Ecography 39: 001–012, 2016

doi: 10.1111/ecog.02191
© 2016 The Authors. Ecography © 2016 Nordic Society Oikos
Subject Editor: Thorsten Wiegand. Editor-in-Chief: Miguel Araújo. Accepted 2 August 2016

Exploring the interaction of avian frugivory and plant spatial

heterogeneity and its effect on seed dispersal kernels using a
simulation model

Andrew P. McKenzie Pegman, George L. W. Perry and Mick N. Clout­

A. P. McK. Pegman (andrewpegman@xtra.co.nz), Laingholm, Auckland, New Zealand. – G. L. W. Perry and APMcKP, School of Environment,
The Univ. of Auckland, Auckland, New Zealand. – M. N. Clout (m.clout@auckland.ac.nz), School of Biological Sciences, The Univ. of Auckland,
Auckland, New Zealand.­

Seed dispersal by avian frugivores is one of the key processes influencing plant spatial patterns, but may fail if there
is disruption of plant–frugivore mutualisms, such as decline in abundance of dispersers, fragmentation of habitat, or
isolation of individual trees. We used simulation model experiments to examine the interaction between frugivore density
and behaviour and the spatial arrangement of fruiting plants and its effect on seed dispersal kernels. We focussed on two
New Zealand canopy tree species that produce large fruits and are dispersed predominantly by one avian frugivore
(Hemiphaga novaeseelandiae). Although the mean seed dispersal distance decreased when trees became more aggregated,
there were more frugivore flights between tree clusters, consequently stretching the tails of the dispersal kernels. Conversely,
when trees were less aggregated in the landscape, mean dispersal distances increased because seeds were deposited over larger
areas, but the kernels had shorter tails. While there were no statistically meaningful changes in kernel parameters when
frugivore density changed, decreases in density did cause a proportional reduction in the total number of dispersed seeds.
However, birds were forced to move further when fruit availability and fruit ripening were low. Sensitivity analysis showed
that dispersal kernels were primarily influenced by the model parameters relating to disperser behaviour, especially those
determining attractiveness based on distance to candidate fruiting trees. Our results suggest that the spatial arrangement
of plants plays an important role in seed dispersal processes – although tree aggregation curbed the mean seed dispersal
distance, it was accompanied by occasional long distance events, and tree dispersion caused an increase in mean dispersal
distance, both potentially increasing the probability of seeds finding suitable habitats for germination and growth. Even
though low frugivore densities did not cause dispersal failure, there were negative effects on the quantity of seed dispersal
because fewer seeds were dispersed.

Dispersal of seeds, for example by avian frugivores, in large
part determines the spatial pattern of seedling recruitment
and is essential for colonisation of landscapes via long
distance events (Wenny 2000, Santamaria et  al. 2007).
However, seed dispersal may fail if plant-disperser mutual-
isms are disrupted, for example, via decline in abundance of
keystone seed dispersers or isolation of trees (Cordeiro and
Howe 2003, Sekercioglu et  al. 2004, Nathan 2005). One
way of testing this assertion is by comparing empirical seed
distributions, obtained by matching dispersed propagules
with their parents, across different plant–frugivore scenarios
(Bullock et  al. 2006, Cortes and Uriarte 2013). However,
such data are difficult to obtain because long distance seed
dispersal is challenging to quantify (Robledo-Arnuncio
et al. 2014) and deposition patterns emerge from multiple
interactions between frugivore behaviour and the distribu-
tion of fruit, both of which vary spatially and temporally
(Levey et al. 2008, Cortes and Uriarte 2013). As an alterna-
tive, virtual in silico (Zurell et  al. 2010) experiments can
be conducted using simulation models in order to recon-
struct seed dispersal distributions and derive dispersal
kernels (including parameter values) directly from informa-
tion describing the natural history of seed production and
behaviour of dispersal agents. The seed dispersal kernel is the
probability of a seed being deposited at a particular distance
from the parent plant (Nathan and Muller-Landau 2000)
and is described in terms of its essential parameters: 1) the
mean seed dispersal distance, and 2) the rate at which the
tail of the distribution decays with distance. In the Weibull
distribution, for example, mean seed dispersal distance and
its variance set the spatial ‘scale’ of dispersal and ‘shape’, the
inverse function of the kurtosis of the kernel, indicates long
distance seed dispersal when it is numerically low (Morales
and Carlo 2006). Seed dispersal effectiveness incorporates
these parameters and is the product of the total number of
dispersed seeds and the probability that a dispersed seed
produces a new adult, all of which influence demographic
features of plant populations, such as the rate and pattern of

