Determining impacts of habitat modification on diversity of tropical forest fauna: the importance of spatial scale

Authors

  • JANE K. HILL,

    Corresponding author
    1. Department of Biology, University of York, York YO10 5YW, UK
      Dr Jane K. Hill, Department of Biology (Area 3), PO Box 373, University of York, York YO10 5YW, UK (e-mail jkh6@york.ac.uk).
    Search for more papers by this author
  • KEITH C. HAMER

    1. Centre for Biodiversity and Conservation, School of Biology, University of Leeds, Leeds LS2 9JT, UK
    Search for more papers by this author

Dr Jane K. Hill, Department of Biology (Area 3), PO Box 373, University of York, York YO10 5YW, UK (e-mail jkh6@york.ac.uk).

Summary

  • 1There is an urgent need to understand the impacts of anthropogenic habitat disturbance on biodiversity in tropical forests, but no consensus has yet emerged. We reviewed the literature for the most frequently studied taxon (birds, 37 studies) and found that increased and decreased diversity in response to disturbance (selective logging and shifting agriculture) were reported with approximately equal frequency.
  • 2The spatial scale at which studies were carried out significantly affected the reported response to disturbance: studies where disturbed and undisturbed habitats were sampled at large spatial scales were more likely to report increased diversity following disturbance, whereas studies that sampled habitats at small spatial scales were more likely to report decreased diversity. These results were not a consequence of sampling method: we divided the studies into those using capture methods and those using observation methods and the same result was obtained when the analysis was restricted to only those studies using observation methods.
  • 3Previously, we have shown that reported impacts of disturbance on Lepidoptera are also affected by the spatial scale of study. We reviewed the Lepidoptera literature published since then and showed that all 12 new studies conformed to the predicted pattern.
  • 4While sampling scale significantly affected the reported responses of both birds and Lepidoptera, there were opposite effects of scale in the two taxa: large-scale bird studies and small-scale Lepidoptera studies were more likely to report increased diversity following disturbance. Bird studies were generally carried out at larger spatial scales than those of Lepidoptera and these opposite impacts of scale were probably due to a non-linear effect of habitat disturbance on habitat heterogeneity at different spatial scales.
  • 5Synthesis and applications. The rapid loss and degradation of tropical forests means that an understanding of the general patterns of responses of species to habitat disturbance is urgently needed. However, there has been little discussion of the most appropriate methods to ensure comparability of results between studies. Data presented here indicate that the spatial scale of sampling chosen in studies has a marked effect on the results obtained, and future studies need to account for this by examining explicitly the effects of disturbance at different spatial scales. The effect of spatial scale differs between taxa, and this may explain why the search for indicator taxa of disturbance effects has so far proved elusive.

Introduction

Degradation of tropical forests through selective logging and shifting agriculture is both widespread and continuing yet, despite several decades of research into this problem, the impacts of such disturbance on the diversity of fauna within tropical forests are only poorly understood. Severe disturbance (e.g. clearfelling and conversion of forest to grassland) usually reduces diversity (Holloway, Kirk-Spriggs & Chey 1992) but impacts of moderate habitat disturbance, such as commercial selective logging where some semblance of forest remains, are unclear. For example, in birds, one of the best studied taxa in tropical regions, both increased and decreased diversity following disturbance have been reported. In addition, responses by one taxon appear to bear little relationship to responses by other taxa, and so the search for indicator taxa of disturbance effects has met with little success (Lawton et al. 1998; Perfecto et al. 2003). Given the rapid loss of tropical forest habitats, ecologists and conservationists urgently need to understand the factors contributing to this lack of consensus. Such understanding could in turn lead to important new insights into the general patterns of species’ responses to habitat disturbance, and so enable reliable predictions to be made of future impacts of disturbance on biodiversity.

In temperate regions, there is some consensus on methods of sampling animal taxa to allow comparison between sites and studies (e.g. standardized transect sampling for butterflies; Pollard 1977). In contrast, there is little consensus in tropical regions, and a wide range of methodologies has been used. This lack of agreement makes it difficult to deduce any general patterns of species’ responses to habitat disturbance among different tropical studies. For instance, the biases involved in different methods for censusing forest birds are well known (Bibby, Burgess & Hill 1992; Sutherland 1996): point counts are generally more efficient in mature forest, whereas mist nets are more efficient in young forest (Whitman, Hagan & Brokaw 1997; Blake & Loiselle 2001; Pagan, Thompson & Burhans 2002). Such biases may affect reported responses of birds to disturbance, but this has not been studied. In addition, there has been little consideration of how other factors such as number and position of sampling points may influence results obtained.

