A necessarily complex model to explain the biogeography of
1 A necessarily complex model to explain the biogeography of Madagascar's amphibians and
2 reptiles
3
4 Jason L. Brown
1,2
, Alison Cameron
3
, Anne D. Yoder
1
, Miguel Vences
4
7
8
5
6
1 Department of Biology, Duke University, Durham, NC 27708 Durham, NC, USA
2 Current address: Department of Biology, The City College of New York, NY, USA
3 School of Biological Sciences, Queen's University Belfast, 97 Lisburn Road, Belfast BT9 7BL, UK
4 Zoological Institute, Technical University of Braunschweig, Mendelssohnstr. 4, 38106 Braunschweig, Germany
9
10 Abstract
11 A fundamental limitation of biogeographic analyses are that pattern and process are inextricably
12 linked, and whereas we can observe pattern, we must infer process. Yet, such inferences are
13 often based on ad-hoc comparisons using a single spatial predictor such as climate, topography,
14 vegetation, or assumed barriers to dispersal without taking into account competing explanatory
15 factors. Here we present an alternative approach, using mixed-spatial models to measure the
16 predictive potential of combinations of spatially explicit hypotheses to explain observed
17 biodiversity patterns. In this study we compiled a comprehensive dataset of 8362 occurrence
18 records from 745 amphibian and reptile species from Madagascar. These data were used to
19 estimate species richness, corrected weighted endemism, and species turnover (based on
20 generalized dissimilarity modeling). We also created or incorporated, when previously available,
21 18 spatially explicit predictions of 12 major diversification and biogeography hypotheses, such
22 as: mid-domain, topographic heterogeneity, sanctuary, and climate-related factors. Our results
23 clearly demonstrate that mixed-models greatly improved our ability to explain the observed
24 amphibian and reptile biodiversity patterns. Hence, the observed biogeographic patterns were
1
25 likely influenced by a combination of diversification processes rather than by a single
26 predominant mechanism. Further, selected genera of Malagasy amphibians and reptiles differed
27 in the major factors explaining their spatial patterns of richness and endemism. These differences
28 suggest that key factors in diversification are lineage specific and vary among major endemic
29 clades. Our study therefore emphasizes the importance of comprehensive analyses across
30 taxonomic, temporal, and spatial scales for understanding the complex diversification history of
31 Madagascar's biota. A "one-size-fits-all" model does not exist.
32
33 Keywords: Conditional Autoregressive Models, Orthogonally Transformed Beta Coefficients,
34 Generalized Dissimilarity Modelling, Species Distribution Modelling
35
36 The spatial distribution of biodiversity is at the core of biogeography, macroecology,
37
evolutionary biology, and conservation biology
. Biodiversity mapping indices are multi-
38 faceted concepts with the main components being local endemism, species richness, and species
39 turnover, of which the two latter correspond to alpha- and beta-diversity as used in community
40
. In different combinations, these components are invoked to identify biogeographic
41 regions 5-7 , prioritize geographic areas for conservation 8 , assess the effects of conservation
42 measures
9
, and/or delimit centers of speciation or extinction. These indices, however, are not
43 independent of one another. For instance, species turnover across an area is closely related to the
44 numbers of endemic species within each geographical unit or community, which in turn is often
45 used to estimate areas of endemism (AOE). These represent the coincident restrictedness of
46 taxa 10-12 and are often used to identify unique geographic areas for biodiversity conservation or
2
47
. Clearly, there is an inescapable circularity to these measures, and thus
48 also to the consequent inferences made regarding biogeographic processes.
49 Inferences of speciation mechanism fall prey to similar limitations. For example, it is
50 generally assumed that species formation and diversification of a range of co-distributed taxa
51 will be triggered or inhibited by similar barriers to gene flow, topographical and geological
52 settings, climatic conditions and shifts, and competition. Accordingly, it is the default
53 expectation that similar barriers (e.g., rivers, ecotones, climatic transitions) will lead to similar
54 patterns of species endemism, turnover, and richness; again, with the underlying assumption that
55 the observation of similar patterns among diverse species reveals a general causal mechanism of
56 diversification across all taxa. But there are additional processes by which species richness may
57 be generated. For example, climatic factors, environmental stability, land area, habitat
58 heterogeneity, paleogeography, and energy available all could be spatially correlated with
59 geographical barriers. Thus, any of these mechanisms might be indirectly, but not causally
60 related to diversification
15
. Patterns of endemism, on the other hand, are generally considered to
61 reflect a particular evolutionary history, with areas of endemism corresponding to centers of
62 diversification
16
and often including some element of stochasticity. Consequently, it can be the
63 case that areas of high endemism often are also characterized by high species richness, though
64 the inverse is not necessarily true.
65 Species distribution models (SDMs) allow sophisticated calculations of centers of
66 historical habitat stability
17
. Yet, their spatial comparison with current patterns usually follows
67 narrative approaches and is similar to classical hypotheses of diversification mechanisms, with
68 no accounting for autocorrelation among the different explanatory variables. Based on either a
69 single explanatory variable or without employing statistics at all, often biogeography researchers
3
70 rely on ad-hoc comparisons with spatial distributions of single environmental factors such as
71 climate, topography, vegetation, or assumed barriers to dispersal. As a sign of progress, many
72 methodological advances are being developed to address the various problems described here.
73 For example, assessments of spatial biodiversity have typically used simple geographic measures
74 as the unit of analysis, such as the distribution range of individual species, though recent
75 methodological refinements include the inclusion of phylogenetic relationships among species
76
and their evolutionary age 2,7 . Moreover, carefully parameterized SDMs can generate accurate
77 estimates of distribution ranges
18
and novel approaches are being developed to translate patterns
78 of species richness, endemism and turnover more objectively for determining those
79
biogeographic regions in greatest need for conservation and protection
80 progress in conceptual and statistical tools, biological explanation of these patterns is still in its
81 methodological infancy.
82 Here we aim to employ the latest techniques for sophisticated and improved statistical
83 methods for identifying the causal mechanisms that have determined the spatial distribution of
84 Madagascar's herpetofauna. Though the search for the drivers of biological diversification was
85 initially focused on the Neotropics, considerable attention has more recently been focused on
86 other areas such as the Australian wet tropics 22 and Madagascar 23 . Madagascar is the world's
87 fourth largest island and hosts an extraordinary number of endemic flora and fauna. For example,
88 100% of the native species of amphibians and terrestrial mammals, 92% of reptiles, 44% of
89 birds, and >90% of flowering plants occur nowhere else
24
. This mega diverse micro-continent,
90 initially part of Gondwana, has been isolated from other continents since the Mesozoic. Its
91 current vertebrate fauna is a mix of only a few ancient Gondwanan clades and numerous
4
92 endemic radiations, originating from Cenozoic overseas colonizers arriving mainly from
93 Africa 25-27 .
94 The extraordinary levels of endemism at the level of entire clades in Madagascar, and
95 their long isolation from their sister lineages, provide a unique opportunity to study the
96 mechanisms driving divergence and diversification in situ
28
. Over the past decade, numerous
97 general mechanisms and models have been formulated to explain biodiversity distribution
98 patterns and species diversification in Madagascar, pertaining to environmental stability (or
99 instability), solar energy input, geographic vicariance triggered by topographic or habitat
100
complexity, intrinsic traits of organisms, or stochastic effects
101 numerous hypotheses, though the evidence has typically been marshalled from limited or
102 phylogenetically-constrained taxa. Comprehensive statistical approaches comparing their relative
103 importance are rare 37 .
104 In this paper we apply an integrative approach to simultaneously test which of several
105 competing and complementary hypotheses are most strongly correlated with empirical
106 biodiversity patterns (Fig. 1). We first translate a total of 12 diversification mechanisms or
107 diversity models into explicit spatial representations. We then use diverse statistical approaches
108 to assess spatial concordance of these predictor variables with species richness, endemism and
109 turnover as calculated from original occurrence data of Madagascar's amphibians and reptiles,
110 with full species-level coverage. Our results best agree with the hypothesis that various
111 assemblages of species are under the influence of differing causal mechanisms. The clear
112 message is that the distribution of diverse organismal lineages will depend upon idiosyncratic
113 factors determined by their specific organismal life-history traits combined with stochastic
5
114 historical factors. Thus, any model that endeavors to explain island-wide patterns must
115 necessarily be complex.
116
117
118 R ESULTS
119 To understand spatial distribution patterns in Madagascar's herpetofauna, we first
120 compared range sizes, and computed species richness and endemism from the modeled
121 distribution areas of amphibians and non-avian reptiles (hereafter reptiles). Mean range size (±
122 standard deviation) in our data set is smaller in amphibians than reptiles taking into account all
123 species (41,673 ± 55,413 km
2
vs. 50,205 ± 84,078 km
2
; t= 3.981, p< 0.001; df=649.7) and after
124 excluding species known from only 1 or 2 localities (64,106 ± 57,532 km
2
vs. 95,294 ± 87,495
125 km 2 ; t= 4.511, p<0.001; df=427.4). Microendemics (species with distributions less than
126 1000km
2
) constitute 36.5% of all amphibian and 33.6 % of all reptile species in Madagascar
127 (difference not significant; Z =0.411, p=0.682).
128 Spatial patterns of species richness are quite similar between the two groups (Fig 2A &
129 E) and reach highest values in the eastern rainforest; in amphibians, richness peaks in the central
130 east, whereas in reptiles, it is more evenly distributed across the rainforest biome, with some
131 areas of high diversity also in the north, west, and southwest. Spatial patterns of endemism in
132 both groups (Fig 2B & F) reveal two centers in the north around the Tsaratanana Massif and in
133 the central east. Endemism values for reptiles are also high in southwestern Madagascar, the
134 most arid region of the island.
135
We applied Generalized Dissimilarity Modelling (GDM) 38,39 to identify areas of
136 endemism on the basis of turnover patterns for non-avian reptiles and amphibians together. The
6
137 major AOE obtained in a 4-class categorization of the originally continuous GDM results (Figure
138 2C & H) largely mirrors the bioclimatic regions of Cornet (1974).
139 Our test includes a total of 12 predictor hypotheses, some of which focus on the
140 geographical pattern in which species diversity is distributed, but without making any clear
141 assumption about how the species originated (e.g., the mid-domain or topography heterogeneity
142 hypotheses). Others explicitly refer to mechanisms of diversification and make predictions about
143 how these processes affected the distribution of species diversity over geographical space (see
144 Supplementary Documents for detailed accounts). We divided all the hypotheses into two
145 categories: one in which predictions for continuous two-dimensional spatial richness and
146 endemism can be derived, and another in which nominal AOE predictions can be derived. The
147 first category includes (1) the Mid-domain Effect, (2) Topographic Heterogeneity, (3) Climatic
148 Refugia, (4) Museum (montane refugia), (5) Disturbance-Vicariance, (6) Climate Stability, (7)
149 Sanctuary and (8) Montane Species Pump. The second category includes (9) the River-Refuge
150 (large river model), (10) Riverine Barrier (minor and major rivers), (11) Climatic Gradient and
151 (12) Watershed. All these hypotheses were transformed into explicit spatial representations
152 (Supplementary Materials) and used as predictor variables for further analyses.
153 We calculated unbiased correlation of the continuous predictor and test variables
154 following the method of Dutilleul
40
, which reduces the degrees of freedom according to the level
155 of spatial autocorrelation between two variables (detailed results in Supplementary Materials
156 Table S3). We found that measures of both reptile and amphibian endemism significantly
157 correlated to the predictor hypotheses of Topographic Heterogeneity, Disturbance-Vicariance
158 and Museum (montane refugia). Reptile endemism (but not amphibian) is also correlated to
159 Sanctuary. Correlations to species richness were not tied to measures of endemisms. Whereas
7
160 reptile species richness is correlated to the Mid-domain Effect (distance) and Sanctuary
161 hypotheses, amphibian richness is correlated to the Sanctuary hypothesis as well as to the
162 Topographic Heterogeneity, Montane Species Pump, Disturbance-Vicariance, Museum, and
163 River-Refuge hypotheses.
164 In the univariate correlation analyses of nominal geospatial data (those related to AOE
165 predictions) we compared the biogeographic zonation of Madagascar as suggested by the GDM
166 analysis of amphibian and reptile distributions with zonations derived from five predictor
167 hypotheses. We found all predictor variables (corresponding to the hypotheses Riverine-major
168 and Riverine-minor, Gradient, River-refuge, and Watershed) to be significantly correlated to the
169 15-class GDM, and all but watershed with the 4-class GDM zonation (Table 1). Both GDM
170 classifications share the most overlap with the Riverine and Gradient hypotheses (between 40.9–
171 54.3% and 56.2–71.1%, respectively; Table 1).
