Impacts of Tropical Forest Disturbance Upon Avifauna on a Small Island with High Endemism : Implications for Conservation

Southeast Asia’s rainforests are facing the highest relative rates of habitat destruction of any major tropical area (Mayaux et al. 2005, Achard et al. 2002) and this has been identifi ed as the major driving force of potential biodiversity loss across the region (Sodhi et al. 2004a). This has severe implication for the biodiversity ‘hotspot’ of Wallacea, a biogeographical region where a complex geological history has facilitated a high prevalence of endemic fauna (Whitten et al. 2002, Myers et al. 2000, Kinnaird 1995). Over 46% of resident vertebrates are restricted entirely to the hotspot, including more than 35% of bird species (Myers et al. 2000), while Sulawesi, the region’s largest island, supports 14 endemic bird genera, the highest of all globally identifi ed endemic bird areas (Stattersfi eld et al. 1998). As with much of Indonesia, Wallacean forests are being subjected to heavy clearance, primarily from


INTRODUCTION
Southeast Asia's rainforests are facing the highest relative rates of habitat destruction of any major tropical area (Mayaux et al. 2005, Achard et al. 2002) and this has been identifi ed as the major driving force of potential biodiversity loss across the region (Sodhi et al. 2004a).This has severe implication for the biodiversity 'hotspot' of Wallacea, a biogeographical region where a complex geological history has facilitated a high prevalence of endemic fauna (Whitten et al. 2002, Myers et al. 2000, Kinnaird 1995).Over 46% of resident vertebrates are restricted entirely to the hotspot, including more than 35% of bird species (Myers et al. 2000), while Sulawesi, the region's largest island, supports 14 endemic bird genera, the highest of all globally identifi ed endemic bird areas (Stattersfi eld et al. 1998).As with much of Indonesia, Wallacean forests are being subjected to heavy clearance, primarily from expanding agricultural activities related to population growth and socio-economic factors (Trainor 2007, Sodhi 2005a), but also from unsustainable logging practices (Marsden 1998).An estimated 55% of original vegetation cover and 85% of original pristine rainforest within the region has been lost or modifi ed (Forest Watch Indonesia/Global Forest Watch 2002).Predictive extinction models estimate that continued habitat alterations on this scale could result in the loss of up to 42% of fl ora and fauna species across Southeast Asia by 2100 (Sodhi et al. 2004a), although consequences could be more severe still in the Wallacea region due to its insular nature, being comprised of 13,500 oceanic islands (Trainor 2007).Island birds have been estimated to possess extinction risks up to 40 times greater than continental species due to their small ranges and population sizes, and consequently are highly vulnerable to habitat destruction (Trevino et al. 2007, Pimm et al. 1995).Indeed, over 90% of recent (post AD 1600) bird extinctions have been island endemics (Clements 2007, Birdlife International 2004), and almost 40% of species currently listed as threatened by the IUCN are restricted to oceanic islands-a highly disproportionate fi gure given the small land mass and contribution to global avian richness these ecosystems represent (IUCN 2009, Trevino et al. 2007, Johnson and Stattersfi eld 1990).Safeguarding the unique biodiversity of this important region is therefore a high conservation priority, and extensive research is required to determine more precisely how habitat loss and change impacts upon biodiversity so that effective measures may be taken to mitigate these impacts.However, due to its isolation, current understanding of the ecological associations of avifaunal communities in this area remains poor.Aside from a few recent surveys (Coates andBishop 1997, White andBruce 1987), the most reliable accounts of bird communities on many Wallacean islands date back to collectors' reports from the late nineteenth century (Trainor 2007) while the bulk of current research has focussed on lowland areas or on larger islands (Lee et al. 2007, Sodhi et al. 2005b, Waltert et al. 2004, Thiollay et al. 2002).Few studies have examined bird communities on small island or montane ecosystems, which are often highly endemic and potentially more ecologically fragile (Macarthur and Wilson 1967).This paper seeks to address this by examining variations in avifaunal communities across a disturbance gradient on a small Wallacean island.The study focuses on Buton, an attendant island of Sulawesi, which is representative of a small island ecosystem with a highly endemic avifauna assemblage.Recent work on these islands focussing on herpetofauna (Gillespie et al. 2005) and butterfl ies (Fermon et al. 2005) has demonstrated the complex nature of the relationships between forest disturbance and the distributions of endemic and habitat-specialist taxa.It has also highlighted the urgent need for more extensive research on the human impact on biodiversity in this region.The study also examines how a select key species, the Knobbed Hornbill Rhyticeros cassidix, responds to environmental disturbance.This species was chosen due to its potential as both a conservation fl agship and an umbrella species (Sutherland 2000, Caro andO'Doherty 1999).It is a highly distinctive, charismatic bird which is widely recognised locally, being the faunal symbol of South Sulawesi province.Further, its function as an important seed disperser, coupled with its extensive habitat requirements (Kinnaird 1998), mean its conservation could have wider benefi ts for other organisms in its range.The relationship between environmental disturbance and the populations of several other large-bodied Sulawesi endemic species is also examined, as is the ecological response of forest specialist species, dietary guilds and regional endemic avifauna overall.

