Raquel Garcia.Pdf

UNCERTAINTY IN PROJECTED IMPACTS OF CLIMATE CHANGE ON BIODIVERSITY

A FOCUS ON AFRICAN VERTEBRATES

RAQUEL A. GARCIA University of Copenhagen | 2014

Uncertainty in projected impacts of climate change on biodiversity A focus on African vertebrates

Raquel A. Garcia PhD thesis | January 2014 University of Copenhagen | Faculty of Science

ACADEMIC ADVISORS Prof Miguel B. Araújo CMEC, University of Copenhagen, Denmark Imperial College, Silwood Park, UK National Museum of Natural History–CSIC, Spain InBio/CIBIO, Évora University, Portugal

Dr Mar Cabeza Metapopulation Research Group, University of Helsinki, Finland

SUBMITTED TO The PhD School of The Faculty of Science, University of Copenhagen, Denmark January 2014

ASSESSMENT COMMITTEE Prof Jane Hill University of York, Department of Biology, UK

Dr Richard Pearson University College London, Research Department of Genetics, Evolution and Environment, UK

Dr David Nogués-Bravo CMEC, University of Copenhagen, Denmark

COPYRIGHT © 2014 Raquel A. Garcia (Synopsis, Design) © 2012 Blackwell Publishing Ltd (Chapter I) © 2014 The Authors Journal of Biogeography Published by John Wiley & Sons Ltd (Chapter II) © 2014 The Authors (Chapter III) © 2014 The Authors (Chapter IV) © 2014 The Authors (Chapter V)

Contents | iii

Contents

Preface v Summary / Resumé vii Acknowledgements x

Synopsis 1 The uncertain nature of assessments under climate change 5 Variation in data and model decisions 9 Robustness to ecological assumptions 11 Broadening the scope of assessments 13 Embracing uncertainty 15 References 18

Chapter I 27 Exploring consensus in 21st century projections of climatically suitable areas for African vertebrates

Chapter II 65 Matching species traits to projected threats and opportunities from climate change

Chapter III 87 Conservation implications of omitting rare and threatened species from climate change impact modelling

Chapter IV 115 Multiple dimensions of climate change and implications for biodiversity

Chapter V 147 Do projections from bioclimatic envelope models and climate change metrics match?

Preface | v

Preface

This thesis is the result of a four-year PhD project based at the Center for Macroecology, Evo- lution and Climate at the University of Copen- hagen, in Denmark, and the National Museum of Natural History (CSIC), in Madrid, Spain. The project was supervised by Prof Dr Miguel Araújo and Dr Mar Cabeza. The thesis work also included various stays at the University of Hel- sinki in Finland, the 'Rui Nabeiro' Biodiversity Chair at Évora University in Portugal, and the South African National Biodiversity Institute in Cape Town, South Africa. The work was funded by a PhD grant from the Portuguese Foundation for Science and Technology. The thesis consists of two parts. The first part is a synopsis describing the background and aims of the thesis, summarising the main findings, and discussing the work presented in a broader perspective. The second part consists of five chapters that form the core of the thesis. They include four analytical chapters and one review chapter. At the time of submission, two were published as scientific articles, and the remaining three were in review.

Summary / Resumé | vii

Summary

Evidence for shifts in the phenologies and dis- need to both integrate uncertainties in assess- tributions of species over recent decades has ments, and reduce or circumvent them where often been attributed to climate change. The possible. Integration of uncertainties is illus- prospect of greater and faster changes in cli- trated in two examples. The first uses ensem- mate during the 21st century has spurred a bles of bioclimatic envelope models for sub- stream of studies anticipating future biodiver- Saharan African vertebrates, built with alterna- sity impacts. Yet, uncertainty is inherent to tive climate data and model algorithms. En- both projected climate changes and their effects semble forecasting provides a means for ex- on biodiversity, and needs to be understood ploring the breadth and spatial variation of before projections can be used. This thesis uncertainties, and for building consensus seeks to elucidate some of the uncertainties among projections. Several consensus method- clouding assessments of biodiversity impacts ologies are compared here, including a newly from climate change, and explores ways to proposed methodology that preserves informa- address them. While the focus is mostly on sub- tion about the variability of projections in the Saharan African vertebrates, the methodologi- ensemble. The second example examines model cal advances and conclusions presented are far- outputs for sub-Saharan African amphibians in reaching and have wider relevance. the light of species' vulnerability to climate Throughout the chapters in this thesis, pro- change. An analytical framework is developed jections under changing climates for sub- for distinguishing between different climatic Saharan African vertebrates, based on biocli- threats and opportunities revealed by the bio- matic envelope models, are shown to be af- climatic envelope models, and analysing how fected by multiple uncertainties. Different they each are altered by the consideration of model algorithms produce different outputs, as specific response-mediating traits. do alternative future climate models and sce- Efforts to reduce uncertainties in biodiver- narios of future emissions of greenhouse gases. sity impact assessments are equally important. Another uncertainty arises due to omission of However, many sources of uncertainty cannot species with small sample sizes, which are easily be reduced, not least the omission of difficult to model. The effect of such bias species that are narrow-ranging, poorly known, against narrow-ranging species is often over- or even unknown to science. This uncertainty looked in assessments of biodiversity impacts, can instead be circumvented through the use of but our results for sub-Saharan African am- alternative approaches to assessing impacts. phibians show that it trickles down to conser- This thesis discusses one candidate approach vation strategies. Finally, assumptions about that is independent of species' data: the use of the climatic tolerance of species, their dispersal climate change metrics. By describing the expo- ability, and other characteristics are also shown sure of regions to multiple changes in the mag- to alter model projections for sub-Saharan nitude, timing, position, or availability of cli- African amphibians. matic conditions, metrics can provide infer- Despite numerous calls to address the un- ences about the potential threats and opportu- certainty challenge, appropriate treatment of nities for the biodiversity in those regions. The uncertainty has yet to be formalised in assess- diversity of existing metrics is reviewed here, ments of biodiversity impacts under climate and the picture that emerges is one of multifac- change. The chapters in this thesis highlight the eted changes in climate, with unequal spatial viii | Summary / Resumé

patterns around the world. To help interpret The uncertainties discussed in this thesis, the diversity of climate change metrics, a con- and many others not covered here, impair the ceptual framework is proposed for using them conservation of biodiversity under changing in biodiversity impact assessments. Early test- climates in Africa and elsewhere. Explicitly ing of this framework, by comparing inferences addressing all uncertainties of projected im- from metrics and from the bioclimatic envelope pacts appears overwhelming. Yet, if model models for sub-Saharan African amphibians, projections are to be useful for conservation suggests that climate change metrics might be a planners, they must be as transparent as possi- useful addition to the biodiversity impact as- ble by including an honest description of their sessment toolbox. level of confidence given the current knowledge.

Resumé

De seneste årtiers evidens for ændringer i konklusioner mere vidt-rækkende og har bred arters fænologi og geografiske udbredelser er relevans. ofte blevet tilskrevet klimaforandringer. En gennemgående tendens i denne Udsigten til hurtigere og mere omfattende afhandling er at forudsigelser for afrikanske ændringer i klimaet i det 21. århundrede har hvirveldyr, baseret på bioclimatic envelope affødt en strøm af studier, der søger at models (klima-baserede modeller for en arts klarlægge de forventede konsekvenser for udbredelse) under klimaforandringer, er fremtidens biodiversitet. Dog ligger der en påvirkede af flere usikkerheder. Forskellige iboende usikkerhed i forudsigelser af model-algoritmer giver forskellige resultater, klimaforandringer, og deres indflydelse på og det gør alternative klimamodeller og biodiversiteten, som det er nødvendig at afklare scenarier for fremtidige drivhusgas- før disse forudsigelser kan anvendes. Denne udledninger også. En anden usikkerhed opstår afhandling sigter mod at udrede nogle af de som følge af at arter med små sample-sizes usikkerheder der forplumrer en vurdering af udelukkes, fordi de er svære at modellere. klimaforandringernes indflydelse på Effekten af denne skævvridning væk fra arter biodiversiteten, og udforsker måder at med små udbredelser overses ofte i håndtere dem på. Selv om fokus hovedsagligt vurderinger af konsekvenser for ligger på afrikanske hvirveldyr syd for Sahara, biodiversiteten, men vores resultater for er de metodologiske fremskridt og padder syd for Sahara viser at der er en effekt,

Summary / Resumé | ix

der afspejles helt ned i konkrete strategier for udbredelser, og som man ved meget lidt eller naturbevarelse. Endelig vises det at antagelser intet om. Denne usikkerhed kan istedet omgåes om arters klimatiske tolerance, spredningsevne ved at anvende alternative tilgange. Denne og andre egenskaber ændrer model- afhandling diskuterer én mulig tilgang, som er forudsigelser for afrikanske padder. uafhængig af artsdata: at anvende specifikke Til trods for adskillige opfordringer til at enheder til at måle klimaforandringer. Ved at håndtere den udfordring som usikkerhed beskrive et områdes eksponering for ændringer udgør, så mangler der stadig en passende i klimatiske forholds udbredelse, timing, formaliseret tilgang til at behandle usikkerhed i placering, eller tilgængelighed, kan sådanne vurderingerne af klimaforandringernes måleenheder bruges til at udlede noget om de konsekvens for biodiversiteten. Kapitlerne i potentielle trusler mod, og muligheder for, denne afhandling understreger biodiversiteten i de pågældende regioner. En nødvendigheden af både at integrere gennemgang af den diverse gruppe af usikkerheden i vurderinger, og at reducere eller eksisterende klimaforandringsmål viser et omgå den når det er muligt. Integration af billede af multifaceterede klimaforandringer usikkerhed illustreres med to eksempler. Det med forskellige rumlige mønstre rundt om i første gør brug af 'ensembles' verden. Som hjælp til at fortolke disse (sammenstillinger) af flere bioclimatic envelope forskellige klimaforandringsmål foreslåes her models for afrikanske vertebrater syd for en konceptuel ramme for hvordan de kan Sahara, konstrueret med alternative klimadata bruges til at vurdere påvirkninger på og model-algoritmer. Ensemble-forudsigelse gør biodiversiteten. Indledende tests af denne det muligt at udforske bredden og den rumlige ramme, foretaget ved at sammenligne variation i graden af usikkerhed, og at etablere fortolkninger af klimaforandringsmål med en konsensus mellem forudsigelser. Her bioclimatic envelope models for afrikanske sammenlignes flere tilgange til at etablere padder syd for Sahara, peger på at konsensus mellem forudsigelser, inklusive en klimaforandringsmål kan blive endnu et nyligt foreslået metode der bevarer information brugbart værktøj til at vurdere påvirkningen om variabiliteten mellem de enkelte på biodiversiteten. forudsigelser i ensemblet. Det andet eksempel De usikkerheder som denne afhandling undersøger model-resultater for afrikanske omhandler, og adskillige andre der ikke er padder syd for Sahara, set i lyset af hver arts omtalt her, vanskeliggør beskyttelsen af sårbarhed overfor klimaforandringer. Der biodiversiteten mod klimaforandringer i Afrika udvikles en analytisk ramme for adskillelse af og andre steder. Det kan forekomme de klimatiske trusler og muligheder der påvises overvældende at skulle imødegå alle disse af bioclimatic ensemble models, og for analyse af usikkerheder eksplicit, men hvis forudsigelser hvordan disse ændrer sig når respons- baseret på modeller skal være nyttige for medierende træk inkorporeres. naturbevarelsen skal de være så gennemsigtige Det er også vigtigt at søge at reducere som muligt. Dette kan opnås ved at inkludere usikkerheden i vurderingen af påvirkninger på en beskrivelse af forudsigelsernes grad af biodiversiteten. Der er dog mange kilder til usikkerhed, som er ærlig og baseret på den usikkerhed der ikke let lader sig reducere, ikke nyeste viden. mindst udeladelsen af arter der har små

x | Acknowledgements

Acknowledgements

This was not a lone journey. Throughout the During the four years I was privileged to four years of my PhD there were various people work in five labs and to meet many incredible pointing me to new paths, encouraging me to people and researchers, young and senior alike. look forward, and reminding me to slow down I kept my status of 'nomad', constantly moving when I needed. between labs and trying to catch the summers My deepest gratitude goes to my supervi- of both hemispheres. Yet, all five labs felt like sors Miguel Araújo and Mar Cabeza. Thank you 'home' to me, for there was always a warm both for trusting from the start (and some- welcome from my colleagues in each one of times, along the way, more than myself) that I them. Gracias to my colleagues at the Museum could do this. Miguel, thank you for instilling in in Madrid for their support during my first year me the capacity to work independently. Part of of science (and Spanish language). Tak to the my research developed because your answers people at CMEC in Copenhagen for their conta- to my questions often broadened, instead of gious enthusiasm for science. Obrigada to all at narrowing, the options. If my 'consultancy style' the Biodiversity Chair in Évora for the warm writing has improved to something closer to environment I always found there (outdoors as scientific writing, I owe it to you. Getting back well). Kiitos to the Metapopulation group in the manuscripts all marked in red was often Helsinki, ever dynamic and helpful. And thanks dispiriting, yet always so helpful. Thank you for to all at SANBI in Cape Town, for their inspiring finding the time to discuss science, advise me connection to nature. My gratitude also goes to on career matters, or simply have a cup of tea. the leaders of these groups for keeping the The barbecues in Alpedrete or Évora, and the doors open (Miguel Araújo, Carsten Rahbek, picnics in the parks of Copenhagen will be Ilkka Hanski, Mar Cabeza, and Guy Midgley), greatly missed. Thank you Miguel, Sonia, Ágata and to all the staff for their support. and Quino for your friendship and company Special thanks to my colleagues who have away from science. made this journey easier in so many different Mar, thank you for teaching me so much ways. It would have been a lot harder to learn about science, the ethics of science, and the fun about statistics or GIS, start writing in R, delve of making science. Above all, thank you for into all the new concepts in ecology, and learn being an inspiration to stay in science. I so en- to read and criticise scientific papers without joyed our discussions about species ranges, Diogo Alagador, Hedvig Nenzén, Márcia Bar- traits and climate in your garden, in the green- bosa, Maria Triviño, Carla Pinto Cruz, Regan house, or while walking Ada to sleep. You have Early, Heini Kujala, Laura Meller, Joaquín gently guided me and helped me to grow as a Hortal, Guida Santos, Pep Serra Díaz, François scientist and as a person – thank you for caring Guilhaumon, Dora Neto, Katie Marske, David about my scientific development beyond pub- Nogués-Bravo, Paco Ferri, Raul García-Valdéz, lishing papers, and for pushing me to continue Sara Varela, Marisa Peláez, Isaac Pozo, Jan going after my dreams. Only someone who Dempewolf, and each and every one in the five knows me well could have found the 'yellow labs! A warm thank you to Phoebe Barnard and cottage' for one of my summer stays in Helsinki John Measey for taking me out to the field, to all – the best house I've lived in during these four in the 'Reserve Selection Journal Club' in Hel- years! Thank you Mar, Tero and Ada for being sinki and in CMEC/iBiochange workshops and the best neighbours, cooking palls, and friends. retreats for many fruitful discussions, and to

Acknowledgements | xi

Res Altwegg for the 'Stats Lunches'. Thanks also Andrea, Mar and Tero, Mats and Sofie, Guida to Heini Kujala, Mar Cabeza, Joaquín Hortal, and Joaquín, Agi and Andy, Maya and Rudi, Jon, Regan Early, Hedvig Nenzén, and Katharine Rean and Tania, Gayane, Manuel, and Virginia Marske for helpful comments on the synopsis, for making my work stays in Portugal, Den- and to Anna-Sofie Stensgaard and Michael mark, Finland, Spain, Switzerland, South Africa, Borregaard for translating the summary into US, New Zealand, and Mozambique so much Danish. more enjoyable. For sharing ideas and making my journey Many have been on this journey with me, so much more rewarding, I thank all those I near or far, but always present. Thank you to all have worked with in manuscripts or projects my precious friends and their beautiful fami- during my PhD. Thanks to all the co-authors in lies: Sol, Joana and Murat, Sofia, Ana, Agi, the papers in this thesis: Mar and Miguel, Wen- Naoko, Zé, Cuncas, Sara and João, Carla, Cristi- dy Foden, Neil Burgess, Carsten Rahbek, Chris- na, Pedro, Diogo M., Eilat and Shaul, Jon, Maya tian Hof, Louis Hansen, Phil Platts, Alexander and Rudi, Linda, Emma, Sergio and Elizabeth, Gutsche, and Res Altwegg. It has also been (and João Carlos, Cláudio, Maral and Mark, Tania and will continue being) fun collaborating with Rean, Andrea, Thierry, Kian, Mats, Philip, Burcu, Jessica Forrest, Guy Midgley, Phoebe Barnard, Ania, Firmas and Ana, Anália, Checa, Patrícia V., and all in the PARCC, HarmBio, RLI, and AECID/ Serpa, Manuel, Patrícia and Pedro, Ana Luísa, Mozambique teams. Thanks to Neil Burgess, Guida and Joaquín H., Márcia, Diogo A., Dora, Miguel Araújo and Mar Cabeza for inviting me Joaquín C., Hedvig, Maria, Helen S., Helen C-B., into new collaborations. and many others. Thank you all for being a Besides Miguel and Mar, I had a few other source of encouragement and fun! official trainers. I am grateful to Richard Pear- Stepping out of the journey every now and son, Steven Phillips, Wilfried Thuiller, Nick then into the yoga classes with Esmé de Wet, Zimmermann, Antoine Guisan, Jens-Christian Elena Mironov, and other wonderful instructors Svenning, David Nogués Bravo, Daniel Kissling, was a life saver. Namaste. Being a child again Signe Normand, and Rob Anderson for teaching while playing with all my nephews and nieces me about the illusions and disillusions of mod- was equally relaxing and even more fun! For els; to Luís Carrascal for making statistics seem their smiles and for reminding me of what is less daunting; to José Alexandre Diniz-Filho for important in life, I thank my goddaughters Mia decrypting (a little) phylogenetics; to Joaquín and Ona Adalís. Hortal for teaching me to question the data; and And, always, thanks to my family. Obrigada to Jan and Camilla for opening up the fascinat- aos meus pais e irmã pelo apoio em todas as ing world of pedagogy to me. Then my turn minhas aventuras com ou sem pára-quedas, came. Teaching the practical lectures about dentro ou fora de Portugal. Aos meus pais, models in an international PhD course in Co- Nanda e David, obrigada por me ensinarem penhagen was one of the most gratifying expe- tanto mesmo sem darem conta. Obrigada à Inês riences during the PhD. Thank you, David, for por estar sempre presente, como amiga e irmã. trusting me, and all the students in the 2012 A toda a família, e aos regulares nos almoços de and 2013 classes for making my job so easy and domingo, obrigada por sorrirem e por me for teaching me so much. fazerem rir. As any nomad, I am grateful to many people for hosting me throughout these four years. This work was funded by the Portuguese Thanks to Inês and Zé, Carla and Luís, Patrícia Foundation for Science and Technology and Pedro, Sol and Jorge, Joana and Murat, (grant SFRH/BD/65615/2009).

'There is something fascinating about science. One gets such wholesale returns of conjecture out of such a trifling investment of fact.'

Mark Twain, Life on the Mississippi, 1883 Synopsis

Uncertainty in projected impacts of climate change on biodiversity - A focus on African vertebrates

RAQUEL A. GARCIA

Synopsis Uncertainty in projected impacts of climate change on biodiversity – A focus on African vertebrates

RAQUEL A. GARCIA1,2,3

Africa's savannahs support a higher diversity of gesting a link between the changes observed in ungulate species than anywhere else in the nature and climate3–6, and attributing it to an- world. It is hard to imagine an Africa without thropogenic climate change7. Faced with the the springbok, the kudu, the giraffe, or the sable prospect of increasing rates of climate change antelope. The Kruger National Park in north- during the 21st century8–11, attention in the eastern South Africa is home to large popula- research community then turned to projecting tions of ungulates, but the numbers of some future potential impacts on biodiversity2. For species began dwindling in about 1988. Re- some of the ungulate populations in the Kruger, searchers Joseph Ogutu and Norman Owen- Ogutu and Owen-Smith projected severe de- Smith examined the data from wildlife census clines by 2016, and even extirpation, should the conducted between 1977 and 1996, and offered same dry season trends persist1,12. As in any a possible explanation for the declines of eleven other field, however, projections are inherently ungulate species1. Extreme reduction in dry uncertain. For projections of the effects of cli- season rainfall, together with warmer tempera- mate change on biodiversity to be improved tures, had led to a decrease in the amount of and judiciously used, it is crucial to appreciate green forage inside the park, while boundary their uncertainty. This thesis is a contribution fencing prevented the animals from finding to understanding some of the uncertainties vegetation elsewhere. clouding familiar assessments of the future of Population declines such as those recorded biodiversity under changing climates, and to for Kruger's ungulates, as well as phenological exploring ways to address them. The focus is shifts and species' distributional changes ob- mostly on sub-Saharan African vertebrates, but served in multiple regions, prompted the first the methodological advances and over-arching studies in climate change ecology2. Together, conclusions are more broadly applicable. these studies created a body of evidence sug-

4 | Synopsis

Box 1 | Glossary Vague or ambiguous language introduces uncertainty in assessments of climate change impacts on biodiversity (Table 1). The terminology of climate change ecology and modelling is constantly evolving, with some terms not yet established and others still contentious. This glossary lists some of these terms, and describes the meaning used for each throughout this thesis, with the aim of enhancing clarity for the reader.

Bioclimatic envelope model: correlative model that uses associations between observations of species' occurrences and climate to define the sets of climatic conditions under which species are likely to maintain viable populations – the Grinnellian niche of species16,17. There is contention on which term best describes what is modelled given the data used and the underlying assumptions. Alternative terms include 'species distribution model', 'ecological niche model', and 'habitat suitability model'18–22. In Chapter III, the term 'species distribution model' is used as a synonym (but see a proposal to distinguish between the two23). Climate change metric: measure of changes in climatic parameters (e.g., temperature, precipitation) in a given site (cell) or across broad regions (sets of cells). Different dimensions of climate change can be measured, including changes in the magnitude, timing, availability, and position of climatic conditions. Ensemble forecasting: modelling framework in which multiple simulations across more than one set of initial conditions, model classes, parameters, and boundary conditions are generated, each combination representing a possible state of the system being forecast24. Ensembles can be combined to generate consensus projections, with the aim of summarising agreement among the individual projections. Exposure: 'the extent of climate change likely to be experienced by a species or locale'25. Future climate projections applied to bioclimatic envelope models yield estimates of exposure of species to climate change, whereas climate change metrics estimate the exposure of regions to climate change and allow for inferences to be made about the level of exposure of the biodiversity in such regions. Narrow-ranging species: species with small geographical ranges. Here, used to refer to species with small numbers of occurrence records such that statistical modelling becomes unviable. Narrow-ranging species in this sense are likely to include the majority of the species typically regarded as 'rare'26 – this term replaces 'narrow-ranging species' in Chapter III, with rarity in this case defined by a relative cutoff point based on eligibility for modelling. Risk: the magnitude and probability associated with the effect of an adverse event on species or populations, for example, by causing their decline, reduced fitness, genetic loss, or extinction. It requires consideration of both threat and vulnerability27. Threat: extrinsic human and natural adverse events occurring in a given area within a given time27. Opportunities brought by extrinsic events can also be considered28. Climatic threats are the focus of the thesis, but non-climatic threats such as those arising from land-use change would fall under the same definition. Trait: 'any morphological, physiological or phenological feature measurable at the individual level, from the cell to the whole-organism level, without reference to the environment or any other level of organization'29. Often replaced with characteristics of species and their ranges which are expected to summarise relevant (yet unavailable) traits. Uncertainty: The lack of certainty or precision. A state of having limited knowledge where it is impos- sible to exactly describe the existing state, a future outcome, or more than one possible outcome. Vulnerability: the intrinsic susceptibility of species to threats27. Here it includes both i) the degree to which species or populations are affected by changes in climate depending on factors such as ecophysi- ology, life history, and microhabitat preferences, and ii) their capacity to adapt to changes through phenotypic plasticity, genetic diversity, evolutionary rates, life history traits, and dispersal and colonization ability25. Whereas exposure refers to the extrinsic threats facing species, vulnerability captures the intrinsic capacity of species to respond.

Uncertainty in projected impacts of climate change | 5

The uncertain nature of assessments certainty), vague or ambiguous expression of under climate change that knowledge (linguistic uncertainty), and subjective human values and preferences asso- Assessing the potential impacts of climate ciated with the system (decision-making uncer- change on biodiversity relies on different ap- tainty)37,42–44. These uncertainties pervade the proaches spanning a continuum from correla- entire modelling process. They affect the three tive to mechanistic models13. At the correlative components of modelling frameworks defined end of the continuum, bioclimatic envelope by Austin45: decisions regarding the generation models14,15 (see Glossary in Box 1) use the asso- of data used in the modelling (what Austin calls ciation between known species' occurrences the 'data model'), decisions regarding the actual and climate to characterise the sets of suitable modelling process and evaluation (the 'statisti- climatic conditions for species (or the realised cal model' in Austin's terminology as applied to subsets thereof). By projecting this characteri- correlative approaches, or, more generally sation to the future, these models describe here, the 'mathematical model'), and the eco- potential changes in the geographical patterns logical knowledge and theory used (the 'eco- of climatic suitability for species, and provide logical model'). Building upon a number of estimates of species' exposure to climate published classifications of uncertainty change. At the opposite end of the continuum, sources37,41–44, this thesis uses a classification mechanistic models explicitly include processes that is based on Austin's components. Three such as the species' physiological constraints broad sources of uncertainty are thus distin- and plastic acclimation capacity30, and the guished, affecting either the 'data model', the demographic dynamics31 underlying the intrin- 'mathematical model', and the 'ecological sic vulnerability of species to climate change. model' (Table 1). Between the two ends of the continuum, ap- First, the data used in assessments are im- proaches of varying complexity combine bio- perfect, inherently complex, and often carry the climatic envelope models with available data uncertainties of the models that such data, in that are expected to characterise the vulner- turn, are based upon. Bioclimatic envelope ability of species32. For example, data on the models rely heavily on species' distributional dispersal ability or physiological climatic toler- data. The quality of such data is affected by ance of species have been used to constrain errors in the identification or location of spe- bioclimatic envelope model projections33,34. cies, by geographical, temporal or taxonomical Whereas the ecological information needed to biases in the sampling, and by the interpolation parameterise mechanistic models is lacking for or expert-based techniques used to map distri- most known species, climate and species' oc- butions46–49. By contrast, mechanistic models currence data of varying quality have become rely on detailed physiological or life-history widely available. As a result, the last decades data that are lacking for most species, and that have seen a surge of studies using bioclimatic are typically affected by intra-specific variation, envelope models to assess how species might subjective judgement by experts, inconsistent become exposed to ongoing and future changes terminology, and the techniques used to fill in climate2,35,36. data gaps41,50,51. Not only are species' geo- Irrespective of the approach used, assess- graphical distributions and biology poorly un- ing the potential impacts of climate changes on derstood (a problem termed the 'Wallacean biodiversity is fraught with uncertainty37–41. shortfall'52), most species on Earth are yet to be Assessments are uncertain because of incom- described53,54 (the 'Linnean shortfall'52). As- plete knowledge of the system (epistemic un- sessments of climate change impacts on biodi-

6 | Synopsis

Table 1 | Major sources of uncertainty in assessments of the impacts of future climate change on biodiversity. Examples are given for three major sources of uncertainty associated with the data used, the mathematical model, and the underlying ecological knowledge. The list is not meant to be exhaustive, but to capture some of the major sources that are relevant in the context of this thesis. For each example, the arrows point to the extreme of the correlative (left) to mechanistic (right) model continuum that is most affected, and the roman numerals refer to the chapters of this thesis where the examples are discussed. 'Data model' uncertainty: decisions about the biological and climatic data used in the modelling C ↔ M Chapter Error, bias or subjective judgement in species occurrence data ← Error, bias, or subjective judgement in mechanistic data → II Omission of species → III Error in interpolated baseline climate data ↔ Error in future climate projections ↔ I Unknown future greenhouse gas emissions ↔ I Error in downscaled climate data ↔ 'Mathematical model' uncertainty decisions about the calibration, evaluation and projection methods C ↔ M Chapter Choice of algorithm ← I Methods for predictor selection, data partitioning, absence estimation, etc. ↔ Choice of evaluation data and technique ↔ Small numbers of species' occurrence ← III Extrapolation to non-analogue conditions ← I Choice of threshold for deriving binary projections ← 'Ecological model' uncertainty decisions about ecological concepts and assumptions underpinning the models C ↔ M Chapter Missing or inadequate biological components ← II Missing or inadequate climate parameters and dimensions, and response forms ↔ IV Missing non-climatic factors ↔ Species-climate equilibrium assumption ← Spatial or temporal scale mismatch/ inadequacy ← Lack of clarity about the object and output of models ← II

versity are thus also crippled by omission of (Box 2). Uncertainty from projections of future many of Earth's species. The deficiencies and climate pervades all assessments and is associ- gaps in biological data are not geographically or ated with assumptions regarding the future taxonomically55,56 random, with the tropics and concentrations of greenhouse gases, imperfect threatened species particularly afflicted (Box understanding of the relevant physical proc- 2). esses, errors in the modelling process, and Uncertainty in climate data is associated with a natural variability58. A host of alternative At- multitude of sources affecting both observa- mosphere-Ocean General Circulation Models tions and projections57,58. Characterisation of (hereafter climate models) are available61, with the current climate relies on interpolated data multiple realisations and under different sce- from weather stations or satellites, with areas narios of future emissions of greenhouse poorly covered59 and spatially heterogeneous60 gases62,63. Most such models are developed at showing decreased data reliability and lower coarse resolutions, but climate data are increas- performance of bioclimatic envelope models ingly being downscaled to resolutions that are

Uncertainty in projected impacts of climate change | 7

Box 2 | The data challenge in the tropics Tropical regions contain most of the world’s biodiversity, but are also where the knowledge of species is poorest. The tropics hold the highest rate of discoveries of new mammal species over the past two decades65, with large numbers of further descriptions expected for Africa up to 203265. Tropical species are also more often narrow-ranging than species in temperate zones66, and are thus more often difficult to model with familiar correlative methods requiring sufficient sample sizes67. The resulting bias against narrow-ranging species in assessments may even be larger than is readily apparent, because such species are less likely to be described currently68. Data quality for known species also poses a challenge55,69 in the tropics, and across most of Africa. For vertebrates at the sub-Saharan African scale, the database held at the Zoological Museum in Den- mark70 (Chapter I) is the most comprehensive collation from multiple sources, albeit not free from sampling bias towards the most biodiverse areas71. Most regional efforts towards systematic data collection and treatment of uncertainty focus on South Africa or Southern Africa. The data generated through the Southern African Bird Atlas Project72 and analysed with occupancy models73 is a case in point. Knowledge of species' traits is also deficient in the tropics, as illustrated by the lower mean number of data entries in the mammalian dataset PanTHERIA that originate in the tropics56, and the disproportionate number of tropical amphibians without data in the Global Amphibian Assessment74. The data challenge in the tropics extends to climate observations and projections. Weather station coverage is poorer in tropical regions75. In comparison to mid-latitudes, the future climates of the tropics, and Africa in particular, are still not fully understood due to more complex systems with strong ocean-atmosphere coupling11,75,76. Precipitation projections are generally less certain, with those for tropical Africa subject to especially large decadal variability relative to the mean precipitation77. In turn, temperature increases relative to natural internal variability are expected to be larger in the tropics than mid-latitudes11, although with high uncertainty associated with future emissions of greenhouse gases78. Reasons linked to wealth, communication and security have been suggested to explain deficiencies in species' data79, and possibly also explain poor climate data. Lack of basic data is reflected in a biased fingerprint of ecological effects of climate change4–6, where evidence is sparse for the tropics in general, and for most African countries. Likewise, existing efforts to anticipate future impacts in the tropics are crippled by lack of data, with most published studies focusing on temperate regions80.

more relevant for conservation planning64. Yet, also introducing uncertainty in assessments. In downscaled products are still constrained by climate change and biological invasion applica- the uncertainty of the coarse-scale data that tions in particular, bioclimatic envelope model drive them39,64,81, as well as new uncertainties projections are affected when models are introduced by the methodologies and assump- forced to extrapolate to climatic conditions tions of downscaling techniques82. outside the range of conditions used to cali- Second, the different mathematical models that brate the models84,89. Different algorithms vary are used in impact assessments each also carry in how they handle truncated functions de- their own methodological uncertainties. In scribing the association between climate and bioclimatic envelope modelling, decisions re- suitability for species, causing model outputs to garding the choice of algorithm83,84, estimation vary. Whereas mechanistic models are more of absence data85,86, and thresholds for trans- robust to extrapolation90, they are subject to forming modelled probabilities of suitability other uncertainties associated with the meth- into projections of species' presence and ab- ods and parameters required to handle their sence87,88, for example, all lead to considerable increased ecological complexity41. variation in model outputs. Species with small The third source of uncertainty in assess- numbers of occurrence records are dificult to ments is associated with the underlying eco- model with familiar correlative techniques67, logical decisions. Knowledge of the causal

8 | Synopsis

mechanisms of species' occurrences is incom- synopsis, and discussed in the broader context plete, including knowledge of relevant aspects of the uncertainty challenge in the last section. of climate91,92 described at adequate spatial and Specifically, the following questions are ad- temporal scales64,93,94; non-climatic threats such dressed: as land-use change and landscape fragmenta- tion95; population, ecophysiological, and evolu- i) How can bioclimatic envelope model uncer- tionary dynamics96–98; and biotic interactions99. tainty be integrated in assessments of the While many such mechanisms are being inte- potential impacts of climate change on biodi- grated in models, uncertainty-free projections versity? are unrealistic, not least because of the com- Chapter I104 compares the uncertainty arising plexity and stochasticity inherent to the eco- from alternative climate projections and bio- logical37 and climatic58 systems. On the correla- climatic envelope model algorithms in assess- tive end of the model continuum, the focus is on ments for sub-Saharan African vertebrates, and the patterns – not the mechanisms – in the explores ways to summarise uncertainty by association between observed species' distribu- building consensus around ensembles of mod- tions and climate13. However, uncertainty can els. In Chapter II105, the focus turns to the un- arise when omitted mechanisms cause such certainty of model projections for sub-Saharan models to depart from working assumptions of African amphibians due to ecological assump- equilibrium between species' distributions and tions. The chapter presents an analytical climate15,100. In addition, lack of clarity about framework for teasing apart projected climatic the ecological underpinnings of bioclimatic threats and opportunities facing species and envelope models pervades most practical ap- examining their robustness to assumptions plications21,101,102, partly because there is still about species' vulnerability. debate about the terminology for describing the assumptions, methods and outputs18–22,103. Such ii) Can assessments of the potential impacts uncertainty can lead to inadequate interpreta- of climate change become more inclusive of tion and misuse of outputs beyond their in- Earth's biodiversity despite incomplete tended purpose102. knowledge of species? This thesis investigates some of the above The omission of species with small sample sizes sources of uncertainty (Table 1), and discusses in bioclimatic envelope modelling studies has ways to address them. Focusing on bioclimatic hitherto been largely overlooked as a source of envelope models, the work presented here uncertainty when projecting the effects of cli- explores how projections are influenced by mate change on biodiversity. Chapter III dem- imperfect climate data, omission of species, the onstrates the conservation implications of such choice of model algorithm, and ecological as- bias against narrow-ranging species of sub- sumptions. Much of the analytical component of Saharan African amphibians, underscoring the the thesis is centred on sub-Saharan African importance of complementing bioclimatic enve- vertebrate species, illustrating some of the lope models with alternative tools that are crucial challenges for assessments in tropical scalable to all species. One such alternative tool and sub-tropical areas of high biodiversity and is the use of simple metrics of climate change as poor knowledge (Box 2). Nevertheless, the proxies for the threats and opportunities facing uncertainties addressed are far-reaching and wholesale biodiversity. Climate change metrics the work should have wider relevance. The are reviewed and illustrated at the global scale thesis is composed of five chapters, summa- in Chapter IV, and compared to bioclimatic rised in the following three sections of this envelope models for sub-Saharan African vertebrates in Chapter V.

Uncertainty in projected impacts of climate change | 9

Variation in data and model decisions data, and thresholds for binary proj- ections88,91,115–121. Although these studies varied Across the wealth of studies assessing the ef- in methodology as well as geographical and fects of future climate change, the uncertainty taxonomic scope, the choice of model algorithm surrounding model projections is often only was consistently identified as the major cause partially addressed or even overlooked. Previ- of uncertainty whenever included in the com- ous continental-scale applications of biocli- parison (with one exception121). None of these matic envelope models to African species106–110 comparative studies applied to Africa or to have considered only limited variation in future multiple taxonomic groups, a gap that was filled climate models, greenhouse gas emissions by Chapter I. Here too, the comparison of mul- scenarios, or bioclimatic envelope model algo- tiple projections of species' temporal rithms. Yet, the effect of some of these and turnover122 resulting from models built under other sources of uncertainty on model projec- different assumptions identified the model tions is likely to be important, as illustrated for algorithms as the largest relative contributor to sub-Saharan African vertebrates throughout overall uncertainty, followed by the climate this thesis (see Box 3 for an overview of uncer- models and emissions scenarios. tainties for amphibians in particular). The bio- The contribution of the input climate data climatic envelope models used in this thesis to the uncertainty around Chapter I projec- (with the exception of Chapter III) are de- tions was larger in southern areas of sub- scribed in Chapter I104. For over 2,500 species Saharan Africa (Box 3). In previous studies of sub-Saharan African birds, mammals, am- projecting the impacts of climate change, phibians, and snakes71, models were built at one treatment of climate-model uncertainty has degree resolution (≈ 111 km at the Equator), varied between building multiple bioclimatic using baseline data111 and projections81,112 for envelope model projections with a suite of three climatic predictors: mean temperatures available climate models109,114 or, less fre- of both warmest and coldest months, and an- quently, applying a multi-model consensus nual precipitation. Seven different algorithms, projection123. The rationale behind consensus is three multi-model climate projections, and that combining an ensemble of projections, three emissions scenarios were used, generat- assumed to be independent and representative ing an ensemble of projections. of the breadth of possible states of the system Ensembles of models built with alternative being forecast, yields lower mean error than 24 assumptions offer a way to quantify the un- any of the individual projections24,124. Building certainty of model projections. The multiple consensus is challenging not only because the projections in an ensemble are seen as plausi- required assumptions are difficult to ble representations of the system under study, meet58,125,126, but also because there is still de- and together express, at least partially, the bate on the best methodology to combine pro- breadth of uncertainty. Ensemble forecasting is jections126. Averaging across entire ensembles common in climatology58, and has more re- of projections is frequently done, but implies cently started to be used in bioclimatic enve- the loss of information about extreme 109,113,114 lope modelling . Previous studies have states126,127. To allay the criticism levelled at examined ensembles of models to compare the simple averages, Chapter I uses a consensus relative importance of uncertainties arising methodology that first groups co-varying cli- from different sources, including the choice of mate models before summarising the central algorithms, climate models, emissions scenar- tendency for each group113. Among the suite of ios, methods for collinearity correction or vari- 17 climate models, those that were closest to able selection, variable sets, species occurrence

10 | Synopsis

Box 3 | Uncertainties for sub-Saharan African amphibians under climate change Throughout this thesis, different sources of uncertainty are shown to affect projections for sub-Saharan African amphibians under climate change (Fig. 1). Model uncertainties are exemplified by the variation in algorithms (Chapter I). The ecological uncertainty is illustrated by the variation between the raw projections and projections that were modified a posteriori by estimates of species' vulnerability according to their dispersal ability and tolerance to climate changes (Chapter II). Climate data uncertainty is shown by the variation across three averaged clusters of climate models, and across three emissions scenarios (Chapter I). Finally, data uncertainty arising from the omission of narrow-ranging species (Chapter III) is revealed by comparing two hypothetical projections, one assuming loss of climatic suitability for all (omitted) species and the other assuming continued suitability. Chapter I presents a spatial assessment of the relative importance of three sources of uncertainty, based on Analysis of Variance (ANOVA) performed for each grid cell116,118 and using the temporal turnover of species122 as response variable. The results show that algorithm-related uncertainty was the largest contributor to overall uncertainty, followed by the variation in climate models and emissions scenarios. When this assessment was updated here to include all five sources of uncertainty examined in this thesis, the ecological uncertainty as estimated here was found to contribute more to overall uncertainty than the variation across algorithms (median [and lower and upper quartile] proportions of the total sum of squares across the study area of 48.0 [30.2–63.4] and 30.4 [17.3–46.7], respectively). Figure 1 | Uncertainties in projections for African amphibians under climate change. Bioclimatic envelope models for 263 sub-Saharan African amphibians were used to project changes in the probability of climatic suitability by mid- century. Projected median absolute (positive and negative) changes are shown for a multi- model climate projection under emissions scenario A1B, and reflecting the median across seven algorithms (a). The coeffi- cient of variation is shown across projections for alternative algorithms (b), assumptions about the climate change vulnerability of species (c), emissions scenarios (d), climate models (e), and assumptions about the impacts on narrow-ranging species excluded from the modelling (f). the median model in sign, magnitude and spa- input data for the modelling, an advantage that tial pattern128 were grouped together and aver- is not trivial in the face of the increasing num- aged. The result was a set of three consensus ber of climate projections available61. climate projections for use in the modelling In contrast to climate data uncertainty, the exercise, retaining the variability of the initial variation in Chapter I projections obtained ensemble. At the same time, the consensus with different bioclimatic envelope model algo- solution presented in Chapter I (subsequently rithms was largest in the Sahel and the South- used in another application129) simplifies the ern Sahara. Algorithm-related uncertainty has

Uncertainty in projected impacts of climate change | 11

previously been associated with the type of sions (Chapter I), they may also be more or species' occurrence data used (presence only less robust to assumptions about ecological versus presence-absence data) and the assump- processes. Notably, consideration of biotic tions made during extrapolation to non- interactions133,134, the dispersal ability of spe- analogous, or novel, climates84. In the case of cies135,136, and the fundamental climatic toler- the model projections built in Chapter I, algo- ances of species33,34,137 have all been previously rithm-related uncertainty affected areas where shown to influence model projections under future temperatures were expected to fall climate change. Such processes are usually above the range of calibration, and was thus omitted in classical bioclimatic envelope mod- possibly due to model extrapolation. Model els, but they are crucial when assessing biodi- projections for these areas are less reliable, as versity risk under climate change. When direct are likely to be any projections in other tropical incorporation of ecological data in mechanistic and sub-tropical regions where emergence of models is not viable, bioclimatic envelope non-analogous climates is expected130 (see model projections can be contrasted with alter- Chapter IV). native assumptions about the vulnerability of The variation in outputs from different al- species138,139. Exposure and vulnerability esti- gorithms has sometimes been summarised by mates are often also combined into indices that building consensus projections109,113,114, but are meant to describe the climate change risk bioclimatic modellers, like climatologists, are facing species138,140. Chapter II105 presents an still debating what is the best methodology to analytical framework to examine projections of combine projections. Chapter I extends previ- changes in climatic suitability with reference to ous comparisons of consensus meth- estimates of species' vulnerability to such odologies24,131,132 by exploring a comprehensive changes. In contrast to previous studies com- suite of five alternative methodologies that paring exposure to vulnerability, Chapter II either measure the central tendency in the argues that the different extrinsic threats and frequency distribution of all projections (mean, opportunities arising from the exposure of weighted mean, and median), measure the species to climate change should be examined central tendency among groups of co-varying separately, and matched to specific intrinsic projections (as above for climate models), or traits that are likely to mediate species' re- select the projection summarising the highest sponses. amount of variation among projections (using In published studies using bioclimatic en- Principal Components Analysis). These five velope models, the degree of species' exposure methodologies were compared by examining to climate change is commonly inferred from their accuracy in predicting the baseline distri- temporal changes in the total size of areas with bution of species. Consensus projections for suitable climatic conditions107,139,141. Species are sub-Saharan African vertebrates were generally considered 'winners' or 'losers' depending on more accurate – and more similar to each other whether such areas become more or less avail- – than individual projections, with the consen- able across a given region. However, such sus method based on co-varying groups yield- summary measures conceal different climatic ing the highest accuracy. opportunities as well as threats, each imposing specific constraints on species28. At the level of each grid cell, climatic suitability may be lost, Robustness to ecological assumptions gained, or become markedly different from neighbour cells, with consequences not only for Not only are bioclimatic envelope model out- the size, but also the position and level of frag- puts sensitive to data and methodological deci-

12 | Synopsis

Box 4 | Teasing apart climatic threats and opportunities The threats and opportunities arising from exposure of species to climate change are normally inferred from bioclimatic envelope models (Chapter II), but can also be gauged with climate change metrics (Chapters IV and V). With the former, the geographical projection of species' bioclimatic envelope is compared across time, and changes in area occupied, distance shifted, and degree of fragmentation can be measured. In turn, climate change metrics depict the spatio- temporal dynamics of climate across broad regions, independently of species' data. Metrics can describe changes in the area occupied by analogous climatic conditions, or in the distance to such conditions, allowing for indirect inferences to be made about the potential implications for the biodiversity in such regions (Chapter IV). Different combinations of the threats and opportunities, measured with models or metrics, have distinct conservation implications (Fig. 2).

Figure 2 | Climatic threats and opportunities for species. Geographical representation of changes in areas of suitable climate for species between two time steps, with reference to different scenarios of threats and opportunities from future climate change inferred from bioclimatic envelope models or measured changes in climate (light and dark blue scales respectively, see Chapter V). Combinations of threats and opportunities relating to both the size and position of areas with given climatic conditions result in different risks for species occupying such areas (panels (a-d)). Using size as sole criterion would lead to classifying species (a) and (b) as 'winners', and species (c) and (d) as 'losers'. However, examining also the changes in position reveals that the future suitable areas for species (b) and (d) are displaced, leaving no overlap with the baseline areas. As a result, species (b) and (d) should also be flagged, particularly if they are poor dispersers. These species thus require the protection of new areas and enhanced connectivity to reach them, whereas species (a) and (c) depend on protection in situ. The axis for area could be decomposed into losses and gains, and a third axis could be added for the threat of increased fragmentation.

mentation of species' bioclimatic envelopes areas of climatic suitability. Previous frame- across broader regions (Box 4). These compo- works28,34 separated the threat of loss from nents of exposure are likely to have distinct opportunities for gains. Chapter II adds spatial distributions and implications for con- fragmentation, a threat to the persistence143 servation guidance142, but are seldom teased and expansion144 of species' ranges that has apart. Using the bioclimatic envelope models hitherto been largely overlooked in climate built in Chapter I for sub-Saharan African am- change ecology (but see a recent example145). phibians as illustration, Chapter II teases apart The different threats and opportunities are projected losses, fragmentation and gains of also likely to interact with specific traits. The

Uncertainty in projected impacts of climate change | 13

specificity in the interaction between extrinsic analysis were the narrowest-ranging species in threats and intrinsic vulnerability has been the database, which were not modelled in previously shown for non-climatic threats such Chapter I (see Chapter III). Indeed, the analy- as habitat destruction or hunting146–149, but has sis presented in Chapter II for sub-Saharan not been sufficiently addressed in climate African amphibians illustrates the data uncer- change ecology. Risk indices used to inform tainty that is typical of tropical regions (Box 2). priorities for conservation under future cli- Not only did model projections exclude narrow- mates150, for example, frequently combine ranging species, but data for traits in the strict summary measures of exposure with summary sense were also unavailable and had to be re- estimates of vulnerability according to available placed with proxies, carrying their own uncer- trait data, irrespective of the potential specific tainty to the analysis. interactions between the two. The analytical framework proposed in Chapter II aims to select specific response-mediating traits for Broadening the scope of assessments each climatic threat and opportunity, based on The uncertainties discussed in Chapters I and theoretical and empirical research on climate II applied to projections of the exposure of sub- change vulnerability151 (see also collated refer- Saharan African vertebrates to future climate ences in Chapter II). Thus, for example, pro- changes. As already alluded to above, such jected losses of climatic suitability can be as- projections were not available for all species. sessed for their robustness to traits influencing Excluded were species with fewer than 15 one- the species' physiological climatic tolerance, degree gridded occurrence records, considered whereas projected gains of newly suitable areas too uncertain to model due to their small sam- can be contrasted with traits associated with ple size67. The modelling thus covered only the dispersal ability or reproductive output of 67% of the 4,092 vertebrate species – and a species. mere 38% of the 741 amphibian species – in For the sub-Saharan African amphibians the African vertebrate database held at the analysed in Chapter II, spatially overlaying Zoological Museum in Denmark70. Far from projected losses, fragmentation or gains of unique to the studies presented in this thesis, climatically suitable areas with vulnerability the bias against narrow-ranging species per- classifications derived from selected trait vades most modelling exercises under climate data140,152 illustrates how interpretation of change. This often overlooked source of uncer- outputs from bioclimatic envelope models can tainty is the focus of Chapter III and a motiva- be altered with consideration of species' cli- tion behind Chapters IV and V, where an alter- mate change vulnerability (Box 3). The differ- native approach is proposed to broaden the ences revealed underscore the need to clearly scope of assessments. communicate in assessments what is being Using sub-Saharan African amphibians as a modelled and what the outputs mean. The case study, Chapter III demonstrates the impli- modifying effect of vulnerability on model pro- cations that omitting narrow-ranging species in jections was stronger for the species in the projections of the impacts of climate change can study sample that were most exposed to cli- have for conservation priorities. The analysis is mate change. For example, for the amphibians based on a new modelling exercise, but the with the smallest geographical overlap between results presented hold for the models of Chap- baseline and future climate envelopes, pro- ter I. The bioclimatic envelope modelling per- jected gains may be unrealisable due to poor formed in Chapter III left out the species with dispersal ability (Box 4). Excluded from the fewer than 10 one-degree gridded occurrence

14 | Synopsis

records, amounting to over half of all species in existing schemes and priorities derived for the the dataset and including the vast majority of modelled species decreased into the future, threatened species. Multivariate ordination suggesting that the use of bioclimatic envelope analysis showed that the subset of modelled models to inform conservation priority setting species is not representative of either the pre- may systematically downplay important areas sent climate space of all known sub-Saharan for rare and threatened species. Irrespective of African amphibians or their future exposure to the actual impacts of ongoing climate changes climate changes. In comparison to modelled on such rare species, most of them face other species, those that were omitted occupy topog- non-climatic, more immediate threats that raphically complex areas with cooler, wetter warrant priority in conservation strategies163. and less seasonal climates, which are projected The results of Chapter III underscore the to experience lower climate anomalies. Previ- importance of complementing models with ous studies also found that climatically and alternative, more generally applicable tools. topographically diverse or distinct regions tend One such alternative tool is the use of simple to contain a disproportionate richness of nar- metrics of climate change, which quantify the row-ranging species153,154. Looking into the exposure of geographical areas to changes in future (Box 3), topographical diversity may climate parameters. Metrics have previously confer climatic stability155 and buffer the effects helped address a diversity of ecological ques- of broader scale climate change93. Yet, other tions such as the potential risks from future dimensions of climate change, including the climate changes to biodiversity130,153,164 and risk of disappearing climates, may especially conservation areas165–167, suggesting that they affect species with narrow climatic tolerances can be a useful tool in impact assessments168,169. such as those omitted from the modelling130,153 However, the variety of existing metrics, and (see below for a discussion of the different their ecological implications, have hitherto not dimensions of climate change reviewed in been fully appreciated. Chapter IV reviews Chapter IV). climate change metrics and classifies them into Bioclimatic envelope models have sup- metrics of local change and regional change. ported analyses of the role of existing conserva- Climate anomalies are the most commonly used tion schemes under changing climates156,157. metric, depicting average changes in climate at However, the range size bias introduced in such the locality (grid cell) level130. Other local met- analyses is counter to traditional frameworks rics capture changes in the magnitude of ex- for identifying priority sites for conservation, treme climates164, the timing of climatic which focus precisely on those taxa most likely events170, or the local velocity of climate167. to be omitted from the modelling158,159. Previ- In turn, regional metrics characterise ous studies160 have shown that including in- changes in the distribution of climatic condi- formation about unknown or poorly sampled tions over broader geographical areas (sets of species influences the selection of reserves. In cells). These metrics begin with the characteri- the case of the sub-Saharan African amphibians sation of climatic conditions across a given examined in Chapter III, conservation priori- region, and can then describe temporal changes ties derived using the modelled species subset in the availability of analogous climatic condi- exhibited lower spatial congruence with exist- tions across that region130,171,172, as well as ing conservation priority schemes (Biodiversity changes in the direction to, or distance be- Hotspots161, Endemic Bird Areas158 and The tween, the positions of analogous conditions171. Global 200162) than those inferred from the Chapter IV highlights the contrasting spatial omitted species subset. Congruence between patterns arising at the global scale from differ-

Uncertainty in projected impacts of climate change | 15

ent climate change metrics. Whereas polar or even extinctions, of species' ranges over regions face reductions in the global availability time182. Such changes often match, with time of similar climatic conditions, the tropics and lag, the increases and decreases in the area of hot arid regions are exposed to average different climatic conditions brought about by changes beyond historical inter-annual vari- the Earth's warming and cooling cycles183–185. At ability, emergence of novel climates, and in- the same time, measured shifts in the position creased frequency of extreme climates. of climatic conditions have been used to explain To help interpret the diversity of metrics, the opportunity for species to track climatic Chapter IV proposes a conceptual framework conditions across broad regions in both palaeo- for classifying metrics into common currencies cological185 and recent3,5 times. of climatic threat and opportunity for species. To test the feasibility of the framework The framework is based on principles linking proposed in Chapter IV for guiding the use of the persistence of populations and species to metrics, Chapter V compares climate change local and regional climatic suitability, respec- metrics for sub-Saharan Africa with bioclimatic tively173. It builds upon published empirical envelope models built in Chapter I for verte- studies associating different aspects of climate brates. For all taxa, local climate anomalies, as change with observed ecological changes in projected by the metrics, were diagnostic of recent or palaeocological time, to establish changes in climatic suitability of grid cells for potential links between climate change metrics species, as projected by the models. In turn, and threats and opportunities for biodiversity. changes in the area and position of analogous Some of the empirical associations previously climates, as measured by the metrics, were found are highlighted here, and further de- indicative of changes in the size and position of scribed in Chapter IV. species' bioclimatic envelopes, respectively. At the local level, decreased climatic suit- The agreement found between metrics and ability can have effects on the physiology, mor- models was stronger for the vertebrate species phology or behavior of the organisms in a with narrower climatic breadths, which depend population174, potentially leading to changes in more strongly on tracking climates130,153 than demography. Reduced activity of individuals, species with more generalist climate prefer- population declines, and mortality have been ences. Nevertheless, the results in Chapter V linked to metrics of gradual175,176 or extreme demonstrate that, when carefully implemented changes177 in local climate. Positive cases of and interpreted, metrics can provide first ap- increased fitness or abundance have also been proximations to the same extrinsic threats and reported178,179, although less frequently. In turn, opportunities typically captured by bioclimatic observed shifts in the timing of seasonal activi- envelope models. ties of populations have been associated with measured changes in the seasonality of climate180. Where dispersal is involved, the Embracing uncertainty velocity of temperature changes has been sug- This thesis illustrates a suite of critical sources gested to pace the rate of population shifts of uncertainty in assessments of biodiversity across the local topography181. impacts under changing climates. Unknown At the regional level, the spatio-temporal future climates (Chapter I), omission of species dynamics of climate can affect the availability (Chapter III), the diversity of model algorithms and distribution of climatically suitable areas (Chapter I), and species-specific vulnerability for species173 (see also Box 4). Long-term eco- to climate change (Chapter II), are shown to logical data reveal expansions and contractions, affect bioclimatic envelope model projections

16 | Synopsis

for sub-Saharan African vertebrates. These There are many other ways to integrate uncer- uncertainties, and many others not covered tainty in assessments, which are not discussed here, impair the conservation of biodiversity in this thesis. Comparing the outputs from dif- under changing climates in Africa and else- ferent models along the correlative-mechanistic where. Despite numerous calls to explicitly continuum, each with their own uncertainties, address this challenge39,41,47,83,84,131, appropriate can help highlight areas of convergence and treatment of uncertainty has yet to be formal- offset uncertainties189–191. Precision estimates ised in impact assessments and conservation for the data used in bioclimatic envelope mod- adaptation. The chapters in this thesis highlight els can be visualised in maps accompanying the need to both integrate uncertainties in model projections47, or incorporated directly in assessments, and reduce or circumvent them the modelling process192,193. For the species' where possible. distributional dataset used in Chapter I and Firstly, precise projections are unattain- many other datasets, explicit information on able, and existing uncertainties should be in- the sampling effort associated with species' corporated in assessments and quantified to occurrence records is lacking. But even in such the extent possible41,186. In answering the first cases, alternative approaches based on qualita- question posed at the outset, this thesis shows tive, objective differences in survey intensity examples of how bioclimatic envelope model across broad, nested regions49 can be used to uncertainty can be integrated in assess- detect potential biases. ments. Building ensembles of models with Secondly, understanding the uncertainty in alternative assumptions is illustrated in Chap- projections helps to direct efforts to reduce it ter I. Here, as in other published studies, the where most needed. Reducing uncertainty is an use of ensembles is constrained by the breadth integral part of ecology and any science. Models of possible future states sampled, but can, at are 'educated guesses about the future'39, and least, give a lower bound on the range of uncer- increasing the precision (and accuracy) of the tainty. Probabilistic forecasting, where site- guesses is a key goal in climate change ecology. specific probability density functions for mod- Narrowing the guesses based on bioclimatic elled suitability account for several uncertain- envelope models rests partly on improving the ties, has only rarely been applied in bioclimatic data on species' occurrences. Efforts towards envelope modelling187,188. When ensembles are enhanced, targeted sampling194 are warranted used to build consensus projections, these particularly in tropical regions (Box 2), where should preserve information about uncer- possible combining observational, experimen- tainty41, for example by relying on methodolo- tal and mathematical lines of research25. En- gies that capture the breadth of the ensemble hancing the precision of future climate projec- (Chapter I). tions, in turn, is seen by climatologists195 as Chapter II gives another example of how bringing smaller benefits than understanding uncertainty can be integrated in assessments, the influence of existing uncertainties on pro- by examining a posteriori the effect of estimates jected impacts. In fact, despite continuous de- of species' vulnerability on model projections. velopment of climate models and greater com- The proposed framework carries several sim- putational capacity since the Fourth Assess- plifying assumptions that require further scru- ment Report of the Inter-governmental Panel tiny. Nevertheless, its application can reveal on Climate Change, model spread has not where, in geographical space, the model projec- changed much with the new generation of tions may be less robust to crucial ecological models195–197. factors not contemplated in the modelling.