172 Given the significant correlation of each of the spatial amphibian and reptile biodiversity
173 patterns with various predictor variables we used mixed conditional autoregressive spatial
174 models (CAR models) to test the influences of various predictors simultaneously. To avoid over-
175 parameterization we used AICc (corrected Akaike Information Criterion), an information-
176 theoretical approach, to compare models with different sets of predictors. We found that complex
177 models including most of the biogeography hypotheses (continuous predictor variables)
178 performed best, based on the lowest AICc values, and consequently used these for further
179 analysis. Detailed contributions of each predictor to the models of richness, endemism and GDM
180 zonation are summarized in Supplementary Materials Table S4. The top-five variables
181 contributed 49.4–75.9% to the models. For a more simplified graphical representation (Fig. 3),
182 we summarized the three Mid-domain Effect hypotheses (latitude, longitude, and distance), the
8
183 three principal components representing the Climate Gradient hypothesis, and three hypotheses
184 focused on topography (Topographic Heterogeneity, Disturbance-vicariance, Montane Species
185 Pump) were categorized together, respectively (Figs 3 & 4) . We found relevant influences of the
186 Mid-domain Effect especially on the GDM and the species richness and endemism of reptiles
187 (30.9%, 32.9% and 45.5%, respectively). However, it is important to point out that almost all the
188 Mid-domain correlation coefficients were negative. Thus, indicating that Mid-domain Effects do
189 not play a key role in determining spatial patterning. Climate Gradient effects influenced all the
190 models of biodiversity equally, contributing roughly a quarter to each (25.1–27.7%), though in
191 many cases the sign of the contribution varied. However in this case, a positive correlation was
192 not expected. The topography variables contributed positively to the richness and endemism
193 models of amphibians and reptiles, with joint influences of 9.1% and 22.4% on richness, and
194 6.5% and 17.3% on endemism. The two unique hypotheses, Sanctuary and Museum, each
195 contributed positively to all models, with Museum contributing between 7.1–17.1% (one of the
196 few hypothesis to contribute >5% and to be positively correlated to all biodiversity
197 measurements). The Sanctuary hypothesis also contributed positively to all hypotheses, though
198 to a lesser degree than the Museum hypothesis (which demonstrated little contribution to reptile
199 endemism).
200 To assess variation in biogeography patterns among major groups of the Malagasy
201 herpetofauna, we calculated mixed CAR models using the same methods for richness and
202 endemism of four exemplar sub-clades: the leaf chameleons ( Brookesia ), tree frogs ( Boophis ),
203 day geckos ( Phelsuma ) and Oplurus iguanas (including the monotypic iguana genus
204 Chalarodon ). The top contributors to the models were drastically different for several of these
205 clades (Fig. 4). For instance, the topography variables had strong influences on Boophis richness,
9
206 with a joint contribution of 24.5%, but contributed much less to explaining the patterns of most
207 other groups. Further, the Sanctuary hypothesis had a strong influence on the Brookesia and
208 iguana models, though it contributed very little to the predictions of endemism in Boophis and
209 Phelsuma . Mid-domain Effects were apparent in most models but the sign on the correlation and
210 the contribution of each Mid-domain hypothesis varied considerably.
211
212
213 D
ISCUSSION
214 The results of this study clearly demonstrate that single-mechanism explanatory
215 hypotheses of spatial patterning in Madagascar's herpetofauna (and presumably, other Malagasy
216 vertebrates) are inadequate. Thus, we propose a novel method for examining and synthesizing
217 spatial parameters such as species richness, endemism, and community similarity. In this
218 framework, biogeographic hypotheses are explanatory variables. The resulting mixed-model
219 geospatial approach to biogeography analyses is both more robust, and more realistic. Our
220 approach has the potential to reduce bias and subjectivity in the search for prevalent factors
221 influencing the distribution of biodiversity, both in Madagascar and elsewhere. Currently,
222 researchers typically approach such questions by univariate and sometimes narrative analyses
223 that examine the fit of the observed patterns to only single explanatory models or mechanisms
224
) or compare a limited number of competing variables in univariate
225 approaches
37
. Such analyses are hampered, however, by spatial autocorrelation of biodiversity
226
patterns and predictor variables thereby inflating type-I errors in traditional statistical tests
227 Several solutions have been proposed for this problem. Some authors attempt to exclude spatial
10
228 autocorrelation from models
44
, whereas others incorporate spatial autocorrelation as a predictive
229 parameter in geospatial models 45-47 , as was applied in this study.
230 The results obtained here for some sub-clades are in agreement with previous analyses,
231 while others are not. For example, the high influence of the mid-domain effect on Boophis
232 treefrogs, one of the most species-rich frog genera in Madagascar, agrees with a previous
233 analysis performed by Colwell & Lees
48
for all Malagasy amphibians (with a high representation
234 of Boophis ). On the contrary, the negative contributions of the mid-domain effects on the
235 biodiversity patterns of the other genera in the analysis are obvious given that their centers of
236 richness and endemism are in either southern or northern Madagascar, but not in central parts of
237 the island. Previous studies postulated a high influence of topography on the diversification of
238 leaf chameleons ( Brookesia
though this is not supported by our analysis. This latter
239 example exemplifies a dilemma of scale, inherent in all comparisons of spatial data sets. In fact,
240 the distribution of Brookesia is highly specific to certain mountain massifs in northern
241 Madagascar while the genus is largely absent from the equally topographically heterogeneous
242 south-east. This absence is probably due to its evolutionary history, with a diversification mainly
243 in the north and limited capacity for range expansion
41
. This historical distribution pattern
244 probably accounts for low influence of the topographic hypotheses on Madagascar-wide
245 Brookesia richness and endemism, while at a smaller spatial scale (northern Madagascar) these
246 hypotheses might well have a strong predictive value.
247
248
While patterns of richness and endemism of the Malagasy herpetofauna have been
analyzed several times for various purposes based on partial data sets
249 turnover of species composition and the definition of biogeographic regions following from such
250 explicit analyses are still in their infancy. For reptiles, Angel's
50
proposal of biogeographic
11
251 regions based on classical phytogeography, i.e., regions based on plant community
252 composition 51 , has usually been adopted 52 . Later, Schatz 53 refined this zonation of Madagascar
253
254 based on explicit bioclimatic analyses, and Glaw & Vences
54
proposed a detailed geographical zonation based on the areas of endemism of Wilmé 33
. The GDM approach herein is the first
255 explicit analysis of a large herpetofaunal dataset to geographically delimit regions distinguished
256 by abrupt changes in their amphibian and reptile communities. This model turned out to agree
257
remarkably well with classical bioclimatic and phytogeographic zonations of Madagascar 51,53 ,
258 strongly correlated to climatic explanatory variables (Fig. 3). Especially in the 4-classes GDM,
259 the regions almost perfectly correspond with those proposed by Schatz
53
based on bioclimate,
260 i.e., eastern humid, central highland/montane, western arid, southwestern subarid zone. Although
261 the coincidence of the precise boundaries of these regions might be methodologically somewhat
262 biased, as we interpolated community distribution using climate variables in the analysis, the
263 model is still mainly based on real distributional information of species and thus provides
264 convincing evidence that amphibian and reptile communities strongly differ among the major
265 bioclimatic zones of Madagascar.
266 Several authors have suggested that the current distribution of biotic diversity in the
267
tropics resulted from a complex interplay of a variety of diversification mechanisms 55,56 . This
268
269 implies that no single hypothesis adequately explains the diversification of broad taxonomic groups
— our results support this assumption. Richness, endemism and turnover of large and
270 heterogeneous groups exemplified by the all-species amphibian and reptile data sets were in all
271 cases best explained by complex CAR models. These models have the advantage of
272 incorporating most or all of the originally included explanatory variables.
12
273 Several alternative explanations may account for this outcome. Patterns of biodiversity
274 may not be strongly correlated to any of the predictor mechanisms simply because none of them
275 provide the causal mechanism underlying the diversification processes. As another consideration,
276 spatial predictions of some of the biodiversity hypotheses may have been inaccurate, though we
277 took great care to avoid such mistakes. In any event, improvements in these methods may result
278 in different outcomes in future analyses.
279 Caveats aside, the results of this study almost certainly support a third explanation, that
280 different clades of organisms are each predominantly influenced by a different set of
281 diversification mechanisms. In turn, these are driven by intrinsic factors, such as morphological
282 or physiological constraints, or to extrinsic factors, such as an initial diversification in an area
283 characterized by a certain topography, climate, or biotic composition. This alternative is
284 supported by the observation that the patterns of several of the smaller subgroups in our analysis
285 were indeed best explained by opposing predominant variables, e.g., topographic heterogeneity
286 and museum ( Boophis endemism) vs. climate stability and sanctuary ( Brookesia endemism). An
287 overarching message is that the taxonomic scale of analysis is of extreme importance when
288 attempting to derive global explanations of biodiversity distribution patterns. Including too many
289 taxa will blur the existing differences among clades and lead to complex explanatory models,
290 whereas patterns within specific clades may be best explained by simple models.
291 The method proposed herein allows for a more objective quantification of the influences
292 of particular diversification mechanisms on biodiversity patterns, compared to traditional,
293 univariate approaches. Further developments of the method should especially focus on including
294 a phylogenetic dimension, and when appropriate (for predictor hypotheses), a temporal
295 component. Geospatial analyses of biodiversity pattern typically use species as equivalent and
13
296 independent data points, though in reality, they are entities with substantial variation in
297 parameters such as evolutionary age, dispersal capacity and population density, and with
298 different degrees of relatedness depending on their position in the tree of life. This multilayered
299 information can be included in various ways in the CAR/OTBC approach, e.g. by plotting
300 richness and endemism of evolutionary history rather than taxonomic identity, calculating
301 turnover only for sister species with adjacent ranges, or repeating the calculations for sets of
302 species defined by particular nodes on a phylogenetic tree. This latter approach— iterating the
303 analysis for successively more inclusive clades
— appears particularly promising for identifying
304 those moments in evolutionary history wherein shifts in prevalent diversification mechanisms
305 have occurred. Given this perspective, we can begin to tease apart the diversification histories of
306 individual clades versus prevailing biogeoclimatic events that shape entire biotas.
307
308
309 M ATERIALS AND M ETHODS
310
311 Biodiversity Estimates
312
313 Species Distribution Modeling
314 Species data consisted of 8362 occurrence records of 745 Malagasy amphibian and
315 reptile species (325 and 420 species, respectively). Species distribution models (SDMs) were
316 limited to species that had, at minimum, 3 unique occurrence points at the 30 arc-second spatial
317 resolution (ca. 1 km). The reduced dataset represented 453 species (consisting of 5440 training
318 points of 248 reptile and 205 amphibian species), with a mean of 12 training points per species
14
319 (max= 131). For 107 amphibian and 119 reptile species with only 1-2 occurrence records a 10km
320 buffer was applied to point localities in place of SDM. The species distribution models were
321 generated in MaxEnt v3.3.3e
57
using the following parameters: random test percentage = 25,
322 regularization multiplier = 1, maximum number of background points = 10000, replicates = 10,
323 replicated run type = cross validate.