Study site
The study focuses on Buton Island, the largest (5,600 km 2 ) attendant island of Sulawesi in the Indonesian archipelago (Figure 1).The island is approximately 100 km long and 42 km wide at its widest point.Altitude ranges from 0-200 m in coastal areas to 400 m along the island's central spine, with  (Whitten et al. 2002, O'Donovan 2001).Buton experiences a tropical monsoon climate with a June-September dry season and a November-April wet season.Mean annual rainfall ranges between 1500-2000 mm, peaking between April and June (Whitten et al. 2002), with mean annual temperatures of 25-27ºC.Recent work using remotely-sensed and GIS data has identifi ed much of the island as being of high conservation value (Cannon et al. 2007).Buton has been shown to support a rich avifauna, with at least 231 bird species including 52 Sulawesi endemics being reported (Catterall 1997).However, the island's forest habitats have undergone signifi cant clearance in recent years.Between 1991-2002 over 13% of land in Southern Buton (27,809 ha) was converted from forest to non-forest land use (Seymour 2004).The primary cause of this deforestation is considered to be agricultural expansion by small-holder farms, occurring due to both population increase and an infl ux of trans-migrant settlers from Bali and the Moluccas (Seymour 2004).Further clearance has resulted from selective logging and asphalt mining (Seymour 2004).Research was conducted in and around the Lambusango Forest (5°10' S, 122°24' E), a 65,000 hectare area of uninhabited lowland tropical forest divided into a 28,510 hectare strict forest reserve and a 35,000 hectare limited production forest (Singer and Purwanto 2006) (Figure 2).A great diversity of tree species occurs within the reserve, with no single family being predominant.The underlying geology of the area is Quaternary karst coral limestone (O'Donovan 2001).

Sample sites
Sampling was conducted in three forest areas located throughout the reserve.Analysis of habitat structure (see below), supported by visual observations and research into local ecological history, suggests that these three forest areas correspond approximately to areas of near-pristine primary forest in the limited production forest; well-regenerated secondary forest in the strict reserve that was subjected to agricultural clearance and logging until the Lambusango conservation area was set up in 1975, and heavily-disturbed secondary forest in the reserve's periphery which has been recently subjected to intermittent logging, shifting cultivation and rattan extraction.The underlying environmental parameters of these last two categories (geology, topography, etc) is expected be similar to the primary forest sites, with disturbance being the only signifi cant cause of difference in vegetation structure.Each of these disturbance classes was far enough removed from other forest types to be considered independent.Additionally, an area of recently-cleared mixed farmland consisting of cassava (Manihot esculenta), papaya (Carica papaya) and rice fi elds (Oryza sp.) was surveyed.Four 900 m linear transects spaced at least 1 km apart were used in each area (Figure 2).Linear transects were used as these presented the most logistically viable way of conducting surveys in the protected forest, as establishing scattered plots throughout the reserve would have required extensive trailcutting which would not have been possible given time and accessibility constraints, and could have been ecological damaging given access trails do not currently exist across much of the reserve.Each transect contained seven sample sites, with each site being spaced 150 m apart.This gave a total of 112 study sites with 28 sites located in each habitat category.The elevation of primary and disturbed secondary forest sites varied between 300-400 m, while elevation of the regenerating secondary forest sites varied between 650-700 m.These altitudinal variations are not expected to be great enough to cause signifi cant systematic changes in vegetation structure, being within the elevation range of lowland forest (Whitten et al. 2002).Farmland sites were located at elevations between 100-150 m.

