Trends in Ecology and Evolution Niche estimation above and below the species level --Manuscript Draft-Manuscript Number:
TREE-D-18-00183R1
Article Type:
Review
Keywords:
ecological niche model; species distribution model; multi-response model; phylogenetic niche conservatism; local adaptation; partial pooling model
Corresponding Author:
Adam B. Smith, Ph.D. Missouri Botanical Garden Saint Louis, MO UNITED STATES
First Author:
Adam B. Smith, Ph.D.
Order of Authors:
Adam B. Smith, Ph.D. William Godsoe, Ph.D. Francisco Rodríguez-Sánchez, Ph.D. Hsiao-Hsuan Wang, Ph.D. Dan Warren, Ph.D.
Abstract:
Ecological niches reflect not only adaptation to local circumstances but also the tendency of related lineages to share environmental tolerances. As a result, information on phylogenetic relationships has underappreciated potential for informing ecological niche modeling. Here we review three strategies for incorporating evolutionary information into niche models: splitting lineages into subunits, lumping across lineages, and partial pooling lineages into a common statistical framework that implicitly or explicitly accounts for evolutionary relationships. We challenge the default practice of modeling at the species level, which ignores the process of niche evolution and erroneously assumes that the species is always the appropriate level for niche estimation. Progress in the field requires re-examining how we assess models of niches versus models of distributions.
Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation
Cover Letter
26 October 2018 Dear Drs. Craze and Lim, Please find enclosed a revised version of our manuscript entitled “Niche estimation above and below the species level” for consideration as a review article in Trends in Ecology and Evolution. We thank you and the three reviewers for the time and attention to helping us improve our submission. We have made changes to the manuscript in response to these concerns as detailed in the Response to Reviewers. Our primary changes involved rewording, minor re-organization, and including new references required by the reviewers. In response to Reviewer #3, we also now include a new Box 1 to disentangle the varied terms given to the types of models we discuss. This work has been prepared exclusively for Trends in Ecology and Evolution and is not being considered for publication elsewhere. It represents the original work of all the authors. Thank you again,
Adam B. Smith, Ph.D. Adam Smith, Ph.D. Assistant Scientist in Global Change Missouri Botanical Garden Saint Louis, MO 63110 314-577-9473 ext. 6314
[email protected]
P.O. Box 299 • St. Louis, MO • 63166-0299 • (314) 577-0871 • fax (314) 577-9596 • www.mobot.org
Manuscript, Glossary, Figure Caption, and Boxes
Click here to access/download;Manuscript;Main Text, Glossary, and Boxes.docx
1
Title
2
Niche estimation above and below the species level
3 4
Authors
5
Adam B. Smith*, Center for Conservation and Sustainable Development, Missouri Botanical
6
Garden, 4344 Shaw Boulevard, Saint Louis, MO 63116, USA
7
William Godsoe, BioProtection Research Centre, Burns Building Lincoln University,
8
Ellesmere Junction Road/Springs Road Lincoln University, Lincoln New Zealand
9
Francisco Rodríguez-Sánchez, Department of Integrative Ecology, Estación Biológica de
10
Doñana, Consejo Superior de Investigaciones Científicas, Avda. Américo Vespucio 26,
11
41092 Sevilla, Spain
12
Hsiao-Hsuan Wang, Department of Wildlife and Fisheries Sciences, Texas A&M University,
13
534 John Kimbrough Blvd., Building 1537, 2258 TAMU, College Station, Texas 77843,
14
USA
15
Dan Warren, Senckenberg Biodiversity and Climate Research Center (SBiK-F), Frankfurt
16
am Main, Germany, and Biodiversity and Biocomplexity Unit, Okinawa Institute of
17
Science and Technology, Japan
18
* Correspondence:
[email protected] (A.B. Smith)
19
Keywords
20
ecological niche model
21
species distribution model
22
multi-response model
23
phylogenetic niche conservatism 1
24
local adaptation
25 26
Abstract
27
Ecological niches reflect not only adaptation to local circumstances but also the tendency of
28
related lineages to share environmental tolerances. As a result, information on
29
phylogenetic relationships has underappreciated potential for informing ecological niche
30
modeling. Here we review three strategies for incorporating evolutionary information into
31
niche models: splitting lineages into subunits, lumping across lineages, and partial pooling
32
lineages into a common statistical framework that implicitly or explicitly accounts for
33
evolutionary relationships. We challenge the default practice of modeling at the species
34
level, which ignores the process of niche evolution and erroneously assumes that the
35
species is always the appropriate level for niche estimation. Progress in the field requires
36
re-examining how we assess models of niches versus models of distributions.
37 38
Main text
39
What is the appropriate phylogenetic level for niche estimation?
40
The niche (see Glossary) is integral to understanding distributions and patterns of
41
biological diversity, identifying critical habitat to mitigate the effects of global change, and
42
reconstructing evolutionary trajectories of clades. Although there are many ways to
43
estimate niches, a common starting point is the correlative ecological niche model, which
44
identifies environmental conditions suitable to a species by finding associations between
45
locations where a species is known to be present and the environment at those sites [1].
46
Although most niche models are constructed at the species level, an increasing 2
47
appreciation for the role of evolutionary processes, especially niche conservatism and
48
local adaptation, has led to modeling of niches above and below the level of species [2-10].
49
These studies raise critical questions. For example, what are the risks of modelling a
50
species as a single undifferentiated entity, ignoring ecotypes and local adaptation? How can
51
we estimate fundamental ecological tolerances if species occupy a subset of inhabitable
52
conditions? And how can we exploit knowledge of hierarchical phylogenetic relationships
53
when estimating niches above and below the level of the species?
54
The justification for considering evolutionary relationships when estimating the
55
fundamental niche is straightforward. Rates of niche evolution vary enormously above
56
and below the species level [11, 12]. When niches show high levels of conservatism, close
57
relatives will likely respond similarly to environmental gradients [13]. In contrast, spatial
58
heterogeneity in environments coupled with reduced gene flow can encourage local
59
adaptation, leading to divergence in niches among closely related lineages. In view of these
60
phenomena, there is little reason to assume a priori that species represent the most
61
appropriate phylogenetic level for modeling. Nevertheless, the large majority of niche
62
estimation exercises are conducted at the species level, which assumes that data on
63
species’ current distributions adequately reflect environmental tolerances and that each
64
species responds to the environment as an undifferentiated entity [14, 15]. These critical
65
assumptions are rarely questioned in practice [13, 16, 17]. Nevertheless, their implications
66
are substantial given the breadth of decisions that these models are used to support such as
67
allocating effort toward in situ versus assisted migration [4, 18], identifying corridors to
68
facilitate climate-induced range shifts [19, 20], and focusing attention on areas vulnerable
69
to invasion by non-native species [21, 22]. 3
70
Here we classify the modeling strategies that have been used to address these
71
concerns into three categories based on dominant evolutionary processes. First, in cases
72
where niche evolution is rapid or local adaptation is suspected, “splitting” lineages into
73
distinct units (e.g., subspecies within a species) and creating separate niche models for
74
each unit may best capture environmental relationships [2, 4, 6, 7, 9, 23]. Second, in
75
situations where related lineages display high levels of niche conservatism, “lumping”
76
presence records within a single model can more accurately represent a lineage’s niche [13,
77
24-27]. Spanning the continuum between lumping and splitting is a third option: “partial
78
pooling” of lineages into a single model [28, 29] that may or may not account for
79
phylogenetic relationships (Box 1).
80
In the following section we review splitting and lumping, the two most commonly
81
used methods for incorporating evolutionary processes into niche modeling. Although
82
these methods show promise, each is based on strong assumptions of either local
83
adaptation (splitting) or niche conservatism (lumping). In the subsequent section we
84
address the more flexible approach of partial pooling, which has to date been largely
85
overlooked as a tool for niche estimation. Finally, we call for insight into what constitutes a
86
robust niche model (versus model of geographic distribution) and how this will be key for
87
advancing the field.
88
Accounting for niche evolution by splitting or lumping
89
Splitting and lumping represent opposing strategies for challenging the assumption that
90
the current distribution of a species is the best surrogate for its niche (Figure 1).
91
Adaptation of populations to local circumstances appears to occur quite frequently [30-34].
92
In these cases a promising strategy is to divide species into locally adapted subpopulations 4
93
and model each as a distinct entity. At the same time, niche conservatism also appears
94
prevalent among diverse taxa [12, 35-38]. If niche overlap is high enough between related
95
taxa then lumping them into a single, undifferentiated unit may provide a more robust
96
depiction of environmental tolerances. Lumping can be especially useful for addressing
97
under-representation of the niche arising from dispersal limitation [25], anthropogenic
98
persecution [26, 43], disturbance [44], biotic interactions [24, 45], and limited sampling
99
[13]. In some instances lumping is also necessitated by the intractability of differentiating
100
taxa to the species level such as when modeling fossil data [39-41] or plant functional types
101
[42].
102
Although lumping and splitting have only become prominent alternatives to species-
103
level modeling in recent years, each of these strategies has a history nearly as long as
104
ecological niche modeling itself. Lumping of species into guilds, life forms, or vegetation
105
types was especially common in the formative period of niche modeling and often
106
motivated by the lack of detailed knowledge of individual species’ distributions [46-48].
107
Some early niche modeling efforts also attempted to account for intra-specific variation by
108
modeling subtaxa delineated on the basis of differences in allozymes, karyotype,
109
morphology, and ecology [33, 49-52].