324 One limitation of presence-only data SDM methods is the effect of sample selection bias,
325 where some areas in the landscape are sampled more intensively than others 58 . MaxEnt requires
326 an unbiased sample. To account for sampling biases, we used a bias file representing a Gaussian
327 kernel-density of all species occurrence localities. The bias file upweighted presence-only data
328 points with fewer neighbors in the geographic landscape
59
. Species distributions were modeled
329 for the current climate using the 19 standard bioclimatic variables derived from interpolation of
330 climatic records between 1950 and 2000 from weather stations around the globe (Worldclim
331 1.4
60
). Non-climatic variables geology, aspect, elevation, solar radiation, and slope were also
332
. All layers were projected to Africa Alber’s Equal-Area Cylindrical projection in
333 ArcMap at a resolution of 0.91 km
2
.
334
335 Correcting SDMs for Over-prediction
336
337
To limit over-prediction of SDMs, a problem common with modeling distributions of
, we clipped each SDM following the approach of Kremen
et al.
8
. This
338 method produces models that represent suitable habitat within an area of known occurrence
339 (based on a buffered MCP), excluding suitable habitat greatly outside of observed range. The
340 size of the buffer was based on the area of the MCP. We used buffer distances of 20km, 40km,
341 and 80km, respectively, for three MCP area classes, 0-200km
2
, 200-1000 km
2
, and >1000 km
2
.
15
342 All corrected SDMs were proofed by taxonomic experts to ensure reliability; if a model did not
343 tightly match knowledge of areas where distributions were well documented, or if little prior
344 information existed regarding a species distribution, or taxonomy was convoluted, and because
345 of, its expected distribution could not be evaluated, the species was excluded from analyses (n=
346 71).
347
348 Range Sizes, Species Richness and Corrected Weighted Endemism
349
350
For descriptive range-size statistics, distribution range-sizes were sampled for all species at 0.01 degrees
2
from corrected SDMs (or buffered point data where applicable) and a student’s
351 t-test with unequal variance was performed between amphibian and reptile species. To assess
352 differences in the frequency of microendemics among the two groups, we converted all
353 distributions that were > or < than 1000 km 2 to a value of 0 and 1, respectively. We then
354 calculated the mean frequency for both groups and ran a binomial test among both groups.
355 Species richness was calculated separately for amphibians and reptiles by summing the
356 respective corrected binary SDMs (based on a maximum training sensitivity plus specificity
357 threshold) and, for species with 1-2 occurrence records, buffered points in ArcGIS. This
358 provided a high resolution estimate of richness that is less affected by spatial scale and
359 incomplete sampling than traditional measurements based solely on occurrence records.
360 Measures of endemism are inherently dependent on spatial scale. We chose a grid scale
361 of 82 x 63 km, separating Madagascar into 24 latitudinal and 8 longitudinal rows, to reduce
362
problems associated with estimating endemism over too small or large areas
363 measured as corrected weighted endemism (CWE), where the proportion of endemics are
364 inversely weighted by their range size (species with smaller ranges are weighted more than those
16
365 with large
63
) and this value divided by the local species richness
12
. CWE emphasizes areas that
366 have a high proportion of animals with restricted ranges, but not necessarily high species
367 richness. We calculated CWE separately for reptiles and amphibians using SMDtoolbox v1
368 (Brown in review).
369
370 Generalized Dissimilarity Modeling
371 Generalized Dissimilarity Modeling (GDM) is a statistical technique extended from
372 matrix regressions designed to accommodate nonlinear data commonly encountered in ecological
373 studies
38
. One use of GDM is to analyze and predict spatial patterns of turnover in community
374 composition across large areas. In short, a GDM is fitted to available biological data (the absence
375 or presence of species at each site and environmental and geographic data) then compositional
376 dissimilarity is predicted at unsampled localities throughout the landscape based on
377 environmental and geographic data in the model. The result is a matrix of predicted
378 compositional dissimilarities (PCD) between pairs of locations throughout the focal landscape.
379 To visualize the PCD, multidimensional scaling was applied, reducing the data to three
380 ordination axes, and in a GIS each axis was assigned a separate RGB color (red, green or blue).
381 Due to computation limitations associated with pairwise comparisons of large datasets,
382 we could not predict composition dissimilarities among all sites in our high resolution
383 Madagascar dataset. To address this, we randomly sampled 2500 points throughout Madagascar
384 from a ca. 10 km
2
grid. We then measured the absence or presence of each of the 679 species at
385 each locality. We used the same high resolution environmental and geography data used in the
386 SDM. These 23 layers were reduced to nine vectors in a principal component analyses which
387 represented 99.4% of the variation of the original data. These data were sampled at the same
17
388 2500 localities. Both data (species presence and environmental data) were input into a
389 generalized dissimilarity model using the R package: GDM R distribution pack v1.1
390 (www.biomaps.net.au/gdm/GDM_R_Distribution_Pack_V1.1.zip). We then extrapolated the
391 GDM into the high resolution climate dataset by assigning ordination scores using k-nearest
392 neighbor classification (k=3, numeric Manhattan distance), calculating each ordination axes
393 independently
38
.
394 The continuous GDM was transformed into a model with four major classes, and each of
395 these was then classified separately into 3-5 minor classes. The numbers of major and minor
396 classes were based on hierarchical cluster analyses in in SPSS v19
64
using a “bottom up”
397 approach. The number of classes equaled the number of dendrogram nodes with relative
398 distances (scaled from 0-1) at 0.71 and 0.63 for major and minor groups, respectively. The
399 distance cut off can be somewhat arbitrary, however in our data there were obvious
400 discontinuities (long dendrogram branches between nodes) at these two values. The resulting
401 classified models were interpolated into high resolution climate space using a k-nearest neighbor
402 classification as described above.
403
404 Biogeography hypotheses
405 We examined which specific spatial predictions for the three biodiversity patterns:
406 species richness, endemism and/or in areas of endemism (AOE- the coincident restrictedness of
407 taxa) in Madagascar could be derived from each of 12 biogeography hypotheses, and then
408 translated these predictions into spatial models in a GIS.
409 In a GIS, spatially explicit predictions of the three biodiversity patterns (species richness,
410 endemism or areas of endemism) were estimated for each biogeography hypothesis. For some of
18
411 the hypotheses not all three metrics of biodiversity were calculated due to lacking, or incomplete,
412 expectations (e.g. not all hypothesis make predictions about AOE). Because of these incomplete
413 biodiversity pattern predictions, comparisons among hypotheses are statistically non-trivial. This
414 is in part because few diversification hypotheses capture all facets of biodiversity (species
415 richness, endemism, AOE). Further, many estimates of biodiversity patterns rely on components
416 of climate or geography, thus some are based on the same data and are not entirely independent
417 of each other. Each hypothesis was generated at the spatial resolution of 30 arc-seconds
418 (matching the resolution of GDM and species richness estimates, later transformed to 0.91 km
2
).
419 For the endemism analyses, each biogeography hypothesis was upscaled to match resolution of
420 the endemism analyses by averaging all values encompassed in cell.
421
422 Spatial Statistics
423 The spatial predictions derived from the various biodiversity hypotheses resulted in either
424 continuous or nominal categorical data. Conducting statistical tests between data types is
425 nontrivial and, in some cases, not logical or impossible. Spatial data are represented in GIS by
426 two different formats: raster and vector. Geospatial raster data are composed of equal sized
427 squares, tessellated in a grid, with each cell representing a value (often continuous data), such as
428 elevation. Spatial vector data (commonly called ‘shapefiles’) can be represented by points, lines,
429 or polygons, such as: localities, roads and countries, respectively. Vector data are non-
430 topological and represent discrete features. They are often used to depict nominal data, where the
431 relationship of data categories to others is unknown or non-linear.
432 Raster data can be converted to vector data (and vice versa) and the data type ( i.e.
433 nominal or continuous) may or may not change when converted. For example, in some cases
19
434 continuous data can be converted to ordered categories (ordinal data) when converted from raster
435 to polygon. However if the same data were converted back to a raster file, it would remain
436 categorical data due to data loss in the first conversion. Regardless of GIS data format, statistical
437 tests need be chosen according to the data types, however GIS data format remain equally
438 important, as often a single data format is required to perform a spatial statistic of interest (i.e.
439 software input limitations).
440 Analyses- Continuous Data
441
442
To assess a global measurement of correlation between continuous data, we calculated
Pearson correlations following the unbiased correlation method of Dutilleul
40
and using the
443 software Spatial Analysis in Macroecology
65
.
444 Analyses- Nominal Categorical Data
445 Comparisons of nominal categorical spatial data ( i.e.
AOE predictions compared to
446 classified GDM) focused on the spatial distributions of the borders between the subunits. We
447 used the following methods to measure similarities and significance: (1) border overlap, and (2)
448 Pearson correlation coefficients (r) with Dutilleul’s spatial correlation (see above).
449 (1) Border overlap was calculated by sampling the landscape at 1 km resolution for the
450 presence of a border. If present, a point was placed. We then measured the spatial overlap of the
451 sampled borders of two landscapes. In all analyses, a 10 km buffer was applied to the overlap
452 calculation, and points datasets that overlap by 10km or less are were considered overlapping
453 boundaries. To account for differences in the level of subdivision of layers, overlap was
454 converted to a percentage and averaged for both layers being compared. Country outline was
455 excluded from all comparisons and thus, only intra-country boundaries were compared.
20
456 (2) To assess global correlation between two polygon shapefiles, each shapefile was
457 converted to a distance raster, measuring the closest distance from any point in the landscape to a
458 boundary. Using these layers we measured a Pearson correlation (unbiased correlation after
459 Dutilleul
40
), where high correlation coefficients represent two landscapes that have congruent
460 areas that are isolated from boundaries and others congruent areas that are adjacent to
461 boundaries. Each distance landscape was evenly sampled by 2000 points and correlations were
462 assessed on the values of these points.
463 Analyses- Mixed Models of Continuous Data
464 To determine the influence of each biogeography hypothesis in predicting the observed
465 biodiversity patterns, we integrated all continuous biogeography hypotheses into a single mixed
466 conditional autoregression model (CAR) using the software Spatial Analysis in Macroecology
65
.
467 To normalize the predictor variables, Box-Cox transformations (Box and Cox 1964) were
468 performed. The lambda parameter was estimated by maximizing the log-likelihood profile in R
469 package GeoR
47
. A Gabriel connection matrix was used to describe the spatial relationship
470 among sample points
66
. Using Gabriel networks, short connections between neighboring points,
471 is preferable (i.e. more conservative
67
) than using inverse-decaying distances because in most
472 empirical datasets the residual spatial autocorrelation tends to be stronger at smaller distance
473 classes
68
.
474 The main goal of our mixed spatial analyses were to determine the combination of
475 biogeography hypotheses that best predict the observed biodiversity patterns. If each explanatory
476 variable was incorporated natively, due to considerable multi-colinearity, often only a few
477 variables would end up contributing to a majority of the model. To estimate the true contribution
478 of each hypothesis in context of a mixed model (even if highly correlated to others), we
21
479 developed a novel approach that removes colinearity from the response variables (but in the
480 process explicit variable identity is temporarily lost). The transformed response variables are
481 then run in a CAR analysis and the resulting standardized model contributions are then
482 transformed back into original response variable identities; reflecting the relative contribution of
483 each in the model. This process is casually referenced here as Orthogonally Transformed Beta
484 Coefficients (OTBCs).
485 Orthogonally Transformed Beta Coefficients
486 Each biogeography hypothesis is standardized from zero to one. This ensured that the
487 component loadings reflected the relative contribution of each biogeography hypothesis. A
488 principal component analysis was performed on the standardized biogeography hypotheses using
489 a covariance matrix. All the resulting principal components (PCs) were extracted and then loaded
490 as explanatory variables in the CAR model. The CAR analyses were run iteratively, starting with
491 all PCs as response variables and then excluding each PC that did not contribute significantly to
492 the model (α = 0.05) until the final model included only PCs that contributed significantly to the
493 model. Because each PC represented a linearly uncorrelated variable, only the relevant,
494 independent data were incorporated into the final CAR model. The resulting standardized beta
495
496 coefficients ( β j from the CAR analyses, Fig. 1 and Equation 1) were then multiplied by the value of the corresponding component loadings (
α ij
from the PCA, see Equation 1). The absolute value
497 of the product reflects the relative contributions of each biogeography hypothesis to each PC,
498 which are weighted by the PC’s contribution in the CAR model (herein termed the weighted
499 component loadings or WCL if
, Equation 1). The weighted component loadings ( WCL if
, Equation
500 1) were then summed for each biogeography hypothesis across all PCs ( H i
) and depict the
501 contributions of each hypothesis in the CAR model. The 𝐻 𝑖
value was then converted to
22
502 percentages ( HP i
) to allow comparison among all CAR analyses. A positive or negative
503 correlation was determined for each biogeography hypothesis by running a separate CAR
504 analysis using the raw biogeography variables as a single response variable (all other parameters
505 were matched).