Early View (EV): 1-EV

spread and colonisation of new areas (Schupp 1993, Cousens
and Rawlinson 2001, Levin et al. 2003, Spiegel and Nathan
2007, Schupp et al. 2010).
To quantify the effect of disruption of plant–frugivore
mutualisms via decline in frugivore density or changes in
tree spatial patterns, we developed a spatially explicit indi-
vidual-based simulation model (SEIBM, DeAngelis and
Mooij 2005) combining disperser foraging and movement
rules with data describing frugivore gut passage time of
seeds and individual tree distributions. We varied frugivore
density across different spatial patterns of trees to evalu-
ate how seed dispersal kernels were affected by changes in
frugivore density or by the effects of forest structure on fru-
givore movements. Based on our results, we discuss what the
demographic effects might be for the tree species we con-
sider. We applied our model to Hemiphaga novaeseelandiae
(New Zealand pigeon or kereru, Columbidae), which
is endemic to New Zealand (NZ) and has a conserva-
tion dependent but increasing status (Clout et  al. 1991,
Miskelly et  al. 2008, Kelly et  al. 2010). This large frugi-
vore is the only widespread avian disperser capable of
dispersing the propagules of large-fruited native trees,
such as our exemplars, Vitex lucens (NZ teak or puriri,
Verbenaceae) and Prumnopitys ferruginea (brown pine or
miro, Podocarpaceae).
Methods
Study system and species
Hemiphaga novaeseelandiae are large (ca 650 g adult) frugivo-
rous pigeons. They live in lowland forests throughout NZ
and are highly mobile, inferred from inter-annual changes
in the numbers in northern forests and radio-telemetry
data, but individuals can also be sedentary (Clout et  al.
1991, Wotton 2007). Vitex lucens are large angiosperm trees
up to 20 m tall, with trunks up to 1.5 m wide, occurring
mainly in temperate coastal areas of NZ (Salmon 1980).
Hermaphroditic flowers are produced all year, with bright
red fruits being present mainly between December and June.
Prumnopitys ferruginea are tall dioecious gymnosperm trees
up to 25 m tall, with trunks up to 1 m wide, occurring in
shady environments throughout NZ, from lowland forest to
ca 1000 m elevation (Salmon 1980). Male trees have cones
and female trees produce large red diaspores, mainly between
December and June, consisting of a single seed surrounded
by fleshy ovuliferous cone scales (McEwen 1988) – for con-
venience, these structures will henceforth also be referred to
as ‘fruit’. The fruit of both tree species are long (V. lucens:
15.4  0.2 mm, n  286 fruit; P. ferruginea: 16.1  0.1 mm,
n  702 fruit; Pegman 2012) and have single seeds that are
hard and ‘woody’ surrounded by nutritious, edible, and
‘fleshy’ pulp, indicating that they are consumed by fruit-
eating animals (Coates-Estrada and Estrada 1988, Tiffney
2004). Because fruits and birds vary in size within species,
mid-sized birds such as tui Prosthemadera novaezelandiae or
blackbird Turdus merula can consume smaller individual
fruits of both tree species, but this occurs infrequently (Kelly
et al. 2010). Mammals may also act as dispersers since seeds
of our study species occasionally showed signs of predation
2-EV
such as open husks and teeth marks; however, this is again
infrequent and Pegman (2012) showed that only 2.2  1.3%
of dispersed P. ferruginea seeds (n  434 seeds across 6 trees)
had open husks, and mammal damage to dispersed V. lucens
seeds was not observed.
We utilised these two tree species because 1) they are
the two large-fruited NZ species for which there are data
to parameterize our model, 2) they differ in their envi-
ronmental requirements, therefore not growing together,
3) they have different reproductive strategies, and 4) they
have differing spatial distributions. Mean annual fruit pro-
duction for P. ferruginea (ca 5000) is around twice that of
V. lucens, and the proportion of annual fruit production that
is actively dispersed by H. novaeseelandiae is ca 46% for both
tree species (Pegman 2012). Field measurements showed
that the distance between P. ferruginea trees (30.2  6.1
m, n  3 sites) is around twice the respective value for
V. lucens (Pegman 2012). The two tree species also have
different shaped seeds (V. lucens is irregular, whereas P.
ferruginea is smooth), affecting frugivore gut passage time
(Table 1, Supplementary material Appendix 1, Traveset
1998). All of these factors may result in different outcomes
for each tree species in our simulations.
Model design
Our model is concerned with how patterns in seed deposi-
tion emerge from interactions between frugivore behav-
iour and the environment, represented by different tree
spatial patterns. The model is semi-mechanistic, spatially
explicit, individual-based, stochastic, and event driven; it
encompasses fruit production and avian frugivore forag-
ing in simulated forests and was ‘built from the ground
up’, having fundamental elements in common with
other models of seed dispersal such as those described
by Morales and Carlo (2006) and Will and Tackenberg
(2008). We implemented our model in NetLogo 5.3
(Wilensky 1999). A complete description of our model
following the ODD (overview, design concepts, details)
protocol of Grimm et  al. (2010) and the code are pro-
vided in the Supplementary material Appendix 1 and 2
(see Supplementary material Appendix 1, Fig. A1 for
schematic overview).
In brief, our model represents the movement of variable
numbers of individual birds (in our case, H. novaeseelan-
diae) through landscapes of fixed abundance of fruiting
plants (in our case, V. lucens or P. ferruginea) whose spatial
arrangement can be varied. Individual birds select the tree
they move to based on the multiplicative interaction of
its available fruit and the distance from the bird’s current
location. The rationale for this interaction is that there is
a positive correlation of frugivory with fruit abundance
(Saracco et  al. 2005) and that frugivores prefer local,
rather than more distantly located, resources. For example,
plants in denser neighborhoods have greater fruit removal
rates than more isolated plants, even when the latter have
desirable fruit (Levey et  al. 1984, Morales et  al. 2012).
However, this rationale provides no indication about the
mathematical function that most closely describes such
relationships. In our model, there is one feeding bout at
Table 1. Baseline simulation model parameters and sources.

Parameter Description Value Source

nK Number of kereru 800 or 2 ha–1 2, 4
v Mean kereru flying speed 7 m s–1 4
e Maximum gut capacity of kereru 30 fruits 4
c Max. consumption rate in hyperbolic functional response 10 fruits visit–1 3, 4
d Half saturation in hyperbolic functional response 2 fruits 3, 4
aPT Scale of gamma distribution of kereru PT 0.67 min 5
bPT Shape of gamma distribution of kereru PT 0.02 5
aGPTP Scale of gamma distribution of miro GPT 13.89 min 1
bGPTP Shape of gamma distribution of miro GPT 0.15 1
aGPTV Scale of gamma distribution of puriri GPT 4.06 min 6
bGPTV Shape of gamma distribution of puriri GPT 0.04 6
aF Factor in tree attractiveness based on number of fruits 0.001 3
bF Exponent in tree attractiveness based on number of fruits 2 3
aD Factor in tree attractiveness based on distance 5  10–5 3
bD Exponent in tree attractiveness based on distance 2 3
pLONG Probability (p) of kereru moving to a random new tree 0.01 4
rtA Roost tree attractiveness at p  0.3 0.6 4
maxFP Maximum number of miro fruits per tree 5000 4
maxFV Maximum number of puriri fruits per tree 2500 4
rF Fruit ripening rate per day 50 3, 4
nT Number of fruiting trees 1200 or 3 ha–1 4
k Number of tree clusters per landscape 120 4
sP Degree of miro tree aggregation 100 m 4
sV Degree of puriri tree aggregation 20 m 4
mF Background fruit loss per day 0.5% 4
Notes: a  scale  mean  mean/variance, b  shape  mean/variance, a  factor, b  exponent, PT  daily perching time, GPT  frugivore
gut passage time of seeds, kereru  H. novaeseelandiae, miro  P. ferruginea, puriri  V. lucens.
Sources: 1Clout and Tilley (1992), 2Greene (2003), 3Morales and Carlo (2006), 4Pegman (2012), 5Wotton (2007), 6Wotton et al. (2008).

