Bee diversity along a disturbance gradient in tropical lowland forests of south-east Asia


  • Lee Hsiang Liow,

    1. The Swedish Biodiversity Centre, CBM, Box 7007, 750 07 Uppsala, Sweden; and the
    2. Department of Biological Sciences, National University of Singapore, Blk S2, 14 Science Drive 4, Singapore 117543, Republic of Singapore
    Search for more papers by this author
  • Navjot S. Sodhi,

    Corresponding author
    1. Department of Biological Sciences, National University of Singapore, Blk S2, 14 Science Drive 4, Singapore 117543, Republic of Singapore
    Search for more papers by this author
  • Thomas Elmqvist

    1. The Swedish Biodiversity Centre, CBM, Box 7007, 750 07 Uppsala, Sweden; and the
    Search for more papers by this author


  • 1 Bees are believed to be dominant pollen vectors in tropical forests, yet studies specific to bees in south-east Asia are rare. Regeneration and restoration of the rapidly disappearing lowland forests of this region are reliant on bees, thus there is an urgent need for forest bee data at the community level.
  • 2 Bee communities of eight forested sites in Johor (Malaysia) and Singapore were surveyed three times each from February to August 1999 at the below-canopy level. These sites ranged from relatively undisturbed primary lowland dipterocarp forests to late secondary forests and exotic forests, including an oil palm plantation. We attempted to elucidate the environmental factors that correlated with the distribution of bees.
  • 3 Bee abundance, in particular that of Apidae, was significantly higher in larger primary forests than other types of forests. However, bee species richness was higher in disturbed forests.
  • 4 The distribution of bees was apparently influenced by variables closely related to forest disturbance and resource abundance, such as the density of big trees (diameter at breast height 30–40 cm), temperature and flowering intensity of trees and shrubs.
  • 5 More stingless bees (Trigona spp.) were found where trees were larger and ambient conditions more constant but flowering intensities lower.
  • 6 The differences between the bee communities in forests of urban Singapore and primary forests in Johor may indicate that ecological processes in the forests of Singapore, in particular pollination, may be changing. However, pollination may not be totally intact in the primary forests surveyed, as their bee communities seemed to be depauperate.
  • 7 The role of important pollinators, especially bees, for the long-term survival of tropical lowland forests is poorly understood. Our study indicates that we urgently need more thorough understanding of pollination and pollinators, as some bee species appear to be disappearing from disturbed tropical lowland forests.


Deforestation in tropical south-east Asia is rapid, reaching 1·6% between 1981 and 1990 compared with 0·9% in the rest of the tropics (Groombridge 1992). In particular, the island state of Singapore has largely been urbanized except for small forested areas totalling about 20 km2 or 3% of Singapore’s total land area (Lum 1999). There have also been intensive land-use changes in Peninsular Malaysia since the 1970s and now the total lowland evergreen broadleaf forest (including disturbed natural forests) stands at 31·7%, of which only 9·0% (2·9% of the total land area of Peninsular Malaysia) is protected (Iremonger, Ravilious & Quinton 1997). In Johor, Peninsular Malaysia, most of the original dipterocarp forests have been logged for timber or cleared for plantations. Oil palm plantations alone accounted for 525 360 ha (26·3%) of the land in Johor in 1990 (Sukaimi et al. 1993). Little is known about how this extensive habitat loss will affect species diversity and ecological processes.

For several reasons, we chose to study the effects of forest disturbance on bees in south-east Asia. First, insects and other invertebrates may be much more important than vertebrates for the maintenance of vital ecosystem processes (Wilson 1987). Despite this, few intensive surveys of tropical insect diversity have ever been made (Holloway, Kirk-Spriggs & Chey 1992; Kremen et al. 1993). Secondly, insect taxa (including butterflies, tiger beetles and termites) are used increasingly as habitat or environmental quality indicators (Holloway & Barlow 1992; Pearson & Cassola 1992; Eggleton et al. 1996; Hamer et al. 1997; Hill 1999; Jones & Eggleton 2000; Kitching et al. 2000). Thirdly, the loss of ecological processes is less apparent than species loss as a negative consequence of anthropogenic habitat changes (Kearns & Inouye 1997), although the loss of ecological processes such as pollination (Buchmann & Nabhan 1996; Allen-Wardell et al. 1998) is at least as destructive as physical changes to a natural ecosystem. Fourthly, bees are believed to be the dominant pollen vector in tropical forests (Bawa 1990; Renner & Feil 1993; Roubik 1993a). For instance, 74% of the insects visiting flowers in Sumatra are apid bees (Inoue et al. 1990).

