You are on page 1of 13

ECOGRAPHY

Research
Diverse temperate forest bird assemblages demonstrate
closer correspondence to plant species composition than
vegetation structure

Bryce T. Adams and Stephen N. Matthews

B. T. Adams (https://orcid.org/0000-0002-7830-6523) ✉ (adams.861@osu.edu) and S. N. Matthews, School of Environment and Natural Resources,
The Ohio State Univ., Columbus, OH 43210, USA.

Ecography The aggregate structure of vegetation has long provided a strong conceptual basis for
42: 1–13, 2019 understanding the ecological separation of faunal species, while correspondence with
doi: 10.1111/ecog.04487 plant species composition remains largely underdeveloped and considered secondary
to structure. Longstanding ecological debate on the matter has likely been sustained by
Subject Editor: Cagan Sekercioglu statistical methods incapable of accommodating an entire species composition in the
Editor-in-Chief: Miguel Araújo explanatory role. We used direct ordination methodologies (predictive co-correspon-
Accepted 22 June 2019 dence and canonical correspondence analyses) that allow the comparison of prediction
levels between composition and structure for understanding avian assemblage com-
position in a temperate forestland in southeastern Ohio. Compositional (birds and
woody plants) and structural (both vertical and horizontal dimensions) data were col-
lected from point samples comprising a spectrum in topography and successional state.
Total woody plant composition (11.22%) explained more (cross-validatory) variation
in avian species composition than structure, quantified with dense LiDAR recordings
(7.35%) and field methods (6.31%). Plant composition assumed an integrative char-
acter, synthesizing aspects of environmental condition, structure, and most likely spe-
cies-specific preferences for tree and shrub species that structural indices alone could
not for predicting avian species composition. These results conflict with the traditional
view that structure be most influential to avian assemblage composition. Instead these
results demonstrate that plant and avian assemblages are closely linked, and that plant
species per se can be a powerful tool for predicting avian habitat. More importantly,
in furthering ecological understanding, it is critical to consider the complex web of
interacting processes that make up temperate forest ecosystems, in which composition
occupies a key position.

Keywords: birds, co-correspondence analysis, composition versus structure,


MacArthur, temperate forests

––––––––––––––––––––––––––––––––––––––––
© 2019 The Authors. Ecography © 2019 Nordic Society Oikos
www.ecography.org

1
Introduction models including both VS and PSC variables that were not
otherwise present among models restricted to VS variables
The cumulative response of individual organisms to ecologi- alone (Hewson et al. 2011). These plant-specific preferences
cal and environmental variation plays a critical role in shaping likely manifest from differences in the provisioning of spe-
biological communities. While individual species respond to cific resources by different tree species, such as food (e.g.
such variation simultaneously across a range of organizational invertebrates, seeds and fruits) and nest sites (Rotenberry
levels, the most direct interactions are expected to occur at 1985, Holmes and Schultz 1988, Shutt et al. 2018). For
relatively fine levels (Wiens 1989, Cushman and McGarigal example, species of the genus Quercus (oaks) often host
2004). Here, biotic conditions defined by vegetation struc- more diverse and abundant lepidopteran assemblages than
ture (VS, hereafter) and plant species composition (PSC, any other co-occurring species in eastern deciduous forests
hereafter) are considered most influential to many species (Summerville et al. 2003, Tallamy and Shropshire 2009).
(Block and Brennan 1993, Quine et al. 2007). Numerous In addition, species-specific foliage architecture and growth
studies support and continue to demonstrate the importance form can influence foraging behavior and concealment from
of such factors on avian assemblages and selected focal spe- predators (Robinson and Holmes 1982, Whelan 2001,
cies in emblematic forest ecosystems among mature stands Wood et al. 2012, Muiruri et al. 2016). Ultimately, these
(Virkkala and Liehu 1990, Rodewald and Abrams 2002, studies support species-specific adaptations that define avian
Lee and Rotenberry 2005, Hewson et al. 2011, Yen et al. niche parameters according to plant-specific food resources
2011, Kellner et al. 2018), but also early-successional stands and foraging substrates (Holmes et al. 1979, Holmes and
(King et al. 2009) and other habitats alike (Wiens et al. 1987, Robinson 1981). Therefore, plant species per se also likely
Rotenberry and Wiens 2009). Of these factors, however, VS contribute to the differential partitioning of bird species in
is often considered the primary driver of avian assemblage some way (Holmes et al. 1979). Methodological expres-
variation (Jones 2001) and other taxa (see Schaffers et al. sions of assemblage variation have likely contributed to
2008 regarding arthropods). Efforts to understand how the the underappreciation of PSC in faunal assemblage studies
interactions between VS and PSC on avian species scale-up to (Rotenberry 1985, Schaffers et al. 2008).
the assemblage-level have contributed to a long-standing eco- Quantifying plant–faunal correspondences have recently
logical debate on the relative importance of vegetation factors experienced a methodological advancement. Inference on
for faunal assemblages (Anderson and Shugart 1974, Wiens this matter has largely followed conclusions made by correla-
and Rotenberry 1981, Rotenberry 1985, Mac Nally 1990, tions between univariate descriptors, such as plant and avian
Fleishman and Mac Nally 2006, Müller et al. 2010). diversity, or independent ordinations/dissimilarity matrices
Observations on the vertical profiling of forest habi- of avian and plant assemblages (MacArthur and MacArthur
tats have favored the importance of structural diversity for 1961, Rotenberry 1985, Lee and Rotenberry 2005). However,
the diversity of avian assemblages for some time (Dunlavy assemblage data are intrinsically multivariate, and such varia-
1935). The pioneering work by MacArthur and MacArthur tion is obscured by simplifying assemblage composition to
(1961), with their structural index relating to the distribu- simple indices (Rotenberry 1985). Specifically, raw species
tion of foliage over height, known as foliage height diver- identities must be preserved to resolve cross-taxon correspon-
sity (FHD), provided the first quantitative assessment of dences among species of either assemblage (Schaffers et al.
such relationships between plant and avian assemblages. 2008). Direct ordinations, particularly canonical corre-
MacArthur’s results and prior observations were interpreted spondence analysis, are better suited to this task (ter Braak
to be the result of the exploitation of individual bird spe- 1986). Until now, however, direct ordinations were limited
cies to the potential increases in niche space afforded by VS, in scope by the number of variables that may be reliably fit to
despite the confounding relationship between physiognomic community data, and, thus, incapable of accommodating a
diversity and plant species diversity (MacArthur 1958, Mac second species assemblage in the predictor role. Predictive co-
Nally et al. 2002). The effects of PSC were thought to oper- correspondence analysis was recently developed by adapting
ate solely through its impacts to VS and contribute little to aspects of canonical correspondence analysis to finally allow
avian niche space directly (MacArthur and MacArthur 1961, a compositional data set (e.g. plants) to predict the species
Mac Nally et al. 2002). This implies that individual tree and composition of another (e.g. birds) (ter Braak and Schaffers
shrub species are otherwise functionally-redundant as long 2004). Quantified prediction levels based on PSC can now be
as the replacement of one species by another imparts little directly compared to those obtained by canonical correspon-
to no change to the structural configuration of vegetation dence analysis parameterized with VS variables, circumvent-
(Rotenberry 1985, Fleishman et al. 2003, Veen et al. 2010). ing the information loss that results from correlating simple
Strong individualistic preferences for certain tree and diversity indices (ter Braak and Schaffers 2004). Moreover,
shrub species are indeed recognized for many bird species ordinations provide a means for the visual inspection of
(Holmes and Robinson 1981, Gabbe et al. 2002, Hartung emergent relationships between species and variables in
and Brawn 2005), even among closely-related species ordination diagrams. Meanwhile, advanced remote sensing
(Veen et al. 2010). Additionally, emergent relationships technologies, such as LiDAR, provide unparalleled windows
between birds and vegetation are often revealed among into the structural arrangement of habitats for examining

