1. Spatial distribution, abundance and habitat requirements of Ragadia makuta (Satyrinae) were studied in Sabah (Borneo) in 1997, in unlogged forest and forest that had been selectively logged 8–9 years ago.
2. Measurement of vegetation structure showed that unlogged forest had significantly larger trees and greater canopy cover than logged forest. A principal components analysis extracted two factors related to forest density and tree size that were also significantly higher in unlogged forest. However, there was significant spatial heterogeneity in vegetation structure within logged forest.
3. In undisturbed forest, a logistic regression model identified suitable habitat for R. makuta as areas of less dense forest close to streams. There were no differences between logged and unlogged habitats in spatial distribution and abundance of R. makuta. Availability of suitable habitat and habitat requirements of butterflies also did not differ between habitats. There was, however, significant heterogeneity in butterfly abundance within logged forest, corresponding with availability of suitable habitat.
4. Fieldwork in 1997 coincided with a severe drought on Borneo, and butterfly spatial distribution and abundance were significantly reduced compared with a year of more normal rainfall (1996); populations in 1997 contracted to areas around streams and to areas with high cover of host-plant.
5. Selectively logged areas can be highly heterogeneous in relation to levels of disturbance. Quantifying the effects of selective logging on forest structure and the availability of suitable habitat was crucial to understanding the responses of R. makuta to habitat disturbance.
Throughout South East Asia, forests are rapidly being logged, and in Malaysia most remaining rain forest is reserved for production and subjected to selective logging on a 35-year cycle (Whitmore 1984, 1991). The Malaysian State of Sabah (Borneo) was originally entirely covered by rain forest, of which 36 000 km2 (50%) was estimated to be remaining in 1985 (Collins, Sayer & Whitmore 1991). Approximately 15% of this forest is under some form of protection, but many protected areas have already been selectively logged, and there will be increasing pressure on remaining areas of forest as timber resources run out (Collins, Sayer & Whitmore 1991). The consequences of selective logging for species and ecosystems are therefore of great current concern. Clear felling and conversion of forest to plantations and agriculture generally results in decreased insect diversity (Holloway, Kirk-Spriggs & Chey 1992), but effects of less severe disturbance are less clear (Wolda 1983; Morse, Stork & Lawton 1988; Barlow & Woiwod 1989; Eggleton et al. 1995, 1996).
Tropical butterfly assemblages are particularly diverse, with many endemic species, most of which are dependent to some extent on forest (Collins & Morris 1985; Sutton & Collins 1991). Data on the effects of selective logging and other forms of forest disturbance on butterfly assemblages (species richness and relative abundance) are accumulating (Bowman et al. 1990; Raguso & Llorente-Bousquets 1990; Kremen 1992; Spitzer et al. 1993, 1997; Hill et al. 1995; Hamer et al. 1997; Lawton et al. 1998), and both increased and decreased butterfly diversity have been reported in response to disturbance. A more consistent response to disturbance, however, is the loss of species with more restricted distributions, and therefore high conservation value, from disturbed areas (Hill et al. 1995; Hamer et al. 1997), probably because these species have more restricted habitat requirements (Thomas 1992). However, these studies have assumed that the presence of adult butterflies in an area reflects local availability of butterfly resources, although availability of these resources was not quantified, and the relationship between the presence of butterflies and their resources, in particular their larval host-plants, has not been tested. Moreover, it has been assumed that butterfly habitat requirements do not vary in relation to disturbance, and the effects of disturbance on resources and habitat requirements have not been considered, making it difficult to interpret changes in species’ distributions following forest disturbance.