506
507 Equation 1
508 𝑾𝑪𝑳
𝒊𝒋
= |𝛽 𝑗
| × |𝛼 𝑖𝑗
| 𝑯 𝒊
= ∑ 𝑊𝐶𝐿 𝑖𝑗 𝑖
𝑯
𝑨𝒍𝒍
= ∑ 𝑊𝐶𝐿 𝑖𝑗 𝑖𝑗
𝑯𝑷 𝒊
𝐻 𝑖
= (
𝐻 𝑎𝑙𝑙
) ∗ 100
509
510
511 Acknowledgments
512 We are grateful to numerous friends and colleagues who provided invaluable assistance
513 during fieldwork and previous discussions of Madagascar's biogeography, we would like to
514 particularly thank Franco Andreone, Parfait Bora, Christopher Blair, Lauren Chan, Sebastian
515 Gehring, Frank Glaw, Steve M. Goodman, Jörn Köhler, Peter Larsen, David C. Lees, Brice P.
516 Noonan, Maciej Pabijan, Ted Townsend, Krystal Tolley, Roger Daniel Randrianiaina
517 Fanomezana Ratsoavina, David R. Vieites, and Katharina C. Wollenberg. Fieldwork of MV was
518 funded by the Volkswagen Foundation. JLB was supported by the National Science Foundation
519 under Grant No. 0905905 and by Duke University start-up funds to ADY.
520
521
522
523
524
23
525 References
559
560
561
562
563
564
565
566
567
545
546
547
548
549
550
551
552
553
554
555
556
557
558
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
1
2
3
4
5
6
7
Kent, M. Biogeography and macroecology. Progress in Physical Geography 29 , 256-
264, doi:10.1191/0309133305pp447pr (2005).
Beck, J.
et al.
What's on the horizon for macroecology? Ecography 35 , 673-683, doi:10.1111/j.1600-0587.2012.07364.x (2012).
Whittaker, R. H. Vegetation of the Siskiyou Mountains, Oregon and California. Ecol
Monogr 30 , 279-338, doi:10.2307/1943563 (1960).
Whittaker, R. H. Evolution and Measurement of Species Diversity. Taxon 21 , 213-251, doi:10.2307/1218190 (1972).
Williams, P. H. Mapping variations in the strength and breadth of biogeographic transition zones using species turnover. Proceedings of the Royal Society B-Biological
Sciences 263 , 579-588, doi:10.1098/rspb.1996.0087 (1996).
Kreft, H. & Jetz, W. A framework for delineating biogeographical regions based on species distributions. Journal of Biogeography 37 , 2029-2053, doi:10.1111/j.1365-
2699.2010.02375.x (2010).
Holt, B. G.
et al.
An Update of Wallace's Zoogeographic Regions of the World. Science
339 , 74-78, doi:10.1126/science.1228282 (2013).
8 Kremen, C.
et al.
Aligning conservation priorities across taxa in Madagascar with highresolution planning tools. Science 320 , 222-226, doi:10.1126/science.1155193 (2008).
Hoffmann, M.
et al.
The Impact of Conservation on the Status of the World's Vertebrates. 9
Science 330 , 1503-1509, doi:10.1126/science.1194442 (2010).
10 Platnick, N. I. On areas of endemism. Australian Systematic Botany 4 , 2pp.-2pp. (1991).
11 Harold, A. S. & Mooi, R. D. Areas of endemism: definition and recognition criteria.
Systematic Biology 43 , 261-266, doi:10.2307/2413466 (1994).
12 Crisp, M. D., Laffan, S., Linder, H. P. & Monro, A. Endemism in the Australian flora.
Journal of Biogeography 28 , 183-198 (2001).
13 Terborgh, J. & Winter, B. A method for siting parks and reserves with special reference to Columbia and Ecuador. Biological Conservation 27 , 45-58 (1983).
14 Ackery, P. R. & Vane-Wright, R. I.
Milkweed butterflies: Their cladistics and biology .
(British Museum of Natural History and Cornell University Press, 1984).
15 Hawkins, B. A.
et al.
Energy, water, and broad-scale geographic patterns of species richness. Ecology 84 , 3105-3117, doi:10.1890/03-8006 (2003).
16 Jetz, W., Rahbek, C. & Colwell, R. K. The coincidence of rarity and richness and the potential signature of history in centres of endemism. Ecology Letters 7 , 1180-1191, doi:10.1111/j.1461-0248.2004.00678.x (2004).
17 Carnaval, A. C., Hickerson, M. J., Haddad, C. F. B., Rodrigues, M. T. & Moritz, C.
Stability Predicts Genetic Diversity in the Brazilian Atlantic Forest Hotspot. Science 323 ,
785-789, doi:10.1126/science.1166955 (2009).
18 Kozak, K. H., Graham, C. H. & Wiens, J. J. Integrating GIS-based environmental data into evolutionary biology. Trends in Ecology & Evolution 23 , 141-148, doi:10.1016/j.tree.2008.02.001 (2008).
19 Lamoreux, J. F.
et al.
Global tests of biodiversity concordance and the importance of endemism. Nature 440 , 212-214, doi:10.1038/nature04291 (2006).
24
589
590
591
592
593
594
595
596
597
598
599
600
601
602
582
583
584
585
586
587
588
603
604
605
606
607
608
609
610
611
612
568
569
570
571
572
573
574
575
576
577
578
579
580
581
20 Linder, H. P.
et al.
The partitioning of Africa: statistically defined biogeographical regions in sub-Saharan Africa. Journal of Biogeography 39 , 1189-1205, doi:10.1111/j.1365-2699.2012.02728.x (2012).
21
Olivero, J., Márquez, A. L. & Real, R. Integrating fuzzy logic and statistics to improve the reliable delimitation of biogeographic regions and transition zones. Systematic biology 62 , 1-21 (2013).
22 Graham, C. H., Moritz, C. & Williams, S. E. Habitat history improves prediction of biodiversity in rainforest fauna. Proceedings of the National Academy of Sciences of the
United States of America 103 , 632-636, doi:10.1073/pnas.0505754103 (2006).
23 Vences, M., Wollenberg, K. C., Vieites, D. R. & Lees, D. C. Madagascar as a model region of species diversification. Trends in Ecology & Evolution 24 , 456-465, doi:10.1016/j.tree.2009.03.011 (2009).
24 Goodman, S. M. & Benstead, J. P. The natural history of Madagascar . (University of
Chicago Press Chicago, 2003).
25 Yoder, A. D. & Nowak, M. D. Has vicariance or dispersal been the predominant biogeographic force in Madagascar? Only time will tell. Annual Review of Ecology,
Evolution, and Systematics , 405-431 (2006).
26 Crottini, A.
et al.
Vertebrate time-tree elucidates the biogeographic pattern of a major biotic change around the K-T boundary in Madagascar. Proceedings of the National
Academy of Sciences of the United States of America 109 , 5358-5363, doi:10.1073/pnas.1112487109 (2012).
27 Samonds, K. E.
et al.
Spatial and temporal arrival patterns of Madagascar's vertebrate fauna explained by distance, ocean currents, and ancestor type. Proceedings of the
National Academy of Sciences of the United States of America 109 , 5352-5357, doi:10.1073/pnas.1113993109 (2012).
28 Yoder, A. D.
et al.
A multidimensional approach for detecting species patterns in
Malagasy vertebrates. Proceedings of the National Academy of Sciences of the United
States of America 102 , 6587-6594, doi:10.1073/pnas.0502092102 (2005).
29 Pastorini, J., Thalmann, U. & Martin, R. D. A molecular approach to comparative phylogeography of extant Malagasy lemurs. Proceedings of the National Academy of
Sciences of the United States of America 100 , 5879-5884, doi:10.1073/pnas.1031673100
(2003).
30 Goodman, S. M. & Ganzhorn, J. U. Biogeography of lemurs in the humid forests of
Madagascar: the role of elevational distribution and rivers. Journal of Biogeography 31 ,
47-55, doi:10.1111/j.1365-2699.2004.00953.x (2004).
31 Yoder, A. D. & Heckman, K. L. Mouse lemur phylogeography revises a model of ecogeographic constraint in Madagascar . (2006).
32 Dewar, R. E. & Richard, A. F. Evolution in the hypervariable environment of
Madagascar. Proceedings of the National Academy of Sciences of the United States of
America 104 , 13723-13727, doi:10.1073/pnas.0704346104 (2007).
33 Wilme, L., Goodman, S. M. & Ganzhorn, J. U. Biogeographic evolution of Madagascar's microendemic biota. Science 312 , 1063-1065, doi:10.1126/science.1122806 (2006).
34 Wollenberg, K. C., Vieites, D. R., Glaw, F. & Vences, M. Speciation in little: the role of range and body size in the diversification of Malagasy mantellid frogs. Bmc Evolutionary
Biology 11 , doi:10.1186/1471-2148-11-217 (2011).
25
634
635
636
637
638
639
640
641
642
643
644
645
646
647
627
628
629
630
631
632
633
648
649
650
651
652
653
654
655
656
657
613
614
615
616
617
618
619
620
621
622
623
624
625
626
35 Wollenberg, K. C.
et al.
Patterns of endemism and species richness in Malagasy cophyline frogs support a key role of mountainous areas for speciation. Evolution; international journal of organic evolution 62 , 1890-1907, doi:10.1111/j.1558-
5646.2008.00420.x (2008).
36 Pabijan, M., Wollenberg, K. C. & Vences, M. Small body size increases the regional differentiation of populations of tropical mantellid frogs (Anura: Mantellidae). Journal of evolutionary biology 25 , 2310-2324, doi:10.1111/j.1420-9101.2012.02613.x (2012).
37 Pearson, R. G. & Raxworthy, C. J. The evolution of local endemism in madagascar: watershed versus climatic gradient hypotheses evaluated by null biogeographic models.
Evolution; international journal of organic evolution 63 , 959-967, doi:10.1111/j.1558-
5646.2008.00596.x (2009).
38 Ferrier, S., Manion, G., Elith, J. & Richardson, K. Using generalized dissimilarity modelling to analyse and predict patterns of beta diversity in regional biodiversity assessment. Diversity and Distributions 13 , 252-264, doi:10.1111/j.1472-
4642.2007.00341.x (2007).
39 Allnutt, T. F.
et al.
A method for quantifying biodiversity loss and its application to a 50year record of deforestation across Madagascar. Conservation Letters 1 , 173-181, doi:10.1111/j.1755-263X.2008.00027.x (2008).
40 Dutilleul, P., Clifford, P., Richardson, S. & Hemon, D. Modifying the t test for assessing the correlation between two spatial processes. Biometrics , 305-314 (1993).
41 Townsend, T. M., Vieites, D. R., Glaw, F. & Vences, M. Testing Species-Level
Diversification Hypotheses in Madagascar: The Case of Microendemic Brookesia Leaf
Chameleons. Systematic Biology 58 , 641-656, doi:10.1093/sysbio/syp073 (2009).
42 Kreft, H. & Jetz, W. Global patterns and determinants of vascular plant diversity.
Proceedings of the National Academy of Sciences of the United States of America 104 ,
5925-5930, doi:10.1073/pnas.0608361104 (2007).
43 Hoeting, J. A. The importance of accounting for spatial and temporal correlation in analyses of ecological data. Ecological Applications 19 , 574-577, doi:10.1890/08-0836.1
(2009).
44 Ohlemuller, R., Walker, S. & Wilson, J. B. Local vs regional factors as determinants of the invasibility of indigenous forest fragments by alien plant species. Oikos 112 , 493-501, doi:10.1111/j.0030-1299.2006.13887.x (2006).
