Edge-effects on tree regeneration in the Colombian Andes

temperature decreased from 17.5 ºC to 16.7 ºC from 1998 to 1999 (Corporación Regional
del Quindío, unpublished data). The region is classified as subandean forest
(Cuatrecasas 1958). This forest is characterized by mesophyll leaf size and a high
percentage of epiphytes, ferns, many typical tropical taxa such as Annonaceae,
Melastomataceae, palms, and Cyclanthaceae, and a low proportion of woody lianas that
are replaced by hemiepiphytic climbers (Webster 1995). Canopy height averages 30 m
except in areas such as the Cairo site, where the wax palm (Ceroxylon alpinum) occurs,
an emergent canopy tree that reaches 40 m or more.
The study region was covered with forest until the middle 1800’s when
colonization began and forests were logged for the establishment of crops on the rich
volcanic soils. At the beginning of the last century, land use turned to cattle grazing due
to the lower expenses associated with ranching. Examination of aerial photographs
dating back to the 1950’s revealed that most of the present day forest fragments have
maintained similar sizes and shapes since that time (pers. obs.). Moreover, the
surrounding landscape matrix has been dominated by pasturelands for at least the past 50
years. Four isolated forest fragments within an area of 40 km² were chosen for this study.
The fragments were remnants of the original continuous forest, completely surrounded
by well-maintained pastures, and selectively logged prior to 1950. The fragments were
chosen based on their accessibility, similarities in fragment size, shape, and elevation,
and the security they offered to monitor plants and the abiotic environment over a one
year period (Table 1).
In each fragment I selected one study location along the edge (Table 1). Edges
had a straight section of at least 50 m length along the forest edge where no cattle
163

disturbance had occurred in the past 20 to 30 years, as verified by on-site inspection and
discussion with landowners. In most cases, there were fences along the edge boundaries.
Edges were sharply defined by trees with DBH (Diameter at Breast Height [1.3 m])
greater than or equal to 20 cm (abrupt edges) and have not changed since time of
fragmentation (cantilevered edges, Ranney et al. 1981). These canopy trees had
overhanging branches toward the pastures that could act as an umbrella shadowing the
edge understory and buffering it from matrix conditions (Ranney et al. 1981). Sites
differed in their edge structure by the degree of vegetation developed at the forest edge
(Table 1). “Sealed” edges have a dense wall of vegetation that consists of a thick
understory and vines that hang from the branches of canopy trees which likely reduce the
impact of wind. In “open” edges the understory vegetation and vines were nearly absent,
while in “relatively open” edges, understory vegetation was scant with some vines
occurring. Edges varied in orientation, with three of the four sites having a northerly
orientation and one having a southerly orientation.
At each site, a plot of 50 by 50 m was located at the forest edge (0 m) and
extended 50 m inside the forests (Figure 2). The plot was extended only 50 m into the
forest interior because my own observation suggested that conditions of forest interior
remained similar after that and other authors also have generally found habitats free of
edge effects in both tropical and temperate forests (Kapos 1989; Williams-Linera 1990;
Jose et al. 1996; Williams-Linera et al. 1998; Didham and Lawton 1999). The study
fragments were previously found to be similar to the continuous forest in structure and
composition of trees and birds (Renjifo 1998). Renjifo’s results further support the idea
that my study fragments still contain interior forest conditions. No trails, streams or gaps
164

occurred within the study plots. Slope within the plots did not change, but varied among
plots from 0 to 60% (Table 1).

Experimental design and study species

To study the effect of edges, light, and conspecific seedling density on seedling
herbivory in each of the 50 x 50 m plots, data were gathered at ten distance intervals in
transects that ran parallel to the forest edge. Transects were 5 m wide and 50 m long and
extended from 0 to 50 m within the forest interior (Figure 2).
Eight tree species were selected to monitor seedling herbivory (2 species per plot)
(Table 2). The species were canopy or subcanopy trees and a treelet, dispersed by
animals and common in old-growth forest. Ideally, tree species would have been
replicated across fragments, but such replication was not possible due to differences in
local seedling abundance and species composition among fragments. Study species
represented a number of plant families and genera. I selected only those species that had
entire leaves with elliptical-ovoid shape to ensure more accurate measurements of
seedling herbivory over time. Voucher specimens of the species identified at only the
genus level (Aiouea sp. novum) are deposited in the herbarium at Missouri Botanical
Garden (MOBOT) (William Vargas 2680, 7394).
To monitor herbivory damage, I selected eight focal seedlings per species (< 1 m
height) with at least four leaves, in each of the ten transects (N = 80 seedlings/species/
plot) (Table 2; Figure 2). Whenever possible, care was taken to select seedlings along
the entire length of the transect, with a minimum distance between focal seedlings of 1 m.
165