110
Lumping and splitting tend to produce very different niche estimates [4]. Split
111
models generally estimate narrower niches compared to their lumped counterparts [7, 53],
112
but splitting does not necessarily result in smaller estimates of occupied geographic area
113
[14] or in projections that suggest greater vulnerability to anticipated climate change [7-9,
114
23, 53-55]. When distributional data is adequate, split models usually have higher apparent
115
predictive accuracy than lumped models [8-10, 15, 54], but if data is incomplete lumped 5
116
models can outperform split models when niche overlap with related taxa is moderately
117
high [13]. Split and lumped models for the same taxon are often sensitive to different
118
environmental factors [2, 4, 6-8, 54-56].
119
Both splitting and lumping require difficult decisions about the appropriate
120
phylogenetic level (roughly speaking, population, species, genus, and so on) at which to
121
lump or split. As a result, it is vital to scrutinize the evidence upon which the decision to
122
combine or divide occurrence data is based. Indeed, a wide variety of data have been used
123
to demarcate (or merge) subunits: neutral genetic markers [4, 7, 8, 10, 23, 53-56] and
124
adaptive genetic variation [2, 6]; morphological, ecotypic, and phenological variation [7, 8,
125
14]; chemical diversity [7, 56]; geographic contact and introgression [27]; community
126
associations with other species [53]; and geographic isolation [55]. While most studies
127
base divisions (or groupings) on more than one type of evidence, very few [8] test
128
alternative schemes for dividing or combining lineages.
129
We advise against justifying splitting or lumping solely on the basis of differences
130
between models or on traditional null model tests of niche overlap. Although splitting and
131
lumping can increase model reliability under specific circumstances, it is difficult to
132
compare models on an equal basis because of factors that are difficult to control across
133
lumped or split units (e.g., range size and geographic extent [59], sample size [60],
134
differences in environments available to each subdivision [61], and autocorrelation
135
between training and test sites that varies in strength depending on whether occurrences
136
are combined or divided [62]). Similarly, variation in predictor importance across split
137
models is not a reliable indicator of the robustness of division since models trained on
138
biologically meaningless subdivisions of a range are often sensitive to different predictors 6
139
[63, 64]. Monte Carlo-based null model tests of niche overlap [65-67] offer a seemingly
140
attractive method by which to evaluate the appropriateness of proposed subdivisions.
141
These tests assess whether the degree of similarity between a pair of taxa is greater or less
142
than what is expected by chance, but this is not the same as asking whether information
143
from one subdivision informs niche estimates of another. For example, a pair of lineages
144
could be more similar than expected given available habitats and yet still be sufficiently
145
different in their environmental tolerances that their niches are best modeled by splitting.
146
In a related vein, an ongoing debate revolves around the prevalence of niche shifts
147
in invaded portions of non-native species’ ranges. Although current evidence suggests
148
niche shifts are rare [37] (and thus lumping across native and invaded ranges is usually
149
warranted [68]), tests for niche shifts are only robust within shared environmental space
150
[65] and thus do not indicate whether niche shifts occur in novel environments as species
151
expand into their invaded ranges.
152
Since the utility of lumping or splitting strongly depends on the degree of niche
153
conservatism among taxa (Figure 1), whenever possible the justification for dividing (or
154
combining) subdivisions should be based on evidence for (or lack of) selectively-based
155
niche differentiation among purported subunits. Such evidence includes results from
156
progeny tests and common garden experiments [6, 34, 57, 58] or association mapping of
157
alleles with observed phenotypic differences across populations [2]. Secondarily, splits can
158
be based on aspects of life history or functional traits expected to experience strong
159
selection [8], although it is important to keep in mind that phenotypic plasticity can
160
obscure lines along which divisions or mergers should be based [7]. Similarly, neutral
161
genetic variation may or may not correspond to selectively-delineated subunits [2, 7]. Of 7
162
course, the utility of lumping or splitting will always depend on the amount and quality of
163
distribution and environmental data. If the available data are unlikely to provide an
164
accurate representation of sub-taxon niches, it may be preferable to lump them in a single
165
model [13] at the expense of misrepresenting sub-taxon niche variation.
166
Regardless of the basis for lumping and splitting, the relative risks of making an
167
incorrect determination should be assessed carefully, especially if model outcomes are
168
used to guide conservation decisions. Following Hällfors and colleagues [4] we recommend
169
that decisions over whether to lump or split be based on the risks and benefits of assuming
170
local adaptation or niche conservatism and that they follow, when possible, a least-regret
171
approach. In some cases, this may involve applying management actions suggested by both
172
lumping and splitting strategies [4]. Given the challenges, it may be tempting to absolve
173
oneself of the decision by taking the traditional approach of modeling at the species level,
174
but we emphasize that the hierarchical nature of evolutionary relationships means that this
175
is nonetheless a decision to both lump (across subpopulations) and split (from other
176
species) [14, 15].
177
Accounting for niche evolution by partial pooling
178
An emerging alternative to lumping and splitting is partial pooling of multiple lineages
179
simultaneously within a single joint model, which permits modeling each lineage’s
180
individual response to the environment while simultaneously “borrowing” information
181
across lineages. The wide variety of partial pooling algorithms makes tracing their history
182
in ecological niche modeling difficult, but generally speaking, pooled niche models which
183
do not explicitly account for phylogenetic relationships have a historical grounding in
8
184
ordination methods [69, 70]. In contrast, models that do explicitly incorporate phylogenetic
185
information are a more recent addition undergoing rapid development [71, 72].
186
We classify partial pooling models into two types. The first, “phylogenetically naïve”
187
partial pooling models, ignore phylogenetic relationships and thus are useful for cases
188
when detailed phylogenetic information is lacking. A wide variety of algorithms can be
189
used in this manner, from standard mixed effects [73] to hierarchical Bayesian models [28]
190
to finite mixture and joint species distribution models [74-83]. In general, these types of
191
models share in common an ability to differentiate the niches of abundant taxa while
192
tending to pull rarer taxa toward a common response (Box 3). For example, finite mixture
193
models, or so-called “archetype” models, group taxa by their relationship to the
194
environment (e.g., linear, unimodal, etc.) while allowing each taxon to have different
195
coefficients describing its particular response [75, 84, 85]. The ability of partial pooling
196
models to accommodate taxa with few occurrences is particularly advantageous for
197
addressing the challenges of modeling data-poor taxa [86, 87]. Nevertheless, the tendency
198
of these models to push rare taxa toward the group mean risks biasing responses of these
199
same taxa [88], so the model structure and appropriate amount of pooling must be
200
considered carefully.
201
The second class of pooling methods comprise “phylogenetically informed” partial
202
pooling models which explicitly account for evolutionary relationships. These models are
203
also algorithmically diverse and thus account for evolutionary relationships in different
204
ways. Some algorithms leverage information on phylogenetic distance using either
205
variance-covariance matrices based on evolutionary distance [71, 72, 89] or by employing
206
phylogenetic distance directly as a predictor [90, 91]. Either of these approaches can 9
207
account for either phylogenetic attraction (a less extreme form of lumping), in which
208
related lineages tend to share environments, or phylogenetic repulsion (a less extreme
209
form of splitting), in which closely related lineages inhabit dissimilar environments,
210
possibly due to competition [71, 93]. Other algorithms implicitly reflect the hierarchical
211
nature of evolutionary relationships by using nested model structures [3, 92] (e.g., by
212
means of nested random effects [92] or more complicated approaches [3]). Mirroring
213
evolutionary relationships within model structure implicitly assumes attraction between
214
lineages in the same subclade since members of a subclade tend to have similar
215
coefficients. As with phylogenetically naïve models the degree of attraction (or repulsion)
216
in phylogenetically informed models is partly determined by the data available for each
217
lineage, and so will likely affect rare lineages more than abundant ones (Box 3).
218
Phylogenetically informed niche models are an area of active development. Recent
219
advances aim not only to increase statistical robustness [94, 95], but also account for
220
multiple levels of relatedness (e.g., above and below the species level) [5], while controlling
221
for imperfect detection [96] and spatial and temporal autocorrelation [93]. Other areas
222
requiring attention include the need for sophisticated cross-validation schemes that
223
account for phylogenetic correlation [3, 62]; uncertainties about the minimum number of
224
taxa necessary for a robust model [28]; whether or not rare species do indeed respond to
225
the environment similarly to more common species [88]; and reliance on simple models of
226
evolution (Box 4).
227
Given the evolutionary basis on which all of biology is built, phylogenetically
228
informed models offer great promise for improving niche estimates [5, 72, 93]. First and
229
foremost, these models have the potential to capture evolutionary processes operating at 10
230
multiple phylogenetic levels [5]. We see this possibility as the “best of both worlds” as it
231
reflects the very real potential for niche differentiation and conservatism to operate
232
simultaneously at different intensities across phylogenetic scales. Second, partial pooling
233
models can accommodate rare lineages that would otherwise be difficult to model. In
234
phylogenetically-informed models, responses of rare lineages will also be influenced by
235
their phylogenetic distance to other lineages, thereby at least partially obviating the
236
concern that niches of rare lineages could become incorrectly drawn toward the overall
237
central tendency. Finally, the sharing of information across lineages can help guard against
238
overfitting and thereby mistaking circumstantial differences in environments inhabited by
239
lineages for adaptively-driven niche differentiation [61].
240
Concluding remarks: Toward better niche estimates
241
Niche models are growing increasingly sophisticated in their ability to account for multiple
242
facets of data. Nevertheless, very few modeling exercises account for phylogenetic
243
relationships despite the foundational importance of evolution in shaping species’
244
responses to the environment. Indeed, the practice of modeling at the individual species’
245
level is so ubiquitous that the underlying assumptions around phylogenetic scale and
246
relatedness are typically not discussed at all [15]. We see this as a missed opportunity. The
247
inclusion of evolutionary information in the niche modeling process has the potential to
248
produce better fundamental niche estimates and thereby greatly improve our
249
understanding of niche evolution and provide a more robust basis for conservation
250
decision-making.
