate linear regressions between the niche median across each environmental gradient (response), and maximum recorded plant height and leaf area for tree species (predictor). Raw values for plant

2 height were used (m), while the log10 (leaf area cm + 1) was used for leaf area, due to the very large leaf areas of a small number of species (hereafter we simply refer to ‘leaf area’). We assessed the strength of all statistical relationships using the coefficients, standardised test statistics and explanatory power from the GAMs and linear models. For the rainfall and nitrogen gradients, we used the S1 coefficient (i.e. the first spline coefficient), F-statistics and deviance explained values.

2 For the temperature gradient, we used the β1 coefficient, T-statistic and adjusted R values.

It is important to emphasise that many of the Wet Tropics species co-occur and thus interact, meaning their environmental niches are not independent from one another. Several methods are available to control for non-independence in geographic space [for example spatially autoregressive models, Meynard et al. (2011); Meynard et al. (2013); Gallagher (2016)]. However, as noted in the introduction, our hypotheses are explicitly formulated in environmental space, meaning we cannot

Chapter 4 Continental species niches link the environmental niche of each species back to a geographic location. More importantly, given the size of the macroecological dataset used, the identification of broad trends across continental environmental gradients takes precedence over measures of statistical significance. Even the most restrictive analyses that attempt to control for spatial structure resulting from biogeographic history

[e.g. Hawkins (2012b, 2012a)] will likely find significant relationships using such large datasets.

There are two additional sources of statistical dependence in this dataset which complicate attempts to infer general patterns of environmental specialisation across continental environmental gradients. First, significant relationships between realised environmental niches and functional traits could emerge simply through the combination of spatial clustering of species in geographic space, and the inherent spatial structure in the environmental variables [e.g. Boucher-Lalonde and Currie

(2016), see Figure 4.2]. Second, the plant species in our analysis share considerable and variable evolutionary history, given the close phylogenetic relationships between some genera and families.

Thus the environmental niches of the Wet Tropics taxa should be considered phylogenetically autocorrelated (Harvey & Pagel, 1998; Freckleton et al., 2002).

We tested the statistical significance of our results using a randomisation procedure to create a distribution against which to compare the observed values (Manly, 2006). First we allocated the species geographic records into all 50 km × 50 km cells across continental Australia (see Figure

4.3). Second, we randomly re-allocated each observed species record to one of the sample locations within the same 50 km × 50 km block in which it originally occurred. The random allocation process was repeated 5000 times to generate 5000 alternate realisations of the sample. The block size of 50km was chosen to balance small differences between the observed and randomised data against larger differences. For example experimentation indicated that smaller block sizes of 10 km

× 10 km do not alter the observed data substantially, thus regression results will be similar, whereas larger block sizes of 100 km × 100 km alter the observed data substantially, changing the regression

Chapter 4 Continental species niches results. Finally, for each random realisation, we re-estimated the realised niche width and niche median of all species and re-ran the regressions (GAMs and linear models). The pseudo P-values for each regression were considered significant if the observed coefficient was in the upper, lower or outer 5% of the random coefficients for one-tail high, one-tail low or two tailed tests, respectively (i.e. a rank relative randomisation, Figure 4.3 d).

Figure 4.3. Visualisation of the randomisation procedure for testing relationships between realised environmental niches and functional traits. Starting with all spatially valid species geographic records (a) within the WT environmental bounds, we allocated each record into 50 km × 50 km blocks across continental Australia (b), including Tasmania. We then randomly shuffled the identity of each record within the same 50 km × 50 km block in which it occurred (c, note the transposition of species i and j). Finally, we re-estimated the realised niche width and median of all species, and then re-ran the regressions (GAMs and linear models) using these randomised data (d). The pseudo P-values for these randomisations were considered significant if the observed coefficient (red line) was in the upper, lower or outer 5% of random coefficients (shaded blue areas of the histogram) for one-tail high, one-tail low or two tailed tests, respectively.

Chapter 4 Continental species niches

To correct for the potential effect of shared evolutionary history on our results, we conducted supplementary analyses using the subset of our Wet Tropics taxa covered by the largest and most up to date angiosperm phylogeny [n = 723, Zanne et al. (2014), noting that no comprehensive phylogeny for the WT taxa has yet been published]. We ran phylogenetically corrected linear models for the relationship between niche width (response) and niche median

(predictor) using the phylolm function in the phylolm package [Tung Ho and Ané (2014), v.2.5; R

Core Team (2015)]. These regressions use phylogenetic models for the covariance in the residuals.

Unfortunately, current methods for conducting phylogenetically independent contrasts cannot account for non-linear relationships between continuous variables such as environmental niches and traits (Quader et al., 2004). This is a key limitation for our dataset, given the clear positive, curvilinear variation in rainfall and nitrogen niches across the gradient (see Figure 4.4).

However, the overall relationship between niche width and niche median for all vascular plants did not change substantially across continental environmental gradients when correcting for phylogenetic relatedness (Appendix 3, Table S3.5 Figure S3.4). Furthermore, the phylolm results for the relationships between niche median and functional traits were also similar to the species- level results (see Appendix 3, Table S3.6). Thus we focused on interpreting the species-level results.

Structural equation models

To test whether a combined measure of niche width, niche volume, was positively related to functional traits in a multivariate framework incorporating interactions, we fitted a structural equation model to data for all tree species (n = 647). Tree species were used to ensure the

Chapter 4 Continental species niches multivariate results were comparable with the bivariate trait analyses. Several methods have been recently developed to quantify niche ‘hyper-volumes’ [e.g. Geange et al. (2011); Blonder et al.

(2014); Swanson et al. (2015)]. Similarly, the individual geographic records of each species could also be used in a principal components analysis to calculate a volumetric hull for each species (i.e. a three dimensional convex hull where each dimension is an environmental condition). However, simple geometric methods, based on our bivariate analyses of realised niche width and relying on the fewest possible assumptions, are sufficient to create low-dimensional estimates of realised niche volumes.

We scaled the original environmental values by the minimum and maximum values of all species geographic records across each environmental gradient, allowing the variables to have different means and standard deviations, but equal niches (Grace, 2006). We then summarised the scaled values into realised niche width (middle 90% of standardised environmental values) and median of the standardised values for each species. The realised niche volume (Nv) was calculated for each species by assuming the scaled niches across each environmental gradient to be elliptical, and using the formula of a three dimensional ellipsoid:

휋 푁 = × 푟푎𝑖푛 푤𝑖푑푡ℎ × 푡푒푚푝 푤𝑖푑푡ℎ × 푛𝑖푡푟표푔푒푛 푤𝑖푑푡ℎ 푣 6

The realised niche volume of all tree species (n = 647) was then used as the primary response variable in a structural equation model, assuming all variables to be measured directly

(noting the results were similar when using all species within the Wet Tropics environmental bounds, n = 1771, see Appendix 3 Figure S3.5). We regressed niche volume against the explanatory

2 variables of scaled rainfall, temperature and nitrogen niche medians, leaf area [log10 (leaf area cm +

Chapter 4 Continental species niches

1)] and maximum plant height (m). Causality was assumed to be from traits to niche medians, and from niche medians to niche volume, while including covariances between niche medians and between traits.

Linear formulae were used given that the primary relationships between niche volume and all other variables had no discernible functional form. The standardised path coefficients represent the expected change in niche volume as a function of the change in the realised niche medians and functional traits in standard deviations (Grace, 2006), being analogous to the coefficients of a standard linear regression with a single explanatory variable. All connections with non-significant path coefficients were excluded, creating a minimum-significant SEM (see Appendix 3 Figure S3.5 for the full SEM). The structural equation models were calibrated in R [lavaan R package, v. 0.5-21;

R Core Team (2015)]. To test whether niche volume was related to niche medians and traits without interactions between predictors, we also ran a supplementary generalised additive model between niche volume (response) and niche medians, leaf area and height (predictors, see Appendix 3, Table

S3.7).

4.3 Results

Realised temperature niche width was moderately, positively and linearly correlated with temperature niche median for all vascular plant species occurring within the Wet Tropics environmental bounds used here (Table 4.1, adjusted R2 = 0.085). However, this relationship for the temperature gradient was non-significant according to the spatially constrained randomisations of the species identities (Table 4.1, P >0.9). In contrast, niche width was more strongly correlated with niche median for all species across rainfall and nitrogen space (Table 4.1, deviance explained =

0.387 for rainfall and 0.513 for nitrogen).

Chapter 4 Continental species niches

There was a generally positive, curvilinear and significant trend towards narrower niche widths being centred in drier and less fertile environments (Table 4.1, Figure 4.4 a, c, e).

Furthermore, the relationship between niche width and niche median for just the tree species was consistent across each environmental gradient with the relationship for all species (Table 4.1, Figure

4.4 b, d, f). In addition, these species-level results for niche width vs. niche median were consistent with the linear regressions that used phylogenetic models for the covariance in residuals (which were run on the subset of 723 Wet Tropics taxa found on the current angiosperm phylogeny by

Zanne et al. (2014), see Appendix 3, Table S3.5, Figure S3.4).

Table 4.1. Bivariate regression results for realised niche width (NW) vs. realised niche median (NM) across each environmental gradient, for all species analysed (n = 1771), and all tree species analysed (n = 647). Explanatory power (EP) is deviance explained 2 for rainfall and nitrogen, and adjusted R for temperature. C is the linear coefficient, β1, for temperature, and the first GAM coefficient, S1, for rainfall and nitrogen. The F-statistic is given for rainfall and nitrogen, and the T-statistic is given for temperature.

P is one minus the fraction of all 5000 randomisations where the observed coefficients were larger than the random coefficients (β1 for temp, and S1 for rainfall and nitrogen).

Species Model N EP Coefficients Test stat P Intercept C All Rainfall log(NW) ~ log(NM) 1771 0.387 7.012 0.499 559.96 0.0002 All Temp NW ~ NM 1771 0.085 -9.665 0.460 12.85 >0.99 All Nitrogen log(NW) ~ log(NM) 1771 0.513 -2.213 0.296 936.01 <0.001 Trees Rainfall log(NW) ~ log(NM) 647 0.322 7.159 0.558 154.29 0.0002 Trees Temp NW ~ NM 647 0.095 -10.956 0.482 8.27 >0.99 Trees Nitrogen log(NW) ~ log(NM) 647 0.473 -2.076 0.406 290.71 0.202

Chapter 4 Continental species niches

Figure 4.4. Plots of species realised niche width vs. niche median for the Wet Tropics vascular flora, for the natural log of precipitation (originally mm), maximum temperature (°C), and the natural log of soil nitrogen (originally %). For each environmental gradient, the regression model is plotted in orange over species niches (blue points), with the explanatory power for each model shown in orange (deviance explained values for rainfall and nitrogen, and adjusted R2 values for temperature). The left column of panels (a, c, e) show all species analysed (n = 1771). The right column of panels (b, d, f) show all tree species analysed (n = 647).

Chapter 4 Continental species niches

Overall, niche median was more strongly related to leaf area than it was to maximum plant height, with the strongest relationship for the trait analyses being positive linear correlations between rainfall median and leaf area (Table 4.2, adjusted R2 = 0.162, Figure 4.5 b). This relationship between rainfall median and leaf area was the only trait model with both a moderate R2 value and a linear coefficient that that differed significantly from the mean of the randomized distribution (Table 4.2). The niche median of Wet Tropics tree species across each environmental gradient was weakly linearly related to maximum plant height, effectively showing no relationship

(e.g. Figure 4.5 a, c, e). Although the pseudo p-values of the spatially-constrained randomisations were not necessarily consistent with the explanatory power of each trait model, the relationships between temperature median and both functional traits were the least significant of all the trait models considered (Table 4.2).

Table 4.2. Bivariate linear regression results for realised niche median in each environmental gradient vs. plant height (maximum 2 recorded height in metres of each tree species) and leaf area [log10 (leaf area cm + 1)] for all tree species analysed (n = 647). For 2 each model, the adjusted R , linear intercept, coefficient (β1) and T-statistic is given. The pseudo P-values for these randomisations were considered significant if the observed coefficient was in the upper, lower or outer 5% of random coefficients for one-tail high, one-tail low or two tailed tests, respectively (i.e. a rank relative randomisation, see Figure 4.3).

Model Adjusted R2 Intercept Coefficient T-stat P Rainfall median ~ height 0.024 1615.84 11.090 4.111 0.219 Temp median ~ height 0.028 32.11 -0.022 -4.425 <0.0002 Nitrogen median ~ height 0.031 0.11 0.001 4.622 <0.0002 Rainfall median ~ leaf area 0.162 688.53 620.706 11.208 <0.0002 Temp median ~ leaf area 0.024 32.55 -0.459 -4.132 <0.0002 Nitrogen median ~ leaf area 0.045 0.08 0.024 5.585 <0.0002

Chapter 4 Continental species niches

Figure 4.5. Plots of realised niche median for all tree species analysed (n = 647) vs. plant height and leaf area. For each environmental gradient, the linear model is plotted in orange over the raw data (blue points), with adjusted R2 values shown in orange in the top right corner of each panel. Height on the x-axis is the maximum recorded height in metres of each tree species, while Log 2 area is the log10 (leaf area cm + 1).

Chapter 4 Continental species niches

The structural equation model of combined niche volume in rainfall, temperature and nitrogen space vs. individual niche medians, height and leaf area for all tree species was generally consistent with the bivariate trait analyses. Although the niche medians were strongly correlated between environmental gradients, the overall explanatory power for niche volume was moderate

(R2 = 0.123, Figure 4.6). Similar to the models of individual niche medians vs. functional traits, the relationship between niche volume and leaf area was stronger than for maximum height (Figure 4.6, standardised path coefficient = 0.174, whereas the path coefficient for maximum height was non- significant and thus excluded from the minimum significant SEM, see Figure 4.6 caption for goodness of fit statistics).

Figure 4.6. Minimum significant structural equation model diagram for realised niche volume vs. niche median across each 2 environmental gradient, leaf area and height for all tree species analysed (n = 647). Leaf area is the log10 (leaf area cm + 1) and height is maximum recorded height (m). Niche volume is a three dimensional ellipsoid, where the dimensions are 0-1 scaled species environmental realised niche widths (rain, temperature and nitrogen). The scaled realised niche medians are shown (originally in mm, °C and % for rainfall, temperature and nitrogen respectively). Single-headed arrows are directional linear relationships, double-headed arrows are co-variances. Arrow thickness is proportional to the standardised path coefficient labelled on each line (blue = positive, red = negative, * <0.05, ** <0.001, *** <0.0001). For the minimum-significant SEM, χ2 = 1.446 (P> 0.229), the root mean square error of approximation (RMSEA) = 0.026 (P = 0.534, lower 90% confidence interval = 0, upper = 0.112). The comparative fit index is >0.99.

Chapter 4 Continental species niches

4.4 Discussion

Temperature niches are relatively constant across continental Australia, whereas

rain and nitrogen niches widen systematically

For the Wet Tropics vascular plant species included in our analysis, wide or narrow temperature niches can occur at any point across continental Australia (Figure 4.4 c, d). This result indicates that the effect of temperature on the physiology of the Wet Tropics species is relatively constant across temperature space (confirming the pattern hypothesised in Fig. 1 b). The lack of variation in temperature niches for the Wet Tropics taxa contrasts with the most analogous study by

Feeley and Silman (2010), which found that realised temperature niches for 2151 plants from the hottest tropical forests in South America were several degrees narrower than those of plants from cooler forests.

Importantly, the pattern for Neotropical plants was not directly attributable to increased physiological specialisation (Feeley & Silman, 2010), being influenced by the truncation of niches at the warmer end of the gradient, but not the colder end. Niche truncation is not limited to the

Neotropics — the Wet Tropics taxa are also restricted by the available temperature space across continental Australia (see Figure 4.2). However, the greater range in mean annual temperature across continental Australia for our study period (4.5-29.64 °C) compared with that of the

Neotropical forests used by Feeley & Silman (23.3-28.3°C) should have increased the chance of observing systematic variation in realised niches. Ultimately, the lack of variation in temperature niche width for the Wet Tropics species could be attributable to biotic factors, in that the lower species diversity in Australian tropical ecosystems relative to Neotropical forests may reduce interspecific competitive pressure across the continental temperature gradient [e.g. Fox (1981);

Manthey et al. (2011)].

Chapter 4 Continental species niches

There is no directly comparable study to ours for Australian plants. However, for the entire

Australian seed flora, higher mean annual temperature is consistently associated with large geographic ranges (Gallagher, 2016). Although Wet Tropics species with large geographic ranges have wider realised temperature niche widths (see Appendix 3, Figure S3.6), the geographic ranges of all species in the Australian seed flora do not vary substantially with latitude (Gallagher, 2016).

If Rapoport’s rule is indeed not as evident in the Australia flora as in other continental floras — presumably due to weaker Australian elevational and latitudinal temperature gradients — we might also expect to see a weaker pattern of variation in physiological specialisation across the continental temperature gradient.

In contrast to the temperature gradient, the Wet Tropics plant species displayed a systematic pattern of increasing niche width across continental gradients of rainfall and soil nitrogen (Figure

4.4 a, b). Indeed this pattern manifested irrespective of functional type (Appendix 3, Figure S3.1,

Figure S3.2). This result suggests that water and nitrogen availability shift from being limiting factors affecting plant growth and reproduction at the dry and infertile end of the gradient, to non- limiting factors at the wet and fertile end. This curvilinear, increasing pattern is likely biogeographic in origin. Australia’s consistent northward drift towards the equator has created a drier, more seasonal monsoon climate, compared with the wetter conditions that prevailed before

Australia separated from Antarctica approximately 38 Mya (White, 1994; Crayn et al., 2015).

Similarly, the relative geological stability of the Australian landmass since the early Neogene, combined with climatic fluctuations during the Quaternary period, have driven strong soil weathering and a gradual decrease in continental soil fertility (McKenzie et al., 2004).

The prolonged drying and soil weathering of Australia ecosystems fostered the adaptive radiation of monsoonal and sclerophyllous taxa (Crisp & Cook, 2013; Pfautsch et al., 2016) which dominate contemporary plant habitats across large areas of tropical Australia. Thus while most

Chapter 4 Continental species niches species of Gondwanan origin have adapted to dry, infertile conditions over tens of millions of years

(left half of Figure 4.4 a, b, e and f), some species may have become geographically restricted to the remaining moister environments (right half of Figure 4.4 a, b, e and f). Similarly, species that have arrived more recently from Southeast Asia via oceanic dispersal (Crisp & Cook, 2013; Crayn et al., 2015) may still be expanding their distributions into niches vacated by the contraction of the mesic biome following convergence of the Australian or South East Asian continental shelves

(Bowman, 2000; Hilbert et al., 2007; Byrne et al., 2011; Crayn et al., 2015). Interestingly, for the subset of 457 species in our analysis which have their biogeographic origins attributed to either

Australia or Southeast Asia (25% of the total 1771 species), there is no difference in rainfall, temperature or nitrogen niche width. That is, species of both origins are spread evenly across all three gradients (Appendix 3, Figure S3.7). This spread may indicate that, independent of biogeographic origin, the variable monsoon climate has selected for wider precipitation ranges in tropical Australia to tolerate annual cycles of drought and inundation (Gallant et al., 2007).

Niche ~ trait relationships vary considerably within environments

For the Wet Tropics tree species included in our analysis, leaf area was larger in wetter environments (Figure 4.5 b), concurring with established empirical relationships between leaf physiognomy and contemporary climate (e.g. http://clamp.ibcas.ac.cn). This pattern suggests that a strategy of greater physiological specialization at both ends of the gradient — smaller leaves at the dry end, and larger leaves at the wet end — indeed increases ecological competitiveness (Westoby

& Wright, 2006). Similarly, assuming wetter conditions with low water stress are more optimal and hence competitive (Reich et al., 2003), our results support trade-offs between hydraulic conductance (being advantageous in wet conditions) and cavitation resistance (being advantageous in dry conditions) along the rainfall gradient (Etterson & Shaw, 2001). The bivariate results were

Chapter 4 Continental species niches also consistent with the structural equation models, which showed that a moderate amount of variation in niche volume was explained by niche medians and leaf area [supported by a supplementary multiple regression between niche volume (response) and niche medians, leaf area and height (predictors), see Appendix 3, Table S3.7]. Thus we conclude that individual and collective environmental niches of the Wet Tropics tree species analysed here are positively related to leaf area.

The pattern of increasing leaf area across the rainfall gradient is diluted by the substantial variation in functional traits between species that occupy the same environment, but utilise different life-history strategies (Westoby et al., 2002; Wright et al., 2004; Westoby & Wright, 2006). This variation creates substantial noise in the relationships between environmental niches and traits, particularly those for plant height. Furthermore, the relationship between the niche median for each environmental gradient and both leaf area and height was similar, regardless of biogeographic origin (for the subset of 457 species where information on biogeographic origin is available, see

Appendix 3, Figure S3.8 and Figure S3.9). Although the biogeographic origin of most Wet Tropics plants remains unknown, this result provides partial evidence that both Australian and Southeast

Asian plants have responded to environmental change in similar ways across continental Australia, but over different time spans.

Improving the quantification of physiological specialisation across macroecological

scales

This analysis represents one approach for analysing continental variation in physiological specialization, approximated by realised environmental niches and functional traits. Our primary limitation — the reliance on species geographic records, mostly collected ad-hoc for herbaria — is

Chapter 4 Continental species niches unavoidable for a macroecological analysis applied across such broad environmental and biological scales. Thus we have simply approximated the realised environmental niche of each species at the spatial resolution (accurate to 1km) and extent (continental Australia) of the geographic records, rather than quantifying species’ fundamental physiological tolerances (Feeley, 2015).

Inevitably, realised niches will underestimate fundamental niches. But more importantly, this underestimation — whether caused by biased environmental sampling, or non-environmental dispersal barriers — will vary with the spatial, environmental and biological scale of analysis. Such principles are well established in the geographic literature (Tobler, 1970; Openshaw, 1983), but continue to preoccupy the increasingly sophisticated methods employed in macroecology. A relative parsimonious avenue for minimising the influence of scale on future analyses could be to use only the best sampled taxa. For example the relationship between realised niche width and niche median for only species with more than 100 geographic records produces a similar pattern to the main results (Appendix 3, Table S3.1, i.e. even when poorly sampled taxa are removed).

The key feature of our approach is the quantification of physiological specialization in environmental space, rather than geographic space [with most analogous studies using geographic measures of ecological specialisation, e.g. Boulangeat et al. (2012); Fort and Inchausti (2013);

Gallagher (2016)]. Thus we cannot link the realised niche median of each species to a geographic location, where proxies of competition such as co-occurring species richness could be quantified.

For example the use of the realised niche median to quantify the niche centre could have biased the analyses, if species are abundant in sub-optimal conditions due increased competition in more optimal conditions. Macroecological models of physiological specialization could be improved by combining continental environmental niches calculated using geographic records with ecological plot data, where community variables such as species, functional and phylogenetic diversity can be quantified. Furthermore, to better estimate realised environmental niches, controlled physiological

100

Chapter 4 Continental species niches experiments (Swarbreck et al., 2011; Tomlinson et al., 2012) could be used to calibrate mechanistic niche models that incorporate non-environmental factors such as competitive exclusion [e.g.

Webber et al. (2011); Cabral and Kreft (2012); Higgins et al. (2012)].