Uncertainty in projected impacts of climate change | 17

More important are perhaps any efforts to be influential201. Yet, increased realism comes reduce the ecological uncertainty arising from at the expense of applicability to many species. the choice of climatic predictors91. Assessments Mechanistic models have only been empirically ignoring influential climate parameters (e.g., tested on few species with available data, and temperature, precipitation) have previously even simple bioclimatic envelope models are been shown to misjudge potential biotic re- not suited to the narrowest-ranging species sponses92. Equally crucial is to measure the (Chapter III). What is more, both correlative relevant dimensions of change in selected pa- and mechanistic models are blind to the major- rameters (e.g., means, extremes, or timing of ity of species on Earth, still unknown54. climatic events)198, an issue that was empha- The second question posed in this thesis sised in Chapter IV by the diversity of spatial asked how assessments can become more patterns, and their potential links to biotic inclusive of Earth's biodiversity. Describing threats, across dimensions. Changes in climate the distributions and auto-ecology of species, extremes, shifts in the timing of climatic events, and modelling their responses, must remain a and climate change velocity, for example, are priority, but gathering data for all species on less frequently considered in bioclimatic enve- Earth is impractical. What is suggested here is lope modelling exercises, but can be integrated that investment in models be accompanied by in the models themselves198 or complement the development of alternative approaches that assessments. Even in large scale, multi-species circumvent the uncertainty associated with modelling exercises such as that in Chapter I, species' data. Climate change metrics are one more ecologically sound choices can be made candidate approach (Chapter IV). Early testing by first grouping species that are likely to be suggests that metrics can provide inferences influenced by similar predictors. What then about extrinsic climatic threats and opportuni- remains a key challenge – in Chapter I and ties in agreement with bioclimatic envelope most other applications199 – is the mismatch in model projections (Chapter V), and metrics spatial scale between available climate predic- should now be compared with available time- tors and species' occurrences64. Whereas coarse series species' data. The use of metrics does not and uncertain global or regional climate models alleviate the need for considering the intrinsic may be useful to inform broad strategies, con- vulnerability of species, but can give first indi- servation planning at fine scales199 may require cations on the potential extrinsic threats facing greater certainty from models of topographic biodiversity. Clearly, providing conservation microclimate200. planners with assessments that are useful re- In an effort to further reduce the ecological quires the integration of multiple approaches. uncertainty of projections, the field of climate In closing, it must be acknowledged that change ecology is becoming more integrated, explicitly addressing all uncertainties of pro- harnessing new disciplines and types of jected impacts appears overwhelming. Yet, if data25,182 to describe the potential behaviour of model projections are to help manage the fu- the ecological system. Complex models are ture impacts of climate change on Earth's bio- valuable for guiding species-based conserva- diversity, they must be as transparent as possi- tion, while also advancing ecological theory. ble by including an honest description of their Indeed, the new uncertainties carried by com- abilities and limitations. Such description must plex models, and even by simplified frame- guide end-users through the ecological mean- works such as that presented in Chapter II, can ing and purpose of the models, and reveal the stimulate new research to identify the circum- level of confidence associated with the outputs stances under which different mechanisms may (Box 3). Projections thus presented can still be

18 | Synopsis

useful to guide decision-making, while avoiding declines among African savanna ungulates. Ecology Letters 6, 412–419 (2003). the risks of over-interpretation and loss of 2. Nabout, J. C. et al. Trends and Biases in Global credibility should additional data or new mod- Climate Change Literature. Natureza & els contradict them in the future. In a rapidly Conservação 10, 45–51 (2012). evolving field as climate change ecology, such 3. Parmesan, C. & Yohe, G. A globally coherent risks are real and could undermine our efforts fingerprint of climate change impacts across natural systems. Nature 421, 37–42 (2003). to adapt to climate change. Ultimately, factoring 4. Parmesan, C. Ecological and Evolutionary 202 uncertainties into decision-making requires Responses to Recent Climate Change. Annual collaboration with social and political scien- Review of Ecology, Evolution, and Systematics 37, tists. Equipped with ensembles of projections, 637–669 (2006). model users have the option of weighing the 5. Chen, I.-C., Hill, J. K., Ohlemüller, R., Roy, D. B. & Thomas, C. D. Rapid Range Shifts of Species risks associated with alternative decisions39,195. Associated with High Levels of Climate Warming. Risk-averse users may opt for the most pessi- Science 333, 1024–1026 (2011). mistic scenario, whereas the middle-term or 6. Rosenzweig, C. et al. Attributing physical and biological impacts to anthropogenic climate the consensus projection may be adequate in a change. Nature 453, 353–357 (2008). 203 risk-tolerant context . When planning for a 7. Root, T. L., MacMynowski, D. P., Mastrandrea, M. D. range of outcomes is viable, assessments under & Schneider, S. H. Human-modified temperatures multiple, plausible hypotheses about future induce species changes: Joint attribution. Proceedings of the National Academy of Sciences of impacts can lead to better conservation deci- the United States of America 102, 7465–9 (2005). 204 sions and may be preferable. 8. IPCC. Climate Change 2001: The Scientific Basis. The picture that emerges from decades of Contribution of Working Group I to the Third climatic research is one of drastic and multifac- Assessment Report of the Intergovernmental Panel on Climate Change. (Cambridge University eted changes in climate. There is still much to Press, 2001). understand about the climate, the distributions 9. IPCC. Climate change 2007: The physical science and biology of species, and the ability of species basis. Contribution of working Group I to the Fourth Assessment Report of the to cope with climate changes and interacting Intergovernemental Panel on Climate Change. threats163,205. In the face of all uncertainties, (Cambridge University Press, 2007). protecting biodiversity is a daunting task. But 10. Peters, G. P. et al. The challenge to keep global are imprecise projections flawed beyond use? warming below 2°C. Nature Climate Change 3, 4–6 (2013). Uncertainty needs to be understood, yet it 11. IPCC. Summary for Policymakers. in Climate should not deter our efforts to anticipate the Change 2013: The Physical Science Basis. impacts of climate change on biodiversity39,195. Contribution of Working Group I to the Fifth Rather, the projections that are attainable with Assessment Report of the Intergovernmental Panel on Climate Change (Stocker, T. F. et al.) the current knowledge should help direct fur- (Cambridge University Press, 2013). 206,207 ther research to priority regions , and 12. Barnosky, A. D. Heatstronke - Nature in an age of prompt us to adapt conservation strategies to Global Warming. 269 (Island Press, 2009). account for uncertain impacts. Sacrificing some 13. Dormann, C. F. et al. Correlation and process in precision to generality may well be needed if species distribution models: bridging a dichotomy. Journal of Biogeography 39, 2119–2131 (2012). we are to rise to the challenge of protecting 14. Guisan, A. & Zimmermann, N. E. Predictive habitat Earth's biodiversity. distribution models in ecology. Ecological Modelling 135, 147–186 (2000). 15. Peterson, A. T. et al. Ecological Niches and Geographic Distributions. (Monographs in References Population Biology, Princeton University Press, 2011). 1. Ogutu, J. O. & Owen-Smith, N. ENSO, rainfall and temperature influences on extreme population 16. Grinnell, J. The Niche-Relationships of the California Thrasher. The Auk 34, 427–433 (1917).

Uncertainty in projected impacts of climate change | 19

17. James, F. C., Johnston, R. F., Wamer, N. O., Niemi, G. 33. Elith, J., Kearney, M. & Phillips, S. The art of J. & Boecklen, W. J. No TitleThe Grinnellian Niche modelling range-shifting species. Methods in of the Wood Thrush. The American Naturalist 124, Ecology and Evolution 1, 330–342 (2010). 17–47 (1984). 34. Arribas, P. et al. Evaluating drivers of vulnerability 18. Kearney, M. Habitat, environment and niche: what to climate change: a guide for insect conservation are we modelling? Oikos 115, 186–191 (2006). strategies. Global Change Biology 18, 2135–2146 (2012). 19. Elith, J. & Leathwick, J. R. Species Distribution Models: Ecological Explanation and Prediction 35. Siqueira, T., Padial, A. A. & Bini, L. M. Mudanças Across Space and Time. Annual Review of Ecology, climáticas e seus efeitos sobre a biodiversidade: Evolution, and Systematics 40, 677–697 (2009). um panorama sobre as atividades de pesquisa. Megadiversidade 5, 17–26 (2009). 20. Warren, D. L. In defense of “niche modeling.”Trends in Ecology & Evolution 27, 497– 36. Guisan, A. et al. Predicting species distributions for 500 (2012). conservation decisions. Ecology Letters 16, 1424– 1435 (2013). 21. McInerny, G. J. & Etienne, R. S. Ditch the niche – is the niche a useful concept in ecology or species 37. Elith, J., Burgman, M. A. & Regan, H. M. Mapping distribution modelling? Journal of Biogeography epistemic uncertainties and vague concepts in 39, 2096–2102 (2012). predictions of species distribution. Ecological Modelling 157, 313–329 (2002). 22. Hortal, J., Lobo, J. M. & Jiménez-Valverde, A. Basic Questions in Biogeography and the (Lack of) 38. Barry, S. & Elith, J. Error and uncertainty in habitat Simplicity of Species Distributions: Putting Species models. Journal of Applied Ecology 43, 413–423 Distribution Models in the Right Place. Natureza & (2006). Conservação 10.2, 108–118 (2012). 39. Wiens, J. A., Stralberg, D., Jongsomjit, D., Howell, C. 23. Anderson, R. P. Harnessing the world’s A. & Snyder, M. A. Niches, models, and climate biodiversity data: promise and peril in ecological change: Assessing the assumptions and niche modeling of species distributions. Annals of uncertainties. Proceedings of the National Academy the New York Academy of Sciences 1260, 66–80 of Sciences 106, 19729–19736 (2009). (2012). 40. Heikkinen, R. K. et al. Methods and uncertainties in 24. Araújo, M. B. & New, M. Ensemble forecasting of bioclimatic envelope modelling under climate species distributions. Trends in Ecology & change. Progress in Physical Geography 30, 751– Evolution 22, 42–47 (2007). 777 (2006). 25. Dawson, T. P., Jackson, S. T., House, J. I., Prentice, I. 41. Beale, C. M. & Lennon, J. J. Incorporating C. & Mace, G. M. Beyond predictions: biodiversity uncertainty in predictive species distribution conservation in a changing climate. Science 332, modelling. Philosophical transactions of the Royal 53–8 (2011). Society of London. Series B, Biological sciences 367, 247–58 (2012). 26. Gaston, K. J. Rarity. (Chapman and Hall, 1994). 42. Regan, H. M., Colyvan, M. & Burgman, M. A. A 27. Araújo, M. B. & Williams, P. H. Selecting areas for taxonomy and treatment of uncertainty for species persistence using occurrence data. ecology and conservation biology. Ecological Biological Conservation 96, 331–345 (2000). Applications 12, 618–628 (2002). 28. Thomas, C. D. et al. A framework for assessing 43. Ascough II, J. C., Maier, H. R., Ravalico, J. K. & threats and benefits to species responding to Strudley, M. W. Future research challenges for climate change. Methods in Ecology and Evolution incorporation of uncertainty in environmental and 2, 125–142 (2011). ecological decision-making. Ecological Modelling 29. Violle, C. et al. Let the concept of trait be 219, 383–399 (2008). functional! Oikos 116, 882–892 (2007). 44. Kujala, H., Burgman, M. A. & Moilanen, A. 30. Kearney, M. & Porter, W. Mechanistic niche Treatment of uncertainty in conservation under modelling: combining physiological and spatial climate change. Conservation Letters 6, 73–85 data to predict species’ ranges. Ecology letters 12, (2012). 334–50 (2009). 45. Austin, M. P. Spatial prediction of species 31. Keith, D. A. et al. Predicting extinction risks under distribution: an interface between ecological climate change: coupling stochastic population theory and statistical modelling. Ecological models with dynamic bioclimatic habitat models. Modelling 157, 101–118 (2002). Biology Letters 4, 560–563 (2008). 46. Hortal, J., Jiménez-Valverde, A., Gómez, J. F., Lobo, 32. Willis, S. G. et al. A framework methodology for J. M. & Baselga, A. Historical bias in biodiversity integrating climate change vulnerability inventories affects the observed environmental assessments from Species Distribution Models and niche of the species. Oikos 117, 847–858 (2008). Trait-based Approaches for use in systematic conservation planning. In preparation.

20 | Synopsis

47. Rocchini, D. et al. Accounting for uncertainty when Bulletin of the American Meteorological Society 93, mapping species distributions: The need for maps 485–498 (2011). of ignorance. Progress in Physical Geography 35, 62. Nakicenovic, N. & Swart, R. Special Report on 211–226 (2011). Emissions Scenarios: A Special Report of Working 48. Kujala, H., Vepsäläinen, V., Zuckerberg, B. & Group III of the Intergovernmental Panel on Brommer, J. E. Range margin shifts of birds Climate Change. 599 (Cambridge University Press, revisited - the role of spatiotemporally varying 2000). survey effort. Global change biology 19, 420–30 63. Moss, R. H. et al. The next generation of scenarios (2013). for climate change research and assessment. 49. Barbosa, A. M., Pautasso, M. & Figueiredo, D. Nature 463, 747–56 (2010). Species–people correlations and the need to 64. Wiens, J. A. & Bachelet, D. Matching the Multiple account for survey effort in biodiversity analyses. Scales of Conservation with the Multiple Scales of Diversity and Distributions 19, 1188–1197 (2013). Climate Change. Conservation Biology 24, 51–62 50. Nakagawa, S. & Freckleton, R. P. Missing inaction: (2010). the dangers of ignoring missing data. Trends in 65. Ceballos, G. & Ehrlich, P. R. Discoveries of new ecology & evolution 23, 592–596 (2008). mammal species and their implications for 51. Fitzsimmons, J. M. How consistent are trait data conservation and ecosystem services. Proceedings between sources? A quantitative assessment. of the National Academy of Sciences 106 , 3841– Oikos 122, 1350–1356 (2013). 3846 (2009). 52. Lomolino, M. V. Conservation Biogeography. in 66. Wiens, J. J., Graham, C. H., Moen, D. S., Smith, S. a & Frontiers of Biogeography: new directions in the Reeder, T. W. Evolutionary and ecological causes geography of nature (Lomolino, M. V. & Heaney, L. of the latitudinal diversity gradient in hylid frogs: R.) 293–296 (Sinauer Associates, Inc., 2004). treefrog trees unearth the roots of high tropical diversity. The American naturalist 168, 579–96 53. May, R. M. & Beverton, R. J. H. How Many Species? (2006). [and Discussion]. Philosophical Transactions of the Royal Society of London. Series B: Biological 67. Stockwell, D. R. B. & Peterson, A. T. Effects of Sciences 330 , 293–304 (1990). sample size on accuracy of species distribution 54. Scheffers, B. R., Joppa, L. N., Pimm, S. L. & models. Ecological Modelling 148, 1–13 (2002). Laurance, W. F. What we know and don’t know 68. Patterson, B. D. Accumulating Knowledge on the about Earth's missing biodiversity. Trends in Dimensions of Biodiversity: Systematic Ecology & Evolution 27, 501–510 (2012). Perspectives on Neotropical Mammals. Biodiversity Letters 2, 79–86 (1994). 55. Feeley, K. J. & Silman, M. R. The data void in modeling current and future distributions of 69. Kamino, L. H. Y. et al. Challenges and perspectives tropical species. Global Change Biology 17, 626– for species distribution modelling in the 630 (2011). neotropics. Biology letters 8, 324–6 (2012). 56. González-Suárez, M., Lucas, P. M. & Revilla, E. 70. Burgess, N., Fjeldså, J. & Rahbek, C. Mapping the Biases in comparative analyses of extinction risk: distributions of Afrotropical vertebrate groups. mind the gap. Journal of Animal Ecology 81, 1211– Species 30, 16–17 (1998). 1222 (2012). 71. Hansen, L. A. Personal communication. (2013). 57. Beaumont, L. J., Hughes, L. & Pitman, A. J. Why is 72. Harrison, J. A., Underhill, L. G. & Barnard, P. The the choice of future climate scenarios for species seminal legacy of the Southern African Bird Atlas distribution modelling important? Ecology letters project. South African Journal of Science 104, 82– 11, 1135–46 (2008). 84 (2008). 58. Knutti, R. et al. A Review of Uncertainties in Global 73. Bled, F., Nichols, J. D. & Altwegg, R. Dynamic Temperature Projections over the Twenty-First occupancy models for analyzing species’ range Century. Journal of Climate 21, 2651–2663 (2008). dynamics across large geographic scales. Ecology 59. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. and Evolution 3, 4896–4909 (2013). G. & Jarvis, A. Very high resolution interpolated 74. IUCN Species Survival Commission, Conservation climate surfaces for global land areas. Science, International Center for Applied International Journal of Climatology 25, 1965– Biodiversity, NatureServe & IUCN. Global 1978 (2005). Amphibian Assessment. (2004). at 60. Fernández, M., Hamilton, H. & Kueppers, L. M. Characterizing uncertainty in species distribution 75. Wilby, R. L. et al. Guidelines for use of climate models derived from interpolated weather station scenarios developed from statistical downscaling data. Ecosphere 4, art61 (2013). methods. (2004). at Overview of CMIP5 and the Experiment Design.

Uncertainty in projected impacts of climate change | 21

76. James, R. & Washington, R. Changes in African 90. Keenan, T., Serra, J. M., Lloret, F., Ninyerola, M. & temperature and precipitation associated with Sabate, S. Predicting the future of forests in the degrees of global warming. Climatic Change 117, Mediterranean under climate change, with niche- 859–872 (2013). and process-based models: CO2 matters! Global Change Biology 17, 565–579 (2011). 77. Hawkins, E. & Sutton, R. The potential to narrow uncertainty in projections of regional 91. Synes, N. W. & Osborne, P. E. Choice of predictor precipitation change. Climate Dynamics 37, 407– variables as a source of uncertainty in continental- 418 (2011). scale species distribution modelling under climate change. Global Ecology and Biogeography 20, 904– 78. Hawkins, E. & Sutton, R. The Potential to Narrow 914 (2011). Uncertainty in Regional Climate Predictions. Bulletin of the American Meteorological Society 90, 92. VanDerWal, J. et al. Focus on poleward shifts in 1095–1107 (2009). species’ distribution underestimates the fingerprint of climate change. Nature Climate 79. Amano, T. & Sutherland, W. J. Four barriers to the Change 3, 239–243 (2013). global understanding of biodiversity conservation: wealth, language, geographical location and 93. Logan, M. L., Huynh, R. K., Precious, R. A. & security. Proceedings of the Royal Society B: Calsbeek, R. G. The impact of climate change Biological Sciences 280, 20122649 (2013). measured at relevant spatial scales: new hope for tropical lizards. Global Change Biology 19, 3093– 80. Felton, A. et al. Climate change, conservation and 3102 (2013). management: an assessment of the peer-reviewed scientific journal literature. Biodiversity and 94. Jackson, S. T., Betancourt, J. L., Booth, R. K. & Gray, Conservation 18, 2243–2253 (2009). S. T. Ecology and the ratchet of events: Climate variability, niche dimensions, and species 81. Tabor, K. & Williams, J. W. Globally downscaled distributions. Proceedings of the National Academy climate projections for assessing the conservation of Sciences 106, 19685–19692 (2009). impacts of climate change. Ecological Applications 20, 554–565 (2010). 95. Brook, B. W., Sodhi, N. S. & Bradshaw, C. J. A. Synergies among extinction drivers under global 82. Pierce, D. W. et al. Probabilistic estimates of future change. Trends in Ecology & Evolution 23, 453– changes in California temperature and 460 (2008). precipitation using statistical and dynamical downscaling. Climate Dynamics 40, 839–856 96. Holt, R. D. The microevolutionary consequences of (2013). climate change. Trends in Ecology & Evolution 5, 311–315 (1990). 83. Thuiller, W. et al. Biodiversity conservation: Uncertainty in predictions of extinction risk. 97. Chevin, L.-M., Lande, R. & Mace, G. M. Adaptation, Nature 430, 145–148 (2004). Plasticity, and Extinction in a Changing Environment: Towards a Predictive Theory. PLoS 84. Pearson, R. G. et al. Model-based uncertainty in Biology 8, e1000357 (2010). species range prediction. Journal of Biogeography 33, 1704–1711 (2006). 98. Hill, J. K., Griffiths, H. M. & Thomas, C. D. Climate Change and Evolutionary Adaptations at Species’ 85. Chefaoui, R. M. & Lobo, J. M. Assessing the effects Range Margins. Annual Review of Entomology 56, of pseudo-absences on predictive distribution 143–159 (2011). model performance. Ecological Modelling 210, 478–486 (2008). 99. Post, E. Ecology of climate change. The importance of biotic interactions. (Princeton University Press, 86. Wisz, M. S. & Guisan, A. Do pseudo-absence 2013). selection strategies influence species distribution models and their predictions? An information- 100. Araújo, M. B. & Pearson, R. G. Equilibrium of theoretic approach based on simulated data. BMC species’ distributions with climate. Ecography 28, Ecology 9, 8 (2009). 693–695 (2005). 87. Jiménez-Valverde, A. & Lobo, J. M. Threshold 101. Nogués-Bravo, D. Predicting the past distribution criteria for conversion of probability of species of species climatic niches. Global Ecology and presence to either–or presence–absence. Acta Biogeography 18, 521–531 (2009). Oecologica 31, 361–369 (2007). 102. Araújo, M. B. & Peterson, A. T. Uses and misuses of 88. Nenzén, H. K. & Araújo, M. B. Choice of threshold bioclimatic envelope modeling. Ecology 93, 1527– alters projections of species range shifts under 1539 (2012). climate change. Ecological Modelling 222, 3346– 103. Peterson, A. T. Uses and requirements of 3354 (2011). ecological niche models and related distribution 89. Williams, J. W. & Jackson, S. T. Novel climates, no- models. Biodiversity informatics 3, 59–72 (2006). analog communities, and ecological surprises. 104. Garcia, R. A., Burgess, N. D., Cabeza, M., Rahbek, C. Frontiers in Ecology and the Environment 5, 475– & Araújo, M. B. Exploring consensus in 21st 482 (2007). century projections of climatically suitable areas

22 | Synopsis

for African vertebrates. Global Change Biology 18, 120. Real, R., Luz Márquez, A., Olivero, J. & Estrada, A. 1253–1269 (2012). Species distribution models in climate change scenarios are still not useful for informing policy 105. Garcia, R. A. et al. Matching species traits to planning: an uncertainty assessment using fuzzy projected threats and opportunities from climate logic. Ecography 33, 304–314 (2010). change. Journal of Biogeography 10.1111/jbi.12257 (2014). 121. Bagchi, R. et al. Evaluating the effectiveness of conservation site networks under climate change: 106. McClean, C. J. et al. African plant diversity and accounting for uncertainty. Global Change Biology climate change. Annals of the Missouri Botanical 19, 1236–1248 (2013). Garden 92, 139–152 (2005). 122. Peterson, A. T. et al. Future projections for 107. Huntley, B. et al. Potential impacts of climatic Mexican faunas under global climate change change upon geographical distributions of birds. scenarios. Nature 416, 626–629 (2002). Ibis 148, 8–28 (2006). 123. Fordham, D. A. et al. Managed relocation as an 108. Thuiller, W. et al. Vulnerability of African adaptation strategy for mitigating climate change mammals to anthropogenic climate change under threats to the persistence of an endangered lizard. conservative land transformation assumptions. Global Change Biology 18, 2743–2755 (2012). Global Change Biology 12, 424–440 (2006). 124. Bates, J. M. & Granger, C. W. J. The Combination of 109. Barbet-Massin, M., Walther, B. A., Thuiller, W., Forecasts. Operational Research Quarterly 20, Rahbek, C. & Jiguet, F. Potential impacts of climate 451–468 (1969). change on the winter distribution of Afro- Palaearctic migrant passerines. Biology Letters 5, 125. Jun, M., Knutti, R. & Nychka, D. W. Spatial Analysis 248–251 (2009). to Quantify Numerical Model Bias and Dependence. Journal of the American Statistical 110. Hole, D. G. et al. Projected impacts of climate Association 103, 934–947 (2008). change on a continent-wide protected area network. Ecology Letters 12, 420–31 (2009). 126. Knutti, R., Furrer, R., Tebaldi, C., Cermak, J. & Meehl, G. A. Challenges in Combining Projections 111. New, M., Lister, D., Hulme, M. & Makin, I. A high- from Multiple Climate Models. Journal of Climate resolution data set of surface climate over global 23, 2739–2758 (2010). land areas. Climate Research 21, 1–25 (2002). 127. Beaumont, L. J., Hughes, L. & Pitman, A. J. Why is 112. Meehl, G. A. et al. in (Solomon S et al.) 747–845 the choice of future climate scenarios for species (Cambridge University Press, 2007). distribution modelling important? Ecology Letters 113. Araújo, M. B., Thuiller, W. & Pearson, R. G. Climate 11, 1135–1146 (2008). warming and the decline of amphibians and 128. Duan, Q. & Phillips, T. J. Bayesian estimation of reptiles in Europe. Journal of Biogeography 33, local signal and noise in multimodel simulations of 1712–1728 (2006). climate change. Journal of Geophysical Research 114. Lawler, J. J. et al. Projected climate-induced faunal 115, D18123 (2010). change in the Western Hemisphere. Ecology 90, 129. Naujokaitis-Lewis, I. R. et al. Uncertainties in 588–597 (2009). coupled species distribution–metapopulation 115. Buisson, L., Thuiller, W., Casajus, N., Lek, S. & dynamics models for risk assessments under Grenouillet, G. Uncertainty in ensemble climate change. Diversity and Distributions 19, forecasting of species distribution. Global Change 541–554 (2013). Biology 16, 1145–1157 (2010). 130. Williams, J. W., Jackson, S. T. & Kutzbach, J. E. 116. Diniz-Filho, J. A. F. et al. Partitioning and mapping Projected distributions of novel and disappearing uncertainties in ensembles of forecasts of species climates by 2100 AD. Proceedings of the National turnover under climate change. Ecography 32, Academy of Sciences of the United States of America 897–906 (2009). 104, 5738–42 (2007). 117. Diniz-Filho, J. A. F. et al. Ensemble forecasting 131. Araújo, M. B., Whittaker, R. J., Ladle, R. J. & Erhard, shifts in climatically suitable areas for Tropidacris M. Reducing uncertainty in projections of cristata (Orthoptera: Acridoidea: Romaleidae). extinction risk from climate change. Global Insect Conservation and Diversity 3, 213–221 Ecology and Biogeography 14, 529–538 (2005). (2010). 132. Marmion, M., Parviainen, M., Luoto, M., Heikkinen, 118. Dormann, C. F., Purschke, O., García Márquez, J. R., R. K. & Thuiller, W. Evaluation of consensus Lautenbach, S. & Schröder, B. Components of methods in predictive species distribution uncertainty in species distribution analysis: a case modelling. Diversity and Distributions 15, 59–69 study of the Great Grey Shrike. Ecology 89, 3371– (2009). 86 (2008). 133. Araújo, M. B. & Luoto, M. The importance of biotic 119. Thuiller, W. Patterns and uncertainties of species’ interactions for modelling species distributions range shifts under climate change. Global Change under climate change. Global Ecology and Biology 10, 2020–2027 (2004). Biogeography 16, 743–753 (2007).

Uncertainty in projected impacts of climate change | 23

134. Kissling, W. D., Field, R., Korntheuer, H., Heyder, U. 148. Murray, K. A., Rosauer, D., McCallum, H. & Skerratt, & Böhning-Gaese, K. Woody plants and the L. F. Integrating species traits with extrinsic prediction of climate-change impacts on bird threats: closing the gap between predicting and diversity. Philosophical Transactions of the Royal preventing species declines. Proceedings of the Society B: Biological Sciences 365 , 2035–2045 Royal Society of London Series B: Biological (2010). Sciences 278, 1515–23 (2011). 135. Midgley, G. F., Hughes, G. O., Thuiller, W. & Rebelo, 149. Fritz, S. A., Bininda-Emonds, O. R. P. & Purvis, A. A. G. Migration rate limitations on climate change- Geographical variation in predictors of induced range shifts in Cape Proteaceae. Diversity mammalian extinction risk: big is bad, but only in and Distributions 12, 555–562 (2006). the tropics. Ecology Letters 12, 538–549 (2009). 136. Engler, R. & Guisan, A. MigClim: Predicting plant 150. Rowland, E., Davison, J. & Graumlich, L. distribution and dispersal in a changing climate. Approaches to Evaluating Climate Change Impacts Diversity and Distributions 15, 590–601 (2009). on Species: A Guide to Initiating the Adaptation 137. Veloz, S. D. et al. No-analog climates and shifting Planning Process. Environmental Management 47, 322–337 (2011). realized niches during the late quaternary: implications for 21st-century predictions by 151. Buckley, L. B. & Kingsolver, J. G. Functional and species distribution models. Global Change Biology Phylogenetic Approaches to Forecasting Species’ 18, 1698–1713 (2012). Responses to Climate Change. Annual Review of 138. Heikkinen, R. K. et al. Assessing the vulnerability Ecology, Evolution, and Systematics 43, 205–226 of European butterflies to climate change using (2012). multiple criteria. Biodiversity and Conservation 19, 152. Foden, W. B. et al. in Wildlife in a Changing World: 695–723 (2009). an analysis of the 2008 IUCN Red List of 139. Triviño, M., Cabeza, M., Thuiller, W., Hickler, T. & Threatened Species. (Vié, J.-C., Hilton-Taylor, C. & Araújo, M. B. Risk assessment for Iberian birds Stuart, S. N.) 77–88 (IUCN, 2008). under global change. Biological Conservation 168, 153. Ohlemüller, R. et al. The coincidence of climatic 192–200 (2013). and species rarity: high risk to small-range species 140. Foden, W. B. et al. Identifying the World’s Most from climate change. Biology Letters 4, 568–572 Climate Change Vulnerable Species: A Systematic (2008). Trait-Based Assessment of all Birds, Amphibians 154. Sandel, B. et al. The influence of Late Quaternary and Corals. PLOS ONE 8, e65427 (2013). climate-change velocity on species endemism. 141. Thuiller, W., Lavorel, S., Araujo, M. B., Sykes, M. T. Science 334, 660–4 (2011). & Prentice, I. C. Climate change threats to plant 155. Fjeldså, J., Ehrlich, D., Lambin, E. & Prins, E. Are diversity in Europe. Proceedings of the National biodiversity `hotspots’ correlated with current Academy of Sciences of the United States of America ecoclimatic stability? A pilot study using the 102, 8245–8250 (2005). NOAA-AVHRR remote sensing data. Biodiversity and Conservation 6, 401–422 (1997). 142. Garcia, R. A. & Araújo, M. B. Planejamento para a Conservação em um Clima em Mudança. 156. Hole, D. G. et al. Projected impacts of climate Natureza&Conservação 8 , 78–80 (English version change on a continent-wide protected area available in Supplementary (2010). network. Ecology Letters 12, 420–431 (2009). 143. Araújo, M. B., Williams, P. H. & Fuller, R. J. 157. Araújo, M. B., Alagador, D., Cabeza, M., Nogués- Dynamics of extinction and the selection of nature Bravo, D. & Thuiller, W. Climate change threatens reserves. Proceedings of the Royal Society of European conservation areas. Ecology letters 14, London. Series B: Biological Sciences 269, 1971– 484–92 (2011). 1980 (2002). 158. Stattersfield, A. J., Crosby, M. J., Long, A. J. & Wege, 144. Hodgson, J. A. et al. Habitat re-creation strategies D. C. Endemic Bird Areas of the World: Priorities for promoting adaptation of species to climate for Biodiversity Conservation. Birdlife change. Conservation Letters 4, 289–297 (2011). Conservation Series no.7. 846 (BirdLife 145. Serra-Diaz, J. M. et al. Bioclimatic velocity: the pace International, 1998). of species exposure to climate change. Diversity 159. Myers, N., Mittermeier, R. A., Mittermeier, C. G., da and Distributions 20, 169–180 (2013). Fonseca, G. A. B. & Kent, J. Biodiversity hotspots 146. González-Suárez, M. & Revilla, E. Variability in life- for conservation priorities. Nature 403, 853–858 history and ecological traits is a buffer against (2000). extinction in mammals. Ecology Letters 16, 242– 160. Bini, L. M., Diniz-Filho, J. A. F., Rangel, T. F. L. V. B., 51 (2013). Bastos, R. P. & Pinto, M. P. Challenging Wallacean 147. Isaac, N. J. B. & Cowlishaw, G. How species respond and Linnean shortfalls: knowledge gradients and to multiple extinction threats. Proceedings of the conservation planning in a biodiversity hotspot. Diversity & Distributions 12, 475–482 (2006). Royal Society of London. Series B: Biological Sciences 271, 1135–1141 (2004).

24 | Synopsis

161. Mittermeier, R. A. et al. Hotspots Revisited: Earth’s 177. Allen, C. D. et al. A global overview of drought and Biologically Richest and Most Endangered heat-induced tree mortality reveals emerging Ecoregions. 390 (CEMEX, 2004). climate change risks for forests. Forest Ecology and Management 259, 660–684 (2010). 162. Olson, D. M. & Dinerstein, E. The Global 200: a representation approach to conserving the Earth’ 178. Chamaillé-Jammes, S., Massot, M., Aragón, P. & s most biologically valuable ecoregions. Clobert, J. Global warming and positive fitness Conservation Biology 12, 502–515 (1998). response in mountain populations of common 163. Tingley, M. W., Estes, L. D. & Wilcove, D. S. lizards Lacerta vivipara. Global Change Biology 12, Ecosystems: Climate change must not blow 392–402 (2006). conservation off course. Nature 500, 271–272 179. Tyler, N. J. C., Forchhammer, M. C. & Øritsland, N. (2013). A. Nonlinear effects of climate and density in the 164. Jiménez, M. A. et al. Extreme climatic events dynamics of a fluctuating population of reindeer. change the dynamics and invasibility of semi-arid Ecology 89, 1675–1686 (2008). annual plant communities. Ecology letters 14, 180. Lane, J. E., Kruuk, L. E. B., Charmantier, A., Murie, J. 1227–35 (2011). O. & Dobson, F. S. Delayed phenology and reduced 165. Iwamura, T., Wilson, K. A., Venter, O. & fitness associated with climate change in a wild Possingham, H. P. A Climatic Stability Approach to hibernator. Nature 489, 554–557 (2012). Prioritizing Global Conservation Investments. 181. Ordonez, A. & Williams, J. W. Climatic and biotic PLOS ONE 5, e15103 (2010). velocities for woody taxa distributions over the 166. Wiens, J. A., Seavy, N. E. & Jongsomjit, D. Protected last 16 000 years in eastern North America. areas in climate space: What will the future bring? Ecology Letters 16, 773–781 (2013). Biological Conservation 144, 2119–2125 (2011). 182. Willis, K. J. & MacDonald, G. M. Long-Term Ecological Records and Their Relevance to Climate 167. Loarie, S. R. et al. The velocity of climate change. Change Predictions for a Warmer World. Annual Nature 462, 1052–1055 (2009). Review of Ecology, Evolution, and Systematics 42, 168. Ohlemüller, R. Running Out of Climate Space. 267–287 (2011). Science 334, 613–614 (2011). 183. Davis, M. B. & Shaw, R. G. Range Shifts and 169. Svenning, J.-C., Fløjgaard, C., Marske, K. A., Nógues- Adaptive Responses to Quaternary Climate Bravo, D. & Normand, S. Applications of species Change. Science 292, 673–679 (2001). distribution modeling to paleobiology. Quaternary Science Reviews 30, 2930–2947 (2011). 184. Blois, J. L., McGuire, J. L. & Hadly, E. A. Small mammal diversity loss in response to late- 170. Burrows, M. T. et al. The Pace of Shifting Climate in Pleistocene climatic change. Nature 465, 771–774 Marine and Terrestrial Ecosystems. Science 334, (2010). 652–655 (2011). 185. Nogués-Bravo, D., Ohlemüller, R., Batra, P. & 171. Ohlemüller, R., Gritti, E. S., Sykes, M. T. & Thomas, Araújo, M. B. Climate predictors of late quaternary C. D. Towards European climate risk surfaces: the extinctions. Evolution 64, 2442–2449 (2010). extent and distribution of analogous and non- analogous climates 1931–2100. Global Ecology 186. Littell, J. S., McKenzie, D., Kerns, B. K., Cushman, S. and Biogeography 15, 395–405 (2006). & Shaw, C. G. Managing uncertainty in climate- driven ecological models to inform adaptation to 172. Ackerly, D. D. et al. The geography of climate climate change. Ecosphere 2, art102 (2011). change: implications for conservation 187. Fronzek, S., Carter, T. R., Räisänen, J., Ruokolainen, biogeography. Diversity and Distributions 16, 476– L. & Luoto, M. Applying probabilistic projections of 487 (2010). climate change with impact models: a case study 173. Jackson, S. T. & Overpeck, J. T. Responses of Plant for sub-arctic palsa mires in Fennoscandia. Populations and Communities to Environmental Climatic Change 99, 515–534 (2010). Changes of the Late Quaternary. Paleobiology 26, 194–220 (2000). 188. Wenger, S. J. et al. Probabilistic accounting of uncertainty in forecasts of species distributions 174. Peñuelas, J. et al. Evidence of current impact of under climate change. Global Change Biology 19, climate change on life: a walk from genes to the 3343–3354 (2013). biosphere. Global Change Biology 19, 2303–2338 189. Hijmans, R. J. & Graham, C. H. The ability of climate (2013). envelope models to predict the effect of climate 175. Foden, W. et al. A changing climate is eroding the change on species distributions. Global Change geographical range of the Namib Desert tree Aloe Biology 12, 2272–2281 (2006). through population declines and dispersal lags. Diversity and Distributions 13, 645–653 (2007). 190. Morin, X. & Thuiller, W. Comparing niche- and process-based models to reduce prediction 176. Sinervo, B. et al. Erosion of Lizard Diversity by uncertainty in species range shifts under climate Climate Change and Altered Thermal Niches. change. Ecology 90, 1301–1313 (2009). Science 328, 894–899 (2010).

Uncertainty in projected impacts of climate change | 25

191. Kearney, M. R., Wintle, B. A. & Porter, W. P. Assessments and Conservation Decisions. Correlative and mechanistic models of species Conservation Biology 27, 1147–1157 (2013). distribution provide congruent forecasts under 204. Meller, L. et al. Ensemble distribution models in climate change. Conservation Letters 3, 203–213 conservation prioritization: from consensus (2010). predictions to consensus reserve networks. 192. Tingley, M. W. & Beissinger, S. R. Detecting range Diversity & Distributions (2014). shifts from historical species occurrences: new doi:10.1111/ddi.12162 perspectives on old data. Trends in Ecology & 205. Hof, C., Levinsky, I., Araújo, M. B. & Rahbek, C. Evolution 24, 625–33 (2009). Rethinking species’ ability to cope with rapid 193. Bierman, S. M., Butler, A., Marion, G. & Kühn, I. climate change. Global Change Biology 17, 2987– Bayesian image restoration models for combining 2990 (2011). expert knowledge on recording activity with 206. Fjeldså, J. How broad-scale studies of patterns and species distribution data. Ecography 33, 451–460 processes can serve to guide conservation (2010). planning in Africa. Conservation biology : the 194. Raxworthy, C. J. et al. Predicting distributions of journal of the Society for Conservation Biology 21, known and unknown reptile species in 659–67 (2007). Madagascar. Nature 426, 837–41 (2003). 207. Kerr, J. T., Kharouba, H. M. & Currie, D. J. The 195. Dessai, S., Hulme, M., Lempert, R. & Pielke, R. Do Macroecological Contribution to Global Change We Need Better Predictions to Adapt to a Solutions. Science 316 , 1581–1584 (2007). Changing Climate? Eos, Transactions American Geophysical Union 90, 111–112 (2009). 196. Knutti, R. & Sedlacek, J. Robustness and uncertainties in the new CMIP5 climate model Other publications or manuscripts in projections. Nature Climate Change 3, 369–373 preparation (2013). 197. Stevens, B. & Bony, S. What Are Climate Models R.A. Garcia and M.B. Araújo (2010). Planejamento para Missing? Science 340 , 1053–1054 (2013). a conservação em um clima em mudança (Conservation planning under changing climates). 198. Zimmermann, N. E. et al. Climatic extremes Natureza & Conservação 8(1): 78-80. improve predictions of spatial patterns of tree species. Proceedings of the National Academy of J. Forrest, N. D. Burgess, R. A. Garcia, A. Carlson, S. Sciences of the United States of America 106, Freeman, C. Loucks, Z. Maritim, S. Olimb, N. 19723–19728 (2009). Olwero, A. Schrag, S. Anstey. An Assessment of the Spatial Vulnerability of Selected Vertebrates to 199. Forrest, J. et al. An Assessment of the Spatial Land Use and Climate Change in the Greater Vulnerability of Selected Vertebrates to Land Use Ruvuma Landscape, Eastern Africa. In and Climate Change in the Greater Ruvuma preparation. Landscape, Eastern Africa. In preparation. S. G. Willis, W. Foden, R.G. Smith, N.D. Burgess, D.J. 200. Bennie, J. et al. Range expansion through Baker, E. Belle, J. Carr, N. Doswald, R. A. Garcia, A. fragmented landscapes under a variable climate. Hartley, C. Hof, T. Newbold, C. Rahbek, P. Visconti, Ecology Letters 16, 921–929 (2013). B. Young , S.H.M. Butchart. A framework 201. Evans, M. R. et al. Predictive systems ecology. methodology for integrating climate change Proceedings of the Royal Society B: Biological vulnerability assessments from Species Sciences 280, 20131452 (2013). Distribution Models and Trait-based Approaches 202. Burgman, M. A., Lindenmayer, D. B. & Elith, J. for use in systematic conservation planning. In Managing Landscapes for Conservation under preparation. Uncertainty. Ecology 86, 2007–2017 (2005). 203. Snover, A. K. et al. Choosing and Using Climate- Change Scenarios for Ecological-Impact

Chapter I

Exploring consensus in 21st century projections of climatically suitable areas for African vertebrates

RAQUEL A. GARCIA, NEIL D. BURGESS, MAR CABEZA, CARSTEN RAHBEK, AND MIGUEL B. ARAÚJO Global Change Biology 18, 1253-1269 (2012)

Exploring consensus in 21st century projections of climatically suitable areas for African vertebrates

RAQUEL A. GARCIA1,2,3, NEIL D. BURGESS2,4, MAR CABEZA1,5, CARSTEN RAHBEK2, and MIGUEL B. ARAÚJO1,2,3

1 Department of Biodiversity and Evolutionary Biology, National Museum of Natural Sciences, CSIC, Madrid, Spain 2 Center for Macroecology, Evolution and Climate, Department of Biology, University of Copenhagen, Denmark 3 Rui Nabeiro Biodiversity Chair, University of Évora, CIBIO, Évora, Portugal 4 WWF US Conservation Science Program, Washington, DC, US 5 Metapopulation Research Group, Department of Biosciences, University of Helsinki, Finland

Global Change Biology 18, 1253-1269 (2012)

Abstract

Africa is predicted to be highly vulnerable to 21st rise agreements among projections from the bio- century climatic changes. Assessing the impacts of climatic envelope models we compare five con- these changes on Africa’s biodiversity is, however, sensus methodologies, which generally increase plagued by uncertainties, and markedly different or retain projection accuracy and provide con- results can be obtained from alternative biocli- sistent estimates of species turnover. Variability matic envelope models or future climate projec- from emissions scenarios increases towards late- tions. Using an ensemble forecasting framework, century and affects southern regions of high spe- we examine projections of future shifts in climatic cies turnover centred in arid Namibia. Two-fold suitability, and their methodological uncertainties, differences in median species turnover across the for over 2,500 species of mammals, birds, amphib- study area emerge among alternative climate ians and snakes in sub-Saharan Africa. To summa- projections and emissions scenarios. Our ensem- rise a priori the variability in the ensemble of 17 ble of projections underscores the potential bias general circulation models, we introduce a con- when using a single algorithm or climate projec- sensus methodology that combines co-varying tion for Africa, and provides a cautious first ap- models. Thus we quantify and map the relative proximation of the potential exposure of sub- contribution to uncertainty of seven bioclimatic Saharan African vertebrates to climatic changes. envelope models, three multi-model climate pro- The future use and further development of bio- jections and three emissions scenarios, and ex- climatic envelope modelling will hinge on the plore the resulting variability in species turnover interpretation of results in the light of methodo- estimates. We show that bioclimatic envelope logical as well as biological uncertainties. Here, we models contribute most to variability, particularly provide a framework to address methodological in projected novel climatic conditions over uncertainties and contextualise results. Sahelian and southern Saharan Africa. To summa-

30 | Chapter I

Introduction considerable time. In face of the increasing number of climate projections and statistical Assessments of the potential effects of 21st functions available, calls have been made to century climatic changes on biodiversity com- explicitly address the methodological uncer- monly rely on bioclimatic envelope models tainties of BEMs so as to quantify the confi- (BEMs). Using correlations between climate dence that can be placed in forecasts (Thuiller and known species occurrences, BEMs estimate et al., 2004a, Araújo et al., 2005b, Araújo et al., future shifts in suitable climate for species. 2006, Pearson et al., 2006, Diniz-Filho et al., Widespread use of BEMs has been accompanied 2009, Wiens et al., 2009, Buisson et al., 2010). by discussions of the biological (e.g. Pearson & The high level of certainty typically required for Dawson, 2003, Araújo & Pearson, 2005, Sinclair policy making can hardly be attained using et al., 2007) and methodological (e.g. Heikkinen correlative models – least of all under unknown et al., 2006, Beaumont et al., 2008) uncertain- future climates – leaving model users with the ties that surround the outputs. BEMs rely on option of exploring the uncertainty in projec- assumptions about the association between tions and weighing the risks associated with climate and species distributions, and their alternative actions (Wiens et al., 2009). One biological realism hinges on additional factors approach to address such uncertainties is to influencing species vulnerability to climatic incorporate several assumptions and explore changes, such as ecophysiological and micro- the resulting range of potential results habitat preferences, phenotypic plasticity, evo- ("ensemble forecasting", Araújo & New, 2007). lutionary rates, dispersal ability (Chevin et al., Here, we provide the most extensive investiga- 2010, Dawson et al., 2011, Hof et al., 2011), and tion to date of methodological uncertainties biotic interactions (Araújo & Luoto, 2007, Suttle associated with ensemble forecasts of climate et al., 2007). In turn, BEM results are sensitive change impacts on sub-Saharan African verte- to the data and statistical functions that are brate species. used to describe the associations between spe- The African continent is predicted to be cies and climate. Alternative algorithms differ one of the most vulnerable to 21st century cli- regarding the data used, variable selection, matic changes (Boko et al., 2007, Collier et al., model parameterisation (Guisan & 2008). Forecasts of warming above the global Zimmermann, 2000, Elith et al., 2006, average (Christensen et al., 2007) are projected Heikkinen et al., 2006), and techniques for to affect African biodiversity and people’s live- extrapolation to novel conditions (Thuiller et lihoods (Velarde et al., 2005, Boko et al., 2007, al., 2004b, Pearson et al., 2006, Elith & Graham, Biggs et al., 2008). Yet, in comparison to well- 2009). BEMs are also sensitive to the green- researched regions such as Europe or North house gas emissions scenarios and climate America, Africa has received limited attention models used to simulate future climates regarding the potential impacts of climate (Beaumont et al., 2008). change on biodiversity (Lovett et al., 2005, Research is ongoing to develop more bio- Felton et al., 2009). Attribution of shifts in spe- logically realistic models (Keith et al., 2008, cies distributions to climate change is difficult Anderson et al., 2009, Brook et al., 2009, in Africa (Chown et al., 2010, Midgley & Huntley et al., 2010, Kearney et al., 2010), but Thuiller, 2011) because changes in water avail- the breath of information and data required to ability – the main determinant of ecological appropriately parameterise them is large. Sim- responses (Hawkins et al., 2003) – are spatially pler approaches such as BEMs are thus likely to complex and difficult to document (MacKellar remain important tools for assessing potential et al., 2007). Increases in temperature, how- impacts of climate change on biodiversity for a

Ensemble forecasting for African vertebrates | 31

ever, have been associated with the observed projected to contract in the Kalahari region, and range extension of the Common Swift (Hockey to shift eastward (Thuiller et al., 2006). For & Midgley, 2009), and are likely to have more birds, forecasts revealed losses in southern and severe impacts for tropical species (Deutsch et eastern Africa for breeding birds (Huntley et al., al., 2008, Wright et al., 2009). 2006) and trans-Saharan migrant passerines Previous studies using BEMs at a continen- (Barbet-Massin et al., 2009), but relatively tal scale have projected substantial geographi- small changes for the former in equatorial and cal shifts in suitable climate for African plants, moist tropical forest habitats and even gains for birds and mammals by late-century (Table 1). the latter in the Sahel and Arabian Peninsula. Specifically, McClean and colleagues (2005) Hole and colleagues (2009) also projected predicted losses of suitable climate for plant higher ensemble turnover for breeding birds in species in the Guineo-Congolian forests of southern Africa, despite high persistence of western and central Africa, and gains in the suitable climate across the network of Impor- surrounding uplands as well as the highlands of tant Bird Areas as a whole. The results pre- Namibia and the South African Drakensberg. sented in these studies are, however, contin- Mammal species ranges around the equatorial gent on the specific BEMs and future climate zone in central Africa were projected to shift projections used. Only one study (Barbet- westward, with contractions in the Congo Ba- Massin et al., 2009) has fitted a range of differ- sin, whereas mammals in southern Africa were ent BEMs. Most of them also spanned a limited

Table 1 | Published continental- and sub-continental-scale studies using correlative models to assess the impacts of climate change on African biodiversity. Species data Extent Res Modelling Future scenarios Ref. olu- approach tion Distributional Sub- 1o . Box model, SGA . 1 GCM: HadCM3 McClean data for 5,197 Saharan and BGA . 1 SRES: B1 et al., plant species (≥2 Africa . 3 time periods: 2025, 2055 2005 records) and 2085

Distributional Sub- 1o . Locally . 3 GCMs: HadCM3, Huntley et data for bird Saharan weighted GFDL_R30, al. 2006 species breeding Africa regression ECHAM4/OPYC3 in Africa . 1 SRES: B2 . 1 time period: 2070-99

Extent of occur- Africa 10’ . GAM . 1 GCM: HadCM3 Thuiller et rence data for 277 . 2 SRES: A2 and B2 al., 2006 mammal species . 2 time periods: 2050 and 2080

Distributional Sub- 1o . CRS and GAM . 3 GCMs: HadCM3, ECHAM4, Hole et al. data for 1,608 Saharan GFDL-R30 2009 bird species (≥5 Africa . 1 SRES: B2a records) . 3 time periods: 2025, 2055 and 2085

Distributional Sub- 0.5o . GLM, GAM, CTA, . 5 GCMs: BCM2, ECHAM5, Barbet- data for 64 bird Saharan ANN, MDA, HadCM3, MIROHIC3_2-HI Massin et species (≥6 re- Africa MARS, GBM, RF, and MK3) al. 2009 cords) MaxEnt . 3 SRES: A1B, B1 and A2 . Consensus . 3 time periods: 2030, 2065 projection and 2100

32 | Chapter I

number of General Circulation Models (GCMs) (Ogawa-Onishi et al., 2010), Australia and emissions scenarios, overlooking the vari- (Crossman & Bass, 2008), and Africa (Barbet- ability among simulations of future climates Massin et al., 2009, Coetzee et al., 2009), and which has been shown to be region-specific and have generally yielded higher accuracy than relatively high for most of Africa south of the single-models. Yet, there is still debate on the Equator (Hawkins & Sutton, 2009, 2011). In- best methodologies for combining BEM projec- vestigation of the level of uncertainty associ- tions and only a few comparisons have been ated with the results was, thus, limited. published (Araújo et al., 2005b, Araújo & New, In this article, we use seven BEM tech- 2007, Marmion et al., 2009). Here, we compare niques to describe the bioclimatic envelopes of five methodologies, including the methodology 284 amphibian, 310 snake, 623 mammal and introduced to combine GCMs. 1,506 bird species in sub-Saharan Africa. To Our projections provide insights into the assess the impacts of climatic changes on the potential exposure of sub-Saharan African ver- modelled species, we project their envelopes to tebrates to 21st century climatic changes. We mid- and late-century climates. We use an en- explicitly address the variability in forecasts of semble of 17 GCMs, under the B1, A1B, and A2 species temporal turnover from alternative emissions scenarios from the Intergovernmen- climate projections and BEMs, or combinations tal Panel on Climate Change (IPCC). As large thereof. More specifically, we investigate: a) the ensembles of projections are difficult to inter- relative contribution of different sources of pret, consensus methodologies can be used to uncertainty in forecasts of species turnover; b) average across ensembles. Multi-model climate the predictive accuracy of forecasts from alter- projections are widely used in climatology, but native BEM consensus methodologies; and c) there is still debate on the best consensus the variation in forecasts with alternative BEM methodologies to combine models (Tebaldi & consensus methodologies and climate projec- Knutti, 2007, Knutti et al., 2010). To retain tions. information about the full variability in our ensemble of GCMs, we introduce a methodology that averages co-varying GCMs based on their Data and Methods similarity in both magnitude and spatial pattern. We thus reduce the ensemble of GCMs to Species and climate data three multi-model projections, and obtain, for The study region covered continental sub- each of the 2,723 species modelled, 126 projec- Saharan Africa, south of 20°N. Species occur- tions (seven BEM techniques, three climate rence data for amphibian (Hansen et al., projections, three emissions scenarios, and two 2007a), snake (Rasmussen et al., 2007), mam- time periods), or 343,098 projections overall. mal (Galster et al., 2007) and terrestrial bird To facilitate interpretation of this ensemble, we species (Hansen et al., 2007b) in sub-Saharan also summarise agreements among projections Africa were used from the 1° resolution (≈111 from the seven BEMs. Consensus methodolo- km x 111 km at the Equator) databases held at gies have been used in previous climate change the Zoological Museum within the University of ecology work in Europe (Thuiller, 2004, Araújo Copenhagen in Denmark. This is the most com- et al., 2006, Buisson & Grenouillet, 2009, Araújo prehensive biodiversity dataset for Africa, et al., 2011, Thuiller et al., 2011), the Americas compiled from multiple sources and continu- (Diniz-Filho et al., 2009, Lawler et al., 2009, ously improved over 15 years (Burgess et al., Marini et al., 2009, Roura-Pascual et al., 2009, 1998, Brooks et al., 2001). Data were available Diniz-Filho et al., 2010, Marini et al., 2010), Asia for 741 amphibians, 477 snakes, 1,085 mam-

Ensemble forecasting for African vertebrates | 33

mals and 1,789 birds. Because of statistical bird species (Huntley 2006) and a variety of difficulties with modelling species with limited other species in Africa (Chown, 2010 and numbers of occurrence records (Stockwell & references therein). The temperature-based Peterson, 2002, Wisz et al., 2008), we excluded variables selected in our study further reflect species with fewer than 15 records over the the important effect of seasonal temperatures study area. Our threshold may introduce uncer- on species’ distributions (Huntley et al., 2006). tainty in the analysis, yet the effect of sample Future climate projections were derived size on model accuracy is a question that re- from 17 GCMs downscaled to 10’ resolution quires further study. More conservative (Tabor & Williams, 2010; see Appendix S2). All thresholds for sample size have been suggested downscaled GCMs were from the World Climate (e.g. Harrell et al., 1984), but some algorithms Research Programme’s Coupled Model Inter- have been shown to achieve 90% of their max- comparison Project phase 3 multi-model data- imum accuracy with ten records (Stockwell & set projections (Meehl et al., 2007) and de- Peterson, 2002). In total 2,723 species were biased using the change-factor technique and modelled, accounting for 67% of the available observational data from the CRU. The datasets data (284 amphibians, 310 snakes, 623 mam- comprise monthly mean temperature and pre- mals and 1,506 birds, with median range sizes cipitation projections for the 2041-60 and of 71, 90, 94 and 162 grid cells respectively). 2081-00 time intervals. Simulations from the Baseline climate data averaged for the 17 GCMs were used for three illustrative green- 1961-90 period were obtained from the Cli- house gas emissions scenarios from the IPCC’s matic Research Unit (New et al., 2002) at a Special Report Emissions Scenarios resolution of 10’ (≈18.6 km x 18.6 km at the (Nakicenovic et al., 2000). We used high- to Equator). Monthly precipitation and mean low-end scenarios (A2, A1B and B1) that reflect temperature values were used to compute 21 different assumptions about demographic, variables that are commonly useful in biocli- socio-economic and technological development matic modelling studies (see Appendix S1 in on greenhouse gas emissions. For each GCM Supporting Information). We applied principal projection and scenario we computed the three components analysis (PCA) to identify sets of selected variables over the study area. To uncorrelated variables that represent major match the species data resolution, both base- climatic gradients over the study area. From the line and future climate datasets were re- first axis we selected the variable with the sampled in ArcGIS (ESRI, 2006), using bilinear highest loading, i.e., annual precipitation. An interpolation, to the 1° grid over sub-Saharan additional variable was selected from the first Africa. Data processing and statistical analyses axis with high loading but opposite sign (corre- were performed in R (R Development Core lation -0.45): temperature of the warmest Team, 2010) version 2.11.1. month. The variable with the highest loading on the second axis, i.e., temperature of the coldest Combining ensembles of climate projections month, was the third variable selected. The two We first summarised the general tendencies first axes explained 74.3% of the variation among the 17 selected GCMs. In climatology, (Appendix S1). Together, precipitation and multi-model ensemble averages have often temperature influence water availability, which been shown to improve the outcome of climate controls biological activity in the tropics and simulations (Phillips & Gleckler, 2006, Tebaldi sub-tropics (Hawkins et al., 2003). Both pre- & Knutti, 2007, Gleckler et al., 2008, Reichler & cipitation- and temperature-based variables are Kim, 2008, Pierce et al., 2009, Knutti et al., important determinants of the distributions of 2010, but see Fordham et al., in press). How-

34 | Chapter I

ever, averaging ensembles can result in the loss cate high similarity between a given model and of higher-order variability reflected in extreme the multi-model median. projections (Beaumont et al., 2008). To avoid Second, the ensemble of 17 GCMs was par- this limitation, we used a hybrid consensus titioned into groups of co-varying projections approach (Araújo et al., 2006) – hereafter re- according to the 2081-00 D and R statistics ferred to as ‘central cluster’ – that groups co- obtained. We used k-means, a clustering tech- varying projections before averaging them. nique that assigns data points to the closest When there is great variation in projections – pre-defined centre. These centres were the as it is often the case with precipitation – this median points of single linkage hierarchical approach also avoids averaging projections that clusters based on the Euclidean distance matrix are very different or even contradictory, by (Venables & Ripley, 2003). The significance of placing them in different groups. the differences between clusters was tested To combine the GCMs, we used three steps with Anosim, a non-parametric test of analysis in R (R Development Core Team, 2010, see of similarity (Clarke & Warwick, 1994). Anosim Appendix S3 for the R scripts). First, as a basis was applied to the dissimilarity matrix of D and for identifying co-varying projections under R values to test whether the distances between each emissions scenario, we assessed similari- clusters were greater than the distances within ties among GCM simulations for each variable clusters. The initial number of clusters to ex- projected in the late-century, when inter- tract was selected so as to minimise inter- simulation spread becomes larger (Stott & cluster distances in the hierarchical trees, and Kettleborough, 2002, Meehl et al., 2007, was increased when needed until the Anosim Hawkins & Sutton, 2009). Similarities were test was statistically significant. For each emis- assessed separately for each variable because sions scenario we obtained a set of clusters, the performance of climate models varies for each with a number of co-varying late-century different variables (Lambert & Boer, 2001, climate models. The same clusters were applied Gleckler et al., 2008). We used model perform- to the baseline and 2041-60 time periods. ance metrics to characterise the agreement Third, we generated summaries of the pat- between individual simulations for each vari- terns of central tendency in each cluster ex- able and the multi-model median ensemble for tracted. For each cluster of co-varying GCMs the same variable. These metrics were spatially obtained for different emissions scenarios and aggregated point-wise measures of regional time slices, un-weighted median consensus deviations (Duan & Phillips, 2010). The spatial forecasts were computed on each variable. pattern Pearson correlation (R) reflects spatial There are contrasting views on the use of agreement between individual simulations of a weights to perform climate ensemble averages. given variable and the median simulation of They have been shown not to systematically that variable. The signed standardised anomaly change the results in some cases (e.g. Pierce et (D), in turn, measures signed differences in al., 2009) and to improve them in other cases magnitude between individual simulations and (e.g. Min & Hense, 2006). Which model per- the median simulation of a given variable, stan- formance metrics to use as weights also re- dardised using the standard deviation of all mains an issue of debate (Tebaldi & Knutti, simulations. D thus reflects whether a simula- 2007, Gleckler et al., 2008). Because the opti- tion tends to under- or over-estimate a given mal performance weights for future projections variable in relation to the median of all simula- are unlikely the same as for baseline climate tions of that variable, and by how much. D val- (Tebaldi & Knutti, 2007), we opted for un- ues close to zero and R values close to 1 indi- weighted averages. In summary, for each emis-

Ensemble forecasting for African vertebrates | 35

sions scenario and time period combination we (Chefaoui & Lobo, 2008). Yet, random selection obtained a set of clusters of GCMs, each with has been shown to result in higher predictive the median simulation of each variable com- power than strategies that select pseudo- puted across GCMs. For each set, the cluster absences from low-suitability regions (Wisz & with the average D closest to zero and the high- Guisan, 2009). est average R captured the maximum consen- For each species, the seven models were sus among projections, corresponding to the built using a calibration subset of 75% of the ‘central cluster’, whereas clusters departing sites selected at random and evaluated with the from the multi-model median ensemble cap- remaining 25% of the sites. This data-splitting tured extreme projections. procedure was repeated five times. Projections of the probability of occurrence of species in Bioclimatic envelope modelling each site were converted to binary format (presence/absence) using a threshold maximis- Models were fitted at 1° resolution using seven ing the True Skill Statistic (TSS, Allouche et al., presence-absence BEM techniques in BIOMOD 2006). The models were evaluated based on (Thuiller et al., 2009), a computing platform for median omission and commission errors and ensemble forecasting that operates in R envi- TSS on the cross-validated data. The calibrated ronment (R Development Core Team, 2010). models were used to generate projections of The techniques included three regression species’ bioclimatic envelopes under each GCM methods (generalised linear models (GLM), cluster and emissions scenario for 2041-60 and generalised additive models (GAM), and multi- 2081-00. The projections were based on the variate adaptive regression splines (MARS)), final runs of the models using 100% of the data, three machine-learning methods (artificial as data partitions have been shown to add sig- neural networks (ANN), Breiman and Cutler's nificant uncertainty to forecasts (Araújo et al., random forest for classification and regression 2005b, 2009). Because we were interested in (RF), and generalised boosting models (GBM)), measuring changes in climatic suitability for and one classification method (flexible dis- species rather than interpreting model projec- criminant analysis, FDA). Due to differences in tions as estimates of the changes in observed quality, species occurrence data were treated species distributions, we adopted an ‘unlimited differently across taxa. Estimated range maps dispersal’ scenario, whereby species are as- for mammals and birds, based on numerous sumed to be able to track shifting suitable cli- records of species across multiple countries, mate over the entire study area. To comple- lend themselves to be treated as presence- ment these projections, the areas where higher absence data. For most amphibians and snakes, proportions of species are projected to retain however, the data comprise confirmed speci- climatic suitability over time (corresponding to men locality records from museum collections a ‘no dispersal’ assumption) were also mapped. and thus were considered to be closer to presence-only data. For the latter taxa, pseudo- Combining ensembles of BEM forecasts absences were randomly generated to allow fitting models that assume the data to be in the For each taxon, we explored the agreement form of presences and absences. The process of among projections from the seven BEM tech- generating random pseudo-absences in BIO- niques using five consensus methodologies. The MOD weighs them to achieve a prevalence of first three methodologies provide a synthetic 0.5. There is a debate on how to select pseudo- measurement of the central tendency in the absences, and the choice of selection method is frequency distribution of the projections ob- dependent on the purpose of the study tained from all BEMs. Implemented within the

36 | Chapter I

BIOMOD package, the ‘ensemble mean’ com- consensus forecasts in the baseline and 2041- putes the un-weighted mean (e.g. Buisson & 60 time periods. Grenouillet, 2009, Diniz-Filho et al., 2010), the The BEM consensus projections were built ‘ensemble weighted mean’ (e.g. Marmion et al., using 100% of the data for the same reasons 2009) uses the TSS values as weights, and the cited above for single-BEM projections. To ‘ensemble median’ calculates the second quar- evaluate the consensus projections we applied tile of the frequency distribution of forecasts the same five consensus methodologies to the from all models (e.g. Araújo et al., 2005b, five evaluation datasets (25% of the data) and Lawler et al., 2009). The fourth and fifth meth- computed the median omission error, commis- odologies, in turn, pre-select projections that sion error and TSS of all cross-validations and, best summarise consensus among them. in the case of amphibians and snakes, all With the fourth method – ‘central model’ – pseudo-absence runs. To assess the level of we selected the model summarising the highest consensus among BEMs, we performed PCAs amount of variation among projections. For for each species on the probabilistic projec- each species, PCAs were performed on the tions, both for all seven BEMs and for the ‘cen- projected probabilities within the BIOMOD tral cluster’ BEMs only. The proportion of vari- package, and the ‘central model’ corresponded ance explained by the first principal component to the one with the highest PCA loading in the axis provided a measure of consensus among first (consensus) axis (e.g. Thuiller et al., 2005, projections (Thuiller, 2004, Araújo et al., 2006, Algar et al., 2009). In the fifth method – ‘central Grenouillet et al., 2011). cluster’ – we investigated patterns of central tendency among groups of co-varying projec- Mapping shifts in climatic suitability and associated tions (e.g. Araújo et al., 2005b, Araújo et al., uncertainties 2006). Following the approach used to combine For each emissions scenario and GCM cluster the ensemble of GCMs, we clustered the BEMs combination, and for the five BEM consensus based on the similarities between single-BEM methodologies applied, baseline and future probabilistic projections and the multi-model species richness and the number of contracting median probabilistic projection. We used the and expanding species were computed. Spatial same measures of D and R, computed for each patterns of change were investigated using species. Unlike the other four consensus meth- measures of species temporal turnover per grid odologies, the grouping was performed for the cell (Peterson et al., 2002). The ‘species turn- set of all species in one taxon rather than indi- over rate’ refers to local dissimilarities between vidually for each species. The same procedure baseline and future sets of species for which a of clustering and Anosim testing we used for given area is projected to be climatically suit- GCM projections was followed. The correspond- able, and thus incorporates both losses and ing binary projections for the BEMs in each gains of climate space. In addition, in situ per- cluster were combined using a majority vote sistence of climatic suitability for species was criterion, whereby a species was considered investigated. present in grid points where more than half of To evaluate and map the relative contribu- the BEMs in the cluster predicted presence. The tions of emissions scenarios, future climates ‘central cluster’ was the closest to the multi- and BEMs to the overall uncertainty in fore- model projection. Both the PCA and cluster casts, we performed analyses of variance analysis were applied to the end-of-century (ANOVA) in R (R Development Core Team, scenario, when divergence is expected to be 2010). Following Diniz-Filho and colleagues highest, and the same model(s) used to derive (2009), we performed a three-way ANOVA

Ensemble forecasting for African vertebrates | 37

without replication for each grid cell, using the the general trend across models (Figure 1a; see turnover rate as the response variable and the below). emissions scenarios, future climate projections Disagreement among BEM forecasts in late- and BEM consensus methodologies as factors. century was mainly concentrated in the north- An ANOVA using single-BEMs, climates and ern half of the study area, particularly in Sahe- emissions scenarios as factors, in turn, pro- lian and southern Saharan Africa (Figure 2). vided indications on the relative contribution of These areas were predicted to experience mean individual BEMs to uncertainty in turnover temperatures of the warmest and coldest projections (before combining the ensembles). months above the calibration range (Figure 3a; To explore BEM uncertainties associated with Appendix S5), forcing the models to extrapolate model extrapolation, we mapped the areas beyond known relationships. Late-century non- experiencing future climates beyond the range analogue climates covered up to 50% of the of climate values used to fit the models for each study area for the most severe climate projec- of the three variables. For each GCM cluster and tion. A comparison of grid points with analogue scenario combination, the sum of these areas and non-analogue climates (Figure 3b) re- corresponded to ‘non-analogue’ areas, where vealed significant differences in the proportion projections become statistically less reliable of the total sum of squares attributed to BEMs (Heikkinen et al., 2006, Williams et al., 2007, for all time periods and future climates (Kol- Fitzpatrick & Hargrove, 2009). mogorov-Smirnov tests p-value<0.001). The northern regions of high BEM uncertainty mostly corresponded to areas with non- Results analogue climates.