Vegetation sampling
A series of vegetation variables were measured within a 20 m radius of each forest study site to provide evidence for differences in habitat structure.Vegetation variables were not measured in farmland sites, as differences in habitat structure here were clearly evident.The sum total of large trees with a diameter at breast height (dbh) of >50 cm was counted within each plot, with the mean dbh of large trees also being recorded at each site.Canopy cover was evaluated at each plot utilising a canopy scope constructed from a perspex square marked with a 5X5 grid of dots separated by 3cm (Brown et al. 2000).At each site fi ve measurements were taken, with density assessed by holding the scope 10 cm away from an observer's eye-level and towards the largest visible canopy gap, with the number of dots unobscured by vegetation being recorded, thus a higher number refl ects a more open canopy.Relative understorey densities were estimated utilising a 1.5 m measuring pole marked with 50 black bands.The pole was held horizontally 1 m off the ground, where a consistent observer counted the number of bands completely unobscured by vegetation at 10 m at four points within each quadrat.Again a higher number is indicative of a sparser understorey density.Coverage of rattan (Calamus sp.), which can be gauged as an indicator of forest quality due to its invasive nature in recently disturbed habitats, was also estimated visually as a percentage of the area of each plot (Table 1).

Bird sampling
Bird communities were surveyed at each study site using 50 m fi xed-radius circular plot point counts (Bibby et al. 2002).Each point count was repeated once, with the total number of species detected after both counts being recorded.Point count surveys were led by the fi rst author and Mr Dani Heryadi of Operation Wallacea, who both had several months' fi eld experience with avifauna in the Lambusango, in addition to several years experience working with tropical birds.The data collection period corresponded to the breeding season for most local bird species.Sampling was conducted each morning between 06:00 and 08:00, this being the period where bird detectibility is highest and mobility is low, reducing the chance of recording contacts multiple times (Marsden 1999, Wunderle 1994).Point count samples were begun on immediate arrival at each study site, with no 'settling in' period being used.This has been shown to allow the recording of any birds disturbed by the surveyors, thereby increasing the number of contacts made per count (Lee and Marsden 2008).A 10 minute sampling period was used, as counts of this length have a reduced likelihood of multiple contact recording, while still being capable of detecting >80% of bird species present in an area (Lynch 1995, Waide andWunderle 1987).All species seen and heard during each count were recorded, excluding those fl ying above the canopy, as these may have been wandering or passage birds not associated with the habitat being assessed.Point counts were not carried out in rain or heavy mist.

