spiral.imperial.ac.uk€¦ · Web viewJack H. HATFIELD*, C. David L. ORME, Joseph A. TOBIAS,...
Transcript of spiral.imperial.ac.uk€¦ · Web viewJack H. HATFIELD*, C. David L. ORME, Joseph A. TOBIAS,...
Trait-based indicators of bird species sensitivity to habitat loss are
effective within but not across datasets
Jack H. HATFIELD1*, C. David L. ORME1, Joseph A. TOBIAS1, Cristina BANKS-LEITE1
1Department of Life Sciences, Imperial College London, Silwood Park Campus, Ascot, SL5
7PY, UK.
*Corresponding author – [email protected]
1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Abstract
Species’ traits have been widely championed as the key to predicting which species are most
threatened by habitat loss, yet previous work has failed to detect trends that are consistent
enough to guide large-scale conservation and management. Here we explore whether traits and
environmental variables predict species sensitivity to habitat loss across two datasets generated
by independent avifaunal studies in the Atlantic Forest of Brazil, both of which detected a similar
assemblage of species, and similar species-specific responses to habitat change, across an over-
lapping sample of sites. Specifically, we tested whether 25 distributional, climatic, ecological,
behavioural and morphological variables predict sensitivity to habitat loss among 196 bird spe-
cies, both within and across studies, and when data were analysed as occurrence or abundance.
We found that 4-9 variables showed high explanatory power within a single study or dataset, but
none performed as strong predictors across all datasets. Our results demonstrate that the use of
species traits to predict sensitivity to anthropogenic habitat loss can produce predictions that are
species- and site-specific and not scalable to whole regions or biomes, and thus should be used
with caution.
Key-words: Atlantic Forest, birds, habitat fragmentation, habitat loss, response traits, species
sensitivity, transferability
2
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Introduction
Species differ in their response to habitat loss (Banks-Leite et al. 2012) and understanding the
basis of this difference is vital to comprehending why species are being lost and how to concen-
trate conservation effort. Many studies have examined which traits make a species more sensitive
to human disturbance (see Henle et al. (2004) for review), with the aim of predicting which spe-
cies will require future conservation intervention. While some studies focused on just a few traits
(Stouffer and Bierregaard 1995a, Castelletta et al. 2000, Ribon et al. 2003, Sodhi et al. 2004),
others have considered a larger numbers of traits (Pakeman 2011, Edwards et al. 2013), trait in-
teractions (Renjifo 1999) and multiple taxa (Faria et al. 2006, Senior et al. 2013). Regardless of
breadth of analyses, taxa and study region, previous studies have not found consistent results
(e.g. Newmark (1991), Senior et al. (2013)). Most importantly, a large number of studies based
their conclusions on observations from a single community or region (Renjifo 1999, Kennedy et
al. 2010), and generally lacked an assessment of model transferability to estimate the reliability
of traits as proxies for species sensitivity.
The most commonly explored traits include those related to population demography, dis-
persal, sociality, body size, trophic level and specialisation, habitat usage, species interactions
and biogeography (Henle et al. 2004). These traits are believed to be reasonable proxies for spe-
cies sensitivity because they are related to species adaptability or the likelihood of local extinc-
tion. Indeed, some authors have found trends to be consistent across studies and regions (Gray et
al. 2007, Williams et al. 2010), such as an increased sensitivity to fragmentation in specialised
taxa, including butterflies with narrow feeding niches (Ockinger et al. 2010). Others, however,
have found that although traits can be useful indicators, the patterns may be specific to certain
areas (Sigel et al. 2010, Thornton et al. 2011, Vetter et al. 2011). For example, birds that join
3
383940
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
mixed species flocks are considered highly sensitive to habitat changes, but Sigel et al. (2010)
only found this to be the case in one of the areas studied. Hence, while this approach may
provide insight into the dynamics of local systems, contradictions arise due to species- and site-
specificity in the relationship between traits and anthropogenic disturbances, or due to the differ-
ent methods used by different authors.
Here, we perform a direct comparison of transferability of trait-based models across two
datasets generated by intensive bird surveys in the Atlantic Forest of Brazil. The Atlantic Forest
has been extensively studied, so provides a rich testing ground for model transferability. The
datasets used follow similar experimental designs and were collected in the same study region at
an overlapping sample of sites, but used different methods for data collection. Thus, we (1) in-
vestigate which traits can be used to predict bird species sensitivity to habitat loss, and then (2)
assess model transferability by testing whether traits that are strong predictors of species sensit-
ivity in one dataset maintain their predictive power in different datasets. We performed the above
tests on four different subsets of the data: occurrence or abundance data of all recorded species or
just the species that were observed in both studies. Lack of model transferability between data-
sets could result from traits being spuriously correlated to species sensitivity, or from differences
in species sensitivities between datasets. The latter is particularly problematic to these analyses
given that different survey methods were used, which influences the probability of species detec-
tion and therefore the probability of detection of ecological trends (Banks-Leite et al. 2014). For
this reason, we also examined the consistency of species sensitivity to habitat change across
studies.