Relative contribution of different sources of Predictive accuracy of forecasts from alternative BEM uncertainty in forecasts of species turnover consensus methodologies Uncertainty in species turnover forecasts was Using the median TSS across all species as an mainly caused by the variability among BEMs. evaluation criterion, we found that consensus In the point-wise ANOVA using BEMs, GCM projections outperformed or equalled all or six clusters, and emissions scenarios as factors, the of the single-BEMs for amphibians and snakes, median proportions of the total sum of squares and between one and seven single-BEMs for across the study area attributed to BEMs mammals and birds (Figure 4a). The ‘central reached between 76% for mammals and 82% cluster’ methodology provided the most robust for snakes by mid-century (Appendix S4). The projections for all taxa, surpassing all single- relative contribution of BEMs to overall uncer- BEMs. This methodology resulted in four clus- tainty decreased by late-century (to between ters for each taxon, with the ‘central cluster’ 61% for mammals and 69% for snakes) due to combining high-accuracy models (GAM and increased divergence among emissions scenar- GLM for mammals, GAM, GBM, GLM and RF for ios. Variability across BEMs was strongly af- birds, and GBM and GLM for amphibians and fected by RF projections that displayed higher snakes for most GCM cluster and scenario com- losses and gains of suitable climate across cli- binations; Anosim p-value<0.05, see Appendix mates and emissions scenarios, departing from S6 for results on 100% of the data). The ‘central cluster’ projections were also the most consen-

38 | Chapter I

Figure 1 | Proportion of suitable climate projected to be lost and gained by species for the seven bioclimatic envelope models (BEM) (a) and the five BEM consensus projections (b) under alternative climate projections. Values are median percentages of grid cells lost or gained for all species of amphibians (n=284), snakes (n=310), mammals (n=623) and birds (n=1,506) in mid- (open circles) and late-century (solid circles). For a given time period, each circle corresponds to one of the nine combinations of three emissions scenarios and three general circulation model (GCM) clusters. The BEMs are Artificial Neural Networks (ANN), Generalised Additive Models (GAM), Generalised Boosting Model (GBM), Generalised Linear Models (GLM), Multivariate Adaptive Regression Splines (MARS), Flexible Discriminant Analysis (FDA), and Random Forests (RF), and the BEM consensus are ensemble mean (EMean), ensemble weighted mean (EWMean), ensemble median (EMed), central model (CMod), and central cluster (CClus). sual, raising the median levels of consensus (Figure 4c; for each taxon, distribution of com- across all species for the four taxa, although to mission errors across species significantly dif- different degrees depending on the number and ferent from the remaining consensus projec- spread of projections in the cluster (Appendix tions according to Wilcoxon signed rank tests, S7). The ‘central model’ methodology, in turn, see Appendix S9). For birds in particular, the yielded high accuracy projections for amphibi- models combined in the ‘central cluster’ projec- ans and snakes but the lowest accuracy of all tions included RF, which incorporate the notion consensus projections for mammals and birds. of ensemble forecasting (Araújo & New, 2007) Whereas for a large fraction of amphibian and and can yield highly accurate projections snake species the ‘central model’ corresponded (Prasad et al., 2006). The extreme discrepancy to high-accuracy models (GBM and ANN), for in commission error of RF models from the mammal and bird species the selection covered other projections, however, suggests that these a wider range of models (see Appendix S8 for models may have over-fitted the training data results on 100% of the data) with varying levels (see Jimenez-Valverde et al., 2008). For am- of accuracy, yielding lower median TSS across phibians and snakes, the measurements of species. commission error may have been affected by High accuracy of consensus projections random pseudo-absences placed in climatically was linked to both low omission and commis- suitable areas (Anderson et al., 2003, Peterson sion errors (Figure 4b and c). The ‘central clus- et al., 2011). ter’ projections displayed lower numbers of known absence points incorrectly predicted

Ensemble forecasting for African vertebrates | 39

Figure 2 | Spatial variation of the relative contribution of future climate projections and bioclimatic envelope models to the variability in species turnover forecasts for amphibian, snake, mammal and bird species. Values shown correspond to the proportion of the total sum of squares accounted for by the bioclimatic envelope models (BEM), general circulation model clusters (GCMcons), emissions scenarios (SRES), and one interaction factor (GCMcons:SRES) in the three-way analysis of variance (ANOVA) performed for each grid cell over the study area (N=1,851) on turnover forecasts for each taxon. Data are shown for mid- and late-century.

Variation in forecasts with alternative BEM terns of species turnover for each taxon, with consensus methodologies and climate projections larger variations for amphibians and snakes between the ‘central cluster’ and the remaining When the BEMs were combined and the rela- consensus projections (Appendix S11). Consis- tive sources of uncertainty re-assessed in a tent across all taxa was a trend towards more point-wise ANOVA, differences emerged across species contracting their bioclimatic envelopes taxa. For amphibians and snakes, mid-century than expanding. Depending on the BEM con- turnover forecasts varied most with the BEM sensus projection, between 54% and 74% of consensus methodologies (median proportions the species in each taxon were consistently of the total sum of squares across the study estimated to lose suitable climate across emis- area of 37% and 43% respectively, decreasing sions scenarios by late-century (Figure 5a). to 24% and 29% by late-century; see Table 2). In contrast, alternative climate projections For these taxa, ‘central cluster’ median esti- had the largest impact on mid-century projec- mates of late-century species turnover were up tions for mammals and birds, with the GCM to 1.6 times higher than those produced by the clusters explaining 34% and 40% respectively most conservative BEM consensus methodolo- of the total sum of squares in the ANOVA (Table gies across all climate projections (Appendix 2). Towards the end of the century, the spread S10). By contrast, only up to 1.2-fold differ- across emissions scenarios increased, becoming ences resulted for mammal and bird species. the major source of uncertainty for all taxa. For For each taxon, however, there were consistent each scenario, the three clusters of GCMs ob- trends among consensus methodologies of tained (Anosim statistics 0.81 (A2), 0.72 (A1B) median losses and gains across climate projec- and 0.81 (B1), P=0.001) reflected a warming tions (Figure 1b). Geographically, the five BEM gradient. The maximum consensus ‘central consensus projections displayed similar pat-

40 | Chapter I

of the coldest month; see Appendix S12). Trends across the GCM clusters were less clear for precipitation forecasts, and sometimes showing contrasting directions of change (Ap- pendix S13), but cluster 3 was consistently the driest for all scenarios. Following the warming gradient, median late-century turnover rates across the study area almost doubled from cluster 1 under B1 to cluster 3 under A2 (Ap- pendix S10). For all taxa, a southern area centred in the arid regions of Namibia emerged with high turnover rates by late-century (Figure 6). How- ever, the geographical extent of this high- turnover effect varied from isolated patches of the Kalahari in Namibia and Botswana and of southern Mozambique for cluster 1 under B1, Figure 3 | Comparison of uncertainties arising to most of inland Namibia, Botswana and from bioclimatic envelope models between ana- southern Mozambique for cluster 3 under A2. A logue and non-analogue climate grid cells, for comparison of the areas projected to remain amphibians, snakes, mammals and birds. The climatically suitable for species over time re- maps (a) show the distribution over the study area (N=1,851) of non-analogue climates (dark vealed more striking differences across taxa grey) for scenarios A2 (N=738 non-analogue grid (Figure 7). For amphibians, it was the West cells), A1B (N=632) and B1 (N=374). The graphs African forests that were projected to remain (b) show the frequency of the distributions of the climatically suitable for the largest proportion proportions of the total sum of squares accounted for by bioclimatic envelope models in the point- of species, irrespective of climate projection. A wise three-way analysis of variance (ANOVA) significant proportion of snake species were performed for each taxon using species turnover also projected to persist in this area, but under projections as response variable. The difference is the B1 scenario also in the Albertine Rift moun- shown between analogue (light grey) and non- analogue (dark grey) climate grid cells over the tain forests and extending around the Congo study area (Kolmogorov-Smirnov tests p- Basin. In comparison, larger climatically stable value<0.001 across taxa). Data refer to the ‘maxi- areas were projected for mammal and bird mum consensus’ general circulation model cluster species, particularly under B1, with the highest (cluster 2) under the three emissions scenarios. proportions of species persisting in the Ethio- pian highland mountain, Albertine Rift, East cluster’ (cluster 2) reflected intermediate levels Africa montane and Eastern Arc forests, as well of warming. The other clusters captured low- as the Angolan scarp and Miombo woodlands (cluster 1) and high-end (cluster 3) tempera- and the Drakensberg and eastern coast of South ture variability across climate models. For the Africa for bird species. Projections for cluster 1 temperature of the warmest month, for exam- under B1 showed the lowest proportions of ple, median values of late-century anomalies species of all taxa consistently predicted to projected over the study area varied between contract across all BEM consensus projections 1.8°C (lower and upper quartiles 1.6-2.0%) for (Figure 5b). the low-end cluster 1 under B1 and 5.1°C (4.6- 5.7°C) for the high-end cluster 3 under A2 (similar patterns emerged for the temperature

Ensemble forecasting for African vertebrates | 41

Figure 4 | Comparison between average True Skill Statistics (TSS) of bioclimatic envelope models (BEM) and BEM consensus projections (a), and density functions of omission error (b) and commission error (c) of BEM consensus projections for all species in each taxon. Omission and commission error and TSS values refer to the median of the validation datasets (and pseudo-absence runs in the case of amphibians and snakes). The TSS plots (a) show the median (full circles) and the upper and lower quartiles (the extremes of the horizontal lines) of the TSS values of all species for each BEM consensus projection, as well as the median TSS values of all species for each BEM projection (black vertical lines). The consensus projections are the ensemble mean (EMean), ensemble weighted mean (EWMean), ensemble median (EMed), central model (CMod) and central cluster (CClus). The CMod and CClus projections shown were built using late-century projections from the GCM cluster 2 under A1B. The individual models are Artifi- cial Neural Networks (ANN), Generalised Additive Models (GAM), Generalised Boosting Model (GBM), Generalised Linear Models (GLM), Multivariate Adaptive Regression Splines (MARS), Flexible Discrimi- nant Analysis (FDA), and Random Forests (RF).

Discussion data, and the thresholds used to convert probabilistic to binary projections (Araújo et al., We explored the spread in estimates of climati- 2005b, Nenzén & Araújo, 2011). Yet, the uncer- cally suitable areas for African vertebrates tainty arising from future climate projections using alternative climate projections and emis- and from BEMs, which differ in how they ex- sions scenarios, as well as BEMs and combina- trapolate to non-analogue climates, has particu- tions thereof. Our aim was to address the lar relevance in the climate change context. methodological uncertainty associated with the results. Other methodological factors that are Sources of uncertainty not accounted for in our analysis are likely to Although comparisons of sources of uncertainty add further uncertainty to the projections. They in forecasts depend on the amount of variability include gaps and biases in species occurrence captured by each source (Diniz-Filho et al., data (Hortal et al., 2008), the selection of pre- 2009), our assessment spanned a wide variabil- dictors (Synes & Osborne, 2011), the temporal ity for the three sources considered: seven (Roubicek et al., 2010) and spatial resolution classes of BEM that perform differently (Kriticos & Leriche, 2010) of baseline climate

42 | Chapter I

Table 2 | Relative contribution of alternative general circulation models and bioclimatic envelope model consensus methodologies to overall uncertainty in species turnover projections for amphibian, snake, mammal and bird species. The values are proportions of the total sum of squares from the three-way analysis of variance (ANOVA) performed for each grid cell over the study area (N=1,851) to evaluate the relative contributions of the bioclimatic envelope model consensus methodologies (BEMcons), the general circulation model clusters (GCMcons), the emissions scenarios (SRES) and interactions among these factors, to the variability in turnover forecasts for each taxon. Values correspond to the median and (in brackets) the lower and upper quartiles of the proportions of the total sum of squares attributed to each factor over the study area, and are shown for both mid- and late-century. Amphibians Snakes Mammals Birds 2041-60 BEMcons 36.5 (20.4-60.5) 42.8 (24.8-60.6) 24.7 (11.8-44.1) 20.6 (11.4-34.9) GCMcons 24.4 (9.2-40.7) 25.7 (11.6-41.3) 34.1 (19.0-50.4) 40.2 (24.5-54.7) SRES 9.8 (4.4-16.3) 10.7 (6.0-16.7) 15.2 (9.5-21.7) 18.2 (11.7-25.7) BEMcons:GCMcons 3.4 (1.7-6.4) 3.1 (1.8-5.2) 3.2 (1.8-5.9) 2.0 (1.1-3.3) BEMcons:SRES 1.9 (0.9-3.8) 1.9 (1.1-3.4) 2.3 (1.3-4.1) 2.0 (0.9-4.7) GCMcons:SRES 4.2 (2-7.6.0) 3.9 (2.1-6.8) 4.3 (2.2-7.5) 4.5 (2.4-8.1) BEMcons:GCMcons:SRES 4.3 (2.2-8.5) 3.7 (2.1-5.9) 4.3 (2.5-7.5) 2.4 (1.3-4.4) 2081-00 BEMcons 23.6 (7.6-48.8) 28.8 (11.3-47) 12.7 (3.6-29.3) 12.4 (3.3-24.7) GCMcons 19.2 (8.9-27.7) 19.8 (11.7-27.2) 24.7 (17.6-31.1) 28.0 (21.3-34.6) SRES 28.5 (13.3-48.7) 32.1 (19.6-48.8) 44.2 (31.4-57.1) 45.8 (32.3-57.3) BEMcons:GCMcons 1.9 (0.9-3.7) 1.5 (0.7-3.2) 1.4 (0.6-2.8) 1.0 (0.4-1.8) BEMcons:SRES 1.8 (0.9-3.7) 1.7 (0.9-3.3) 1.9 (0.8-3.7) 1.4 (0.6-2.7) GCMcons:SRES 4.3 (2.2-8.2) 3.8 (2.0-6.9) 3.6 (2.0-6.1) 4.0 (2.0-6.8) BEMcons:GCMcons:SRES 3.4 (1.5-8.5) 2.6 (1.5-5.9) 2.4 (1.1-4.7) 1.2 (0.7-2.3)

Figure 5 | Percentage of species predicted to lose or gain suitable climate consistently across the three emissions scenarios (a) and across the five BEM consensus methodologies (b). The percentages of species of amphibians (n=284), snakes (n=310), mammals (n=623) and birds (n=1,506) that are consistently projected to contract (lighter tones, left side of graphs) or expand (darker tones, right side of graphs) their bioclimatic envelopes for the ‘maximum consensus’ general circulation model cluster across all emissions scenarios (A2, A1B and B1) are shown for each consensus methodology (a). The percentages of species in each taxon that are consistently projected to contract or expand across all BEM consensus projections (EMean=ensemble mean, EWMean=ensemble weighted mean, EMed=ensemble median, CMod=central model and CClus=central cluster) are shown for each climate projection (b). Data are shown for mid- and late-century.

Ensemble forecasting for African vertebrates | 43

Figure 6 | Projected late-century species turnover rate (%) for amphibian, snake, mammal and bird species under the alternative emissions scenarios (A2, A1B and B1) and general circulation model clusters (Clusters 1, 2 and 3). Data refer to the ensemble median of all bioclimatic envelope model projections.

when projected to the future (Thuiller et al., 2004a, Araújo et al., 2006, Pearson et al., 2006), three emissions scenarios that span a good portion of the range of the six IPCC scenarios (Manning et al., 2010), and 17 of the 23 IPCC fourth Assessment Report GCMs. In line with previous studies (Araújo et al., 2005b, Pearson et al., 2006, Diniz-Filho et al., 2009, 2010, Nenzén & Araújo, 2011), mid-century projections were most affected by the choice of BEM (Figure 2). The relative amount of variation introduced by BEMs, however, decreases over time (Buisson et al., 2010 and this study), as divergence among emissions scenarios increases (Stott & Kettleborough, 2002, Hawkins & Sutton, 2009). By late-century, two-fold dif- Figure 7 | Percentage of species predicted to ferences emerged in species turnover projec- retain climatic suitability under each emissions tions with alternative GCM clusters and emis- scenario. The proportion of the total numbers of species of amphibians (n=284), snakes (n=310), sions scenarios. mammals (n=623) and birds (n=1,506) that are Differences among projections from alter- projected to retain climatic suitability in each native climate projections were especially im- location are shown for the median ensemble of all portant in southern Africa (Figure 2), where bioclimatic envelope models and for the ‘maximum consensus’ general circulation model cluster some of the greatest changes in species turn- under the A2, A1B and B1 emissions scenarios. over were projected (Figure 6). In contrast, variability from BEMs was especially high in the northern half of the study area (Figure 2),

44 | Chapter I

coinciding with projected non-analogue cli- likely to be higher (Platts et al., 2008, mates (Figure 3). The problem of model ex- Fitzpatrick & Hargrove, 2009, Elith et al., 2010, trapolation into 21st century novel climates has Araújo et al., 2011, Roberts & Hamann, 2011) been overlooked in most continental-scale and using a range of BEMs that make different studies in Africa and elsewhere, although there assumptions about the responses of species in have been efforts to quantify it at different those areas (Pearson et al., 2006). geographical scales, for example in Europe (Saetersdal et al., 1998, Thuiller et al., 2004b, Role of consensus approaches Araújo et al., 2011), North America (Roberts & Although multi-model averages of GCMs have Hamann, 2011), and Australia (Fitzpatrick & rarely been used by ecologists (but see Hargrove, 2009, Elith et al., 2010). Projected Beaumont et al., 2011, Roberts & Hamann, novel climate conditions are unevenly distrib- 2011), their value for global change studies has uted worldwide (Williams et al., 2007). In Af- been recognised (Beaumont et al., 2008, rica, forecasts of severe warming increase the Fordham et al., in press). Our ensemble of GCMs risk of non-analogue climates, particularly reflects availability, and may thus not sample under high-end emissions scenarios (Williams the full range of uncertainty or guarantee inde- et al., 2007; Appendix S5). pendence of models (Meehl et al., 2007, Tebaldi Solutions to the problem of extrapolation & Knutti, 2007, Knutti et al., 2008), but the into non-analogue climates include classifying large number of models included is expected to such areas as species absences (Austin & minimise potential biases by model choice Meyers, 1996, Thuiller et al., 2004b), excluding (Knutti et al., 2010). Our approach to combine them from the analysis (Saetersdal et al., 1998), groups of co-varying GCMs enabled us to retain or using a larger calibration area before pro- information about the full variability of projec- jecting to the region of interest (Pearson et al., tions, including those that are extreme, and 2006). Yet, such solutions may lead to mislead- minimise the effect of combining often diver- ing results when the bioclimatic envelope of the gent precipitation projections. In face of the species has not been fully captured by the cali- wealth of climate simulations available, this bration data or when the species is not in equi- approach might prove increasingly useful in librium with climate (Svenning & Skov, 2004, ecological modelling studies. Araújo & Pearson, 2005). With long time hori- How alternative methodologies to combine zons, novel climates are expected to become BEMs affect forecasts of species bioclimatic more widespread (Williams et al., 2007), and envelopes has rarely been investigated (but see the problem of extrapolation may persist even Araújo et al., 2005b, Araújo et al., 2006, when using larger calibration areas, and all the Marmion et al., 2009). Measurements of model more when the species response curves are accuracy in the baseline context do not neces- high or increasing where truncated (Anderson sarily provide an indication of the models’ abil- & Raza, 2010, Webber et al., 2011). The ideal ity to transfer into future conditions (Araújo et solution of using ecological or physiological al., 2005a). However, in the absence of inde- knowledge of the species to classify non- pendent data for evaluation of the models, they analogue climate areas as either presences or can be used as benchmark for verification of the absences (Elith et al., 2010) is not feasible for consistency of alternative consensus method- most species where these data are lacking. The ologies. Using a variety of methodologies, we approach used in our study may thus reflect the found that consensus projections displayed current best practice: mapping non-analogue greater consistency in accuracy for amphibians climates to identify where uncertainties are and snakes. Whereas the three consensus

Ensemble forecasting for African vertebrates | 45

methodologies averaging the full ensemble of tat structure further limit the response of ver- BEM projections were generally consistent for tebrates. African habitats are controlled by the all taxa, for mammals and birds they diverged interaction among climate, atmospheric CO2 from those methodologies combining only pre- and disturbances like fire (Scheiter & Higgins, selected models (Figure 4a). This distinction 2009). In the case of grass-dominated savannas, across taxa resulted from the greater spread of the low levels of CO2 that triggered their devel- projection accuracy from the seven BEMs for opment during the last glacial are clearly being mammals and birds. For all taxa, however, the surpassed, with expected changes in tree cover ‘central cluster’ methodology stood out as the (Bond et al., 2003, Kgope et al., 2010) and asso- most accurate projection. In the case of am- ciated fauna (Sirami et al., 2009). The use in phibians and snakes, accurate and constrained BEMs of vegetation predictors derived from baseline ‘central cluster’ projections led to mechanistically-based dynamic vegetation higher species turnover rates than the remain- models (Triviño et al., In press) could capture ing consensus methodologies (Appendix S11). these effects and, to some extent, reduce uncertainties arising in non-analogue situations. Interpreting BEM outputs Exposure of species to climatic changes was measured in our study by three climatic The climatically suitable areas identified in this variables. The realism of our models thus de- study are areas where the exposure of species pends on the relationship between species to climatic changes can potentially allow the distributions and these variables. This relation- persistence of vertebrates through time. As- ship is unlikely to be constant through time, not sessing the potential response of species to only for its complexity but also for the changing these changes would necessitate mechanistic correlations among climatic variables (Morin & information about their sensitivity and adap- Lechowicz, 2008). Decoupling of patterns of tive capacity. Correlative studies have used covariance between predictor and proximal such information as a source of data to infer variables may undermine the value of BEMs absence (Elith et al., 2010) or presence points when extrapolating in time (Jackson et al., (Hijmans & Graham, 2006), as model predictors 2009, Elith et al., 2010). Our variables also (Rödder et al., 2009), or as complementary discount climatic variability nested across tem- information (Morin & Thuiller, 2009), but simi- poral and spatial scales. Multi-decadal climatic lar applications to large datasets like ours are means fail to capture fluctuations and rapid limited by data availability. Information on transitions between climate states (Jackson et species’ dispersal capacity, for example, would al., 2009). Projected changes in mean precipita- determine whether new suitable climate space tion in Southern Africa, for example, differ be- is accessible to species. Given the known varia- tween seasons and do not always parallel those tion in dispersal capacity among the four taxa, in extreme precipitation (Shongwe et al., 2009). effective range shifts would diverge more However, it is the interplay between temporal across taxa than the projected gains of suitable variability and species survival thresholds that climate (Figure 1). Mapping areas that remain determines the effects on species (Jackson et al., climatically suitable for most species through 2009). By the same token, the coarse spatial time reflects potential persistence of species in resolution used here overlooks microclimates situ (Figure 7), whereas integrating dispersal provided by topographical or vegetation fea- rate estimates in BEMs (Midgley et al., 2006) or tures (e.g. Shoo et al., 2011). combining BEM and migration model projec- BEM outputs need to be interpreted in the tions (Iverson et al., 2004) would indicate po- light of the methodological uncertainty ex- tential range shifts. Biotic factors such as habi-

46 | Chapter I

plored and the biological limitations discussed roecology, Evolution and Climate, and NDB also above. If carefully implemented, BEMs can, thanks WWF US. therefore, provide a first-order, parsimonious assessment of the changes in the distribution of suitable climate for species (Thuiller et al., References 2008, Jackson et al., 2009, Chevin et al., 2010, Algar AC, Kharouba HM, Young ER, Kerr JT (2009) Huntley et al., 2010). A new generation of mod- Predicting the future of species diversity: els that couple correlative with mechanistic macroecological theory, climate change, and direct tests of alternative forecasting methods. Ecography, approaches (Keith et al., 2008, Anderson et al., 32, 22-33. 2009, Brook et al., 2009, Huntley et al., 2010) is Allouche O, Tsoar A, Kadmon R (2006) Assessing the required to allow predictions of population accuracy of species distribution models: prevalence, persistence that have direct relevance to local kappa and the true skill statistic (TSS). Journal of Applied Ecology, 43, 1223-1232. or regional conservation (Jackson et al., 2009, Anderson BJ, Akcakaya HR, Araújo MB, Fordham DA, Chevin et al., 2010). Yet our ensemble forecast- Martinez-Meyer E, Thuiller W, Brook BW (2009) ing implementation of BEMs provides general Dynamics of range margins for metapopulations insights into the potential exposure of sub- under climate change. Proceedings of the Royal Society B-Biological Sciences, 276, 1415-1420. Saharan African vertebrates to climate change Anderson RP, Lew D, Peterson AT (2003) Evaluating at the continental scale, as well as a critical predictive models of species' distributions: criteria background for those seeking to interpret these for selecting optimal models. Ecological Modelling, 162, 211-232. results and use them as the basis for decision- making at this large scale. Anderson RP, Raza A (2010) The effect of the extent of the study region on GIS models of species geographic distributions and estimates of niche evolution: preliminary tests with montane rodents (genus Nephelomys) in Venezuela. Journal of Biogeography, Acknowledgements 37, 1378-1393. We are grateful to Thomas Phillips for advice Araújo MB, Pearson RG (2005) Equilibrium of species’ distributions with climate. Ecography, 28, 693-695. regarding the comparison of General Circula- Araújo MB, Pearson RG, Thuiller W, Erhard M (2005a) tion Models, and to Diogo Alagador for assis- Validation of species–climate impact models under tance in statistical analyses and R coding. climate change. Global Change Biology, 11, 1504- 1513. Thanks to Hedvig Nenzén and Sara Varela for Araújo MB, Whittaker RJ, Ladle RJ, Erhard M (2005b) commenting on the manuscript, and to three Reducing uncertainty in projections of extinction risk anonymous reviewers for constructive criti- from climate change. Global Ecology and cism. RAG is funded through a PhD studentship Biogeography, 14, 529-538. from the Portuguese Foundation for Science Araújo MB, Thuiller W, Pearson RG (2006) Climate warming and the decline of amphibians and reptiles and Technology with co-funding from the in Europe. Journal of Biogeography, 33, 1712-1728. European Social Fund Araújo MB, Luoto M (2007) The importance of biotic (SFRH/BD/65615/2009). Research by MBA is interactions for modelling species distributions funded through the FCT PTDC/AAC- under climate change. Global Ecology and Biogeography, 16, 743-753. AMB/98163/2008 project. RAG, MC and MBA Araújo MB, New M (2007) Ensemble forecasting of acknowledge the Spanish Research Council species distributions. Trends in Ecology & Evolution, (CSIC) for support, and RAG and MBA also 22, 42-47. thank the ‘Rui Nabeiro/Delta’ Biodiversity Araújo MB, Thuiller W, Yoccoz NG (2009) Reopening the Chair. MBA and MC also received support climate envelope reveals macroscale associations with climate in European birds. Proceedings of the through the RESPONSES project. NDB, CR, RAG National Academy of Sciences, 106, E45-E46. and MBA thank the Danish National Research Araújo MB, Alagador D, Cabeza M, Nogués-Bravo D, Foundation for supporting the Center of Mac- Thuiller W (2011) Climate change threatens

Ensemble forecasting for African vertebrates | 47

European conservation areas. Ecology Letters, 14, Chevin L-M, Lande R, Mace GM (2010) Adaptation, 484-492. Plasticity, and Extinction in a Changing Environment: Towards a Predictive Theory. PLoS Biol, 8, e1000357. Austin MP, Meyers JA (1996) Current approaches to modelling the environmental niche of eucalypts: Chown S, Hoffmann A, Kristensen T, Angilletta MJ, implication for management of forest biodiversity. Stenseth N, Pertoldi C (2010) Adapting to climate Forest Ecology and Management, 85, 95-106. change: a perspective from evolutionary physiology. Climate Research, 43, 3-15. Barbet-Massin M, Walther BA, Thuiller W, Rahbek C, Jiguet F (2009) Potential impacts of climate change Chown SL (2010) Temporal biodiversity change in on the winter distribution of Afro-Palaearctic transformed landscapes: a southern African migrant passerines. Biology Letters, 5, 248-251. perspective. Philosophical Transactions of the Royal Society B-Biological Sciences, 365, 3729-3742. Beaumont LJ, Hughes L, Pitman AJ (2008) Why is the choice of future climate scenarios for species Christensen JH, Hewitson B, Busuioc A et al. (2007) distribution modelling important? Ecology Letters, Regional Climate Projections. In: Climate Change 11, 1135-1146. 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of Beaumont LJ, Pitman A, Perkins S, Zimmermann NE, the Intergovernmental Panel on Climate Change (eds Yoccoz NG, Thuiller W (2011) Impacts of climate Solomon S, Qin D, Manning M, Chen Z, Marquis M, change on the world's most exceptional ecoregions. Averyt K B, Tignor M, Miller H L) pp 847-940. Proceedings of the National Academy of Sciences, 108, Cambridge University Press, Cambridge, United 2306-2311. Kingdom and New York, NY, USA. Biggs R, Simons H, Bakkenes M, Scholes RJ, Eickhout B, Clarke KR, Warwick RM (1994) Similarity-based testing van Vuuren D, Alkemade R (2008) Scenarios of for community pattern: the two-way layout with no biodiversity loss in southern Africa in the 21st replication. Marine Biology, 118, 167-176. century. Global Environmental Change-Human and Policy Dimensions, 18, 296-309. Coetzee BWT, Robertson MP, Erasmus BFN, Rensburg BJv, Thuiller W (2009) Ensemble models predict Boko M, Niang I, Nyong A et al. (2007) Africa. In: Climate Important Bird Areas in southern Africa will become Change 2007: Impacts, Adaptation and Vulnerability. less effective for conserving endemic birds under Contribution of Working Group II to the Fourth climate change. Global Ecology and Biogeography, 18, Assessment Report of the Intergovernmental Panel on 701-710. Climate Change (eds Parry M L, Canziani O F, Palutikof J P, Van Der Linden P J, Hanson C E), pp Collier P, Conway G, Venables T (2008) Climate change 433-467. Cambridge University Press, Cambridge UK. and Africa. Oxford Review of Economic Policy, 24, 337- 353. Bond WJ, Midgley GF, Woodward FI (2003) The importance of low atmospheric CO2 and fire in Crossman ND, Bass DA (2008) Application of common promoting the spread of grasslands and savannas. predictive habitat techniques for post-border weed Global Change Biology, 9, 973-982. risk management. Diversity and Distributions, 14, 213-224. Brook BW, Akçakaya HR, Keith DA, Mace GM, Pearson RG, Araújo MB (2009) Integrating bioclimate with Dawson TP, Jackson ST, House JI, Prentice IC, Mace GM population models to improve forecasts of species (2011) Beyond Predictions: Biodiversity extinctions under climate change. Biology Letters, 5, Conservation in a Changing Climate. Science, 332, 53- 723-725. 58. Brooks T, Balmford A, Burgess N et al. (2001) Toward a Deutsch CA, Tewksbury JJ, Huey RB, Sheldon KS, Blueprint for Conservation in Africa. Bioscience, 51, Ghalambor CK, Haak DC, Martin PR (2008) Impacts of 613-624. climate warming on terrestrial ectotherms across latitude. Proceedings of the National Academy of Buisson L, Grenouillet G (2009) Contrasted impacts of Sciences, 105, 6668-6672. climate change on stream fish assemblages along an environmental gradient. Diversity and Distributions, Diniz-Filho JAF, Mauricio Bini L, Fernando Rangel T, 15, 613-626. Loyola RD, Hof C, Nogués-Bravo D, Araújo MB (2009) Partitioning and mapping uncertainties in ensembles Buisson L, Thuiller W, Casajus N, Lek S, Grenouillet G of forecasts of species turnover under climate (2010) Uncertainty in ensemble forecasting of change. Ecography, 32, 897-906. species distribution. Global Change Biology, 16, 1145- 1157. Diniz-Filho JAF, Nabout JC, Bini LM, Loyola RD, Rangel TF, Nogues-Bravo D, Araújo MB (2010) Ensemble Burgess N, Fjeldså J, Rahbek C (1998) Mapping the forecasting shifts in climatically suitable areas for distributions of Afrotropical vertebrate groups. Tropidacris cristata (Orthoptera: Acridoidea: Species, 30, 16-17. Romaleidae). Insect Conservation and Diversity, 3, Chefaoui RM, Lobo JM (2008) Assessing the effects of 213-221. pseudo-absences on predictive distribution model Duan Q, Phillips TJ (2010) Bayesian estimation of local performance. Ecological Modelling, 210, 478-486. signal and noise in multimodel simulations of climate

48 | Chapter I

change. Journal of Geophysical Research, 115, Hawkins BA, Field R, Cornell HV et al. (2003) Energy, D18123. water, and broad-scale geographic patterns of species richness. Ecology 84(12), 3105–3117 Elith J, Graham CH, Anderson RP et al. (2006) Novel methods improve prediction of species' distributions Hawkins E, Sutton R (2009) The Potential to Narrow from occurrence data. Ecography, 29, 129-151. Uncertainty in Regional Climate Predictions. Bulletin of the American Meteorological Society, 90, 1095- Elith J, Graham CH (2009) Do they? How do they? WHY 1107. do they differ? On finding reasons for differing performances of species distribution models. Hawkins E, Sutton R (2011) The potential to narrow Ecography, 32, 66-77. uncertainty in projections of regional precipitation change. Climate Dynamics, 37, 407-418. Elith J, Kearney M, Phillips S (2010) The art of modelling range-shifting species. Methods in Ecology and Heikkinen RK, Luoto M, Araújo MB, Virkkala R, Thuiller Evolution, 1, 330-342. W, Sykes MT (2006) Methods and uncertainties in bioclimatic envelope modelling under climate ESRI (2006) ArcGIS. Environmental Systems Research change. Progress in Physical Geography, 30, 751-777. Institute (ESRI). Redlands CA. Hijmans RJ, Graham CH (2006) The ability of climate Felton A, Fischer J, Lindenmayer D et al. (2009) Climate envelope models to predict the effect of climate change, conservation and management: an change on species distributions. Global Change assessment of the peer-reviewed scientific journal Biology, 12, 2272-2281. literature. Biodiversity and Conservation, 18, 2243- 2253. Hockey PAR, Midgley GF (2009) Avian range changes and climate change: a cautionary tale from the Cape Fitzpatrick M, Hargrove W (2009) The projection of Peninsula. Ostrich, 80, 29-34. species distribution models and the problem of non- analog climate. Biodiversity and Conservation, 18, Hof C, Levinsky I, Araújo MB, Rahbek C (2011) 2255-2261. Rethinking species' ability to cope with rapid climate change. Global Change Biology, 17, 2987-2990. Fordham D, Wigley T, Brook B (in press) Multi-model climate projections for biodiversity risk assessments. Hole DG, Willis SG, Pain DJ et al. (2009) Projected Ecological Applications, doi:10.1890/1811- impacts of climate change on a continent-wide 0314.1891. protected area network. Ecology Letters, 12, 420-431. Galster S, Burgess ND, Fjeldså J, Hansen LA, Rahbek C Hortal J, Jiménez-Valverde A, Gómez JF, Lobo JM, Baselga (2007) One degree resolution databases of the A (2008) Historical bias in biodiversity inventories distribution of 1085 species of mammals in Sub- affects the observed environmental niche of the Saharan Africa. On-line data source-Version 1.00. species. Oikos, 117, 847-858. Zoological Museum, University of Copenhagen, Huntley B, Collingham YC, Green RE, Hilton GM, Rahbek Denmark. C, Willis SG (2006) Potential impacts of climatic Gleckler PJ, Taylor KE, Doutriaux C (2008) Performance change upon geographical distributions of birds. Ibis, metrics for climate models. Journal of Geophysical 148, 8-28. Research, 113, D06104. Huntley B, Barnard P, Altwegg R et al. (2010) Beyond Grenouillet G, Buisson L, Casajus N, Lek S (2011) bioclimatic envelopes: dynamic species' range and Ensemble modelling of species distribution: the abundance modelling in the context of climatic effects of geographical and environmental ranges. change. Ecography, 33, 621-626. Ecography, 34, 9-17. Iverson LR, Schwartz MW, Prasad AM (2004) Potential Guisan A, Zimmermann NE (2000) Predictive habitat colonization of newly available tree-species habitat distribution models in ecology. Ecological Modelling, under climate change: an analysis for five eastern US 135, 147-186. species. Landscape Ecology, 19, 787-799. Hansen AJ, Burgess ND, Fjeldså J, Rahbek C (2007a) One Jackson ST, Betancourt JL, Booth RK, Gray ST (2009) degree resolution databases of the distribution of 739 Ecology and the ratchet of events: Climate variability, species of amphibians in Sub-Saharan Africa. On-line niche dimensions, and species distributions. data source-version 1.00. Zoological Museum, Proceedings of the National Academy of Sciences, 106, University of Copenhagen, Denmark. 19685-19692. Hansen LA, Fjeldså J, Burgess ND, Rahbek C (2007b) One Jimenez-Valverde A, Lobo JM, Hortal J (2008) Not as degree resolution databases of the distribution of 1789 good as they seem: the importance of concepts in resident birds in Sub-Saharan Africa. On-line data species distribution modelling. Diversity and source-Version 1.00. Zoological Museum, University Distributions, 14, 885-890. of Copenhagen, Denmark. Kearney MR, Wintle BA, Porter WP (2010) Correlative Harrell FE, Lee KL, Califf RM, Pryor DB, Rosati RA (1984) and mechanistic models of species distribution Regression modelling strategies for improved provide congruent forecasts under climate change. prognostic prediction. Statistics in Medicine, 3, 143- Conservation Letters, 3, 203-213. 152.

Ensemble forecasting for African vertebrates | 49

Keith DA, Akçakaya HR, Thuiller W et al. (2008) Intergovernmental Panel on Climate Change (eds Predicting extinction risks under climate change: Solomon S, Qin D, Manning M, Chen Z, Marquis M, coupling stochastic population models with dynamic Averyt K B, Tignor M, Miller H L), pp 747-845. bioclimatic habitat models. Biology Letters, 4, 560- Cambridge University Press, Cambridge, United 563. Kingdom and New York, NY, USA. Kgope BS, Bond WJ, Midgley GF (2010) Growth Midgley GF, Hughes GO, Thuiller W, Rebelo AG (2006) responses of African savanna trees implicate Migration rate limitations on climate change-induced atmospheric [CO2] as a driver of past and current range shifts in Cape Proteaceae. Diversity and changes in savanna tree cover. Austral Ecology, 35, Distributions, 12, 555-562. 451-463. Midgley G, Thuiller W (2011) Potential responses of Knutti R, Allen MR, Friedlingstein P et al. (2008) A terrestrial biodiversity in Southern Africa to Review of Uncertainties in Global Temperature anthropogenic climate change. Regional Projections over the Twenty-First Century. Journal of Environmental Change, 11, 127-135. Climate, 21, 2651-2663. Min S-K, Hense A (2006) A Bayesian approach to climate Knutti R, Furrer R, Tebaldi C, Cermak J, Meehl GA (2010) model evaluation and multi-model averaging with an Challenges in Combining Projections from Multiple application to global mean surface temperatures Climate Models. Journal of Climate, 23, 2739-2758. from IPCC AR4 coupled climate models. Geophysical Research Letters, 33, 5pp. Kriticos DJ, Leriche A (2010) The effects of climate data precision on fitting and projecting species niche Morin X, Lechowicz MJ (2008) Contemporary models. Ecography, 33, 115-127. perspectives on the niche that can improve models of species range shifts under climate change. Biology Lambert SJ, Boer GJ (2001) CMIP1 evaluation and Letters, 4, 573-576. intercomparison of coupled climate models. Climate Dynamics, 17, 83-106. Morin X, Thuiller W (2009) Comparing niche- and process-based models to reduce prediction Lawler JJ, Shafer SL, White D, Kareiva P, Maurer EP, uncertainty in species range shifts under climate Blaustein AR, Bartlein PJ (2009) Projected climate- change. Ecology, 90, 1301-1313. induced faunal change in the Western Hemisphere. Ecology, 90, 588-597. Nakicenovic N, Alcamo J, Davis G et al. (2000) Special Report on Emissions Scenarios: A Special Report of Lovett JC, Midgely GF, Barnard P (2005) Climate change Working Group III of the Intergovernmental Panel on and ecology in Africa. African Journal of Ecology, 43, Climate Change, 599pp. Cambridge University Press, 167–169. Cambridge, U.K. MacKellar NC, Hewitson BC, Tadross MA (2007) Nenzén HK, Araújo MB (2011) Choice of threshold alters Namaqualand’s climate: recent historical changes projections of species range shifts under climate and future scenarios. Journal of Arid Environments, change. Ecological Modelling, 222, 3346– 3354. 70, 604-614. New M, Lister D, Hulme M, Makin I (2002) A high- Manning MR, Edmonds J, Emori S et al. (2010) resolution data set of surface climate over global land Misrepresentation of the IPCC CO2 emission areas. Climate Research, 21, 1-25. scenarios. Nature Geoscience, 3, 376-377. Ogawa-Onishi Y, Berry PM, Tanaka N (2010) Assessing Marini MÂ, Barbet-Massin M, Lopes LE, Jiguet F (2009) the potential impacts of climate change and their Major current and future gaps of Brazilian reserves conservation implications in Japan: A case study of to protect Neotropical savanna birds. Biological conifers. Biological Conservation, 143, 1728-1736. Conservation, 142, 3039-3050. Pearson RG, Dawson TP (2003) Predicting the impacts of Marini MÂ, Barbet-Massin M, Martinez J, Prestes NP, climate change on the distribution of species: are Jiguet F (2010) Applying ecological niche modelling bioclimate envelope models useful? Global Ecology to plan conservation actions for the Red-spectacled and Biogeography, 12, 361-371. Amazon (Amazona pretrei). Biological Conservation, 143, 102-112. Pearson RG, Thuiller W, Araújo MB et al. (2006) Model- based uncertainty in species range prediction. Marmion M, Parviainen M, Luoto M, Heikkinen RK, Journal of Biogeography, 33, 1704-1711. Thuiller W (2009) Evaluation of consensus methods in predictive species distribution modelling. Diversity Peterson AT, Ortega-Huerta MA, Bartley J, Sanchez- and Distributions, 15, 59-69. Cordero V, Soberón J, Buddemeier RH, Stockwell DRB (2002) Future projections for Mexican faunas under McClean CJ, Jon CL, Küper W et al. (2005) African Plant global climate change scenarios. Nature, 416, 626- Diversity and Climate Change. Annals of the Missouri 629. Botanical Garden, 92, 139-152. Peterson AT, Soberón J, Pearson RG, Anderson RP, Meehl GA, Stocker TF, Collins WD et al. (2007) Global Martínez-Meyer E, Nakamura M, Araújo MB (2011) Climate Projections. In: Climate Change 2007: The Ecological Niches and Geographic Distributions, Physical Science Basis. Contribution of Working Group 328pp. Monographs in Population Biology 49, I to the Fourth Assessment Report of the Princeton University Press.

50 | Chapter I

Phillips TJ, Gleckler PJ (2006) Evaluation of continental under Global Warming. Part I: Southern Africa. precipitation in 20th century climate simulations: Journal of Climate, 22, 3819-3837. The utility of multimodel statistics. Water Resources Shoo LP, Storlie C, Vanderwal J, Little J, Williams SE Research, 42, 10pp. (2011) Targeted protection and restoration to Pierce DW, Barnett TP, Santer BD, Gleckler PJ (2009) conserve tropical biodiversity in a warming world. Selecting global climate models for regional climate Global Change Biology, 17, 186-193. change studies. Proceedings of the National Academy Sinclair SJ, White MD, Newell GR (2007) How Useful Are of Sciences, 106, 8441-8446. Species Distribution Models for Managing Platts PJ, McClean CJ, Lovett JC, Marchant R (2008) Biodiversity under Future Climates? Ecology and Predicting tree distributions in an East African Society, 15, 8. biodiversity hotspot: model selection, data bias and Sirami C, Seymour C, Midgley G, Barnard P (2009) The envelope uncertainty. Ecological Modelling, 218, 121- impact of shrub encroachment on savanna bird 134. diversity from local to regional scale. Diversity and Prasad A, Iverson L, Liaw A (2006) Newer Classification Distributions, 15, 948-957. and Regression Tree Techniques: Bagging and Stockwell DRB, Peterson AT (2002) Effects of sample Random Forests for Ecological Prediction. size on accuracy of species distribution models. Ecosystems, 9, 181-199. Ecological Modelling, 148, 1-13. R Development Core Team (2010) R: A language and Stott PA, Kettleborough JA (2002) Origins and estimates environment for statistical computing. R Foundation of uncertainty in predictions of twenty-first century for Statistical Computing, Vienna, Austria. temperature rise. Nature, 416, 723-726. Rasmussen JB, Hansen LA, Burgess ND, Fjeldså J, Rahbek Suttle KB, Thomsen MA, Power ME (2007) Species C (2007) One degree resolution databases of the Interactions Reverse Grassland Responses to distribution of 467 species of snakes in Sub-Saharan Changing Climate. Science, 315, 640-642. Africa. On-line data source-Version 1.00. Zoological Museum, University of Copenhagen, Denmark. Svenning J-C, Skov F (2004) Limited filling of the potential range in European tree species. Ecology Reichler T, Kim J (2008) How Well Do Coupled Models Letters, 7, 565-573. Simulate Today's Climate? Bulletin of the American Meteorological Society, 89, 303-311. Synes NW, Osborne PE (2011) Choice of predictor variables as a source of uncertainty in continental- Roberts DR, Hamann A (2011) Predicting potential scale species distribution modelling under climate climate change impacts with bioclimate envelope change. Global Ecology and Biogeography, 20, 904- models: a palaeoecological perspective. Global 914. Ecology and Biogeography, DOI: 10.1111/j.1466- 8238.2011.00657.x. Tabor K, Williams JW (2010) Globally downscaled climate projections for assessing the conservation Rödder D, Schmidtlein S, Veith M, Lötters S (2009) Alien impacts of climate change. Ecological Applications, Invasive Slider Turtle in Unpredicted Habitat: A 20, 554-565. Matter of Niche Shift or of Predictors Studied? PLoS ONE, 4, e7843. Tebaldi C, Knutti R (2007) The use of the multi-model ensemble in probabilistic climate projections. Roubicek AJ, VanDerWal J, Beaumont LJ, Pitman AJ, Philosophical Transactions of the Royal Society A: Wilson P, Hughes L (2010) Does the choice of climate Mathematical, Physical and Engineering Sciences, 365, baseline matter in ecological niche modelling? 2053-2075. Ecological Modelling, 221, 2280-2286. Thuiller W (2004) Patterns and uncertainties of species' Roura-Pascual Nr, Brotons Ls, Peterson A, Thuiller W range shifts under climate change. Global Change (2009) Consensual predictions of potential Biology, 10, 2020-2027. distributional areas for invasive species: a case study of Argentine ants in the Iberian Peninsula. Biological Thuiller W, Araújo MB, Pearson RG, Whittaker RJ, Invasions, 11, 1017-1031. Brotons L, Lavorel S (2004a) Biodiversity conservation: Uncertainty in predictions of extinction Saetersdal M, Birks HJB, Peglar SM (1998) Predicting risk. Nature, 430, 145-148. changes in Fennoscandian vascular-plant species richness as a result of future climatic change. Journal Thuiller W, Brotons L, Araújo MB, Lavorel S (2004b) of Biogeography, 25, 111-122. Effects of restricting environmental range of data to project current and future species distributions. Scheiter S, Higgins SI (2009) Impacts of climate change Ecography, 27, 165-172. on the vegetation of Africa: an adaptive dynamic vegetation modelling approach. Global Change Thuiller W, Lavorel S, Araujo MB, Sykes MT, Prentice IC Biology, 15, 2224-2246. (2005) Climate change threats to plant diversity in Europe. Proceedings of the National Academy of Shongwe ME, van Oldenborgh GJ, van den Hurk BJJM, de Sciences of the United States of America, 102, 8245- Boer B, Coelho CAS, van Aalst MK (2009) Projected 8250. Changes in Mean and Extreme Precipitation in Africa

Ensemble forecasting for African vertebrates | 51

Thuiller W, Broennimann O, Hughes G, Alkemade JRM, performance of species distribution models. Diversity Midgley GF, Corsi F (2006) Vulnerability of African and Distributions, 14, 763-773. mammals to anthropogenic climate change under Wright SJ, Muller-Landau HC, Schipper JAN (2009) The conservative land transformation assumptions. Future of Tropical Species on a Warmer Planet. Global Change Biology, 12, 424-440. Conservation Biology, 23, 1418-1426. Thuiller W, Albert C, Araújo MB et al. (2008) Predicting global change impacts on plant species' distributions: Future challenges. Perspectives in Plant Ecology Evolution and Systematics, 9, 137-152. Supporting Information Thuiller W, Lafourcade B, Engler R, Araujo MB (2009) BIOMOD - a platform for ensemble forecasting of Appendix S1. Principal Components Analysis to species distributions. Ecography, 32, 369-373. select predictor variables. Thuiller W, Lavergne S, Roquet C, Boulangeat I, Araújo Appendix S2. General Circulation Models used MB (2011) Consequences of climate change to the Tree of Life in Europe Nature, 470, 531-534. in the study. Triviño M, Thuiller W, Cabeza M, Hickler T, Araújo MB Appendix S3. R scripts used to build the ‘cen- (In press) The Contribution of Vegetation and tral cluster’ consensus projections. Landscape Configuration for Predicting Appendix S4. Sources of uncertainty in the Environmental Change Impacts on Iberian Birds. ensemble of forecasts from individual BEMs. Velarde SJ, Malhi Y, Moran D, Wright J, Hussain S (2005) Valuing the impacts of climate change on protected Appendix S5. Non-analogue climate maps. areas in Africa. Ecological Economics, 53, 21-33. Appendix S6. Single-BEMs selected for the Venables WN, Ripley BD (2003) Modern Applied ‘central clusters’ and Anosim results. Statistics with S, xi+495pp. Springer-Verlag, New Appendix S7. Level of consensus among all York. BEMs and ‘central cluster’ BEMs. Webber BL, Yates CJ, Le Maitre DC et al. (2011) Modelling horses for novel climate courses: insights Appendix S8. Single-BEMs selected as ‘central from projecting potential distributions of native and model’ across species. alien Australian acacias with correlative and Appendix S9. mechanistic models. Diversity and Distributions, 17, Statistical differences in the dis- 978-1000. tributions of TSS and omission and commission Wiens JA, Stralberg D, Jongsomjit D, Howell CA, Snyder error among the five BEM consensus MA (2009) Niches, models, and climate change: projections. Assessing the assumptions and uncertainties. Proceedings of the National Academy of Sciences, 106, Appendix S10. Frequency distribution of spe- 19729-19736. cies turnover for alternative GCM clusters, Williams JW, Jackson ST, Kutzbach JE (2007) Projected emissions scenarios and BEM consensus distributions of novel and disappearing climates by methodologies. 2100 AD. Proceedings of the National Academy of Sciences, 104, 5738-5742. Appendix S11. Late-century species turnover Wisz MS, Guisan A (2009) Do pseudo-absence selection for alternative BEM consensus methodologies. strategies influence species distribution models and Appendix S12. Frequency distribution of cli- their predictions? An information-theoretic approach mate anomalies over the study area. based on simulated data. BMC Ecology, 9, 8. Appendix S13. Climate anomaly maps for the Wisz MS, Hijmans RJ, Li J, Peterson AT, Graham CH, Guisan A (2008) Effects of sample size on the three variables.

52 | Chapter I

Appendix S1: Principal Components Analysis to select predictor variables

21 predictor variables were computed using baseline monthly precipitation and temperature data for 1961-60 from the Climatic Research Unit (CRU, New et al., 2002). A principal components analysis (PCA) was performed on the values of these variables over the study area (N=1,851) in order to select a smaller set of variables and minimise co-linearity. The loadings for the first three principal components (PC1 to 3) are shown. The last three rows show the standard deviation, proportion of variance explained and cumulative variance explained by each PC.