45 Bacaro, G. & Ricotta, C. A spatially explicit measure of beta diversity. Community
Ecology 8 , 41-46 (2007).
46 Bacaro, G.
et al.
Geostatistical modelling of regional bird species richness: exploring environmental proxies for conservation purpose. Biodiversity and Conservation 20 , 1677-
1694 (2011).
47 Diggle, P. J. & Ribeiro, P. J. J. Model-based Geostatistics . (Springer, 2007).
48 Colwell, R. K. & Lees, D. C. The mid-domain effect: geometric constraints on the geography of species richness. Trends in Ecology & Evolution 15 , 70-76, doi:10.1016/s0169-5347(99)01767-x (2000).
49 Raxworthy, C. J. & Nussbaum, R. A. Systematics, speciation and biogeography of the dwarf chameleons (Brookesia; Reptilia, Squamata, Chamaeleontidae) of northern
Madagascar. Journal of Zoology 235 , 525-558 (1995).
50 Angel, F.
Les Lézards de Madagascar
. (Academie Malgache, 1942).
26
679
680
681
682
683
684
685
686
687
688
672
673
674
675
676
677
678
689
690
691
692
693
694
695
696
697
658
659
660
661
662
663
664
665
666
667
668
669
670
671
51
Humbert, H. Les territoires phytogéographiques de Madagascar. Année Biologique 31
,
439–448 (1955).
52 Glaw, F. & Vences, M. Amphibians and Reptiles of Madagascar . (Vences, M. and Glaw
Verlags, F. GbR., 1994).
53 Schatz, G. E. in Diversity and Endemism in Madagascar (eds W.R. Lourenço & S.M.
Goodman) 1–9 (Société de Biogéographie, MNHN, ORSTOM, 2000).
54 Glaw, F. & Vences, M. Field Guide to the Amphibians and Reptiles of Madagascar .
Third Edition edn, (Vences and Glaw Verlag, 2007).
55 Bush, M. B. Amazonian speciation: a necessarily complex model. Journal of
Biogeography 21 , 5-17, doi:10.2307/2845600 (1994).
56 Oneal, E., Otte, D. & Knowles, L. L. Testing for biogeographic mechanisms promoting divergence in Caribbean crickets (genus Amphiacusta). Journal of Biogeography 37 ,
530-540, doi:10.1111/j.1365-2699.2009.02231.x (2010).
57 Phillips, S. J., Anderson, R. P. & Schapire, R. E. Maximum entropy modeling of species geographic distributions. Ecological Modelling 190 , 231-259, doi:10.1016/j.ecolmodel.2005.03.026 (2006).
58 Phillips, S. J.
et al.
Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. Ecological Applications 19 , 181-
197, doi:10.1890/07-2153.1 (2009).
59 Elith, J.
et al.
A statistical explanation of MaxEnt for ecologists. Diversity and
Distributions 17 , 43-57, doi:10.1111/j.1472-4642.2010.00725.x (2011).
60 Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. & Jarvis, A. Very high resolution interpolated climate surfaces for global land areas. International Journal of Climatology
25 , 1965-1978, doi:10.1002/joc.1276 (2005).
61 Moat, J. & Du Puy, D. (ed Kew Royal Botanic Gardens) (1997).
62 Jarvis, A., Reuter, H. I., Nelson, A. & Guevara, E. (CGIAR-CSI SRTM 90m Database,
2008).
63 Williams, P. H. Some properties of rarity scores for site-quality assessment. British
Journal of Entomology and Natural History 13 , 73-86 (2000).
64 IBM SPSS Statistics for Windows v. 19.0 (IBM Corporation, Armonk, NY, 2010).
65 Rangel, T. F., Diniz-Filho, J. A. F. & Bini, L. M. SAM: a comprehensive application for
Spatial Analysis in Macroecology. Ecography 33 , 46-50, doi:10.1111/j.1600-
0587.2009.06299.x (2010).
66 Legendre, P. & Legendre, L. Numerical ecology . Vol. 2nd (Elsevier, 1998).
67 Griffith, D. A. in Practical handbook of spatial statistics (ed S. L. Arlinghaus) 65-82
(CRC Press, 1996).
68 Bini, L. M.
et al.
Coefficient shifts in geographical ecology: an empirical evaluation of spatial and non-spatial regression. Ecography 32 , 193-204, doi:10.1111/j.1600-
0587.2009.05717.x (2009).
698
699
700
27
701
702 Figure 1. Overview of work protocol and dataflow.
Three types of original data were input
703 into the analyses: (1) biogeography hypotheses, (2) geography and climate data and (3) species
704 locality data. These data were used to predict distribution of species, and the distribution models
705 used to calculate biodiversity patterns (species richness, corrected weighted endemism,
706 turnover). We then tested for correlation of these biodiversity patterns with spatial predictions
28
707 derived from biogeography hypotheses, and used a multivariate model to simultaneously test the
708 influences of these hypotheses on the biodiversity patterns. *The response variables constituted
709 standardized principal components of the raw biogeography hypotheses ** The CAR models
710 were iterated until only response variables that contributed significantly the model were
711 included.
712
713
714
715
29
716 Figure 2 . Observed Biodiversity Data.
Reptile and amphibian species richness (A & E)
717 measures the number of species present. Endemism (B & F), based on a corrected weighted
718 calculation, reflects the proportion of unique species present within certain areas. The
719 generalized dissimiliarity model (GDM) analyzes compositional turnover of communities (here
720 jointly for amphibians and reptiles) and predicts dissimilarity throughout the landscape based on
721 an interpolation based on variation in climate and geographic data. The continuous GDM (G) can
722 then be classified into 4 major and 15 minor areas of endemism (H & C).
723
724
30
725
726 Figure 3. Explanatory contribution of continuous biogeography hypotheses to a conditional
727 autoregressive spatial model of each observed biodiversity measurements. Only hypotheses
728 contributing >5% are shown; see Supplementary Table S4 for complete and detailed data).
Mid-
729 domain: I. latitude, II. longitude, III. distance. Climate-Gradient: I. PC1, II. PC2, III. PC3.
730 Climate-Stability: I. Precipitation Stability, II. Climate Stability (temperature and precipitation).
731 Topography: I. Topographic Heterogeneity, II. Disturbance-vicariance, III. Montane Species
732 Pump. An asterisk marks hypotheses that contributed negatively to the mixed CAR model.
733
734
735
736
737
738
31
739
740
741 Figure 4. Contribution of continuous biogeography hypotheses to a conditional
742 autoregressive spatial model of species richness and endemism for four focal groups.
Only
743 hypotheses contributing >5% are shown; see Supplementary Table S4 for complete and detailed
744 data). Mid-domain: I. latitude, II. longitude, III. distance. Climate-Gradient: I. PC1, II. PC2, III.
745 PC3. Climate-Stability: I. Precipitation Stability, II. Climate Stability (temperature and
746 precipitation). Topography: I. Topographic Heterogeneity, II. Disturbance-vicariance, III.
747 Montane Species Pump. An asterisk marks hypotheses that contributed negatively to the mixed
32
748 CAR model. (1) Brookesia chameleons (number of species = 27, number of original distribution
749 points = 178). (2) Boophis treefrogs (n of spp.= 77, n of pts.=460) (3) Phelsuma day geckos (n of
750 spp.= 28, n of pts.= 304) (4) oplurid iguanas (Oplurus plus the monotypic Chalarodon ; n of
751 spp.= 7, n of pts.= 147). Subgroups and colors of pie charts as in Fig. 3. An asterisk marks
752 hypotheses that contributed negatively CAR. For all hypotheses (with exception of the climate-
753 gradient variables) a positive correlation was expected between biodiversity metrics.
754
755
756
757
33
Table 1. Correlations of nominal biodiversity hypotheses to Generalized Dissimilarity Models (upper: 15-class; lower: 4-class transformation of the original model). The left three columns show calculations based on the percentage of overlapping cells between boundaries. High mean values depict high levels of shared boundaries among GDM and AOE (derived from the respective hypothesis). R-values reflect non-spatial Pearson product-moment correlation coefficients. To assess significance of raster data, we used an unbiased correlation following the method of Dutilleul (1993) that reduces the degrees of freedom according to the level of spatial autocorrelation between the two variables.
Hypothesis Correlation to Generalized
Dissimilarity Model
Riverine- Major
Riverine- Minor
Gradient
Percent overlapping cells
(10km buffer)
GDM 15 classes
77.11%
72.43%
74.63%
Hypothesis
35.31%
49.41%
67.51%
Mean
56.21%
60.92%
71.07% r
0.427
0.503
0.403
F-stat
13.984
28.599
24.974 df
62.537
84.415
128.641 p
<.001
<.001
<.001
34
Watershed
GDM -4
Riverine – Refuge
Riverine- Major
Riverine- Minor
Gradient
Watershed
GDM -15
Riverine – Refuge
60.14%
100.00%
71.00%
GDM- 4 classes
25.92%
38.71%
20.69%
43.03%
69.35%
45.85%
0.229
0.551
0.440
5.887
26.257
11.450
Hypothesis Mean Pearson's r F-stat
106.824
60.171
47.693 df
0.017
<.001
0.001 p
41.6%
39.4%
39.6%
33.2%
52.1%
30.3%
40.3% 40.9%
58.1% 48.7%
69.0% 54.3%
33.3% 33.2%
100.0% 76.1%
22.1% 26.2%
0.556 12.519 28.042 0.001
0.51 16.345 46.550 <.001
0.378 13.012 77.926 <.001
0.213 3.775 79.701 0.056
0.551 26.257 60.171 <.001
0.837 52.355 22.433 <.001
35
SUPPLEMENTARY MATERIALS
Hypotheses
Climate Stability Hypothesis (Fig S2.J)
Climate stability is thought to create greater climatic stratification across environmental gradients (Dynesius and Jansson 2000, Jansson and Dynesius 2002). In stable climates, orbitally forced species’ range dynamics (ORD) are low, allowing localized populations to persist, and thus become highly specialized and differentiated. Like the Gradient Hypothesis (following), this model also focuses on bioclimatic disparities, but additionally incorporates climate stability. This hypothesis states that areas of climate stability, particularly those with climatic stratification, should possess higher species richness and endemism.
To estimate this model for Madagascar, coliniarity was measured for all BIOCLIM layers
(Bio1–19) using a Pearson Coefficient. If R
2
values exceeded 0.5, one of the layers was excluded. We preferentially selected layers based on raw data (e.g. selecting mean annual precipitation over seasonality). The following layers were excluded: Bio3, Bio7, Bio9, Bio13–
18. For each remaining BIOCLIM layer we calculated the standard deviation of each cell throughout the four time periods for which climate data were available (0 kya, 6kya, 21 kya, 120 kya). The resulting standard deviation of each BIOCLIM layer was standardized to 1 to account
36
for different units in raw data. All standardized stability layers were summed to create the final climate stability layer; with lower values representing higher climate stability through time.
Disturbance-Vicariance Hypothesis (Fig S2.I)
Under this model, the major factor contributing to diversification was temperature fluctuations
(Colinvaux 1993, Bush 1994, Haffer 1997), rather than fluctuations in precipitation and forest fragmentation (as in the preceding Refuge and River hypotheses). This hypothesis states cyclic fluctuations of temperature during the Quaternary caused reoccurring displacement of taxa towards lower and higher altitudes (during cool and warm periods, respectively). Range retractions of taxa into the highlands occasionally resulted in allopatric divergence into new species. The gradual displacement of temperature specialized taxa would cause reoccurring invasions and counter-invasions of heterogeneous landscapes by both montane and lowland taxa.
This hypothesis predicts species richness and endemism would be highest in areas of topographic heterogeneity and temperature instability.
This hypothesis was generated by measuring colinearity between all BIOCLIM layers corresponding to temperature (Bio1– Bio11, see Climate Stability above for details). The following layers were excluded: Bio3, Bio7 and Bio11. For each remaining BIOCLIM layer we calculated the standard deviation of each cell throughout the four time periods for which climate data were available (0 kya, 6kya, 21 kya, 120 kya). This layer was standardized from 0 to 1 to account for different units between layers. All standardized stability layers were summed to create the final temperature stability layer with lower values representing higher temperature stability. This layer was inverted and multiplied by a standardized version of the final
37
topographic heterogeneity layer (see above). Higher values represented areas with high temperature instability through time and high topographic heterogeneity.