251 252
While we advocate for greater use of partial pooling to incorporate evolutionary information into niche modeling, we also recognize that in some situations this approach 11
253
may be neither desirable nor practical. For instance, phylogenetically informed models are
254
difficult to construct, evaluate and communicate, and may not bring significant advantages
255
over simpler approaches (i.e., non-phylogenetic models) for some datasets (e.g., cases with
256
poorly resolved phylogenies or few species). Likewise, independent (split) species-level
257
models may perform as well as partial pooling if taxa are well-sampled. Finally, even
258
though partial pooling has the potential to ameliorate uncertainty in decisions over
259
whether to lump or split units of analysis, practitioners must still make the decision over
260
how to combine or divide occurrences used for modeling [40, 41, 97, 98].
261
In a conservation context, the utility of splitting, lumping, and partial pooling will be
262
determined by the decision-making framework in which they are applied. For example, the
263
splitting and partial pooling methods may be particularly apt for identifying critical habitat
264
of sub-specific taxa under the US Endangered Species Act which allows listing of individual
265
vertebrate populations. On the other hand, most existing conservation plans have been
266
developed around the species concept, so updating plans using methods that subdivide
267
species may incur unacceptable costs if sub-specific models and species-level models give
268
different outcomes [99]. We urge caution when applying splitting or partial pooling models
269
to species that have become genetically differentiated due to anthropogenic fragmentation
270
since they may overstate differences that are likely important to dissolve through
271
management action [100]. On the same hand, these models could be used to guide efforts to
272
merge anthropogenically-differentiated populations by identifying populations that are
273
most similar in their response to the environment and thereby reduce risks from managed
274
genetic introgression.
12
275
In this review we have discussed methodological issues surrounding the inclusion of
276
evolution into the modeling process, but the focus on geographic distributions creates
277
cultural inertia within the field that must also be addressed, particularly with respect to
278
model evaluation. The use of lumping, splitting, or partial pooling may in many cases result
279
in niche estimates that are more biologically accurate, but which poorly estimate the
280
current geographic distribution of the lineages included in the model. The current
281
emphasis on model selection based solely on geographic predictions of niche models will
282
therefore tend to reject biologically realistic models of niches in favor of
283
overparameterized models that include spurious correlations with other processes (e.g.,
284
dispersal limitation, biotic interactions [101-103]). The move toward evolutionarily-
285
informed models therefore requires a shift in how we think about and evaluate niche
286
models: they must be assessed based on their biological plausibility rather than the
287
geographic distributions we make by projecting them. Promising avenues include
288
comparing models’ ability to estimate known physiological limits [17, 104-106] and
289
generation of relatively “smooth” response curves which match theoretical expectations
290
[103, 107, 108].
291
Niche estimation has largely proceeded by assuming neither anything above nor
292
anything below the species level is relevant to understanding species’ relationships with
293
the environment. Each of the strategies we review here offers promise for addressing this
294
shortcoming. Of them, phylogenetically informed partial pooling holds substantial yet
295
unrealized promise, but we anticipate all of these methods will remain useful tools so long
296
as outstanding issues can be addressed (see Outstanding Questions). Looking forward, we
297
anticipate the development of flexible modeling frameworks that account for hierarchical 13
298
evolutionary relationships above and below the species level, variable rates of niche
299
evolution, and possibly uncertainty in estimation of the phylogeny itself.
300 301
Acknowledgements
302
This manuscript emerged from the Investigative Workshop “Species’ Range Shifts in a
303
Warming World” at the US National Institute for Mathematical and Biological Synthesis
304
(NIMBioS; http://www.nimbios.org/workshops/WS_rangeshifts), sponsored by the
305
National Science Foundation through NSF Award #DBI-1300426, with additional support
306
from the University of Tennessee, Knoxville. We thank Sean Hoban, Allan Strand, Andria
307
Dawson, and Michelle Lawing for organizing, and NIMBioS for logistical and financial
308
support. F. Rodríguez-Sánchez was supported by a postdoctoral fellowship from the
309
Spanish Ministry of Economy and Competitiveness (FPD-2013-16756), W. Godsoe by the
310
New Zealand Tertiary Education Commission CoRE grant to the Bio-Protection Research
311
Centre, and A.B. Smith by the Alan Graham Fund in Global Change. We thank Stephen J.
312
Murphy for helpful comments on a draft version of the manuscript.
313 314
References
315
1. Peterson, A.T. et al. (2011) Ecological Niches and Geographic Distributions (MPB-49),
316
Princeton University Press.
317
2. Banta, J.A. et al. (2012) Climate envelope modelling reveals intraspecific relationships
318
among flowering phenology, niche breadth and potential range size in Arabidopsis
319
thaliana. Ecology Letters 15 (8), 769-777.
14
320
3. Godoy, B.S. et al. (2018) When phylogeny and ecology meet: Modleing the occurrence of
321
Trichoptera with environmental and phylogenetic data. Ecology and Evolution 8, 5313-
322
5322.
323
4. Hällfors, M.H. et al. (2016) Addressing potential local adaptation in species distribution
324
models: Implications for conservation under climate change. Ecological Applications 26
325
(4), 1154-1169.
326 327 328 329 330
5. Joly, S. et al. (2018) On the importance of accounting for intraspecific genomic relatedness in multi-species studies. bioRxiv, doi: 10.1101/321927. 6. Koralewski, T.E. et al. (2015) Plants on the move: Assisted migration of forest trees in the face of climate change. Forest Ecology and Management 344, 30-37. 7. Maguire, K.C. et al. (2018) Intraspecific niche models for Ponderosa Pine (Pinus
331
ponderosa) suggest potential variability in population-level response to climate change.
332
Systematic Biology, doi: 10.1093/sysbio/syy017.
333
8. Marcer, A. et al. (2016) Tackling intraspecific genetic structure in distribution models
334
better reflects species geographical range. Ecology and Evolution 6 (7), 2084-2097.
335
9. Pearman, P.B. et al. (2010) Within‐taxon niche structure: Niche conservatism, divergence
336 337
and predicted effects of climate change. Ecography 33 (6), 990-1003. 10. Serra‐Varela, M. et al. (2015) Does phylogeographical structure relate to climatic niche
338
divergence? A test using maritime pine (Pinus pinaster Ait.). Global Ecology and
339
Biogeography 24 (11), 1302-1313.
340
11. Kellermann, V. et al. (2012) Phylogenetic constraints in key functional traits behind
341
species’ climate niches: Patterns of desiccation and cold resistance across 95 Drosophila
342
species. Evolution 66 (11), 3377-3389. 15
343 344 345
12. Peterson, A.T. (2011) Ecological niche conservatism: A time‐structured review of evidence. Journal of Biogeography 38 (5), 817-827. 13. Qiao, H. et al. (2017) Using data from related species to overcome spatial sampling bias
346
and associated limitations in ecological niche modelling. Methods in Ecology and
347
Evolution 8 (12), 1804-1812.
348
14. Mota-Vargas, C. and Rojas-Soto, O.R. (2016) Taxonomy and ecological niche modeling:
349
Implications for the conservation of wood partridges (genus Dendrortyx). Journal for
350
Nature Conservation 29, 1-13.
351
15. Romero, D. et al. (2014) Uncertainty in distribution forecasts caused by taxonomic
352
ambiguity under climate change scenarios: A case study with two newt species in
353
mainland Spain. Journal of Biogeography 41 (1), 111-121.
354 355 356 357 358 359
16. Qiao, H. et al. (2017) Accessible areas in ecological niche comparisons of invasive species: Recognized but still overlooked. Scientific Reports 7 (1), 1213. 17. Qiao, H. et al. (2018) An evaluation of transferability of ecological niche models. Ecography, doi:10.1111/ecog.03986. 18. Wang, H.-H. et al. (2015) Species distribution modelling for conservation of an endangered endemic orchid. AoB PLANTS 7, plv039-plv039.
360
19. Alagador, D. et al. (2016) Climate change, species range shifts and dispersal corridors:
361
an evaluation of spatial conservation models. Methods in Ecology and Evolution 7 (7),
362
853-866.
363
20. Carroll, C. et al. (2018) Climatic, topographic, and anthropogenic factors determine
364
connectivity between current and future climate analogs in North America. Global
365
Change Biology 24 (11), 5318-5331. 16
366 367 368 369 370
21. Peterson, A.T. (2003) Predicting the Geography of Species’ Invasions via Ecological Niche Modeling. The Quarterly Review of Biology 78 (4), 419-433. 22. Thuiller, W. et al. (2005) Niche-based modelling as a tool for predicting the risk of alien plant invasions at a global scale. Global Change Biology 11 (12), 2234-2250. 23. Prasad, A.M. and Potter, K.M. (2017) Macro-scale assessment of demographic and
371
environmental variation within genetically derived evolutionary lineages of eastern
372
hemlock (Tsuga canadensis), an imperiled conifer of the eastern United States.
373
Biodiversity and Conservation 26 (9), 2223-2249.
374
24. Alvarez‐Martínez, J.M. et al. (2015) Can Eltonian processes explain species distributions
375
at large scale? A case study with Great Bustard (Otis tarda). Diversity and Distributions
376
21 (2), 123-138.
377 378 379
25. Bush, A. et al. (2018) Truncation of thermal tolerance niches among Australian plants. Global Ecology and Biogeography 27 (1), 22-31. 26. Martínez‐Freiría, F. et al. (2016) Contemporary niche contraction affects climate change
380
predictions for elephants and giraffes. Diversity and Distributions 22 (4), 432-444.
381
27. Thorne, J.H. et al. (2013) Alternative biological assumptions strongly influence models
382
of climate change effects on mountain gorillas. Ecosphere 4 (9), 108.
