ArticlePDF Available

Abstract and Figures

Abstract Questions Niche differentiation is a central explanation for the co-existence and distribution patterns of numerous tree species in tropical forests, but functional equivalence leading to neutral dynamics has been proposed as an alternative explanation. This niche vs neutral debate is fuelled by the highly variable results yielded by studies of the association between tree species distributions and environmental factors, where some studies find strong associations but others do not. Here, we ask how differences in sampling scale between studies contribute to this variation. Location Barro Colorado Island, Panama. Methods Using distribution maps of canopy-statured individuals, we evaluated patterns of habitat association in five tropical tree species on Barro Colorado Island across a wide range of sampling scales (from 50 to 1600 ha). We investigated the scale-dependency of species clumping patterns (Ripley's K) and the association of species distributions with important environmental variables (forest age, topography and geological formation) using point pattern analyses. Results Clump size and clump density had high variances within and among spatial scales. Significant habitat associations were found in all five species, with the number of habitat associations generally increasing with the sampling scale. Ignoring dispersal constraints inflated the number of significant habitat associations. Conclusions We found that patterns of habitat association (and hence conclusions on the importance of niche vs neutral processes) are strongly affected by the choice of sampling scale and location. Explicit inclusion of the effect of spatial scale is critical for studies of habitat association and the main processes that structure communities of tropical trees.
Content may be subject to copyright.
Journal of Vegetation Science 25 (2014) 349–362
Effects of sampling scale on patterns of habitat
association in tropical trees
Carol X. Garzon-Lopez, Patrick A. Jansen, Stephanie A. Bohlman, Alejandro Ordonez &
Han Olff
Keywords
Dispersal limitation; Habitat association;
Niche differentiation; Panama; Point pattern
analysis; Species co-existence; Tropical moist
forest
Nomenclature
Correa et al. (2004)
Received 3 April2012
Accepted 5 April 2013
Co-ordinating Editor: Sam Scheiner
Garzon-Lopez, C.X. (corresponding author
c.x.garzon@gmail.com), Jansen, P.A.
(patrick.jansen@wur.nl), Ordonez, A.
(ordonezglori@wisc.edu) & Olff, H.
(h.olff@rug.nl): Community and Conservation
Ecology, Center of Life Sciences, University of
Groningen, Nijenborgh 7, 9747 AG Groningen,
The Netherlands
Jansen, P.A. : Department of Environmental
Sciences, Wageningen University,PO Box 47,
6700 AA Wageningen, The Netherlands
Jansen, P.A. &Bohlman, S.A.
(sbohlman@ufl.edu): Smithsonian Tropical
Research Institute, Apartado 084303092,
Balboa, Ancon, Republicof Panama
Bohlman, S.A.: School of Forest Resources
and Conservation, University of Florida, P.O.
Box 110410, Gainesville, FL 32611-0410, USA
Ordonez, A. : Nelson Institutefor
Environmental Studies Center for Climatic
Research (CCR), 53706-1695 Madison, WI, USA
Abstract
Questions: Niche differentiation is a central explanation for the co-existence
and distribution patterns of numerous tree species in tropical forests, but func-
tional equivalence leading to neutral dynamics has been proposed as an alterna-
tive explanation. This niche vs neutral debate is fuelled by the highly variable
results yielded by studies of the association between tree species distributions
and environmental factors, where some studies find strong associations but oth-
ers do not. Here, we ask how differences in sampling scale between studies con-
tribute to this variation.
Location: Barro Colorado Island, Panama.
Methods: Using distribution maps of canopy-statured individuals, we evaluated
patterns of habitat association in five tropical tree species on Barro Colorado
Island across a wide range of sampling scales (from 50 to 1600 ha). We investi-
gated the scale-dependency of species clumping patterns (Ripley’s K) and the
association of species distributions with important environmental variables (for-
est age, topography and geological formation) using point pattern analyses.
Results: Clump size and clump density had high variances within and among
spatial scales. Significant habitat associations were found in all five species, with
the number of habitat associations generally increasing with the sampling scale.
Ignoring dispersal constraints inflated the number of significant habitat associa-
tions.
Conclusions: We found that patterns of habitat association (and hence conclu-
sions on the importance of niche vs neutral processes) are strongly affected by
the choice of sampling scale and location. Explicit inclusion of the effect of spa-
tial scale is critical for studies of habitat association and the main processes that
structure communities of tropical trees.
Introduction
Tropical forests are among the most diverse ecosystems
in the world (Givnish 1999). Various hypotheses have
been proposed to explain the high level of species co-
existence that maintains this diversity (Wright 2002).
One of the most important explanations is niche theory,
which argues that species adaptation to specific condi-
tions determines the distribution of different species
along environmental gradients in space and time (Whit-
taker et al. 1975; Pulliam 2000; Hubbell 2001; Wright
2002). Thus, each species occupies a specific niche
formed by a combination of environmental conditions
(light, soil factors) that allow their establishment and
survival (Pulliam 2000). However, recent studies indicate
that dispersal limitation is an additional determinant of
the spatial distribution of tree species (e.g. Svenning
2001; Vormisto et al. 2004; Svenning et al. 2006; John
et al. 2007; Bohlman et al. 2008) that may contribute to
the co-existence of species by preventing the local domi-
nance of a single species. The degree to which dispersal
limitation vs habitat specialization drives species spatial
distributions along environmental gradients is currently
an active subject of scientific investigation.
A simple indicator of the importance of niche differenti-
ation is the percentage of species having a significant
Journal of Vegetation Science
Doi: 10.1111/jvs.12090©2013 International Association for Vegetation Science 349
association with different habitats or environmental fac-
tors. However, studies of habitat associations in tropical
forest trees have yielded highly variable results in terms of
the percentage of species with significant habitat associa-
tions (Appendix S1). Possible causes for this inconsistency
include differences among studies in: (1) the set of envi-
ronmental factors examined, which vary from soil chemis-
try to soil geologic origin and parental material (Appendix
S1); (2) the statistical techniques used to detect associa-
tions, which range from Mantel tests and randomization
tests to multivariate analyses (Itoh et al. 2010); (3) species-
specific characteristics (e.g.body size, dispersal strategy;
Nathan & Muller-Landau 2000); (4) the spatial scale of the
study (Lam & Quattrochi 1992; Wu 2004; Cottenie 2005;
McGill 2010), which varies from as little as 0.3 ha (Balva-
nera et al. 2002) to the 25- and 50-ha scale typically used
for long-term forest dynamics plots (Harms et al. 2001;
Russo et al. 2005) to as much as 158 ha (Phillips et al.
2003).
Spatial scale (i.e. the spatial extent of the area studied,
rather than the level of detail, resolution or ‘grain’ of sam-
pling; van Gemerden et al. 2005) is especially noteworthy
because different environmental factors show heterogene-
ity at different spatial scales (Whittaker et al. 2001; Kneitel
& Chase 2004; Snyder & Chesson 2004; Ricklefs 2008; Wu
& Li 2009). At large scales (ca. 100010 000 ha), species
distributions may reflect climatic gradients (Rhode 1992),
whereas at small scales (<10), they may rather reflect soil-
related heterogeneity with respect to nutrient or water
availability (which may be partially driven by individual
trees) and heterogeneity in canopy openness (Keddy 1982;
Ceccon et al. 2003). Spatial heterogeneity at intermediate
scales (ca. 101000 ha) may be caused by environmental
factors (Clark et al. 1995), such as geology, topography
and historical events (e.g.forest management history). To
date, no study has evaluated how the sampling scale, rang-
ing from local to intermediate spatial scales, affects the
detection of habitat associations.
Differences in dispersal capacity of species can also
strongly affect the spatial structure of communities, espe-
cially at intermediate scales. This can further confound the
detection of environmental effects (Duque et al. 2002; Gil-
bert & Lechowicz 2004; Alonso et al. 2006). Seed dispersal
patterns determine the initial potential distribution of indi-
viduals, but if habitat characteristics are also important for
survival, the final distribution will resemble the potential
distribution modified by the habitat requirements of the
species. To discriminate between the effects of environ-
mental factors and dispersal limitations, it is essential to
incorporate them in appropriate statistical models (Nathan
& Muller-Landau 2000).
In this paper, we consider the effect of sampling scale on
the detection of habitat associations in the tropical moist
forest of Barro Colorado Island (BCI), Panama. Four stud-
ies of habitat association have previously been conducted
on BCI; these used different sampling scales and have
drawn different conclusions. The two studies of the well-
studied BCI 50-ha plot (Harms et al. 2001; John et al.
2007) found that 64% and 29% of 171 and 75 tree species,
respectively, had significant habitat associations and that
topography and soil chemistry were important factors.
Two studies at the scale of the whole island (Svenning
et al. 2004, 2006), which sampled 32 and 7 ha of forest,
respectively, found that 25% and 68% of the 94 and 26
species studied, respectively, had significant habitat associ-
ations and that forest age was an important factor (Appen-
dix S1). Focusing on five tree species, the work presented
here not only encompasses the spatial scales of these previ-
ous studies, but also uses a much larger range of the size of
forest sampled (i.e. 50, 100, 200, 400, 800 and 1600 ha)
than in any previously published study. This is possible
because we develop a map of all canopy-statured individu-
als for these five species across all of BCI using high-resolu-
tion aerial photographs.
