B I O L O G I C A L C O N S E RVAT I O N 1 4 2 ( 2 0 0 9 ) 4 1 4 –4 2 3

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/biocon

The effects of forest fragmentation on euglossine bee

communities (Hymenoptera: Apidae: Euglossini)

Berry J. Brosi*
Department of Biology, Stanford University, 385 Serra Mall, Stanford, CA 94305, USA

A R T I C L E I N F O A B S T R A C T

Article history: Despite scientific and media attention on pollinator declines, there is still only a rudimen-
Received 22 July 2008 tary understanding of the response of bees—the most important group of pollinators
Received in revised form worldwide—to ongoing land use changes. Euglossine bees are an ecologically important
28 October 2008 Neotropical clade of forest-dependent pollinators. Despite the fact that euglossines are
Accepted 4 November 2008 well studied relative to other groups of tropical bees, only three previous studies, all from
Available online 17 December 2008 Brazil, address the response of euglossines to forest fragmentation. In this study, I tripled
the maximum sample size of previous efforts by sampling male euglossines in 22 forest
Keywords: fragments ranging in area from 0.25 ha to 230 ha in southern Costa Rica, using chemically
Forest fragmentation baited Van Someren traps. Abundance of euglossine bees was significantly positively
Costa Rica related to forest fragment size, negatively related to shape (edge:area ratio), and margin-
Orchid bees ally related to fragment isolation. Euglossine species richness showed similar, but weaker
Pollinators trends: richness was significantly positively related to the quantity of forest edge, margin-
Tropical wet forest ally negatively related to fragment area, and not related to fragment isolation. The positive
Van Someren trapping relationship between euglossine richness and abundance and forest fragment edge is
consistent with other studies that have found high euglossine density in secondary or
disturbed forest. The data suggest that individual euglossines move between forest frag-
ments, as has been shown in other systems. Still, forest fragmentation appears to affect
euglossine bees more strongly than other bee groups in the study region. Their large flight
range and positive relationship with forest edges may help to buffer the negative effects of
fragmentation, allowing euglossines to utilize even the very smallest forest fragments in
the study area.
Ó 2008 Elsevier Ltd. All rights reserved.

‘‘The dwindling and extinction of orchid bees, and their ties and the success of the human agricultural enterprise, it is
plants, is one of the most striking traits of the decay of for- imperative that we better understand their responses to
est fragments melting on the Neotropical agroscape ongoing global changes. This is particularly true in light of
today’’—Daniel Janzen (in Roubik and Hanson, 2004, p. 12) the lack of scientific consensus on putative pollinator declines
(e.g. Cane, 2001; Ghazoul, 2005; Steffan-Dewenter et al., 2005).
Though land-use changes such as deforestation could hypo-
1. Introduction thetically have major impacts on bee communities (Buch-
mann and Nabhan, 1996; Allen-Wardell et al., 1998),
Given the importance of bees—the most important taxon of empirical studies of the effects of habitat fragmentation on
pollinators—to both the persistence of native plant communi- bees are scarce (Cane, 2001). Unfortunately, this paucity of

* Tel.: +1 650 450 3715; fax: +1 650 723 5920.

E-mail address: [email protected]
0006-3207/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.biocon.2008.11.003
 B I O L O G I C A L C O N S E RVAT I O N 1 4 2 ( 2 0 0 9 ) 4 1 4 –4 2 3 415