Methods
Study area: Our data originates from two independent studies conducted in the Atlantic Plateau
4
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
of the State of São Paulo, Brazil. The survey area is broadly the same for both studies, including
19 overlapping study sites (Appendix S1: Fig. S1). Elevation ranges from 700 to 1000 m and the
original vegetation is lower montane rainforest. Both studies were designed with similar spatial
sampling protocols, comparing paired 10,000 ha blocks of continuous and fragmented forest
landscape, but differed in the main method of data collection:
i) Mist-net study (Banks-Leite et al. 2011). Data were collected from 2001 to 2007 in 65
study sites varying from 5 to 100% of forest cover (800 m radius). Nearly 8000 individuals of
140 species were sampled during 41,000 hours of mist netting (Appendix S1).
ii) Point count study (Develey 2004). Data were collected from 2000 to 2003 in 32 sites
varying from 20 to 100% of forest cover (800 m radius). Birds were sampled using 10 minute
point counts, with 20 point counts carried out per site, resulting in estimates of abundance for
154 species. (Appendix S1)
Trait data: Bird surveys produced a total list of 197 species, including 43 species unique to mist
netting, 57 unique to point counts, and 97 shared across datasets. One species, Picumnus cir-
ratus, was removed from the analysis because of missing data. Our final dataset contained in-
formation on 25 ecological or phenotypic variables for 196 bird species (Appendix S2: Table
S1). These variables included information on geographical distribution, as well as a range of
traits chosen to reflect the morphology and ecology of species, including their foraging habits
and behaviour. A similar mix of environmental variables and species traits are widely used to
predict sensitivity to habitat change (Turner 1996, Henle et al. 2004). For convenience, we use
the term ‘traits’ when environmental variables and species traits are referred to collectively.
Global species’ distribution area was measured using ESRI ArcGIS v.10.0. from standard
range polygons (BirdLife International and NatureServe 2012). We then used MaxEnt v.3.3.3k
5
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
(Phillips et al. 2006) to predict the bio-climatic suitability of each survey site for all species. Spe-
cies distribution models were produced using the MaxEnt auto features option and the full set of
19 temperature and rainfall derived Bioclim variables (Hijmans et al. 2005). Cost distance ana-
lysis was used to find distance to nearest terrestrial distributional range edge (i.e. proximity to
the edge of distribution; PED) from the centre point of both study areas. This was conducted us-
ing R v.3.3.2 (R Core Team 2016) and the packages raster (Hijmans and van Etten 2013), gdis-
tance (Van Etten 2012) and maptools (Bivand and Lewin-Koh 2013), as well as ESRI ArcGIS
v.10.0., defining coastlines using the admin98 boundaries layer. Proximity to range edge (PED)
and distributional range size was logarithmically transformed (ln +1 for PED).
Morphological traits were compiled by measuring museum specimens following standard
protocols (Bregman et al. 2015, Bregman et al. 2016, Ulrich et al. 2017). We used three bill
measurements (length, width and depth), tarsus length, tail length and wing length (for a more
detailed description of measurements, see Appendix S2: Table S1). We also calculated Hand
Wing Index as an index of dispersal ability using a combination of wing length and first second-
ary length, following Claramunt et al. (2012). Measurements were accurate to 0.01mm, except
tail and wing which were to the nearest mm. Where possible, each measurement was averaged
from four specimens with an even sex ratio. Where trait information was unavailable, values
were taken from closely related and ecologically similar congeneric species, or, in the case of
some hummingbird species in monotypic genera, tarsus measurements were taken from similar
sized species in closely related genera. Body mass was compiled from literature (see Appendix
S2: Table S1). All morphological traits including body mass were logarithmically transformed
for analyses, except HWI.
Diet composition information (Appendix S2: Table S1) was subject to a PCA on the whole
6
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
species pool to obtain independent variables summarising the diet composition of each species
relative to the rest of the community using the package vegan (Oksanen et al. 2013).
Statistical analysis: All statistical analyses were run using R v.3.3.2 (R Core Team 2016) with
packages vegan (Oksanen et al. 2013), caper (Orme et al. 2013), ape (Paradis et al. 2004) and
phytools (Revell 2012).
We used a novel approach to select the traits and calculate a species-specific score of
sensitivity to habitat changes for each of species’ abundance and occurrence (Fig. 1). The ap-
proach was based on weighted averages ordination (Gauch 1982), in which Bray-Curtis and
Sørensen dissimilarity between sites was first calculated from each of the community abundance
and incidence matrices, respectively. A Principal Coordinates Analysis (PCoA) was then con-
ducted on each dissimilarity matrix. We assumed that the position of each site along the first
PCoA axis reflects its position on a gradient of habitat change based on the previous finding that
over 90% of variance in the PCoA axis 1 of the mist-nest data is explained by landscape metrics
(Banks-Leite et al. 2011). The sensitivity of each species, or the weighted average, was then
taken as the average PCoA score of the sites in which it was found, hence representing its mean
position along the habitat change gradient. Where abundance data was considered, the value for a
species at each site was calculated by weighting the site score by the species abundance at that
site. The directionality of the ordination axes were checked (and inverted if required) to ensure
that an increasing sensitivity score indicated increasing sensitivity to habitat change. This
method produces results that are qualitatively similar to fourth-corner methods and RLQ (Dray et
al. 2014) (Appendix S3), but has the advantage of allowing us to control for phylogenetic non-in-
dependence and allows for more flexibility in model building.
First, we investigated which species traits can be used to predict species’ sensitivity to
7
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
fragmentation using Phylogenetic Generalized Least Squares (PGLS) models with traits as ex-
planatory variables. Phylogenetic trees were obtained as phylogeny subsets from BirdTree.org
(Jetz et al. 2012): this dataset provides a posterior distribution of phylogenies from Bayesian ana-
lyses. For each set of species, we therefore selected 1,000 trees and calculated a consensus tree
using a 10% burn-in, maximum clade credibility, median node heights and a posterior probabil-
ity limit of 0.5, using TreeAnnotator v.1.8.3 and FigTree v. 1.4.2. from the software BEAST
(Drummond et al. 2012). Consensus trees calculated from a large number of trees have been
found to provide the same results as model averaging over multiple trees (Rubolini et al. 2008).
In this case a consensus tree was used as it allowed greater clarity in model simplification and
the influence of phylogeny used was found to be relatively low for the majority of analyses.