2
organismic response to VS (Hinsley et al. 2006, Müller et al. ridgetops and southwestern hillslopes (n = 96 points) and
2010, Swatantran et al. 2012). transition to mesophytic assemblages on opposing, mesic,
There is developing interest in the role of PSC on faunal 2) northeastern hillslopes (n = 59 points) and 3) bottom-
assemblages and identifying and reevaluating the factors lands (n = 55 points), with considerable species-level varia-
considered important to avian resource utilization and tion in between (Fig. 2a). In fact, the region is particularly
community structure is important for adapting manage- speciose, harboring 75 tree species (Prasad et al. 2007). In
ment response to environmental and ecological change. addition, the overall sampling design was inclusive to a wide
Specifically, concrete efforts in establishing plant–fau- spectrum of native forest type (<1% plantation) and growth
nal links at local scales could provide immense support stage, including open-canopy stands like MacArthur and
to ongoing efforts examining faunal response to apparent MacArthur (1961), maximizing the ecological coverage and
proximate signals of shifting vegetation pressures, including comprehensiveness of our data set.
climate change (Rodenhouse et al. 2008, Matthews et al.
2011, Buchanan et al. 2016, Iverson et al. 2018). We used Compositional data sets
predictive ordination techniques to compare PSC and VS
for explaining avian assemblage composition in a temper- Bird surveys followed the standard point count protocol
ate forestland in southeastern Ohio. Our objectives are to (Ralph et al. 1995). Trained observers performed counts
establish a better link between plant and avian assemblages for 10 min over six surveys, three in each year (2015–
in temperate forestlands and demonstrate the utility of PSC 2016), from mid-May to early-July on mornings (half-hour
for predicting avian habitat. We expect that some of the before sunrise to the hour of 10:30 EST) with minimal
associations that emerge be the result of the interrelation- wind or precipitation interference. We used detections
ships between PSC and VS (Mac Nally 1990, Fleishman ≤50 m from the observation point to ensure that only
and Mac Nally 2006, Hewson et al. 2011). However, we birds using the habitat being sampled were included in
predict that the effects of PSC not operate solely through the analysis. Recent studies support the use of fixed-radius
variation in VS, and these tools support clarity into the counts over model-based solutions, particularly for lim-
multiple interacting pathways that structure avian response ited observation radii and comparably large sample sizes
to forest habitats across successional and environmental (Johnson 2008, Hutto 2016). Thus, we used raw counts
gradients. as an index of relative abundance and limited our analysis
to only passerines because members of this group (except
for American crow Corvus brachyrhynchos and blue jay
Methods Cyanocitta cristata) feature similar body sizes, insectivo-
rous diets during the breeding period, and defense of rela-
Study area and sampling design tively small discrete territories. We summed all detections
(auditory and visual) over the six surveys and arranged
We sampled the composition of bird and plant assemblages these data into a matrix of bird species composition by
from point samples distributed across five study sites in point (hereafter denoted as [B]).
an actively managed, mixed-used forest landscape (~1300 We assessed PSC through three compositional matrices:
km2) in southeastern Ohio’s Central Hardwoods (Fig. 1a, tree, shrub and total woody plant composition. Live woody
Supplementary material Appendix 1). A total of 210 points stems (≥1 m height) were sampled at each point loca-
were distributed >150 m from one another and <400 m tion from either two or three 11.3-m radius circular plots
from roads to facilitate landscape-scale coverage according to (~400 m2), depending on logistical constraints, and in two
a generalized random tessellation stratified sampling design stem size classes (i.e. physiognomies): large-diameter stems
(Fig. 1b) (Stevens and Olsen 2004). Floristic assemblages (i.e. trees) were ≥8 cm diameter at breast height (DBH),
closely track the topographic contours of the unglaciated and small-diameter stems (i.e. shrubs and young trees)
Allegheny Plateaus physiographic region, which features were <8 cm DBH. Plots were arranged into 3 × 3 arrays
dissected terrain with <100 m local relief. The connection on point center, in which the center plot was sampled first
between floristic variation and topography has constituted and additional plots (either one or two) selected randomly.
a fundamental component of ecological classification in the Large stems were counted and DBH recorded within the full
study area based on an ecological landtype (ELT) concept plot, while small stems were counted (DBH omitted) from
(Hix and Pearcy 1997, Iverson et al. 2018). Though not a nested subplot (5-m radius; ~100 m2). All woody stems
used directly for predicting avian species composition, each were identified to species following Braun (1961), except
point sample was labeled according to three ELTs to support for members of the genera Carya, Crataegus, Salix, Smilax,
interpretation of avian and floristic gradients (Fig. 1c, addi- Vaccinium and Vitis due to difficulties in field identification.
tional information on species associations and an indicator Stem counts were summarized across the multiple plots per
species analysis can be found in Supplementary material point into local stem densities (no. stems ha−1) for each plant
Appendix 1). In general, mixed-oak assemblages, primarily taxon, generating two physiognomically-defined matrices,
composed of Quercus and Carya genera, dominate the 1) dry representing the composition of the shrub/regeneration

3
Figure 1. Study area (a) and distribution of 210 point samples located in southeastern Ohio (b). Panel (c) shows and an example of three
points distributed across ecological landtypes that support diverse forest types and used here for interpretation of results. Panel (d) shows
LiDAR returns for an example point (StructureLiDAR, [L] matrix) with horizontal lines indicating vertical layers used to quantify foliage
height diversity (FHD). LiDAR-derived vertical profiles of points (including sample sizes) by ecological landtype (gray line indicates the
profile of the example point) (e). Finally, violin plots in panels (f ) and (g) show stem densities (no. stems ha−1) and canopy cover (%) of
points by ecological landtype (StructureField, [F]).

layer by the small-diameter size class (i.e. shrubs, [S]) and densities, [S] + [T]) representing total woody plant composi-
midstory and overstory tree composition by the large- tion ([A]). The number of sample plots (either two or three)
diameter size class (i.e. trees, [T]), and a combined matrix had no effect on the total species richness summarized per
including all stems from both size classes (summed species point (p = 0.453).

4
(a) Indicator species analysis

Acer rubrum
Acer saccharinum
Acer saccharum
Amelanchier arborea
Betula nigra
Cornus amomum
Fraxinus pennsylvanica
Lindera benzoin
Nyssa sylvatica
Oxydendrum arboreum Indicator value
Populus grandidentata 10
20
Quercus alba
30
Quercus coccinea 40
Quercus montana 50
Quercus velutina 60
Rhus copallinum
Rhus glabra
Rosa multiflora
Sassafras albidum
Smilax spp
Toxicodendron radicans
Ulmus americana
Ulmus rubra
Vaccinium spp
Viburnum dentatum

Northeastern Bottomlands points Ridgetops and


hillslopes points southwestern
hillslopes points

(b) Total woody plant composition, [A] (c) Bird composition, [B]

Ridgetops and
southwesternv
hillslopes

Northeastern
hillslopes

Bottomlands

10 15 20 25 30 35 40 10 15 20 25

Woody plant richness (NO species) Bird richness (NO species)

Figure 2. Compositional variation, including indicator values of diagnostic woody plant species (a) and species richness (no. species) of
plants (total composition, [A]) (b) and birds (birds, [B]) (c) by ecological landtype.