Seedlings were individually labeled and monitored every six weeks over a one year
period (July 1998 – August 1999). Here, I present only the results of accumulated
herbivory after one year, rather than rates, since plants were followed over the same time
period (1 year). In July 1998, I selected the two youngest and undamaged leaves not
completely expanded on each of the 80 seedlings per species. I tagged each leaf with
colored plastic cables, and traced the outline of each leaf on acetate sheets every six
weeks. Tracings of leaf damage were limited to missing leaf area only; damage due to
other sources was not included in this study (i.e., pathogens, leaf miners). Tracings were
scanned, and total leaf area and area lost to herbivory in square centimeters, were
calculated using image analysis software (SigmaScan ™ 1993). When the leaf area was
difficult to determine due to considerable herbivory, I used the area of the same leaf
during the previous survey, unless the remaining part of the leaf indicated that the leaf
had grown substantially between surveys. In such cases, I used as the total leaf area, the
area of the opposite leaf of the same individual (C. brasiliense and P. angustifolia), or the
average area from five leaves of similar sizes within the survey for species with alternate
leaves (other six study species). I calculated the percentage of leaf herbivory (PH) over 1
year using measurements from the final survey as:
PH = [Leaf Area Lost (cm²) * 100] / [Total Leaf Area (cm²)]
I used the average PH from the two leaves per seedling as the response variable. In
seedlings where a leaf disappeared between surveys, only the PH of the remaining leaf
was used. PH in these seedlings may be overestimated, but not as much as if the missing
leaf was assumed with 100% herbivory, as leaves usually fall before they are completely
eaten (pers. obs.; de la Cruz & Dirzo 1987). Most seedlings had both leaves with or
166

without herbivory, rather than just one leaf with herbivory (Appendix 1B). I assumed
that leaf herbivory was the result of insect attack since ground-dwelling vertebrate
herbivores in the region, such as white-tailed deer (Odocoileus virginianus), spectacled
bear (Tremarctos ornatus ), and montane tapir (Tapirus pinchaque) are locally extinct
(pers. obs.). Further, there was no evidence of browsing damage on leaves.
I measured understory light availability to determine the effect of light on
seedling herbivory. Light availability was estimated at 50 cm above the forest floor in
each transect using hemispherical photos (180º Nikkor fish-eye lens). Hemispherical
photographs were taken every 3 m along each transect during uniform overcast sky, for a
total of 10 photos per transect. Because the degree of deciduousness is very low in this
type of forest, and canopy cover does not change noticeably across seasons (pers. obs.), I
measured light only once (wet season). Scanned images from these black-and-white
photos were analyzed with Winphot ™ software (ter Steege 1996) to estimate mean daily
percentage of photosynthetically active photon flux density (%PPFD), calculated as
global site factor (GSF = [(T_dif x P_dif) + (T_dir x P_dir)] x 100). T_dif is the diffuse site factor
(proportion of diffuse PPFD under the canopy relative to that above the canopy), T_dir is
the direct site factor (proportion of direct PPFD under the canopy relative to that above
the canopy), and P_dif and P_dir are the proportions of incident PPFD received above the
canopy as either diffuse sky radiation or direct radiation, respectively (Machado 1999).
Because light measures were not taken directly above each focal seedling, I assigned a
light value per seedling, from the closest hemispherical photograph taken less than 5 m
away from the focal seedling.
To examine the effect of seedling density on seedling herbivory, I calculated
167

conspecific seedling densities around each of the focal seedlings. To determine seedling
densities for each study species, I mapped all the focal seedlings and all conspecific
individuals less than 1.5 m height in the 50 x 50 m plots (Figure 2). Using GIS software
(ESRI 1994), conspecific seedling densities around focal plants were determined for a
range of neighborhood sizes (1 m², 9 m², 25 m², 49 m² and 100 m²) to assess scaling
effects of density-dependence (Figure 2). Neighborhoods had square shapes (1x1, 3x3,
5x5, 7x7, and 10x10 m). Focal plants in which neighborhoods extended outside the 50 x
50 m plot boundaries were eliminated from analyses, since conspecific seedling densities
could not be accurately measured; thus, sample size for neighborhood analyses were
reduced in some cases. A range of neighborhood sizes was used since the spatial scale at
which herbivores may respond to seedling densities is unknown. I did not measure
density dependent herbivory effects beyond 10 m, following Clark & Clark (1985),
Condit et al. (1994), and Cintra (1997).

Data analysis

To determine the extent to which distance from the forest edge, understory light
availability, and density of conspecific seedlings help to explain variation in seedling
herbivory in eight tree species, I used multiple regression analyses (SPSS Inc. 1999).
Regression models for each species were run separately with focal seedlings as the
experimental units. The number of focal seedlings varied among species in the
regression models (Table 3). Several seedlings were eliminated from analyses reducing
168