a random point during the time spent at each tree dur-
ing a mean of 120 minutes per day of perching time
(Wotton 2007). A gut passage time for each feeding event
is drawn from an appropriate probability distribution
(Supplementary material Appendix 1), seeds are deposited
after that time elapses, and in due course the seed disper-
sal distribution emerges from this activity (Fig. 3b). There
are no interactions among birds in the model with respect
to their seed dispersal service (they do not flock due to
current low densities, although they likely did so in the
past) and so the final seed deposition pattern emerges from
independent superposition of individual dispersal events
across all birds. The parameters of the seed dispersal kernel
are then fitted to the simulated seed deposition pattern
(see below).
Other elements in our model reflect the biology of the
systems we considered and include: 1) ‘roost’ trees i.e.
trees that are attractive to birds but do not provide a food
source (Beveridge 1964 and Johnson 2001 describe the use
of Weinmannia racemosa in this way) and 2) the occasional
longer distance random movement by individual dispersers
within the landscape, rather than on the basis of the attrac-
tiveness of trees. We also included background fruit loss
from trees due to wind, abscission, predation, mishandling
by birds, etc.
Model parameterisation, sensitivity and structural
analyses, and in silico experiments
The model parameterization was informed by data describing
the foraging behavior of H. novaeseelandiae and the
physiognomy of V. lucens and P. ferruginea, obtained through
experiments, field observations, and from the relevant
literature (Table 1). We conducted a univariate sensitivity
analysis, following the methods described by Hamby (1994),
by imposing a  25% change on each specified baseline
model input parameter and then calculating absolute ratios
(Sy, x) of change in kernel parameters to assess how these seed
dispersal characteristics changed with model parameteriza-
tion (Table 2 and Supplementary material Appendix 4). The
model parameters we tested were chosen because they were
a priori deemed to be the most relevant to the questions
we were focussing on or were the most uncertain. Because
the sensitivity analysis suggested that it was a crucial driver
of the model’s dynamics, we also conducted a structural
analysis on the mathematical function governing frugivores’
choice of tree based on the distance from its current loca-
tion (Supplementary material Appendix 1, Eq. A2). Since
we were assessing whether the degree of tree aggregation
(sigma, s) was a major driver of seed distributions, we used
the entire biologically reasonable range (i.e. s  2 to 100 m)
in the assessment of its sensitivity.
We then conducted a series of in silico plant–frugivore
experiments to explore the extent to which seed dispersal
kernels are influenced by interactions between tree spa-
tial patterns and frugivore density. A factorial design was
employed in which the density of H. novaeseelandiae (nK)
varied between 0.1 and 10 ha–1, representing the range of
plausible values (Pegman 2012). In support of this, Greene
(2003) calculated H. novaeseelandiae densities of 0.1 to 2.6
ha–1 in Pureora Forest Park, NZ, reflecting seasonal move-
ments linked to food availability. The degree of aggregation

3-EV
Table 2. Univariate sensitivity analysis (exemplified here by V. lucens; see Supplementary material Appendix 4 for P. ferruginea) according to
the methods of Hamby (1994). All model runs comprised 30 replicates, each of 14 d. Rank denotes rank of S ratio (the ratio of the relative
change in the state variable y for a given relative change in the value of parameter x) based on mean of the ratios. Cases where Sy, x  1 are
in bold. See Table 1 for description of baseline model parameter abbreviations.

Scale parameter Shape parameter

Baseline model parameter Sy, x (–25%) Sy, x ( 25%) Rank Sy, x (–25%) Sy, x ( 25%) Rank
bD  2 6.473 2.521 1 1.856 1.298 1
aD  5  10–5 0.634 0.426 2 0.260 0.195 2
aPT  0.67 min. 0.400 0.291 3 0.027 0.082 4
bGPT  0.04 0.414 0.243 4 0.021 0.014 9
bPT  0.02 0.313 0.314 5 0.017 0.034 5
aGPT  4.06 min 0.305 0.280 6 0.021 0.014 10
pLONG  0.01 0.105 0.134 7 0.168 0.086 3
bF  2 0.017 0.050 8 0.024 0.021 6
PT  120 min 0.052 0.014 9 0.007 0.010 13
maxF  2500 0.024 0.022 10 0.031 0.010 8
rtA  0.60 0.008 0.035 11 0.014 0.003 12
rF  50 0.021 0.022 12 0.021 0.003 11
aF  0.001 0.015 0.028 13 0.021 0.000 14
e  30 fruit 0.016 0.025 14 0.010 0.000 15
No. of roost trees  100 0.033 0.001 15 0.003 0.003 16
s  20 m 0.002 0.004 16 0.030 0.026 7

of trees (sigma, s) was varied by tightening or relaxing clus-
ters in the Thomas point process, used to simulate patterns
of individual trees from aggregated (low s) to diffuse (high
s), with ca 86% of all trees of a given cluster located within
2s of the cluster centre (Illian et  al. 2008, Supplementary
material Appendix 1, Fig. A2). We varied s from 10 to 200
m (representing nearest fruiting tree neighbors at ca 11 to
28 m) and held abundance constant at 1200 fruiting trees in
120 clusters per simulated landscape of 400 ha (i.e. 3 trees
ha–1), based on field estimates of the density of both tree
species (Pegman 2012). We also conducted experiments to
assess the combined effects of low fruit availability and low
fruit ripening on the seed dispersal kernels.
There were 30 simulations for each combination of fac-
tors, with each replicate comprising 120 min of daily frugi-
vore perching time (time from arrival until departure from a
tree, including foraging, Wotton 2007) across 14 d, adjust-
ing frugivore gut passage time of seeds and fruit produc-
tion according to the tree species examined (Table 1). Our
analyses used seed dispersal events that occurred during the
last three days of the model runs; this minimised the risk of
‘burn in’ effects, but still had ample sample size for robust
parameter estimation. Each individual bird started the simu-
lation at a randomly chosen tree, and the number of fruit
removed from each tree and the dispersal distance of every
seed were recorded.
Seed dispersal distributions were summed across individ-
ual birds to characterize the population level pattern for each
tree species. A Weibull probability density function (Weibull
1951, Eq. 1) was fitted to the distributions by maximum
likelihood estimation using the ‘fitdistr’ command in the
MASS library in R–3.1.1 (R Development Core Team). We
used the Weibull distribution because it is flexible, approxi-
mating a number of other probability distribution functions
that have been used to summarize seed dispersal kernels, and
it quantifies the properties of dispersal kernels that we are
interested in i.e. scale (mean seed dispersal distance) and
shape (i.e. kurtosis–1; low shape values equate to high kurto-
sis and ‘fat’ long tails and vice versa).
β x β−1
f ( x | α, β ) =   e − ( x / α )β (1)
α  α
where x  distance from tree, a  scale parameter, and
b   shape parameter, all  0.
For each analysis, we estimated the scale and shape param-
eters of the seed dispersal kernels; for these data, we were
most interested in the scenario analyses that used empiri-
cally derived ‘field’, or ‘observed pattern’, estimates of tree
aggregations while varying disperser densities, and vice versa.
We also quantified the total number of dispersed seeds, seed
rain map plots, frugivore foraging ranges (via convex hulls,
computed in the R Spatstat library, Baddeley and Turner
2005) and the flight distances that resulted in them, and
performed a spatial analysis of seed rain using a bivariate
pair correlation function (Illian et al. 2008) based on spatial
association between trees and deposition events.
Data available from the Dryad Digital Repository:
< http://dx.doi.org/10.5061/dryad.15d8v > (Pegman et  al.
2016).
Results
Sensitivity and structural analyses
The scale and shape parameters of the seed dispersal kernels
were both strongly influenced by the model parameters
relating to frugivore movement and also by gut passage
time. In particular, seed dispersal kernels were highly sensi-
tive to the coefficient determining attractiveness based on
distance to candidate fruiting tree (bD, Table 2). Otherwise,
all other model parameters were robust (absolute sensitivity
values  1) and outcomes were symmetric (i.e. increases and
decreases had similar relative effects). As expected, sigma was
robust when held constant while testing the effect of differ-