Variables PC1 PC2 PC 3 mean temperature of the coldest month 0.022 0.365 0.016 mean temperature of the warmest month 0.277 0.194 -0.023 annual mean temperature 0.196 0.305 -0.028 annual sum of precipitation -0.295 0.161 0.095 annual mean precipitation -0.295 0.161 0.095 precipitation of the driest month -0.194 0.080 -0.558 precipitation of the wettest month -0.231 0.172 0.348 precipitation of the coldest month -0.143 0.171 0.036 precipitation of the warmest month -0.250 0.023 -0.208 mean temperature of the driest month 0.050 0.345 -0.027 mean temperature of the wettest month 0.268 0.161 -0.162 mean temperature of the coldest quarter 0.042 0.364 0.022 mean temperature of the warmest quarter 0.272 0.203 -0.038 precipitation of the driest quarter -0.209 0.097 -0.538 precipitation of the wettest quarter -0.246 0.158 0.343 precipitation of the coldest quarter -0.160 0.190 0.050 precipitation of the warmest quarter -0.270 0.020 -0.108 mean temperature of the driest quarter 0.068 0.355 -0.010 mean temperature of the wettest quarter 0.272 0.165 -0.147 temperature seasonality 0.159 -0.275 -0.057 precipitation seasonality 0.282 -0.010 0.167

Standard Deviation 2.936 2.642 1.301 Proportion of Variance Explained 0.410 0.332 0.081 Cumulative Proportion of Variance 0.410 0.743 0.823

Ensemble forecasting for African vertebrates | 53

Appendix S2: General Circulation Models used in the study

17 General Circulation Models (GCM) were used in this study. Their native resolution is shown in the table below, but downscaled projections at 10’ (Tabor & Williams, 2010) were used here. Model name abbreviations used in our study are also shown. The GCM clusters in which each GCM was included are given in the shaded columns for the three emissions scenarios (A2, A1B and B1).

Source Model Abbrev. Resolution Clusters y x A2 A1B B1 Bjerknes Centre for Climate Research BCCR-BCM2.0 bc2 2.80 2.80 2 1 3 National Center for Atmospheric Research CCSM3 cs3 1.41 1.41 2 1 1 Canadian Centre for Climate Modelling and CGCM3.1(T47) t47 3.75 3.75 3 3 3 Analysis Centre National de Recherches CNRM-CM3 cm3 2.80 2.80 3 3 3 Météorologiques, Météo-France Commonwealth Scientific and Industrial CSIRO-MK3.0 m30 1.90 1.90 1 1 1 Research Organisation (CSIRO) Atmospheric Research Commonwealth Scientific and Industrial CSIRO-MK3.5 m35 1.90 1.90 3 2 2 Research Organisation (CSIRO) Atmospheric Research Max Planck Institute for Meteorology ECHAM5/MPI- eh5 1.90 1.90 3 2 2 OM Meteorological Institute of the University of ECHO-G ecg 3.90 3.90 2 3 3 Bonn, Meteorological Research Institute of the Korea Meteorological Administration (KMA), and Model and Data Group Geophysical Fluid Dynamics Laboratory GFDL-CM2.0 cm2 2.00 2.50 3 2 3 (GFDL), National Oceanic and Atmospheric, Administration (NOAA), U.S. Department of Commerce Geophysical Fluid Dynamics Laboratory GFDL-CM2.1 c21 2.00 2.50 3 2 3 (GFDL), National Oceanic and Atmospheric, Administration (NOAA), U.S. Department of Commerce Goddard Institute for Space Studies (GISS), GISS-ER mer 4.00 5.00 2 3 3 National Aeronautics and Space Administra- tion (NASA) Institute for Numerical Mathematics INM-CM3.0 im3 4.00 5.00 2 1 3 Institut Pierre Simon Laplace IPSL-CM4 cm4 2.50 3.75 3 2 2 Center for Climate System Research (Univer- MIROC3.2 32m 2.80 2.80 1 3 3 sity of Tokyo), National Institute for Environ- (medium reso- mental Studies, and Frontier Research Center lution) for Global Change Meteorological Research Institute MRI-CGCM2.3.2 c23 2.80 2.80 1 1 1 National Center for Atmospheric Research PCM pc1 2.80 2.80 1 1 1 Hadley Centre for Climate Prediction UKMO-HadCM3 hd3 2.50 3.75 3 2 2

54 | Chapter I

Appendix S3: R scripts used to build the “central cluster” consensus projections

Available online.

Appendix S4: Sources of uncertainty in the ensemble of forecasts from individual BEMs

A point-wise three-way analysis of variance (ANOVA) without replication was performed for each grid cell over the study area (N=1,851) to evaluate the relative contributions of bioclimatic envelope models (BEM), the General Circulation Model clusters (GCMcons), the emissions scenarios (SRES), and interactions among these factors, to the variability in turnover forecasts for each taxon. The values in the table correspond to the median value and (in brackets) the lower and upper quartiles of the proportion of the total sum of squares attributed to each factor and are shown for both mid- and late-century, for each taxon.

Amphibians Snakes Mammals Birds 2041-60 BEMcons 76.5 (61.1-87.4) 81.7 (70.9-88.9) 76.2 (61.8-86.4) 79.8 (61.7-87.9) GCMcons 7.3 (2.7-16.2) 6.8 (3.0-13.4) 7.6 (3.7-16.5) 8.2 (4.2-19.1) SRES 3.3 (1.3-6.7) 3.1 (1.6-5.6) 4.0 (2.1-7.6) 4.0 (2.2-7.7) BEMcons:GCMcons 3.2 (1.6-5.8) 2.2 (1.3-4.1) 2.8 (1.5-5.4) 2.2 (1.2-4.3) BEMcons:SRES 1.4 (0.8-2.6) 1.1 (0.6-1.8) 1.2 (0.7-2.2) 1.0 (0.5-1.7) GCMcons:SRES 1.1 (0.4-2.2) 0.9 (0.4-1.7) 1.0 (0.5-2.2) 0.9 (0.4-1.9) BEMcons:GCMcons:SRES 2.2 (1.2-4.0) 1.6 (0.9-2.8) 1.8 (1-3.4.0) 1.2 (0.6-2.1) 2081-00 BEMcons 63.4 (38.7-83.1) 69.0 (48.2-82.5) 60.6 (37.3-76.6) 62.6 (34.1-76.7) GCMcons 7.4 (2.6-16.2) 7.5 (4.0-14.1) 9.6 (5.1-16.4) 10.2 (6.2-18.2) SRES 12.0 (3.8-30.9) 13.5 (6.1-27.6) 18.2 (9.5-33.7) 17.0 (9.1-35.2) BEMcons:GCMcons 2.4 (1.4-3.9) 1.7 (1.1-2.6) 1.9 (1.2-3.2) 1.8 (1.1-2.7) BEMcons:SRES 2.5 (1.5-4.3) 2.0 (1.2-3.1) 2.3 (1.4-3.8) 2.2 (1.4-3.6) GCMcons:SRES 1.4 (0.5-2.8) 0.9 (0.4-2.0) 1.1 (0.5-2.3) 1.0 (0.5-2.4) BEMcons:GCMcons:SRES 2.4 (1.4-4.2) 1.7 (1.1-2.8) 1.8 (1.1-3.0) 1.4 (0.9-2.3)

Ensemble forecasting for African vertebrates | 55

Appendix S5: Non-analogue climate maps

For both mid- and late-century, non-analogue climates correspond to the sum of the areas where mean temperature of the warmest month, mean temperature of the coldest month and annual precipitation values beyond the observed ranges in the baseline period are projected. Areas in red were projected to experience future climate conditions lacking baseline analogues. The percentage of the total area projected to experience non-analogue climates is indicated in red and the same percentage for each variable (tcm=temperature of the coldest month, tw=temperature of the warmest month, ps=annual precipitation) in black underneath each map. Data are shown for the three General Circulation Model clusters under each emissions scenario (A2, A1B and B1).

56 | Chapter I

Appendix S6: Single-BEMs selected for the ‘central clusters’ and Anosim results

For each taxon and climate projection, the seven bioclimatic envelope models were clustered based on similarities in magnitude and spatial pattern between single-model probabilistic projections and the multi-model median probabilistic projection for late-century. The k-means clustering resulted in four clusters, shown below for the three General Circulation Model clusters under each emission scenario (A2, A1B and B1). Cluster 1 is the maximum consensus ‘central cluster’. The analysis of similarity (Anosim) test was used to test the statistical significance of the clustering.

BEM clusters Anosim SRES GCM ANN GAM GBM GLM MARS FDA RF Statistics P-value A2 Cluster 1 4 2 1 1 3 3 3 0.912 0.004 Cluster 2 4 2 1 1 3 3 3 0.765 0.013 Cluster 3 2 4 1 1 3 3 1 0.971 0.004 A1B Cluster 1 4 2 1 2 1 3 1 0.853 0.004 Cluster 2 4 2 1 1 3 3 3 0.853 0.004

Amphibians Cluster 3 2 4 1 1 3 3 3 0.735 0.027 B1 Cluster 1 4 2 1 1 3 3 3 0.824 0.008 Cluster 2 4 2 1 2 1 3 1 0.853 0.004 Cluster 3 4 2 1 1 3 3 3 0.912 0.004

A2 Cluster 1 3 2 1 1 4 4 4 0.853 0.004 Cluster 2 2 4 1 1 3 3 3 0.882 0.004 Cluster 3 2 4 1 1 3 3 1 0.941 0.004

Snakes A1B Cluster 1 3 2 1 1 4 4 4 0.853 0.004 Cluster 2 2 4 1 1 3 3 3 0.853 0.004 Cluster 3 2 4 1 1 3 3 3 0.882 0.004 B1 Cluster 1 3 2 1 1 4 4 4 0.824 0.002 Cluster 2 3 2 1 1 4 4 4 0.853 0.004 Cluster 3 3 2 1 1 4 4 4 0.882 0.004

A2 Cluster 1 4 1 2 1 2 3 2 0.853 0.003 Cluster 2 4 1 2 1 3 2 2 0.941 0.003 Cluster 3 4 2 2 1 3 2 2 0.912 0.003 A1B Cluster 1 4 1 2 1 2 3 2 0.794 0.006

Mammals Cluster 2 4 1 2 1 2 3 2 0.618 0.045 Cluster 3 4 1 1 2 3 3 1 1.000 0.003 B1 Cluster 1 4 1 2 1 2 3 2 0.933 0.034 Cluster 2 4 1 2 1 2 3 2 0.765 0.016 Cluster 3 4 1 2 1 2 3 2 1.000 0.003

A2 Cluster 1 2 1 1 1 4 3 1 0.933 0.032 Cluster 2 2 1 1 1 4 3 1 0.911 0.032

Birds Cluster 3 2 1 1 1 4 3 1 1.000 0.003 A1B Cluster 1 2 1 1 1 4 3 1 0.978 0.032 Cluster 2 2 1 1 1 4 3 1 0.933 0.032 Cluster 3 2 1 1 1 4 3 1 0.882 0.011 B1 Cluster 1 4 1 2 1 2 3 2 0.911 0.032 Cluster 2 2 1 4 1 4 3 4 0.889 0.032 Cluster 3 2 1 1 1 4 3 1 0.933 0.032

Ensemble forecasting for African vertebrates | 57

Appendix S7: Level of consensus among all BEMs and ‘central cluster’ BEMs

To evaluate the degree of consensus achieved in bioclimatic envelope model (BEM) consensus projections, Principal Components Analysis (PCA) was performed for each species on the probabilistic projections. The proportion of variance explained by the first principal component axis reflects the degree of consensus among projections. PCAs were performed on baseline and late- century projections, both on the seven single-BEMs and on the ‘central cluster’ BEMs. The graphs show the frequency distributions across all amphibian (n=284), snake (n=310), mammal (n=623) and bird (n=1,506) species, with the degree of consensus increasing on a scale from 0 to 1. For future projections calculations, the ‘maximum consensus’ General Circulation Model cluster (cluster 2, the ‘central cluster’) under the emissions scenario A1B was used.

Appendix S8: Single-BEMs selected as ‘central model’ across species

With the ‘central model’ methodology to combine the ensembles of bioclimatic envelope models (BEM), we selected the model summarising the highest amount of variation among projections for each species. A Principal Components Analysis (PCA) was performed for each species on the late- century projected probabilities and the ‘central model’ corresponded to the one with the highest PCA loading in the first (consensus) axis. The graphs show the percentage of species for which each of the seven single BEMs was selected as the ‘central model’. Results are shown for the projections built with 100% of the data.

58 | Chapter I

Appendix S9: Statistical differences in the distributions of TSS and omission and commission error among the five BEM consensus projections

Pair-wise differences in True Skill Statistics (TSS), omission and commission error frequencies for the five bioclimatic envelope model (BEM) consensus projections for amphibian (n=284), snake (n=310), mammal (n=623) and bird (n=1,506) species were tested using the Wilcoxon signed rank test. EMean is the ensemble mean, EWMean the ensemble weighted mean, EMed the ensemble median, CMod the central model, and CClus the central cluster. For each taxon and accuracy measure, Bonferroni-corrected P-values (number of tests=10) are shown above the diagonal and the test statistics below.

Method Omission error Commission error EMean EWMean EMed CMod CClus EMean EWMean EMed CMod CClus EMean 1.70E-02 3.36E-11 1.37E-01 8.15E-09 7.89E-12 5.70E-35 2.63E-19 5.47E-47 EWMean 4049 6.62E-15 1.13E-03 7.08E-05 9519 8.73E-15 6.73E-09 1.57E-45 EMed 13681 2802 2.57E-01 1.33E-18 2689 29204 1.07E+00 4.57E-41 CMod 8373 7593 13164 2.36E-13 32400 28041 21704 2.02E-21

Amphibians CClus 14402 12482 20381 16303 40149 39831 38580 33304

EMean 1.19E+00 3.83E-12 1.81E-01 1.03E-01 3.27E-04 2.30E-29 3.04E-15 2.14E-51 EWMean 9137 3.45E-13 2.96E-02 4.01E-02 16091 1.51E-13 6.91E-09 3.99E-51 EMed 20619 6024 2.63E-02 9.26E-03 5680 35393 4.65E+00 3.97E-45 Snakes CMod 13201 11260 20768 3.46E+00 35288 32048 24302 4.80E-33 CClus 12536 12079 19925 15602 47891 47827 46097 41596

EMean 2.05E-21 2.36E-08 1.84E-41 8.97E-12 6.93E+00 4.45E-15 4.88E-06 2.57E-97 EWMean 9805 5.53E-38 9.47E-57 3.85E+00 77856 1.47E-09 1.18E-02 1.71E-97 EMed 37665 10056 6.88E-31 5.21E-28 49203 110835 8.24E-01 3.60E-99

Mammals CMod 18247 12514 24231 7.71E-62 113297 102932 95942 2.87E-66 CClus 64478 41076 77061 125512 189618 188631 189976 167655

EMean 1.62E-52 9.46E-16 4.79E-82 3.13E-33 4.00E-02 4.20E-27 1.10E-04 9.08E-241 EWMean 79127 1.08E-84 1.32E-132 9.92E-01 435115 3.27E-09 2.67E-01 9.77E-238 Birds EMed 239289 80336 7.06E-65 3.40E-70 294998 591882 5.57E+00 1.19E-239 CMod 136689 73731 156865 1.53E-134 593707 559791 503122 1.39E-189 CClus 404898 274275 530221 738702 1108284 1098301 1104396 1037136

Method TSS EMean EWMean EMed CMod CClus EMean 4.31E-06 7.55E-01 2.78E-02 1.55E-44 EWMean 12609 2.54E-02 9.44E+00 3.89E-43 EMed 17772 15934 1.43E+00 1.56E-43 CMod 24215 19997 22112 1.39E-34

Amphibians CClus 39855 39539 39629 37250

EMean 7.54E-02 4.88E-01 1.11E+00 7.77E-43 EWMean 19213 3.59E+00 9.37E+00 3.64E-35 EMed 20990 22359 5.44E+00 1.43E-39 Snakes CMod 25957 23608 24426 1.56E-30 CClus 46034 43936 45166 41583

EMean 2.11E-47 6.11E-01 2.96E-54 2.63E-77 EWMean 27902 9.20E-39 1.71E-85 1.44E-55 EMed 98747 35386 1.43E-58 2.29E-82

Mammals CMod 25971 7915 23224 1.63E-100 CClus 180356 166971 183086 193325

EMean 1.15E-118 1.15E-02 1.54E-140 6.55E-218 EWMean 159544 2.88E-98 2.10E-208 3.64E-173 Birds EMed 583286 198652 8.45E-141 1.41E-223 CMod 134970 44482 135868 2.66E-241 CClus 1090226 1032510 1102239 1125877

Ensemble forecasting for African vertebrates | 59

Appendix S10: Frequency distribution of species turnover for alternative GCM clusters, emissions scenarios and BEM consensus projections

Frequency distribution of species turnover rates over the study area (N=1851) for amphibians, snakes, mammals and birds, for mid- and late-century. Data are shown for the five bioclimatic envelope model consensus projections (EMean=ensemble mean, EWMean=ensemble weighted mean, EMed=ensemble median, CMod=central model, and CClus=central cluster), and for the three Gen- eral Circulation Model clusters under emissions scenarios A2, A1B and B1.

60 | Chapter I

Ensemble forecasting for African vertebrates | 61

Appendix S11: Late-century species turnover for alternative BEM consensus methodologies

Projected late-century turnover rate (%) for amphibian, snake, mammal and bird species for alternative BEM consensus projections. The five consensus projections are: ensemble mean (EMean), ensemble weighted mean (EWMean), ensemble median (EMed), central model (CMod), and central cluster (CClus). Projections refer to the ‘maximum consensus’ General Circulation Model cluster (cluster 2) under the A1B emission scenario.

62 | Chapter I

Appendix S12: Frequency distribution of climate anomalies over the study area

Boxplots of mean temperature of the warmest and coldest month and annual precipitation anomalies over the study area (N=1,851). Data are shown for the mid- (white bars) and late-century (grey bars), for the three General Circulation Model clusters under emissions scenarios A2, A1B and B1.

Ensemble forecasting for African vertebrates | 63

Appendix S13: Climate anomaly maps for the three variables

Difference between future (mid- and late-century) projections and baseline data for the three variables (annual precipitation and temperature of the coldest and warmest months), for the different General Circulation Model clusters (1 to 3) under alternative emissions scenarios (A2, A1B and B1). Note that for annual precipitation, the scale (in mm) covers both negative values (blue tones, referring to decreased precipitation in the future) and positive values (red tones, referring to increased precipitation in the future), whereas for the temperature-based variables the scale (in degrees Celsius) has positive values only (blue through to red tones indicating increasingly warmer future temperatures).

References

Clarke KR, Warwick RM (1994) Similarity-based testing Tabor K, Williams JW (2010) Globally downscaled for community pattern: the two-way layout with no climate projections for assessing the conservation replication. Marine Biology, 118, 167-176. impacts of climate change. Ecological Applications, 20, 554-565. Duan Q, Phillips TJ (2010) Bayesian estimation of local signal and noise in multimodel simulations of climate Venables WN, Ripley BD (2003) Modern Applied change. J. Geophys. Res., 115, D18123. Statistics with S, New York, Springer-Verlag. New M, Lister D, Hulme M, Makin I (2002) A high- resolution data set of surface climate over global land areas. Climate Research, 21, 1-25.

64 | Chapter I

Addendum

Figure 7 in Garcia et al. (2012) shows the per- alternative figure might be perceived as provid- centage of species projected to retain climatic ing more useful information for examining the suitability under each emissions scenario. For regional impacts of climate change on climatic each location i, the proportion R of species suitability for sub-Saharan African vertebrates. retaining climatic suitability is given by , where ri is the number of species that retain climatic suitability in location i under climate change and N is the total number of species in sub-Saharan Africa in the baseline period. Differences across taxa and across emissions scenarios are visible. But because the proportions of retention throughout the region are relative to the same total number of species, it is the areas of high baseline richness that are emphasised. As this figure may lead to misinterpreta- tions, we show below an alternative figure where in situ retention is calculated in relation to local (rather than total) species richness. In this case, the proportion R of species retaining climatic suitability in location i is given by , where ri is the number of species that retain climatic suitability in location i under climate change and ni is the num- Percentage of species predicted to retain cli- ber of species in location i in the baseline sce- matic suitability under each emissions scenario. nario. Here the variability among scenarios is The proportion of the local numbers of species still clear, as is the greater potential impact of of amphibians, snakes, mammals and birds that climate change on in situ climatic suitability for are projected to retain climatic suitability in amphibians and snakes in comparison to each location are shown for the median ensemble of all bioclimatic envelope models and for mammals and birds. However, by normalising the ‘maximum consensus’ general circulation the retention of climatic suitability to local model cluster under the A2, A1B and B1 emis- species richness, the alternative maps empha- sions scenarios. sise the areas where higher proportions of local richness are projected to retain climatic suitability, irrespective of total richness. As such, the

Chapter II

Matching species traits to projected threats and opportunities from climate change

RAQUEL A. GARCIA, MIGUEL B. ARAÚJO, NEIL D. BURGESS, WENDY B. FODEN, ALEXANDER GUTSCHE, CARSTEN RAHBEK, AND MAR CABEZA, Journal of Biogeography doi:10.1111/jbi.12257 (2014)

Matching species traits to projected threats and opportunities from climate change

RAQUEL A. GARCIA1,2,3, MIGUEL B. ARAÚJO1,2,3,4 , NEIL D. BURGESS2,5,6, WENDY B. FODEN7,8, ALEXANDER GUTSCHE9, CARSTEN RAHBEK2,4, and MAR CABEZA10

1 Center for Macroecology, Evolution and Climate, Department of Biology, University of Copenhagen, Denmark 2 Department of Biogeography and Global Change, National Museum of Natural Sciences, CSIC, Madrid, Spain 3 InBio/CIBIO, University of Évora, Évora, Portugal 4 Imperial College London, Silwood Park, Ascot, Berkshire, United Kingdom 5 WWF US Conservation Science Program, Washington, DC, US 6 United Nations Environment Programme World Conservation Monitoring Centre, Cambridge, UK 7 Global Species Programme, International Union for Conservation of Nature (IUCN), Cambridge, UK 8 Animal, Plant and Environmental Sciences, University of the Witwatersrand, Johannesburg, South Africa 9 Museum für Naturkunde, Leibniz Institute for Research on Evolution and Biodiversity, Berlin, Germany 10 Metapopulation Research Group, Department of Biosciences, University of Helsinki, Finland

Journal of Biogeography doi:10.1111/jbi.12257 (2014)

Abstract

Aim Climate change can lead to decreased climatic all species, and individually for groups of species suitability within species’ distributions, increased with different combinations of threats and oppor- fragmentation of climatically suitable space, tunities. and/or emergence of newly suitable areas outside Results In the Congo Basin and arid Southern present distributions. Each of these extrinsic Africa, projected losses for wide-ranging amphibi- threats and opportunities potentially interacts ans were compounded by sensitivity to climatic with specific intrinsic traits of species, yet this variation, and expected gains were precluded by specificity is seldom considered in risk assess- poor dispersal ability. The spatial overlap be- ments. We present an analytical framework for tween exposure and vulnerability was more pro- examining projections of climate change-induced nounced for species projected to have their cli- threats and opportunities with reference to traits mate space contracting in situ or shifting to distant that are likely to mediate species’ responses, and geographical areas. Our results exclude the poten- illustrate the applicability of the framework. tial exposure of narrow-ranging species to shrink- Location Sub-Saharan Africa. ing climates in the African tropical mountains. Methods We applied the framework to 195 sub- Main conclusions We illustrate the application of Saharan African amphibians with both available a framework combining spatial projections of bioclimatic envelope model projections for the climate change exposure with traits that are likely mid-21st century and trait data. Excluded were to mediate species’ responses. Although the pro- 500 narrow-ranging species mainly from montane posed framework carries several assumptions areas. For each of projected losses, increased that require further scrutiny, its application adds a fragmentation and gains of climate space, we degree of realism to familiar assessments that selected potential response-mediating traits and consider all species to be equally affected by cli- examined the spatial overlap with vulnerability mate change-induced threats and opportunities. due to these traits. We examined the overlap for

68 | Chapter II

Introduction of local populations (e.g. Sinervo et al., 2010). Second, even where suitable climate space Fingerprints of recent climate change impacts persists, it may become more fragmented. Ar- on species’ distributions are already apparent eas of suitable climate may lose contiguity, with (Parmesan & Yohe, 2003), but predicting future fragments becoming more isolated and smaller climate change impacts is still a major scientific in area. Third, gains of climatic suitability out- challenge (Pereira et al., 2010). Predictions of side current distributions of species bring po- species’ exposure to climate change have to tential opportunities for colonization. These date relied mostly on bioclimatic envelope components of exposure are likely to have models, and are increasingly being comple- distinct spatial distributions and conservation mented with available trait data to estimate implications, but are seldom teased apart (but species’ vulnerability (e.g. Heikkinen et al., see Midgley et al., 2003; Heikkinen et al., 2009; 2009). Assessing the risk of species’ extinction Araújo et al., 2011). or decline requires an understanding of threats, Species also vary in their degree of intrinsic or extrinsic adverse events, and vulnerability, sensitivity and capacity to adapt to exposure or the intrinsic susceptibility of species to (Williams et al., 2008; Chevin et al., 2010), and threats (Araújo & Williams, 2000). It is thus traits can capture this variation. Traits are important to understand the interaction be- meant as ‘any morphological, physiological or tween threats and vulnerability, as specific phenological feature measurable at the individ- traits are likely to mediate species’ responses to ual level, from the cell to the whole-organism different threats (Isaac & Cowlishaw, 2004; level, without reference to the environment or Fritz et al., 2009; Murray et al., 2011; González- any other level of organization’ (sensu Violle et Suárez et al., 2013). For example, under habi- al., 2007, p. 884). Examples are limb or wing tat-modifying processes such as agriculture and length, and clutch size. In many cases, ecologi- logging, small-sized, habitat specialist mam- cal characteristics of species and their geo- mals are most affected, whereas under proc- graphical or environmental ranges are used as esses that directly affect survival, such as hunt- proxies for traits, in the expectation that they ing, the most susceptible are mammals with summarize combinations of traits. Examples large body size and small litter size (González- are species’ geographical range size and cli- Suárez et al., 2013; see also Isaac & Cowlishaw matic breadth. 2004). In climate change risk assessments, Previous research has identified general however, this specificity in the interaction be- traits that predispose species to extinction (e.g. tween threat and vulnerability has not been Purvis et al., 2000), and specific traits that me- addressed sufficiently. diate the effect of particular threats on species In studies using bioclimatic envelope mod- (Isaac & Cowlishaw, 2004; Murray et al., 2011; els, the level of species’ exposure to climate González-Suárez et al., 2013). Under changing change is commonly inferred from temporal climates, poor dispersal ability and habitat or changes in the overall size of species’ climati- climatic specialization, for example, have been cally suitable space (e.g. Thuiller et al., 2005b; suggested to increase vulnerability (Peters & Araújo et al., 2006; Huntley et al., 2006; Feeley Darling, 1985), and have been found to corre- et al., 2012; Triviño et al., 2013). Such summary late positively with empirical data on range measures conceal different opportunities as contractions (Beaumont & Hughes, 2002; Botts well as threats, each imposing specific con- et al., 2013). Generally, traits have accounted straints on species (Thomas et al., 2011). First, for a significant but small amount of the varia- loss of climatic suitability within existing dis- tion in climate change-induced range shifts tributions is expected to affect the persistence

Species traits and climatic threats and opportunities | 69

(Buckley & Kingsolver, 2012). Yet, changes in exacerbate projected losses and increased the size, level of fragmentation and position of fragmentation of climatically suitable areas, or species’ climate space each represent a distinct restrain projected gains of newly suitable areas. threat or opportunity under changing climates, We test the practicality of the framework and thus are likely to interact with particular on wide-ranging sub-Saharan African amphibi- sets of ‘response-mediating’ traits (sensu Luck ans, using available bioclimatic envelope model et al., 2012). and trait data. Our application of the frame- Besides evolution on longer time-scales, work reflects situations where only proxies for three main determinants of climate change traits are available, and where bioclimatic enve- vulnerability can be distinguished (Williams et lope model projections are available only for al., 2008; Chevin et al., 2010). First, traits de- wide-ranging species, in our case leading to the scribing plasticity of individual phenology, exclusion of most narrow-ranging endemics behaviour or physiology affect the potential of typical of biodiverse montane regions. World- individuals to persist in situ under changing wide, amphibian populations are declining due climates. For example, species able to physio- to a multitude of threats that include habitat logically tolerate a wide range of climatic varia- destruction, climate change and the pathogeni- tion (Huey et al., 2012), or adapt their behav- cal fungal disease chytridiomycosis (Blaustein iour to lessen exposure to unsuitable climates & Kiesecker, 2002; Hof et al., 2011; Li et al., (Chown, 2012), have, all else being equal, 2013). Climate change, often in tandem with higher chances of persistence in situ under land-use change, is expected to affect large climate change. Second, traits influencing the areas of tropical Africa in the future (Hof et al., potential of individuals to disperse affect their 2011; Foden et al., 2013). While our aim here is capacity to colonize newly suitable environ- to present a spatially explicit framework for ments (Pöyry et al., 2009) or move between linking threats or opportunities and vulnerabil- fragments of suitable climate. Third, life-history ity in climate-change risk assessments, the traits influencing population growth, although practical application we show can also contrib- not affected by environmental change, may ute to a better understanding of climate change constrain the rate of dispersal or in situ adapta- risks facing wide-ranging sub-Saharan African tion. In the case of reproductive traits, for ex- amphibians. ample, frequent or early reproduction and high fecundity, should increase colonization opportunities (Angert et al., 2011). Material and Methods Here, we present an analytical framework for examining projections of climate change- Study species induced threats or opportunities for species Our study sample consists of 195 amphibian with reference to the vulnerability of species. species (see Appendix S1 in Supporting Infor- Each threat or opportunity – exposure to loss, mation) restricted in their distributions to sub- fragmentation and gain of climate space – is Saharan Africa, and with available projections matched to specific response-mediating traits of species bioclimatic envelope models (Garcia that potentially render species vulnerable, et al., 2012) and trait data (Foden et al., 2008, according to expectations from theory and 2013). From all 695 species in the original empirical evidence (Fig. 1). Areas of spatial species distribution (Hansen et al., 2007) and overlap between threats or opportunities and trait (Foden et al., 2008, 2013) databases, only associated vulnerability are then identified. 272 had model projections available (excluded These are areas where traits can potentially were 423 species with fewer than 15 gridded

70 | Chapter II

Exposure to

seline climate space (left) to -mediating traits associated with spe- across taxa. by dashed lines, and the component of exposure under consider- responses to climate change-induced threats and opportunities. tunities, examples are given of potential response areas are each illustrated with diagrams of the shift from ba imate space represented mography, where supported by theoretical and empirical studies | Examples of traits that potentially mediate species’ Figure 1 loss, increased fragmentation and gain of climatically suitable cl baseline the with lines, closed (right, space climate future ation highlighted in black). For each of these threats or oppor cies’ plasticity, dispersal ability and/or de

Species traits and climatic threats and opportunities | 71

occurrence records), and, of these, 195 also had First, for each species we considered local data for all traits (excluded were 77 species losses in pixels projected to be climatically with some trait data missing). Our sample is, suitable in the baseline period but unsuitable in therefore, restricted to the widest-ranging the future. Second, for fragmentation we used a species in the dataset, and may not be repre- distance-based measure of contagion of suit- sentative of the overall taxonomic and geo- able climate space. Contagion was measured as graphical amphibian diversity patterns in Af- the weighted average of the number of suitable rica (Appendix S2a–c). In particular, our data pixels among a set of ki neighbours of a central exclude most species endemic to biodiverse pixel yi, where the weight given to the grid cell montane areas such as the Cameroon/ Nigerian yj is wij = 1/dij, and dij is the great-circle distance highlands and the Eastern Afromontane biodi- between grid cells yi and yj (Araújo et al., 2002; versity hotspot, and most species in the highest equation 1). Owing to poor dispersal ability of threat categories of the IUCN Red List (Appen- most amphibians, we considered only the first- dix S2d). order neighbours (maximum = 8) adjacent to the central pixel. Changes in contagion were Extrinsic threats and opportunities from climate given by the difference between future and change baseline contagion, with negative values indicating reduced contagion, i.e. increased frag- We used published baseline (1961–90) and mentation. mid-century (2041–60) projections of climatically suitable areas for our 195 amphibians (1) species, at one degree latitudinal–longitudinal resolution (c. 111 km × 111 km at the equator), according to mean temperatures of warmest and coldest month and annual precipitation Third, local gains corresponded to pixels (for detailed methods see Garcia et al., 2012). projected to be climatically unsuitable in the Future projections were for a multi-model baseline period but suitable in the future. climate ensemble under the A1B greenhouse Fourth, for species projected to gain newly gas emissions scenario (Nakicenovic & Swart, suitable areas in mid-century, we also com- 2000). We used consensus projections obtained puted, for each pixel of newly suitable climate, by computing the median among seven biocli- the great-circle distance to the nearest pixel of matic envelope modelling techniques, in pres- baseline suitable climate. ence–absence format, and assuming unlimited dispersal in future projections. Here, to charac- Intrinsic vulnerability to climate change terize climate change-induced threats and op- We sourced the trait data from the IUCN’s trait- portunities, we computed four types of pro- based climate change vulnerability assessment jected changes in climatic suitability: local loss, for amphibians (Foden et al., 2008, 2013). fragmentation and gain of climatic suitability, These data are mainly ecological characteristics and distance to new areas gained. Each metric of species or their ranges rather than traits in a was quantified per pixel, as described below, strict sense (sensu Violle et al., 2007). From the yielding maps of changes for each species. data available, we selected characteristics that Composite maps for our species sample were are likely to summarize response-mediating also obtained by summing the number of spe- traits under climate change (Fig. 1), and that cies with projected local loss, increased frag- are not strongly correlated. We thus selected mentation, or gain of suitable climate in each tolerance to temperature and precipitation pixel. change, dependence on precipitation cues,

72 | Chapter II

dispersal ability and reproductive output (Ta- traits and exposure (Fig. 1), we matched plas- ble 1; see Foden et al., 2008, 2013, for details). ticity characteristics to local losses, dispersal For each of these characteristics, we used the characteristics to increased fragmentation, and climate change vulnerability classification of both dispersal and demography characteristics species by Foden et al. (2008, 2013): species to gains (see Table 1). For climatic tolerance, were assigned ‘high’, ‘lower’ or ‘unknown’ we considered for each species either the toler- scores of vulnerability based on the ranking of ance to temperature or to precipitation de- all sub-Saharan African species in the IUCN pending on the variable of highest importance database or on pre-defined criteria (see Table in the bioclimatic envelope models (assessed 1). with permutations within the BIOMOD computing platform in R; Thuiller et al., 2009). Spatial overlap between threats or opportunities and We thus identified where projected losses were vulnerability for species with high vulnerability according to plasticity traits, where projected fragmentation We spatially assessed where climate-induced was for species with high vulnerability accord- threats or opportunities, defined with the met- ing to dispersal traits, and where projected rics of species’ exposure to climate change, gains were for species with high vulnerability overlapped with high climate change vulner- according to dispersal or demography traits. ability of species according to our selected The assessment of spatial overlap was first traits. Following expected interactions between

Table 1 | Species traits used as estimates of intrinsic vulnerability to climate change exposure for sub-Saharan African amphibians. Three climate change-induced threats and opportunities for species were defined based on bioclimatic envelope model projections (Garcia et al., 2012): loss, increased fragmentation, and gain of suitable climate space. For each, different sets of species traits or characteristics of species and their ranges (Foden et al., 2008, 2013) were selected that are likely to mediate species’ responses (see Fig. 1).

Intrinsic vulnerability to climate change

Plasticity . Tolerance to temperature change: average absolute deviation for all cells in species’ refined range for the monthly means; high risk if ≤ 1.06 °C, i.e. 25% of all 704 sub-Saharan African species in the IUCN dataset with the narrowest tolerance ranges. . Tolerance to precipitation change: average absolute deviation for all cells in species’ refined

Loss range for the monthly means; high risk if ≤ 46.89 mm, i.e. 25% of all 704 sub-Saharan African species in the IUCN dataset with the narrowest tolerance ranges. . Dependence on environmental cues: high risk if dependent on rainfall or increased water availability for mass breeding (excludes species buffered by occurring in forests).

Dispersal . Dispersal ability: high risk if not known to have become established outside their natural ranges, not associated with flowing water, and have small ranges (≤ 4000 km2, i.e. 25% of all 704 sub- Saharan African species in the IUCN dataset with the smallest ranges). Gain Demography Extrinsic threats and opportunities from climate change . Reproductive output: high risk if ≤ 50 offspring annually (where known) or viviparous.

Species traits and climatic threats and opportunities | 73

conducted for all species in our sample, and distant areas; and ‘expanding’ species were then individually for groups of species pro- projected to retain most of their baseline suit- jected to experience different combinations of ability and have large and distant gains outside threats and opportunities. To identify these their distributions. groups, we ranked species according to each species’ overall projected losses, gains and distance to gains of suitable climate. For each Results species, the overall local losses of climatic suit- Our results concern 195 sub-Saharan African ability L were quantified using the proportion amphibians with wider ranges, significantly of baseline area of suitable climate (ai) pro- larger temperature tolerances (P-value < 0.05, jected to be lost in the future (li; equation 2). Student’s t-test), and a larger proportion of The overall opportunity for gains G was meas- species with higher reproductive output than ured as the proportion of baseline area of suit- the remaining 500 species in the dataset (Ap- able climate (ai) projected to be gained in the pendix S2c,e). For this subset, the spatial pat- future (gi; equation 3). The surface area across terns of climate change exposure (greyscale the A pixels of the study area was measured maps in Fig. 2) showed widespread local losses taking into account the curvature of the Earth. of climatically suitable space, but a concentra- The overall distance to new areas D for each tion of increased fragmentation in montane species was the mean of the minimum great- areas and local gains in the Congo Basin. Spe- circle distances dib between each pixel gained i cies were also not randomly distributed across and the baseline suitable areas b computed sub-Saharan Africa with regard to climate across the N pixels gained (equation 4). change vulnerability due to selected traits or

ecological characteristics of species ranges (2) (Appendix S3a), leading to generally well- defined spatial patterns of overlap between

exposure and vulnerability (red and blue scale (3) maps in Fig. 2). Local losses of climatic suitability were

projected throughout most of the study area, ∀ (4) with the Congo Basin and the species-poor arid areas of the Sahel and Namibia/Botswana showing the highest proportions of species For each of the metrics considered, we re- losing local suitability in the future (Fig. 2a). tained the 25% of species with the smallest Geographical areas with greatest proportions of change values, and the 25% of species with the losses overlapping with vulnerability due to largest change values. We grouped these spe- traits varied across the three selected traits: the cies depending on the combination of change Congo Basin and coastal West Africa for tem- level thus defined for the three metrics, with perature tolerance (red shaded areas in Fig. the intention of highlighting the extremes of the 2b), South Africa and especially Namibia for distribution of changes. ‘Contracting in situ’ precipitation tolerance (red shaded areas in species were projected to suffer the largest Fig. 2c), and few scattered areas in the Sahel, losses of baseline climatic suitability, while the Albertine Rift and Namibia for dependence having little opportunity to move to new suit- on precipitation cues (red shaded areas in Fig. able areas; ‘obligate shifting’ species also faced 2d). Most projected increases in fragmentation large in situ losses but gained suitability in new, of climate space (Fig. 2e) were for species with

74 | Chapter II

Figure 2 | Overlap of climate change exposure and intrinsic vulnerability for a subset of 195 wide- ranging sub-Saharan African amphibians. The greyscale maps show pixel-based proportions of species exposed to losses, increased fragmentation, or gains of climate space: i.e. the proportions of species with baseline climatic suitability in a pixel that lose suitability in that pixel in the future (a), the proportions of species with suitability through time in a pixel that suffer increased fragmentation of climate space around that pixel (e), and the proportions of species with future suitability in a pixel that had no suitability in that pixel in the baseline (g), respectively. The red and blue scale maps compare, for each pixel, the proportions of losses (b–d), increased fragmentation (f) or gains (h–i) that correspond to species with higher vulnerability versus species with lower vulnerability due to selected traits (see Table 1). Different shades on the maps thus indicate dominance of losses, fragmentation or gains for species highly vulnerable (red), species less vulnerable (blue), both species with high and lower vulnerability (black) or none (white). Maps were drawn using quantile classification. low vulnerability due to dispersal traits (blue Student’s t-test; see Appendix S3b). For each shaded areas in Fig. 2f). Projected gains (Fig. group, we compared their modelled future 2g) may have been overestimated due to poor distributions of climatic suitability to future dispersal ability in the Congo Basin, and espe- projections modified in the following way: cially in areas extending from West Africa to losses for species with lower vulnerability due the Ethiopian highlands as well as western to climatic tolerance were converted into pres- South Africa (grey and red shaded areas in Fig. ences, and gains for species with high vulner- 2h). Gains for species with lower reproductive ability due to dispersal ability were trans- output were fewer and more scattered formed in absences (Fig. 3). throughout the same areas (red shaded areas in ‘Contracting in situ’ wide-ranging species Fig. 2i). occurred mainly in montane areas, which also Among the three groups of species with dif- hold the majority of the narrow-ranging species ferent combinations of threats and opportuni- excluded from our analysis. Although exposed ties, ‘contracting in situ’ and ‘obligate shifting’ to large overall losses, species in this group had species had significantly smaller geographical greater tolerance to climatic variation, and thus range sizes than ‘expanding’ species or those the modified future projections were more species not in the three groups (P-value < 0.05, conservative. For ‘obligate shifting’ species,

Species traits and climatic threats and opportunities | 75

Figure 3 | Potential effect of species’ climate change vulnerability on projections of climate change exposure for three groups of wide-ranging sub- Saharan African amphibians. Species were classified into ‘contracting in situ’ (n = 10), ‘obligate shifting’ (n = 6) and ‘expanding’ (n = 21) based on projected threats and opportunities from climate change. The first node of the tree classifies species based on overall losses of suitable climate, and the second node is based on overall gains of climatic suitability and distances to newly suitable areas. For each metric, only the extreme cases are considered, i.e. species below the 25th and above the 75th percentiles of the distribution of values for all species. The three groups thus obtained are illustrated with diagrams of the shift from baseline climate space (left, white circles) to future climate space (right, dark circles, with the baseline climate space represented by white circles with dashed lines). For each group, the maps show the projected suitability in the baseline (first row) and future (second row) time periods, and the future suitability modified by omitting losses for species with lower vulnerability due to climatic tolerance and the gains for species with poor dispersal ability (third row).

occurring in West and East African coastal ing the few losses of species with lower vulner- forests and along the eastern border of the ability due to the plasticity trait and the gains of Congo Basin, dispersal traits had the potential a quarter of species with poor dispersal ability, to modify projections of exposure. Whereas future projections for this group remained very losses in West Africa were concordant with the similar. Overall, correlation between projected species’ high vulnerability due to narrow toler- and modified future suitability was lowest for ance to climatic variation, gains in the Congo ‘obligate shifting’ species and ‘contracting in Basin were partly associated with poor dis- situ’ species, and highest for ‘expanding’ species persers and may thus have been overestimated. (Pearson’s product–moment correlation coeffi- Species in the ‘expanding’ group occurred along cient 0.60, 0.74 and 0.97, respectively, P-value < a broad band extending from West Africa to 0.05). west of the Ethiopian highlands. After discount-

76 | Chapter II

Discussion (704 sub-Saharan African species). Classes of high and lower vulnerability were defined Our analysis for wide-ranging sub-Saharan based on the quantile distribution of trait val- African amphibians shows how a simple ues across this wider pool of species, yielding a framework can be applied that combines famil- relative classification for all species that is iar projections of climate change exposure with unlikely to reflect the real vulnerability of indi- response-mediating traits, to help deliver more vidual species. realistic climate change risk assessments. The Second, the case of sub-Saharan African framework teases apart the threats and oppor- amphibians illustrates the framework’s applica- tunities resulting from exposure of species to tion when traits in the strict sense (sensu Violle climate change, and identifies key traits that et al., 2007) are largely unavailable, a situation potentially mediate species’ responses to each. that is common for many taxonomic groups Its application is contingent on the availability (e.g. González-Suárez et al., 2012). Some of the and quality of both exposure and trait data, as data used were derived from the characteriza- our analysis for sub-Saharan African amphibi- tion of known distributions of species (Foden et ans clearly illustrates. al., 2008, 2013) as proxies for traits. One exam- Availability of bioclimatic envelope model ple is tolerance to climatic variation, inferred projections of future climatic suitability is lim- with statistical approaches relating species ited by the number of existing records of spe- ranges to climate variables. Whereas previous cies occurrence (Feeley & Silman, 2011), and studies (e.g. Thuiller et al., 2005a; Feeley et al., has biased our sample towards wide-ranging 2012) used similar approaches, such proxies do species. Trait data availability was a further not strictly summarize traits but the interaction limitation, although it affected a smaller num- between traits and the environment. Climatic ber of species. Together, these limitations have tolerance inferred with these approaches may skewed our sample towards larger geographi- represent under-estimates when climatic cal range sizes, lower level of current threat, niches realized in the present are truncated narrower climatic tolerance breadth, and larger (Feeley & Silman, 2010). The finding that upper reproductive outputs (Appendix S2). Such bi- thermal limits tend to be highly conserved ases reduce the representativeness of the re- while lower limits are highly variable across sults and limit the scope for conservation guid- organisms (Araújo et al., 2013) further indi- ance. Indeed, most threatened amphibians cates that such proxies may be misleading. Only were excluded, particularly those from the the physiological limits of species could indi- Cameroon highlands and Eastern Afromontane cate their full capacity to adapt to climatic centres of diversity. Phylogenetic inference changes through plastic adaptation. One exem- methods exist that could circumvent the bias in plar study is that of Arribas et al. (2012), where the trait data (Nakagawa & Freckleton, 2008; an experimental approach was applied to esti- Buckley & Kingsolver, 2012), but, for the bulk mating the safety thermal limits and acclima- of the species excluded here, new approaches tion capacity of water beetles. Likewise, esti- that overcome limitations of correlative models mates of species’ dispersal abilities derived are needed to assess exposure of narrow- from empirical data on organism movement ranging species to climate change. (e.g. Gamble et al., 2007), phylogenetic dis- Application of our framework is also de- tances (Arribas et al., 2012), or morphological pendent on the quality of trait data. First, for or life-history traits (e.g. Baselga et al., 2012; some of the traits the classification of vulner- Whitmee & Orme, 2013) would more reliably ability (Foden et al., 2008, 2013) is contingent predict the ability of species to track suitable on the initial pool of species used for scoring

Species traits and climatic threats and opportunities | 77

climates than estimates based on known geo- excluded from this study probably mirrors graphical ranges of species. more closely that of our ‘contracting in situ’ Despite the shortcomings of the trait and group, which includes species with ranges exposure data used here, our results illustrate among the smallest of our sample and encom- how interpretation of spatial projections of passing mountain regions (Fig. 3). For narrow- species' exposure to climate change can be ranging species, contraction of the available altered with consideration of species’ climate marginal climatic conditions that are suitable change vulnerability. Projections under climate for them may play an important role (Williams change have been shown elsewhere to vary et al., 2007; Ohlemüller et al., 2008). At the because of assumptions regarding the thermal same time, in the topographically diverse re- tolerance of species (Feeley et al., 2012), and gions where these species occur, microclimates differences in dispersal capacity (Urban et al., not captured at the coarse scale used here may 2012). Our analysis for wide-ranging sub- facilitate adaptation to changing climates Saharan African amphibians highlights the (Pearson, 2006). Besides the potential threat Congo Basin and arid regions of Southern Af- from climate change, the high-elevation, range- rica, where projected losses were compounded restricted species with low fecundity that were by species’ sensitivity to climatic variation, and excluded from our study (Appendix S2) are also expected gains were precluded by poor disper- susceptible to declines associated with the sal ability (Fig. 2). Tropical ectotherms have chytridiomycosis disease (Bielby et al., 2008). been highlighted for their vulnerability to cli- By contrast, future projections for ‘expanding’ mate change because they are living close to species remained largely unchanged after con- their upper thermal limits (Deutsch et al., 2008; sideration of traits, although this may result Huey et al., 2009) and have narrower thermal from our simplistic approach whereby we re- breadths (Sunday et al., 2012). The lowland tained all gains by good dispersers irrespective tropics in particular have been suggested to of the distances involved. Nevertheless, the hold a high concentration of ectotherms shar- current levels of loss and fragmentation of ing vulnerability traits (Huey et al., 2012), and natural ecosystems are likely to pose an impor- to face biotic attrition in the future (Colwell et tant challenge to dispersal (Opdam & Wascher, al., 2008). The shallower temperature gradient 2004). in tropical lowlands (Colwell et al., 2008) in- We propose that application of our frame- creases distances required to track suitable work can provide information about the causes, climates, potentially imposing a challenge to spatial distribution and conservation implica- poor dispersers. tions of climate change risk (Thomas et al., Among the groups of wide-ranging am- 2011). Conservation needs will differ between phibians exposed to different combinations of species projected to partly retain suitability threats and opportunities, those most exposed where they occur and those for which suitabil- – ‘contracting in situ’ and ‘obligate shifting’ ity shifts to new areas. In the latter case, gains species – showed the strongest modifying effect in new areas may compensate losses and even of vulnerability on projections of exposure (Fig. lead to an increased, displaced, climate space. 3). Whereas poor dispersal ability rendered However, if newly suitable areas are distant, projected gains to be unlikely for ‘obligate shift- poor dispersal ability may place these ‘obligate ing’ species, losses for ‘contracting in situ’ spe- shifting’ species at risk. In risk assessments cies were discordant with the group’s lower based on changes in total area of climatic suit- vulnerability with respect to plasticity. Expo- ability (e.g. Thuiller et al., 2005b; Araújo et al., sure of the narrow-ranging montane species 2006; Huntley et al., 2006; Feeley et al., 2012;

78 | Chapter II

Triviño et al., 2013) such species may be classi- on species. Important response-mediating fied as ‘winners’ and thus be overlooked. In- traits are likely to vary across taxa, and further deed, the ‘obligate shifting’ amphibians in our studies of trait correlates of observed changes study were projected to increase their overall in ranges under changing climates can help climate space, but were flagged for the low in expand and adapt our list of examples (Fig. 1). situ persistence and the large discontinuity Where available, more precise and reliable between baseline and future climate space, estimates of response-mediating traits allow compounded by poor dispersal ability. for a closer coupling of bioclimatic envelope Previous frameworks to guide conservation models with traits, leading to projections that under climate change separated the threat of are more appropriate for conservation plan- loss from opportunities for gains of climate ning. For example, measures of dispersal capac- space (Thomas et al., 2011; Arribas et al., ity can be used to filter, pixel by pixel, projected 2012). The framework we present here consid- gains of climate space depending on their dis- ers an additional extrinsic factor that is seldom tance from present distributions (see Bateman explored – changes in the level of fragmenta- et al., 2013, for a review of options of dispersal tion of climate space (but see Serra-Diaz et al., scenarios in predictive modelling). Similarly, 2013). Besides influencing the probability and physiological climatic limits can provide the speed of range expansion (Hodgson et al., bounds for species persistence in modelling 2011), the level of aggregation of species re- exercises (Arribas et al., 2012; Summers et al., cords has been found to be a strong covariate of 2012). Projections of climate change exposure local extinction risk of bird species in Britain typically discount other important factors such (Araújo et al., 2002). At the coarse resolution of as biotic interactions, local population adapta- our study, our measure of fragmentation and tions and landscape structure, but assessing the the correspondent trait provide some indica- robustness of projections to the effect of re- tion on the risk of isolation from surrounding sponse-mediating traits is one crucial step areas of suitable climate, although with more towards increased realism. limitations in topographically diverse regions. The level of fragmentation characterized through the metric of contagion is particularly Acknowledgements important at fine scales, where the risk from We are grateful to Louis Hansen for help match- both increased isolation and decreased area of ing the taxonomy between the two datasets, fragments of suitable climate becomes more Zhiheng Wang for assistance with Fig. 2, and evident. Measures of contagion that consider Lucas Joppa for drawing our attention to the both effects could thus be used at finer scales, importance of increased fragmentation of cli- borrowing from metapopulation and landscape mate space. We thank John Measey, Anni Arpo- theory (Hanski, 2005). Yet the parallel between nen, Johanna Eklund, Laura Meller, Antti Tako- within-generation habitat fragmentation at the lander and three anonymous referees for com- landscape level, on one hand, and increased ments on the manuscript. R.A.G. thanks the fragmentation of climate space at larger spatial Metapopulation Research Group at the Univer- and temporal scales like the ones used here, on sity of Helsinki for providing the supportive the other hand, is not clear-cut and needs more environment in which much of this work was attention. developed, and the reserve-selection journal Our framework relies on the identification club members for helpful discussion. R.A.G. is and quantification of traits that potentially funded through a FCT PhD studentship mediate the effect of climate change exposure (SFRH/BD/65615/2009), M.C. through the

Species traits and climatic threats and opportunities | 79

RESPONSES project, and M.B.A. through the Beaumont, L.J. & Hughes, L. (2002) Potential changes in the distributions of latitudinally restricted Australian FCT PTDC/AAC-AMB/98163/2008 project. butterfly species in response to climate change. R.A.G., M.B.A., N.D.B. and C.R. thank the Danish Global Change Biology, 8, 954–971. National Research Foundation for support to Betzholtz, P.-E., Pettersson, L.B., Ryrholm, N. & Franzén, the Center for Macroecology, Evolution and M. (2013) With that diet, you will go far: trait-based analysis reveals a link between rapid range Climate. M.B.A. also acknowledges support expansion and a nitrogen-favoured diet. Proceedings from the Integrated Program of IC&DT Call Nº of the Royal Society B: Biological Sciences, 280, 20122305. 1/SAESCTN/ALENT-07-0224-FEDER-001755. Bielby, J., Cooper, N., Cunningham, A.A., Garner, T.W.J. & Purvis, A. (2008) Predicting susceptibility to future declines in the world’s frogs. Conservation Letters, 1, References 82–90. Blaustein, A.R. & Kiesecker, J.M. (2002) Complexity in Angert, A.L., Crozier, L.G., Rissler, L.J., Gilman, S.E., conservation: lessons from the global decline of Tewksbury, J.J. & Chunco, A.J. (2011) Do species’ amphibian populations. Ecology Letters, 5, 597–608. traits predict recent shifts at expanding range edges? Ecology Letters, 14, 677–689. Botts, E.A., Erasmus, B.F.N. & Alexander, G.J. (2013) Small range size and narrow niche breadth predict Araújo, M.B. & Williams, P.H. (2000) Selecting areas for range contractions in South African frogs. Global species persistence using occurrence data. Biological Ecology and Biogeography, 22, 567–576. Conservation, 96, 331–345. Buckley, L.B., Hurlbert, A.H., & Jetz, W. (2012) Broad- Araújo, M.B., Williams, P.H. & Fuller, R.J. (2002) scale ecological implications of ectothermy and Dynamics of extinction and the selection of nature endothermy in changing environments. Global reserves. Proceedings of the Royal Society B: Ecology and Biogeography, 21, 873–885. Biological Sciences, 269, 1971–1980. Buckley, L.B. & Kingsolver, J.G. (2012) Functional and Araújo, M.B., Thuiller, W. & Pearson, R.G. (2006) Climate phylogenetic approaches to forecasting species’ warming and the decline of amphibians and reptiles responses to climate change. Annual Review of in Europe. Journal of Biogeography, 33, 1712–1728. Ecology, Evolution, and Systematics, 43, 205–226. Araújo, M.B., Guilhaumon, F., Neto, D.R., Pozo, I. & Chevin, L.-M., Lande, R. & Mace, G.M. (2010) Adaptation, Calmaestra, R.G. (2011) Impactos, vulnerabilidad y plasticity, and extinction in a changing environment: adaptación al cambio climático de la biodiversidad towards a predictive theory. PLoS Biology, 8, española. 2. Fauna de vertebrados. Dirección General e1000357. de Medio Natural y Política Forestal, Ministerio de Medio Ambiente y Medio Rural y Marino & Museo Chown, S.L. (2012) Trait-based approaches to Nacional de Ciencias Naturales (CSIC), Madrid. conservation physiology: forecasting environmental change risks from the bottom up. Philosophical Araújo, M.B., Ferri-Yáñez, F., Bozinovic, F., Marquet, P.A. Transactions of the Royal Society B: Biological & Valladares, F. (2013) Heat freezes niche evolution. Sciences, 367, 1615–1627. Ecology Letters, 16, 1206–1219. Colwell, R.K., Brehm, G., Cardelús, C.L., Gilman, A.C. & Arribas, P., Abellán, P., Velasco, J., Bilton, D.T., Millán, A. & Longino, J.T. (2008) Global warming, elevational Sánchez-Fernández, D. (2012) Evaluating drivers of range shifts, and lowland biotic attrition in the wet vulnerability to climate change: a guide for insect tropics. Science, 322, 258–261. conservation strategies. Global Change Biology, 18, 2135–2146. Cowling, R.M. & Pressey, R.L. (2001) Rapid plant diversification: planning for an evolutionary future. Barbaro, L. & Van Halder, I. (2009) Linking bird, carabid Proceedings of the National Academy of Sciences USA, beetle and butterfly life-history traits to habitat 98, 5452–5457. fragmentation in mosaic landscapes. Ecography, 32, 321–333. Deutsch, C.A., Tewksbury, J.J., Huey, R.B., Sheldon, K.S., Ghalambor, C.K., Haak, D.C. & Martin, P.R. (2008) Baselga, A., Lobo, J.M., Svenning, J.-C., Aragón, P. & Impacts of climate warming on terrestrial ectotherms Araújo, M.B. (2012) Dispersal ability modulates the across latitude. Proceedings of the National Academy strength of the latitudinal richness gradient in of Sciences USA, 105, 6668–6672. European beetles. Global Ecology and Biogeography, 21, 1106–1113. Donnelly, M.A. & Crump, M.L. (1998) Potential effects of climate change on two Neotropical amphibian Bateman, B.L., Murphy, H.T., Reside, A.E., Mokany, K. & assemblages. Climatic Change, 39, 541–561. VanDerWal, J. (2013) Appropriateness of full-, partial- and no-dispersal scenarios in climate change Feeley, K.J. & Silman, M.R. (2010) Biotic attrition from impact modelling. Diversity and Distributions, 19, tropical forests correcting for truncated temperature 1224–1234. niches. Global Change Biology, 16, 1830–1836.