The bee fauna of the oriental region is the poorest (89 genera) in the world (Michener 1979) and, accordingly, the bee fauna in tropical south-east Asia is also species poor, despite an extremely high plant species richness (Whitmore 1984). However, tropical bee communities are more ecologically diverse (Roubik 1989). This means that each species potentially pollinates more plant species and may have a greater part in maintaining fertilization in the angiosperms than bees in other regions (Michener 1979). Despite their purported role as dominant pollen vectors, studies specific to bees in tropical south-east Asia are rare, in contrast to the neotropics (Bawa et al. 1985; Roubik 1993b; Frankie et al. 1997 and references therein). There are many botanical and anecdotal accounts of bee-pollinated plants in south-east Asia (Jackson 1981; Appanah 1987; Momose et al. 1997; Ghazoul, Liston & Boyles 1998; Sakai et al. 1999) but few studies deal specifically with the pollinators themselves. Exceptions include studies done in Sarawak (Kato 1996; Inoue & Hamid 1997; Nagamitsu & Inoue 1997), Peninsular Malaysia (Appanah 1981), Brunei (Roubik 1996) and Sumatra (Sakagami, Inoue & Salmah 1990; Salmah, Inoue & Sakagami 1990), but these are concerned mainly with the family Apidae only.

There is no comprehensive inventory, systematic comparison between sites or study on the habitat requirements of bees in Malaysian and Singaporean forests. We attempted to partly fill these gaps. First, we asked if there are differences in the assemblages of bee species found at different locations, in particular between undisturbed and disturbed sites. We tested the hypothesis that primary forests are richer than disturbed forests in the species richness and abundance of bees. Secondly, we asked whether bees show habitat preferences. We tested whether vegetation structure and/or microclimate are correlated with the distribution of the bee species. Thirdly, we attempted to make generalizations about the bees found at the below-canopy level in forests of varying disturbance in the southern Malay Peninsula and discussed the implications of this study for conservation in this region.

Methods and materials

Eight sites were surveyed, five in Singapore and three in Johor, Malaysia (Fig. 1), to represent a range of tropical lowland forests of varying degrees of anthropogenic disturbance. A summary of the forest types (definitions based on Whitmore 1984), location and background information of the eight sites is given in Table 1.

Figure 1.

The eight study sites. BK, Bekok; BL, Belumut; BT, Bukit Timah Nature Reserve; HW, Holland Woods; KR, Kent Ridge; MC, MacRitchie; NS, Nee Soon; UMP, UMP oil palm plantation. The reference sites (undisturbed forests) have been bracketed in all other figures.

Table 1. Summary of information on study sites surveyed for bees from February to August 1999 including the names, state and countries of the study sites and abbreviations used in the text; the description of the forest type (based on Whitmore 1984); the landscape of the matrix surrounding the surveyed forest; the geographical co-ordinates of the site; the size of the sites; the dates of the bee surveys; transect lengths at each site; and approximate age of the forests
Name and location of site and abbreviation used in textDescriptionSurrounding matrixCoordinatesSize (ha)Dates of bee surveys (1999)Transect length, km (hours surveyed)Approximate age of forest
Belumut, Johor Malaysia (BL)Primary hillOil palm and rubber plantations and logged forests2°03·90′N 103°31·57′E>20009–11 February, 25–27 May, 3–5 August1·2 (16·04), 1·1 (4·63), 0·5 (13·99)Never logged
 Dipterocarp forest      
Bekok, Johor Malaysia (BK)Primary lowland and hillOil palm and rubber plantations and logged forests2°20·84′N 103°09·41′E>200017–19 June, 13–15 July, 11–13 August1 (13·80), 1 (16·59), 1 (13·26)Never logged
 Dipterocarp forest      
Bukit Timah Nature Reserve, Singapore (BT)Primary hillUrban areas1°20′N 103°50′E 873–5 March, 20–22 April, 28–30 June1 (12·24), 1 (12·80), 1 (11·39)Never logged
 Dipterocarp forest      
MacRitchie, Singapore (MC)Secondary forest with primary patchesUrban areas and secondary forest1°20′N 103°50′E521·024–26 February, 12–14 April, 22–24 June1 (12·04), 1 (10·32), 1 (12·41)60–80 years
Nee Soon, Singapore (NS)Secondary forest with primary patchesUrban areas and secondary forest1°20′N 103°50′E794·616–18 March, 26–28 April, 6–8 July1 (16·16), 1 (14·34), 1 (14·64)60–80 years
Holland Woods, Singapore (HW)Secondary forest mixed (Albizia spp·) exotic forestUrban areas1°20′N 103°50′E41·55–7 May, 8–10 June, 26–28 July1·8 (24·42)20–40 years
Kent Ridge Park, Singapore (KR)Secondary forest and open parklandUrban areas1°20′N 103°50′E27·230 March–1 April, 18–20 May, 1–3 July1 (10·96), 1 (11·88)20–40 years
UMP oil palm plantation, Johor (UMP)Uniform oil palm stand (Elaeis guineensis)Pineapple, rubber and fruit tree plantations1°35·81′ N103°27·57′E2105·612–14 March, 12–14 May, 20–22 July1 (11·75), 1 (12·45), 1 (13·13)Cleared 1946/7