4-EV
ent frugivore densities (Table 2), but we found that there was
an increase (although less than proportional) of both shape
and scale parameters in both tree species when s increased
across all of its plausible values (Supplementary material
Appendix 3, Table A7). We assumed that there was a strong
decline in attractiveness with distance by frugivores for fruit-
ing trees, hence we used a hyperbolic function to determine
the effect of distance on tree selection; this function has been
used in other similar models that have produced plausible
biological outcomes, e. g. Morales and Carlo (2006). There
is no empirical data on bird movements at a sufficient reso-
lution to derive empirical fits for this chosen function, or to
parameterize it. However, our use of a hyperbolic function
was justified by the fact that our frugivore foraging ranges,
and movements which resulted in them, were similar to
those derived from empirical observation; we focused on
movement data and foraging ranges, rather than mean seed
dispersal distance, because there is ample empirical data for
the first two factors (see below), against which we could vali-
date our model, but little for the latter. Nevertheless, because
of the importance of this functional relationship for model
results and because it utilised bD (Supplementary material
Appendix 1, Eq. A2), which was both a sensitive and uncer-
tain model parameter (Table 2), we conducted a structural
analysis to assess the robustness of the model outcomes to
this decision. In the structural analysis we assessed three
other distance-attractiveness functions: 1) a linear decline,
2) a negative exponential decline (to zero attractiveness at
250 m), and 3) a null model (where all trees were equally
attractive irrespective of distance). The linear and negative
exponential relationships produced similar foraging range
sizes to the hyperbolic (linear  hyperbolic ∼ negative
exponential), but those from a null model were much larger
(Supplementary material Appendix 3, Table A1, exemplified
by V. lucens). Morales and Carlo (2006) also rejected a null
diffusion model because they found that mean seed disper-
sal distances were not linearly related to the degree of plant
aggregation. The fact that the outcome from the negative
exponential is similar to the hyperbolic also suggests that our
results do not depend on an arbitrary decision to use a spe-
cific function for distance attractiveness.
Seed dispersal kernels
Across both tree species, our simulated seed dispersal dis-
tributions ranged between being exponential to gamma in
appearance (Fig. 6), both of which can be described by the
Weibull distribution. The shape parameters (b), obtained by
fitting the Weibull, corresponded to mainly fat tailed curves
at high tree aggregation (b  1), exponential at b  1, and
‘hump’ shaped at low tree aggregation (b  1.1 to 1.4), but
none corresponded to simple diffusion i.e. b  2.
Variation of the distribution of trees in the landscape trans-
lated into clear differences in the scale and shape parameters
of the seed dispersal kernels. Because the movement rules are
fixed, these differences were the outcome of emergent inter-
actions between seed dispersers and the landscape structure.
Increased tree aggregation resulted in shorter mean seed dis-
persal distances and kernels with long and ‘fat’ tails, as evi-
denced by low shape parameters. Conversely, tree populations
that were less aggregated resulted in longer mean seed dis-
persal distances, which were associated with high value shape
parameters and kernels with short tails (Fig. 1a, b, 2a, b show
results at ‘observed pattern’ bird densities; Supplementary
material Appendix 3, Table A3 to A6 show seed dispersal ker-
nel parameter values obtained in different landscape scenarios
in the factorial experiments). However, the total number of
dispersed seeds was not significantly affected by the spatial
distribution of either tree species.
Variation of disperser density did not significantly
influence mean seed dispersal kernel parameters when
fruit availability was high, except in V. lucens forests at
the highest simulated frugivore density (i.e. 10 ha–1), but
these comprised a difference of only a few metres and so are
unlikely to be ecologically important (Fig. 1c, d, 2c, d show
results in ‘observed pattern’ forests; Supplementary material
Appendix 3, Table A3 to A6 show dispersal kernel param-
eter values obtained in different frugivore density scenarios
in the factorial experiments). The total number of dispersed
seeds changed directly in proportion to H. novaeseelandiae
density. However, inter-realisation variance of individual
kernel shape and scale parameters increased with decreasing
disperser density (Fig. 1d and 2d), suggesting that in these
scenarios, long distance seed dispersal events will become
increasingly stochastic, reducing the predictability of esti-
mates relying on it e.g. plant connectivity, spatial distribu-
tions, and dynamics (Nathan 2005). When we concurrently
lowered fruit availability and fruit ripening, we found that
seed dispersal distances increased when frugivore densities
increased because birds had to fly further to locate fruit
(Supplementary material Appendix 3, Table A2, exemplified
by V. lucens).
Frugivore foraging ranges
Under ‘observed pattern’, or baseline, conditions, our simulated
frugivore foraging ranges (mean  one SEM  27.3  2.3
ha, statistical range  1.2 to 134.2 ha, n  100) are consis-
tent, both in area and time frame, with those determined
in other studies (Fig. 3a, b, Pierce and Graham 1995,
Bell 1996, Hill 2003, Schotborgh 2005, Campbell 2006,
Powlesland et al. 2011). For example, Hill (2003) reported
values from ca 14 to 704 ha, Bell (1996) reported ca 20
to 30 ha, and Schotborgh (2005) reported ca 2 to 22 ha,
all similar to our values. Powlesland et al. (2011) reported
much higher values (ca 619 to 31 732 ha) because their
data captured infrequent, possibly seasonal, long-distance
movements between distinct habitats (from mainland NZ
to an offshore island). Hemiphaga novaeseelandiae had larger
foraging ranges in simulated P. ferruginea landscapes than
in V. lucens landscapes since the former were less aggregated
and therefore the birds had to fly further to forage. Our
model produced mean flight distances between 60 and
100 m, consistent with a mean  one SEM of 77.0  7.0 m in
Wotton and Kelly (2012) and ca 11 to 102 m in Powlesland
et  al. (2011). The short flight distances and rapid flying
speed of frugivores (ca 7 ms–1, Pegman 2012) meant that
most seeds were deposited at trees rather than during flight
in the model, consistent with the observed behaviour of
H. novaeseelandiae (McEwen 1978).
5-EV
Figure 1. Seed dispersal kernels and corresponding scale and shape parameters for V. lucens using the Weibull distribution, n  30 model
runs  14 d; (a) and (b) are across sigma (s, degree of tree aggregation) of 10 m (highly aggregated trees, dark blue) to 200 m (least
aggregated, red) at a disperser density of two H. novaeseelandiae (kereru) ha–1, and (c) and (d) are across kereru densities of 0.1 ha–1 (dark
blue) to 10 ha–1 (red) at s  20 m.
Seed rain map plots and spatial analysis
In highly aggregated populations of either tree species, maps
of the seed rain show that deposition mainly occurred around
tree clusters, but was anisotropic and spatially non-uniform
across the landscape; the numbers of seeds deposited differed
markedly between adjacent areas, with some receiving none,
and deposition occurred only occasionally near roost trees
(Fig. 4 and 5). Empirical measurements of seed deposition
at the sites of individual fruiting trees of our study species
showed that it is also anisotropic (Pegman 2012), reflect-
ing biotic factors such as canopy shape, perching behaviour
of birds, and overlapping seed shadows, plus abiotic effects
such as micro-topography (Bullock et al. 2006).
When fruiting trees were less aggregated, seed rain was
more uniform across the landscape, reflecting different pat-
terns of disperser movement; anisotropy was still evident, but
seed rain occurred more frequently near roost trees. In both
scenarios, seed distribution patterns were independent of bird
densities, but the total number of dispersed seeds changed
directly in proportion to disperser numbers. These outcomes
are in agreement with the seed dispersal kernel results pre-
sented above, where it was shown that tree patterns, but not
bird densities, significantly influenced the main parameters
of seed dispersal kernels when resources were not scarce. A
spatial association between trees and seed deposition events
using the bivariate pair correlation function (Illian et  al.
6-EV
2008) based on all seed dispersal events, showed that the
distances over which seed deposition occurred with respect
to fruiting trees increased when trees became less aggregated
(Fig. 6).
Discussion
We aimed to investigate how seed dispersal is influenced by
the interaction between frugivore behaviour and plant spa-
tial patterns, using in silico experiments. We varied frugivore
density across different spatial arrangements of large-fruited
tree species prevalent in NZ temperate forests and examined
the resulting changes in the main characteristics of seed dis-
persal kernels, and how these might influence the dynamics
of plant populations.
Tree spatial distribution influenced bird movement
patterns, resulting in clear differences in the main statisti-
cal characteristics of the seed dispersal kernels. As forests
became more aggregated, the mean seed dispersal distance
(or scale of the dispersal kernel) decreased because birds
were restricted mainly to clusters of high tree density. One
outcome of this limited dispersal distance was reduced seed
deposition near roost trees. Previous simulation studies have
also suggested that clumped plant spatial patterns influ-
ence disperser or pollinator behaviour by confining them to
monospecific patches (Morales and Carlo 2006, Hanoteaux
Figure 2. Seed dispersal kernels and corresponding scale and shape parameters for P. ferruginea using the Weibull distribution, n  30 model
runs  14 d; (a) and (b) are across sigma (s, degree of tree aggregation) of 10 m (highly aggregated trees, dark blue) to 200 m (least
aggregated, red) at two H. novaeseelandiae (kereru) ha–1, and (c) and (d) are across kereru densities of 0.1 ha–1 (dark blue) to 10 ha–1 (red)
at s  100 m.