Conclusions

For vascular plants occurring within the Australian Wet Tropics, we found strong evidence that wide or narrow realised temperature niches can occur at any point across the continental temperature gradient. Conversely, species occupying wetter and more fertile environments have wider niches and larger leaves, with niche-trait patterns confirming established relationships between leaf physiognomy and contemporary climate. Further work is required to clarify the role of adaptation and competition in shaping relationships between continental environmental distributions and functional traits in the Australian context.

4.5 Supporting information

Additional supporting information submitted to the Journal of Biogeography is included here as:

 Appendix 3: Supplementary material for Chapter 4 — Additional figures and tables for

the relationship between realised niche width and niche median, the relationship between

niche median and traits, and the relationship between niche volume, niche medians and

traits.

101

Chapter 4 Continental species niches

102

103

5 Primary productivity is related to niche width in the Australian Wet Tropics

Hugh M. Burley, Karel Mokany, Shawn W. Laffan, Kristen J. Williams, Dan Metcalfe, Helen T.

Murphy, Andrew Ford, Tom D. Harwood and Simon Ferrier

This Chapter is in review for Global Ecology and Biogeography.

Contributions: Hugh Burley developed the hypotheses, wrote the R code to perform the analyses and wrote the article. Karel Mokany helped develop the hypotheses, wrote code to create a variable rainfall threshold between site-pairs and provided supervisory guidance. Shawn Laffan wrote code to create the site-pair datasets, helped with code development and provided supervisory advice.

Kristen Williams provided additional environmental layers and Dan Metcalfe, Helen Murphy and

Andrew Ford provided the ecological plot data. Tom Harwood provided additional environmental layers and suggestions on the use of the primary productivity layers. Simon Ferrier helped develop the hypotheses and advised on the publication strategy. All co-authors provided editorial suggestions for the article

104

Chapter 5 Bioregional productivity and niche width

Abstract

Aim A key debate in contemporary ecology is whether ecosystem functions are distinctly influenced by biological diversity, across broad scales. Although recent work has emphasised the importance of links between ecosystem functions and measures of ecological specialisation as proxies of biodiversity, few studies have empirically analysed net, macroecological relationships in diverse environments. We tested whether gross primary productivity (GPP) in the Australian Wet

Tropics (WT) was distinctly related to community-level measures of the ecological specialisation of component tree species across environmental space, after accounting for environmental drivers.

Location WT biogeographic region, Australia.

Methods Using all geographically valid herbarium records for 948 WT tree species, we quantified realised environmental niche widths using continental surfaces for maximum temperature of the warmest period and total annual rainfall. The median realised niche width for all tree species occurring at 510 sites was used to approximate ecological specialisation within communities. To partial out the direct effects of abiotic environment on GPP, we applied a novel analysis based on the difference in GPP and the difference in median community realised niche width between site- pairs with similar environmental conditions. Linear models were then run on the difference in GPP between site-pairs (response) and the difference in environmental niche widths (predictor).

Results For environmentally similar sites in drier areas, GPP was higher in sites comprising species with narrower temperature niches (average R2 = 0.087, average t-statistic = -3.45). Conversely, for environmentally similar sites in drier areas, GPP was lower in sites comprising species with narrower rainfall niches (average R2 = 0.171, average T-statistic = 5.06).

105

Chapter 5 Bioregional productivity and niche width

Main conclusions. Wet Tropics sites with more thermal specialists had higher productivity, whereas sites with more moisture specialists had lower productivity. These mixed findings suggest that physiological specialisation across environmental space can influence primary productivity at broad scales, but in inconsistent ways.

5.1 Introduction

The notion that the ecosystem functions at a location — for example primary productivity

— could be influenced not only by the environment, but also by the diversity of the biota itself, goes back to at least the writings of Darwin (1859), and probably much earlier. Ecosystem functions have a rather convoluted history of description and quantification, much like measures of biological diversity. Nonetheless, the intuitive definition provided by Ghilarov (2000): ‘stocks and fluxes of matter and energy derived from biological activity’, has achieved wide currency. Many ecologists have used this definition to examine the hypothesis that greater biological diversity at a location (of species, functional traits or genes) leads to a greater magnitude and lower variability of ecosystem functions such as primary productivity, at least up to a point (Loreau et al., 2003; Cardinale et al.,

2012). This overarching prediction comes in many guises, but is the essence of the concept of biological complementarity (Grime, 1998), and underpins the classic biodiversity-ecosystem function (B-EF) argument in the ecological literature [e.g. Tilman and Downing (1994); Tilman

(1999); Tilman et al. (2012)].

In order to isolate relationships between ecosystem functions and biodiversity from the direct effects of environmental conditions, B-EF experiments are usually conducted on small plots in the same environment, subject to a range of biodiversity treatments. Typically, a small number of taxa (e.g. <100 species, and sometimes including species functional traits and genes) are analysed at

106

Chapter 5 Bioregional productivity and niche width fine spatial resolutions (on the order of square metres), across small geographic extents (on the order of square kilometres) and over brief time periods [<10 years, e.g. Cadotte (2015a); Isbell et al.

(2015b); Chiang et al. (2016)]. This relatively narrow, aspatial scope — compared with the spatio- temporal variation in natural ecosystems — limits the direct applicability of much B-EF research for regional, continental and global conservation assessment (Spasojevic et al., 2016). This is because the potential effect of the significant biogeographic shifts that are predicted under global environmental change scenarios on ecosystem functions are largely ignored by focusing exclusively on biological α-diversity (Mokany et al., 2016). Although some studies have recently tested the complementarity concept at broader spatial and biological extents [e.g. Paquette and Messier

(2011); Chisholm et al. (2013); Ruiz-Benito et al. (2014); Jing et al. (2015); Poorter et al. (2015);

Liang et al. (2016); Spasojevic et al. (2016)], the results do not universally support the complementarity hypothesis. Thus recent work has begun to focus on potential relationships between ecosystem functions and biological diversity at macroecological scales.

A broader macroecological complementarity hypothesis was recently proposed to incorporate the spatial dimension of biodiversity into B-EF analyses (Pasari et al., 2013; Wang &

Loreau, 2014; Burley et al., 2016b; Wang & Loreau, 2016). Macroecological complementarity centres on relationships between ecosystem functions and community-level measures of ecological specialisation. In biogeographic regions where the biota has become adapted to current environmental conditions through deterministic evolutionary processes (such as competitive assembly), species should be finely partitioned across environmental space, developing refined physiology that operates most effectively under a narrow suite of conditions. At any particular location within such regions, the average environmental niche width of all species present should therefore be relatively narrow, reflecting greater physiological optimisation to contemporary environments. The macroecological complementarity hypothesis states that locations with a greater degree of ecological specialisation should display higher magnitudes and lower variabilities of

107

Chapter 5 Bioregional productivity and niche width ecosystem functions than locations with less specialisation. However, species can also become restricted to narrow portions of environmental space purely through stochastic evolutionary processes (for example vicariance), without becoming physiologically optimised to current conditions (Chase & Myers, 2011). Similarly, a species may appear highly specialized simply because it has been excluded from some portion of an environmental gradient where it could otherwise persist by other more competitive taxa. Thus, summary measures of environmental niche width for the species within a community — which more directly quantify environmental specialisation than measures such as beta diversity? — may be necessary to test the macroecological complementarity hypothesis in natural systems.

Given the divergent processes by which contemporary biodiversity patterns have been formed, the argument that diversity must be preserved at both local and macroecological scales in order to maintain current and future ecosystem functions can only be properly tested across broad spatio-temporal and biological scales. A recent analysis of European forest plots found that floristic

β-diversity was positively related to multiple ecosystem functions at the landscape scale (Van Der

Plas et al., 2016). However, attempts to establish empirical links between ecosystem functions and ecological specialisation across broader environmental and biological scales have thus far proven equivocal [e.g. for continental floras, see Burley et al. (2016a)]. Establishing these empirical links is a crucial first step before practical applications of the macroecological complementarity concept can be considered — for example systematic conservation planning that maximises the degree of ecological specialisation retained across continental scales [see Bush et al. (2016)]. This is a seemingly obvious point, yet is often overlooked in the B-EF literature.

Previous macroecological B-EF analyses have attempted to control for direct relationships between ecosystem functions and environmental conditions by using all locations where variables were measured, and including environmental conditions as co-variates. For example structural

108

Chapter 5 Bioregional productivity and niche width equation models have been used to specify causal relationships between different measures of environment, photosynthetic activity (for example primary productivity) and biological diversity

[e.g. Grace et al. (2016); Spasojevic et al. (2016)]. Similarly, spatially explicit models could be used to isolate the relationship between ecosystem functions and diversity from confounding environmental factors that are correlated with the geographic distance between sites [e.g. Gouveia et al. (2013); Mellin et al. (2014); Roll et al. (2015)]. A conceptually parsimonious alternative for testing the macroecological complementarity mechanism — which is also in keeping with traditional plot-scale B-EF experiments — is to analyse sets of sites that have relatively similar environments, yet different levels of ecological specialisation (i.e. to analyse site-pairs). Although defining such locations across continental and global extents is problematic due to a lack of survey plot data, the flora of many bioregions have observational and survey data of sufficient quality and quantity to allow plausible analyses. For example the highly diverse, endemic vascular flora within and around the Australian Wet Tropics (WT) has been relatively well sampled (Metcalfe & Ford,

2009; Crayn et al., 2015).

Here we conduct a novel empirical test of the macroecological complementarity hypothesis.

For pairs of Wet Tropics sites, we assess whether sites comprising species with collectively narrower niches – those that have a greater degree of ecological specialisation – are more productive, after removing the influence of site environment. We use gross primary productivity

(GPP) as a measure of ecosystem functioning. GPP is defined as the photosynthetic flux of carbon between vegetation and the atmosphere (Monteith, 1972; Donohue et al., 2014). We predict that pairwise GPP will be negatively linearly related to pairwise niche width (Figure 5.1).

109

Chapter 5 Bioregional productivity and niche width

Figure 5.1. We tested whether sites with greater ecological specialisation were more productive, after removing the influence of site environment. Site-level niche width (a) was calculated as the median of the continental environmental niches for all tree species occurring at each site. We hypothesise (b) that the relationship between pairwise difference in GPP, and pairwise difference in niche width, should be negative (represented by the solid red line). If this prediction holds between site-pairs with similar environmental conditions, it would indicate that ecological specialisation can have a positive influence on site productivity, independent of abiotic environmental conditions.

110

Chapter 5 Bioregional productivity and niche width

5.2 Methods

Overview

Our study region encompasses the Wet Tropics bioregion of Australia (Thackway &

Cresswell, 1995) and the surrounding area within a 100 km buffer [approximately 113,003 km2, with the extent top: 14° 35' 51.746" S, left: 143° 48' 12.463" E, bottom: 20° 27' 55.175" S, right:

147° 35' 0.561" E, see Figure 5.2]. To test the pairwise relationship between GPP and ecological specialisation for the Wet Tropics flora (Figure 5.1), we combined spatial surfaces of environmental conditions with ecological survey plots and observational species records. First we derived the average monthly GPP, monthly total rainfall and monthly mean daily maximum temperature for the period 2001-2012 using existing surfaces (Donohue et al., 2014; Hutchinson et al., 2014). These values were calculated for 510 Wet Tropics plots where a complete inventory of the vascular flora has been conducted. For all geographically valid herbarium records of all tree species occurring at these 510 sites, we quantified the continental realised rainfall and temperature niches over a longer time period (1976-2005) than site environment was quantified over (2001-2012). The longer time period was used to better characterise the realised niche of each species. For all possible site-pairs, we calculated the difference in GPP, site environment and median community realised niche width between sites. Site-pairs where selected for analysis based on environmental similarity (pairs with a difference > 0.5°C were excluded, while pairs from drier areas could have smaller differences than wetter areas). Pairs were then subsampled so that each site appeared only once in the final set of site-pairs, repeating the subsampling 500 times. For each subsample, linear models were fitted to the difference in GPP between site-pairs (response variable) as a function of the difference in either rainfall or temperature niche width between site-pairs (predictor variable, Figure 5.1).

111

Chapter 5 Bioregional productivity and niche width

5.2.1 Analysis variables used to test the macroecological complementarity mechanism

Gross primary productivity

We employed monthly 250 m resolution estimates of GPP from January 2001 to December

2012. These surfaces use the shortwave irradiance, the diffuse sunlight fraction and MODIS-derived foliage cover values as inputs to a parsimonious light use efficiency model, that quantifies the monthly photosynthetic flux of carbon across the Australian continent [g C m-2 month-1, Donohue et al. (2014)]. Site productivity was calculated by averaging the GPP values of each 250 m grid cell across all 144 months (2001-2012, Figure 5.2 a).

Environmental conditions at the Wet Tropics sites

The factors shaping the distribution of rainforest in Australia are not purely climatic, and encompass complex interactions between edaphic conditions and disturbances such as fires and cyclones (Bowman, 2000). Nonetheless, the GPP of individual sites in the humid, monsoonal climate of the Wet Tropics will be strongly influenced by water stress and extremes of temperature across inter-annual wet-dry cycles. Thus the explanatory factors for GPP were specified as the total monthly rainfall (mm month-1) and the mean daily maximum temperature of each month (°C). We used the ANUCLIM 1.0 surfaces to quantify rainfall and temperature at 1 km × 1 km resolution from January 2001 to December 2012 (Hutchinson et al., 2014). We again calculated the average monthly rainfall and temperature across all 144 months.

Site-level GPP will also be influenced by non-climatic factors such as soil nutrient status, competition and disturbance history. For example some areas within the Wet Tropics can be very

112

Chapter 5 Bioregional productivity and niche width wet, yet have low productivity when nutrients are more limiting for plant growth than water.

However, we omitted soils from the analysis to avoid circularity, as the best available continental estimates for Australian soil conditions contain MODIS input [e.g. Viscarra Rossel and Bui (2016)]

— as do the GPP surfaces. Moreover, although competition between plant species will be reflected to some extent in their degree of ecological specialization — as outlined in the introduction — the effects of competition on specialisation are difficult to distinguish from other stochastic processes at macroecological scales. For all Wet Tropics sites where the vascular flora had been comprehensively surveyed, we extracted the average GPP, rainfall and temperature values across the 2001-2012 time series (Figure 5.2 b, c).

Continental environmental niches for the tree species occurring at the Wet Tropics

sites

To quantify ecological specialisation in environmental space, we obtained data for 510 sites in the Wet Tropics where a comprehensive survey of the vascular flora had been conducted, for either 50 m × 20 m plots or 50 m × 10 m plots [Mokany et al. (2014), Figure 5.2 d]. 2375 species of vascular plants were recorded at these sites, 948 of which were trees. Given that primary productivity will mainly be influenced by the largest plants, we focused on quantifying the ecological specialisation of trees. For all 948 tree species occurring at the survey plots, we downloaded all geographic herbarium records across mainland Australia and Tasmania with spatial error ≤1000 m from the Atlas of Living Australia (www.ala.org.au, accessed 23 June 2015, yielding a median accuracy of 100 m per record).

The realised environmental niche of each species was then estimated using average values from 1976-2005 of total annual rainfall (mm) and maximum temperature of the warmest period

113

Chapter 5 Bioregional productivity and niche width

(°C) at 250 m × 250 m spatial resolution (see Chapter 4). We used the 9 arc-second Australian digital elevation model in ANUCLIM [version 6.1, Xu and Hutchinson (2013)] to run separate interpolations for continental Australia and the Wet Tropics separately. The two climate surfaces were then merged to account for the stronger effects of topography on climate in the Wet Tropics.

The longer time period was used for the species environmental niches (1976-2005) than for the site environment (2001-2012), to better quantify the realised environmental niche of each species across continental Australia. Similarly, the average maximum of the monthly daytime temperature was considered a better proxy of thermal tolerance — and hence a better measure of niche width — than mean annual temperature. This is because the conditions influencing photosynthesis during the day are partitioned from those influencing respiration overnight, rather than averaging the overnight and daytime temperatures over each month of the year. These climate interpolations were applied to continental Australia and the Wet Tropics separately, and the two data series were merged to account for the stronger effects of topography on climate in the Wet Tropics.

We defined the realised environmental niche width of each tree species as the middle 90% of temperature and rainfall values for all continental geographic records of that species [i.e. the 95th percentile minus the 5th percentile for the 1976-2005 surfaces, Figure 5.2 e]. Because the spatial density of the geographic species records in Southeast Asia and the Pacific is lower than the records from within Australia, we have quantified environmental niches within Australia only. Our proxy of ecological specialisation at the community level (Devictor et al., 2010) was calculated by taking the median realised rainfall and temperature niche width for all species occurring at each site — i.e. site-level niche width (see Figure 5.1 a). Median site niche width was strongly, positively and linearly correlated with both mean site niche width and the geometric mean site niche width.

114

Chapter 5 Bioregional productivity and niche width

Figure 5.2. We tested the macroecological complementarity hypothesis by combing continental environmental surfaces with ecological plot surveys and observational plant records. Maps of average gross primary productivity (GPP) for the Wet Tropics (gC m-2 month-1 2001-2012 in brown, a, used to quantify site productivity), and for eastern Australia. Maps of total monthly rainfall for the Wet Tropics (mm month-1 2001-2012 in blue, b, used to quantify site rainfall), and of annual rainfall across continental Australia (mm 1976-2005, used to quantify median site rainfall niche width). Maps of mean monthly daily maximum temperature for the Wet Tropics (°C 2001-2012 in orange, c, used to quantify site temperature), and of the maximum temperature of the warmest period across continental Australia (°C 1976-2005, used to quantify median site temperature niche width). Histograms of environmental values for the Wet Tropics are shown below each surface. The top row of histograms are for the whole Wet Tropics study region, and the bottom row are for only the 510 WT sites (d). The niche width of each tree species occurring at the sites was defined as the middle 90% of environmental values for all Australian records of that species [e.g. Ficus congesta, e, with the rain niche width in blue (RW) and temperature niche width in orange (TW)].

115

Chapter 5 Bioregional productivity and niche width

5.2.2 Statistical Analyses

Creating site pairs and filtering by environmental differences

We tested the macroecological complementarity hypothesis by quantifying the relationship between the difference in GPP for pairs of Wet Tropics sites, and the difference in community-level summaries of species rainfall and temperature niche width (Figure 5.1 , for site-level relationships, see Appendix 4, Table S4.1, Table S4.2 and Table S4.3). For all possible 129,795 pairwise combinations, we calculated the difference in site GPP (gC m-2 month-1, 2001-2012), site temperature (°C, 2001-2012), site rainfall (mm month-1, 2001-2012), median site temperature niche width (°C, 1976-2005) and median site rainfall niche width (mm, 1976-2005). We ordered calculation of the difference so that the site in the pair (either site i or j) with the maximum niche width (NW) always came first (equations 1 and 2):

퐺푃푃 퐷𝑖푓푓푖푗 = 퐺푃푃max(푁푊푖, 푁푊푗) − 퐺푃푃min(푁푊푖, 푁푊푗) (1)

푁𝑖푐ℎ푒 푊𝑖푑푡ℎ 퐷𝑖푓푓푖푗 = 푁푊푀푎푥(푁푊푖, 푁푊푗) − 푁푊푀𝑖푛(푁푊푖, 푁푊푗) (2)

This ordering ensured that a continuous regression between pairwise GPP difference

(response variable) and pairwise niche difference (predictor variable) tested the hypothesis that sites with narrower environmental niches have higher GPP (Figure 5.1). The pairwise calculations were run using three separate rainfall subsets of the original sites: all 510 sites (ranging from 42-516 mm of mean monthly rainfall), 255 drier sites (ranging from 42-138 mm of rainfall) and 255 wetter sites

(ranging from 139-516 mm of rainfall, see “Subset” column in Table 5.1). To partial out relationships between GPP and environmental niches, we then filtered the unique site-pairs by the

116

Chapter 5 Bioregional productivity and niche width difference in temperature and rainfall between sites, using the monthly values averaged across

2001-2012. We used a fixed allowable difference in temperature between site-pairs (maximum ΔT

= 0.5 °C), given that temperature varies less across the Wet Tropics than rainfall (Figure 5.2 c) and does not correlate strongly with GPP for our time series data (Figure 5.3 a). For rainfall, we used a variable allowable difference between site-pairs, to incorporate the saturating relationship between site GPP and site rainfall into the pairwise analyses (Figure 5.3 b). To determine the variable rainfall limit, we modelled site GPP as a negative exponential function of site rainfall — a positively monotonically increasing non-linear regression (equation 3):

퐺푃푃 = 푎(1 − 푒−푏.푟푎푖푛푓푎푙푙) (3)

Where a = 277.49 and b = 0.019 (R2 = 0.66, P-value = <0.001), based on monthly site rainfall (2001-2012). Using the non-linear regression parameters and a maximum allowable difference in predicted GPP at each site (ΔGPP, set at 20), we generated lower and upper limits of rainfall differences between any site-pair along the rainfall gradient (i.e. red and blue lines on

Figure 5.3 c, respectively). This approach allowed the rainfall difference between site-pairs to vary along the gradient, being largest for pairs in the wettest areas (where additional rainfall has no discernible effect on GPP). The candidate set of environmentally similar site-pairs for regression analysis was created by excluding any site-pair where the rainfall of the second site (site j in the ij pair) was either less than the lower rainfall limit (red line on Figure 5.3 c), or greater than the upper rainfall limit (blue line on Figure 5.3 c), according to the negative exponential prediction.

117

Chapter 5 Bioregional productivity and niche width

Figure 5.3. Scatterplots of site-level relationships between gross primary productivity (GPP), environmental conditions and environmental niche widths. Scatterplots of site GPP vs. all variables in the analysis (a), RNW: median rainfall niche width of all tree species occurring at the site (mm 1976-2005), TNW: median temperature niche width of all tree species at the site (°C 1976-2005). The 510 Wet Tropics sites (b) used to generate pairwise variables were considered in three sets; all sites, drier sites receiving ≤138 mm monthly rainfall (mostly in savannah environments) and wetter sites receiving ≥139 mm monthly rainfall (mostly in rainforest environments). The upper and lower limits of allowable differences in rainfall between site-pairs vary across the gradient based on a negative exponential model of site GPP (gC m-2 month-1) as a function of site rainfall (c, blue and red lines respectively, see methods). It is also important to note that using the response variable (GPP) to constrain the explanatory variable (niche width) raises the problem of circularity. Particularly, the variable rainfall limit constrains the difference in GPP between pairs of sites to a relatively low level. This could reduce the effect of niche-width on GPP (similar to the bias towards wetter and more productive sites observed in chapter 3). However, using a different relationship for calculating the variable rainfall limit (for example a global GPP– rainfall relationship) is equally problematic, given that this would forgo the advantages of calibrating the analysis to the biogeography of Wet Tropics. A simple alternative is to either use a fixed limit (as was used for temperature), or to use a standard multiple regression at the site level (so just using 510 sites, with site GPP ~ site environment + site niche width, see Appendix 4, Tables S4.1, S4.2 and S4.3). Also note that in chapter 4, species-level niche width (in mm) was log transformed to clarify the relationship between species niche width and niche position (see Figure 4.4). In chapter 5, site-level niche width was not transformed (i.e. left in the original units of mm), because a different question is being tested: whether site niche width affects GPP after environmental effects have been accounted. If site rainfall and temperature are in the same units as site rain niche and site temp niche, the test is more logical and rigorous.