Preliminary studies were conducted between December 1998 and January 1999 to optimize sampling methods. Three yellow funnel traps and three yellow floating platforms on Petri dishes were set up at MacRitchie, Singapore, with baits (honey solutions, sugar solutions and banana pulp) in December 1998 to test the feasibility of trapping and/or attracting bee individuals. The traps were set 50 m apart with funnels alternating with the Petri dishes. Some bees were trapped after entering the funnel or by the surface tension of the liquid. Both methods rendered the bees wet, making taxonomic identification difficult. The average trap rates (± SE) were 9·0 ± 2·4 and 8·3 ± 4·3 for individuals trap−1 day−1 for funnel traps and Petri dishes, respectively. Bees were not attracted to banana pulp.

Honey baiting on vegetation was also tested. Honey–salt–water and sugar–salt–water solutions of varying concentrations were sprayed on vegetation 1–2 m from the ground at two sites. At MacRitchie there were three baiting points of about 1 m in diameter for each concentration of honey or sugar solutions, while at Bekok there was one of each. The baiting points were about 10 m apart at each site.

A solution of honey diluted at a ratio of 1 : 2 (honey : water, v:v) and solutions diluted at a ratio of 1 : 4 (honey : water and sugar : water), with salt concentration constant at 4·00 ± 0·25 g l−1 of solution sprayed on vegetation, did not have significantly different recruitment rates for different morphospecies of bees. It was also found that the method of baiting on vegetation attracted more bees than baited funnel type traps and Petri dishes. During each check, approximately 20 ± 3 stingless bees (Trigona spp.) were counted at individual vegetation baiting points.

For the actual bee surveys, one to three transects of 0·5–1 km were set up at each site (see Table 1 for transect lengths). Honey solution (1 : 4, honey : water) with 2 cm3 salt 500 ml−1 of solution, standardized with a hand refractometer, was used to attract bees. Thirty jets (20 ml) of this solution were sprayed on vegetation marked with coloured flagging tape every 100 m along the transects. The baited spots were about 1 m in diameter and between 1 and 2 m above the ground. The baits were replenished on alternate transect checks by spraying 15 jets of the solution on the same previously sprayed spots. If the solution was washed away by rain, the baiting spots were sprayed at least half an hour before the transects were surveyed again. Bees attracted to each of these baited spots were caught with a standard insect net four times a day (between 07:30 and 17:00 h) during a maximum of 3 min. The time lapse between bait application and collection ranged from 30 min to 16 h (e.g. when the bait was left overnight). The surveyor waited at each baited spot for 20 s to scan the area for bees and moved on if there was none. Bees spotted when the surveyor was walking along the transects were also collected and the substrates (including flowers, leaves and resins) on which they were caught were recorded. Each collection cycle at a particular site ran for 3 consecutive days. We visited each study site three times between February and August 1999 (Table 1). Temperature, humidity and light intensity were measured every 200 m along the transact during the first and third checks of the transact during the day, using a thermohygrometer and an illumination meter.

The bees collected were identified with the help of R. W. Brooks (Snow Entomological Museum at the Kansas State University) and are now deposited at the Raffles Museum of Biodiversity Research (RMBR) of the National University of Singapore and the Snow Collections of the Kansas State University.

To determine if floral resource availability affected the numbers of bees caught, the flowering intensities every 200 m along the transects were recorded on the second day of each bee collecting cycle. Binoculars were used to scan a circular area (8 m radius = approximately 200 m2 ground area) and trees (more than 2 m in height) with open flowers were counted. A 50-m2 area was divided into 10 sectors and the number of sectors with flowering shrubs recorded. In both cases, open flowers were defined as single flowers or parts of inflorescence with anthers and/or stigmas no longer being concealed by the corolla and/or calyx but before the corolla and/or calyx had fallen off. When the perianth was not obvious, open flowers were taken to be single flowers or parts of the inflorescence with mature anthers or stigmas.