We tested the hypothesis that the percentage of species
with significant habitat associations varies with sampling
scale. In particular, we expect the percentage of significant
associations to increase with sampling scale (from 50 to
1600 ha) as greater heterogeneity in environmental vari-
ables is captured as well as a greater number of individual
trees are sampled. We expect that the number of signifi-
cant habitat associations detected will decrease across all
spatial scales if species-specific clumping, likely arising
from dispersal limitation, has been taken into account. This
decrease should be highest at small spatial scales (50
200 ha) where such clumping is the highest. Support for
our overall expectation that detection of habitat associa-
tion varies with scale would indicate that sampling scale
and species-specific aggregation patterns need to be explic-
itly considered when evaluating the relative importance of
niche- vs neutral-based explanations for tropical forest bio-
diversity.
Methods
Study site
Barro Colorado Island (BCI), Panama (9.9°N, 79.51°W),
is a 1560-ha island that became isolated from the sur-
rounding mainland tropical forest during 19101914,
when the Chagres River was dammed to form the cen-
tral part of the Panama Canal (Leigh 1999). The island
is covered with tropical moist forest and has a mean
annual rainfall of 2623 mm, with a 4-month wet season
from December to April (Leigh 1999). BCI is home to a
50-ha forest dynamics plot, established in 1982 (Condit
1998), and the island has been the focus of many stud-
Journal of Vegetation Science
350 Doi:10.1111/jvs.12090 ©2013 International Association for Vegetation Science
Scale-dependence of habitat association in tropical trees C.X. Garzon-Lopez et al.
ies, including vegetation structure, geology, hydrology,
soil dynamics and tropical ecology (Losos & Leigh
2004).
The island is heterogeneous at various spatial scales and
in several aspects including forest age, soil type and topog-
raphy. About half of the forest is 90135 yrs old while the
remaining is old-growth forest (Enders 1935; Leigh 1999;
Fig. 1). BCI has a variety of soil types that vary systemati-
cally with the type of underlying rock and the topography
(Leigh 1999). The island hosts two main geologic forma-
tions dating from the Oligocene, known as Bohio and Cai-
mito, where the latter is separated into marine and
volcanic facies (Fig. 1). The top of the island is covered
with igneous Andesite flows (Johnson & Stallard 1989;
Barthold et al. 2008), the only non-sedimentary lithology.
Moreover, there is a large variation in slope, and the eleva-
tion ranges from 27 to 160 m a.s.l. (Svenning et al. 2004),
a type of variation that can be classified into five topogra-
phy classes (Fig. 1).
Tree distributions
Across the entire study area, the spatial distributions of
canopy-statured individuals of the five study species (three
arborescent palms Attalea butyraceae, Astrocaryum stand-
leyanum, Oenocarpus mapora and two large-canopy tree
species Jacaranda copaia, Tabebuia guayacan;Table 1)were
assessed from high-resolution aerial photographs (Garzon-
lopez et al. 2012). The species were chosen because they
could be easily distinguished in the photos, were widely
distributed across the island, and represented a wide spec-
trum of shade tolerance and dispersal characteristics, such
as dispersal mode and seed size (Table 1).
The photographs were taken in April 2005 and April
2006. Flights were flown in overlapping northsouth
swathes at an altitude of 400 m in 2005, 700 m in 2006
and 8001000 m in 2007. In 2005, each photo, on aver-
age, covered 8.6 ha, with a spatial resolution of
0.085 mpixel
1
. In 2006, coverage and resolution aver-
aged 15.9 ha and 0.114 mpixel
1
. The aerial photographs
were registered to a georeferenced Quickbird satellite
image of BCI (DigitalGlobe, Longmont, CO, US), captured
in March 2004, using the ERDAS IMAGINE v8.7 program
(Leica Geosystems, Norcross, GA, US). Features visible in
both aerial photos and satellite images, including large tree
crowns and telemetry towers, were used as registration
points for warping and georeferencing the individual pho-
tographs to produce an island-wide orthorectified mosaic.
Following the criteria of Trichon (2001), a key for iden-
tifying the crowns of the tree species was developed based
on crown structure and known positions of trees of differ-
ent species based on the stem map of the 50-ha forest
dynamics plot (Condit 1998; Hubbell et al. 1999; Hubbell
2005). All aerial photos were surveyed to map canopy-
statured individuals belonging to the five tree species. Vali-
dation conducted using 75 ha of ground-mapped distribu-
tions, including those in the 50-ha forest dynamics plot,
confirmed that the distribution maps reflected the actual
distributions of adult trees of each species (Garzon-lopez
et al. 2012). The r
2
of the spatial regression between stem
and crown maps was on average 0.67, increasing from
0.51 to 0.84 with grid sizes of 0.64ha.
Environmental factors
Three environmental factors geological formation, forest
age and topography class were used to test for associa-
tions with the distributions of canopy individuals
(Table 2). These are a subset of the most common variables
used in habitat association studies (Appendix S1). Geologi-
cal formation (mapped at the 1:15000 scale) has a large
impact on soil nutrient availability, as well as other soil-
related characteristics (e.g. particle size, water retention
capacity, pH). Forest age indicates different stages of suc-
cession, different canopy density and complexity, and liana
coverage. Information on slope, elevation (mapped at a
1:25000 scale) and distance to shore were combined into
one categorical topographic variable with five habitat types
(Table 2) as a means to describe the biological response of
evaluated species to topographical heterogeneity. By com-
bining these factors, we were able to capture the variability
in soil moisture, soil run-off and other hydrological factors,
known to be collinear, using a single explanatory variable.
Determining the level of spatial aggregation and habitat
associations
Sampling
We used six different sizes of rectangular plots representing
increasing sampling scales (i.e. 50-, 100-, 200-, 400 and
800-ha, and the entire island, 1560 ha). Each sampling
scale was replicated by placing polygons at randomly gen-
erated points, with the exception of the two largest catego-
ries. Overlap among plots at the same sampling scale was
not allowed so that the same configuration of habitats and
species distributions was not replicated in multiple plots.
Overlap was avoided by randomly removing one of any
pair of overlapping plots. The number of replicates was ten
for 50- and 100-ha plots, five for 200-ha plots, three for
400-ha and one for the 800-ha plot and the whole island
(1560 ha). Plots had a range of heterogeneity in the levels
of environmental variables. Not every plot had each level
of each environmental variable present. Just three plots
(all 50-ha in size), however, had no heterogeneity in a par-
ticular environmental variable (two for geology; one for
forest age). Additionally, in one case (one 50-ha plot sam-
Journal of Vegetation Science
Doi: 10.1111/jvs.12090©2013 International Association for Vegetation Science 351
C.X. Garzon-Lopez et al. Scale-dependence of habitat association in tropical trees
Fig. 1. Spatial distribution of five tropical tree species on Barro Colorado Island, Panama, in relation to environmental factors.
Journal of Vegetation Science
352 Doi:10.1111/jvs.12090 ©2013 International Association for Vegetation Science
Scale-dependence of habitat association in tropical trees C.X. Garzon-Lopez et al.
ple for D. panamensis), a subsample contained less than ten
individuals of the target species and was excluded from the
analysis. Each replicate was evaluated as an independent
sample.
Spatial aggregation
We used Ripley’s K(d) function (representing the aggrega-
tion of trees at distance d) as a measure of spatial aggrega-
tion (Ripley 1976). Ripley’s Kmeasures the second-order
properties of a spatial point process, and describes the way
that spatial interactions change through space (Baddeley &
Silverman 1984). We used Ripley’s Kinstead of alternative
summaries (e.g. mean nearest-neighbour distance or
cumulative density function of distances from random
points to their nearest neighbours) as it allows the simulta-
neous evaluation of cross-scale patterns (e.g. clustering at
large scales and regularity at small scales) by assessing the
relationship between the observed intensity (mean num-
ber of observed trees per unit area) and the expected inten-
sity (maximum number of points at a given distance d
assuming a mean density) of a point process. The K(d)
function was transformed to L(d)/d = (K(d)/pd)/d and
plotted against distance to determine deviations from a
random distribution (Besag & Diggle 1977). Values of L(d)/
d>1 indicate clumping, and <1 indicate even distribu-
tions. We measured aggregation for the entire island and
at each of the seven smaller spatial scales in order to deter-
mine the effect of sampling scale on estimates of K(d). Both
clump density (q, defined as the number of clumps per
unit area) and clump size (r, defined as the SD of the mean
distance of an observation to the centre of the cluster, with
66% of the observations being at rmetres from the cluster
centre) were estimated to describe the clumping character-
istics of each species at each sampled scale. We used the
approach of Plotkin et al. (2000) to estimate K(d) by
assuming that its distribution follows a Poisson point pro-
cess described by the function:
KðdÞPCP ¼pd2þq11exp d2
4r2

ð1Þ
Parameters rand qwere estimated at each spatial scale
using the ‘pcp’ function in the R-package ‘splancs’ (R
Foundation for Statistical Computing, Vienna, AT). Clump
size (r) is expressed in meters and clump density (q)in
clumps per m
2
.
Simulations
To determine the effect of the environmental and dispersal
constraints on the observed spatial patterns, we used two
simulation approaches (i.e. complete spatial randomness
CSR and a Poisson cluster process PCP) to determine if
habitat associations were significant even when clumping
patterns were taken into account. For the PCP simulations,
we simulated the spatial distributions of adult trees based
on the aggregation characteristics (rand q) of each species.