studies is especially pronounced in the tropics, where nearly sampled male euglossine bees in 22 forest fragments ranging
all tropical-rainforest trees are animal-pollinated (Bawa, 1990) over four orders of magnitude in area in a largely deforested
and where pollen limitation can be particularly severe landscape in southern Costa Rica, thus tripling the sample
(Vamosi et al., 2006). size of any previous study. I hypothesized that euglossines
The euglossine, or orchid, bees (Hymenoptera: Apidae: would be more diverse and abundant in forest patches that
Euglossini) are a particularly important group of Neotropical were: (1) larger in area; (2) less isolated from other forest; (3)
pollinators. Euglossines pollinate many tropical plant groups, relatively low in edge:area ratio; and (4) in areas with more
including many economic plants (Dressler, 1968; Dressler, surrounding forest cover.
1982), and are ecologically critical in their role in long-dis-
tance transport of pollen between widely spaced individuals 2. Methods
of low-density tropical plants (Janzen, 1971; Williams and
Dodson, 1972). Most euglossines spend the bulk of their time 2.1. Site selection
in tropical forest habitats (Dressler, 1982), and as such are
thought to be particularly vulnerable to the effects of defores- I sampled male euglossines in the landscape around the Las
tation (Roubik and Hanson, 2004, p. 38). This is especially true Cruces Biological Station (8°47’N, 82°57’W) in the Valle de Coto
relative to other bee groups that extensively nest and forage Brus, Puntarenas province, Costa Rica (Fig. 1). The 22 sampled
in human-altered habitats (Banaszak, 1992; Marlin and La- forest fragments ranged in area from 0.25 ha to 290 ha and
Berge, 2001; Tylianakis et al., 2005; Cane et al., 2006; Brosi in altitude from 900 to 1400 m above sea level (Table 1). All
et al. 2007, 2008). Still, a long-term effort in a region of Pan- study fragments were situated in a matrix of primarily cattle
ama with relatively stable land cover found no declines in pasture (>60% of perimeter boundary). I chose sites separated
euglossine bees over 21 years of regular sampling (Roubik, by at least 500 m (maximum site separation 13.5 km) and with
2001). Euglossines are also well suited for forest fragmenta- a minimum distance of 500 m to the nearest other forest frag-
tion studies, because they are well described taxonomically ment of any size. The study fragments were all isolated by
in relation to many other bee groups (particularly for the tro- deforestation approximately 40 years ago, when a wave of
pics), and because euglossine males are attracted to chemical European settlement in the 1960s reduced forest cover in
baits and thus can be sampled thoroughly and easily (Dress- the region to the approximately 15% that it is today. Locally
ler, 1982; Ackerman, 1983). collected pollen records, however, indicate several thousand
Despite the importance of euglossine bees, only three pub- years of forest clearing, burning, and agriculture by indige-
lished studies have directly examined how they are affected nous people (Clement and Horn, 2001). The predominant land
by forest fragmentation (Powell and Powell, 1987; Becker uses in the region are cattle pastures, coffee fields, and small
et al., 1991; Tonhasca et al., 2002a,b). These studies were well agricultural plots for subsistence crops like corn, beans, and
designed and executed, but wider inference from them is lim- bananas. See Brosi et al. (2008) for more details on the study
ited in several ways. First, all three took place in Brazil, area and sampled fragments.
including two (Powell and Powell, 1987; Becker et al., 1991) Euglossine bees are strongly forest-associated in this study
at the same site, the biological dynamics of forest fragments region. Extensive and taxonomically broader bee community
(BDFF) project near Manaus (Bierregaard, 2001). The work of surveys in the same landscape (Brosi et al., 2007, 2008), sam-
Powell and Powell (1987) took place immediately after experi- pled no euglossines outside of forests in work that tallied
mental deforestation at the BDFF site, and the acute distur- >4000 bee individuals. In thousands of hours of field time in
bance of forest clearing was likely the predominant factor the study region, I have observed (non-baited) euglossine bees
shaping their results (Becker et al., 1991; Cane, 2001). Second, foraging outside of tree-covered areas fewer than five times.
all three studies were based on a small number of sample At least some species of euglossines, however, can be at-
sites (Powell and Powell, four; Becker et al., seven; and Tonha- tracted to chemical baits set out in deforested areas (B. Brosi,
sca et al., nine, three in forest fragments and six in continu- pers. obs.; Milet-Pinheiro and Schlindwein, 2005).
ous forest). Third, while all three studies examined the While an ideal study design would sample each fragment
effects of forest fragment size, none of them assessed other proportionally to its area, limited time and resources dictated
potentially important spatial factors, such as fragment shape, that I maximize the true sample size of the study (number of
isolation, or landscape context. Fourth, results from the stud- forest fragments sampled) by sampling each fragment in one
ies, taken as a whole, are inconclusive. Becker et al. and place. The sole exception was the large (230 ha) forest frag-
Tonhasca et al. found no significant relationships between ment at the Las Cruces Biological Station, which I sampled in
forest fragment area and euglossine diversity or abundance, two sites 1 km apart.
while Powell and Powell (whose results were likely dominated
by disturbance) did find a significant relationship. Finally, all 2.2. Trapping
took place in landscapes dominated by large, contiguous
patches of forest. Since some euglossine bees possess flight I sampled bees with traps, which allowed me to sample in
ranges of tens of kilometers (Janzen, 1971), these studies multiple sites simultaneously. I did not sample by hand-net-
may tell us little about the impacts of fragmentation on ting at chemical baits, since in test trials a large proportion
euglossine bees in much more extensively deforested of bees (commonly 20% or more) eluded either capture or
landscapes. in-flight identification. Instead, I used Van Someren traps,
In order to address these issues to improve understanding which are commonly used to sample butterflies (e.g. Horn-
of the responses of euglossine bees to forest fragmentation, I er-Devine et al., 2003) and which consist of a cylinder of mesh
416 B I O L O G I C A L C O N S E RVAT I O N 1 4 2 ( 2 0 0 9 ) 4 1 4 –4 2 3

nicaragua

atlantic ocean

san josé

costa rica
pacific ocean

las cruces

panamá

Fig. 1 – Map of study site within Costa Rica.

Table 1 – Study sites. Sites 20 and 21 both represent the forest fragment at the Las Cruces Biological Station; all other
samples occur in unique fragments.

Site Area (Ha) Perimeter (m) Isolation (m) Elevation (m asl)

1 0.24 242.99 1203.4 1180

2 0.47 419.98 829.8 1089
3 0.53 354.54 778.0 963
4 0.65 409.06 832.9 1195
5 1.05 538.86 1838.3 1100
6 1.06 484.91 874.2 1027
7 1.13 543.01 1076.3 1115
8 1.18 509.40 1461.9 1160
9 1.46 711.29 507.0 1387
10 2.14 617.75 770.6 1284
11 3.27 1000.86 571.6 1184
12 3.39 879.19 850.2 898
13 4.67 1273.65 505.0 1000
14 6.83 1115.19 587.0 899
15 8.36 1706.42 569.7 1103
16 24.76 3730.86 512.0 1479
17 24.96 2791.96 519.0 1181
18 28.79 3532.11 769.4 916
19 55.19 5800.42 571.9 942
20 236.62 9737.84 599.5 1201
21 236.62 9737.84 548.4 1273
22 296.49 17,134.14 812.2 1414

cloth, raised 4 cm over a platform on which bait is placed. traps for sampling bees; Smith-Pardo (1999) used such traps,
Insects are attracted to the bait and fly upward, becoming baited with rotten fish, to sample bee communities (but not
trapped in the mesh cylinder, from when they can be col- specifically euglossines) in Colombian dry forest. I did not
lected alive. This is the second recorded use of Van Someren use McPhail traps (Bennett, 1972) because of their low catch
 B I O L O G I C A L C O N S E RVAT I O N 1 4 2 ( 2 0 0 9 ) 4 1 4 –4 2 3 417

