You may notice some images loading slow across the Open Collections website. Thank you for your patience as we rebuild the cache to make images load faster.

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Ecological niche divergence and evolution in western North American monkeyflowers Li, Qin 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


24-ubc_2018_february_li_qin.pdf [ 3.97MB ]
JSON: 24-1.0360769.json
JSON-LD: 24-1.0360769-ld.json
RDF/XML (Pretty): 24-1.0360769-rdf.xml
RDF/JSON: 24-1.0360769-rdf.json
Turtle: 24-1.0360769-turtle.txt
N-Triples: 24-1.0360769-rdf-ntriples.txt
Original Record: 24-1.0360769-source.json
Full Text

Full Text

ECOLOGICAL NICHE DIVERGENCE AND EVOLUTION INWESTERN NORTH AMERICAN MONKEYFLOWERSbyQin LiM.Sc., Beijing Normal University, China, 2011B.Sc., Beijing Normal University, China, 2008A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinTHE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES(Botany)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)November 2017c• Qin Li, 2017AbstractAbstractThe ecological niche is an essential concept for studies in ecology, evolution and biogeogra-phy. Geographic distributions are largely determined by species’ ecological niches. In turn,niches evolve via selection stemming from where species occur, which has implications for co-existence and the breadth of environmental tolerance. With modern comparative methods,we can improve our understanding of interactions among niche, range and diversificationacross spatial scales.To select variables for quantifying niche properties, I first applied generalized linear modelswith occurrence data of 71 western North American monkeyflowers (Mimulus sensu lato).Then I evaluated the relative importance of four bioclimatic variables by ranking thembased on the magnitudes of model-averaged regression coecients. Thus three out of fourbioclimatic variables were identified as important predictors in determining geographic dis-tributions of Mimulus species, while one variables was negligible due to its small eect.To determine how geographic overlap aects niche divergence, I quantified niche divergencefor 16 closely related Mimulus species pairs. I found that macrohabitat niche divergencedecreased with increasing range overlap, consistent with environmental filtering operatingin sympatry and divergent selection operating in allopatry. For species pairs with partiallyoverlapping ranges, greater microhabitat niche divergence was found in sympatry, consistentwith competition driving divergence where species interact. Phylogenetic distance was pos-itively related to niche divergence for two macrohabitat axes but negatively related for onemicrohabitat axis. This suggests increasing coarse-scale niche similarity with increasing sym-patry following allopatric speciation, while greater local-scale niche divergence accumulatesthrough time.Given dierences in evolutionarily lability of niche axes across spatial scales, I next examinedevolutionary trends in niche breadth. For 82Mimulus species, I converted niche breadths intobinary states, generalist or specialist. Then I tested whether niche breadth aected diversi-fication rate and explored evolutionary transitions. My results showed higher diversificationrates for generalists and weak generalist-to-specialist trends for three bioclimatic variables,but higher diversification rates for specialists and weak specialist-to-generalist trends for twomicrohabitat variables.iiAbstractTogether, these results suggest that ecology plays an essential role in diversification processes,but underlying mechanisms might dier across spatial scales.iiiLay SummaryLay SummaryA species’ geographic distribution can be determined by its preference for environmentalconditions (ecological niche). In turn, its niche can evolve via selection stemming from whereit occurs. In this dissertation, I aimed to contribute to the understanding of the pattern ofniche evolution at various spatial scales and its interactions with geographic distribution andorigins of new species. I conducted my research with western North American monkeyflowers.In the first study, I identified three bioclimatic variables that showed relatively large eects onspecies’ distributions. In the second study, I quantified niche divergence between 16 closelyrelated species, and tested how it was aected by range overlap and divergence time. In thethird study, I quantified niche breadth for 82 species and tested its eect on diversificationand evolutionary switches to specialization. Together, this dissertation shows that patternsof niche evolution depend on the spatial scale at which niche variables are measured.ivPrefacePrefaceThe work described in this dissertation is the culmination of research from September 2012through June 2017. It consists of three separate yet related research projects, correspondingto Chapter 2, Chapter 3 and Chapter 4.• Chapter 3 -The eect of range overlap on niche divergence depends on spatialscale was based on the following unpublished manuscript:Qin Li, Dena Grossenbacher, Amy Angert (under review)– Q.L. conceived the initial idea for this project and further developed with A.A.Q.L. and D.G. conducted fieldwork and microhabitat data collection. Q.L. per-formed the statistical analyses with critical input from A.A. and D.G.. Q.L. ledmanuscript writing; A.A. and D.G. contributed important manuscript revisions.• Chapter 4 - The evolution of niche breadth in Mimulus was based on the fol-lowing unpublished manuscript:Qin Li, Amy Angert– Q.L. conceived this project in collaboration with A.A. Q.L. conducted data col-lection, performed all analyses and wrote the original manuscript. A.A. con-tributed important manuscript revisions.vTable of ContentsTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiiDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 The concept of niche . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.1 Hutchinson’s niche and geographic distribution . . . . . . . . . . . . 11.1.2 The BAM niche diagram . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Feedbacks between niche and range . . . . . . . . . . . . . . . . . . . . . . . 21.2.1 Niche stasis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2.2 Niche evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.2.3 Factors aecting niche dynamics . . . . . . . . . . . . . . . . . . . . . 51.3 Niche across spatial scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61.4 The evolution of niche breadth . . . . . . . . . . . . . . . . . . . . . . . . . . 71.5 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 The relative importance of predictive variables for species distributions 102.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122.3.1 Occurrence data and bioclimatic variables . . . . . . . . . . . . . . . 122.3.2 Model fitting and relative variable importance . . . . . . . . . . . . . 132.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14viTable of Contents2.4.1 The rankings of —* estimations . . . . . . . . . . . . . . . . . . . . . 142.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.5.1 Bioclimatic variables and niche hypervolume . . . . . . . . . . . . . . 152.5.2 Caveats for model fitting . . . . . . . . . . . . . . . . . . . . . . . . . 162.5.3 Other variable selection approaches . . . . . . . . . . . . . . . . . . . 173 Niche divergence driven by range overlap and time in Mimulus . . . . . 203.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233.3.1 Study system and phylogeny reconstruction . . . . . . . . . . . . . . 233.3.2 Occurrence data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243.3.3 Niche axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243.3.4 Range overlap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253.3.5 Niche divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263.3.6 Relationships among range overlap, phylogenetic distance and nichedivergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273.3.7 Niche divergence in sympatry versus in allopatry . . . . . . . . . . . . 273.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283.4.1 PCA of niche variables . . . . . . . . . . . . . . . . . . . . . . . . . . 283.4.2 Eects of range overlap and phylogenetic distance on niche divergence 283.4.3 Niche divergence in sympatry vs in allopatry . . . . . . . . . . . . . . 293.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303.5.1 Association between niche divergence and range overlap . . . . . . . . 303.5.2 Association between niche divergence and phylogenetic distance . . . 323.5.3 Implications for the geography of speciation . . . . . . . . . . . . . . 334 The evolution of niche breadth in Mimulus . . . . . . . . . . . . . . . . . . 394.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424.3.1 Study system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424.3.2 Bioclimatic niche variables and designation of niche breadth state . . 424.3.3 Microhabitat niche variables and designation of niche breadth state . 444.3.4 Binary trait correlation . . . . . . . . . . . . . . . . . . . . . . . . . . 464.3.5 State-dependent diversification and niche breadth evolution . . . . . . 464.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484.4.1 States of niche breadth and their correlated evolution . . . . . . . . . 484.4.2 Diversification analyses . . . . . . . . . . . . . . . . . . . . . . . . . . 484.4.3 The evolutionary direction of niche breadth . . . . . . . . . . . . . . 504.4.4 The mode of state transition . . . . . . . . . . . . . . . . . . . . . . . 514.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514.5.1 Diversification in generalists versus specialists . . . . . . . . . . . . . 514.5.2 No consistent evolutionary trend across spatial scales . . . . . . . . . 524.5.3 The designation of niche breadth states . . . . . . . . . . . . . . . . . 53viiTable of Contents4.5.4 Model inadequacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555 General discussion and conclusions . . . . . . . . . . . . . . . . . . . . . . . 655.1 Major findings and and implications . . . . . . . . . . . . . . . . . . . . . . . 655.2 Limitations of the research and future directions . . . . . . . . . . . . . . . . 67Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83A Supplementary material for Chapter 2 . . . . . . . . . . . . . . . . . . . . . 83B Supplementary material for Chapter 3 . . . . . . . . . . . . . . . . . . . . . 89B.1 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89B.1.1 Occurrence data filtering associated with nomenclature changes . . . 89B.1.2 Alternative methods to estimate range extent and range overlap . . . 90B.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90B.2.1 Correlations among range overlap methods . . . . . . . . . . . . . . . 90B.2.2 The results of sensitivity analysis of alternative range overlap methodson the relationship between niche divergence and range overlap . . . 91B.2.3 The eect of background choices on niche divergence estimates in al-lopatry versus in sympatry . . . . . . . . . . . . . . . . . . . . . . . . 92C Supplementary material for Chapter 4 . . . . . . . . . . . . . . . . . . . . . 101viiiList of TablesList of Tables2.1 The rankings of predictors across Mimulus species . . . . . . . . . . . . . . 183.1 Multiple linear regression fits for niche axes . . . . . . . . . . . . . . . . . . 354.1 Candidate ClaSSE models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 574.2 Binary trait correlation among five niche variables . . . . . . . . . . . . . . 584.3 Top models supported for five niche variables . . . . . . . . . . . . . . . . . 59A.1 Correlations among predictors . . . . . . . . . . . . . . . . . . . . . . . . . 83A.2 Estimations of —* for four bioclimatic variables for Mimulus species . . . . . 84B.1 Summary of 16 Mimulus species pairs . . . . . . . . . . . . . . . . . . . . . 93B.2 Multiple linear regression model fits between niche divergence and two co-variates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94C.1 Summary of locality and specimen records of Mimulus species . . . . . . . . 101C.2 Word list for habitat water anity . . . . . . . . . . . . . . . . . . . . . . . 104C.3 Word list for soil type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105C.4 Estimated median rates of compound parameters . . . . . . . . . . . . . . . 106C.5 Estimated 95% credibility intervals of compound parameters . . . . . . . . . 107ixList of FiguresList of Figures2.1 Estimates of —* for bioclimatic variables in Mimulus . . . . . . . . . . . . . 193.1 Four possible patterns between niche divergence and range overlap . . . . . 363.2 Eects of range overlap and phylogenetic distance on niche divergence . . . 373.3 Niche divergence in allopatry versus niche divergence in sympatry . . . . . 384.1 Rank plot of niche breadth estimates of Mimulus species . . . . . . . . . . . 604.2 Distribution of generalist and specialist lineages on Mimulus phylogeny . . . 614.3 Posterior distributions for diversification rates for five niche variables . . . . 624.4 Posterior distributions for parameters of evolutionary trend . . . . . . . . . 634.5 Posterior distributions for parameters of evolutionary asymmetry in mode . 64A.1 Estimates of —* for another set of predictors in Mimulus . . . . . . . . . . . 87A.2 Estimated Kernel densities of Mimulus species in comparison with the back-ground environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88B.1 Mimulus pairs selected from the phylogeny . . . . . . . . . . . . . . . . . . 95B.2 PCA results for macrohabitat and microhabitat axes . . . . . . . . . . . . . 96B.3 The relationship between range overlap and phylogenetic distance . . . . . . 97B.4 The relationship between niche divergence and range distance . . . . . . . . 98B.5 The eect of background choice on macrohabitat niche divergence . . . . . . 99B.6 The eect of background choice on microhabitat niche divergence . . . . . . 100xList of AbbreviationsList of AbbreviationsAIC Akaike’s information criterionAICc A second-order AICAridity aridity of growing season—* standardized regression coecientBAM The heuristic diagram of factors aecting the distribution of a species:biotic (B), abiotic (A) and movement (M)BIC Bayesian information criterionBiSSE Binary-State Speciation and Extinction modelCFP the California Floristic ProvinceCI confidence intervalsClaSSE Cladogenetic State change Speciation and Extinction modelD niche overlapDEC Dispersal-Extinction-Cladogenesis modelDEMs digital elevation modelsENM ecological niche modelingFiSSE Fast, intuitive State-dependent Speciation- Extinction analysisG generalistG-to-S generalist-to-specialistGBIF Global Biodiversity Information FacilityGDD0 growing degree days above 0 Celsius degreeGLM generalized linear modelHiSSE Hidden State Speciation and Extinction modelHWA habitat water anityIT information-theoryLM linear modelLRT likelihood ratio testMCMC Markov chain Monte CarloML Maximum likelihood method of estimationND niche divergencePC principle componentPCA principle component analysisPCA-env principle component analysis calibrated on the entire environmental spaceof the study areaPseason precipitation seasonalityPD phylogenetic distancePDmcct phylogenetic distance from the maximum clade credibility treePDmean average phylogenetic distance based on 1000 phylogeniesr Pearson correlation coecientRO range overlapROmax range overlap estimation with the larger range as the denominatorxiList of AbbreviationsROmin range overlap estimation with the smaller range as the denominatorROsum range overlap estimation with the union of two species’ ranges as thedenominatorS specialistS-to-G specialist-to-generalistSDM species distribution modelingSoil soil typeSW sum of weightsSW sum of weightsTcold temperature of coldest monthTPsyn temperature-precipitation synchronicityz corrected species occupancy of a given environmentxiiAcknowledgementsAcknowledgementsDuring my doctoral studies in the past five year, so many people have contributed to theexistence of this dissertation present here. I would like to express my sincere gratitude tomy supervisor, Dr. Amy Angert, for the constant encouragement and inspiration she gaveduring my PhD program. I especially appreciate her support of my work at the later stagethat deviated from the original proposal. All my studies has benefited from her insightfulsuggestions and mentorship. I deeply thank my dissertation committee, Dr. Dolph Schluter,Dr. Jeannette Whitton, and Dr. Jill Jankowski, who have kindly provided valuable andtimely feedback on my research and editorial advice. I would also like to thank Dr. MatthewPennell for his plentiful advice and pleasant discussions. Appreciation is extended to Dr.Sarah Otto and Dr. Greg Henry for their insightful questions during my oral defence.This dissertation research was supported by a Discovery Grant to A. Angert from the Na-tional Science and Engineering Research Council of Canada (NSERC), China ScholarshipCouncil scholarship, the Department of Botany and the Biodiversity Research Centre atUniversity of British Columbia.It is fortunate to have the opportunity to work with great members in Angert lab, currentand past: Megan Bontrager, Seema Sheth, Barb Gass, Matthew Bayly, Julie Lee-Yaw, SamPironon, Anna Mária Csergö, Rachel Wilson, Chris Muir, Chris Kopp, Anna Hargreaves, andRachel Germain. I really enjoy the interaction with them during lab meetings, collaborativeprojects, and field trips. I would express my special gratitude to Xiaojun Kou, who was oneof former advisors of my master study and a visitor scholar in Angert lab, for his inspiration,key conceptual feedback and statistical support for the first study of my dissertation.My dissertation also benefited from discussions with researchers who studiedMimulus exten-sively before and during the development of my research. Seema Sheth, Dena Grossenbacher,Steve Schoenig, and Megan Peterson generously oered detailed information on locations ofMimulus species populations. Dena Grossenbacher and Justen Whittall kindly supplied thegenetic sequences used in the second study to reanalyze phylogenetic relationships. DenaGrossenbacher (a co-author on the second study of my dissertation) shared microhabitatmeasurement data. Naomi Fraga oered important taxonomic insights. The conversationswith Jay Sexton on lots of my work were quite helpful.Many thanks to the people who have kindly oered me technical and logistic help. I thankBarb Gass and Matthew Bayly, who helped a lot with my field survey during the first twosummer seasons. I also thank D. Grossenbacher, B. Green and J. Smith for helping with fielddata collection. I appreciate the support from the National Forest Service, National ParkService, Bureau of Land Management in United States for field data collection. I also thankJennifer Chen, Lisa Lin, Megan Bontrager, Matthew Bayly, Adam Wilkinson and WinnieCheung for greenhouse assistance, and Seema Sheth, Jay Sobel, Nancy Emery, and DenaGrossenbacher for technique suggestions on plant germination and experimental settings,xiiiAcknowledgementsthough the project did not work out in the end.For the great work atmosphere, I would like to thank the tradition and organizers for aca-demic activities in Beaty Building. It has been a pleasure for me to join the BiodiversityDiscussion Group, Florum (Discussions in Plant Ecology), personal meeting and graduatestudent lunch with speakers after Biodiversity Research Seminar, to share my passion inbiodiversity and communicate with people in similar fields.My friends have made Vancouver an enjoyable place to stay, and are alway here for theups and downs. I thank Micah Scholer, who shared the oce with me, who always makesme feel so warm. I thank members of "Wednesday Running Group", Micah Scholer, ChengChen, Mannfred Boehm, Stef Fri, and Luke Busta. It was a truly memorable experience tokeep running in woods and along beach near UBC. Appreciation is also expressed to KaichiHuang, with whom I started birdwatching and got to know the city from another angle. Iam also glad to have Yichun Qiu, Xiaoou Dong, Chipan Zhu, and Shuang Liu around, whoshared the happiness and pain of the PhD life.Last, but not least, I would like to thank my family: my parents, Xingquan Li and JianhuaZhou, and my brother, Bingfan Li, for their love and support over many years. Theirencouragement will carry me forward to the next stage of life.xivxv 		 Dedicated to my beloved parents, Xingquan Li and Jianhua Zhou, and my brother, Bingfan Li, for their endless support and patience as I found my own path. 	IntroductionChapter 1 Introduction1.1 The concept of nicheThe niche concept has provided an essential framework in ecological and evolutionary re-search. It is a natural bridge among ecological, evolutionary and biogeographic processes,and studying niche dynamics provides a way to translate between current and historicaltime scales (Pearman et al., 2008). The term "niche" was originally formulated by Grin-nell (1917) nearly a century ago to describe the preference for certain climates and habitatsby the California thrasher, in relation to its geographic distribution. Further developmentsadded consideration of biotic interactions and resource-consumer dynamics (known as theEltonian niche), as well as the idea of multidimensionality, resulting in a diversity of usagesand applications (Elton, 1927; Hutchinson, 1957; Leibold, 1995; Chase and Leibold, 2003).1.1.1 Hutchinson’s niche and geographic distributionPerhaps the most widely applied conception of the niche is credited to Hutchinson, who en-visioned the niche as a multidimensional hypervolume within which populations of a specieshave a positive net growth rate (births exceeds deaths) and hence can exist indefinitelythrough time (Hutchinson, 1957). The fundamental niche is generally a combination ofabiotic variables, which is normally related to physiological tolerance of a species and de-termined by its genetic variation. The realized niche is usually considered as a reduction ofthe fundamental niche due to negative biotic interactions. The idea of multidimensionalityprovided a practical approach to connect a species’ niche to its distribution. The observeddistribution can be viewed as mapping the species’ ecological niche to geographic space byoccupying suitable habitats, as suggested by the equilibrium hypothesis (Pulliam, 2000),but potentially modified by dispersal and biotic interactions (Gaston, 2003; Peterson et al.,2011).1.1.2 The BAM niche diagramDispersal limitations and historical contingencies might cause species to be absent from somesuitable locations, while strong dispersal ability might allow species to occupy unsuitable lo-cations (Pulliam, 2000). These further complicate the relationship between species’ niche1Introductionand range (Soberón, 2007). On one hand, local sink populations with intrinsic growth rateless than one could be maintained by recurrent immigration from source populations (Pul-liam, 1988, 2000). On the other hand, meta-population dynamics suggest that species mightexperience recurring local extinction due to demographic stochasticity and environmentalvariability (Hanski, 1998). The disequilibrium hypothesis proposes that, because of limiteddispersal ability or time-lags in dispersal relative to rates of environmental change, speciesmight not fully colonize all suitable habitats (Svenning and Skov, 2004). Hence the BAMdiagram (Soberón and Peterson, 2005; Soberón, 2007) incorporates three elements (circles)in gridded geographic space to describe the relationship between species’ niche and range:abiotic factors (A), biotic interactions (B), and movement (M, dispersal ability). Areas inA are abiotic factors (i.e., species’ requirements) corresponding to the fundamental Grin-nellian niche at a coarse scale as species’ requirements. Areas in B focus on local-scaleresource-consumer dynamics, allowing species persist under biotic interactions. M depictsareas accessible to the species within a time period of interest. The intersection of thesethree circles represents areas of distribution of species with non-negative growth rate (Gas-ton, 2003), and therefore the realized niche can be quantified by mapping species distributionback to the environmental space (Peterson et al., 1999). However, natural occurrences, suchas records from field surveys or museum specimens without further information on populationdemographic dynamics, can be observed within the region of M. Species occurrences in Mbut outside of the intersection of A and B represent sink populations with negative intrinsicgrowth rates and/or populations that are undergoing competitive exclusion (Soberón, 2007).Therefore, whether the estimated realized niche is a reduction or beyond the fundamentalniche depends on the relative importance of dispersal and biotic interactions in modifyingthe fundamental niche, and how they interact across multiple spatial scales (Peterson et al.,2011). In practice, especially for species distribution modeling (SDM) or ecological nichemodeling (ENM), people assume non-negative population growth rates for occurrence data,and assume species are in dispersal equilibrium with M large enough to enclose the inter-section of A and B (pseudo-equilibrium; Guisan and Theurillat, 2000; Guisan and Thuiller,2005). Independent tests of ENM predictions, based on field translocations, have shownthat range limits are often in concordance with niche limits and species show reduced fitnessbeyond current range edges (Sexton et al., 2009; Lee-Yaw et al., 2016).1.2 Feedbacks between niche and rangeFeedbacks between ecological niches and species ranges are essential in shaping biodiversitypatterns and aecting diversification processes (Donoghue and Moore, 2003; Wiens, 2011).2IntroductionThe geographic patterns in species distribution and diversity that we observe today are deter-mined not only by contemporary ecological factors but also strongly aected by evolutionaryhistory. The BAM diagrams depict the relationship between the distribution of species andthe distribution of suitable habitats on a landscape, which are further influenced by his-torical contingencies and the geography of speciation (Myer, 1963; Coyne and Orr, 2004).The environmental heterogeneity encompassed by species’ ranges and geographic variationin biotic interactions across the ranges could drive niche evolution (Holt and Gaines, 1992;Schemske et al., 2009). In turn, the evolution of environmental tolerance and biotic inter-actions can aect range dynamics over time. Considerable work has been carried out toexplore the important niche variables in determining species distributions across taxa andregions, especially in shaping range boundaries (Gaston, 2003; Araújo and New, 2007; Sextonet al., 2009; Peterson et al., 2011). Similarly, many studies have been done to quantify nichedivergence where species’ ranges overlap with other species as compared to non-overlappingareas to test the eect of biotic interactions (Pigot and Tobias, 2013; Kraft et al., 2015) andto estimate the rate of niche evolution in relation to historical distributions and range shifts(Eiserhardt et al., 2013; Sedio et al., 2013).Another class of studies has attempted to infer the biogeography of speciation based on cur-rent distributions, which ignores niche evolution during and after speciation. These studiessuggest that in many animal clades allopatric speciation is more common, with range over-lap increasing as the divergence time increases (Barraclough and Vogler, 2000; Fitzpatrickand Turelli, 2006). Progenitor-derivative species or budding speciation (an initially smallcolonizing population becomes reproductively isolated within the range, or near, or beyondthe range edge of the larger-ranged species) was suggested to be common in plants, due togreater range overlap and dramatic range asymmetry between younger sister species pairs(Crawford, 2010; Anacker and Strauss, 2014). The problem here is the underlying assump-tion that post-speciation ranges are conserved and range shifts have not obscured patternsof speciation. But frequent range shifts and changes in range overlap in history might blurthe biogeography pattern in the past, and a changing environment itself could drive dynamicdivergence in the geographic context (Losos and Glor, 2003; Harrison, 2012).1.2.1 Niche stasisOver evolutionary time, no change in the BAM diagram means niche stasis, while nicheshifts (expansion, shrinkage, and shift) can be reflected via changes of position and shape inthree elements in BAM diagram (Peterson et al., 2011). Niche stasis (niche conservatism)3Introductionhas been demonstrated across many taxa and regions. Theoretical work suggests that sta-bilizing selection could pose an evolutionary constraint that leads to niche stasis, underwhich species maintain ancestral states (Holt, 1996; Pulliam, 2000). In variable environ-ments, species can track suitable habitats in space and time, instead of adapting to newlyencountered ones, which can promote stabilizing selection (Harvey and Pagel, 1991; Ackerly,2003). Other factors, such as a lack of genetic variation for selected traits, trade-os amongtraits, and insucient time, can also contribute to niche stasis (Harvey and Pagel, 1991).Evolutionary stasis at the species level can be demonstrated by similar ecological preferencein the present and past (Jackson and Overpeck, 2000; Webb et al., 2002). Disjunct closerelatives distributed on dierent continents have shown conserved ecological traits, rangesizes and latitudinal positions over millions of years (Ricklefs and Latham, 1992; Petersonet al., 1999). Comparative analysis of ecological variation, as well as comparisons based onENMs, have revealed high degrees of similarity in phenotypic traits and ecological charac-teristics (Peterson, 2011). Niche stasis and conservatism in related traits were detect even ata large phylogenetic scale (genus- or family-level), with variation among species associatedwith phylogenetic relationships, or even more conserved than expected (Ackerly and Reich,1999; Peterson et al., 1999; Prinzing et al., 2001; Wiens and Graham, 2005; Cavender-Bareset al., 2006).1.2.2 Niche evolutionIn contrast, previous studies have also suggested that niche evolution has not been