Describing and explaining patterns of species diversity and understanding the consequences of diversity loss are central themes in ecology (Huston 1994). One commonly observed pattern is the relationship between species richness and size of area sampled (species–area relationship; Gaston & Blackburn 2000). Thus, spatial scale is important in estimating diversity and there is increasing evidence that the perceived importance of other ecological variables and processes also depends crucially on the spatial scale at which variables are measured (Bellehumeur & Legrandre 1998; He & Gaston 2000; Rahbek & Graves 2000; Robinson, Brawn & Robinson 2000; Lennon et al. 2001). Not only are measures of diversity per se scale-dependent, but the perceived pattern of diversity change following rainforest modification may also be dependent on the spatial scale at which studies are carried out. It has been assumed in previous studies that sampling similar sized areas in different habitats is sufficient to control for any area effects when investigating impacts of disturbance, but this may be a false assumption if habitat modification affects the relationship between area sampled and number of species recorded. Previously, we suggested for Lepidoptera that studies carried out at small spatial scales were more likely to report increased diversity following forest disturbance, whereas larger-scale studies reported decreased diversity (Hamer & Hill 2000). These effects suggest that the spatial scale chosen for sampling may influence the results obtained, but it is not clear whether these scale effects are widespread or also affect other taxa. Birds are the best studied animal taxon in tropical regions, and we analytically reviewed the published literature to investigate whether or not the spatial scale at which studies were carried out has affected the recorded responses of birds to disturbance. In addition, there have been a number of new Lepidoptera studies published recently and we investigated how many of these new studies are in agreement with predictions from our previous findings. We also compared the effects of spatial scale in studies of birds and Lepidoptera.

Methods

sources of data

We used the ISI Web of Knowledge database (http://wok.mimas.ac.uk) to search for tropical studies of birds for each year from 1981 to 2003. We then used reference lists in these studies to search for pre-1981 studies. We included only studies investigating impacts of anthropogenic forest modification (selective logging, agriculture), and we excluded studies (or parts of studies) that compared forest with non-forest habitats; we excluded data from clearfelling, farm fallow and early successional habitats, as well as young forest habitats without a canopy. We have included information in Table 1 explaining which habitats in specific studies were included in our analyses. Studies primarily investigating forest fragmentation effects were normally excluded because impacts of disturbance would be confounded by fragment size. The only exception was a study of impacts of fragmentation on birds in Brazilian Atlantic forest (Marsden, Whiffin & Galetti 2001). This study was included because there was no effect of fragment size or isolation, and so variation in species diversity could be attributed unequivocally to variation in the disturbance regime within forest fragments. Multiple published studies by the same author from a single location and sampling method were included only once (e.g. Shankar Raman 2001; Shankar Raman, Rawat & Johnsingh 1998; Johns 1986, 1989). Separate data sets from single studies using different sampling methods (Beehler, Krishna Raju & Shahid 1987) or from different geographical locations (e.g. different islands in Indonesia; Jones, Marsden & Linsley 2003) were included. Studies where disturbance effects were confounded with altitude effects were excluded (e.g. Cresswell et al. 1999), as were studies where there was insufficient methodological information for the spatial scale of the study to be determined with any precision (e.g. Thiollay 1995, 1999; Fjeldså 1999). We confined our review to studies published in peer-reviewed journals (i.e. we did not include studies in the grey literature) but, with the exception of those constraints indicated above, we imposed no other restrictions for inclusion of studies in the analyses. We assumed that the peer-review process would ensure that only those studies with adequate sampling and reliable results would be published.