80 | Chapter II

Feeley, K.J. & Silman, M.R. (2011) The data void in butterflies to climate change using multiple criteria. modeling current and future distributions of tropical Biodiversity and Conservation, 19, 695–723. species. Global Change Biology, 17, 626–630. Henle, K., Davies, K.F., Kleyer, M., Margules, C. & Settlele, Feeley, K.J., Malhi, Y., Zelazowski, P. & Silman, M.R. J. (2004) Predictors of species sensitivity to (2012) The relative importance of deforestation, fragmentation. Biodiversity and Conservation, 13, precipitation change, and temperature sensitivity in 207–251. determining the future distributions and diversity of Hodgson, J.A., Thomas, C.D., Cinderby, S., Cambridge, H., Amazonian plant species. Global Change Biology, 18, Evans, P. & Hill, J.K. (2011) Habitat re-creation 2636–2647. strategies for promoting adaptation of species to Foden, W.B., Mace, G.M., Vié, J.-C., Angulo, A., Butchart, climate change. Conservation Letters, 4, 289–297. S.H.M., DeVantier, L., Dublin, H.T., Gutsche, A., Stuart, Hof, C., Araújo, M.B., Jetz, W. & Rahbek, C. (2011) S.N. & Turak, E. (2008) Species susceptibility to Additive threats from pathogens, climate and land- climate change impacts. Wildlife in a changing world: use change for global amphibian diversity. Nature, an analysis of the 2008 IUCN Red List of Threatened 480, 516–519. Species (ed. by J.-C. Vié, C. Hilton-Taylor and S.N. Stuart), pp. 77–88. IUCN, Gland, Switzerland. Huey, R.B., Deutsch, C.A., Tewksbury, J.J., Vitt, L.J., Hertz, P.E., Álvarez Pérez, H.J. & Garland, T. (2009) Why Foden, W.B., Butchart, S.H.M., Stuart, S.N., Vié, J.-C., tropical forest lizards are vulnerable to climate Akçakaya, H.R., Angulo, A., DeVantier, L.M., Gutsche, warming. Proceedings of the Royal Society B: A., Turak, E., Cao, L., Donner, S.D., Katariya, V., Biological Sciences, 276, 1939–1948. Bernard, R., Holland, R.A., Hughes, A.F., O’Hanlon, S.E., Garnett, S.T., Şekercioğlu, Ç.H. & Mace, G.M. (2013) Huey, R.B., Kearney, M.R., Krockenberger, A., Holtum, Identifying the world’s most climate change J.A.M., Jess, M. & Williams, S.E. (2012) Predicting vulnerable species: a systematic trait-based organismal vulnerability to climate warming: roles of assessment of all birds, amphibians and corals. PLoS behaviour, physiology and adaptation. Philosophical ONE, 8, e65427. Transactions of the Royal Society B: Biological Sciences, 367, 1665–1679. Fritz, S.A., Bininda-Emonds, O.R.P. & Purvis, A. (2009) Geographical variation in predictors of mammalian Huntley, B., Collingham, Y.C., Green, R.E., Hilton, G.M., extinction risk: big is bad, but only in the tropics. Rahbek, C. & Willis, S.G. (2006) Potential impacts of Ecology Letters, 12, 538–549. climatic change upon geographical distributions of birds. Ibis, 148, 8–28. Gamble, L.R., McGarigal, K. & Compton, B.W. (2007) Fidelity and dispersal in the pond-breeding Isaac, N.J.B. & Cowlishaw, G. (2004) How species amphibian, Ambystoma opacum: implications for respond to multiple extinction threats. Proceedings of spatio-temporal population dynamics and the Royal Society B: Biological Sciences, 271, 1135– conservation. Biological Conservation, 139, 247–257. 1141. Garcia, R.A., Burgess, N.D., Cabeza, M., Rahbek, C. & Li, Y., Cohen, J.M. & Rohr, J.R. (2013) Review and Araújo, M.B. (2012) Exploring consensus in 21st synthesis of the effects of climate change on century projections of climatically suitable areas for amphibians. Integrative Zoology, 8, 145–161. African vertebrates. Global Change Biology, 18, 1253– Lips, K.R., Reeve, J.D. & Witters, L.R. (2003) Ecological 1269. traits predicting amphibian population declines in González-Suárez, M., Lucas, P.M. & Revilla, E. (2012) Central America. Conservation Biology, 17, 1078– Biases in comparative analyses of extinction risk: 1088. mind the gap. Journal of Animal Ecology, 81, 1211– Luck, G.W., Lavorel, S., McIntyre, S. & Lumb, K. (2012) 1222. Improving the application of vertebrate trait-based González-Suárez, M., Gómez, A. & Revilla, E. (2013) frameworks to the study of ecosystem services. Which intrinsic traits predict vulnerability to Journal of Animal Ecology, 81, 1065–1076. extinction depends on the actual threatening Midgley, G.F., Hannah, L., Millar, D., Thuiller, W. & Booth, processes. Ecosphere, 4, art76. A. (2003) Developing regional and species-level Hansen, A.J., Burgess, N.D., Fjeldså, J. & Rahbek, C. (2007) assessments of climate change impacts on One degree resolution databases of the distribution of biodiversity in the Cape Floristic Region. Biological 739 species of amphibians in Sub-Saharan Africa. On- Conservation, 112, 87–97. line data source-version 1.00. Zoological Museum, Murray, K.A., Rosauer, D., McCallum, H. & Skerratt, L.F. University of Copenhagen, Copenhagen. (2011) Integrating species traits with extrinsic Hanski, I. (2005) The shrinking world: ecological threats: closing the gap between predicting and consequences of habitat loss. International Ecology preventing species declines. Proceedings of the Royal Institute, Oldendorf/Luhe. Society B: Biological Sciences, 278, 1515–1523. Heikkinen, R.K., Luoto, M., Leikola, N., Pöyry, J., Settele, J., Nakagawa, S. & Freckleton, R.P. (2008) Missing inaction: Kudrna, O., Marmion, M., Fronzek, S. & Thuiller, W. the dangers of ignoring missing data. Trends in (2009) Assessing the vulnerability of European Ecology and Evolution, 23, 592–596.

Species traits and climatic threats and opportunities | 81

Nakicenovic, N. & Swart, R. (2000) Special report on (2011) A framework for assessing threats and emissions scenarios. Cambridge University Press, benefits to species responding to climate change. Cambridge, UK. Methods in Ecology and Evolution, 2, 125–142. Ohlemüller, R., Anderson, B.J., Araújo, M.B., Butchart, Thuiller, W., Lafourcade, B., Engler, R. & Araújo, M.B. S.H.M., Kudrna, O., Ridgely, R.S. & Thomas, C.D. (2009) BIOMOD – a platform for ensemble (2008) The coincidence of climatic and species rarity: forecasting of species distributions. Ecography, 32, high risk to small-range species from climate change. 369–373. Biology Letters, 4, 568–572. Thuiller, W., Lavorel, S. & Araújo, M.B. (2005a) Niche Opdam, P. & Wascher, D. (2004) Climate change meets properties and geographical extent as predictors of habitat fragmentation: linking landscape and species sensitivity to climate change. Global Ecology biogeographical scale levels in research and and Biogeography, 14, 347–357. conservation. Biological Conservation, 117, 285–297. Thuiller, W., Lavorel, S., Araújo, M.B., Sykes, M.T. & Pannell, J.R. & Barrett, S.C.H. (1998) Baker’s law Prentice, I.C. (2005b) Climate change threats to plant revisited: reproductive assurance in a diversity in Europe. Proceedings of the National metapopulation. Evolution, 52, 657–668. Academy of Sciences USA, 102, 8245–8250. Parmesan, C. & Yohe, G. (2003) A globally coherent Triviño, M., Cabeza, M., Thuiller, W., Hickler, T. & Araújo, fingerprint of climate change impacts across natural M.B. (2013) Risk assessment for Iberian birds under systems. Nature, 421, 37–42. global change. Biological Conservation, 168, 192-200. Pearson, R.G. (2006) Climate change and the migration Urban, M.C., Tewksbury, J.J. & Sheldon, K.S. (2012) On a capacity of species. Trends in Ecology & Evolution, 21, collision course: competition and dispersal 111–113. differences create no-analogue communities and cause extinctions during climate change. Proceedings Pereira, H.M., Leadley, P.W., Proença, V. et al. (2010) of the Royal Society B: Biological Sciences, 279, 2072– Scenarios for global biodiversity in the 21st century. 2080. Science, 330, 1496–1501. Van Bocxlaer, I., Loader, S.P., Roelants, K., Biju, S.D., Perry, A.L., Low, P.J., Ellis, J.R. & Reynolds, J.D. (2005) Menegon, M. & Bossuyt, F. (2010) Gradual adaptation Climate change and distribution shifts in marine toward a range-expansion phenotype initiated the fishes. Science, 308, 1912–1915. global radiation of toads. Science, 327, 679–82. Peters, R.L. & Darling, J.D.S. (1985) The greenhouse Violle, C., Navas, M.-L., Vile, D., Kazakou, E., Fortunel, C., effect and nature reserves. BioScience, 35, 707–717. Hummel, I. & Garnier, E. (2007) Let the concept of Pöyry, J., Luoto, M., Heikkinen, R.K., Kuussaari, M. & trait be functional! Oikos, 116, 882–892. Saarinen, K. (2009) Species traits explain recent Whitmee, S. & Orme, C.D.L. (2013) Predicting dispersal range shifts of Finnish butterflies. Global Change distance in mammals: a trait-based approach. Journal Biology, 15, 732–743. of Animal Ecology, 82, 211–21. Purvis, A., Gittleman, J.L., Cowlishaw, G. & Mace, G.M. Williams, J.W., Jackson, S.T. & Kutzbach, J.E. (2007) (2000) Predicting extinction risk in declining species. Projected distributions of novel and disappearing Philosophical transactions of the Royal Society B: climates by 2100 AD. Proceedings of the National Biological Sciences, 267, 1947–1952. Academy of Sciences USA, 104, 5738–5742. Serra-Diaz, J.M., Franklin, J., Ninyerola, M., Davis, F.W., Williams, S.E., Shoo, L.P., Isaac, J.L., Hoffmann, A.A. & Syphard, A.D., Regan, H.M. & Ikegami, M. (2013) Langham, G. (2008) Towards an integrated Bioclimatic velocity: the pace of species exposure to framework for assessing the vulnerability of species climate change. Diversity and Distributions, 20, 169– to climate change. PLoS Biology, 6, e325. 180. Sinervo, B., Méndez-de-la-Cruz, F., Miles, D.B. et al. (2010) Erosion of lizard diversity by climate change and altered thermal niches. Science, 328, 894–899. Supporting Information Summers, D.M., Bryan, B.A., Crossman, N.D. & Meyer, W.S. (2012) Species vulnerability to climate change: Appendix S1 List of sub-Saharan African am- impacts on spatial conservation priorities and phibians included in the study. species representation. Global Change Biology, 18, 2335–2348. Appendix S2 Representativeness of the subset Sunday, J.M., Bates, A.E. & Dulvy, N.K. (2012) Thermal of sub-Saharan African amphibians used in the tolerance and the global redistribution of animals. study. Nature Climate Change, 2, 686–690. Appendix S3 Comparison of climate change Thomas, C.D., Hill, J.K., Anderson, B.J., Bailey, S., Beale, vulnerability traits across different groups of C.M., Bradbury, R.B., Bulman, C.R., Crick, H.Q.P., Eigenbrod, F., Griffiths, H.M., Kunin, W.E., Oliver, T.H., sub-Saharan African amphibians. Walmsle,y C.A., Watts, K., Worsfold, N.T. & Yardley, T.

82 | Chapter II

Appendix S1

List of sub-Saharan African amphibians included in the study. Species taxonomy follows Frost (2013), and codes in brackets refer to the IUCN Red List status: Extinct (EX), Regionally Extinct (RE), Extinct in the Wild (EW), Critically Endangered (CR), Endangered (EN), Vulnerable (VU), Near Threatened (NT), Least Concern (LC), and Data Deficient (DD).

Acanthixalus spinosus (LC) Amietophrynus poweri (LC) Chiromantis rufescens (LC)

Afrixalus aureus (LC) Amietophrynus rangeri (LC) Chiromantis xerampelina (LC)

Afrixalus crotalus (LC) Amietophrynus regularis (LC) Cryptothylax greshoffii (LC)

Afrixalus delicatus (LC) Amietophrynus xeros (LC) Geotrypetes seraphini (LC)

Afrixalus dorsalis (LC) Arthroleptis adelphus (LC) Heleophryne natalensis (LC)

Afrixalus enseticola (VU) Arthroleptis poecilonotus (LC) Hemisus guineensis (LC)

Afrixalus equatorialis (LC) Arthroleptis stenodactylus (LC) Hemisus marmoratus (LC)

Afrixalus fornasini (LC) Arthroleptis sylvaticus (LC) Hemisus microscaphus (LC)

Afrixalus fulvovittatus (LC) Arthroleptis taeniatus (LC) Hemisus olivaceus (LC)

Afrixalus laevis (LC) Arthroleptis variabilis (LC) Hylarana albolabris (LC)

Afrixalus leucostictus (LC) Arthroleptis xenochirus (LC) Hylarana amnicola (LC)

Afrixalus nigeriensis (NT) Arthroleptis xenodactyloides (LC) Hylarana darlingi (LC)

Afrixalus osorioi (LC) Breviceps adspersus (LC) Hylarana galamensis (LC)

Afrixalus paradorsalis (LC) Breviceps mossambicus (LC) Hylarana lemairei (LC)

Afrixalus quadrivittatus (LC) Breviceps poweri (LC) Hylarana lepus (LC)

Afrixalus stuhlmanni (LC) Breviceps verrucosus (LC) Hymenochirus boettgeri (LC)

Afrixalus vittiger (LC) Bufo pentoni (LC) Hyperolius acuticeps (LC)

Afrixalus weidholzi (LC) Cacosternum boettgeri (LC) Hyperolius argus (LC)

Afrixalus wittei (LC) Cacosternum nanum (LC) Hyperolius benguellensis (LC)

Alexteroon hypsiphonus (LC) Cardioglossa elegans (LC) Hyperolius cinnamomeoventris (LC)

Alexteroon obstetricans (LC) Cardioglossa escalerae (LC) Hyperolius concolor (LC)

Amietia angolensis (LC) Cardioglossa gracilis (LC) Hyperolius fusciventris (LC)

Amietia fuscigula (LC) Cardioglossa gratiosa (LC) Hyperolius glandicolor (LC)

Amietophrynus garmani (LC) Cardioglossa leucomystax (LC) Hyperolius guttulatus (LC)

Amietophrynus gutturalis (LC) Chiromantis kelleri (LC) Hyperolius kivuensis (LC)

Amietophrynus maculatus (LC) Chiromantis petersii (LC) Hyperolius lamottei (LC)

Hyperolius langi (LC) Phrynobatrachus bullans (LC) Ptychadena obscura (LC)

Hyperolius lateralis (LC) Phrynobatrachus calcaratus (LC) Ptychadena oxyrhynchus (LC)

Hyperolius marginatus (LC) Phrynobatrachus cornutus (LC) Ptychadena perreti (LC)

Hyperolius marmoratus (LC) Phrynobatrachus francisci (LC) Ptychadena porosissima (LC)

Hyperolius mitchelli (LC) Phrynobatrachus guineensis (NT) Ptychadena pumilio (LC)

Hyperolius mosaicus (LC) Phrynobatrachus gutturosus (LC) Ptychadena schillukorum (LC)

Species traits and climatic threats and opportunities | 83

Hyperolius nasutus (LC) Phrynobatrachus hylaios (LC) Ptychadena subpunctata (LC)

Hyperolius nitidulus (LC) Phrynobatrachus liberiensis (NT) Ptychadena superciliaris (NT)

Hyperolius parallelus (LC) Phrynobatrachus mababiensis (LC) Ptychadena taenioscelis (LC)

Hyperolius parkeri (LC) Phrynobatrachus natalensis (LC) Ptychadena tellinii (LC)

Hyperolius phantasticus (LC) Phrynobatrachus parvulus (LC) Ptychadena tournieri (LC)

Hyperolius pusillus (LC) Phrynobatrachus perpalmatus (LC) Ptychadena trinodis (LC)

Hyperolius semidiscus (LC) Phrynobatrachus phyllophilus (NT) Ptychadena upembae (LC)

Hyperolius tuberilinguis (LC) Phrynobatrachus plicatus (LC) Ptychadena uzungwensis (LC)

Hyperolius viridiflavus (LC) Phrynobatrachus tokba (LC) Pyxicephalus adspersus (LC)

Hyperolius zonatus (NT) Phrynobatrachus villiersi (VU) Pyxicephalus edulis (LC)

Kassina arboricola (VU) Phrynomantis affinis (LC) Schismaderma carens (LC)

Kassina cassinoides (LC) Phrynomantis annectens (LC) Semnodactylus wealii (LC)

Kassina cochranae (NT) Phrynomantis bifasciatus (LC) Silurana epitropicalis (LC)

Kassina fusca (LC) Phrynomantis microps (LC) Spelaeophryne methneri (LC)

Kassina kuvangensis (LC) Phrynomantis somalicus (LC) Strongylopus fasciatus (LC)

Kassina maculata (LC) Poyntonophrynus beiranus (LC) Strongylopus grayii (LC)

Kassina maculifer (LC) Poyntonophrynus dombensis (LC) Tomopterna cryptotis (LC)

Kassina maculosa (LC) Poyntonophrynus fenoulheti (LC) Tomopterna delalandii (LC)

Kassina senegalensis (LC) Poyntonophrynus hoeschi (LC) Tomopterna krugerensis (LC)

Kassina somalica (LC) Poyntonophrynus kavangensis (LC) Tomopterna marmorata (LC)

Kassinula wittei (LC) Poyntonophrynus vertebralis (LC) Tomopterna natalensis (LC)

Leptopelis viridis (LC) Ptychadena anchietae (LC) Tomopterna tandyi (LC)

Mertensophryne lindneri (LC) Ptychadena ansorgii (LC) Tomopterna tuberculosa (LC)

Mertensophryne micranotis (LC) Ptychadena bibroni (LC) Vandijkophrynus gariepensis (LC)

Mertensophryne taitana (LC) Ptychadena bunoderma (LC) Xenopus andrei (LC)

Nectophryne afra (LC) Ptychadena erlangeri (NT) Xenopus borealis (LC)

Nectophryne batesii (LC) Ptychadena grandisonae (LC) Xenopus clivii (LC)

Opisthothylax immaculatus (LC) Ptychadena guibei (LC) Xenopus fraseri (LC)

Paracassina obscura (LC) Ptychadena keilingi (LC) Xenopus laevis (LC)

Phrynobatrachus acridoides (LC) Ptychadena longirostris (LC) Xenopus muelleri (LC)

Phrynobatrachus alleni (NT) Ptychadena mascareniensis (LC) Xenopus petersii (LC)

Phrynobatrachus auritus (LC) Ptychadena mossambica (LC) Xenopus pygmaeus (LC)

Phrynobatrachus batesii (LC) Ptychadena neumanni (LC) Xenopus tropicalis (LC)

References Frost, D.R. (2013) Amphibian Species of the World: an Online Reference. Version 5.6 (9 January 2013). Available at: http://research.amnh.org/herpetology/amphibia/index.html.

84 | Chapter II

Appendix S2

Representativeness of the subset of sub-Saharan African amphibians used in the study. The first four panels compare all species in our dataset (n=695), the subset of species with bioclimatic envelope model projections available (n=272), and the subset of species with both model projections and trait data available (n=195, the subset used in the study), in terms of geographical distribution (a), taxonomy (b), range size (c), and IUCN Red List status (d). The last panel (e) compares species included (n=195) and excluded (n=500) in the study in terms of climate change vulnerability according to the selected traits and ecological characteristics of species and their ranges (Foden et al., 2008, 2013). Boxplots compare the distributions of values for continuous traits, and barplots compare the proportion of species assigned high, lower or unknown vulnerability for categorical traits.

(a)

(b)

Species traits and climatic threats and opportunities | 85

(c)

(d)

(e)

86 | Chapter II

Appendix S3

Comparison of climate change vulnerability traits across different groups of sub-Saharan African amphibians. The spatial distribution of species is compared between those assigned high (top row) and lower (bottom row) vulnerability according to five traits or characteristics of species and their ranges (a). High-vulnerability species are those with the lowest temperature or precipitation tolerances, dispersal ability, and reproductive output, and with the strongest dependence on precipitation cues (Foden et al., 2008, 2013). Vulnerability due to the same traits, as well as geographical range size, is also compared across three groups of species exposed to different combinations of extrinsic threats and opportunities from climate change ('contracting in situ' [n=10], 'obligate shifting' [n=6], and 'expanding' [n=21] species), the remaining species included in the study (n=158), and those excluded (n=500) (b). For excluded species, closed bars refer to proportions of excluded species with data that are assigned high vulnerability, and open bars refer to proportions of all excluded species that are assigned high or unknown vulnerability.

(a)

(b)

Chapter III

Conservation implications of omitting rare and threatened species from climate change impact modelling

PHILIP J. PLATTS, RAQUEL A. GARCIA, CHRISTIAN HOF, WENDY B. FODEN, LOUIS HANSEN, CARSTEN RAHBEK, AND NEIL D. BURGESS

Manuscript in review

Conservation implications of omitting rare and threatened species from climate change impact modelling

PHILIP J. PLATTS1*, RAQUEL A. GARCIA2,3,4*, CHRISTIAN HOF5, WENDY B. FODEN6,7, LOUIS HANSEN2, CARSTEN RAHBEK2 ,and NEIL D. BURGESS2,8,9*

1 Environment Department, University of York, Heslington, York, UK 2 Center for Macroecology, Evolution and Climate, Department of Biology, University of Copenhagen, Denmark 3 Department of Biogeography and Global Change, National Museum of Natural Sciences, CSIC, Madrid, Spain 4 InBio/CIBIO, University of Évora, Évora, Portugal 5 Biodiversity and Climate Research Centre (BiK-F) & Senckenberg Gesellschaft für Naturforschung, Frankfurt, Germany 6 Animal, Plant and Environmental Sciences, University of the Witwatersrand, Johannesburg, South Africa 7 Climate Change Specialist Group, IUCN Species Survival Commission 8 United Nations Environment Programme World Conservation Monitoring Centre, Cambridge, UK 9 World Wildlife Fund–US, Washington, USA * Contributed equally to the manuscript

Manuscript in review

Abstract

Aim Projected impacts of climate change on spe- Results Over half of sub-Saharan African amphib- cies distributions are increasingly factored into ians are too rare for large-scale SDM, including conservation plans. Species distribution modelling 94% of those threatened with extinction. The (SDM) is commonly used to predict impacts on omitted species occupy topographically complex sets of taxa with sufficient distributional records, areas with cooler, wetter and less seasonal cli- while omitting rare species of high conservation mates, which are projected to experience lower importance. We investigate whether this dichot- rates of change. Spatial priorities for these species omy: restricts SDM calibration to an unrepre- exhibit higher congruence with existing conserva- sentative sample of present-day conditions tion schemes than do priorities derived from SDM. and/or exposure to change; and has potential to Congruence between existing schemes and SDM undermine existing conservation priorities for priorities decreases into the future, with globally Africa. important sites such as the Eastern Arc Mountains Location Sub-Saharan Africa. being omitted. Methods We use multivariate ordination to char- Main conclusions Future conservation priorities acterise the environmental niches of 767 African derived using SDM may systematically downplay amphibian species, distinguishing between those important areas for rare and threatened species, ‘eligible’ for large-scale SDM (>10 records at 1° which currently underpin global allocation of resolution) and the remaining ‘rare’ species. Eligi- conservation funds. This issue spans taxonomic ble species’ distributions are projected to 1980 groups and is only partially mitigated by model- and 2080 using MaxEnt. Empirical priorities are ling at finer spatial resolutions. Effective biodiver- then derived, based on species richness, rarity and sity conservation relies on our capacity to predict irreplaceability, and compared with spatial pat- climate change impacts on all species, and thus a terns for rare species and three existing conserva- wider range of approaches beyond SDM is essen- tion schemes: Biodiversity Hotspots, Endemic Bird tial. Areas and Global 200.

90 | Chapter III

Introduction 2002). In tropical regions, which contain most of the world’s biodiversity, species are more Limited funds for tackling unprecedented rates often narrow-ranging than in temperate zones of biodiversity loss dictate that spatial priori- (Wiens et al., 2006) and even common species’ ties for conservation must be identified and distributions tend to be less well documented periodically revised (Margules & Pressey, 2000; (da Fonseca et al., 2000; Küper et al., 2006; Butchart et al., 2010). Large-scale priority Feeley & Silman, 2011). Consequently, those schemes have focussed on measures of species rare and threatened species that have tradi- richness and threat (Myers et al., 2000), re- tionally been used to define conservation prior- stricted ranges and threat (Stattersfield et al., ities are the ones most likely to be omitted for 1998) and other ways to maximise species correlative SDM. An urgent question for con- conservation at minimal cost (Wilson et al., servation planning is whether this systematic 2006). In general, biologically rich areas with bias toward more widespread species results in high irreplaceability, which are also vulnerable a spurious shift, now or in the future, toward or to degradation and loss, are given the highest away from sites prioritised under existing con- rank. In recent years, the predicted (modelled) servation schemes. impact of climate change on species distribu- We address this question using African tions has become an additional consideration amphibians as an exemplar group. Amphibians for conservation priority setting (Hannah et al., contain many narrow-ranging and threatened 2002; Huntley et al., 2006; Foden et al., 2013). species, typical of most tropical species groups, The collection of biological data is time but also wide-ranging species that are more consuming and expensive, particularly for rare typical of species’ distributions at higher lati- species (Ahrends et al., 2011). Despite its im- tudes. A further motivation for studying am- portance for conservation, such work is chroni- phibians is the high rates of threat they are cally underfunded (Balmford & Gaston, 1999; predicted to face from climate change, habitat da Fonseca et al., 2000). In prioritising sites for loss and disease, especially in Africa (Sodhi et conservation, the near term solution is to pro- al., 2008; Hof et al., 2011a; Foden et al., 2013). ject existing distributional data across space Separating species that can be modelled using and, increasingly, into the future, using envi- correlative SDM from those that must be omit- ronmental correlates (Pressey et al., 2000). ted due to rarity, we aim to establish whether Correlative models that predict species distri- this dichotomy: (1) systematically restricts butions under climate change are frequently model calibration to an unrepresentative sam- applied in the scientific literature, and are ple of present-day conditions and/or exposure widely cited by conservation planners, who to climate change, and (2) has potential to un- seek to ensure that priorities remain valid undermine or reinforce existing conservation der future climates (Williams et al., 2005; Hole priority schemes across sub-Saharan Africa. We et al., 2009; Shoo et al., 2011). consider three schemes widely used by interna- Considerable literature exists on species tional organisations to allocate conservation distribution modelling (SDM), its strengths and funds: Conservation International’s Hotspots, limitations (e.g., Wiens et al., 2009; Araújo & BirdLife International’s Endemic Bird Areas Peterson, 2012; Franklin, 2013). For conserva- and the World Wildlife Fund’s Global 200. The tion prioritisation, perhaps the most fundamen- extent to which our results may be generalised tal constraint is that only species with adequate across spatial resolutions and taxonomic numbers of spatially distinct occurrence rec- groups is discussed, and avenues for uniting ords can be modelled (Stockwell & Peterson, established conservation prioritisation proce-

Omission of narrow-ranging species | 91

dures with SDM projections under climate with the addition of 26 recently described spe- change are identified. cies that are included here, but not yet assessed on the IUCN Red List (Table S2). The species data are reliable to 1° resolution (111 km at the Material and Methods equator), similar to that of the highest resolution general circulation model (MIROC ver. 3.2, Amphibian data 1.1° × 1.1°) and consistent with most studies of continental-scale biodiversity pattern and pri- Distributional data were collated for 767 spe- oritisation under climate change (e.g., McClean cies of amphibians found in mainland Africa et al., 2005; Huntley et al., 2006; Hole et al., south of the Sahara (Fig. 1a and Table S1 in 2009; Garcia et al., 2012). Supporting Information; Hansen et al., 2007, Each amphibian was allocated to one of updated to November 2011 by the University of two species sets: those that are eligible for Copenhagen, Denmark). The taxonomic basis correlative SDM at 1° resolution (N=335) and was Frost et al. (2006), cross-checked against those that are not (N=432), hereafter referred the IUCN list of amphibians for the same region, to respectively as ‘eligible’ and ‘rare’ amphibi-

Figure 1 | (a) Mountainous areas in sub-Saharan Africa (Kapos et al., 2000), overlaid with the distribution of records for ‘rare’ (≤10 records at 1° resolution) and more widespread (>10 records) amphibians, the latter being eligible for species distribution modelling (SDM), while the former are omitted due to statistical constraints. Stacked bar chart (b) details the number of distinct 1° grid cells known to be occupied by African amphibians, including a break-down of threat in each range-size category, according to the IUCN Red List (www.iucnredlist.org, accessed June 2011).

92 | Chapter III

ans (Fig. 1 and Table S1). We used a ten record niche differences in environmental niche space cutoff for the minimum number of records between eligible and rare species, both under required for SDM. Stricter restrictions have present conditions and projected future chang- been suggested, such as 50 records in total or es. The OMI analysis identifies the ordination ten records per predictor variable in the model axes that optimise separation between species (Harrell et al., 1984; Stockwell & Peterson, occurrences, and quantifies the niche position 2002; Wisz et al., 2008), but ten is the more and niche breadth for each species along those commonly applied cutoff. For both species sets, axes. Niche position, or marginality, is meas- we identified the proportion of species listed as ured as the deviation of the mean environmen- Threatened on the IUCN Red List under differ- tal conditions occupied by a species (its cen- ent threat categories (www.iucnredlist.org, troid) from the mean conditions of the study accessed June 2011; Tables S1 and S3). area (origin of the hyperspace). The larger the deviation, the more marginal the species’ niche Environmental data position is in relation to the available environmental space. Niche breadth, or tolerance, is We selected four predictor variables commonly quantified as the dispersion of environmental used in SDM: mean annual temperature, annual conditions occupied by species, with larger rainfall, rainfall seasonality and elevational dispersion values indicating higher tolerance. range within 1° pixels (according to a 30 arc- We performed a first OMI analysis on the second elevation model; Farr et al., 2007). The four predictor variables representing present- rainfall seasonality index (PSI; Walsh & Lawler, day conditions, and a second on future climate 1981) reflects rainfall contrasts across seasons anomalies. Anomalies for mean annual tem- (pres) in relation to the total annual precipita- perature and rainfall seasonality were com- tion (preT): puted by subtracting the future from the pre-

sent values; annual rainfall anomalies were .. given by the ratio of future to present values. For both OMI analyses, we performed randomi- Climatic data for the baseline period sation tests (1000 permutations) to determine (1980s, mean of 1970-1999) were from the whether species’ niche positions differed sig- World Climate Research Programme’s Coupled nificantly from random expectations. Wilcoxon Model Intercomparison Project (Meehl et al., signed-rank one-sided tests were used to test 2007). For future climatic conditions (2080s, whether rare species have higher marginality mean of 2070-2099), we used data from the and/or lower tolerance than eligible species. MIROC Global Circulation Model (ver. 3.2) un- Calculations were performed in R using the der IPCC-AR4 emissions scenario A1B. ade4 package (Chessel et al., 2004; R-Core- Team, 2012). Niche differences between species Implicit in studies that project future conserva- Species distributions under climate change tion priorities using correlative SDM are the Correlative SDMs were constructed for all eligi- assumptions that the eligible species ble amphibians using MaxEnt (ver. 3.3.3e; subsetrepresents similar environmental space Phillips & Dudik, 2008), chosen for its wide- to that of omitted rare species, and that pro- spread application and favourable performance jected exposure to change will also be similar. in comparative studies (Hernandez et al., Using multivariate ordination (Outlying Mean 2006). Models were calibrated using the same Index, OMI; Dolédec et al., 2000), we assessed four predictor variables described above, and

Omission of narrow-ranging species | 93

then projected for the years 1980 and 2080. empty cells and reduced it, one cell at a time, Ten replicates were performed for each species, until no further reduction was possible without with the final prediction for a given pixel being excluding one or more species. The procedure the mean value over these ten runs. To assess was randomised and repeated 100 times. The generality, we cross-validated the area under final priority metric was the union over all the receiver-operating characteristic curve optimal solutions. (AUC), repeating ten times and recording the For context, we derived and compared the average. Continuous occurrence probabilities same three metrics for the rare species set. were then dichotomised into presence-absence Since correlative SDM was not possible for maps by maximising the sum of sensitivity and these species, their distributions were estimat- specificity (Liu et al., 2013). Otherwise, the ed using multidimensional niche envelopes MaxEnt default settings were accepted, so that (MDNE). Correlative SDM, such as maximum background data were distributed across all entropy and statistical regression, describe cells in sub-Saharan Africa (N = 1954; similar to gradients of suitability as well as the relative the overall sample coverage for the class importance of predictors in a model. The amphibia; Fig. 1a). MDNE, however, classifies all conditions within Future predictions were constrained under the observed environmental range as uniformly a no-dispersal scenario. This is appropriate due viable for a species, and conditions beyond as to the coarse resolution of the 1° grid (Bateman wholly unsuitable. This simplistic approach et al., 2013). Considered at this scale, climate means that only two distinct records of occur- change velocities are likely to outpace dispersal rence are required for spatial prediction (Platts capabilities for most African amphibians (Hof et et al., 2013). Accordingly, we mapped the dis- al., 2011a, 2011b): to traverse a single pixel, the tributions of 256 of the 432 rare amphibians, leading edge of a population would need to and from these derived the same priority met- migrate more than 1 km/y, subject to continu- rics described above. ous corridors for dispersal and a steady change Empirically derived metrics for both pre- in climate. Yet, habitat fragmentation and tem- sent and future conditions were checked for poral climatic variability are likely to curtail congruence with three large-scale conservation future rates and patterns of range expansion priority schemes: Biodiversity Hotspots relative to the past (Early & Sax, 2011; Bennie (Conservation International; Mittermeier et al., et al., 2013; Pyron & Wiens, 2013). 2004), Endemic Bird Areas (BirdLife International; Stattersfield et al., 1998) and The Conservation priorities Global 200 (World Wildlife Fund US; Olson & Dinerstein, 1998). Congruence was defined as From the MaxEnt predictions for both present the percentage of empirically prioritised cells and future time periods, we derived three spa- that fell within existing schemes. tial priority metrics widely used in the conservation planning literature (Williams, 1998; Williams et al., 2005): (1) the 100 grid cells Results with the highest species richness; (2) the 100 grid cells with the highest range-size rarity, defined as the sum of inverse range sizes of all Degree of (non-climatic) threat species in a grid cell; (3) minimum sets of grid Of the 767 amphibian species in sub-Saharan cells that represent all species at least once Africa, one quarter are designated Threatened (greedy complementarity). To identify mini- on the IUCN Red List: 71 (38%) Vulnerable, 88 mum sets, we begun with a full set of non- (47%) Endangered, 26 (14%) Critically Endan-

94 | Chapter III

gered and one Extinct in the Wild (now rein- restricted to species with statistically signifi- troduced to original locality). Because restrict- cant values of marginality (Fig. S1a). ed range is one of the criteria for Red Listing species, the degree of threat is significantly Exposure to change skewed toward taxa with fewer occurrence The OMI analysis on climate anomalies also records at the 1° model resolution (χ2 = 150.97, revealed contrasting patterns between eligible p < 0.001). Thus, only 11 of the 186 threatened and rare species. The first axis, the main envi- amphibian species were eligible for correlative ronmental gradient, explains niche differentia- SDM, whereas the remaining 175 threatened tion between species occupying areas that are species, including nearly all of those in the projected to warm less and decrease in rainfall highest threat categories (Endangered and seasonality, compared with those expected to Critically Endangered), were too rare to model experience stronger warming and higher rain- (Fig. 1b and Table S3). fall seasonality. The second axis represents a gradient from negative to positive changes in Niche differences between species annual rainfall. Together, they explain 75% of the total variance in species' exposure to cli- Rare amphibians have been recorded predomi- mate change (Table S4). Rare species were nantly in the tropical highlands (regions 1-8 in found to be significantly more marginal and Fig. 1a), with further centres of rarity in the less tolerant than eligible species (Fig. 2f and Cape Fynbos and Drakensberg of South Africa Table S5; Wilcoxon signed-rank one-sided tests, (regions 9-10). By contrast, at least one eligible p < 0.05), occupying areas projected to be less amphibian has been recorded in the majority of exposed to warming, drying and increased grid cells across sub-Saharan Africa (Fig. 1a). seasonality (Fig. 2d-e). The same patterns hold when the comparison on both OMI axes is re- Current conditions stricted to species with significant marginality The first two ordination axes of the OMI analy- (Fig. S1b). sis explained 81% of niche separation between species, with the first axis capturing rainfall Conservation priorities gradients and the second representing temperature and elevation gradients (Table S4). Eligi- Model performance ble and rare species were found to be signifi- The predictive performance of MaxEnt models cantly different in both their marginality and for eligible species was good, with median tolerance, with respect to the first two ordina- cross-validated AUC of 0.95 (interquartile tion axes (Fig. 2c and Table S5; Wilcoxon range [IQR], 0.91-0.97). At the presence- signed-rank one-sided tests, p < 0.05). Whereas absence threshold, the predictions had median eligible species are more ubiquitous on the sensitivity of 0.94 (IQR, 0.91-0.97) and median study area, rare species tend to occupy envi- specificity of 0.92 (IQR, 0.88-0.95). The degree ronmental conditions that deviate from the to which models extrapolated beyond the pa- average conditions for sub-Saharan Africa (Fig. rameter range of the training data was general- 2a-b). These marginal environments are char- ly minimal, the exception being novel high tem- acterised by higher annual rainfall with lower peratures in the Sahelian zone, and a small area rainfall seasonality (axis 1), and by cooler tem- of high uncertainly in the eastern Congo rain- peratures amidst more complex topography forests (Fig. S2). The most important predictor (axis 2). Rare species also have environmental in MaxEnt models was annual rainfall (median niches that are narrower than those of eligible contribution, 34%), followed by annual tem- species. These results hold when the analysis is

Omission of narrow-ranging species | 95

Figure 2 | Niche differences between species that are eligible for SDM and the remaining rare species set. Scatterplots (a-b, d-e) overlay all cells in the study region (grey) with observed species richness across environmental space, represented by the first two ordination axes of the Outlying Mean Index. Insets show representations for two example species, selected as those having median marginality among eligible and rare species sets, respectively. The further the centroid of the species from the origin of the hyperspace, the more marginal the species; i.e., the more dissimilar the conditions it occupies. The wider the ellipse, the wider is the species’ niche-breadth (tolerance). Boxplots (c, f) compare marginality and tolerance for both sets of species. Whiskers extend up to 1.5 times the interquartile range from each box. As indicated by non-overlapping notches, eligible and rare species sets are significantly different (Wilcoxon signed-rank, p < 0.05; see also Fig. S1).

perature (25%), rainfall seasonality (23%) and median specificity of 0.94 (IQR, 0.77-0.99), elevation range (3%), although this varied indicating that MaxEnt extrapolates farther considerably between species (respectively, from the observed distributions than MDNE. IQR: 23-46%, 10-42%, 10-42% and 2-8%; Fig. For species richness and range-size rarity, S3). MDNE and MaxEnt results were highly corre- The MDNE maps for rare species could not lated (Spearman’s ρ = 0.72 and ρ = 0.68; Fig. be validated directly: by definition, there are no S4), although compared with MaxEnt, priority errors of omission against the training data, sites (top 100 cells) derived using MDNE un- while specificity is unknown. To gauge compa- derestimated congruence with existing conser- rability of priorities derived using MDNE with vation priority schemes, especially with respect priorities derived using correlative SDM, we to Biodiversity Hotspots and Endemic Bird instead mapped MDNEs for eligible species, and Areas. Using minimum sets to identify priority used MaxEnt predictions as test data. Accord- sites, MDNE and MaxEnt results for eligible ingly, MDNE maps for eligible species had me- species were similar (Fig. 4). dian sensitivity of 0.87 (IQR, 0.63-0.98) and

96 | Chapter III

Priority metrics versus existing schemes under current climate, despite the exclusion of Spatial priorities derived from correlative SDM local endemics from correlative SDM, but under for eligible species identified hotspots of spe- future climate these regions were omitted en- cies richness and range-size rarity in West tirely from species richness and range-size Africa, particularly the Cameroon Highlands rarity priorities (Fig. 3). (region 2 in Fig. 1a). Together with the For present-day species richness, range- Albertine Rift and Eastern Congo (region 6), size rarity and – to a lesser extent – minimum these sites maintained their priority status sets, congruence with existing priority schemes under future climate (Fig. 3). Other important was higher on the rare species set than on the regions were missed by SDM priority metrics eligible species set, despite the opposing bias of for both 1980 and 2080, such as the South Afri- MDNE (cf. MaxEnt; Fig. 4). For rare amphibians, can Fynbos and Drakensberg (regions 9-10), 70-87% (richness) and 61-79% (range-size although not when using minimum sets to pri- rarity) of prioritised cells coincided with exist- oritise sites. Globally important centres of en- ing schemes, compared with eligible species demism such as the Eastern Arc Mountains and congruence of 5-49% and 15-55% (MDNE), or Southern Rift (regions 7-8) were identified 43-77% and 43-70% (MaxEnt), respectively.

Rare species Eligible species Eligible species

MDNE (1980s) SDM (1980s) SDM (2080s) Species richness

Range-size rarity

Minimum sets

Figure 3 | Conservation priorities derived empirically from data on amphibian distributions, projected using simple niche envelopes (MDNE for rare species, 1980 only) and correlative SDM (MaxEnt for eligible species, 2080 according to IPCC-AR4 scenario A1B). For species richness and range-size rarity, the top 100 scoring cells are selected as priority sites. Minimum sets identify the smallest possible subsets of cells that represent all species at least once; where the solution is not unique, the union across solutions is mapped. Ellipses highlight key sites projected to lose (solid lines) or gain (dashed lines) priority status under future climate, according to correlative SDM. For a key to regions, see Fig. 1a.

Omission of narrow-ranging species | 97

Figure 4 | Spatial congruence between empirically derived amphibian priorities and three existing conservation priority schemes in sub-Saharan Africa. Bar heights equal the mean percentage of cells that coincide with existing schemes (stacking shows relative contributions to the mean). Underlying species distributions were mapped on 1° grids using simple niche envelopes (MDNE) and – for eligible (E) species – MaxEnt (SDM). Rare species (R) are defined as those having too few records for correlative SDM. Projections of congruence under 2080 climate are according to MaxEnt models under IPCC-AR4 emissions scenario A1B.

For both species sets, the degree of congruence size rarity to prioritise sites, congruence with with The Global 200 was higher than with Bio- Biodiversity Hotspots reduced from 43% in diversity Hotspots or Endemic Bird Areas. Cur- 1980 to 30-32% by 2080, while congruence rent priorities according to minimum sets were with Endemic Bird Areas reduced from 48-55% more similar for rare versus eligible species, to 37-43%. For minimum sets, congruence with suggesting that this metric may be more robust Biodiversity Hotspots and The Global 200 de- to the omission of rare species from correlative creased slightly, while congruence with Endem- SDM (Figs. 3-4). ic Bird Areas increased (Fig 4). Due to higher Using SDM to project conservation priori- irreplaceability, minimum sets for rare species ties under future climate, congruence with were larger than for eligible species (65 cells, existing schemes reduced on average, such that cf. 20 [MDNE] or 15 [MaxEnt]). Under future future priorities diverged further from rare climate, more cells were required to capture species patterns (Figs. 3 and 4). Under all three each eligible species at least once, so that the metrics, the proportion of prioritised cells with- minimum set size increased from 15 to 26 cells, in The Global 200 remained relatively stable while the number of unique solutions de- though time. Using species richness and range-

98 | Chapter III

creased from seven to one, reducing the overall point out, any SDM must be underpinned by coverage of the union across solutions (Fig. 3). sufficient data on a species’ distribution to avoid spurious predictions – a prerequisite that Discussion is, almost by definition, not fulfilled by many taxa of highest conservation concern, particu- Natural habitats are changing at a rate unprec- larly at the coarse spatial resolutions dictated edented in human history (Butchart et al., by species and climate data at global or conti- 2010). Faced with finite resources for address- nental scales. ing the rapid decline in global biodiversity, In this study, we have highlighted that over scientists have developed analytical frame- half of all known amphibian species in sub- works to locate priority sites for conservation Saharan Africa are too rare for large-scale SDM, investment (e.g., Olson & Dinerstein, 1998; and that these species are disproportionately Stattersfield et al., 1998; Mittermeier et al., Threatened on the IUCN Red List, with 94% of 2004). Without these global priorities, re- all Threatened amphibians being ineligible for sources would be spread too thinly to have any modelling. Crucially for studies that infer con- meaningful impact on species conservation servation priorities from SDM, we find that the (Pimm et al., 2001; Wilson et al., 2006). Large- omitted amphibians occupy significantly differ- scale priority schemes, such as Biodiversity ent niche space to eligible species: their ob- Hotspots, Endemic Bird Areas and The Global served distributions are characterised by high- 200, were developed directly from the available er annual rainfall with lower rainfall seasonali- data on species occurrence, rarity and threats, ty, and by cooler temperatures and more com- guided by expert opinion. More recently, spe- plex topography. This is consistent with previ- cies distribution modelling (SDM) has become ous studies, which also find that climatically an almost ubiquitous technique in the conser- and topographically diverse/distinct areas tend vation scientist’s toolbox, accelerated in popu- to contain a disproportionate richness of nar- larity by free internet access to large databases row-ranging species, both at continental (de of herbaria specimens and other species infor- Klerk et al., 2002; Ohlemüller et al., 2008) and mation (Graham et al., 2004), digitised envi- global (Sandel et al., 2011; Pyron & Wiens, ronmental data, and a latent demand for pre- 2013) scales. dictive spatial models, particularly with respect Topographic complexity confers climatic to the impacts of anthropogenic climate change stability over short-, medium- and long-term (Williams et al., 2005; Hole et al., 2009; Shoo et climatic cycles (Fjeldså, 1994; Fjeldså et al., al., 2011). 1997). Here, we show that sites occupied by From simple envelope approaches, which rare species may be exposed to lower rates of employ concise rule-sets to define species future warming, drying and increased seasonal- range boundaries (e.g., BIOCLIM, Busby, 1991; ity than those occupied by eligible species. HABITAT, Walker & Cocks, 1991; GARP, These results are consistent with the theory Stockwell & Peters, 1999), more complex mod- that long-term climatic stability, combined with elling procedures have emerged (e.g., MARS geographic isolation gives rise to, and may and BRT, Friedman, 1991, 2001; MaxEnt, continue supporting, high rates of rarity and Phillips & Dudik, 2008). Due to their superior endemism (Sandel et al., 2011; Platts et al., capacity to fit the observed species data, 2013). At sub-1° scales, climate change veloci- housed within user-friendly desktop applica- ties in mountainous regions are expected to be tions, such advances have been widely advocat- lower still (Loarie et al., 2009), buffering the ed (Elith et al., 2006; Franklin, 2010). However, effects of broader scale climatic change (Sandel as the authors of these tools are the first to

Omission of narrow-ranging species | 99

et al., 2011). Yet, some of the highest elevation biotic interaction or adaptive potential – all taxa may still run out of climate space even if important for species persistence (Hof et al., dispersal speeds are viable (Williams et al., 2010) – this remains the default resolution for 2007; Colwell et al., 2008). This risk of disap- climate change impact studies at continental pearing climates is greater for rare species, scales, especially in highly biodiverse but acute- which are endemic to a narrower range of cli- ly data-deficient regions like Africa. Given the matic conditions (less tolerant) than eligible highly complex and scale-dependent nature of species. At any elevation, the probability of rare rare species’ persistence under changing cli- species persistence may be lower than in the mate, we conclude that condemning sites to past, due to anthropogenic fragmentation of high rates of extinction, based on correlative previously connected habitats (Early & Sax, SDM at continental scales, has potential to seri- 2011; Bennie et al., 2013; Pyron & Wiens, ously undermine ongoing conservation efforts 2013). in existing, expert-derived priorities for inter- Together, these results suggest that large- national conservation investment, and that the scale SDM may be calibrated using an unrepre- focus on these sites ought to be maintained sentative sample of both current and future (Ricketts et al., 2005; Hodgson et al., 2009; (and likely past) environmental conditions, and Iwamura et al., 2013). that this bias is counter to more traditional Conclusions drawn from SDM are clearly frameworks for identifying priority sites for sensitive to the spatial scale of analysis (Trivedi conservation, which focus precisely on those et al., 2008). Thus, the issues discussed here taxa most likely to be omitted from SDM. Con- could potentially be mitigated by the progres- servation priorities derived empirically from sion of climate change scenarios to horizontal the rare species set using simple climate enve- grid resolutions of tens, rather than hundreds lopes exhibit higher congruence with existing of kilometres under IPCC-AR5, or to even finer priority schemes than those inferred using scales thereafter (Platts et al., 2013). However, correlative SDM. This said, under current cli- much of the available species data, especially mate, SDM on the eligible species set does a for the tropics, are not reliably georeferenced reasonable job of identifying existing conserva- to sub-1° resolutions, prohibiting robust analy- tion priorities, except for the South African sis at finer scales (Graham et al., 2004; Wiens & Fynbos and the mountains of the Drakensberg, Bachelet, 2010; Feeley & Silman, 2011). In cas- which are identified only by the minimum sets es where high resolution species data are avail- metric. Importantly, however, when using SDM able, a disproportionate percentage of threat- to infer conservation priorities under future ened species are still omitted from modelling climate, congruence with existing schemes procedures. In the Eastern Arc Mountains, for generally decreases. The Eastern Arc Moun- example, 66% of amphibian species have ten or tains in Tanzania and Kenya, for example, are fewer records on a ten arc-minute grid correctly identified as priority sites under pre- (Burgess et al., 2007). sent climate, but are omitted entirely under Moreover, our findings are unlikely to be future climate, despite substantial evidence unique to this taxonomic group. The same da- that these ecosystems have historically been tabase of African vertebrate species used here buffered against broader-scale change contains 29% of snakes, 33% of mammals and (Marchant et al., 2006; Mumbi et al., 2008; 10% of birds with fewer than ten 1° occurrence Finch et al., 2009; Platts et al., 2013). records. In a recent study of climate change While the 1° resolution of our analysis does impacts on sub-Saharan Africa’s vertebrates not capture local variations in microclimate, (Garcia et al., 2012), only 67% of the species in

100 | Chapter III

the database could be assessed using SDM, and future climate space is highest within leaving out those with fewer than 15 records. known centres of endemism, as indicated here As reported here for amphibians, we find that for rare amphibians. regions of importance differ between the omit- Another approach that has application to ted taxa and the modelled species set (Fig. S5). the rare species conundrum is to use simple In a study of all vascular plant species strictly measures of species' exposure to climate endemic to the Eastern Arc Mountains (Platts et change that do not rely on SDM (Ohlemüller et al., 2013), none had sufficient herbarium data al., 2008; Ohlemüller, 2011). These methods for SDM on a 1° grid and, even at 30 arc-second can be combined with biological trait-based resolution (1 km), 90% were too rare to model, assessments of species’ sensitivity (ability to the majority of these species being either as- cope in situ) and adaptive capacity (ability to sessed as Threatened under Red Listing criteria escape by dispersal or evolution) to changing (R. E. Gereau & W. R. Q. Luke, pers. comm., climatic conditions (Dawson et al., 2011). Phys- 2013) or scheduled for Red List assessment by iological, ecological, genetic and functional the Eastern African Plant Red List Authority characteristics, combined with simple (Gereau et al., 2010). For restricted range taxa measures of exposure, are currently being em- that are eligible for SDM, extrapolations beyond ployed to ascribe a measure of climate change dispersal barriers, such as mountains, valleys vulnerability to thousands of species of am- and water bodies, are especially uncertain phibian, bird and coral, for eventual inclusion in (Platts et al., 2010); within known extents of the IUCN Red List Categories and Criteria ver- occurrence, discriminative power may appear sion 9.0 (Foden et al., 2008, 2013; IUCN, 2011). to be good, but model stability can be highly Through methods such as these, conserva- sensitive to errors/changes in the distribution- tion planners may embrace valuable insights al record (Hernandez et al., 2006; Platts et al., provided by popular SDM methodologies, while 2008). addressing their inherent limitations with re- Novel approaches to address these chal- spect to those rare and endangered species that lenges, both within and beyond the field of SDM have traditionally underpinned global alloca- are, however, emerging. Current knowledge on tion of conservation funds. Effective climate rare species distributions could be systemati- change adaptation planning will rely on ad- cally improved by iterative application of sim- vancing our capacity to predict impacts on all ple predictive models and targeted fieldwork species, and not just those which fulfil biologi- (Williams et al., 2009; Platts et al., 2010). Mak- cally arbitrary occurrence thresholds required ing the most of small sample sizes, Lomba et al. by one tool in the box. (2010) suggest combining large numbers of bivariate models, built individually using all pairwise combinations of predictors. In con- Acknowledgements trast to a species-level focus, hierarchical ap- We thank the many taxonomists and amphibian proaches, combining species-specific and com- experts who contributed to the African amphib- munity models (Loarie et al., 2008; Ovaskainen ian database, in particular through the IUCN & Soininen, 2011), or spatial modelling of cli- Species Survival Commission Specialist Group matically associated species pools (Golicher et on African Herpetology. Additional distribution al., 2008) can help to detect shared patterns of maps were provided by A. Schiøtz (tree frogs), response across range-restricted taxa. At the J. Poynton (bufonids), A. Channing (southern ecoregion level, Iwamura et al. (2013) have African), S. Loader (Caecilians), M. Rödel (West recently shown that overlap between current Africa), M. Menegon, K.M. Howell, D. Broadley

Omission of narrow-ranging species | 101

and N. Doggart (Eastern Arc and Coastal For- Watson R. (2010) Global biodiversity: indicators of recent declines. Science, 328, 1164–8. ests). P.J.P. was funded by the Ministry for For- Chessel D., Dufour A.B., & Thioulouse J. (2004) The ade4 eign Affairs of Finland and by the Leverhulme package-I-One-table methods. R News, 4, 5–10. Trust, R.A.G. by the Portuguese Foundation for Colwell R.K., Brehm G., Cardelús C.L., Gilman A.C., & Science and Technology and by the European Longino J.T. (2008) Global warming, elevational Social Fund, W.F. by the IUCN Species Pro- range shifts, and lowland biotic attrition in the wet tropics. Science, 322, 258–261. gramme, C.H. by LOEWE and N.D.B. by the Dawson T.P., Jackson S.T., House J.I., Prentice I.C., & Mace WWF-US Conservation Science Programme. G.M. (2011) Beyond predictions: biodiversity N.D.B., R.A.G., C.H. and C.R. acknowledge the conservation in a changing climate. Science, 332, 53– support of the Danish National Research Foun- 8. dation. Dolédec S., Chesse D., & Gimaret-Carpentier C. (2000) Niche separation in community analysis: a new method. Ecology, 81, 2914–2927. Early R. & Sax D.F. (2011) Analysis of climate paths References reveals potential limitations on species range shifts. Ecology Letters, 14, 1125–33. Ahrends A., Rahbek C., Bulling M.T., Burgess N.D., Platts Elith J., H Graham C., P Anderson R., Dudík M., Ferrier S., P.J., Lovett J.C., Kindemba V.W., Owen N., Sallu A.N., & Guisan A., J Hijmans R., Huettmann F., R Leathwick J., Marshall A.R. (2011) Conservation and the botanist Lehmann A., Li J., G Lohmann L., A Loiselle B., Manion effect. Biological Conservation, 144, 131–140. G., Moritz C., Nakamura M., Nakazawa Y., McC M Araújo M.B. & Peterson A.T. (2012) Uses and misuses of Overton J., Townsend Peterson A., J Phillips S., bioclimatic envelope modeling. Ecological Society of Richardson K., Scachetti-Pereira R., E Schapire R., America, 93, 1527–1539. Soberón J., Williams S., S Wisz M., & E Zimmermann N. (2006) Novel methods improve prediction of Balmford A. & Gaston K.J. (1999) Why biodiversity species’ distributions from occurrence data. surveys are good value. Nature, 398, 204–205. Ecography, 29, 129–151. Bateman B.L., Murphy H.T., Reside A.E., Mokany K., & Farr T.G., Rosen P.A., Caro E., Crippen R., Duren R., VanDerWal J. (2013) Appropriateness of full-, partial- Hensley S., Kobrick M., Paller M., Rodriguez E., Roth and no-dispersal scenarios in climate change impact L., Seal D., Shaffer S., Shimada J., Umland J., Werner modelling. Diversity and Distributions, 19, 1224– M., Oskin M., Burbank D., & Alsdorf D. (2007) The 1234. shuttle radar topography mission. Reviews of Bennie J., Hodgson J.A., Lawson C.R., Holloway C.T.R., Roy Geophysics, 45, RG2004. D.B., Brereton T., Thomas C.D., & Wilson R.J. (2013) Feeley K.J. & Silman M.R. (2011) The data void in Range expansion through fragmented landscapes modeling current and future distributions of tropical under a variable climate. Ecology Letters, 16, 921– species. Global Change Biology, 17, 626–630. 929. Finch J., Leng M.J., & Marchant R. (2009) Late Quaternary Burgess N., Butynski T., Cordeiro N., Doggart N., Fjeldsa vegetation dynamics in a biodiversity hotspot, the J., Howell K., Kilahama F., Loader S., Lovett J., & Uluguru Mountains of Tanzania. Quaternary Mbilinyi B. (2007) The biological importance of the Research, 72, 111–122. Eastern Arc Mountains of Tanzania and Kenya. Biological Conservation, 134, 209–231. Fjeldså J. (1994) Geographical patterns for relict and young species of birds in Africa and South-America Busby J.R. (1991) BIOCLIM – a bioclimatic analysis and and implications for conservation priorities. prediction system. Nature Conservation: Cost Effective Biodiversity and Conservation, 3, 207–226. Biological Surveys and Data Analysis (ed. by C.R. Margules and M.P. Austin), pp. 64–68. CSIRO, East Fjeldså J., Ehrlich D., Lambin E., & Prins E. (1997) Are Melbourne, Australia. biodiversity `hotspots’ correlated with current ecoclimatic stability? A pilot study using the NOAA- Butchart S.H.M., Walpole M., Collen B., van Strien A., AVHRR remote sensing data. Biodiversity and Scharlemann J.P.W., Almond R.E. a, Baillie J.E.M., Conservation, 6, 401–422. Bomhard B., Brown C., Bruno J., Carpenter K.E., Carr G.M., Chanson J., Chenery A.M., Csirke J., Davidson Foden W.B., Butchart S.H.M., Stuart S.N., Vié J.-C., N.C., Dentener F., Foster M., Galli A., Galloway J.N., Akçakaya H.R., Angulo A., DeVantier L.M., Gutsche A., Genovesi P., Gregory R.D., Hockings M., Kapos V., Turak E., Cao L., Donner S.D., Katariya V., Bernard R., Lamarque J.-F., Leverington F., Loh J., McGeoch M. a, Holland R. a., Hughes A.F., O’Hanlon S.E., Garnett S.T., McRae L., Minasyan A., Hernández Morcillo M., Şekercioğlu Ç.H., & Mace G.M. (2013) Identifying the Oldfield T.E.E., Pauly D., Quader S., Revenga C., Sauer World’s Most Climate Change Vulnerable Species: A J.R., Skolnik B., Spear D., Stanwell-Smith D., Stuart Systematic Trait-Based Assessment of all Birds, S.N., Symes A., Tierney M., Tyrrell T.D., Vié J.-C., & Amphibians and Corals. PLoS ONE, 8, e65427.