To determine if vegetation structure was a factor related to bee species distribution, the following variables were measured and recorded along transects at 150-m intervals. (i) The canopy cover of the forest, using a spherical densiometer, according to Lemmon (1957). (ii) The diameter at breast height (d.b.h.) of trees in which this exceeded 2 cm. We also estimated visually whether these trees were more or less than 10 m in height. (iii) The number of trees of less than 2 cm d.b.h. (iv) The number of dead trees. (v) The number of palms. (vi) The percentage ground cover. (vii) The leaf litter depth, measured by gently inserting a ruler vertically into the leaf litter at 12 random points in each plot.

The Kruskal–Wallis (KW) test was used to determine if the catch rate of bee species and number of bee individuals at the eight study sites were significantly different. On finding that there were significant differences, the catch rates were manually ranked and Duncan’s multiple range test was used to determine which sites were different from the others. The two catching methods (netting at baits and netting along transects) were compared using the Mann–Whitney U-test. All these tests were performed using SAS version 6.12 (SAS Institute Inc. 1990). Bee diversity indices (Margalef’s, Menhinick’s, Simpson’s, Berger–Parker, McIntosh’s, Brioullin’s and Shannon’s indices) were hand calculated according to Magurran (1988). Each set of indices was ranked and Kendall’s coefficient of concordance, W, was calculated according to Siegel & Castellan (1988).

The bee species presence–absence data from the 21 transects in the eight study sites were subjected to a cluster analysis using PC-ORD version 2.0 (McCune & Mefford 1995). This analysis used Euclidean distance and Ward’s method. Similarly, the presence–absence data of the Apidae species from our eight study sites and those from Brunei (Roubik 1996) and Sumatra (Salmah, Inoue & Sakagami 1990) were also subjected to a cluster analysis.

Twenty-four environmental variables (the mean d.b.h. and density of trees < 10 m height plot−1; the mean density of palms < and > 10 m height plot−1; the mean d.b.h. and density of trees > 10 m height plot−1; the mean number of trees < 2 cm d.b.h. plot−1; the mean number of dead trees plot−1; shrub and canopy cover; leaf litter depth; the density of trees with d.b.h. < 10 cm, between 10 and 20 cm, 20 and 30 cm, 30 and 40 cm and > 40 cm; the number of flowering trees m−2; the flowering density index of shrubs; mean and standard deviations of temperature, humidity and light intensity) were correlated using SAS (SAS Institute Inc. 1990). These variables formed six groups according to whether they were correlated with each other (Pearson correlation coefficient > 0·50) or not. Only one variable from each group was retained for ordination analysis. The retained variables were the ones that were most biologically relevant: the mean density of trees < 10 m in height, leaf litter depth, mean density of trees with d.b.h. between 30 and 40 cm, number of flowering trees m−2, flowering density index of shrubs and mean temperature.

To determine the factors affecting the distribution of bee species, canonical correspondence analysis (CCA; ter Braak 1986) in PC-ORD version 2.0 (McCune & Mefford 1995) was performed using data matrices of log-transformed bee species abundance data and the six environmental variables retained. Axis scores were standardized using Hill’s (1979) method and scaled to optimize the representation of species.


Species present and richness

Forty-five morphospecies and 1613 individuals from five bee families (Anthophoridae, Apidae, Colletidae, Halictidae and Megachilidae) were collected during 71 days (279 h). The three most abundant species in the eight sites combined were Trigona (Tetragonula) geissleri Friese 1918, T. (T.) melina Gribodo 1918 and T. (T.) laeviceps Smith 1857, with 412, 202 and 546 individuals collected, respectively, while seven species, T. (Lepidotrigona) ventralis Smith 1857, T. (Geniotrigona) thoracica Smith 1857, two species of Nomia (Maculonomia), two species of Lipotriches and one species of Halictidae were collected only once each during the collection period. The total number of species collected at each site ranged from four at Nee Soon to 22 at MacRitchie. The total number of individuals collected ranged from 63 individuals at Kent Ridge Park to 444 individuals at Belumut.

The cumulative collection curves at Bekok, Belumut, Bukit Timah, Holland Woods and Kent Ridge appeared to have reached their asymptotes during this study (Fig. 2). However, those at MacRitchie, Nee Soon and the UMP oil palm plantation were still rising. Only bees of the honey bee family, Apidae, were encountered at the two large primary forest tracts in Johor, while the family Halictidae dominated in the oil palm plantation, an exotic forest (Fig. 3a,b). Apis mellifera Linn. 1758, the globally widespread exotic honey bee, was not found.

Figure 2.

The cumulative number of bee species collected at the eight study sites. For site abbreviations see Fig. 1.

Figure 3.

(a) Proportion of bee families collected at each site, based on the number of individuals collected, and (b) proportion of bee families collected at each site, based on the number of species collected. For site abbreviations seeFig. 1.