Table 1.  Effects of habitat modification of tropical forests on bird species diversity. Studies are listed by increasing size of sampling area. ‘Change in diversity’ is the result reported by authors or based on our reanalysis of data published in appendices. Sampling method ‘transects’ incorporates all observational sampling methods (i.e. walk-and-count transects, point counts). Sampling scale = area over which birds were sampled and at which data were analysed (assuming a 250-m sampling radius for mist nets; see text)
Change in diversitySampling methodRegionSampling scale (ha)SourceNotes
No effectTransectsNeotropics0·2Reitsma, Parrish & McLarney (2001)Comparing number of species in forest with abandoned cacao plantations; 25 m-radius point counts, data analysed per count
DecreaseTransectsNeotropics0·25Thiollay (1992)Data analysed per 50 × 50-m observation area. Data also presented for total species richness across all samples; these data were not analysed but they indicated decreased diversity at large scales. Other studies by Thiollay (1995, 1999) were not included as there was insufficient information to calculate spatial scale
No effectTransectsNeotropics0·3Aleixo (1999)Scale estimate assumes ‘unlimited’ point counts have 30-m radius. Data analysed per count
DecreaseTransectsAsia0·3Marsden (1998)30-m radius point counts, data analysed per count
DecreaseTransectsNeotropics0·3Marsden, Whiffin & Galetti (2001)30-m radius point counts, data analysed per count. Comparing forest reserve and disturbed forest fragments
DecreaseTransectsNeotropics0·3Estrada, Coates-Estrada & Meritt (1997)Comparing undisturbed and disturbed forest sites. 30-m radius point counts, data analysed per count
DecreaseTransectsAsia0·4Bowman et al. (1990)Minimum of two 25 × 80-m transects per forest site
DecreaseTransectsAsia0·6Beehler, Krishna Raju & Shahid (1987)Data in appendix reanalysed separately for mist nets and transects; results for two methods included separately. Two 30-m radius point counts per site
DecreaseTransectsAsia0·8Shankar Raman & Sukumar (2002)Authors report decrease in species richness with disturbance, but not tested significantly. Data from 50-m radius point counts, analysed per point count
DecreaseTransectsNeotropics0·8Perfecto et al. (2003)Decrease reported at one disturbed site. Authors report results from second site are likely to be confounded by proximity to forest reserve. Data from four 25-m radius point counts
DecreaseTransectsAsia1·0Wang & Young (2003)Data from a total length of 1 km of transects (10 m width)
DecreaseTransectsAfrica1·0Lawton et al. (1998)No difference between primary and old-second growth sites, but decrease in diversity across all sites in relation to disturbance. Scale estimate based on 1-ha sample plots
IncreaseTransectsAsia1·1Jones, Marsden & Linsley (2003)Data from two independent studies included separately. Data from 60-m radius point counts, analysed per point count
No effectTransectsAsia1·1Jones, Marsden & Linsley (2003)Comparing sites where logged and unlogged habitats co-occur. Data from 60-m radius point counts, analysed per point count
DecreaseTransectsAsia5Shankar Raman, Rawat & Johnsingh (1998)Data also published in Shankar Raman (2001) but only included once. Data from 500 × 100-m transects
DecreaseTransectsAsia10Lambert (1992)Data reanalysed for transects only, no information on number of mist nets used. Minimum transect length = 1580 m
DecreaseTransectsNeotropics11Hutto (1989)Data in appendix reanalysed (combining point counts) comparing unlogged and tall secondary forest. Scale estimate based on 54 25-m radius point counts
No effectTransectsAfrica15Sekercioglu (2002)Data from appendix reanalysed comparing undisturbed and logged sites. Author reports higher richness in logged site, but not tested statistically. Data combined from 60 0·25-ha point counts
DecreaseTransectsAsia18Johns (1989)Data also published in Johns (1986) but only included once. Decrease reported 6–7 years after logging. Scale estimate assumes one 3-km transect per site
No effectNets and countsNeotropics21Whitman, Hagan & Brokaw (1998)Data combined for point counts and mist nets; scale estimate based on method giving largest sampling scale (three nets per sample, data analysed per sample)
DecreaseMist netsNeotropics22Pearman (2002)Most bird guilds declined in disturbed forest. Five nets set in a line per site
No effectNets and countsNeotropics22Terborgh & Weske (1969)Scale estimate based on five mist nets set in a continuous straight line per site
DecreaseNets and countsNeotropics24Canaday (1997)Scale estimate based on 10 10-m nets set in a continuous straight line. Decline in insectivorous birds with increasing disturbance
IncreaseMist netsAfrica24Catry et al. (1999)Scale estimate based on 72 m of mist net set in a continuous straight line per site. Minimum of one net line per site. Reanalysis of data in appendix showed significant difference between primary and three out of four second growth sites (fourth site showed no difference)
IncreaseTransectsNeotropics26Woltman (2003)Two 650-m transects per habitat; birds were recorded up to 100 m either side of transect. No formal statistical analysis, but author reports higher species richness in disturbed sites
No effectMist netsAsia27Beehler, Krishna Raju & Shahid (1987)Scale estimate based on 13 nets set in a continuous straight line per habitat
No effectTransectsAfrica28Owiunji & Plumptre (1998)Scale estimate based on 30-m radius point counts. Data summed from 100 counts per habitat
IncreaseTransectsAsia30Johns (1996)Mist net data not included by author in analysis. Scale estimate based on 5-km transects per site (author quotes length of multiple repeats of transects)
No effectsMist netsAustralia42Crome, Thomas & Moore (1996)Distance between nets varies within habitat and is often < 100 m. Scale estimate based on 36 nets (four nets from nine sites) set in a continuous straight line, but scale may be larger
No effectTransectsNeotropics43Johns (1991)Reanalysis of data in appendix comparing unlogged and logged sites. Scale estimate based on 36 km of transects split equally among five sites
IncreaseMist netsAfrica51Dranzoa (1998)Scale estimate based on two 120-m lines of mist nets per site (but scale may be larger)
IncreaseMist netsNeotropics52Blake & Loiselle (1991)Reanalysis of data in appendix comparing primary and old-second growth; 30 nets placed within a 4·8-ha area will sample over a larger area, scale estimate based on nets sampling 250 m beyond edge of 4·8-ha study area
IncreaseMist netsNeotropics60Andrade & Rubio-Torgler (1994)Authors report little difference but reanalysis of data in appendix shows significant difference between primary and old secondary growth forest. Scale estimate based on a minimum of three nets per habitat
No effectMist netsNeotropics60Mason (1996)Scale estimate based on two sites with continuous lines of 18 nets per habitat. No difference between primary and three out of four logged sites (fourth logged site showed increased diversity)
DecreaseMist netsAsia72Wong (1986)Author reports 60 individual nets set up within a 16-ha study site, but nets would sample over a larger area. Scale estimate based on 850 × 850-m sampling area
IncreaseMist netsAfrica78Owiunji (2000)Scale estimate based on method (nets) giving largest sample area. Scale estimate based on 100 mist nets set 100 m apart in each habitat (50-m radius of sampling per net)
IncreaseTransectsNeotropics100Anderson (2001)1-km2 point counts from vantage points above the canopy