102 | Chapter III

Foden W.B., Mace G.M., Vié J.-C., Angulo A., Butchart Zoological Museum, University of Copenhagen, S.H.M., DeVantier L., Dublin H.T., Gutsche A., Stuart Copenhagen. S.N., & Turak E. (2008) Species susceptibility to Harrell F.E., Lee K.L., Califf R.M., Pryor D.B., & Rosati R. a climate change impacts. Wildlife in a Changing World: (1984) Regression modelling strategies for improved an analysis of the 2008 IUCN Red List of Threatened prognostic prediction. Statistics in Medicine, 3, 143– Species (ed. by J.-C. Vié, C. Hilton-Taylor, and S.N. 52. Stuart), pp. 77–88. IUCN, Gland. Hernandez P.A., Graham C.H., Master L.L., Albert D.L., & Da Fonseca G.A.B., Balmford A., Bibby C., Boitani L., Corsi The A.D.L. (2006) The effect of sample size and F., Brooks T., Gascon C., Olivieri S., Mittermeier R.A., species characteristics on performance of different Burgess N., Dinerstein E., Olson D., Hannah L., Lovett species distribution modeling methods. Ecography, J., Moyer D., Rahbek C., Stuart S., & Williams P. (2000) 29, 773–785. … following Africa’s lead in setting priorities. Nature, 405, 393–394. Hodgson J.A., Thomas C.D., Wintle B.A., & Moilanen A. (2009) Climate change, connectivity and Franklin J. (2010) Mapping Species Distributions. conservation decision making: back to basics. Journal Cambridge University Press, Cambridge. of Applied Ecology, 46, 964–969. Franklin J. (2013) Species distribution models in Hof C., Araújo M.B., Jetz W., & Rahbek C. (2011a) conservation biogeography: developments and Additive threats from pathogens, climate and land- challenges. Diversity and Distributions, 19, 1217– use change for global amphibian diversity. Nature, 1223. 480, 516–9. Friedman J.H. (1991) Multivariate Adaptive Regression Hof C., Levinsky I., Araújo M.B., & Rahbek C. (2011b) Splines. Annals of Statistics, 19, 1–67. Rethinking species’ ability to cope with rapid climate Friedman J.H. (2001) Greedy function approximation: A change. Global Change Biology, 17, 2987–2990. gradient boosting machine. Annals of Statistics, 29, Hof C., Rahbek C., & Araújo M.B. (2010) Phylogenetic 1189–1232. signals in the climatic niches of the world’s Frost D.R., Grant T., Faivovich N., Bain R.H., Haas A., amphibians. Ecography, 33, 242–250. Haddad C.F.B., de Sa R.O., Channing A., Wilkinson M., Hole D.G., Willis S.G., Pain D.J., Fishpool L.D., Butchart Donnellan S.C., Raxworthy C.J., Campbell J.A., Blotto S.H.M., Collingham Y.C., Rahbek C., & Huntley B. B.L., Moler P., Drewes R.C., Nussbaum R.A., Lynch J.D., (2009) Projected impacts of climate change on a Green D.M., & Wheeler W.C. (2006) The amphibian continent-wide protected area network. Ecology tree of life. Bulletin of the American Museum of Letters, 12, 420–31. Natural History, 297, 1–370. Huntley B., Collingham Y.C., Green R.E., Hilton G.M., Garcia R. a., Burgess N.D., Cabeza M., Rahbek C., & Araújo Rahbek C., & Willis S.G. (2006) Potential impacts of M.B. (2012) Exploring consensus in 21st century climatic change upon geographical distributions of projections of climatically suitable areas for African birds. Ibis, 148, 8–28. vertebrates. Global Change Biology, 18, 1253–1269. IUCN (2011) Guidelines for Using the IUCN Red List Gereau R.E., Taylor C.M., Bodine S., & Kindeketa W.J. Categories and Criteria. Version 9.0. Prepared by the (2010) Plant Conservation Assessment in the Eastern IUCN Standards and Petitions Subcommittee. . Arc Mountains and Coastal Forests of Tanzania and Kenya. Available at Iwamura T., Guisan A., Wilson K. a., & Possingham H.P. http://www.mobot.org/MOBOT/Research/tanzania/ (2013) How robust are global conservation priorities cepf.shtml/. . to climate change? Global Environmental Change, In press. Golicher D.J., Cayuela L., Alkemade J.R.M., González- Espinosa M., & Ramírez-Marcial N. (2008) Applying Kapos V., Rhind J., Edwards M., Price M.F., & Ravilious C. climatically associated species pools to the modelling (2000) Developing a map of the world’s mountain of compositional change in tropical montane forests. forests. Forests in sustainable mountain development: Global Ecology and Biogeography, 17, 262–273. a state of knowledge report for 2000 (ed. by M.F. Price and N. Butt), pp. 4–9. CAB International, Wallingford. Graham C.H., Ferrier S., Huettman F., Moritz C., & Peterson a T. (2004) New developments in museum- De Klerk H.M., Crowe T.M., Fjeldså J., & Burgess N.D. based informatics and applications in biodiversity (2002) Patterns of species richness and narrow analysis. Trends in Ecology & Evolution, 19, 497–503. endemism of terrestrial bird species in the Afrotropical region. Journal of Zoology, 256, 327–342. Hannah L., Midgley G.F., Lovejoy T., Bond W.J., Bush M., Lovett J.C., Scott D., & Woodward F.I. (2002) Küper W., Sommer J.H., Lovett J.C., & Barthlott W. (2006) Conservation of biodiversity in a changing climate. Deficiency in African plant distribution data – Conservation Biology, 16, 264–268. missing pieces of the puzzle. Botanical Journal of the Linnean Society, 150, 355–368. Hansen, L.A., Burgess, N.D., Fjeldså, J., Rahbek C. (2007) One degree resolution databases of the distribution of Liu C., White M., & Newell G. (2013) Selecting thresholds 739 amphibians in sub-Saharan Africa (version 1.0). for the prediction of species occurrence with

Omission of narrow-ranging species | 103

presence-only data. Journal of Biogeography, 40, Phillips S.J. & Dudik M. (2008) Modeling of species 778–789. distributions with Maxent: new extensions and a comprehensive evaluation. Ecography, 31, 161–175. Loarie S.R., Carter B.E., Hayhoe K., McMahon S., Moe R., Knight C. a, & Ackerly D.D. (2008) Climate change and Pimm S.L., Ayres M., Balmford A., Branch G., Brandon K., the future of California’s endemic flora. PloS ONE, 3, Brooks T., Bustamante R., Costanza R., Cowling R., e2502. Curran L.M., Dobson A., Farber S., Fonseca G.A.B., Gascon C., Kitching R., Mcneely J., Lovejoy T., Loarie S.R., Duffy P.B., Hamilton H., Asner G.P., Field C.B., Mittermeier R.A., Myers N., Patz J.A., Raffle B., & Ackerly D.D. (2009) The velocity of climate change. Rapport D., Raven P., Roberts C., Rodríguez J.P., Nature, 462, 1052–5. Rylands A.B., Tucker C., Safina C., Samper C., Stiassny Lomba A., Pellissier L., Randin C., Vicente J., Moreira F., M.L.J., Supriatna J., Wall D.H., & Wilcove D. (2001) Can Honrado J., & Guisan A. (2010) Overcoming the rare we defy nature’s end? Science, 293, 2207–2208. species modelling paradox: a novel hierarchical Platts P.J., Ahrends A., Gereau R.E., McClean C.J., Lovett framework applied to an Iberian endemic plant. J.C., Marshall A.R., Pellikka P.K.E., Mulligan M., Biological Conservation, 143, 2647–2657. Fanning E., & Marchant R. (2010) Can distribution Marchant R., Mumbi C., Behera S., & Yamagata T. (2006) models help refine inventory-based estimates of The Indian Ocean dipole – the unsung driver of conservation priority? A case study in the Eastern climatic variability in East Africa. African Journal of Arc forests of Tanzania and Kenya. Diversity and Ecology, 45, 4–16. Distributions, 16, 628–642. Margules C.R. & Pressey R.L. (2000) Systematic Platts P.J., Gereau R.E., Burgess N.D., & Marchant R. conservation planning. Nature, 405, 243–253. (2013) Spatial heterogeneity of climate change in an McClean C.J., Lovett J.C., Küper W., Hannah L., Sommer Afromontane centre of endemism. Ecography, 36, J.H., Barthlott W., Termansen M., Smith G.F., 518–530. Tokumine S., & Taplin J.R.D. (2005) African plant Platts P.J., McClean C.J., Lovett J.C., & Marchant R. (2008) diversity and climate change. Annuals of the Missouri Predicting tree distributions in an East African Botanical Garden, 92, 139–152. biodiversity hotspot : model selection, data bias and Meehl G. a., Covey C., Taylor K.E., Delworth T., Stouffer envelope uncertainty. Ecological Modelling, 218, R.J., Latif M., McAvaney B., & Mitchell J.F.B. (2007) 121–134. THE WCRP CMIP3 Multimodel Dataset: A New Era in Pressey R.L., Hager T.C., Ryan K.M., Schwarz J., Wall S., Climate Change Research. Bulletin of the American Ferrier S., & Creaser P.M. (2000) Using abiotic data Meteorological Society, 88, 1383–1394. for conservation assessments over extensive regions: Mittermeier R.A., Gil P.R., Hoffmann M., Pilgrim J., Brooks quantitative methods applied across New South T., Mittermeier C.G., Lamoreux J., & da Fonseca G.A.B. Wales, Australia. Biological Conservation, 96, 55–82. (2004) Hotspots Revisited: Earth’s Biologically Richest Pyron R.A. & Wiens J.J. (2013) Large-scale phylogenetic and Most Endangered Ecoregions. CEMEX, Mexico analyses reveal the causes of high tropical amphibian City. diversity. Proceedings of the Royal Society B, 280, Mumbi C.T., Marchant R., Hooghiemstra H., & Wooller 1471–2954. M.J. (2008) Late Quaternary vegetation R-Core-Team (2012) R: A language and environment for reconstruction from the Eastern Arc Mountains, statistical computing. R Foundation for Statistical Tanzania. Quaternary Research, 69, 326–341. Computing, Vienna. http://www.r-project.org/. Myers N., Mittermeier R.A., Mittermeier C.G., Da Fonseca Ricketts T.H., Dinerstein E., Boucher T., Brooks T.M., G.A., & Kent J. (2000) Biodiversity hotspots for Butchart S.H.M., Hoffmann M., Lamoreux J.F., conservation priorities. Nature, 403, 853–858. Morrison J., Parr M., Pilgrim J.D., Rodrigues A.S.L., Ohlemüller R. (2011) Running out of climate space. Sechrest W., Wallace G.E., Berlin K., Bielby J., Burgess Science, 334, 613–614. N.D., Church D.R., Cox N., Knox D., Loucks C., Luck G.W., Master L.L., Moore R., Naidoo R., Ridgely R., Ohlemüller R., Anderson B.J., Araújo M.B., Butchart Schatz G.E., Shire G., Strand H., Wettengel W., & S.H.M., Kudrna O., Ridgely R.S., & Thomas C.D. (2008) Wikramanayake E. (2005) Pinpointing and The coincidence of climatic and species rarity: high preventing imminent extinctions. Proceedings of the risk to small-range species from climate change. National Academy of Sciences of the United States of Biology Letters, 4, 568–72. America, 102, 18497–18501. Olson D.M. & Dinerstein E. (1998) The Global 200: a Sandel B., Arge L., Dalsgaard B., Davies R.G., Gaston K.J., representation approach to conserving the Earth’ s Sutherland W.J., & Svenning J.-C. (2011) The most biologically valuable ecoregions. Conservation influence of Late Quaternary climate-change velocity Biology, 12, 502–515. on species endemism. Science, 334, 660–4. Ovaskainen O. & Soininen J. (2011) Making more out of Shoo L.P., Storlie C., Vanderwal J., Little J., & Williams S.E. sparse data: hierarchical modeling of species (2011) Targeted protection and restoration to communities. Ecology, 92, 289–95. conserve tropical biodiversity in a warming world. Global Change Biology, 17, 186–193.

104 | Chapter III

Sodhi N.S., Bickford D., Diesmos A.C., Lee T.M., Koh L.P., Changing World (ed. by G.M. Mace, A. Balmford, and Brook B.W., Sekercioglu C.H., & Bradshaw C.J. a J.R. Ginsberg), pp. 211–249. Cambridge University (2008) Measuring the meltdown: drivers of global Press, Cambridge. amphibian extinction and decline. PloS ONE, 3, Wilson K.A., McBride M.F., Bode M., & Possingham H.P. e1636. (2006) Prioritizing global conservation efforts. Stattersfield, A., Crosby, M. J., Long, A. J. and Wege D.C. Nature, 440, 337–40. (1998) Endemic Bird Areas of the world: priorities for Wisz M.S., Hijmans R.J., Li J., Peterson a. T., Graham C.H., biodiversity conservation. BirdLife International, & Guisan a. (2008) Effects of sample size on the Cambridge. performance of species distribution models. Diversity Stockwell D. & Peters D. (1999) The GARP modelling and Distributions, 14, 763–773. system: problems and solutions to automated spatial prediction. International Journal of Geogrpahical Information Science, 13, 143–158. Stockwell D.R.. & Peterson A.T. (2002) Effects of sample Supporting Information size on accuracy of species distribution models. Ecological Modelling, 148, 1–13. Figure S1. Comparison of Outlying Mean Index Trivedi M.R., Berry P.M., Morecroft M.D., & Dawson T.P. results for eligible versus rare species. Results (2008) Spatial scale affects bioclimate model are shown for the first two OMI axes together, projections of climate change impacts on mountain plants. Global Change Biology, 14, 1089–1103. on each axis individually, for all species, and for Walker P.A. & Cocks K.D. (1991) HABITAT - A procedure the subset of species with statistically signifi- for modeling a disjoint environmental envelope for a cant marginality plant or animal species. Global Ecology and Figure S2. Spatial variation in the dissimilarity Biogeography Letters, 1, 108–118. of future climatic conditions compared with the Walsh R.P.D. & Lawler D.M. (1981) Rainfall seasonality: description, spatial patterns and change through (present-day) calibration range used in MaxEnt time. Weather, 36, 201–208. models. Wiens J.A. & Bachelet D. (2010) Matching the multiple Figure S3. Histograms summarising variable scales of conservation with the multiple scales of contributions in MaxEnt models. climate change. Conservation Biology, 24, 51–62. Figure S4. Estimates of species richness and Wiens J.A., Stralberg D., Jongsomjit D., Howell C. a, & Snyder M. a (2009) Niches, models, and climate range-size rarity based on the eligible species change: assessing the assumptions and uncertainties. set, compared for MaxEnt predictions versus Proceedings of the National Academy of Sciences of the United States of America, 106 Suppl, 19729–36. simple niche envelopes. Figure S5. Conservation priorities, now and in Wiens J.J., Graham C.H., Moen D.S., Smith S. a, & Reeder T.W. (2006) Evolutionary and ecological causes of the future, derived empirically for snakes, the latitudinal diversity gradient in hylid frogs: mammals and birds in sub-Saharan Africa treefrog trees unearth the roots of high tropical diversity. The American Naturalist, 168, 579–96. (adapted from Garcia et al., 2012). Williams J.N., Seo C., Thorne J., Nelson J.K., Erwin S., Table S1. Species of sub-Saharan African am- O’Brien J.M., & Schwartz M.W. (2009) Using species phibians included in the study, including their distribution models to predict new occurrences for range size and status on the IUCN Red List. rare plants. Diversity and Distributions, 15, 565–576. Table S2: Main amphibian data sources. Williams J.W., Jackson S.T., & Kutzbach J.E. (2007) Projected distributions of novel and disappearing Table S3. Summary of IUCN Red List status for climates by 2100 AD. Proceedings of the National all amphibian species included in the study, Academy of Sciences of the United States of America, detailed separately for species eligible or too 104, 5738–42. rare for SDM. Williams P., Hannah L., Andelman S., Midgley G., Araújo M., Hughes G., Manne L., Martinez-Meyer E., & Table S4. Environmental variable loadings and Pearson R. (2005) Planning for Climate Change: variance explained by the two first axes of the Identifying Minimum-Dispersal Corridors for the Cape Proteaceae. Conservation Biology, 19, 1063– Outlying Mean Index Analysis, on present-day 1074. climate and on future climate anomalies. Williams P.H. (1998) Key sites for conservation: area- selection methods for biodiversity. Conservation in a

Omission of narrow-ranging species | 105

Supporting Information

(a) Present-day climate conditions

(a) Future climate conditions

Figure S1 | Comparison of Outlying Mean Index (OMI) results for eligible versus rare species, on present-day conditions (a) and on future climate anomalies (b). Boxplots show the distribution of OMI values (marginality) and tolerance on the two first OMI axes together, and on each axis individually. Results are shown for all species, and for the subset of species with statistically significant OMI values. Significant differences between rare and eligible species are denoted with an asterisk (Wilcoxon signed- rank one-sided tests to examine whether rare species have higher marginality and lower tolerance than eligible species, p < 0.05).

Figure S2 | Spatial variation in the dissimilarity of future climatic conditions compared with the (present-day) calibration range used in MaxEnt models. Areas in red (negative values) have one or more environmental variables that are beyond the calibration range by 2090 (IPCC-AR4 emissions scenario A1B) - predictions in those areas should be treated with caution (Platts et al., 2008; Elith et al., 2010).

106 | Chapter III

Elevational range Mean annual rainfall

Mean annual temperature Rainfallseasonality

Figure S3 | Histograms summarising variable contributions in MaxEnt models (for details of derivation, see Phillips & Dudik, 2008).

Figure S4 | Estimates of species richness and range-size rarity based on the eligible species set, compared for MaxEnt predictions versus simple niche envelopes (MDNE – also possible for rare species). Comparison includes all cells in sub-Saharan Africa.

Omission of narrow-ranging species | 107

Rare species Eligible species Eligible species MDNE (1980) SDM (1980) SDM (2080)

Snakes

Mammals

Birds

Figure S5 | Conservation priorities derived empirically for snakes, mammals and birds in sub-Saharan Africa (adapted from Garcia et al., 2012). Priorities based on the top 100 scoring cells for species richness are shown, derived using simple niche envelopes (1980) for rare species and from an ensemble of SDM (1980 and 2080) for eligible species. Future projections (2080) are for a multi-model climate ensemble under the IPCC-AR4 emissions scenario A1B. Following Garcia et al. (2012), eligibility for SDM is here defined as 15 records on a 1° grid (cf. 11 records in the main article).

108 | Chapter III

Table S1 | Species of sub-Saharan African amphibians included in the study, their IUCN Red List Status (www.iucnredlist.org, accessed June 2011) and range size (on a 1° grid). Red Ran Red Ran Genus and species Family List ge Genus and species Family List ge Acanthixalus sonjae HYPEROLIIDAE NT 5 Hyperolius quadratomaculatus HYPEROLIIDAE DD 1 Acanthixalus spinosus HYPEROLIIDAE LC 68 Hyperolius quinquevittatus HYPEROLIIDAE LC 141 Amietia amieti PYXICEPHALIDAE DD 1 Hyperolius raveni HYPEROLIIDAE DD 1 Amietia angolensis PYXICEPHALIDAE LC 394 Hyperolius reesi HYPEROLIIDAE LC 2 Amietia desaegeri PYXICEPHALIDAE DD 3 Hyperolius rhizophilus HYPEROLIIDAE DD 1 Amietia fuscigula PYXICEPHALIDAE LC 103 Hyperolius rhodesianus HYPEROLIIDAE LC 2 Strongylopus grayii PYXICEPHALIDAE LC 90 Hyperolius riggenbachi HYPEROLIIDAE VU 11 Amietia inyangae PYXICEPHALIDAE EN 2 Hyperolius robustus HYPEROLIIDAE DD 2 Amietia johnstoni PYXICEPHALIDAE EN 1 Hyperolius rubrovermiculatus HYPEROLIIDAE EN 1 Amietia ruwenzorica PYXICEPHALIDAE DD 2 Hyperolius sankuruensis HYPEROLIIDAE DD 1 Amietia vandijki PYXICEPHALIDAE LC 3 Hyperolius schoutedeni HYPEROLIIDAE LC 2 Amietia wittei PYXICEPHALIDAE DD 5 Hyperolius seabrai HYPEROLIIDAE DD 1 Afrixalus aureus HYPEROLIIDAE LC 26 Hyperolius semidiscus HYPEROLIIDAE LC 30 Afrixalus brachycnemis HYPEROLIIDAE LC 64 Hyperolius sheldricki HYPEROLIIDAE LC 2 Afrixalus clarkei HYPEROLIIDAE VU 6 Hyperolius soror HYPEROLIIDAE DD 1 Afrixalus crotalus HYPEROLIIDAE LC 55 Hyperolius spatzi HYPEROLIIDAE NA 25 Afrixalus delicatus HYPEROLIIDAE LC 87 Hyperolius spinigularis HYPEROLIIDAE LC 7 Afrixalus dorsalis HYPEROLIIDAE LC 126 Hyperolius steindachneri HYPEROLIIDAE LC 6 Afrixalus dorsimaculatus HYPEROLIIDAE VU 1 Hyperolius stenodactylus HYPEROLIIDAE DD 1 Afrixalus enseticola HYPEROLIIDAE VU 12 Hyperolius substriatus HYPEROLIIDAE LC 77 Afrixalus equatorialis HYPEROLIIDAE LC 55 Hyperolius swynnertoni HYPEROLIIDAE LC 4 Afrixalus fornasini HYPEROLIIDAE LC 130 Hyperolius sylvaticus HYPEROLIIDAE LC 52 Afrixalus fulvovittatus HYPEROLIIDAE LC 26 Hyperolius tannerorum HYPEROLIIDAE EN 1 Afrixalus knysnae HYPEROLIIDAE EN 4 Hyperolius tornieri HYPEROLIIDAE DD 1 Afrixalus lacteus HYPEROLIIDAE EN 3 Hyperolius torrentis HYPEROLIIDAE EN 2 Afrixalus laevis HYPEROLIIDAE LC 152 Hyperolius tuberculatus HYPEROLIIDAE LC 191 Afrixalus leucostictus HYPEROLIIDAE LC 23 Hyperolius tuberilinguis HYPEROLIIDAE LC 133 Afrixalus lindholmi HYPEROLIIDAE DD 1 Hyperolius veithi HYPEROLIIDAE NA 1 Afrixalus morerei HYPEROLIIDAE VU 1 Hyperolius vilhenai HYPEROLIIDAE DD 1 Afrixalus nigeriensis HYPEROLIIDAE NT 33 Hyperolius viridiflavus HYPEROLIIDAE LC 175 Afrixalus orophilus HYPEROLIIDAE VU 8 Hyperolius viridigulosus HYPEROLIIDAE VU 10 Afrixalus osorioi HYPEROLIIDAE LC 89 Hyperolius viridis HYPEROLIIDAE DD 15 Afrixalus paradorsalis HYPEROLIIDAE LC 74 Hyperolius wermuthi HYPEROLIIDAE NT 15 Afrixalus quadrivittatus HYPEROLIIDAE LC 349 Hyperolius xenorhinus HYPEROLIIDAE DD 1 Afrixalus schneideri HYPEROLIIDAE DD 1 Hyperolius zonatus HYPEROLIIDAE NT 27 Afrixalus septentrionalis HYPEROLIIDAE LC 5 Idiocranium russeli CAECILIIDAE DD 3 Afrixalus spinifrons HYPEROLIIDAE VU 14 Kassina arboricola HYPEROLIIDAE VU 18 Afrixalus stuhlmanni HYPEROLIIDAE LC 9 Kassina cassinoides HYPEROLIIDAE LC 204 Afrixalus sylvaticus HYPEROLIIDAE EN 4 Kassina cochranae HYPEROLIIDAE NT 22 Afrixalus uluguruensis HYPEROLIIDAE EN 8 Kassina fusca HYPEROLIIDAE LC 115 Afrixalus upembae HYPEROLIIDAE DD 2 Kassina kuvangensis HYPEROLIIDAE LC 68 Afrixalus vibekensis HYPEROLIIDAE NT 4 Kassina lamottei HYPEROLIIDAE VU 5 Afrixalus vittiger HYPEROLIIDAE LC 155 Kassina maculata HYPEROLIIDAE LC 116 Afrixalus weidholzi HYPEROLIIDAE LC 239 Kassina maculifer HYPEROLIIDAE LC 95 Afrixalus wittei HYPEROLIIDAE LC 117 Kassina maculosa HYPEROLIIDAE LC 78 Alexteroon hypsiphonus HYPEROLIIDAE LC 69 Kassina mertensi HYPEROLIIDAE DD 77 Alexteroon jynx HYPEROLIIDAE CR 1 Kassina schioetzi HYPEROLIIDAE LC 11 Alexteroon obstetricans HYPEROLIIDAE LC 59 Kassina senegalensis HYPEROLIIDAE LC 111 6 Altiphrynoides malcolmi BUFONIDAE EN 4 Kassina somalica HYPEROLIIDAE LC 132 Amietia dracomontana PYXICEPHALIDAE LC 4 Kassina wazae HYPEROLIIDAE DD 1 Strongylopus hymenopus PYXICEPHALIDAE LC 6 Kassinula wittei HYPEROLIIDAE LC 39 Amietia vertebralis PYXICEPHALIDAE NT 7 Lanzarana largeni PTYCHADENIDAE NT 33 Vandijkophrynus amatolicus BUFONIDAE CR 2 Laurentophryne parkeri BUFONIDAE DD 1 Amietia amieti PYXICEPHALIDAE DD 2 Leptodactylodon ventrimarmoratus ARTHROLEPTIDAE VU 5 Vandijkophrynus angusticeps BUFONIDAE LC 15 Leptodactylodon albiventris ARTHROLEPTIDAE VU 4 Amietophrynus asmarae BUFONIDAE LC 36 Leptodactylodon axillaris ARTHROLEPTIDAE EN 1 Poyntonophrynus beiranus BUFONIDAE NA 25 Leptodactylodon bicolor ARTHROLEPTIDAE VU 7 Amietophrynus blanfordii BUFONIDAE LC 57 Leptodactylodon blanci ARTHROLEPTIDAE DD 1 Amietophrynus brauni BUFONIDAE EN 11 Leptodactylodon boulengeri ARTHROLEPTIDAE VU 10 Amietophrynus buchneri BUFONIDAE DD 4 Leptodactylodon bueanus ARTHROLEPTIDAE VU 1 Amietophrynus camerunensis BUFONIDAE LC 66 Leptodactylodon erythrogaster ARTHROLEPTIDAE EN 2 Amietophrynus chudeaui BUFONIDAE DD 1 Leptodactylodon mertensi ARTHROLEPTIDAE EN 3 Amietophrynus cristiglans BUFONIDAE DD 2 Leptodactylodon ornatus ARTHROLEPTIDAE EN 3 Poyntonophrynus damaranus BUFONIDAE DD 29 Leptodactylodon ovatus ARTHROLEPTIDAE NT 11 Amietophrynus danielae BUFONIDAE DD 1 Leptodactylodon perreti ARTHROLEPTIDAE EN 5

Omission of narrow-ranging species | 109

Red Ran Red Ran Genus and species Family List ge Genus and species Family List ge Amietophrynus funereus BUFONIDAE LC 4 Leptodactylodon polyacanthus ARTHROLEPTIDAE VU 4 Poyntonophrynus damaranus BUFONIDAE DD 96 Leptodactylodon stevarti ARTHROLEPTIDAE EN 2 Poyntonophrynus dombensis BUFONIDAE LC 35 Leptodactylodon wildi ARTHROLEPTIDAE EN 1 Poyntonophrynus fenoulheti BUFONIDAE LC 81 Leptopelis anchietae ARTHROLEPTIDAE LC 30 Amietophrynus fuliginatus BUFONIDAE LC 26 Leptopelis argenteus ARTHROLEPTIDAE LC 9 Amietophrynus funereus BUFONIDAE LC 239 Leptopelis aubryi ARTHROLEPTIDAE LC 109 Vandijkophrynus gariepensis BUFONIDAE LC 76 Leptopelis aubryioides ARTHROLEPTIDAE NA 77 Amietophrynus garmani BUFONIDAE LC 174 Leptopelis barbouri ARTHROLEPTIDAE VU 5 Amietophrynus gracilipes BUFONIDAE LC 78 Leptopelis bequaerti ARTHROLEPTIDAE DD 2 Poyntonophrynus grandisonae BUFONIDAE DD 3 Leptopelis bocagii ARTHROLEPTIDAE LC 341 Amietophrynus gutturalis BUFONIDAE LC 507 Leptopelis boulengeri ARTHROLEPTIDAE LC 98 Poyntonophrynus hoeschi BUFONIDAE LC 23 Leptopelis brevirostris ARTHROLEPTIDAE LC 78 Vandijkophrynus inyangae BUFONIDAE EN 1 Leptopelis broadleyi ARTHROLEPTIDAE LC 40 Poyntonophrynus kavangensis BUFONIDAE LC 27 Leptopelis bufonides ARTHROLEPTIDAE LC 155 Amietophrynus kerinyagae BUFONIDAE LC 55 Leptopelis calcaratus ARTHROLEPTIDAE LC 184 Amietophrynus kisoloensis BUFONIDAE LC 14 Leptopelis christyi ARTHROLEPTIDAE LC 50 Amietophrynus langanoensis BUFONIDAE DD 4 Leptopelis concolor ARTHROLEPTIDAE LC 23 Amietophrynus latifrons BUFONIDAE LC 84 Leptopelis crystallinoron ARTHROLEPTIDAE DD 1 Amietophrynus lemairii BUFONIDAE LC 40 Leptopelis cynnamomeus ARTHROLEPTIDAE LC 108 Mertensophryne lindneri BUFONIDAE LC 41 Leptopelis fenestratus ARTHROLEPTIDAE DD 1 Mertensophryne lonnbergi BUFONIDAE NT 8 Leptopelis fiziensis ARTHROLEPTIDAE DD 2 Poyntonophrynus lughensis BUFONIDAE LC 88 Leptopelis flavomaculatus ARTHROLEPTIDAE LC 101 Amietophrynus maculatus BUFONIDAE LC 421 Leptopelis gramineus ARTHROLEPTIDAE LC 28 Bufo xeros BUFONIDAE LC 7 Leptopelis jordani ARTHROLEPTIDAE DD 1 Mertensophryne melanopleura BUFONIDAE LC 4 Leptopelis karissimbensis ARTHROLEPTIDAE EN 2 Mertensophryne mocquardi BUFONIDAE DD 2 Leptopelis kivuensis ARTHROLEPTIDAE NT 13 Mertensophryne nairobiensis BUFONIDAE DD 3 Leptopelis lebeaui ARTHROLEPTIDAE DD 3 Mertensophryne nyikae BUFONIDAE VU 1 Leptopelis mackayi ARTHROLEPTIDAE DD 1 Amietophrynus pantherinus BUFONIDAE EN 3 Leptopelis macrotis ARTHROLEPTIDAE NT 43 Amietophrynus pardalis BUFONIDAE LC 20 Leptopelis marginatus ARTHROLEPTIDAE DD 1 Poyntonophrynus parkeri BUFONIDAE LC 17 Leptopelis millsoni ARTHROLEPTIDAE LC 189 Bufo pentoni BUFONIDAE LC 158 Leptopelis modestus ARTHROLEPTIDAE LC 35 Amietophrynus perreti BUFONIDAE VU 1 Leptopelis mossambicus ARTHROLEPTIDAE LC 64 Amietophrynus poweri BUFONIDAE LC 106 Leptopelis natalensis ARTHROLEPTIDAE LC 13 Amietophrynus rangeri BUFONIDAE LC 114 Leptopelis nordequatorialis ARTHROLEPTIDAE LC 14 Amietophrynus reesi BUFONIDAE DD 1 Leptopelis notatus ARTHROLEPTIDAE LC 176 Amietophrynus regularis BUFONIDAE LC 641 Leptopelis occidentalis ARTHROLEPTIDAE NT 30 Vandijkophrynus robinsoni BUFONIDAE LC 10 Leptopelis ocellatus ARTHROLEPTIDAE LC 116 Mertensophryne schmidti BUFONIDAE DD 1 Leptopelis oryi ARTHROLEPTIDAE LC 9 Amietophrynus steindachneri BUFONIDAE LC 40 Leptopelis parbocagii ARTHROLEPTIDAE LC 67 Amietophrynus superciliaris BUFONIDAE LC 78 Leptopelis parkeri ARTHROLEPTIDAE VU 9 Amietophrynus superciliaris BUFONIDAE LC 9 Leptopelis parvus ARTHROLEPTIDAE DD 3 Amietophrynus superciliaris BUFONIDAE LC 24 Leptopelis ragazzii ARTHROLEPTIDAE VU 23 Amietophrynus taiensis BUFONIDAE CR 1 Leptopelis rufus ARTHROLEPTIDAE LC 46 Mertensophryne taitana BUFONIDAE LC 77 Leptopelis spiritusnoctis ARTHROLEPTIDAE LC 75 Amietophrynus togoensis BUFONIDAE NT 35 Leptopelis susanae ARTHROLEPTIDAE EN 1 Amietophrynus tuberosus BUFONIDAE LC 101 Leptopelis uluguruensis ARTHROLEPTIDAE VU 9 Amietophrynus turkanae BUFONIDAE DD 2 Leptopelis vannutellii ARTHROLEPTIDAE VU 9 Amietophrynus urunguensis BUFONIDAE DD 1 Leptopelis vermiculatus ARTHROLEPTIDAE VU 10 Mertensophryne uzunguensis BUFONIDAE VU 5 Leptopelis viridis ARTHROLEPTIDAE LC 280 Poyntonophrynus vertebralis BUFONIDAE LC 43 Leptopelis xenodactylus ARTHROLEPTIDAE EN 3 Amietophrynus villiersi BUFONIDAE EN 4 Leptopelis yaldeni ARTHROLEPTIDAE NT 7 Amietophrynus vittatus BUFONIDAE DD 8 Leptopelis zebra ARTHROLEPTIDAE NT 10 Amietophrynus xeros BUFONIDAE LC 350 Mertensophryne micranotis BUFONIDAE LC 19 Hylarana albolabris RANIDAE LC 358 Microbatrachella capensis PYXICEPHALIDAE CR 3 Hylarana amnicola RANIDAE LC 38 Morerella cyanophthalma HYPEROLIIDAE VU 1 Hylarana asperrima RANIDAE EN 5 Natalobatrachus bonebergi PYXICEPHALIDAE EN 8 Hylarana darlingi RANIDAE LC 106 Nectophryne afra BUFONIDAE LC 61 Hylarana fonensis RANIDAE DD 5 Nectophryne batesii BUFONIDAE LC 37 Hylarana galamensis RANIDAE LC 318 Nectophrynoides asperginis BUFONIDAE EW 1 Hylarana lemairei RANIDAE LC 42 Nectophrynoides cryptus BUFONIDAE EN 2 Hylarana lepus RANIDAE LC 71 Nectophrynoides frontierei BUFONIDAE DD 1 Hylarana longipes RANIDAE VU 9 Nectophrynoides laevis BUFONIDAE DD 1 Hylarana occidentalis RANIDAE EN 5 Nectophrynoides laticeps BUFONIDAE EN 1 Hylarana parkeriana RANIDAE DD 3 Nectophrynoides minutus BUFONIDAE EN 3 Anhydrophryne rattrayi PYXICEPHALIDAE EN 4 Nectophrynoides paulae BUFONIDAE CR 1 Arlequinus krebsi HYPEROLIIDAE EN 2 Nectophrynoides poyntoni BUFONIDAE CR 1 Arthroleptella bicolor PYXICEPHALIDAE LC 2 Nectophrynoides pseudotornieri BUFONIDAE EN 1

110 | Chapter III

Red Ran Red Ran Genus and species Family List ge Genus and species Family List ge Arthroleptella drewesii PYXICEPHALIDAE DD 1 Nectophrynoides tornieri BUFONIDAE LC 14 Anhydrophryne hewitti PYXICEPHALIDAE LC 12 Nectophrynoides vestergaardi BUFONIDAE EN 1 Arthroleptella landdrosia PYXICEPHALIDAE NT 4 Nectophrynoides viviparus BUFONIDAE VU 10 Arthroleptella lightfooti PYXICEPHALIDAE NT 2 Nectophrynoides wendyae BUFONIDAE CR 1 Anhydrophryne ngongoniensis PYXICEPHALIDAE EN 3 Petropedetes yakusini PETROPEDETIDAE EN 1 Arthroleptella rugosa PYXICEPHALIDAE CR 1 Nimbaphrynoides liberiensis BUFONIDAE CR 1 Arthroleptella subvoce PYXICEPHALIDAE DD 1 Nimbaphrynoides occidentalis BUFONIDAE CR 1 Arthroleptella villiersi PYXICEPHALIDAE LC 4 Nothophryne broadleyi PYXICEPHALIDAE EN 4 Petropedetes dutoiti PETROPEDETIDAE CR 2 Nyctibates corrugatus ARTHROLEPTIDAE LC 18 Petropedetes martiensseni PETROPEDETIDAE EN 2 Opisthothylax immaculatus HYPEROLIIDAE LC 86 Petropedetes yakusini PETROPEDETIDAE EN 5 Paracassina kounhiensis HYPEROLIIDAE LC 11 Arthroleptis adelphus ARTHROLEPTIDAE LC 70 Paracassina obscura HYPEROLIIDAE LC 26 Arthroleptis adolfifriederici ARTHROLEPTIDAE LC 2 Parhoplophryne usambarica MICROHYLIDAE CR 1 Arthroleptis affinis ARTHROLEPTIDAE LC 11 Petropedetes cameronensis PETROPEDETIDAE NT 4 Arthroleptis brevipes ARTHROLEPTIDAE DD 1 Petropedetes euskircheni PETROPEDETIDAE NA 1 Arthroleptis carquejai ARTHROLEPTIDAE DD 1 Petropedetes johnstoni PETROPEDETIDAE NT 6 Arthroleptis fichika ARTHROLEPTIDAE EN 1 Petropedetes juliawurstnerae PETROPEDETIDAE NA 1 Arthroleptis francei ARTHROLEPTIDAE EN 1 Petropedetes natator PETROPEDETIDAE NT 12 Arthroleptis kidogo ARTHROLEPTIDAE NA 1 Petropedetes newtoni PETROPEDETIDAE LC 37 Arthroleptis krokosua ARTHROLEPTIDAE EN 1 Petropedetes palmipes PETROPEDETIDAE EN 4 Arthroleptis langeri ARTHROLEPTIDAE NA 1 Petropedetes parkeri PETROPEDETIDAE LC 25 Arthroleptis lonnbergi ARTHROLEPTIDAE DD 5 Petropedetes perreti PETROPEDETIDAE EN 3 Arthroleptis nguruensis ARTHROLEPTIDAE NA 1 Petropedetes vulpiae PETROPEDETIDAE NA 9 Arthroleptis nikeae ARTHROLEPTIDAE EN 1 Phlyctimantis boulengeri HYPEROLIIDAE LC 26 Arthroleptis nlonakoensis PHRYNOBATRACHIDAE DD 1 Phlyctimantis boulengeri HYPEROLIIDAE LC 20 Arthroleptis_perreti ARTHROLEPTIDAE NA 3 Phlyctimantis keithae HYPEROLIIDAE VU 1 Arthroleptis poecilonotus ARTHROLEPTIDAE LC 138 Phlyctimantis leonardi HYPEROLIIDAE LC 48 Arthroleptis reichei ARTHROLEPTIDAE NT 6 Phlyctimantis verrucosus HYPEROLIIDAE LC 61 Arthroleptis stenodactylus ARTHROLEPTIDAE LC 223 Phrynobatrachus latifrons PHRYNOBATRACHIDAE LC 141 Arthroleptis stridens ARTHROLEPTIDAE DD 1 Phrynobatrachus acridoides PHRYNOBATRACHIDAE LC 128 Arthroleptis tanneri ARTHROLEPTIDAE VU 4 Phrynobatrachus acutirostris PHRYNOBATRACHIDAE VU 3 Arthroleptis tuberosus ARTHROLEPTIDAE DD 5 Phrynobatrachus africanus PHRYNOBATRACHIDAE LC 88 Arthroleptis variabilis ARTHROLEPTIDAE LC 115 Phrynobatrachus albolabris PHRYNOBATRACHIDAE DD 1 Arthroleptis wahlbergii ARTHROLEPTIDAE LC 13 Phrynobatrachus albomarginatus PHRYNOBATRACHIDAE DD 1 Astylosternus batesi ARTHROLEPTIDAE LC 72 Phrynobatrachus alleni PHRYNOBATRACHIDAE NT 40 Astylosternus diadematus ARTHROLEPTIDAE VU 7 Phrynobatrachus annulatus PHRYNOBATRACHIDAE EN 4 Astylosternus fallax ARTHROLEPTIDAE EN 3 Phrynobatrachus anotis PHRYNOBATRACHIDAE DD 5 Astylosternus laurenti ARTHROLEPTIDAE EN 4 Phrynobatrachus asper PHRYNOBATRACHIDAE DD 2 Astylosternus montanus ARTHROLEPTIDAE NT 16 Phrynobatrachus auritus PHRYNOBATRACHIDAE LC 146 Astylosternus nganhanus ARTHROLEPTIDAE CR 1 Phrynobatrachus batesii PHRYNOBATRACHIDAE LC 35 Astylosternus occidentalis ARTHROLEPTIDAE LC 29 Phrynobatrachus bequaerti PHRYNOBATRACHIDAE VU 6 Astylosternus perreti ARTHROLEPTIDAE EN 4 Phrynobatrachus breviceps PHRYNOBATRACHIDAE DD 1 Astylosternus ranoides ARTHROLEPTIDAE EN 4 Phrynobatrachus brevipalmatus PHRYNOBATRACHIDAE DD 1 Astylosternus rheophilus ARTHROLEPTIDAE VU 8 Phrynobatrachus ogoensis PHRYNOBATRACHIDAE DD 1 Astylosternus schioetzi ARTHROLEPTIDAE EN 1 Phrynobatrachus bullans PHRYNOBATRACHIDAE LC 29 Astylosternus sp nov ARTHROLEPTIDAE NA 1 Phrynobatrachus calcaratus PHRYNOBATRACHIDAE LC 138 Aubria masako PYXICEPHALIDAE LC 24 Phrynobatrachus chukuchuku PHRYNOBATRACHIDAE NA 1 Aubria occidentalis PYXICEPHALIDAE LC 44 Phrynobatrachus cornutus PHRYNOBATRACHIDAE LC 61 Aubria subsigillata PYXICEPHALIDAE LC 21 Phrynobatrachus cricogaster PHRYNOBATRACHIDAE VU 5 Balebreviceps hillmani BREVICIPITIDAE EN 1 Phrynobatrachus cryptotis PHRYNOBATRACHIDAE DD 5 Boulengerula boulengeri CAECILIIDAE LC 2 Phrynobatrachus dalcqi PHRYNOBATRACHIDAE DD 1 Boulengerula changamwensis CAECILIIDAE DD 3 Phrynobatrachus dendrobates PHRYNOBATRACHIDAE LC 11 Boulengerula denhardti CAECILIIDAE DD 2 Phrynobatrachus elberti PHRYNOBATRACHIDAE DD 1 Boulengerula fischeri CAECILIIDAE DD 1 Phrynobatrachus francisci PHRYNOBATRACHIDAE LC 192 Boulengerula niedeni CAECILIIDAE CR 1 Phrynobatrachus fraterculus PHRYNOBATRACHIDAE LC 9 Boulengerula taitana CAECILIIDAE DD 1 Phrynobatrachus gastoni PHRYNOBATRACHIDAE DD 1 Boulengerula uluguruensis CAECILIIDAE LC 3 Phrynobatrachus ghanensis PHRYNOBATRACHIDAE EN 7 Breviceps acutirostris BREVICIPITIDAE LC 9 Phrynobatrachus giorgii PHRYNOBATRACHIDAE DD 1 Breviceps adspersus BREVICIPITIDAE LC 276 Phrynobatrachus graueri PHRYNOBATRACHIDAE LC 12 Breviceps bagginsi BREVICIPITIDAE DD 2 Phrynobatrachus guineensis PHRYNOBATRACHIDAE NT 17 Breviceps fichus BREVICIPITIDAE LC 3 Phrynobatrachus gutturosus PHRYNOBATRACHIDAE LC 27 Breviceps fuscus BREVICIPITIDAE LC 14 Phrynobatrachus hylaios PHRYNOBATRACHIDAE LC 49 Breviceps gibbosus BREVICIPITIDAE VU 4 Phrynobatrachus inexpectatus PHRYNOBATRACHIDAE DD 2 Breviceps macrops BREVICIPITIDAE VU 9 Phrynobatrachus intermedius PHRYNOBATRACHIDAE CR 1 Breviceps verrucosus BREVICIPITIDAE LC 3 Phrynobatrachus irangi PHRYNOBATRACHIDAE EN 2 Breviceps montanus BREVICIPITIDAE LC 11 Phrynobatrachus kakamikro PHRYNOBATRACHIDAE NA 1 Breviceps mossambicus BREVICIPITIDAE LC 208 Phrynobatrachus keniensis PHRYNOBATRACHIDAE LC 5 Breviceps namaquensis BREVICIPITIDAE LC 12 Phrynobatrachus kinangopensis PHRYNOBATRACHIDAE LC 3

Omission of narrow-ranging species | 111

Red Ran Red Ran Genus and species Family List ge Genus and species Family List ge Breviceps poweri BREVICIPITIDAE LC 111 Phrynobatrachus krefftii PHRYNOBATRACHIDAE EN 2 Breviceps rosei BREVICIPITIDAE LC 7 Phrynobatrachus liberiensis PHRYNOBATRACHIDAE NT 33 Breviceps sopranus BREVICIPITIDAE DD 8 Phrynobatrachus mababiensis PHRYNOBATRACHIDAE LC 360 Breviceps sylvestris BREVICIPITIDAE VU 7 Phrynobatrachus maculiventris PHRYNOBATRACHIDAE NA 2 Breviceps verrucosus BREVICIPITIDAE LC 32 Phrynobatrachus manengoubensis PHRYNOBATRACHIDAE DD 1 Cardioglossa leucomystax ARTHROLEPTIDAE LC 75 Phrynobatrachus minutus PHRYNOBATRACHIDAE LC 13 Cardioglossa leucomystax ARTHROLEPTIDAE LC 39 Phrynobatrachus nanus PHRYNOBATRACHIDAE DD 1 Cacosternum boettgeri PYXICEPHALIDAE LC 254 Phrynobatrachus natalensis PHRYNOBATRACHIDAE LC 862 Cacosternum capense PYXICEPHALIDAE VU 4 Phrynobatrachus ogoensis PHRYNOBATRACHIDAE DD 1 Cacosternum karooicum PYXICEPHALIDAE DD 7 Phrynobatrachus pallidus PHRYNOBATRACHIDAE LC 3 Cacosternum kinangopensis PYXICEPHALIDAE LC 1 Phrynobatrachus parkeri PHRYNOBATRACHIDAE LC 3 Cacosternum leleupi PYXICEPHALIDAE DD 5 Phrynobatrachus parvulus PHRYNOBATRACHIDAE LC 212 Cacosternum namaquense PYXICEPHALIDAE LC 11 Phrynobatrachus perpalmatus PHRYNOBATRACHIDAE LC 51 Cacosternum nanum PYXICEPHALIDAE LC 52 Phrynobatrachus petropedetoides PHRYNOBATRACHIDAE LC 3 Cacosternum parvum PYXICEPHALIDAE LC 12 Phrynobatrachus phyllophilus PHRYNOBATRACHIDAE NT 17 Cacosternum platys PYXICEPHALIDAE LC 14 Phrynobatrachus pintoi PHRYNOBATRACHIDAE DD 1 Cacosternum plimptoni PYXICEPHALIDAE LC 13 Phrynobatrachus plicatus PHRYNOBATRACHIDAE LC 48 Cacosternum poyntoni PYXICEPHALIDAE DD 1 Phrynobatrachus pygmaeus PHRYNOBATRACHIDAE DD 1 Cacosternum sp. 1 PYXICEPHALIDAE NA 21 Phrynobatrachus rouxi PHRYNOBATRACHIDAE DD 1 Cacosternum striatum PYXICEPHALIDAE DD 7 Phrynobatrachus rungwensis PHRYNOBATRACHIDAE LC 15 Callixalus pictus HYPEROLIIDAE VU 5 Phrynobatrachus scapularis PHRYNOBATRACHIDAE LC 8 Callulina dawida BREVICIPITIDAE NA 1 Phrynobatrachus scheffleri PHRYNOBATRACHIDAE LC 30 Callulina kisiwamistu BREVICIPITIDAE EN 2 Phrynobatrachus steindachneri PHRYNOBATRACHIDAE VU 4 Callulina kreffti BREVICIPITIDAE LC 1 Phrynobatrachus sternfeldi PHRYNOBATRACHIDAE DD 1 Callulina laphami BREVICIPITIDAE NA 1 Phrynobatrachus stewartae PHRYNOBATRACHIDAE DD 1 Callulina shengena BREVICIPITIDAE NA 1 Phrynobatrachus sulfureogularis PHRYNOBATRACHIDAE DD 1 Callulina sp1 BREVICIPITIDAE NA 13 Phrynobatrachus taiensis PHRYNOBATRACHIDAE DD 1 Callulina stanleyi BREVICIPITIDAE NA 2 Phrynobatrachus tokba PHRYNOBATRACHIDAE LC 35 Capensibufo rosei BUFONIDAE VU 4 Phrynobatrachus ukingensis PHRYNOBATRACHIDAE DD 16 Capensibufo tradouwi BUFONIDAE LC 7 Phrynobatrachus uzungwensis PHRYNOBATRACHIDAE VU 4 Cardioglossa alsco ARTHROLEPTIDAE CR 1 Phrynobatrachus versicolor PHRYNOBATRACHIDAE VU 5 Cardioglossa aureoli ARTHROLEPTIDAE EN 1 Phrynobatrachus villiersi PHRYNOBATRACHIDAE VU 18 Cardioglossa cyaneospila ARTHROLEPTIDAE DD 5 Phrynobatrachus vogti PHRYNOBATRACHIDAE DD 1 Cardioglossa elegans ARTHROLEPTIDAE LC 24 Phrynobatrachus werneri PHRYNOBATRACHIDAE LC 5 Cardioglossa escalerae ARTHROLEPTIDAE LC 51 Phrynobatrachus sandersoni PHRYNOBATRACHIDAE LC 12 Cardioglossa gracilis ARTHROLEPTIDAE LC 80 Phrynomantis affinis MICROHYLIDAE LC 99 Cardioglossa gratiosa ARTHROLEPTIDAE LC 75 Phrynomantis annectens MICROHYLIDAE LC 79 Cardioglossa manengouba ARTHROLEPTIDAE NA 1 Phrynomantis bifasciatus MICROHYLIDAE LC 462 Cardioglossa melanogaster ARTHROLEPTIDAE EN 4 Phrynomantis microps MICROHYLIDAE LC 240 Cardioglossa nigromaculata ARTHROLEPTIDAE NT 10 Phrynomantis somalicus MICROHYLIDAE LC 18 Hylarana occidentalis RANIDAE EN 17 Poyntonia paludicola PYXICEPHALIDAE NT 4 Cardioglossa oreas ARTHROLEPTIDAE EN 3 Probreviceps durirostris BREVICIPITIDAE EN 1 Cardioglossa pulchra ARTHROLEPTIDAE EN 10 Probreviceps loveridgei BREVICIPITIDAE VU 4 Cardioglossa schioetzi ARTHROLEPTIDAE EN 1 Probreviceps macrodactylus BREVICIPITIDAE VU 2 Cardioglossa trifasciata ARTHROLEPTIDAE CR 2 Probreviceps rhodesianus BREVICIPITIDAE EN 2 Cardioglossa venusta ARTHROLEPTIDAE EN 4 Probreviceps rungwensis BREVICIPITIDAE VU 2 Chiromantis kelleri RHACOPHORIDAE LC 115 Probreviceps uluguruensis BREVICIPITIDAE VU 2 Chiromantis petersii RHACOPHORIDAE LC 57 Pseudhymenochirus merlini PIPIDAE LC 8 Chiromantis rufescens RHACOPHORIDAE LC 183 Ptychadena aequiplicata PTYCHADENIDAE LC 97 Chiromantis xerampelina RHACOPHORIDAE LC 268 Ptychadena anchietae PTYCHADENIDAE LC 532 Chlorolius koehleri HYPEROLIIDAE LC 18 Ptychadena ansorgii PTYCHADENIDAE LC 112 Chrysobatrachus cupreonitens HYPEROLIIDAE DD 2 Ptychadena arnei PTYCHADENIDAE DD 3 Churamiti maridadi BUFONIDAE CR 1 Ptychadena bibroni PTYCHADENIDAE LC 370 Conraua alleni PETROPEDETIDAE VU 4 Ptychadena broadleyi PTYCHADENIDAE EN 1 Conraua alleni PETROPEDETIDAE VU 2 Ptychadena bunoderma PTYCHADENIDAE LC 41 Conraua alleni PETROPEDETIDAE VU 8 Ptychadena christyi PTYCHADENIDAE DD 19 Conraua beccarii PETROPEDETIDAE LC 28 Ptychadena chrysogaster PTYCHADENIDAE LC 7 Conraua crassipes PETROPEDETIDAE LC 91 Ptychadena cooperi PTYCHADENIDAE LC 10 Conraua derooi PETROPEDETIDAE CR 3 Ptychadena erlangeri PTYCHADENIDAE NT 15 Conraua goliath PETROPEDETIDAE EN 14 Ptychadena filwoha PTYCHADENIDAE DD 2 Conraua robusta PETROPEDETIDAE VU 4 Ptychadena gansi PTYCHADENIDAE LC 15 Crotaphatrema tchabalmbaboensis CAECILIIDAE DD 1 Ptychadena grandisonae PTYCHADENIDAE LC 124 Crotaphatrema bornmuelleri CAECILIIDAE DD 1 Ptychadena guibei PTYCHADENIDAE LC 144 Crotaphatrema lamottei CAECILIIDAE DD 1 Ptychadena harenna PTYCHADENIDAE DD 1 Cryptothylax greshoffii HYPEROLIIDAE LC 145 Ptychadena ingeri PTYCHADENIDAE DD 3 Cryptothylax minutus HYPEROLIIDAE DD 1 Ptychadena keilingi PTYCHADENIDAE LC 18 Didynamipus sjostedti BUFONIDAE EN 4 Ptychadena longirostris PTYCHADENIDAE LC 52 Ericabatrachus baleensis PYXICEPHALIDAE EN 1 Ptychadena mahnerti PTYCHADENIDAE LC 6

112 | Chapter III

Red Ran Red Ran Genus and species Family List ge Genus and species Family List ge Geotrypetes angeli CAECILIIDAE DD 3 Ptychadena mapacha PTYCHADENIDAE DD 2 Geotrypetes pseudoangeli CAECILIIDAE DD 2 Ptychadena mascareniensis PTYCHADENIDAE LC 447 Geotrypetes seraphini CAECILIIDAE LC 112 Ptychadena mossambica PTYCHADENIDAE LC 219 Heleophryne hewitti HELEOPHRYNIDAE CR 2 Ptychadena nana PTYCHADENIDAE DD 1 Heleophryne natalensis HELEOPHRYNIDAE LC 34 Ptychadena neumanni PTYCHADENIDAE LC 35 Heleophryne orientalis HELEOPHRYNIDAE LC 2 Ptychadena obscura PTYCHADENIDAE LC 44 Heleophryne purcelli HELEOPHRYNIDAE LC 7 Ptychadena oxyrhynchus PTYCHADENIDAE LC 649 Heleophryne regis HELEOPHRYNIDAE LC 5 Ptychadena perplicata PTYCHADENIDAE LC 47 Heleophryne rosei HELEOPHRYNIDAE CR 1 Ptychadena perreti PTYCHADENIDAE LC 72 Hemisus barotseensis HEMISOTIDAE DD 6 Ptychadena porosissima PTYCHADENIDAE LC 144 Hemisus brachydactylus HEMISOTIDAE DD 1 Ptychadena pujoli PTYCHADENIDAE DD 3 Hemisus guineensis HEMISOTIDAE LC 601 Ptychadena pumilio PTYCHADENIDAE LC 186 Hemisus guttatus HEMISOTIDAE VU 12 Ptychadena retropunctata PTYCHADENIDAE DD 4 Hemisus marmoratus HEMISOTIDAE LC 102 Ptychadena schillukorum PTYCHADENIDAE LC 38 5 Hemisus microscaphus HEMISOTIDAE LC 28 Ptychadena stenocephala PTYCHADENIDAE LC 8 Hemisus olivaceus HEMISOTIDAE LC 43 Ptychadena straeleni PTYCHADENIDAE LC 3 Hemisus perreti HEMISOTIDAE DD 8 Ptychadena submascareniensis PTYCHADENIDAE DD 3 Hemisus wittei HEMISOTIDAE DD 3 Ptychadena subpunctata PTYCHADENIDAE LC 126 Herpele multiplicata CAECILIIDAE DD 1 Ptychadena superciliaris PTYCHADENIDAE NT 34 Herpele squalostoma CAECILIIDAE LC 87 Ptychadena taenioscelis PTYCHADENIDAE LC 161 Hildebrandtia macrotympanum PTYCHADENIDAE LC 57 Ptychadena tellinii PTYCHADENIDAE LC 192 Hildebrandtia ornata PTYCHADENIDAE LC 362 Ptychadena tournieri PTYCHADENIDAE LC 47 Hildebrandtia ornatissima PTYCHADENIDAE DD 5 Ptychadena trinodis PTYCHADENIDAE LC 212 Hoplobatrachus occipitalis DICROGLOSSIDAE LC 514 Ptychadena upembae PTYCHADENIDAE LC 88 Hoplophryne rogersi MICROHYLIDAE EN 3 Ptychadena uzungwensis PTYCHADENIDAE LC 173 Hoplophryne uluguruensis MICROHYLIDAE VU 3 Ptychadena wadei PTYCHADENIDAE DD 1 Hymenochirus boettgeri PIPIDAE LC 185 Pyxicephalus adspersus PYXICEPHALIDAE LC 335 Hymenochirus boulengeri PIPIDAE DD 2 Pyxicephalus edulis PYXICEPHALIDAE LC 277 Hymenochirus curtipes PIPIDAE LC 1 Pyxicephalus obbianus PYXICEPHALIDAE LC 16 Hymenochirus feae PIPIDAE DD 1 Rana lubrica PYXICEPHALIDAE NA 1 Hyperolius acuticephalus HYPEROLIIDAE DD 1 Rana tenuoplicata PYXICEPHALIDAE NA 1 Hyperolius acuticeps HYPEROLIIDAE LC 578 Amietia viridireticulata PYXICEPHALIDAE DD 1 Hyperolius acutirostris HYPEROLIIDAE NT 11 Schismaderma carens BUFONIDAE LC 295 Hyperolius ademetzi HYPEROLIIDAE NT 2 Schistometopum gregorii CAECILIIDAE LC 8 Hyperolius argus HYPEROLIIDAE LC 111 Arthroleptis crusculum ARTHROLEPTIDAE EN 1 Hyperolius atrigularis HYPEROLIIDAE DD 4 Arthroleptis discodactylus ARTHROLEPTIDAE DD 1 Hyperolius balfouri HYPEROLIIDAE LC 131 Arthroleptis hematogaster ARTHROLEPTIDAE DD 2 Hyperolius baumanni HYPEROLIIDAE LC 2 Arthroleptis lameerei ARTHROLEPTIDAE LC 55 Hyperolius benguellensis HYPEROLIIDAE LC 222 Arthroleptis loveridgei ARTHROLEPTIDAE DD 1 Hyperolius bicolor HYPEROLIIDAE DD 1 Tomopterna milletihorsini PYXICEPHALIDAE DD 1 Hyperolius bobirensis HYPEROLIIDAE EN 2 Arthroleptis mossoensis ARTHROLEPTIDAE DD 1 Hyperolius bocagei HYPEROLIIDAE NA 103 Arthroleptis nimbaensis ARTHROLEPTIDAE DD 1 Hyperolius bolifambae HYPEROLIIDAE LC 41 Arthroleptis phrynoides ARTHROLEPTIDAE DD 1 Hyperolius bopeleti HYPEROLIIDAE NT 4 Arthroleptis pyrrhoscelis ARTHROLEPTIDAE NT 2 Hyperolius brachiofasciatus HYPEROLIIDAE DD 1 Arthroleptis schubotzi ARTHROLEPTIDAE LC 9 Hyperolius camerunensis HYPEROLIIDAE LC 4 Arthroleptis spinalis ARTHROLEPTIDAE DD 1 Hyperolius castaneus HYPEROLIIDAE VU 5 Arthroleptis sylvaticus ARTHROLEPTIDAE LC 51 Hyperolius chlorosteus HYPEROLIIDAE NT 32 Arthroleptis taeniatus ARTHROLEPTIDAE LC 72 Hyperolius chrysogaster HYPEROLIIDAE VU 5 Arthroleptis troglodytes ARTHROLEPTIDAE CR 1 Hyperolius cinereus HYPEROLIIDAE DD 5 Arthroleptis vercammeni ARTHROLEPTIDAE DD 1 Hyperolius cinnamomeoventris HYPEROLIIDAE LC 363 Arthroleptis xenochirus ARTHROLEPTIDAE LC 70 Hyperolius concolor HYPEROLIIDAE LC 126 Arthroleptis xenodactyloides ARTHROLEPTIDAE LC 12 Hyperolius cystocandicans HYPEROLIIDAE VU 5 Arthroleptis xenodactylus ARTHROLEPTIDAE VU 93 Hyperolius diaphanus HYPEROLIIDAE DD 4 Arthroleptis zimmeri ARTHROLEPTIDAE DD 1 Hyperolius dintelmanni HYPEROLIIDAE EN 1 Scolecomorphus kirkii CAECILIIDAE LC 11 Hyperolius discodactylus HYPEROLIIDAE VU 6 Scolecomorphus uluguruensis CAECILIIDAE LC 2 Hyperolius endjami HYPEROLIIDAE VU 5 Scolecomorphus vittatus CAECILIIDAE LC 8 Hyperolius fasciatus HYPEROLIIDAE DD 1 Scotobleps gabonicus ARTHROLEPTIDAE LC 40 Hyperolius ferreirai HYPEROLIIDAE DD 1 Semnodactylus wealii HYPEROLIIDAE LC 66 Hyperolius ferrugineus HYPEROLIIDAE DD 1 Silurana epitropicalis PIPIDAE LC 177 Hyperolius frontalis HYPEROLIIDAE VU 8 Silurana tropicalis PIPIDAE LC 143 Hyperolius fumosus HYPEROLIIDAE NA 1 Spelaeophryne methneri BREVICIPITIDAE LC 25 Hyperolius fusciventris HYPEROLIIDAE LC 2 Altiphrynoides osgoodi BUFONIDAE VU 8 Hyperolius fusciventris HYPEROLIIDAE LC 87 Mertensophryne anotis BUFONIDAE EN 2 Hyperolius ghesquieri HYPEROLIIDAE DD 1 Mertensophryne loveridgei BUFONIDAE LC 12 Hyperolius glandicolor HYPEROLIIDAE LC 83 Mertensophryne usambarae BUFONIDAE EN 1 Hyperolius gularis HYPEROLIIDAE DD 1 Strongylopus bonaespei PYXICEPHALIDAE LC 12 Hyperolius guttulatus HYPEROLIIDAE LC 84 Strongylopus fasciatus PYXICEPHALIDAE LC 95

Omission of narrow-ranging species | 113

Red Ran Red Ran Genus and species Family List ge Genus and species Family List ge Hyperolius horstockii HYPEROLIIDAE VU 10 Strongylopus fuelleborni PYXICEPHALIDAE LC 15 Hyperolius hutsebauti HYPEROLIIDAE DD 1 Strongylopus hymenopus PYXICEPHALIDAE LC 6 Hyperolius inornatus HYPEROLIIDAE DD 1 Strongylopus kilimanjaro PYXICEPHALIDAE DD 1 Hyperolius kachalolae HYPEROLIIDAE LC 13 Strongylopus kitumbeine PYXICEPHALIDAE VU 1 Hyperolius kibarae HYPEROLIIDAE DD 5 Strongylopus merumontanus PYXICEPHALIDAE VU 1 Hyperolius kihangensis HYPEROLIIDAE EN 1 Strongylopus rhodesianus PYXICEPHALIDAE VU 6 Hyperolius kivuensis HYPEROLIIDAE LC 238 Strongylopus springbokensis PYXICEPHALIDAE LC 6 Hyperolius kuligae HYPEROLIIDAE LC 60 Strongylopus wageri PYXICEPHALIDAE NT 9 Hyperolius lamottei HYPEROLIIDAE LC 40 Sylvacaecilia grandisonae CAECILIIDAE DD 4 Hyperolius langi HYPEROLIIDAE LC 19 Tomopterna cryptotis PYXICEPHALIDAE LC 389 Hyperolius lateralis HYPEROLIIDAE LC 29 Tomopterna damarensis PYXICEPHALIDAE DD 1 Hyperolius laurenti HYPEROLIIDAE VU 6 Tomopterna delalandii PYXICEPHALIDAE LC 33 Hyperolius leleupi HYPEROLIIDAE EN 1 Tomopterna krugerensis PYXICEPHALIDAE LC 136 Hyperolius leucotaenius HYPEROLIIDAE EN 2 Tomopterna luganga PYXICEPHALIDAE LC 7 Hyperolius lucani HYPEROLIIDAE DD 1 Tomopterna marmorata PYXICEPHALIDAE LC 99 Hyperolius maestus HYPEROLIIDAE DD 1 Tomopterna natalensis PYXICEPHALIDAE LC 71 Hyperolius major HYPEROLIIDAE LC 5 Tomopterna tandyi PYXICEPHALIDAE LC 184 Hyperolius marginatus HYPEROLIIDAE LC 196 Tomopterna tuberculosa PYXICEPHALIDAE LC 159 Hyperolius mariae HYPEROLIIDAE LC 12 Trichobatrachus robustus ARTHROLEPTIDAE LC 50 Hyperolius marmoratus HYPEROLIIDAE LC 301 Werneria bambutensis BUFONIDAE EN 2 Hyperolius minutissimus HYPEROLIIDAE VU 4 Werneria iboundji BUFONIDAE CR 1 Hyperolius mitchelli HYPEROLIIDAE LC 92 Werneria mertensiana BUFONIDAE EN 9 Hyperolius montanus HYPEROLIIDAE LC 7 Werneria preussi BUFONIDAE EN 2 Hyperolius mosaicus HYPEROLIIDAE LC 30 Werneria submontana BUFONIDAE EN 1 Hyperolius nasutus HYPEROLIIDAE LC 101 Werneria tandyi BUFONIDAE EN 1 Hyperolius nasutusis HYPEROLIIDAE NA 2 Wolterstorffina chirioi BUFONIDAE CR 1 Hyperolius nienokouensis HYPEROLIIDAE EN 1 Wolterstorffina mirei BUFONIDAE EN 2 Hyperolius nimbae HYPEROLIIDAE EN 1 Wolterstorffina parvipalmata BUFONIDAE VU 10 Hyperolius nitidulus HYPEROLIIDAE LC 294 Xenopus amieti PIPIDAE NT 6 Hyperolius obscurus HYPEROLIIDAE DD 1 Xenopus andrei PIPIDAE LC 20 Hyperolius occidentalis HYPEROLIIDAE LC 14 Xenopus borealis PIPIDAE LC 29 Hyperolius ocellatus HYPEROLIIDAE LC 201 Xenopus boumbaensis PIPIDAE DD 1 Hyperolius parallelus HYPEROLIIDAE LC 160 Xenopus clivii PIPIDAE LC 51 Hyperolius pardalis HYPEROLIIDAE LC 70 Xenopus fraseri PIPIDAE LC 235 Hyperolius parkeri HYPEROLIIDAE LC 40 Xenopus gilli PIPIDAE EN 3 Hyperolius phantasticus HYPEROLIIDAE LC 125 Xenopus itombwensis PIPIDAE CR 1 Hyperolius pickersgilli HYPEROLIIDAE EN 5 Xenopus laevis PIPIDAE NA 558 Hyperolius picturatus HYPEROLIIDAE LC 41 Xenopus laevis PIPIDAE LC Hyperolius pictus HYPEROLIIDAE LC 12 Xenopus largeni PIPIDAE DD 2 Hyperolius platyceps HYPEROLIIDAE LC 132 Xenopus longipes PIPIDAE CR 1 Hyperolius polli HYPEROLIIDAE DD 2 Xenopus muelleri PIPIDAE LC 248 Hyperolius polystictus HYPEROLIIDAE VU 1 Xenopus muelleri PIPIDAE LC 210 Hyperolius protchei HYPEROLIIDAE DD 1 Xenopus petersii PIPIDAE LC 217 Hyperolius pseudargus HYPEROLIIDAE LC 2 Xenopus pygmaeus PIPIDAE LC 32 Hyperolius punctulatus HYPEROLIIDAE DD 3 Xenopus ruwenzoriensis PIPIDAE DD 2 Hyperolius pusillus HYPEROLIIDAE LC 118 Xenopus vestitus PIPIDAE LC 8 Hyperolius pustulifer HYPEROLIIDAE DD 1 Xenopus victorianus PIPIDAE LC 9 Hyperolius pyrrhodictyon HYPEROLIIDAE LC 12 Xenopus wittei PIPIDAE LC 8

Table S2 | Main amphibian data sources.