This analysis reduces the likelihood that ‘artificial’ habitat
associations are found, which in reality arise from large,
dispersal-limited clumps of trees occurring in one or a few
specific habitat patches, causing a type of pseudo-replica-
tion effect.
First, we simulated 1000 distributions for each one of
the subsamples with complete spatial randomness (CSR),
Table 1. Characteristics of the tree species studied. Crown diameter is the mean sun-exposed crown diameter calculated from the aerial photos.
Species Attalea butyracea Astrocaryum standleyunum Jacaranda copaia Tabebuia guayacan Oenocarpus mapora
Dispersal mode Animal Animal Wind Wind Animal
Height (m) 30 20 30 30 20
Characteristics Palm tree Palm tree Deciduous Deciduous Palm tree
Shade tolerance Low Medium Low Low High
Individuals mapped 3121 2456 977 688 7555
Seed Size (cm) 4 92392291.8 3 91292
Crown diameter (m) 8.9 6.4 15.1 17.3 3.8
Density (ha
1
) 2.05 1.73 0.67 0.45 5.41
Table 2. Environmental variables considered in this studyand their levels.
Variable Levels Description Source
Geologic
formation
Bohio volcanic Barthold
et al. (2008)Caimito volcanic
Caimito marine
Andesite
Forest age Secondary
growth
regenerated
after 1880
Enders (1935)
Old growth regenerated
before 1880
Topography Shore <150 m to shore Johnson &
Stallard
(1989)
Flat >150 m to shore; <5°
slope; <63 m elevation
Ridge >150 m to shore; <5°
slope>63 m elevation
Shallow >150 m to shore;
510°slope
Steep >150 m to shore;
>10°slope
Journal of Vegetation Science
Doi: 10.1111/jvs.12090©2013 International Association for Vegetation Science 353
C.X. Garzon-Lopez et al. Scale-dependence of habitat association in tropical trees
which is determined only by the number of trees. Second,
we generated 1000 spatially clumped distributions at each
spatial scale, assuming a Thomas cluster process (a special
case of a Poisson Cluster Process; De Smith et al. 2007) in
which spatial patterns are determined by species-specific
clumping parameters determined from the Ripley’s Kanal-
ysis. For this, we implemented the approach of Plotkin
et al. (2000), using both the clump density (q,as defined
above) and size (r, as defined above) estimated form K(d)
at each scale as input parameters. PCP distributions were
simulated by placing a number of clusters equal to
q
i
*A + 0.5, where Acorresponds to the area of the sample
for the ith species. The number of stems per cluster was
given by n
i
/q
i
*A, where ncorresponds to the total number
of individuals in species i. Simulations based on PCP
resembled the observed distributions more closely than
simulations with CSR, showing the importance of dispersal
limitation as a basic determinant of species distributions
(Appendix S2).
Habitat associations
To test for specieshabitat associations at each scale, we
used the Gamma test (Plotkin et al. 2000), which is a modi-
fied version of a Chi-square test. This metric quantified the
likelihood that habitat associations were found at a particu-
lar sampling scale (i.e. the frequency of observed occur-
rences at a particular habitat level was significantly higher
or lower than expected based on a CSR or a PCP model).
A standard Chi-square test is not suitable in this case due to
its assumption of independence in the locations of conspe-
cific trees. The Gamma statistic is defined as the proportion
of trees found on habitat type jth from the Javailable habi-
tats, which is defined as (n
i,j
/n
i,
);wheren
i
is the number of
trees of species iin the sample, and n
i,j
is the number of
trees of species iin the sample occurring on environmental
factor jth. We first quantified the strength of the habitat
associations on a given level of an environmental variable
(e.g. ‘young forest’ is a level within the ‘forest age’ environ-
mental variable) within a single plot as the proportion of
the 1000 comparisons to the CSR and PCP expectations for
which observed occurrences exceeded the expected occur-
rence of the species on that habitat level (Gotelli & Graves
1996). We then determined the direction of the occur-
rencesenvironment associations (positive or negative) by
comparing observed Gamma values to the distributions of
Gamma values generated in 1000 PCP and CSR simula-
tions. If observed Gamma values fell outside the 95% con-
fidence interval of simulated Gamma values, habitat
associations were considered significant and either nega-
tive (Observed
Gamma
<Simulated
Gamma
)orpositive
(Observed
Gamma
>Simulated
Gamma
). Our Gamma test dif-
fers from a traditional Chi-square test in that the expected
frequency of observations within each category is not
based on a theoretical Chi-square distribution, but rather
drawn from the expected frequencies derived from a pro-
cess-based null expectation (i.e. CSR of a PCP process). This
approach was chosen, as opposed to a general linear
model-based approach, because it allowed the comparison
between the expectation based on a completely random
model (CSR) and the expectation of a model constrained
by clumping patterns due to dispersal (i.e. PCP). We per-
formed all the analyses using the Splancs and grdevices
package in the R program.
Results
We mapped a total of 15 209 crowns of the five tree spe-
cies, where the density ranged from 0.45 (T. guayacan)to
5.4 treesha
1
(O. mapora) (Table 1). Of the five species, O.
mapora had the highest L(d)/d value, indicating the strong-
est aggregation (individuals per clump), while T. guayacan
had the lowest aggregation in the range from 0 to 1500 m
(Appendix S2). J. copaia had a peak L(d)/d at the largest dis-
tance (12 m) of all species, which suggested it had the larg-
est clump size (r; Appendix S2). Visual inspection of the
island-wide distribution maps suggested that all species
had biased/clustered spatial distributions, with more indi-
viduals occurring in some habitat types than in others
(Fig. 1). For example, O. mapora was associated with
ridges, J. copaia with old-growth forest and A. butyracea
with secondary forest.
The spatial clumping parameters showed several consis-
tent trends among species (Fig. 2). While the size of
clumps and the number of trees per clump increased with
sampling scale, the clumps per ha decreased with plot size
(Fig. 2). The variance in clumps per ha decreased with plot
size for all species, but the variance in cluster size and trees
per clump showed no clear pattern with plot size (data not
shown).
Habitat associations with CSR model
The probability of detecting associations increased with
sampling scale (Fig. 3). When disaggregated across levels
for an environmental variable (Fig. 4), the level of the
environmental variables with the highest number of habi-
tat associations differed among species, but tended to be
consistent for each species across different sampling scales.
For example, A. butyracea showed a consistent association
with young forest, and the strength of this association
increased with sampling scale, while J. copaia showed a
consistent association with old-growth forest that
increased with sampling scale.
At the scale of the entire island, there were more signifi-
cant habitat associations (higher percentage of positive or
Journal of Vegetation Science
354 Doi:10.1111/jvs.12090 ©2013 International Association for Vegetation Science
Scale-dependence of habitat association in tropical trees C.X. Garzon-Lopez et al.
negative associations) than expected under complete spa-
tial randomness (CSR). The percentage of samples that
yielded significant habitat associations increased with sam-
pling scale for all five species and for all environmental
variables (Fig. 3; Appendix S3). At sampling scales
400 ha, all species had significant associations with
40% of the individual levels of the environmental
variables (e.g. young vs old-growth forest for the forest age
variable). The three palm species had significant associa-
tions, with 40% of levels within each environmental
variables starting at plot sizes of 100 ha. J. copaia and
T. guayacan had relatively few associations with
(a)
(b)
(c)
Fig. 2. Relationship with sampling scale of (a) clump size (m), (b) number of clumps per ha and (c) number of individuals per clump for five tropical tree
species on Barro Colorado Island, Panama. Clumping parameters were determined by a Poisson cluster analysis. Box plots summarize the mean and
interquartile variability across replicates at a particular sampling scale. Numbers of replicates per scale are 10 for 50 and 100 ha, six for 200 ha, three for
400 ha, and one for 800 and 1600 ha.
Journal of Vegetation Science
Doi: 10.1111/jvs.12090©2013 International Association for Vegetation Science 355
C.X. Garzon-Lopez et al. Scale-dependence of habitat association in tropical trees
topography. For T. guayacan, the highest number of associ-
ations was with forest age.
Habitat associations on the basis of the PCP model
We found that the probability of detecting habitat associa-
tions also increased with sampling scale under the PCP
model (Fig. 3), but the variables driving this pattern chan-
ged between scales (Fig. 5). The simulations under PCP,
which incorporated species-specific aggregation patterns,
yielded fewer habitat associations than CSR simulations
(Fig. 3; Appendix S4). The decrease in associations from
CSR to PCP simulations was highest for geological forma-
tions and topography (Fig. 3). The number of significant
associations was nearly the same under CSR and PCP for
forest age. For simulations under PCP, forest age had the
most significant associations for nearly all the species, espe-
cially at large plot sizes. For association with geologic for-
mation, and to a lesser degree for association with
topography, the divergence between PCP and CSR simula-
tions increased at larger plot sizes. For A. butyracea and
O. mapora, the PCP simulations showed a drop in the per-
centage of significant correlations at the largest spatial
sizes. O. mapora showed significant reductions in the per-
centage of significant habitat associations between PCP
and CSR simulations at all spatial scales. At the 800-ha to
1600-ha scales, A. butyracea, A. standleyanum and J. copaia
showed the largest decreases in the percentage of habitat
associations under PCP simulations.
With the CSR model, there was a high variability in the
direction of habitat associations (positive or negative)
within and among sampling scales (Appendix S3).