Vegetation structure data sets each point, included composite (i.e. across all species) stem
density (combined size classes) (Fig. 1f ), composite basal area
Vegetation structure, emphasizing both vertical and hori- (m2 ha−1) (large size class only), (mean) canopy cover (%)
zontal dimensions, was derived from publicly available (Fig. 1g) and (mean) canopy height (m). Canopy cover was
LiDAR data ([L]) (<http://ogrip.oit.ohio.gov/Home.aspx>; estimated at each vegetation plot using a convex, spherical
accessed 13 October 2014), collected in 2007, and through densiometer and averaged for each point. The LiDAR data
the field ([F]) methods for two structural predictor matrices set, including filtered ground and vegetation returns, had an
(Supplementary material Appendix 2). We chose to develop average spacing and density of 1.27 m and 0.27 returns m−2,
structural variable sets from different sources to leverage the respectively. We computed the maximum, mean and SD
potential unique information provided by conventional field of the canopy surface for each point from a 5-m resolution
methods and remote sensing as part of our overall evalua- canopy height model. Three penetration ratios (number
tion and to foster robust comparisons with PSC. Field- of LiDAR returns <2 m divided by the number of returns
derived variables, collected among the vegetation plots at <50 and <10 m, including returns <2 m; and the number

5
of returns <1 m divided by the number of returns <5 m, solutions in which model fit varies according to the num-
including returns <1 m) were also computed to charac- ber of ordination axes. To help facilitate the selection of
terize the degree of canopy openness (0–2/0–50 m ratio), final predictive models, permutation tests (999 permuta-
amount of midstory foliage (0–2/0–10 m ratio) and amount tions) were used to determine the significance of each axis
of understory foliage (0–1/0–5 m ratio) (Müller et al. 2010, combination. In all cases, we chose the number of ordina-
Melin et al. 2018). Finally, we computed FHD in a manner tion axes for final interpretation according to the maximum
similar to MacArthur and MacArthur (1961) by clustering predictive fit (i.e. global maximum) (Schaffers et al. 2008)
returns within three vertical layers, 0–5, 5–25 and >25 m, among those statistically significant (p < 0.05). Finally, the
and calculating the diversity (Shannon’s index) of vegeta- statistical difference in model fit between the individual pre-
tion hits throughout the vertical profile of each point loca- dictor sets was determined with two-sided randomization
tion (Melin et al. 2018) (Fig. 1c–d): −∑pi ln pi; where pi tests (999 permutations) (van der Voet 1994). All analyses
is the proportion of the total LiDAR returns within the ith were conducted in the R statistical environment (< https://
vertical layer. cran.r-project.org/ >). Specifically, co-correspondence
analysis utilized various functions in the cocorresp package
Data analysis (Simpson 2016), whereas canonical correspondence analysis
was implemented by adapting functions in cocorresp and
The study area includes active forest management in which scripts provided in supplementary materials of ter Braak and
timber harvests occurred at some of the point locations since Schaffers (2004).
the acquisition of LiDAR. We excluded those points, as
recommended in studies featuring non-coincident LiDAR Data deposition
and wildlife survey data (Vierling et al. 2014), and points
<30 m from roads for the final sample of 210 points used Data are available from the Dryad Digital Repository:
in this study. We applied log (x + 1) transformations to the < http://dx.doi.org/10.5061/dryad.k48h616 > (Adams and
compositional ([B], [S], [T] and [A]) and structural ([L] Matthews 2019).
and [F]) data sets before analysis. In addition, we used
only species recorded in ≥5 (2%) point locations for each
compositional data set to reduce noise associated with the Results
effect of rare species on the exaggeration of the distinctive-
ness of sample units according to the X2 distance measure Our final data set included 48 bird species and 65 plant
(McCune et al. 2002). taxa (Supplementary material Appendix 3). Mean richness
We used a predictive version of canonical correspondence was 16.58 (± 3.97 SD) bird species and 19.76 (± 6.03 SD)
analysis (partial least squares [PLS] extension; ter Braak and woody plant taxa per point (Fig. 2b–c). The five predictive
Verdonschot 1995, ter Braak and Schaffers 2004) to pre- data sets based on PSC and VS included cross-validatory fits
dict species composition with the structural data ([L] or greater than zero, exceeding the null expectation of no rela-
[F]), and (predictive) co-correspondence analysis (ter Braak tionship with bird species composition. Plotting model fit
and Schaffers 2004, Schaffers et al. 2008) to predict spe- percentages against the number of ordination axes revealed
cies composition (such as [B]) with the compositional data unique trends among each predictor set (Fig. 3; filled sym-
of another assemblage (such as [A]). Both approaches are bols indicate significant axis solutions). While only margin-
similar in many ways, such as assuming a unimodal spe- ally greater than local maxima in some instances, we selected
cies response (ter Braak and Schaffers 2004, Schaffers et al. the final predictive models based on the global maximum
2008). Model fit was determined by a predictive, leave-one- for each data set. Final models included six, three, four, five
out cross-validatory procedure. Accordingly, a series of PLS and four axes solutions for [L], [F], [T], [S] and [A], respec-
regression models were tuned, each time withholding an tively (Table 1). The decision to use global maxima did not
individual point location for use as test data, until all point qualitatively change the results. Total woody plant composi-
samples were used in validation. Prediction accuracies (i.e. tion ([A]) explained the most variation in [B] at 11.22%,
cross-validated explained variance) were quantified by com- which was significantly greater than the remaining predic-
paring the predicted composition of each point to the actual tor sets according to the randomization tests: [L] (p = 0.048),
species data using the cross-validatory fit: 100 × (1 − sspa/ [F] (p = 0.015), [S] (p = 0.001) and [T] (p = 0.001) (see
ssp0); where sspa is the sum of squared prediction errors for Supplementary material Appendix 4 for randomization test
the model and a given axis solution, and ssp0 is the squared results). Structural predictor sets, [L] and [F], exhibited
prediction error under the null expectation of equal relative moderate predictive capacity, according to the range in fits,
abundances. This robust estimator produces model fits that at 7.35% and 6.31%, respectively. The fit of [S] was slightly
are usually lower than those customary for explanatory meth- greater than the two structural data sets at 9.11%, but statis-
ods, but do not imply weak relationships (Schaffers et al. tically indifferent. The weakest prediction level was observed
2008). In fact, models are implicitly validated when cross- for [T] at 3.45%. The difference between [S] and [T] was
validatory fit exceeds zero (ter Braak and Schaffers 2004, significant (p = 0.014), while the fits between [L], [F] and [T]
Schaffers et al. 2008). These methods yield multidimensional were similar. Correspondingly, total PSC, considering both

6
Figure 3. Prediction levels (% cross-validatory fit) of bird species composition by the structural and compositional predictor data sets and
methods (CCA-PLS = canonical correspondence analysis-partial least squares extension; CoCA = co-correspondence analysis). Displayed is
the cross-validatory fit plotted against the number of ordination axes for 12 axes. The filled symbols indicate significant axis solutions deter-
mined by permutation tests. Final models are indicated by horizontal bars and were selected according to the global maximum.