con-strained (Schluter, 1996) and correspondingly interacts with range dynamics. Niche shiftsmight happen due to ecological processes, such as newly formed or eliminated biotic inter-actions, which largely aect the realized niche but not the fundamental niche. Evolutionaryprocesses could drive niche shifts in response to selection, which would largely aect thefundamental niche. Theoretical work implied that the interaction between dispersal andselection can determine niche expansion, especially at range edges (Kirkpatrick and Barton,1997; Orr and Smith, 1998; Holt and Keitt, 2005). Some invasive plant or animal specieshave shown rapid niche shifts during the process of invading new geographic areas over arelatively short time period (Broennimann et al., 2007; Fitzpatrick et al., 2007). Trait andniche evolution (e.g., dispersal ability) could happen in response to climate change and accel-erate range expansion (Thomas et al., 2001) or be driven by biotic interactions and facilitatespecies coexistence (Silvertown, 2004; Losos, 2009). Niche evolution can be rapid in animals(Dormann et al., 2010; Ahmadzadeh et al., 2016) and plants (Evans et al., 2009; Nakazatoet al., 2010) during the process of diversification. Similarly, comparative phylogenetic anal-4Introductionyses have uncovered high lability of some niche axes across various taxonomic groups (Lososet al., 2003; Rice and Emery, 2003; Knouft et al., 2006; Lovette and Hochachka, 2006).1.2.3 Factors aecting niche dynamicsThe interactions among ecological and evolutionary processes can be inferred not only fromcontemporary patterns of distribution and diversity, but also from evolutionary history asrevealed by phylogenetic patterns. From an evolutionary perspective, the pattern of currentdistributions and niche properties of extant species is a snapshot in time, the outcome ofintertwined ecological and biogeographic processes (dispersal, competition, adaptation, andspeciation) operating over time and space (Wiens and Donoghue, 2004; Wiens, 2011). Howthe niche evolves can play an important role in explaining many biogeographic patterns.Niche conservatism could shape past and current distribution along environmental gradients(Wiens and Graham, 2005). If climatic niche properties are strongly conserved, species willfail to disperse into novel climates or habitats. Evidence for niche conservatism was found toexplain high tropical richness patterns across taxa in animals and plants (reviewed in Wienset al., 2010). Other studies have suggested that the latitudinal diversity pattern may bethe result of high speciation rates in the tropics and/or high extinction rates in temperatezones (reviewed in Mittelbach et al., 2007), but limited dispersal due to evolutionary con-straints was still important for generating such geographic patterns. Conversely, transitionsto new niche optima or expansion of niche breadth could foster expansion or dispersal to newbiogeographic areas. For example, an ‘out of the tropics’ colonization hypothesis suggestsan oak species moved into the temperate zone by the acquisition of freezing tolerance butleft its sister species in the tropics (Cavender-Bares et al., 2011). In addition, the rate ofclimatic niche evolution may depend on biotic factors. Kozak and Wiens (2010a) found geo-graphic overlap has a negative eect on the rate of climate niche evolution in 16 major cladesof plethodontid salamanders by preventing dispersal. Including information of presence ofcompetitor as a predictor in species distribution modeling can improve the predictive power(Anderson et al., 2002), implying that the eect of biotic interactions can be detected at acoarse scale and range-wide extent. However, how biotic interactions constrain or promoteniche evolution at the range-wide scale remain poorly explored (Wiens, 2011).Correspondingly, biogeographical processes can strongly influence niche evolution. How newdaughter species inherit ranges from their common ancestor during speciation often is associ-ated with niche dynamics (Schluter, 2000b; Coyne and Orr, 2004). Under one circumstance,when two sets of populations are separated by geographic barriers or unsuitable habitats,5Introductionthe tendency to maintain the ancestral niche was suggested to contribute to reproductiveisolation by failure to adapt into the barriers (Wiens, 2004; Hua and Wiens, 2013). In thiscase, intrinsic reproductive isolation forms as a by-product of independent evolution dueto the accumulation of post-zygotic incompatibilities over time by drift (Lessios, 2008). Ifenvironmental heterogeneity exists, reproductive isolation could form by divergent selectiondue to occupying dissimilar environments (ecological speciation in allopatry) (Rundle andNosil, 2005; Schluter, 2001, 2009). Under another circumstance, when two ranges inheritedby daughter species have complete overlap (or at least complete for one), strong divergentselection is often involved in sympatry to eciently overcome gene flow and promote re-productive isolation (Coyne and Orr, 2004). The sources for such divergent selection canbe ecological due to local environmental dierences, involving sexual selection or ecologicalinteractions (ecological speciation facing gene flow Schluter, 2001; Nosil, 2012). It has beensuggested that there is no general relationship between the geographic modes of speciationand character evolution, at least in allopatric speciation (often involving adaptive divergencein vicariance but not in founder-induced and peripatric speciation) (Losos and Glor, 2003).However, divergence is more likely to happen with adaptive processes, and even when themechanism of initial divergence in both situations is non-ecological, natural selection andreinforcement (in sympatry or secondary sympatry) can favour niche evolution at a laterstage to enhance reproductive isolation (Schluter, 2001; Losos and Glor, 2003). Further-more, descendants with diering range sizes might be directly diverged in terms of nichebreadth, since range size is normally thought to be positively correlated with niche breadth(Slatyer et al., 2013).1.3 Niche across spatial scalesNiche variables at dierent spatial scales could show varying degrees of evolutionary labil-ity, though no predominant pattern has been revealed in nature, with mixed support inthe literature. Macrohabitats represent the set of environmental conditions at a coarse spa-tial scale, such as macroclimate and topography, which are suggested to largely determinespecies distributions (Gaston, 2003). Microhabitats describe the immediate surroundings ofan organism, such as soil moisture and nutrients, micro-topographical and light characteris-tics, which may dier from the average conditions nearby. The hierarchical niche concept istightly associated with biological processes to explain species coexistence and richness acrossspatial scales (Whittaker et al., 2001; Pearson and Dawson, 2003). Climatic niches at thecoarse scale and/or niches related to microhabitat use have been shown to evolve rapidlyand diverge across phylogenies (Losos et al., 2003; Ackerly et al., 2006; Silvertown et al.,6Introduction2006; Kozak and Wiens, 2010b; Wiens et al., 2010; Emery et al., 2012). Niche divergence ata variety of spatial scales can directly contribute to habitat isolation, whether divergence ismicro-spatial (e.g., species are broadly sympatric but locally allopatric) and/or macro-spatial(e.g., species are allopatric), which in turn could enhance reproductive isolation (Coyne andOrr, 2004). Especially, divergence in microhabitats that creates micro-spatial isolation, suchas specialization to dierent edaphic or soil moisture conditions, can be often found in sym-patry (Peterson et al., 2013; Sobel, 2014). However, the evolution of niche breadth acrossspatial scales has been still understudied in a phylogenetic framework (Schluter, 2000b).1.4 The evolution of niche breadthEstimating the rate and magnitude of niche evolution in clades could help us to identify thepropensity for niche change, in other words, the trend of niche evolution. Niche breadthrefers to species’ preference or tolerance along an environmental axis (Futuyma and Moreno,1988). Both intrinsic and extrinsic factors constrain the degree of specialization, resultingin a nested concept of niche breadth from the individual level to the species level. A wideniche breadth at the species level could be an outcome of plasticity at the individual levelthat maintains high fitness across various conditions or specialized genotypes within oneor more populations that each have high fitness in a subset of environmental conditions(Futuyma and Moreno, 1988; Ackerly, 2003). Hence, genetic variation in niche-related traitsand fitness tradeos among traits may both constrain the evolution of niche breadth, aswell as range expansion (MacArthur, 1972; Huey and Hertz, 1984; Futuyma and Moreno,1988; Kellermann et al., 2009). Externally, among-locality and within-locality (both spatialor temperal) variation in environment contribute to the likelihood of specialization (Levins,1968; Stevens, 1989; Quintero and Wiens, 2012).Specialization generally means a narrower niche breadth relative to related species (Simpson,1953; Schluter, 2000b). Specialization is often thought to be an evolutionary dead end(Colles et al., 2009; Day et al., 2016). It is often associated with restricted distributions andreduced genetic variation, leading to reduced speciation rates or higher risk of extinctionunder environmental changes (Day et al., 2016). However, macroevolutionary analyses haveobtained mixed results regarding diversification rates in generalists and specialists, and theevolutionary direction of niche breadth was suggested to be unpredictable as well (Schluter,2000b). A long-standing generalist-to-specialist hypothesis is that ecological generalists giverise to specialists (Simpson, 1953). This is because generalists have a greater potential tocolonize new habitats and give rise to new lineages, and descendants partition ancestral7Introductionniches as diversification proceeds (Freckleton and Harvey, 2006; Janz and Nylin, 2008). Butspecialists may expand into new adaptive zones and eventually become generalists that areobserved at the current time. Meta-analyses have shown that reversals from specialists togeneralists are not impossible (Schluter, 2000b; Vamosi et al., 2014). Moreover, speciesmight be specialized dierently across various niche axes (Litsios et al., 2014), and it isunclear whether the degrees of specialization are consistent across spatial scales. Therefore,it is still an open question whether evolutionary trends in niche breadth are associated withdiversification processes across dierent spatial scales.1.5 ObjectivesUnderstanding of the intertwined biological processes that aect niche and range dynamics isessential for extending our knowledge of where species live, how species coexist, and factorsshaping biodiversity patterns over evolutionary time (Holt, 2009). Meanwhile, it helps toimprove our ability to predict species’ responses and vulnerabilities to future changes inclimates. In my dissertation, I focused on two main aspects of niche dynamics across scales(both phylogenetic and spatial): 1) niche divergence in relation to the degree of range overlapof closely related species pairs, and 2) the evolution of niche breadth in relation to thediversification process across an entire clade.I conducted three studies using the western North American monkeyflowers, Mimulus sensulato (Phrymaceae). Western North American Mimulus contains about 90 described species(Beardsley et al., 2004), or about 150 taxa in a recent revision (Barker et al., 2012). Itis a model system in evolutionary ecology with great variation in range size and habitatpreference, as well as climatic niche breadth (Wu et al., 2008; Sheth et al., 2014). In my firstchapter, I applied multi-model averaging to empirical Mimulus occurrence data to selectimportant climatic niche variables in determining species geographic distributions. Nichevariables of high relative importance in determining distributions were used for later analysesof niche divergence and evolution.Species show variance not only in range size and overlap, but also in the degree of nichedivergence. Negative biotic interactions in sympatry might drive niche divergence, whileenvironmental filtering might prefer niche similarity. In my second chapter, to assess theeect of range overlap on niche divergence in a phylogenetic framework, I quantified nichedivergence, both for macrohabitat and microhabitat variables, between 16 closely relatedspecies pairs with varying degrees of range overlap. Results from the second study suggested8Introductioncoarse- and local-scale niche axes show dissimilar patterns of divergence, in relation to varyingdegrees of range overlap.Niche variables also showed dierent degrees of evolutionary lability. Thus, my third chaptertested whether there was a significant dierence in diversification of generalists and specialistsin Mimulus, focusing on niche breadth across spatial scales. To do so, I classified threecoarse-scale climatic variables and two local-scale microhabitat variables into binary states.I then explored the association between their evolution and the diversification process, withconsideration of niche shifts happening during and after speciation events. In the Conclusionssection, I discuss the overall implications and future directions of my dissertation work.9Relative variable importanceChapter 2 The relative importance ofpredictive variables for speciesdistributions2.1 SummaryA fundamental goal of scientific research is to identify the underlying variables that governcrucial processes of a system. This is especially dicult in ecology, which is intrinsically richin candidate predictors. One approach is to quantify the relative importance of candidatepredictors by applying statistical procedures to empirical data. Here we applied the stan-dardized regression coecient method (—*) from generalized linear models with an empiricalspecies presence/background dataset for the plant genus Mimulus. We treated weighted re-gression coecients as the eect sizes of candidate bioclimatic variables in a multi-modelframework, and ranked them to evaluate their relative importance in species distributions.This study sensibly identified three important predictors with high credibility in modelinggeographical distributions of 71 Mimulus species. Thus, this approach reasonably reducedthe dimensionality of niche hypervolume without losing interpretive power.2.2 IntroductionOne of the fundamental motivations of scientific research in natural or social systems isto reveal the underlying forces that govern crucial processes within studied systems of in-terest. However, identifying those important drivers is not an easy task, especially forsuch complicated systems that may contain a large number of candidate predictor variables.One approach is to rely upon common knowledge, intuition, or long-term research customs.This kind of conventional wisdom is usually system-specific, thus beyond the intention ofthis paper. An alternative approach is to apply general statistical procedures to empiricaldata at hand to identify the important variables (Burhnam and Anderson, 2002; Grueberet al., 2011). The goals of statistical models can be diverse, from gleaning basic under-standing about the properties of a system to making future predictions. Therefore, theyrequire the ability to extract as much information as possible from data collected. In the10Relative variable importancelast decades, information-theoretic (IT)-based model selection has been widely adopted inecological studies (Burhnam and Anderson, 2002; Burnham et al., 2011; Zsolt Garamszegi,2011). The IT approach fits and compares models with dierent sets of predictive variablessimultaneously based on various criteria, such as Akaike’s information criterion (AIC), asmall-sample version of AIC (AICc), or Bayesian information criterion (BIC). Therefore,it avoids common criticisms of other well-known approaches (e.g., stepwise reduction tech-niques, fully-parameterized global model), such as overfitting, arbitrary variable selection,or low predictability (Whittingham et al., 2006). Moreover, IT-based model selection hasessential and unique advantages. Each candidate model could be viewed as a hypothesis.Hence the comparison of multiple models simultaneously can be viewed as hypothesis rank-ing by minimizing information loss, based on data at hand. When model probability (Akaikeweight) is not 1, it provides the opportunity for multi-model inference with quantification ofthe uncertainty in model selection, to summarize information captured across models (e.g.,model averaging, Cade, 2015).Among candidate predictive variables, researchers often aim to determine which ones influ-ence the response variable the most. By selecting the most important variables, one couldnarrow down potential processes that generate observed patterns. Moreover, by excludingnegligible variables, one could reduce the dimensionality of predictors under consideration.There is a large body of literature in which various indices for the relative variable im-portance have been developed or discussed in applied statistics, social science and ecology(Gröemping, 2006; Murray and Conner, 2009; Giam and Olden, 2015). Intuitively, "impor-tance" has two levels of meaning. The first is whether each variable in the selected model hasan eect on the response, no matter how small it is (Burhnam and Anderson, 2002). Thesecond is about eect size, which can be understood as how strong a change in response isdue to a change in a predictor, or the contribution of a prediction to the total variance in theresponse. Based on measurements of eect size, one can rank candidate variables accordingto their importance to select the most important ones.Existing approaches for determining variable importance fall into two realms. The first ap-proach is the sum of weights method (SW) from the information theory realm, and it has beenwidely adopted (Burhnam and Anderson, 2002). However, SW was recently criticized for itswidespread misinterpretation (Murray and Conner, 2009; Galipaud et al., 2014). Moreover,simulation experiments demonstrated that SW performs poorly to distinguish the relativeimportance among predictors (Galipaud et al. 2014; but see Giam and Olden 2015). In itspure mathematical sense, SW is an estimate of the probability of one predictive variable toappear in the best model (Burnham et al., 2011; Symonds and Moussalli, 2010). Therefore,11Relative variable importanceit is not an estimate of eect size of the corresponding predictor on the response. The secondapproach contains two basic indices from the realm of multiple linear regression, oers: stan-dardized regression coecients (—*) and the variance partitioning components (Bring, 1994).These indices could be interpreted as measurements of eect size. The simplest —* methodin LM can achieve nearly the same power in identifying dominant predictors as those fromsophisticated methods, such as squared partial correlation, squared semi-partial correlation,and Budescu’s Dominance Analysis method (Whittaker et al., 2002). Though the —* methodhas been commonly used with a global model (with all candidate variables under consider-ation), the practice of multi-model comparison and model averaging for parameters oersrobust estimations, accounting for model uncertainties (Burhnam and Anderson, 2002).A species’ distribution is, at least in large part, the realization of its niche hypervolumein geographic space (Hutchinson, 1957; Pulliam, 2000). Studies on species-environmentalrelationships have been crucial in ecology, and climatic factors have been frequently usedto explain species distributions at a coarse scale (Peterson et al., 2011). As a parametricapproach, generalized linear models (GLM) with respective link functions are commonlyapplied for the correlative analysis of species presence and environmental variables (Guisanand Theurillat, 2000). However, selecting climatic predictors from the myriad of possiblevariables remains a major challenge. Here we used —* to evaluate the relative importance ofcandidate variables in predicting species geographical distributions. GLM was applied to fitto occurrences of western North American Mimulus species, and evaluated the importanceof four bioclimatic variables by ranking their eect sizes for each species. Results showedthat this approach was able to provide reasonable basis for us to select three out for fourcandidates with high relative importance across Mimulus species.2.3 Methods2.3.1 Occurrence data and bioclimatic variablesMimulus sensu lato (Phrymaceae) is a diverse genus with most member species distributed inwestern North America, having a hotspot of species richness and endemism in the CaliforniaFloristic Province (CFP) and especially the Sierra Nevada mountain ranges. Relatively fewspecies with broad distributions (e.g., M. guttatus, M. floribundas) have expanded beyondthe CFP. Mediterranean climates are assumed to be a major factor confining species to theCFP. However, this hypothesis has never been thoroughly tested.12Relative variable importanceGeo-referenced occurrence data of 71 targeted taxa were compiled from Global BiodiversityInformation Facility (GBIF,, accessed in January 2017). We furtherremoved duplicate points within a 1-km grid cell after Albers equal area projection (fi-nal numbers across species ranged 10 - 1077, median 83). The five bioclimatic variableswere: temperature of the coldest month (Tcold), growing degree days above 0 Celsius de-gree (GDD0 ), precipitation seasonality (Pseason, coecient of variation), aridity of growingseason (Aridity), and temperature-precipitation synchronicity (TPsyn). The first four arefactors commonly used in plant ecological studies, and the last one is the one we hypoth-esized to be important for Mimulus specifically. TPsyn is an index to denote the correla-tion coecient between monthly temperature and monthly precipitation (Kou et al., 2011).Relatively low TPsyn is the most characteristic attribute of Mediterranean climates. Biocli-matic layers of Tcold and Pseason were downloaded at a 30-second resolution from WorldClim(, and derived other variables from monthly temperature and monthlyprecipitation data from WorldClim (see method from Kou et al., 2011). Here we tested ifTPsyn could be identified as the most important variable in predictingMimulus distributions.The presence/background technique was applied as commonly practiced in the species distri-bution modeling (SDM) community. Therefore, we firstly created a common background bymerging all buered species occurrence points with a 100-km radius, and then from it 10,000points were randomly sampled as background data for each species. According to valuesof bioclimatic variables across this background, Tcold and GDD0 were highly correlated (r= 0.92, Table A.1); thus we carried out downstream model fitting and estimated — by justkeeping one of them (Tcold here) in the candidate predictor set. In addition, we kept GDD0and excluded Tcold in downstream analyses, and the result was similar (Figure B.1.1).2.3.2 Model fitting and relative variable importanceGeneralized linear models (GLM), were applied to model the distributions for 71 Mimulusspecies with four bioclimatic variables. With a binomial distribution and logit-link function,GLM used here was the logistic regression, modeling the log odds of presence/absence foreach species. Regression coecients were estimated to provide strong information aboutthe relative variable importance in determining the geographic distributions. Each biocli-matic variable was standardized before fitting models to assure they had the same scale forcomparison. Regression coecients were estimated in a multi-model framework. For eachspecies, all possible models (a global model and subset models) were ranked by AICc values.Then regression coecients were weighted by composite model weights and averaged over allmodels in which the predictor appeared (p154, Anderson et al., 2002). For each of four bio-13Relative variable importanceclimatic variables, we reported the distribution of regression coecient estimations across 71species. Because the same climatic variable could have opposite, yet equally strong, eectson dierent species’ probabilities of occurrence (e.g., one species is more likely to occur withincreasing values of a variable while another species is less likely to occur), we consideredthe absolute values of regression coecient estimates (as their eects) for each species. Thenrelative frequencies of possible orders were calculated to see which predictors frequently havelarger eects (i.e., more important) than others.2.4 Results2.4.1 The rankings of —* estimationsDuring the model fitting procedure, the Akaike weight of the top model was often not close to1; therefore, model averaging provides a relatively more robust inference. The distributionsof —* estimations of four candidate bioclimatic variables showed variations across 71Mimulusspecies (Figure 2.1). The estimations of —* for TPsyn had the widest range, from -87.71 to2.03; while the estimations of —* for Aridity had the narrowest range, from -3.63 to 2.52 withvalues clustered around zero for most species (Table A.2). The estimations of —* for Tcoldand Pseason had similar ranges, from -16.40 to 12.93, and from -4.94 to 33.35, respectively.With absolute values of —* estimates, rankings of four candidate variables were made acrossall species considered here (Table 2.1). The most frequent ranking was "TPsyn > Tcold >Pseason > Aridity", which occurred for 22 out of 71 species. The second frequent rankingwas "TPsyn > Pseason > Tcold > Aridity", which occurred for 15 species with TPsyn andAridity being at the same positions as the most frequent ranking. Across all rankings, thevariable occupying the first position (i.e., with the largest absolute eect) was frequentlyTPsyn (the relative frequency, 77.5%), while the variable occupying the last position (i.e.,with the smallest absolute eect) was frequently Aridity (the relative frequency, 71.8%).Tcold and Pseason were sometimes occupying the first position (relative frequencies, 11.3% and9.9%, respectively). But more often, they occupied the second positions, together reachinga relative frequency of 87.3%. The median absolute —* estimations were: 5.31, 2.39, 2.02,and 0.48 for TPsyn, Tcold, Pseason, and Aridity, respectively. Thus, the first three bioclimaticvariables were identified as of relatively large importance: TPsyn, Tcold, and Pseason. The lastvariable, Aridity could be considered as a negligible one with the smallest eect.Visual inspection of the kernel density of 71 Mimulus species’ preference in comparison with14Relative variable importancetheir environmental background (Figure A.2) confirmed that Mimulus were highly selectiveto the lower end of the variable of TPsyn, as shown by considerable dierence from thebackground points; while there were fewer contrasts with background in Aridity. Thesevisual patterns were consistent with results derived from rankings based on —* here.2.5 DiscussionWe applied model-averaged standardized regression coecients to evaluate the relative vari-able importance via general linear models with a dataset of species occurrence records andcandidate bioclimatic variables. According to ranking frequencies, three bioclimatic vari-ables showed relatively large importance across 71 Mimulus species. Results conformed toour expectation that the temperature-precipitation synchronicity (TPsyn) had the highestrelative importance in terms of determining species distributions in western North America.2.5.1 Bioclimatic variables and niche hypervolumeEcological niche modeling (ENM) has been always an essential issue in ecological studies toexplore the relationship between species’ geographic distributions and crucial environmentalfactors (Guisan and Theurillat, 2000; Peterson et al., 2011). Over evolutionary time, a speciesadapts to tolerate some portion of the global environmental space, which is determined bythe evolutionary potential of its physiological characteristics, as well as species interactionsand dispersal. Candidate bioclimatic variable were chosen because they represent generalclimatic means, variation, and extremes. The distribution of a species is viewed as occupy-ing suitable habitats across landscape. In western North America, a striking feature of theCalifornia Floristic Province (CFP) is its high species richness and endemism under currentMediterranean climates. Though the onset of winter-wet, summer-dry Mediterranean cli-mates might not be the driver of high diversification rate (Lancaster and Kay, 2013), thecomplexity of climates in combination with topography in this region provides plenty ofopportunities for high biodiversity (Raven and Axelrod, 1978). Both species with restricteddistributions in Sierra Nevada ranges (e.g., M. bicolor, M. douglasii, M. norrisii) and specieswith relative broad distributions beyond CFP (e.g., M. cardinalis, M. floribundus, M. gut-tatus) showed TPsyn had the largest eect on their distributions in comparison with othervariables. Also the negative eect of TPsyn (the direction of —* estimations) demonstratedspecies’ preference for Mediterranean climates. The few species for which TPsyn was the thirdmost important variables have their distributions farther away from CFP (e.g., M. eastwood-iae, M. parryi, M. rupicola), implying other climatic factors might be more important for15Relative variable importancethem.Researchers are using these parametric and additional non-parametric and machine learningmethods to understand which environmental variables are most important in determininggeographical distributions across taxa and regions. Many approaches besides GLM, such asgeneralized additive models (Hastie et al., 2001), random forest (Breman, 2001), boostedregression trees (Elith et al., 2008) and MaxEnt (Phillips et al., 2006) have been commonlyapplied with presence-absence or presence-only data. Whatever approach is applied, selectingrelatively important variables, based on their eect sizes, has explicit advantages on reduc-ing dimensionality