118

Chapter 5 Bioregional productivity and niche width

Subsampling of site-pairs for regression analysis

To minimise the influence of any particular site on the results, we randomised the subset of environmentally similar site-pairs 500 times, and then subsampled from these pairs so that each site appeared only once in the final subsample set of site-pairs. Thus each site occurred only once in each subsample, but the same sites occurred across multiple subsamples. For the final dataset of each subsampling iteration — ordered by one explanatory variable at a time — we ran linear regressions between the pairwise GPP difference (response) and the pairwise explanatory variable difference (see Figure 5.1 b):

lm (GPP difference between sites ~ temperature niche width difference between sites) lm (GPP difference between sites ~ rainfall niche width difference between sites)

For each linear model, we tested the relationship between pairwise GPP difference and pairwise niche difference (Figure 5.1) using three diagnostic measures, which were averaged across all 500 subsamples [using the R stats package, version 1.8.4, R Core Team (2015)]. The linear coefficients (β1) and t-statistics (T) indicated the direction and size of the linear effects, while the adjusted R2 values (Adj. R2) indicated the strength of the linear relationships. Statistical significance

(P) was calculated for each linear model under the null hypothesis of no relationship between pairwise GPP difference and pairwise niche width difference, at α = 0.01. Using one unique subsample of site-pairs for each model, P was calculated as the fraction of 1000 bootstrapped t- statistics which were more extreme than the average t-statistic from all 500 subsamples. Preliminary analyses also demonstrated that once the site-pairs had been filtered by environmental similarity, there was no clear pattern in the residuals of the linear models vs. either the fitted values, or the

119

Chapter 5 Bioregional productivity and niche width

Euclidean distance between the sites in each pair. Thus spatially explicit models were not needed to partial out the effect of site environment.

5.3 Results

For site-pairs calculated from all original 510 sites — spanning the entire rainfall gradient from 42-516 mm of monthly rainfall — GPP was moderately related to both temperature and rainfall niche width. The strength of these relationships was similar for both rainfall and temperature niches across the entire rainfall gradient (Table 5.1, average adjusted R2 across 500 subsamples = 0.087 for temperature niche width and 0.083 for rainfall niche width). For site-pairs calculated from only the 255 sites spanning the drier portion of the gradient (from 42-138 mm of monthly rainfall), the relationship between GPP difference and temperature niche width difference was unchanged compared with all sites. However, GPP difference was more strongly related to rainfall niche width difference for drier site-pairs (Table 5.1, average adjusted R2 = 0.171 for rainfall niche width). In contrast, for site-pairs calculated from only the 255 sites spanning the wetter portion of the gradient (from 139-516 mm of monthly rainfall), GPP difference was moderately related to temperature niche width difference, and weakly related to rainfall niche difference (Table 5.1, average adjusted R2 = 0.08 and 0.017 for temperature and rainfall niche width respectively).

120

Chapter 5 Bioregional productivity and niche width

Table 5.1. Average results for 500 linear models run on subsamples of environmentally similar, unique site-pairs, where pairwise differences in gross primary productivity (GPP, gC m-2 month-1 2001-2012) are a function of pairwise differences in median site rainfall and temperature niche widths (mm and °C respectively). All site-pair differences are ordered by the response variable, so that GPP difference values are always positive. ‘N’ is the average number of site- pairs used across 500 subsamples, ‘Int’ is the average linear intercept, ‘β1’ is the average linear coefficient, ‘T’ is the average t-statistic and ‘R2’ is the average adjusted R2. ‘P’ is the fraction of 1000 bootstrapped t-statistics — from one unique subsample under the null hypothesis —which were more extreme than the average t-statistic from all 500 subsamples.

2 Precip. subset (mm) Model N Int β1 T R P All(42-516) GPP ~ temp niche 245.2 -1.113 -14.136 -4.87 0.087 <0.001 All(42-516) GPP ~ rain niche 245.4 -0.977 0.031 4.76 0.083 0.001 Drier(42-138) GPP ~ temp niche 121.2 -0.881 -14.595 -3.45 0.087 <0.001 Drier(42-138) GPP ~ rain niche 121.3 -2.056 0.049 5.06 0.171 <0.001 Wetter(139-516) GPP ~ temp niche 123.7 -1.606 -13.605 -3.34 0.080 <0.001 Wetter(139-516) GPP ~ rain niche 123.7 0.156 0.014 1.64 0.017 0.045

The effect sizes of the relationships between GPP and niche width were generally consistent with the explanatory power of each model. For site-pairs calculated from all original sites, the effect sizes of temperature and rainfall niche width were of similar magnitude, but opposing directions

(Table 5.1, average t-statistic across 500 subsamples = -4.87 for temperature niche width and 4.76 for rainfall niche width). GPP was always negatively related to temperature niche width, regardless of the site-pair subset used (Figure 5.4 a, c, e). Conversely, the relationship between GPP and rainfall niche width was largely positive (Figure 5.5, a, c), albeit effectively neutral for wetter site- pairs (Figure 5.5, e). Moreover, the positive effect of rainfall niche width was appreciably stronger than the negative effect of temperature niche width for site-pairs calculated from drier site-pairs

(Table 5.1, average t-statistic for rainfall niche width = 5.06 for drier sites, and average t-statistic for temperature niche width = -3.45).

Overall, the mean explanatory power and effect sizes across 500 subsamples were consistent with our bootstrapped P-values (P). Using one unique subsample of site-pairs for each model, P was calculated as the fraction of 1000 bootstrapped t-statistics which were more extreme than the average t-statistic from all 500 subsamples. The effects of both temperature and rainfall niche width

121

Chapter 5 Bioregional productivity and niche width were significant when analysed for site-pairs subsampled from all original sites, and drier sites

(Table 5.1, P <0.001). The effect of rainfall niche width was non-significant at α = 0.01 when analysed for site-pairs subsampled from wetter original sites (Table 5.1, P = 0.045). Overall, the pairwise relationships between GPP difference and niche width difference were consistent with the original site-level relationships between GPP and niche widths (see Figure 5.3 a). Site GPP had a negatively linear relationship with the median temperature niche width of all tree species occurring at each site. Conversely, site GPP had a positive, saturating relationship with the median rainfall niche width of all tree species occurring at each site — reflecting stronger rainfall niche width effects at drier sites (Figure 5.3 b, also see Appendix 4, Table S4.1, Table S4.2 and Table S4.3 for site-level multiple regression results).

122

Chapter 5 Bioregional productivity and niche width

Figure 5.4. Plots of linear models for 500 subsamples of environmentally similar, unique site-pairs, where pairwise differences in gross primary productivity (GPP, gC m-2 month-1 2001-2012) are a function of pairwise differences in median site temperature niche width (°C). The slopes of all 500 linear models are plotted in grey, and their mean slope is plotted in orange. Panel a) plots relationships for site-pairs derived from all original 510 sites (42-516 mm). Panel c) plots relationships for site-pairs derived from only the 255 driest sites (42-138 mm). Panel e) plots relationships for site-pairs derived from only the 255 wettest sites (139-516 mm). The right column of panels (b, d, and f) plot histograms of 1000 bootstrapped t-statistics from one subsample of each model under the null hypothesis in pink, with the average t-statistic from all 500 subsamples plotted in green.

123

Chapter 5 Bioregional productivity and niche width

Figure 5.5. Plots of the linear models for 500 subsamples of environmentally similar, unique site-pairs, where pairwise differences in gross primary productivity (GPP, gC m-2 month-1 2001-2012) are a function of pairwise differences in median site rainfall niche width (mm). The slopes of all 500 linear models are plotted in grey, and their mean slope is plotted in orange. Panel a) plots relationships for site-pairs derived from all original 510 sites (42-516 mm). Panel c) plots relationships for site-pairs derived from only the 255 driest sites (42-138 mm). Panel e) plots relationships for site-pairs derived from only the 255 wettest sites (139-516 mm). The right column of panels (b, d, f) plot histograms of 1000 bootstrapped t-statistics from one subsample of each model under the null hypothesis (in grey), with the average t-statistic from all 500 subsamples plotted in green.

124

Chapter 5 Bioregional productivity and niche width

5.4 Discussion

Higher primary productivity is related to greater temperature specialisation

Our pairwise analyses of environmentally similar sites in the Australian Wet Tropics support the macroecological complementarity hypothesis when considering ecological specialisation in temperature space (see Figure 5.1). Ostensibly, the consistently negative relationship between GPP and temperature niche width (Figure 5.4, a, c, e) implies that Wet Tropics sites containing tree species more specialised to certain thermal environments have higher productivity. This link between productivity and ecological specialisation is consistent with predictions in the literature regarding the magnitude of ecosystem functions (Loreau et al., 2003;

Pasari et al., 2013; Burley et al., 2016b; Van Der Plas et al., 2016; Wang & Loreau, 2016), and may have a basis in biogeographic and evolutionary patterns at the species level.

For example a recent global analysis of vertebrates found higher net diversification rates in species with narrower temperature niches (Rolland & Salamin, 2016). If higher temperatures in the tropics mean tropical ectotherms perform more optimally over a narrower thermal range than species from higher latitudes (Payne & Smith, 2017), the same principle may apply to plants.

Assuming thermal tolerances have evolved deterministically, site niche width may indeed approximate physiological performance at each point in environmental space. Thus our results for temperature niche width could indicate that species with narrower temperature distributions can outperform those with wider distributions, as approximated by the difference in the gross photosynthetic flux of carbon between sites.

Whether analysed at the site or the site-pair level, GPP consistently decreased with temperature niche width across the rainfall gradient (Figure 5.3 a, Figure 5.4 a, c, e). Despite the restrictive nature of our pairwise analyses (which were run on relatively small subsets of all

125

Chapter 5 Bioregional productivity and niche width possible site-pairs), the temperature niche width results were consistent with the original site-level relationships (Figure 5.3 a). The narrow niched species driving the negative relationship between

GPP and temperature niche width are necessarily rare in geographic and environmental space.

Although some authors have suggest that common taxa should influence ecosystem functions more than rare taxa [e.g. Grime (1998); Winfree et al. (2015)], experimental evidence suggests rare taxa can play a greater role when a wider range of environmental conditions is considered (Isbell et al.,

2011; Turnbull et al., 2016). Our Wet Tropics analyses — covering strong gradients of elevation, rainfall and temperature — certainly fit these criteria (Figure 5.2). Given that site GPP had a weak, idiosyncratic relationship with site temperature (Appendix 4, Table S4.1), the consistently negative linear relationship between pairwise GPP and pairwise temperature niche width could represent a genuine physiological signal of the macroecological complementarity mechanism.

Lower primary productivity is related to greater rainfall specialisation

In contrast to the temperature dimension, our pairwise results reject the macroecological complementarity hypothesis when considering ecological specialisation along the rainfall gradient.

The generally positive relationship between GPP and rainfall niche width demonstrates that Wet

Tropics sites containing tree species that are more specialised to certain rainfall conditions have lower productivity (Figure 5.5 , a, c). Assuming that site niche width approximates physiological performance, this result suggests that rainfall generalists can outperform specialists at a given point in environmental space, contradicting our findings for temperature niche width. The key to understanding these contrasting relationships between productivity and specialisation in temperature and rainfall space could lie in the evolution of environmental niches for the Wet

Tropics flora in response to climatic variability.

126

Chapter 5 Bioregional productivity and niche width

An important biogeographic feature of contemporary Australian environments is the strong inter-annual variability in rainfall (Nicholls et al., 1997; Power et al., 1999; Gallant et al., 2007;

Murphy & Timbal, 2008). Moreover, in the Wet Tropics — which along with southwest Tasmania is the wettest region of Australia — mean rainfall variability across our time series is greater than mean temperature variability (see Appendix 4, Fig. S4.1). Assuming this disparity in variability between rainfall and temperature has persisted over evolutionary time-scales, Wet Tropics trees that became specialised to narrow portions of rainfall space would have faced greater physiological risks than those that became specialised to narrow portions of temperature space.

This logic concurs with simulations and experiments suggesting that environmental variability may have greater negative effects on the survival of species with narrow environmental distributions than on those with wide distributions (Schemske et al., 1994; Drake & Lodge, 2004;

Bartholomeus et al., 2011). If the negative impact of rainfall variability on rare species holds among the Wet Tropics taxa, wide rainfall niches could indeed optimise their physiological performance over multiple years. Thus when species niches are scaled to the community level, our results aggregated across 12 years imply that Wet Tropics sites containing more species with wide rainfall niches can maintain greater productivity over time (Figure 5.3, Figure 5.5). Furthermore, given this aggregated result, a temporally subset analysis could explicitly test whether sites with more rainfall generalists better maintain their productivity within each year, and potentially within each season

[e.g. Shriver (2016)].

The positive relationship between productivity and niche width is also influenced by the fact that wetter sites contain more tree species with wider rainfall niches than drier sites. This bias in median site rainfall niche width is in turn driven by the systematic widening of the continental rainfall niches for species in wetter areas for our Wet Tropics taxa (see Appendix 4, Fig. S4.2).

Conversely, the temperature niche widths of our Wet Tropics taxa can be wide or narrow across the

127

Chapter 5 Bioregional productivity and niche width continental temperature gradient (Appendix 4, Fig. S4.2). Thus high GPP sites occur in areas with both higher rainfall, and where greater numbers of species have wider continental rainfall niches.

Isolating relationships between GPP and rainfall niche width is therefore more problematic for the rainfall dimension than for temperature, even if using alternative analysis methods (for example spatially explicit models, structural equation models or mixed effects models).

The macroecological complementarity effect depends on the biogeographic context

Unfortunately, we cannot comprehensively quantify a key assumption of our hypothesis — that the degree of ecological specialisation at each site was formed through adaptive, rather than stochastic evolutionary processes (Chase, 2010; Chase & Myers, 2011). Nonetheless, our contrasting results demonstrate that greater ecological specialisation across environmental space does not consistently lead to greater magnitudes of primary productivity in the Wet Tropics of north eastern Australia. This further underscores that empirical relationships between ecosystem functions and ecological specialisation are difficult to generalise across broad spatial and biological scales

[e.g. Wardle et al. (1997); Huston et al. (2000); Jonsson et al. (2001)].

The explanatory power of our relatively simple linear models was moderate, and these results were commensurate with recent broad-scale studies using analogous data. For example the productivity response to fire of 6603 remotely sensed pixels in the US Southwest had only moderate relationships with the functional diversity of regeneration and seed mass functional traits for the coincident vegetation types in those pixels (Spasojevic et al., 2016). Given the considerably greater variability of natural vs. controlled ecosystems — and the different processes by which ecological specialisation can be formed — the macroecological complementarity effect could be positive, negative or neutral, depending on ecological context (Burley et al., 2016b). Care must therefore be

128

Chapter 5 Bioregional productivity and niche width taken by researchers to avoid simply generating empirical measures of ecological specialisation until they find ones that confirm their hypotheses (our hypothesis being just one example among many in the broader B-EF literature).

We quantified ecological specialisation by combining plot data with continental environmental niche widths, rather than using measures which classify species as specialists or generalists based on geographic patterns of co-occurrence [e.g. Fridley et al. (2007); Boulangeat et al. (2012); Vimal and Devictor (2015)]. We argue that our approach is preferable for testing the macroecological complementarity hypothesis, given the observed lack of relationship between productivity and biological turnover in geographic space (Burley et al., 2016a). Nevertheless, we must acknowledge that realised niches will always underestimate fundamental niches, because the full environmental tolerances of most plant species are unknown (Feeley & Silman, 2011; Feeley,

2015), particularly in diverse tropical ecosystems such as those of the Wet Tropics.

The interrelationships we observed between productivity, environment and ecological specialisation were ultimately shaped by the biogeographic confines of the Wet Tropics.

Environmental space is skewed towards drier and hotter conditions, both across our Wet Tropics study region, and across continental Australia. Although this bias is surprisingly well captured by our relatively small sample of 510 sites, the range of community-level niche widths is nonetheless

‘truncated’ by the available rainfall and temperature niche space (Feeley & Silman, 2010). This environmental truncation is especially relevant for those taxa that are endemic to the Wet Tropics, and may limit the scope for niche width to influence productivity. The effects of ecological specialisation on ecosystem functions, both positive and negative, could therefore be stronger when evaluated for sites spanning a broader range of biogeographic conditions and niche widths. For example, the macroecological complementarity hypothesis could be tested for multiple global

129

Chapter 5 Bioregional productivity and niche width bioregions with similar contemporary environments, yet different biogeographic histories [e.g.

Manthey et al. (2011)].

Despite the contrasting nature of our results, the negative relationship between productivity and temperature specialisation nonetheless warrants further consideration in light of climate projections. The increased temperatures forecast under global climate change scenarios [for example +2.49 °C to +4.93 °C under Representative Concentration Pathway 4.5, England et al.

(2015)] may cause the extinction of those Wet tropics taxa that are specialised to narrow portions of temperature space (Williams et al., 2009; Cabrelli et al., 2014), assuming an equivalent transpiration response. If locations with greater temperature specialisation have higher productivity under current conditions, the loss of specialised taxa could then reduce the primary productivity at those locations. To assess such predictions, further analyses could test whether the physiological performance of individual species is related to realised niche width under current and future climates, either through meta-analyses, or new studies linking laboratory and field data [e.g. Austin et al. (2009); Müller et al. (2017)]. If ecological generalists outperform specialists at the centre of the specialists’ niche, this would contradict the hypothesis that narrower niches at the community level facilitate greater magnitudes of ecosystem functions.

Conclusions

We demonstrated that primary productivity was related to community-level measures of ecological specialisation at macroecological scales in natural environments. Wet Tropics sites with more ecological specialists in temperature space had higher productivity, whereas sites with more ecological specialists in rainfall space had lower productivity. Future analyses could focus on the degree to which species’ physiological performance is related to their realised niche width under

130

Chapter 5 Bioregional productivity and niche width current and future climates, to directly test the physiological basis for the hypothesis that ecological specialisation increases productivity.

5.5 Supporting information

Additional supporting information is included here as:

 Appendix 4: Supplementary material for Chapter 5 — Additional figures of

environmental variability and species level niche widths. Additional tables for site-level

relationships, for the effects of rainfall and temperature on pairwise GPP, for the full results

at multiple rainfall thresholds, and for the effects of mean niche widths on pairwise GPP.

131

6 Discussion

In the preceding Chapters I have tested empirical relationships between gross primary productivity and ecological specialisation at continental and regional extents, as well as analysing variation in species’ realised environmental niches across continental Australia. In this Chapter I first summarise the key findings of each analytical research component (Chapters 3-5), and then discuss potential avenues for further testing the hypotheses originally described in Chapter 2.

6.1 Continental primary productivity and geographic beta diversity

The continental analyses described in Chapter 3 demonstrated that remotely sensed primary productivity was effectively unrelated to taxonomic β-diversity of vascular plants in the Australian context (Figure 6.1). Furthermore, adding β-diversity to simple environmental models of primary productivity contributed no meaningful improvement in model fit or explanatory power. Recent theoretical and experimental studies have argued that the loss of β-diversity could increase the variability and decrease the magnitude of ecosystem functions at regional scales (Pasari et al., 2013;

Wang & Loreau, 2014; Wang & Loreau, 2016). Similarly, a recent empirical analysis of forest plots demonstrated that the magnitudes of multiple combined ecosystem functions — measured from boreal to Mediterranean ecosystems — had consistently positive relationships with the taxonomic

β-diversity between plots (Van Der Plas et al., 2016). Here I reiterate that my analyses have only one response variable, GPP, rather than several as in many B-EF studies [e.g. Pasari et al. (2013);

Van Der Plas et al. (2016)].

132

Chapter 6 Discussion

Nonetheless, contrary to these empirical findings, and counter to the expectations of Pasari et al. (2013) and Turnbull et al. (2016), this research has demonstrated that considering more realistic, broad-scale environmental and biological gradients in real ecosystems did not increase the strength of relationships between primary productivity and taxonomic floristic β-diversity.

Figure 6.1. Summary of the key results for Chapter 3. Example of a random subsample of cells or “communities” (a), generated with the same methods as that described in Chapter 3, section 3.2 (approximately 1000 cells). These subsampled cells must contain more than 10 species, have more than 10 locations within a 10km radius, and a reasonable probability that α-diversity was well sampled (see section 3.2.2). Plots of GPP vs. β-diversity for one the continental generalised additive models from Chapter 3 (b), showing raw observations for one random subsample (blue points, n = 1000 across continental Australia). GPP is the mean monthly photosynthetic flux of carbon (gC m-2 month-1, January 2001-December 2012, see section 3.2.1), GPP CV is the ratio of the standard deviation in GPP to the mean.

133

Chapter 6 Discussion

The weak relationships observed in Chapter 3 may be based on the logic that the macroecological complementarity mechanism only manifests in locations with a low degree of ecological specialisation. If this principle holds, the taxonomic turnover between most of the grid cells used in Chapter 3 (e.g. Figure 6.1) would have been high enough to negate any effects of changes in β-diversity on productivity. It is possible that stronger relationships between productivity and β-diversity could arise at finer spatial resolutions than 1 km × 1 km, at scales where increasing biological turnover may not be functionally redundant (Poorter et al., 2015). However, as outlined in Chapters 1 and 3, biological turnover in geographic space is only an indirect proxy of ecological specialisation (Burley et al., 2016b) — and based on the evidence in Chapter 3, a fairly weak proxy at that.

These continental results may also have implications for predictions that biological homogenization may decrease ecosystem functioning in natural systems across broad scales, as recently argued by Van Der Plas et al. (2016). If biological ‘saturation’ occurs at macroecological scales across real ecosystems — whereby additional biological richness and heterogeneity are redundant — the significant relationships found in experimental analyses are unlikely to be borne out at macroecological scales. This logic may find anecdotal support in global bioregions were high levels of biological heterogeneity are associated with relatively low levels of productivity, among other ecosystem functions (for example the globally significant hotspot of plant diversity in southwestern Australia). Ultimately, it is reasonable to conclude from Chapter 3 that macro-scale measures of both productivity and taxonomic diversity may be better suited as response variables

(on the y-axes), responding to many of the same environmental conditions — as suggested by

Lavers and Field (2006).

From a methodological perspective, the effect of the concurrent declines in rainfall, productivity, α-diversity and sampling intensity with increasing distance from the Australian

134

Chapter 6 Discussion coastline on these analyses must also be considered. Indeed the reliance on observational plant records in Chapter 3 has undoubtedly biased the results towards the wettest, most productive and most species rich areas (see Appendix 2, Figure 6.1). Although a seemingly mundane point, it is worth emphasising the need for more high quality plot data to be measured in the greatest possible range of ecosystems, both across Australia and around the world. Questions about broad-scale interrelationships between ecosystem functions, environmental conditions and measures of biological diversity and ecological specialisation cannot be answered decisively without more reliable, widespread knowledge about which plants occur in particular ecosystems. Unsurprisingly, this fundamental information is still lacking for the more remote regions of many continents such as

Australia, making it difficult to evaluate the practical relevance of B-EF arguments for continental- scale conservation and management initiatives.

6.2 Continental variation in species-level niche width

As noted in Chapters 2 and 3, the range of environmental conditions across which a plant species is known to occur — its realised environmental niche width — may approximate how physiologically and ecologically specialised that species has become to current conditions. Chapter

4 tested the hypothesis that the realised environmental niche width of vascular plants was positively related to the niche median for 1771 plant species known to occur in the Wet Tropics. This analysis demonstrated that across all 1771 species, temperature niche width was weakly related to niche median (Figure 6.2 c). In contrast, rainfall and nitrogen niches were wider in wetter and more fertile areas, forming a positive curvilinear relationship between niche width and median (Figure 6.2 b. d).