rates and the wetted, damaged bee specimens they yield, any size, and calculated isolation distances using the Nearest
which can be difficult to identify (Becker et al., 1991; Roubik, Features 3.8a (Jenness, 2004) extension for ArcView 3.2 (ESRI,
2001). Redlands, CA, USA). I defined landscape context as the pro-
For each sampling event, I set one Van Someren trap portion of forest from landsat data (classification details in
15 m inside the forest fragment between 7:00 and 10:30am Daily et al., 2001) surrounding sample points at radii from
and collected it approximately 3 h later, before daily rains 200 to 2000 m, which I quantified using ArcMap 9.1.
commenced. I rotated the temporal order of site sampling
within days, as well as the choice of essential oil bait, to min- 2.4. Data analysis
imize biases. I divided the sites into four ‘‘routes’’ of five or six
nearby sites and sampled each route on the same day, thus I compared forest characteristics (forest fragment area,
running traps simultaneously in the sites nearest one shape, isolation, and context) with bee relative abundance
another. (raw number of trapped bees) and diversity (measured by
I baited the Van Someren traps with either methyl salicy- both species richness and the Shannon–Weiner diversity in-
late (wintergreen oil, four drops) or cineole (eucalypus oil, dex, hereafter SWDI). I used mixed-effects modeling to ana-
six drops) placed on paper towels in a small tupperware-style lyze data because of the hierarchical nature of the sampling
plastic container on the trap base. I choose these two baits be- (repeated trapping sub-samples within sites). Such models
cause in combination they attract nearly all recorded species allow for the use of all data while correcting for pseudore-
in the nearby (though lower-elevation) Osa Peninsula (Janzen, plication below the site level (e.g. Crawley, 2002, 669). Forest
1981). I purposefully utilized smaller quantities of essential oil fragment area, shape index, and isolation distance were all
baits than have been used in some previous studies. In typical significantly correlated with one another, which would have
euglossine sampling, a drink coaster or paper towel is com- led to poorly-conditioned multivariate models. I therefore
pletely saturated with essential oil, and bees are netted by tested these landscape factors in separate models and Bon-
hand. In many contexts, this technique can lure hundreds ferroni-corrected the resulting P-values. The type of essen-
of bees in a very short time frame (e.g. Janzen, 1981). Because tial oil bait (cineole vs. methyl salicylate) was strongly
destructive sampling was necessary to ensure accurate iden- related to both euglossine diversity and abundance in indi-
tifications, a larger bait quantity combined with the repeated vidual trapping trials, so I included bait type as an explan-
sampling protocol I used could have had significant deleteri- atory variable along with the individual landscape elements
ous impacts on local euglossine populations (Roubik, 1993). in each of the analyses. Since the data are count-based and
Additionally, though the attractive radius of essential oil baits sample variance increased with sample mean, I used gener-
has not been studied, a smaller bait quantity should hypo- alized linear models with Poisson errors in the mixed-effect
thetically have a smaller radius of attraction, thus allowing models (e.g. Olofsson and Shams, 2007), with site as a ran-
for a more precise spatial focus on the targeted forest frag- dom effect.
ments (rather than luring individuals from more distant, To assess the effect of landscape context on the euglossine
non-target areas). community, I calculated the correlation between the propor-
I sampled bees from June to September 2004, during the tion of forest surrounding sample points and euglossine spe-
wet season, the time of year in which Becker et al. (1991) cies richness and abundance, at scales from 200 to 2000 m
found the greatest richness and abundance of orchid bees, surrounding sample points (Ricketts et al., 2001; Steffan-Dew-
and repeated trapping six times in each site. I collected all enter et al., 2002). To assess differences in community compo-
bees within the traps and retained all specimens to maximize sition, I used Mantel and Partial Mantel tests to compare pair-
the quality of identifications and to minimize bias from bees wise, standardized differences in landscape factors with Mor-
returning to traps. I identified bee specimens using Roubik isita–Horn community similarity matrices. I used Moran’s I to
and Hanson (2004); I. Hinojosa, University of Kansas, reviewed test for spatial autocorrelation in bee diversity and abun-
and corrected species determinations and D. Roubik, Smith- dance between sites.
sonian Tropical Research Institute, reviewed the final species To avoid pseudoreplication, I averaged data within sites for
list. Some specimens are deposited at the University of Kan- all analyses except the mixed-effects models, in which I spec-
sas Entomological Museum (SEMC); the remainder are held ified the hierarchical nature of the data without averaging. I
at the Biology Department, Stanford University. analyzed data using the R statistical programming language
(R Development Core Team, 2008), using the lme4 package
2.3. Measurement of forest characteristics (Bates, 2008) for mixed-effects generalized linear models
and the vegan package (Oksanen et al., 2006) for community
I measured area and perimeter by outlining forest fragments ecological analyses.
from georeferenced, orthorectified aerial photographs (cour-
tesy G. Durán, Organization for Tropical Studies) in ArcMap 3. Results
9.1 (ESRI, Redlands, CA, USA). For fragments <1 ha, I com-
pared these measurements to those made on the ground with 3.1. Overview
compass, laser range finder, and clinometer; the two methods
generated measurements within 5% of one another. As a I sampled 412 euglossine bees in the genera Euglossa (18 spe-
measure of forest fragment shape, I used the ratio of perime- cies) and Eulaema (5 species), Table 2. Two species, Euglossa
ter2:area, which is constant for circles of varying area. I de- championi and Eulaema bombiformis, accounted for almost half
fined isolation as distance to the nearest forest fragment of of the sampled bee individuals; Fig. 2 shows the rank abun-
418 B I O L O G I C A L C O N S E RVAT I O N 1 4 2 ( 2 0 0 9 ) 4 1 4 –4 2 3

Table 2 – Species list. Species in alphabetical order, sorted by genus. Proportion is proportion of total captures represented
by that species. See Roubik and Hanson (2004) for nomenclatural authorities.