Between the 50- and 400-ha sampling scales, there were
62 examples in which the same species showed both a
positive and negative association with the same level of
environmental variable between two plots of the same
size. For example, there were both significantly positive
and negative associations at the 100-ha and 200-ha scales
Fig. 3. Variation in habitat associations with different sampling scales assuming PCP (Poisson cluster process) and CSR (complete spatial randomness) in
generating the simulated data sets for five tropical tree species on Barro Colorado Island, Panama. The y-axis (% habitat associations) is the mean number of
levels (e.g. young forest within the forest age variable) within the given environmental variable (forest age, geology, topography) showing a significant
association (either positive or negative) with the species distribution (Appendices S3 and S4). The error bars indicate the SE of significant associations
among plots at the given sampling scale.
Journal of Vegetation Science
356 Doi:10.1111/jvs.12090 ©2013 International Association for Vegetation Science
Scale-dependence of habitat association in tropical trees C.X. Garzon-Lopez et al.
found for A. butyracea with the ridge habitat type, depend-
ing on plot location. In contrast, we found only six exam-
ples of this under the PCP model (Appendix S4). This
indicates that clumping, most likely related to dispersal
constraints, not only increases the number of significant
habitat associations, but also does so in an arbitrary man-
ner with respect to habitat type. This also indicates that the
actual spatial placement of plots had an influence (via
environmental heterogeneity, clustering characteristics,
species density, etc.) on the observed association patterns.
Forest age had consistently positive or consistently nega-
tive associations for both PCP and CSR at nearly all spatial
scales and for all species. Geologic substrate and topogra-
phy had more instances of having both negative and posi-
tive associations for the same species within and between
spatial scales, especially with the CSR simulations.
Under PCP simulations, several combinations of species
and environmental variables showed some consistency
within or among spatial scales in terms of direction and sig-
nificance of association (Appendices S3, S5). All palm spe-
cies plus T. guayacan were positively associated with the
secondary forest and negatively associated with old-
growth forest (Fig. 3). In contrast, J. copaia was positively
associated with the old-growth forest and negatively
associated with secondary forest. These associations with
forest age were most consistent at the largest plot sizes
Fig. 4. Strength of habitat associations based on comparison to a complete spatial randomness (CSR) model. Habitat associations were determined using
an observed vs expected approach based on comparison to 1000 CSR simulations for each level of the evaluated environmental variable. Strength of the
relationship was measured as the number of the 1000 contrasts where observed occurrence on a certain habitat was higher than the expected occurrence
on that habitat. Box plots summarize the median and interquartile variability across replicates at a particular sampling scale. Numbers of replicates per
scale are 10 for 50 and 100 ha, six for 200 ha, three for 400 ha, and one for 800 and 1600 ha.
Journal of Vegetation Science
Doi: 10.1111/jvs.12090©2013 International Association for Vegetation Science 357
C.X. Garzon-Lopez et al. Scale-dependence of habitat association in tropical trees
(8001600 ha). O. mapora had the most numerous habitat
associations. It was significantly and positively associated
with the Andesite and Caimito volcanic types, regrowth
forest and the shore and ridge habitats, but negativelyasso-
ciated with the old-growth forest and steep habitats.
Discussion
Habitat association is a distinctive feature of niche special-
ization (Harms et al. 2001). However, comparisons across
previous studies yielded many inconsistencies in the num-
ber and strength of habitat associations (Appendix S1;
Cottenie 2005). The results of our study strongly suggest
that an important part of these inconsistencies may be
attributed to the variation in the sampling scale across
studies and/or to failure to account for species-specific
aggregation patterns that are typically due to dispersal limi-
tation. For the five tree species evaluated in this work, the
percentage of detected habitat associations varied substan-
tially with the sampling scale. However, differences among
plot sizes for a particular species decreased when species-
specific clumping patterns had been taken into account.
Changes in the percentage of habitat associations
depended on the environmental factor considered. A con-
sistently large number of associations were observed across
spatial scales for forest age (especially spatial scales
400 ha), both with and without incorporating dispersal
limitation, but not for geological formation and topogra-
Fig. 5. Strength of habitat associations based on contrast to a Poisson cluster process (PCP) model. Habitat associations were determined by using an
observed vs expected approach based on comparison to 1000 PCP simulations for each level of the environmental variables. Strength of the relationship
was measured as the number of the 1000 contrasts where observed occurrence on a certain habitat was higher than the expected occurrence on that
habitat. Box plots summarize the median and interquartile variability across replicates at a particular sampling scale. Numbers ofreplicatesper scale are 10
for 50 and 100 ha, six for 200 ha, three for 400 ha, and one for 800 and1600 ha.
Journal of Vegetation Science
358 Doi:10.1111/jvs.12090 ©2013 International Association for Vegetation Science
Scale-dependence of habitat association in tropical trees C.X. Garzon-Lopez et al.
phy. This result is consistent with Svenning et al. (2004),
who found that forest age had the highest number of sig-
nificant relationships with plant distributions on BCI of all
environmental variables considered. The fact that for some
species and environmental factors, no associations were
found at particular spatial scales (especially smaller sam-
pling scales where placement of plots is most critical) does
not necessarily mean that no relationship exists. Rather,
consistent results showing habitat associations for canopy-
statured tropical forest species may require large sampling
scales (>200 ha). This indicates the scale of the study
should match the scale at which the variables of interest
vary across space (in this case, the largest plot sizes).
Across plots with a 50-ha size, the standard in studies of
tropical tree diversity and distribution patterns, there was
much variation in environmental and species arrange-
ment. The mean percentage of significant habitat associa-
tions was lowest for the 50-ha plot scales for nearly all
combinations of species and environmental factors. Vari-
ance in percentage of significant habitat associations
among 50-ha plots was also high. This variability indicates
that the particular placement of a single plot can determine
the results of a study. Therefore, 50 ha may not be a suit-
able scale for identifying consistent trends of habitat associ-
ations that could be applicable at landscape scales for many
species and environmental variables, especially for the
canopy-statured species and the environmental variables
(forest age, geology and topography, which vary on local
to intermediate spatial scales) studied here. Additionally,
different plot sizes may yield entirely opposite conclusions
on the importance of niche differentiation, especially if
clumping characteristics are not adequately characterized.
Analyses at larger spatial scales may reveal habitat associa-
tions that are not detectable at smaller scales, simply
because the chance for many factors to vary at small scale
is limited.
The point pattern analysis used in this study identified
pronounced species-specific aggregation patterns, in agree-
ment with Condit et al. (2000) and Plotkin et al. (2002).
However, the calculated clumping characteristics changed
when larger sampling scales were considered. In general,
larger sample scales yielded larger clump sizes. This proba-
bly occurred because large clumps are relatively rare, and
thus more likely picked up at the largest scales. If there
were a single set of ‘best’ clump characteristics (clump size,
clumps per ha, individuals per clump; Fig. 2) for these spe-
cies, we would expect clumping characteristics to remain
the same for at least several adjacent sampling scales,
which was clearly not the case. Also at smaller plot sizes,
large clumps may be quantified inaccurately due to bound-
ary effects. The variation in clumping characteristics with
sampling size suggests different mechanisms are causing
spatial aggregation depending on the spatial scale. For
example, different dispersal agents may cause recruits to
aggregate at different scales. For animal-dispersed seeds,
seeds may initially clump near parent trees. For species
with multiple animal dispersers, such as A. butyracea, dif-
ferent dispersers (agoutis, squirrels) may move seeds differ-
ent distances from the parent tree. Also, tree-fall gap
formation may generate clumps of characteristics sizes,
especially for light-demanding species. Four of the five spe-
cies in this study have low shade tolerance, and thus
recruitment is elevated in tree-fall gaps (Table 1). Previous
studies (Plotkin et al. 2002; Wiegand et al. 2007, 2009)
have quantified clustering properties at different spatial
scales, but all within plots of 50 ha. Our results suggest
different critical clustering sizes continue to develop at lar-
ger spatial scales and imply that it would be useful in future
studies to apply techniques such as those of Plotkin et al.
(2002) and Wiegand et al. (2007) to landscape-scale data
sets, so as to analyse a larger range of spatial patterns and
their underlying causes.
Some of the clump sizes were quite large at the larg-
est spatial scales, ranging from 120 m for A. standleya-
num to 500 m for J. copaia at the whole-island scale
(1560 ha). It is debatable whether these large clump
sizes represent dispersal limitation (‘self-organized
clumping’) or clumping due to habitat associations
(‘imposed clumping’). However, it must be remembered
that the trees studied here are canopy-sized individuals
with large crowns that exist at low densities inter-
spersed mainly with crowns of other species. Therefore,
a clump of 20 trees of one of the target species may
cover many hectares. However, J. copaia at the plot size
400 ha had clump sizes exceeding 200 m and more
than 40 trees per clump. In these cases, the PCP
method may be characterizing clumping patterns related
to habitat associations rather than dispersal.
Once the effect of spatial aggregation (predominantly
due to dispersal limitation) was accounted for, habitat
associations that emerged in tests under CSR often disap-
peared. While some previous studies of habitat association
have treated trees as independent units without account-
ing for the natural clumping that may arise from dispersal
limitation (e.g.Baillie et al. 1987; Clark et al. 1999; Dalle
et al. 2002; Blundell & Peart 2004; Eilu et al. 2004; Costa
et al. 2005), recent studies that do include dispersal limita-
tion find that aggregation patterns have a strong effect on
the number of planthabitat associations found for a spe-
cies (Plotkin et al. 2000; Svenning et al. 2006; Bohlman
et al. 2008; Leithead et al. 2009). Unique to this study is
evidence of how clumping patterns operate at multiple
spatial scales. In fact, we found the largest reduction in
habitat associations between CSR and PCP occurred at the
largest spatial scales, which were rarely used in previous
studies.