Gradient Hypothesis (Fig S1.E)
According to the Gradient Hypothesis, diversification of Malagasy taxa was driven by bioclimatic disparities throughout the island (i.e. between the east and west), causing parapatry of populations along environmental gradients. This hypothesis was first formulated by Endler
(1982), however more recently it was adapted and formulated in detail for Madagascar by
Vences et al.
(2010). Vences et al.
focused on the climatic stratification longitudinally between the dry west and humid east of Madagascar, terming this species case the Ecogeographic hypothesis. Under the Gradient hypothesis populations adapt to local ecotones, diverging from a generalist ancestor (or one of broader ecological tolerance). Due to local specialization, gene flow within ecological similar sites is higher than those distributed at ecologically different sites across a gradient. The subdivision of populations create a scenario where drift or selection can override gene flow among ecogeographic subpopulations (Fisher 1930) and daughter species occupy separate, adjacent niches. The exact mechanism of speciation is controversial (e.g. prezygotic isolation, behavioral isolation or reproductive barriers) and beyond the focus of this study. This hypothesis invokes no barriers or mechanism of allopatry and predicts species richness and endemism to be highest in areas of high bioclimatic stratification. This GIS prediction was obtained from Pearson & Raxworthy (2009). In our continuous CAR-OBTC analyses, this hypothesis was represented by the first 3 PCs from a PCA on the 19 current
BIOCLIM data for Madagascar (Hijmans et al. 2005).
38
Mid-domain Effect Hypothesis (Fig S2. A-C)
Initially described to explain latitudinal trends in species richness, this mathematical hypothesis demonstrates that if species’ ranges are distributed randomly between northern and southern geographic limits, the highest overlap of species ranges would be in the middle (Lees 1996, Lees and Colwell 2007). For Madagascar, this hypothesis (in two dimensions) results in increased overlap of species toward the center of country, over the Ankaratra highlands, resulting from the sum of random overlapping species ranges. We explore four variants of this hypothesis: 1 dimension (the mid-domain of latitudinal, altitude or longitudinal gradients) and 2 dimensions, as described above (the mid-domain of both latitudinal and longitudinal gradients). Note the middomain of altitude is the same GIS calculation of the Museum hypothesis. Thus, throughout the manuscript this hypothesis is referred to as the Museum hypothesis rather than the mid-domain altitude hypothesis. In one dimension, the latitude hypothesis predicts that species richness would be highest around 18° S, or in two dimensions, centered around 18° S and 46.5° E. The mid-domain hypotheses are included as a null hypothesis for spatial variation in species richness; these hypotheses invoke no barriers or ecological/ habitat heterogeneity and rely solely on the random distribution of species in a defined geographic space.
Montane Species Pump Hypothesis (Fig S2.M)
This hypothesis predicts that montane regions have higher species richness because of their topographic complexity and climatic zonation - both increase opportunities for allopatric and parapatric speciation (e.g., Moritz et al. 2000; Rahbek and Graves 2001; Hall 2005; Fjeldsa° and Rahbek 2006; Kozak and Wiens 2007). This hypothesis predicts that speciation should be highest in areas of high topographic and climatic heterogeneity. Within those habitats, rates of
39
speciation should be highest in mid-elevations. This hypothesis predicts species richness and endemism would be highest in areas of topographic and climatic heterogeneity.
To create a spatially explicit model of this hypothesis we calculated the mean standard deviation of the first three climate PCs (based on Bioclim current data) at ca. 10 km
2
square neighborhood of each cell. Each was summed together and then the product was standardized from 0-1. The resulting layer was then multiplied by a standardized (0-1) topographic variation hypothesis. High values depict areas of high topographic and climatic heterogeneity.
Museum Hypothesis (Fig S2.D)
According to this hypothesis speciation occurred in montane habitats. In the Montane Museum hypothesis, more species exist at intermediate elevation because these elevations were simply occupied the longest, because of this, there has been more time for speciation and the accumulation of species in these habitats when compared to habitats at lower and higher elevations (Stebbins 1974; Stephens and Wiens 2003). This hypothesis predicts levels of endemism and richness will be highest in the middle elevations. Note that in execution, the prediction for this hypothesis is identical to a Mid-domain hypothesis of altitude. The GIS representation of this hypothesis for Madagascar represents the median elevation (378m), where the overlap of species should peak. This elevation was given a value of one and from this elevation, values linearly transitioned to zero at the maximum and minimum elevations.
Paleogeographic Hypothesis
This hypothesis states that vicariate differentiation of Malagasy lineages is associated with formation of geologic barriers to dispersal. Each hypothesis is specific to the focal
40
paleogeographic event. There are two major classes of geological events that have occurred since the separation from India (ca. 65 MYA). The first, directly a result from separation from India, is the orogeny of eastern escarpment. This hypothesis states that deep lineages, pre-65Mya, should exist between the east and western species. A second, more local barrier is the volcanisms of several regions: Tsaratanana, Manongarivo, Ampasindava (ca. 50 Mya), Ankaratra (phase 1 ca.
28 Mya, phase 2 ca.15 Mya, phase 3 ca. 2 Mya) and Ambre (ca. 2 Mya, Krause 2007). Clades should exhibit breaks around volcanic activity, though given localization of these barriers; biota in most cases should have been able to dispersal around, though perhaps experiencing reduced gene flow. All paleogeographic hypotheses are difficult to test because more strongly than other hypotheses, they are dependent on the location of ancestral populations and the timing of geological events relative to a clade's diversification. Given the variation of ages of origins in
Madagascar's vertebrates, it is likely that some clades were affected by most of the major paleogeographical events while others were only affected by some. Because of these factors, it is unlikely that species exhibit congruent biogeographic patterns. Thus in this paper, we were unable to test this hypothesis.
Refuge Hypothesis (Fig S2.D) This hypothesis holds that episodic fragmentation of forests resulted in isolated patches of wet forest and this caused vicariant differentiation between adjacent patches. These transitions were driven by periodic changes (every 20-100 Ky) in insolation associated with Milankovitch cycles (aberrations of the orbit of the earth around the sun due to the slight asymmetrical shape of the earth). Recurrent changes in insolation caused various dry periods followed by humid periods, particularly pronounced at tropical latitudes. The evolutionary consequences of paleoecological changes in climate likely depend on the regional
41
topography, predominant weather patterns and the impact on the ecosystem (for example, a slight reduction in rain may have different biological consequences in a spiny forest versus rainforest). In Amazonia during the late Tertiary and Quaternary during dry climatic periods, it has been argued that extensive humid forests survived due to subtle topographic variation that facilitated rainfall gradients adjacent to the Andes, Guianan highlands, Rondonia and hilly areas east of Pará (Haffer 1969; Vanzolinii1970, 1973; Brown et al., 1974; Prance, 1982, 1996).
In addition to changes in insolation associated with Milankovitch cycles, the gradual drifting of Madagascar northward towards the equator likely led to habitat changes, presenting another source of refugial isolation. Thus, it is plausible that refugia persisted and species with narrow ecological niches became isolated and diverged into separate species. To create a continuous spatial representation of this for Madagascar, first colinearity was measured for each
BIOCLIM precipitation layer (Bio12 – Bio19) using a Pearson Correlation. If R 2 values exceeded 0.5, one of the layers was excluded. This resulted in the exclusion of Bio13, Bio15,
Bio16 and Bio17. We preferentially selected layers based on raw data (e.g. selecting mean precipitation over seasonality). For each remaining BIOCLIM layer, we calculated the standard deviation of each cell throughout the four time periods for which climate data were available (0 kya, 6kya, 21 kya, 120 kya). This layer was standardized from 0 to 1 to account for different units/decimal places in raw data between layers. All standardized stability layers were summed to create the final precipitation stability layer with lower values representing higher precipitation stability.
Riverine Barrier Hypothesis (Fig S1. A,B).
42
Malagasy rivers are thought to have acted as barriers separating populations, resulting in intrariverine species and subspecies. As the geology of Madagascar has remained relatively stable since absolute isolation major rivers should have persisted (at least seasonally). Under this hypothesis, we expect vicariate differentiation of Malagasy lineages associated with large tributaries. There have been several criticisms of this hypothesis, such that rivers frequently change course, causing land and its inhabitants to passively transfer across the barrier, rivers cease to act as barriers due to the lack geographic separation at headwaters (Wallace 1852) or temporal fluctuations of climate causing rivers to change in size (e.g. the sizes of any rivers were dramatically reduced during the Pleistocene).
To create a spatial prediction for this hypothesis, we selected all major permanent rivers that have headwaters above 1000m and created polygons from the lowland areas between major rivers and 1000m contour line. A second calculation focused on major riverine units composed of areas between rivers with headwaters above 2000m and created biogeographic units from lowland areas between rivers and 1000m contour line.
River-Refuge Hypothesis (FigS1.C)
The River-Refuge hypothesis, initially described by Haffer (1992, 1993) for the Neotropics, was more recently proposed as a model of Malagasy diversification (Craul et al.
2007). This hypothesis combines aspects of the Riverine Barrier hypothesis and the Refuge hypothesis under which it states that lowland vicariant speciation occurs in refugia separated by broad lowland rivers and by considerable unsuitable terrain in the headwaters.
To estimate this model, we combined the Riverine and a binary refuge hypothesis. The continuous refuge layer was converted to a binary model by converting the top quartile to 1 and
43
all other values to 0. To account for differences between precipitation stability in arid areas
(versus wet), we excluded areas with less than 50cm in any of the time periods. Areas above
1000m were also excluded. Regions of the Riverine layer and binary refuge layer were then combined into smaller river-refuge subunits.
Sanctuary Hypothesis (Fig S2.L)
Past climate changes greatly altered the distributions of organisms through time; causing local extinctions, bottlenecks, isolation, range expansion and contraction of populations. Sanctuaries represent specific areas of habitat stability that have remained present through time, differing from refugia (which do not invoke geospatial consistency) and species track suitable habitat across geography (Recuero & García-París 2011). We estimated sanctuaries in Madagascar by calculating SDMs for 453 species (a subset of the GDM dataset, using species with three or more unique occurrence localities). The SDM was projected into three historic time periods: LGM,
120KYA, 6 KYA. All SDMs were converted to binary models using the maximum training sensitivity plus specificity threshold. The maximum training sensitivity plus specificity threshold: 0.080 (SD +-0.047), area: 0.316 (SD +-0.196), training omission: 0.005 (SD +- 0.017), number of training samples (mean: 12.321, max 137, min 3, SD 14.414). For each species, all four binary SDMs were summed. The resulting layer was reclassified so that values of 3 and below were converted to zero and values of 4 were converted to 1. Under this classification, areas where the species was present for all four time periods were considered ‘sanctuaries’. This was repeated for all species and all sanctuaries were summed to estimate areas of higher species richness and endemism.
44
Topographic Heterogeneity (Fig S2.E)
In several studies, the level of topographic variation has been observed to be positively correlated to species richness patterns and centers of endemism (Kerr & Packer 1997; Rahbek and Graves 20001; Jetz & Rahbek 2002; Jetz et al. 2004). To characterize this hypothesis for
Madagascar we measured the standard deviation of elevation at ca. 10 km
2
of each pixel.
Watershed Hypothesis (Fig S1.D)
One of the more recent diversification hypotheses is the Watershed hypothesis (Wilmé et al.
2006). According to this model, climatic changes caused retraction of forests to the surrounding major rivers. If the headwaters of adjacent rivers were at lower elevations, the intervening areas between rivers (the watersheds) become arid and forests populations became isolated, serving as areas of speciation. By contrast, if the headwaters of rivers were higher elevations, the watershed served as areas of retreat and forest refugia remained connected among rivers. These watersheds are expected to contain proportionally much lower diversity and endemism. This GIS prediction was obtained from Wilmé et al. (2006).
Additional references.