383
28. Gelman, A. and Hill, J. (2007) Data Analysis Using Regression and
384
Multilevel/Hierarchical Models, Cambridge University Press.
385 386
29. Qian, S.S. et al. (2010) On the application of multilevel modeling in environmental and ecological studies. Ecology 91 (2), 355-361.
17
387
30. Bocedi, G. et al. (2013) Effects of local adaptation and interspecific competition on
388
species’ responses to climate change. Annals of the New York Academy of Sciences 1297
389
(1), 83-97.
390
31. Lascoux, M. et al. (2016) Local Adaptation in Plants. In eLS, John Wiley & Sons, Ltd.
391
32. Moran, E.V. and Ormond, R.A. (2015) Simulating the interacting effects of intraspecific
392
variation, disturbance, and competition on climate-driven range shifts in trees. PLoS
393
ONE 10 (11), e0142369.
394 395
33. Peterson, A.T. and Holt, R.D. (2003) Niche differentiation in Mexican birds: Using point occurrences to detect ecological innovation. Ecology Letters 6 (8), 774-782.
396
34. Valladares, F. et al. (2014) The effects of phenotypic plasticity and local adaptation on
397
forecasts of species range shifts under climate change. Ecology Letters 17 (11), 1351-
398
1364.
399
35. Culumber, Z.W. and Tobler, M. (2018) Correlated evolution of thermal niches and
400
functional physiology in tropical freshwater fishes. Journal of Evolutionary Biology 31
401
(5), 722-734.
402 403 404 405 406 407
36. Mesquita, D.O. et al. (2016) Life-history patterns of lizards of the world. The American Naturalist 187 (6), 689-705. 37. Petitpierre, B. et al. (2012) Climatic niche shifts are rare among terrestrial plant invaders. Science 335 (6074), 1344-1348. 38. Pie, M.R. et al. (2017) The evolution of climatic niches in squamate reptiles. Proceedings of the Royal Society B: Biological Sciences 284 (1858).
18
408
39. Maguire, K.C. et al. (2016) Controlled comparison of species- and community-level
409
models across novel climates and communities. Proceedings of the Royal Society B:
410
Biological Sciences 283 (1826), doi: 10.1098/rspb.2015.2817.
411
40. Moreno-Amat, E. et al. (2017) Incorporating plant fossil data into species distribution
412
models is not straightforward: Pitfalls and possible solutions. Quaternary Science
413
Reviews 170, 56-68.
414
41. Rodrıguez-Sánchez, F. and Arroyo, J. (2011) Cenozoic Climate Changes and the Demise
415
of Tethyan Laurel Forests: Lessons for the Future from an Integrative Reconstruction of
416
the Past. In Climate Change, Ecology and Systematics (Hodkinson, T.R. et al. eds), pp.
417
280-303, Cambridge University Press.
418 419 420 421 422
42. Snell, R.S. et al. (2014) Using dynamic vegetation models to simulate plant range shifts. Ecography 37 (12), 1184-1197. 43. Faurby, S. and Araújo, M.B. (2018) Anthropogenic range contractions bias species climate change forecasts. Nature Climate Change 8 (3), 252-256. 44. Tinner, W. et al. (2013) The past ecology of Abies alba provides new perspectives on
423
future responses of silver fir forests to global warming. Ecological Monographs 83 (4),
424
419-439.
425 426 427 428 429 430
45. Mpakairi, K.S. et al. (2017) Missing in action: Species competition is a neglected predictor variable in species distribution modelling. PLoS ONE 12 (7), e0181088. 46. Bonn, A. and Schröder, B. (2001) Habitat models and their transfer for single and multi species groups: a case study of carabids in an alluvial forest. Ecography 24 (4), 483-496. 47. Box, E.O. (1981) Predicting physiognomic vegetation types with climate variables. Vegetatio 45 (2), 127-139. 19
431
48. Woodward, F.I. (1987) Climate and Plant Distribution, Cambridge University Press.
432
49. Austin, M.P. et al. (1990) Measurement of the realized qualitative niche: Environmental
433
niches of five Eucalyptus species. Ecological Monographs 60 (2), 161-177.
434
50. Costa, J. et al. (2002) Ecologic niche modeling and differentiation of populations of
435
Triatoma brasiliensis Neiva, 1911, the most important Chagas' disease vector in
436
northeastern Brazil (Hemiptera, Reduviidae, Triatominae). The American Journal of
437
Tropical Medicine and Hygiene 67 (5), 516-520.
438
51. Kohlmann, B. et al. (1988) Environmental predictions and distributional limits of
439
chromosomal taxa in the Australian grasshopper Caledia captiva (F.). Oecologia 75 (4),
440
483-493.
441 442
52. Thuiller, W. et al. (2004) Relating plant traits and species distributions along bioclimatic gradients for 88 Leucadendron taxa. Ecology 85 (6), 1688-1699.
443
53. Cacciapaglia, C. and van Woesik, R. (2018) Marine species distribution modelling and
444
the effects of genetic isolation under climate change. Journal of Biogeography 45 (1),
445
154-163.
446 447 448 449 450
54. Ikeda, D.H. et al. (2017) Genetically informed ecological niche models improve climate change predictions. Global Change Biology 23 (1), 164-176. 55. Malone, S.L. et al. (2018) The future of subalpine forests in the Southern Rocky Mountains: Trajectories for Pinus aristata genetic lineages. PLoS ONE 13 (3), e0193481. 56. Shinneman, D.J. et al. (2016) Exploring climate niches of ponderosa pine (Pinus
451
ponderosa Douglas ex Lawson) haplotypes in the western United States: Implications
452
for evolutionary history and conservation. PLoS ONE 11 (3), e0151811.
20
453
57. Benito-Garzón, M. et al. (2013) Interspecific differences in tree growth and mortality
454
responses to environmental drivers determine potential species distributional limits in
455
Iberian forests. Global Ecology and Biogeography 22 (10), 1141-1151.
456
58. Mendola, M.L. et al. (2015) The role of ecotypic variation and the environment on
457
biomass and nitrogen in a dominant prairie grass. Ecology 96 (9), 2433-2445.
458
59. Smith, A.B. (2013) On evaluating species distribution models with random background
459
sites in place of absences when test presences disproportionately sample suitable
460
habitat. Diversity and Distributions 19, 867-872.
461
60. Soultan, A. and Safi, K. (2017) The interplay of various sources of noise on reliability of
462
species distribution models hinges on ecological specialisation. PLoS ONE 12 (11),
463
e0187906.
464 465 466 467 468 469 470
61. Godsoe, W. (2010) Regional variation exaggerates ecological divergence in niche models. Systematic Biology 59 (3), 298-306. 62. Roberts, D.R. et al. (2017) Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure. Ecography 40 (8), 913-929. 63. Raes, N. (2012) Partial versus full species distribution models. Natureza and Conservação 10, 127-138. 64. Titeux, N. et al. (2017) The need for large‐scale distribution data to estimate regional
471
changes in species richness under future climate change. Diversity and Distributions 23
472
(12), 1393-1407.
473 474
65. Broennimann, O. et al. (2012) Measuring ecological niche overlap from occurrence and spatial environmental data. Global Ecology and Biogeography 21 (4), 481-497.
21
475 476 477
66. Nunes, L.A. and Pearson, R.G. (2017) A null biogeographical test for assessing ecological niche evolution. Journal of Biogeography 44 (6), 1331-1343. 67. Warren, D.L. et al. (2008) Environmental niche equivalency versus conservatism:
478
Quantitative approaches to niche evolution. Evolution 62 (11), 2868-2883.
479
68. Broennimann, O. and Guisan, A. (2008) Predicting current and future biological
480
invasions: Both native and invaded ranges matter. Biology Letters 4 (5), 585-589.
481
69. Ramensky, L.G. (1929) К методике сравнительной обработки и систематизации
482
списков растительности и других объектов, определяемых несколькими
483
несходно действующими факторами. (On methods for comparative analysis and
484
ordering of plant lists and other objects determined by multiple factors). Leningrad:
485
Trudy Soveščaniya Geobotanikov-lugovedov, 1-26.
486 487 488 489 490
70. ter Braak, C.J.F. (1994) Canonical community ordination. Part I: Basic theory and linear methods. Écoscience 1 (2), 127-140. 71. Ives, A.R. and Helmus, M.R. (2011) Generalized linear mixed models for phylogenetic analyses of community structure. Ecological Monographs 81 (3), 511-525. 72. Ives, A.R. and Garland, T. (2014) Phylogenetic Regression for Binary Dependent
491
Variables. In Modern Phylogenetic Comparative Methods and Their Application in
492
Evolutionary Biology (Garamszegi, L.Z. ed), pp. 231-261, Springer-Verlag Berlin
493
Heidelberg.
494 495 496 497
73. Bolker, B.M. et al. (2009) Generalized linear mixed models: A practical guide for ecology and evolution. Trends in Ecology & Evolution 24 (3), 127-135. 74. Clark, J.S. et al. (2017) Generalized joint attribute modeling for biodiversity analysis: Median-zero, multivariate, multifarious data. Ecological Monographs 87 (1), 34-56. 22
498
75. Hui, F.K.C. et al. (2013) To mix or not to mix: Comparing the predictive performance of
499
mixture models vs. separate species distribution models. Ecology 94 (9), 1913-1919.
500
76. Nieto‐Lugilde, D. et al. (2018) Multiresponse algorithms for community‐level modelling:
501
Review of theory, applications, and comparison to species distribution models. Methods
502
in Ecology and Evolution 9 (4), 834-848.