Genus species Captures

Cineole Meth. Salic. Proportion (%) # Sites

Euglossa allosticta 1 0.2 1

Euglossa asarophora 1 0.2 1
Euglossa bursigera 1 0.2 1
Euglossa championi 12 107 26.1 18
Euglossa cognata 1 0.2 1
Euglossa crassipunctata 1 0.2 1
Euglossa cybelia 19 4.2 11
Euglossa despecta 1 0.2 1
Euglossa erythrochlora 83 18.2 16
Euglossa flammea 19 4.2 12
Euglossa gorgonensis 8 1.8 7
Euglossa heterosticta 2 2 0.9 3
Euglossa imperialis 1 2 0.7 3
Euglossa maculilabris 39 1 8.8 8
Euglossa mixta 25 5.5 14
Euglossa oleolucens 6 1.3 4
Euglossa sapphirina 3 0.7 3
Euglossa villosiventris 1 0.2 1
Eulaema bombiformis 106 23.2 19
Eulaema cingulata 1 0.2 1
Eulaema meriana 9 2.0 8
Eulaema nigrita 3 0.7 3
Eulaema speciosa 1 0.2 1
Additional species caught with hand netting at baits (not analyzed statistically)
Euglossa cyanura
Euglossa deceptrix
Euglossa dodsoni
Euglossa igniventris
Euglossa tridentata

140 cies in the system. The start time and duration of individual
trapping trials were not related statistically with euglossine
120
bee diversity or abundance.
100

80 3.2. Spatial and altitudinal autocorrelation

60

40 Bee abundance was significantly spatially autocorrelated

(Moran’s I = 0.10, P = 0.030), but bee diversity did not show
20
any pattern of spatial autocorrelation. Community similarity,
0 as measured by the Morisita–Horn index, was also signifi-
cantly spatially autocorrelated (Mantel r = 0.21; P = 0.013). I
found no relationships between euglossine bee diversity or
abundance and the elevation of sample sites. Community
similarity was, however, strongly related to elevation (Mantel
standardized r = 0.38; P < 0.001).
Fig. 2 – Rank abundance curve. Because pair-wise site proximities and elevations were
also strongly correlated to one another (Mantel r = 0.43,
P < 0.001), I also assessed the partial correlations of elevation
dance curve. Sub-samples were marked by high variability and geographic distance on euglossine bee community simi-
within sites. For example, in one site a trap yielded four indi- larity. Holding geographic distance constant, community sim-
vidual bees of three species; 5 days later at the same site, the ilarity and elevation were still significantly correlated (Mantel
trap yielded 38 bees of five species. Overall, there was a 5-fold r = 0.33, P = 0.001). In contrast, holding elevation constant, bee
difference in species richness and a 15-fold difference in community similarity and geographic distance were no long-
abundance in the highest and lowest samples within a site. er significantly correlated (Mantel r = 0.052, P = 0.28). There-
Species accumulation curves showed a flattening trend fore, the spatial autocorrelation in bee community similarity
(Fig. 3), but also indicate that I likely did not detect every spe- appears to be governed by elevation, not geographic distance
 B I O L O G I C A L C O N S E RVAT I O N 1 4 2 ( 2 0 0 9 ) 4 1 4 –4 2 3 419

Number of species 25 Euglossine Species Accumulation 4. Discussion

20 Euglossine bee communities are affected by forest fragmenta-
tion in southern Costa Rica. Euglossine abundance was signif-
15
icantly positively related to fragment area and forest edge,
10 and marginally related to fragment isolation. Euglossine spe-
cies richness showed similar but slightly weaker trends, with
5 a significant positive association with forest fragment edge, a
marginal positive relationship to fragment area, and no rela-
0
0 20 40 60 80 100 120 140 tionship with fragment isolation. Euglossine community sim-
Number of samples ilarity was related only to site elevation, when the correlation
between site proximity and elevation was considered.
Fig. 3 – Species accumulation curve, based on randomizing
This work was intended to examine the differences in
sample days 500 times without replacement.
euglossine communities between different forest fragments
in a relatively small region, and not to estimate the true spe-
per se. In subsequent analyses, I assumed that sites repre- cies richness or diversity for the euglossine community as a
sented independent samples. whole in the study region. Especially because of the novel
sampling method, the purposefully small bait quantities
3.3. Landscape variables used, and the incomplete species-accumulation curve from
the sampling (Fig. 3), the results of this work should not be
Table 3 and Fig. 4 show the results for relationships between used to compare the species density of this region with other
forest characteristics and euglossine abundance and species areas. In addition, this work should be primarily considered in
richness. Forest fragment area was significantly positively re- the context of the common species that were sampled, as
lated to euglossine abundance (Fig. 4a), and marginally re- many of the rare species were trapped in such low numbers
lated to species richness (Fig. 4b). Shape (edge:area ratio) of as to make interpretation difficult (Table 2).
fragments was strongly related to both abundance (Fig. 4c) The appropriate spatial pattern of sampling of euglossine
and species richness (Fig. 4d) of euglossines, though in some- bees is the subject of some controversy, and could have
what unexpected ways. Overall euglossine abundance in- potentially impacted my results. Armbruster (1993) hypothe-
creased with increasing edge of forest fragments, as did sized that euglossine males have relatively fixed discrete ter-
species richness. The direction of the response of species ritories over which they forage, and thus that sampling in
richness, however, varied when samples were separated by proximal sites within the same forest could yield substan-
bait type. For cineole, richness maintained the same pattern tially different euglossines abundance and diversity. He sub-
of increasing with fragment edge, but for methyl salycilate, stantiated this hypothesis with data from paired sample
richness followed the previously expected pattern of decreas- sites 200 m apart in several Neotropical regions, which
ing species richness with increasing edge. Fragment isolation showed statistically significant differences in the proportions
was marginally related to euglossine abundance (Fig. 4e), and of euglossine taxa arriving at baits. This is of particular inter-
was unrelated to species richness (Fig. 4f). Euglossine abun- est here because in the interest of maximizing the true sam-
dance was significantly positively related to proportion of for- ple size, I sampled each forest fragment in one place, except
est surrounding sample sites at small scales, in radii around for the 230 ha patch at the Las Cruces Biological Station,
sample points from 200 to 400 m (Fig. 5). At this scale, how- which I sampled in two sites.
ever, landscape context is essentially related only to the area The balance of published evidence, however, goes against
of the target fragment, since each study forest patch was sep- Armbruster’s ‘‘within-habitat heterogeneity’’ hypothesis. Ton-
arated from other forested areas by at least 500 m. hasca et al. (2002b) disagreed with Armbruster’s choice of sta-