of niche hypervolume. Ordination techniques, like principle componentanalysis, could be used before model fitting to make fewer orthogonal composite predictors,but they introduce issues with the diculty in later interpretation of the biological meaningsof predictive variables.2.5.2 Caveats for model fittingLimitations associated with the —* method include inequivalence of changes of one stan-dard deviation among predictors due to its property of being unitless, and diculties ininterpretation when multicollinearity exists among predictors (Murray and Conner, 2009).During model building procedures, multicollinearity among predictors is always a dicultissue, introducing problems like inflated variability in estimation of regression coecients(Whittaker et al., 2002). The bias would be that a predictor obtained a high importanceestimate because it was highly correlated with another predictor that was actually correlatedwith the response. Based on —*, Johnson (2000) proposed an index of relative weight (Á) thatattempted to solve the issue of correlated predictors by creating orthogonal representationsof the original predictors in linear regression. Recently Tonidandel and LeBreton (2010) ex-tended the relative weight analysis with —* into logistic regression, which may give us someimplications for our empirical practice. We did not consider multicollinearity among predic-tors here, which needs further investigation. Rather, we chose to only include one of twohighly correlated variables, Tcold or GDD0, in our model fitting. For later analysis of nicheproperties, these two variables can be combined into one composite variable by extractingtheir first principal component (representing 97% variation in the background). Therefore,reducing environmental dimensions from five to three can make hypervolume niche analysismore feasible.Furthermore, using standardized coecients only applies to first order linear regressionmodel; it cannot deal with non-linearity and among-predictor interactions. Interaction terms16Relative variable importancewere not included in above models. Adding an interaction might artificially increase theimportance estimate of corresponding variables, since they might have more chance to beincluded in models with higher support.2.5.3 Other variable selection approachesAlong with estimating the relative variable importance, studies have applied various ap-proaches to select predictors in study systems of interest. These include (but are not limitedto): stepwise procedures, subset IT-based multimodel inference with AIC or BIC criteria, andclassification/regression trees. Little dierence between stepwise and all-subset approacheswas detected in terms of the number of variables in selected models and predictive ability bycross validation in 12 empirical datasets (Murtaugh, 2009). But stepwise procedures havebeen criticized due to their inability to rank alternative models (i.e., competing hypotheses),and estimate model selection uncertainty. Accounting for the influence of multicollinearity,there are some methods to quantify eect sizes of predictors (both direct eect on the re-sponse, and joint eect contingent on other predictors). One among them is hierarchicalpartitioning (Chevan and Sutherland, 1991; Mac Nally, 2002; Bi, 2012), which is equivalentto Dominance Analysis computationally (Budescu, 1993), but it emphasizes global models.Giam and Olden (2015) recently has extended the idea of hierarchical partitioning to a multi-model framework and used the model weights in averaging partitioned R squared in linearmodels (Iweight). Although eorts in modifying these indices with multimodel comparisonscan improve their statistical properties, some methods are still too computationally tediousto be readily used in empirical studies (Johnson, 2000, 2004).In summary, statistical models are particularly useful for explanation purpose of the correl-ative relationships between species occurrences and plausible causal environmental factors.The main goal of measuring the eect sizes of candidate predictors is to quantify their contri-butions to the variance in species distributions and provide critical information on variableselection for reducing niche dimensionality. In light of the growing body of species localitydata and global-scale environmental layers, better understandings of the relationship be-tween distributions and environments will advance our capability to quantitatively describeof niche properties, as well as predictions for future range shifts under climate change.17Relative variable importanceTable 2.1: The rankings of four bioclimatic variables based on absolute values of —* estima-tions across 71 Mimulus species, and corresponding relative frequencies. Tcold: temperatureof coldest month; Pseason: precipitation seasonality; TPsyn: temperature-precipitation syn-chronicity; and Aridity: aridity of growing season.rankings species case relative frequencyTPsyn > Tcold > Pseason > Aridity 22 0.31TPsyn > Pseason > Tcold > Aridity 15 0.21TPsyn > Tcold > Aridity > Pseason 10 0.14TPsyn > Pseason > Aridity > Tcold 7 0.10Tcold > TPsyn > Pseason > Aridity 5 0.07Pseason > Tcold > TPsyn > Aridity 4 0.06Tcold > Pseason > TPsyn > Aridity 3 0.04Pseason > TPsyn > Tcold > Aridity 2 0.03Pseason > TPsyn > Aridity > Tcold 1 0.01TPsyn > Aridity > Pseason > Tcold 1 0.01Aridity > Pseason > TPsyn > Tcold 1 0.0118Relative variable importance●● ●●●●●●●●●●● ●●●●●●● ●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●● ●●●●● ●●● ●●●●●●●● ●● ● ●●●● ●●●● ●●●● ● ●●● ●●●● ●●●●●●●● ●●●●● ●●● ●● ●● ●● ●● ●● ●●●● ●●●●● ●●●●TPsyn Arid i tyTcold Pseason−500−500Species rankEstimations of β*Figure 2.1: The distributions of —* estimations for four candidate bioclimatic variables across 71Mimulus species, ranked from low to high values. Dashed lines depict the eect size of zero. Tcold:temperature of coldest month; Pseason: precipitation seasonality; TPsyn: temperature-precipitationsynchronicity; and Aridity: aridity of growing season.19Niche divergence and range overlapChapter 3 Niche divergence driven byrange overlap and time inMimulus3.1 SummaryExamining patterns of niche divergence and geographic range overlap of closely relatedspecies provides insights into the processes of niche evolution and speciation. When rangesoverlap, environmental filtering and shared selective pressures may preserve niche similaritybetween close relatives, particularly along coarse-scale macrohabitat niche axes. Alterna-tively, competitive interactions may drive greater niche divergence, particularly along local-scale microhabitat niche axes. We tested these hypotheses in 16 pairs of western NorthAmerican monkeyflowers (Mimulus) using data on species’ niches, geographic ranges anda robust phylogeny. We found that coarse-scale macrohabitat niche divergence decreasedwith increasing range overlap and phylogenetic distance, consistent with environmental fil-tering that operates as range overlap increases following allopatric speciation. No significantrelationships were detected for local-scale microhabitat niche divergence among all species,except for a positive association with phylogenetic distance along one microhabitat axis re-lated to vegetation cover. For the subset of species pairs with partially overlapping ranges,greater microhabitat divergence was found in sympatry than in allopatry along at least oneniche axis per pair, consistent with competitive interactions driving greater niche divergencein sympatry. Thus, coarse- and local-scale niche axes show dissimilar patterns of evolu-tionarily lability, perhaps because the relative importance of environmental filtering versuscompetitive interactions depends on spatial scale.3.2 IntroductionComplicated feedbacks between niche divergence, range overlap and the geography of spe-ciation play an essential role in generating patterns of biodiversity. Niche divergence canlead to dierences in geographic range position and extent. In turn, dierences in geo-graphic range among species (and hence environments experienced) can cause selection for20Niche divergence and range overlapniche divergence. The initial stages of niche and range divergence are intimately related tothe geography of speciation, while subsequent niche and range evolution can impact whichspecies assemble into a community (Ricklefs and Jenkins, 2011; Wiens, 2011). Thus, pat-terns of niche divergence are predicted to dier depending on whether species are sympatricor allopatric.Closely related species can have varying degrees of range overlap, with potentially vary-ing degrees of niche dissimilarity. In sympatry, alternative predictions of niche divergencecould be reached depending on the relative roles of environmental filtering and competi-tion. Environmental filtering describes how abiotic factors allow or prevent organisms fromestablishing and persisting at a given location (Kraft et al., 2015). Given strong environ-mental filtering, species that co-occur should have higher niche similarity (Webb et al., 2002;Kraft et al., 2008; Cavender-Bares et al., 2009; Mouillot and Gaston, 2009). Conversely,higher range overlap potentially increase encounter frequencies and competition in sympatryconsequently increase competition intensity. Therefore, higher range overlap could generateselection for greater niche divergence to reduce resource competition or reproductive inter-ference, i.e., ecological or reproductive character displacement (Brown and Wilson, 1956;Silvertown, 2004; Beans, 2014). In addition, apparent niche dierentiation may also becaused by plasticity of resource utilization in response to competition (Schluter, 2000a).In allopatry, predictions about the degree of niche divergence or similarity between closerelatives depend on a variety of factors. When geographic barriers of unsuitable environmentslead to allopatric speciation (Barraclough and Vogler, 2000; Fitzpatrick and Turelli, 2006),species may retain similar niches because of stabilizing selection in similar habitats. Sucha pattern of niche conservatism (the tendency to maintain ancestral niches) is supportedacross many taxa and geographic regions (Peterson et al. 1999; Wiens and Graham 2005;Pearman et al. 2008; Wiens et al. 2010; and reviewed in Peterson 2011). Alternatively, ifallopatric species experience dierent environmental conditions, then they are more likely toexperience divergent selection pressures leading to increased niche divergence (McCormacket al., 2010; Schemske, 2010).By examining niche divergence in multiple pairs of species that vary in their degree of rangeoverlap, we can gain insight into which of the processes described above are dominant. Forexample, a negative relationship between niche divergence and range overlap suggests envi-ronmental filtering in sympatry and divergent selection in allopatry (abiotic factors dominate;Figure 3.1a). In contrast a positive relationship between niche divergence and range over-lap suggests niche stasis in allopatry and competition driving niche divergence in sympatry21Niche divergence and range overlap(biotic interactions dominate; Figure 3.1d). However, various processes might be at play si-multaneously, leading to no clear positive or negative pattern (Figure 3.1b and Figure 3.1c),which might depend on the species involved. Empirical studies are required to determinewhich patterns are prevalent in nature.Furthermore, the relationship between niche divergence and range overlap could dier among-predictor macro- and microhabitat niche axes. Macrohabitats represent the conditions underwhich a species can tolerate and persist at a coarse spatial scale, such as range-wide macro-climate and topography. Microhabitats describe local-scale environments of an organismor population, such as soil moisture and nutrients, micro-topographical and light charac-teristics. Environmental filtering is predicted to favor similarity along coarse-scale nicheaxes while competitive interactions will favor divergence along local-scale niche axes (Swen-son et al., 2007; HilleRisLambers et al., 2012). For example, various ecomorphs of Anolisspecies generally co-occur in the Greater Antilles and are able to coexist by partitioningresources at the fine scale by specializing on structural microhabitat (Losos, 2009, e.g., treecrown/trunk/twig/ground,). This partitioning leads to an expectation of more divergencein allopatry for macrohabitat axes, but more divergence in sympatry for microhabitat axesto allow coexistence. Therefore, a negative relationship was expected between niche diver-gence and range overlap at the coarse scale, but a positive one at the fine scale. However,how biotic interactions constrain or promote niche evolution at the range-wide scale remainpoorly explored (Wiens, 2011).Because evolutionary changes are assumed to accumulate through time, the degree of nichedivergence is expected to positively correlate with phylogenetic distance (relative divergencetime, Harvey and Pagel, 1991). Hence, in order to test the relationship between nichedivergence and range overlap, we must account for divergence time. However, niche axescould exhibit dierent degrees of lability, and empirical results show no clear pattern formacro- and microhabitat niche (Losos and Glor, 2003; Peterson and Holt, 2003; Knouft et al.,2006). How range overlap aects the evolutionary rate and the magnitude of divergence forniche axes at various scales is still understudied.Here we explored the eects of range overlap and phylogenetic distance on multiple nicheaxes for closely related species pairs in Western North America monkeyflowers of the plantgenus Mimulus sensu lato (Phrymaceae). Among species pairs, we asked: 1) accounting fordivergence time, what are the relationships with the magnitude of niche divergence and thedegree of range overlap? 2) does the relationship dier across spatial scales or among nicheaxes? and 3) for pairs with partial range overlap, is niche divergence in sympatry larger22Niche divergence and range overlapthan that in allopatry, or vice versa? To test such patterns in a phylogenetic context acrossspatial scales can be useful to infer interactions between niche evolution and range shift overevolutionary time, as well as the role of ecology in diversification.3.3 Methods3.3.1 Study system and phylogeny reconstructionThe plant genus Mimulus sensu lato is a model system in evolutionary ecology (Wu et al.,2008). Western North American Mimulus contains about 75% of the 120 described speciesworldwide (Beardsley et al., 2004); or about 150 taxa in revised classification (Barker et al.,2012), with great variations in range size, habitat preference, and climatic niche breadth(Sheth et al., 2014). Recently there has been a taxonomic revision in this genus (Barker et al.,2012), but most changes are nomenclatural modifications that do not aect our major results(Appendix B.1.1). This study focused on 16 pairs of closely related species (Table B.1),because recently diverged pairs allow us to carry out independent contrasts in a phylogeneticframework.To identify sister pairs and estimate their divergence times, we reconstructed time-calibratedphylogenies with near-complete taxon sampling for North America (N = 120 accessions, in-cluding three outgroup species, and for some species there were multiple samples) in BEASTv1.8.2 (Drummond et al., 2012) on CIPRES ( following Grossenbacheret al. (2014). We applied the GTR+G substitution model for three DNA regions (ITS, ETS,and trnL-F) of Beardsley et al. (2004) and Whittall et al. (2006), a Bayesian uncorrelatedlog-normal relaxed clock models and birth-death speciation model. We conducted threeruns, each for 100 million generations with sampling every 2000 generations. Posterior sam-ples were summarized by Tracer v1.6 (Rambaut et al., 2014) to check the stationary statusof converged chains and eective sampling sizes. Then we used LogCombiner v1.8.2 andTreeAnnotator v1.8.2 (Drummond et al., 2012) to combine samples with 40% burn-in andcompute a maximum clade credibility tree (Figure ??). We also sampled 1000 trees fromthe posterior distribution to account for phylogenetic uncertainty. We scaled the entire phy-logeny to a root depth of 1.0 and expressed branch lengths in relative units. Phylogeneticdistance (PD) of each pair was calculated as the sum of their branch lengths divided by2. Values from the maximum clade credibility tree (PDmcct) and averages based on 1000samples (PDmean) were estimated; the correlation between them was high (r = 0.996). Weused PDmcct in our main analysis below.23Niche divergence and range overlap3.3.2 Occurrence dataOccurrence records were collected mainly from the Global Biodiversity Information Facility( Additional data were collected from several local databases: the Con-sortium of California Herbaria (, the Consortium ofPacific Northwest Herbaria (, the Southwest Environmental In-formation Network (, and Canadensys ( accessed March 2014). After removing duplicate records and observations, we filtereddata of herbarium specimens by excluding records without georeferences, or with mismatchesbetween location descriptions and georeferenced coordinates, or in gardens rather than nativehabitats. For each species, if there was more than one occurrence record in a 1-km-resolutiongrid cell after Albers equal area projection, we further restricted occurrences to one per cellby deleting records at random. The final number of records per species ranged from 7 to2834 (mean: 226; Table B.1).3.3.3 Niche axesThe conception of the niche employed here includes the essential features of Hutchinson’sfundamental niche in multidimensional environmental space and the availability of nichespace relative to the background environment. Therefore, it describes both the conditionswithin which a species persists across spatial scales and the species’ usages, which can bequantified as frequency distributions along various component axes (Broennimann et al.,2012). We separated niche variables into macrohabitat and microhabitat axes according tothe hierarchical scale at which the variables were measured.Three macrohabitat axes were mainly estimated based on range-wide occurrences at thecoarse scale: bioclimatic, edaphic and topographic. Five contemporary bioclimatic variableswere derived from monthly temperature and precipitation data of the time period 1950-2000 at 30-second resolution from WorldClim (, Hijmans et al.2005; for methods see Kou et al. 2011). These included the average temperature of thecoldest month, growing degree days above 0 Celsius degree, precipitation seasonality, thesynchronicity of temperature and precipitation (TPsyn), and growing season aridity. TPsyn(calculated as the Pearson correlation coecient between monthly temperatures and pre-cipitation, ranging from -1 to 1) is an eective index for distinguishing between two majorclimatic classes, the Mediterranean-type climate (low TPsyn) and the monsoon-type cli-mate (high TPsyn) (Kou et al., 2011). Eight edaphic variables mainly describing texturesat 1-km resolution were downloaded from SoilGrids (mean values at the depth of 5-15cm,24Niche divergence and range overlap, including bulk density, coarse fragments volumetric, fractionsand, fraction silt, fraction clay, soil organic carbon content (log-transformed), soil pH, andcation exchange capacity. Four topographical variables derived from 90m-resolution digi-tal elevation models (DEMs, included elevation, slope, roughness,and hillshade (a relative solar radiation index based on aspect and slope), all being log-transformed except hillshade.The microhabitat axes captured habitat attributes at the local scale of individuals withinpopulations. Microhabitat data were collected for 14 species pairs during the growing sea-sons of 2008, 2009, 2013 and 2014. For each pair, populations were sampled from areas ofrange overlap (i.e., regional sympatry) and non-overlap (allopatry; hereafter site type), basedon buered polygon ranges (see below). For each species, 3-5 populations were chosen toencompass latitudinal variation within site types. The number of sites and plots for eachsite type per species ranged 1-7 and 10-68, respectively (Table B.1). we measured slope(by clinometer), canopy (by densitometer), the percent of total vegetation cover (includingoverstory and understory), the percent of rocky ground (rock diameter > 0.2 cm), and thepercent of bare ground (sand and soil), measured within scaled plots (3 * focal plant height)centered on focal individuals of each species. We conducted principal components analysis(PCA) to macro- and microhabitat niche axes separately before quantifying niche divergence(see below).3.3.4 Range overlapThe range area for each species was estimated from a buered polygon, formed by mergingcircles of 50km radius around each occurrence point and then reducing it from the edge by40 km. The 50km radius was chosen to avoid gaps among regionally adjacent circles. Theperimeter was reduced from 50km to 10km to avoid overestimating range extent. The rangeoverlap (ROsum) was a ratio, known as the Jaccard similarity coecient (Phillimore et al.,2008), with the area of overlap between two species’ ranges as the numerator and the areaof the union of two species’ ranges as the denominator. We conducted a sensitivity analysisby evaluating results using alternative methods for estimating range (e.g., minimum convexpolygon) and range overlap (e.g., nestedness) (Appendix B.1.2, B.2.1 and B.2.2).25Niche divergence and range overlap3.3.5 Niche divergenceNiche divergence (ND) between species pairs was quantified for all niche axes separately.For each macrohabitat axis (bioclimatic, edaphic, and topographic), ND was calculated as1 - D, where D is a metric of niche overlap (Schoener, 1970; Warren et al., 2008), and NDranges between 0 (no divergence) and 1 (complete divergence). To estimate D for macro-habitat axes, we used the covariance technique (PCA-env, Broennimann et al., 2012), whichmakes comparisons directly in environmental space, independent of resolution and samplingeort. Additionally, species’ preferences are weighted by the availability of environmentalconditions within its distribution, which avoids systematic underestimation of niche overlapdue simply to dierences in range placement. We created a global Mimulus range by merg-ing all 32 species’ ranges, and sampled 10,000 random points from it. Then we conductedPCA with niche variables associated with background points and created a two-dimensionalenvironmental space as a grid of 100 ◊ 100 cells. Kernel smoothing was applied to projectthe occurrence density for each species across all gridded cells. Similarly, kernel smoothingwas used to estimate environmental availability across random background points withineach species’ range (N = 429 - 9910). The ratio of the density of species to the density ofthe environment in each cell was then standardized by dividing by the maximum ratio. Thestandardized ratio (z), representing corrected species’ occupancy of a given environment,was calculated as 1 minus half of summed absolute dierences in z between two speciesacross all cells. To estimate uncertainty in the degree of niche overlap, we used bootstrap-ping to resample occurrences and background points (75% of original dataset) 200 times, or,when the number of occurrence records was smaller than 10 (for one species), we used jack-knife resampling. Then, we calculated the mean ND value and its bootstrapped standarddeviation.For microhabitat ND, we used a dierent method because background data were not avail-able. We conducted PCA across all species’ microhabitat data (log-transformed first) tocollapse them to first three PC axes. For each axis, the average of each site was calculatedbased on plot-level measurements, and we calculated the average of each species based on sitemeans. We then calculated ND as the absolute dierence between species. For estimatesof ND between allopatry and sympatry within each pair, we calculated species means inallopatry or sympatry separately, using sites falling into the corresponding site types. To ac-count for the uncertainty from our limited site measurements and to avoid any false positivesignificance, we used a simulation method to estimate variation in ND. For each pair alongeach PC axis, mixed eect models were applied to fit the relationship between PC scoresand species using the R package lme4 (v1.1-10, Bates et al., 2015), with species as the fixed26Niche divergence and range overlapeect and sites as the random eect. The model returned means and variances for both fixedand random eects for all sites, which were used to simulate normal distributions for eachsite separately. We then sampled 30 values per site to calculate the site mean. The speciesmean was then calculated as the mean of the site means. ND was estimated as the absolutedierence between species means. We repeated this procedure 200 times, and calculated theaverages of ND and the standard deviation.3.3.6 Relationships among range overlap, phylogenetic distanceand niche divergenceFor each niche axis, we used a multiple linear regression model to test for any significantassociation between ND and RO, with PD being included as a covariate. We built a fullmodel including an interaction term between RO and PD, and a reduced model without it.Models were ranked by AICc and the one with a lower value was reported here.Spatial autocorrelation might confound the association between ND and RO at the coarsescale, leading to an appearance that species pairs distributed closely share more similarniche properties. Though this problem is alleviated to some extent by accounting for thebackground availability of environments when estimating species’ corrected occupancy z, wefurther explored its potential influence on our results by measuring the correlation betweenND and geographic distance between species pairs’ ranges. The geographic distance wascalculated as the distance between the centroids of the two range polygons.3.3.7 Niche divergence in sympatry versus in allopatryFive species pairs have partial range overlap, which means both species have allopatricand sympatric portions of their ranges. Tests for dierences in niche divergence betweensympatry and allopatry were conducted for these pairs. Due to insucient microhabitatdata, one pair was excluded for microhabitat niche comparison. For each macrohabitat axis,ND was estimated (with the same method above) in allopatry and sympatry separatelyusing occurrences and random background data falling into each site type. We also appliedbootstrapping to estimate mean ND and 95% confidence intervals (CI, 95% percentiles).For microhabitat axes, we added site type as the second fixed eect in mixed eect models,everything else being kept the same, and returned mean ND and 95% CI based on 200simulations. For each pair along each niche axis, we tested if mean ND in sympatry waslarger than mean ND in allopatry, and if 95% CI of ND in sympatry was not overlapped with27Niche divergence and range overlapthat in allopatry. If both criteria were met, we concluded there was significantly greater NDin sympatry than in allopatry. We conducted sensitivity tests for macrohabitat with pair-specific range as PCA-env background choice, and for microhabitat with PCA conductedwithin each pair (Appendix B.2.3).3.4 Results3.4.1 PCA of niche variablesFor macrohabitat niche variables, the first two principal component axes summarizing coarse-scaled environmental variation accounted for the majority of variation in the backgroundregion: 84.60%, 77.84% and 83.34% for bioclimatic, edaphic and topographic niches, respec-tively (Figure B.2a-c). For bioclimatic variables, PC1 described variation mainly related togrowing degree days, the average temperature in the coldest month and precipitation sea-sonality (54.20%), and PC2 described variation related to aridity and the synchronicity oftemperature and precipitation (30.40%). For edaphic variables, PC1 was mainly related tovariation in soil pH and coarse and fine particle contents (47.33%), while PC2 was mainlyrelated to medium particle contents and cation exchange capacity (30.51%). For topographicvariables, PC1 captured major variation related to roughness (58.77%), which is highly cor-related with slope, and PC2 captured variation related to hillshade and elevation (24.57%).For the microhabitat niche variables, since there was no sampling for the background spacefrom the whole Mimulus distribution, PCA was conducted for five microhabitat attributesacross all measurements of all species. The first three PC axes explained 84.16% variationacross all surveyed plots (Figure B.2d-f). PC1 (38.94%) was mainly related to % groundthat was bare, % ground that was rock, slope, and canopy (from flat, open and bare to steep,closed and rocky areas). PC2 (24.63%) was mainly related to vegetation coverage, canopyand ground-level rock coverage (closed vegetated vs. open rocky areas), and PC3 (16.06%)was related to slope and vegetation coverage (steep open vs. flat closed areas).3.4.2 Eects of range overlap and phylogenetic distance on nichedivergenceFor every niche axis, ND was better explained by a reduced linear model with only maineects of RO and PD, compared to the full model containing their interaction (Table B.2).ND decreased significantly as RO increased for macrohabitat axes (all P < 0.05), implying28Niche divergence and range overlapspecies pairs share more coarse-scale niche properties when they have more sympatric areas(Table 3.1; Figure 3.2a). ND also decreased with PD for bioclimatic and topographic axes,meaning younger species pairs had larger niche divergence, but no such pattern existed forthe edaphic axis (Table 3.1, Figure 3.2c). Similar patterns held if using PDmean insteadof PDmcct (data not shown). For microhabitat axes, there was no significant associationbetween ND and RO (Table 3.1; Figure 3.2b). Instead, we detected increasing ND withincreasing PD for microhabitat PC2 axis only (related to total vegetation cover, canopy, androcky ground; adjusted R2 = 0.229, P = 0.039), indicating greater niche divergence overtime along this axis. No such relationship was detected for PC1 or PC3 (Figure 3.2d).No significant relationship was detected between RO and PD (adjusted R2 = 0, P = 0.408;nor when using nestedness as RO, Figure