503 504 505
77. Ovaskainen, O. and Soininen, J. (2011) Making more out of sparse data: Hierarchical modeling of species communities. Ecology 92 (2), 289-295. 78. Ovaskainen, O. et al. (2016) Uncovering hidden spatial structure in species communities
506
with spatially explicit joint species distribution models. Methods in Ecology and
507
Evolution 7 (4), 428-436.
508
79. Ovaskainen, O. et al. (2017) How to make more out of community data? A conceptual
509
framework and its implementation as models and software. Ecology Letters 20 (5), 561-
510
576.
511
80. Pollock, L.J. et al. (2014) Understanding co-occurrence by modelling species
512
simultaneously with a Joint Species Distribution Model (JSDM). Methods in Ecology and
513
Evolution 5 (5), 397-406.
514
81. Thorson, J.T. et al. (2016) Joint dynamic species distribution models: A tool for
515
community ordination and spatio-temporal monitoring. Global Ecology and
516
Biogeography 25 (9), 1144-1158.
517 518 519 520
82. Warton, D.I. et al. (2015) So many variables: Joint modeling in community ecology. Trends in Ecology & Evolution 30 (12), 766-779. 83. Zhang, C. et al. (2018) Comparing the prediction of joint species distribution models with respect to characteristics of sampling data. Ecography, doi:10.1111/ecog.03571. 23
521
84. Dunstan, P.K. et al. (2013) Finite mixture of regression modeling for high-dimensional
522
count and biomass data in ecology. Journal of Agricultural, Biological, and
523
Environmental Statistics 18 (3), 357-375.
524 525
85. Hui, F.K. et al. (2015) Multi-species distribution modeling using penalized mixture of regressions. The Annals of Applied Statistics 9 (2), 866-882.
526
86. Breiner, F.T. et al. (2015) Overcoming limitations of modelling rare species by using
527
ensembles of small models. Methods in Ecology and Evolution 6 (10), 1210-1218.
528
87. Lomba, A. et al. (2010) Overcoming the rare species modelling paradox: A novel
529
hierarchical framework applied to an Iberian endemic plant. Biological Conservation
530
143 (11), 2647-2657.
531 532 533
88. Pacifici, K. et al. (2014) Guidelines for a priori grouping of species in hierarchical community models. Ecology and Evolution 4 (7), 877-888. 89. Hadfield, J.D. and Nakagawa, S. (2010) General quantitative genetic methods for
534
comparative biology: Phylogenies, taxonomies and multi-trait models for continuous
535
and categorical characters. Journal of Evolutionary Biology 23 (3), 494-508.
536 537 538
90. Morales‐Castilla, I. et al. (2017) Combining phylogeny and co‐occurrence to improve single species distribution models. Global Ecology and Biogeography 26 (6), 740-752. 91. Oke, O.A. et al. (2014) Integrating phylogenetic community structure with species
539
distribution models: An example with plants of rock barrens. Ecography 37 (7), 614-
540
625.
541 542
92. Bates, A.E. et al. (2010) Deep-sea hydrothermal vent animals seek cool fluids in a highly variable thermal environment. Nature Communications 1, 14.
24
543 544 545 546 547 548 549
93. Kaldhusdal, A. et al. (2015) Spatio‐phylogenetic multispecies distribution models. Methods in Ecology and Evolution 6 (2), 187-197. 94. Bürkner, P.C. (In press) Advanced Bayesian multilevel modeling with the R package brms. The R Journal. 95. Pearse, W.D. et al. (2015) pez: phylogenetics for the environmental sciences. Bioinformatics 31 (17), 2888-2890. 96. Frishkoff, L.O. et al. (2017) Phylogenetic occupancy models integrate imperfect
550
detection and phylogenetic signal to analyze community structure. Ecology 98 (1), 198-
551
210.
552 553
97. Nogués-Bravo, D. et al. (2016) Amplified plant turnover in response to climate change forecast by Late Quaternary records. Nature Climate Change 6, 1115-1119.
554
98. Veloz, S.D. et al. (2012) No-analog climates and shifting realized niches during the late
555
quaternary: Implications for 21st-century predictions by species distribution models.
556
Global Change Biology 18 (5), 1698-1713.
557
99. Malaney, J.L. and Cook, J.A. (2013) Using biogeographical history to inform
558
conservation: the case of Preble's meadow jumping mouse. Molecular Ecology 22 (24),
559
6000-6017.
560 561 562 563
100. Ralls, K. et al. (2018) Call for a paradigm shift in the genetic management of fragmented populations. Conservation Letters 11 (2), e12412. 101. Escobar, L.E. et al. (2018) Ecological niche modeling re‐examined: A case study with the Darwin's fox. Ecology and Evolution 8 (10), 4757-4770.
25
564
102. Jarnevich, C.S. et al. (2017) Minimizing effects of methodological decisions on
565
interpretation and prediction in species distribution studies: An example with
566
background selection. Ecological Modelling 363, 48-56.
567 568 569
103. Merow, C. et al. (2014) What do we gain from simplicity versus complexity in species distribution models? Ecography 37 (12), 1267-1281. 104. Austin, M.P. et al. (2009) Physiological responses and statistical models of the
570
environmental niche: A comparative study of two co-occurring Eucalyptus species.
571
Journal of Ecology 97 (3), 496-507.
572
105. Martínez, B. et al. (2015) Combining physiological threshold knowledge to species
573
distribution models is key to improving forecasts of the future niche for macroalgae.
574
Global Change Biology 21 (4), 1422-1433.
575
106. Rodríguez-Sánchez, F. and Arroyo, J. (2008) Reconstructing the demise of Tethyan
576
plants: Climate-driven range dynamics of Laurus since the Pliocene. Global Ecology and
577
Biogeography 17 (6), 685-695.
578
107. Bell, D.M. and Schlaepfer, D.R. (2016) On the dangers of model complexity without
579
ecological justification in species distribution modeling. Ecological Modelling 330, 50-
580
59.
581
108. Guevara, L. et al. (2018) Toward ecologically realistic predictions of species
582
distributions: A cross-time example from tropical montane cloud forests. Global Change
583
Biology 24 (4), 1511-1522.
584
109. Soberón, J. and Nakamura, M. (2009) Niches and distributional areas: Concepts,
585
methods, and assumptions. Proceedings of the National Academy of Sciences 106
586
(Supplement 2), 19644-19650. 26
587 588 589 590 591 592 593 594
110. Scheffer, M. et al. (2018) Toward a unifying theory of biodiversity. Proceedings of the National Academy of Sciences 115 (4), 639-641. 111. Ferrier, S. and Guisan, A. (2006) Spatial modelling of biodiversity at the community level. Journal of Applied Ecology 43 (3), 393-404. 112. Soberón, J. and Peterson, A.T. (2011) Ecological niche shifts and environmental space anisotropy: A cautionary note. Revista Mexicana de Biodiversidad 82 (4), 1384-1355. 113. Mooers, A.Ø. et al. (1999) Using phylogenies to test macroevolutionary hypotheses of trait evolution in cranes (Gruinae). The American Naturalist 154 (2), 249-259.
595 596
27
597
Glossary
598
Ecological niche model (ENM): A statistical method that identifies environmental
599
conditions conducive to persistence of a taxon. As used here, ENMs estimate niches
600
whereas species distribution models estimate geographic distributions.
601
Fundamental niche: The set of environmental conditions in which a taxon can survive and
602
reproduce ignoring biotic interactions.
603
Lineage: A set of genetically related populations that are typically monophyletic and
604
typically assumed to be adaptive units. Lineages may form above, at, or below the level of
605
the species.
606
Local adaptation: Higher fitness of genotypes in their environment of origin relative to
607
their fitness in other environments.
608
Lumping: As used here, modeling multiple taxa or sub-taxa together as a single statistical
609
sample.
610
Niche conservatism: The tendency of closely related taxa to have similar ecological
611
tolerances and preferences. In the current context we are specifically interested in
612
environmental requirements.
613
Niche: Generally, the set of requirements necessary for persistence of a lineage. As used in
614
this review, “niche” refers to favorable environmental conditions.
615
Partial pooling: Modeling multiple taxa simultaneously in a single statistical framework so
616
that information is shared across taxa. There are many kinds of partial pooling models,
617
which may or may not explicitly account for phylogenetic relationships.
28
618
Species distribution model (SDM): A statistical tool that estimates the environmental
619
tolerances of a species. As used here, SDMs estimate geographic distribution while
620
ecological niche models estimate niches.
621
Splitting: Dividing lineages (genera, species, subspecies, etc.) into subunits and modeling
622
each separately.
623 624
29
625
Figure 1. Choosing an Appropriate Niche Modeling Strategy. The “biotic-abiotic-
626
movement” (BAM) diagram is commonly used to conceptualize niche modeling (top) [1, 13,
627
109]. The set of suitable abiotic conditions (A1) for a species is represented in blue. The
628
goal of most niche modeling studies is to delineate the environmental conditions that fall
629
within this region. However, taxa often only occur within a subset of the abiotically suitable
630
environmental conditions, as their distributions are also limited to the set of locations that
631
have favorable abiotic environments (A1) and favorable biotic interactions (B1), and are
632
reachable by dispersal (M1). As a result, although the taxon can tolerate all conditions with
633
A1, it will only be possible to collect occurrence data from the intersection of A1, B1, and M1,
634
represented by the white Reuleaux triangle. In the middle panel niche evolution is
635
sufficiently slow that a pair of closely related taxa effectively have the same fundamental
636
niche (A1 + A2). These taxa can therefore tolerate the same set of environments, but differ
637
in their realized niches due to differences in biotic interactions (B1 and B2) and dispersal
638
(M1 and M2). In this case a lumped or partially pooled model will produce a better niche
639
estimate than is possible for either taxon alone, as the joint distribution of both taxa does a
640
much better job of sampling the set of suitable environments. In the bottom panel a pair of
641
taxa have diverged ecologically to the point where they have entirely non-overlapping
642
abiotic niches (A1 and A2). In this case, a split model is advised because a partially pooled or
643
lumped niche estimate would likely be misleading, even if the taxa were restricted to the
644
same set of biotic interactions (B1 + B2) and able to reach the same locations through
645
dispersal (M1 + M2).