The mean numbers (± SE) of bees caught at baits and along transects were 5·62 ± 2·65 and 1·08 ± 0·39 individuals h−1, respectively. The baiting method was significantly more efficient at catching bees than just netting bees sighted along transects (Mann Whitney’s U = 10·43, d.f. = 7,7, P < 0·001). The mean number of Apidae caught at baits (3·83 ± 4·63 individuals per hour) was significantly higher than the mean number of non-Apidae bees caught at baits (0·09 ± 0·07) (U = 29·011, d.f. = 7,7, P < 0·0001). Of all the bees caught while they were approaching or foraging on flowers, 80·2% were caught on only four species of continuously flowering plants. These were Dillenia suffruticosa (Griff.) Martelli, Melastoma malabatricum Linn., Asystasia gangetica (L.) T. Anderson and Stachytarpheta indica (L.) Vahl; of these the first two are plants of early secondary forests and the last two are weeds found in open areas. All are native plants except S. indica, which originates from South America.

The number of individuals and species collected per hour varied from 12·15 ± 1·48 and 1·63 ± 0·22 (at Bekok and MacRitchie) to 0·95 ± 0·26 and 0·38 ± 0·11 (at Bukit Timah), respectively (Fig. 4a,b). The mean numbers of bee species caught per hour differed significantly among sites (KW = 44·763, d.f. = 7, P < 0·0001). Duncan’s multiple range test showed that the ranked mean numbers of bee species caught per hour at Belumut, NeeSoon and Bukit Timah were significantly lower than those at Holland Woods, MacRitchie, Bekok and Kent Ridge (P < 0·05). The mean number of individuals caught per hour differed significantly among sites (KW = 40·673, d.f. = 7, P < 0·0001). Duncan’s multiple range test showed that the ranked mean numbers of bee individuals caught per hour at Bekok and Belumut were not significantly different from each other but were significantly higher than those at the remaining sites (P < 0·05), with the exception of Nee Soon.

Figure 4.

(a) Mean number of species collected per hour at each site, and (b) mean of number of individuals collected per hour at each site. Bars represent standard error. For site abbreviations see Fig. 1.

Bee communities

Of the 11 diversity indices calculated, six gave the most disturbed forest, the oil palm plantation, the highest diversity value and Nee Soon, a late secondary forest, the lowest diversity value (Table 2). The Kendall coefficient of concordance, W, was significant (W = 0·453, P < 0·0001), indicating reasonably good consensus among indices. Averaging the ranks of the sites (using the 11 indices and the total number and species of bees collected at each site), the most diverse site was MacRitchie, a secondary forest, while the least diverse site was Bukit Timah Nature Reserve, a small primary forest (Table 2).

Table 2. Bee species richness (S), absolute numbers of bees (N) collected at each site (see Table 1 for site abbreviations) and diversity indices calculated according to Magurran (1988). D (Mg), Margalef’s index; S (Mn), Menhinick’s index; 1/D(S), inverse of Simpson’s index; N/Nmax, inverse of Berger-Parker’s index; U (McIn), McIntosh’s index in general form; D (McIn), McIntosh’s dominance measure; E (McIn), McIntosh’s evenness; HB, Brillouin index; E of HB, Brillouin’s evenness index; H′, Shannon’s index and E of H′, Shannon’s evenness index
SitesSND (Mg)D (Mn)1/D (S)N/NmaxU (McIn)D (McIn)E (McIn)HBE of HBH′E of H′
BK 94191·320·443·392·07228·070·480·680·3730·711·5430·256
BL 74440·980·332·121·77305·590·330·50·660·270·8810·145
BT 5 421·070·771·61·27 33·440·240·360·2820·20·7650·205
NS 52320·730·331·11·05221·090·921·550·0980·140·2490·046
HW 91111·70·853·852·41 57·30·530·730·6430·761·5980·339
KR10 632·171·264·123 31·730·570·730·6220·741·6080·388
UMP17 643·862·138·884·92 22·580·740·850·8810·822·2590·543

Three of the more disturbed forests (Kent Ridge, Holland Woods and the UMP oil palm plantation) and transects A and B from MacRitchie (a secondary forest) formed the first main cluster in the dendrogram drawn using presence–absence data of all bee species found. The remaining sites, consisting of the three primary forest sites (Bekok, Belumut and Bukit Timah), one secondary forest site (Nee Soon) and transect C from MacRitchie, formed another main cluster. Five out of six transects from the two large tracts of primary forests (Bekok and Belumut) were very similar, while many transects from the closed forests in Singapore grouped together (Fig. 5).