The traits to be included in the PGLS models were checked for collinearity. All traits
that were not highly correlated (r < 0.7 (Dormann et al. 2013)) were fitted to form the maximal
model. The lambda value of the maximal model was determined using maximum likelihood and
subsequently fixed to allow for model simplification. Step-wise simplification was conducted
based upon p-values, by using the ‘anova’ function to compare between models. The minimum
adequate model was considered to be that in which all terms were significant at the 0.05 level.
This method was preferred over AIC model selection because of difficulties comparing models
with different random factors (i.e. phylogenetic covariance structure). We did not fit interaction
terms as the dataset was not large enough.
Model Transferability: We first obtained the R2 from the minimum adequate PGLS model gener-
ated by one dataset, and then applied the same set of selected predictors to obtain the R2 from the
other dataset. The difference in R2 between these models was used to assess the retention of ex-
planatory power for transferred predictions. This process was conducted in both directions (Fig
8
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
1). We initially fitted the models using all of the species detected in each dataset to better under-
stand the extent to which we can generalise knowledge on response traits from a given bird com-
munity. To ensure that any differences to R2 were not simply a product of changes in phylogen-
etic covariance, we also re-ran models restricted to the 96 species common to both datasets. We
tested for the correlation between sensitivity scores for this 96-species sample using the occur-
rence data and a Pearson correlation to examine whether survey method influences our measures
of species sensitivity or confounds model transferability.
Results
Species Sensitivity: For the shared sampled (n = 96), species’ sensitivity scores were found to be
highly correlated between datasets (r = 0.86) indicating that observed species responses to habitat
changes are not dependent on the sampling method, and thus that sampling method is unlikely to
influence model transferability.
All Species: For occurrence data, the minimum adequate model for the mist-net data contained 8
variables (adjusted R2 = 0.34) and the model for the point count data contained 6 variables (adj.
R2 = 0.43). Some variables (number of habitats used, open habitat usage, bill width, tarsus
length) were common to both models. When transferred to the other dataset, the explanatory
power of the model parameterised on the mist-net data increased to 0.38 (despite the inclusion of
non-significant variables) but the explanatory power of the model parameterised on the point
count data fell to 0.26. Hence we detected a 12% increase in explanatory power for one model,
and a 38% decrease for the other model.
Using abundance data, seven traits were retained in the minimum adequate models of
species sensitivity using the mist-net data (adj. R2 = 0.27), and five traits were retained using the
point count data (adj. R2 = 0.38). When transferred, both models lost performance, with explanat-
9
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
ory power falling to R2 values of 0.18 and 0.18, respectively (representing a drop of 35% and
53% in explanatory power). Again, the set of variables differed between models, and only ant
following behaviour and understory usage were significant in both models.
No variables were retained across all models (Appendix S2: Table S2).
Species present in both datasets: For occurrence data, the minimum adequate model for the mist-
net data contained 6 variables (adj. R2 = 0.46) while the model for the point count data contained
9 variables (adj. R2 = 0.51). The variables that were consistent in significance and sign to both
models were: ant following behaviour, diet composition, understory usage, open habitat usage
and number of habitats used. When transferred to the other dataset, the explanatory power of the
model parameterised on the mist-net data dropped to 0.39 and the explanatory power of the
model parameterised on the point count data fell slightly to 0.47. Thus, a 14% decrease in ex-
planatory power was found for the mist-net model and a 7% decrease for the point count model.
For abundance data, the minimum adequate model for mist-net data contained 4 variables
(adj. R2 = 0.30), while the minimum adequate model for the point count data contained 5 vari-
ables (adj. R2 = 0.44). Again there was a considerable difference between the sets of variables
(Appendix S2: Table S2); only ant following behaviour was significant in both datasets. When
we tested for model transferability, we found the explanatory power of these two models fell to
0.22 and 0.25, respectively (representing a 28% and 43% drop in explanatory power). Across
models of abundance and occurrence data on both datasets, only ant following behaviour was
found to be consistently significant and positively associated with the disturbed to intact habitat
gradient.
Phylogenetic non-independence: In general, models showed little evidence of phylogenetic sig-
nal in the data (λ, typically 1 x 10-6 or lower, and not significantly different from zero). Values of
10
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
λ were also consistent through model simplification, suggesting that variable combinations do
not differ markedly in their phylogenetic pattern. The abundance data models using point count
data were exceptions. The model for all species showed a range of λ values between the maximal
and minimal models (0.547-0.562), as did the model for the species common to both datasets
(0.358-0.344). However, this had minimal effects on the eventual model structure; all terms were
still retained using a λ value calculated under maximum likelihood. This is likely due to the un-
certainty associated with the phylogenetic structure of the species in this dataset, but only had a
slight effect upon the results.
Discussion
Our results show that response variables can be strongly correlated to measures of species sensit-
ivity to habitat loss and fragmentation at local scales, in line with previous studies (Stouffer and
Bierregaard 1995b, Kennedy et al. 2010). However, although it is often proposed that species
traits can therefore be used to predict sensitivity at wider spatial and temporal scales (Castelletta
et al. 2000, Ockinger et al. 2010), our results indicate that this type of extrapolation may not be
possible given the large (up to 53%) reduction in explanatory power when transferring models
between datasets.
In general, we found evidence that a wide range of distributional, behavioural, dietary and
morphological variables can be used to indicate species sensitivity at least locally, corroborating
the findings of previous studies (Henle et al. 2004, Sekercioglu et al. 2004, Gray et al. 2007).
The best indicators in our study appear to be number of habitats used, open habitat usage and ant
following behaviour, as these variables were the most consistently effective across different data-
sets. Nonetheless, there was a lack of consistency of trait predictors between datasets and data
types, with no single trait performing as a significant predictor for all dataset or data type com-
11
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
binations.