et  al. 2013). Over the long-term, shortened seed dispersal
distances may amplify any existing local tree aggregation and
increase neighborhood competition and negative density-
dependent effects (Augspurger and Kelly 1984). In sup-
port of this expectation, some spatially explicit models have
shown that localized seed dispersal drives the emergence of
aggregated plant spatial distributions (Holmes and Wilson
1998, Fedriani et  al. 2010, Rodriguez-Perez et  al. 2012).
In our case, there were also occasional traverses between
tree clusters in aggregated forest settings, resulting in long

Figure 3. (a) Example of distribution of foraging ranges (ha) of 100 H. novaeseelandiae after 14 d in V. lucens and P. ferruginea forests, and
(b) example of emergence of a foraging range of a single H. novaeseelandiae after 14 d; colored points depict foraging at trees, with colours
representing visitation (blue to red, with unvisited in grey) – in this case the area visited was 22.5 ha.

7-EV
Figure 4. Seed deposition by H. novaeseelandiae across landscapes of 400 ha (2000  2000 m, showing ‘patches’, each 100 m2) with 1200
V. lucens trees (black points) and 100 roost trees (white points) after one fruiting season, at varying disperser density (D, H. novaeseelandiae
ha–1) and tree aggregation i.e. s (m), low values indicate high tree aggregation and vice versa. Legends indicate the total number of seeds
deposited ha–1 (note they vary across sub plots).

distance seed dispersal events, consequently stretching the
tail of the dispersal kernel; such events, even if rare, enhance
seed dispersal efficiency and allow plant population spread
(Howe and Miriti 2004). Conversely, when the distribu-
tion of trees was less aggregated, mean dispersal distances
increased because birds had larger foraging ranges and there-
fore had the opportunity to disperse seeds at longer distances
and in a more uniform pattern, including under roost trees.
Variation in frugivore density did not significantly
influence seed dispersal kernels in our factorial experi-
ments. Increased disperser densities had no direct effect on
seed dispersal kernels because they arose from independent
superposition of individual seed shadows of all birds, and
resources did not become scarce, so birds did not have to
venture out of fruiting patches to obtain more fruit. Both
of the tree species we considered have high annual fruit
production, with up to 83% of V. lucens fruit and 70% of
P. ferruginea fruit falling uneaten beneath their respective
canopies, due in part to low disperser numbers in modern
ecosystems (Clout and Hay 1989, Wotton and Kelly 2011,
Pegman 2012). However, fruit depletion in some ecosystems
can be very high (Forget 1992) and in such cases, frugivore
movement patterns are different. For example, the plant
taxon represented in the model by Morales and Carlo (2006)
was frequently depleted of fruit, resulting in longer frugivore
foraging movements and an increase in mean seed dispersal
distance and kurtosis of the dispersal kernels. Decreased dis-
perser abundance in our simulations did, however, result in
a proportional reduction in the total number of dispersed
seeds, with the commensurate effect of decreasing the prob-
ability of long distance seed dispersal, thus reducing seed
dispersal efficiency and potentially diminishing population
spread (Spiegel and Nathan 2007, Nathan et al. 2008). In
support of this finding, Wotton and Kelly (2011) showed
that lack of seed dispersal has resulted in low recruitment
rates in large-fruited NZ tree species. Conversely, increased
disperser abundance resulted in a proportional increase in
the total number of dispersed seeds in our simulations. We
also tested a mechanism of frugivore density that indirectly
had the potential to change the shape of the seed rain, by