stem size classes ([A]), provided the best explanation of bird 2.61–6.07 m (95% C.I.) and 1.86–5.42 m (95% C.I.) less
species composition ([B]) (Table 1). than that of northeastern hillslopes and bottomlands, respec-
Examining the model with most support reveals sensible tively, explaining the positioning of those points towards
ecological gradients in ordination space (Fig. 4; points are the lower-right of the diagram with respect to the ELT
shaded according to ELT and vectors of structural variables and structural gradients. Plant species typical of inundated
are fitted post hoc to assist interpretation (Oksanen et al. floodplains, such as the indicator species Acer saccharinum
2018)). The primary and secondary axes display topographic and Betula nigra (Supplementary material Appendix 1), and
and structural gradients, respectively. However, these factors wetland bird species, red-winged blackbird (Agelaius phoeni-
appear to interact and vary in a more diagonal fashion with ceus) and common yellowthroat (Geothlypis trichas), occupy
respect to the origin, where shifts in ELT relates to some the upper-left section of the diagram, while the other end
variation in VS and vice versa. For example, canopy height includes taxa typical of drier site conditions and sparse, oak-
among dry ridgetops (and southwestern hillslopes) was dominated canopies, such as Vaccinium spp., Quercus coccinea

Table 1. Prediction levels (% cross-validatory fit) of the final models (dimensionality given in parentheses) for the taxonomic composition of
avian and plant assemblages constrained by vegetation structure or plant composition. Statistical comparisons in model fit (determined
by two-sided randomization tests; p < 0.05) are coded by letters where statistically similar fits are represented by similar letters
(CCA-PLS = canonical correspondence analysis-partial least squares extension; CoCA = co-correspondence analysis).

Dependent matrix
Independent matrix (model) Tree composition, [T] Shrub composition, [S] Total composition, [A] Bird composition, [B]
StructureLiDAR, [L] (CCA-PLS) 1.71 (2) 2.35 (2) 2.66 (2) 7.35 (6) a, b
StructureField, [F] (CCA-PLS) 1.23 (1) 3.56 (2) 3.45 (2) 6.31 (3) a, b
Tree composition, [T] (CoCA) – 8.72 (8) – 3.45 (4) a
Shrub composition, [S] (CoCA) 13.41 (7) – – 9.11 (5) b
Total composition, [A] (CoCA) – – – 11.22 (4) c

7
8
Bird composition, [B] Total woody plant composition, [A]
35
a Acadian Flycatcher (Empidonax virescens) 1 boxelder (Acer negundo)
b American Crow (Corvus brachyrhynchos) 2 red maple (Acer rubrum)
c American Goldfinch (Spinus tristis) 3 silver maple (Acer saccharinum) 25
d American Redstart (Setophaga ruticilla) 4 sugar maple (Acer saccharum)
e American Robin (Turdus migratorius) 5 yellow buckeye (Aesculus flava)
f Baltimore Oriole (Icterus galbula) 6 tree−of−heaven (Ailanthus altissima) 15
g Black−and−white Warbler (Mniotilta varia) 7 downy serviceberry (Amelanchier arborea)
h Blue−gray Gnatcatcher (Polioptila caerulea) 8 paw paw (Asimina triloba)
i Brown−headed Cowbird (Molothrus ater) 9 river birch (Betula nigra)

Canopy height (m)


5
j Blackburnian Warbler (Setophaga fusca) 10 American hornbeam (Carpinus caroliniana)
k Blue Jay (Cyanocitta cristata) 11 hickory (Carya spp)
l Brown Thrasher (Toxostoma rufum) 12 American chestnut (Castanea dentata) Ridgetops and southwestern hillslopes points StructureLiDAR, [L]
m Black−throated Green Warbler (Setophaga virens) 13 hackberry (Celtis occidentalis) Northeastern hillslopes points StructureField, [F]
n Blue−winged Warbler (Vermivora cyanoptera) 14 eastern redbud (Cercis canadensis) Bottomlands points
o Carolina Chickadee (Poecile carolinensis) 15 alternate leaf dogwood (Cornus alternifolia) 3
p Carolina Wren (Thryothorus ludovicianus) 16 silky dogwood (Cornus amomum)
Structural ea
q Cedar Waxwing (Bombycilla cedrorum) 17 flowering dogwood (Cornus florida) _m s
r Cerulean Warbler (Setophaga cerulea) 18 American hazelnut (Corylus americana) vector key: e ig ax _ba
s Chipping Sparrow (Spizella passerina) 19 hawthorn (Crataegus spp) h m m ea
n_ _ te en a _m

chm_std
t Common Yellowthroat (Geothlypis trichas) 20 autumn olive (Elaeagnus umbellata) 3 ca hm g_s Ev me vr
c i g _ co
u Eastern Phoebe (Sayornis phoebe) 21 American beech (Fagus grandifolia) b e m _
v h n
c a
v Eastern Towhee (Pipilo erythrophthalmus) 22 white ash (Fraxinus americana) c 0
_1
w Eastern Wood−Pewee (Contopus virens) 23 green ash (Fraxinus pennsylvanica) n2
2 pe
x Eastern Tufted Titmouse (Baeolophus bicolor) 24 witch hazel (Hamamelis virginiana)
y Field Sparrow (Spizella pusilla) 25 shrubby St. Johns wort (Hypericum spathulatum)
8
z Great Crested Flycatcher (Myiarchus crinitus) 26 black walnut (Juglans nigra)
A Gray Catbird (Dumetella carolinensis) 27 spicebush (Lindera benzoin) −0.75 53
pen1_5
B Hooded Warbler (Setophaga citrina) 28 tulip poplar (Liriodendron tulipifera)

CoCA axis 2
C Indigo Bunting (Passerina cyanea) 29 Japanese honeysuckle (Lonicera japonica) 16 6
D Kentucky Warbler (Geothlypis formosa) 30 amur honeysuckle (Lonicera maackii) 59
E Louisiana Waterthrush (Parkesia motacilla) 31 black gum (Nyssa sylvatica) 13
F Northern Cardinal (Cardinalis cardinalis) 32 eastern hophornbeam (Ostrya virginiana)
1

_
−0.75 5723

0
G Northern Parula (Setophaga americana) 33 sourwood (Oxydendrum arboreum)

m
_5 n

e
st
H Ovenbird (Seiurus aurocapilla) 34 virginia creeper (Parthenocissus quinquefolia) 27

n2 de
R 2 = 0.05 G

ll_
pe
I Pine Warbler (Setophaga pinus) 35 pitch pine (Pinus rigida)