We extracted and quantified the following information from published studies.

  • 1Impact of disturbance: diversity decreased, increased or unaffected following disturbance.
  • 2Sampling technique: capture, i.e. mist net, or observation, i.e. transects and point counts.
  • 3Type of disturbance: selective logging or agriculture.
  • 4Status of least disturbed habitat surveyed: primary/near-primary forest vs. secondary or selectively logged. This variable controlled for the possibility that reported responses to disturbance depended on the initial state of the study site before the disturbance event. For instance, according to the intermediate disturbance hypothesis (Connell 1978; Molino & Sabatier 2001), undisturbed sites might be more likely to showed increased diversity following disturbance whereas previously disturbed sites might be more likely to show decreased diversity following further disturbance.
  • 5Elevation of study site: lowland, i.e. < 1000 m a.s.l., or montane, i.e. = 1000 m. This controlled for the possibility that responses to disturbance could be a function of initial species richness, which tends to be higher at lower altitudes (May 1972; Gaston & Blackburn 2000);
  • 6Geographical location: south-east Asia, Africa or neotropics. This controlled for the possibility that regions that experience high levels of natural disturbance (such as neotropical hurricane zones) might be expected to be more resistant to anthropogenic disturbance.
  • 7Time span of the study: short-term, < 7 months, vs. long-term, ≥ 7 months, based on the frequency distribution of studies with different time spans. This variable was an index of sampling effort and investigated whether shorter-term studies might have produced less reliable conclusions and so contributed to the lack of consensus in findings.
  • 8Spatial scale at which the study was carried out and at which data were analysed: area of transects/point counts, or sampling area of traps; see below.

It was not possible to include a variable quantifying the degree of habitat disturbance, because many studies did not measure vegetation changes or give information on timber extraction rates (for recording levels of disturbance resulting from selective logging see recommendations in Greiser Johns 1997). However, we restricted our analysis to those studies, or parts of studies, that were within forest (see above) and so all studies were investigating impacts of moderate habitat disturbance. It was not possible to include a variable quantifying the spatial extent of studies (to control for possible confounding effects of β diversity prior to disturbance; Jones, Marsden & Linsley 2003), although, in practice, most studies compared habitats that were in fairly close proximity to each other, and so any such impacts of β diversity were unlikely to have been great.