Despite being highly diverse, tropical butterfly communities typically contain few species present in large enough numbers for the autecological studies necessary for investigating resource requirements; most common species are usually either widespread generalist species of low conservation value, or migrants (Hill et al. 1995). An exception comprises species of the satyrid genus Ragadia, which are often relatively common and are dependent on closed-canopy forest (Corbet & Pendlebury 1992; Spitzer et al. 1993). This study investigated the habitat requirements of Ragadia makuta Fruh. Its distribution is confined to Sundaland (Borneo, Sumatra, Java and peninsular Malaysia; Corbet & Pendlebury 1992) and it is reported to be a poor flier (Corbet & Pendlebury 1992), suggesting particular vulnerability to forest disturbance. In addition, both adult butterflies and larval food plants (Selaginella spp.) can be recorded reliably from ground-based surveys, making them excellent model organisms for this type of study. Transect techniques were used to investigate butterfly spatial distribution and abundance and to quantify habitat requirements in undisturbed forest habitats. Changes in availability of suitable habitat, butterfly spatial distribution, and butterfly habitat requirements were studied in areas that had been selectively logged 8–9 years previously. Field work coincided with one of the most serious droughts in recent years on Borneo (coinciding with a marked El Niño event), and although this drought was severe, such droughts are by no means uncommon in this region (Walsh 1996). Many satyrid butterfly species are known to be sensitive to changes in humidity (Braby 1995), and butterfly distributions and abundances during the drought were compared with data from a year of more normal rainfall (1996).
Materials and methods
Fieldwork took place at Danum Valley Field Centre, adjacent to the Danum Valley Conservation Area and the Ulu Segama Forest Reserve, Sabah (5°N, 117° 50′ E), between 22 August and 29 September 1997. The study area is within lowland evergreen rain forest, dominated by dipterocarp tree species (Marsh & Greer 1992), and average temperatures (26·7 °C annual mean) and rainfall (2822 mm year–1) are typical of the moist tropics (Marsh & Greer 1992), with little monthly variation (Heydon & Bulloh 1997). The Danum Valley Conservation Area covers approximately 428 km2 of protected, unlogged forest (Collins, Sayer & Whitmore 1991), and is surrounded by extensive areas of production forest, most of which have been selectively logged. During the 1980s, logging methods in the study area followed a modified uniform system (Whitmore 1984) in which all commercial stems > 0·6 m diameter were removed using high lead cable and tractor extraction methods. Sabah timber yields are generally high (averaging 118 m3 ha–1; Marsh & Greer 1992), and although only about 7% of trees are taken for commercial purposes, associated damage can be severe, with between 60% and 80% of remaining trees destroyed (Lambert 1992).
Transects were established along existing paths and trails in two areas of forest: undisturbed forest (transects 1 and 2, total length 4·3 km) and forest that had been selectively logged 8–9 years previously, in 1988 and 1989 (transects 3 and 4, total length 4·2 km; Fig. 1). Observation stations were marked at 100-m intervals along each transect (81 stations in total). To allow ordination of differences in disturbance and vegetation structure in the two areas, the following data were recorded within a 30-m radius of every station: number, circumference at breast height, and distance from the station of the 10 nearest trees (excluding trees with circumference less than 0·6 m at breast height), and estimated vegetation cover (%) at ground, low (2 m above ground), understorey and canopy levels. In addition, estimated cover (%) of larval host-plant (Selaginella spp.) within 10 m of every station and distance (m) to the nearest stream were also measured at every station. These measurements were used to calculate nine variables (Table 1) which were normalized where necessary (including arc-sine transformation of percentages) and analysed by a principal components analysis (PCA), nested anova and logistic regression.
Table 1. Mean scores for nine variables relating to vegetation structure and distribution of larval host-plants, and mean scores for three main factors from PCA. Means followed by * are significantly different at the 5% level (anova nesting transect within habitat)
Ground cover (%)
Low-level cover (%)
Understorey cover (%)
Canopy cover (%)
Factor 1 (forest density)
Factor 2 (tree size)
Selaginella cover (%)
Distance to stream (m)
Ragadia makuta was surveyed along transects using methods similar to those described for butterflies in temperate regions by Pollard (1977) and used in previous studies (Hill et al. 1995; Hamer et al. 1997). All adults of R. makuta observed during a 5-min period within a 10-m radius of stations were recorded, and butterflies seen within 5 m of the path were also recorded whilst the observer was walking between stations. Surveys were only carried out between 10.00 and 14.00 h, corresponding with peak flight activity, and only during good weather. Although R. makuta is one of the most common butterfly species locally, it is nonetheless recorded relatively rarely and, to avoid possible errors in estimates of presence/absence and relative abundance, each survey was repeated four times and data were combined for analysis. Total numbers of butterflies recorded at each station also included butterflies seen from paths, up to 50 m either side of stations. Data from transects were used to investigate butterfly habitat requirements and changes in butterfly spatial distribution and abundance resulting from selective logging.