Statistical analysis
The mean number of species detected at sample sites after one repetition was calculated and compared for each habitat category using non-parametric Kruskal-Wallis one-way ANOVA and Mann-Whitney U-tests (Zar 1999).The mean number of Sulawesi endemic and forest specialist species per site (identifi ed after Coates and Bishop 1997), were also compared using the same non-parametric analysis, as was the mean number of dedicated frugivorous and insectivorous species, as well as endemic frugivores and insectivores.These feeding guilds were chosen as they have been hypothesised to be particularly susceptible to habitat modifi cation.Diversity of forest frugivores is often linked to the richness of fruiting plant species, which are most diverse in undisturbed forest ecosystems, and birds of this feeding guild often require large foraging areas due to the spatial and temporal scarcity of yearround fruit resources; thus degradation and fragmentation of forest ecosystems impacts strongly upon these species (Gray et al. 2009, Sodhi et al. 2004b).Insectivores have also been considered to be strongly affected by habitat modifi cation due to their dependence on specifi c foraging microhabitats and possessing limited dispersal capabilities (Sodhi et al. 2004b).Feeding guilds were assigned following BirdLife International ( 2008), Coates and Bishop (1997), and White and Bruce (1987).All these comparisons were tested using a 95% confi dence interval.Sample-based rarefaction curves plotting numbers of individuals recorded against number of species detected were calculated utilising the software package EstimateS (Colwell 2006).A further three non-parametric species-richness estimators (ACE, Chao 2, and MMMeans) were calculated using 50 randomisation runs on the software package EstimateS (Colwell 2006).These estimators were chosen as they have been shown to be appropriate for extrapolating tropical bird community richness (Herzog et al. 2002).The mean value of the three estimators were taken as an estimation of species richness in each habitat class as the effectiveness of different estimators is expected to vary with different data sets (Walther and Moran 1998).This was included to corroborate the results of the accumulation curve, as information in larger samples may be lost when 'rarefi ed' to the size of the smallest sample in the analysis (Lee et al. 2007, Sodhi et al. 2005a).

/ Martin and Blackburn
Comparisons of community structure between habitat categories were examined by constructing similarity matrices using PC-ORD version 5.0 and comparing congruence using a series of Mantel tests.Signifi cance of these tests was determined by a Monte-Carlo procedure utilising 999 permutations (McCune and Grace 2002).Kruskal-Wallis one-way ANOVA analysis was also utilised to compare mean abundance of R. cassidix individuals per study site in each habitat category, along with two other selected large-bodied endemic species; the insectivorous Pied Cuckoo Shrike (Coracina bicolor), classifying as Near Threatened by the IUCN (2009), and the frugivorous Golden-mantled Racquet-tail Parrot (Prioniturus platurus).

RESULTS
A total of 67 species and 1701 individual birds were recorded in the point count surveys.Table 2 summarises species detected within each habitat category.Figure 3 demonstrates that mean species richness per study site increases signifi cantly between the primary forest and regenerating secondary forest sites, rising from a mean of 7.2 ± 1.2 to 9.3 ± 1.4 species per site (Mann-Whitney U=1138, P≤0.01).As anthropogenic disturbance levels increase further, however, richness decreases again, with a drop in mean species per site to 7.2 ± 1.1 in disturbed secondary forest.This is signifi cantly lower than that of regenerating secondary forest (Mann-Whitney U=1125, P≤0.01), although not signifi cantly different from primary forest (Mann-Whitney U=1553, P=0.93).Species richness per site in cleared agricultural land drops to 6.3 ± 1.0; signifi cantly less than all forest habitat classes (P≤0.05).Figure 4 shows similar patterns occurring within feeding guilds.The mean number of frugivorous and insectivorous species per site is not signifi cantly different between primary and disturbed secondary forest (Mann-Whitney P≥0.05 for both).Richness of both feeding guilds is slightly higher in regenerating secondary forest, although no signifi cant differences occurs between this habitat and primary forest (Mann-Whitney P≥0.05 for Tropical forest disturbance on a small island / 131  Scarlet Honeyeater 0 (0) 0 ( 0 both guilds).Cleared farmland has signifi cantly less species of both feeding guilds per site compared to all forest habitats (Mann-Whitney P≤0.01 for all tests).While mean aggregations of species per site suggest a relatively impoverished avifaunal assemblage in cleared farmland, other statistical analyses provide an alternative view.Average species richness estimates displayed in Table 3 were calculated as 31.8(95% confi dence interval) for primary forest, 37.3 for regenerating secondary forest, 39.4 for disturbed secondary forest and 54.2 for farmland.This would suggest a trend towards increasing species richness with anthropogenic disturbance.Rarefaction curves in Figure 5 support this, predicting primary forest to have the most impoverished avifauna, with farmland having the highest number of species and species density.
Regenerating secondary forest displays similar predicted overall species richness to disturbed secondary forest, which again implies that no simple direct correlation exists between avian richness and level of disturbance.
No signifi cant difference in the number of species per site occurs between forest habitats in each of these categories (Mann-Whitney P≥0.05 for all comparisons), although cleared farmland had signifi cantly less species per site than all forest classes (Mann-Whitney P≤0.01 for all comparisons).Figure 6 demonstrates that abundance of R. cassidix decreases rapidly with increasing disturbance, falling from a mean abundance of 0.57 individuals per sample in primary forest to 0.39 individuals per sample in regenerating secondary forest, and levelling off at a value of virtual absence in disturbed secondary forest and farmland.The decline of R. cassidix across the four habitat categories shares a strong positive correlation with the number of large trees per plot at each site (r 2 =0.87).Similar trends of decline are also found in other selected large-bodied Sulawesi endemics; P. platurus and C. bicolor.