B.3), implying no clear pattern regarding thegeographic mode of speciation with this dataset. Furthermore, for all niche axes we obtainednon-significant correlation between ND and range distance (all adjusted R2 = 0, P > 0.4;Figure B.4), and hence we concluded that spatial autocorrelation was not the major driverof patterns with range overlap, though it could still aect the relationship between ND andRO to some degree.3.4.3 Niche divergence in sympatry vs in allopatryFor partially overlapping pairs, the pattern of ND in sympatry compared to ND in allopatrydiered between macrohabitat and microhabitat axes (Figure 3.3). Generally, macrohabitatND in sympatry was mostly significantly smaller than or similar to that in allopatry acrosspairs. There was only one case where mean ND was larger in sympatry (M. congdonii andM. douglasii on topographic axis), but 95% CI were overlapping. Results were similar whenND was calculated based on pair-specific ranges, showing it was not sensitive to the choiceof environmental background (Figure B.5).On the contrary, three species pairs that had greater ND in sympatry than in allopatry, withone pair per microhabitat axis (M. douglasii and M. congdonii on PC1, M. suksdorfii andM. montioides on PC2, M. breweri and M. bicolor on PC3). Two more pairs showed largermean ND on PC3, but the 95% CI were overlapping. The PC axes along which closelyrelated species showed larger ND in sympatry diered among pairs. This suggested thatspecies within pairs might interact with each other dierently and diverge idiosyncraticallywith respect to microhabitat partitioning. When ND was calculated based on pair-specificPCA, results were similar, where pairs showed greater ND in sympatry mainly for PC1 andPC2 (Figure B.6).29Niche divergence and range overlap3.5 DiscussionAmong studies exploring the association between niche divergence and range overlap, quan-titative inferences are still rare in a phylogenetic context. Here our results suggest that en-vironmental filtering plays a role in Mimulus at the coarse scale by selecting for co-occurringspecies with a similar macrohabitat preference. At the fine scale, we did not find a singleoverall signal of microhabitat divergence among all species. However, we detected moremicrohabitat divergence in sympatry than in allopatry when we focused on species pairswith partial range overlap. This is consistent with potential competitive interactions driv-ing closely related species to diverge locally through ecological character displacement (orplasticity), but the specific niche axis that showed divergence was idiosyncratic to each pair.When niche divergence was related to phylogenetic distance, the relationships showed oppo-site patterns across spatial scales.3.5.1 Association between niche divergence and range overlapThe contrasting relationships between niche divergence and range overlap for macrohabitatand microhabitat imply that the processes underlying niche dynamics depend on spatialscale. Inferring dierences between coarse and fine spatial scales would be strengthened byformal statistical tests between niche categories. For example, one could do the Fisher’s ex-act test for patterns observed for both macrohabitat and microhabitat, which would test thesigns (positive or negative) of the dierence in ND between allopatry and sympatry. Suchtest marginally supported the idea of more ND in allopatry for macrohabitats or more NDin sympatry for microhabitats (P = 0.060, data not shown here). For now, conclusions werebased on qualitative dierences between directions of significant relationships within eachscale category. At a coarse scale, negative patterns suggested that species share ecologicalconditions in sympatry and adapt to divergent environments in allopatry. These results areconsistent with environmental filtering determining species’ broad distributions. Similarly,Steinbauer et al. (2016) found climatic niche dierentiation in allopatry was either strongerthan or similar to that in sympatry for six plant clades in the Canary Islands. Kozak andWiens (2010a) suggested range overlap, representing local biotic interactions, had a negativeeect on the rate of climatic niche evolution for 16 clades of plethodontid salamanders, be-cause dispersal of one clade was restricted by the presence of others. Spatial autocorrelationcould potentially lead to higher niche divergence estimates with greater geographic distanceeven if niches are not diverged. For example, organisms, occurring at dierent locations,prefer microhabitats that maintain similar abiotic conditions despite overall macroclimatic30Niche divergence and range overlapdierences. The way we estimated ND by PCA-env technique accounted for the availabilityof environments and spatial resolution, therefore corrected biases in geographic space to somedegree (Broennimann et al., 2012). Besides, we found niche divergence was not significantlyassociated with geographic distance between species’ range centroids. Furthermore, dissim-ilar environmental conditions in dierent locations are indeed potential agents of divergentselection on isolated populations.At the local scale, if closely related species have high range overlap, character displacementis expected to cause microhabitat niche divergence. Contrary to this, in some plant groupsmicrohabitat axes are more conserved than macrohabitat niche (Ackerly et al., 2006; Emeryet al., 2012). Our analyses failed to detect global patterns for any microhabitat axis. Thereare various possible reasons. Firstly, divergence might occur along one or a few niche axes,and such key axes could vary among pairs depending on how species interact. The results forniche divergence in sympatry versus allopatry were consistent with this idea. For example,M. bicolor occurs on very steep slopes in sympatry while M. breweri occurs in flatter areas(personal observ., also supported by divergence on PC3 related to slope); in contrast, bothM. bicolor and M. breweri occur in flatter areas in allopatry. Another pair, M. congdoniioccurs at disturbed or sloped runo areas, while M. douglasii occurs on bare clay, serpentineor granitic soils (Baldwin et al., 2012, also supported by divergence on PC1 related to bareground and slope). When testing the significance of comparisons by estimate whether zerowas excluded by the 95% confidence intervals the dierence in ND between allopatry andsympatry, the results obtained similar inferences. Even two more species pairs were foundwith significantly more ND in sympatry along microhabitat PC3 axis (data not shown here).Together, these could possibly explain no overall pattern when testing across all species pairs.Secondly, increasing sampling eort of local populations and examining more niche axes couldpotentially increase the power to detect divergence at the local scale. The microhabitat datawe collected might not capture relevant factors reflecting niche partitioning. For example,shifts between soil types (e.g., serpentine and non-serpentine) were suggested to be commonfor angiosperms in the California Floristic Province (Anacker and Strauss, 2014; Baldwin,2014). Thirdly, coexistence theory suggests that competitive outcomes among species aredetermined by both stabilizing niche dierences and fitness dierences, such that speciescan coexist by weak stabilization when they have small fitness inequalities (Chesson, 2000;Adler et al., 2007). Close relatives often have similar fitness (Godoy et al., 2014), implyingthat slight niche dierentiation (perhaps less than what we could detect here) would allowcoexistence. Lastly, here we used range overlap as a surrogate for biotic interactions, butit was not a direct metric of interaction intensity. Regionally sympatric species might not31Niche divergence and range overlapco-occur locally, by spatial or temporal partitioning (e.g., dierent substrate patches orflowering times).We applied dierent ways to quantify niche divergence for macrohabitat and microhabitataxes. To account for uncertainty in limited field data and avoid false positive outcomes, thesimulation method was a remedy to variation in estimates of niche divergence. However,this study would have benefited by collecting background microhabitat data at field sitesacross species ranges and quantifying niche divergence relative to a common backgroundniche space.3.5.2 Association between niche divergence and phylogenetic dis-tanceAssociations between range overlap and niche divergence do not imply unidirectional causal-ity, since complex feedbacks can occur through evolutionary time (Donoghue and Moore,2003; Warren et al., 2014). In particular, how daughter species inherit niche properties fromtheir common ancestor during speciation is often associated with range dynamics (Schluter,2001; Coyne and Orr, 2004). In theory, niche divergence could promote speciation in sym-patry, but post-speciation range shifts and niche evolution might remove the signature ofinitial divergence (Barraclough and Vogler, 2000; Losos and Glor, 2003). Likewise, speciesmight retain high niche similarity during allopatric speciation, but adaptation to distant en-vironments and competition during secondary contact might obscure the signature of initialconservatism. Although the pattern of niche conservatism is common in plants (Prinzinget al., 2001), studies have also showed evidence of rapid niche evolution in multiple taxa,either for climatic variables (Broennimann et al., 2007; Kozak and Wiens, 2010a) or micro-habitat (Losos et al., 2003; Silvertown et al., 2006; Wiens et al., 2010). Moreover, it mightdepend on the phylogenetic scale of investigation, as niche conservatism is more apparentwhen including greater phylogenetic diversity (Cavender-Bares et al., 2006; Evans et al.,2009; Peterson, 2011).Here we found that the relationship between niche divergence and phylogenetic distancedepended on the spatial scale. Macrohabitat niche divergence was high between the closestrelatives. Opposite to conservatism, this implies that macrohabitat axes are highly labile,though this might be restricted by the shallow phylogenetic scale applied here. In supportof lability in Mimulus, there was no detectable phylogenetic signal for other bioclimatic vari-ables (Grossenbacher et al., 2014; Sheth et al., 2014). In contrast, microhabitat divergence32Niche divergence and range overlapincreased with phylogenetic distance only along one axis. This suggested that biotic interac-tions at a local scale drive increasing niche divergence over time, either due to accumulatingdierences over time or more intense interactions happening because of secondary contact.When further testing the overall relationship between niche divergence and phylogenetic dis-tance across niche axes, one could average the estimates at the same spatial scale. The meanniche divergence could be averaged with relative estimations (by dividing values by the max-imum for each axis) for three macrohabitat axes or for three macrohabitat axes separately,and then fit them in the multiple regression with range overlap and phylogenetic distance asexplanatory variables. This additional analysis confirmed the negative relationship betweenmacrohabitat niche divergence and phylogenetic distance, but no association for microhab-itat niche (data not shown). To be noted here, this depended on the combination of nicheaxes analyzed in the study.3.5.3 Implications for the geography of speciationLabile macrohabitat niches, with more divergence in younger pairs with low range overlap,and diverging microhabitat over time, suggests both ecological dierentiation and geographicisolation were critical for diversification in monkeyflowers. Decreasing macrohabitat nichedivergence with increasing range overlap and increasing phylogenetic distance is consistentwith environmental filtering or convergent selection operating at a coarse scale, as rangeoverlap increases following allopatric speciation. The lack of significant relationship betweenrange overlap and phylogenetic distance implies either no single dominant speciation mode orprevalent post-speciation range shifts. However, range overlap was generally low (maximum0.23 km2 of overlap per km2 of unioned range), indicating a high degree of geographicisolation consistent with prevalent allopatric speciation. These inferences are seemingly atodds with recent analyses of the geography of speciation in the Californian flora (Anackerand Strauss, 2014) and monkeyflowers in particular (Grossenbacher et al., 2014). Thesestudies showed that younger species pairs had higher asymmetry in niche breadth and rangesize, suggesting a dominance of “budding” speciation. Such niche breadth asymmetry couldcontribute substantially to niche divergence, though we didn’t consider it here. Furthermore,we applied dierent methods of calculating range (instead of minimum convex polygon) andrange overlap (instead of nestedness), which likely are less prone to overestimating range sizeand overlap, but more prone to concluding low range overlap even for fully nested specieswith asymmetric range sizes. Our findings could be consistent with previous conclusions ofthe prevalence of budding speciation if budding events tend to occur in peripheral isolates33Niche divergence and range overlap(i.e., peripatrically, a special case of allopatry).Probabilistic modeling of range evolution to reconstruct ancestral areas can supply di-rect inference of biogeographic processes by reconstructing ancestral areas (e.g., Dispersal-Extinction-Cladogenesis model, Donoghue and Moore, 2003; Ree and Smith, 2008). More-over, modeling of diversification processes while incorporating ecological divergence explicitlywill further test the role of ecology in diversification (e.g., Binary-State Speciation and Ex-tinction model and its extensions, Maddison et al., 2007; FitzJohn, 2010). Future studiescan harness these types of analyses to better understand feedbacks between niche and rangedynamics.34Niche divergence and range overlapTable 3.1: Multiple linear regression fits for testing eects of range overlap (ROsum) and phyloge-netic distance (PDmcct) on niche divergence for each of three coarse-scale macrohabitat axes andthree fine-scale microhabitat PC axes. Significant P values were showed in bold.niche axes slope slope SE p-value adjusted R2macrohabitatbioclimatic ROsum -1.063 0.472 0.042PDmcct -1.841 0.85 0.0495 0.413edaphic ROsum -1.242 0.381 0.006PDmcct -1.16 0.686 0.114 0.497topographic ROsum -0.935 0.235 0.002PDmcct -1.287 0.423 0.009 0.667microhabitatmicro_PC1 (bare ground)† ROsum -1.246 3.289 0.712PDmcct -5.218 5.967 0.401 0micro_PC2 (total vegetation)† ROsum -0.236 2.731 0.932PDmcct 11.602 4.956 0.039 0.229micro_PC3 (slope)† ROsum 1.074 1.869 0.577PDmcct 1.967 3.392 0.574 0† indicated the microhabitat attribute loaded strongly for each PC axis.35Niche divergence and range overlapRONDRONDRONDRONDDivergent selectionEnvironmental filteringStasisCharacter displacementProcess dominating in sympatryProcess dominating in allopatry (a) (b)(c) (d)Figure 3.1: Illustrations of four possible patterns between niche divergence (ND) and range over-lap (RO), predicted under combinations of contrasting processes in sympatry and in allopatryrespectively: (a) environmental filtering in sympatry and divergent selection in allopatry result ina negative relationship (i.e., abiotic factors dominate); (b) character displacement in sympatry anddivergent selection in allopatry result in a non-significant relationship with relatively high nichedivergence; (c) environmental filtering in sympatry and evolutionary stasis in allopatry result ina non-significant relationship with relatively low niche divergence; (d) character displacement insympatry and evolutionary stasis in allopatry result in a positive relationship (i.e., biotic factorsdominate). Range overlap is treated as a continuous variable, with complete sympatry and completeallopatry at its two extremes.36Niche divergence and range overlap●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●bioclimatic edaphic topographic0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 overlapniche divergence(a)●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●micro_PC1 (bare ground) micro_PC2 (total vegetation) micro_PC3 (slope)0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 overlapniche divergence(b)●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●bioclimatic edaphic topographic0.05 0.10 0.05 0.10 0.05 distanceniche divergence(c)●●●●●● ●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●micro_PC1 (bare ground) micro_PC2 (total vegetation) micro_PC3 (slope)0.05 0.10 0.05 0.10 0.05 distanceniche divergence(d)Figure 3.2: Conditional regression plots show the relationships between niche divergence andindividual covariates (range overlap and phylogenetic distance) from multiple regression modelsin Mimulus. (a)(c) for three coarse-scale macrohabitat axes and (b)(d) for first three fine-scalemicrohabitat PC axes. Error bars are estimations of standard deviation from (a)(c) bootstrap and(b)(d) mixed-eect model-based simulation. Best-fit lines are from significant linear regressions(P<0.05; correct slopes as in Table 3.1), while controlling for the other covariate (being set asmedian values); gray shading denotes 95% confidence intervals.37Nichedivergenceandrangeoverlap●●●●●●●● ●●●●bioclimatic edaphic topographicallo sym allo sym allo sym0.200.400.600.800.200.400.600.800.200.400.600.801.00niche divergence species pairs●●M. breweri, M. bicolorM. cardinalis, M. parishiiM. constrictus, M. whitneyiM. douglasii, M. congdoniiM. suksdorfii, M. montioides(a)●●●●●●●●●●●●micro_PC1 (bare ground) micro_PC2 (total vegetation) micro_PC3 (slope)allo sym allo sym allo sym0.000.501. divergencespecies pairs●●M. breweri, M. bicolorM. constrictus, M. whitneyiM. douglasii, M. congdoniiM. suksdorfii, M. montioides(b)Figure 3.3: Niche divergence in allopatry (allo) versus niche divergence in sympatry (sym) for Mimulus species pairs with partial rangeoverlap. (a) for five pairs along three coarse-scale macrohabitat axes; (b) for four pairs along first three microhabitat PC axes. Errorbars are estimations of 95% confidence intervals from (a) bootstrap and (b) mixed-eect model-based simulation. Solid lines denotesignificantly more niche divergence in comparison.38The evolution of niche breadthChapter 4 The evolution of niche breadthin Mimulus4.1 SummarySpecies show remarkable variation in niche breadth, but the directionality of niche breadthevolution remains a question for various niche variables across spatial scales. Testing theassociation between niche breadth evolution and the process of species diversification couldshed light on the role of ecology in speciation. Here we applied Cladogenetic State changeSpeciation and Extinction models in western North American monkeyflowers (Mimulus sensulato) to estimate the eects of generalization versus specialization along coarse-scale biocli-matic and local-scale microhabitat variables on diversification rates. Dierent niche variablesshowed opposite patterns across spatial scales. Higher diversification rates were detectedfor coarse-scale bioclimatic generalists and for local-scale microhabitat specialists. Further-more, we found a weak generalist-to-specialist trend for bioclimatic niche breadth, but aweak specialist-to-generalist trend for microhabitat niche breadth. Both cladogenetic andanagenetic changes were comparable hence important in the evolution of niche breadth inthis genus. Together, it suggested that the underlying mechanisms of niche breadth evo-lution might dier across spatial scales, and that change in niche breadth could be tightlyassociated with diversification process.4.2 IntroductionNiche breadth (or niche width) describes a species’ preference or tolerance along one envi-ronmental axis (Slatyer et al., 2013). It has two components, variation among individualswithin populations and variation among populations across the geographic distribution (Fu-tuyma and Moreno, 1988); here we focus on the latter. Specialization generally describeshaving a narrower niche breadth relative to related species (Simpson, 1953; Schluter, 2000b).As a consequence of fitness trade-os, speciation and extinction rates are hypothesized tobe unequal between specialists and generalists. Specialization may be advantageous if it isselected for adaptation in a certain environment, while generalization may promote adap-tation across a wider range of environments but by sacrificing high fitness in any given39The evolution of niche breadthenvironment (i.e., a jack-of-all trades is master of none). However, specialization is oftenthought to be an evolutionary dead end because organisms are always facing unpredictableand unfavorable environments, and hence narrowed niche breadth could make lineages proneto extinction (Colles et al., 2009; Day et al., 2016). Specialists are often associated with re-stricted distributions and insucient genetic variation for adaptation to new niches, leadingto reduced speciation rate and higher risk of extinction under environmental changes (Dayet al., 2016). In contrast, generalists could have a greater potential to colonize new habitatsand give rise to new lineages. However, thus far, macroevolutionary analyses on generaliza-tion/specialization have obtained mixed results regarding the eects of specialization andgeneralization on diversification rate (Schluter, 2000b; Day et al., 2016).In addition to potential dierences in diversification rates, generalists and specialists mightshow evolutionary biases in the niche characteristics of descendant species. A long-standinggeneralist-to-specialist hypothesis is that ecological generalists give rise to specialists (Simp-son, 1953) more often than the reverse direction. There are several reasons to propose a biastowards specialist descendants. Due to competition for limited resources, descendants parti-tion ancestral niches as diversification proceeds, with progressively narrower niche breadthsdue to increasing number of species (Schluter, 2000b). This also leads to decreasing vari-ance as more ecological niches have already been occupied (Freckleton and Harvey, 2006).Moreover, because of strong selection favoring specialization, it is dicult and unlikely to re-verse to generalists, because specialists might lack genetic variation needed for generalization(Whitlock, 1996). Conversely, generalist lineages are not the only source of new species, andmeta-analyses have shown that reversals from specialists to generalists are not impossible(Schluter, 2000b; Vamosi et al., 2014). Specialists may expand into new adaptive zones, andeventually become generalists over time. Interspecific competition was even shown to pro-mote generalization in host specificity of parasites on birds (Johnson et al., 2009). Moreover,specialized lineages might shift in specialty (e.g., colonization of new habitats, host shifts),which could be accompanied with speciation events without major changes in breadth. Thepossible mechanisms could be either specialization promotes speciation by reducing geneflow or speciation in itself generates specialization (Hardy and Otto, 2014). Meta-analysesof 19 clades suggested that the evolutionary direction of niche breadth is unpredictable, andthe state of ancestors might not necessarily be generalists (Schluter, 2000b).The evolution of niche breadth could happen as cladogenetic changes during speciation orhappen as anagenetic changes over time. Testing how often changes in niche breadth arecladogenetic could reveal the role of ecology during speciation. The advances in phylogeneticmethods (Maddison et al., 2007; Goldberg and IgiÊ, 2012) and computational abilities have40The evolution of niche breadthrealized the estimations of relevant parameters. Both stages could be greatly intertwinedwith the geographic modes of speciation and post-speciation range shifts. Previous studieshave suggested that most plants in California speciated in restricted areas near the edge orinside of the ancestral range (budding speciation, Anacker and Strauss, 2014; Grossenbacheret al., 2014). Though the strength might vary depending on niche axes and niche positions,range size is normally thought to be positively correlated with niche breadth (Slatyer et al.,2013). Therefore, budding speciation implies a potential evolutionary trend from generalistancestors to specialist oshoots that is concentrated at phylogenetic nodes, at least forrange-wide, coarse-scale niche variables (e.g., bioclimatic). The descendants might maintainspecialized status or might gradually become generalists by niche expansion in novel orvarying environments and give birth to more species, leading to an evolutionary trend fromspecialization to generalization along branches. At a local scale, microhabitat preferencemight shift in the same way for descendants during budding speciation. But there couldbe a possibility that wide-ranging ancestor had a narrow microhabitat preference, whiledescendants might obtain new traits that help them occupy more microhabitat types eventhough they had restricted ranges.Furthermore, most studies of generalization and specialization have focused on niche vari-ables at a fine scale (such as host, pollinator, diet, and habitat type, etc.) and have notcome to an agreement (Schluter, 2000b; Anderson et al., 2014; Hardy and Otto, 2014). Yet,due to dissimilarity between species among niche variables across spatial scales (both coarseand local, van der Niet and Johnson, 2009), species might perform dierently across variousniche axes, e.g., a specialist along some axes but a generalist along others (Litsios et al.,2014). It is unclear whether the degrees of specialization are consistent among niche axes atthe same spatial scale or across spatial scales. Meanwhile it has been understudied whetheran evolutionary trend towards specialization exists across spatial scales or not.Here we applied a phylogenetic comparative approach to investigate the association betweendiversification and the evolution of niche breadth in a very diverse plant group, western NorthAmerican monkeyflowers (Mimulus sensu lato, Phrymaceae). Niche breadth was treatedas a binary state, generalist or specialist, and I focused on three coarse-scale bioclimaticvariables and two local-scale microhabitat variables. We fitted Cladogenetic State changeSpeciation and Extinction models (ClaSSE, Goldberg and IgiÊ, 2012), which can jointlyestimate diversification and character change processes. We asked: 1) do diversification ratesdier between generalists or specialists? 2) are transition rates from generalist to specialist(G-to-S) or generalist to specialist (S-to-G) more common? 