We analysed data using binary logistic regression. We chose this method, rather than more formal meta-analysis, because studies presented data on impacts of disturbance using a wide variety of statistical techniques and diversity indices and so we conservatively used qualitative categories (diversity increased, decreased or no effect on diversity) to characterize impacts of disturbance. In those cases where data were published in appendices but diversity had not been calculated and/or tested statistically, we reanalysed data (using Species Diversity and Richness software; Pisces Conservation Ltd, Lymington, Hants, UK) to determine whether or not there was a significant impact of disturbance on diversity. Regression models were built by forwards stepwise selection of significant variables, and model reliability was then checked by backwards stepwise deletion of non-significant variables. All eight variables listed above were categorical except sampling area, which was square-root transformed for analysis. We carried out separate regression analyses coding studies reporting no effect of disturbance on diversity in both the increased diversity category (i.e. increased diversity or no effect vs. decreased diversity) and in the decreased diversity category (i.e. decreased diversity or no effect vs. increased diversity). Coding of studies reporting no effect and use of either forwards or backwards stepwise regression did not qualitatively affect our results, and we quote results for models using forwards stepwise regression where no effect cases were coded with studies reporting increased diversity following disturbance.

measurement of spatial scale

For transects and point counts, spatial scale was estimated from the maximum distance to the observer within which birds were recorded (generally 30–50 m). In the few cases where this distance was not stated (Beehler, Krishna Raju & Shahid 1987; Aleixo 1999), transect width was assumed to be 60 m (30 m either side of path) and a 30-m radius was assumed for point counts, because many bird species in tropical forests are difficult to detect at distances > 30 m from the observer and calls made by many species attenuate beyond this distance (Owiunji & Plumptre 1998; Shankar Raman & Sukumar 2002). For those studies that had unequal sampling effort in different habitat types, or where sampling effort changed during the study, we calculated spatial scale values based on the site with the smallest sampling effort. If studies combined data from both point counts and mist nets in a single analysis, we calculated spatial scale values for the sampling method resulting in the largest scale estimate. This was always the spatial scale of the mist nets as the spatial scale of transects/point counts was entirely subsumed within this value. If studies combined data from several sample points prior to analysis, then we calculated spatial scale as the sum of the sampling areas of all the sample points. For example, a study sampling birds at 80 point counts in primary forest and 50 point counts in logged forest would have a scale measure of 0·28 ha (= 302 × π/10 000) if data were analysed per point count, or 14 ha (= 0·28 × 50) if data from all point counts in a habitat were combined for analysis.

There was little information on the size of area over which traps sampled individuals, and so we used two methods to estimate the sampling area of traps. First, we assumed that mist nets (each net usually being 12 m long) sampled birds over a 250-m radius, i.e. the sample area of a single mist net = 20 ha. This was supported by studies that indicated, using semi-variograms, that mist nets > 200 m apart were statistically independent (Whitman, Hagan & Brokaw 1998) and by data indicating that the composition of bird species sampled in mist nets is related to vegetation within a 200–600-m radius (Pearman 2002). Thus, we estimated that a study using 15 nets in a single line would sample an area of 28 ha. This is because a single mist net operated on its own has a circular sampling area of radius 250 m, and 15 nets joined in a line would stretch this circle out into a rectangle of width 500 m and length 168 m (14 × 12 m), with convex ends each of 250 m radius. In contrast, another study that combined data from three mist nets set up > 500 m apart would sample 60 ha.

Although our estimated sampling area of 20 ha has some statistical support, many understorey birds in tropical forests have territories smaller than this (e.g. median of 111 neotropical species = 9 ha; Terborgh et al. 1990), which suggests that 20 ha may be an overestimate of sampling area for birds holding territories. Accordingly, our second method assumed that mist nets sample over a radius of 125 m, giving a sampling area of 5 ha for a single net, which is less than the territory sizes of most species. Using this second method, we estimated that a study using 15 nets in a line would sample 9 ha, and a study combining data from three separate mist nets would sample 15 ha.

In a few cases our estimate of sampling area was larger than that quoted by the authors. For example, Blake & Loiselle (2001) placed 30 nets over a 4·8-ha grid, giving a suggested sampling area of 4·8 ha. However, birds can enter nets from either side (i.e. not only from within the area enclosed by the grid) and so in practice we calculated that the nets would have sampled over a larger area (52 ha assuming a 250-m sampling radius for mist nets, or 22 ha assuming a 125-m radius; the latter was approximately the area of disturbed habitat in the study). Many bird species do not cross obvious habitat boundaries (e.g. forest to pasture) but the disturbed area sampled by Blake & Loiselle (2001) was embedded within a forest matrix where the boundary between primary and secondary forest was unlikely to have been great and so was probably permeable to many species. Similarly, Wong (1986) set up a total of 60 nets in a 16-ha plot but we estimated that the nets would have sampled over a larger area of forest (72 ha assuming a 250-m sampling radius for mist nets; Table 1). Across all studies, estimated sampling areas for mist nets were generally larger than those for point counts or transects, corresponding with the much longer time for which mist nets were operated (typically > 10 h for mist nets, < 0·5 h for point counts). Notes in Table 1 give extra information on how we determined impacts of disturbance or calculated spatial scale where this was not straightforward.