Figure 5.

A dendrogram of bee species recorded along the transects of the eight study sites calculated with cluster analysis in PC-ORD using Euclidean distance and Ward’s method (see Table 1 for abbreviations).

A dendrogram of bees of the family Apidae (Trigona and Apis), drawn using presence–absence data, separated the sites from this study (Singapore and Johor) from the Sumatran sites (data from Salmah, Inoue & Sakagami 1990) and those in Brunei (data from Roubik 1996) (Fig. 6). Within the sites from the present study, the primary forests in Johor were clearly distinguished from the rest of the sites in Singapore and the oil palm plantation in Johor. Within the Sumatran and Bornean sites, the pristine primary forests formed a separate cluster (Fig. 6).

Figure 6.

A dendrogram of the distribution of Apidae in 14 sites, calculated with cluster analysis in PC-ORD using Euclidean distance and Ward’s method, where BK, BL, BT, MC, NS, HW, KR and UMP are from this study (see Fig. 1 for abbreviations). Sum P1, primary forest; Sum P2, primary forest mixed with secondary forest; Sum S1, secondary forest; Sum S2, secondary forest mixed with other disturbed habitats; Sum D1, coconut plantations and orchards; Sum D2, villages (Sumatran sites from Salmah, Inoue & Sakagami 1990); Bru, Brunei primary forest (Roubik 1996).

A CCA was performed on a primary data matrix consisting of log-transformed numbers of bee individuals collected per hour for 45 bee species, and a secondary matrix with six environmental variables. The first two CCA axes explained 28·2% and 22·4% of the variation in the data sets, respectively (Table 3). Both the first and second axes had high loadings of the mean temperature of the sites, the density of large trees (d.b.h. 30–40 cm) and the flowering density index of shrubs (Table 3). Five groups of bees could be distinguished, namely the honey bees (group 1), the first group of stingless bees (group 2, including T. melina, itama, reepeni, ventralis and terminata), the second group of stingless bees (group 3, including T. geissleri and laeviceps), the first group of megachilids and anthophorids (group 4, including Nomia, Ceratina, Amegilla and Xylocopa) and the second group of megachilids and anthophorids (group 5, including Lipotriches and Lasioglossum) (Fig. 7). Both the groups of stingless bees increased in abundance with the increase in the number of big trees (the density of trees with d.b.h. between 30 and 40 cm), and decreased with increasing temperature and flowering intensity of both trees and shrubs. The presence of honey bees (Apis) was not strongly related to the measured variables. Group 5 increased in abundance with increasing flowering intensity and temperature and tolerated the low density of larger trees (Fig. 7).

Table 3. Summary statistics of the first two CCA axes of the CCA performed with a primary matrix of the log-transformed numbers of bees collected per hour and a secondary matrix of six environmental variables and their canonical coefficients
Canonical axis12
Eigenvalue 0·668 0·529
% of variance in species data explained28·222·4
Cumulative percentage of variance in species data explained28·250·6
Pearson correlation (species–environmental variable) 0·995 0·999
Environmental variables (abbreviation)CanonicalCoefficients
Density of trees < 10 m height (A)−0·121 0·600
Leaf litter depth (B) 0·499−0·326
Density of trees with d.b.h. between 30 and 40 cm (C) 2·080−1·896
Flowering tree m−2 (D)−0·582−0·143
Flowering density index of shrubs (E)−2·013 1·670
Temperature (F) 2·935−4·204
Figure 7.

The ordination diagram from the CCA of a primary matrix of the log-transformed numbers of bees collected per hour and a secondary matrix of six environmental variables and their canonical coefficients. Solid circles refer to sites (for abbreviations see Table 1). Small diamonds refer to bee species. Arrows A–F refer to the environmental variables A, density of trees < 10 m in height; B, leaf litter depth; C, density of trees with d.b.h. between 30 cm and 40 cm; D, flowering trees m−2; E, flowering density index of shrubs; F= temperature.


Disturbed forests tended to have greater absolute bee species richness (α-diversity). These disturbed habitats may also attract more ‘tourists’, i.e. bees that do not reside within the habitat under investigation and those with potentially large foraging ranges (e.g. the trap-lining Amegilla and Xylocopa spp.). However, the relatively undisturbed lowland dipterocarp primary and secondary forests had high absolute abundance of bees.

The baits used tended to attract mainly recruiting species: the honey bees (Apis) and stingless bees (Trigona). But clearly, in the disturbed forests, many Trigona spp. were not found despite the sampling method, which was biased towards them. On the other hand, many non-recruiting species foraging high up in the canopy of the undisturbed forests may have been missed, so we cannot say that the total species richness of undisturbed forests is lower than disturbed ones, although that of the lower canopy and shrub level may be lower in undisturbed forests as shown by this study.