DISCUSSION
While mean species richness per sample demonstrates a general pattern of decline with increasing environmental disturbance, this is not a simple progressive relationship as has been found in other studies (Sodhi et al. 2005a, Thiollay et al. 2002), as richness rises signifi cantly between primary and regenerated secondary forest before declining in the more disturbed sites.The rise in species richness between primary and regenerated secondary forest could perhaps be explained by the ecological history of the latter.These sites have been subjected to heavy disturbance in the past, but subsequently left undisturbed for over 25 years.Past disturbance would have led to major disruption of ecological niches, and while it is likely that a decline in the total diversity of avifaunal assemblages would have occurred due to local deterministic loss of specialist rainforest species (Pullin 2002), previous studies have shown that smaller numbers of generalist species are often able to colonise new niches created by disturbance that did not previously exist in the area (Sodhi et al. 2005a).Over time and with regeneration of the forest, niches may again have appeared which accommodate forest specialists, but it is possible that certain generalist species have become established and remain; hence the higher species richness.This relates to the Theory of Intermediate Disturbance (Connell 1978), and may explain why total richness is signifi cantly higher in regenerating secondary forest, but richness of endemics and forest species is not.Species richness in disturbed secondary forest is somewhat higher than could be expected from fi ndings of previous studies (Sodhi et al. 2005a), being statistically similar to that of primary forest.This, again, could possibly occur because of an overlap of opportunistic, generalist species with specialist forest species returning to the forest as it regenerates, as hypothesised for the regenerated secondary forest.However, the vegetation survey indicates that the disturbed secondary forest has not recovered so extensively, so it could be that fewer niches have become re-available for forest species, thus accounting for differences in richness between the two types of secondary forests.