Table 3 – Landscape factors and euglossine bee species richness and relative abundance. Because of strong correlations
between landscape factors, each was tested separately in a model including bait type. Thus, the ‘‘bait type’’ columns
report on the significance of bait type in the model for each landscape factor (rows), and the ‘‘interaction’’ column reports
on the bait type*landscape factor interaction. For all models, df = 20 (22 sample sites), and the Bonferroni-corrected
a = 0.017.

Factor Factor Bait type Interaction

Z P Z P Z P

Species richness Log area 1.73 0.083 5.86 <0.001*** 1.35 0.177
Shape 2.32 0.021 5.39 <0.001*** 2.48 0.013*
Isolation 1.05 0.295 1.57 0.117 0.99 0.322

Relative abundance Log area 3.13 0.002** 8.79 <0.001*** 1.75 0.080.
Shape 4.49 <0.001*** 9.52 <0.001*** 4.93 <0.001***
Isolation 1.74 0.082. 3.23 0.001** 0.81 0.417
420 B I O L O G I C A L C O N S E RVAT I O N 1 4 2 ( 2 0 0 9 ) 4 1 4 –4 2 3

abundance species richness

a Cineole Methyl Salicylate b Cineole Methyl Salicylate

species richness per subsample

abundance per subsample

6 6
30 30
fragment area

4 4
20 20

10 10 2 2

0 0 0 0

1 10 100 1 10 100 1 10 100 1 10 100

area (Hectares; log scale) area (Hectares; log scale)

c Cineole Methyl Salicylate d Cineole Methyl Salicylate

species richness per subsample

aabundance per subsample

6 6
30 30
fragment shape

4 4
20 20

10 2 2
10

0 0 0 0

20 40 60 80 100 20 40 60 80 100 20 40 60 80 100 20 40 60 80 100

shape (edge2:area ratio) 2
shape (edge :area ratio)

e Cineole Methyl Salicylate f Cineole Methyl Salicylate

species richness per subsample

6 6
abundance per subsample

30 30
fragment isolation

4 4
20 20

10 10 2 2

0 0 0 0

500 1000 1500 500 1000 1500 500 1000 1500 500 1000 1500

isolation (m) isolation (m)

Fig. 4 – Euglossine bee responses to landscape factors and bait type. While all subsamples are shown, these were grouped by
site for statistical analyses. Left column is abundance; right column is species richness. (a, b) Log10 area of forest (in hectares).
(c, d) Fragment shape (edge2: area ratio). (e, f) Fragment isolation (meters to nearest forest).
 B I O L O G I C A L C O N S E RVAT I O N 1 4 2 ( 2 0 0 9 ) 4 1 4 –4 2 3 421

0.6 by water) has shown that where habitat isolation distances

and euglossine richness or abundance)
are greater than typical flight ranges, isolation can have
abundance
0.5 strong effects on euglossine communities, particularly in
(r-value; between % forest

p = 0.05 significance line terms of species richness (e.g. Janzen, 1981). Fifth, it is unli-
0.4 kely that up to several hundred euglossines would arrive at
correlation