Habitat requirements of R. makuta were investigated in undisturbed forest using logistic regression, by relating presence/absence of butterflies at each station to five habitat variables: three factor scores from the PCA, percentage cover of Selaginella and distance to nearest stream. Variables were entered into the model by forward stepwise selection, with the significance level of inclusion/removal of variables set at 5%.
The effects of selective logging on habitat requirements of R. makuta were investigated by comparing its observed distribution within logged forest with that predicted on the basis of habitat availability. This was to investigate whether any changes in abundance in different habitats were due to differences in availability of suitable habitat, or to butterflies altering their requirements in different habitats.
Effects of drought
Data on the distribution of R. makuta were also collected on transects in September 1996, using the same protocol as that described above, at 65 of the 81 stations in logged and unlogged forest. These data were compared with data collected in 1997, to investigate the effects of drought on butterfly distribution and abundance. Because data had been collected from fewer stations in 1996, and because no habitat variables were measured in 1996, habitat requirements of R. makuta were determined using data from 1997 only. Data for mean monthly rainfall from March 1996 to August 1997 were obtained from a meteorological station at the Danum Valley Field Centre.
Table 1 shows the mean values for variables relating to vegetation structure, distance to nearest stream and cover of larval host-plants in the two forest areas. Unlogged forest had significantly larger trees (as measured by girth) and greater canopy cover (anova nesting transect within habitat). Seven vegetation variables were analysed by a PCA. Distance to the nearest stream was not included as a variable in the PCA because it was not a direct measure of vegetation structure. Percentage cover of Selaginella was not closely correlated with any of the other vegetation variables and so was also not included in the PCA (Norusis 1990). PCA extracted three components of variation from seven vegetation variables which accounted for 74·9% of the variability in the data set. Factor 1 accounted for 38% of the variance and increased with (in order from greatest to least important) increasing canopy and understorey cover, decreasing cover at ground level and increasing number and density of trees. A high factor 1 score thus represented dense forest with a closed canopy. Factor 2 accounted for a further 22% of the variability of the data set and increased with increasing girth of trees and increasing canopy cover. Factor 1 therefore primarily measured density of forest whereas factor 2 measured tree size. Factor 1 and factor 2 scores were significantly higher in unlogged forest than logged forest (anova nesting transect within habitat; Table 1). Factor 3 accounted for a further 15% of the variance in the data set but did not differ between areas.
Results from anovas showed significant differences in vegetation measures within transects after accounting for habitat effects. There were no differences in vegetation structure between transects in unlogged forest (P > 0·1 in all cases), but in logged forest transect 3 was in forest with significantly larger but fewer trees, and was closer to streams, compared with transect 4 (Table 2). In addition, factor 2 from the PCA, which is also a measure of tree size (see above), was also higher on transect 3 compared with transect 4.
Table 2. Mean scores for nine variables relating to vegetation structure and distribution of larval host-plants, and mean scores for three main factors from PCA from two transects in logged forest. Means followed by * are significantly different at the 5% level (t-test following nested anova)
Transect 3 (logged 1989)
Transect 4 (logged 1988)
Ground cover (%)
Low-level cover (%)
Understorey cover (%)
Canopy cover (%)
Factor 1 (forest density)
Factor 2 (tree size)
Selaginella cover (%)
Distance to stream (m)
Distribution and abundance of r. makuta
Sixty R. makuta adults were seen at 32 of the 81 stations surveyed. There was no difference between habitats in the number of stations where butterflies occurred (chi-square = 1·00, 1 d.f., P = 0·32) or mean number of butterflies per station (anova nesting transect within habitat, F1,77 = 1·25, P = 0·3). However, there was a significant difference between transects within habitats (F2,77 = 4·81, P = 0·011) and butterflies occurred most frequently on transect 3 in logged forest (chi-square = 9·15, 3 d.f., P = 0·027) but least frequently on transect 4, also in logged forest, with transects in unlogged forest having intermediate values.