This level of inconsistency may help to explain why our models generally had low transfer-
ability, performing most consistently when transferred between datasets containing exactly the
same species (particularly when using occurrence data). There are a number of possible reasons
for limited transferability between datasets. Survey method, and its associated effects on species-
specific detection probabilities, may account for some of the reduced transferability, but not all.
The high correlation in species sensitivity scores between datasets shows that the use of different
survey methods is not biasing or confounding our ability to detect species sensitivity to habitat
changes. Thus, the fact that we found strong explanatory power of response variables but low
transferability of models suggests that trait-sensitivity correlations may be spurious and not gen-
eralizable. Another issue is the high collinearity of the response variables, which may be the
reason why under certain circumstances model transfer does not lead to a large drop in explanat-
ory power but instead to conflicting sets of variables included in final models. This was observed
with occurrence data for the 96 species where the model parameterised on the point count data
only suffered a minor drop in explanatory power using the mist-net data, yet the mist-net data
had yielded a minimal model containing six variables, whereas the point count model contained
nine variables, with five variables common to both.
The results of this study add to the growing consensus of site specificity in the literature. For
example, Vetter et al. (2011) shows that that vertebrate responses to fragmentation can be highly
site specific. Thornton et al. (2011) went further to show that certain traits of Neotropical mam-
mals were more or less influential across different sites in Guatemala and Mexico even when the
same species are considered, thus reinforcing the idea that traits may not be consistent predictors
of species responses. As for Neotropical birds, Sigel et al. (2010) showed that some response
12
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
traits can be extremely site specific and that patterns of variation were best interpreted in the
local context.
To conclude, our results suggest that response traits should be used with caution if the aim is
to understand mechanisms of species sensitivity to habitat loss and fragmentation. In particular,
our analyses indicate that even when species traits predict sensitivity to habitat change, these pat-
terns may be inconsistent even within study communities or regions, and cannot simply be extra-
polated to predict species sensitivity at broader spatial or taxonomic scales. Thus, although trait-
based proxies have become popular short-cuts in conservation forecasting and priority-setting
exercises, their application across multiple taxa, regions and time (Henle et al. 2004) may be
more complex than first anticipated. The validity and transferability of trait-based models of spe-
cies sensitivity to habitat change requires more thorough investigation.
Acknowledgements
We thank P. F. Develey for the point count data; and A. Aleixo, T. Bregman, B. Darski and C.
Sheard for help compiling biometric data. We are also grateful to the Natural History Museum,
UK, the American Museum of Natural History, USA, and the Museu Paraense Emílio Goeldi,
Brazil, for access to specimens. This research was supported by the Natural Environment Re-
search Council (NE/H016228/1, NE/K016393/1 and NE/K016431/1), and a Marie Curie Interna-
tional Incoming Fellowship in the 7th European Community Framework Programme.
References
Banks-Leite, C., R. M. Ewers, V. Kapos, A. C. Martensen, and J. P. Metzger. 2011. Comparing
species and measures of landscape structure as indicators of conservation importance.
Journal of Applied Ecology 48:706-714.
Banks-Leite, C., R. M. Ewers, and J. P. Metzger. 2012. Unraveling the drivers of community dis-
13
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287288289
290
291
292
similarity and species extinction in fragmented landscapes. Ecology 93:2560-2569.
Banks-Leite, C., R. Pardini, D. Boscolo, C. R. Cassano, T. Puttker, C. S. Barros, and J. Barlow.
2014. Assessing the utility of statistical adjustments for imperfect detection in tropical
conservation science. Journal Applied Ecology 51:849-859.
BirdLife International and NatureServe. 2012. Bird species distribution maps of the world.
BirdLife International, Cambridge, UK and NatureServe, Arlington, USA.
Bivand, R., and N. Lewin-Koh. 2013. maptools: Tools for reading and handling spatial objects.
R package version 0.8-23. URL http://CRAN.R-project.org/package=maptools.
Bregman, T. P., A. C. Lees, H. E. MacGregor, B. Darski, N. G. de Moura, A. Aleixo, J. Barlow,
and J. A. Tobias. 2016. Using avian functional traits to assess the impact of land-cover
change on ecosystem processes linked to resilience in tropical forests. Proceedings of the
Royal Society B: Biological Sciences 283.
Bregman, T. P., A. C. Lees, N. Seddon, H. E. MacGregor, B. Darski, A. Aleixo, M. B. Bonsall,
and J. Tobias. 2015. Species interactions regulate the collapse of biodiveristy and ecosys-
tem function in tropical forest fragments. Ecology 96:2692-2704.
Castelletta, M., N. S. Sodhi, and R. Subaraj. 2000. Heavy extinctions of forest avifauna in Singa-
pore: Lessons for biodiversity conservation in Southeast Asia. Conservation Biology
14:1870-1880.
Claramunt, S., E. P. Derryberry, J. V. Remsen, Jr., and R. T. Brumfield. 2012. High dispersal
ability inhibits speciation in a continental radiation of passerine birds. Proceedings of the
Royal Society B: Biological Sciences 279:1567-1574.
Develey, P. F. 2004. Efeitos da fragmentação e do estado de conservação da floresta na diversi-
dade de aves de Mata Atlântica. Universidade de Sãu Paulo, Sãu Paulo.
14
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
Dormann, C. F., J. Elith, S. Bacher, C. Buchmann, G. Carl, G. Carré, J. R. G. Marquéz, B. Gru-
ber, B. Lafourcade, P. J. Leitão, T. Münkemüller, C. McClean, P. E. Osborne, B. Reinek-
ing, B. Schröder, A. K. Skidmore, D. Zurell, and S. Lautenbach. 2013. Collinearity: a re-
view of methods to deal with it and a simulation study evaluating their performance.