a single baiting station within a few hours (Janzen, 1981) if

0.3 richness they were more or less restricted to a small within-habitat
territory.
0.2 While it is thus likely that euglossines are moving be-
tween forest patches—a result that should hypothetically
0.1
weaken the relationships between forest characteristics
and the euglossine community—I still found strong relation-
ships between euglossines and both patch area and shape.
0
200 400 600 800 1000 1200 1400 1600 1800 2000 In terms of fragment area, given that nesting and foraging
spatial scale resources are patchily distributed in tropical forests, more
(circle radius in meters) bees should be found in larger fragments due to resource
availability alone. My results are consistent with this idea,
Fig. 5 – Correlation between landscape context (proportional
given that I found a significant positive relationship be-
area of forest surrounding sample sites) and euglossine
tween forest fragment area and euglossine abundance, and
species richness and abundance, at scales from 200 to
a marginal relationship with euglossine species richness.
2000 m. Horizontal gray line shows cutoff for significant
The previous studies of euglossines and forest fragmenta-
correlations at P = 0.05 (r = 0.43); thus, only the correlation
tion were focused on fragment area (as opposed to other
between abundance and landscape context at radii from 200
landscape factors), and in general have suggested—while
to 400 m is significant.
not providing definitive evidence—that patch area impacts
euglossine bee communities. Neither Becker et al. (1991)
tistical test, re-analyzed his data, and found no statistically nor Tonhasca et al. (2002a,b) found significant differences
significant differences between the euglossine communities in euglossine bee abundance or diversity with forest frag-
in Armbruster’s within-habitat site pairs. Otero and Sandino ment area, though the smallest fragments in both works
(2003) sampled four close-by sites in each of three distinct had the lowest levels of euglossine abundance and diversity.
habitats repeatedly over nearly two years in Colombia. While In both studies, the small number of fragments may have
they did not directly address the question of within-habitat limited their power to find statistical differences. In con-
heterogeneity in their paper, Otero and Sandino’s data show trast, Powell and Powell (1987) did find significant relation-
much larger differences between habitats relative to between ships between fragment area and euglossine richness and
sample points within the same habitat. The within-habitat abundance, but they sampled immediately after a major dis-
samples appear from the tables and charts within their work turbance (experimental deforestation), which is thought to
to have stabilized at relatively similar communities. Finally, be the primary factor affecting their results (Becker et al.,
they marked and released each of the >2000 individuals they 1991; Cane, 2001).
caught, and yet recorded no recaptures. Of the landscape factors I tested, forest fragment shape
My results and the literature are consistent with the idea had the strongest associations with both euglossine species
that individual euglossines routinely travel between forest richness and abundance. Euglossines are significantly more
fragments in the area, rather than remaining resident in abundant and speciose in forest fragments with more edge
individual forest fragments (or territories within fragments), relative to their area. This relationship also had a significant
for several reasons. First, Tonhasca et al. (2003) recorded interaction term with bait type for euglossine species rich-
marked euglossines (including some of the same species I ness. For bees attracted by cineole, the relationship was in
sampled) traveling between forest fragments in Brazil, and the same direction as the overall pattern (richness increased
it seems likely that euglossines would have similar behavior with increasing edge), but for bees attracted by methyl salic-
in Central America. Second, I recorded euglossines even in ylate, the direction of the response changed, and species
the very smallest fragments in the study region, which are richness decreased with increasing edge. This difference in
less than half of a hectare in area. Fragments of this size the direction of the response of species richness to edge
are unlikely to provide all of the life-history requirements was surprising, because abundance and richness are typi-
needed for stable populations of several species of large, cally positively related. This pattern could be explained in
fast-flying bees. Third, euglossines are known to have large part by the fact that the three dominant euglossine species
flight ranges, being capable of traveling up to 30 km in a (which together comprised 2/3 of all sampled bees) were
day (Janzen, 1971) and the fragments in this study area have all attracted primarily to methyl salicylate, and also were
short isolation distances (less than 2 km in all cases). all positively related to increasing forest edge. If in general
Fourth, I did not find strong patterns of euglossine species forest edges provide greater resource density, the dominant
richness or abundance with isolation distance, which would species could potentially be competitively excluding less-
have been expected if euglossines were largely limited to common species from the ‘‘edgier’’ fragments. Alternatively,
single fragments and only occasionally dispersed between less-common euglossine species attracted by methyl salicy-
fragments. For example, research from true islands (isolated late could prefer forest-interior habitats, which are in scar-
422 B I O L O G I C A L C O N S E RVAT I O N 1 4 2 ( 2 0 0 9 ) 4 1 4 –4 2 3