a
R 2 = 0.25 9 5 d r m
J Prairie Warbler (Setophaga discolor) 36 eastern white pine (Pinus strobus) 30 24
R 2 = 0.55 58 E 4
K Rose−breasted Grosbeak (Pheucticus ludovicianus) 37 American sycamore (Platanus occidentalis)
L Red−eyed Vireo (Vireo olivaceus) 38 bigtooth aspen (Populus grandidentata) E M 50
4 p z 22 21
M Red−winged Blackbird (Agelaius phoeniceus) 39 black cherry (Prunus serotina)
t 37 V 64 10
K 32 Ia e
N Scarlet Tanager (Piranga olivacea) 40 white oak (Quercus alba) 1 28S55
A 34 F 15 xh jPL N
O Summer Tanager (Piranga rubra) 41 scarlet oak (Quercus coccinea) 0 60 u i11 H R
bD 14 UB 2w
62
P White−breasted Nuthatch (Sitta carolinensis) 42 shingle oak (Quercus imbricaria) 26 29 c f k g44 31
Q White−eyed Vireo (Vireo griseus) 43 chestnut oak (Quercus montana) 22 21 o 19 43
56 65q18 v 493517 40 712
O
R Worm−eating Warbler (Helmitheros vermivorum) 44 red oak (Quercus rubra) z C 63
l52 39 45 33
S Wood Thrush (Hylocichla mustelina) 45 black oak (Quercus velutina) Ia Q 51 54
10K e
T Yellow−breasted Chat (Icteria virens) 46 winged sumac (Rhus copallinum) 32 61
U Yellow−throated Vireo (Vireo flavifrons) 47 smooth sumac (Rhus glabra) h 48 41
28 S 42 20
V Yellow−throated Warbler (Setophaga dominica) 48 staghorn sumac (Rhus typhina) 15 55
x j PL
N 36
49 black locust (Robinia pseudoacacia) H n 25
50 multiflora rose (Rosa multiflora) −1 D 60 i 11 R 38
b u 47 s
51 blackberry (Rubus allegheniensis) 14
U B 2w
6231
52 raspberry (Rubus occidentalis)
f k g 44 T 46
53 elderberry (Sambucus canadensis)
17 43 J
54 sassafras (Sassafras albidum) 1819 49 35 40 7
55 green briar (Smilax spp) v
5q 12
56 steeplebush spirea (Spiraea tomentosa) O
57 American basswood (Tilia americana) y
63
39 45 33
58 poison ivy (Toxicodendron radicans) −2 l
59 American elm (Ulmus americana) 52
54
60 red elm (Ulmus rubra) 1
61 blueberry (Vaccinium spp)
62 mapleleaf viburnum (Viburnum acerifolium)
63 arrowwood viburnum (Viburnum dentatum)
64 blackhaw viburnum (Viburnum prunifolium) −5 −4 −3 −2 −1 0 1 5 15 25 35
65 grape vine (Vitis spp)
CoCA axis 1 Canopy height (m)

Figure 4. Predictive co-correspondence analysis ordination diagram of bird species composition constrained by total woody plant composition, [A]. Bird species are positioned
according to their normalized sites scores derived from the plants. Structural vectors (not used in the ordination) and the coloring of point samples by ecological landtype is used
to facilitate interpretation (see Supplementary material Appendix 2 for variable descriptions). In addition, variation in canopy height (m) is emphasized along each axis by the verti-
cal and horizontal bars. Two inset figures display a structural vector key and zoomed in view of the center of the ordination.
and worm-eating warbler Helmitheros vermivorum. The or not (Hawkins and Porter 2003, Gioria et al. 2010). Here
second gradient, moving diagonally towards the upper- again, ordination diagrams provide meaningful context to
right of the diagram, corresponds to a structural gradient unpack the synthesizing nature of PSC. Topographic driv-
from open- to closed-canopy habitats. Early-seral forests/ ers of PSC manifested along the co-correspondence analy-
shrublands, comprising species such as Rhus copallinum, sis, while this relationship was not portrayed when VS was
Hypericum spathulatum, prairie warbler Setophaga discolor, used to constrain bird species composition (regular canoni-
and yellow-breasted chat Icteria virens, dominate areas with cal correspondence analysis results in Fig. 5). Additionally,
comparably low axes scores. These assemblages transition to projecting structural variables onto the co-correspondence
structurally-complex stands, appearing in riparian habitats, results produced directional loadings towards expected plant
characterized by species such as Asimina triloba and north- assemblages according to their observed structural proper-
ern parula Setophaga americana, and northeastern hillslopes, ties. These visual observations suggest that PSC synthesized
characterized by Fraxinus pennsylvanica and cerulean warbler some aspects of VS, highlighting the integrative character of
Setophaga cerulea. these data. This is not unlike the patterns observed in other
studies, demonstrating correspondence between PSC and
VS, albeit with univariate descriptors of diversity (MacArthur
Discussion and MacArthur 1961). It is indeed the plants themselves that
support structural variation of the vegetation.
Bird species composition was best explained by the taxo- Interestingly, however, VS, either LiDAR- or field-derived,
nomic identity of vegetation rather than structure across the was not a strong predictor of woody PSC (Table 1). These
floristically-speciose forest landscape in southeastern Ohio. data domains typically obtained similar predictive fits, ulti-
The predictive ordination techniques provided a unique mately indicating that, while the different domains likely
basis for resolving relationships between bird and plant captured some different aspects of VS, no basis was more
assemblages. In doing so, however, our analysis revealed that useful than the other for predicting bird species composition
associations with PSC were, as predicted, highly-complex. and even PSC in our study. This follows two realizations: 1)
Direct floristic effects likely explain much of the predictive despite the vast quantities of data obtained from LiDAR,
power of PSC in our study. For instance, numerous examples simple metrics such as canopy height are often strong
identify species-specific selection of certain plants by birds correlates of more complex metrics such as vertical vegetation
in our study area, such as the cerulean warbler utilizing density (Jakubowski et al. 2013, Roussel et al. 2017); and 2)
Vitis spp. bark for nest construction (Weakland and Wood metrics, such as the penetration ratios, serve as proxies for
2005, Bakermans and Rodewald 2009) and the worm-eat- in situ measurements of horizontal vegetation density. This
ing warbler preferentially foraging among Quercus montana expresses the intrinsic gross structural relationships of forest
trees (Greenberg 1987). Ordination diagrams are powerful habitats that are expected along successional pathways and
tools for displaying these patterns and it is noteworthy the supports that utility of advanced remote sensing technolo-
proximity of these species in the co-correspondence results. gies to efficiently describe VS (Falkowski et al. 2009). Why
In other systems, such as bird-seed dispersal networks, the VS did not capture PSC better could have resulted from a
phenology and abundance of fruiting resources (Machado- high incidence of stands composed of similar gross structures
de-Souza et al. 2019) and sometimes morphological comple- amid an overall more pronounced floristic gradient along the
mentarity of birds and plants (e.g. fruit diameter and bill successional pathway in the data set. Here, to some degree,
length) (González-Castro et al. 2015) supports the under- subtle compositional gradients do not appear to result in
standing of the temporal and spatial distribution of species aggregate structural change for certain plant assemblages
interactions, demonstrating the often tight-linkage between detectable by gross VS. For example, predictions based on VS
plant–avian assemblages. failed to partition the sparsely-treed stands, particularly open
The strength in this relationship, however, need not be floodplains and regenerating clear-cuts, from one another,
entirely due to direct floristic effects. This relationship could which are expected to function differently and host different
emerge from the ability of PSC to also synthesize other fac- avian assemblages. These otherwise structurally-similar, albeit
tors that may be responsible for determining faunal resource floristically-different, habitats were closely related according
use, in addition to species-specific preferences. This is to their quantified structure, which resulted in weak predic-
because plants themselves are responding to environmental tions of their corresponding avian assemblages as constrained
condition, as well as defining and modifying their environ- by VS. Additionally, predictions of avian species composition
ment through growth (i.e. affecting VS), litter deposition by VS failed to distinguish mixed-oak and mesophytic for-
and competition (as interpreted by Schaffers et al. 2008). est types (noted by the overlap in ELT ellipses in Fig. 5),
Consequently, the correspondence could result from shared which can be a strong determinant of avian community
responses to the same set of conditions, or the conditions structure (Rodewald and Abrams 2002). This largely follows
responsible for shaping plant assemblages may appear to the misinterpretation in the traditional view of the impor-
manifest as drivers of associated avian assemblages, despite tance of VS addressed by other authors that implies a cer-
whether they directly interact with individual bird species tain degree of functional-redundancy in tree species as long

9
35

height (m)
Canopy
20

StructureLiDAR, [L]
Bird composition, [B]
2
a Acadian Flycatcher (Empidonax virescens)
b American Crow (Corvus brachyrhynchos)
c American Goldfinch (Spinus tristis)
d American Redstart (Setophaga ruticilla)
1