3) is the transition more likelyto happen during speciation or over time (cladogenetic vs. anagenetic)? By answering41The evolution of niche breadthsuch questions, insights can be obtained about the direction and mode of niche evolutionand explore the role of ecology in diversification process across spatial scales. We expecthigher diversification rates in state G and a G-to-S trend for coarse-scale niche variables,especially at cladogenesis; but for local-scale niche variables it might be the same pattern,or an opposite one. We also expect to see higher rates of cladogenetic than anageneticchanges because dramatic changes in range size during budding speciation could lead tocorresponding changes in niche breadth.4.3 Methods4.3.1 Study systemThe plant genus Mimulus sensu lato (Phrymaceae) is a model system in evolutionary ecol-ogy (Barker et al., 2012). Western North American Mimulus contains about 75% of the 120described species worldwide (Beardsley et al., 2004) or ~ 140 out of 178 taxa in revised classi-fication (two major genera, Diplacus and Erythranthe, Barker et al., 2012). Recent taxonomicrevision in this genus suggested many nomenclatural modifications (Barker et al., 2012), aswell as new species delimitations. About 20% of 82 species involved in this study were basedon revised species circumscriptions. For example, some varieties of M. aurantiacus are nowat species status in Diplacus, and five revised species were proposed for M. montioides andM. palmeri (previously subspecies, Fraga, 2012); we used the recent, narrower circumscrip-tions here. We made respective changes on the phylogeny when new species were describedor when species were taken out of synonymy. Thus, we retained the genus name Mimulushere for simplicity but included recent taxonomic conversions to our compiled data (TableC.1). Nonetheless, this diverse wildflower group shows great variation in life history, habitatpreference, range size, and climatic niche position and breadth (Barker et al., 2012; Shethet al., 2014; Grossenbacher et al., 2014). Putting niche properties on a robust phylogeny forMimulus allows us to carry out macroevolutionary analyses to test the association betweenniche evolution and diversification.4.3.2 Bioclimatic niche variables and designation of niche breadthstateNiche variables in this study included two categories at dierent spatial scales: one relatedto bioclimatic variables at a coarse scale and the other related to microhabitat preference at42The evolution of niche breadtha local scale (see below). Niche breadth was quantified as species’ preferred range along eachcontinuous environmental gradient (or the count of number of types of an environmentalgradient with discrete categories). Along each niche variable, quantitative estimates of nichebreadth were later converted into binary states, generalist or specialist, based on a certainthreshold (see details below).Three bioclimatic niche variables were obtained based on range-wide locality data for eachspecies. Occurrence data for 82 species were compiled mainly from the Global BiodiversityInformation Facility (GBIF, Additional data were collected from sev-eral local databases: the Consortium of California Herbaria (CCH,,the Consortium of Pacific Northwest Herbaria (CPNWH, andthe Southwest Environmental Information Network (SEINet, (allaccessed January 2017). Records for all species in genusMimulus, Erythranthe, and Diplacuswere downloaded to capture the uncertainty from taxonomic revision. After removing dupli-cate records and observations (i.e., those without physical specimens), we filtered specimendata for quality by excluding records without georeferences or with mismatched coordinatesbetween the location descriptions and the georeferenced ones. We also confirmed that local-ities spanned most of the species’ range (i.e., localities were not clumped in a small part ofthe range). We then applied Albers equal area projection to locality data. For each species,if there was more than one occurrence in a 1-km-resolution grid cell, localities were furtherrestricted to one per cell by deleting extra records at random. We finally obtained 11,688records, and the number of locality per species ranged from 3 to 1074 (mean = 224, median= 84; Table C.1). We created 100-km-radius circles around each locality and merged allinto a single polygon as the common Mimulus distribution, from which we generated 10,000random background points.According to a previous study on predictor importance (chapter 2), we decided here tofocus on three bioclimatic variables that are closely related to plant life history character-istics and predict Mimulus distributions well: Pseason (precipitation seasonality), TPsyn (thesynchronicity of temperature and precipitation), and the first PC axis of two highly corre-lated variables, Tcold (the average temperature of the coldest month) and GDD0 (growingdegree days above 0 Celsius degree). These variables reflect thermal constraints, seasonalwater availability, and the interrelationship between temperature and precipitation, all es-sential for organismal distributions. Bioclimatic layers of Pseason, Tcold, monthly tempera-ture and monthly precipitation were downloaded at a 30-second resolution from WorldClim( TPsyn was a normalized index to denote the correlation co-ecient between monthly temperature and monthly precipitation (Kou et al., 2011), which43The evolution of niche breadthdistinguishes regional patterns of summer green versus Mediterranean vegetation distribu-tions. GDD0 was calculated with 0 degree as a threshold indicating the start of growthseason. Four derived bioclimatic layers were clipped by the common Mimulus backgroundand values for both species’ localities and background points were extracted. For Tcold andGDD0, we then conducted PCA with background points, and obtained values of PC1 forspecies localities (denoted as GDD0 for simplification hereafter).For each bioclimatic variable, values of background points were used to create the environ-mental gradient with the range being the upper and lower bounds. This gradient was dividedinto 200 bins and a kernel smoothing function was applied to project the occurrence densityof each species along it. Niche breadth was quantified as the preference range along an envi-ronmental gradient. In practice, we counted the number of bins until the accumulative kerneldensity reached 0.9 (i.e., the number of bins covering 90% under the kernel density curve).Another approach was to consider niche breadth as the variation along a gradient, by resam-pling species occurrences along the 200-bin gradient with probability being the correspondingkernel density, and calculating the variance (sample size = 200). Results of niche breadthestimates were quite consistent between both approaches (R = 0.972, 0.940, and 0.959 forthree variables separately, all P < 0.001), and we used the first one for downstream analysesbecause it was technically consistent with quantification of microhabitat niche breadth. Wechose the mean niche breadth (log-transformed) across 82 species as the threshold to convertniche breadth into binary states. Species with niche breadth equal to or above the thresholdwere coded as generalists (G), while species with niche breadth below the threshold werecoded as specialists (S) (see Figure 4.1a for distribution of continuous niche breadth valuesin Mimulus with thresholds being means).4.3.3 Microhabitat niche variables and designation of niche breadthstateTwo microhabitat niche variables were habitat water anity (HWA) and substrate type(Soil). These two variables are tightly associated with individual growth of organisms ata local scale, therefore having a powerful influence on species distribution and abundance.We used HWA to describe species preferences across a dry-wet soil moisture gradient. Weused Soil to capture species preferences across various substrate types, emphasizing chemicalproperties with potential eects on plant nutrient availability.Text analysis was conducted on habitat descriptions from range-wide herbarium specimen44The evolution of niche breadthrecords. Available habitat descriptions were gathered after removing duplicates among dif-ferent sources via accession number and removing records at the same locality via longi-tude/latitude. We also added records from GBIF if they were not included in above threesources as described for climatic variables. We obtained 25521 records in total, ranging from2 to 5137, and average 311 records per species (Table C.1). We applied the tm package in R(v0.7-1, Feinerer et al., 2017) to convert raw text into lower case and to remove punctuations,numbers and white spaces, prior to text analysis on cleaned data. We kept any originallymisspelled words, but corrected them during later quantification of word frequencies. Weobtained 10,209 words in total, and a word matrix was generated with frequency for eachword for each species. We extracted words and categorized them into either HWA-relatedor Soil-related groups (242 and 133 words, respectively).For each variable, we created a spectrum with discrete bins. For HWA, the spectrum hadfive bins based on a dry-to-wet gradient. Each HWA-related word was assigned into a singlebin. Meanwhile the word was also assigned a weight value: 1 for direct adjectives describingmoisture conditions (e.g., wet, dry, moist, damp, etc.); 0.5 for indirect but relevant adjectives(e.g., saturated, vernally, seasonal, etc.), and 0.1 for nouns that were related to moisture(e.g., river, pool, lake, etc.). Nouns were assigned low weights because nouns were likelyto carry redundant information in one specimen record, e.g., "on wet soil along riverbank",and they are also more likely to describe general locations rather than actual soil moistureconditions, e.g., "near Lake Almador" (see Table C.2 for details on word weighting). Withthe word matrix, we counted word frequency in each bin, multiplied the frequency and thecorresponding weight, calculated the relative frequency in each bin as the sum of the weightedfrequencies of all words, and further standardized it by dividing it by summed frequencies overall five bins. For each species, we estimated HWA index as the average relative frequencyin each bin through bootstrap resampling. We then calculated the cumulative frequencystarting with the bin with the highest HWA index. When the cumulative frequency reacheda threshold of 0.90, we set HWA index of bins left to be 0. The average number of consecutivedry-to-wet bins covered was 3.2 across species. We finally converted HWA to a binary state.We coded a species to be G if it had non-zero HWA index covering 4-5 consecutive bins(where there might be one or two empty bins in the middle); while we coded a species to beS if it had non-zero HWA index covering three or fewer consecutive bins (Figure 4.1b).Soil types were defined based on parent materials, which determine the predominant mineralsin the soil and therefore has a direct impact on soil chemistry. For Soil, we used discrete binsto represent separate substrate types (e.g., serpentine, granite, limestone, etc.). We refinedthe soil spectrum by excluding a few soil types from the original pool, because the summed45The evolution of niche breadthword frequency of each type over all species was fewer than five. Four species occurringon these excluded soil types were not strictly restricted to them. Finally, we obtained 19discrete bins, and each Soil-related word was assigned into a single type bin without weighting(Table C.3). We estimated relative frequencies in each bin for each species. Similarly, afterthe cumulative frequency reached 0.90, the relative frequency was set to be 0 for remainingbins. Then we estimated Soil index as the number of non-zero bins for each species. Forfour species with no information about soil in their habitat descriptions due to small samplesizes of locality records, we set Soil index to be one for them (M. nelsonii, M. rupestris,M. wiensii, M. yecorensis). The averaged Soil index was 2.9 over 82 species. Finally, weconverted Soil into a binary state by assigning a species to be G if its Soil index was equalto or larger than three and assigning a species to be S if its Soil index was no larger thantwo (Figure 4.1b).4.3.4 Binary trait correlationCorrelated evolution between niche variables might lead to similar associations with diver-sification processes. Before fitting state-dependent diversification models, we applied thefunction fitPagel in R package phytools (v0.6-00, Revell, 2012) to test if any combinationof two binary niche variables were correlated, meaning the evolution of one character state(G or S) was aected by (or depended on) the second one (Pagel, 1994; Revell, 2012). Aset of 100 trees of Mimulus from a BEAST posterior sample (from chapter 3) were appliedin this study to accounting for uncertainties in phylogeny topology and branch length. Weexpected bioclimatic variables would show some degree of correlation, since bioclimatic nichebreadth was suggested to predict range size and various bioclimatic variables might posi-tively associated with each other via geographic distribution (though one species being Gon one axis might not necessarily be G on other axes). We expected no correlation betweenbioclimatic niche and microhabitat niche, or between two microhabitat variables.4.3.5 State-dependent diversification and niche breadth evolutionWe tested whether the diversification process was state-dependent for each niche variablewith the ClaSSE model, and estimated compound parameters of interest for evolution trendsand modes in the R package Diversitree (v0.9-10, FitzJohn, 2010). The ClaSSE model is anextension of BiSSE model (equivalent to BiSSEness, Maddison et al., 2007; Magnuson-Fordand Otto, 2012), allowing parameter estimations for cladogenetic changes of character statesduring speciation. We implemented the "skeleton tree" approach to correct for incomplete46The evolution of niche breadthtaxonomic sampling (with sampling fraction for 82 out of 140 species: 0.586), which assumedthat missing taxa are randomly distributed across the tree (FitzJohn et al., 2009). Weconsidered a model with eight free parameters and referred to it as the "full" model: twospeciation rates (⁄S and ⁄G), two additional speciation rates associated with state change(⁄SSG and ⁄GGS), two extinction rate (µS and µG), and two transition rates (qSG and qGS).Here we did not consider the scenario under which both descendants obtained dierent statescompared to their ancestor (⁄GSS = ⁄SGG = 0), assuming at least one descendant kept thesame state as its ancestor.Firstly, we conducted model fitting with maximum likelihood (ML) methods and comparedsix candidate models. For each niche variable, we created a full ClaSSE model with eight freeparameters and five restricted models (Table 4.1), to assess whether a simpler model would bemore supported by the data. Within those restricted models, we either set symmetric ratesof speciation, or set asymmetric transition between generalist and specialist, or excludedcladogenetic or anagenetic mode of change. Specifically, we developed two restricted modelsto test the evolutionary direction between two states: G-to-S and S-to-G models. These tworestricted models represented two opposite scenarios of completely irreversible evolution. Inthe G-to-S model, we allowed changes only to happen from generalist to specialist (i.e., ⁄GGSand qGS), by setting rates for the reverse direction to be zero (i.e., ⁄SSG and qSG). In the S-to-G model, we allowed changes only to happen from specialist to generalist, with completelyopposite settings. We fit all models to each of 100 trees. Each time, we fitted a modelwith 10 random starting points to account for uncertainty of multimodality in the likelihoodsurface and kept the one with the maximum likelihood. The set of top models for each treecomprised the model with the lowest Akaike information criterion (AIC) value and thosewithin two AIC units from it. Furthermore, we fitted models with the same extinctionrates for both states (just for top models for each niche variables), and then compared withcorresponding top models by likelihood ratio test (LRT).Since there was no single simpler model that could explain the data (see details in Results),we secondly applied Bayesian analysis with a full ClaSSE model. For each variable, weapplied Bayesian techniques via Markov chain Monte Carlo (MCMC) with 10,000 steps oneach of 100 trees. Compound parameters of interest were calculated within each step. TheMCMC runs were further combined across 100 trees after removing a burn-in of 2000 steps pertree to form posterior distributions of parameters, accounting for phylogenetic uncertainties.We estimated diversification rates for generalist and specialist separately (i.e., rG vs. rS,or ⁄GGG + ⁄GGS ≠ µG vs. ⁄SSS + ⁄SSG ≠ µS), and used the hdr (highest density region)function in Diversitree to calculate the 95% credibility interval. We calculated the dierence47The evolution of niche breadthin diversification rates between two states (r = rG ≠ rS). To test the evolutionary trendof niche breadth, we estimated the total evolutionary transition direction between generalistand specialist (G-to-S vs. S-to-G), combining both cladogenetic and anagenetic changes(i.e., ⁄GGS + qGS vs. ⁄SSG + qSG). Meanwhile, we also separated evolutionary trend forcladogenetic and anagenetic events (e.g., ⁄GGS vs. ⁄SGS for during speciation). To testthe dominant evolutionary mode, we estimated total cladogenetic changes versus anageneticchanges, combining both evolutionary directions (i.e., ⁄GGS+⁄SSG vs. qGS+qSG). Meanwhile,we also separated evolution modes for generalist and specialist (e.g., ⁄GGS vs. qGS for G).A statistical comparison between any two rates was conducted by calculating the dierencebetween them within each MCMC step and estimating the 95% credibility interval of theposterior distribution. If zero was excluded from the 95% credibility interval, two rates wereconsidered significantly dierent.4.4 Results4.4.1 States of niche breadth and their correlated evolutionThe encoding of generalist or specialist (G or S) based on a threshold of mean niche breadthresulted a relatively even distribution of states across the monkeyflower phylogeny, withoutobvious clusters (Figure 4.2). The ratios of G/S were 44/38, 33/49, 46/36, 61/21, and 45/37,for Pseason, TPsyn, GDD0, HWA, and Soil, respectively.Tests of binary trait correlations returned dierent results among ten combinations of nichevariables (Table 4.2). Pseason and TPsyn showed significant correlation among 100 trees (allP from LRT < 0.05), implying possibly dependent evolutionary histories. The same patternheld between Pseason and GDD0, and between HWA and Soil (all P < 0.05). Significantcorrelation also existed between GDD0 and HWA (92 out of 100 trees), and between Pseasonand Soil (38 out of 100 trees). No significance was detected for other combinations, especiallybetween bioclimatic variables and microhabitat variables.4.4.2 Diversification analysesMaximum likelihood model comparison across six candidate models returned dierent resultsfor dierent niche variables (Table 4.3). For Pseason, the BiSSE model obtained the high-est support in terms of the number of appearances among top models across 100 trees (>80%). For TPsyn, the G-to-S model (mG2S) was commonly supported (> 90%), and the sec-48The evolution of niche breadthondly supported one was the cladogenetic-only model. For GDD0, the full ClaSSE model,the cladogenetic-only model and the G-to-S model (mG2S) were generally supported (>75%). For HWA, the full ClaSSE model, the cladogenetic-only model and the G-to-S model(mG2S) were generally preferred (> 60%). For Soil, the full ClaSSE, the cladogenetic-onlymodel and the S-to-G model (mS2G) were generally supported (Ø 60%). For most variables,models with asymmetric cladogenetic changes were generally preferred, while models withsymmetric or no cladogenetic changes were less supported. The only exception was Pseason,for which the full BiSSE model with no cladogenetic change was generally more supported.Two simpler models, the anagenetic-only model (m.BiSSE) and the cladogenetic-only model(m.clado), were not generally rejected across 100 trees. However, our model comparison pro-cedure did not identify a single simpler model across all variables, nor even for each variable.Moreover, multiple non-nested models (e.g., m.sym vs. G-to-S model) were compatible andnot significantly dierent from the full model (across ~ 90% trees by LRT; data not shown).Therefore, considering the equivalence among top models, we focused on Bayesian MCMCanalyses on the full ClaSSE model to test the direction and mode of niche breadth evolutionwhile making comparisons across niche variables.Parameters estimated via the full ClaSSE models showed opposite results for niche vari-ables depending on spatial scale (Table C.4). For three coarse-scale bioclimatic variables(Pseason, TPsyn, and GDD0 ), the median rates indicated a higher diversification in gener-alist lineages. However, posterior distributions for diversification rates in two states acrossthe MCMC steps were overlapping to various extends (Figure 4.3a). Moreover, the 95%credibility intervals for the dierence in these two rates included zero (rG ≠ rS), meaningthey were not significantly dierent (Table C.5). For GDD0, the overlapping area betweenposterior distributions for diversification rates in two states was quite narrow, which couldbe further explained by the dierence in speciation rates in two states (Figure 4.3b), thoughthis dierent was not supported by the 95% credibility interval for GDD0 (Table C.5). Theextinction rates in two states were not significantly dierent, which was supported by quiteclose median values, highly overlapping posterior distributions of estimations (Figure 4.3c),and the 95% credibility intervals. On the contrary, for two fine-scale microhabitat variables,the median estimations indicated higher diversification rates in specialist lineages. Whenfurther exploring the components of diversification process, we found that posterior distri-butions for speciation rates of generalists and specialists across the MCMC steps were highlyoverlapping for HWA, but not for Soil (Figure 4.3b). Meanwhile, the 95% credibility inter-vals for the dierence in speciation rates of two states for Soil excluded zero, suggestinga significantly higher speciation of specialist lineages (Table C.4; Table C.5). The extinc-49The evolution of niche breadthtion rates in two states for microhabitat variables were not significantly dierent, similar tobioclimatic variables.Thus, estimates of diversification rates for specialists versus generalists were comparablewithin each niche category (i.e. similar among all coarse-scaled bioclimatic variables andsimilar between two fine-scale microhabitat variables), but opposite between scales (Figure4.3).4.4.3 The evolutionary direction of niche breadthSimilar to diversification rates, estimations of the evolutionary direction showed oppositetrends between niche variables across spatial scales. We observed a weak G-to-S trendfor three bioclimatic niches, while we observed a weak S-to-G trend for two microhabitatvariables (Figure 4.4).Combining both cladogenetic and anagenetic changes, median compound rates of total G-to-S trend were nearly 1.5-2.5 times higher than the reverse S-to-G direction for bioclimaticvariables (Table C.4). Conversely, microhabitat variables showed 2.5-5 times higher mediancompound rates of total S-to-G trend than the reverse direction. However, these patternswere not strongly supported by analyses across MCMC samples, which showed these twocompound rates were not significantly dierent from each other (Figure 4.4), because the 95%credibility intervals for the dierence in two rates included zero in the posterior distributions(Table C.4).When partitioning transition asymmetries during speciation from those occurring post-speciation processes (i.e., cladogenetic vs. anagenetic), we found that the asymmetries de-scribed above were more enhanced during speciation events (Figure 4.4b). More specifically,for three bioclimatic variables, we found relatively stronger patterns during speciation forthe G-to-S trend (Table C.4). Rates of cladogenetic G-to-S transition (⁄GGS) were nearly2.5-6 times higher than rates of cladogenetic S-to-G transition (⁄SSG). For anagenetic tran-sitions, the rates of G-to-S and S-to-G were close. Similarly, for two microhabitat variables,the relatively stronger S-to-G trends were more evident during speciation too (Table C.4).Rates of cladogenetic S-to-G transition ⁄SSG) were more than 6-fold greater than rates of thereverse transition (⁄GGS) at cladogenesis. Rates of anagenetic S-to-G transition were alsohigher, but with a lower magnitude. However, due to overlapping posterior distributions,these trends in transition rates were not significant (Figure 4.4c, Table C.5).50The evolution of niche breadth4.4.4 The mode of state transitionFurther comparisons between transition modes (i.e., cladogenetic vs. anagenetic) showedindistinguishable patterns. Overall, the median rates associated with cladogenetic changes(including both G-to-S and S-to-G transitions) were comparable to the rates associated withanagenetic changes (Table C.4). The posterior distributions of rates of the two modes werehighly overlapping (Figure 4.5a). When partitioning modes for each transition direction,we did not find any stronger pattern for five niche variables (Figure 4.5b and 4.5c). Morespecifically, for the G-to-S transition, cladogenetic changes happened slightly more oftenthan anagenetic changes for bioclimatic variables (⁄GGS > qGS, except for Pseason). Con-versely, anagenetic changes happened more often for two microhabitat variables with ratesabout 2.5 times higher (⁄GGS < qGS). For S-to-G transitions, bioclimatic variables showedslightly higher median rates of anagenetic changes (3-5 times higher; ⁄SSG < qSG), whereas,two microhabitat variables showed higher median rates of cladogenetic changes (1.3-7 timeshigher; ⁄SSG > qSG). However, posterior distributions from MCMC steps for all of theabove partitioning comparisons cannot distinguish the significance of dierence between thecladogenetic and anagenetic modes (Table C.5). The result that ML model comparison didnot reject the cladogenetic-only model or the anagenetic-only model also complicated theinference of such speciational and gradual changes in this system.4.5 DiscussionTogether, we found dierent evolutionary patterns for niche breadth evolution across spatialscales in Mimulus. Here we showed higher diversification rates and a weak G-to-S trend,though not significant, for coarse-scale bioclimatic niches, especially during speciation. Incontrast, we found higher diversification rates and a weak S-to-G trend for local-scale mi-crohabitat niches, though not significant. However, we did not detect a predominant mode(cladogenetic or anagenetic) of temporal change during niche breadth evolution when nichebreadth states were ignored.4.5.1 Diversification in generalists versus specialistsIf generalist lineages (G) diversify more rapidly than specialist lineages (S), we may mistak-enly conclude that the evolutionary trend is more likely from specialist to generalist, or viceversa (Maddison, 2006; Maddison et al., 2007). We were trying to disentangle the abovetwo processes with estimations of anagenetic and cladogenetic changes. Though through51The evolution of niche breadthML model comparison procedure, the cladogenetic-only model or the anagenetic-only modelwere not completely rejected across five niche variables considered here, we applied the fullClaSSE model fitting here to fully consider both potential changes. The estimations of di-versification rates between the two states showed contrasting patterns for niche variablesquantified at dierent spatial scales. Generalists had higher median rates of diversificationfor bioclimatic niches. This is consistent with the idea that coarse-scale temperature andprecipitation factors determine species’ broad geographic distributions (Slatyer et al., 2013;Sheth et al., 2014), and wide-ranging species have a higher chance to give rise to new species.On the other hand, higher median rates of diversification in microhabitat specialists couldindicate dierent mechanisms of speciation. Firstly, it could be related to the spatial distri-butions of macroclimate versus microhabitat. Macroclimates could tend to be more spatiallyautocorrelated and gradually changing as species evolves, while microhabitat might be highlypatchy. Specialization to distinct microhabitats might lead to fewer opportunities for geneflow because of patchiness, hence higher possibility for speciation (Nosil, 2012). Secondly,previous studies show specialized evolutionary outcomes across various plant systems, wherepreference on soil moisture and soil type might be associated with character shifts due tobiotic interactions, and restricted preference on microhabitat might provide evolutionaryrefuges (Kiang and Hamrick, 1978; Ivey and Carr, 2012, e.g., drought escape with selfing,etc.). Hence specialization might not necessarily be a dead end in this sense.4.5.2 No consistent evolutionary trend across spatial scalesNiche dierentiation along various axes could happen at dierent rates and stages as evolu-tion proceeds. Changes of niche breadth could occur cladogenetically or anagenetically, con-tributing to the overall trends. On the one hand, range shifts during speciation (e.g., buddingspeciation, or progenitor-derivative relationships Gottlieb, 2004) might indicate punctuatedniche breadth shifts. Specialist descendants could be formed from a generalist ancestor di-rectly by speciation events when they occupy restricted patches, under which divergence wasaccompanied by speciation. Asymmetric range sizes for two descendants, along with asym-metric niche breadths, indicate a way of resource subdivision. Whereas, from a specializedancestor, descendants might acquire some novel traits or exhibit trait reversals that enableadaptation into newly-encountered habitats or frequently changing environments, thus ex-panding niche breadths at a local scale. On the other hand, a specialist could graduallybecome a generalist through anagenetic changes by range expansion as well as niche expan-sion; while a generalist can become specialized due to lack of appropriate genetic variationunder climate change with range shrinkage (Schluter, 2000b). Biotic interactions might lead52The evolution of niche breadthto narrowed niche breadth in a scenario of secondary contact following allopatric speciation,in which niches of closely related species might diverge and become much narrower due tointense competition. Therefore, subdivision and continuous expansion of the ancestral nichebreadth probably operate dierently across spatial scales. Historical climatic fluctuations,as well as the ecologically and topographically complex landscape in western North Amer-ica, oered many opportunities for diversification (Stebbins and Major, 1965; Ackerly, 2009;Baldwin, 2014).Evolutionary lability of mean preference and niche breadth introduce lots of uncertainty inancestral state reconstruction. Therefore, it is dicult to detect any association betweenthe evolutionary trend of niche breadth and diversification process using data only fromthe present time. However, the opposite, though weak, trends across spatial scales mightsuggest dierent underlying mechanism of niche evolution and the contribution to diver-sification. Nonetheless, all five niche variables showed non-significant dierence betweentwo evolutionary trend with overlapping posterior distributions of transition rates, whichmight suggest there really are no dierences in evolutionary outcomes of diversification ingeneralists or specialists.Furthermore, divergence in reproductive traits could potentially be important in the diver-sification process (Kiester et al., 1984), which we did not consider here. Particularly, theshifts in plant mating system were suggested to be directional (from self-incompatible toself-compatible) with selfing species arising predominantly upon speciation (Goldberg andIgiÊ, 2012). In combination with the association between mating system and range size, suchthat selfers generally have larger ranges (Grossenbacher et al., 2015), mating system changesmight potentially be linked to the evolution of niche breadth.4.5.3 The designation of niche breadth statesA gray area still remains for defining the degree of generalization or specialization, often withsomewhat arbitrary thresholds needed along continuous niche axes. Furthermore, consider-ation is needed to choose relevant niche variables to explore (Hardy and Otto, 2014; Medinaand Langmore, 2016). A judgment of relative degree of specialization may be more reliableif specialists’ preference or tolerance ranges are included within that of generalists’. Herewe estimated realized niche breadth based on occurrence records, which may be narrowerthan the fundamental niche that is more nearly a manifestation of the genetic properties ofspecies. Moreover, within-locality variation (temporal/seasonal/interannual), which we didnot estimate here, might contribute more to niche breadth than among-locality variation53The evolution of niche breadth(Quintero and Wiens, 2012). Also, niche breadth along one niche dimension may not re-lated to status along another dimension. It would be interesting to estimate niche breadthfrom multidimensional niche space, like the volume of a convex hull encompassing all speciesoccurrences (Cornwell et al., 2006; Emery et al., 2012).The way to assign G/S for the two microhabitat variables was dierent from continuousbioclimatic variables due to the discrete nature of microhabitat characters (e.g. soil