30
646
Box 1. “Splitting,” “Lumping,” and “Partial Pooling”
647
In ecology the practice of combining samples goes by a diverse set of terms often used
648
interchangeably including “lumping”, “pooling”, “grouping”, “combining,” “integrating”, and
649
“assimilating”, amongst others [13, 39, 88, 110]. Depending on context, any of these
650
designations can mean either combining occurrences from different classes (e.g., species I a
651
genus) and treating them as if they belonged to a single class (as often done in paleoecology
652
when identification of fossil records to the species level is difficult [40, 41]), or combining
653
occurrences in a single multi-response model that respects sample class [76]. Here we use
654
“lumping” to refer to modeling occurrences of different lineages as if they were a single
655
lineage, “splitting” to fitting a model to each lineage independently, and “partial pooling” to
656
modeling occurrences of related lineages in a single mixed effects or multi-response model
657
[28].
658
Confusingly, multi-response models themselves go by different names in ecology,
659
including “community-level”, “multispecies,” and “joint species” models [76]. A somewhat
660
more focused set of methods is encompassed by the term “vegetation modeling,” which
661
usually refers to combining taxa with similar functional forms for estimating community-
662
level properties [42]. Regardless of name, all of these types of models are typically applied
663
to collections of taxa irrespective of their phylogenetic relatedness [111] except in the
664
broadest sense (e.g., all plants). Moreover, these models typically focus on spatial
665
distribution, diversity, assembly mechanisms, and functional traits of potentially co-
666
occurring species [76, 78, 82, 83]. In contrast, the goal of niche modeling is to estimate
667
fundamental environmental tolerances [1, 112], often of lineages that are allopatric and
668
which may even comprise extant and extinct lineages. Our focus is on modeling niches of 31
669
lineages that are relatively closely related regardless of whether they co-occur in time and
670
space.
32
671
Box 2. Lumping Versus Splitting
672
Figure I. Lumped versus Split Niche Models. The wood partridge genus Dendrortyx
673
comprises three species, two of which are further subdivided into several subspecies (left).
674
Lumping species into a genus-level model results in broader niches than individual species-
675
level models which are indicated by the filled polygons in the middle panel. In turn,
676
species-levels models estimate broader niches than subspecies models which are indicated
677
by the filled polygons in the right panel (only results for Dendrortyx macroura are shown).
678
The left and middle panels are adapted from Mota-Vargas and Rojas-Soto [14].
679 680
Niche models are commonly applied at the species level but this practice can obscure
681
relevant information about at other phylogenetic levels. To illustrate, we highlight work on
682
wood partridges (genus Dendrortyx) endemic to Central America. The genus is divided into
683
three species, two of which are further divided into subspecies based on morphological
684
variation and geographic separation, though wide variation within and across populations
685
makes taxonomic designation uncertain.
686
To explore this situation, Mota-Vargas and Rojas-Soto [14] used ecological niche
687
models to estimate niches at three taxonomic levels (genus, species, and subspecies; Figure
688
I). They found striking differences between lumped and split niches. In particular,
689
temperature was the most important variable in the genus-level model, whereas
690
temperature and precipitation had equal influence in the species- and subspecies-level
691
models. Niches at one level were as broad as, and sometimes broader, than combined
692
niches estimated for the next-lower level. The greater breadth of the higher-level niche was
693
generally—but not always—reflected in the geographic area estimated to be suitable. For 33
694
example, the genus-level model predicted a suitable geographic area 39% larger than the
695
sum of the species-level models, and the species-level model for D. macroura predicted an
696
area 59% larger than the sum of its sub-species models. In contrast, the species-level model
697
for D. leucrophrys predicted a suitable area 29% smaller than the combination of its three
698
subspecies models. Differences between split models were used to corroborate proposed
699
locations of environmental barriers conducive to morphological (and thus taxonomic)
700
differentiation.
34
701
Box 3 Comparing Splitting, Lumping, and Partial Pooling Models
702
Figure I. Simulated Example Illustrating Different Modeling Approaches. Split, lumped,
703
phylogenetically naïve, and phylogenetically informed partial pooling models depict
704
different environmental relationships for the same set of related taxa.
705 706
To illustrate the general behavior of each modeling approach we simulated the niches of six
707
related taxa (Figure I) across an artificial environmental gradient (hereafter,
708
“temperature”) from 0 to 20°C (see Online Supplemental Material). The six taxa vary in
709
their sensitivity to temperature according to their location in the phylogeny (green and
710
yellow taxa being more warm-adapted than bluer taxa). In this example, the splitting
711
strategy (i.e., fitting six independent binomial generalized linear models, or GLMs, one to
712
each taxon) recovers different niches among taxa at the expense of mis-estimating relative
713
suitability, especially at colder temperatures. The lumping strategy (fitting a single GLM to
714
all taxa together) finds a common response while ignoring inter-taxon differences. Partial
715
pooling implemented with a binomial generalized linear mixed model (GLMM) that ignores
716
phylogenetic relationships allows each taxon to respond differently while nonetheless
717
drawing them toward a common response. Finally, a phylogenetically informed generalized
718
linear mixed effects model (PGLMM) implemented with phylogenetic “attraction” causes
719
closely-related taxa to be more similar in their estimated response (compare to the
720
phylogenetically-naïve GLMM).
35
721
Box 4. Models of Evolution
722
Figure I. A Diversity of Evolutionary Modes. Phylogenetic partially pooled models typically
723
assume that changes in a lineage’ niche can be modeled as a random walk (Brownian
724
motion; top). More sophisticated models allow that niches in a clade frequently may tend to
725
converge to some common value (Ornstein-Uhlenbeck; middle). Finally, “speciational”
726
models of evolution (bottom) are able to account for differences between the distributions
727
of parent and daughter lineages which can emerge from allopatric parapatric or peripatric
728
speciation. In each panel niche evolution is assumed to begin with a common ancestor at
729
the bottom and proceed through time toward the top.
730 731
Phylogenetically informed partial pooling models typically assume that lineages that are
732
more closely related tend to have more similar or dissimilar niches. This requires an
733
explicit model of how the niche changes over time, but there are open questions about the
734
best way to represent niche evolution. Existing approaches assume a random (Brownian
735
motion) model of evolution where the environments occupied by a lineage diverge
736
gradually over time with no directional trend (Figure I top). However, there may be
737
situations where niche specialization is disfavored, say if there are constraints in the
738
degree to which a lineage can adapt to extreme temperatures. This leads to a constrained
739
random walk (an Ornstein-Uhlenbeck model; Figure I middle), which is commonly used in
740
macro-evolution but less commonly in phylogenetically informed ecological niche
741
modeling [71]. Macroevolutionary studies have developed approaches to account for rapid
742
evolution during speciation, the simplest of which is to model divergence based on the
743
number of speciation events separating lineages (versus divergence time between 36
744
lineages), leading to a “speciational” model of evolution (Figure I bottom; [3, 113]).
745
Allopatric speciation can alter lineages’ distributions even when they do not alter niches
746
[61], an important effect which may be missed by current models of evolution.
37
Outstanding Questions
1
Outstanding Questions
2
● How should lumped and split models be post-processed to obtain predictions at the
3 4 5
appropriate level (usually species or populations) for stakeholders? ● At what level is phylogenetic level (i.e., corresponding roughly to genus, family, etc.) no longer informative about species-levels niches?
6
● Rare taxa can “borrow strength” from common taxa in partial pooled frameworks,
7
allowing the modeling of taxa that would erstwhile have too few presences for
8
modeling in “single-taxon” frameworks. However, the degree to which rare taxa truly
9
conform to the responses of common species remains an open question. Ergo, do these
10
methods mis-specifying the niches of rare taxa and thereby risk mis-representing their
11
vulnerability to anticipated global change?
12
● Niches could be more or less similar than expected by chance but splitting or lumping,
13
respectively, might still yield better niche estimates. How do we avoid conflating
14
allopatry with dissimilarity in environmental tolerances?
15 16 17
● Do phylogenetically informed partial pooling methods perform better than other methods when extrapolating to non-analog climates? ● How can we perform model selection for lumping/splitting/pooling applied to the same
18
set of taxa while controlling for confounding factors that vary according to the strategy
19
employed (e.g., sample size, study region extent, range size, etc.)?
20
● How do niches scale along phylogenetic hierarchies? Are constraints fundamentally
21
imposed at certain levels, and do lineages below that level adaptively “wander” within
22
those constraints?
1
23
● How can we capture niche similarities and differences arising simultaneously from
24
stabilizing selection and diversifying selection operating at different phylogenetic
25
levels?
2
Response to Reviewers
Ref.: Ms. No. TREE-D-18-00183 25 Sep 2018 Dear Adam, Thank you for your manuscript "Niche estimation above and below the species level". I am now enclosing three referee reports on your manuscript. Overall the referees' comments are very favourable, but they do raise a few points that I would be grateful if you could address. In addition to the comments of the referees I also have some minor editorial remarks which are either outlined below or on the electronically marked copy of the manuscript associated with your submission. To view the marked up manuscript: 1. Log into the Trends in Ecology & Evolution Editorial Manager site https://tree.editorialmanager.com/ Your username is: adamlilith If you need to retrieve your password, please click on the Send Username/Password link on the login page. 2. Go to "submissions needing revision" which will take you to your submission 3. Select file inventory in the action links, the marked up copy is labelled "Smith_et al_Editor's_comments". The comments can be displayed by selecting the track changes function. Please let me know if you have any problems. I hope you will be able to return the revised article by 16 Oct 2018. If you need longer, please let me know. Please also take this time to update your article with any new relevant papers. To submit your revised article, please go to "submissions needing revision" and click on the action link "submit revision". Please submit a word file of your text (i.e. not a pdf) including all box texts and figure legends. All figures should be submitted as separate files and NOT EMBEDDED in the text, stating the computer and software used in the preparation of the figures; please also provide a viewable version of the figures (ppt, pdf, jpeg or tif).