11__55
e American Robin (Turdus migratorius)
y
f Baltimore Oriole (Icterus galbula) J

ppeenn
g Black−and−white Warbler (Mniotilta varia) O 00
h Blue−gray Gnatcatcher (Polioptila caerulea) T ppeenn22_ s 22__11
i Brown−headed Cowbird (Molothrus ater) n _5500 v BH I
w ppeenn
Q o bz giL U

CCA2
j Blackburnian Warbler (Setophaga fusca) 0 q xk P
RNSa de r chm
C f p Dh K vveech m_mea
E _mea
cchm

ppee
k Blue Jay (Cyanocitta cristata) O ggE
l Brown Thrasher (Toxostoma rufum) A
l c F
u
V
j G E hm m__ vveenn
t mmaa
m Black−throated Green Warbler (Setophaga virens) xx

cchhm
n Blue−winged Warbler (Vermivora cyanoptera) s
−1

m__s
o Carolina Chickadee (Poecile carolinensis) Iw ppee M
5500

sttdd
p Carolina Wren (Thryothorus ludovicianus)
v iB H
q Cedar Waxwing (Bombycilla cedrorum) U
LPRN
r Cerulean Warbler (Setophaga cerulea) b a e r
z q D kg S
s Chipping Sparrow (Spizella passerina) −2 x
h d
t Common Yellowthroat (Geothlypis trichas) K
f p
u Eastern Phoebe (Sayornis phoebe)
v Eastern Towhee (Pipilo erythrophthalmus) V
w Eastern Wood−Pewee (Contopus virens) E
x Eastern Tufted Titmouse (Baeolophus bicolor) −3 F j G
Field Sparrow (Spizella pusilla)
u
y
z Great Crested Flycatcher (Myiarchus crinitus)
A Gray Catbird (Dumetella carolinensis)
−3 −2 −1 0 1 2 5 20 35
B Hooded Warbler (Setophaga citrina)
C Indigo Bunting (Passerina cyanea) Canopy
D Kentucky Warbler (Geothlypis formosa) CCA1 height (m)
E Louisiana Waterthrush (Parkesia motacilla)
F Northern Cardinal (Cardinalis cardinalis)
G Northern Parula (Setophaga americana)
H Ovenbird (Seiurus aurocapilla) 35
height (m)

I Pine Warbler (Setophaga pinus)


Canopy

J Prairie Warbler (Setophaga discolor)


K Rose−breasted Grosbeak (Pheucticus ludovicianus) 20
L Red−eyed Vireo (Vireo olivaceus)
M Red−winged Blackbird (Agelaius phoeniceus)
N Scarlet Tanager (Piranga olivacea) 5
O Summer Tanager (Piranga rubra)
P White−breasted Nuthatch (Sitta carolinensis)
Q White−eyed Vireo (Vireo griseus) StructureField, [F]
R Worm−eating Warbler (Helmitheros vermivorum) 2
S Wood Thrush (Hylocichla mustelina)
T Yellow−breasted Chat (Icteria virens)
U Yellow−throated Vireo (Vireo flavifrons)
eenn

V Yellow−throated Warbler (Setophaga dominica)


__dd

1
m
eem

ccaann
sstt

__hhee f
ll__

iigg__m
meea q s
aall

a m K D
zp j l
d F
O C Q
0 can_covr_mea r aw
can_covr_mea S
R
P
N g uv
BixM TJ
CCA2

e EULhk c t
G H b Ao n
ee

Ridgetops and southwestern bbaass I f


sstt

emm__ V
ll__

__sstte q
aall

hillslopes points −1
bbiig
g K
D
m
Northeastern hillslopes points d zp
j
F
O l
r v C
Bottomlands points mea
mea PBi xgMu
SR
awEUN h
e HL k t
StructureLiDAR, [L] vectors −2
G
c
A o
b
StructureField, [F] vectors s
I V
−3

−2 −1 0 1 2 3 4 5 20 35
Canopy
CCA1 height (m)

Figure 5. Regular canonical correspondence analysis ordination diagrams of bird species composition constrained by either LiDAR-
(StructureLiDAR, [L]) (top) or field-derived (StructureField, [F]) (bottom) vegetation structural predictor sets (see Supplementary material
Appendix 2 for variable descriptions).

10
their replacement by another species does not impart changes Robinson and Holmes 1982, Gabbe et al. 2002, Wood et al.
to the structural configuration of vegetation (Holmes et al. 2012). This aspect could be missing in less diverse for-
1979, Holmes and Robinson 1981, Fleishman et al. 2003). ests in which VS alone may be most important in leverag-
It is possible that higher density LiDAR data could have pre- ing competitive interactions among different bird species
dicted PSC better (Müller et al. 2010). However, in applica- (MacArthur 1958).
tion, it may be possible to overestimate the importance of Examining the three PSC predictor matrices, we found
gross VS if such metrics characterize PSC so closely that they interesting support for the shrub layer ([S]) in comparison
are entirely redundant with PSC. In such case, predictions to trees ([T]). This is not surprising, considering that many
could be falsely attributed to VS when in fact they are cap- forest bird species in our study area forage and nest among
turing a floristic signal more closely, considering if species- the lower strata and utilize a variety of shrub species as sea-
specific growth form and leaf architecture are detectable with sonal resources throughout the breeding period (Vitz and
high-resolution remote sensing data. In total, however, in Rodewald 2007). Additionally, this size class was present
situ measurement of PSC provided the best explanation of across all point samples (i.e. present in early- and late-seral
bird species composition in this study, demonstrating that stands) and likely presented greater potential for synthesizing
this vegetation dimension should be explicitly considered, no VS, as the different understory species respond to variation
matter the means used, when examining avian assemblage in light regimes determined by canopy structure and open-
composition. ness (Müller et al. 2010). In total, however, both stem size
It was not our intention to combine environmental classes ([A]) provided the best prediction of avian assemblage
parameters with VS. However, combining VS with ELT had composition, suggesting that total woody PSC is necessary
little effect on the predictive accuracy of bird species compo- when evaluating resource use of bird species.
sition for the LiDAR- (7.31% fit) and field-derived (6.71% We demonstrate that statistical methods that account for
fit) data sets in a post hoc examination, expecting that the the total species composition of vegetation can provide novel
inclusion of ELT data could help to characterize PSC that VS insight into the factors that support local avian assemblages.
alone could not. However, these results suggest that the com- The underappreciation for PSC has likely manifested from the
bination of VS and landscape position does not provide the simplification of compositional data into univariate descrip-
information necessary for accurate predictions of bird species tors (Rotenberry 1985). In furthering the debate, it is critical
composition as does highly-detailed PSC data. While ELT to consider the multiple interacting processes that underpin
data are suitable for generalizing PSC into syntaxonomic patterns in avian assemblages. Despite recognizing the associ-
units, avian assemblages appear to respond more closely ation between VS and PSC, these factors have generally been
to subtle compositional gradients apparent among stands considered rather separate. Our results support a perspective
of similar units. This is because PSC likely unfolds along a in which variation in bird species composition and VS mani-
continuum in topo-edaphic gradients rather than arbitrary fests from PSC that otherwise unfolds along environmental
breaks in topographic contours. Moreover, the presence of gradients and successional pathways. Most importantly, PSC
the ELT gradient, thus, likely emerged because of individual occupies a key position in determining and describing this
plant species response rather than a direct determinant of variation. Thus, PSC, being an important integrative indi-
bird species. cator, has great potential for explaining avian assemblages
These techniques support the power of PSC for the at local scales in temperate forestlands (Rotenberry 1985,
prediction of arthropod assemblages as well (Schaffers et al. Wiens et al. 1987, Mac Nally 1990, Fleishman et al. 2003).
2008, Gioria et al. 2010, Nyafwono et al. 2015). The strong In addition, our synthesis suggests that such relationships may
correspondence between arthropod and plant assemblages be context-dependent, contingent on the degree of functional
could also explain much of the variation in direct floristic significance of individual tree species, notably in the presence
effects likely occurring in this study, being that arthropods of diverse hardwood assemblages. While VS has proven to be
make up a large portion of avian diets during the breeding a strong generalizable framework in assessing avian–habitat
period (Rotenberry 1985, Holmes and Schultz 1988, relationships and understanding of the ecological separation
Gunnarsson et al. 2018). In another case examining predic- of species, PSC harbors a practical synthesizing character,
tions by PSC and VS in a Bavarian forestland, VS emerged as accounting for the many decisive factors that determine fau-
the best predictor of bird species composition (Müller et al. nal resource use, including VS. Finally, plant species per se
2010). The difference in predictive powers based on tree/ do indeed support considerable faunal niche space, consider-
shrub PSC between studies is particularly noteworthy ing the strength in the relationship between avian and plant
(1.91% versus 11.22%). These differences suggest that the assemblages, and that VS, being an emergent property of PSC
factors that underpin compositional turnover of avian assem- in most forests, is only one piece of the puzzle.
blages may depend partly on the floristic variation, and, most
importantly, functional significance of individual plant spe- Acknowledgements – We thank Kaley Donovan, James Hanks,
cies within different forest ecosystems (Gabbe et al. 2002, Garrett Evans, Sara Zaleski, Alex Eberts and Daniel Hodges for
Muiruri et al. 2016). Indeed, considerable evidence for spe- their field assistance. We thank Laura Kearns, Louis Iverson, Robert
cies-specific preferences for certain tree species comes from Gates and Christopher Tonra for their input on early versions of
work in temperate forestlands (Holmes and Robinson 1981, the manuscript.