lepidoptera studies

Our previous analysis (Hamer & Hill 2000) suggested that studies at small scales (< 1 ha) would report increased diversity in disturbed forest whereas larger scale studies (> 3 ha) would report decreased diversity following disturbance. To test this prediction, we used Web of Knowledge to search for studies of butterflies or moths published since our previous study. The methods used to calculate spatial scale were the same as for birds except that we assumed the sample radius of a point count was 10 m and that transect width was 10 m, unless otherwise stated. As in our previous study, we assumed that traps sampled individuals over a radius of at least 100 m (Chey, Holloway & Speight 1997), giving a minimum catchment area of 3·1 ha for a single trap.

Results

birds

We extracted data suitable for analysis from 37 data sets; nine studies reported increased diversity, 17 reported decreased diversity and 11 reported no change in diversity following disturbance (Table 1). The range of sampling scales used in studies ranged from 0·2 ha (Reitsma, Parrish & McLarney 2001) to 100 ha (Anderson 2001). These data revealed that reported responses to disturbance were strongly related to the spatial scale at which the studies were carried out; studies at small spatial scales were more likely to report decreased diversity following disturbance whereas large-scale studies were more likely to report increased diversity (Table 1; logistic regression model described 78% of cases correctly; model χ2 = 11·1, 1 d.f., P < 0·001; no other variables significant, P > 0·2 in all cases). This result was qualitatively the same if mist nets were assumed to sample over a radius of 125 m rather than 250 m as previously assumed (logistic regression model described 70% of cases correctly; model χ2 = 9·0, 1 d.f., P = 0·003; no other variables significant, P > 0·1 in all cases).

The sampling method did not significantly affect the reported response to disturbance (logistic regression model, P > 0·9) but inspection of Table 1 shows that most studies using mist nets were at a large spatial scale. Given that the precise distance over which nets sample birds is poorly understood, we checked the reliability of our findings by analysing studies using observation methods (point counts and transects; 24 studies; spatial scale range 0·2–100 ha) separately from those using capture methods (mist nets; 13 studies; spatial scale range 21–78 ha). For observational studies, these data confirmed our previous findings that the likelihood of observing increased diversity following disturbance was strongly positively related to the spatial scale of sampling (logistic regression model described 71% of cases correctly; model χ2 = 6·7, P= 0·01; no other variables significant, P > 0·3 in all cases). However, for mist net studies none of the variables was significantly related to response to disturbance (P > 0·1 in all cases, regardless of whether we assumed mist nets sampled over a radius of 125 m or 250 m). There was no evidence that changes in efficiency of sampling methods among habitats contributed to our findings, as there was a similar range of disturbance impacts in capture and observational studies (Table 1).

lepidoptera

Twelve new Lepidoptera studies suitable for analysis had been published since Hamer & Hill (2000; Table 2); four of these studies reported increased diversity following disturbance, five reported decreased diversity and three reported no effect. Reported results were in agreement with predictions from previous findings; all studies reporting increased diversity were at a small spatial scale and all studies reporting decreased diversity were at a large scale.