The lack of systematic variation in temperature niche width suggests that, although temperature is clearly important to plant physiology, its marginal effect is relatively constant across

135

Chapter 6 Discussion the continental gradient. This interpretation is consistent with the macroecological evidence that across Australia, temperature-related variables are generally less limiting for plant growth and reproduction than water-related variables [e.g. Donohue et al. (2014)]. Thus, it is plausible that a range of realised temperature niche widths are equally likely to occur at any point along the continental temperature gradient (Figure 6.2 c). Conversely, the systematic widening of realised rainfall and nitrogen niches across environmental space implies that available moisture and fertility were important for plants, and their effects on physiology were variable across continental gradients. Such an increasing relationship between niche width and median for both moisture and fertility seems logical in the Australian context (Figure 6.2 b, d). Water and nutrient availability are likely to gradually shift from being limiting to non-limiting factors affecting plant growth and reproduction as plants occupy wetter and more fertile environments.

Figure 6.2. Summary of the key results for Chapter 4. The records of 1771 vascular plants known to occur in the Wet Tropics (a) used to analyse species environmental niches described in Chapter 4, section 4.2 (366,111 occurrences of 1771 species, plotted in blue). Plots of vascular plant species niche width vs. niche median (n = 1771) for the natural log of precipitation (b, originally mm), maximum temperature (c, °C), and the natural log of soil nitrogen (d, originally %). The continental niche width is defined as the middle 90% of environmental values for all records of each species across mainland Australia and Tasmania, and niche position is the median of environmental values for all records of each species. For each environmental gradient, the regression model is plotted in orange over species niches (blue points), with the explanatory power for each model shown in orange (deviance explained values for rainfall and nitrogen, and adjusted R2 values for temperature).

136

Chapter 6 Discussion

These divergent patterns of variation in realised environmental niches necessarily reflect the biogeographic history of the diverse taxa that comprise the Wet Tropics flora. Interestingly, the subset of species in Chapter 4 which are known to originate in either Australia or South East Asia

(457 out of the total 1771 species studied) have similar niche widths (see Appendix 3, Figure S3.7).

Given that the biogeographic attribution of these taxa is an ongoing project, further analyses are required to test if the relationships between niches and functional traits differ based on how long each taxa has been persisting in Australian ecosystems.

The systematic widening of species realised niches in rainfall space, driven by biogeographic history, also has implications for testing community-level relationships between ecosystem functions and realised niche width discussed in Chapters 2, 3 and 5. Because species with wider rainfall niches primarily occur in wetter areas for the Wet Tropics flora — rather than a mix of niche widths occurring across the gradient, as for temperature niches — community level measures of niche width will also be wider in wetter areas. If the median realised niche width of all species comprising a community approximates the degree of ecological specialisation at that location, as argued in Chapters 2 and 3, direct relationships between community ecosystem function and rainfall niche width will then be more difficult to isolate from environmental effects than for the temperature dimension. Thus tests of relationships between ecosystem functions and niche width using these taxa should carefully control for variations in realised community niche width across environmental space. Importantly, these variations in realised niches would be lost if the ecological specialisation of each species was defined solely using geographic patterns of co-occurrence [e.g.

Boulangeat et al. (2012); Vimal and Devictor (2015)], potentially leading to confounding of the relationships between ecosystem functions, ecological specialisation and environmental conditions.

137

Chapter 6 Discussion

6.3 Bioregional primary productivity and community-level niche width

The macroecological complementarity hypothesis states that locations with a greater degree of ecological specialisation should display higher magnitudes of ecosystem functions than locations with less specialisation. The analyses presented in Chapter 5 are the first empirical test of this hypothesis using measures of ecosystem function and ecological specialisation at broad spatio- temporal and biological scales in real ecosystems. Ecological specialisation was quantified as the median of the rainfall and temperature niche widths — across continental Australia — of all tree species occurring at 510 Wet Topics sites. These community level analyses demonstrated that site niche width can influence ecosystem functions at macroecological scales, even when removing the direct effect of site environment on primary productivity. Wet Tropics sites with more ecological specialists in temperature space had higher primary productivity, whereas sites with more ecological specialists in rainfall space had lower productivity (Figure 6.3).

138

Chapter 6 Discussion

Figure 6.3. Summary of the key results for Chapter 5. Scatterplots of site-level relationships (a) between gross primary productivity (GPP, gC m-2 month-1), rainfall niche width (RNW) and temperature niche width (TNW) for the 510 Wet Tropics sites used to generate pairwise variables. Bottom panels: plots of linear models for 500 subsamples of environmentally similar, unique site-pairs calculated from the drier 255 sites (42-138 mm of monthly rainfall), where pairwise differences in GPP are a function of pairwise differences in median site temperature niche width (b, °C) and median site rainfall niche width (c, mm). The slopes of all 500 linear models are plotted in grey, and their mean slope is plotted in orange. The allowable difference in predicted GPP between site-pairs, ΔGPP = 20, is based on a negative exponential model of site GPP as a function of rainfall. The allowable temperature difference between site-pairs is fixed at ΔT = 0.5°C.

139

Chapter 6 Discussion

These contrasting relationships between productivity and specialisation in temperature and rainfall space could be explained by the evolution of environmental niches in response to climatic variability, both in the Wet Tropics, and across Australia. The results in Chapter 5 demonstrated that Wet Tropics sites with wider median rainfall niches tended to also have narrower median temperature niches (Figure 6.3 a). Moreover, the mean rainfall variability at the Wet Tropics sites is greater than mean temperature variability across 2001-2012 (see Appendix 4, Fig S4.1). Hence the evolution of physiological specialisation to particular levels of moisture availability could be negated by a more generalist ecological strategy, whereby tree species are successful in areas where they can continue to grow, despite variable rainfall. This is a similar rationale to that advanced in

Chapter 4 to explain why the rainfall niches of individual species are widest in the wettest environments — to tolerate the high inter-annual variability in rainfall that accompanied the onset of monsoonal conditions in tropical Australia. Ultimately, the community-level results in chapter 5 suggest that wide rainfall niches are more physiologically optimal in the Australian Wet Tropics than narrow rainfall niches, contradicting the logic of the macroecological complementarity hypothesis described in Chapters 2, 3 and 5.

The results for Chapter 5 highlight the importance of analysing bivariate relationships between individual ecosystem functions and ecological specialisation using the simplest possible methods. This point is especially relevant given the recent emphasis in the B-EF literature on the use of ‘multi-functionality’ indexes as response variables. These indexes combine multiple ecosystem functions into a single value that can be explained by various measures of biological diversity [e.g. Pasari et al. (2013); Jing et al. (2015); Soliveres et al. (2016); Thompson and

Gonzalez (2016); Van Der Plas et al. (2016)]. Such methods could have masked the contrasting bivariate results for rainfall and temperature niches observed in Chapter 5, potentially obscuring biogeographic and evolutionary interpretations of these patterns. Similarly, the thus far equivocal results of macroecological B-EF analyses [e.g. Chisholm et al. (2013); Burley et al. (2016a);

140

Chapter 6 Discussion

Spasojevic et al. (2016)] caution against simply generating measures of ecosystem functions and ecological specialisation until the hypotheses in question are accepted. If the empirical relationships between ecosystem functions and specialisation are weak and idiosyncratic in natural ecosystems at macroecological scales, it is questionable how these relationships could be managed for across the broad spatio-temporal scales relevant to global conservation initiatives. As discussed in Chapter 2, section 2.3, the relationships between any particular measure of ecosystem function and ecological specialisation are just as likely to be negative as they are to be positive in natural ecosystems. Thus arguments that biodiversity must be conserved in order to maintain current and future levels of ecosystem functions may present a double edged sword, at least when tested at broad scales without experimental controls.

6.4 Future implications

The contrasting results found at the continental (Chapter 3) and bioregional scale (Chapter

5) do not conclusively refute or confirm the macroecological complementarity hypothesis. As noted in section 6.1, the continental analyses used β-diversity across geographic space to approximate ecological specialisation (calculated with observational plant records, which are biased towards the best sampled areas, see Figure 6.1). Alternatively, the bioregional analyses used site-level niche width to more directly approximate ecological specialisation — a kind of environmental β-diversity

— but across a smaller biogeographic range than the continental analysis, constrained by the available ecological plots with similar environments inside the Wet Tropics. Thus the results of chapters 3 and 5 are not directly comparable. Yet despite these methodological differences, the uncertainty surrounding the continental and bioregional patterns still opens the door to further analyses.

141

Chapter 6 Discussion

Importantly, realised niche width at either the species or community level is still only an indirect proxy of both physiological performance and ecological specialisation, because it does not differentiate between specialisation formed through adaptive, rather than stochastic evolutionary processes (Chase, 2010; Chase & Myers, 2011). Unfortunately it is impractical to expect such differentiation in spatially explicit datasets across macroecological scales. Nonetheless, further analyses could appreciably improve the proxy of specialisation by testing whether species’ realised niches are related to their physiological performance under current and future climates. It seems plausible that some ecological generalists — for example highly productive invasive plants (Levine

& D'Antonio, 1999; Srivastava & Vellend, 2005) — could outperform specialists under certain conditions. The key point here is that empirical tests of B-EF theories often ignore circumstances where relationships between ecosystem functions and biodiversity could be either neutral, or indeed negative.

Recent decades have seen a proliferation in the creation of phylogenies for an increasing proportion of global taxa, including vascular plants [e.g. Mishler et al. (2014); Nagalingum et al.

(2015); Rosauer et al. (2015); Thornhill et al. (2015); Chase et al. (2016)]. Consequently, phylogenies are now incorporated into many B-EF analyses to help explain variation in ecosystem functions [e.g. Maherali and Klironomos (2007); Connolly et al. (2011); Cadotte (2013, 2015b);

Thompson et al. (2015)]. Thus there is considerable scope for using measures such as phylogenetic turnover and endemism to approximate ecological specialisation [e.g. Laffan et al. (2016);

Thornhill et al. (2016)]. However, a more biogeographically informative avenue could be to quantify the environmental niche widths of particular genes across continental scales. This could be achieved by combining plot data with observational records and species-level phylogenetic trees for the flora of diverse regions such as the Wet Tropics. Such analyses could potentially incorporate the influence of biogeographic history on ecosystem functions, which up to this point has largely been omitted from macroecological B-EF studies.

142

Chapter 6 Discussion

This thesis has only briefly touched on the hypothesis that ecological specialisation could affect future ecosystem functions — the macroecological spatio-temporal compensation mechanism (outlined in Chapter 2, section 2.3). The results from Chapter 5 indicate that tests of this hypothesis could begin by quantifying relationships between the variability of ecosystem functions, the variability of environmental conditions and different measures of ecological specialisation under current conditions (including phylogenetic and functional measures of specialisation), and projecting these relationships under future climate scenarios. Furthermore, directional environmental changes such as land clearing — which predominantly affect the rarest taxa, and are linked to socioeconomic inequities within and between countries — should be incorporated into these models. Ultimately, determining whether certain ecosystem processes will be negatively or positively affected by ecological specialisation may be less useful than improving our understanding of how changes in community composition — for example spatio-temporal turnover in functional traits — influence ecosystem functions.

6.5 Conclusions

This research illustrates that the nature of macroecological relationships between ecosystem functions and ecological specialisation depends on how the variables are quantified. Furthermore, it highlights how little is empirically known about fundamental ecological relationships at broad scales. Using taxonomic floristic turnover in geographic space to approximate ecological specialisation produced a neutral relationship with productivity across continental Australia

(Chapter 3). By contrast, using community-level niche width to approximate ecological specialisation in the Australian Wet Tropics produced both negative and positive relationships with productivity for temperature and rainfall niches, respectively (Chapter 5). These contrasting relationships between productivity and specialisation at the bioregional scale likely reflect the

143

Chapter 6 Discussion evolution of species environmental niches in response to climatic variability across continental

Australia.

A final point should be made about the ethical dimension of investigating relationships between ecosystem functions and biological diversity, however these concepts are defined, quantified and analysed. Such analyses may produce ambiguous results, and there may be no economic or otherwise utilitarian benefits to conserving biodiversity by any reasonable quantitative measure. Moreover, even if strong links between ecosystem functions and diversity can be demonstrated, they are unlikely to convince bureaucrats or politicians to increase conservation efforts, particularly where economic objectives entail environmental transformation and degradation. But regardless of the evidence for relationships between ecosystem functions and biological diversity, compelling philosophical, ethical and moral arguments can be made that humans have no right to extinguish even small components of the marvellous natural world of which we are all part, and to which we are all heirs.

144

References

Adler, P.B., Seabloom, E.W., Borer, E.T., Hillebrand, H., Hautier, Y., Hector, A., et al. (2011) Productivity is a poor predictor of plant species richness. Science, 333, 1750-1753. Austin, M.P., Nicholls, A.O. & Margules, C.R. (1990) Measurement of the realized qualitative niche: Environmental niches of five Eucalyptus species. Ecological Monographs, 60, 161- 177. Austin, M.P., Smith, T.M., Van Niel, K.P. & Wellington, A.B. (2009) Physiological responses and statistical models of the environmental niche: A comparative study of two co-occurring Eucalyptus species. Journal of Ecology, 97, 496-507. Baltzer, J.L., Davies, S.J., Noor, N.S.M., Kassim, A.R. & Lafrankie, J.V. (2007) Geographical distributions in tropical trees: Can geographical range predict performance and habitat association in co-occurring tree species? Journal of Biogeography, 34, 1916-1926. Bar-Massada, A. (2015) Complex relationships between species niches and environmental heterogeneity affect species co-occurrence patterns in modelled and real communities. Proceedings of the Royal Society of London B: Biological Sciences, 282, 20150927. Barlow, B.A. & Hyland, B.P.M. (1988) The origins of the flora of Australia's wet tropics. The ecology of Australia's wet tropics, 15, 1-17. Bartholomeus, R.P., Witte, J.P.M., Van Bodegom, P.M., Van Dam, J.C. & Aerts, R. (2011) Climate change threatens endangered plant species by stronger and interacting water-related stresses. Journal of Geophysical Research: Biogeosciences, 116, G04023. Barton, P.S., Cunningham, S.A., Manning, A.D., Gibb, H., Lindenmayer, D.B. & Didham, R.K. (2013) The spatial scaling of beta diversity. Global Ecology and Biogeography, 22, 639- 647. Barton, P.S., Lentini, P.E., Alacs, E., Bau, S., Buckley, Y.M., Burns, E.L., et al. (2015) Guidelines for Using Movement Science to Inform Biodiversity Policy. Environmental Management, 56, 791-801. Baselga, A. (2010) Partitioning the turnover and nestedness components of beta diversity. Global Ecology and Biogeography, 19, 134-143. Baselga, A. (2013) Multiple site dissimilarity quantifies compositional heterogeneity among several sites, while average pairwise dissimilarity may be misleading. Ecography, 36, 124-128. Bengtsson, J. (1998) Which species? What kind of diversity? Which ecosystem function? Some problems in studies of relations between biodiversity and ecosystem function. Applied Soil Ecology, 10, 191-199. Blonder, B., Lamanna, C., Violle, C. & Enquist, B.J. (2014) The n-dimensional hypervolume. Global Ecology and Biogeography, 23, 595-609. Boucher-Lalonde, V. & Currie, D.J. (2016) Spatial autocorrelation can generate stronger correlations between range size and climatic niches than the biological signal - A demonstration using bird and mammal range maps. PLoS ONE, 11, e0166243. Boulangeat, I., Lavergne, S., Van Es, J., Garraud, L. & Thuiller, W. (2012) Niche breadth, rarity and ecological characteristics within a regional flora spanning large environmental gradients. Journal of Biogeography, 39, 204-214. Bowman, D.M.J.S. (2000) Australian Rainforests: Islands of Green in a Land of Fire. Cambridge University Press, Cambridge.

145

Bradford, M.A., Wood, S.A., Bardgett, R.D., Black, H.I.J., Bonkowski, M., Eggers, T., et al. (2014) Discontinuity in the responses of ecosystem processes and multifunctionality to altered soil community composition. Proceedings of the National Academy of Sciences of the United States of America, 111, 14478-14483. Broennimann, O., Fitzpatrick, M.C., Pearman, P.B., Petitpierre, B., Pellissier, L., Yoccoz, N.G., et al. (2012) Measuring ecological niche overlap from occurrence and spatial environmental data. Global Ecology and Biogeography, 21, 481-497. Brunsdon, C., Fotheringham, S. & Charlton, M. (1998) Geographically weighted regression - Modelling spatial non-stationarity. Journal of the Royal Statistical Society Series D: The Statistician, 47, 431-443. Burley, H.M., Laffan, S.W. & Williams, K.J. (2012) Spatial non-stationarity and anisotropy of compositional turnover in eastern Australian Myrtaceae species. International Journal of Geographical Information Science, 26, 2065-2081. Burley, H.M., Mokany, K., Ferrier, S., Laffan, S.W., Williams, K.J. & Harwood, T.D. (2016a) Primary productivity is weakly related to floristic alpha and beta diversity across Australia. Global Ecology and Biogeography, 25, 1294-1307. Burley, H.M., Mokany, K., Ferrier, S., Laffan, S.W., Williams, K.J. & Harwood, T.D. (2016b) Macroecological scale effects of biodiversity on ecosystem functions under environmental change. Ecology and Evolution, 6, 2579-93. Bush, A., Harwood, T., Hoskins, A.J., Mokany, K. & Ferrier, S. (2016) Current Uses of Beta- Diversity in Biodiversity Conservation: A response to Socolar et al. Trends in Ecology and Evolution, 31, 337-338. Byrne, M., Steane, D.A., Joseph, L., Yeates, D.K., Jordan, G.J., Crayn, D., et al. (2011) Decline of a biome: Evolution, contraction, fragmentation, extinction and invasion of the Australian mesic zone biota. Journal of Biogeography, 38, 1635-1656. Cabral, J.S. & Kreft, H. (2012) Linking ecological niche, community ecology and biogeography: Insights from a mechanistic niche model. Journal of Biogeography, 39, 2212-2224. Cabrelli, A., Beaumont, L. & Hughes, L. (2014) The impacts of climate change on Australian and New Zealand flora and fauna. Austral Ark: The State of Wildlife in Australia and New Zealand, pp. 65-82. Cambridge University Press, Cambridge. Cadotte, M.W. (2013) Experimental evidence that evolutionarily diverse assemblages result in higher productivity. Proceedings of the National Academy of Sciences of the United States of America, 110, 8996-9000. Cadotte, M.W. (2015a) Phylogenetic diversity and productivity: Gauging interpretations from experiments that do not manipulate phylogenetic diversity. Functional Ecology, 29, 1603- 1606. Cadotte, M.W. (2015b) Phylogenetic diversity-ecosystem function relationships are insensitive to phylogenetic edge lengths. Functional Ecology, 29, 718-723. Cadotte, M.W., Carscadden, K. & Mirotchnick, N. (2011) Beyond species: Functional diversity and the maintenance of ecological processes and services. Journal of Applied Ecology, 48, 1079- 1087. Cardinale, B.J., Duffy, J.E., Gonzalez, A., Hooper, D.U., Perrings, C., Venail, P., et al. (2012) Biodiversity loss and its impact on humanity. Nature, 486, 59-67. Ceballos, G., Ehrlich, P.R., Barnosky, A.D., García, A., Pringle, R.M. & Palmer, T.M. (2015) Accelerated modern human–induced species losses: Entering the sixth mass extinction. Science Advances, 1, e1400253. Chase, J.M. (2010) Stochastic community assembly causes higher biodiversity in more productive environments. Science, 328, 1388-1391. Chase, J.M. & Myers, J.A. (2011) Disentangling the importance of ecological niches from stochastic processes across scales. Philosophical Transactions of the Royal Society B: Biological Sciences, 366, 2351-2363.

146

Chase, M.W., Christenhusz, M.J.M., Fay, M.F., Byng, J.W., Judd, W.S., Soltis, D.E., et al. (2016) An update of the Angiosperm Phylogeny Group classification for the orders and families of flowering plants: APG IV. Botanical Journal of the Linnean Society, 181, 1-20. Chiang, J.M., Spasojevic, M.J., Muller-Landau, H.C., Sun, I.F., Lin, Y., Su, S.H., et al. (2016) Functional composition drives ecosystem function through multiple mechanisms in a broadleaved subtropical forest. Oecologia, 182, 829-840. Chisholm, R.A., Muller-Landau, H.C., Abdul Rahman, K., Bebber, D.P., Bin, Y., Bohlman, S.A., et al. (2013) Scale-dependent relationships between tree species richness and ecosystem function in forests. Journal of Ecology, 101, 1214-1224. Connolly, J., Cadotte, M.W., Brophy, C., ÁIne, D., Finn, J., Kirwan, L., et al. (2011) Phylogenetically diverse grasslands are associated with pairwise interspecific processes that increase biomass. Ecology, 92, 1385-1392. Corlett, R.T. & Westcott, D.A. (2013) Will plant movements keep up with climate change? Trends in Ecology and Evolution, 28, 482-488. Costion, C.M., Edwards, W., Ford, A.J., Metcalfe, D.J., Cross, H.B., Harrington, M.G., et al. (2015) Using phylogenetic diversity to identify ancient rain forest refugia and diversification zones in a biodiversity hotspot. Diversity and Distributions, 21, 279-289. Crayn, D.M., Costion, C. & Harrington, M.G. (2015) The Sahul-Sunda floristic exchange: Dated molecular phylogenies document Cenozoic intercontinental dispersal dynamics. Journal of Biogeography, 42, 11-24. Crisp, Laffan, Linder & Monro (2001) Endemism in the Australian flora. Journal of Biogeography, 28, 183-198. Crisp, M.D. & Cook, L.G. (2013) How was the Australian flora assembled over the last 65 million years? A molecular phylogenetic perspective. Annual Review of Ecology, Evolution, and Systematics, 44, 303-324. Crisp, M.D., Burrows, G.E., Cook, L.G., Thornhill, A.H. & Bowman, D.M.J.S. (2011) Flammable biomes dominated by eucalypts originated at the Cretaceous-Palaeogene boundary. Nature Communications, 2 Darwin, C. (1859) On the origin of species. Oxford University Press, New York. De Keersmaecker, W., Lhermitte, S., Honnay, O., Farifteh, J., Somers, B. & Coppin, P. (2014) How to measure ecosystem stability? An evaluation of the reliability of stability metrics based on remote sensing time series across the major global ecosystems. Global Change Biology, 20, 2149-2161. Devictor, V., Clavel, J., Julliard, R., Lavergne, S., Mouillot, D., Thuiller, W., et al. (2010) Defining and measuring ecological specialization. Journal of Applied Ecology, 47, 15-25. Donohue, R.J., Hume, I.H., Roderick, M.L., McVicar, T.R., Beringer, J., Hutley, L.B., et al. (2014) Evaluation of the remote-sensing-based DIFFUSE model for estimating photosynthesis of vegetation. Remote Sensing of Environment, 155, 349-365. Drake, J.M. & Lodge, D.M. (2004) Effects of environmental variation on extinction and establishment. Ecology Letters, 7, 26-30. Dunstan, P.K., Foster, S.D., Hui, F.K.C. & Warton, D.I. (2013) Finite Mixture of Regression Modeling for High-Dimensional Count and Biomass Data in Ecology. Journal of Agricultural, Biological, and Environmental Statistics, 18, 357-375. Ebdon, D. (1985) Statistics in Geography: A Practical Approach - Revised with 17 Programs. Wiley, New York. England, M.H., Kajtar, J.B. & Maher, N. (2015) Robust warming projections despite the recent hiatus. Nature Climate Change, 5, 394-396. ESRI (2015) ArcGIS Desktop: Release 10.2.2. Environmental Systems Research Institute, Redlands, California. Etterson, J.R. & Shaw, R.G. (2001) Constraint to adaptive evolution in response to global warming. Science, 294, 151-154.