sample sizes: for example if at least one leaf was not monitored during the entire study
year such as when both leaves were missing or found dead or seedlings were dead. In
addition focal seedlings were eliminated in some analyses when portions of the
neighborhoods extended out of the 50 x 50 m plots (Appendix 1A and B).
Seedling herbivory (PH) was the dependent variable, which only included focal
seedlings that survived to the end of the experiment. Distance from the edge (m), light
(% PPFD), and conspecific seedling density in different neighborhoods were the
independent variables. To satisfy assumptions of normality, seedling herbivory was
arcsin transformed in all species. Graphical diagnostics based on standardized residuals
of preliminary models were used to determine departures from the model assumptions
and to suggest appropriate transformations. I examined the presence of outliers in scatter
plots and their disproportionately large influence on the estimated model based on
Cook’s distance and leverage of each case (SPSS Inc. 1999). I also used scatter plots and
bivariate Pearson’s correlations between all pairs of variables to determine potential
collinearity and interactions among independent variables (Appendix 2). I used a
sequential Bonferroni test (Rice 1989) to assess potential table-wide type I errors at the α
= 0.05. The corrected alpha level was 0.008 for correlations of herbivory with distance,
light, and seedling density in different neighborhoods. Significant interactions between
independent variables were included in preliminary regression models. These significant
interactions were represented in the models by new variables. I followed the method of
Philippi (1983) to create the new variables which equal the product of interacting
variables, but before multiplication, the mean is subtracted to these variables to center
them at cero. These variables should not have their variances standardized to 1. Because
169

interactions did not show a significant contribution in preliminary models, they were not
included in final models.
I used two different approaches of model building to obtain the best fit model for
each species. First, I produced preliminary models that used all independent variables.
These analyses provided collinearity diagnoses that identified highly correlated variables
(see above). Based on these diagnoses, I tested sets of reduced models, eliminating
highly collinear variables, and including new variables that reflected interactions based
on correlations. Selection of best models was based on the highest R² (coefficients of
multiple determination) and lowest probability (P) of the regression coefficients (slopes).
Secondly, I used a stepwise method for model building with F probability for variable
entry 0.05 and for variable removal 0.1. Models obtained from these two approaches
were compared, and the model with the highest R² and lowest P value was selected as the
best fit model for each species.

General herbivory patterns
RESULTS

Mean percentage herbivory per seedling species ranged between 2 and 22%, with
S. laurifolia having the lowest mean herbivory damage and N. purpurea the highest
(Table 3). Coefficients of variation of herbivory ranged between 122-295%, indicating
high intraspecific variation (Table 2). The majority of seedlings did not experience
herbivory damage, or had less than 1% or 5% herbivory damage, while others
experienced from low to high herbivory damage (left skewed distribution) (Figure 3). C.
170

brasiliense had the greatest number of undamaged individuals and P. angustifolia the
least. Most seedlings damaged by leaf-chewing insects experienced no more than 10%
herbivory after one year, except for N. purpurea where 38% of the seedlings had 20% or
more of leaf area lost to herbivory. However, most species did have some individuals
with greater than 50% herbivory.

Light and edge-mediated effect on seedling herbivory

Mean daily percentage of light available for seedlings of each species ranged
between 0.6 and 3% (Table 3). The lowest levels of light in the forest understory were
found for seedlings in Cairo, and the greatest levels for seedlings in Gironda.
Light availability was significantly related to seedling herbivory in S. laurifolia
only. In this species there was a tendency for greater herbivory under high levels of light
(Table 4, Figure 4, Appendix 2). The stepwise model, which retained only light in the
regression explained 21% of the variation in herbivory and the regression coefficient was
low (Table 4). Light availability for S. laurifolia and Aiouea sp. novum, both from El
Cairo site was significantly correlated with distance from the forest edge (Appendix 2).
In the regression models of S. laurifolia or any of the study species, however, light was
not significant in combination with distance from the forest edge indicating that light did
not act as an edge-mediated effect on seedling herbivory (Table 4).

Conspecific seedling density and edge-mediated effect on herbivory

Mean number of conspecific seedlings varied among species and increased with
171

larger neighborhood areas (Table 3). Seedling numbers were greater in C. brasiliense, B.
utile and S. trophoides, which had from 3 to 16 times more conspecific individuals in all
neighborhood sizes, than the other study species (Table 3).
Seedling numbers among neighborhoods of different areas were highly correlated
in all species (Appendix 2). Therefore, the final model for each species (Table 4)
included the seedling numbers in the neighborhood that had the highest relationship with
percent herbivory.
Conspecific seedling numbers were not significantly related to herbivory in six of
the species in any neighborhood area (Table 4, Figure 5). In C. brasiliense and A. coto,
herbivory was significantly related to seedling density, with a trend toward greater
herbivory in seedlings with higher numbers of conspecific seedlings in neighborhoods of
1 m² (Table 4, Figure 5, Appendix 2). The stepwise models, which retained seedling
density in A. coto and seedling density and distance from the edge in C. brasiliense
explained 20% of the variation in herbivory, and the regression coefficients were low
(Table 4). In A. coto, two data points may be driving the observed pattern, so I ran an
alternative model without the two “outliers” and the new model showed the same
qualitative results [n = 63, R² = 0.1, P = 0.02, constant = 0.05, significant regression
coefficient (1m² neighborhood) = 0.02, P = 0.02]. In these two species, most of the focal
seedlings occurring in the lowest density conspecific seedling neighborhoods escape or
experience little herbivory. Seedling density was correlated with distance from the forest
edge in N. purpurea, B. utile, A. coto, and P. angustifolia (Appendix 2). In the regression
models, however, plant density was only significant in combination with distance from
the forest edge in C. brasiliense indicating that plant density acted as an edge-mediated
172