Table 2.  Effects of habitat modification of tropical forests on species diversity of Lepidoptera. Designation of small (< 1 ha) and large (= 3·1 ha) sampling scales is based on criteria used by Hamer & Hill (2000). Studies cited below were published after the review by Hamer & Hill (2000) and are listed by increasing size of sampling area (range = 0·1–40 ha)
Change in diversitySampling methodRegionSampling scale (ha)SourceNotes
IncreaseTransectsAsiaSmall (0·1)Walpole & Sheldon (1998)Four 250 × 6-m transects per habitat; data analysed per transect
IncreaseTransectsAsiaSmall (0·3)Willott et al. (2000)Authors report increased diversity in logged site, but effect was marginal (P = 0·059). Results from this study are incorrectly reported by Ghazoul (2002)
IncreaseTransectsAsiaSmall (0·5)Vu & Decheng (2003)Comparing sites at high elevation. Low-elevation sites had little forest. Scale estimate based on one 10 × 500-m transect per habitat
IncreaseTransectAsiaSmall (0·9)Cleary (2003)Scale estimate based on 300 × 30-m transects; data analysed per transect. Data also analysed by author at large scale (combining data from 11 to 18 transects per habitat) and showed little difference between sites
DecreaseFruit trapsAsiaLarge (3·1)Beck & Schulze (2000)Scale estimate based on one trap per habitat
DecreaseLight trapsAsiaLarge (3·1)Beck et al. (2002)Scale estimate based on one trap per habitat; canopy traps (placed directly above ground traps) not included in estimate and so scale may be larger
No effectTraps and netsAfricaLarge (3·1)Stork et al. (2003)Data included in Lawton et al. (1998). This study has been reclassified from Hamer & Hill (2000): a statistical analysis indicates no differences among sites and details of fruit traps are included. Scale estimate based on method (trap) sampling over largest scale; one trap per habitat
No effectFruit trapsNeotropicsLarge (6·3)Lewis (2001)Scale estimate based on eight traps per habitat, each with a sampling radius of 50 m
DecreaseLight trapsAsiaLarge (9·3)Willott (1999)Scale estimate based on three traps in logged site (five traps were operated in unlogged forest). Data include understorey and canopy samples from different locations
DecreaseFruit trapsNeotropicsLarge (12·4)Perfecto et al. (2003)Scale estimate based on four traps per habitat; canopy traps (placed directly above ground traps) not included in estimate and so scale may be larger
No effectFruit trapsAsiaLarge (31)Hamer et al. (2003)Scale estimate based on 40 traps per habitat. Traps placed 100 m apart so each trap had a 50-m sampling radius
DecreaseTransectAsiaLarge (40)Ghazoul (2002)Scale estimate based on 20 500 × 40-m transects per habitat

comparison between taxa

There were opposite effects of spatial scale between taxa. For birds, relatively small-scale studies were more likely to report a decrease in diversity following disturbance than were large-scale studies. For Lepidoptera, small-scale studies were more likely to report an increase in diversity following disturbance than were larger scale studies. This difference between taxa corresponded with a difference in the spatial scale at which studies were carried out; the bird studies sampled over a significantly larger area than the Lepidoptera studies [birds, mean area = 24·0 ha, n= 37, SD = 27·5; Lepidoptera (including studies in Hamer & Hill 2000), mean = 7·4 ha, n= 25, SD = 11·0; t-test (unequal variance estimate) comparing scale estimates for birds and Lepidoptera, t59·3 = 3·02, P= 0·004].

Discussion

impacts of spatial scale

Our study showed that reported impacts of habitat modification on avian diversity are heavily scale-dependent: data collected and analysed at a scale of less than c. 25 ha (assuming a 250-m radius of sampling for mist nets) tended strongly to record lower diversity in disturbed forest, whereas larger scale studies tended strongly to record the opposite effect (Table 1). Excluding studies that reported no effects of disturbance on diversity, only two out of 23 bird studies (Wong 1986; Jones, Marsden & Linsley 2003) did not fit this pattern. Most large-scale bird studies use mist net sampling, indicating that scale effects might be confounded with method. However, impacts of scale on diversity were also evident when the analysis was restricted to only those studies using observational methods, confirming the strong effect of scale on reported response to disturbance. The effect was not observed when only studies using mist nets were analysed, but these studies spanned a relatively small range of spatial scales (21–78 ha) compared with observational studies (0·2–100 ha; Table 1) and so this analysis may have lacked statistical power. Across all studies, our analyses were little affected by changes in our method of estimating the sampling area of mist nets, and so we are confident that our results did not depend on the precise spatial scale estimates for different studies.

Of the 10 bird studies that analysed impacts of disturbance at a small spatial scale (< 1 ha), five studies, all of which recorded lower diversity in disturbed forest, also presented summary data combining samples within each habitat, allowing effects of disturbance on diversity to be assessed at a larger scale. In four of these cases (Thiollay 1992; Estrada, Coates-Estrada & Meritt 1997; Aleixo 1999; Reitsma, Parrish & McLarney 2001) disturbance also reduced diversity at the larger scale. With one exception (Thiollay 1992), the larger scale in these studies was < 24 ha and so would still be expected to record lower diversity in disturbed forest (Table 1). In the remaining study (Marsden, Whiffin & Galetti 2001), where combined data covered a relatively large spatial scale (c. 60 ha), species richness was 50% higher in disturbed forest (20 species and 29·5 species per 100 point counts in undisturbed and disturbed forest, respectively; table 2 in Marsden, Whiffin & Galetti 2001), supporting our analysis. The balance of evidence from comparisons within individual bird studies is therefore consistent with the pattern derived from comparison between studies (Table 1). This was also the case with butterfly studies. For example, Cleary (2003) analysed the same butterfly data set at large and small spatial scales, and showed an increase in diversity following disturbance at small spatial scales but little difference at large scales, in agreement with the pattern observed from comparison among studies (Table 2; Hamer & Hill 2000).