types).Another way we tried to characterize microhabitat specialization was to identify keywords inthe habitat descriptions from multiple floras (e.g., Jepson Manual, the Flora of North Amer-ica, Calflora, etc.) to see if a species was described as having a wide preference (consideredas G), or a restricted preference (considered as S). However, there were many uncertaintiesassociated with this approach. A source often provided only one description for a givenmicrohabitat variable for a species, which might not cover variation across its range. Thedescriptions also often skipped some microhabitat variables altogether for some species. And,sometimes the descriptions were inadequate for us to decide if one species should be assignedG or S. Therefore, we gave up this arbitrary approach and instead used a relatively moreobjective way by text mining method. However, one issue here was that information aboutsubstrate type was largely missing in the specimen records (low proportion of habitat recordswith Soil-related words, average 0.27 across species; in comparison, the average was 0.78 forHWA-related words). Sparse information about soil preferences could introduce uncertaintyin quantification of soil type preference. Another issue was that soil might have similarchemical properties even they were developed from dierent types of parent material, whichprobably would decrease the number of soil types. A more restricted way for the G/S clas-sification would be to assign species with only one soil type to be S, which would returnfewer specialists. The eect of fitting models with a more restricted definition of S or afterpruning tips with no information on substrate needs further investigation.Similarly, the designation of G/S for bioclimatic variables was based on a threshold thatwas the average of niche breadth estimated continuously. This threshold chosen here toconvert niche breadth into a binary trait was relatively conservative, and our ClaSSE modelfitting was able to distinguish the dierence in diversification rates and to return a weakG-to-S trend. If the threshold was set more restrictively, e.g., specialists were species withniche breadth values less than one standard deviation from the mean, it would result in fewerspecialists and more generalists at the tips of the phylogeny, which could alter the estimationsof transition rates. Moreover, information on variation in niche breadth and its eect ondiversification was lost when treat niche breadth as a binary trait. Further investigation ofthe relationship between niche breadth and diversification rate can be achieved with niche54The evolution of niche breadthbreadth as a continuous trait.Another source of uncertainty came from the data quality. Both bioclimatic and micro-habitat niche variables were quantified based on specimen collection. The possibility ofmisidentification of taxa cannot be eliminated, especially among morphologically similarspecies. Furthermore, standard criteria for recording habitat details at collection locationscan improve our ability for comparison across species.4.5.4 Model inadequacyRecent critiques of tests of binary trait correlations and the BiSSE model and its exten-sions (Maddison and FitzJohn, 2015; Rabosky and Goldberg, 2015, 2017) suggest high falsepositive rates. One reason for false positives is pseudoreplicates of correlations between di-versification rates and character states on phylogenies. Another cause could be violatingthe assumption of constant rates over time. By visualizing the Mimulus phylogeny andG/S states of niche variables tested here, there was no obvious clustering of generalists orspecialists on the phylogeny, which would not lead to pseudoreplication. Another critiqueconcerned the violation of the assumption of constant rates. This needs further direct esti-mates of temporal shifts in diversification rate to test whether there was one or more burstsof speciation on this pruned tree. Future investigations could apply recent extensions thatincorporate a hidden trait as the important driver of diversification or a nonparametric sta-tistical test (e.g. HiSSE: Murray and Conner 2009; Galipaud et al. 2014 and FiSSE Raboskyand Goldberg 2017), which might oer more comprehensive understanding for this study.Lastly, it should also be noted, that the sampling fractions of two major clades here wereactually dierent, with a lower ratio of species in Erythranthe were present on the phylogeny(~ 0.5), in comparison with Diplacus (~ 0.7). However, this work was conducted at a rela-tive low taxonomy level (genus), at which species are closely related and share lots of similarecophysiological characters and life history traits. These imply no dramatic trait shift asso-ciated with diversification on this phylogeny. We doubt that the eect of dierent samplingfractions could influence our major results, but future work would benefit from improvedevenness of sampling across the phylogeny.Studying the degree to which ecological niche breadths are conserved across spatial scales canincrease our understanding of eco-evolutionary patterns and underlying processes, from howbiodiversity is generated, to future response to climate change. Illuminating the relationshipsbetween the evolution of ecological niche breadth and diversification processes will facilitateour understanding of the role of ecology in speciation and diversification. Future studies are55The evolution of niche breadthneeded on niche evolution across spatial scales in other taxa groups.56The evolution of niche breadthTable 4.1: Five candidate ClaSSE models with their model names, the numbers of free pa-rameters, corresponding parameter constraintsmodel df parameter constraintsm.full 8 the full model with asymmetric speciation and anagenetictransition ratesm.sym 5 symmetric speciation and anagenetic transition ratesm.clado 6 no anagenetic change; asymmetric cladogenetic transitionratesm.BiSSE 6 no cladogenetic change; asymmetric anagenetic transitionrates (equivalent to a full BiSSE model)m.G2S 6 only G-to-S trend (no S-to-G) for both cladogenetic andanagenetic changesm.S2G 6 only S-to-G trend (no G-to-S) for both cladogenetic andanagenetic changes57The evolution of niche breadthTable 4.2: Binary trait correlation test results among five niche variables: Pseason (pre-cipitation seasonality), TPsyn (temperature-precipitation synchronicity), GDD0 (growingdegree days above 0 Celsius degree), HWA (habitat water anity) and Soil (soil type) inMimulus. Values were the numbers of trees showing significant correlation in combina-tions.Pseason TPsyn GDD0 HWA SoilPseasonTPsyn 100GDD0 100 0HWA 0 0 0Soil 16 0 47 8858The evolution of niche breadthTable 4.3: Top models supported across 100 trees for five niche variables: Pseason (precipitationseasonality), TPsyn (temperature-precipitation synchronicity), GDD0 (growing degree daysabove 0 Celsius degree), HWA (habitat water anity) and Soil (soil type). Model fittings andcomparisons were based on maximum-likelihood technique (values were the number of treeswith the corresponding model in the top set). Models with values in bold were top modelssupported over 50 trees.niche variable m.full m.sym m.clado m.BiSSE mG2S mS2GPseason 25 15 44 88 36 38TPsyn 34 3 89 46 97 8GDD0 77 11 83 51 87 5HWA 61 0 69 45 65 4Soil 87 1 65 43 0 6259Theevolutionofnichebreadth●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●Pseason TPsyn GDD012345Species rankNiche breadth(a)(b)●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●HWA Soi l246Species rankNiche breadthFigure 4.1: Rank plot of niche breadth estimates of 82 Mimulus species from narrow to broad, for five niche variables: (1) Pseason(precipitation seasonality), TPsyn (temperature-precipitation synchronicity), and GDD0 (growing degree days above 0 Celsius degree),with all data log-transformed; (b) HWA (habitat water anity) and Soil (soil type). Horizontal dashed lines corresponds to the meanvalues of niche breadth.60The evolution of niche breadthjungermannioideswashingtonensisbreviflorushymenophylluspatulusmoschatuspulsiferaefloribundusnorrisiidudleyilatidensexiguusmicranthusnudatustilingiiglaucescensguttatusyecorensisdentilobuswiensiidentatusgemmiparusalsinoidescardinalisverbenaceuslewisii sparishiilewisii nrupestriseastwoodiaenelsoniibicolorbrewerifilicaulisrubellusprimuloidesandrosaceusshevockiipurpureusgracilipespalmerisuksdorfiimontioidesinconspicuusaurantiacuslongifloruscalycinusrutilusaustralisariduspuniceusflemingiigrandiflorusclevelandiiangustatuspulchellustricolorpygmaeuscongdoniidouglasiikelloggiitorreyipictusparryirupicolapilosusbigeloviibolandericonstrictuswhitneyimephiticuslayneaeviscidusfremontiijohnstoniibrevipesrattaniimohavensiscusickiinanusjepsoniiclivicola0.2Niche BreadthGeneralistSpecialistFigure 4.2: Distribution of generalist and specialist lineages onMimulus phylogeny (82 tips) for fiveniche variables: Pseason (precipitation seasonality), TPsyn (temperature-precipitation synchronic-ity), GDD0 (growing degree days above 0 Celsius degree), HWA (habitat water anity) and Soil(soil type), from left to right, with red squares for generalists (G) and blue for specialists (S).61Theevolutionofnichebreadth-20 -10 0 10 20 300. densityP season(a) diversification-20 -10 0 10 20 300. densityTP syn-20 -10 0 10 20 300.000.040.08Probability densityGDD0-20 -10 0 10 20 30 400.000.040.08Probability densityHWA-20 -10 0 10 20 300.000.040.08Probability densitySoilGS0 5 10 15 20 25 300. densityP season(b) speciation0 5 10 15 20 25 300. densityTP syn0 5 10 15 20 25 300.000.100.20Probability densityGDD00 10 20 30 40 500. densityHWA0 10 20 30 400.000.100.20Probability densitySoilGS0 5 10 15 20 25 300. densityP season(c) extinction0 5 10 15 20 25 300.000.040.08Probability densityTP syn0 5 10 15 20 25 300. densityGDD00 10 20 30 40 500. densityHWA0 10 20 30 400. densitySoilGSRate of interestProbability densityFigure 4.3: Posterior distributions for diversification rates from ClaSSE models for five niche variables: Pseason (precipitation seasonality),TPsyn (temperature-precipitation synchronicity), GDD0 (growing degree days above 0 Celsius degree), HWA (habitat water anity) andSoil (soil type), (a) diversification rate, (b) speciation rate, and (c) extinction rate, with red for generalists (G) and blue for specialists(S).62Theevolutionofnichebreadth0 10 20 30 400. transition rateProbability densityPseason(a) Total0 10 20 30 400.000.040.08Total transition rateProbability densityTPsyn0 10 20 30 400. transition rateProbability densityGDD00 10 20 30 40 500. transition rateProbability densityHWA0 10 20 30 400. transition rateProbability densitySoilG−to−SS−to−G0 5 10 15 20 25 300. transition rateProbability densityPseason(b) Cladogenetic0 5 10 15 20 25 300. transition rateProbability densityTPsyn0 5 10 15 20 25 300. transition rateProbability densityGDD00 10 20 30 400. transition rateProbability densityHWA0 5 10 15 20 25 300. transition rateProbability densitySoilG−to−SS−to−G0 5 10 15 20 25 300. transition rateProbability densityPseason(c) Anagenetic0 5 10 15 20 25 300. transition rateProbability densityTPsyn0 5 10 15 20 25 300. transition rateProbability densityGDD00 5 10 15 20 25 300. transition rateProbability densityHWA0 5 10 15 20 25 300. transition rateProbability densitySoilG−to−SS−to−GTransition rateProbability densityFigure 4.4: Posterior distributions for parameters of evolutionary trend estimated from ClaSSE model fitting for five niche variables:Pseason (precipitation seasonality), TPsyn (temperature-precipitation synchronicity), GDD0 (growing degree days above 0 Celsius degree),HWA (habitat water anity) and Soil (soil type). Parameters of interest includes transition rates for both directions, with red for G-to-Sand blue for S-to-G: (a) total transition rates; (b) cladogenetic transition rates; (c) anagenetic transition rates.63Theevolutionofnichebreadth0 10 20 30 400. in modeProbability densityPseason(a) Total asymmetry in mode0 10 20 30 400. in modeProbability densityTPsyn0 10 20 30 400. in modeProbability densityGDD00 10 20 30 400.000.040.08Asymmetry in modeProbability densityHWA0 10 20 30 400. in modeProbability densitySoilcladogeneticanagenetic0 5 10 15 20 25 300. in modeProbability densityPseason(b) Asymmetry for G−to−S trend0 5 10 15 20 25 300. in modeProbability densityTPsyn0 5 10 15 20 25 300.000.040.08Asymmetry in modeProbability densityGDD00 5 10 15 20 25 300. in modeProbability densityHWA0 5 10 15 20 25 300. in modeProbability densitySoilcladogeneticanagenetic0 5 10 15 20 25 300. in modeProbability densityPseason(c) Asymmetry for S−to−G trend0 5 10 15 20 25 300. in modeProbability densityTPsyn0 5 10 15 20 25 300. in modeProbability densityGDD00 10 20 30 400. in modeProbability densityHWA0 5 10 15 20 25 300. in modeProbability densitySoilcladogeneticanageneticTransition rateProbability densityFigure 4.5: Posterior distributions for parameters of evolutionary asymmetry in mode estimated from ClaSSE model fitting for five nichevariables. : Pseason, TPsyn, GDD0, HWA and Soil. Parameters of interest includes cladogenetic changes (red) and anagenetic changes(blue): (a) total asymmetry in mode for both evolutionary directions; (b) asymmetry in mode for G-to-S direction; (c) asymmetry inmode for S-to-G direction.64General discussion and conclusionsChapter 5 General discussion and con-clusions5.1 Major findings and and implicationsThe overall goal of my dissertation was to study the interactions among niche evolution, rangeoverlap, and diversification processes across spatial scales. I used western North AmericanMimulus species as a study system, with consideration of bioclimatic variables in determiningbroad species distributions and microhabitat variables in determining local distributions.In the first study of my dissertation (chapter 2), I applied generalized linear models to fitto occurrences of 71 Mimulus species, and used regression coecients to evaluate the eectsof candidate bioclimatic variable in determining geographic distributions. The estimates ofmodel-averaged —* in a multimodel inference framework, as an index of the relative vari-able index, were ranked based on their absolute values for each species. The frequencies ofrankings across species allowed us to select the most important predictors. Specifically, myresults suggest that one bioclimatic variable, temperature-precipitation synchronicity, wasthe most importance predictor for geographic distributions in western North American mon-keyflowers, as we speculated prior to modeling. This approach showed great advantages inevaluating predictor importance, and hence reducing dimensions without losing interpretivepower in comparison with other ordination techniques (e.g., PCA).In the second study of my dissertation (chapter 3), I tested the eect of range overlapon niche divergence across spatial scales, accounting for divergence time (phylogenetic dis-tance). I hypothesized a negative association at a coarse scale, but a positive relationship ata fine scale. For 16 closely related Mimulus species, I quantified niche divergence for threemacrohabitat axes and three microhabitat axes. Consistent with environmental filtering op-erating in sympatry and divergent selection operating in allopatry, the results showed thatmacrohabitat niche divergence decreased with increasing range overlap at a coarse scale.Meanwhile, the negative relationships between niche divergence and phylogenetic distancefor two macrohabitat axes (except the edaphic axis), might suggest allopatric speciationand increasing niche similarity over time, presumably with increasing sympatry. This workcomplements my conclusions from the first study by showing the connection between coarse-scaled climatic variables and broad distributions of species. At the local scale, I found no65General discussion and conclusionssignificant relationship between microhabitat niche divergence and range overlap. This mightsuggest that multiple countervailing processes operate at the same time or that the eectof biotic interactions could not be detected across species’ ranges in this system. However,for the subset of species pairs with partially overlapping ranges, greater microhabitat nichedivergence was found in sympatry than in allopatry This was consistent with competitive in-teractions driving greater niche divergence in sympatry. More specifically, each pair showedthis pattern along one niche axis, but which axis diered among species pairs. Thus, to somedegree this might explain why there was no overall pattern between niche divergence andrange overlap for all 16 species together, because certain pairs diverged along dierent nicheaxes. A positive association was detected between phylogenetic distance and one microhab-itat axis related to vegetation cover, suggesting more divergence was accumulated throughtime. Thus, course- and local-scale niche axes show dissimilar patterns of evolutionarilylability. My results contribute to a growing number of studies demonstrating the interactionbetween niche divergence and geographic distributions among close relatives. It is importantto note that, when quantifying niche divergence, consideration of background environmentalavailability is an essential step, though here I could only do so for macrohabitat niche axes.Although dierent patterns of evolutionarily lability existed for niche axes across spatialscales, it remains an open question whether evolutionary changes between generalizationand specialization depend on the scale of niche variables. In the third study of my disser-tation (Chapter 4), I extended the phylogenetic scale to test the evolution of niche breadthwith 82 Mimulus species. I hypothesized an evolutionary trend from generalist to specialistfor niche variables at a coarse scale because of higher possibility of diversification with wide-ranging species and the tight association between niche breadth and range size; while forniche variables at a local scale, the pattern might be the same or the opposite. By convertingniche breadth into binary states, I assigned generalist or specialist states to every speciesalong each of five niche axes (three bioclimatic and two microhabitat variables). A specialistalong one axis does not necessarily have to be a specialist along other niche axes, and this wasobserved for all five niche axes involved in this study. However, species’ states along nicheaxes tended to be consistent within a given spatial scale, i.e., among bioclimatic variablesor between microhabitat variables, respectively. Diversification modeling results showeda weak generalist-to-specialist trend for three bioclimatic variables, but a weak specialist-to-generalist trend for two microhabitat variables. Meanwhile, generalist lineages showedsignificantly higher diversification rate for one bioclimatic variable (growing-season degreedays), while specialist lineages showed significantly higher diversification rate for one micro-habitat variable (soil type). Combined together, the inconsistent and often non-significant66General discussion and conclusionsevolutionary trends across spatial scales on the same phylogeny could suggest possibly nodierences in net evolutionary outcomes for generalists and specialists. Moreover, modelresults suggested no dominant mode for shifts of niche breadth, with both cladogenetic andanagenetic changes being of roughly equal importance. It also confirmed lability for all nicheaxes, and highlighted the diculty and uncertainty in estimation of ancestral states withcurrent data at hand. One major contribution of this study is the approach I took for quan-tifying specialization for continuous niche variables. As far as I know, this is the first studythat compiled herbarium specimen records to extract microhabitat preference across speciesranges via text analysis, and it has shown to be a promising approach when field surveys arehard to realize for a large number of species across a broad geographic extent.5.2 Limitations of the research and future directionsChapter 2 in this dissertation applied model-averaged standardized regression coecients(—*) to identify important variables for geographic distributions, though potential problemsexisted in theory and in practice. I excluded highly correlated predictive variables beforefitting models. However, multicollinearity issue was not fully solved and I did not considerinteraction terms here. Multicollinearity among predictors could introduce problems likeinflated variability in estimation of regression coecients and inaccurate estimations of rel-ative importance. Meanwhile, using standardized regression coecients was only applicableto first order linear regression model, which cannot deal with non-linearity and among-predictor interactions. Therefore it needs further investigation about the performance of —*in generalized linear models under above circumstances. Moreover, though the coarse-scalevariables showed their importance for species distributions, further studies including bioticinteractions (e.g., competition, pollinators) would be interesting to incorporate and evaluatetheir relative contributions to range-wide distributions.Chapter 3 and Chapter 4 here mainly focused on niche variables across spatial scales, whichneed to be extended into other taxonomic groups. In Chapter 3, I quantified niche diver-gence through dierent methods across spatial scales. When estimating niche divergence formacrohabitat variables, I considered the availability of background environments and cor-rected niche divergence between species based on availability. In other words, backgroundavailability correction could avoid overestimating niche divergence and make the compari-son independent of resolution and sampling eort. But the background environments formicrohabitat variables were not available here. This study might benefit from further fieldmeasurements of plot-level microhabitats. Furthermore, the degree of range overlap was just67General discussion and conclusionsa surrogate of the intensity of biotic interactions. To understand the mechanisms drivingspecies divergence, it is needed to clearly define the potential driver in the first place. Con-trolled experiments with appropriate study systems can oer insights on the cause-eectrelationship, beyond the detection of observable correlation patterns. For example, pair-wise or community-level competition experiments could be designed to test the intensity ofinteractions among populations from dierent locations and the eects on fitness (Pacalaand Roughgarden, 1982; Schluter, 1994; Godoy et al., 2014), to answer whether currentco-occurrence was an outcome of competition from the past.In Chapter 4, uncertainties associated with the methods applied here need further consid-erations. First, patterns might be sensitive to the threshold I used to convert continuousestimates of niche breadth into binary states of generalists and specialist. Dierent thresh-olds could alter the distribution of states across the phylogeny, and the eect on downstreamanalyses need further investigation. Moreover, the phylogenetic comparative methods ap-plied here have come under criticism recently because of the possibility of false positives.Though this study here might be not aected by the concern of pseudoreplication, furtheranalyses on temporal shifts of rates of interest are needed. Lastly, the dissimilar patternsacross spatial scales might indicate no dierence in evolutionary outcomes for generalizedand specialized niche breadth, especially when tested at a relatively small phylogenetic scale(genus level).Western North America is diverse with respect to complex conditions of climate and to-pography, as well as historical fluctuations. In particular, the California Floristic Provincehas distinctive floristic elements, with high biodiversity and a large proportion of endemicspecies (Stebbins and Major, 1965; Raven and Axelrod, 1978; Ackerly, 2009; Baldwin, 2014).Mimulus (Phrymaceae) and other systems, such as clades in Orobanchaceae, Compositae andLeguminosae, provide an excellent opportunity to explore how diversification is associatedwith abiotic and biotic factors. Further assessment of the role of ecology and biogeogra-phy in diversification will improve our understanding of how biodiversity is generated andmaintained across time and space. In the last decade, probabilistic modeling of range evo-lution has been employed to reconstruct ancestral areas, and to explore biogeographic andecological processes (McPeek, 2008; Donoghue and Moore, 2003; Ronquist and Sanmartin,2011). For example, Ree et al. (2005) and Ree and Smith (2008) applied the Dispersal-Extinction-Cladogenesis (DEC) model and found a prevalent pattern of rapid divergencefollowing range expansion in Hawaiian Psychotria. Sedio et al. (2013) found contemporaryprecipitation-related and microhabitat niches were associated with climatic conditions in ar-eas of ancestral origin, and this historical signature on the community structure could be68General discussion and conclusionsexplained by ecologically conservative dispersal, with a slow rate of niche evolution relativeto distribution changes. By reconstructing biogeographic histories for 19 plant clades, Xingand Ree (2017) found evidence supporting an up-lifting hypothesis for diversification; theyfound increasing in situ diversification rate in the Hengduan Mountains around 8Ma, con-gruent with independent estimates of the time period of orogeny. The direct inference ofrange dynamics, combined with niche dynamics, can oer a new perspective to explore therole of geology and ecology in evolutionary history. The accumulation of case studies will, inthe long run, enable meaningful meta-analyses of correlations among niche evolution, rangedynamics and diversification.69BibliographyBibliographyAckerly, D. D., 2003. Community assembly, niche conservatism, and adaptive evolution inchanging environments. International Journal of Plant Sciences 164:S165–S184.———, 2009. Evolution, origin and age of lineages in the Californian and Mediterraneanfloras. Journal of Biogeography 36:1221–1233.Ackerly, D. D. and P. B. Reich, 1999. Convergence and correlations among leaf size andfunction in seed plants: a comparative test using independent contrasts. American Journalof Botany 86:1272–1281.Ackerly, D. D., D. W. Schwilk, and C. O. Webb, 2006. Niche evolution and adaptive radiation:testing the order of trait divergence. Ecology 87:S50–61.Adler, P. B., J. HilleRisLambers, and J. M. Levine, 2007. A niche for neutrality. EcologyLetters 10:95–104.Ahmadzadeh, F., M. Flecks, M. A. Carretero, W. Böhme, F. Ihlow, P. Kapli, A. Miraldo,and D. Rödder, 2016. Separate histories in both sides of the Mediterranean: phylogenyand niche evolution of ocellated lizards. Journal of Biogeography 43:1242–1253.Anacker, B. L. and S. Y. Strauss, 2014. The geography and ecology of plant speciation:range overlap and niche divergence in sister species. Proceedings of the Royal Society ofLondon B: Biological Sciences 281:20132980.Anderson, B., P. Ros, T. J. Wiese, and A. G. Ellis, 2014. Intraspecific divergence andconvergence of floral tube length in specialized pollination interactions. Proceedings ofthe Royal Society of London B: Biological Sciences 281:20141420.Anderson, R. P., A. T. Peterson, and M. Gómez Laverde, 2002. Using niche-based GISmodeling to test geographic predictions of competitive exclusion and competitive releasein South American pocket mice. Oikos 98:3–16.Araújo, M. B. and M. New, 2007. Ensemble forecasting of species distributions. Trends inEcology and Evolution 22:42–47.Baldwin, B. G., 2014. Origins of plant diversity in the California Floristic Province. AnnualReview of Ecology, Evolution, and Systematics 45:347–369.Baldwin, B. G., D. H. Goldman, K. J. David, R. Patterson, T. J. Rosatti, and D. Wilken,2012. The Jepson manual: vascular plants of California. University of California Press.Barker, W. R., G. L. Nesom, P. M. Beardsley, and N. S. Fraga, 2012. A taxonomic conspectusof Phrymaceae: a narrowed circumscription for Mimulus, new and resurrected genera, andnew names and combinations. Phytoneuron 39:1–60.70BibliographyBarraclough, T. G. and A. P. Vogler, 2000. Detecting the geographical pattern of speciationfrom species-level phylogenies. American Naturalist 155:419–434.Bates, D., M. Maechler, B. Bolker, S. Walker, R. H. B. Christensen, H. Singmann, B. Dai,G. Grothendieck, and P. Green, 2015. lme4: Linear Mixed-Eects Models using ’Eigen’and S4. R package version 1.1-13 .Beans, C. M., 2014. The case for character displacement in plants. Ecology and Evolution4:852–865.Beardsley, P. M., S. E. Schoenig, J. B. Whittall, and R. G. Olmstead, 2004. Patterns ofevolution in western North AmericanMimulus (Phrymaceae). American Journal of Botany91:474–489.Bi, J., 2012. A review of statistical methods for determination of relative importance ofcorrelated predictors and identification of drivers of consumer liking. Journal of SensoryStudies 27:87–101.Breman, L., 2001. Random forest. Machine Learning 45:5–32.Bring, J., 1994. How to Standardize Regression-Coecients. American Statistician 48:209–213.Broennimann, O., M. C. Fitzpatrick, P. B. Pearman, B. Petitpierre, L. Pellissier, N. G.Yoccoz, W. Thuiller, M.