Journal of Vegetation Science
Doi: 10.1111/jvs.12090©2013 International Association for Vegetation Science 359
C.X. Garzon-Lopez et al. Scale-dependence of habitat association in tropical trees
We propose the following conceptual model (Fig. 6)
that integrates the probability of detecting habitat associa-
tions based on (1) species-specific scales of dispersal and
population density; (2) scale-dependent heterogeneity of
the environmental variable being studied; and (3) whether
species- and scale-dependent clumping patterns are
included in the analysis of habitat associations (in this case,
CSR vs PCP model). The probability of detecting habitat
associations increases with sampling scale as the amount of
habitat heterogeneity and number of individuals of the tar-
get species increases until thresholds are reached for both
habitat heterogeneity (specific to the spatial arrangement
of the environmental variable under study) and number of
individuals evaluated (dependent of the spatial arrange-
ment and density of the tree species studied). Above these
thresholds, the probability of detecting habitat associations
levels off. Below the maximum dispersal length of the tar-
get species, CSR will have a higher probability of detections
than PCP because CSR will be erroneously attributing
some clumping patterns due to dispersal to habitat associa-
tions. Above the habitat heterogeneity threshold, some of
the clumping patterns quantified by PCP may not be
related to dispersal, but rather to habitat association, thus
decreasing the probability of detecting habitat associations
for PCP vs CSR. Additional studies are needed to distin-
guish whether clumping patterns, especially those at large
spatial scales, are due to dispersal characteristics or habitat
heterogeneity.
Our study supports the emerging view that the spatial
distribution of tropical tree species involves both niche dif-
ferentiation and dispersal limitation, where relative impor-
tance of both factors varies with spatial scale. Rather than
being two contrasting processes, niche differentiation and
dispersal limitation interact with each other. This implies
that dispersal limitation needs to be accounted for when
niche differences are studied and vice versa.
Acknowledgements
We thank Helene Muller-Landau for her valuable help
with the aerial photos taken by Marcos Guerra, and the
Smithsonian Tropical Research Institute for their logistical
support. We are very grateful to Joe Wright for his vital
support. We wish to thank Sara Pinzon-Navarro, Noelle
Beckman and Eduardo Medina for all their support and
help in the analysis. This work was funded by the Ubbo
Emmius scholarship programme of the University of Gron-
ingen (CXG).
References
Alonso, D., Etienne, R.S. & McKane, A.J. 2006. The merits of
neutral theory. Trends in Ecology & Evolution 21: 451457.
Baddeley, A.J. & Silverman, B.W. 1984. A cautionary example
on the use of second-order methods for analyzing point pat-
terns. Biometrics 40: 10891093.
Baillie, I.C., Ashton, P.S., Court, M.N., Anderson, J.A.R., Fitzpa-
trick, E.A. & Tinsley, J. 1987. Site characteristics and the dis-
tribution of tree species in mixed Dipterocarp forest on
tertiary sediments in Central Sarawak, Malaysia. Journal of
Tropical Ecology 3: 201220.
Balvanera, P., Lott, E., Segura, G., Siebe, C. & Islas, A. 2002. Pat-
terns of b-diversity in a Mexican tropical dry forest. Journal of
Vegetation Science 13: 145158.
Barthold, F.K., Stallard, R.F. & Elsenbeer, H. 2008. Soil nutrient
landscape relationships in a lowland tropical rainforest in
Panama. Forest Ecology and Management 255: 11351148.
Besag, J.E. & Diggle, P.J. 1977. Simple Monte Carlo tests for spa-
tial pattern. Applied Statistics 26: 327333.
Blundell, A.G. & Peart, D.R. 2004. Density-dependent popula-
tion dynamics of a dominant rain forest canopy tree. Ecology
85: 704715.
Bohlman, S., Laurance, W., Laurance, S.G., Nascimento,
H.E.M., Fearnside, P.M. & Andrade, A. 2008. Importance of
soils, topography and geographic distance in structuring cen-
tral Amazonian tree communities. Journal of Vegetation Science
19: 863874.
Ceccon, E., Huante, P. & Campo, J. 2003. Effects of nitrogen and
phosphorus fertilization on the survival and recruitment of
seedlings of dominant tree species in two abandoned tropical
dry forests in Yucat
an, Mexico. Forest Ecology and Management
182: 387402.
Fig. 6. Conceptual model for the detection of habitat associations in
relation tosampling scale. The probability ofdetecting habitat associations
increases with sampling scale, as the amount of habitat heterogeneity and
population size of trees increases, until a threshold is reached for both
tree population size and habitat heterogeneity, at which the probability of
detecting habitat association levels off. Below the maximum dispersal
length (dark grey area), CSR will yield a higher probability of detections
than PCP because CSR will attribute some clumping patterns to habitat
associations rather than dispersal limitation. Above the habitat
heterogeneity threshold (light grey area), PCP will start to attribute some
habitat associations to dispersal-related clumping, thus decreasing the
probability of detecting habitat associations for PCP vs CSR.
Journal of Vegetation Science
360 Doi:10.1111/jvs.12090 ©2013 International Association for Vegetation Science
Scale-dependence of habitat association in tropical trees C.X. Garzon-Lopez et al.
Clark, D.A., Clark, D.B., Sandoval, R. & Castro, M.V. 1995.
Edaphic and human effects on landscape-scale distributions
of tropical rain forest palms. Ecology 76: 25812594.
Clark, D.B., Palmer, M.W. & Clark, D.A. 1999. Edaphic factors
and the landscape-scale distributions of tropical rain forest
trees. Ecology 80: 26622675.
Condit, R. 1998. Tropical forest census plots: methods and results from
Barro Colorado Island. Springer, Berlin, DE.
Condit, R., Ashton, P.S., Baker, P.S., Buyavejchewim, S.,
Gunatilleke, S., Gunatilleke, N., Hubbell, S.P., Foster, R.B.,
Itoh, A. (...) & Yamakura, T. 2000. Spatial patterns in the
distribution of tropical tree species. Science 288: 14141418.
Correa, M.D.A., Galdames, C. & Stapt, M. 2004. Cat
alogo de las
plantas vascula res de Panam
a. Quebecor World Bogota, S.A.,
Bogota, Colombia.
Costa, F.R.C., Magnusson, W.E. & Luizao, R.C. 2005. Mesoscale
distribution patterns of Amazonian understory herbs in rela-
tion to topography, soil and watersheds. Journal of Ecology
93: 863878.
Cottenie, K. 2005. Integrating environmental and spatial pro-
cesses in ecological community dynamics. Ecology Letters 8:
11751182.
Dalle, S.P.B., Lopez, H., Diaz, D., Legendre, P. & Potvin, C.
2002. Spatial distribution and habitats of useful plants:
an initial assessment for conservation on an indigenous
territory, Panama. Biodiversity and Conservation 11: 637
667.
De Smith, M.J., Goodchild, M.F. & Longley, P.A. 2007. Geospatial
analysis: a comprehensive guide to principles, techniques and soft-
ware tools.TroubadorPublishing,Leicester,UK.
Duque, A., Sanchez, M., Cavelier, J. & Duivenvoorden, J.F.
2002. Different floristic patterns of woody understorey and
canopy plants in Colombian Amazonia. Journal of Tropical
Ecology 18: 499525.
Eilu, G., Hafashimana, D.L.N. & Kasenene, J.M. 2004. Density
and species diversity of trees in four tropical forests of the
Albertine rift, western Uganda. Diversity and Distributions 10:
303312.
Enders, K. 1935. Mammalian life histories from Barro Colorado
Island, Panama. Bulletin of the Museum of Comparative Zoology
78: 385502.
Garzon-lopez, C.X., Bohlman, S.A., Olff, H. & Jansen, P.A. 2012.
Mapping tropical forest trees using high-resolution aerial dig-
ital photographs. Biotropica. 45: 308316.
Gilbert, B. & Lechowicz, M.J. 2004. Neutrality, niches, and dis-
persal in a temperate forest understory. Proceedings of the
National Academy of Sciences of the United States of America 101:
76517656.
Givnish, T.J. 1999. On the causes of gradients in tropical tree
diversity. JournalofEcology87: 193210.
Gotelli, N.J. & Graves, G.R. 1996. Null models in ecology. Smithso-
nian Institution, Washington, DC, US.
Harms, K.E., Condit, R., Hubbell, S.P. & Foster, R.B. 2001. Habi-
tat associations of trees and shrubs in a 50-ha neotropical for-
est plot. Journal of Ecology 89: 947959.
Hubbell, S.P. 2001. The unified neutral theory of biodiversity and bio-
geography. Princeton University Press, Princeton, NJ, US.
Hubbell, S.P. 2005. Neutral theory in community ecology and
the hypothesis of functional equivalence. Functional Ecology
19: 166172.
Hubbell, S.P., Foster, R.B., O’Brien, S.T., Harms, K.E., Condit, R.,
Wechsler, B., Wright, S.J. & de Lao, S. 1999. Light gap distur-
bances, recruitment limitation, and tree diversity in a neo-
tropical forest. Science 283: 554557.
Itoh, A., Ohkubo, T., Nanami, S., Tan, S. & Yamakura, T. 2010.