cer supply in ‘‘edgier’’ forest fragments. Why such a pattern Acknowledgements

would not be shown for the other attractant (cineole), how-
ever, is puzzling. I thank the many Costa Rican families who allowed me to
Evidence does exist that euglossines will readily utilize re- work on their land. G. Daily contributed crucial support, ad-
sources from less-than-pristine forest. Otero and Sandino vice, and funding at all stages of this work. G. Daily, P. Ehrlich,
(2003) sampled three habitat types—farm, secondary forest, K. Levy, and R. Pringle gave helpful suggestions on earlier
and old logged forest—in coastal Colombia and found euglos- drafts of the manuscript. I. Hinojosa, University of Kansas,
sine abundance was positively related to land-use intensity. introduced me to orchid bee biology and identification,
The farm sites had the highest euglossine abundance, fol- helped inspire this study, and confirmed my species determi-
lowed by secondary forest, while the low-disturbance old nations. D. Roubik reviewed and corrected the final species
logged forest had the lowest euglossine abundance. Similarly, list. J. DeNoyer, K. Frangioso, B. Graham, J. Ilama, F. Oviedo,
Rincon et al. (1999) found that logged forest had higher and T. Shih provided superb field assistance. The staff of the
euglossine abundance than did less-disturbed forest in Costa Las Cruces Biological Station and the Organization for Tropi-
Rica. Both of these results are consistent with the idea that cal Studies (especially R. Quirós, E. Ramirez, and Z. Zahawi)
forest edges may be sites of relatively high resource-density provided cheerful field research support. G. Durán of Las Cru-
for euglossine bees. ces provided key spatial data. J. Ranganathan provided GIS ad-
While euglossines are positively associated with forest vice, data, and assistance. I am grateful for funding from the
edge, they still appear to be more affected by forest frag- Anne M. and Robert T. Bass Stanford Graduate Fellowship in
mentation than other clades of bees in the region. Sampling Science and Engineering, a Teresa Heinz Scholarship for Envi-
across all groups of bees in the same region, Brosi et al. ronmental Research, and the Moore Family Foundation; the
found that bee abundance and diversity were not related Stanford University Field Studies and Human Biology Re-
to distance from forest (2007) or forest fragment size search Experiences for Undergraduates (HB-REX) Programs;
(2008). Species composition, however, did change with both and Grants to the Center for Conservation Biology at Stanford
distance to forest and forest fragment size. Net sampling University from the Koret, McDonnell, Sherwood, and Win-
results from both within forest fragments and in paired slow Foundations and Peter and Helen Bing.
pasture habitats directly adjacent to them showed that
15% of sampled bees within forest were euglossines, com-
pared to 0% outside of forest (Brosi et al., 2008). Given this R E F E R E N C E S
and the positive significant relationship between forest frag-
ment size and euglossine abundance, and the marginal
relationship with species richness, euglossines are appar- Ackerman, J.D., 1983. Diversity and seasonality of male euglossine
ently more affected by forest fragmentation than other bees (Hymenoptera: Apidae) in central Panama. Ecology 64,
groups of bees. 274–283.
A feature of both the work presented here and of previous Allen-Wardell, G., Bernhardt, P., Bitner, R., Burquez, A.,
studies (e.g. Roubik, 2001) is high sample variability. My sam- Buchmann, S., Cane, J., Cox, P.A., Dalton, V., Feinsinger, P.,
Ingram, M., Inouye, D., Jones, C.E., Kennedy, K., Kevan, P.,
ples, taken typically within four to 6 days of one another in
Koopowitz, H., Medellin, R., Medellin-Morales, S., Nabhan, G.P.,
the same site and repeated six times, show a similar magni-
Pavlik, B., Tepedino, V., Torchio, P., Walker, S., 1998. The
tude of variation as studies conducted over several years (Rou- potential consequences of pollinator declines on the
bik, 2001) and indicate that this variability is not necessarily conservation of biodiversity and stability of food crop yields.
due to fluctuations in population size. Euglossines may move Conservation Biology 12, 8–17.
large distances in the landscape over short time scales in re- Armbruster, W.S., 1993. Within-habitat heterogeneity in baiting
sponse to floral or other resource availability, and/or may be samples of male euglossine bees: possible causes and
implications. Biotropica 25, 122–128.
affected by relatively subtle changes in weather that are re-
Banaszak, J., 1992. Strategy for conservation of wild bees in an
flected in their presence at traps. Thus, while long-term mon- agricultural landscape. Agriculture Ecosystems &
itoring is indisputably valuable, perhaps the most important Environment 40, 179–192.
factor for characterizing a euglossine bee community is re- Bates, D. 2008. LME4: linear mixed-effects models using S4
peated sampling of the same sites, even if those samples are classes. <http://lme4.r-forge.r-project.org/Version 0.99875-9>
repeated over a relatively short space of time. (accessed April 2008).
Understanding euglossine responses to habitat fragmen- Bawa, K., 1990. Plant-pollinator interactions in tropical rain-
forests. Annual Review of Ecology and Systematics 21, 399–
tation has been hampered by a small number of studies,
422.
and future work should consider and potentially even focus Becker, P., Moure, J.S., Peralta, F.J.A., 1991. More about euglossine
on the larger context of study sites. For example, research bees in Amazonian forest fragments. Biotropica 23, 586–591.
in landscapes that are more extensively deforested may re- Bennett, F.D., 1972. Baited McPhail fruit fly traps to collect
veal stronger negative responses of euglossine species rich- euglossine bees. Journal of the New York Entomological
ness than previous work. Euglossine bees are a critically Society 80, 137–145.
Bierregaard, R.O., 2001. Lessons from Amazonia: The Ecology and
important group of tropical forest-dependent pollinators,
Conservation of a Fragmented Forest. Yale University Press,
and given the increasing concern over pollinator declines
New Haven, CT.
worldwide, we must better understand how ongoing land- Brosi, B.J., Daily, G.C., Ehrlich, P.R., 2007. Bee community shifts
use changes affect their abundance, diversity, and commu- with landscape context in a tropical countryside. Ecological
nity composition. Applications 17, 418–430.
 B I O L O G I C A L C O N S E RVAT I O N 1 4 2 ( 2 0 0 9 ) 4 1 4 –4 2 3 423