Thank you once again for your interesting and well written article. If you have any questions or problems, please don't hesitate to get in contact. I look forward to reading the revised manuscript. Best wishes, Caeul 1
------------------------------------------Caeul, Acting Editor Trends in Ecology & Evolution | Cell Press E-mail:
[email protected] Journal metrics: IF2017 15.938 | Citescore2017 10.35 | SJR2017 8.634 Follow @Trends_Ecol_Evo on Twitter Check out Figure360, an author-narrated animation of select figures, and offer your feedback here: www.cell.com/figure360 ------------------------------------------Editorial comments When revising the manuscript 1. Please clarify/consider the points highlighted by the referees and detail your response/the changes you introduced in a letter accompanying your revision. 2. From an editorial perspective, I thought that your article is overall very well written. I only have some minor points relating to our house-style. Please see the marked up version for specific comments. >>Thank you very much for your time in handling the manuscript. We made the changes you noted in the marked-up version (esp. typos, wording, and house style). In that regard please note that we have reconfigured what was Box 2 to Figure 1. The new Box 1 is included in response to a request for clarification about terminology by reviewer #1. >>Also note that we have revised the figure from former Box 1 (now Figure 1) so that readers are less likely to be confused between panel labels (A, B, and C) and the figures’ elements. You had originally suggested we use S1 and S2 (species 1, species 2) instead of B, A, and M, but a combination of three elements (B, A, M) refers to a single species, so we elected instead to use subscripts on the B, A, and M labels to refer to species. We also discarded the panel labels (A, B, and C) and instead use “top”, “middle” and “bottom”. This should make the difference between panels labels and element labels more distinct. >>We now list several applications of ENMs to issues conservation to emphasize that understanding niches has critical application in real-world situations (LL 65-69). 3.
Please note that each figure in the boxes should have a short explanatory title.
>>Done
2
4. Please note that no additional editorial work is carried out on supplementary material once the article is accepted. You should therefore ensure that the text is as accurate as possible and is consistent with the main article Please note that all references (peer-reviewed articles) should be numbered in order of appearance in the format [S1], and all other sources (such as software links) listed separately under “Resources” >>Done. 5. Optional Features While you are performing revisions on your article, I want to encourage you to make use of two new special features, the multimedia Figure360 and Interactive Questions. a. Figure360 Create a narrated, animated version of one of your figures that helps the reader zoom in on the most important take-home message in two minutes or less. The video should contain elements from only one figure, and include minimal introduction. For further guidelines and examples, please click here (http://www.cell.com/figure360). Then, simply upload your Figure360 along with your revised manuscript files. b. Interactive Questions (IQs) Maximize the educational value of your article by using the online “Interactive Questions” tool (http://authortools.elsevier.com/iq/verification). The tool helps you to create multiple-choice questions (usually 3-4) based on your review that are displayed in an interactive quiz next to your published article. Answers can also include True or False or All of the above and the (brief) explanation for the correct answer can be either the author’s thoughts or a key reference. You can see an example from another journal here http://authortools.elsevier.com/iq/verification/Sample_IQ.pdf Once you create the Q&A with the online authoring tool, please save the file as PDF and do not change the file name. Then upload the PDF together with your revised manuscript under the corresponding designation in the dropdown menu. To get more information about the tool and see examples please go to the "Extra Features" tab here: http://www.cell.com/trends/ecology-evolution/authors.
6. Please note that it is your responsibility to obtain permission to reproduce copyrighted material (i.e. figures, tables or excerpts that have been published online or in print) from the publishers of the original material. This is also relevant to figures that have been altered in any way (however, if additional information is added to figures it is sufficient to cite the original material in the figure legend). Note that while permission is usually always given, almost all publishers still require a formal request be made to reproduce material from their publications. Depending on the publisher, permission can be obtained in various ways as explained in our author guidelines. You should retain
3
the completed form on its return from the copyright holder. Please note that it is courteous to inform the author of the original material of your intent to use their published work. >>Elements of Figure I in Box 2 are from Mota-Vargas, C. and Rojas-Soto, O.R. 2016. Taxonomy and ecological niche modeling: Implications for the conservation of wood partridges (genus Dendrortyx). Journal for Nature Conservation 29:1-13. Reproduction rights are covered by an agreement to which Cell Press/Elsevier is a party (https://www.stm-assoc.org/copyright-legal-affairs/permissions/permissionsguidelines/). Also please note that before the initial submission we contacted and received well-wishes from Dr. Mota-Vargas in adopting the figure from his publication, and that he supplied us the photograph of the wood partridge. 7. The Publisher now requires authors to declare any conflicts of interest that relate to papers accepted for publication in this Journal. For more details see our Editorial Policies at http://www.cell.com/trends/editorial-policies#/ . There is no need to add any statement, if you have no conflict of interest. >>None declared. Referee Comments Reviewer #1: This manuscript covers an important and current topic. A summary like this is lacking but needed to push development and new thinking forward. This paper provides a balanced overview of the stateof-the-art, common pitfalls, and the most common options and best solutions available instead of conventionally modelling at the species level. It goes further to recognize the best ways forward and where the concentration on development should lay. The boxes are informative and easy to follow also for non-specialists. This paper would be of interest to a wide audience, ranging from ecological modellers and conservation practitioners to students and ecologists in large. >>We thank you for your time and enthusiasm for our work! The main scope of the manuscript does thus not need improvement. There are, however, some sentences and definitions, in the beginning of the abstract and in the glossary, which could be more precise. I find that some of the concepts are defined quite vaguely and unconventionally. I wonder if you could, e.g., re-evaluate the two first sentences in the abstract where you define the formation of the ecological niche and how this relates to ecological niche models. E.g., should the latter part of the first sentence, concerning ecological niches, perhaps adhere more to the evolutionary history of the specific species/OTU in question instead of whether it happens to share environmental preferences with other related species or not?
4
>> Thank you for pointing this out. The original statement did indeed seem to intonate that "niches are *defined* as adaptation plus niche conservatism". We have re-worded to indicate that niches are shaped by these processes, rather than being defined by them. LL 27-30. In the glossary, the definitions are not always very general but rather refer to a small part of what the specific term contains. I like the way you define niche conservatism - could you perhaps change the others more towards this direction, with a general part and a specific part, if needed, for conveying how this term relates to this topic? >> The Glossary now includes a general definition for most terms followed by a more nuanced definition specific to how we use the term in the manuscript (LL 597+). In addition, we now include a new Box 1 (LL 646+) which disentangles a large number of terms (especially with regards to what we now call “partial pooling”). Also, please note that in response to advice from another reviewer, we now use “partial pooling” to refer to multi-response modeling (though occasionally use “pooling” to avoid overuse of alliteration with “p”). Line 67: You say that three modelling strategies have been used to address the concern of including evolutionary relationships in ecological modelling. I understand that this is not a literature review, but I am still curious to know how you have identified these three main methods? And what is the temporal scale of this development - can you give a small summary of the history of recognizing the need to account for other units than species within the field of ecological niche modelling? >>We have reworded the introduction to this sentence so that it indicates we broadly classify strategies for modeling above and below the species level (as opposed to the previous wording which intonated that these are the entire set of strategies; LL 70-71). Also, we now briefly recapitulate the history of each strategy in niche modeling with the respective sections (splitting/lumping: LL 102-109, pooling: LL 181-184). A few minor comments: Line 43: consider exchanging "environments" with "conditions" or "environmental conditions" >>Done. Line 105-106. Is there a word (models?) missing somewhere? >>Done. Line 127: "it may BE preferable"? >>Done. Line 205-206: "the need TO make"? 5
>>Done. Line 491: I suggest removing first names of authors in the reference >>Done. Line 519: "niches THAT differ"? >>Done.