Comparison of statistical tests for habitat associations in
tropical forests: a case study of sympatric dipterocarp trees
in a Bornean forest. Forest Ecology and Management 259:
323332.
John, R., Dalling, J.W., Harms, K.E., Yavitt, J.B., Stallard, R.F.,
Mirabello, M., Hubbell, S.P., Valencia, R., Navarrete, H., Val-
lejo, M. & Foster, R.B. 2007. Soil nutrients influence spatial
distributions of tropical tree species. Proceedings of the National
Academy of Sciences of the United States of America 104: 864869.
Johnson, M.J. & Stallard, R.F. 1989. Physiographic controls on
the composition of sediments derived from volcanic and sed-
imentary terrains on Barro Colorado Island, Panama. Journal
of Sedimentary Petrology 59: 768781.
Keddy, P.A. 1982. Population ecology on an environmental gra-
dient: Cakile edentula on a sand dune. Oecologia 52: 348355.
Kneitel, J.M. & Chase, J.M. 2004. Trade-offs in community ecol-
ogy: linking spatial scales and species coexistence. Ecology Let-
ters 7: 6980.
Lam, N.S. & Quattrochi, D.A. 1992. On the issues of scale, resolu-
tion, and fractal analysis in the mapping sciences. The Profes-
sional Geographer 44: 8898.
Leigh, E.G. 1999. Tropical forest ecology: a view from Barro Colorado
Island. Oxford University Press, Oxford, UK.
Leithead, M.D., Anand, M. & Deeth, L. 2009. A synthetic
approach for analyzing tropical tree spatial patterns through
time. Community Ecology 10: 4552.
Losos, E. & Leigh, E.G. 2004. Tropical forest diversity and dynamism.
University of Chicago Press, Chicago, IL, US.
McGill, B.J. 2010. Matters of scale. Science 328: 575576.
Nathan, R. & Muller-Landau, H.C. 2000. Spatial patterns of seed
dispersal, their determinants and consequences for recruit-
ment. Trends in Ecology & Evolution 15: 278285.
Phillips, O.L., Nu~
nez, P., Monteagudo, A.L., Cruz, A.P., Chuspe,
M., Sanchez, W.G., Yli-Halla, M. & Rose, S. 2003. Habitat
association among Amazonian tree species: a landscape scale
approach. Journal of Ecology 91: 757
775.
Plotkin, J.B., Potts, M.D., Leslie, N., Manokaran, N., LaFrankie,
J. & Ashton, P.S. 2000. Speciesarea curves, spatial aggrega-
tion, and habitat specialization in tropical forests. Journal of
Theoretical Biology 207: 8199.
Plotkin, J.B., Chave, J. & Ashton, P.S. 2002. Cluster analysis of
spatial patterns in Malaysian tree species. The American Natu-
ralist 160: 629644.
Pulliam, H.R. 2000. On the relationship between niche and dis-
tribution. Ecology Letters 3: 349361.
Journal of Vegetation Science
Doi: 10.1111/jvs.12090©2013 International Association for Vegetation Science 361
C.X. Garzon-Lopez et al. Scale-dependence of habitat association in tropical trees
Rhode, K. 1992. Latitudinal gradients in species diversity: the
search for the primary cause. Oikos 65: 514527.
Ricklefs, R.E. 2008. Disintegration of the ecological community.
The American Naturalist 172: 741750.
Ripley, B. 1976. The second order analysis of stationary point
processes. Journal of Applied Probabilities 13: 255266.
Russo, S.E., Davies, S.J., King, D.A. & Tan, S. 2005. Soil-related
performance variation and distributions of tree species in a
Bornean rain forest. Journal of Ecology 93: 879889.
Snyder, R.E. & Chesson, P. 2004. How the spatial scales of dis-
persal, competition and environmental heterogeneity inter-
act to affect coexistence. The American Naturalist 164: 633
650.
Svenning, J.C. 2001. Environmental heterogeneity, recruitment
limitation and the mesoscale distribution of palms in a tropi-
cal montane rain forest (Maquipucuna, Ecuador). Journal of
Tropical Ecology 17: 97113.
Svenning, J.C., Kinner, D.A., Stallard, R.F., Engelbrecht, B.M.J.
& Wright, S.J. 2004. Ecological determinism in plant com-
munity structure across a tropical forest landscape. Ecology
85: 25262538.
Svenning, J.C., Engelbrecht, B.M.J., Kinner, D.A., Kursar, T.A.,
Stallard, R.F. & Wright, S.J. 2006. The relative roles of envi-
ronment, history and local dispersal in controlling the distri-
butions of common tree and shrub species in a tropical forest
landscap e, Panama. Journal of Tropical Ecology 22: 575586.
Trichon, V. 2001. Crown typology and identification of rainforest
trees on large-scale aerial photographs. Plant Ecology 153:
301312.
van Gemerden, B.S., Etienne, R.S., Olff, H., Hommel, P.W.F.M.
& van Langevelde, F. 2005. Reconciling methodologically
different biodiversity assessments. Ecological Applications 15:
17471760.
Vormisto, J., Svenning, J.C., Hall, P. & Balslev, H. 2004. Diversity
and dominance in palm (Aracaceae) communities in terra
firme forests in the western Amazon basin. Journal of Ecology
92: 577588.
Whittaker, R.H., Levin, S.A. & Root, R.B. 1975. Reasons for dis-
tinguishing niche, habitat and ecotope. The American Natural-
ist 109: 479482.
Whittaker, R.J., Willis, K.J. & Field, R. 2001. Scale and species
richness: towards a general, hierarchical theory of species
diversity. Journal of Biogeography 28: 247254.
Wiegand, T., Gunatilleke, S. & Gunatilleke, N. 2007. Species
associations in a heterogeneous Sri Lankan Dipterocarp for-
est. The American Naturalist 170: 7795.
Wiegand, T., Martinez, I. & Huth, A. 2009. Recruitment in tropi-
cal tree species: revealing complex spatial patterns. The Amer-
ican Naturalist 174: 106140.
Wright, J.S. 2002. Plant diversity in tropical forests: a review of
mechanisms of species coexistence. Oecologia 130: 114.
Wu, J. 2004. Effects of changing scale on landscape pattern anal-
ysis: scaling relations. Landscape Ecology 19: 125138.
Wu, H. & Li, Z.-H. 2009. Scale issues in remote sensing: a review
on analysis, processing and modeling. Sensors 9: 17681793.
Supporting Information
Additional supporting information may be found in the
online version of this article:
Appendix S1. Methods and outcomes of published
studies on habitat associations of tropical forest trees.
Appendix S2. Ripley’s Kfunctions for five species on
Barro Colorado Island, Panama.
Appendix S3. Table of habitat associations in CSR-
generated simulations.
Appendix S4. Table of habitat associations for PCP-
generated simulations.
Journal of Vegetation Science
362 Doi:10.1111/jvs.12090 ©2013 International Association for Vegetation Science
Scale-dependence of habitat association in tropical trees C.X. Garzon-Lopez et al.
... As a way for analyzing the pattern of trees, SPPA allows one to characterize forest structure (i.e., dispersed, random, clustered) and test ecological hypotheses about underlying natural processes (Law et al. 2009, Ripley 1981, Velázquez et al. 2016, Wiegand & Moloney 2013. For example, SPPA has been used to investigate the effects of seed dispersal mechanisms (Garzon-Lopez et al. 2014, Lee et al. (2022)/Math. Comput. ...
... The result of SPPA is influenced by many factors. For example, spatial scale (size of sampling site) for the detection of tree patterns can be important, as larger sampling scales allow one to better detect a spatial pattern, such as clustering, that are not evident at smaller scales (Carrer et al. 2018, Garzon-Lopez et al. 2014. Further, using the heterogeneous Poisson model as a null model for emulating CSR may provide more reliable results than using the homogeneous Poisson model, since constant intensity across a study site may not be guaranteed (Carrer et al. 2018). ...
... Current SPPA studies do not consider the inherent error in positions determined by GNSS receivers, potentially leading to analysis error when spatial relationships are based on the distance between objects (i.e., distances between objects or counts of objects within certain distance). To overcome these challenges, different approaches to estimate the positions of trees have been explored, including the use of base maps facilitated by satellite images (Atkinson et al. 2007, Moustakas et al. 2008, aerial images from aircraft or unmanned vehicles (Garzon-Lopez et al. 2014, Moustakas et al. 2008, Xu et al. 2019, or LiDAR (light detection and ranging) including terrestrial and airborne laser scanning (Trochta et al. 2013). However, there are limitations in estimating the location of tree stems from these types of images, due to image quality, spatial resolution, and feature displacement. ...