Data source Resolution Location Type

Literature based confirmed location Copenhagen http://130.225.211.158/subsa 1 degree grid databases, with range interpreta- database haranafrica/subsaharan.htm tion for most widely spread species

Expert and literature based Extent Global Amphibi- http://www.iucnredlist.org/in Polygons of Occurrence map polygons that an Assessment itiatives/amphibians join known locations

114 | Chapter III

Table S3 | Summary of IUCN Red List status for 767 species of amphibian in sub-Saharan Africa (www.iucnredlist.org, accessed June 2011). The distributions of species with more than ten records of occurrence on a 1° grid (eligible species) were modelled using MaxEnt. Those with 1-10 records (rare species) could not be modelled, but are a key component of other conservation priority schemes. The 186 species in the categories ‘Extinct in the Wild, Critically Endangered, Endangered and Vulnerable’ are regarded as ‘threatened with extinction’ by the IUCN.

Rare species Eligible species All species (N = 432) (N = 335) (N = 767) Red List category Mean range = 3.1 Mean range = 110.8 Mean range = 50.1 Median range = 2.0 Median range = 64.0 Median range = 7 Extinct in the wild (EW) 1 0.23% 0 0.00% 1 0.13% Critically Endangered (CR) 26 6.02% 0 0.00% 26 3.39% Endangered (EN) 85 19.68% 3 0.90% 88 11.47% Vulnerable (VU) 63 14.58% 8 2.39% 71 9.26% Near-threatened (NT) 18 4.17% 20 5.97% 38 4.95% Data deficient (DD) 155 35.88% 6 1.79% 161 20.99% Least Concern (LC) 64 14.81% 292 87.16% 356 46.41% Not evaluated (NE) 20 4.63% 6 1.79% 26 3.39% NT: Species not quite meeting criteria as threatened DD: Species listed that may be threatened, but there are not enough data to evaluate them LC: Species not regarded as threatened according to the IUCN Red List NE: Species missing from the IUCN Red List (newly described species)

Table S4 | Environmental variable loadings and variance explained by the first two axes of the Outlying Mean Index Analysis (OMI), on present-day climate and on future climate anomalies.

Present-day climate Future anomalies Variable (1980) (2080) OMI axis 1 OMI axis 2 OMI axis 1 OMI axis 2 Mean temperature -0.05 0.78 0.64 0.42 Annual precipitation 0.70 0.27 -0.33 0.91 Precipitation seasonality -0.70 0.11 0.70 0.05 Elevation range 0.12 -0.56 - -

Variance explained 48% 33% 46% 29%

Phillips S.J. & Dudik M. (2008) Modeling of species Supporting references distributions with Maxent: new extensions and a comprehensive evaluation. Ecography, 31, 161–175. Elith J., Kearney M., & Phillips S. (2010) The art of modelling range-shifting species. Methods in Ecology Platts P.J., McClean C.J., Lovett J.C., & Marchant R. (2008) and Evolution, 1, 330–342. Predicting tree distributions in an East African biodiversity hotspot : model selection, data bias and Garcia R. a., Burgess N.D., Cabeza M., Rahbek C., & Araújo envelope uncertainty. Ecological Modelling, 218, M.B. (2012) Exploring consensus in 21st century 121–13. projections of climatically suitable areas for African vertebrates. Global Change Biology, 18, 1253–1269. Chapter IV

Multiple dimensions of climate change and their implications for biodiversity

RAQUEL A. GARCIA, MAR CABEZA, CARSTEN RAHBEK, AND MIGUEL B. ARAÚJO Manuscript in review

Multiple dimensions of climate change and their implications for biodiversity

RAQUEL A. GARCIA1,2,3, MAR CABEZA4, CARSTEN RAHBEK1, and MIGUEL B. ARAÚJO1,2,3,5

Manuscript in review

The 21st century is projected to witness to aspects of climate. By contrast, mechanistic climatic changes unprecedented in human models explicitly consider demographic (5) and history, with greater warming often report- physiological (6) processes to predict species' ed for northern latitudes. Yet, climate responses to climate change exposure. Given change can be measured in a variety of the limited availability of demographic and ways, reflecting distinct dimensions of physiological data, most assessments of climate change with unequal spatial patterns across change effects on biodiversity use statistical the world. For example, extreme and novel rather than process-based models. Yet, for most climates will be most common in the tropics, species on Earth distributional data are also whereas polar climates will shrink in area. lacking so that even the simplest statistical What are the implications for biodiversity model cannot be fitted. An alternative is to use from each of these multiple changes? We simple metrics of climate change to quantify the review existing climate change metrics and exposure of areas to different dimensions of discuss how they relate to threats and op- change and relate these dimensions to different portunities for biodiversity. For unknown or threats and opportunities for biodiversity (e.g., poorly described species, which represent 7–10). the majority of Earth's biodiversity, simple Among climate change metrics, the anoma- metrics of climate change can identify lies in temperature or rainfall at any given lo- where species are most exposed to the dif- cality over time are commonly used. However, a ferent challenges arising from changing diversity of metrics exists, including temporal climates. differences in extreme events such as droughts, Accurate forecasts of climate change effects decreases or increases in total area with given on biodiversity are required to address broader climatic conditions, and the velocity at which societal commitments towards natural re- climate moves across space and time. Climate source management and conservation (1–3). To change metrics have been applied to a variety address this problem, researchers have devel- of questions. When coupled with climate oped a range of bioclimatic models. Statistical hindcasts, they have been used to examine the models (4) measure species' exposure to cli- role of historical climatic changes in species mate change by relating species' distributions richness gradients (11–14), the geographical

118 | Chapter IV

Box I | Metrics of climate change

Local change metrics Panel I | Schematic representation of local change metrics. Climatic conditions at a given locality (e.g., a gridded cell) can change in magnitude, timing or position over time. From time period t1 to t2, the reference cell (lower left corner) experiences change in the magnitude of climate or timing of climatic events. The velocity at which climate in t1 moves across space and time depends on the magnitude of change and the local spatial climatic gradient. Dimensions Examples of metrics Examples of ecological applications Magnitude Anomalies: the difference in climatic parame- Assessing the role of Last Glacial Maxima climate ters at a given locality over time. on European dung beetle (13) and amphibian and reptile (11) diversity patterns. Standardized anomalies*: the Euclidean dis- Predicting the level of future climate changes in tance between baseline and future climate at a relation to current biodiversity pattern across the given locality, standardized by historical world (22, 23). inter-annual climate variability (22). Change in the probability of extremes*: the Examining the influence of extreme climates on difference over time in the probability of invasion susceptibility of plant communities in occurrence of the most extreme historical North-central Chile over the past decades (25). climatic event at a given locality. Timing Changes in seasonality: The difference in the Assessing the role of delayed snowmelt on the timing of climatic events. hibernation emergence date of ground squirrels in Canada (33). Position Climate change velocity*: the ratio of the Predicting climate residence time in global protect- temporal climatic gradient at a given locality ed areas (8); comparing past velocities to Australian to the local spatial climatic gradient (8). bird distribution shifts (18), and to global vertebrate endemism (15).

Regional change metrics Panel II | Schematic representation of regional change metrics. The distribution of climatic conditions across a set of localities (e.g., set of gridded cells) can change in area availability or position. For the reference cell (lower left corner), the amount of area experiencing the same climate decreases from four cells in t1 to two cells in t2 (dotted outline), and average distance to cells experiencing the same climate increases from t1 to t2 (compare the length of the arrows to similar conditions in t1 and t2). The lightest shade in t1 and the darkest shade in t2 correspond to disappearing and novel climates respectively. Dimensions Metrics Examples of ecological applications Availability Disappearing (novel) climates*: the minimum Testing the agreement of dissimilarities relative to Euclidean distance between a given baseline present between climate and plant associations in (future) climate and all future (baseline) eastern North America over the past 18 thousand climates (22). years (31, 67). Change in area of analogous climates*: the Exploring climate determinants of centers of rarity change over time in area experiencing similar for Northern Hemisphere plant, bird and butterfly climates [differing less than set thresholds species (7); assessing the role of changes in Late (22, 34), or belonging to the same set class Quaternary climate availability on mammal extinc- (35, 105) or cluster (106)]. tions (20). Position Change in distance to analogous climates*: the Assessing the role of Late Quaternary climate change over time in the distance to similar movements on mammal extinctions (20). climates (as defined above). * Climate change metric implemented in our global comparison of climate change patterns

Dimensions of climate change | 119

distribution of centers of endemism (15) or termed local metrics), or measure shifts in the past climatic refugia (16, 17), the pace of distri- distribution of climatic conditions over space butional shifts of species over past decades (18) and time (hereafter termed regional metrics). or millennia (19), the risk of species extinctions Metrics characterize different dimensions of (20), and the degree of biotic network speciali- climate change, including changes in the magni- zation across latitude (21). When coupled with tude, timing, position, and availability of cli- climate change forecasts, they have supported mate. Local metrics include temporal changes risk assessments for biodiversity (7, 9, 10, 22– in the magnitude of climatic averages (22, 32) 26) and conservation areas (8, 10, 27–29), while or climatic extremes (9, 25), shifts in the timing also enabling the identification of potential of seasonal climatic conditions (12, 33), and the regional pools of species (30). velocity at which climate shifts its position (8, Despite the increased use of climate change 15). Regional metrics begin with the characteri- metrics in basic and applied sciences, the varie- zation of climatic conditions across a given ty of existing metrics and their ecological impli- region, and then measure temporal changes in cations have not been fully appreciated. Studies the availability of analogous climatic conditions spanning different geographical areas have across that region (22, 34, 35), as well as chang- used single metrics in isolation (e.g., 8, 10, 25, es in the direction to, or distance between, the 31) more frequently than multiple metrics in positions of analogous climatic conditions (34). combination (20, 22, 23). A comprehensive Metrics can be applied to a single climatic pa- comparison of metrics, including the analysis of rameter, such as temperature or precipitation, their global patterns and implications for biodi- or multiple parameters. versity, is thus needed to gain understanding of the properties of these metrics and guide their use. Here, we provide a review of the climate Mapping multiple dimensions of change metrics commonly used in biodiversity climate change assessments. We first describe the different metrics and illustrate their variety by imple- Implementation of six common climate change menting six of the most common metrics glob- metrics worldwide reveals that each has differ- ally with 21st century climate projections. We ent, sometimes opposing, patterns worldwide then explore how the different metrics capture (Fig. 1). Local climate anomalies are projected distinct dimensions of climate change and carry to affect mainly the tropics and sub-tropics, complementary information that is useful for with more than half the area of tropical cli- understanding the impacts of ongoing climate mates experiencing large changes in relation to changes on biodiversity. Finally, we outline a historical inter-annual variability (for more conceptual framework for linking the different details see Fig. S3). Changes in temperature are metrics to assessments of threat and oppor- the main cause for such large anomalies in the tunity for species. tropics. The patterns of change in the probability of local extremes also highlight the tropics, with ca. 60% of the area of tropical climates Climate change metrics facing increased likelihood of extreme drying and warming events. Existing metrics of climate change generally fall In turn, regional metrics of change in the into one of two categories (Box I; Table S1). area of climate types reveal a trend towards They measure temporal changes in climate expansion of hot arid climates of the Sahara, variables at individual localities (hereafter southern Africa and Australia, accompanied by

120 | Chapter IV

Figure 1 | Projected global climate change according to six metrics. The maps show projections of change in annual mean temperature and total precipitation between the baseline and an end-of-century multi-model ensemble under the A1B emissions scenario (107). The metrics illustrated cover changes in magnitude [standardized climate anomalies (22), and change in the probability of extreme climates (108)], availability [change in area of analogous climates (34, 105), and novel climates (22)], and position [change in distance to analogous climates (34, 105), and climate change velocity (8)]. Correlation between these metrics is generally lower than expected by chance (Table S2). In each panel, the main maps show changes for temperature and precipitation combined, and the smaller maps show changes for each variable individually. The scales were defined using quantiles and reflect a gradient from small or positive (dark blue) to large or negative (dark brown) changes. Local anomalies, novel climates and climate change velocity values were converted to logarithmic scale for visualization. See Fig. S1 and S2 for analysis of sensitivity to alternative climate models.

Dimensions of climate change | 121

high in such high latitudes, but also in the tropical regions of central Africa, the western Ama- zon, and southern Australia. Examining metrics in combination shows that different areas of the world are exposed to different dimensions of climate change. The global patterns are far more complex than the often reported disproportionate warming for land areas at higher northern latitudes (36). In the polar regions, reductions in the global availability of similar climatic conditions are by far the greatest threat among all three dimensions examined (Fig. 2). The tropics and, to a lesser extent, hot arid regions are exposed to both emergence of novel climates and changes in average and extreme climates. Temperate regions are exposed almost equally to all dimensions of change, and cold regions are particularly exposed to large or high-velocity shifts in the position of prevailing climates.

Figure 2 | Exposure of the world’s climatic regions to different climate change dimensions. The star plot (upper panel) shows the pro- Linking metrics to threats and portion of area of different climatic regions of the world that are exposed to large changes in six opportunities for biodiversity climate change metrics capturing changes in mag- If different climate change metrics have differ- nitude, availability and position of climate (107). The line colors on the star plot correspond to the ent spatial patterns around the world, what are Köppen-Geiger broad climatic regions represent- the implications of each for biodiversity? We ed on the map (bottom panel; see Fig. S4). See Fig. outline a conceptual framework that describes S5 for analysis of sensitivity to alternative climate the links between metrics of climate change models. and potential threats and opportunities for

biodiversity (Fig. 3). The framework is based on reductions in the area of polar and mountain two principles. Firstly, under changing climates, climates. The climatic conditions over the polar persistence of local populations is more likely region and major mountains across the world where climatic conditions remain suitable (37, are projected to shrink up to 66%. By contrast, 38). Secondly, the survival of species depends novel climates are more likely to emerge in the on the continued availability of suitable climatic tropics and sub-tropics, again due to warmer conditions either within or outside the present temperatures: end-of-century temperatures for ranges of species (39). Under these general 62% of the world's tropical areas are projected principles, local changes can be seen as a proxy to be the most dissimilar from today’s tempera- for demographic threats and opportunities tures. In about a third of the temperate and acting at the population level, whereas regional polar regions, distances to similar climatic changes in climate can have negative or posi- conditions will increase more than elsewhere tive implications for the size and position of on the globe. Local velocity of changes in tem- entire species' ranges. perature and precipitation are forecasted to be

122 | Chapter IV

Threats and opportunities from local climate change tive seasons (41), and the mortality and population decline at the retracting margin of Aloe All else being equal, large changes in local cli- trees in the Namib desert, owing to desiccation mate are more likely to result in large changes stress (42). Changes in extreme climates may in local suitability for populations than small pose even greater threats on populations than changes, particularly when climate changes gradual trends (43). Increased tree mortality exceed past local variability (22). Decreased across the globe over the last decades has been local climatic suitability can have effects on the attributed to drought- and heat-induced stress physiology, morphology or behavior of the (44). Positive changes in local suitability for organisms in a population (40), potentially populations under warming are less frequently leading to changes in population demography. documented, but include reports of higher Two examples are the reported local extinc- fitness in common lizards in Southern France tions of lizards in Mexico, attributed to reduced resulting from an increase in body size (45), activity of individuals during warmer reproduc-

Figure 3 | Metrics of climate change and associated threats and opportunities for species. Metrics of climate change either quantify changes at local (locality) level or at regional (set of localities) level (Box I). Links are established between metrics capturing different dimensions of change and potential threats and opportunities for population dynamics, species occurrence, and species assemblages. increased growth of vascular plants in tundra of snowmelt date, for example, have led to de- ecosystems (46), and increased abundance of lays in the hibernation emergence date of Svalbard reindeer during phases of population ground squirrels in Canada (33). growth (47). Where populations are exposed to de- In turn, shifts in the timing of climatic creased climatic suitability, responses can in- events can lead to populations altering the clude adaptation in situ or dispersal. The veloci- timing of seasonal activities such as migration, ty of climate change provides an indication of flowering, or breeding. Such phenological shifts the opportunities that populations have to are commonly observed in response to climate track suitable climates across the local topog- change (48), and have been linked to the sea- raphy (51). In eastern North America, for ex- sonality of climate (33) but also to trends in ample, the rate of northward shifts in the dis- climatic means (49) and extremes (50). Delays tribution of woody taxa populations over the

Dimensions of climate change | 123

last 16 thousand years was paced by the veloci- discussed above was also greater in continents ty of temperature changes (19). The coinci- where suitable cold and dry conditions retreat- dence of global centers of endemism and areas ed farther (20). Over more recent decades, of low velocity temperature change since the average range shifts observed for various taxo- Last Glacial Maximum, 21 thousand years ago, nomic groups, mostly in the Northern Hemi- further testifies to the potential role of short- sphere, closely matched expectations from the distance climate tracking for in situ persistence poleward displacement of isotherms (48, 62). of species, particularly less vagile ones (15). Dispersal challenges are exacerbated when the direction of climate displacement crosses topo- Threats and opportunities from regional climate graphical barriers, fragmented habitat (63), or change unsuitable climate across time or space (64, 65). Shifts in the regional distribution of climates Predicting the effects of climate change on can affect the availability and distribution of whole assemblages of species remains challeng- climatically suitable areas for species across ing. The emergence of novel climates has been time and space (39). Increases or decreases in suggested to give rise to new assemblages lack- the area of climatic conditions that are suitable ing modern analogs (66), such as it happened for species can lead to range expansions or with vegetation assemblages in eastern North contractions (22). Long-term ecological data America during the last deglaciation (31, 67), reveal that species' distributions and abun- whereas disappearing climates can lead to the dances in both North and South Hemispheres disaggregation of extant assemblages (22), In often match, with time lag, warming and cool- general, spatio-temporal shifts in the distribu- ing cycles (52–55). The extensive fossil record tion of climate can result in the rearrangement in North America and Eurasia provides evi- of assemblages, with potential implications for dence of increases in the distribution or abun- species interactions and community reshuf- dance of plant species like elm and pine (56), fling. Non-analog communities (68) can arise and expansion of oaks out of glacial refugia (57) under changing climates either when species' as the end of the last glacial period brought individualistic responses lead to the disruption warmer temperatures. By contrast, Alpine trees of existing interactions, or when species with (58) and other cold-adapted species of birds non-overlapping ranges come into contact and and butterflies (7), and small mammals like the interact with one another (69–72). Examples mountain beaver and the western pocket go- include the climate-induced spread of patho- pher (59) became confined to small and isolat- gens (73, 74) or invasive species (75), insect ed pockets of cool conditions during the same outbreaks (44), decline of prey species (76), warming period. Climatic changes can also and temporal changes in host presence (77). explain the extinction of large mammals such as Despite the complexity of community dynamics the wooly rhino towards the end of the Late under climate change, measuring the multiple Quaternary (60). Indeed, continents losing dimensions of climate change and their likely more area representative of cold and dry condi- consequences for individual species (Fig. 3) is a tions owing to climate changes witnessed more first step towards a better understanding of the extinctions of megafauna (20). potential threats and opportunities for biodi- In turn, shifts in the position of climatic versity. conditions affect the ability with which species track climatic conditions (61) and modulate the risk of range fragmentation. The magnitude of the Late Quaternary megafaunal extinctions

124 | Chapter IV

The limits of climate change metrics duction of individuals, often mediating population demographic changes, vary across seasons Linking multiple dimensions of climate change and life-history stages (81). Also, long to threats and opportunities for species is but a timeframes conceal finer-scale climatic varia- first step towards understanding the effects of bility which can drive pulses of expansion or climate change on biodiversity. Ultimately, of contraction of species' ranges, as has been course, the effects of climate change on organ- shown for North American trees during the isms are also dependent on their intrinsic abili- Holocene (84). The spatial scale at which cli- ties to cope with, or adapt to, the different chal- mate changes are quantified also has implica- lenges they face (78). While climate change tions for the interpretation of metrics, particu- metrics can be used as proxies for threats and larly regional ones. For example, the expansion, opportunities facing biodiversity in general, a movement or emergence of novel climates number of considerations relating to the char- across broad spatial areas may represent acteristics of individual species are important threats and opportunities for widespread spe- when interpreting metric outputs. cies better than for geographically constrained A first consideration when using metrics of species (22). climate change in biodiversity assessments is that the vulnerability of species to a given threat, or the capacity to seize a given oppor- A combined view of threats and tunity, will vary across species. For example, opportunities individuals with higher genetic diversity or higher capacity to change key morphological, Simple metrics of climate change are often used physiological or behavioral traits in response to to predict the degree to which species, eco- adverse local climate changes are better regions, or protected areas are exposed to fu- equipped to persist in situ (78). In turn, changes ture climate changes (8–10). Given that alterna- in the regional distribution of climates can tive metrics represent different threats and disproportionately affect species with more opportunities for species, how should they be specialized climatic requirements, as they need used for supporting assessments of climate to follow suitable climatic conditions more change impacts on biodiversity? Threats or closely (7, 22). Likewise, poor dispersers are opportunities for biodiversity should increase more vulnerable to changes in the position of where species are simultaneously exposed to climates (15). A second consideration when several dimensions of climate change (85). using metrics is that different species respond Thus, examining multiple metrics in combina- to different climatic parameters (79). As the tion shows how different dimensions of climate spatial patterns arising for a given metric vary change interact to exacerbate or lessen species' between temperature and precipitation, for exposure to climate change. The selection of example (18, 80 and this review), the threats metrics to use is contingent on the question at and opportunities inferred from metrics may be hand but should cover all dimensions of change more or less significant for individual species (see Fig. 3) deemed relevant. depending on which parameter was measured. For illustration, we consider three metrics Finally, the temporal and spatial scales at representing changes in the magnitude, availa- which climate changes are measured have im- bility and position of climate. Standardized plications for the interpretation of metrics in anomalies, change in area of analogous cli- biodiversity assessments (81, 82), as processes mates, and climate change velocity, are ex- related to climate are scale-dependent (83). pected to discriminate between threats or op- Climatic thresholds for the survival or repro- portunities for local population demographic

Dimensions of climate change | 125

changes, for species' range contraction or ex- their optimal, they are likely threatened by pansion, and for species' range shifts, respec- local changes (88, 89). Opportunities for adap- tively (as per Fig. 3). When these three metrics tation despite local changes, particularly for are spatially overlaid, polar and mountain areas poor dispersers (15), are greater where micro- are highlighted as particularly exposed to mul- climates can offer refugia (90) than in lowland tiple threats from contemporary climate change areas such as in the Amazon and Congo Basin, (Fig. 4). Around 5% of the global land surface is where projected rates of local climate dis- exposed to large changes from the three met- placement are faster (cf. the two lightest orange rics (darkest brown shades in Fig. 4), mostly in shades in Fig. 4). In lowlands, shallower latitu- pockets of the Nordic tundra and mountain dinal gradients of temperature (91) may also ranges on both hemispheres. Both positive and prevent longer-distance dispersal or expansion negative effects of local warming have been into newly suitable areas. Dispersal challenges reported on population dynamics in polar areas will be greater for poor-dispersing and slow- (46, 47, 86), but regional shrinkage of similar reproducing species (92, 93), and in human- climates, compounded by the high velocities dominated landscapes (94). Range expansion required to track these climates, may place may also be possible for species in scattered polar species under the threat of range contrac- areas of expanding climates mostly in arid re- tion. gions including large portions of North-eastern Mountain areas offer more opportunities Australia and Sudan, especially for those with for species to track slow-moving climates over good dispersal abilities (the two darkest shades the complex mosaic of microclimates provided of blue in Fig. 4). by the local topography. However, range contraction still appears as a threat at the regional level (bright orange shades in Fig. 4). Where Challenges for impact assessment and suitable climates disappear altogether within conservation the species' reach, threat of extinction from habitat and ecosystem loss will be greater for No single unifying metric of climate change species endemic to those climates (7, 22, 34). exists. The conceptual framework proposed Small-ranged species of tropical mountain are- herein can help to interpret the outputs of dif- as with high levels of endemism, and species ferent metrics by classifying them into common lacking the genetic variability to survive in currencies of threat and opportunity. While the remnant isolated populations (87), thus appear framework can be applied to a range of sectors particularly threatened. Species in cold climate including agricul ture, forestry and health, our areas of the Northern Hemisphere face similar focus is on impacts of climate change on biodi- decreases in availability of suitable climates versity. (the two light blue shades in Fig. 4). Lower Projecting the level of exposure of biodi- anomalies may pose relatively smaller threats versity to climate changes is an integral part of locally, but at the regional level, species that are any framework to assess risks and select ap- specialized on these cold climates may face the propriate conservation strategies. Some of the threat of range contraction. existing frameworks for conservation under By contrast, most of the remaining tropical changing climates rely on single metrics of and sub-tropical regions are exposed to cli- climate change as proxies for exposure, includ- mates that are expanding regionally as well as ing the velocity of climate change (95), local changing locally. If tropical species require anomalies (96), and an index of future climatic specialized climatic conditions or live closer to stability that reflects availability of specific

126 | Chapter IV

Figure 4 | Spatial overlap of climate change metrics. The three metrics overlaid in space on the map (107)—standardized local anomalies, change in area of baseline-analogous climates, and climate change velocity—capture different climate-induced threats and opportunities for biodiversity (Fig. 3). For each metric, two classes were defined to differentiate between either positive and negative values, or below and above the median of values across the world. The different shades on the map and the diagram reflect interactions between local and regional threats or opportunities for biodiversity, with warm shades indicating a relatively higher level of local change than cold shades. See Fig. S6 for analysis of sensitivity to alternative climate models, and Fig. S7 for alternative combinations of climate change metrics. climatic conditions (10). Each of these frame- (97) can be vital in areas where shrinking cli- works thus addresses specific threats and op- mates reduce opportunities for species survival portunities for biodiversity. However, integrat- outside present ranges, particularly for climati- ing multiple metrics into a single framework cally specialized species. By contrast, promot- would allow for the identification of comple- ing landscape connectivity (98) and restoring mentary conservation strategies that may be new target habitat will be warranted wherever required. For example, mitigating local impacts local changes threaten in situ persistence but through abatement of habitat loss and en- low velocities of climate change enable species hancement of habitat quality or heterogeneity to track suitable climates (95).

Dimensions of climate change | 127

Although the actual effects of climate Acknowledgements change on biodiversity are extremely difficult to predict owing to the complexity of species We thank Hari Prasad for assisting with base- and community dynamics, climate change met- line climate data handling, and Brody Sandel for rics are likely to remain an important assess- sharing the R script for the spatial gradient of ment tool for at least three reasons. Firstly, climate change. R.A.G. is funded through a FCT models assessing the effects of climate change PhD studentship (SFRH/BD/65615/2009), on biodiversity are typically calibrated to cli- M.B.A. through the FCT PTDC/AAC- mate means, and little attention has been paid AMB/98163/2008 project and the Integrated to alternative dimensions of local change such Program of IC&DT Call Nº 1/SAESCTN/ALENT- as climate extremes or the timing of climate 07-0224-FEDER-00175, and M.C. through the events. Teasing apart alternative dimensions of RESPONSES project. C.R., R.A.G. and M.B.A. change to complement model-based assess- thank the Danish National Research Foundation ments, or integrating them into models (99, for support to the Center for Macroecology, 100), can bring valuable additional information. Evolution and Climate, and C.R. and M.B.A. also Secondly, metrics can be used for assessing thank the Imperial College London’s Grand climate change exposure when insufficient data Challenges in Ecosystems and Environment are available on the distributions of species. initiative for support of their research. Such is the case for the majority of known species across taxa (101), and particularly for the species-rich and highly impacted tropical areas References and Notes (102). Research on climate change risk to bio- 1. H. M. Pereira et al., Scenarios for Global Biodiversity in diversity has to date been geographically and the 21st Century, Science 330, 1496–1501 (2010). taxonomically biased (103), often excluding the 2. S. M. McMahon et al., Improving assessment and species that are currently most threatened for modelling of climate change impacts on global terrestrial biodiversity, Trends Ecol. Evol. 26, 249– lack of sufficient data points for modeling. 259 (2010). Hence, it is not a trivial feature of metric-based 3. D. Purves et al., Time to model all life on Earth, Nature assessments that they can be applied when and 493, 295–297 (2013). where limited knowledge of biodiversity exists. 4. A. T. Peterson et al., Ecological Niches and Geographic Distributions (Monographs in Population Biology, Thirdly, the sheer number of undescribed Princeton University Press, NewJersey, 2011). and undiscovered species (104) means that 5. D. A. Keith et al., Predicting extinction risks under assessments relying on available data represent climate change: coupling stochastic population a very small proportion of the total existing models with dynamic bioclimatic habitat models, Biol. Lett. 4, 560–563 (2008). biodiversity. For example, global inventories 6. M. Kearney, W. Porter, Mechanistic niche modelling: are estimated to cover only 20% of all existing combining physiological and spatial data to predict insects (104), a group with significant influence species’ ranges, Ecoll. Lett. 12, 334–350 (2009). on ecosystem functioning and services. When 7. R. Ohlemüller et al., The coincidence of climatic and knowledge of biodiversity is poor, simple met- species rarity: high risk to small-range species from climate change, Biol. Lett. 4, 568–572 (2008). rics of climate change, carefully implemented 8. S. R. Loarie et al., The velocity of climate change, and linked to threats and opportunities, can Nature 462, 1052–1055 (2009). provide a first-order assessment of the poten- 9. L. J. Beaumont, A. Pitman, S. Perkins, N. E. tial effects on the biota as a whole. Zimmermann, N. G. Yoccoz, Impacts of climate change on the world’ s most exceptional ecoregions, Proc. Natl. Acad. Sci. U.S.A. 108, 2306–2311 (2010). 10. J. E. M. Watson, T. Iwamura, N. Butt, Mapping vulnerability and conservation adaptation strategies

128 | Chapter IV

under climate change, Nat. Clim. Change 3, 989–994 plant communities, Ecoll. Lett. 14, 1227–1235 (2013). (2011). 11. M. B. Araújo et al., Quaternary climate changes 26. J. S. Li et al., Global Priority Conservation Areas in the explain diversity among reptiles and amphibians, Face of 21st Century Climate Change, PLoS ONE 8, Ecography 31, 8–15 (2008). e54839 (2013). 12. M. T. Burrows et al., The Pace of Shifting Climate in 27. T. Iwamura, K. A. Wilson, O. Venter, H. P. Possingham, Marine and Terrestrial Ecosystems, Science 334, A Climatic Stability Approach to Prioritizing Global 652–655 (2011). Conservation Investments, PLoS ONE 5, e15103 (2010). 13. J. Hortal et al., Ice age climate, evolutionary constraints and diversity patterns of European dung 28. J. a. Wiens, N. E. Seavy, D. Jongsomjit, Protected areas beetles., Ecoll. Lett. 14, 741–748 (2011). in climate space: What will the future bring?, Biol. Conserv. 144, 2119–2125 (2011). 14. S. F. Gouveia, J. Hortal, F. a. S. Cassemiro, T. F. Rangel, J. A. F. Diniz-Filho, Nonstationary effects of 29. T. Iwamura, A. Guisan, K. A. Wilson, H. P. Possingham, productivity, seasonality, and historical climate How robust are global conservation priorities to changes on global amphibian diversity, Ecography climate change?, Global Environ. Change 23, 1277– 36, 104–113 (2013). 1284 (2013). 15. B. Sandel et al., The Influence of Late Quaternary 30. J. Bergmann et al., The Iberian Peninsula as a Climate-Change Velocity on Species Endemism, potential source for the plant species pool in Science 334, 660–664 (2011). Germany under projected climate change, Plant Ecology 207, 191–201 (2010). 16. M. B. Ashcroft, J. R. Gollan, D. I. Warton, D. Ramp, A novel approach to quantify and locate potential 31. J. W. Williams, B. N. Shuman, I. I. I. T. Webb, microrefugia using topoclimate, climate stability, Dissimilarity analyses of late-Quaternary vegetation and isolation from the matrix, Global Change Biol. and climate in eastern North America, Ecology 82, 18, 1866–1879 (2012). 3346–3362 (2001). 17. F. P. Werneck, C. Nogueira, G. R. Colli, J. W. Sites, G. C. 32. F. Giorgi, Climate change hot-spots, Geophys. Res. Costa, Climatic stability in the Brazilian Cerrado: Lett. 33, L08707 (2006). implications for biogeographical connections of 33. J. E. Lane, L. E. B. Kruuk, A. Charmantier, J. O. Murie, F. South American savannas, species richness and S. Dobson, Delayed phenology and reduced fitness conservation in a biodiversity hotspot, J. Biogeogr. associated with climate change in a wild hibernator, 39, 1695–1706 (2012). Nature 489, 554–557 (2012). 18. J. VanDerWal et al., Focus on poleward shifts in 34. R. Ohlemüller, E. S. Gritti, M. T. Sykes, C. D. Thomas, species’ distribution underestimates the fingerprint Towards European climate risk surfaces: the extent of climate change, Nature Clim. Change 3, 239–243 and distribution of analogous and non-analogous (2013). climates 1931–2100, Glob. Ecol. Biogeogr. 15, 395– 19. A. Ordonez, J. W. Williams, Climatic and biotic 405 (2006). velocities for woody taxa distributions over the last 35. D. D. Ackerly et al., The geography of climate change: 16 000 years in eastern North America, Ecoll. Lett. implications for conservation biogeography, Divers. 16, 773–781 (2013). Distrib. 16, 476–487 (2010). 20. D. Nogués-Bravo, R. Ohlemüller, P. Batra, M. B. 36. G. A. Meehl et al., in Solomon S et al., Eds. (Cambridge Araújo, Climate predictors of late Quaternary University Press, Cambridge, United Kingdom and extinctions, Evolution 64, 2442–2449 (2010). New York, NY, USA, 2007), pp. 747–845. 21. B. Dalsgaard et al., Specialization in plant- 37. L. R. Holdridge, Determination of world plant hummingbird networks is associated with species formations from simple climatic data, Science 105, richness, contemporary precipitation and 367–368 (1947). quaternary climate-change velocity, PLoS ONE 6, e25891 (2011). 38. R. H. Whittaker, Communities and ecosystems (Macmillan, New York, ed. 2nd, 1975). 22. J. W. Williams, S. T. Jackson, J. E. Kutzbach, Projected distributions of novel and disappearing climates by 39. S. T. Jackson, J. T. Overpeck, Responses of Plant 2100 AD, Proc. Natl. Acad. Sci. U.S.A. 104, 5738–5742 Populations and Communities to Environmental (2007). Changes of the Late Quaternary, Paleobiology 26, 194–220 (2000). 23. C. D. Thomas et al., Exporting the ecological effects of climate change, EMBO Rep 9, S28–S33 (2008). 40. J. Peñuelas et al., Evidence of current impact of climate change on life: a walk from genes to the 24. S. J. Wright, H. C. Muller-Landau, J. Schipper, The biosphere, Global Change Biol. 19, 2303–2338 future of tropical species on a warmer planet, (2013). Conserv. Biol. 23, 1418–26 (2009). 25. M. A. Jiménez et al., Extreme climatic events change the dynamics and invasibility of semi-arid annual

Dimensions of climate change | 129

41. B. Sinervo et al., Erosion of lizard diversity by climate north america: scaling from taxa to biomes, Ecol. change and altered thermal niches, Science 328, Monogr. 74, 309–334 (2004). 894–9 (2010). 57. M. B. Davis, R. G. Shaw, Range shifts and adaptive 42. W. Foden et al., A changing climate is eroding the responses to Quaternary climate change, Science geographical range of the Namib Desert tree Aloe 292, 673–679 (2001). through population declines and dispersal lags, 58. H. J. B. Birks, K. J. Willis, Alpines, trees, and refugia in Divers. Distrib. 13, 645–653 (2007). Europe, Plant Ecology & Diversity 1, 147–160 43. A. Jentsch, C. Beierkuhnlein, Research frontiers in (2008). climate change: Effects of extreme meteorological 59. J. L. Blois, J. L. McGuire, E. A. Hadly, Small mammal events on ecosystems, Comptes Rendus Geoscience diversity loss in response to late-Pleistocene 340, 621–628 (2008). climatic change, Nature 465, 771–774 (2010). 44. C. D. Allen et al., A global overview of drought and 60. E. D. Lorenzen et al., Species-specific responses of heat-induced tree mortality reveals emerging Late Quaternary megafauna to climate and humans, climate change risks for forests, Forest Ecology and Nature 479, 359–64 (2011). Management 210, 660–684 (2010). 61. L. F. Pitelka et al., Plant migration and climate 45. S. Chamaillé-Jammes, M. Massot, P. Aragón, J. Clobert, change, American Scientist 85, 464–474 (1997). Global warming and positive fitness response in mountain populations of common lizards Lacerta 62. I.-C. Chen, J. K. Hill, R. Ohlemüller, D. B. Roy, C. D. vivipara, Global Change Biol. 12, 392–402 (2006). Thomas, Rapid Range Shifts of Species Associated with High Levels of Climate Warming, Science 333, 46. G. B. Hill, G. H. R. Henry, Responses of High Arctic wet 1024–1026 (2011). sedge tundra to climate warming since 1980, Global Change Biol. 17, 276–287 (2011). 63. P. Opdam, D. Wascher, Climate change meets habitat fragmentation: linking landscape and 47. N. J. C. Tyler, M. C. Forchhammer, N. A. Øritsland, biogeographical scale levels in research and Nonlinear effects of climate and density in the conservation, Biol. Conserv. 117, 285–297 (2004). dynamics of a fluctuating population of reindeer, Ecology 89, 1675–1686 (2008). 64. R. Early, D. F. Sax, Analysis of climate paths reveals potential limitations on species range shifts, Ecoll. 48. C. Parmesan, G. Yohe, A globally coherent fingerprint Lett. 14, 1125–1133 (2011). of climate change impacts across natural systems, Nature 421, 37–42 (2003). 65. J. Bennie et al., Range expansion through fragmented landscapes under a variable climate, Ecoll. Lett. 16, 49. T. L. Root et al., Fingerprints of global warming on 921–929 (2013). wild animals and plants, Nature 421, 57–60 (2003). 66. J. W. Williams, S. T. Jackson, Novel climates, no- 50. A. Jentsch, J. Kreyling, C. Beierkuhnlei, Beyond analog communities, and ecological surprises, Front. gradual warming: extreme weather events alter Ecol. Environ. 5, 475–482 (2007). flower phenology of European grassland and heath species, Global Change Biol. 15, 837–849 (2009). 67. J. T. Overpeck, R. S. Webb, T. Webb III, Mapping eastern North American vegetation change of the 51. D. L. Peterson, E. G. Schreiner, N. M. Buckingham, past 18 ka: No-analogs and the future, Geology 20, Gradients, Vegetation and Climate: Spatial and 1071–1974 (1992). Temporal Dynamics in the Olympic Mountains, U.S.A, Glob. Ecol. Biogeogr. Letters 6, 7–17 (1997). 68. R. W. Graham et al., Spatial Response of Mammals to Late Quaternary Environmental Fluctuations, 52. L. R. Flenley, Tropical Forests Under the Climates of Science 272, 1601–1606 (1996). the Last 30,000 Years, Clim. Change 39, 177–197 (1998). 69. E. Post et al., Ecological dynamics across the Arctic associated with recent climate change, Science 325, 53. R. J. Morley, M. Bush, J. Flenley, W. Gosling, in 1355–1358 (2009). Tropical Rainforest Responses to Clim. Change, M. B. Bush, J. R. Flenley, W. D. Gosling, Eds. (Springe, 70. S. E. Gilman, M. C. Urban, J. Tewksbury, G. W. Verlag Berlin Heidelberg, 2011), pp. 1–34. Gilchrist, R. D. Holt, A framework for community interactions under climate change, Trends Ecol. Evol. 54. K. J. Willis, G. M. MacDonald, Long-Term Ecological 25, 325–31 (2010). Records and Their Relevance to Climate Change Predictions for a Warmer World, Annu. Rev. Ecol. 71. G.-R. Walther, Community and ecosystem responses Evol. Syst. 42, 267–287 (2011). to recent climate change, Philos. Trans. R. Soc. London Ser. B 365, 2019–2024 (2010). 55. K. J. Willis, K. D. Bennett, S. L. Burrough, M. Macias- Fauria, C. Tovar, Determining the response of 72. E. Post, Ecology of climate change. The importance of African biota to climate change: using the past to biotic interactions (Princeton University Press, model the future, Philos. Trans. R. Soc. London Ser. B Princeton, New Jearsey, 2013). 368 (2013), doi:10.1098/rstb.2012.0491. 73. J. A. Pounds et al., Widespread amphibian extinctions 56. J. W. Williams, B. N. Shuman, T. Webb, P. J. Bartlein, P. from epidemic disease driven by global warming, L. Leduc, Late-Quaternary vegetation dynamics in Nature 439, 161–167 (2006).

130 | Chapter IV

74. J. R. Rohr, T. R. Raffel, Linking global climate and 89. J. M. Sunday, A. E. Bates, N. K. Dulvy, Global analysis temperature variability to widespread amphibian of thermal tolerance and latitude in ectotherms, declines putatively caused by disease, Proc. Natl. Proc. R. Soc. London Ser. B 278, 1823–1830 (2011). Acad. Sci. U.S.A. 107, 8269–8274 (2010). 90. S. Z. Dobrowski, A climatic basis for microrefugia: the 75. J. M. Diez et al., Will extreme climatic events facilitate influence of terrain on climate, Global Change Biol. biological invasions?, Front. Ecol. Environ. 10, 249– 17, 1022–1035 (2011). 257 (2012). 91. R. K. Colwell, G. Brehm, C. L. Cardelús, A. C. Gilman, J. 76. I. Durance, S. J. Ormerod, Evidence for the role of T. Longino, Global Warming, Elevational Range climate in the local extinction of a cool-water triclad, Shifts, and Lowland Biotic Attrition in the Wet J. North Am. Benthol. Soc. 29, 1367–1378 (2010). Tropics, Science 322, 258–261 (2008). 77. J. F. McLaughlin, J. J. Hellmann, C. L. Boggs, P. R. 92. J. Pöyry, M. Luoto, R. K. Heikkinen, M. Kuussaari, K. Ehrlich, Climate change hastens population Saarinen, Species traits explain recent range shifts of extinctions., Proc. Natl. Acad. Sci. U.S.A. 99, 6070–4 Finnish butterflies, Global Change Biol. 15, 732–743 (2002). (2009). 78. L.-M. Chevin, R. Lande, G. M. Mace, Adaptation, 93. A. L. Angert et al., Do species’ traits predict recent Plasticity, and Extinction in a Changing shifts at expanding range edges?, Ecoll. Lett. 14, Environment: Towards a Predictive Theory, PLoS 677–689 (2011). Biol. 8, e1000357 (2010). 94. C. A. Schloss, T. A. Nuñez, J. J. Lawler, Dispersal will 79. M. W. Tingley, W. B. Monahan, S. R. Beissinger, C. limit ability of mammals to track climate change in Moritz, Birds track their Grinnellian niche through a the Western Hemisphere, Proc. Natl. Acad. Sci. U.S.A. century of climate change, Proc. Natl. Acad. Sci. U.S.A. 109, 8606–8611 (2012). 106, 19637–16943 (2009). 95. L. Gillson, T. P. Dawson, S. Jack, M. A. McGeoch, 80. A. Ordonez, J. W. Williams, Projected climate Accommodating climate change contingencies in reshuffling based on multivariate climate- conservation strategy, Trends Ecol. Evol. 28, 135– availability, climate-analog, and climate-velocity 142 (2013). analyses: implications for community 96. W. B. Foden et al., Identifying the World’s Most disaggregation, Clim. Change 119, 659–675 (2013). Climate Change Vulnerable Species: A Systematic 81. S. T. Jackson, J. L. Betancourt, R. K. Booth, S. T. Gray, Trait-Based Assessment of all Birds, Amphibians and Ecology and the ratchet of events: Climate Corals, PLoS ONE 8, e65427 (2013). variability, niche dimensions, and species 97. L. P. Shoo et al., Engineering a future for amphibians distributions, Proc. Natl. Acad. Sci. U.S.A. 106, under climate change, J. Appl. Ecol. 48, 487–492 19685–19692 (2009). (2011). 82. M. L. Logan, R. K. Huynh, R. A. Precious, R. G. 98. P. Williams et al., Planning for Climate Change: Calsbeek, The impact of climate change measured at Identifying Minimum-Dispersal Corridors for the relevant spatial scales: new hope for tropical lizards, Cape Proteaceae, Conserv. Biol. 19, 1063–1074 Global Change Biol. 19, 3093–3102 (2013). (2005). 83. C. Rahbek, The role of spatial scale and the 99. R. Altwegg et al., Novel methods reveal shifts in perception of large-scale species-richness patterns, migration phenology of barn swallows in South Ecoll. Lett. 8, 224–239 (2005). Africa, Proc. R. Soc. London Ser. B 279, 1485–1490 84. S. T. Gray, J. L. Betancourt, S. T. Jackson, R. G. Eddy, (2012). Role of multidecadal climate variability in a range 100. N. E. Zimmermann et al., Climatic extremes improve extension of Pinyon Pine, Ecology 87, 1124–1130 predictions of spatial patterns of tree species, Proc. (2006). Natl. Acad. Sci. U.S.A.106, 19723–19728 (2009). 85. B. W. Brook, N. S. Sodhi, C. J. a Bradshaw, Synergies 101. M. V. Lomolino, in Frontiers of Biogeography: new among extinction drivers under global change, directions in the geography of nature, M. V. Lomolino, Trends Ecol. Evol. 23, 453–60 (2008). L. R. Heaney, Eds. (Sinauer Associates, Inc., 86. K. L. Laidre et al., Quantifying the sensitivity of arctic Sunderland, Massachussets, 2004), pp. 293–296. marine mammals to climate-induced habitat change, 102. K. J. Feeley, M. R. Silman, The data void in modeling Ecology 18, S97–S125 (2008). current and future distributions of tropical species, 87. J. Overpeck, C. Whitlock, B. Huntley, in Paleoclimate, Global Change Biol. 17, 626–630 (2011). Global Change and the Future, K. D. Alverson, R. S. 103. A. Felton et al., Climate change, conservation and Bradley, T. F. Pedersen, Eds. (Springer, Berlin, 2003), management: an assessment of the peer-reviewed pp. 81–103. scientific journal literature, Biodiversity and 88. R. B. Huey et al., Why tropical forest lizards are Conservation 18, 2243–2253 (2009). vulnerable to climate warming Why tropical forest 104. B. R. Scheffers, L. N. Joppa, S. L. Pimm, W. F. lizards are vulnerable to climate warming, Proc. R. Laurance, What we know and don’t know about Soc. London Ser. B 276, 1939–1948 (2009).

Dimensions of climate change | 131

Earth's missing biodiversity, Trends Ecol. Evol. 27, Supplementary Material 501–510 (2012). 105. K. Fraedrich, F. W. Gerstengarbe, P. C. Werner, Materials and Methods Climate Shifts during the Last Century, Clim. Change 50, 405–417 (2001). Figs. S1–S7 106. W. W. Hargrove, F. M. Hoffman, Potential of Tables S1 and S2 multivariate quantitative methods for delineation Supplementary References (108–150) and visualization of ecoregions, Environmental Management 34, S39–S60 (2004). 107. Materials and Methods are available as Supplementary Material.

132 | Chapter IV

Materials and Methods ECHO-G from the Meteorological Institute of the University of Bonn, Meteorological Research 1. Climatic data Institute of the Korea Meteorological Admin- istration (KMA), and Model and Data Group; We used 30-year averages of mean annual MIROC3.2 (medium resolution) from the Center temperature and total annual precipitation at a for Climate System Research (University of spatial resolution of 10 minutes. Baseline Tokyo), National Institute for Environmental (1961-90) climate data were obtained from the Studies, and Frontier Research Center for Glob-

Climatic Research Unit (109). For 2081-00, we al Change; IPSL-CM4 from the Institut Pierre sourced 15 downscaled General Circulation Simon Laplace; and UKMO-HadCM3 from the Models (GCM) (110) for the A1B greenhouse Hadley Centre for Climate Prediction. gas emissions scenarios from the World Cli- Time-series data, composed of monthly da- mate Research Programme’s Coupled Model ta for 30-year periods, were sourced from the Intercomparison Project phase 3 multi-model CRU (112) at 0.5° resolution for the baseline dataset projections. We followed the methodol- period (1961-90), and from the World Data ogy described by Garcia and colleagues (111) to Center for Climate (http://cera-www.dkrz.de/) build multi-model ensembles of similar GCMs. (113–126) for IPCC AR4 simulations for the The main ensemble (GCM1) used in the study same 15 General Circulation Models (2069-98; combined nine models and the alternative en- run 1) at their native resolution. The same semble (GCM2) used for sensitivity analysis ensembles of nine (GCM1) and six (GCM2) combined six models. models were derived after resampling the pro- The nine models in the main ensemble jections to the lowest common resolution (4° x used in the study (GCM1) are: BCCR-BCM2.0 5°) using bilinear interpolation. The same from the Bjerknes Centre for Climate Research; resampling was applied to the baseline time- CGCM3.1(T47) from the Canadian Centre for series. Calculations were performed in R ver- Climate Modelling and Analysis; CNRM-CM3 sion 12.2 and 12.5.1 (127) and mapping in from the Centre National de Recherches ArcGIS 9.3 (128). Météorologiques, Météo-France; CSIRO-MK3.0 from the Commonwealth Scientific and Indus- 2. Computing climate change metrics for tempera- trial Research Organisation (CSIRO) Atmos- ture and precipitation pheric Research; GFDL-CM2.0 from the Geo- physical Fluid Dynamics Laboratory (GFDL), Six climate change metrics were computed for National Oceanic and Atmospheric, Administra- temperature and precipitation. Whereas tem- tion (NOAA), U.S. Department of Commerce; perature and precipitation can each have direct GFDL-CM2.1 from the Geophysical Fluid Dy- physiological impacts or indirect impacts on namics Laboratory (GFDL), National Oceanic habitat and resource requirements, for some and Atmospheric, Administration (NOAA), U.S. species it is the interaction between the two Department of Commerce; INM-CM3.0 from the that becomes critical (129). When different Institute for Numerical Mathematics; MRI- methods exist to compute a given metric (see CGCM2.3.2 from the Meteorological Research Supplementary Table 1), we selected methods Institute; and PCM from the National Center for commonly used in biodiversity assessments. Atmospheric Research. Standardised local anomalies The six GCMs in the alternative ensemble To quantify local changes in magnitude we used (GCM2) are: CCSM3 from the National Center standardised anomalies. For each cell, we com- for Atmospheric Research; ECHAM5/MPI-OM puted the sum of the standardised temperature from the Max Planck Institute for Meteorology;

Dimensions of climate change | 133

and precipitation anomalies between the base- classify each cell in both the baseline and future line and future periods following Williams and time periods. Each cell was checked against the colleagues (22). The temporal differences for Köppen-Geiger rules for classification (130), each climatic parameter were standardised based on temperature and precipitation, and using the local inter-annual (baseline) standard assigned a climatic class in R (127). To quantify deviation for that parameter. Where the stand- the change in area of analogous climates (clas- ard deviation was zero, we set the standardised ses), we computed the change in area occupied anomaly to the value of the anomaly. The base- by a given class between the baseline and fu- line standard deviation was computed using the ture periods (34, 105). These calculations were time-series climatic data for 1961-90 at half- performed using the raster package in R (131), degree resolution and resampled to t10 taking into account the curvature of the earth. minutes using the nearest neighbour method. For a given cell with a given climate class, the change in area of analogous climates represents Change in probability of local climate extremes the ratio (in percentage) of the difference be- As daily data were unavailable, the analysis of tween future and baseline area of that class to extreme climates (108) was based on the time- the baseline area of the same class. Positive series monthly data (see "Climatic data" above). values indicate gains in area, negative values For each cell, we calculated the 5th and 95th indicate losses, and null values reflect no percentiles of the distributions of precipitation change. and temperature, respectively, over the 30 years of the baseline period. For each cell, the Novel climates percentiles of the future distributions over a Following Williams and colleagues (22), we 30-year period that correspond to the extreme computed the dissimilarities between baseline baseline values were then computed. These and future climates. For each cell, we computed percentiles correspond to the probability that the standardised Euclidean distances between the historical extremes will be exceeded (in the the future climate of that cell and the baseline case of temperature, the probability that the climates of all cells, retaining the minimum of baseline 95th percentile will be exceeded is those distances. Similarly to the standardised given by one minus the probability calculated). anomalies above, the inter-annual standard To obtain a measure of the probability of occur- deviation for each variable was used for the rence of either of the two extreme events (tem- standardisation. These calculations were per- perature and precipitation), for each cell we formed using the analogues R package (132). summed the two probabilities and subtracted the product of the two probabilities to avoid Change in distance to analogous climates counting probabilities twice. The future proba- For a given cell, we calculated the distances to bility of historical extreme climates was then all cells with analogous climates in the baseline subtracted from the probability of baseline period, i.e., belonging to the same climate class extreme climates to obtain the changes in the of the given cell as defined above (34, 105). We probability of extreme warm and dry climates. also calculated the distances to all cells that Positive values indicate increased probability in were projected to experience analogous cli- the future, whereas negative values indicate a mates in the future. Using the raster package in decrease. R (131) we computed, for each cell, the median of the great-circle distances below the 10th Change in area of analogous climates percentile, for both baseline and future periods, We used a modified version of the Köppen- and mapped the change over time. We show Geiger climatic classification (Fig. S4) (130) to distances in km. Negative values indicate a

134 | Chapter IV

temporal decrease in distance, whereas posi- precipitation separately, and subtracted the tive values indicate an increase. future from the baseline probabilities.