Reviewer #2: The manuscript "Niche Estimation Above and Below the Species Level" reviews the pros and cons of estimating niches at different taxonomic levels and describes three methods that researchers employ to do so: splitting taxa into subunits, lumping across taxa, and pooling taxa. The authors make the argument that phylogenetically informed pooling methods are superior and they provide a simple simulation to illustrate their points. This is a well written manuscript describing an emerging topic. While using multi-response models is becoming more common, it is rarely used to estimate niches. The authors do an excellent job summarizing where the field is now and cite the appropriate papers for this topic. Overall, I approve this manuscript for publication but I do have some minor comments: >> We thank you for the positive assessment of our manuscript and especially for your time. 1. The term 'pooling' is often used in niche modeling studies in the same way 'lumping' is used in this manuscript. In addition, in the paleontological literature, 'pooling' refers to using occurrences from multiple time periods. Given this, I suggest finding a different term for multi-response methods, cite papers that use the 'pooling' terminology in reference to these models, or clearly indicate how pooling used in this manuscript is different than how it is used in other niche analysis studies. >>We agree that the existing terminology is imperfect. In the new manuscript we now try to mitigate this ambiguity by using the term “partial pooling” (though sometimes use “pooling” for short). We believe that this minor change emphasizes the distinction from lumping. We have also added a new Box 1 which disentangles the varied terms used for this type of modeling (LL 646+). Although there are variety of works we could cite to justify use of any one of these terms, we have chosen to use “partial pooling” based on Gelman & Hill (2007, Data Analysis Using Regression and Multilevel/Hierarchical Models, Cambridge—cited in our main text), an often-cited and definitive text. 2. While I agree that using phylogenetically informed 'pooling' methods is the superior method, a discussion on the inherent assumptions, biases, and potential pitfalls of this approach is needed. In
6
addition, is some cases it is appropriate to model a niche at the species level given the question of interest, the system and the taxon. This should be explained following Line 227. >>Thank you for offering a check to our enthusiasm for this technique. Partial pooling methods are not a panacea for all circumstances. We now include a new paragraph highlighting some of the potential pitfalls and disadvantages of partial pooling (LL 251-260). 3. There are several examples of lumping taxa in the paleo-niche literature and some of these studies also incorporate or infer evolutionary processes into niche estimation. Citing this literature would enhance the discussion. >>Thank you for noting this omission. In response to these comments and to comments from Reviewer #3 we now include discussion (and citations) of issues in the paleo literature that have benefited from these kinds of strategies (especially lumping but also partial pooling; examples: LL 99-101 and LL 651652). 4. Line 165-167 - This statement is not exclusive - niche similarities can arise from processes other than shared phylogenetic history. >> We agree and now make clear that the ability to account for shared phylogenetic history is one advantage, not the only advantage (LL 227-239). Edits: Line 55 - change to "…high levels of niche conservatism" >>Done Line 74 - add in citation(s) for the term "pooling" if available. >>Done Line 105 - change to "Splits for the same taxon…" >>Done Line 126 - change to "…provide an accurate…" >>Done Line 205 - change to "…pooling obviates the need to make strong…" >>Done 7
Line 233-237 - Excellent point, thank you for highlight this. Suggestions for testing models then? >>Thanks! We now suggest several methods by which to assess the ability of models to recreate niches (vs. distributions; LL 287-290). We also emphasize this in the last bullet point in the Highlights. This is obviously a topic that deserves further, lengthy treatment. Line 519 - change to "…may have realized niches that differ." >>Done Line 523 - change to "In panel (C)…" >>Done
Reviewer #3: Overall - This paper is a nice review of the literature to date on the state of niche modeling with regard to taxonomy. The paper is well written and the citations seem sufficient. >>Thank you very much for your time and careful consideration of our manuscript. However, I think the discussion is a bit incomplete. It would be helpful to readers if you added a discussion of what these models are typically used for, and what the impacts to previous conservation actions might be from a recapitulation of niche/SDMs to date that incorporates the approaches reviewed here. Perhaps there are situations where incorporation of the approaches outlined would be particularly helpful or would hinder? >>In response to this comment and from Editor Caeul we have included examples of applications of niche models in conservation (LL 65-69). The main part of the manuscript does indeed discuss the potential advantages of each approach (e.g., splitting: LL 91-93; lumping: LL 94-99; and partial pooling: LL 227-239). However, to ensure balance we now also note that even if these methods produce better niche estimates, applying them may impose a high cost for existing conservation plans and policies which tend to be focused on the species level (LL 261-274). There have also been initiatives in vegetation science that attempt to model vegetation ensembles, often in response to climate change. While this is not the focus of your paper, I wonder whether more explicit recognition of the taxonomic range of modeling would help identify the niche of this paper. >>Thanks for noting this—vegetation type modeling is indeed beyond the scope of our paper yet some of the models used for this type of work are amenable to partial pooling (e.g., multinomial models). We now note the similarities between vegetation modeling and lumping (LL 104-106 and 659-662). Also, in 8
response to other reviewers’ comments we include a new Box 1 which disentangles the many terms used for modeling communities (for vegetation modeling see especially LL 659-662). Box 1 emphasizes that we are focused on sets of relatively closely-related species (versus, say, all plants in a set of locations; LL 662-670). Abstract Suggest replacing "preferences" with "requirements" or "tolerances" or something similar - maybe "suitable environmental conditions" >>Done If you have space, suggest stronger ending by suggesting when alternate phylogenetic levels might be more suitable than species for niche estimation. OK if you don't have words to do so. >>As noted above the main text emphasizes the potential advantages of each approach, but we now include a paragraph discussing potential pitfalls of employing split, lumped, and partially pooled models (LL 251-274). Line 55-56 - insert "will likely respond…" >>Done Line 63 - suggest you also give some credit to the records held, which typically are available at the species level and may have therefore restricted researchers in their modeling efforts. Note that there also has been some commentary from people such as Art Shapiro about the conservation costs of taxonomic revisions - if lands have already been set aside for conservation because of assessments at the species level (prior to phylogenetic assessments) what is the cost of a reclassification. The sequential editions of California's Jepson Flora are an example where a large percentage of the flora was re-classed. These considerations may be appropriate in the discussion. >>We now note that decisions to model at a particular level are often necessitated by the available data (LL 163-165 and indirectly in 215-217). Line 103 - "when distributional data is adequate, split models…" >>Done 191 -192 - Is this assertion always going to be true? I'm trying to think of cases where taxa in the same subclade might represent competitive interactions and therefore might be themselves a negative influence on the other. Maybe Darwin's finches' beaks represent a group who would not be exclusionary to themselves, but are there examples (serpentine species perhaps?) where they might be. I agree that the assumption you relay is correct, but wonder if this point needs expanded. 9
>>Original statement: “Mirroring phylogenetic structure in the model structure inherently assumes attraction since taxa in the same subclade are more likely to have similar coefficients.” >> New version: “Other algorithms implicitly reflect the hierarchical nature of evolutionary relationships by using nested model structures [3, 92] (e.g., by means of nested random effects [92] or more complicated approaches [3]). Mirroring evolutionary relationships within model structure implicitly assumes attraction between lineages in the same subclade since members of a subclade tend to have similar coefficients.”) LL 210-215 >>We agree that in many cases repulsion between related taxa may be the norm. However, if phylogenetic structure is mirrored by model structure, then in most cases related taxa will tend to have similar coefficients. We are thinking especially of Godoy et al. (2018--in references) who used a Bayesian hierarchical model in which each level reflected (roughly speaking) family, superfamily, etc. At each MCMC iteration the proposal value (mean, precision) for a given level was used to construct the prior distribution for the next-lower level. In this manner taxa tended to have similar coefficients since they came from the same hierarchical family of distributions. Bates et al. (2010) used a similar approach. >>It is probably possible to reflect phylogenetic repulsion in model structure (as opposed to, say, doing so using the VCV matrix), but this seems difficult to do. For example, the model structure that would tend to impart the most distinctive niche estimates would probably be a mixed-effects framework with each taxon represented as a random slope and/or intercept. However, even in this case the responses (to the fixed effects) should be more similar than if taxa were modeled completely apart as in a split framework (please compare the “Splitting” and “Pooling” panels in Figure I of Box 3). 216 - "if" should be "of" >>Done Highlights - I think the last bullet is rather too strongly worded. It seems to confound phylogeny with biological plausibility. Are the two both influential and therefore calling this "an important" criterion would be appropriate? Or renaming biological plausibility to something closer to the review focus (taxonomic level) be useful? >>We think we see where our original wording caused confusion… We have re-phrased the last bullet point to emphasize that the “biological plausibility” against which niche models should be judged refers to the plausibility of the estimated niche envelope. >>Original statement: “Biological plausibility should be the primary criterion by which niche models are evaluated. Presently models are usually assessed solely on their ability to recreate geographic distributions, which can be shaped by factors unrelated to the niche.”
10
>>Revised: “The advancement of evolutionarily-informed niche models will be accelerated by emphasizing their ability to generate biologically plausible niche envelopes over their ability to re-create geographic distributions which can be shaped by factors unrelated to the fundamental niche.”
11
Highlights
1
Highlights
2
● Modeling niches at the species level disregards information about evolutionary
3 4 5 6 7 8 9 10 11
processes above and below the level of species. ● Species-level models ignore the potential for locally adaptive responses and assume current distributions reflect the entire set of suitable conditions. ● In many cases splitting taxa into subunits and modeling each separately or lumping related taxa can improve niche estimates. ● Partial pooling of lineages into a single multi-response framework has underutilized potential for niche estimation, especially when accounting for phylogenetic relationships. ● The advancement of evolutionarily-informed niche models will be accelerated by
12
emphasizing their ability to generate biologically plausible niche envelopes over their
13
ability to re-create geographic distributions which can be shaped by factors unrelated
14
to the fundamental niche.
1
Author Supplementary Material
Click here to access/download
Author Supplementary Material Online Supplemental Material.pdf
Figure 1
Click here to access/download;Figure;Figure 1.jpg
Box 2 Figure I
Box 2 Figure I
Dendrortyx macroura
Genus- and species-level niches genus
0
-3
4
Climate Axis 2
Climate Axis 2
3
Photo credit: Karla Patricia Parra Noguez
Niches of D. macroura and its subspecies
-3
0
Climate Axis 1
3
species
2
0
-2
-4
-2
0
Climate Axis 1
2
Box 3 Figure I
Phylogeny
True niches
1.00
Suitability
0.75
0.50
0.25
0.00 0
5
10
15
20
Temperature
1.00
1.00
0.75
0.75
0.50
0.50
0.25
0.25
0.00
0.00 0
5
10
15
20
0
10
15
20
Temperature
Pooling (GLMM)
Phylogenetic Pooling (PGLMM)
1.00
1.00
0.75
0.75
0.50
0.50
0.25
0.25
0.00
0.00 0
5
Temperature
Suitability
Suitability
Lumping (GLM)
Suitability
Suitability
Splitting (GLM)
5
10
Temperature
15
20
0
5
10
Temperature
15
20
Box 4 Figure I
Time
Brownian motion
Time
Orstein-Uhlenbeck
Speciation events
Speciational
Temperature