Ecography 36:27-46.
Dray, S., P. Choler, S. Doledec, P. R. Peres-Neto, W. Thuiller, S. Pavoine, and C. J. ter Braak.
2014. Combining the fourth-corner and the RLQ methods for assessing trait responses to
environmental variation. Ecology 95:14-21.
Drummond, A. J., M. A. Suchard, D. Xie, and A. Rambaut. 2012. Bayesian phylogenetics with
BEAUti and the BEAST 1.7. Molecular Biology and Evolution 29:1969-1973.
Edwards, F. A., D. P. Edwards, K. C. Hamer, and R. G. Davies. 2013. Impacts of logging and
conversion of rainforest to oil palm on the functional diversity of birds in Sundaland. Ibis
155:313-326.
Faria, D., R. R. Laps, J. Baumgarten, and M. Cetra. 2006. Bat and Bird Assemblages from
Forests and Shade Cacao Plantations in Two Contrasting Landscapes in the Atlantic For-
est of Southern Bahia, Brazil. Biodiversity and Conservation 15:587-612.
Gauch, H. 1982. Multivariate Analysis in Community Ecology. Cambridge University Press,
New York.
Gray, M. A., S. L. Baldauf, P. J. Mayhew, and J. K. Hill. 2007. The response of avian feeding
guilds to tropical forest disturbance. Conservation Biology 21:133-141.
Henle, K., K. F. Davies, M. Kleyer, C. Margules, and J. Settele. 2004. Predictors of species sen-
sitivity to fragmentation. Biodiversity and Conservation 13:207-251.
Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones, and A. Jarvis. 2005. Very high resolution
15
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
interpolated climate surfaces for global land areas. International Journal of Climatology
25:1965-1978.
Hijmans, R. J., and J. van Etten. 2013. Raster: Geographic data analysis and modelling. R pack-
age version 2.1-2.5 URL http://CRAN.R-project.org/package=raster.
Jetz, W., G. H. Thomas, J. B. Joy, K. Hartmann, and A. O. Mooers. 2012. The global diversity of
birds in space and time. Nature 491:444-448.
Kennedy, C. M., P. P. Marra, W. F. Fagan, and M. C. Neel. 2010. Landscape matrix and species
traits mediate responses of Neotropical resident birds to forest fragmentation in Jamaica.
Ecological Monographs 80:651-669.
Newmark, W. D. 1991. Tropical Forest Fragmentation and the Local Extinction of Understory
Birds in the Eastern Usambara Mountains, Tanzania. Conservation Biology 5:67-78.
Ockinger, E., O. Schweiger, T. O. Crist, D. M. Debinski, J. Krauss, M. Kuussaari, J. D. Petersen,
J. Poyry, J. Settele, K. S. Summerville, and R. Bommarco. 2010. Life-history traits pre-
dict species responses to habitat area and isolation: a cross-continental synthesis. Ecology
Letters 13:969-979.
Oksanen, J., F. Guillaume Blanchet, R. Kindt, P. Legendre, P. R. Minchin, R. B. O'Hara, G. L.
Simpson, P. M. Solymos, H. H. Stevens, and H. Wagner. 2013. Vegan: Community Ecol-
ogy Package. Version 2.0-7. URL http://CRAN.R-project.org/package=vegan.
Orme, C. D., R. Freckleton, G. Thomas, T. Petzoldt, S. Fritz, N. Isaac, and W. Pearse. 2013. ca-
per: Comparative Analyses of Phylogenetics and Evolution in R.
Pakeman, R. J. 2011. Multivariate identification of plant functional response and effect traits in
an agricultural landscape. Ecology 92:1353-1365.
Paradis, E., J. Claude, and K. Strimmer. 2004. APE: analyses of phylogenetics and evolution in
16
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
R language. Bioinformatics 20:289-290.
Phillips, S. J., R. P. Anderson, and R. E. Schapire. 2006. Maximum entropy modeling of species
geographic distributions. Ecological Modelling 190:231-259.
R Core Team. 2016. R: A language and environment for statistical computing. R Foundation for
Statistical Computing, Vienna, Austria.
Renjifo, L. M. 1999. Composition changes in a subandean avifauna after long-term forest frag-
mentation. Conservation Biology 13:1124-1139.
Revell, L. J. 2012. phytools: An R package for phylogenetic comparative biology (and other
things). Methods in Ecology and Evolution 3:217-223.
Ribon, R., J. E. Simon, and G. T. De Mattos. 2003. Bird extinctions in Atlantic forest fragments
of the Vicosa region, southeastern Brazil. Conservation Biology 17:1827-1839.
Rubolini, D., A. Liker, L. Garamszegi, A. Moller, and N. Saino. 2008. Using the BirdTree.org
website to obtain robust phylogenies for avian comparative studies: A primer. Current
Zoology 61:959-965.
Sekercioglu, C. H., G. C. Daily, and P. R. Ehrlich. 2004. Ecosystem consequences of bird de-
clines. Proceedings of the National Academy of Sciences USA 101:18042-18047.
Senior, M. J. M., K. C. Hamer, S. Bottrell, D. P. Edwards, T. M. Fayle, J. M. Lucey, P. J. May-
hew, R. Newton, K. S. H. Peh, F. H. Sheldon, C. Stewart, A. R. Styring, M. D. F. Thom,
P. Woodcock, and J. K. Hill. 2013. Trait-dependent declines of species following conver-
sion of rain forest to oil palm plantations. Biodiversity and Conservation 22:253-268.
Sigel, B. J., D. W. Robinson, and T. W. Sherry. 2010. Comparing bird community responses to
forest fragmentation in two lowland Central American reserves. Biological Conservation
143:340-350.