147

Faith, D.P. (2008) Biodiversity. In: The Stanford Encyclopedia of Philosophy (ed. E.N. Zalta). The Metaphysics Research Lab Stanford. Faith, D.P. (2016) A general model for biodiversity and its value. The Routledge Handbook of Philosophy of Biodiversity. Routledge, New York. Faith, D.P., Magallón, S., Hendry, A.P., Conti, E., Yahara, T. & Donoghue, M.J. (2010) Evosystem services: An evolutionary perspective on the links between biodiversity and human well- being. Current Opinion in Environmental Sustainability, 2, 66-74. Fang, Y., Sun, G., Caldwell, P., McNulty, S.G., Noormets, A., Domec, J.-C., et al. (2016) Monthly land cover-specific evapotranspiration models derived from global eddy flux measurements and remote sensing data. Ecohydrology, 9, 248-266. Feeley, K.J. (2015) Moving forward with species distributions. American Journal of Botany, 102, 173-175. Feeley, K.J. & Silman, M.R. (2010) Biotic attrition from tropical forests correcting for truncated temperature niches. Global Change Biology, 16, 1830-1836. Feeley, K.J. & Silman, M.R. (2011) The data void in modeling current and future distributions of tropical species. Global Change Biology, 17, 626-630. Feeley, K.J. & Rehm, E.M. (2012) Amazon's vulnerability to climate change heightened by deforestation and man-made dispersal barriers. Global Change Biology, 18, 3606-3614. Fernandez-Going, B.M., Harrison, S.P., Anacker, B.L. & Safford, H.D. (2013) Climate interacts with soil to produce beta diversity in Californian plant communities. Ecology, 94, 2007- 2018. Ferrier, S., Manion, G., Elith, J. & Richardson, K. (2007) Using generalized dissimilarity modelling to analyse and predict patterns of beta diversity in regional biodiversity assessment. Diversity and Distributions, 13, 252-264. Flores-Moreno, H. & Moles, A.T. (2013) A Comparison of the Recruitment Success of Introduced and Native Species Under Natural Conditions. PLoS ONE, 8, e72509. Fonseca, C.R., Overton, J.M., Collins, B. & Westoby, M. (2000) Shifts in trait-combinations along rainfall and phosphorus gradients. Journal of Ecology, 88, 964-977. Fort, H. & Inchausti, P. (2013) Tropical forests are non-equilibrium ecosystems governed by interspecific competition based on universal 1/6 niche width. PLoS ONE, 8 Fortin, M.J. & Dale, M.R.T. (2005) Spatial Analysis: A Guide for Ecologists. Cambridge University Press, Cambridge. Fotheringham, A.S., Crespo, R. & Yao, J. (2015) Exploring, modelling and predicting spatiotemporal variations in house prices. Annals of Regional Science, 54, 417-436. Fox, B.J. (1981) Niche parameters and species richness. Ecology, 62, 1415-1425. Fraser, L.H., Pither, J., Jentsch, A., Sternberg, M., Zobel, M., Askarizadeh, D., et al. (2015) Worldwide evidence of a unimodal relationship between productivity and plant species richness. Science, 349, 302-305. Freckleton, R.P., Harvey, P.H. & Pagel, M. (2002) Phylogenetic analysis and comparative data: A test and review of evidence. American Naturalist, 160, 712-726. Fridley, J.D., Vandermast, D.B., Kuppinger, D.M., Manthey, M. & Peet, R.K. (2007) Co- occurrence based assessment of habitat generalists and specialists: A new approach for the measurement of niche width. Journal of Ecology, 95, 707-722. Gaitán, J.J., Bran, D., Oliva, G., Maestre, F.T., Aguiar, M.R., Jobbagy, E., et al. (2014) Plant species richness and shrub cover attenuate drought effects on ecosystem functioning across Patagonian rangelands. Biology Letters, 10, 20140673. Gallagher, R.V. (2016) Correlates of range size variation in the Australian seed-plant flora. Journal of Biogeography, 43, 1287-1298. Gallant, A.J., Hennessy, K.J. & Risbey, J. (2007) Trends in rainfall indices for six Australian regions: 1910-2005. Australian Meteorological Magazine, 56, 223-241.

148

Gallant, A.J.E., Reeder, M.J., Risbey, J.S. & Hennessy, K.J. (2013) The characteristics of seasonal- scale droughts in Australia, 1911-2009. International Journal of Climatology, 33, 1658- 1672. Garcia, R.A., Cabeza, M., Rahbek, C. & Araújo, M.B. (2014) Multiple dimensions of climate change and their implications for biodiversity. Science, 344, 1247579-1 - 1247579-10. Gaston, K.J. & Fuller, R.A. (2009) The sizes of species' geographic ranges. Journal of Applied Ecology, 46, 1-9. Gauch, H.G., Jr. & Whittaker, R.H. (1972) Coenocline Simulation. Ecology, 53, 446-451. Gauch, H.G., Jr. & Whittaker, R.H. (1976) Simulation of community patterns. Vegetatio, 33, 13-16. Geange, S.W., Pledger, S., Burns, K.C. & Shima, J.S. (2011) A unified analysis of niche overlap incorporating data of different types. Methods in Ecology and Evolution, 2, 175-184. Ghilarov, A.M. (2000) Ecosystem functioning and intrinsic value of biodiversity. Oikos, 90, 408- 412. González-Orozco, C.E., Laffan, S.W. & Miller, J.T. (2011) Spatial distribution of species richness and endemism of the genus Acacia in Australia. Australian Journal of Botany, 59, 600-608. Gouveia, S.F., Hortal, J., Cassemiro, F.A.S., Rangel, T.F. & Diniz-Filho, J.A.F. (2013) Nonstationary effects of productivity, seasonality, and historical climate changes on global amphibian diversity. Ecography, 36, 104-113. Grace, J.B. (2006) Structural Equation Modeling and Natural Systems. Cambridge University Press, New York. Grace, J.B., Adler, P.B., Stanley Harpole, W., Borer, E.T. & Seabloom, E.W. (2014) Causal networks clarify productivity-richness interrelations, bivariate plots do not. Functional Ecology, 28, 787-798. Grace, J.B., Anderson, T.M., Seabloom, E.W., Borer, E.T., Adler, P.B., Harpole, W.S., et al. (2016) Integrative modelling reveals mechanisms linking productivity and plant species richness. Nature, 529, 390-393. Grierson, P.F., Adams, M.A. & Attiwill, P.M. (1992) Estimates of carbon storage in the aboveground biomass of Victoria's forests. Australian Journal of Botany, 40, 631-640. Grime, J.P. (1973) Competitive exclusion in herbaceous vegetation. Nature, 242, 344-347. Grime, J.P. (1998) Benefits of plant diversity to ecosystems: Immediate, filter and founder effects. Journal of Ecology, 86, 902-910. Gross, K., Cardinale, B.J., Fox, J.W., Gonzalez, A., Loreau, M., Wayne Polley, H., et al. (2014) Species richness and the temporal stability of biomass production: A new analysis of recent biodiversity experiments. American Naturalist, 183, 1-12. Harfoot, M., Tittensor, D.P., Newbold, T., McInerny, G., Smith, M.J. & Scharlemann, J.P.W. (2014) Integrated assessment models for ecologists: The present and the future. Global Ecology and Biogeography, 23, 124-143. Harvey, P.H. & Pagel, M.D. (1998) The comparative method in evolutionary biology. Oxford University Press, New York. Haverd, V., Raupach, M.R., Briggs, P.R., Canadell, J.G., Davis, S.J., Law, R.M., et al. (2013) The Australian terrestrial carbon budget. Biogeosciences, 10, 851-869. Hawkins, B.A. (2012a) Eight (and a half) deadly sins of spatial analysis. Journal of Biogeography, 39, 1-9. Hawkins, B.A. (2012b) Are multiple regression models of spatially structured data to be trusted? Journal of Biogeography, 39, 998-998. Higgins, S.I., O'Hara, R.B., Bykova, O., Cramer, M.D., Chuine, I., Gerstner, E.M., et al. (2012) A physiological analogy of the niche for projecting the potential distribution of plants. Journal of Biogeography, 39, 2132-2145. Hilbert, D.W., Graham, A. & Hopkins, M.S. (2007) Glacial and interglacial refugia within a long- term rainforest refugium: The Wet Tropics Bioregion of NE Queensland, Australia. Palaeogeography, Palaeoclimatology, Palaeoecology, 251, 104-118.

149

Hooper, D.U., Chapin Iii, F.S., Ewel, J.J., Hector, A., Inchausti, P., Lavorel, S., et al. (2005) Effects of biodiversity on ecosystem functioning: A consensus of current knowledge. Ecological Monographs, 75, 3-35. Hopper, S.D. & Gioia, P. (2004) The Southwest Australian floristic region: Evolution and conservation of a global hot spot of biodiversity. In: Annual Review of Ecology, Evolution, and Systematics, pp. 623-650 Hughes, L. (2003) Climate change and Australia: Trends, projections and impacts. Austral Ecology, 28, 423-443. Huston, M.A., Aarssen, L.W., Austin, M.P., Cade, B.S., Fridley, J.D., Garnier, E., et al. (2000) No consistent effect of plant diversity on productivity. Science, 289, 1255. Hutchinson, M., Kesteven, J. & Xu, T. (2014) Monthly total precipitation: ANU Climate 1.0, 0.01 degree, Australian Coverage, 1970-2012. In. Australian National University, Canberra. Made available by the Ecosystem Modelling and Scaling Infrastructure (eMAST, http://www.emast.org.au) of the Terrestrial Ecosystem Research Network Canberra, Australia. Isbell, F., Tilman, D., Polasky, S. & Loreau, M. (2015a) The biodiversity-dependent ecosystem service debt. Ecology Letters, 18, 119-134. Isbell, F., Calcagno, V., Hector, A., Connolly, J., Harpole, W.S., Reich, P.B., et al. (2011) High plant diversity is needed to maintain ecosystem services. Nature, 477, 199-202. Isbell, F., Craven, D., Connolly, J., Loreau, M., Schmid, B., Beierkuhnlein, C., et al. (2015b) Biodiversity increases the resistance of ecosystem productivity to climate extremes. Nature, 526, 574-7. IUCN (1980) World Conservation Strategy: living resource conservation for sustainable development. International Union for Conservation of Nature and Natural Resources (IUCN). In, Gland, Switzerland. Jackson, S.T. & Sax, D.F. (2010) Balancing biodiversity in a changing environment: extinction debt, immigration credit and species turnover. Trends in Ecology and Evolution, 25, 153- 160. Jing, X., Sanders, N.J., Shi, Y., Chu, H., Classen, A.T., Zhao, K., et al. (2015) The links between ecosystem multifunctionality and above- and belowground biodiversity are mediated by climate. Nature Communications, 6, 8159. Jones, D.A., Wang, W. & Fawcett, R. (2009) High-quality spatial climate data-sets for Australia. Australian Meteorological and Oceanographic Journal, 58, 233-248. Jonsson, L.M., Nilsson, M.C., Wardle, D.A. & Zackrisson, O. (2001) Context dependent effects of ectomycorrhizal species richness on tree seedling productivity. Oikos, 93, 353-364. Kanniah, K.D., Beringer, J. & Hutley, L.B. (2013) Response of savanna gross primary productivity to interannual variability in rainfall: Results of a remote sensing based light use efficiency model. Progress in Physical Geography, 37, 642-663. Kattge, J., Díaz, S., Lavorel, S., Prentice, I.C., Leadley, P., Bönisch, G., et al. (2011) TRY - a global database of plant traits. Global Change Biology, 17, 2905-2935. Keith, H., Barrett, D. & Keenan, R. (2000) Review of Allometric Relationships for Estimating Woody Biomass for New South Wales, the Australian Capital Territory, Victoria, Tasmania and South Australia: National Carbon Accounting System Technical Report No. 5B. In. Australian Greenhouse Office, Canberra. Körner, C. & Cochrane, P.M. (1985) Stomatal responses and water relations of Eucalyptus pauciflora in summer along an elevational gradient. Oecologia, 66, 443-455. Kraft, N.J.B., Comita, L.S., Chase, J.M., Sanders, N.J., Swenson, N.G., Crist, T.O., et al. (2011) Disentangling the drivers of β diversity along latitudinal and elevational gradients. Science, 333, 1755-1758. Kujala, H., Moilanen, A., Araújo, M.B. & Cabeza, M. (2013) Conservation Planning with Uncertain Climate Change Projections. PLoS ONE, 8, e53315.

150

Laffan, S.W. & Crisp, M.D. (2003) Assessing endemism at multiple spatial scales, with an example from the Australian vascular flora. Journal of Biogeography, 30, 511-520. Laffan, S.W., Rosauer, D.F., Di Virgilio, G., Miller, J.T., González-Orozco, C.E., Knerr, N., et al. (2016) Range-weighted metrics of species and phylogenetic turnover can better resolve biogeographic transition zones. Methods in Ecology and Evolution, 7, 580-588. Laliberté, E., Wells, J.A., Declerck, F., Metcalfe, D.J., Catterall, C.P., Queiroz, C., et al. (2010) Land-use intensification reduces functional redundancy and response diversity in plant communities. Ecology Letters, 13, 76-86. Lamb, E.G., Mengersen, K.L., Stewart, K.J., Attanayake, U. & Siciliano, S.D. (2014) Spatially explicit structural equation modeling. Ecology, 95, 2434-2442. Lavers, C. & Field, R. (2006) A resource-based conceptual model of plant diversity that reassesses causality in the productivity-diversity relationship. Global Ecology and Biogeography, 15, 213-224. Lefcheck, J.S., Byrnes, J.E.K., Isbell, F., Gamfeldt, L., Griffin, J.N., Eisenhauer, N., et al. (2015) Biodiversity enhances ecosystem multifunctionality across trophic levels and habitats. Nature Communications, 6, 6936. Legendre, P. & De Cáceres, M. (2013) Beta diversity as the variance of community data: Dissimilarity coefficients and partitioning. Ecology Letters, 16, 951-963. Levine, J.M. & D'Antonio, C.M. (1999) Elton revisited: A review of evidence linking diversity and invasibility. Oikos, 87, 15-26. Liang, J., Crowther, T.W., Picard, N., Wiser, S., Zhou, M., Alberti, G., et al. (2016) Positive biodiversity-productivity relationship predominant in global forests. Science, 354 Lindenmayer, D.B., Likens, G.E., Andersen, A., Bowman, D., Bull, C.M., Burns, E., et al. (2012) Value of long-term ecological studies. Austral Ecology, 37, 745-757. Loreau, M. (1998) Biodiversity and ecosystem functioning: A mechanistic model. Proceedings of the National Academy of Sciences of the United States of America, 95, 5632-5636. Loreau, M. (2000) Biodiversity and ecosystem functioning: Recent theoretical advances. Oikos, 91, 3-17. Loreau, M. (2010) From Populations to Ecosystems: Theoretical Foundations for a New Ecological Synthesis (MPB-46). Princeton University Press, New Jersey. Loreau, M. & de Mazancourt, C. (2013) Biodiversity and ecosystem stability: A synthesis of underlying mechanisms. Ecology Letters, 16, 106-115. Loreau, M., Mouquet, N. & Gonzalez, A. (2003) Biodiversity as spatial insurance in heterogeneous landscapes. Proceedings of the National Academy of Sciences of the United States of America, 100, 12765-12770. Lovejoy (1980) Preface. Conservation biology: an evolutionary-ecological perspective (ed. by M.E. Soulé and B.A. Wilcox), p. 395. Sinauer Associates, Massachusetts. Mac Nally, R. (2000) Regression and model-building in conservation biology, biogeography and ecology: The distinction between - and reconciliation of - 'predictive' and 'explanatory' models. Biodiversity and Conservation, 9, 655-671. Maestre, F.T., Quero, J.L., Gotelli, N.J., Escudero, A., Ochoa, V., Delgado-Baquerizo, M., et al. (2012) Plant species richness and ecosystem multifunctionality in global drylands. Science, 335, 214-218. Maherali, H. & Klironomos, J.N. (2007) Influence of phylogeny on fungal community assembly and ecosystem functioning. Science, 316, 1746-1748. Manly, B.F.J. (2006) Randomization, Bootstrap and Monte Carlo Methods in Biology, Third Edition. Chapman and Hall, Boca Raton. Manthey, M., Fridley, J.D. & Peet, R.K. (2011) Niche expansion after competitor extinction? A comparative assessment of habitat generalists and specialists in the tree floras of southeastern North America and south-eastern Europe. Journal of Biogeography, 38, 840-853.

151

Matthiessen, B. & Hillebrand, H. (2006) Dispersal frequency affects local biomass production by controlling local diversity. Ecology Letters, 9, 652-662. McDonald, P.G., Fonseca, C.R., Overton, J.M. & Westoby, M. (2003) Leaf-size divergence along rainfall and soil-nutrient gradients: Is the method of size reduction common among clades? Functional Ecology, 17, 50-57. McKenzie, N., Jacquier, D., Isbell, R. & Brown, K. (2004) Australian Soils and Landscapes: An Illustrated Compendium. CSIRO PUBLISHING, Melbourne. Mellin, C., Mengersen, K., Bradshaw, C.J.A. & Caley, M.J. (2014) Generalizing the use of geographical weights in biodiversity modelling. Global Ecology and Biogeography, 23, 1314-1323. Metcalfe, D.J. & Ford, A.J. (2009) Floristics and Plant Biodiversity of the Rainforests of the Wet Tropics. Living in a Dynamic Tropical Forest Landscape (ed. by N.E. Stork), pp. 123-132. Blackwell, Melbourne. Meynard, C.N., Devictor, V., Mouillot, D., Thuiller, W., Jiguet, F. & Mouquet, N. (2011) Beyond taxonomic diversity patterns: How do α, β and γ components of bird functional and phylogenetic diversity respond to environmental gradients across France? Global Ecology and Biogeography, 20, 893-903. Meynard, C.N., Lavergne, S., Boulangeat, I., Garraud, L., Van Es, J., Mouquet, N., et al. (2013) Disentangling the drivers of metacommunity structure across spatial scales. Journal of Biogeography, 40, 1560-1571. Miller, J., Franklin, J. & Aspinall, R. (2007) Incorporating spatial dependence in predictive vegetation models. Ecological Modelling, 202, 225-242. Milly, P.C.D., Dunne, K.A. & Vecchia, A.V. (2005) Global pattern of trends in streamflow and water availability in a changing climate. Nature, 438, 347-350. Mishler, B.D., Knerr, N., González-Orozco, C.E., Thornhill, A.H., Laffan, S.W. & Miller, J.T. (2014) Phylogenetic measures of biodiversity and neo-and paleo-endemism in Australian acacia. Nature Communications, 5, 4473. Mittelbach, G.G., Steiner, C.F., Scheiner, S.M., Gross, K.L., Reynolds, H.L., Waide, R.B., et al. (2001) What is the observed relationship between species richness and productivity? Ecology, 82, 2381-2396. Mokany, K., Harwood, T.D. & Ferrier, S. (2013a) Comparing habitat configuration strategies for retaining biodiversity under climate change. Journal of Applied Ecology, 50, 519-527. Mokany, K., Burley, H.M. & Paini, D.R. (2013b) beta diversity contributes to ecosystem processes more than by simply summing the parts. Proceedings of the National Academy of Sciences of the United States of America, 110, E4057. Mokany, K., Harwood, T.D., Williams, K.J. & Ferrier, S. (2012) Dynamic macroecology and the future for biodiversity. Global Change Biology, 18, 3149-3159. Mokany, K., Westcott, D.A., Prasad, S., Ford, A.J. & Metcalfe, D.J. (2014) Identifying priority areas for conservation and management in diverse tropical forests. PLoS ONE, 9, e89084. Mokany, K., Ferrier, S., Connolly, S.R., Dunstan, P.K., Fulton, E.A., Harfoot, M.B., et al. (2016) Integrating modelling of biodiversity composition and ecosystem function. Oikos, 125, 10- 19. Monteith, J.L. (1972) Solar Radiation and Productivity in Tropical Ecosystems. Journal of Applied Ecology, 9, 747-766. Mori, A.S., Furukawa, T. & Sasaki, T. (2013) Response diversity determines the resilience of ecosystems to environmental change. Biological Reviews, 88, 349-364. Morin, X., Fahse, L., de Mazancourt, C., Scherer-Lorenzen, M. & Bugmann, H. (2014) Temporal stability in forest productivity increases with tree diversity due to asynchrony in species dynamics. Ecology Letters, 17, 1526-1535. Müller, L.L.B., Albach, D.C. & Zotz, G. (2017) ‘Are 3 °C too much?’: thermal niche breadth in Bromeliaceae and global warming. Journal of Ecology, 105, 507-516.

152

Murphy, B.F. & Timbal, B. (2008) A review of recent climate variability and climate change in Southeastern Australia. International Journal of Climatology, 28, 859-879. Nagalingum, N.S., Knerr, N., Laffan, S., González-Orozco, C.E., Thornhill, A.H., Miller, J., et al. (2015) Continental scale patterns and predictors of fern richness and phylogenetic diversity. Frontiers in Genetics, 6, 132. Nicholls, N., Drosdowsky, W. & Lavery, B. (1997) Australian rainfall variability and change. Weather, 52, 66-72. Nicotra, A.B., Leigh, A., Boyce, C.K., Jones, C.S., Niklas, K.J., Royer, D.L., et al. (2011) The evolution and functional significance of leaf shape in the angiosperms. Functional Plant Biology, 38, 535-552. O'Sullivan, D. & Unwin, D.J. (2010) Geographic Information Analysis: Second Edition. John Wiley & Sons, New Jersey. Oliver, T.H., Heard, M.S., Isaac, N.J., Roy, D.B., Procter, D., Eigenbrod, F., et al. (2015) Biodiversity and Resilience of Ecosystem Functions. Trends in Ecology and Evolution, 30, 673-84. Openshaw, S. (1983) The modifiable areal unit problem. Concepts and Technique in Modern Geography 38. Geo Books, Norwich. Orchard, A.E. (1999) Introduction. Flora of Australia: Volume 1, Introduction, second edition (ed. by A.E. Orchard and H.S. Thompson), pp. 1-9. ABRS/CSIRO Australia, Melbourne. Osborne, P.E., Foody, G.M. & Suárez-Seoane, S. (2007) Non-stationarity and local approaches to modelling the distributions of wildlife. Diversity and Distributions, 13, 313-323. Paquette, A. & Messier, C. (2011) The effect of biodiversity on tree productivity: From temperate to boreal forests. Global Ecology and Biogeography, 20, 170-180. Pasari, J.R., Levi, T., Zavaleta, E.S. & Tilman, D. (2013) Several scales of biodiversity affect ecosystem multifunctionality. Proceedings of the National Academy of Sciences, 110, 10219-10222. Payne, N.L. & Smith, J.A. (2017) An alternative explanation for global trends in thermal tolerance. Ecology Letters, 20, 70-77. Petchey, O.L. (2003) Integrating methods that investigate how complementarity influences ecosystem functioning. Oikos, 101, 323-330. Petter, G., Wagner, K., Wanek, W., Sánchez Delgado, E.J., Zotz, G., Cabral, J.S., et al. (2016) Functional leaf traits of vascular epiphytes: Vertical trends within the forest, intra- and interspecific trait variability, and taxonomic signals. Functional Ecology, 30, 188-198. Pfautsch, S., Harbusch, M., Wesolowski, A., Smith, R., Macfarlane, C., Tjoelker, M.G., et al. (2016) Climate determines vascular traits in the ecologically diverse genus Eucalyptus. Ecology Letters, 19, 240-248. Poorter, L., van der Sande, M.T., Thompson, J., Arets, E.J.M.M., Alarcón, A., Álvarez-Sánchez, J., et al. (2015) Diversity enhances carbon storage in tropical forests. Global Ecology and Biogeography, 24, 1314-1328. Potter, K.M. & Woodall, C.W. (2014) Does biodiversity make a difference? Relationships between species richness, evolutionary diversity, and aboveground live tree biomass across U.S. forests. Forest Ecology and Management, 321, 117-129. Power, S., Casey, T., Folland, C., Colman, A. & Mehta, V. (1999) Inter-decadal modulation of the impact of ENSO on Australia. Climate Dynamics, 15, 319-324. Pressey, R.L., Cabeza, M., Watts, M.E., Cowling, R.M. & Wilson, K.A. (2007) Conservation planning in a changing world. Trends in Ecology and Evolution, 22, 583-592. Pulsford, I., Worboys, G., Howling, G. & Barrett, T. (2012) Great Eastern Ranges, Australia. Climate and Conservation: Landscape and Seascape Science, Planning, and Action (ed. by J.A. Hilty, Chester, C.C., Cross, M.S.), pp. 202-216. Island Press-Center for Resource Economics, Washington D.C.