Brown Jr., K.S., Sheppard, P.M., Turner, J.R.G. 1974. Quaternary refugia in tropical America: evidence from race formation in Heliconius butterflies. Proc. R. Soc. Lond. B. 187: 369-378.
Bush, M.B. 1994. Amazonian speciation: a necessarily complex model. Journal of Biogeography 21::5–17.
Colinvaux, P.A. 1993. Pleistocene biogeography and diversity in tropical forests of South America P. Goldblatt
(Ed.), Biological Relationships between Africa and South America, Yale University Press, New Haven, CT.
Craul, M., Zimmermann, E., Rasoloharijaona, S., Randrianambinina, B., Radespiel, U. 2007. Unexpected species diversity of Malagasy primates ( Lepilemur spp.) in the same biogeographical zone: a morphological and molecular approach with the description of two new species. BMC Evolutionary Biology 7:83.
45
Dynesius, M., Jansson, R. 2000. Evolutionary consequences of changes in species_ geographical distributions driven by Milankovitch climate oscillations. Proc. Natl Acad. Sci. U.S.A. 97:9115–9120.
Endler, J. 1982. Pleistocene forest refuges: fact or fancy? In Prance, G.T. (Ed.). Biological Diversification in the
Tropics. New York: Columbia University Press, p. 179-200.
Fisher, R.A. 1930. The Genetical Theory of Natural Selection. Clarendon Press.
Goodman, S. M. & Benstead, J. P. The natural history of Madagascar. (University of Chicago Press Chicago, 2003).
Fjeldsa°, J., Rahbek, C. 2006. Diversification of tanagers, a species-rich bird group, from the lowlands to montane regions of South America. Integr. Comp. Biol. 46:72–81. (doi:10.1093/icb/icj009)
Haffer, J. 1969. S peciation in Amazonian forest birds. Science 165: 131-137.
Haffer, J. 1997.Alternative models of vertebrate speciation in Amazonia: an overview Biodiversity and
Conservation, 6:451–476.
Haffer,, J. 1992. On the “river effect” in some forest birds of southern Amazonia. Bol. Mus. Para. Emilio Goeldi, sér. Zool. 8:217-245.
Haffer, J. 1993. Time’s cycle and time’s arrow in the history of Amazonia. Biogeographica 69:15-45.
Hall, J. P. 2005. Montane speciation patterns in Ithomiola butterflies (Lepidoptera: Rhiodinidae): are they consistently moving up in the world? Proc. R. Soc. B 272:2457–2466. (doi:10.1098/rspb.2005.3254)
Jansson, R., Dynesius, M. 2002. The fate of clades in a world of recurrent climatic change: Milankovitch oscillations and evolution. Annu. Rev. Ecol. Syst. 33:741–777.
Jetz, W., Rahbek, C. 2002. Geographic range size and determinants of avian species richness. Science 297:1548–
1551.
Jetz, W., Rahbek, C., Colwell, R.C. 2004. The coincidence of rarity and richness and the potential signature of history in centers of endemism. Ecol. Lett. 7:1180–1191.
Kerr, J.T. Packer, L. 1997. Habitat heterogeneity determines mammalian species richness in high energy environments. Nature. 385:252-254..
Kozak, K.H., Wiens, J.J. 2007. Climatic zonation drives latitudinal variation in speciation mechanisms. Proc. R.
Soc. B 274:2995-3003. doi: 10.1098/rspb.2007.1106
46
Krause, D. W. 2003. Late Cretaceous vertebrates of Madagascar: A window into Gondwanan biogeography at the end of the Age of Dinosaurs. Pp. 40-47 in S. M. Goodman and J. P. Benstead (eds.), The Natural History of
Madagascar. University of Chicago Press, Chicago.
Lees, D. C. 1996. The Périnet effect? Diversity gradients in an adaptive radiation of butterflies in Madagascar
(Satyrinae: Mycalesina ) compared with other rainforest taxa, Pages 479-490 in W. R. Lourenço, ed.
Biogéographie de Madagascar. Paris, Editions de l'ORSTOM.
Lees, D. C., Colwell R.K. 2007. A strong Madagascan rainforest MDE and no equatorward increase in species richness: Re-analysis of 'The missing Madagascan mid-domain effect', by Kerr J.T., Perring M. & Currie D.J
(Ecology Letters 9:149-159, 2006). Ecology Letters 10:E4-E8.
Moritz, C., Patton, J. L., Schneider, C. J., Smith, T. B. 2000. Diversification of rainforest faunas: an integrated molecular approach. Annu. Rev. Ecol. Syst. 31:533–563. (doi:10. 1146/annurev.ecolsys.31.1.533)
Pearson, R.G., Raxworthy C.J. 2009. The evolution of local endemism in Madagascar: watershed versus climatic gradient hypotheses evaluated by null biogeographic models. Evolution 63:959–967.
Prance, G.T. 1982. Biological diversification in the tropics. New York: Columbia University.
Prance, G.T. 1996. Islands in Amazonia. Phil. Trans. R. Soc. Lond. B 351: 823-833.
Rahbek, C. 1997. The relationship among area, elevation, and regional species richness in Neotropical birds. Am.
Nat. 149:875–902.
Rahbek, C., Graves, G. R. 2001. Multiscale assessment of patterns of avian species richness. Proc. Natl Acad. Sci.
USA 98:4534–4539. (doi:10.1073/pnas.071034898)
Recuero, E., García-París, M. 2011. Evolutionary history of Lissotriton helveticus : Multilocus assessment of ancestral vs. recent colonization of the Iberian Peninsula. Molecular Phylogenetics and Evolution 60: 170-
182.
Stebbins, G. L. 1974. Flowering plants: evolution above the species level. Harvard University Press, Cambridge,
Mass.
Stephens, P. R., Wiens, J.J. 2003. Explaining species richness from continents to communities: the time-forspeciation effect in emydid turtles. American Naturalist 161:112–128.
Vanzolini, PE. 1970. Zoología sistemática, geografía e a origem das espécies. São Paulo: Instituto Geográfico de
São Paulo. 56 p
47
Vanzolini, P.E. 1973. Paleoclimates, relief, and species multiplication in equatorial forests. In Meggers, B.J.,
Ayensu E.S.. Duckworth, W.D. (Eds.). Tropical forest ecosystems in Africa and South America: A comparative review. Washington: Smithsonian Institution. p. 255-258.
Wallace, A.R. 1852. On the monkeys of the Amazon. London. Proc. Zool. Soc., 1852:107-110.
Wilmé, L., Goodman, S.M., Ganzhorn, J.U. 2006. Biogeographic evolution of Madagascars microendemic biota.
Science 312:1063– 1065.
48
Supplementary Figure S1. Nominal Biogeography hypotheses.
These hypotheses are comprised of non-order categories from which relationships among contained data are not known
49
or non-linear. Four hypotheses fell into this category: The Riverine (major and minor), River-
Refuge, Watershed and Gradient. See table S1 for summary of each hypotheses.
50
51
Supplementary Figure S2. Continuous Biogeography hypotheses.
Nine hypotheses fell into this category: Mid-domain (longitude, latitude and distance), Museum, Topographic
Heterogeneity, Gradient* (PC1-PC3), Disturbance-vicariance, Climate Stability, Precipitation
Stability, Sanctuary and Montane Species Pump (see table S1 for summary of each hypotheses).
Inlayed tables represent the percent contribution of each corresponding hypothesis in the CAR model with the lowest AICc of each observed biodiversity measurements. *The values of the three climate principal components are not necessary assumed to reflect a positive correlation to endemism and species richness, however, we are assuming each reflects a prediction of a linear correlation (either positive or negative). The inclusion of the three climate principal components is the result of the power CAR models and the ability to include multiple explanatory variables.
Due to the statistical limitations associated with nominal hypotheses, if a hypothesis could be depicted by continuous data (even if it required several variables) they were converted to this format and incorporated into the CAR. For example, if nominal data represented classified continuous data, we include the continuous data (such as the 3 PCs of climate data).
52
Supplementary Table S1. Major Biogeographic Predictions Relevant to Madagascar.
Hypothesis
Climate Stability Climate stability creates
(Fig S2.J) greater climatic stratification across environmental gradients
Disturbancevicariance
(Fig S2.L)
(Fig S1.E)
Gradient
Description
Allopatry results from altitudinal range retractions caused by temperature fluctuations.
Parapatry of populations due to environmental gradients
Key Factors
Stable climate; both seasonally and through geologic time; no barrier is necessary
Effects
In stable climates, orbitally forced species’ range dynamics
(ORD) are low, allowing localized populations to persist, and thus become highly specialized and differentiated.
Divergent selection and an environmental
Parapatric speciation across climate space
Predictions for Reptiles and Amphibians
Higher levels of endemism in areas of climatic stability
GIS Prediction
Use GIS and spatiotemporal explicit climate data to estimate climate stability; stable areas should harbor higher species richness and endemism.
Temperature fluctuations, changes in CO
2 and habitat heterogeneity
(usually associated with changes in altitude)
Decreased temperatures allow cool adapted species to disperse
Diversity with south. Cyclic fluctuations in temperature cause populations to region. Most common habitat track attitudinally, when associated with a single ancestor of sister clades
Estimate areas of high monophyletic lineages are temperate fluctuations adjacent areas of slope; higher values reflect higher species richness temperatures are at their highest, date to Pleistocene. populations become isolated on sky islands
Sister taxa are found in different habitats along an current climate data to environmental gradient
Cluster analyses of all estimate areas of endemism.
Temporal
Scope
None
Key Citations
Dynesius & Jansson,
2000; 2002
Quaternary Colinvaux 1993,
Bush 1994, Haffer
1997; Raxworthy
Nussbaum 1995.
None Endler 1982
53
Mid-domain
Effect
(Fig S2. A-C )
Montane Species
Pump
(Fig S2. M) gradient; no barrier is necessary
Species’ ranges are distributed randomly between
Geographic space northern and southern geographic limits, the highest overlap of species ranges would be in the middle.
Topographic complexity and climatic zonation of mountains increase opportunities for allopatric and parapatric speciation
Topographic and climatic heterogeneity
No mechanism invoked
Allopatric and parapatric speciation across elevations
Species richness should be Richness is highest in the None highest in mid-domains mid-domain of latitude, longitude and elevation.
Sister taxa share common ancestry with montane ancestor. Extant montane species display higher levels of intraspecific genetic variation.
Estimate areas of topographic and climatic
None heterogeneity- high values reflect centers of high endemism and species richness
Museum
( Fig S2.D)
More species occur at intermediate elevations simply because these elevations were occupied the longest and there has been more time for speciation and
Extended time occupying in midelevations
Increased differentiation at midelevations
More species occur at intermediate elevations
Use GIS to calculate the median elevation of because these elevations Madagascar which were occupied the longest should possess the highest species richness and endemism
None
Lees 1996, Lees and
Colwell 2007
Moritz et al. 2000;
Rahbek and Graves
2001; Hall 2005;
Fjeldsa° and Rahbek
2006; Kozak and
Wiens 2007, Roy et al 1997, Fjeldsa et al
1999
Stephens and Wiens
2003
54
the accumulation of species in these habitats relative to those at lower and higher elevations
Paleogeographical Vicariant differentiation of
Malagasy lineages is associated with formation of geologic barriers to dispersal.
Each hypothesis is specific to the focal paleogeographic event.
Geological changes Vicariant differentiation across resulting in vicariant barriers events
Distinct east and west lineages associated with the central mountains, and between the southern/central/northern massifs, tsingys.
Not Calculated Specific to each geological event
Refuge
(Fig S2. K)
Allopatry due to retraction of wet habitats
Riverine (Fig Allopatry due to rivers acting
S1. A, B) as barriers.