8-EV
Figure 5. Seed deposition by H. novaeseelandiae across landscapes of 400 ha (2000  2000 m, showing ‘patches’, each 100 m2) with 1200
P. ferruginea trees (black points) and 100 roost trees (white points) after one fruiting season, at varying disperser density (D, H. novaeseelan-
diae ha–1) and tree aggregation i.e. s (m), low values indicate high tree aggregation and vice versa. Legends indicate the total number of
seeds deposited ha–1 (note they vary across sub plots).

concurrently reducing the availability and ripening of fruits
on trees in the model. As a result, at high frugivore den-
sities, birds were forced to move further afield to forage,
as evidenced by significantly longer mean seed dispersal
distances.
Other simulations that investigated how plant–frugivore
mutualisms affect seed dispersal kernels have produced dif-
ferent outcomes, possibly due to the different ways in which
they represented disperser activity or landscape structure.
For example, despite clear changes in bird movement in
response to landscape configuration, Uriarte et  al. (2011)
stated that there was negligible effect of landscape struc-
ture on seed dispersal distances. However, their model used
fragmented forest patches that did not take into account
aggregation of individual trees. Will and Tackenberg (2008)
found that seed dispersal kernels, based on epizoochorous
and endozoochorous dispersal by mammals, are influenced
more by changes in the animal disperser than on the varia-
tion of vegetation, but this was determined using random
rather than directional disperser movement. Wotton and
Kelly (2012) reported that seed dispersal distances were
established primarily by frugivore movements, but did not
comment on the role of landscape structure in this process.
Our sensitivity analyses revealed that the model param-
eters having the largest influence on the results were those
relating to bird behaviour, while the effect of landscape
characteristics (sigma) was much smaller (Table 2 and
Supplementary material Appendix 4). Mean seed dispersal
distances increased disproportionately when the parameter
(i.e. bD) that was used to calculate the hyperbolic decline in
tree attractiveness with distance varied (shape also increased,
but to a lesser degree, Table 2). The sensitivity of this model
parameter suggests that frugivore movement plays an impor-
tant role in the emergence of the seed dispersal kernel; this
finding is consistent with other studies, which have shown
that spatial patterns of recruitment are influenced by move-
ment patterns of animal dispersers (Wenny 2000, Santamaria
et al. 2007). Although we found little empirical information
with which to estimate bD for our study taxa, our hyperbolic
functional response (which utilised bD, see above) produced

9-EV
Figure 6. The left hand column shows seed dispersal kernels generated by 30 randomly selected H. novaeseelandiae from the total pool (grey
shaded), with the red curved line showing the kernel density estimated across all frugivores, with s, the degree of tree aggregation,
increasing from top to bottom (highly aggregated to diffuse trees). The right hand column shows the spatial association between trees and
seed deposition events using the bivariate pair correlation function (PCF, Illian et al. 2008) for each of 30 replicates (grey lines) for each
value of s. The red straight line is the theoretical expection (g[r] ≡ 1) under a null homogeneous Poisson model, and the distance (x-axis)
at which the grey lines approach this value gives an estimate of the distances over which seed deposition occurred with respect to fruiting
trees.

frugivore foraging ranges, and movements that resulted in
them, that were closest to empirical values, providing sup-
port for our representation of this process.
Frugivore gut passage time of seeds (GPT) and daily fru-
givore perching time (PT) also influenced the scale (but not
the shape) of the dispersal kernels. As expected, an increase
in GPT caused dispersal distances to increase because
seeds passed through the frugivore gut more slowly and
therefore had the opportunity to travel further. When the
shape parameter of the probability distribution describing
GPT increased, there was a reduction in mean seed disper-
sal distances because the shortened tail of the GPT curve
did not allow longer passage times, and so seeds typically
were expelled at shorter distances. An increase in PT caused
a decrease in the mean seed dispersal distances because
the longer frugivores remained at a given perch, the more
likely they were to disperse seeds under the parent tree or
in the immediate vicinity of it. When the shape parameter
of the distribution of PT increased, there was a concurrent
increase in mean seed dispersal distance, consistent with the
shorter perching time resulting from shorter tail of the PT
curve. Morales and Carlo (2006) showed in their scenarios
that mean seed dispersal distance was primarily influenced
by the parameters of the distributions of GPT and PT, but
found that plant attractiveness based on distance was less
important, on the basis of their sensitivity analysis of aD,
the factor in tree attractiveness based on distance (Table 1
and Supplementary material Appendix 1, Eq. A2). However,
Wotton and Kelly (2012) showed in their simulations that
seed dispersal distances were not determined by GPT since
birds never flew for longer than seed retention times, as they
did in our scenarios.
Simulation models that link plant spatial patterns and
the behaviour of organisms are uncommon (Morales and
Carlo 2006, Will and Tackenberg 2008), and ours is one
of few that include frugivore roosting behaviour. The only
previous in silico model of seed dispersal by H. novaeseelan-
diae is a phenomenological model that combined gut pas-
sage time of seeds with radio-telemetry data (Wotton and
Kelly 2012). Incorporating other indigenous large-fruited
co-fruiting tree species could enhance our model by intro-
ducing seasonal variation. The simultaneous representa-
tion of multiple tree species would create dynamics that are
more realistic because there are usually multiple fruit sources
in forests, and the dietary preferences of frugivores may
affect where seeds are deposited (Morales and Carlo 2006,
Morales et al. 2013). For example, Pegman (2012) showed
that seeds of three other large-fruited NZ indigenous tree
species (Beilschmiedia tarairi, B. tawa, and Corynocarpus
laevigatus) were frequently deposited under P. ferruginea and