153

Quader, S., Isvaran, K., Hale, R.E., Miner, B.G. & Seavy, N.E. (2004) Nonlinear relationships and phylogenetically independent contrasts. Journal of Evolutionary Biology, 17, 709-715. R Core Team (2015) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Rabinowitz, D. (1981) Seven Forms of Rarity. The Biological aspects of rare plant conservation (ed. by H. Synge), pp. 205-217. Wiley, New York. Raison, J., Keith, H., Barrett, D., Burrows, B. & Grierson, P. (2003) Estimates of Biomass in ‘mature’ Native Vegetation: National Carbon Accounting System Technical Report No. 44. In. Australian Greenhouse Office, Canberra. Reich, P.B., Wright, I.J., Cavender-Bares, J., Craine, J.M., Oleksyn, J., Westoby, M., et al. (2003) The evolution of plant functional variation: Traits, spectra, and strategies. International Journal of Plant Sciences, 164, S143-S164. Roll, U., Geffen, E. & Yom-Tov, Y. (2015) Linking vertebrate species richness to tree canopy height on a global scale. Global Ecology and Biogeography, 24, 814-825. Rolland, J. & Salamin, N. (2016) Niche width impacts vertebrate diversification. Global Ecology and Biogeography, 25, 1252-1263. Rosauer, D.F., Catullo, R.A., VanDerWal, J., Moussalli, A. & Moritz, C. (2015) Lineage range estimation method reveals fine-scale endemism linked to pleistocene stability in Australian rainforest herpetofauna. PLoS ONE, 10, e0126274. Roughgarden, J. (1972) Evolution of Niche Width. The American Naturalist, 106, 683-718. Ruiz-Benito, P., Gómez-Aparicio, L., Paquette, A., Messier, C., Kattge, J. & Zavala, M.A. (2014) Diversity increases carbon storage and tree productivity in Spanish forests. Global Ecology and Biogeography, 23, 311-322. Savage, J. & Vellend, M. (2015) Elevational shifts, biotic homogenization and time lags in vegetation change during 40 years of climate warming. Ecography, 38, 546-555. Schaminée, H.J.J., Hennekens, M.S., Chytrý, M. & Rodwell, J. (2009) Vegetation-plot data and databases in Europe: An overview. Preslia, 81, 173-185. Schelker, J., Öhman, K., Löfgren, S. & Laudon, H. (2014) Scaling of increased dissolved organic carbon inputs by forest clear-cutting - What arrives downstream? Journal of Hydrology, 508, 299-306. Schemske, D.W., Husband, B.C., Ruckelshaus, M.H., Goodwillie, C., Parker, I.M. & Bishop, J.G. (1994) Evaluating approaches to the conservation of rare and endangered plants. Ecology, 75, 584-606. Schluter, D. (2015) Speciation, ecological opportunity, and latitude (American society of naturalists address). American Naturalist, 187, 1-18. Schwartz, M.W., Brigham, C.A., Hoeksema, J.D., Lyons, K.G., Mills, M.H. & Van Mantgem, P.J. (2000) Linking biodiversity to ecosystem function: Implications for conservation ecology. Oecologia, 122, 297-305. Secretariat of the Convention on Biological Diversity (2005) Handbook of the Convention on Biological Diversity Including its Cartagena Protocol on Biosafety, 3rd edition Secretariat of the CBD, Montreal, Canada. Sengupta, A. & Cressie, N. (2013) Hierarchical statistical modeling of big spatial datasets using the exponential family of distributions. Spatial Statistics, 4, 14-44. Shriver, R.K. (2016) Quantifying how short-term environmental variation leads to long-term demographic responses to climate change. Journal of Ecology, 104, 65-78. Siefert, A., Ravenscroft, C., Weiser, M.D. & Swenson, N.G. (2013) Functional beta-diversity patterns reveal deterministic community assembly processes in eastern North American trees. Global Ecology and Biogeography, 22, 682-691. Slatyer, C., Rosauer, D. & Lemckert, F. (2007) An assessment of endemism and species richness patterns in the Australian Anura. Journal of Biogeography, 34, 583-596.

154

Slatyer, R.A., Hirst, M. & Sexton, J.P. (2013) Niche breadth predicts geographical range size: A general ecological pattern. Ecology Letters, 16, 1104-1114. Soliveres, S., Maestre, F.T., Bowker, M.A., Torices, R., Quero, J.L., García-Gómez, M., et al. (2014) Functional traits determine plant co-occurrence more than environment or evolutionary relatedness in global drylands. Perspectives in Plant Ecology, Evolution and Systematics, 16, 164-173. Soliveres, S., Van Der Plas, F., Manning, P., Prati, D., Gossner, M.M., Renner, S.C., et al. (2016) Biodiversity at multiple trophic levels is needed for ecosystem multifunctionality. Nature, 536, 456-459. Sørensen, T. (1948) A method of establishing groups of equal amplitude in plant sociology based on similarity of species and its application to analyses of the vegetation on Danish commons. Kongelige Danske Videnskabernes Selskab, 5, 1-34. Spasojevic, M.J., Bahlai, C.A., Bradley, B.A., Butterfield, B.J., Tuanmu, M.N., Sistla, S., et al. (2016) Scaling up the diversity-resilience relationship with trait databases and remote sensing data: The recovery of productivity after wildfire. Global Change Biology, 22, 1421- 1432. Srivastava, D.S. & Vellend, M. (2005) Biodiversity-ecosystem function research: Is it relevant to conservation? Annual Review of Ecology, Evolution, and Systematics, 36, 267-294. Strassburg, B.B., Kelly, A., Balmford, A., Davies, R.G., Gibbs, H.K., Lovett, A., et al. (2010) Global congruence of carbon storage and biodiversity in terrestrial ecosystems. Conservation Letters, 3, 98-105. Suppiah, R. & Hennessy, K.J. (1998) Trends in total rainfall, heavy rain events and number of dry days in Australia, 1910-1990. International Journal of Climatology, 18, 1141-1164. Swanson, H.K., Lysy, M., Power, M., Stasko, A.D., Johnson, J.D. & Reist, J.D. (2015) A new probabilistic method for quantifying n-dimensional ecological niches and niche overlap. Ecology, 96, 318-324. Swarbreck, S.M., Sudderth, E.A., St.Clair, S.B., Salve, R., Castanha, C., Torn, M.S., et al. (2011) Linking leaf transcript levels to whole plant analyses provides mechanistic insights to the impact of warming and altered water availability in an annual grass. Global Change Biology, 17, 1577-1594. Thackway, R. & Cresswell, I. (1995) An interim biogeographic regionalisation for Australia : a framework for setting priorities in the National Reserves System, Cooperative Program Version 4.0. Australian Nature Conservation Agency, Canberra. Thackway, R. & Cresswell, I.D. (1997) A bioregional framework for planning the national systems of protected areas in Australia. Natural Areas Journal, 17, 241-247. Thompson, P.L. & Gonzalez, A. (2016) Ecosystem multifunctionality in metacommunities. Ecology, 97, 2867-2879. Thompson, P.L., Davies, T.J. & Gonzalez, A. (2015) Ecosystem functions across trophic levels are linked to functional and phylogenetic diversity. PLoS ONE, 10, e0117595. Thornhill, A.H., Ho, S.Y.W., Külheim, C. & Crisp, M.D. (2015) Interpreting the modern distribution of Myrtaceae using a dated molecular phylogeny. Molecular Phylogenetics and Evolution, 93, 29-43. Thornhill, A.H., Mishler, B.D., Knerr, N.J., González-Orozco, C.E., Costion, C.M., Crayn, D.M., et al. (2016) Continental-scale spatial phylogenetics of Australian angiosperms provides insights into ecology, evolution and conservation. Journal of Biogeography, 43, 2085-2098. Tilman, D. (1982) Resource Competition and Community Structure. Princeton University Press, New Jersey. Tilman, D. (1999) The ecological consequences of changes in biodiversity: A search for general principles. Ecology, 80, 1455-1474. Tilman, D. & Downing, J.A. (1994) Biodiversity and stability in grasslands. Nature, 367, 363-365.

155

Tilman, D., Reich, P.B. & Isbell, F. (2012) Biodiversity impacts ecosystem productivity as much as resources, disturbance, or herbivory. Proceedings of the National Academy of Sciences of the United States of America, 109, 10394-10397. Tilman, D., Isbell, F. & Cowles, J.M. (2014) Biodiversity and ecosystem functioning. Annual Review of Ecology, Evolution, and Systematics,, 45, 471-493. Tobler, W.R. (1970) A computer movie simulating urban growth in the Detroit region. Economic geography, 46, 234-240. Tomlinson, K.W., Sterck, F.J., Bongers, F., da Silva, D.A., Barbosa, E.R.M., Ward, D., et al. (2012) Biomass partitioning and root morphology of savanna trees across a water gradient. Journal of Ecology, 100, 1113-1121. Tung Ho, L.S. & Ané, C. (2014) A linear-time algorithm for gaussian and non-gaussian trait evolution models. Systematic Biology, 63, 397-408. Tuomisto, H. (2010a) A diversity of beta diversities: Straightening up a concept gone awry. Part 2. Quantifying beta diversity and related phenomena. Ecography, 33, 23-45. Tuomisto, H. (2010b) A diversity of beta diversities: Straightening up a concept gone awry. Part 1. Defining beta diversity as a function of alpha and gamma diversity. Ecography, 33, 2-22. Turnbull, L.A., Isbell, F., Purves, D.W., Loreau, M. & Hector, A. (2016) Understanding the value of plant diversity for ecosystem functioning through niche theory. Proceedings of the Royal Society B: Biological Sciences, 283, 20160536. Turner, B., Wells, K., Bauhus, J., Carey, G., Brack, C. & Kanowski, P. (1999) Woody biomass methods for estimating change: National Carbon Accounting System,Technical Report No. 3. In. Australian Greenhouse Office, Canberra. Van Der Plas, F., Manning, P., Soliveres, S., Allan, E., Scherer-Lorenzen, M., Verheyen, K., et al. (2016) Biotic homogenization can decrease landscape-scale forest multifunctionality. Proceedings of the National Academy of Sciences of the United States of America, 113, 3557-3562. Vanderwal, J., Murphy, H.T., Kutt, A.S., Perkins, G.C., Bateman, B.L., Perry, J.J., et al. (2013) Focus on poleward shifts in species' distribution underestimates the fingerprint of climate change. Nature Climate Change, 3, 239-243. Vellend, M., Baeten, L., Myers-Smith, I.H., Elmendorf, S.C., Beauséjour, R., Brown, C.D., et al. (2013) Global meta-analysis reveals no net change in local-scale plant biodiversity over time. Proceedings of the National Academy of Sciences of the United States of America, 110, 19456-19459. Vimal, R. & Devictor, V. (2015) Building relevant ecological indicators with basic data: Species and community specialization indices derived from atlas data. Ecological Indicators, 50, 1- 7. Violle, C., Navas, M.L., Vile, D., Kazakou, E., Fortunel, C., Hummel, I., et al. (2007) Let the concept of trait be functional! Oikos, 116, 882-892. Viscarra Rossel, R.A. & Bui, E.N. (2016) A new detailed map of total phosphorus stocks in Australian soil. Science of the Total Environment, 542, 1040-1049. Viscarra Rossel, R.A., Chen, C., Grundy, M.J., Searle, R., Clifford, D. & Campbell, P.H. (2015) The Australian three-dimensional soil grid: Australia's contribution to the GlobalSoilMap project. Soil Research, 53, 845-864. Vitousek, P.M., Mooney, H.A., Lubchenco, J. & Melillo, J.M. (1997) Human domination of Earth's ecosystems. Science, 277, 494-499. Waide, R.B., Willig, M.R., Steiner, C.F., Mittelbach, G., Gough, L., Dodson, S.I., et al. (1999) The relationship between productivity and species richness. Annual Review of Ecology and Systematics, 30, 257-300. Wang, S. & Loreau, M. (2014) Ecosystem stability in space: α, β and γ variability. Ecology Letters, 17, 891-901.

156

Wang, S. & Loreau, M. (2016) Biodiversity and ecosystem stability across scales in metacommunities. Ecology Letters, 19, 510-8. Wang, X., Wiegand, T., Swenson, N.G., Wolf, A.T., Howe, R.W., Hao, Z., et al. (2015) Mechanisms underlying local functional and phylogenetic beta diversity in two temperate forests. Ecology, 96, 1062-1073. Wardle, D.A., Bonner, K.I. & Nicholson, K.S. (1997) Biodiversity and plant litter: Experimental evidence which does not support the view that enhanced species richness improves ecosystem function. Oikos, 79, 247-258. Warren, D.L., Cardillo, M., Rosauer, D.F. & Bolnick, D.I. (2014) Mistaking geography for biology: inferring processes from species distributions. Trends Ecol Evol, 29, 572-80. Webber, B.L., Yates, C.J., Le Maitre, D.C., Scott, J.K., Kriticos, D.J., Ota, N., et al. (2011) Modelling horses for novel climate courses: Insights from projecting potential distributions of native and alien Australian acacias with correlative and mechanistic models. Diversity and Distributions, 17, 978-1000. Westoby, M. & Wright, I.J. (2006) Land-plant ecology on the basis of functional traits. Trends in Ecology and Evolution, 21, 261-268. Westoby, M., Falster, D.S., Moles, A.T., Vesk, P.A. & Wright, I.J. (2002) Plant ecological strategies: Some leading dimensions of variation between species. Annual Review of Ecology and Systematics, 33, 125-159. White, M.E. (1994) After the greening : the browning of Australia. Kangaroo Press, Sydney. Whittaker, R.H. (1960) Vegetation of the Siskiyou Mountains, Oregon and California. Ecological Monographs, 30, 407-407. Williams, K.J., Belbin, L., Austin, M.P., Stein, J.L. & Ferrier, S. (2012) Which environmental variables should I use in my biodiversity model? International Journal of Geographical Information Science, 26, 2009-2047. Williams, K.J., Ford, A., Rosauer, D.F., De Silva, N., Mittermeier, R., Bruce, C., et al. (2011) Forests of East Australia: The 35th Biodiversity Hotspot. Biodiversity Hotspots: Distribution and Protection of Conservation Priority Areas (ed. by F.E. Zachos and J.C. Habel), pp. 295-310. Springer Berlin Heidelberg, Berlin, Heidelberg. Williams, S.E., Isaac, J.L. & Shoo, L.P. (2009) The Impact of Climate Change on the Biodiversity and Ecosystem Functions of the Wet Tropics. Living in a Dynamic Tropical Forest Landscape, pp. 282-294. Wilson, E.O. (1988) Biodiversity: Foreword. (ed. by E.O. Wilson). National Academies Press, Washingnton D.C. Wilson, K.A., Lulow, M., Burger, J., Fang, Y.-C., Andersen, C., Olson, D., et al. (2011) Optimal restoration: accounting for space, time and uncertainty. Journal of Applied Ecology, 48, 715-725. Winfree, R., Fox, J.W., Williams, N.M., Reilly, J.R. & Cariveau, D.P. (2015) Abundance of common species, not species richness, drives delivery of a real-world ecosystem service. Ecology Letters, 18, 626-635. Wood, S.N. (2013) On p-values for smooth components of an extended generalized additive model. Biometrika, 100, 221-228. Wright, I.J., Reich, P.B., Westoby, M., Ackerly, D.D., Baruch, Z., Bongers, F., et al. (2004) The worldwide leaf economics spectrum. Nature, 428, 821-827. Xu, T. & Hutchinson, M.F. (2013) New developments and applications in the ANUCLIM spatial climatic and bioclimatic modelling package. Environmental Modelling and Software, 40, 267-279. Xu, X., Wang, Z., Rahbek, C., Sanders, N.J. & Fang, J. (2016) Geographical variation in the importance of water and energy for oak diversity. Journal of Biogeography, 43, 279-288. Zanne, A.E., Tank, D.C., Cornwell, W.K., Eastman, J.M., Smith, S.A., Fitzjohn, R.G., et al. (2014) Three keys to the radiation of angiosperms into freezing environments. Nature, 506, 89-92.

157

Zuppinger-Dingley, D., Schmid, B., Petermann, J.S., Yadav, V., De Deyn, G.B. & Flynn, D.F.B. (2014) Selection for niche differentiation in plant communities increases biodiversity effects. Nature, 515, 108-111.

158

Appendix 1: Supplementary material for Chapter 2 - Burley et al. (2016a)

Additional figures and tables for Chapter 2, showing the vegetation communities used in the case study and the biomass estimates for each tree species used.

159

Appendix 1 Supplementary material for Chapter 2

Figure S1.1. Maps showing a), the 1km extant vegetation mask for all of Australia in grey, b), the Interim Biogeographic regions (Liang et al.) and c), the Major Vegetation Groups (MVG) intersected by the 1 km × 500 km transect centred on latitude -36.48 (red line in all three maps).

160

Appendix 1 Supplementary material for Chapter 2

Table S1.1. The 30 most common tree species in south eastern Australia used to create Fig. 3 in the main text of Chapter 2, showing the number of georeferenced records from the Atlas of living Australia (ALA) with a spatial error <2km, occurring within native vegetation and recorded since 1970. The average above ground biomass (t ha-1) of plant communities dominated by each species is derived from a variety of published and unpublished sources [e.g. Grierson et al. (1992); Turner et al. (1999); Keith et al. (2000); Raison et al. (2003)].

Species ALA records Aboveground biomass (Tonnes per ha-1) Angophora floribunda 6875 176 Callitris glaucophylla 9789 39 Corymbia maculate 3579 287 Corymbia gummifera 7049 261 c agglomerate 2415 327 Eucalyptus albens 3772 161 Eucalyptus bridgesiana 1942 96 Eucalyptus camaldulensis 7788 47 Eucalyptus cypellocarpa 5141 274 Eucalyptus dalrympleana 5,448 133 Eucalyptus dealbata 2,214 50 Eucalyptus delegatensis 7108 447 Eucalyptus dives 3796 193 Eucalyptus dwyeri 1858 12 Eucalyptus elata 1584 176 Eucalyptus fastigata 2302 340 Eucalyptus largiflorens 2933 50 Eucalyptus macrorhyncha 5868 136 Eucalyptus microcarpa 2565 61 Eucalyptus mannifera 2257 61 Eucalyptus melliodora 3836 176 Eucalyptus muelleriana 2642 327 Eucalyptus obliqua 17,933 493 Eucalyptus pauciflora 5174 143 Eucalyptus radiata 6196 133 Eucalyptus rossii 1903 62 Eucalyptus rubida 2,077 130 Eucalyptus sideroxylon 1368 114 Eucalyptus sieberi 7,597 298 Eucalyptus viminalis 11258 282

161

Appendix 1 Supplementary material for Chapter 2

Appendix 1 references

Grierson, P.F., Adams, M.A. & Attiwill, P.M. (1992) Estimates of carbon storage in the aboveground biomass of Victoria's forests. Australian Journal of Botany, 40, 631-640. Keith, H., Barrett, D. & Keenan, R. (2000) Review of Allometric Relationships for Estimating Woody Biomass for New South Wales, the Australian Capital Territory, Victoria, Tasmania and South Australia: National Carbon Accounting System Technical Report No. 5B. In. Australian Greenhouse Office, Canberra. Raison, J., Keith, H., Barrett, D., Burrows, B. & Grierson, P. (2003) Estimates of Biomass in ‘mature’ Native Vegetation: National Carbon Accounting System Technical Report No. 44. In. Australian Greenhouse Office, Canberra. Turner, B., Wells, K., Bauhus, J., Carey, G., Brack, C. & Kanowski, P. (1999) Woody biomass methods for estimating change: National Carbon Accounting System,Technical Report No. 3. In. Australian Greenhouse Office, Canberra.

162

Appendix 2: Supplementary material for Chapter 3 - Burley et al. (2016b)

Sensitivity analyses for chapter 3

We conducted sensitivity analyses to determine how robust our methods are to the key analysis settings used in Chapter 3. We emphasise that the data quality thresholds used in the main manuscript are lenient in the Australian context. Australia’s flora comprises approximately 20,000 species of vascular plants (Orchard, 1999), c. 12,500 of which are represented here. Many of these species are under sampled, particularly across the majority of the Australian landmass which is either arid or semi-arid. Therefore it is highly unlikely that that any 1 km × 1 km grid cell in

Australia truly contains fewer than 10 plant species, even in the arid biome. Analysing only those grid cells with the best quality biodiversity estimates has undoubtedly biased the analyses towards areas of high productivity (see discussion). To illustrate the effect of this bias, we removed the thresholds for species richness (10) and the number of locations within a 10km radius (also 10), but still omitted cells where α-diversity was <25% of the Smax for that cell. Smax is the 50km neighbourhood mean of the average maximum number of species recorded within a 20 km radius around each cell (see Figure 3.2, section 3.2.2, Continental subsampling). Using this approach, the retained grid cells are more spread out across the continent encompassing a wider range of productivity values (Figure S2.1 b).

163

Appendix 2 Supplementary material for Chapter 3

Figure S2.1. Maps showing the values of GPP for different subsets of the biological data. a). The approximately 400,000 1 km × 1 km grid cells that were aggregated from the original species occurrence records. b). The initial grid cells (approximately 32,000) where α-diversity is >25% of the Smax. Smax is the 50km neighbourhood mean of the average maximum number of species recorded in any cell within a 20 km radius of each cell (see Figure 3.2, section 3.2.1, Alpha and beta diversity of vascular plants). c). A random subsample from the 32,000 cells in b), where one grid cell is selected within from each 20 km × 20 km spatial block containing data, making approximately 5,500 cells. d). Example of a random subsample generated with the same methods as the main analysis (approximately 1000 cells). These subsampled cells must contain more than 10 species, have more than 10 locations within a 10km radius, and a reasonable probability that α-diversity was well sampled (see section 3.2.2, Continental subsampling).