-J. Fortin, C. Randin, N. E. Zimmermann, C. H. Graham, andA. Guisan, 2012. Measuring ecological niche overlap from occurrence and spatial environ-mental data. Global Ecology and Biogeography 21:481–497.Broennimann, O., U. A. Treier, H. Mueller-Schaerer, W. Thuiller, A. T. Peterson, andA. Guisan, 2007. Evidence of climatic niche shift during biological invasion. EcologyLetters 10:701–709.Brown, W. L. and E. O. Wilson, 1956. Character Displacement. Systematic Zoology 5:49.Budescu, D. V., 1993. Dominance analysis: A new approach to the problem of relativeimportance of predictors in multiple regression. Psychological bulletin 114:542–551.Burhnam, K. P. and D. R. Anderson, 2002. Model selection and multimodel inference: apractical Information-Theoretic approach. Second ed. Springer, New York.Burnham, K. P., D. R. Anderson, and K. P. Huyvaert, 2011. AIC model selection and multi-model inference in behavioral ecology: some background, observations, and comparisons.Behavioral Ecology and Sociobiology 65:23–35.Cade, B. S., 2015. Model averaging and muddled multimodel inferences. Ecology 96:2370–2382.Cavender-Bares, J., A. Gonzalez-Rodriguez, A. Pahlich, K. Koehler, and N. Deacon, 2011.Phylogeography and climatic niche evolution in live oaks (Quercus series Virentes) fromthe tropics to the temperate zone. Journal of Biogeography 38:962–981.71BibliographyCavender-Bares, J., A. Keen, and B. Miles, 2006. Phylogenetic structure of Floridian plantcommunities depends on taxonomic and spatial scale. Ecology 87:S109–22.Cavender-Bares, J., K. H. Kozak, P. V. A. Fine, and S. W. Kembel, 2009. The merging ofcommunity ecology and phylogenetic biology. Ecology Letters 12:693–715.Chase, J. M. and M. A. Leibold, 2003. Ecological Niches. Linking Classical and Contempo-rary Approaches. University of Chicago Press.Chesson, P., 2000. Mechanisms of Maintenance of Species Diversity. Annual Review ofEcology and Systematics 31:343–366.Chevan, A. and M. Sutherland, 1991. Hierarchical Partitioning. American Statistician45:90–96.Colles, A., L. H. Liow, and A. Prinzing, 2009. Are specialists at risk under environmentalchange? Neoecological, paleoecological and phylogenetic approaches. Ecology Letters12:849–863.Cornwell, W. K., D. W. Schwilk, and D. D. Ackerly, 2006. A trait-based test for habitatfiltering: convex hull volume. Ecology 87:1465–1471.Coyne, J. A. and H. A. Orr, 2004. Speciation. Sunderland, MA: Sinauer Associates.Crawford, D. J., 2010. Progenitor-derivative species pairs and plant speciation. Taxon59:1413–1423.Day, E. H., X. Hua, and L. Bromham, 2016. Is specialization an evolutionary dead end?Testing for dierences in speciation, extinction and trait transition rates across diversephylogenies of specialists and generalists. Journal of Evolutionary Biology 29:1257–1267.Donoghue, M. J. and B. R. Moore, 2003. Toward an integrative historical biogeography.Integrative and Comparative Biology 43:261–270.Dormann, C. F., B. Gruber, M. Winter, and D. Herrmann, 2010. Evolution of climate nichesin European mammals? Biology Letters 6:229–232.Drummond, A. J., M. A. Suchard, D. Xie, and A. Rambaut, 2012. Bayesian phylogeneticswith BEAUti and the BEAST 1.7. Molecular biology and evolution 29:1969–1973.Eiserhardt, W. L., J.-C. Svenning, W. J. Baker, Thomas L. P. Couvreur, and H. Balslev,2013. Dispersal and niche evolution jointly shape the geographic turnover of phylogeneticclades across continents. Scientific reports 3:1164.Elith, J., J. R. Leathwick, and T. Hastie, 2008. A working guide to boosted regression trees.Journal of Animal Ecology 77:802–813.Elton, C. S., 1927. Animal ecology. University of Chicago Press.72BibliographyEmery, N. C., E. J. Forrestel, G. Jui, M. S. Park, B. G. Baldwin, and D. D. Ackerly,2012. Niche evolution across spatial scales: climate and habitat specialization in CaliforniaLasthenia (Asteraceae). Ecology 93:S151–S166.Evans, M. E. K., S. A. Smith, R. S. Flynn, and M. J. Donoghue, 2009. Climate, nicheevolution, and diversification of the "bird-cage" evening primroses (Oenothera, sectionsAnogra and Kleinia). American Naturalist 173:225–240.Feinerer, I., K. Hornik, and Artifex Software, Inc., 2017. tm: Text Mining Package. Rpackage version 0.7-1 .FitzJohn, R. G., 2010. Quantitative traits and diversification. Systematic Biology 59:619–633.FitzJohn, R. G., W. P. Maddison, and S. P. Otto, 2009. Estimating trait-dependent spe-ciation and extinction rates from Incompletely resolved phylogenies. Systematic Biology58:595–611.Fitzpatrick, B. M. B. and M. M. Turelli, 2006. The geography of mammalian speciation:mixed signals from phylogenies and range maps. Evolution 60:601–615.Fitzpatrick, M. C., J. F. Weltzin, N. J. Sanders, and R. R. Dunn, 2007. The biogeographyof prediction error: why does the introduced range of the fire ant over-predict its nativerange? Global Ecology and Biogeography 16:24–33.Fraga, N., 2012. A Revision of Erythranthe montioides and Erythranthe palmeri (Phry-maceae), with descriptions of five new species from California and Nevada, USA. Aliso:A Journal of Systematic and Evolutionary Botany Pp. 49–68.Freckleton, R. P. and P. H. Harvey, 2006. Detecting non-Brownian trait evolution in adaptiveradiations. PLoS Biology 4:e373.Futuyma, D. J. and G. Moreno, 1988. The evolution of ecological specialization. AnnualReview of Ecology and Systematics 19:207–233.Galipaud, M., M. A. F. Gillingham, M. David, and F.-X. Dechaume-Moncharmont, 2014.Ecologists overestimate the importance of predictor variables in model averaging: a pleafor cautious interpretations. Methods In Ecology And Evolution 5:983–991.Gaston, K. J., 2003. The structure and dynamics of geographic ranges. Oxford UniversityPress.Giam, X. and J. D. Olden, 2015. Quantifying variable importance in a multimodel inferenceframework. Methods In Ecology And Evolution 7:388–397.Godoy, O., N. J. B. Kraft, and J. M. Levine, 2014. Phylogenetic relatedness and the deter-minants of competitive outcomes. Ecology Letters 17:836–844.Goldberg, E. E. and B. IgiÊ, 2012. Tempo and mode in plant breeding system evolution.Evolution 66:3701–3709.73BibliographyGottlieb, L. D., 2004. Rethinking classic examples of recent speciation in plants. Newphytologist 161:71–82.Grinnell, J., 1917. Field tests of theories concerning distributional control. American Natu-ralist 51:115–128.Gröemping, U., 2006. Relative importance for linear regression in R: The package relaimpo.Journal of Statistical Software 17:1–27.Grossenbacher, D., R. Briscoe Runquist, E. E. Goldberg, and Y. Brandvain, 2015. Geo-graphic range size is predicted by plant mating system. Ecology Letters 18:706–713.Grossenbacher, D. L., S. D. Veloz, and J. P. Sexton, 2014. Niche and range size patterns sug-gest that speciation begins in small, ecologically diverged populations in North Americanmonkeyflowers (Mimulus spp.). Evolution 68:1270–1280.Grueber, C. E., S. Nakagawa, R. J. Laws, and I. G. Jamieson, 2011. Multimodel inference inecology and evolution: challenges and solutions. Journal of Evolutionary Biology 24:699–711.Guisan, A. and J. P. Theurillat, 2000. Equilibrium modeling of alpine plant distribution:how far can we go? Phytocoenologia 30:353–384.Guisan, A. and W. Thuiller, 2005. Predicting species distribution: oering more than simplehabitat models. Ecology Letters 8:993–1009.Hanski, I., 1998. Metapopulation dynamics. Nature 396:41–49.Hardy, N. B. and S. P. Otto, 2014. Specialization and generalization in the diversification ofphytophagous insects: tests of the musical chairs and oscillation hypotheses. Proceedingsof the Royal Society of London B: Biological Sciences 281:20132960.Harrison, R. G., 2012. The language of speciation. Evolution 66:3643–3657.Harvey, P. H. and M. D. Pagel, 1991. The comparative method in evolutionary biology.Oxford university press, Oxford.Hastie, T., R. Tibshirani, and J. Friedman, 2001. The Elements of Statistical Learning:Data Mining, Inference, and Prediction. Springer, New York.Hijmans, R. J., S. E. Cameron, and J. L. Parra, 2005. Very high resolution interpolatedclimate surfaces for global land areas. International Journal of Climatology 25:1965–1978.HilleRisLambers, J., P. B. Adler, W. S. Harpole, J. M. Levine, and M. M. Mayfield, 2012.Rethinking Community Assembly through the Lens of Coexistence Theory. Annual Reviewof Ecology, Evolution, and Systematics 43:227–248.Holt, R. D., 1996. Demographic constraints in evolution: Towards unifying the evolutionarytheories of senescence and niche conservatism. Evolutionary Ecology 10:1–11.74Bibliography———, 2009. Bringing the Hutchinsonian niche into the 21st century: ecological and evo-lutionary perspectives. Proceedings of the National Academy of Sciences of the UnitedStates of America 106 (Suppl 2):19659–19665.Holt, R. D. and M. S. Gaines, 1992. Analysis of Adaptation in Heterogeneous Landscapes -Implications for the Evolution of Fundamental Niches. Evolutionary Ecology 6:433–447.Holt, R. D. and T. H. Keitt, 2005. Species’ borders: a unifying theme in ecology. Oikos108:3–6.Hua, X. and J. J. Wiens, 2013. How does climate influence speciation? American Naturalist182:1–12.Huey, R. B. and P. E. Hertz, 1984. Is a Jack-of-all-temperatures a master of none? Evolution38:441.Hutchinson, G. E., 1957. Concluding remarks. Cold Spring Harbor Symposia on QuantitativeBiology 22:415–427.Ivey, C. T. and D. E. Carr, 2012. Tests for the joint evolution of mating system and droughtescape in Mimulus. Annals of Botany 109:583–598.Jackson, S. T. and J. T. Overpeck, 2000. Responses of plant populations and communitiesto environmental changes of the late Quaternary. Paleobiology 26:194–220.Janz, N. and S. Nylin, 2008. The Oscillation Hypothesis of Host-Plant Range and Speciation.Pp. 203–215, in Specialization, Speciation, and RadiationThe Evolutionary Biology ofHerbivorous Insects. University of California Press.Johnson, J. W., 2000. A heuristic method for estimating the relative weight of predictorvariables in multiple regression. Multivariate Behavioral Research 35:1–19.———, 2004. Factors Aecting Relative Weights: The Influence of Sampling and Measure-ment Error. Organizational Research Methods 7:283–299.Johnson, K. P., J. R. Malenke, and D. H. Clayton, 2009. Competition promotes the evolutionof host generalists in obligate parasites. Proceedings of the Royal Society of London B:Biological Sciences 276:3921–3926.Kellermann, V., B. van Heerwaarden, C. M. Sgrò, and A. A. Homann, 2009. Fundamen-tal evolutionary limits in ecological traits drive drosophila species distributions. Science325:1244–1246.Kiang, Y. T. and J. L. Hamrick, 1978. Reproductive isolation in the Mimulus guttatus M.nasutus Complex. American Midland Naturalist 100:269–276.Kiester, A. R., R. Lande, and D. W. Schemske, 1984. Models of coevolution and speciationin plants and their pollinators. American Naturalist 124:220–243.75BibliographyKirkpatrick, M. and N. H. Barton, 1997. Evolution of a species’ range. The AmericanNaturalist 150:1–23.Knouft, J. H., J. B. Losos, R. E. Glor, and J. J. Kolbe, 2006. Phylogenetic analysis of theevolution of the niche in lizards of the Anolis sagrei group. Ecology 87:S29–S38.Kou, X., Q. Li, and S. Liu, 2011. High-resolution bioclimatic dataset derived from futureclimate projections for plant species distribution modeling. Ecological Informatics 6:196–204.Kozak, K. H. and J. J. Wiens, 2010a. Accelerated rates of climatic-niche evolution underlierapid species diversification. Ecology Letters 13:1378–1389.———, 2010b. Niche conservatism drives elevational diversity patterns in Appalachian sala-manders. American Naturalist 176:40–54.Kraft, N. J. B., P. B. Adler, O. Godoy, E. C. James, S. Fuller, and J. M. Levine, 2015.Community assembly, coexistence and the environmental filtering metaphor. FunctionalEcology 29:592–599.Kraft, N. J. B., R. Valencia, and D. D. Ackerly, 2008. Functional traits and niche-based treecommunity assembly in an Amazonian forest. Science 322:580–582.Lancaster, L. T. and K. M. Kay, 2013. Origin and diversification of the California flora:re-examining classic hypotheses with molecular phylogenies. Evolution 67:1041–1054.Lee-Yaw, J. A., H. M. Kharouba, M. Bontrager, C. Mahony, A. M. Cserg, A. M. E. Noreen,Q. Li, R. Schuster, and A. L. Angert, 2016. A synthesis of transplant experiments andecological niche models suggests that range limits are often niche limits. Ecology Letters19:710–722.Leibold, M. A., 1995. The niche concept revisited: mechanistic models and communitycontext. Ecology 76:1371–1382.Lessios, H. A., 2008. The great American schism: divergence of marine organisms afterthe rise of the Central American Isthmus. Annual Review of Ecology, Evolution, andSystematics 39:63–91.Levins, R., 1968. Evolution in changing environments. Princeton University Press, Princeton.Litsios, G., A. Kostikova, and N. Salamin, 2014. Host specialist clownfishes are environmentalniche generalists. Proceedings of the Royal Society of London B: Biological Sciences281:20133220.Losos, J. B., 2009. Lizards in an evolutionary tree: ecology and adaptive radiation of Anoles.University of California Press.Losos, J. B. and R. E. Glor, 2003. Phylogenetic comparative methods and the geography ofspeciation. Trends in Ecology and Evolution 18:220–227.76BibliographyLosos, J. B., M. Leal, R. E. Glor, K. De Queiroz, P. E. Hertz, L. R. Schettino, A. C. Lara,T. R. Jackman, and A. Larson, 2003. Niche lability in the evolution of a Caribbean lizardcommunity. Nature 424:542–545.Lovette, I. J. and W. M. Hochachka, 2006. Simultaneous eects of phylogenetic niche con-servatism and competition on avian community structure. Ecology 87:S14–S28.Mac Nally, R., 2002. Multiple regression and inference in ecology and conservation biol-ogy: further comments on identifying important predictor variables. Biodiversity andConservation 11:1397–1401.MacArthur, R. H., 1972. Geographical ecology. Princeton University Press.Maddison, W. P., 2006. Confounding asymmetries in evolutionary diversification and char-acter change. Evolution 60:1743–1746.Maddison, W. P. and R. G. FitzJohn, 2015. The unsolved challenge to phylogenetic corre-lation tests for categorical characters. Systematic Biology 64:127–136.Maddison, W. P., P. E. Midford, and S. P. Otto, 2007. Estimating a binary character’s eecton speciation and extinction. Systematic Biology 56:701–710.Magnuson-Ford, K. and S. P. Otto, 2012. Linking the investigations of character evolutionand species diversification. American Naturalist 180:225–245.McCormack, J. E., A. J. Zellmer, and L. L. Knowles, 2010. Does niche divergence accompanyallopatric divergence in Aphelocoma jays as predicted under ecological speciation? Insightsfrom tests with niche models. Evolution 64:1231–1244.McPeek, M. A., 2008. The ecological dynamics of clade diversification and communityassembly. American Naturalist 172:E270–84.Medina, I. and N. E. Langmore, 2016. The evolution of host specialisation in avian broodparasites. Ecology Letters 19:1110–1118.Mittelbach, G. G., D. W. Schemske, H. V. Cornell, A. P. Allen, J. M. Brown, M. B. Bush,S. P. Harrison, A. H. Hurlbert, N. Knowlton, H. A. Lessios, C. M. McCain, A. R. McCune,L. A. McDade, M. A. McPeek, T. J. Near, T. D. Price, R. E. Ricklefs, K. Roy, D. F. Sax,D. Schluter, J. M. Sobel, and M. Turelli, 2007. Evolution and the latitudinal diversitygradient: speciation, extinction and biogeography. Ecology Letters 10:315–331.Mouillot, D. and K. Gaston, 2009. Spatial overlap enhances geographic range size conser-vatism. Ecography 32:671–675.Murray, K. and M. M. Conner, 2009. Methods to quantify variable importance: implicationsfor the analysis of noisy ecological data. Ecology 90:348–355.Murtaugh, P. A., 2009. Performance of several variable-selection methods applied to realecological data. Ecology Letters 12:1061–1068.77BibliographyMyer, E., 1963. Animal species and evolution. Harvard University Press.Nakazato, T., D. L. Warren, and L. C. Moyle, 2010. Ecological and geographic modes ofspecies divergence in wild tomatoes. American Journal of Botany Pp. 680–693.van der Niet, T. and S. D. Johnson, 2009. Patterns of plant speciation in the Cape floristicregion. Molecular Phylogenetics and Evolution 51:85–93.Nosil, P., 2012. Ecological speciation. Oxford University Press.Orr, M. R. and T. B. Smith, 1998. Ecology and speciation. Trends in Ecology and Evolution13:502–506.Pacala, S. and J. Roughgarden, 1982. Resource partitioning and interspecific competition intwo two-species insular anolis lizard communities. Science 217:444–446.Pagel, M., 1994. Detecting correlated evolution on phylogenies: a general method for thecomparative analysis of discrete characters. Proceedings of the Royal Society of LondonB: Biological Sciences 255:37–45.Pearman, P. B., A. Guisan, O. Broennimann, and C. F. Randin, 2008. Niche dynamics inspace and time. Trends in Ecology and Evolution 23:149–158.Pearson, R. G. and T. P. Dawson, 2003. Predicting the impacts of climate change onthe distribution of species: are bioclimate envelope models useful? Global Ecology andBiogeography 12:361–371.Peterson, A., S. n J, and V. Sánchez-Cordero, 1999. Conservatism of ecological niches inevolutionary time. Science 285:1265–1267.Peterson, A. T., 2011. Ecological niche conservatism: a time-structured review of evidence.Journal of Biogeography 38:817–827.Peterson, A. T. and R. D. Holt, 2003. Niche dierentiation in Mexican birds: using pointoccurrences to detect ecological innovation. Ecology Letters 6:774–782.Peterson, A. T., J. Soberón, R. G. Pearson, R. P. Anderson, E. Martínez-Meyer, M. Naka-mura, and M. B. Araújo, 2011. Ecological niches and geographic distributions (MPB-49).Princeton University Press.Peterson, M. L., K. J. Rice, and J. P. Sexton, 2013. Niche partitioning between close relativessuggests trade-os between adaptation to local environments and competition. Ecologyand Evolution 3:512–522.Phillimore, A. B., C. D. L. Orme, G. H. Thomas, T. M. Blackburn, P. M. Bennett, K. J.Gaston, and I. P. F. Owens, 2008. Sympatric speciation in birds is rare: insights fromrange data and simulations. American Naturalist 171:646–657.Phillips, S., R. Anderson, and R. Schapire, 2006. Maximum entropy modeling of speciesgeographic distributions. Ecological Modelling 190:231–259.78BibliographyPigot, A. L. and J. A. Tobias, 2013. Species interactions constrain geographic range expan-sion over evolutionary time. Ecology Letters 16:330–338.Prinzing, A., W. Durka, S. Klotz, and R. Brandl, 2001. The niche of higher plants: evidencefor phylogenetic conservatism. Proceedings of the Royal Society of London B: BiologicalSciences 268:2383–2389.Pulliam, H. R., 1988. Sources, sinks, and population regulation. The American Naturalist132:652–661.———, 2000. On the relationship between niche and distribution. Ecology Letters 3:349–361.Quintero, I. and J. J. Wiens, 2012. What determines the climatic niche width of species?The role of spatial and temporal climatic variation in three vertebrate clades. GlobalEcology and Biogeography 22:422–432.Rabosky, D. L. and E. E. Goldberg, 2015. Model inadequacy and mistaken inferences oftrait-dependent speciation. Systematic Biology 64:340–355.———, 2017. FiSSE: A simple nonparametric test for the eects of a binary character onlineage diversification rates. Evolution 106:13410–1442.Rambaut, A., M. A. Suchard, D. Xie, and A. J. Drummond, 2014. Tracer v1.6, Availablefrom Retrieved on March 10, 2015 .Raven, P. H. and D. I. Axelrod, 1978. Origin and relationships of the California flora.University of California Press.Ree, R. H., B. R. Moore, C. O. Webb, and M. J. Donoghue, 2005. A likelihood framework forinferring the evolution of geographic range on phylogenetic trees. Evolution 59:2299–2311.Ree, R. H. and S. A. Smith, 2008. Maximum likelihood inference of geographic range evolu-tion by dispersal, local extinction, and cladogenesis. Systematic Biology 57:4–14.Revell, L. J., 2012. phytools: an R package for phylogenetic comparative biology (and otherthings). Methods In Ecology And Evolution 3:217–223.Rice, K. J. and N. C. Emery, 2003. Managing microevolution: restoration in the face ofglobal change. Frontiers In Ecology And The Environment 1:469–478.Ricklefs, R. E. and D. G. Jenkins, 2011. Biogeography and ecology: towards the integrationof two disciplines. Philosophical Transactions of the Royal Society of London B BiologicalSciences 366:2438–2448.Ricklefs, R. E. and R. E. Latham, 1992. Intercontinental correlation of geographical rangessuggests stasis in ecological traits of relict genera of temperate perennial herbs. AmericanNaturalist 139:1305–1321.Ronquist, F. and I. Sanmartin, 2011. Phylogenetic methods in biogeography. Annual Reviewof Ecology, Evolution, and Systematics 42:441.79BibliographyRundle, H. D. and P. Nosil, 2005. Ecological speciation. Ecology Letters 8:336–352.Schemske, D. W., 2010. Adaptation and the origin of species. American Naturalist 176:S4–S25.Schemske, D. W., G. G. Mittelbach, H. V. Cornell, J. M. Sobel, and K. Roy, 2009. Is there alatitudinal gradient in the importance of biotic interactions? Annual Review of Ecology,Evolution, and Systematics Pp. 245–269.Schluter, D., 1994. Experimental evidence that competition promotes divergence in adaptiveradiation. Science 266:798–801.———, 1996. Ecological causes of adaptive radiation. American Naturalist 148:S40–S64.———, 2000a. Ecological character displacement in adaptive Radiation. American Naturalist156:S4–S16.———, 2000b. The ecology of adaptive radiation. Oxford University Press.———, 2001. Ecology and the origin of species. Trends in Ecology and Evolution 16:372–380.———, 2009. Evidence for ecological speciation and its alternative. Science 323:737–741.Schoener, T. W., 1970. Nonsynchronous spatial overlap of lizards in patchy habitats. Ecology51:408–418.Sedio, B. E., J. R. Paul, C. M. Taylor, and C. W. Dick, 2013. Fine-scale niche structureof Neotropical forests reflects a legacy of the Great American Biotic Interchange. Naturecommunications 4:2317.Sexton, J. P., P. J. McIntyre, A. L. Angert, and K. J. Rice, 2009. Evolution and ecology ofspecies range limits. Annual Review of Ecology, Evolution, and Systematics 40:415–436.Sheth, S. N., I. Jiménez, and A. L. Angert, 2014. Identifying the paths leading to vari-ation in geographical range size in western North American monkeyflowers. Journal ofBiogeography 41:2344–2356.Silvertown, J., 2004. Plant coexistence and the niche. Trends in Ecology and Evolution19:605–611.Silvertown, J., K. McConway, D. Gowing, M. Dodd, M. F. Fay, J. A. Joseph, and K. Dolphin,2006. Absence of phylogenetic signal in the niche structure of meadow plant communities.Proceedings of the Royal Society of London B: Biological Sciences 273:39–44.Simpson, G. G., 1953. The major features of evolution. Columbia University Press, NewYork.Slatyer, R. A., M. Hirst, and J. P. Sexton, 2013. Niche breadth predicts geographical rangesize: a general ecological pattern. Ecology Letters 16:1104–1114.80BibliographySobel, J. M., 2014. Ecogeographic isolation and speciation in the genus Mimulus. AmericanNaturalist 184:565–579.Soberón, J., 2007. Grinnellian and Eltonian niches and geographic distributions of species.Ecology Letters 10:1115–1123.Soberón, J. and A. T. Peterson, 2005. Interpretation of models of fundamental ecologicalniches and species’ distributional areas. Biodiversity Informatics 2:1–10.Stebbins, G. L. and J. Major, 1965. Endemism and speciation in the California flora. Eco-logical Monographs 35:1–35.Steinbauer, M. J., R. Field, J. M. Fernández Palacios, S. D. H. Irl, R. Otto, H. Schaefer, andC. Beierkuhnlein, 2016. Biogeographic ranges do not support niche theory in radiatingCanary Island plant clades. Global Ecology and Biogeography 25:792–804.Stevens, G. C., 1989. The latitudinal gradient in geographical range: how so many speciescoexist in the tropics. American Naturalist 133:240–256.Svenning, J. C. and F. Skov, 2004. Limited filling of the potential range in European treespecies. Ecology Letters 7:565–573.Swenson, N. G., B. J. Enquist, J. Thompson, and J. K. Zimmerman, 2007. The influence ofspatial and size scale on phylogenetic relatedness in tropical forest communities. Ecology88:1770–1780.Symonds, M. R. E. and A. Moussalli, 2010. A brief guide to model selection, multimodelinference and model averaging in behavioural ecology using Akaike’s information criterion.Behavioral Ecology and Sociobiology 65:13–21.Thomas, C. D., E. J. Bodsworth, R. J. Wilson, A. D. Simmons, Z. G. Davies, M. Musche,and L. Conradt, 2001. Ecological and evolutionary processes at expanding range margins.Nature 411:577–581.Tonidandel, S. and J. M. LeBreton, 2010. Determining the relative importance of predictorsin logistic regression: an extension of relative weight analysis. Organizational ResearchMethods 13:767–781.Vamosi, J. C., W. S. Armbruster, and S. S. Renner, 2014. Evolutionary ecology of special-ization: insights from phylogenetic analysis. Proceedings of the Royal Society of LondonB: Biological Sciences 281:20142004.Warren, D. L., M. Cardillo, D. F. Rosauer, and D. I. Bolnick, 2014. Mistaking geography forbiology: inferring processes from species distributions. Trends in Ecology and Evolution29:572–580.Warren, D. L., R. E. Glor, and M. Turelli, 2008. Environmental niche equivalency versusconservatism: quantitative approaches to niche evolution. Evolution 62:2868–2883.81BibliographyWebb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue, 2002. Phylogenies andCommunity Ecology. Annual Review of Ecology and Systematics 33:475–505.Whitlock, M. C., 1996. The Red Queen Beats the Jack-Of-All-Trades: The Limitations onthe Evolution of Phenotypic Plasticity and Niche Breadth. American Naturalist 148:S65–S77.Whittaker, R. J., K. J. Willis, and R. Field, 2001. Scale and species richness: towards ageneral, hierarchical theory of species diversity. Journal of Biogeography 28:453–470.Whittaker, T. A., R. T. Fouladi, and N. J. Williams, 2002. Determining predictor importancein multiple regression under varied correlational and distributional conditions. Journal ofModern Applied Statistical Methods 1:354–366.Whittall, J. B., M. L. Carlson, P. M. Beardsley, R. J. Meinke, and A. Liston, 2006. TheMimulus moschatus alliance (Phrymaceae): molecular and morphological phylogeneticsand their conservation implications. Systematic Botany 31:380–397.Whittingham, M. J., P. A. Stephens, R. B. Bradbury, and R. P. Freckleton, 2006. Whydo we still use stepwise modelling in ecology and behaviour? Journal of Animal Ecology75:1182–1189.Wiens, J. J., 2004. Speciation and ecology revisited: phylogenetic niche conservatism andthe origin of species. Evolution 58:193–197.———, 2011. The niche, biogeography and species interactions. Philosophical Transactionsof the Royal Society of London B Biological Sciences 366:2336–2350.Wiens, J. J., D. D. Ackerly, A. P. Allen, B. L. Anacker, L. B. Buckley, H. V. Cornell, E. I.Damschen, T. Jonathan Davies, J.-A. Grytnes, S. P. Harrison, B. A. Hawkins, R. D. Holt,C. M. McCain, and P. R. Stephens, 2010. Niche conservatism as an emerging principle inecology and conservation biology. Ecology Letters 13:1310–1324.Wiens, J. J. and M. J. Donoghue, 2004. Historical biogeography, ecology and species richness.Trends in Ecology and Evolution 19:639–644.Wiens, J. J. and C. H. Graham, 2005. Niche conservatism: integrating evolution, ecology, andconservation biology. Annual Review of Ecology, Evolution, and Systematics 36:519–539.Wu, C. A., D. B. Lowry, A. M. Cooley, K. M. Wright, Y. W. Lee, and J. H. Willis, 2008.Mimulus is an emerging model system for the integration of ecological and genomic studies.Heredity 100:220–230.Xing, Y. and R. H. Ree, 2017. Uplift-driven diversification in the Hengduan Mountains, atemperate biodiversity hotspot. Proceedings of the National Academy of Sciences of theUnited States of America 114:E3444–E3451.Zsolt Garamszegi, L., 2011. Information-theoretic approaches to statistical analysis in be-havioural ecology: an introduction. Behavioral Ecology and Sociobiology 65:1–11.82AppendicesAppendicesAppendix A Supplementary materialfor Chapter 2Table A.1: Correlations among five candidate bioclimatic variables across the commonMimu-lus environmental background. Tcold: temperature of coldest month; GDD0 : growing de-gree days above 0 Celsius degree; Pseason: precipitation seasonality; TPsyn: temperature-precipitation synchronicity; and Aridity: aridity of growing season.Tcold GDD0 Pseason TPsyn AridityTcold 1.00 0.92 0.74 -0.19 0.18GDD0 0.92 1.00 0.54 0.00 0.38Pseason 0.74 0.54 1.00 -0.52 -0.27TPsyn -0.19 0.00 -0.52 1.00 0.33Aridity 0.18 0.38 -0.27 0.33 1.0083AppendicesTable A.2: Estimations of weighted —* based on generalized linear models for fourbioclimatic variables for 71 Mimulus species. Tcold: temperature of coldest month,Pseason: precipitation seasonality, TPsyn: temperature-precipitation synchronicity,and Aridity: aridity of growing seasonspecies Tcold Pseason TPsyn AridityM. alsinoides 0.93 -0.82 -11.40 -0.91M. androsaceus -0.76 1.95 -2.28 0.07M. angustatus -0.16 10.80 -25.31 -2.16M. aridus 2.55 -0.73 -2.75 0.28M. aurantiacus 7.76 8.05 -18.99 -2.20M. australis 10.00 -3.29 -13.76 -0.48M. bicolor -5.34 9.44 -46.48 -1.87M. bigelovii 2.24 -1.08 -1.85 1.02M. bolanderi -4.28 8.89 -11.71 -0.64M. breviflorus -1.38 0.28 -3.76 0.18M. brevipes 4.93 -0.78 -6.51 -0.25M. breweri -4.43 2.93 -4.11 0.06M. calycinus 1.53 0.50 -9.10 0.62M. cardinalis 2.39 0.25 -4.28 -0.19M. clevelandii 3.29 -0.22 -4.48 -0.39M. congdonii 0.34 4.06 -5.61 -0.44M. constrictus -4.44 5.92 -2.47 0.34M. cusickii -0.30 -0.80 -2.79 0.79M. dentatus 6.27 -0.05 -13.62 -3.63M. douglasii 1.28 2.59 -9.04 -0.65M. dudleyi 2.28 2.02 -21.11 0.65M. eastwoodiae 0.02 -2.29 1.51 2.52M. exiguus -1.02 1.20 -1.57 -0.20M. filicaulis -6.95 9.06 -43.50 -1.03M. flemingii 7.96 33.35 -7.47 -0.12M. floribundus -0.13 1.00 -1.46 0.08M. fremontii 1.97 0.43 -4.03 0.08M. glaucescens 3.30 2.11 -72.84 -1.62M. gracilipes -5.57 13.87 -42.07 -1.23M. grandiflorus 1.10 9.23 -87.71 -3.55M. guttatus 0.06 0.16 -1.23 -0.1284AppendicesTable A.2: Continued from previous page.species Tcold Pseason TPsyn AridityM. inconspicuus -4.77 7.49 -16.37 -0.21M. jepsonii -6.57 4.11 -21.20 -0.11M. johnstonii -5.07 7.25 -1.74 -0.10M. kelloggii -0.47 5.08 -22.27 -1.46M. latidens 1.74 1.00 -15.52 0.71M. layneae -4.82 5.89 -20.05 -0.60M. lewisii_n -2.67 0.38 -3.00 -0.56M. lewisii_s -9.92 7.51 -11.52 0.89M. longiflorus 4.25 0.26 -5.31 -0.26M. mephiticus -2.81 1.74 -3.20 0.68M. micranthus -0.05 0.77 -2.76 -0.14M. mohavensis 3.02 -1.85 -6.03 1.56M. montioides -16.40 17.08 -8.01 0.76M. moschatus -0.72 -0.05 -2.72 -0.45M. nanus -0.99 0.04 -3.33 0.43M. norrisii 2.19 4.14 -21.32 -0.05M. nudatus -3.38 13.95 -4.39 -1.18M. palmeri 0.33 1.57 -2.93 0.16M. parishii 1.90 -0.51 -3.59 0.24M. parryi 3.84 -3.51 -2.95 0.35M. patulus 0.10 -1.61 -1.15 0.09M. pictus 1.52 1.46 -13.72 1.20M. pilosus 1.12 0.48 -2.65 0.07M. primuloides -4.52 2.77 -4.46 0.34M. pulchellus -3.63 6.03 -45.08 -1.39M. pulsiferae -3.03 2.11 -23.58 -0.31M. puniceus 12.93 -3.79 -21.76 -0.43M. purpureus -2.06 1.99 -2.00 -0.30M. pygmaeus -5.40 2.31 -48.49 0.65M. rattanii -0.41 12.29 -2.34 -1.21M. rubellus 0.46 -0.13 -0.92 0.21M. rupicola 5.02 -4.94 -4.89 1.51M. shevockii 1.32 0.09 -10.48 1.22M. suksdorfii -1.75 0.64 -2.17 0.8985AppendicesTable A.2: Continued from previous page.species Tcold Pseason TPsyn AridityM. tilingii -3.67 1.99 -3.27 0.21M. torreyi -8.78 8.54 -50.96 -0.92M. tricolor 1.07 0.99 -33.64 0.20M. viscidus -1.74 5.81 -50.46 -1.13M. whitneyi -7.05 6.58 -4.86 0.34M. yecorensis 0.80 1.62 2.03 -1.7886Appendices●●●●●●●●●●●●●● ●●●●●●●●●● ●● ●●●●●●● ●●●●●●●●●●●●●●●●● ●●●●●● ●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●● ●●●● ●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●● ●●●●●● ●●●●●●●● ●●●●●●●●●●●●● ●●●●● ●●●●●●●● ●●●●●●●●●●● ●●●●● ●●● ●●●● ●●●● ●● ●●●●●●●●● ●●●●●●●● ●●●● ●●●●● ●●●●●●●● ●●●●●● ● ●● ●● ●● ●●● ●●● ●●● ●●●●TPsyn Arid i tyGDD0 Pseason−500−500Species rankEstimations of β*Figure A.1: The distributions of —* estimations for another set of candidate biocli-matic variables across 71 Mimulus species, ranked from low to high values. Dashedlines depict the eect size of zero. GDD0 : growing degree days above 0 Celsiusdegree; Pseason: precipitation seasonality; TPsyn: temperature-precipitation syn-chronicity; and Aridity: aridity of growing season.87Appendices●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●−300 −200 −100 0 100 200 3000.