17
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
Sodhi, N. S., L. H. Liow, and F. A. Bazzaz. 2004. Avian extinctions from tropical and subtropi-
cal forests. Annual Review of Ecology Evolution and Systematics 35:323-345.
Stouffer, P. C., and R. O. Bierregaard. 1995a. Effects of Forest Fragmentation on Understory
Hummingbirds in Amazonian Brazil. Conservation Biology 9:1085-1094.
Stouffer, P. C., and R. O. Bierregaard. 1995b. Use of Amazonian Forest Fragments by Under-
story Insectivorous Birds. Ecology 76:2429-2445.
Thornton, D., L. Branch, and M. Sunquist. 2011. Passive sampling effects and landscape location
alter associations between species traits and response to fragmentation. Ecological Appli-
cations 21:817-829.
Turner, I. M. 1996. Species loss in fragments of tropical rain forest: A review of the evidence.
Journal of Applied Ecology 33:200-209.
Ulrich, W., C. Banks-Leite, G. De Coster, J. C. Habel, H. Matheve, W. D. Newmark, J. A. To-
bias, and L. Lens. 2017. Environmentally and behaviourally mediated co-occurrence of
functional traits in bird communities of tropical forest fragments. Oikos.
Van Etten, J. 2012. gdistance: distances and routes on geographical grids. R package version 1.1-
4. URL http://CRAN.R-project.org/package=gdistance.
Vetter, D., M. M. Hansbauer, Z. Végvári, and I. Storch. 2011. Predictors of forest fragmentation
sensitivity in Neotropical vertebrates: a quantitative review. Ecography 34:1-8.
Williams, N. M., E. E. Crone, T. H. Roulston, R. L. Minckley, L. Packer, and S. G. Potts. 2010.
Ecological and life-history traits predict bee species responses to environmental distur-
bances. Biological Conservation 143:2280-2291.
18
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
Figure 1 – Schematic of the modelling and transferability testing process. The PGLS analysis
uses the species weights (which are derived from a species by site table using weighted aver-
ages), a species phylogeny (consensus tree) and a trait by species table as input data. After
PGLS, traits are selected using stepwise backwards model selection, and finally models are
transferred only across similar datasets.
19
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
Figure 1
20
427
428
429
430
431
432
433
434
435
436
Supplementary Information
Trait-based indicators of bird species sensitivity to habitat loss are
effective within but not across datasets
Jack H. HATFIELD*, C. David L. ORME, Joseph A. TOBIAS, Cristina BANKS-LEITE
*Corresponding author – [email protected]
Appendix S1
Both mist-net and point count studies were conducted in the Atlantic Plateau of the State of São
Paulo. Within this region, there were four fragmented landscapes and three continuous land-
scapes areas across a 150 km-wide area (Fig. S1). The fragmented areas varied in the total
amount of forest cover (11%, 20%, 31%, 49% of forest cover in a defined area of 10,000 ha),
and all habitat patches were composed of a patchwork of secondary vegetation at different suc-
cessional stages (Lira et al 2012). The continuously forested landscapes were composed of late
secondary to old-growth native vegetation. There were between 8 and 23 fragments sampled in
each fragmented landscape, and 4 sites per continuously forested area (Fig. S1). Altogether, there
were 19 sites that were sampled by both mist-net and point count studies. On average (± SD),
the distance among sampling sites within fragmented areas was 4.4 km (± 2.2) for the mist-net
landscapes and 5.1 km (± 3.2) for the point count landscapes.
More specifically, the mist-net study surveyed the 11, 31 and 49% forest cover landscapes
(between 17 to 19 fragments each), and in all continuously-forested landscapes (four sites each).
Forest fragments in each of these landscapes varyed in size from 2 ha to 150ha (Banks-Leite et
al. 2011). The point count study surveyed eight forest fragments in each of the 20, 31 and 49%
21
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
forest cover landscapes, as well as 4 sites in two of the continuously-forested landscapes (Deve-
ley 2004). Fragment area varied from 14 ha to 491 ha. At each of the 32 sites a 20 ha plot was es-
tablished, and within each plot four point counts were conducted and visited five times, totaling
20 counts per site. Point counts with a detection radius of 100m were conducted at least 100m
from the forest edge and at least 200m apart.
Figure S1 – Map of study site locations used in the mist-net (MN) and point-count (PC)
samples, including an overlap of 19 sites covered by both types of survey. Forest cover (shown
in grey) denotes areas with >50% canopy closure in 2000. Data source: Hansen/UMD/Google/
USGS/NASA (Licensed under a Creative Commons Attribution 4.0 International License - ht-
22
458
459
460
461
462
463
464
465
466
467
tps://creativecommons.org/licenses/by/4.0/). Produced using ESRI ArcGIS v.10 using adminis-
trative boundaries from http://www.gadm.org/. Location of study sites based on figures in
Banks-Leite et al. (2011) and Develey (2004).
23
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
Appendix S2
Table S1 – Description of variables used and information source.