Diversity indices such as the ones calculated in this study do not adequately reflect the ecological status of the habitat in consideration. The primary forest sites were ranked as less diverse by these indices, while the absolute abundance of bees and their ecological function may be more important for maintaining the ecosystem than the absolute number of species. Lumping all bees together and gauging bee species richness per se is probably not a good surrogate for measuring the level of disturbance in these forests, although this study was not an attempt at doing so, but rather an attempt at finding the bee compositional differences between known levels of forest disturbance as dictated by the recorded histories of those sites.

Apidae, especially the genus Trigona, are ubiquitous in rain forest (Appanah, Willemstein & Marshall 1986) and its species are important pollinators (Sakai et al. 1999), especially in the understorey (Bawa & Opler 1975; Appanah 1981). Both the primary and secondary forests harboured a much greater proportion of Apidae: the more disturbed the forests were by human activity, the less Apidae bees occurred in the understorey. In fact, only Apis and Trigona were recorded in the understoreys of Bekok and Belumut. In similar but smaller forests in Singapore, the Apidae were also relatively dominant (99·1% of all bees caught in Nee Soon and 95·2% in Bukit Timah Nature Reserve).

The two more abundant families of non-Apidae bees were Anthophoridae and Halictidae. Colletidae and Megachilidae were rarely collected or observed during this study. Anthophoridae and Halictidae seemed to favour the more open and disturbed forests. They were frequently caught on flowers of common early secondary successional plants and widespread weeds in open habitats

There is no reason to believe that Sumatra and Brunei have a different bee composition from Peninsular Malaysia (as all belong to the shallow Sunda Shelf), with the exception of a few high elevation endemic species. However, 13 out of 27 species of Trigona and Apis found in Sumatra and Brunei, and thus expected during this study, were never encountered. It is not known why this was so, despite a comparatively lengthy collection period that included months with a higher flowering intensity in the lowland forests (March to July; Medway 1972) and similar collection methods. Perhaps some species are becoming so rare, even in the primary forests of Johor, that chance encounters are very low. It must be born in mind that these primary forests of Johor are also forest fragments, albeit large ones, surrounded by plantation forests. The alternative explanation is that population numbers of certain species may be low during intergeneral flowering periods (see Sakai et al. 1999 and below), hence decreasing the number of chance encounters. On a positive note, however, the feral A. mellifera was never encountered during this study even in the most disturbed forest habitats.

General flowering (Sakai et al. 1999) or mass flowering (Appanah 1985) in the lowland dipterocarp forests of west Malaysia is a well-recorded phenomenon, only reported in these forests. However, this phenomenon is not well understood ecologically (Corlett 1990). During general flowering (occurring at irregular intervals of 3–10 years), plant species from many different families flower together (Sakai et al. 1999). During the intergeneral flowering years, however, the level of flowering is more or less constant as a result of the blossoming of a changing assortment of species, and the peak flowering period during any year appears to be between March to July (Medway 1972). Our study spanned over this period of high flowering intensity although it did not coincide with a general flowering period. Hence it sheds light on the bee community structure during intergeneral flowering periods in Bekok, Belumut, Bukit Timah, Nee Soon and MacRitchie, all of which have at least some resemblance to the undisturbed forests described by Medway (1972). The Apidae must have some mechanisms for surviving periods of low flowering and these may include storage of floral resources (Sakai et al. 1999 and references therein) and/or the use of resources other than nectar and pollen (e.g. extra-floral nectaries). It is also possible that the bees may migrate to greener pastures or hibernate to survive unfavourable conditions.

In the highly disturbed forests Kent Ridge Park and Holland Woods, and the UMP oil palm plantation, where plant species composition is different and plant species diversity lower, flowering of both trees and shrubs is relatively constant. Numerous solitary bee species of Anthophoridae and Halictidae found in these forests thus have a constant supply of open flowers.

Many of the environmental variables correlated with the abundance or presence of bees are related to the level of disturbance in the forest. In the relatively undisturbed forests, temperature and light intensity were lower, humidity higher, and all three variables more constant than in the rest of the sites. This is due to a closed canopy that results from the higher density of subcanopy and canopy trees. Very few understorey plants or trees were observed flowering during this study. Under these conditions, only bees of the Apidae were collected in the understorey. The species numbers within Apidae were higher in Bekok and Belumut (nine and seven, respectively) than in the other sites. Out of the total of 14 species of Apidae collected at these two sites, eight were not collected elsewhere. In particular, T. melina, an abundant species collected at Bekok, appears to be a species restricted to primary forests, as also indicated by a Sumatran study (Salmah, Inoue & Sakagami 1990).