Article
Full-text available
Global navigation satellite systems (GNSS) can provide valuable spatial information for effectively mapping and navigating through complex terrain and forest conditions. Relatively accurate positional information is essential for certain algorithms and models that base analyses on the spatial arrangement of trees, and for management of forestry operations. The accuracy of GNSS receivers has been well-tested under many environmental conditions. Depending on the technology selected and conditions within which it is employed, different amounts of variation will occur in the determination of a horizontal position. However, studies involving the spatial pattern and distribution of tree locations (point positions) observed by independent GNSS receivers generally have not considered the horizontal position error inherent in the spatial data. We conducted this study to investigate whether tree locations determined by recreation-and mapping-grade GNSS receivers can adequately represent the real point pattern of trees in a forest. The study area was a pine seed orchard located at the Whitehall Forest in Athens, Georgia (USA), that consisted of a regular pattern of trees. We tested three different GNSS receivers: one mapping-grade receiver and two recreation-grade receivers (traditional, handheld-type, and non-traditional types, GPS watch). With each receiver we determined tree locations at cardinal points around the stems of 112 trees (at North, South, East, and West, sides of the stems) and estimated the middle point measurement of two cardinal points (North-South and East-West). In addition, we used the average of all cardinal points (All) to determine tree locations. We compared these observed tree locations to actual tree locations, which were determined through precise field measurements and high-precision GPS base points. This study confirmed that the horizontal positional error of mapping grade receivers was significantly lower than those of recreation grade receivers, regardless of measurement method. However, the observed point pattern of trees from the GNSS observations of both recreation-and mapping-grade receivers failed to adequately represent the actual regular point pattern of the trees, as the positional error observed was not consistently projected in the same direction and with the same magnitude.
... Habitat filtering and dispersal limitations are regarded as the two major mechanisms of population distribution. The importance of habitat filtering increases as the spatial scale increases, while the importance of diffusion limitation decreases (Garzon-Lopez et al. 2014). Our study found that the homogeneous Thomas model fitted the spatial patterns of F. pashanica better than the other null models in all three plots (Figure 2, Table 2) and these results indicate that the aggregated patterns of F. pashanica could be caused by the dispersal limitation process in all three plots, especially at small scales (< 5 m). ...
... Furthermore, we found that the habitat filtering hypothesis can explain the pattern on a large scale (> 20 m) in the middle-and highestelevation plots ( Figure 2). As the scale increases, the importance of environmental heterogeneity is suggested to increase in community assembling, while the importance of random drift and seed dispersal decreases (Wiens 1989;Legendre et al. 2009;Garzon-Lopez et al. 2014). Habitat filtering accounted for patterns of phylogenetic and functional beta diversity at larger scales (150-250 m) in two temperate forests . ...
Article
Background Examining the spatial patterns of species distributions and their underlying processes is important for characterising population dynamics and can provide novel insights for conservation management. However, little attention has been paid to spatial distribution patterns of endangered species. Aims We quantified the effects of plant interactions and environmental heterogeneity on the spatial distribution of endangered Fagus pashanica in communities, to reveal the processes which may account for its population dynamics. Methods We collected spatial coordinates of each tree in communities at three elevations and evaluated the effects of plant interactions and environmental heterogeneity using point pattern analysis. Results Dispersal limitation rather than habitat filtering shaped the spatial patterns of F. pashanica at all three elevations. Intraspecific competition in F. pashanica was found to be significant at middle- and high elevations; interspecific interactions were not significant at any of the three elevations. Conclusions Intraspecific competition significantly affected the spatial patterns of F. pashanica. Dispersal limitation appear to lead to aggregation, while at small spatial scales intraspecific interactions are likely to decrease aggregation due to potential density-dependent thinning effects.
... To evaluate how tree losses vary across the landscape, we generated a regular grid of 50-ha hexagonal cells overlaying the entire concession areas (Supplementary material Fig. S5) and estimated the percentage of basal area loss for each dispersal mode within each cell. We use a 50 ha cell size as this is the standard large tree plot size in tropical forest studies (Condit, 1995;Garzon-Lopez et al., 2014). All analyses were conducted in R version 3.6.3 ...
Article
Full-text available
Fruits and seeds are key food resources for most Amazonian mammals and birds. Selective logging is an increasingly dominant land use in the region that can deplete these resources over large areas. However, this potential impact remains poorly studied. Here we assess potential losses of animal-dispersed (endozoochorous and synzoochorous) trees resulting from reduced-impact logging in Amazonian forest concessions. We use data from forestry surveys conducted by concession companies that include the location, identity and fate (logged or not) of large (≥ 40 cm diameter at breast height) individual trees within concessions to quantify absolute and relative losses of animal-dispersed trees in the landscape. We found that most individual trees (66%) within concessions belong to animal-dispersed genera. However, despite their predominance these trees were significantly less targeted for logging than abiotically-dispersed trees so that their losses were much lower than expected based on overall harvest intensities. However, at least ten percent of all large animal-dispersed trees were lost from the entire landscape, with site-level (50-ha plots) losses sometimes exceeding one third of all animal-dispersed trees. Results suggest that the relatively low level of logging for animal-dispersed trees can still deplete frugivore resources in selectively logged forests.
... Point pattern analysis based on spatial locations of plant individuals obtained from highresolution optical images or lidar data has been extensively used for inferring plant-plant interactions (Inference III) (Atkinson et al., 2007;Garzon-Lopez et al., 2014;Moustakas et al., 2008). Local facilitation tends to lead to higher frequencies of co-occurrence of the interacting plants. ...
Article
Full-text available
1. Rapid technological advancements and increasing data availability have improved the capacity to monitor and evaluate Earth’s ecology via remote sensing. However, remote sensing is notoriously ‘blind’ to fine‐scale ecological processes such as interactions among plants, which encompass a central topic in ecology. 2. Here, we discuss how remote sensing technologies can help infer plant‐plant interactions and their roles in shaping plant‐based systems at individual, community, and landscape levels. At each of these levels, we outline the key attributes of ecosystems that emerge as a product of plant‐plant interactions and could possibly be detected by remote sensing data. We review the theoretical bases, approaches, and prospects of how inference of plant‐plant interactions can be assessed remotely. 3. At the individual level, we illustrate how close‐range remote sensing tools can help to infer plant‐plant interactions, especially in experimental settings. At the community level, we use forests to illustrate how remotely‐sensed community structure can be used to infer dominant interactions as a fundamental force in shaping plant communities. At the landscape level, we highlight how remotely‐sensed attributes of vegetation states and spatial vegetation patterns can be used to assess the role of local plant‐plant interactions in shaping landscape ecological systems. 4. Synthesis. Remote sensing extends the domain of plant ecology to broader and finer spatial scales, assisting to scale ecological patterns and search for generic rules. Robust remote sensing approaches are likely to extend our understanding of how plant‐plant interactions shape ecological processes across scales – from individuals to landscapes. Combining these approaches with theories, models, experiments, data‐driven approaches, and data analysis algorithms will firmly embed remote sensing techniques into ecological context and open new pathways to better understand biotic interactions.
... In our study, concordance was highly scale-dependent ( Fig. 3B), losing significance when the resolution was reduced beyond 4000/4,500 m. Scale-dependency has been pointed as an explanation to inconsistencies in habitat association studies of plant communities, highlighting the importance of adequately accounting for taxon-specific aggregating patterns resulting from dispersal limitations (Garzon-Lopez et al., 2014). Scale certainly accounts for some of the inconsistencies in cross-taxa concordance (Westgate et al., 2014), and our methodology was able to depict this effect in the evaluated surrogate taxa. ...
Book
Full-text available
Este compêndio de trabalhos técnico-científicos elaborados na Gerência de Espeleologia e Tecnologia trata-se de uma síntese dos principais artigos publicados em periódicos nacionais e internacionais e congressos científicos entre os anos de 2018 e 2019. No intuito de se aprimorar os conhecimentos da dinâmica e funcionamento dos ecossistemas cavernícolas ao entorno das atividades de mineração, os artigos aqui apresentados correspondem a uma representação do status do conhecimento técnico da equipe de Espeleologia da Vale em temas correlatos a espeleologia, como taxonomia, ecologia, biologia molecular, geofísica, geotecnia e sismografia; todas aplicáveis à conservação e/ou uso sustentável de cavernas em litologias ferríferas. Os dados coletados nos últimos anos através de pesquisas de campo, monitoramentos contínuos e avançados sistemas de transmissão remota são encaminhados a rigorosos processos de validação e posteriormente submetidos a estudos de modelagem, resultando em importantes interpretações para subsidiar projetos pertinentes a definição de áreas de influência de cavidades, estudos de avaliação de impactos e otimização de atividades de mineração concomitantes à manutenção da integridade física e ecológica das cavernas situadas ao entorno das minas em operação de lavra. A busca pela constante interação com o desenvolvimento científico e acadêmico no tema espeleologia em áreas de mineração e aqui representados pelas publicações em revistas especializadas e eventos científicos relacionados a Espeleologia é uma das ações onde a Vale reafirma seu compromisso socioambiental, demonstrando o cumprimento do seu papel de operador sustentável nas regiões onde atua. Boa leitura! Iuri Viana Brandi Gerente Espeleologia e Tecnologia VALE
... Several studies have suggested that variation in species distribution and composition can be explained more by environmental variables at broader scales (Arellano et al., 2016;Garzon-Lopez et al., 2014;Xing & He, 2019), likely because greater environmental heterogeneity is encompassed as the sampling extent increases (Chase, 2014;Viana & Chase, 2019). Our results not only clearly support such scale dependency but also demonstrate that these patterns are quite general across the globe (Figure 2). ...