We then sampled one cell from within each 20 km × 20 km block as per the main analysis

(section 3.2.2, Continental subsampling), but this time drawing from the larger pool of approximately 32,000 grid cells where neither the α-diversity of each cell nor the number of neighbouring cells were restricted. We did not use the probabilities that α- or β-diversity were well sampled as subsampling criteria for this component of the sensitivity analyses (Pα and Pβ, section

Continental subsampling), given that these methods are predicated on the initial data quality thresholds being applied. The block sampling left approximately 5,500 cells in each subsample,

164

Appendix 2 Supplementary material for Chapter 3 with 1000 subsamples chosen as per the main analysis. For each subsample, we ran generalised additive models and calculated the additional percentage of deviance explained (ΔDE) induced by adding each diversity measure (section 3.2.2, Statistical modelling). The sensitivity of our main results was simply calculated as the absolute difference between the unrestricted results and those in the main analyses (Table S2.1). These comparisons clearly show that neither GPP magnitude nor variability has a meaningful relationship with α- or β-diversity, even when considering a wider range of environmental conditions (i.e. despite the bias of our main methods towards locations with higher GPP).

Table S2.1. Results of the sensitivity analysis for the additional contribution of α- and β-diversity measures to GPP, while removing the threshold for >10 species recorded in a grid cell (i.e. minimum α-diversity is 1) and the threshold for >10 locations within a 10km radius surround each grid cell (i.e. minimum number of pairs, N.pairs, is 1). These thresholds were relaxed while still omitting those cells where α-diversity was <25% of Smax (i.e. the 50km neighbourhood mean of the average maximum number of species recorded within a 20 km radius of each cell). N is the number of cells in the final subsample, with subsamples chosen 1000 times. ΔDE is the average additional percentage of deviance explained to the generalised additive models by adding each diversity variable. ΔDE sensitivity is the absolute change in ΔDE between the results for the supplementary analysis and those in Table 3 of the main manuscript.

Response Comparison α-diversity N.pairs N ΔDE (%) ΔDE sensitivity (%) GPP AV env + α vs. env 1 1 c. 5,500 0.085 0.670 env + β vs. env 1 1 c. 5,500 0.049 0.297 env + α + β vs. env + α 1 1 c. 5,500 0.001 0.002 GPP CV env + α vs. env 1 1 c. 5,500 0.056 0.889 env + β vs. env 1 1 c. 5,500 0.472 0.160 env + α + β vs. env + α 1 1 c. 5,500 0.005 0.001

165

Appendix 2 Supplementary material for Chapter 3

Varying the radii for calculating β-diversity

In addition to the data quality thresholds outlined above, both our choice of the Simpson dissimilarity to quantify β-diversity, and the scale at which we calculated it – a 10km radius around each point – could also have affected the results. There any many different ways to calculate β- diversity (Baselga, 2010; Tuomisto, 2010b; Barton et al., 2013; Baselga, 2013; Legendre & De

Cáceres, 2013), with different approaches suited to testing particular hypotheses, depending on the nature of the data. For example, the Simpson index could have been calculated as a site contribution to β-diversity (Legendre & De Cáceres, 2013), or the β-diversity value for each site-pair could have been related to the GPP difference between pairs.

As outlined in the main manuscript, we have used simple geographic measures of biological turnover between site pairs as a proxy for the degree of ecological specialisation at each site.

Ecological specialisation is the key mechanism, which on the balance of the relevant literature

(Devictor et al., 2010; Pasari et al., 2013; Wang & Loreau, 2014), we argue should underpin the macroecological complementarity mechanism. Therefore the effect of the overall biological turnover between locations on GPP – quantified by the Simpson dissimilarity – is the most appropriate measures of β-diversity for our test. Although using simple indices such as the

Sorensen’s or Simpson’s dissimilarity does not directly quantify ecological specialisation, this is true of any of the multitude of other β-diversity indices, so long as they are quantifying purely geographic taxonomic turnover. Thus in our second sensitivity analysis, we have kept Simpson’s dissimilarity as the measure of β-diversity, but have varied the radii across which Simpson’s was calculated (Table S2.2, 10, 20, 40 and 60km). All other settings are the same as the main analysis, demonstrating that the magnitude and stability of GPP is weakly related to β-diversity regardless of the radius across which it is calculated (Table S2.2).

166

Appendix 2 Supplementary material for Chapter 3

Table S2.2. Results of the sensitivity analysis for the additional contribution of α- and β-diversity to GPP when varying the radius for calculating β-diversity around each grid cell, but keeping the data quality thresholds from the main analysis. N is the number of cells in the final subsample, with subsamples chosen 1000 times. ΔDE represents the average additional percentage of deviance explained by adding that variable. ΔDE sensitivity represents the absolute change in ΔDE between the results for the supplementary analysis and those in the main manuscript.

Response Comparison Radius (km) N ΔDE (%) ΔDE sensitivity (%) GPP AV env + α vs. env 10 c. 1,000 0.756 NA env + β vs. env 10 c. 1,000 0.346 NA env + α + β vs. env + α 10 c. 1,000 0.002 NA GPP CV env + α vs. env 10 c. 1,000 0.945 NA env + β vs. env 10 c. 1,000 0.312 NA env + α + β vs. env + α 10 c. 1,000 0.003 NA GPP AV env + α vs. env 20 c. 3,000 0.307 0.449 env + β vs. env 20 c. 3,000 0.394 0.048 env + α + β vs. env + α 20 c. 3,000 0.004 0.001 GPP CV env + α vs. env 20 c. 3,000 0.418 0.527 env + β vs. env 20 c. 3,000 1.763 1.452 env + α + β vs. env + α 20 c. 3,000 0.019 0.016 GPP AV env + α vs. env 40 c. 5,500 0.096 0.659 env + β vs. env 40 c. 5,500 0.283 0.064 env + α + β vs. env + α 40 c. 5,500 0.003 0.001 GPP CV env + α vs. env 40 c. 5,500 0.055 0.890 env + β vs. env 40 c. 5,500 1.787 1.475 env + α + β vs. env + α 40 c. 5,500 0.019 0.016 GPP AV env + α vs. env 60 c. 6,500 0.062 0.693 env + β vs. env 60 c. 6,500 0.323 0.023 env + α + β vs. env + α 60 c. 6,500 0.004 0.001 GPP CV env + α vs. env 60 c. 6,500 0.054 0.891 env + β vs. env 60 c. 6,500 2.012 1.701 env + α + β vs. env + α 60 c. 6,500 0.021 0.018

Using spatial coordinates as an alternative to subsampling

Accounting for spatio-temporal variation in ecological relationships is an ongoing research challenge, as clearly illustrated by Fortin and Dale (2005). With the spatio-temporal perspective in mind, an alternative approach to subsampling the best quality grid cells is to use all valid data points, but also include the spatial coordinates as explanatory variables. This approach has been taken in other studies using similar datasets to test related questions [e.g.Soliveres et al. (2014)],

167

Appendix 2 Supplementary material for Chapter 3 and will reduce the chance of discarding data points of reasonable quality. However, in dealing with spatio-temporal variation in the Australian context, we must consider the interrelationships between rainfall, productivity, temperature and taxonomic diversity measures at continental scales. As noted in the discussion, rainfall, productivity and α-diversity decay with distance from the Australian coastline (Lavers & Field, 2006), with the additional complication that sampling effort also declines with distance from the coast. Therefore including the spatial coordinates would simply approximate environmental conditions, albeit indirectly, effectively ‘soaking up’ explanatory power without necessarily adding any biogeographic insight. Indeed geographic distances themselves do not explain biogeographic variations, rather they poorly approximate aspects of biogeographic history which cannot be quantified over broad spatio-temporal scales [e.g. Burley et al. (2012); Warren et al. (2014)].

To demonstrate this circularity, we have re-run our generalised additive models using the subset of approximately 32,000 grid cells depicted in Figure S2.1 b, but this time including splines for the spatial coordinates as explanatory variables. This simple re-analysis shows that although latitude and longitude alone explain more than 50% of the variation in productivity (Table S2.3,

Figure S2.2), this explanatory power is due to strong covariance with rainfall and temperature

(Figure S2.2). Furthermore, including the coordinates does not alter the minimal contribution of α- and β-diversity to productivity (Table S2.3). More spatially explicit techniques, such as geographically weighted regression (Xu et al., 2016) and spatial structural equation modelling

(Lamb et al., 2014), would likely further reduce the explanatory power of α- and β-diversity by decreasing overall residual variation. Given that a much simpler method shows no relationship between GPP magnitude or variability and α- or β-diversity, we retained the approach outlined in the main manuscript.

168

Appendix 2 Supplementary material for Chapter 3

Table S2.3. Total percentage (%) of deviance explained (DE) for mean gross primary productivity (GPP) as a function of environmental conditions, spatial coordinates (i.e. latitude and longitude) and α- and β-diversity (shaded colours). Analyses were performed on the initial set is depicted in Figure S2.1 b with approximately 32,000 cells. DE values are shown for each predictor variable by itself and the four model combinations used in the continental analyses. Rain denotes total monthly rainfall (mm month-1 2001-2012), and Temp denotes mean monthly maximum temperature (°C). ΔDE is the absolute change in deviance explained (DE) between the results for this supplementary analysis, and those in Table 1 of the main manuscript.

Predictors GPP Rain Temp α-diversity β-diversity Lat/long Subset n DE (%) ΔDE (%) initial c. 32,000 77.104 0.403 initial c. 32,000 74.562 1.888 initial c. 32,000 74.502 1.548 initial c. 32,000 74.426 1.274 initial c. 32,000 64.047 0.027 initial c. 32,000 50.364 NA initial c. 32,000 21.622 6.012 initial c. 32,000 7.120 4.640 initial c. 32,000 1.720 1.270

169

Appendix 2 Supplementary material for Chapter 3

Figure S2.2. Plots of a continental generalised additive model of average gross primary productivity (GPP, raw observations in blue points), using approximately 32,000 grid cells where α-diversity was >25% of Smax (the 50km neighbourhood mean of the average maximum number of species recorded within a 20 km radius around each cell). Predicted average GPP for models using only that predictor [e.g. GPP = f (rain) for top row of panel] are plotted in green. Predicted GPP values for models holding all other variables at their mean [e.g. GPP = f (rain + mean temp + mean α + mean β) for top row of panel] are also plotted in orange. GPP is the mean monthly photosynthetic flux of carbon (gC m-2 month-1, 2001/01-2012/12, section 3.2.2, Continental subsampling).

170

Appendix 2 Supplementary material for Chapter 3

Table S2.4. Results of the sensitivity analysis for the additional contribution of α- and β-diversity to the magnitude and variability of GPP, while including a spline with five knots for latitude and longitude, in addition to rainfall and temperature. The initial subset is depicted in Figure S2.1 b, and the best blocks set is depicted in Figure S2.1 d from the main manuscript. The best blocks subsamples were chosen 1000 times, as per the main analysis. ΔDE sensitivity is the absolute change in ΔDE between the results for the supplementary analysis and those in Table 3.3 of the main manuscript.

response comparison subset n ΔDE (%) ΔDE sensitivity (%) GPP AV env + s(x,y) + α vs. env + s(x,y) initial c. 32,000 0.044 0.712 env + s(x,y) + β vs. env + s(x,y) initial c. 32,000 0.248 0.098 env + s(x,y) + α + β vs. env + s(x,y) + α initial c. 32,000 0.002 0.000 GPP CV env + s(x,y) + α vs. env + s(x,y) initial c. 32,000 0.054 0.891 env + s(x,y) + β vs. env + s(x,y) initial c. 32,000 0.110 0.202 env + s(x,y) + α + β vs. env + s(x,y) + α initial c. 32,000 0.001 0.002 GPP AV env + s(x,y) + α vs. env + s(x,y) best blocks c. 1,000 0.313 0.443 env + s(x,y) + β vs. env + s(x,y) best blocks c. 1,000 0.548 0.202 env + s(x,y) + α + β vs. env + s(x,y) + α best blocks c. 1,000 0.005 0.002 GPP CV env + s(x,y) + α vs. env + s(x,y) best blocks c. 1,000 0.996 0.051 env + s(x,y) + β vs. env + s(x,y) best blocks c. 1,000 0.097 0.215 env + s(x,y) + α+ β vs. env + s(x,y) + α best blocks c. 1,000 0.001 0.003

171

Appendix 2 Supplementary material for Chapter 3

Appendix 2 references

Barton, P.S., Cunningham, S.A., Manning, A.D., Gibb, H., Lindenmayer, D.B. & Didham, R.K. (2013) The spatial scaling of beta diversity. Global Ecology and Biogeography, 22, 639- 647. Baselga, A. (2010a) Multiplicative partition of true diversity yields independent alpha and beta components; additive partition does not. Ecology, 91, 1974-1981. Baselga, A. (2010b) Partitioning the turnover and nestedness components of beta diversity. Global Ecology and Biogeography, 19, 134-143. Baselga, A. (2013) Multiple site dissimilarity quantifies compositional heterogeneity among several sites, while average pairwise dissimilarity may be misleading. Ecography, 36, 124-128. Burley, H.M., Laffan, S.W. & Williams, K.J. (2012) Spatial non-stationarity and anisotropy of compositional turnover in eastern Australian Myrtaceae species. International Journal of Geographical Information Science, 26, 2065-2081. Burley, H.M., Mokany, K., Ferrier, S., Laffan, S.W., Williams, K.J. & Harwood, T.D. (2016) Macroecological scale effects of biodiversity on ecosystem functions under environmental change. Ecology and Evolution, in press, DOI 10.1002/ece3.2036. Devictor, V., Clavel, J., Julliard, R., Lavergne, S., Mouillot, D., Thuiller, W., et al. (2010) Defining and measuring ecological specialization. Journal of Applied Ecology, 47, 15-25. Fortin, M.J. & Dale, M.R.T. (2005) Spatial Analysis: A Guide for Ecologists. Cambridge University Press, Cambridge Frainer, A., McKie, B.G. & Malmqvist, B. (2014) When does diversity matter? Species functional diversity and ecosystem functioning across habitats and seasons in a field experiment. Journal of Animal Ecology, 83, 460-469. Lamb, E.G., Mengersen, K.L., Stewart, K.J., Attanayake, U. & Siciliano, S.D. (2014) Spatially explicit structural equation modeling. Ecology, 95, 2434-2442. Lavers, C. & Field, R. (2006) A resource-based conceptual model of plant diversity that reassesses causality in the productivity-diversity relationship. Global Ecology and Biogeography, 15, 213-224. Legendre, P. & De Cáceres, M. (2013) Beta diversity as the variance of community data: Dissimilarity coefficients and partitioning. Ecology Letters, 16, 951-963. Orchard, A.E. (1999) Introduction. Flora of Australia: Volume 1, Introduction, second edition (ed. by A.E. Orchard and H.S. Thompson), pp. 1-9. ABRS/CSIRO Australia, Melbourne. Pasari, J.R., Levi, T., Zavaleta, E.S. & Tilman, D. (2013) Several scales of biodiversity affect ecosystem multifunctionality. Proceedings of the National Academy of Sciences, 110, 10219-10222. Soliveres, S., Maestre, F.T., Bowker, M.A., Torices, R., Quero, J.L., García-Gómez, M., et al. (2014) Functional traits determine plant co-occurrence more than environment or evolutionary relatedness in global drylands. Perspectives in Plant Ecology, Evolution and Systematics, 16, 164-173. Tuomisto, H. (2010a) A diversity of beta diversities: Straightening up a concept gone awry. Part 2. Quantifying beta diversity and related phenomena. Ecography, 33, 23-45. Tuomisto, H. (2010b) A diversity of beta diversities: Straightening up a concept gone awry. Part 1. Defining beta diversity as a function of alpha and gamma diversity. Ecography, 33, 2-22. Wang, S. & Loreau, M. (2014) Ecosystem stability in space: α, β and γ variability. Ecology Letters, 17, 891-901. Warren, D.L., Cardillo, M., Rosauer, D.F. & Bolnick, D.I. (2014) Mistaking geography for biology: inferring processes from species distributions. Trends in Ecology and Evolution, 29, 572-80. Xu, X., Wang, Z., Rahbek, C., Sanders, N.J. & Fang, J. (2016) Geographical variation in the importance of water and energy for oak diversity. Journal of Biogeography, 43, 279-288.

172

Appendix 2 Supplementary material for Chapter 3

173

Appendix 3: Supplementary material for Chapter 4

Additional figures and tables for the relationship between realised niche width and niche median, the relationship between niche median and traits, and the relationship between niche volume, niche medians and traits.

174

Appendix 3 Supplementary material for Chapter 4

Figure S3.1. Plots of species realised niche width vs. niche median for the Wet Tropics vascular flora by life form, showing the natural log of precipitation (originally mm), maximum temperature (originally °C), and the natural log of soil nitrogen (originally %). For each environmental gradient, the regression model is plotted in orange over species niches (blue points), with the explanatory power for each model shown in orange (deviance explained values for rainfall and nitrogen, and adjusted R2 values for temperature). The left column of panels show relationships for shrub species only (n = 334), and the right column of panels shows relationships for herbaceous species only (n = 450).

175

Appendix 3 Supplementary material for Chapter 4

Figure S3.2. Plots of species realised niche width vs. niche median for the Wet Tropics vascular flora by life form, showing the natural log of precipitation (originally mm), maximum temperature (originally °C), and the natural log of soil nitrogen (originally %). For each environmental gradient, the regression model is plotted in orange over species niches (blue points), with the explanatory power for each model shown in orange (deviance explained values for rainfall and nitrogen, and adjusted R2 values for temperature). The left column of panels show relationships for vine species only (n = 187), and the right column of panels show relationships for fern species only (n = 92).

176

Appendix 3 Supplementary material for Chapter 4

Figure S3.3. Plots of genus-level realised niche width vs. niche median for the Wet Tropics vascular flora, showing the natural log of precipitation (originally mm, RW = rainfall niche width. RM = rainfall niche median), maximum temperature (°C, TW = temperature niche width, TM = temperature niche median) and log of soil nitrogen (originally %, NW = nitrogen niche width, NM = nitrogen niche median). For each plot, the regression model is plotted in orange over the niches (blue points), with the explanatory power for each model shown in orange (deviance explained values for rainfall and nitrogen, and adjusted R2 values for temperature). The left column of panels show relationships for genus-level niches of all genera (n = 788), and the right column of panels show relationships for tree genera (n = 268).

177

Appendix 3 Supplementary material for Chapter 4

Figure S3.4. Comparison plots of species regressions (left column of panels) vs. phylogenetically corrected linear models (right column of panels) for species realised niche width vs. niche median for the natural log of precipitation (originally mm), maximum temperature (°C), and the natural log of soil nitrogen (originally %). For each environmental gradient, the regression model is plotted in orange over species niches (blue points). The left column of panels show all species analysed (n = 1771). The right column of panels show all species on the Zanne et al. (2014) angiosperm phylogeny (n = 723). For the right column, the phylogenetically corrected linear model is plotted in orange, and the standard linear model is plotted in green. Phylogenetic regressions were run using the phylolm function in the phylolm package (Tung Ho & Ané, 2014).

178

Appendix 3 Supplementary material for Chapter 4

Figure S3.5. Structural equation model diagram for realised niche volume vs. realised niche median across each environmental gradient, leaf area and height for all species analysed in chapter 4 (n = 1771). Leaf area is the log10 (leaf area cm2 + 1) and height is maximum recorded height (m). Niche volume is a three dimensional ellipsoid, where the dimensions are 0-1 scaled species environmental realised niche widths (rain, temperature and nitrogen). The scaled realised niche medians are shown (originally in mm, °C and % for rainfall, temperature and nitrogen respectively). Single-headed arrows are directional linear relationships, double-headed arrows are co-variances. Arrow thickness is proportional to the standardised path coefficient labelled on each line (blue = positive, red = negative, dashed = non-significant at α = 0.05, * <0.05, ** <0.001, *** <0.0001).

179

Appendix 3 Supplementary material for Chapter 4

Figure S3.6. Plots of the realised niche width and median (response) against the geographic niche of all species in the analysis (predictor, n = 1771). The geographic niche plotted here is the natural log of the count of all 10 km × 10 km cells that each species occurred in across continental Australia, including Tasmania, multiplied by the area of one grid cell (100 km2).

180

Appendix 3 Supplementary material for Chapter 4

Figure S3.7. Plots of species realised niche width vs. niche median for the subset of the Wet Tropics vascular flora which have their biogeographic origin attributed to either Australia (n = 229, blue points), or Southeast Asia (n = 228, purple points). For each environmental gradient, the regression model for Australian-origin species is plotted in orange over species niches (points), and the model for Asian-origin species is plotted in green (the explanatory power of the model for all species is plotted in black). The right column of panels show boxplots of the realised niche width across each environmental gradient for species from Australia (blue) and Asia (purple). P-values are for two-sample T-tests of the hypothesis that the difference in mean niche width between Australian and Asian species ≠ 0. Biogeographic attributions by Samantha Yap.

181

Appendix 3 Supplementary material for Chapter 4

Figure S3.8. Plots of species realised niche width vs. niche median for the subset of the Wet Tropics vascular flora which have their biogeographic origin attributed to either Australia (n = 229, blue points), or Southeast Asia (n = 228, purple points). For each environmental gradient, the regression model for Australian-origin species is plotted in orange over species niches (points), and the model for Asian-origin species is plotted in green (the explanatory power of the model for all species is plotted in black). The right column of panels show boxplots of the realised niche width across each environmental gradient for species from Australia (blue) and Asia (purple). P-values are for two-sample T-tests of the hypothesis that the difference in mean niche width between Australian and Asian species ≠ 0. Biogeographic attributions by Samantha Yap.

182

Appendix 3 Supplementary material for Chapter 4

2 Figure S3.9. Plots of species realised niche median vs. leaf area [log10 (leaf area cm + 1)] for the subset of 457 Wet Tropics species which have their biogeographic origins attributed to either Australia (n = 229, blue points), or South East Asia (n = 228, purple points). For each environmental gradient, the regression model for Australian-origin species is plotted in orange over species niches (points), and the model for Asian-origin species is plotted in green. The explanatory power of the model for all species (n = 457) is plotted in black. The right panel shows boxplots of maximum plant height for species from Australia (blue) and Asia (purple). The P- value is for a two-sample T-test of the hypothesis that the difference in log leaf area between Australian and Asian species ≠ 0. Biogeographic attributions by Samantha Yap.

183

Appendix 3 Supplementary material for Chapter 4

Table S3.1. Bivariate regression results for realised niche width (NW) vs. niche median (NM) across each environmental gradient for vascular plants in the analysis, by data subset: all species without restriction (All, n = 4162), all species within the WT environmental bounds (Env, n = 2293), all species within the WT environmental bounds and with >20 records (Env >20, n = 1771), all species within the WT environmental bounds and with >100 records (Env >100, n = 697). Explanatory power (EP) is deviance explained for 2 rainfall and nitrogen, and adjusted R for temperature. C is the linear coefficient, β1, for temperature, and the first GAM coefficient,

S1, for rainfall and nitrogen. The F-statistic is given for rainfall and nitrogen, and the T-statistic is given for temperature. Standard p- values for the tests statistics are given, but are unreliable due to strong spatial autocorrelation in the biological and environmental data (see section ‘Bivariate regression models’).