The CCA indicates that many of the stingless bees favour forests with larger trees and lower temperatures. But whether this is due to a preference for size and temperature rather than plant species composition or other factors not measured, is unknown. However, in the case of some Trigona it has been shown that nest sites (in large trees) are limiting factors for their density (Inoue et al. 1990). There are also indications from temperate studies that plant species richness and nest site availability are important indicators of bee species richness and abundance (Tscharntke, Gathmann & Steffan-Dwenter 1998).

In the intermediately disturbed forests of Singapore, physical parameters were more variable as there were more gaps in the canopy. Open area species such as Dicranopteris linearis (Burm.f.) Underw, a fern, and Clidemia hirta (L.) D. Don, an exotic shrub, were also apparent. MacRitchie in particular had abundant shrubs and treelets that were constantly in flower. Both Nee Soon and Bukit Timah are, in a physiognomic and a bee compositional sense (mainly Apidae), similar to the primary forests in Johor, although their species richness is lower. MacRitchie, consisting of a mix of open and closed habitats, is much more bee species rich, as can be expected of a mixed habitat (Niemelä 1996; Pandey & Shukla 1999). It also has a more even distribution of bee families. Bukit Timah Nature Reserve is notable because, despite its primary forest status, it is very small (87 ha with a lot of edge habitats) and apparently very poor in both species richness and abundance. Trigona laeviceps, which has been described as the most commonly found trigonid species (Inoue et al. 1984) especially in disturbed areas, dominated both Bukit Timah and Nee Soon.

At the other end of the spectrum, Kent Ridge, Holland Woods and the UMP oil palm plantation are more or less dominated by exotic plants (by ornamental plants in Kent Ridge, Albizia spp. in Holland Woods and Elaeis guineensis Jacq. in the oil palm plantation). Under these conditions, the family Halictidae and Anthophoridae were more commonly caught. Their occurrence correlated with higher temperatures and light intensity, lower humidity levels and greater flowering intensities.

We emphasize that baiting methods are a biased means of sampling (Southwood 1978). In particular, this study could not investigate the bee community in the canopy of trees, especially in the less disturbed forests where trees were commonly more than 20 m in height. It was also biased by the non-randomness of the transects chosen. However, the collection methods were standardized throughout the eight sites and it is possible to compare the bees collected from the sites of varying disturbances. The physical architecture of forests is determined mainly by plants, but the underlying perpetuation of plants is dependent on pollinators, especially in the tropics where many plants are dioecious (Renner & Feil 1993) and where wind-pollination is very rare (Whitmore 1984). Animal pollination is an important ecological process dominated by bees (Inoue et al. 1990). However, the extent of negative effects of habitat change on bees is not fully understood, although inferences can be made from this and previous studies.

We question whether there are enough pollinators surviving for the process of pollination to continue in forest fragments or regenerated logged forests. The five Singaporean forest sites studied are quite different in bee community structure compared with the primary forests in Johor, which may indicate that ecological processes in the forests of Singapore have changed. Stingless bees, dominant in closed forests, were absent from the oil palm plantation, a habitat that is rapidly becoming all-pervading in Peninsular Malaysia. It is possible that disturbed forests such as secondary forests do not contain adequate numbers of pollinators required for regenerating forests. These negative trends, which can also be seen in the South American tropics (Frankie et al. 1997), should be rectified before the level of pollination becomes dangerously low. On the basis of our present knowledge of tropical ecology we suggest that larger tracts of undisturbed tropical lowland forests must be preserved.


We are grateful to Björn Cederberg, Mats Pettersson and Linus Svensson and two anonymous referees for their constructive criticisms and suggestions, D.H. Murphy for suggesting this research topic, Lars-Åke Janzon for teaching L. H. Liow about bees, Robert Brooks for helping with the bee identification, H. Tan for helping in plant identification, and S. Appanah for a copy of his PhD thesis.

We also wish to thank Gard Otis, Marjorie Castellatta and Börge Pettersson for kindly providing information, Liow’s field assistants for their patience in the field, Arthur Lim for allowing Liow to carry out research in his oil palm plantation (UMP), and the local community at UMP for providing her with a conducive work environment.

This research would not have been possible without the loan and gifts of equipment from the Ecology and Systematics Lab and the Reef Ecology Lab of the Department of Biological Sciences, National University of Singapore (NUS) and RMBR of NUS. This research was funded by the Swedish Biodiversity Centre and a NUS research grant no. RP 960316 to N. S. Sodhi.

Received 12 November 1999; revision received 5 August 2000