11
Funding – This study was funded by the State Wildlife Grants Hartung, S. C. and Brawn, J. D. 2005. Effects of savanna restoration
Program, administered jointly by the US Fish and Wildlife Service on the foraging ecology of insectivorous songbirds. – Condor
and the Ohio Division of Wildlife, through the Ohio Biodiversity 107: 879–888.
Conservation Partnership. Additional support was provided by the Hawkins, B. A. and Porter, E. E. 2003. Does herbivore diversity
School of Environment and Natural Resources and the Terrestrial depend on plant diversity? The case of California butterflies.
Wildlife Ecology Laboratory at The Ohio State University. – Am. Nat. 161: 40–49.
Hewson, C. M. et al. 2011. Species-specific responses of woodland
birds to stand-level habitat characteristics: the dual importance
References of forest structure and floristics. – For. Ecol. Manage. 261:
1224–1240.
Adams, B. and Matthews, S. N. et al. 2019. Data from: diverse Hinsley, S. A. et al. 2006. The application of lidar in woodland bird
temperate forest bird assemblages demonstrate closer ecology: climate, canopy structure and habitat quality.
correspondence to plant species composition than vegetation – Photogramm. Eng. Remote Sens. 72: 1399–1406.
structure. – Dryad Digital Repository, <http://dx.doi. Hix, D. M. and Pearcy, J. N. 1997. Forest ecosystems of the
org/10.5061/dryad.k48h616>. Marietta Unit, Wayne National Forest, southeastern Ohio:
Anderson, S. H. and Shugart, H. H. 1974. Habitat selection of multifactor classification and analysis. – Can. J. For. Res. 27:
breeding birds in an east Tennessee deciduous forest. – Ecology 1117–1131.
55: 828–837. Holmes, R. T. and Robinson, S. K. 1981. Tree species preferences
Bakermans, M. H. and Rodewald, A. D. 2009. Think globally, of foraging insectivorous birds in a northern hardwoods forest.
manage locally: the importance of steady-state forest – Oecologia 48: 31–35.
features for a declining songbird. – For. Ecol. Manage. 258: Holmes, R. T. and Schultz, J. C. 1988. Food availability for forest
224–232. birds: effects of prey distribution and abundance on bird
Block, W. M. and Brennan, L. A. 1993. The habitat concept in foraging. – Can. J. Zool. 66: 720–728.
ornithology. – In: Power, D. M. (ed.), Current ornithology. Holmes, R. T. et al. 1979. Guild structure of the Hubbard Brook
Plenum Press, pp. 35–91. bird community: a multivariate approach. – Ecology 60:
Braun, E. L. 1961. The woody plants of Ohio: trees, shrubs and 512–520.
woody climbers, native, naturalized and escaped. – Ohio State Hutto, R. L. 2016. Should scientists be required to use a model-
Univ. Press. based solution to adjust for possible distance-based detectability
Buchanan, M. L. et al. 2016. Response of bird populations to bias? – Ecol. Appl. 26: 1287–1294.
long-term changes in local vegetation and regional forest cover. Iverson, L. R. et al. 2018. Spatial modeling and inventories for
– Wilson J. Ornithol. 128: 704–718. prioritizing investment into oak–hickory restoration. – For.
Cushman, S. A. and McGarigal, K. 2004. Hierarchical analysis of Ecol. Manage. 424: 355–366.
forest bird species–environment relationships in the Oregon Jakubowski, M. K. et al. 2013. Tradeoffs between lidar pulse density
coast range. – Ecol. Appl. 14: 1090–1105. and forest measurement accuracy. – Remote Sens. Environ.
Dunlavy, J. C. 1935. Studies on the phyto-vertical distribution of 130: 245–253.
birds. – Auk 52: 425–431. Johnson, D. H. 2008. In defense of indices: the case of bird surveys.
Falkowski, M. J. et al. 2009. Characterizing forest succession with – J. Wildl. Manage. 72: 857–868.
lidar data: an evaluation for the inland northwest, USA. Jones, J. 2001. Habitat selection studies in avian ecology: a critical
– Remote Sens. Environ. 113: 946–956. review. – Auk 118: 557–562.
Fleishman, E. and Mac Nally, R. 2006. Patterns of spatial autocor- Kellner, K. F. et al. 2018. Local-scale habitat components driving
relation of assemblages of birds, floristics, physiognomy and bird abundance in eastern deciduous forests. – Am. Midl. Nat.
primary productivity in the central Great Basin, USA. – Divers. 180: 52–65.
Distrib. 12: 236–243. King, D. I. et al. 2009. Habitat use and nest success of scrub–shrub
Fleishman, E. et al. 2003. Effects of floristics, physiognomy and birds in wildlife and silvicultural openings in western Massa-
non-native vegetation on riparian bird communities in a Mojave chusetts, USA. – For. Ecol. Manage. 257: 421–426.
Desert watershed. – J. Anim. Ecol. 72: 484–490. Lee, P.-Y. and Rotenberry, J. T. 2005. Relationships between bird
Gabbe, A. P. et al. 2002. Tree-species preferences of foraging insec- species and tree species assemblages in forested habitats of
tivorous birds: implications for floodplain forest restoration. eastern North America. – J. Biogeogr. 32: 1139–1150.
– Conserv. Biol. 16: 462–470. Mac Nally, R. C. 1990. The roles of floristic and physiognomy
Gioria, M. et al. 2010. The conservation value of farmland ponds: in avian community composition. – Aust. J. Ecol. 15:
predicting water beetle assemblages using vascular plants as a 321–327.
surrogate group. – Biol. Conserv. 143: 1125–1133. Mac Nally, R. et al. 2002. How well do ecosystem-based planning
González-Castro, A. et al. 2015. Relative importance of phenotypic units represent different components of biodiversity? – Ecol.
trait matching and species’ abundances in determining plant Appl. 12: 900–912.
– avian seed dispersal interactions in a small insular community. MacArthur, R. H. 1958. Population ecology of some warblers of
– AoB Plants 7: plv017. northeastern coniferous forests. – Ecology 39: 599–619.
Greenberg, R. 1987. Seasonal foraging specialization in the MacArthur, R. H. and MacArthur, J. W. 1961. On bird species
worm-eating warbler. – Condor 89: 158–168. diversity. – Ecology 42: 594–598.
Gunnarsson, B. et al. 2018. Predation by avian insectivores on Machado-de-Souza, T. et al. 2019. Local drivers of the structure of
caterpillars is linked to leaf damage on oak (Quercus robur). a tropical bird-seed dispersal network. – Oecologia 189:
– Oecologia 188: 733–741. 421–433.