Article
Full-text available
Beta(β)‐diversity, or site‐to‐site variation in species composition, generally decreases with increasing latitude, and the underlying processes driving this pattern have been challenging to elucidate because the signals of community assembly processes are scale‐dependent. In this meta‐analysis, by synthesising the results of 103 studies that were distributed globally and conducted at various spatial scales, we revealed a latitudinal gradient in the detectable assembly processes of vascular plant communities. Variations in plant community composition at low and high latitudes were mainly explained by geographic variables, suggesting that distance decay and dispersal limitations causing spatial aggregation are influential in these regions. In contrast, variation in species composition correlated most strongly with environmental variables at mid‐latitudes (20–30°), reflecting the importance of environmental filtering, although this unimodal pattern was not statistically significant. Importantly, our analysis revealed the effects of different spatial scales, such that the correlation with spatial variables was stronger at smaller sampling extents, and environmental variables were more influential at larger sampling extents. We concluded that plant communities are driven by different community assembly processes in distinct biogeographical regions, suggesting that the latitudinal gradient of biodiversity is created by a combination of multiple processes that vary with environmental and species size differences. We found a clear unimodal latitudinal trend in the relative importance of the community assembly processes; that is, spatial processes were more important in lower and higher latitudes, whereas environmental processes were most influential in the mid‐latitudinal regions.
... In the hitherto most exhaustive review of the effects of soil factors on floristic composition in tropical forests, Sollins (1998) indicated that the most important edaphic variables that affect evergreen lowlands forests were, in a decreasing order of importance: phosphorus availability, aluminum content, moisture (in terms of drainage and water-holding capacity) and availability of base-metal cations (potassium, calcium, and magnesium). Recent studies, however, provide only partial support to these results, with differences arising mostly from variation in the soil variables measured, sampling methods, analytical approaches and spatial scales considered, making generalizations difficult to establish (Chave, 2008;Baraloto et al., 2013;Garzon-Lopez et al., 2014). ...
Article
Full-text available
A vast literature indicates that environment plays a paramount role in determining floristic composition in tropical forests. However, it remains unclear which are the most important environmental factors and their relative effect across different spatial scales, plant life forms or forest types. This study reviews the state of knowledge on the effect of soil and climate on floristic composition in tropical forests. From 137 publications, we collated information regarding: (1) spatial scale, continent, country, life form, and forest type; (2) proportion of variance in floristic composition explained by soil and climatic variables and how it varies across spatial scales; and (3) which soil and climate variables had a significant relationship on community composition for each life form and forest type. Most studies were conducted at landscape spatial scales (67%) and mainly in South America (74%), particularly in Brazil (40%). Studies majorly focused on trees (82%) and on lowland evergreen tropical forests (74%). Both soil and climate variables explained in average the same amount (14% each) of the variation observed in plant species composition, although soils appear to exert a stronger influence at smaller spatial scales while climate effect increases toward larger ones. Temperature, precipitation, seasonality, soil moisture, soil texture, aluminum, and base cations-calcium and magnesium-and their related variables (e.g., cation exchange capacity, or base saturation) were frequently reported as important variables in structuring plant communities. Yet there was variability when comparing different life forms or forest types, which renders clues about certain ecological peculiarities. We recommend the use of standardized protocols for collecting environmental and floristic information in as much as possible, and to fill knowledge gaps in certain geographic regions. These actions will be especially beneficial to share uniform data between researchers, conduct analysis at large spatial scales and get a better understanding of the link between soils and climate gradients and plant strategies, which is key to propose better conservation policies under the light of global change.
Article
Ecological communities are assembled through a series of multiple processes, including dispersal, abiotic and biotic filtering, and ecological drift. Although these assembly processes act in concert to structure local communities, their relative importance is considerably variable among study systems. While such contingency of community assembly has been widely appreciated, the empirical and theoretical evidence is scattered around in the literature, and few efforts have been made to synthesize it. In this mini‐review, we summarize the accumulated evidence of the context‐dependency of community assembly rules, to reach a rough generalization of the contingency. Specifically, we argue that spatial and temporal dimensions can serve as general axes that regulate the relative importance of assembly processes. To this end, we synthesize the current understanding of how the relative importance of multiple assembly processes changes with spatial scales and complexity, and with time in the long and short terms. This review concludes that spatial and temporal dimensions can be common currencies of community assembly rules that are shared across various systems. “Mapping” the relative importance of community assembly processes onto these two dimensions will help us predict how communities are assembled in a given spatiotemporal context.
Article
Forest landscapes pattern and development are affected by environment and disturbance. Disentangling their effects is important to understanding current landscape and predicting future changes. Such studies are limited by short-term observation and sparse disturbance-history data. Spatially-explicit forest landscape modeling represents a solution to these limitations. Here, we reconstructed the 300-year-time-series (1710–2010) of post-volcanic-eruption forest landscapes experiencing periodic-typhoons in Changbai Mountain, China, using LANDIS forest landscape model. We used a factorial simulation design to quantify the main and interactive effects of environment and typhoon on forest landscape recovery. Results showed environment had dominant effects (>80%) on early recovery (1710–1760), suggesting early forest development follows deterministic community-assembly processes governed by environment. However, as forest matured, disturbance became dominant (>50%) at later-recovery stages (1860–2010). This study showed that historical landscape reconstruction reveals the full spectrum of interplays of environment, disturbance, and succession in forest ecosystems, which may not be captured by short-term studies.
Article
Full-text available
We evaluated the relative importance of environmental and spatial variables in structuring tadpole metacommunities at three hierarchical spatial scales in two ecoregions of the Atlantic Forest. Tadpole communities were generally more spatially structured at larger spatial scales. Environmental filtering processes seemed to be equally important at both large and small scales, however, spatially structured environmental variables were more important at larger scales, whereas non-spatially structured ones were more important at smaller scales. The importance of local environmental filters was also greater in the ecoregion that exhibits harsher environmental conditions, such as higher intensity of land-use patterns and dryer climate. This may indicate a greater role for stochastic assembly processes in more benign environments. Species co-occurrence patterns at the larger spatial scale mostly indicated biogeographical differences between the two ecoregions in a Clementsian structure, exhibiting two clear groups of species composition. Co-occurrence patterns at the smallest spatial scale were coherent only when strong environment gradients such as canopy cover were present. However, at the intermediate spatial scale the metacommunities became consistently non-coherent suggesting that species may be responding to processes acting at different spatial scales.
Article
This paper provides a rigorous foundation for the second-order analysis of stationary point processes on general spaces. It illuminates the results of Bartlett on spatial point processes, and covers the point processes of stochastic geometry, including the line and hyperplane processes of Davidson and Krickeberg. The main tool is the decomposition of moment measures pioneered by Krickeberg and Vere-Jones. Finally some practical aspects of the analysis of point processes are discussed.
Article
In the humid tropical environment source rocks of similar lithological character produce sediments of differing compositions in different physiographic environments. Moreover, sediments derived from source rocks of dissimilar mineralogical composition but produced on similar physiographic terrains may be nearly indistinguishable. Physiography influences the composition of sediments released from the source areas by controlling the duration of chemical weathering. On relatively flat terrains, rates at which loose solids are generated by weathering may exceed rates at which slope processes can remove this material. On steeper slopes, potential solid erosion rates exceed weathering rates. Residence times of minerals within soils are short, resulting in weathering products that are cation rich and contain an abundance of unstable phases. -from Authors
Article
Ongoing large-scale habitat disturbance requires quick identification of conservation priorities such as targeting sites rich in species and/or endemics. Biodiversity assessments are time consuming and expensive, so surveys often rely on partial sampling. Optimal use should be made of all currently available sources of information, but methodological differences between surveys hamper direct comparison. Because diversity depends on spatial scale, diversity characteristics of different sites are best compared on the basis of species-area relationships. As a result of the incompleteness of sampling, the observed species-area relationship deviates from the "true" species-area relationship. In this paper, we identify five key factors affecting the shape of the species-area relationship due to incomplete sampling: (1) the total spatial extent of the observations, (2) the spatial distribution of the observations, (3) the proportion of the total extent sampled, (4) the proportion of the individuals in the sampled area included in the survey, and (5) the proportion of the included individuals successfully identified. We outline how methodologically different surveys can be combined to optimize the use of existing data in the evaluation of conservation needs, particularly for tropical forests. As an illustration, we analyzed four methodologically different botanical surveys in the same area of old growth lowland forest in South Cameroon with the aim of reconciling these surveys. The four surveys were (1) reconnaissance scale vegetation mapping, (2) detailed botanical assessment (all individuals), (3) incomplete botanical assessment (10% individuals), and (4) herbarium collections. By correcting for the five key factors we were able to match the results of the four different biodiversity surveys. The five key factors affected the recorded number of species and endemics differently; partial sampling of extent (3) and individuals (4) and partial identification of individuals (5) were the three most important factors. We conclude that reconciliation of biodiversity assessments is possible if the differences between methods can be accounted for. We advocate reliable documentation of survey methods, especially the five key factors, because it greatly enhances the potential of combining methodologically different surveys for comparative biodiversity analyses.
Article
Light gap disturbances have been postulated to play a major role in maintaining tree diversity in species-rich tropical forests. This hypothesis was tested in more than 1200 gaps in a tropical forest in Panama over a 13-year period. Gaps increased seedling establishment and sapling densities, but this effect was nonspecific and broad-spectrum, and species richness per stem was identical in gaps and in nongap control sites. Spatial and temporal variation in the gap disturbance regime did not explain variation in species richness. The species composition of gaps was unpredictable even for pioneer tree species. Strong recruitment limitation appears to decouple the gap disturbance regime from control of tree diversity in this tropical forest.
Article
It is not true that all important information about a pattern of points is encapsulated in its second-order description. As an example, we construct a random point process which exhibits exactly the same second-order characteristics as a Poisson process, but which differs manifestly and visibly from such a process.