Repeated cycles of drastic fluctuation in forests resulting in isolated precipitation
Episodic fragmentation of patches of wet forest causing
Evolutionary lineages associated with refugia vicariant differentiation between precipitation relative to adjacent patches
(areas of continued regional mosaic of habitats)
Permanent large rivers
Vicariant differentiation of
Malagasy lineages associated with large tributaries
Use GIS and spatiotemporal explicit
Reciprocal monophyly of clades on opposite sides of areas which are areas of river
Measured inter-riverine endemism
Cenozoic
(Tertiary climate data to estimate and stable wet habitats, these Quaternary
) areas reflect centers of endemism
None
Goodman &
Benstead 2003
Haffer 1969m 1990,
1999, Endler (1982),
Brown (1987) Nores
(1999)
Wallace 1853,
Capparella 1991,
Patton et al.
1994,
55
River-Refuge
(Fig S1.C)
Sanctuary
(Fig S2.L)
Topographic
Heterogeneity
(Fig S1. D)
Allopatry due the restriction of wet habitats to lower elevations; higher elevation habitats and intervening rivers act as barriers
Extinction occurs more often in instable habitats; thus stable areas accumulate species.
Reduced precipitation, maintenance of permanent rivers
Similar to Riverine Hypothesis; fragmental faunal distributions into intra-riverine corridors, isolation is associated with increased aridity adversely affecting habitat suitability at headwater regions
Climate fluctuations Areas of niche stability through time across accumulate species through time. heterogeneous landscapes
The level of topographic variation has been observed to be positively correlated to species richness patterns and centers of endemism
Topography No mechanism invoked
Sister taxa are found in adjacent intra-riverine corridors
Combine the Riverine
Hypothesis subunits and a binary precipitation stability /low elevation layer. Resulting areas depict areas of high predicted endemism.
Late tertiary
(Post-
Miocene)
Areas of niche stability
(specific aspects of
Use GIS and SDM to estimate stable areas in climate vs. overall climate each species distribution
None in Climate stability) provide sanctuary for species though time. through time; areas of highest stability should harbor higher species richness and endemism
Species richness should be Areas of high highest in areas of high topographic heterogenety topographic heterogeneity harbor higher species richness
None
Goodman &
Ganzhorn, 2004
Haffer 1992; 1993,
Craul et al 2007
Recuero & García-
Paris 2011
Kerr & Packer 1997;
Rahbek and Graves
20001; Jetz &
Rahbek 2002; Jetz et al. 2004
56
Watershed
(Fig S1.D)
Allopatry results from altitudinal range retractions caused by simultaneous decreases in temperature and
Repeated cycles of Dispersal during warm-humid drastic simultaneous periods allows lowland species increases of both temperature and to disperse attitudinally, across headwater habitat (previously precipitation. Lower elevation precipitation. Large too arid to occupy). Allopatry rivers act as barriers. permanent rivers and mountains occurs as climate cools and the species depends into lowlands adjacent to lowlands. and lowland rivers prevent gene flow between populations, now occurring on both sides of river.
Sister taxa are found in adjacent intra-riverine corridors. Most common ancestor of sister clades date to Pleistocene.
See Wilmé et al. 2006.
Quaternary Wilmé et al 2006
57
Supplementary Table S3.
Correlations of biodiversity hypotheses to observed biodiversity patterns. R-values reflect non-spatial
Pearson product-moment correlation coefficients. To assess significance of raster data, we used an unbiased correlation following the method of Dutilleul (1993). This method reduced the degrees of freedom according to the level of spatial autocorrelation between the two variables.
Hypothesis
Correlation to Observed Endemism R F-stat
Reptile df p r
Amphibian
F-stat df p
Mid-domain: Distance
Topographic Heterogeneity
Refuge
Montane Species Pump
Disturbance-vicariance
Climate stability
Sanctuary
Museum
-0.148
0.343
0.036
0.303
0.379
0.241
0.296
0.285
0.539
5.425
0.04
2.639
4.101
0.62
2.438
4.971
24.037
40.738
32.328
48.488
22.413
14.580
25.313
56.429
0.470
0.025*
0.848
0.107
0.055
0.444
0.131
0.030*
-0.026
0.662
0.154
0.613
0.616
0.356
0.606
0.335
0.018
21.664
0.552
13.852
15.758
2.033
7.611
17.709
26.753
27.724
22.722
22.990
22.829
15.124
13.080
49.836
0.894
<.001*
0.465
0.001*
<.001*
0.174
0.016*
0.015*
58
River-Refuge (binary)
Correlation to Observed Richness
Mid-domain: Distance
Topographic Heterogeneity
Refuge
Montane Species Pump
Disturbance-vicariance
Climate stability
Sanctuary
Museum
River-Refuge (binary)
0.207 3.87
R F-stat
-0.480 10.265
0.140 2.398
-0.101 0.332
0.056 0.507
0.019 2.297
0.233 1.506
0.419 10.079
0.234 4.037
0.108 2.378
86.869 0.052 0.066 0.954 216.959 0.330 df p r F-stat df p
34.353 0.003* -0.204 1.072 24.698 0.310
119.852 0.124 0.307 6.769 64.952 0.011*
31.948 0.598 -0.093 0.216 24.605 0.646
160.576 0.477 0.296 5.506 57.196 0.022*
62.251 0.135 0.303 5.286 41.554 0.027*
26.191 0.231 0.210 0.723 15.672 0.408
47.267 0.003* 0.818 34.069 16.885 <.001*
69.472 0.048* 0.250 4.481 66.989 0.038*
201.187 0.125 0.234 4.94 85.022 0.029*
59
Supplementary Table S4. Mixed CAR spatial models of observed biodiversity data. A principal component analyses was performed on the standardized biogeography hypotheses. All the resulting principal components (PCs) were extracted and then loaded as explanatory variables. The CAR analyses were run iteratively, starting with all PCs as response variables and then excluding each PC that did not contribute significantly to the model (α = 0.05) until the final model included only PCs that contributed significantly to the model. The standardized beta coefficients ( β ) were then used to calculate contributions of each biogeography hypothesis (see methods on OTBCs) in the final CAR analysis. To compare the contributions of each biogeography hypothesis among models of observed biodiversity patterns (richness, endemism, GDM), β coefficents from each OTBC/CAR analyses were converted to the percentage of contribution. *The mean of the 3 MDS vectors loadings were calculated and contributed as a single value to the total mean.
Percent Contribution to CAR
Mid-domain- Latitude
Mid-domain- Longitude
Mid-domain- Distance***
Endemism
Amphibian
0.0%
Reptile
27.7%
Richness
Amphibian
3.5%
Reptile
3.6%
MDS-D1
5.3%
GDM
MDS-D2
20.7%
MDS-D3
1.6%
Mean*
9.2%
Mean contribution to all observed biodiversity models*
8.8%
8.1%
15.1%
8.0%
9.7%
6.7%
7.7%
9.2%
20.0%
10.9%
14.4%
10.2% 8.1% 9.7%
5.7% 15.8% 11.9%
8.4%
12.9%
60
Climate- PC1
Climate- PC2
Climate- PC3
Refuge
Climate Stability***
Topographic Heterogeneity
Disturbance-vicariance
Montane Species Pump
Sanctuary
Museum
CAR Model Summary
Explained by Predictor
Variables: r
2
(AICc)
Total Explained (Predictor and
Space): r 2 (AICc)
15.4%
10.1%
2.2%
1.4%
4.7%
6.5%
5.9%
5.0%
10.9%
14.8%
0.482
(-372.3)
0.646
( -426.0
)
0.1%
12.7%
12.8%
4.7%
4.6%
1.7%
4.8%
0.0%
0.3%
11.4%
8.5%
4.6%
6.9%
7.4%
7.1%
8.1%
7.1%
13.7%
14.5%
11.0%
1.8%
0.0%
3.0%
3.7%
3.1%
2.3%
12.6%
15.1%
10.1%
4.4%
3.8%
0.0%
3.9%
3.4%
2.9%
11.7%
4.5% 13.8% 11.1%
11.7%
11.9%
10.5%
1.0%
0.5%
2.2%
0.0%
3.1%
7.4%
3.9%
0.0%
3.8%
8.0%
6.8%
7.2%
4.7%
9.7%
6.7%
4.8%
1.6%
4.2%
4.1%
3.4%
6.5%
13.0%
0.303
(-468.5)
0.626
(-556.4)
7.1% 15.1% 14.3% 18.1% 19.0% 17.1%
0.624
(7861.9)
0.309 0.869 0.435 0.659
(21849.9) (-4037.8) (-1084.2) (-3327.7)
0.849
(6349.3)
0.745 0.954 0.942 0.868
(19364.2) (-6659.7) (-6760.4) (-5700.9)
10.5%
10.4%
5.6%
3.5%
4.2%
4.6%
5.2%
3.6%
8.8%
13.4%
61
Model significance: n, F, p-val 141, 42.5, 141, 9.7,
<0.001 <0.001
2501,
303.5
<0.001
2501,
278.1,
<0.001
2501,
1265.9,
<0.001
2501,
147.2,
<0.001
2501,
396.8,
<0.001
NA
62
Supplementary Table S5. Mixed CAR spatial models of focal subgroups. A principal component analyses was performed on the standardized biogeography hypotheses. All the resulting principal components (PCs) were extracted and then loaded as explanatory variables. The CAR analyses were run iteratively, starting with all PCs as response variables and then excluding each PC that did not contribute significantly to the model (α = 0.05) until the final model included only PCs that contributed significantly to the model. The standardized beta coefficients ( β ) were then used to calculate contributions of each biogeography hypothesis (see methods on OTBCs) in the final CAR analysis. To compare the contributions of each biogeography hypothesis among models of observed biodiversity patterns (richness, endemism, GDM), β coefficents from each OTBC/CAR analyses were converted to the percentage of contribution.
*The mean of the 3 MDS vectors loadings were calculated and contributed as a single value to the total mean.
Percent Contribution to CAR Endemism
Boophis Brookesia Oplurus* Phelsuma Boophis
Richness
Brookesia Oplurus* Phelsuma
Mid-domain- Latitude
Climate- PC1
Climate- PC2
Climate- PC3
Refuge
Mid-domain- Longitude
Mid-domain- Distance***
3.0%
8.4%
16.0%
12.2%
6.0%
5.2%
0.0%
4.0%
9.9%
9.9%
12.2%
14.1%
4.3%
10.5%
2.5%
10.6%
6.3%
13.5%
13.2%
2.8%
11.3%
17.0%
11.8%
6.4%
5.8%
7.0%
18.3%
2.3%
0.0%
6.1%
10.6%
14.4%
13.0%
1.6%
9.9%
5.1%
8.3%
12.0%
11.7%
11.2%
7.9%
4.5%
2.3%
6.3%
11.0%
15.9%
8.9%
0.0%
3.3%
4.2%
10.2%
20.4%
13.1%
9.7%
3.9%
0.0%
63
Climate Stability***
Sanctuary
Museum
Topographic Heterogeneity
Disturbance-vicariance
Montane Species Pump
CAR Model Summary
3.6%
9.0%
7.0%
8.4%
0.1%
21.1%
8.8%
2.2%
4.7%
0.6%
18.9%
0.0%
11.2%
1.0%
4.1%
0.0%
22.6%
0.8%
5.9%
5.1%
7.2%
4.3%
0.0%
8.8%
Explained by Predictor
Variables: r 2 (AICc)
Total Explained (Predictor and
Space): r
2
(AICc)
Model significance: n, F, p-val
0.251
(-252.2)
0.524
( -316.0
)
0.303
(-176.1)
0.685
(-235.3)
0.432
(-317.9)
0.756
(-436.8)
0.209
(-253.4)
0.759
(-354.6)
0.569
(9710.5)
0.863
(6851.0)
141, 46.6, 141, 49.7, 141, 20.5,
<0.001 <0.001 <0.001
141, 20.1,
<0.001
2501, 364.7,
<0.001
8.9%
2.8%
5.8%
2.7%
20.3%
3.8%
0.444
(5392.1)
0.812
(2680.8)
2501, 220.6,
<0.001
3.6%
2.1%
2.5%
0.0%
17.2%
13.9%
5.8%
6.5%
6.8%
4.5%
23.6%
5.2%
0.178
(9789.1)
0.504
(8530.6)
0.559
(6731.4)
0.794
(4837.3)
2501, 550.1, 2501, 632.4,
<0.001 <0.001
1.8%
4.1%
2.7%
2.7%
7.9%
19.3%
64