Climate change velocity Change in area of analogous climates We computed the climate change velocity Because the method described above could not (km/year; 8) as the ratio of the temporal cli- be applied to each climatic parameter individu- mate gradient (units of climate parame- ally, we followed the approach of Ackerly and ter/year) to the spatial climate gradient (units colleagues (35) instead. We constructed histo- of climate parameter/km). The two climate grams of the baseline climate space for each parameters were rescaled (to range from zero parameter, setting the width of the bins to the to one) and averaged to obtain a multi-variate lower quartile of the distribution of the inter- climate parameter. The temporal gradient was annual variability in the baseline period over given by the local difference between baseline the study area. Due to a highly skewed distribu- and future values of the multi-variate parame- tion for precipitation, we used logarithmic ter. We followed Sandel and colleagues (15) to values for this parameter. This procedure re- calculate the spatial gradient as the local slope sulted in 64 bins for mean annual temperature of the multi-variate parameter surface for the and 6 bins for annual precipitation. For both baseline period. To avoid dividing by zero or observed and projected climate, each cell could values close to zero, the distribution of the thus be assigned to a specific bin of tempera- spatial gradient values was truncated at the ture and precipitation. The computation of the lower end. All values below 0.00005 rescaled change in area of analogous climates was done units/km were set to 0.00005 rescaled as described above for combined temperature units/km. We show velocities in km/year. and precipitation changes.

3. Computing climate change metrics for each cli- Novel climates matic parameter individually The method described above was applied individually to each climatic parameter. Different climatic parameters, including temperature, precipitation and derived bioclimatic Change in distance to analogous climates parameters, can be important in different situa- For each climatic parameter we used the same tions and for different species. We thus also computation described above for the two pa- computed all metrics for each parameter indi- rameters. However, similarly to the computa- vidually. Comparing both sets of results pro- tion of changes in area of analogous climates, vides a better understanding of which parame- we used the climate classification based on ter drives the patterns of combined tempera- histograms. ture and precipitation change. Climate change velocity Standardised local anomalies The same method described for the two param- The method described above for local changes eters was applied to the (non-scaled) values of in magnitude of temperature and precipitation each climatic parameter separately. was also applied individually to each climatic parameter. 4. Correlation among climate change metrics

Change in probability of local climate extremes To test for correlation among metrics, we com- We used the local probabilities of extreme cli- puted pair-wise spatial correlation tests. We mates calculated above for temperature and used the SpatialPack package in R (133) and adapted the modified.ttest function so as to

Dimensions of climate change | 135

perform Spearman correlations. We performed puted the same overlap but replacing climate modified versions of the test based on correc- change velocity with the change in distance to tions of the number of degrees of freedom to analogous climates, and replacing standardised account for existing spatial auto-correlation. local anomalies with the change in probability of local climate extremes (Fig. S7). 5. Overlap among climate change metrics 6. Analysis of sensitivity to alternative climate Areas of simultaneous change in multiple met- models rics were identified by overlapping three metrics selected to depict three main facets of risk The climate change metrics were computed for species: standardised local anomalies, using a multi-model ensemble of nine GCMs change in area of analogous climates, and cli- that were considered to form a cluster of GCMs mate change velocity. The values for each indi- similar to the median of a total of 15 GCMs vidual metric were reclassified on a two-class (111) (GCM1). We repeated the analyses for scale using as threshold either the median of the alternative cluster of six GCMs for the same the distribution of values, or zero for metrics emissions scenario (GCM2). We computed the with positive and negative values. Our maps of same six metrics for temperature and precipita- overlap use a three-dimensional colour gradi- tion (Fig. S1 and S2), the comparison across ent to depict the range of combinations among climate regions of the proportion of area affect- the three metrics. To test whether the results ed by large changes (Fig. S5), and the overlap of are sensitive to the choice of metrics, we com- the three selected metrics (Fig. S6).

136 | Chapter IV

Supplementary Figures

Figure S1 | Projected global climate change according to six metrics for the alternative multi- model ensemble (GCM2). The maps show projections of change in mean annual temperature and total annual precipitation between the baseline and the end-of-century alternative multi-model ensemble GCM2 for the A1B emissions scenario. The classes were defined using quantiles and reflect a gradient from small or positive (dark blue) to large or negative (dark brown) changes. Local anomalies and novel climates values were converted to logarithmic scale for visualization.

Figure S2 | Comparison of climate change metrics for two multi-model climate ensembles. Boxplots are shown for six climate change metrics computed using two alternative multi-model climate ensembles (GCM1 and GCM2). Outliers are excluded.

Dimensions of climate change | 137

Figure S3 | Quantile distribution of climate change metrics. For each metric, the maps show the distribution of the three classes of change (below the 25th percentile, between the 25th and 75th percentiles, and above the 75th percentile of the global distribution of values of each metric), and the barplots the proportion of area occupied by the three classes within each broad Köppen-Geiger climate region. The results shown are for the main multi-model ensemble (GCM1).

138 | Chapter IV

Figure S4 | Köppen-Geiger climatic classification for the baseline period. Each location is assigned to one of the Köppen-Geiger climatic classes.

Dimensions of climate change | 139

Figure S5 | Comparison of the exposure of the world’s climatic regions to different climate change components for two multi-model climate ensembles. For each Köppen-Geiger broad climatic region depicted in the map, the star plots show the percentage of area exposed to values above the 75th percentile of the distribution of values for six climate change metrics. The star plots for the main (GCM1) and alternative (GCM2) multi-model ensembles are superimposed.

140 | Chapter IV

Figure S6 | Spatial overlap among climate change metrics for the alternative multi-model ensemble (GCM2). The three metrics used (standardized local anomalies, change in area of baseline-analogous climates and climate change velocity) capture three main facets of climate-induced risks to biodiversity.

Dimensions of climate change | 141

Figure S7 | Comparison of the spatial overlap among climate change metrics using different combinations of metrics. The maps show the overlap among metrics capturing three components of change in the magnitude of local climate, regional availability of climates, and regional position of climates for GCM1. For comparison with the combination of metrics used in the text, the climate change velocity is replaced with the change in distance to analogous climates (a), and the standardized local anomalies are replaced with the change in probability of local climate extremes (b).

142 | Chapter IV

Supplementary Tables

Table S1 | Examples of climate change metrics used in published studies. Descriptions refer to changes in climate between two time periods, t1 and t2. The list is not meant to be exhaustive, but rather shows the diversity of metrics by covering a diversity of approaches and computation methods.

Dimension Metric Ref. Changes in local climate Magnitude Changes in average climate: . anomalies for a climatic parameter between t1 and t2, sum of normalized anoma- (13, 26, 32, lies for all parameters, or weighted sum of anomalies for all parameters. 134–137) . the Euclidean distance between the t1 and t2 values of a climatic parameter, (9, 22, 23) standardized by the standard deviation of t1 inter-annual variability; the squared sum of standardized Euclidean distances for multiple parameters; or the weighted sum of Euclidean distances measured along the axes of a Principal Component Analysis performed on climate. Changes in climate extremes: . the probability of occurrence in t2 of the most extreme event in t1 for a given (25, 108, parameter; or weighted average of probabilities for multiple parameters. 138) . additional number of occurrences in t2 of the “1 in 20 years” extreme event of t1. (139) . the variation between t1 and t2 in the length of the period where a given climatic (140) parameter is above or below quantile or threshold. . the average distance of t2 values for a climatic parameter from those of t1, where (9) distance is measured as standard deviations from the mean of t1; extreme values are defined as exceeding two standard deviations of the baseline mean. . the number of sub-units of time in t2 with extreme patterns of a climatic parame- (9) ter, i.e. exceeding a pre-determined number of standard deviations departing from the mean of t1. . the change in inter- or intra-annual variability for a climatic parameter between (26, 32, t1 and t2. 140) . the inter-annual variation in maximum (minimum) values for a climatic parame- (16) ter, calculated as the difference in the 95th (5th) quantiles of maximum (minimum) values between t1 and t2. Timing Changes in climate seasonality: . difference in units of time in the date of climatic events. (33) . shifts in seasonal timing of a climatic parameter, measured by the ratio of the (12) long term trend for that parameter to the seasonal rate of change in the same parameter. Position Velocity of climate change: . the ratio of the rate of temporal change in a given climate parameter to the spa- (8, 18, 80, tial gradient for the same parameter (km/year). 134)

Dimensions of climate change | 143

Table S1 (continued) Dimension Metric Ref. Changes in regional climate Availability Degree of similarity in climate: . the minimum of the Euclidean distances between a given t1 (t2) climate value (22, 31, 80, and all t2 (t1) climate values in a region; minimum distances above pre-defined 134, 141) thresholds represent disappearing (novel) climates. . similarity between a given t1 climate and all t2 climates, calculated using percen- (142) tiles of the t1 climatic range but allowing for negative values to indicate climates outside the range. Change in area of analogous climates: . the change over time in area experiencing climates that differ less than set (7, 20, 34) thresholds. . the change over time in area experiencing climates that belong to the same cli- (35, 80) mate domain in k-variate histograms of t1 climate. . the change over time in area experiencing climates with the same cluster mem- (30, 106, bership in hierarchical clustering of t1 climates or multivariate geographic clus- 143, 144) tering of climates through time. . the change over time in area experiencing climates that belong to the same t1 (105, 137, climate class defined according to classification rules (e.g. Köppen-Geiger climate 145–150) classification). . areas of overlap and non-overlap between t1 and t2 climate profiles, where (10, 27– climate profiles are defined through Principal Components Analysis of climate 29) parameters across the region; t2 climates inside (outside) the t1 climate profile are persisting (novel) climates, and t1 climates outside the t2 climate profile are disappearing climates. Position Distance between analogous climates: . the change over time in the average, minimum or a given percentile of the geo- (34) graphical distances between a given climate and all climates that differ less than set thresholds. . the geographical distance between a given t1 climate and the t2 climate with the (80, 134) highest similarity (minimum Euclidean distance) to t1 climate. Direction to analogous climates: . the bearing between a given t1 climate and the t2 climate with the highest simi- (80, 134) larity (minimum Euclidean distance) to t1 climate. . or average direction between a given climate in t1 and all locations that in t2 (34) have climates that differ less than set thresholds from that climate. Velocity of climate change: . the ratio of the distance between analogous climates in t1 and t2 to the length of (19) the time interval.

Table S2 | Pair-wise correlations between climate change metrics. The correlation statistics for each pair of metrics is given above the diagonal line; the p-value and degrees of freedom, corrected to take into account spatial auto-correlation, are given below the diagonal line.

anomalies extremes area novel distance velocity anomalies 0.31 0.22 0.08 -0.03 -0.13

extremes P=6.68E- 0.31 0.21 0.08 -0.15 30, dof=1296 area P=5.67E- P=1.88E- 0.28 0.11 -0.03 18, dof=1497 04, dof=134 novel P=1.41E- P=6.92E- P=7.55E- 0.10 -0.06 34, dof=26046 27, dof=2634 29, dof=1485 distance P=2.41E- P=4.08E- P=1.56E- P=7.79E- 0.07 03, dof=13126 03, dof=1404 05, dof=1595 30, dof=11966 velocity P=2.50E- P=1.15E- P=6.09E- P=3.66E- P=4.67E- 12, dof=2742 02. dof=273 01, dof=373 04, dof=3801 05, dof=3118

144 | Chapter IV

120. Volodin, IPCC DDC AR4 INM-CM3.0 SRESA1B run1. Supplementary References World Data Center for Climate. CERA-DB “INM_CM3.0_SRESA1B_1,” (2005). 86, 1124–1134 (2005). 121. IPCC DDC AR4 NCAR-CCSM3 SRESA1B run1. World 109. M. New, D. Lister, M. Hulme, I. Makin, A high- Data Center for Climate. CERA-DB resolution data set of surface climate over global “NCAR_CCSM3_SRESA1B_1,” http://cera- land areas, Climate Research 21, 1–25 (2002). www.dkrz.de/WDCC/ui/Compact.jsp?acronym=NCAR 110. K. Tabor, J. W. Williams, Globally downscaled _CCSM3_SRESA1B_1 (2005). climate projections for assessing the conservation 122. Roeckner, IPCC DDC AR4 ECHAM5/MPI-OM impacts of climate change., Ecol. Appl. 20, 554–65 SRESA1B run1. World Data Center for Climate. (2010). CERA-DB “EH5_MPI_OM_SRESA1B_1,” (2005). projections of climatically suitable areas for African 123. Min, IPCC DDC AR4 ECHO-G SRESA1B run1. World vertebrates, Global Change Biol. 18, 1253–1269 Data Center for Climate. CERA-DB (2012). “ECHO_G_SRESA1B_1,” http://cera- 112. I. H. University of East Anglia Climatic Research www.dkrz.de/WDCC/ui/Compact.jsp?acronym=ECHO Unit (CRU) [Phil Jones, CRU Time Series (TS) high _G_SRESA1B_1 (2005). resolution gridded datasets, [Internet] (2008). 124. Nozawa, IPCC DDC AR4 CCSR-MIROC3.2_(med-res) 113. Collier, IPCC DDC AR4 CSIRO-Mk3.0 SRESA1B run1. SRESA1B run1. World Data Center for Climate. World Data Center for Climate. CERA-DB CERA-DB “MIROC3.2_mr_SRESA1B_1,” http://cera- “CSIRO_Mk3.0_SRESA1B_1,” (2005). 125. Denvil, IPCC DDC AR4 IPSL-CM4 SRESA1B run1. 114. N.A., IPCC DDC AR4 GFDL-CM2.1 SRESA1B run1. World Data Center for Climate. CERA-DB World Data Center for Climate. CERA-DB “IPSL_CM4_SRESA1B_1,” http://cera- “GFDL_CM2.1_SRESA1B_1,” (2005). 126. Lowe, IPCC DDC AR4 UKMO-HadCM3 SRESA1B 115. Flato, IPCC DDC AR4 CGCM3.1-T47_(med-res) run1. World Data Center for Climate. CERA-DB SRESA1B run1. World Data Center for Climate. “UKMO_HadCM3_SRESA1B_1,” http://cera- CERA-DB “CGCM3.1_T47_SRESA1B_1,” (2005). 127. R Development Core Team, R: A language and 116. D. Salas, IPCC DDC AR4 CNRM-CM3 SRESA1B run1. environment for statistical computing (2010) World Data Center for Climate. CERA-DB (available at URL http://www.R-project.org). “CNRM_CM3_SRESA1B_1,” (2005). 129. C. M. McCain, R. K. Colwell, Assessing the threat to 117. N.A., IPCC DDC AR4 NCAR-PCM SRESA1B run1. montane biodiversity from discordant shifts in World Data Center for Climate. CERA-DB temperature and precipitation in a changing climate, “NCAR_PCM_SRESA1B_1,” (2005). 130. M. C. Peel, B. L. Finlayson, T. A. McMahon, Updated world map of the Köppen-Geiger climate 118. N.A., IPCC DDC AR4 MRI-CGCM2.3.2 SRESA1B run1. classification, Hydrology and Earth System Sciences World Data Center for Climate. CERA-DB 11, 1633–1644 (2007). “MRI_CGCM2.3.2_SRESA1B_1,” (2005). analysis and modeling with raster data. R package version >=2.11. http://CRAN.R- 119. N.A., IPCC DDC AR4 BCCR_BCM2.0 SRESA1B run1. project.org/package=raster (2012). World Data Center for Climate. CERA-DB “BCCR_BCM2.0_SRESA1B_1,” (2006). 133. F. Osorio, R. Vallejos, F. Cuevas, SpatialPack: Package for analysis of spatial data. R package

Dimensions of climate change | 145

version 0.2. http://CRAN.R- 142. J. Elith, M. Kearney, S. Phillips, The art of modelling project.org/package=SpatialPack (2012). range-shifting species, Methods in Ecology and Evolution 1, 330–342 (2010). 134. S. Veloz et al., Identifying climatic analogs for Wisconsin under 21st-century climate-change 143. E. Saxon, B. Baker, W. Hargrove, F. Hoffman, C. scenarios, Clim. Change 112, 1037–1058. Zganjar, Mapping environments at risk under different global climate change scenarios, Ecoll. Lett. 135. C. Enquist, D. Gori, T. N. C. in N. M. Climate Change 8, 53–60 (2005). Ecology & Adaptation Program, A climate change vulnerability assessment for biodiversity in New 144. F. M. Hoffman, W. W. Hargrove, D. J. Erickson, R. J. Mexico, Part I: Implications of Recent Climate Oglesby, Using Clustered Climate Regimes to Change on Conservation Priorities in New Mexico Analyze and Compare Predictions from Fully (2008), p. 69. Coupled General Circulation Models, Earth Interactions 9, 1–27 (2005). 136. I. Mahlstein, R. Knutti, Regional climate change patterns identified by cluster analysis, Clim. Dyn. 35, 145. M. de Castro, C. Gallardo, K. Jylha, H. Tuomenvirta, 587–600 (2010). The use of a climate-type classification for assessing climate change effects in Europe from an ensemble 137. B. Baker, H. Diaz, W. Hargrove, F. Hoffman, Use of of nine regional climate models, Clim. Change 81, the Köppen–Trewartha climate classification to 329–341 (2007). evaluate climatic refugia in statistically derived ecoregions for the People’s Republic of China, Clim. 146. C. Beck, J. Grieser, M. Kottek, F. Rubel, B. Rudolf, Change 98, 113–131 (2009). Characterizing Global Climate Change by means of Köppen Climate Classification, Klimastatusbericht 138. P. Frich et al., Observed coherent changes in 2005 , 139–149 (2005). climatic extremes during the second half of the twentieth century, Climate Research 19, 193–212 147. S. Feng et al., Evaluating observed and projected (2002). future climate changes for the Arctic using the Köppen-Trewartha climate classification, Clim. Dyn. 139. M. B. Baettig, M. Wild, D. M. Imboden, A climate 38, 1359–1373 (2011). change index: Where climate change may be most prominent in the 21st century, Geophys. Res. Lett. 34, 148. H. F. Diaz, J. K. Eischeid, Disappearing “alpine L01705 (2007). tundra” Köppen climatic type in the western United States, Geophys. Res. Lett. 34, L18707 (2007). 140. C. Tebaldi, K. Hayhoe, J. M. Arblaster, G. a. Meehl, Going to the Extremes, Clim. Change 79, 185–211 149. J. Kalvová, T. Halenka, K. Bezpalcová, I. Nemešová, (2006). Köppen Climate Types in Observed and Simulated Climates, Studia Geophysica et Geodaetica 47, 185– 141. D. R. Roberts, A. Hamann, Predicting potential 202 (2003). climate change impacts with bioclimate envelope models: a palaeoecological perspective, Glob. Ecol. 150. M. Wang, J. E. Overland, Detecting Arctic Climate Biogeogr. 21, 121–133 (2012). Change Using Köppen Climate Classification, Clim. Change 67, 43–62 (2004).

Chapter V

Do projections from bioclimatic envelope models and climate change metrics match?

RAQUEL A. GARCIA, MAR CABEZA, RES ALTWEGG, AND MIGUEL B. ARAÚJO Manuscript in review

Do projections from bioclimatic envelope models and climate change metrics match?

RAQUEL A. GARCIA1,2,3, MAR CABEZA4, RES ALTWEGG5, and MIGUEL B. ARAÚJO1,2,3,6

Manuscript in review

Abstract

Bioclimatic envelope models are widely used to the metrics, qualitatively matched changes in the assess the potential exposure of species to future climatic suitability of grid cells for species, as climate changes. They do so by first describing the projected by the models. In turn, changes in the association between species' geographical distri- area and position of analogous climates, as meas- butions and climate, and then projecting this ured by the metrics, were indicative of changes in association under climate change scenarios. In the size and position of species' bioclimatic enve- turn, climate change metrics characterise the lopes, respectively. Agreement between the two spatio-temporal dynamics of climate. While less approaches was found for all taxa, although it was common, they have also been used as indicators of stronger for species with narrower climatic toler- potential threats and opportunities for biodiver- ances. When carefully implemented and inter- sity. However, it remains unclear whether the two preted, it seems that climate change metrics allow approaches provide qualitatively similar infer- inferences about the exposure of biodiversity to ences about species' exposure to climate changes. climate changes, independently of species' data. Here, we compared climate change metrics for Assessments based on metrics can thus fill an sub-Saharan Africa with bioclimatic envelope important gap left by existing methodologies, models for over two thousand species of amphibi- allowing inferences about the potential exposure ans, snakes, mammals and birds in the region. We of species that are unknown or poorly known. found that local climate anomalies, as projected by

150 | Chapter V

Introduction analyses of climate change exposure, i.e., climate-induced threats and opportunities for Earth's climate is known to have varied over species (see for review Garcia et al., in review). millennia, with recent decades witnessing rapid Bioclimatic envelope modelling has been exten- changes (Mann et al., 1998) that are set to con- sively used in climate change risk assessments tinue (IPCC, 2013). As temperature and precipi- for biodiversity at global (e.g., Thomas et al., tation regimes change over time, not only is the 2004; Hof et al., 2011) and regional scales (e.g., climate at any given locality altered, but climate Thuiller et al., 2006; Araújo et al., 2011). As- conditions are also redistributed at broader suming that species are at equilibrium with spatial scales. Across a given region, particular climate (a working assumption rather than a climate conditions may become more or less theoretical or empirical expectation; see for available, and may shift closer or farther in discussion Araújo & Peterson, 2012), these position. Such spatio-temporal dynamics of models use statistical associations between climate have been suggested to influence biodi- observed species occurrences and climate pa- versity, by causing local population declines rameters to define the envelope of climatically (Foden et al., 2007; Allen et al., 2010) that even- suitable areas for species across a given region. tually translate into changes in range area or Future suitability across space is then assessed position (Nogués-Bravo et al., 2010; Chen et al., depending on whether projected climates are 2011). Metrics of climate change that reflect within the defined envelope or not. The biocli- such spatio-temporal climate dynamics thus matic envelope modelling approach can thus help to explain the role of past climate changes provide statistical inferences of the exposure of in shaping biodiversity patterns, as well as to individual species to climate change, by quanti- forecast the exposure of biodiversity to ex- fying expected losses, gains or fragmentation of pected future climate changes. Indeed, climate climatically suitable areas, and shifts required change metrics have been used to assess the to track suitable climates (Midgley et al., 2003; potential threats and opportunities that species Heikkinen et al., 2009; Garcia et al., 2014). may face under future climates (e.g., Ohlemüller In turn, metrics of climate change quantify et al., 2006; Williams et al., 2007; Loarie et al., the level of exposure of geographical areas to 2009; Ackerly et al., 2010), and the conserva- climate changes, allowing for post hoc infer- tion measures required (e.g., Watson et al., ences about the potential implications of these 2013). Such metrics can be seen as an alterna- changes for biodiversity. Both changes at any tive to commonly used bioclimatic envelope given locality and changes in the distribution of models (Guisan & Zimmermann, 2000; Peter- climates across broader regions can be meas- son et al., 2011), yet it remains unclear whether ured (Fig. 1; for a review of existing metrics see the two approaches provide qualitatively simi- Garcia et al., in review). Different metrics depict lar assessments of species' exposure to climate distinct dimensions of climate change, from change. local changes in average or extreme climates The impacts of climate change on species (Katz et al., 2005; Williams et al., 2007), to re- depend on the degree to which species' ranges gional changes in the availability or position of are exposed to changes in relevant aspects of particular climate conditions (Ohlemüller et al., climate, their intrinsic vulnerability to these 2006; Ackerly et al., 2010). Each of these di- changes (Williams et al., 2008), and ultimately mensions of change is expected to impose dis- also on biotic interactions (Post, 2013) and tinct threats and opportunities for species synergies with other extrinsic threats (Brook et (Garcia et al., in review). At the local level, al., 2008). Both bioclimatic envelope models changes in average or extreme climatic condi- and climate change metrics can support spatial

Bioclimatic models vs. climate change metrics | 151

tions can be used as a proxy for the threat of provided by metrics can be compared with local demographic changes. At the regional species' spatio-temporal turnover data (mostly level, decreases or increases in the area avail- unavailable, but see Araújo et al., 2005) or, in able of analogous climatic conditions provide their absence, with the results of other model- indications on the degree of threat of loss or ling approaches (as done for bioclimatic enve- opportunity for gains of species' climate envelope models and mechanistic models, e.g., lopes across a given region, whereas shifts in Kearney et al., 2010). the location of climatic conditions can indicate Here, we provide the first comparison be- the need for geographical shifts of species' tween the inferences provided by climate envelopes. change metrics and by bioclimatic envelope models, for over two thousand species of sub- Saharan African amphibians, snakes, mammals and birds. If the two approaches provide similar information, we predict that 1) projected changes in climatic suitability from bioclimatic envelope models, at the grid cell level, are greater where local anomalies projected by the metrics are also greater, 2) projected losses or gains of species' envelopes across regions, according to bioclimatic envelope models, are larger where climate change metrics project shrinking or expanding climates, respectively, and 3) projected shifts in the position of spe-

cies' climate envelopes across regions, accord- Figure 1 | Metrics of local and regional changes in climate. Local metrics refer to changes occur- ing to bioclimatic envelope models, are greater ring over time at any given locality (dotted line in where climates are projected by the metrics to the schematic representation), and regional met- move farther. We test these predictions indi- rics refer to changes in the distribution of climate vidually for each taxon. As climate change met- conditions across broader regions (closed line). rics are applied independently of species' data, For example, the climate conditions in the upper right cell in t1 change to a darker shade in t2 (local we assess the influence of the breadth of spe- change), whereas the total area available across cies' bioclimatic envelopes on the level of the region with the cell's climate condition in t1 agreement between the two approaches. decreases from five cells in t1 to three cells in t2, and moves farther in space (regional changes). Materials and methods Climate change metrics have the potential to add useful information to traditional biocli- Bioclimatic Envelope Models matic envelope models (Ohlemüller, 2011), and may be the only alternative when species' iden- We used published baseline (1961-90) and tities or their locations are poorly known or not late-century (2081-2100) projections of cli- known at all (Garcia et al., in review). Yet, in matically suitable areas for sub-Saharan African order to ensure the appropriate use of climate species of birds (1,506), mammals (623), am- change metrics as an alternative or comple- phibians (284) and snakes (310), at one degree mentary tool to make inferences about the resolution (≈ 111 km x 111 km at the Equator) exposure of species to climate change, it is (Garcia et al., 2012). The models were built crucial to examine their outputs. The inferences with three climatic variables: mean tempera-

152 | Chapter V

tures of both warmest and coldest months, and mate relative to the species' bioclimatic enve- annual precipitation. Baseline data for these lope breadth. In turn, projections by the models variables were from the Climate Research Unit of changes in the area or position of the biocli- (New et al., 2002), and future projections from matic envelope of a given species are contin- a multi-model ensemble of nine General Circu- gent on the regional availability and location of lation Models (Meehl et al., 2007; Tabor & Wil- the climatic conditions characterising the enve- liams, 2010) under the A1B emissions scenario lope. (see Garcia et al., 2012 for detailed methods). Local anomalies were estimated by com- The projections used in our study reflected the puting, for each cell, the multivariate Euclidean consensus obtained by computing the median distance between baseline and future climates. among seven modelling techniques, and as- Metrics of regional change relied on the defini- sumed unlimited dispersal of species. tion of analogous climates. Following Oh- Based on the species' bioclimatic envelopes lemüller and colleagues (2006), we considered projected for the baseline and future periods, climatic conditions to be analogous across we computed four types of changes: local space and time if they differed by less than pre- changes in climatic suitability, and losses, gains defined thresholds. To define the optimal and shifts of the species' bioclimatic envelopes. thresholds for each climate variable, we tested For each species, local (cell-based) change in a sequence of 20 thresholds spanning from half suitability was computed as the absolute value of the mean historical inter-annual variability of the difference between the baseline and (1961-90) across sub-Saharan Africa to ten future probabilities of climatic suitability of any times that value (Ackerly et al., 2010). For each given cell, to account for both potential im- threshold, we derived a classification of all cells provement and deterioration in suitability. across the study area, whereby cells that dif- Using the model projections in binary form, fered by less than the set thresholds for all envelope losses or gains of suitability were variables were assigned the same class. To quantified for each species as the proportion of select the threshold yielding the optimal classi- the baseline climate envelope projected to be fication, we compared the 20 classifications lost or gained in the future across sub-Saharan obtained to the Köppen-Geiger climatic classifi- Africa. Envelope shifts in position were meas- cation (Peel et al., 2007), using two approaches. ured for each species by the great-circle dis- With the ANOSIM test (Clarke & Warwick, tance between the centroids of baseline and 1994), we assessed with 999 permutations future climate envelopes. whether our classes of cells differed more between Köppen-Geiger classes than within the Climate change metrics same Köppen-Geiger class. With the True Skill Statistic (TSS) test (Allouche et al., 2006), we We used the selected climatic variables to com- assessed the accuracy of our classification to pute three climate change metrics: local anoma- discriminate between Köppen-Geiger classes. lies, regional changes in area of baseline- For each Köppen-Geiger class, we computed the analogous climates, and regional changes in probability that the climatic differences be- distance to baseline-analogous climates. These tween cells in that class and cells in different three metrics were selected as those most classes were greater than the set thresholds closely associated to the concepts underpinning (true negative fraction) and the probability that bioclimatic envelope models. Whether a given the differences between cells within the same locality is projected by the models to remain Köppen-Geiger class were smaller than the set suitable or become unsuitable for a species thresholds (true positive fraction). We then depends on the degree of local change in cli-

Bioclimatic models vs. climate change metrics | 153

computed the median TSS across all Köppen- bioclimatic envelope models co-varied with the Geiger classes. For regional metric calculations, dimensions of climate change measured with we thus used the threshold maximising the the metrics (Table 1). First, local (cell) changes ANOSIM and TSS statistics simultaneously. in climatic suitability for species were com- Using the optimal threshold, we identified, pared to local climate anomalies. Using the for each cell, all other cells across sub-Saharan median absolute changes in climatic suitability Africa with analogous climates in the baseline across species in each cell, we asked whether period. We repeated this procedure for the this median was greater in grid cells exposed to future period, and calculated, for each cell, the large climate anomalies than in grid cells ex- temporal change in area. Positive values indi- posed to small anomalies. The median value of cated expanding climates, negative values indi- anomalies across sub-Saharan Africa was used cated shrinking climates, and zero reflected no to differentiate large from small anomalies. change. We also calculated the change in dis- Second, changes in species' climate enve- tance to baseline-analogous climatic conditions. lopes for each taxon were compared to regional For each cell, we computed the median of the changes in climate. We asked whether the me- great-circle distances to all cells with analogous dian projected loss of climate envelopes across climates, in both baseline and future periods, species occurring in shrinking climates was and retained the change over time. Negative greater than the same median across species in values indicated that similar climates were expanding climates. We repeated the analysis projected to move closer, whereas positive for projected median gains of climate space, values indicated they were projected to move with the expectation that they would be greater farther. for species occurring in areas of expanding climates. Finally, we asked whether the median Comparison of bioclimatic envelope models and projected shift in position of envelopes was climate change metrics greater for species occurring in climates projected by the metrics to move farther than for For each taxon, we assessed whether the species occurring in climates moving closer. changes in climatic suitability projected by the

Table 1 | Comparisons between bioclimatic envelope models and metrics of climate change. The different temporal changes in climatically suitable areas for species, derived from bioclimatic envelope models for sub-Saharan African amphibians, snakes, mammals and birds, were each compared to relevant metrics of future climate change for sub-Saharan Africa. The arrows indicate whether the two approaches are expected to be positively related (arrows pointing in the same direction) or negatively related (arrows pointing in opposite directions).

Bioclimatic envelope models Metrics of climate change

Local (cell) absolute changes in climatic suit- Local anomalies (multivariate Euclidean dis- ↑↑ ability for species over time tance between baseline and future climate)

Proportion of species' climatic envelope lost Regional change in total area available of ↑↓ over time baseline-analogous climates over time

Proportion of species' climatic envelope Regional change in total area available of ↑↑ gained over time baseline-analogous climates over time

Distance between centroids of baseline and Regional change in median distance to base- future species' envelope ↑↑ line-analogous climates over time

154 | Chapter V

For each comparative analysis, we used between the two approaches to be greater, i.e., Wilcoxon signed-rank tests to assess the statis- larger, positive, effect sizes measured with tical difference between the two groups of grid Cliff's delta, for species with smaller climatic cells exposed to different degrees of climate tolerances. Species endemic to more restricted change. To assess the magnitude and direction ranges of climatic conditions likely depend of the differences, we also measured effect size. more strongly on tracking climates (Williams et We used Cliff's delta (Cliff, 1993), a non- al., 2007; Ohlemüller et al., 2008) than species parametric alternative measure of effect size with more generalist climate preferences. that is robust to violations of the normality assumption. Cliff's delta estimates the probability that a value selected from one of the groups Results being compared is greater than a value selected Metrics of climate change revealed that tropical from the other group, minus the reverse prob- areas of Africa were the most exposed to large ability. It varies from –1 to +1, with zero indi- local (cell) anomalies in mean temperatures of cating complete overlap between the distribu- the coldest and warmest months and annual tions of the two groups, and values farther from precipitation (Fig. 2a). In turn, bioclimatic enve- zero reflecting smaller overlap. The sign of the lope models forecast larger absolute changes in delta estimate reflects which group dominates. local climatic suitability for species in tropical In our tests, positive values indicated domi- areas extending into West Africa as well as the nance according to expectations (see Table 1), Ethiopian highlands (Fig. S1 in Supporting whereas negative values indicated dominance Information). For all four taxonomic groups, contrary to expectations. projections by bioclimatic envelope models and Sensitivity analysis to differences in species' enve- climate change metrics were consistent with lope properties our expectations that changes in local (cell) climatic suitability for species were greater As climate change metrics are applied inde- where climate anomalies were also greater (Fig pendently of species' data, we investigated 2b; Wilcoxon signed rank-test, p-values<0.05; whether the comparisons performed were Cliff's delta [confidence interval], 0.45 [0.40 to affected by the properties of species' biocli- 0.49] for birds, 0.49 [0.44 to 0.53] for mam- matic envelopes. We first assessed species' mals, 0.44 [0.39 to 0.48] for snakes, and 0.45 envelope breadth using the Outlying Mean [0.41 to 0.50] for amphibians). The same quali- Index analysis (OMI; Dolédec et al., 2000). The tative conclusion held when the comparison OMI analysis identifies the ordination axes that was done separately for positive and negative optimise the separation between species' oc- changes in local suitability (Fig. S2). currences, and quantifies the envelope position For regional metric calculations, we used and breadth for each species along those axes. the values of three times the inter-annual vari- Envelope breadth, or tolerance, is quantified as ability of each climatic parameter across sub- the dispersion of environmental conditions Saharan Africa as thresholds to define analo- occupied by species, with larger dispersion gous climates (Fig. S3). Regional metrics values indicating higher tolerance. We then showed a decrease in area available for most defined two groups of species across the four conditions across the study region, with the taxa with climatic envelope tolerance above exception of a narrow strip from West Africa to and below the median tolerance, and repeated the Ethiopian highlands where the prevailing the comparative tests above for each group climatic conditions were projected to expand in individually. We expected the correspondence

Bioclimatic models vs. climate change metrics | 155

Figure 2 | Comparison of projected changes in climatic suitability for species between areas exposed to different levels of climate change. The results from the bioclimatic envelope modelling for sub-Saharan African amphibians (n=284), snakes (n=310), mammals (n=623) and birds (n=1,506) were compared between groups of one-degree grid cells with different levels of climate change as measured with metrics of climate change. Metric-based projections of local anomalies (a) were grouped into small (blue) and large (orange) anomalies, and the model-based median absolute change in local (cell) climatic suitability across species was compared between areas of small and large anomalies (b). Metric-based projected changes in area of analogous climates (c) were grouped into shrinking (orange) or expanding (blue) climates, and the median proportion of climate envelope projected by the bioclimatic envelope models to be lost (d) or gained (e) was compared between areas of expanding and shrinking climates. Metric-based projected changes in distance to analogous climates (f) were grouped into areas where analogous climates are projected to move farther (orange) and closer (blue) in the future, and the median distance shifted by the species' climate envelope according to the models was compared between the two groups (g). The maps were drawn using quantile classification.

area over sub-Saharan Africa (Fig. 2c). These S4a-b). The projections from the bioclimatic expanding climatic conditions were character- envelope models built for the four taxonomic ised by high temperatures of the coldest and groups showed greater losses than gains of warmest months across the study area (Fig. climatic suitability between the baseline and

156 | Chapter V

future periods (Fig. S1). Median losses of cli- across the four taxa with more specialised climatic envelopes across species were greater in matic envelopes (Fig. 3). For all comparative Southern Africa and the Eastern African moun- tests involving local and regional metrics, Cliff's tains. By contrast, species in West Africa and delta values were higher (positive) for the the Sahel region were projected, on average, to group of species with wider climatic tolerance. gain higher percentages of their baseline climatic envelopes and undergo larger displace- ments of their envelopes. For all taxonomic groups, the median percentages of species' climate envelope lost were higher for species occurring in shrinking climates than for species occurring in expanding climates (Fig. 2d; Wilcoxon signed rank-test, p- values<0.05; Cliff's delta [confidence interval], 0.49 [0.41 to 0.57] for birds, 0.45 [0.37 to 0.53] for mammals, 0.41 [0.33 to 0.48] for snakes, and 0.45 [0.38 to 0.51] for amphibians). By contrast, median percentages of suitable climate area gained, according to the bioclimatic envelope models, were higher for species in areas of expanding climates (Fig. 2e; Wilcoxon signed rank-test, p-values<0.05; Cliff's delta [confidence interval], 0.27 [0.21 to 0.33] for birds, 0.39 [0.32 to 0.45] for mammals, 0.47 [0.41 to 0.53] for snakes, and 0.55 [0.48 to 0.62] Figure 3 | Comparison between bioclimatic envelope models and climate change metrics for amphibians). for groups of species with different climatic Regional metrics of change in the position tolerances. For sub-Saharan African vertebrates of climates generally revealed a tendency for (n=2,723), changes in areas of climatic suitability increased distances between similar climatic for species, projected by bioclimatic models, were compared between areas exposed to different conditions (Fig. 2f), but for the montane areas levels of climate change, as measured by metrics of Ethiopia, Eastern African and South Africa of climate change. The effect size estimations the distances to similarly colder conditions using Cliff's delta are shown here for the compara- were projected to decrease (Fig. S4c-d). In tive tests performed for two groups of species these montane areas, the median distances that individually: species with wider (open circles) and narrower (closed circles) climatic tolerances. the species' climate envelopes were projected Median absolute changes in local (cell) climatic by the models to shift were significantly smaller suitability for species were compared between than the median distances in the remaining areas of large and small local anomalies; median areas (Fig. 2g; Wilcoxon signed rank-test, p- projected losses and gains of climate envelope across species were compared between areas of values<0.05; Cliff's delta [confidence interval], shrinking and expanding climates, and median 0.26 [0.19 to 0.33] for birds, 0.26 [0.18 to 0.33] projected shifts of climate envelopes were com- for mammals, 0.35 [0.28 to 0.42] for snakes, pared between areas with climates moving farther and 0.35 [0.28 to 0.41] for amphibians). vs. closer. Lines to the left and right of the circles indicate the lower and upper confidence intervals The correspondence between the two ap- of the calculated Cliff's delta. proaches to assessing climate change exposure investigated here was stronger for the species

Bioclimatic models vs. climate change metrics | 157

Discussion contains the full range of suitable conditions for species (Peterson et al., 2011; Anderson, 2013). A critical first step towards understanding the At the same time, the reliance of climate change potential consequences of 21st century cli- metrics on climate data alone, independently of mates on biodiversity is to assess the extent to information on the biogeography of species which species or areas are likely to become ranges, limits their interpretation in two ways. exposed to climate changes (exposure sensu First, climate change metrics disregard any Dawson et al., 2011). Bioclimatic envelope information about the climatic ranges of indi- models are well suited to assessing the expo- vidual species. In a given cell, changes in cli- sure of species to climate changes (Dawson et mate can result in lost (or gained) climatic al., 2011; Moritz & Agudo, 2013), whereas cli- suitability for a species with a narrow climatic mate change metrics can help quantifying the envelope, while it may enable another species exposure of geographical areas and inferring with more generalist climatic preferences to about the exposure of species occurring in remain within (or outside) its climate envelope. those areas (Garcia et al., in review; Ohlemüller, Likewise, reduced availability of given climatic 2011). Although the two approaches have fun- conditions across a region pose a greater threat damental differences in the data that are re- to species that depend on such conditions than quired and the quantities that are measured, in to species that inhabit a broader variety of our study for sub-Saharan African vertebrates climatic conditions. These differences may they provided qualitatively similar assess- explain why we found a better match between ments. models and metrics for species with narrower Projections by bioclimatic envelope models bioclimatic envelopes (Fig. 3). Yet, alternative qualitatively matched projections by climate explanations may rest on the poorer model change metrics (Fig. 2). That is, grid cells where performance that is typical for species with bioclimatic models projected the greatest in- wide geographical (and likely climatic) ranges creases or decreases of climate suitability for (Stockwell & Peterson, 2002; Segurado & species were also the cells with greater climate Araújo, 2004) anomalies as described by the metrics. In turn, Second, in disregarding information about larger decreases or increases in area of analo- the geographical ranges of species, climate gous climates were indicative of larger losses change metrics are blind to associated meas- and gains of species' climate envelopes, respec- ures of species richness or complementarity. tively. Finally, the larger displacement of analo- That is, they identify areas where climate- gous climates was an indicator for larger shifts induced threats are expected to be greatest, of species' climate envelopes. irrespective of the conservation importance of Congruence between projections of biocli- those areas (e.g., Watson et al., 2013). One pos- matic envelope models and climate change sibility to bring in such information is to over- metrics is consistent with the view that metrics lay climate change metric outputs with layers of might be useful for assessing the exposure of species richness (e.g., Ohlemüller et al., 2008) biodiversity to climate changes. Interpreting or protected areas (e.g., Loarie et al., 2009; metric outputs as indicators of climatic threats Gillson et al., 2013), or to restrict the computa- and opportunities for species requires some of tion of metrics to areas of conservation impor- the same working assumptions underpinning tance (e.g., Beaumont et al., 2010; Wiens et al., bioclimatic envelope models. It needs to be 2010). By contrast, bioclimatic envelope mod- assumed that the selected climatic variables els add precision to priority setting in that they and their spatial and temporal scales are rele- also consider the numbers or the irreplaceabil- vant for species, and that the region of study

158 | Chapter V

ity of affected species (Williams et al., 2005; related research that has led to this article, and Araújo et al., 2011; Kujala et al., 2013). to Guy Midgley and Phoebe Barnard for stimu- Importantly, climate change metrics can be lating discussions. R.A.G. is supported by a FCT applied when and where limited knowledge of PhD studentship (SFRH/BD/65615/2009), biodiversity exists, thereby broadening the M.B.A. by the FCT PTDC/AAC-AMB/98163/ scope of exposure assessments to species that 2008 project and the Integrated Program of are known, poorly known, or even unknown. IC&DT Call Nº 1/SAESCTN/ALENT-07-0224- Familiar applications of bioclimatic envelope FEDER-00175, M.C. by the RESPONSES project, models exclude species with small sample sizes, and R.A. by the National Research Foundation due to the statistical limitations of the models of South Africa (Grant 85802). R.A.G. and M.B.A. (e.g., Stockwell & Peterson, 2002). Most as- thank the Danish National Research Foundation sessments of the effects of climate change rely- for support to the Center for Macroecology, ing on such models are thus biased against Evolution and Climate, and M.B.A. also thanks narrow-ranging species. Such bias is particu- the Imperial College London’s Grand Challenges larly acute in the tropics (Feeley & Silman, in Ecosystems and Environment initiative for 2011), and has potential consequences for support of his research. conservation priority setting under climate change (Platts et al., in review). Appreciating the full array of dimensions of References climate change captured by metrics can also Ackerly DD, Loarie SR, Cornwell WK, Weiss SB, Hamilton help to make climate change impact assess- H, Branciforte R, Kraft NJB (2010) The geography of ments more complete. In this study, we consid- climate change: implications for conservation biogeography. Diversity and Distributions, 16, 476–487. ered the three metrics of climate change that Allen CD, Macalady AK, Chenchouni H et al. (2010) A are most closely associated with the bioclimatic global overview of drought and heat-induced tree envelope models used, capturing local changes mortality reveals emerging climate change risks for in average climates and regional changes in the forests. Forest Ecology and Management, 210, 660– 684. area and position of analogous climates. Yet, Allouche O, Tsoar A, Kadmon R (2006) Assessing the other metrics such as local changes in climate accuracy of species distribution models: prevalence, extremes or variability, the timing of specific kappa and the true skill statistic (TSS). Journal of Ap- climate events, and the velocity at which cli- plied Ecology, 43, 1223–1232. mates are displaced over the local topography Altwegg R, Broms K, Erni B, Barnard P, Midgley GF, Underhill LG (2012) Novel methods reveal shifts in could be equally important (for a critical review migration phenology of barn swallows in South Af- see Garcia et al., in review). Such dimensions of rica. Proceedings of the Royal Society B-Biological Sci- ences, 279, 1485–1490. change can be integrated in bioclimatic envelope models (e.g., Zimmermann et al., 2009; Anderson RP (2013) A framework for using niche models to estimate impacts of climate change on species Altwegg et al., 2012) or complement model- distributions. Annals of the New York Academy of Sci- based assessments. Undoubtedly, the field of ences, 1297, 8–28. climate change ecology can only advance with Araújo MB, Peterson AT (2012) Uses and misuses of bioclimatic envelope modeling. Ecology, 93, 1527– the integration of multiple approaches and 1539. tools. Araújo MB, Pearson RG, Thuiller W, Erhard M (2005) Validation of species-climate impact models under climate change. Global Change Biology, 11, 1504– Acknowledgements 1513. Araújo MB, Alagador D, Cabeza M, Nogués-Bravo D, Thuiller W (2011) Climate change threatens Euro- R.A.G. thanks the South African National Biodi- pean conservation areas. Ecology Letters, 14, 484– versity Institute in Cape Town for supporting 492.

Bioclimatic models vs. climate change metrics | 159

Beaumont LJ, Pitman A, Perkins S, Zimmermann NE, IPCC (2013) Summary for Policymakers. In: Climate Yoccoz NG (2010) Impacts of climate change on the Change 2013: The Physical Science Basis. Contribution world’ s most exceptional ecoregions. Proceedings of of Working Group I to the Fifth Assessment Report of the National Academy of Sciences, 108, 2306–2311. the Intergovernmental Panel on Climate Change (eds Stocker TF, Qin D, Plattner G-K, Tignor M, Allen SK, Brook BW, Sodhi NS, Bradshaw CJA (2008) Synergies Boschung J, Nauels A, Xia Y, Bex V, Midgley PM). among extinction drivers under global change. Cambridge University Press, Cambridge, United Trends in Ecology & Evolution, 23, 453–460. Kingdom and New York, NY, USA. Chen I-C, Hill JK, Ohlemüller R, Roy DB, Thomas CD Katz RW, Brush GS, Parlange MB (2005) Statistics of (2011) Rapid Range Shifts of Species Associated with extremes: modeling ecological disturbances. Ecology, High Levels of Climate Warming. Science, 333, 1024– 86, 1124–1134. 1026. Kearney MR, Wintle BA, Porter WP (2010) Correlative Clarke KR, Warwick RM (1994) Similarity-based testing and mechanistic models of species distribution pro- for community pattern: the two-way layout with no vide congruent forecasts under climate change. Con- replication. Marine Biology, 118, 167–176. servation Letters, 3, 203–213. Cliff N (1993) Dominance statistics - ordinal analyses to Kujala H, Moilanen A, Araújo MB, Cabeza M (2013) answer ordinal questions. Psychological Bulletin, 114, Conservation planning with uncertain climate change 494–509. projections. PloS one, 8, e53315. Dawson TP, Jackson ST, House JI, Prentice IC, Mace GM Loarie SR, Duffy PB, Hamilton H, Asner GP, Field CB, (2011) Beyond predictions: biodiversity conserva- Ackerly DD (2009) The velocity of climate change. tion in a changing climate. Science, 332, 53–8. Nature, 462, 1052–1055. Dolédec S, Chesse D, Gimaret-Carpentier C (2000) Niche Mann ME, Bradley RS, Hughes MK (1998) Global-scale separation in community analysis: a new method. temperature patterns and climate forcing over the Ecology, 81, 2914–2927. past six centuries. Nature, 392, 779–787. Feeley KJ, Silman MR (2011) The data void in modeling Meehl GA, Stocker TF, Collins WD et al. (2007) Global current and future distributions of tropical species. Climate Projections. In: Climate Change 2007: The Global Change Biology, 17, 626–630. Physical Science Basis. Contribution of Working Foden W, Midgley GF, Hughes G et al. (2007) A changing Group I to the Fourth Assessment Report of the In- climate is eroding the geographical range of the Na- tergovernmental Panel on Climate Change. (eds mib Desert tree Aloe through population declines Solomon S, Qin D, Manning M, Chen Z, Marquis M, and dispersal lags. Diversity and Distributions, 13, Averyt K B, Tignor M, Miller H L), pp. 747–845. Cam- 645–653. bridge University Press, Cambridge, United Kingdom Garcia RA, Cabeza M, Rahbek C, Araújo MB Multiple and New York, NY, USA. dimensions of climate change and their implications Midgley GF, Hannah L, Millar D, Thuiller W, Booth A for biodiversity. In review. (2003) Developing regional and species-level as- Garcia RA, Burgess ND, Cabeza M, Rahbek C, Araújo MB sessments of climate change impacts on biodiversity (2012) Exploring consensus in 21st century projec- in the Cape Floristic Region. Biological Conservation, tions of climatically suitable areas for African verte- 112, 87–97. brates. Global Change Biology, 18, 1253–1269. Moritz C, Agudo R (2013) The Future of Species Under Garcia RA, Araújo MB, Burgess ND, Foden WB, Gutsche A, Climate Change: Resilience or Decline? Science , 341 , Rahbek C, Cabeza M (2014) Matching species traits to 504–508. projected threats and opportunities from climate New M, Lister D, Hulme M, Makin I (2002) A high- change. Journal of Biogeography, 10.1111/jbi.12257. resolution data set of surface climate over global land Gillson L, Dawson TP, Jack S, McGeoch MA (2013) Ac- areas. Climate Research, 21, 1–25. commodating climate change contingencies in con- Nogués-Bravo D, Ohlemüller R, Batra P, Araújo MB servation strategy. Trends in ecology & evolution, 28, (2010) Climate predictors of late quaternary extinc- 135–142. tions. Evolution, 64, 2442–2449. Guisan A, Zimmermann NE (2000) Predictive habitat Ohlemüller R (2011) Running Out of Climate Space. distribution models in ecology. Ecological Modelling, Science, 334, 613–614. 135, 147–186. Ohlemüller R, Gritti ES, Sykes MT, Thomas CD (2006) Heikkinen RK, Luoto M, Leikola N et al. (2009) Assessing Towards European climate risk surfaces: the extent the vulnerability of European butterflies to climate and distribution of analogous and non-analogous change using multiple criteria. Biodiversity and Con- climates 1931–2100. Global Ecology and Biogeogra- servation, 19, 695–723. phy, 15, 395–405. Hof C, Araújo MB, Jetz W, Rahbek C (2011) Additive Ohlemüller R, Anderson BJ, Araújo MB, Butchart SHM, threats from pathogens, climate and land-use change Kudrna O, Ridgely RS, Thomas CD (2008) The coinci- for global amphibian diversity. Nature, 480, 516–9. dence of climatic and species rarity: high risk to

160 | Chapter V

small-range species from climate change. Biology Let- Wiens JA, Seavy NE, Jongsomjit D (2010) Protected areas ters, 4, 568–572. in climate space: What will the future bring? Biologi- cal Conservation, 144, 2119–2125. Peel MC, Finlayson BL, McMahon TA (2007) Updated world map of the Köppen-Geiger climate classifica- Williams P, Hannah L, Andelman S et al. (2005) Planning tion. Hydrology and Earth System Sciences, 11, 1633– for Climate Change: Identifying Minimum-Dispersal 1644. Corridors for the Cape Proteaceae. Conservation Biol- ogy, 19, 1063–1074. Peterson AT, Soberón J, Pearson RG, Anderson RP, Martínez-Meyer E, Nakamura M, Araújo MB (2011) Williams JW, Jackson ST, Kutzbach JE (2007) Projected Ecological Niches and Geographic Distributions. distributions of novel and disappearing climates by Monographs in Population Biology, Princeton Uni- 2100 AD. Proceedings of the National Academy of Sci- versity Press, NewJersey. ences, 104, 5738–5742. Platts PJ, Garcia RA, Hof C, Foden W, Hansen LA, Rahbek Williams SE, Shoo LP, Isaac JL, Hoffmann A a, Langham G C, Burgess ND Conservation implications of omitting (2008) Towards an integrated framework for assess- rare and threatened species from climate change im- ing the vulnerability of species to climate change. pact modelling. In review. PLoS Biologyiology, 6, 2621–6. Post E (2013) Ecology of climate change. The importance Zimmermann NE, Yoccoz NG, Edwards TC et al. (2009) of biotic interactions. Princeton University Press, Climatic extremes improve predictions of spatial pat- Princeton, New Jearsey. terns of tree species. Proceedings of the National Academy of Sciences of the United States of America, Segurado P, Araújo MB (2004) An evaluation of methods 106, 19723–19728. for modelling species distributions. Journal of Bio- geography, 31, 1555–1568. Stockwell DRB, Peterson AT (2002) Effects of sample size on accuracy of species distribution models. Eco- Supporting Information logical Modelling, 148, 1–13. Tabor K, Williams JW (2010) Globally downscaled cli- Figure S1 | Projected changes in climatic suit- mate projections for assessing the conservation im- ability for sub-Saharan African vertebrate spe- pacts of climate change. Ecological Applications, 20, 554–565. cies Thomas CD, Cameron A, Green RE et al. (2004) Extinc- Figure S2 | Comparison of projected changes in tion risk from climate change. Nature, 427, 145–8. local climatic suitability for species between Thuiller W, Broennimann O, Hughes G, Alkemade JRM, areas exposed to different levels of local Midgley GF, Corsi F (2006) Vulnerability of African anomalies mammals to anthropogenic climate change under conservative land transformation assumptions. Figure S3 | Selection of optimal threshold to Global Change Biology, 12, 424–440. define analogous climates. Watson JEM, Iwamura T, Butt N (2013) Mapping vulner- Figure S4 | Climatic characterisation of sites ability and conservation adaptation strategies under climate change. Nature Climate Change, 3, 989–994. across classes of projected climate changes

Bioclimatic models vs. climate change metrics | 161

Supporting Figures

Figure S1 | Projected changes in climatic suitability for sub-Saharan African vertebrate species. For amphibians (n=284), snakes (n=310), mammals (n=623) and birds (n=1,506), the maps show the results from bioclimatic envelope modelling under late-century climates in terms of absolute cell changes in climatic suitability (first column), proportion of climate envelope lost (second column), percentage of climate envelope gained (third column), and distance shifted by species' climate envelopes (fourth column). Shown are the median values across species occurring in each pixel.

162 | Chapter V

Figure S2 | Comparison of projected changes in local climatic suitability for species between areas exposed to different levels of local anomalies. The results from the bioclimatic envelope modelling for sub-Saharan African amphibians (n=284), snakes (n=310), mammals (n=623) and birds (n=1,506) were compared across areas with different levels of local anomalies. Metric-based projections of local anomalies (a) are grouped into small (blue) and large (orange) anomalies, and the model-based median negative (b) or positive (c) changes in local (cell) climatic suitability across species were compared between areas of small and large anomalies. The differences were significant and in the expected direction for both negative changes (Wilcoxon signed rank-test, p-values<0.05; Cliff's delta [confidence interval], 0.25 [0.20 to 0.30] for birds, 0.31 [0.25 to 0.35] for mammals, 0.36 [0.31 to 0.41] for snakes, and 0.44 [0.39 to 0.48] for amphibians) and positive changes (Wilcoxon signed rank-test, p-values<0.05; Cliff's delta [confidence interval], 0.33 [0.27 to 0.37] for birds, 0.35 [0.30 to 0.40] for mammals, 0.41 [0.36 to 0.46] for snakes, and 0.40 [0.35 to 0.44] for amphibians).

Bioclimatic models vs. climate change metrics | 163

(a) (b)

(c) Thresholds for analogous climates 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 tcm 0.69 1.39 2.08 2.77 3.47 4.16 4.85 5.54 6.24 6.93 twm 0.60 1.19 1.79 2.39 2.99 3.58 4.18 4.78 5.38 5.97 psum 135.0 270.0 405.1 540.1 675.1 810.1 945.1 1080.2 1215.2 1350.2

5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0 9.5 10.0 tcm 7.62 8.32 9.01 9.70 10.4 11.09 11.78 12.47 13.17 13.86 twm 6.57 7.17 7.77 8.36 8.96 9.56 10.15 10.75 11.35 11.95 psum 1485.2 1620.2 1755.2 1890.3 2025.3 2160.3 2295.3 2430.3 2565.3 2700.3

Figure S3 | Selection of optimal threshold to define analogous climates. To define analogous climates, 20 thresholds were defined for each climatic parameter (tcm=temperature of the coldest month in °C, twm=temperature of the warmest month in °C, and psum=total annual precipitation in mm) spanning from 0.5 to 10 times the inter-annual variability of each climatic parameter across sub-Saharan Africa (c). The climatic classifications obtained with each threshold were compared to the Köppen- Geiger climatic classification using two tests, one based on the True Skill Statistics (TSS; (a)) and the other on the ANOSIM test ((b); see main text for details). The values of 3 times the inter-annual variability for each climatic parameter were selected as the optimal thresholds, based on the results from both tests.