10-EV
V. lucens. The mean seed dispersal distances produced by our
simulations may therefore need further refinement because
the presence of other local attractive tree species in the land-
scape may act to reduce seed dispersal distances, assuming
all trees are equally attractive. Conversely, if tree species that
have low abundance in the landscape are highly favoured,
this may act to extend dispersal distances. Based on data
from study sites that had multiple fruiting species, Wotton
and Kelly (2012) calculated, via simulation, a shorter mean
seed dispersal distance for V. lucens than ours, using iden-
tical gut passage times of seeds and perching times for
H. novaeseelandiae (their value was ca 98 m versus ours of
ca 151 m, Supplementary material Appendix 3, Table A3,
using nK  2, s  20; nevertheless, up to 20% of V. lucens
seeds in their simulations were dispersed  100 m).
Conclusion
It is challenging to evaluate the link between plant demog-
raphy and behavior of organisms because they interact
reciprocally. Our experiments show the potential of sim-
ulation-based approaches for exploring plant–organism
interactions and identified that plant spatial heterogeneity is
a key determinant of bird behaviour and the seed dispersal
kernels that arise from it. Although this outcome is specific
to the scenarios we examined, we believe our results apply
to other similar mutualisms. Perhaps the most critical result
from our simulations is that the loss of dispersers resulted
in fewer seeds being transported, decreasing the probabil-
ity of rare long distance seed dispersal events, reducing seed
dispersal effectiveness and therefore ultimately diminishing
population spread, genetic connectivity, and survival of tree
species (Nathan 2005). Such understanding is essential in
view of declining animal dispersers and habitats worldwide.
Acknowledgements – This study was funded by a Univ. of Auckland
Doctoral Scholarship (awarded to APMcKP), the School of Envi-
ronment Graduate Research Fund, and the Auckland Regional
Council. We thank the editor and reviewers for useful comments
on the manuscript and the birds for dispersing seeds.­­­­­­­
References
Augspurger, C. K. and Kelly, C. K. 1984. Pathogen mortality of
tropical tree seedlings: experimental studies of the effects
of dispersal distance, seedling density, and light conditions.
– Oecologia 61: 211–217.
Baddeley, A. and Turner, R. 2005. Spatstat: an R package for
analyzing spatial point patterns. – J. Stat. Softw. 12,
< www.jstatsoft.org/ >.
Bell, R. 1996. Seed dispersal by kereru (Hemiphaga novaeseelandiae)
at Wenderholm Regional Park. – MSc thesis, Auckland Univ.,
New Zealand.
Beveridge, A. E. 1964. Dispersal and destruction of seed in central
North Island podocarp forests. – Proc. N. Z. Ecol. Soc. 11:
48–56.
Bullock, J. M. et al. 2006. Measuring plant dispersal: an introduc-
tion to field methods and experimental design. – Plant Ecol.
186: 217–234.
Campbell, K. L. 2006. A study of home ranges, movements, diet
and habitat use of kereru (Hemiphaga novaeseelandiae) in the
southeastern sector of Banks Peninsula, New Zealand. – MSc
thesis, Lincoln Univ., Canterbury, New Zealand.
Clout, M. N. and Hay, J. R. 1989. The importance of birds as
browsers, pollinators and seed dispersers in New Zealand
forests. – N. Z. J. Ecol. 12 (Suppl.): 27–33.
Clout, M. N. and Tilley, J. A. V. 1992. Germination of miro
(Prumnopitys ferruginea) seeds after consumption by New
Zealand pigeons (Hemiphaga novaeseelandiae). – N. Z. J. Bot.
30: 25–28.
Clout, M. N. et  al. 1991. Seasonal movements of New Zealand
pigeons from a lowland forest reserve. – Notornis 38: 37–47.
Coates-Estrada, R. and Estrada, A. 1988. Frugivory and seed
dispersal in Cymbopetalum baillonii (Annonaceae) at Los
Tuxtlas, Mexico. – J. Trop. Ecol. 4: 157–172.
Cordeiro, N. J. and Howe, H. F. 2003. Forest fragmentation severs
mutualism between seed dispersers and an endemic African
tree. – Proc. Natl Acad. Sci. USA 100: 14052–14056.
Cortes, M. C. and Uriarte, M. 2013. Integrating frugivory and
animal movement: a review of the evidence and implications
for scaling seed dispersal. – Biol. Rev. 88: 255–272.
Cousens, R. D. and Rawlinson, A. A. 2001. When will plant
morphology affect the shape of a seed dispersal ‘kernel’? – J.
Theor. Biol. 211: 229–238.
DeAngelis, D. L. and Mooij, W. M. 2005. Individual-based
modeling of ecological and evolutionary processes. – Annu.
Rev. Ecol. Syst. 36: 147–168.
Fedriani, J. M. et al. 2010. Spatial pattern of adult trees and the
mammal-generated seed rain in the Iberian Pear. – Ecography
33: 545–555.
Forget, P. M. 1992. Seed removal and seed fate in Gustavia superba
(Lecythidaceae). – Biotropica 24: 408–414.
Greene, T. 2003. Distance sampling populations of New Zealand
birds – estimating kaka and kereru abundance. – < www.doc.
govt.nz/Documents/science–and.../SciencePoster64 >.
Grimm, V. et  al. 2010. The ODD protocol: a review and first
update. – Ecol. Model. 221: 2760–2768.
Hamby, D. M. 1994. A review of techniques for parameter sensitiv-
ity analysis of environmental models. – Environ. Monitoring
Assess. 32: 135–154.
Hanoteaux, S. et  al. 2013. Effects of spatial patterns on the
pollination success of a less attractive species. – Oikos 122:
867–880.
Hill, M. T. 2003. Diet, dispersal and distribution of kereru
(Hemiphaga novaeseelandiae) in a lowland podocarp-hardwood
forest. – MSc thesis, Massey Univ., Palmerston North,
New Zealand.
Holmes, E. E. and Wilson, H. B. 1998. Running from trouble:
long-distance dispersal and the competitive coexistence of
inferior species. – Am. Nat. 151: 578–586.
Howe, H. F. and Miriti, M. N. 2004. When seed dispersal matters.
– Bioscience 54: 651–660.
Illian, J. et  al. 2008. Statistical analysis and modeling of spatial
point patterns. – John Wiley and Sons.
Johnson, S. 2001. Prumnopitys ferruginea (miro) seedling recruit-
ment patterns in the Catlins, New Zealand, and dispersal by