12
Matthews, S. N. et al. 2011. Changes in potential habitat of 147 Stevens, D. L. and Olsen, A. R. 2004. Spatially balanced sampling
North American breeding bird species in response to redistribu- of natural resources. – J. Am. Stat. Assoc. 99: 262–278.
tion of trees and climate following predicted climate change. Summerville, K. S. et al. 2003. Community structure of arboreal
– Ecography 34: 933–945. caterpillars within and among four tree species of the eastern
McCune, B. et al. 2002. Analysis of ecological communities. deciduous forest. – Ecol. Entomol. 28: 747–757.
– MjM Software Design. Swatantran, A. et al. 2012. Mapping migratory bird prevalence
Melin, M. et al. 2018. Living on the edge: utilising lidar data to using remote sensing data fusion. – PLoS One 7: e28922.
assess the importance of vegetation structure for avian diversity Tallamy, D. W. and Shropshire, K. J. 2009. Ranking lepidopteran
in fragmented woodlands and their edges. – Landscape Ecol. use of native versus introduced plants. – Conserv. Biol. 23:
33: 895–910. 941–947.
Muiruri, E. W. et al. 2016. Do birds see the forest for the trees? ter Braak, C. J. F. 1986. Canonical correspondence analysis: a new
Scale-dependent effects of tree diversity on avian predation of eigenvector technique for multivariate direct gradient analysis.
artificial larvae. – Oecologia 180: 619–630. – Ecology 67: 1167–1179.
Müller, J. et al. 2010. Composition versus physiognomy of ter Braak, C. J. F. and Schaffers, A. P. 2004. Co-correspondence
vegetation as predictors of bird assemblages: the role of lidar. analysis: a new ordination method to relate two community
– Remote Sens. Environ. 114: 490–495. compositions. – Ecology 85: 834–846.
Nyafwono, M. et al. 2015. Tree community composition and ter Braak, C. J. F. and Verdonschot, P. F. M. 1995. Canonical
vegetation structure predict butterfly community recovery in a correspondence analysis and related multivariate methods in
restored Afrotropical rain forest. – Biodivers. Conserv. 24: aquatic ecology. – Aquat. Sci. 57: 255–289.
1473–1485. van der Voet, H. 1994. Comparing the predictive accuracy of
Oksanen, J. et al. 2018. vegan: community ecology package. models using a simple randomization test. – Chemom. Intell.
– <https://cran.r-project.org/package=vegan>. Lab. Syst. 25: 313–323.
Prasad, A. M. et al. 2007. A climate change atlas for 134 forest tree Veen, T. et al. 2010. Temporal differences in food abundance
species of the eastern United States [database]. – <www.nrs. promote coexistence between two congeneric passerines.
fs.fed.us/atlas/tree>. – Oecologia 162: 873–884.
Quine, C. P. et al. 2007. Stand management: a threat or opportu- Vierling, K. T. et al. 2014. How much does the time lag between
nity for birds in British woodland? – Ibis 149: 161–174. wildlife field-data collection and LiDAR-data acquisition
Ralph, C. J. et al. 1995. Monitoring bird populations by point matter for studies of animal distributions? A case study using
counts. – US Dept of Agriculture, Forest Service Gen. Tech bird communities. – Remote Sens. Lett. 5: 185–193.
Rep. PSW-GTR-149. Virkkala, R. and Liehu, H. 1990. Habitat selection by the Siberian
Robinson, S. K. and Holmes, R. T. 1982. Foraging behavior of tit Parus cinctus in virgin and managed forests in northern
forest birds: the relationships among search tactics, diet and Finland. – Ornis Fenn. 67: 1–12.
habitat structure. – Ecology 63: 1918–1931. Vitz, A. C. and Rodewald, A. D. 2007. Vegetative and fruit
Rodenhouse, N. L. et al. 2008. Potential effects of climate change resources as determinants of habitat use by mature-forest birds
on birds of the Northeast. – Mitig. Adapt. Strateg. Global during the postbreeding period. – Auk 124: 494–507.
Change 13: 517–540. Weakland, C. A. and Wood, P. B. 2005. Cerulean warbler (Dendroica
Rodewald, A. D. and Abrams, M. D. 2002. Floristics and avian cerulea) microhabitat and landscape-level habitat characteristics
community structure: implications for regional changes in in southern West Virginia. – Auk 122: 497–508.
eastern forest composition. – For. Sci. 48: 267–272. Whelan, C. J. 2001. Foliage structure influences foraging of
Rotenberry, J. T. 1985. The role of habitat in avian community insectivorous forest birds: an experimental study. – Ecology 82:
composition: physiognomy or floristics? – Oecologia 67: 219–231.
213–217. Wiens, J. A. 1989. Spatial scaling in ecology. – Funct. Ecol. 3:
Rotenberry, J. T. and Wiens, J. A. 2009. Habitat relations of 385–397.
shrubsteppe birds: a 20-year retrospective. – Condor 111: Wiens, J. A. and Rotenberry, J. T. 1981. Habitat associations and
401–413. community structure of birds in shrubsteppe environments.
Roussel, J.-R. et al. 2017. Removing bias from LiDAR-based – Ecol. Monogr. 51: 21–42.
estimates of canopy height: accounting for the effects of pulse Wiens, J. A. et al. 1987. Habitat occupancy patterns of North
density and footprint size. – Remote Sens. Environ. 198: 1–16. American shrubsteppe birds: the effects of spatial scale. – Oikos
Schaffers, A. P. et al. 2008. Arthropod assemblages are best predicted 48: 132–147.
by plant species composition. – Ecology 89: 782–794. Wood, E. M. et al. 2012. Birds see the trees inside the forest:
Shutt, J. D. et al. 2018. The effects of woodland habitat and the potential impacts of changes in forest composition on
biogeography on blue tit Cyanistes caeruleus territory occupancy songbirds during spring migration. – For. Ecol. Manage. 280:
and productivity along a 220 km transect. – Ecography 12: 176–186.
1967–1978. Yen, J. D. L. et al. 2011. To what are woodland birds responding?
Simpson, G. L. 2016. cocorresp: co-correspondence analysis Inference on relative importance of in-site habitat variables
ordination methods. – <https://cran.R-project.org/package= using several ensemble habitat modelling techniques.
cocorresp>. – Ecography 34: 946–954.

Supplementary material (available online as Appendix ecog-


04487 at < www.ecography.org/appendix/ecog-04487 >).
Appendix 1–4.

13

You might also like