Brosi, B.J., Daily, G.C., Shih, T.M., Oviedo, F., Durán, G., 2008. The Powell, A.H., Powell, G.V.N., 1987. Population-dynamics of male
effects of forest fragmentation on bee communities in tropical euglossine bees in Amazonian forest fragments. Biotropica 19,
countryside. Journal of Applied Ecology 45, 773–783. 176–179.
Buchmann, S.L., Nabhan, G.P., 1996. The forgotten pollinators. R Development Core Team, 2008. R: a language and environment
Island Press for Shearwater Books, Washington, DC. for statistical computing. R Foundation for Statistical
Cane, J.H., 2001. Habitat fragmentation and native bees: a Computing, Vienna, Austria (http://www.R-project.org).
premature verdict? Conservation Ecology 5 (article 3). Ricketts, T., Daily, G., Ehrlich, P., Fay, J., 2001. Countryside
Cane, J.H., Minckley, R.L., Kervin, L.J., Roulston, T.H., Williams, biogeography of moths in a fragmented landscape:
N.M., 2006. Complex responses within a desert bee guild biodiversity in native and agricultural habitats. Conservation
(Hymenoptera: Apiformes) to urban habitat fragmentation. Biology 15, 378–388.
Ecological Applications 16, 632–644. Rincon, R.M., Roubik, D.W., Finegan, B., Delgado, D., Zamora, N.,
Clement, R., Horn, S., 2001. Pre-Columbian land-use history in 1999. Regeneration in tropical rain forest managed for timber
Costa Rica: a 3000-year record of forest clearance, agriculture production: understory bees and their floral resources in a
and fires from Laguna Zoncho. Holocene 11, 419–426. logged and silviculturally treated Costa Rican forest. Journal of
Crawley, M.J., 2002. Statistical Computing: An Introduction to the Kansas Entomological Society 72, 379–393.
Data Analysis using S-plus. John Wiley and Sons, New York. Roubik, D.W., 1993. Tropical pollinators in the canopy and
Daily, G., Ehrlich, P., Sanchez-Azofeifa, G., 2001. Countryside understory – field data and theory for stratum preferences.
biogeography: use of human-dominated habitats by the avifauna Journal of Insect Behavior 6, 659–673.
of southern Costa Rica. Ecological Applications 11, 1–13. Roubik, D.W., 2001. Ups and downs in pollinator populations:
Dressler, R.L., 1968. Pollination by euglossine bees. Evolution 22, when is there a decline? Conservation Ecology 5 (art. no.-2).
202–210. Roubik, D.W., Hanson, P.E., 2004. Orchid bees of tropical America:
Dressler, R.L., 1982. Biology of the orchid bees (Euglossini). Annual biology and field guide. Instituto Nacional de Biodiversidad,
Review Ecology and Systematics 13, 373–394. Santo Domingo de Heredia, Costa Rica.
Ghazoul, J., 2005. Buzziness as usual? Questioning the global Smith-Pardo, A., 1999. Evaluación de cuatro diferentes métodos
pollination crisis. Trends in Ecology & Evolution 20, 367–373. de captura de abejas silvestres en el área de influencia del
Horner-Devine, M., Daily, G., Ehrlich, P., Boggs, C., 2003. Embalse Porce II. Revista Facultad de Ciencias Agropecuarias,
Countryside biogeography of tropical butterflies. Conservation Medellı́n, Colombia 52, 435–450.
Biology 17, 168–177. Steffan-Dewenter, I., Munzenberg, U., Burger, C., Thies, C.,
Janzen, D.H., 1971. Euglossine bees as long distance pollinators of Tscharntke, T., 2002. Scale-dependent effects of landscape
tropical plants. Science 171, 203–205. context on three pollinator guilds. Ecology 83, 1421–1432.
Janzen, D.H., 1981. Reduction in euglossine bee species richness Steffan-Dewenter, I., Potts, S.G., Packer, L., 2005. Pollinator
on Isla del Caño, a Costa Rican offshore island. Biotropica 13, diversity and crop pollination services are at risk. Trends in
238–239. Ecology & Evolution 20, 651–652.
Jenness, J., 2004. Nearest features (nearfeat.avx) extension for Tonhasca, A., Blackmer, J.L., Albuquerque, G.S., 2002a.
ArcView 3.x, v. 3.8a. Available at: <http:// Abundance and diversity of euglossine bees in the
www.jennessent.com/arcview/nearest_features.htm>. fragmented landscape of the Brazilian Atlantic Forest.
Jenness Enterprises, Flagstaff, AZ (accessed August 2007). Biotropica 34, 416–422.
Marlin, J.C., LaBerge, W.E., 2001. The native bee fauna of Tonhasca, A., Blackmer, J.L., Albuquerque, G.S., 2002b. Within-
Carlinville, Illinois, revisited after 75 years: a case for habitat heterogeneity of euglossine bee populations: a re-
persistence. Conservation Ecology 5, 1–24. evaluation of the evidence. Journal of Tropical Ecology 18, 929–
Milet-Pinheiro, P., Schlindwein, C., 2005. Do euglossine males 933.
(Apidae, Euglossini) leave tropical rainforest to collect Tonhasca, A., Blackmer, J.L., Albuquerque, G.S., 2003. Dispersal of
fragrances in sugarcane monocultures? Revista Brasileira de euglossine bees between fragments of the Brazilian Atlantic
Zoologia 22, 853–858. Forest. Journal of Tropical Ecology 19, 99–102.
Oksanen, J., Kindt, R., Legendre, P., O’Hara, R. 2006. Vegan: Tylianakis, J.M., Klein, A.M., Tscharntke, T., 2005. Spatiotemporal
Community Ecology Package. Ordination methods and other variation in the diversity of Hymenoptera across a tropical
useful functions for community and vegetation ecologists. habitat gradient. Ecology 86, 3296–3302.
<http://cc.oulu.fi/jarioksa/> (accessed August 2007). Vamosi, J.C., Knight, T.M., Steets, J.A., Mazer, S.J., Burd, M.,
Olofsson, J., Shams, H., 2007. Determinants of plant species Ashman, T.L., 2006. Pollination decays in biodiversity hotspots.
richness in an alpine meadow. Journal of Ecology 95, 916– Proceedings of the National Academy of Sciences of the United
925. States of America 103, 956–961.
Otero, J.T., Sandino, J.C., 2003. Capture rates of male euglossine Williams, N.H., Dodson, C.H., 1972. Selective attraction of male
bees across a human intervention gradient, Chocó region, euglossine bees to orchid floral fragrances and its importance
Colombia. Biotropica 35, 520–529. in long distance pollen flow. Evolution 26, 84–95.