impacts of disturbance on relationships between αand βdiversity

It is well known from studies of species–area relationships that measures of diversity are dependent on the area sampled (Gaston & Blackburn 2000) but the question of how habitat disturbance affects species–area relationships has not been properly addressed. Previous work has indicated that the relationship between rainforest butterfly diversity and size of area sampled is significantly affected by habitat disturbance (Hamer & Hill 2000) and the results of the current study suggest that this is also the case for birds. Future studies need to account for this effect by explicitly examining the impacts of disturbance at a number of different spatial scales.

Changes in species–area relationships in relation to disturbance reflect changes in the pattern and scale of habitat heterogeneity within tropical forests, which alter the relationship between point diversity at each sampling location (α diversity) and species turnover between locations (β diversity; Condit et al. 2002). Disturbance is known to create novel opportunities for new species not found in undisturbed forest (Connell 1978) and, by opening up areas of closed-canopy dense forest as well as creating large artificial gaps that are quickly colonized by pioneer tree species, disturbance also results in more homogeneous vegetation structure compared with undisturbed forest (Hill et al. 1995; Hamer et al. 2003). Thus, compared with undisturbed habitats, studies in disturbed habitats would be expected to report increased diversity at small spatial scales (< 1 ha) but to report decreased diversity at larger spatial scales, due to lower β diversity in more homogeneous disturbed forest. This is consistent with the pattern of results for Lepidoptera (Hamer & Hill 2000; Table 2). Studies of birds also indicated decreased diversity in response to disturbance at intermediate (1–25 ha) spatial scales, but did not record higher diversity in disturbed forest at the smallest spatial scales (< 1 ha). This may have been because the greater mobility of birds resulted in these studies effectively sampling over a larger area than similarly small-scale studies of Lepidoptera. If this is the case, then even studies at similar spatial scales may not be directly comparable between taxa.

Whereas disturbance reduces habitat heterogeneity at small to intermediate spatial scales, at larger spatial scales disturbance increases habitat heterogeneity by creating a mosaic of disturbance intensities ranging from areas of severe disturbance (e.g. along logging roads, around timber collection points or in recently abandoned crop fields) to areas of minimal disturbance and patches of undisturbed forest (e.g. around river catchments and on steep slopes; Whitmore 1990; Hill 1999). Thus studies at large spatial scales would be expected to report increased diversity following disturbance, due to elevated β diversity across disturbed sites. This is consistent with the pattern of results for birds, which generally were sampled over larger scales than Lepidoptera. The reason this pattern of increased diversity at large spatial scales was not observed in Lepidoptera was probably because no butterfly studies were at a sufficiently large scale (none were > 40 ha; Table 2). Overall, our findings suggest a non-linear pattern of diversity change in relation to spatial scale. Further data are needed, however, to confirm how habitat heterogeneity and β diversity change with spatial scale in different landscapes.

implications for conservation

Measures of diversity are important tools for assessing community-level responses to disturbance and are commonly used in highly diverse ecosystems such as tropical forests, where few species are sufficiently abundant for autecological studies. Diversity measures, however, give little information on changes in the composition of species assemblages following disturbance. There is increasing evidence that species’ ecologies affect their sensitivity to anthropogenic environmental change (McKinney & Lockwood 1999; Hamer et al. 2003). For example, in butterflies sedentary habitat specialists are particularly vulnerable to anthropogenic impacts (Warren et al. 2001), leading to the prediction that habitats in the future are likely to be dominated by widespread generalist species while specialists will continue to decline. In tropical regions, restricted-range species (which have the highest conservation value; Vane-Wright, Humphries & Williams 1991) are often shown to be vulnerable to moderate habitat disturbance, even in studies that report increased diversity after disturbance (Hamer et al. 1997). This indicates that diversity measures on their own may not be sufficient indices of the conservation value of communities.

Understanding relationships between local species richness and species turnover is crucial for interpreting impacts of disturbance on diversity. Our results suggest that these relationships are taxon-specific, as well as scale-dependent, and so it is perhaps not surprising that studies sampling a range of taxa at a single spatial scale show little similarity in patterns of response to disturbance (Lawton et al. 1998; Perfecto et al. 2003) and that there has been little success to date in finding indicator taxa of disturbance effects (McGeoch 1998).

Ancillary