Hemiphaga novaeseelandiae (New Zealand pigeon). – MSc
thesis in Environmental Science, Univ. of Otago, New Zealand.
Kelly, D. et al. 2010. Mutualisms with the wreckage of an avifauna:
the status of bird pollination and fruit-dispersal in New Zealand.
– N. Z. J. Ecol. 34: 66–85.
Levey, D. J. et al. 1984. Fruit choice in neotropical birds: the effect
of distance between fruits on choice patterns. – Ecology 65:
844 – 850.
Levey, D. J. et al. 2008. Modeling long-distance seed dispersal in
heterogeneous landscapes. – J. Ecol. 96: 599–608.
Levin, S. A. et al. 2003. The ecology and evolution of seed dispersal:
a theoretical perspective. – Annu. Rev. Ecol. Evol. Syst. 34:
575–604.
11-EV
McEwen, W. N. 1978. The food of the New Zealand pigeon
(Hemiphaga novaeseelandiae novaeseelandiae). – N. Z. J. Ecol.
1: 99–108.
McEwen, W. N. 1988. Cone and seed phenology in several
New Zealand conifer tree species. – Tuatara 30: 66–76.
Miskelly, C. M. et al. 2008. Conservation status of New Zealand
birds. – Notornis 55: 117–135.
Morales, J. M. and Carlo, T. A. 2006. The effects of plant distribu-
tion and frugivore density on the scale and shape of dispersal
kernels. – Ecology 87: 1489–1496.
Morales, J. M. et al. 2012. Neighborhood effects on seed dispersal
by frugivores: testing theory with a mistletoe–marsupial system
in Patagonia. – Ecology 93: 741–748.
Morales, J. M. et al. 2013. Frugivore behavioural details matter for
seed dispersal: a multi-species model for Cantabrian thrushes
and trees. – PLoS One 8:e65216.
Nathan, R. 2005. Long-distance dispersal research: building a
network of yellow brick roads. – Divers. Distrib. 11:
125–130.
Nathan, R. and Muller-Landau, C. 2000. Spatial patterns of seed
dispersal, their determinants and consequences for recruit-
ment. – Trends Ecol. Evol. 15: 278–285.
Nathan, R. et  al. 2008. Mechanisms of long distance dispersal.
– Trends Ecol. Evol. 23: 638–647.
Pegman, A. P. McK. 2012. Reconstruction of seed dispersal via
modeling, seedling recruitment, and dispersal efficiency of
Hemiphaga novaeseelandiae in Vitex lucens and Prumnopitys
ferruginea in New Zealand. – PhD thesis, Auckland Univ.,
New Zealand.
Pegman, A. P. McK. et al. 2016. Data from: Exploring the interac-
tion of avian frugivory and plant spatial heterogeneity and its
effect on seed dispersal kernels using a simulation model.
– Dryad Digital Repository, < http://dx.doi.org/10.5061/
dryad.15d8v >.
Pierce, R. J. and Graham, P. J. 1995. Ecology and breeding biol-
ogy of kukupa (Hemiphaga novaeseelandiae) in Northland.
– Science and Research Series 91, Dept of Conservation,
Wellington.
Powlesland, R. G. et  al. 2011. Satellite tracking of kereru
(Hemiphaga novaeseelandiae) in Southland, New Zealand:
impacts, movements and home range. – N. Z. J. Ecol. 35:
229–235.
Robledo-Arnuncio, J. J. et al. 2014. Space, time and complexity in
plant dispersal ecology. – Movement Ecol. 2: 16.
Rodriguez-Perez, J. et  al. 2012. Adult proximity and frugivore’s
activity structure the spatial pattern in an endangered plant.
– Funct. Ecol. 26: 1221–1229.
Salmon, J. T. 1980. Native trees of New Zealand. – Reed Methuen
Publishers.
Supplementary material (Appendix ECOG-02191 at < www.
ecography.org/appendix/ecog-02191 >). Appendix 1–4.­
12-EV
Santamaria, L. et  al. 2007. Predicting spatial patterns of plant
recruitment using animal-displacement kernels. – PLoS One
2: e1008.
Saracco, J. F. et al. 2005. Crop size and fruit neighbourhood effects
on visitation to fruiting trees. – Biotropica 37: 80–86.
Schotborgh, H. M. 2005. An analysis of home ranges, movements,
foods and breeding of kereru (Hemiphaga novaeseelandiae) in a
rural–urban landscape on Banks Peninsula, New Zealand.
– MSc thesis, Lincoln Univ., Canterbury, New Zealand.
Schupp, E. W. 1993. Quantity, quality and the effectiveness of seed
dispersal by animals. – Vegetatio 107: 15–29.
Schupp, E.W. et al. 2010. Seed dispersal effectiveness revisited: a
conceptual review. – New Phytol. 188: 333–353.
Sekercioglu, C. H. et  al. 2004. Ecosystem consequences of bird
declines. – Proc. Natl Acad. Sci. USA 101: 18042–18047.
Spiegel, O. and Nathan, R. 2007. Incorporating dispersal distance
into the disperser effectiveness framework: frugivorous birds
provide complementary dispersal to plants in a patchy environ-
ment. – Ecol. Lett. 10: 718–728.
Tiffney, B. H. et  al. 2004. Vertebrate dispersal of seed plants
through time. – Annu. Rev. Ecol. Evol. Syst. 35: 1–29.
Traveset, A. 1998. Effect of seed passage through vertebrate
frugivores’ guts on germination: a review. – Perspect. Plant
Ecol. Evol. Syst. 1: 151–190.
Uriarte, M. et al. 2011. Disentangling the drivers of reduced long-
distance seed dispersal by birds in an experimentally fragmented
landscape. – Ecology 92: 924–937.
Weibull, W. 1951. A statistical distribution function of wide
applicability. – J. Appl. Mech. 18: 293–297.
Wenny, D. G. 2000. Seed dispersal, seed predation, and seedling
recruitment of a neotropical montane tree. – Ecol. Monogr.
70: 331–351.
Wilensky, U. 1999. NetLogo. – Center for Connected Learning
and Computer-Based Modeling, Northwestern Univ., Evanston,
IL, USA, < http://ccl.northwestern.edu/netlogo/ >.
Will, H. and Tackenberg, O. 2008. A mechanistic simulation model
of seed dispersal by animals. – J. Ecol. 96: 1011–1022.
Wotton, D. M. 2007. Consequences of dispersal failure: kereru and
large seeds in New Zealand. – PhD thesis, Canterbury Univ.,
New Zealand.
Wotton, D. M. and Kelly, D. 2011. Frugivore loss limits recruitment
of large-seeded trees. – Proc. R. Soc. B 278: 3345–3354.
Wotton, D. M. and Kelly, D. 2012. Do larger frugivores move seeds
further? Body size, seed dispersal distance, and a case study of
a large, sedentary pigeon. – J. Biogeogr. 39: 1973–1983.
Wotton, D. M. et al. 2008. Seed retention times in the New Zealand
pigeon, Hemiphaga novaeseelandiae. – N. Z. J. Ecol. 32: 1–6.
Zurell, D. et al. 2010. The virtual ecologist approach: simulating
data and observers. – Oikos 119: 622–635.