Trait Description SourceDistribution variablesBio-climaticSuitability
Relative probability presence value for site calculated using MaxEnt
Maps from BirdLife In-ternational and Nature-Serve, 2012
Endemism Whether the species is restricted to the Brazilian Atlantic Forest. 1 = True, 0 = False
Parker et al., 1996
Proximity to Edge (PED)
Distance to nearest terrestrial range edge in km from site
Maps from BirdLife In-ternational and Nature-Serve, 2012
Range Size Total range size in km2 Maps from BirdLife In-ternational and Nature-Serve, 2012
Relative Abundance
1 = uncommon, patchily distributed and/or rare 2 = fairly common 3 = common
Parker et al., 1996
Foraging traitsAnt Following Whether the species utilises army ant
swarms while foraging. 1 = True 0 = Falsedel Hoyo et al., 1992-2011; Parker et al., 1996
Diet Composi-tion
Percentage of diet composed of inverteb-rates, fruit, nectar, seeds, plants and ver-tebrates. A PCA was conducted on these percentages to produce two independent variables summarising diet composition
Wilman et al., 2014
Foraging Stratum
Foraging strata used by species classified as – Terrestrial, Understory, Midstorey, Can-opy and Aerial. 1 = Used 0 = Not used
Parker et al., 1996
Mixed Flock Whether species joins mixed species flocks. 1 = True 0 = False
del Hoyo et al., 1992-2011; Develey, 2000
Behavioural and ecological traitsMigration 1 = sedentary 2 = partially migratory 3 = mi-
gratoryTobias et al., 2016
Number of Habitats
Number of habitat types used Parker et al., 1996
Open Habitat 1 = dense habitats 2 = semi-open habitats 3 = open habitats
Tobias et al., 2016
Territoriality 1= permanently territorial 2= seasonally or weakly territorial 3= non-territorial
Tobias et al., 2016
Morphological traitsBill Length Length of culmen from skull to bill tip See Bregman et al.,
2015; Pigot et al., 2016; Bregman et al., 2016.
24
486
487
Bill Depth Depth of bill at anterior nostril edge See Bregman et al., 2015; Pigot et al., 2016; Bregman et al., 2016.
Bill Width Distance across bill at anterior nostril edge See Bregman et al., 2015; Pigot et al., 2016; Bregman et al., 2016.
Body Mass Mass of species in grams. Based on Dun-ning (1993) with missing values obtained from congeners or regression methods
Ramirez et al., 2008
Tail Length of central tail feathers (base to tip length of longest rectrix)
See Bregman et al., 2015; Pigot et al., 2016; Bregman et al., 2016.
Tarsus Length of tarsus bone (centre of the rear ankle joint to the end of the last ac-rotarsium scale).
See Bregman et al., 2015; Pigot et al., 2016; Bregman et al., 2016.
Wing Length from bend of wing (carpel joint) to the tip of the longest primary.
See Bregman et al., 2015; Pigot et al., 2016; Bregman et al., 2016.
Hand Wing In-dex (HWI)
Kipp’s distance controlling for wing length. For calculation method see Claramunt et al. (2012)
See Bregman et al., 2015; Pigot et al., 2016; Bregman et al., 2016.
25
488
489
490
Table S2 – Response variables assessed and results obtained from simplified models for each
dataset. The direction of correlation with species sensitivity is indicated as positive or negative,
e.g. endemic species have higher sensitivity values than non-endemics, and sensitivity values are
negatively correlated with PED. “Overall” column indicates the traits that were considered ro-
bust as predictors of species sensitivity because they were selected in all models and had the
same sign. Values show the directionality and the size of the t-statistic from the final model with
a fixed lambda value.
Data Type Variable Mist-netAll
Point CountAll
Mist-netBoth
Point Count Both Overall
Occurrence Bio-climatic Suitability 2.61 4.54Proximity to Range Edge
(PED) -3.77 -3.35
Ant Follower 2.98 2.77DietComp1 -2.47 -3.13 -3.07Understory -3.01 -2.46 -3.12
Mixed Flock 2.43Number of Habitats -2.34 -5.41 -3.56 -3.44 Y
Open Habitat -3.26 -3.33 -3.68 -2.12 YTerritoriality -2.16 -2.00Bill DepthBill Width 2.02 3.79 2.45
Tarsus -2.00 -3.73 -3.00
Abundance Bio-climatic Suitability 4.34 4.91Endemism 3.03 2.67
Proximity to Range Edge (PED)
-3.63 -2.51
Relative Abundance -2.02Ant Follower 3.34 2.92 3.91 4.48 Y
Canopy -2.10 2.15DietComp1 -4.64 -4.15Understory -2.29 -2.88
Number of Habitats -4.69 -2.86Bill Length 2.10
Tail -2.36
26
491
492
493
494
495
496
497
498
499
Appendix S3
Methods Comparison
Methods
To explore whether our results were biased by the use of weighted averages, we re-assessed species re-
sponses to habitat change using other standard techniques: Fourth corner and RLQ analyses (Dray et al.
2014). For fourth corner analysis, we deemed important variables to be those that were significantly as-
sociated with all landscape characteristics (i.e. patch size, forest cover at 800 m around sampling point
and forest cover at the 10,000ha landscape scale). We used 50,000 replicates for all permutation tests.
The combined RLQ fourth corner method proposed by Dray et al. (2014) was used to test for significant
relationships between species variables and RLQ axes. RLQ axis 1 was used as this explained 99.9% of
the variance and was significantly correlated with all three landscape metrics. Only species variables sig-
nificantly correlated with RLQ axis 1 were considered important. The weighted averages model used a
significance level of 0.05. For this sensitivity analysis, we only present the results obtained with the point
count data (see main text for details), as it contains more species and is therefore a better sampled rep-
resentation of the local species assemblage and its constituent traits. We then compared which vari-
ables were identified as important by each method.
Landscape metrics were calculated from maps in Banks-Leite et al. (2011). Maps were analysed using
ESRI ArcGIS v10.0 and its base map imagery, with metric calculation conducted using Fragstats v4
(McGarigal et al. 2012). Control forest sites were in continuous forest and thus were assigned maximum
values for all landscape variables.
Results
The variables Bioclimatic suitability, DietComp1, Understory and Number of Habitats were found to be
27
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
important by all methods (Table S1). Other variables were only found to be important by one or more
methods but not all. In total 17 different variables were found to be important in at least one method.