Figure 5 Sample-based rarefaction curves displaying number of individuals against number of species recorded in primary forest (notated as 1), regenerating secondary forest (notated at 2), disturbed secondary forest (notated as 3) and cleared agricultural land (notated as 4) sample sites within the Lambusango forest reserve, Buton Island, Southeast Sulawesi
Although overall richness varies across study sites, both secondary forest classes have a similar number of endemic species per site to primary forest, including frugivorous and insectivorous species.These fi ndings would appear to support the conclusions of Veddeler et al. (2005) who, in their work on butterfl ies in central Sulawesi, recognised the signifi cance of secondary forests, especially in older successional stages, in the conservation of tropical biodiversity.The conservation value of regenerating forest has also been demonstrated elsewhere in Indonesia, such as the Harapan Forest in Sumatra (BirdLife International 2008).
The results indicating a low mean number of species per sample in the farmland result from habitat structure.The farmland sites were largely cleared land with no signifi cant tree growth, canopy or understorey, thus representing a heavily modifi ed ecosystem to which forest species have not adapted, therefore creating deterministic extinctions as described by Sutherland (2000) and Pullin (2002).Further, it could be argued that as the farmland consists largely of monoculture crops, plant diversity in any given area would be comparatively low in relation to the mosaic of microhabitats represented in any pristine forest ecosystem.This lack of niches would limit the number of generalist species that could exploit the habitat change, and hence species richness is low.These results are in concordance with the fi ndings of Sodhi et al. (2005a), who recorded mean avifaunal species numbers in farmed plantations as 15.67 ± 1.07 in contrast to a mean species richness of 31.99 ± 1.38 in primary forest, and of Trainor (2007), who reported comparable results.However, species estimates obtained from the non-parametric tests and the rarefaction curves predict cleared farmland to contain the highest total number of species, and primary forest the lowest.This could result from the heterogeneous nature of the cleared farmland.While the mean number of species per sampling point was low in the farmland, the high species richness predicted to occur across the entire agricultural area could result from the high spatial variability in vegetation and land-use; this would create a broad range of habitat niches on a larger spatial scale which in turn could support a more diverse avifauna than the more spatially contiguous primary forest.The high number of species predicted in the cleared farmland could also be a function of its proximity to more intact forest environments; this could result in forest specialists not typically found in agricultural land occurring as foragers based in roosts in adjacent forest habitats.Despite this proximity, the cleared farmland has a very low richness of endemics and forest species.This is in concordance with the fi ndings of Peh et al. (2005), where only 28-32% of birds found in primary forest were also found in mixed-rural habitats, with estimates of species richness in agricultural land being even lower.These fi ndings support the arguments noted by Fermon et al. (2005) richness does not imply higher species distinctiveness, and that the contribution of land-use systems to global biodiversity should be evaluated with caution, even when high species richness estimates are found.This is further demonstrated by the mantel cross-similarity community comparisons, which indicate that no signifi cant similarity exists between the compositions of farmland and forest bird assemblages.Species summaries show that community structures in these farmland sites are dominated by wide-ranging generalist species, including representatives from several families that were completely absent in forest habitats (i.e.Ardeiidae, Rallidae, Hirundinidae).This persistence of generalist species in these habitats over specialist insectivores and frugivores is concurrent with the fi ndings of Sodhi et al. (2005a), while the apparent high vulnerability of regional endemics to heavy disturbance is concomitant with studies by Trainor (2007) and Posa and Sodhi (2006).Indeed, our results would appear to suggest that the specialisation of many of these endemic species is extremely fi ne, as in the areas of heavy disturbance certain endemic species are replaced by very similar generalist species which fulfi l a comparable ecological role.
Examples of this effect can be seen in population comparisons of endemics and widespread generalist species in the same genus: the Bay Coucal (Centropus celebensis) and Lesser Coucal (Centropus bengalensis), and the Sulawesi Whiteeye (Zosterops celebense) with the Lemon-bellied Whiteeye (Zosterops consobrinorum) (Figure 7).These results are in accordance with those for the abundance of endemics overall, which suggest that while many endemic species are abundant in both primary and secondary forests, they are virtually absent from the areas of heaviest disturbance, being replaced by generalist species not found in naturally vegetated areas.This further demonstrates the vulnerability to heavy anthropogenic disturbance of Buton's endemic avifauna and highlights the critical differences between the effects of intermittent logging and shifting cultivation which permit the regeneration of secondary forest and clearance for agriculture which largely eliminates the forest habitat.Results indicating R. cassidix as highly vulnerable to even moderate environmental disturbance confl ict with fi ndings in certain other studies.Sodhi et al. (2005a), for example, found that high numbers of the species could be found even in degraded mixed-rural habitats, and Cahill (2003), while acknowledging the reliance of hornbill populations on large trees, describes R. cassidix as being fairly catholic in habitat preference.However, the negative correlation between frequency of large trees and R. cassidix abundance found in this study is very strong, and similar trends, at least in cleared farmland, have been found in the population dynamics of similar species such as Blyth's Hornbill (Rhyticeros plicatus) and Rhinoceros Hornbill (Buceros rhinoceros) (Marsden andPilgrim 2003, Anggraini et al. 2001).Similar patterns of decline have also been found in other large-bodied Sulawesi endemics.
Strong negative relationships between abundance of largebodied forest birds and increasing disturbance have been reported and discussed in numerous other studies.Suggestions for the apparent extirpation vulnerability of these species include occurrence at naturally low densities, large habitat patch requirements, low reproductive rates and increased vulnerability to hunting (Boyer 2008, Sodhi et al. 2004a, Gaston and Blackburn 1995).In the case of R. cassidix, a likely explanation may relate to deterministic extinction processes.Cahill (2003), Anggraini et al. (2000) and Coates and Bishop (1997) describe how hornbill species are dependent on large, mature fruit-bearing trees to provide perennial feeding grounds and roosts, thus as habitat disturbance reduces the availability of these large trees, so the abundance of R. cassidix declines correspondingly.Findings here accord with those from previous work in Central Sulawesi that has demonstrated the importance of trees in structuring tropical forest habitats and in providing resources (Kessler et al. 2005).