Species Model N EP Coefficients Test stat P Intercept C All Rainfall log(NW) ~ log(NM) 4162 0.184 6.830 -0.04760655 511.383 <0.001 Env Rainfall log(NW) ~ log(NM) 2293 0.191 6.880 0.500973505 272.025 <0.001 Env >20 Rainfall log(NW) ~ log(NM) 1771 0.387 7.012 0.499241615 559.961 <0.001 Env >100 Rainfall log(NW) ~ log(NM) 697 0.652 7.070 0.358733489 652.769 <0.001 All Temp NW ~ NM 4162 0.011 2.358 0.085204099 6.793 <0.001 Env Temp NW ~ NM 2293 0.067 -9.570 0.441728918 12.895 <0.001 Env >20 Temp NW ~ NM 1771 0.085 -9.665 0.460043166 12.850 <0.001 Env >100 Temp NW ~ NM 697 0.073 -5.582 0.374322148 7.457 <0.001 All Nitrogen log(NW) ~ log(NM) 4162 0.463 -2.307 0.205137987 1797.332 <0.001 Env Nitrogen log(NW) ~ log(NM) 2293 0.307 -2.331 0.317698484 511.444 <0.001 Env >20 Nitrogen log(NW) ~ log(NM) 1771 0.513 -2.213 0.295852764 936.013 <0.001 Env >100 Nitrogen log(NW) ~ log(NM) 697 0.601 -2.123 -4.8616E-09 1045.082 <0.001

184

Appendix 3 Supplementary material for Chapter 4

Table S3.2. Bivariate regression results for realised niche width (NW) vs. different measures of niche midpoint across each environmental gradient for all vascular plants in the analysis (n = 1771). NMD = niche median, NMN = niche mean, NC = the centre of the niche width (i.e. the centre of the middle 90% of environmental values used in the main manuscript). Explanatory power (EP) 2 is deviance explained for rainfall and nitrogen, and adjusted R for temperature. C is the linear coefficient, β1, for temperature, and the first GAM coefficient, S1, for rainfall and nitrogen. The F-statistic is given for rainfall and nitrogen, and the T-statistic is given for temperature. Standard p-values for the tests statistics are given, but are unreliable due to strong spatial autocorrelation in the biological and environmental data (see section ‘Bivariate regression models’).

Species Model N EP Coefficients Test stat P Intercept C All Rainfall log(NW) ~ log(NMD) 1771 0.387 7.012 0.499 559.961 <0.001 All Rainfall log(NW) ~ log(NMN) 1771 0.457 7.012 0.511 744.724 <0.001 All Rainfall log(NW) ~ log(NC) 1771 0.536 7.012 0.420 1015.923 <0.001 All Temp NW ~ NMD 1771 0.085 -9.665 0.460 12.850 <0.001 All Temp NW ~ NMN 1771 0.102 -10.674 0.491 14.221 <0.001 All Temp NW ~ NC 1771 0.109 -10.218 0.477 14.732 <0.001 All Nitrogen log(NW) ~ log(NMD) 1771 0.513 -2.213 0.296 936.013 <0.001 All Nitrogen log(NW) ~ log(NMN) 1771 0.628 -2.213 0.388 1492.730 <0.001 All Nitrogen log(NW) ~ log(NC) 1771 0.758 -2.213 0.380 2779.321 <0.001

185

Appendix 3 Supplementary material for Chapter 4

Table S3.3. Bivariate regression results for realised niche width (NW) vs. realised niche median (NM) across each environmental gradient for all species in the analysis, by life form: trees (n = 647), herbs (n = 450), shrubs (n = 334), ferns (n = 187) and vines (n = 2 92). Explanatory power (EP) is deviance explained for rainfall and nitrogen, and adjusted R for temperature. C is the linear coefficient, β1, for temperature, and the first GAM coefficient, S1, for rainfall and nitrogen. The F-statistic is given for rainfall and nitrogen, and the T-statistic is given for temperature. Standard p-values for the tests statistics are given, but are unreliable due to strong spatial autocorrelation in the biological and environmental data (see section ‘Bivariate regression models’).

Species Model N EP Coefficients Test stat P Intercept C

All Rainfall log(NW) ~ log(RM) 1771 0.387 7.012 0.499 559.96 <0.001 All Temp NW ~ NM 1771 0.085 -9.665 0.460 12.85 <0.001 All Rainfall log(NW) ~ log(RM) 1771 0.513 -2.213 0.296 936.01 <0.001 Trees Rainfall log(NW) ~ log(RM) 647 0.322 7.159 0.558 154.29 <0.001 Trees Temp NW ~ NM 647 0.095 -10.956 0.482 8.27 <0.001 Trees Rainfall log(NW) ~ log(RM) 647 0.473 -2.076 0.406 290.71 <0.001 Herbs Rainfall log(NW) ~ log(RM) 450 0.325 6.866 0.239 111.289 <0.001 Herbs Temp NW ~ NM 450 -0 5.326 0.038 0.555 0.579 Herbs Nitrogen log(NW) ~ log(RM) 450 0.411 -2.398 0.291 158.573 <0.001 Shrubs Rainfall log(NW) ~ log(RM) 334 0.33 6.763 0.617 82.932 <0.001 Shrubs Temp NW ~ NM 334 0.052 -6.458 0.357 4.381 <0.001 Shrubs Nitrogen log(NW) ~ log(RM) 334 0.488 -2.339 0.064 186.701 <0.001 Ferns Rainfall log(NW) ~ log(RM) 187 0.404 7.262 0.378 63.823 <0.001 Ferns Temp NW ~ NM 187 0.089 -11.316 0.498 4.378 <0.001 Ferns Nitrogen log(NW) ~ log(RM) 187 0.552 -2.087 0.231 117.310 <0.001 Vines Rainfall log(NW) ~ log(RM) 92 0.389 7.442 0.321 26.504 <0.001 Vines Temp NW ~ NM 92 0.12 -16.192 0.679 3.659 <0.001 Vines Nitrogen log(NW) ~ log(RM) 92 0.505 -1.869 0.118 53.065 <0.001

186

Appendix 3 Supplementary material for Chapter 4

Table S3.4. Bivariate regression results for realised niche width (NW) vs. niche median (NM) across each environmental gradient, for all genera (n = 788), and all tree genera (n = 268). Explanatory power (EP) is deviance explained for rainfall and nitrogen, and 2 adjusted R for temperature. C is the linear coefficient, β1, for temperature, and the first GAM coefficient, S1, for rainfall and nitrogen. The F-statistic is given for rainfall and nitrogen, and the T-statistic is given for temperature. Standard p-values for the tests statistics are given, but are unreliable due to strong spatial autocorrelation in the biological and environmental data (see section ‘Bivariate regression models’).

Species Model N EP Coefficients Test stat P Intercept C All genera Rain log(NW) ~ log(NM) 788 0.252 7.206 <0.000 264.376 <0.001 All genera Temp NW ~ NM 788 0.001 8.377 -0.054 -1.287 0.198 All genera Nitrogen log (NW) ~ log(NM) 788 0.537 -2.010 0.000 913.038 <0.001 Tree genera Rain log(NW) ~ log(NM) 268 0.142 7.363 0.503 20.780 <0.001 Tree genera Temp NW ~ NM 268 -0.001 8.235 -0.077 -0.894 0.372 Tree genera Nitrogen log(NW) ~ log(NM) 268 0.471 -1.846 0.215 122.286 <0.001

Table S3.5. Comparison of bivariate regression results for species realised niche width vs. niche median across each environmental gradient, for all Wet Tropics species on the Zanne et al. (2014) angiosperm phylogeny (N = 723) using linear models (Lm) and phylogenetically corrected linear models (Phy). Phylogenetic regressions were run using the phylolm function in the phylolm package (Tung Ho & Ané, 2014). RW/RM: realised rainfall niche width and niche median. TW/TM: realised temperature niche width and niche median. NW/NM: realised nitrogen niche width and niche median. EP: adjusted R2 value of the standard linear models of species niches for each environmental gradient (i.e. uncorrected for phylogeny). The intercepts, coefficients, t-statistics and p-values (P) from the linear (Lm) and phylogenetically corrected models (Phy) are given.

Model N EP Intercept Coefficient T-statistic P Lm Phy Lm Phy Lm Phy Lm Phy log(RW) ~ log(RM) 723 0.335 1.608 4.525 0.761 0.393 19.100 10.599 <0.001 <0.001 TW ~ TM 723 0.068 -8.957 -1.342 0.445 0.229 7.310 5.470 <0.001 <0.001 log(NW) ~ log(NM) 723 0.498 -0.105 -0.530 0.893 0.641 26.767 14.842 <0.001 <0.001

187

Appendix 3 Supplementary material for Chapter 4

Table S3.6. Comparison of regression results for the realised niche median across each environmental gradient vs height and leaf area, for all Wet Tropics tree species on the Zanne et al. (2014) angiosperm phylogeny (N = 319) using linear models (Lm) and phylogenetically corrected linear models (Phy). Phylogenetic regressions were run using the phylolm function in the phylolm package (Tung Ho & Ané, 2014). RM: realised rainfall niche median. TM: realised temperature niche median. NM: realised nitrogen niche median. EP: adjusted R2 value of the standard linear models of species niches for each environmental gradient (i.e. uncorrected for phylogeny). The intercepts, coefficients, t-statistics and p-values (P) from the linear (Lm) and phylogenetically corrected models (Phy) are given.

Model N EP Intercept Coefficient T-statistic P Lm Phy Lm Phy Lm Phy Lm Phy RM ~ height 319 0.010 1683.05 1506.10 6.903 22.846 2.013 16.125 0.0449 0.0000 TM ~ height 319 0.009 31.95 29.72 -0.013 0.071 -1.951 16.856 0.0519 0.0000 NM ~ height 319 0.007 0.12 0.15 0.001 -0.001 1.819 -4.470 0.0698 0.0000 RM ~ leaf area 319 0.132 853.19 2979.14 516.965 -499.854 7.027 -4.094 0.0000 0.0001 TM ~ leaf area 319 0.001 32.03 35.96 -0.173 -2.422 -1.110 -6.849 0.2678 0.0000 NM ~ leaf area 319 0.007 0.10 0.10 0.012 0.017 1.808 1.819 0.0715 0.0699

Table S3.7. Total percentage (%) of deviance explained (DE), t-statistics and Bayesian information criterion values (BIC) for generalised additive models of species realised niche volume (Nv) as a function of realised niche median across each environmental gradient, height and leaf area for all tree species analysed (n = 647). Models are ranked by BIC values in ascending order. RM: realised rainfall niche median, TM: realised temperature niche median. NM: realised nitrogen niche median. Approximate hypothesis tests of whether the total deviance of each model was different to that of the full model [Nv = f(RM + TM + NM + height + area)] were performed using the anova.gam function in the mgcv R package (*** <0.001, NA for full model and NS = non-significant). 2 Height is the maximum recorded height in metres of each tree species, and leaf area is the log10 (leaf area cm + 1).

Predictors Realised niche volume (Nv) RM TM NM Height Area DE T.stat BIC P.value 23.367 -146.65 -5143.09 0.0012 23.366 -149.28 -5142.30 0.0126 23.366 -149.28 -5142.28 <0.0001 23.367 -149.28 -5142.24 0.0279 19.851 -155.34 -5127.72 <0.001 14.222 -152.52 -5102.03 <0.0001 15.333 -142.85 -5087.40 <0.0001 12.207 -136.11 -5080.30 <0.0001 8.490 -141.11 -5067.49 <0.0001 6.658 -145.41 -5037.07 <0.0001 1.015 -138.41 -5008.88 <0.0001

188

Appendix 4: Supplementary material for Chapter 5

Additional figures and tables for Chapter 5, showing the variability of rainfall and temperature across the Wet Tropics, site-level relationships, the effects of rainfall and temperature on pairwise

GPP, the full results at multiple rainfall thresholds and the effects of mean niche widths on pairwise

GPP.

189

Appendix 4 Supplementary material for Chapter 5

Fig. S4.1. Maps and plots of environmental variability in the Wet Tropics for the period of analysis between January 2001 and December 2012. The coefficient of variation (CV, the standard deviation divided by the mean) in total monthly rainfall (mm month-1 2001-2012) across the Wet Tropics (in blue, a), and the CV of mean monthly daily maximum temperature (in orange, b). Comparative histograms of the coefficient of variation distribution for rainfall and temperature at all 510 Wet Tropics sites across the time series (2001-2012, rainfall histograms in blue, temperature histograms in orange). The difference in variability between rainfall and temperature was similar to that depicted in panel c when considered across the whole Wet Tropics.

190

Appendix 4 Supplementary material for Chapter 5

Fig. S4.2. Plots of continental temperature and rainfall niche width vs. continental niche median for all 948 tree species occurring at 510 Wet Tropics sites. The niche width of each species is the middle 90% of temperature and rainfall values for all continental geographic records of that species. The niche median is the median of environmental values for all geographic records of that species. For rainfall, the natural log of niches are plotted (originally mm). For each environmental gradient, the regression model is plotted in orange over the species niches (blue points), with the explanatory power for each model shown in orange (deviance explained values for the generalised additive model of rainfall, and adjusted R2 for the linear model of temperature).

191

Appendix 4 Supplementary material for Chapter 5

Table S4.1. Total percentage (%) of deviance explained (DE), t-statistics and Bayesian information criterion values (BIC) for generalised additive models of site gross primary productivity (GPP, gC m-2 month-1 2001-2012) as a function of site environment and median site niche widths (shaded colours) for all 510 sites (42-516 mm of monthly rainfall). Models are ranked by BIC values in ascending order. RNW: rainfall niche width, TNW: temperature niche width. Approximate hypothesis tests of whether the total deviance of each model was different to that of the full model [GPP = f(rainfall + RNW + temperature + TNW)] were performed using the anova.gam function in the mgcv R package (*** <0.001, NA for full model and NS = non-significant).

Predictors GPP Rain RNW Temp TNW DE(%) T statistic BIC 79.3 1233.6 4735.8 NS 79.4 1238.9 4737.0 NA 74.9 1125.7 4812.6*** 75.1 1138.0 4814.2*** 71.8 1050.9 4876.2*** 69.1 1020.6 4902.0*** 67.1 981.8 4927.9*** 55.9 866.2 5076.8*** 55.7 828.7 5079.3*** 8.2 624.8 5452.8***

Table S4.2. Total percentage (%) of deviance explained (DE), t-statistics and Bayesian information criterion values (BIC) for generalised additive models of site gross primary productivity (GPP, gC m-2 month-1) as a function of environmental conditions and niche widths (shaded colours) for only the 255 driest sites (42-138 mm of monthly rainfall). Models are ranked by BIC values in ascending order. RNW: rainfall niche width, TNW: temperature niche width. Approximate hypothesis tests of whether the total deviance of each model was different to that of the full model [GPP = f(rainfall + RNW + temperature + TNW)] were performed using the anova.gam function in the mgcv R package (*** <0.001, NA for full model and NS = non-significant).

Predictors GPP Rain RNW Temp TNW DE(%) T-statistic BIC 79.5 741.3 2388.2 NS 79.5 744.1 2391.5 NA 75.0 683.0 2408.2*** 72.4 637.0 2448.1*** 73.3 660.2 2455.7*** 65.7 587.6 2491.3***

63.1 561.5 2504.6*** 54.5 514.8 2558.7***

52.5 484.4 2569.6*** 33.7 412.6 2639.1***

192

Appendix 4 Supplementary material for Chapter 5

Table S4.3. Total percentage (%) of deviance explained (DE), t-statistics and Bayesian information criterion values (BIC) for generalised additive models of site gross primary productivity (GPP, gC m-2 month-1) as a function of environmental conditions and niche widths (shaded colours) for only 255 wetter sites (139-516 mm of monthly rainfall). Models are ranked by BIC values in ascending order. RNW: rainfall niche width, TNW: temperature niche width. Approximate hypothesis tests of whether the total deviance of each model was different to that of the full model [GPP = f(rainfall + RNW + temperature + TNW)] were performed using the anova.gam function in the mgcv R package (*** <0.001, NA for full model and NS = non-significant).

Predictors GPP Rain RNW Temp TNW DE T.stat BIC 20.4 1212.6 2274.8 NA NS 20.2 1210.0 2277.1 NS 20.5 1207.6 2286.7 20.5 1208.0 2286.8 NS 20.4 1206.7 2288.0 NS NS 7.8 1123.3 2320.6 NS 9.3 1130.1 2327.0 NS 2.8 1098.8 2328.7 NS 4.3 1103.5 2329.3 7.2 1115.7 2335.3 NS

Table S4.4. Average results for 500 linear models run on subsamples of environmentally similar, unique site-pairs, where pairwise differences in gross primary productivity (response, GPP gC m-2 month-1 2001-2012) are a function of pairwise differences in site rainfall and site temperature (predictor, mm and °C respectively, 2001-2012). All site-pair differences are ordered by the predictor variable, so that the predictor is always positive. ΔGPP is the allowable difference in predicted GPP between site-pairs, based on a negative exponential model of site GPP as a function of rainfall, showing subsampling results for two different ΔGPP values (1, 20). ΔT is the allowable temperature difference between site-pairs, set at 0.5°C. N is the average number of site-pairs used across 500 subsamples, Int is the average

2 2 linear intercept, β1 is the average linear coefficient, T is the average t-statistic and R is the average adjusted R . None of the relationships were significant at the 0.05 level using the standard lm p-values).

2 Precip. subset (mm) Model ΔGPP N Int β1 T R All(42-516) GPP ~ temp site 1 186.5 -2.789 6.535 0.46 -0.002 All(42-516) GPP ~ rain site 1 186.6 0.133 0.043 0.16 -0.003 All(42-516) GPP ~ temp site 20 245.4 0.643 -11.203 -0.74 0.002 All(42-516) GPP ~ rain site 20 245.3 4.976 0.027 0.47 -0.001 Drier(42-138) GPP ~ temp site 1 77.8 -2.088 -18.006 -0.70 -0.003 Drier(42-138) GPP ~ rain site 1 77.7 -0.837 -0.573 -0.08 -0.009 Drier(42-138) GPP ~ temp site 20 121.3 3.079 -25.912 -1.05 0.008 Drier(42-138) GPP ~ rain site 20 121.3 0.971 0.616 1.28 0.011 Wetter(139-516) GPP ~ temp site 1 108.5 -1.812 15.143 0.94 0.005 Wetter(139-516) GPP ~ rain site 1 108.5 1.296 -0.008 0.00 -0.005 Wetter(139-516) GPP ~ temp site 20 123.7 -0.581 0.447 0.02 -0.001 Wetter(139-516) GPP ~ rain site 20 123.7 3.459 0.037 0.64 0.001

193

Appendix 4 Supplementary material for Chapter 5

Table S4.5. Average results for 500 linear models run on subsamples of environmentally similar, unique site-pairs, where pairwise differences in gross primary productivity (response, GPP, gC m-2 month-1 2001-2012) are a function of pairwise differences in median site rainfall and temperature niche widths (predictor, mm and °C respectively, 1976- 2005). All site-pair differences are ordered by the predictor variable, so that the predictor is always positive. ΔGPP is the allowable difference in predicted GPP between site-pairs, based on a negative exponential model of site GPP as a function of rainfall, showing subsampling results for two different ΔGPP values (1, 20). ΔT is the allowable temperature difference between site-pairs, set at 0.5°C. N is the average number of site-pairs used at each ΔGPP across

2 500 subsamples, Int is the average linear intercept, β1 is the average linear coefficient, T is the average t-statistic and R is the average adjusted R2. P is the fraction of 1000 bootstrapped t-statistics — from one unique subsample under the null hypothesis — which were more extreme than the average t-statistic from all 500 subsamples. GPP had no meaningful relationship with a combined measured of niche area (derived by scaling the environmental values from the 1976-2005 surfaces, then calculating the median site of π × scaled rainfall niche width × scaled temperature niche width), and we do not report the results.

2 Precip. subset (mm) Model ΔGPP N Int β1 T R P All(42-516) GPP ~ temp niche 1 186.5 -0.978 -9.739 -3.18 0.048 0.004 All(42-516) GPP ~ rain niche 1 186.5 -2.528 0.037 5.62 0.142 0.001 All(42-516) GPP ~ temp niche 20 245.2 -1.113 -14.136 -4.87 0.087 <0.001 All(42-516) GPP ~ rain niche 20 245.4 -0.977 0.031 4.76 0.083 0.001 Drier(42-138) GPP ~ temp niche 1 77.8 1.672 -10.266 -2.35 0.058 0.009 Drier(42-138) GPP ~ rain niche 1 77.7 -1.506 0.050 5.58 0.281 <0.001 Drier(42-138) GPP ~ temp niche 20 121.2 -0.881 -14.595 -3.45 0.087 <0.001 Drier(42-138) GPP ~ rain niche 20 121.3 -2.056 0.049 5.06 0.171 <0.001 Wetter(139-516) GPP ~ temp niche 1 108.4 -2.776 -10.264 -2.10 0.035 0.022 Wetter(139-516) GPP ~ rain niche 1 108.4 -1.919 0.016 1.65 0.019 0.051 Wetter(139-516) GPP ~ temp niche 20 123.7 -1.606 -13.605 -3.34 0.080 <0.001 Wetter(139-516) GPP ~ rain niche 20 123.7 0.156 0.014 1.64 0.017 0.045

194

Appendix 4 Supplementary material for Chapter 5

Table S4.6. Average results for 500 linear models run on subsamples of environmentally similar, unique site-pairs, where pairwise differences in gross primary productivity (response, GPP, gC m-2 month-1) are a function of pairwise differences in mean site rainfall and temperature niche widths (predictor, mm and °C, respectively). All site-pair differences are ordered by the predictor variable, so that the predictor is always positive. ΔGPP is the allowable difference in predicted GPP between site-pairs, based on a negative exponential model of site GPP as a function of rainfall, showing subsampling results for two different ΔGPP values (1, 20). ΔT is the allowable temperature difference between site-pairs, set at 0.5°C. N is the average number of site-pairs used across 500 subsamples, Int is the average

2 2 linear intercept, β1 is the average linear coefficient, T is the average t-statistic and R is the average adjusted R . P is the fraction of 1000 bootstrapped t-statistics — from one unique subsample under the null hypothesis — which were more extreme than the average t-statistic from all 500 subsamples.

2 Precip. subset (mm) Model ΔGPP N Int β1 T R P All(42-516) GPP ~ temp niche 1 186.5 -2.599 -8.723 -2.67 0.033 0.002 All(42-516) GPP ~ rain niche 1 186.5 -2.375 0.047 5.25 0.126 <0.001 All(42-516) GPP ~ temp niche 20 245.2 0.426 -15.285 -4.66 0.080 <0.001 All(42-516) GPP ~ rain niche 20 245.3 -0.641 0.038 4.62 0.079 <0.001 Drier(42-138) GPP ~ temp niche 1 77.7 -0.957 -9.557 -2.03 0.041 0.012 Drier(42-138) GPP ~ rain niche 1 77.9 -1.466 0.059 5.22 0.255 0.001 Drier(42-138) GPP ~ temp niche 20 121.3 0.513 -16.323 -3.42 0.086 <0.001 Drier(42-138) GPP ~ rain niche 20 121.2 1.014 0.055 4.79 0.156 0.001 Wetter(139-516) GPP ~ temp niche 1 108.4 -4.483 -7.143 -1.42 0.015 0.077 Wetter(139-516) GPP ~ rain niche 1 108.5 -0.369 0.016 1.01 0.004 0.084 Wetter(139-516) GPP ~ temp niche 20 123.7 -0.638 -12.767 -2.81 0.057 0.004 Wetter(139-516) GPP ~ rain niche 20 123.7 -0.486 0.016 1.34 0.011 0.085

195

196