●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●0 2000 4000 6000 8000 100000.●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●0 20 40 60 80 1000.●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●−100 −50 0 50 1000.●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●−200 −100 0 100 200 3000. A.2: Estimated Kernel density curves of 71Mimulus species (depicted as greylines) along the environmental gradient, in comparison with the background Kerneldensity (depicted as a black line) for each of five bioclimatic variables: Tcold (tem-perature of coldest month), GDD0 (growing degree days above 0 Celsius degree),Pseason (precipitation seasonality), TPsyn (temperature-precipitation synchronicity),and Aridity (aridity of growing season), respectively.88AppendicesAppendix B Supplementary materialfor Chapter 3B.1 MethodsB.1.1 Occurrence data filtering associated with nomenclature changesMost changes of the taxonomic revision in Mimulus (Barker et al. 2012) do not aect ourmajor results here, similar to Sheth et al.’s (2014). Nomenclatural modifications retain themajor phylogenetic hypothesis and occurrence data filtering. For some species where nomen-clatural changes are associated with taxonomic issues and geographic range changes, we tookthe following actions. For a few widespread species (e.g. M. guttatus, M. floribundus), newdelineations of close relatives are narrowly distributed oshoots with few records, so theirinclusion does not have a major impact on any estimations of range or niche properties. Afew records in the northwestern corner of the Sierra Nevada of California were described asa new species, M. filicifolius, occupying similar habitats to its close relatives M. laciniatus;therefore, those occurrence data were excluded for M. laciniatus. For M. lewisii, we split itinto two species, one in Sierra Nevada region (Erythranthe erubescens) and the other in thenorthern region (E. lewisii) (Beardsley et al. 2003, Sheth et al. 2014). For M. cardinalis,we kept occurrence data ranging from northern Baja, California to central Oregion (E. car-dinalis), and excluded occurrences in southwestern USA and central Mexico (Beardsley etal. 2003, Sheth et al. 2014). Finally, both M. palmeri and M. montiodes clades have re-cently undergone taxonomic changes (Fraga, 2012), each being split into five species. Thesechanges did not alter the estimations of phylogenetic distance with their close relatives theywere paired with in our analysis (M. gracilipes and M. suksdorfii, respectively). Thus, wedidn’t further split occurrence data; rather we kept them all for M. palmeri and M. mon-tiodes to estimate the overall niche divergence (ND) with the close relatives supported bythe phylogenetic hypothesis.89AppendicesB.1.2 Alternative methods to estimate range extent and rangeoverlapThe range extent for each species was estimated from occurrence data in two major ways: 1)by merging circles with a certain radius (10 km or 50 km) around each occurrence point intoa merged buered polygon (BP: BP10k or BP50k); and 2) by forming a minimum convexpolygon (MCP) containing all occurrence points without or with a 10-km buer (MCP orMCP10k). The BP method with 10-km radius included two dierent sub-methods, one tocreate the circles with 10 km (BP10k1), the other to merge circles with a bigger radius, 50km, then reduce polygon from the edge by 40 km (BP10k2). There were then three methodsto estimate the range overlap (RO) between species, using the area of range overlap betweenthe two species’ ranges as the numerator. Three options for the denominator were: 1)the area of the union of the ranges of both species (ROsum) known as Jaccard similaritycoecient (Phillimore et al. 2008), 2) the area of the smaller range (ROmin, also knownas nestedness), or 3) the area of the larger range (ROmax). Estimate results using BP10k2and BP10k1 were highly correlated (r = 1.000). We preferred to use the BP range methodbecause MCP tended to overestimate range size. We also focus on results with ROsum fortwo reasons. First, ROmin was more sensitive to small ranged species’ overlap status and didnot reflect the prevalence of interactions because the large-ranged species will not encounterthe other species in most of its range. Second, ROsum and ROmax were generally highlycorrelated for both BP and MCP methods (r: 0.798-0.998), especially for pairs having highrange asymmetry. There were five dierent ways to estimate range extent (three BP methodsand two MCP methods), and three ways to quantify range overlap ratio (ROsum, ROmax, andROmin). Together, there were 15 combinations of all possible RO methods. Nonetheless,we conducted a sensitivity analysis by evaluating results using all alternative methods forestimating range and range overlap.B.2 ResultsB.2.1 Correlations among range overlap methodsRange sizes for each species from five range extent methods were highly correlated (R: 0.918-1.000). For each method of estimating range extent, two indices of range overlap, ROsumand ROmax, were highly correlated with one another (R: 0.798-0.995 for BP and 0.985-0.998for MCP), but neither ROsum nor ROmax was highly correlated with ROmin (R: 0.078-0.451for BP and 0.151-0.213 for MCP). For each of three methods of range overlap calculation,90Appendicescorrelations among values from five range extent methods were high (R: 0.611-0.0.993 forROsum, 0.576-0.0.994 for ROmax, and 0.635-0.0.991 for ROmin).B.2.2 The results of sensitivity analysis of alternative range over-lap methods on the relationship between niche divergenceand range overlapSimilar patterns to models using ROsum based on BP10k2 also held when using ROmax(adjusted R2 ranged 0.405 - 0.640), but when using ROmin, slopes were much flatter and notsignificantly dierent from 0 (adjusted R2 ranged 0.255 - 0.397). Models using ROsum andROmax under estimations of dierent range extent methods returned similar results, implyingmore negative associations between RO and PD than models using ROmin.Since macrohabitat niche was quantified based on dierent range extent methods, therewere five estimates for it, corresponding to five ways to estimate range extent. For threemacrohabitat niche axes, multiple linear regression models using ROsum and models usingROmax with BP range extents returned very similar results, where ND was consistentlysignificantly associated with range overlap (all P < 0.05; except one insignificant result fortopographic niche and RO based on BP50k method). Such relationships became insignificantwhen the buer size became larger (i.e., using MCP methods; P ranged from 0.111 to 0.903).Models using ROmin returned less significant results, where RO based on five range extentmethods were all insignificantly associated with ND for bioclimatic and topographic niches(P ranged from 0.072 to 0.831), while for edaphic niches, RO based on BP50k method weresignificantly associated with ND (P = 0.006). For microhabitat ND, there were no significantrelationships between any RO estimate and any PC axis.When accounting for the eect of phylogenetic distance, we found for macrohabitat nicheaxes, PDmcct was significantly (or at least marginally) associated with two estimates of bio-climatic ND based on BP10K1 and BP10k2 methods in models using ROsum and ROmax (Pranged from 0.042 to 0.052), and five estimates of ND in models with ROmin (P ranged from0.015 to 0.039). PDmcct was significantly associated with one estimate of edaphic ND basedon BP10k1 method in models with ROsum and ROmin (P = 0.046, 0.006), and two estimatesof topographic niche based on BP10K1 and BP10k2 methods in models with ROsum, ROmaxand ROmin (P ranged from 0.005 to 0.034). Besides, we noticed no significant relationshipwas detected between ROmin. and PD (adjusted R2 = 021, P = 0.270) neither, implying noclear pattern regarding the geographic model of speciation when using nestedness.91AppendicesB.2.3 The eect of background choices on niche divergence esti-mates in allopatry versus in sympatryFor macrohabitat niche, an alternative choice of background in estimating of macrohabitatND was to set it specific to each species pair, corresponding to the pair-specific union ofthe two species ranges, rather than using the background from all Mimulus pairs. Afterwe recalculated ND in this way, results remained the similar, in which no sign for moreND in sympatry (Figure B.5). For microhabitat niche, the alternative way was to conductPCA of microhabitat attributes for each species pair separately, rather than for all speciestogether. After we did so, there were still a few cases of more divergence in sympatry (FigureB.6). One pair, M. breweri and M. bicolor, still showed more divergence in sympatry forone PC axis. One pairs, M. douglasii and M. congdonii, no longer showed more divergencealong any PC axis. One pair, M. suksdorfii and M. montioides, showed more divergence insympatry on one more PC axis (two in total). The last pair, M. constrictus and M. whitneyi,which did not show more divergence in sympatry when PCA was conducted across all pairs,showed more divergence in sympatry along one PC axis. Cases of significant divergence weredetected all along the first two PC axes. This was probably because most variation wasdiscriminated along first one or two PC axes when conducting PCA within pairs.92AppendicesTable B.1: Summary of 16 Mimulus species pairs and revised taxonomic names (Baker et al. 2012), with information on numberof locality records (given as species 1/species 2), number of study populations for both, range overlap (ROsum) and phylogeneticdistance (PDmcct) between species 1 revised name species 2 revised name # locality # field site ROsum PDmcctp1 M. guttatus Erythranthe guttata M. laciniatus E. laciniata 2834/59 13/3 0.011 0.011‡p2 M. floribundus E. floribunda M. norrisii E. norrisii 624/10 3/2 0.006 0.048p3 M. fremontii Diplacus fremontii M. johnstonii D. johnstonii 296/70 4/3 0.126 0.076p4 M. androsaceus E. androsacea M. shevockii E. shevockii 66/14 4/3 0.063 0.007p5 M. lewisii_n E. lewisii† M. lewisii_s E. erubescens† 272/108 5/7 0.000 0.037p6 M. bigelovii D. bigelovii M. bolanderi D. bolanderi 471/119 4/6 0.003 0.072p7 M. angustatus D. angustatus M. pulchellus D. pulchellus 47/30 4/3 0.002 0.085p8 M. verbenaceus E. verbenacea M. eastwoodiae E. eastwoodiae 45/14 3/3 0.000 0.046p9 M. washingtonensis E. washingtonensis M. jungermannioides E. jungermannioides 25/7 3/3 0.000 0.023p10 M. parryi D. parryi M. rupicola D. rupicola 23/23 0/0 0.000 0.074p11 M. breweri E. breweri M. bicolor E. bicolor 622/148 10/11 0.088 0.013p12 M. douglasii D. douglasii M. congdonii D. congdonii 124/61 4/4 0.130 0.105p13 M. cardinalis E. cardinalis† M. parishii E. parishii 436/77 4/0 0.074 0.024p14 M. suksdorfii E. suksdorfii M. montioides E. montioides† 192/85 3/3 0.050 0.131p15 M. constrictus D. constrictus M. whitneyi D. whitneyi 88/81 3/3 0.228 0.071p16 M. palmeri E. palmeri† M. gracilipes E. gracilipes 141/10 1/2 0.030 0.023† associated with changes in range; otherwise just nomenclatural changes.‡ used PD between M. guttatus and M. nudatus instead due to no genetic sequence of M. laciniatus for phylogeney reconstruction.93AppendicesTable B.2: Multiple linear regression model fits between niche divergence (ND) and twocovariates, range overlap (RO) and phylogenetic distance (PD). The full model included aninteraction term while the restricted model did not. For each niche axis, models were rankedby AICc values.niche axes model formula # parameter adjusted R2 AICc AICcmacrohabitatbioclimatic ND ≥ RO + PD 4 0.413 -15.015ND ≥ RO ◊ PD 5 0.371 -10.839 4.176edaphic ND ≥ RO + PD 4 0.497 -21.895ND ≥ RO ◊ PD 5 0.47 -17.986 3.91topographic ND ≥ RO + PD 4 0.667 -37.367ND ≥ RO ◊ PD 5 0.685 -35.144 2.223microhabitatmicro_PC1 ND ≥ RO + PD 4 0 41.797(bare ground)† ND ≥ RO ◊ PD 5 0 45.687 3.891micro_PC2 ND ≥ RO + PD 4 0.229 36.595(total vegetation)† ND ≥ RO ◊ PD 5 0.152 41.649 5.054micro_PC3 ND ≥ RO + PD 4 0 25.978(slope)† ND ≥ RO ◊ PD 5 0 30.691 4.713† indicated the microhabitat attribute loaded strongly for each PC axis.94Appendicesjungermannioideswashingtonensisbreviflorushymenophylluspatulusmoschatuspulsiferaefloribundusnorrisiidudleyilatidensexiguusmicranthuslaciniatusguttatusglaucescenstilingiiyecorensisdentilobuswiensiidentatusgemmiparusalsinoidescardinalisverbenaceuslewisii sparishiilewisii nrupestriseastwoodiaenelsoniibicolorbrewerifilicaulisrubellusprimuloidesandrosaceusshevockiipurpureusgracilipespalmerisuksdorfiimontioidesinconspicuusaurantiacuslongifloruscalycinusrutilusaustralisariduspuniceusflemingiigrandiflorusclevelandiiangustatuspulchellustricolorpygmaeuscongdoniidouglasiikelloggiitorreyipictusparryirupicolapilosusbigeloviibolandericonstrictuswhitneyimephiticuslayneaeviscidusfremontiijohnstoniibrevipesrattaniimohavensiscusickiinanusjepsoniiclivicolaFigure B.1: 16 closely related pairs of Mimulus selected from a maximum cladecredibility tree (in bold). The entire phylogeny was scaled to a root depth of 1.0.95AppendicesT_coldGDD0P_seasonTP_synAridityPC1 = 54.2%  PC2 = 30.4%(a)BLDCECCLYPPTCRFVOLORCDRCPHIHOXSLTPPTSNDPPTPC1 = 47.33%  PC2 = 30.51%(b)elevationroughnesshillshadeslopePC1 = 58.77%  PC2 = 24.57%(c)slopecanopyveg.totalgr.rockgr.barePC1 = 38.94%  PC2 = 24.63%(d)slopecanopyveg.totalgr.rockgr.barePC1 = 38.94%  PC3 = 16.06%(e)slopecanopyveg.totalgr.rockgr.barePC2 = 24.63%  PC3 = 16.06%(f)Figure B.2: The correlation circles of PCA for (a-c) three coarse-scale macrohabitataxes (bioclimatic, edaphic, and topographic, respectively); and (d-f) combinationsof first three PCs for microhabitat axes (PC1 and PC2, PC1 and PC3, and PC2and PC3, respectively). The variation explained by PC axes is shown in percentagesunder the circle diagrams. Variable abbreviations are: (a) Tcold = the average tem-perature of the coldest month ◊ 10, GDD0 = growing degree days above 0 Celsiusdegree, Pseason = precipitation seasonality, TPsyn = the synchronicity of tempera-ture and precipitation, and Arid = growing season aridity; (b) BLD = bulk density,CRFVOL = coarse fragments volumetric, SNDPPT = fraction sand, SLTPPT =fraction silt, CLYPPT = fraction clay, ORCDRC = soil organic carbon content(log-transformed), PHIHOX = soil pH ◊ 10, and CEC = cation exchange capacity;(d-f) = total vegetation cover, gr.rock = % rocky ground cover, gr.bare =% bare ground cover.96Appendices●●●●●●●●●●●● ● ●●● 0.10phylogenetic distancerange overlapFigure B.3: The relationship between range overlap and phylogenetic distance for16 Mimulus species pairs. No significance was detected here.97Appendices●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●bioclimatic edaphic topographic11 12 13 11 12 13 11 12 distance of range centroid)niche divergence(a)●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●micro_PC1 (bare ground) micro_PC2 (total vegetation) micro_PC3 (slope)11 12 13 11 12 13 11 12 distance of range centroid)niche divergence(b)Figure B.4: The relationship between niche divergence and range distance for 16Mimulus species pairs. Range distance was calculated as the log-transformed dis-tance between centroids of range polygons. No significance was detected here, (a) forthree coarse-scale macrohabitat axes and (b) for first three fine-scale microhabitatPC axes.98Appendices●●●●●●●●●●●●micro_PC1 (bare ground) micro_PC2 (total vegetation) micro_PC3 (slope)allo sym allo sym allo sym0.000.501. divergencespecies pairs●●M. breweri, M. bicolorM. constrictus, M. whitneyiM. douglasii, M. congdoniiM. suksdorfii, M. montioides(a)●●●●●●●●●●●●micro_PC1 (bare ground) micro_PC2 (total vegetation) micro_PC3 (slope)allo sym allo sym allo sym0.000.501.001.500.000.501.001.502. divergencespecies pairs●●M. breweri, M. bicolorM. constrictus, M. whitneyiM. douglasii, M. congdoniiM. suksdorfii, M. montioides(b)Figure B.5: The eect of background space choices on estimation of niche divergenceof three coarse-scaled macrohabitat axes in sympatry (sym) compared to that inallopatry (allo) for species pairs with partial range overlap. (a) background commonto all pairs. (b) background specific to each species pair. Error bars are estimations of95% confidence intervals from the bootstrap method. Solid lines denote significantlymore niche divergence in comparison.99Appendices●●●●●●●● ●●●●bioclimatic edaphic topographicallo sym allo sym allo sym0.200.400.600.800.200.400.600.800.200.400.600.801.00niche divergence species pairs●●M. breweri, M. bicolorM. cardinalis, M. parishiiM. constrictus, M. whitneyiM. douglasii, M. congdoniiM. suksdorfii, M. montioides(a)●●●●●●●● ●●●●bioclimatic edaphic topographicallo sym allo sym allo sym0.400.600.800.200.400.600.800.250.500.751.00niche divergence species pairs●●M. breweri, M. bicolorM. cardinalis, M. parishiiM. constrictus, M. whitneyiM. douglasii, M. congdoniiM. suksdorfii, M. montioides(b)Figure B.6: The eect of background space choices on niche divergence of threemicrohabitat axes in sympatry (sym) compared to that in allopatry (allo) for pairswith partial overlap. (a) PCA conducted across all species. (b) PCA conducted sep-arately for each species pair. Error bars are estimations of 95% confidence intervalsfrom the simulation method. Solid lines denote significantly more niche divergencein comparison.100AppendicesAppendix C Supplementary materialfor Chapter 4Table C.1: Summary of revised taxonomy, the number of locality data and specimen habitatrecords of 82 Mimulus species select from the phylogeny, and the average number of relatedwords for habitat water anity (HWA) and soil type (Soil).tree tip species revised genus # locality # specimen # HWA word # Soil wordalsinoides E. alsinoides Erythranthe 160 217 117.5 16.5androsaceus E. androsacea Erythranthe 82 146 82.0 36.5angustatus D. angustatus Diplacus 55 88 86.3 14.1aridus D. aridus Diplacus 38 70 9.0 25.3aurantiacus D. aurantiacus Diplacus 668 1349 298.6 141.4australis D. australis Diplacus 164 164 20.9 58.7bicolor E. bicolor Erythranthe 152 239 142.0 41.6bigelovii D. bigelovii Diplacus 602 857 409.5 176.6bolanderi D. bolanderi Diplacus 134 211 76.5 53.9breviflorus E. breviflora Erythranthe 104 100 89.0 8.3brevipes D. brevipes Diplacus 505 606 184.3 98.1breweri E. breweri Erythranthe 571 763 708.4 83.9calycinus D. calycinus Diplacus 124 68 12.7 24.7cardinalis E. cardinalis Erythranthe 573 796 865.3 79.5clevelandii D. clevelandii Diplacus 64 122 24.8 13.5clivicola D. clivicola Diplacus 56 72 42.1 21.7congdonii D. congdonii Diplacus 65 110 28.4 34.9constrictus D. constrictus Diplacus 107 214 46.8 39.8cusickii D. cusickii Diplacus 113 192 53.8 45.3dentatus E. dentata Erythranthe 96 110 73.9 1.4dentilobus E. dentiloba Erythranthe 48 54 52.0 5.8douglasii D. douglasii Diplacus 137 205 76.0 55.7dudleyi E. geniculata Erythranthe 44 27 18.4 14.3eastwoodiae E. eastwoodiae Erythranthe 65 69 36.9 22.8exiguus E. exigua Erythranthe 19 34 38.2 1.7filicaulis E. filicaulis Erythranthe 22 46 27.2 7.2flemingii D. parviflorus Diplacus 75 212 22.6 24.5floribundus E. floribunda Erythranthe 788 951 940.3 142.2fremontii D. fremontii Diplacus 380 600 144.6 111.7gemmiparus E. gemmipara Erythranthe 6 11 12.7 14.5glaucescens E. glaucescens Erythranthe 86 119 116.1 14.7gracilipes E. gracilipes Erythranthe 10 11 1.8 9.1grandiflorus D. grandiflorus Diplacus 88 159 47.2 26.2101AppendicesTable C.1: Continued from previous page.tree tip species revised genus # locality # specimen # HWA word # Soil wordguttatus E. guttata Erythranthe 3724 5137 5023.4 416.9hymenophyllus E. hymenophylla Erythranthe 6 9 8.0 5.2inconspicuus E. inconspicua Erythranthe 57 40 11.9 6.2jepsonii D. jepsonii Diplacus 76 140 52.4 26.8johnstonii D. johnstonii Diplacus 94 108 35.0 19.3jungermannioides E. jungermannioides Erythranthe 8 7 3.5 1.6kelloggii D. kelloggii Diplacus 170 291 90.4 57.2latidens E. latidens Erythranthe 61 103 144.3 5.2layneae D. layneae Diplacus 274 485 185.7 164.0lewisii_n E. lewisii Erythranthe 845 1122 1176.3 56.6lewisii_s E. erubescens Erythranthe 109 162 162.8 16.8longiflorus D. longiflorus Diplacus 588 944 260.6 134.5mephiticus D. mephiticus Diplacus 499 633 165.8 189.5micranthus E. arvensis Erythranthe 66 35 34.1 4.2mohavensis D. mohavensis Diplacus 31 45 14.6 6.5montioides E. montioides Erythranthe 27 44 22.0 4.2moschatus E. moschata Erythranthe 931 1199 1239.4 53.1nanus D. nanus Diplacus 425 465 129.2 90.7nelsonii E. nelsonii Erythranthe 6 8 14.4 0.0norrisii E. norrisii Erythranthe 10 30 13.3 32.7nudatus E. nudata Erythranthe 40 73 56.0 42.8palmeri E. palmeri Erythranthe 40 49 41.9 7.7parishii E. parishii Erythranthe 110 145 174.5 16.0parryi D. parryi Diplacus 62 73 19.0 12.7patulus E. patula Erythranthe 33 31 36.4 11.9pictus D. pictus Diplacus 45 62 4.9 30.1pilosus M. pilosa Mimetanthe 666 778 621.6 115.4primuloides E. primuloides Erythranthe 592 785 902.4 52.1pulchellus D. pulchellus Diplacus 36 32 25.0 5.9pulsiferae E. pulsiferae Erythranthe 106 160 113.5 25.3puniceus D. puniceus Diplacus 281 446 106.5 71.6purpureus E. purpurea Erythranthe 32 58 35.3 2.2pygmaeus D. pygmaeus Diplacus 26 45 65.0 2.9rattanii D. rattanii Diplacus 56 97 11.4 19.6rubellus E. rubella Erythranthe 371 462 197.5 101.7rupestris E. rupestris Erythranthe 4 2 1.0 0.0rupicola D. rupicola Diplacus 23 19 3.0 15.0rutilus D. rutilus Diplacus 7 11 5.2 2.8shevockii E. shevockii Erythranthe 19 75 14.3 34.6suksdorfii E. suksdorfii Erythranthe 422 457 253.4 90.4tilingii E. tilingii Erythranthe 445 575 543.0 43.5torreyi D. torreyi Diplacus 250 407 178.2 42.5tricolor D. tricolor Diplacus 181 277 294.6 8.1102AppendicesTable C.1: Continued from previous page.tree tip species revised genus # locality # specimen # HWA word # Soil wordverbenaceus E. verbenacea Erythranthe 143 104 80.6 2.4viscidus D. viscidus Diplacus 42 72 14.4 8.1washingtonensis E. washingtonensis Erythranthe 30 40 49.1 5.5whitneyi D. whitneyi Diplacus 90 181 72.6 49.8wiensii E. madrensis Erythranthe 5 2 0.9 0.0yecorensis E. pallens Erythranthe 21 4 1.5 0.0103AppendicesTable C.2: The list of words for habitat water anity (HWA) from herbarium specimen habitat records, and correspondingcategories and weights.category weight = 1 weight = 0.5 weight = 0.1wet hydric, saturatedwet,seepswet,substratewet, wet,wetmoss, wetness,wetter, wettestaquatic,boggy,flooded,flooding,running,saturatedbar, bar, barstreamside, bog, bog, brook, brooklet, brookside, channel,creek, creekbank, creekbed, creeklet, creekseep, creekside, crk, crkside,drainage, drainageway, falls, flood, floodbed, floodchannel,flooddeposited, floodplain, floodplains, floods, freshwater, highwater, lake,lakelet, lakeshore, lakeside, marsh, marshpond, pond, pond, pondside,pool, rainpool, riparian, riparianstream, river, riverbank, riverbar,riverbed, riverbottom, riverside, riverwash, rivulet, rivulets,runningwater, steambank, steamside, stream, streambank, streambed,streambeds, streambottom, streamlet, streammarsh, streamriver,streamside, streamsidemixed, streamvalley, streamwashed, streamway,swamp, tributary, water, waterfall, watershed, wetland, wetlandmeadowmediate wet mesic, mesichydric,moistwet, semiwet,vernallywet, wetmoistboggyseepy,dripping,marshy, seepy,springybank, ditch, draw, mudflat, mudflow, mud, outflow, seep, seepage,seepagespring, seepbr, seepephemeral, seepmeadow, seepspring, shore,shoreline, spring, springbed, springpond, streammeadowsmoist basalticmoist, damp,damper, dampish,draining,graniticdamp,humusmoist, moist,moistened, moister,moistsoil, moisture,woodlandmoistseasonal,vernal,vernallydepression, gulley, meadow, runo, semiriparianmediate dry dryish, semidry,semimoistdrained,dryingarroyo, arroyos, wash, washed, washesdry arid, dried, drier, dry,xeric, xerophyticdesertscrub104AppendicesTable C.3: The list of words for substrate (Soil) from herbarium specimen habitat records,and corresponding types.soil type rock material wordandesite igneous rock andesite, metadacite, pumiceandesitic, trachyandesiteash igneous rock ash, ashtu, ashy, gravellyashy, sandyashy, tubasalt igneous rock basalt, basaltclay, basaltderived, basaltic, basalticmoist,loambasalt, sandstonebasaltclastic sedimentary rock breccia, conglomerate, conglomeratesandy, pyroclastic,siliciclastic, tuconglomerate, volcaniclastic,volcaniclasticsgneiss metamorphic rock gneissgranite igneous rock alluviumgranitic, granite, granitealluvial,graniteboulder, granitederived, granitediorite,granitegneiss, graniteloam, granitequartzite, granites,granitesand, granitic, graniticcalcareous, graniticclay,graniticdamp, graniticloamy, graniticmud, granitico,graniticpumice, granitoid, granodiorite, quartz,quartzite, quartziteclay, quartziteclay, rockygranite,sandygranitic, semigraniticlimestone sedimentary rock limestone, limestonederived, limestonegravel,limestonelike, limestonemetamorphic, riparianlimestone,sandstonelimestonemarble metamorphic rock marblemetamorphic metamorphic rock limestonemetamorphic, metadacite, metamorphic,metamorphicrock, metamorphics, metamorphosed,metasediment, metasedimentary, metavolcanic,metavolcanicderivedmudstone sedimentary rock coal, peat, mudstoneobsidian igneous rock obsidianpumice igneous rock graniticpumice, pumice, pumiceandesitic, pumiceyrhyolite igneous rock rhyolitesandstone sedimentary rock sandstone, sandstonebasalt, sandstonederived,sandstonelimestone, sandstones, sandstonesandyschist metamorphic rock schist, schistassociated, schistoseserpentine metamorphic rock alkaliserpentine, serpentine, serpentinite, serpentinousshale sedimentary rock shale, shalecampito, shaleclay, shaledecomposing,shalelike, shaleloamy, shaley, shaleyclayey,shaleygravellyslate sedimentary rock slate, slatelikevolcano igneous rock lava, lavaderived, lavarocky, metavolcanic,metavolcanicderived, scoria, volcanic, volcanicderived,volcaniclastic, volcanicsoil105AppendicesTable C.4: Estimated median rates of compound parameters from nearly full ClaSSE model for five niche variables: Pseason,TPsyn, GDD0, HWA and Soil.Pseason TPsyn GDD0 HWA SoilDiversification in G, rG 9.96 12.28 10.81 1.19 -1.36Diversification in S, rS -0.81 0.23 -1.78 15.32 12.99Dierence in diversification, rG ≠ rS 11.09 12.46 12.80 -14.56 -14.64Dierence in speciation, ⁄G ≠ ⁄S 12.72 14.32 15.46 -21.82 -17.42Dierence in extinction, µG ≠ µS 0.09 0.71 1.92 -5.74 -2.30G-to-S cladogenetic change, ⁄GGS 5.60 8.65 7.61 1.27 1.35S-to-G cladogenetic change, ⁄SSG 2.18 1.48 1.41 19.72 8.60G-to-S anagenetic change, qGS 8.02 7.45 5.30 3.11 3.82S-to-G anagenetic change, qSG 7.35 4.34 6.69 2.70 6.55Total G-to-S transition, ⁄GGS + qGS 14.54 16.89 13.54 4.74 5.65Total S-to-G transition, ⁄SSG + qSG 10.87 6.64 8.83 23.50 15.63Transition asymmetry, ⁄GGS + qGS - (⁄SSG + qSG) 4.02 10.25 4.48 -18.47 -9.69Cladogenetic transition asymmetry, ⁄GGS ≠ ⁄SSG 3.12 6.88 5.80 -18.29 -6.83Anagenetic transition asymmetry, qGS ≠ qSG 0.66 2.86 -1.26 0.22 -2.53Total cladogenetic changes, ⁄GGS + ⁄SSG 9.01 10.92 9.68 21.20 10.57Total anagenetic changes, qGS + qSG 15.40 12.00 12.16 6.37 10.59Asymmetry in mode, ⁄GGS + ⁄SSG - (qGS + qSG) -5.96 -0.85 -2.12 14.03 0.11Asymmetry in mode of G-to-S trend, ⁄GGS ≠ qGS -2.43 0.89 2.38 -1.82 -2.27Asymmetry in mode of S-to-G trend, ⁄SSG ≠ qSG -4.66 -2.49 -4.87 16.05 2.12106AppendicesTable C.5: Estimated 95% credibility intervals of compound parameters from nearly full ClaSSE model for five niche variables:Pseason, TPsyn, GDD0, HWA and Soil.Pseason TPsyn GDD0 HWA SoilDiversification in G, rG (-5.17, 19.50) (-7.05, 23.78) (2.18, 19.18) (-7.87, 11.12) (-11.16, 7.01)Diversification in S, rS (-14.59, 15.74) (-10.44, 13.08) (-14.10, 7.34) (-11.20, 33.68) (2.04, 22.08)Dierence in diversification, rG ≠ rS (-20.26, 32.78) (-19.46, 32.56) (-4.58, 31.76) (-39.74, 21.29) (-31.36, 4.5)Dierence in speciation, ⁄G ≠ ⁄S (-18.53, 25.80) (-15.46, 28.95) (-1.22, 26.94) (-43.90, 12.54) (-28.95, -1.46)Dierence in extinction, µG ≠ µS (-17.00, 17.26) (-12.90, 17.36) (-14.14, 17.32) (-25.73, 7.23) (-17.46, 11.30)G-to-S cladogenetic change, ⁄GGS (0.27, 14.87) (0.50, 20.31) (1.42, 16.54) (0.05, 5.42) (0.05, 8.50)S-to-G cladogenetic change, ⁄SSG (0.07, 13.71) (0.05, 9.98) (0.05, 9.49) (0.62, 38.29) (1.50, 17.39)G-to-S anagenetic change, qGS (0.55, 24.36) (0.46, 23.16) (0.27, 20.00) (0.28, 9.39) (0.40, 12.2)S-to-G anagenetic change, qSG (0.59, 22.22) (0.35, 15.22) (0.65, 22.60) (0.10, 12.32) (0.35, 20.27)Total G-to-S transition, ⁄GGS + qGS (3.55, 30.07) (5.12, 33.57) (6.70, 27.93) (1.64, 11.49) (1.61, 16.63)Total S-to-G transition, ⁄SSG + qSG (2.44, 27.38) (1.38, 19.78) (1.97, 26.14) (2.18, 42.59) (8.59, 27.91)Transition asymmetry, ⁄GGS + qGS - (⁄SSG + qSG) (-12.79, 16.01) (-6.17, 23.92) (-7.59, 14.74) (-37.48, 3.11) (-20.33, 0.5)Cladogenetic transition asymmetry, ⁄GGS ≠ ⁄SSG (-11.50, 13.23) (-7.35, 18.97) (-4.42, 15.04) (-37.26, 2.36) (-16.15, 3.43)Anagenetic transition asymmetry, qGS ≠ qSG (-10.41, 12.33) (-6.4, 15.16) (-12.59, 8.78) (-8.84, 7.01) (-14.3, 5.67)Total cladogenetic changes, ⁄GGS + ⁄SSG (2.15, 20.20) (3.22, 22.98) (3.05, 20.31) (3.01, 39.59) (3.20, 20.39)Total anagenetic changes, qGS + qSG (3.08, 43.64) (2.37, 35.25) (2.60, 39.85) (1.28, 17.77) (2.31, 29.39)Asymmetry in mode, ⁄GGS + ⁄SSG - (qGS + qSG) (-36.86, 10.50) (-26.35, 14.28) (-31.61, 11.07) (-5.69, 33.30) (-22.19, 13.88)Asymmetry in mode of G-to-S trend, ⁄GGS ≠ qGS (-21.22, 11.29) (-18.14, 16.25) (-15.68, 13.64) (-8.29, 3.79) (-10.35, 5.27)Asymmetry in mode of S-to-G trend, ⁄SSG ≠ qSG (-19.83, 9.63) (-12.94, 6.84) (-20.6, 4.74) (-3.44, 35.86) (-16.38, 14.82)107


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items