CONCLUSION
This study has demonstrated that the response of avifaunal assemblages to anthropogenic disturbance in the Lambusango reserve is subject to considerable complexity.Overall, avian species richness per study site increases between primary and secondary forest and then decreases through disturbed secondary forest to a level comparable with primary forest.Such variations may be attributable to ecological history and the relative abundances of endemics and generalists throughout the phases of initial forest disturbance, establishment and maturation of secondary forest.Further work is needed to examine such dynamics in greater detail and this will benefi t from the ongoing long-term monitoring programme in Buton.This research suggests that forest species and many specialist endemic species with a high conservation priority are highly intolerant of heavy disturbance, although many species persist in moderately disturbed forest ecosystems.Such responses have also been reported in other locations in Central Sulawesi (Sodhi et al. 2005, Thiollay andRahman 2002) as well as for species in other taxonomical groups within Wallacea; notably Herpetofauna (Gillespie et al. 2005).This suggests that secondary forest ecosystems can have an important role in supporting endemics as well as overall avian diversity, although disturbed secondary forest is shown to be ineffective in conserving some of the region's larger-bodied endemic species with lower ecological tolerance.Secondary forest in later successional stages is indicated to have a particularly high conservational value, possessing a community structure similar to that found in primary forest and supporting an equally high numbers of endemics, insectivores, and frugivores as well as sizable populations of the large-bodied species evaluated.This is encouraging, given the large areas of disturbed forest ecosystems remaining on Sulawesi (Cannon et al. 2007), and that studies in other continental tropical ecosystems often indicate these habitats as being depauperate in endemics and forest specialists (Peh et al. 2008, Barlow et al. 2007, Burgess et al. 2002, Fjeldsa 1999, Canaday 1996).These findings therefore suggest the importance of allowing adequate regeneration of disturbed areas, a conservation priority also advocated by BirdLife International's habitat restoration initiatives in Sumatra (BirdLife International 2008).However, further work is needed to determine the extent to which populations within the secondary forest are dependent upon continual recruitment from adjacent primary forest refugia, and whether critical proportions of primary and secondary forests are needed for sustainability at the landscape scale.Hence, given the biological vulnerability of islands such as Buton, which are small in size and separated from the larger land masses, strong protection of the regions remaining primary forest should also be a focus for conservation efforts if viable populations of these range-restricted endemics are to be successfully maintained in Wallacea.

Figure 1
Figure 1 The Indonesian archipelago.Inset displays location of Sulawesi and Buton island

Figure 7
Figure 7 Mean abundance per plot of: a) Coucal species Centropus celebensis and Centropus bengalensis, and b) White-eye species Zosterops consobrinorum and Zosterops chloris in primary forest, regenerating secondary forest, disturbed secondary forest and cleared agricultural land within the Lambusango forest reserve, Buton Island, Southeast Sulawesi