Conclusions
The use of different methods can produce different results (Table S1). Only 4 were common to all mod-
els, although in some cases this may be due to exclusion because of high correlation with other vari-
ables. These results consolidate the implications from the model transferability testing – the majority of
species variables tested cannot be used as robust and reliable predictors of species sensitivity to habitat
fragmentation.
28
522
523
524
525
526
527
528
529
Table S1– Table showing which variables were found to be important using the different methods. Only
variables found to be important by at least one method were included. Variables labelled NA were not
modelled using weighted averages ordination due to high correlation with other explanatory variables.
Trait GroupFourth Corne
r
RLQ Fourth corner com-
bination
Weighted Averages
Bioclimatic Suitability Distributional + + +
Endemism Distributional + +
PED Distributional - -
Range Size Distributional - - NA
Relative Abundance Distributional - -
Ant Fol-lowing Foraging +
DietComp1 Foraging - - -
DietComp2 Foraging -
Understory Foraging - - -
Number of Habitats Behavioural - - -
Body Mass Morphological + + NA
Bill Length Morphological +
Bill Width Morphological + +
Bill Depth Morphological + + NA
HWI Morphological + + NA
29
530
531
532
Tail Morphological + +
Wing Morphological + + NA
30
533
534
535
536
537
538
539
540
541
542
543
R eferences for supplementary material
Banks-Leite, C., R. M. Ewers, V. Kapos, A. C. Martensen, and J. P. Metzger. 2011. Comparing
species and measures of landscape structure as indicators of conservation importance.
Journal of Applied Ecology 48:706-714.
BirdLife International and NatureServe. 2012. Bird species distribution maps of the world.
BirdLife International, Cambridge, UK and NatureServe, Arlington, USA.
Bregman, T. P., A. C. Lees, H. E. MacGregor, B. Darski, N. G. de Moura, A. Aleixo, J. Barlow,
and J. A. Tobias. 2016. Using avian functional traits to assess the impact of land-cover
change on ecosystem processes linked to resilience in tropical forests. Proceedings of the
Royal Society B: Biological Sciences 283.
Bregman, T. P., A. C. Lees, N. Seddon, H. E. MacGregor, B. Darski, A. Aleixo, M. B. Bonsall,
and J. Tobias. 2015. Species interactions regulate the collapse of biodiveristy and ecosys-
tem function in tropical forest fragments. Ecology 96:2692-2704.
Claramunt, S., E. P. Derryberry, J. V. Remsen, Jr., and R. T. Brumfield. 2012. High dispersal
ability inhibits speciation in a continental radiation of passerine birds. Proceedings of the
Royal Society B: Biological Sciences 279:1567-1574.
del Hoyo, J., A. Elliott, J. Sargatal, and D. A. Christie. 1992-2011. Handbook of the birds of the
world. Lynx Edicions.
Develey, P. F. 2004. Efeitos da fragmentação e do estado de conservação da floresta na diversi-
dade de aves de Mata Atlântica. Universidade de Sãu Paulo, Sãu Paulo.
Develey, P. F., and C. A. Peres. 2000. Resource seasonality and the structure of mixed species
bird flocks in a coastal Atlantic forest of southeastern Brazil. Journal of Tropical Ecology
16:33-53.
31
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
Dray, S., P. Choler, S. Doledec, P. R. Peres-Neto, W. Thuiller, S. Pavoine, and C. J. ter Braak.
2014. Combining the fourth-corner and the RLQ methods for assessing trait responses to
environmental variation. Ecology 95:14-21.
Dunning, J. B. 1993. CRC Handbook of avian body masses. CRC Press, Boca Raton, FL.
Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D.
Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L.
Chini, C. O. Justice, and J. R. G. Townshend. 2013. High-Resolution Global Maps of
21st-Century Forest Cover Change. Science 342: 850–53. Data available on-line
from:http://earthenginepartners.appspot.com/science-2013-global-forest.
Lira, P. K., R. M. Ewers, C. Banks-Leite, R. Pardini, J. P. Metzger, and J. Barlow. 2012. Evalu-
ating the legacy of landscape history: extinction debt and species credit in bird and small
mammal assemblages in the Brazilian Atlantic Forest. Journal of Applied Ecology
49:1325-1333.
McGarigal, K., S. A. Cushman, and E. Ene. 2012. FRAGSTATS v4: Spatial Pattern Analysis
Program for Categorical and Continuous Maps. University of Massachusetts, Amherst.
Parker, T. A. I., D. F. Stotz, and J. W. Fitzpatrick. 1996. Ecological and distributional database.in
D. F. Stotz, J. W. Fitzpatrick, T. A. I. Parker, and D. K. Moskovits, editors. Neotropical
birds: ecology and conservation Chicago University Press, Chicago.
Pigot, A. L., C. H. Trisos, and J. A. Tobias. 2016. Functional traits reveal the expansion and
packing of ecological niche space underlying an elevational diversity gradient in passer-
ine birds. Proceedings of the Royal Society B: Biological Sciences 283.
32
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
Ramirez, L., J. A. F. Diniz-Filho, and B. A. Hawkins. 2008. Partitioning phylogenetic and adap-
tive components of the geographical body-size pattern of New World birds. Global Ecol-
ogy and Biogeography 17:100-110.
Tobias, J. A., C. Sheard, N. Seddon, A. Meade, A. J. Cotton, and S. Nakagawa. 2016. Territorial-
ity, Social Bonds, and the Evolution of Communal Signaling in Birds. Frontiers in Ecol-
ogy and Evolution 4.
Wilman, H., J. Belmaker, J. Simpson, C. de la Rosa, M. M. Rivadeneira, and W. Jetz. 2014. El-
tonTraits 1.0: Species-level foraging attributes of the world's birds and mammals. Ecol-
ogy 95:2027-2027.
33
588
589
590
591
592
593
594
595
596
597