Effects of logging on habitat requirements of r. makuta
Habitat requirements of R. makuta in unlogged forest (n = 41 stations; 14 occupied vs. 27 unoccupied stations) were described from transect data using logistic regression. Variables included in the analysis were three factor scores from the PCA (see above), percentage cover of Selaginella, and distance to nearest stream. The model predicted 82·9% of presence/absences correctly (chi-square = 19·22, 2 d.f., P < 0·001). Ragadia makuta was more likely to occur at stations close to streams in less dense forest, and the probability (p) of R. makuta presence was described by the following equation:
The predicted probability (P) of R. makuta being present at each station in logged and unlogged habitats was then calculated as:
P = 1/1 + e–z(eqn 2)
where z = habitat requirements of R. makuta in unlogged forest (equation 1).
Using equation 2, the predicted probability of R. makuta being present did not differ significantly between logged and unlogged areas (anova nesting transect within habitat, F1,77 = 1·19, P = 0·3) but differed between transects (F2,77 = 4·85, P = 0·01) and was significantly higher on transect 3 in logged forest than elsewhere (Table 3). In order to investigate whether habitat requirements were different in logged forest compared with unlogged forest, the predicted probability of R. makuta being present was set at P > 0·5 (there were only six stations with probability values ± 0·05 of this cut-off point), and the probability of R. makuta being present/absent was compared with recorded presence/absence data from transects. There was no difference between habitats in predicted and recorded presence/absence of R. makuta (chi-square = 2·59, 1 d.f., P = 0·11) but there was a difference among transects (Table 3; chi-square = 8·51, 3 d.f., P = 0·037). Transect 3 in logged forest had a greater number of stations where the predicted presence of R. makuta did not agree with recorded presence (‘wrong’ in Table 3). These instances (11 stations) included both stations where the butterfly was predicted to occur but did not (four stations), and stations where the butterfly was recorded but was not predicted to occur (seven stations). Differences between observed and predicted occurrences on transect 3 indicate changes in habitat requirements, and so differences between these 11 ‘wrong’ stations and 13 ‘correct’ stations were investigated in more detail. Stations where the butterfly was present but predicted absent had significantly higher probabilities of occurrence (mean probability = 0·34) than stations where the butterfly was both predicted and recorded absent (mean probability = 0·22; t = –3·09, 10 d.f., P = 0·012). In contrast, there was no significant difference in probability values between stations where the butterfly was predicted and recorded present, and stations where it was predicted to occur but was absent (mean probability = 0·81 and 0·78, respectively).
Table 3. Differences among four transects in unlogged and logged forest in predicted and recorded presence/absence of R. makuta at 81 observation stations. P is mean probability of occurrence. Means followed by the same letter are not significantly different at the 5% level. ‘Correct’ is the number of stations where R. makuta was both recorded and predicted present (using P > 0·5 as criterion for occurrence, see text), and also stations where it was not observed and not predicted to occur. ‘Wrong’ are cases when either R. makuta was present but not predicted to occur, or when it was absent but predicted to occur
Effects of drought
During the 6 months from March to August 1997, total rainfall at Danum was approximately 27% less and there were approximately 28% fewer raindays compared with the same period in 1996 (Table 4). Corresponding with changes in rainfall, R. makuta was nearly three times more abundant in 1996 (total number recorded in 1996 = 71; total in 1997 = 26) and was more widely distributed, being present at 26 stations (40%) in 1996, compared with only 14 stations (17%) in 1997. Stations where R. makuta occurred were also more isolated in 1997 (mean distance between occupied stations = 313·3 m, SD = 292·5) compared with 1996 (mean = 142·3 m, SD = 98·7; t-test with unequal variance, t = –2·19, 15·86 d.f., P = 0·043). The pattern of presence/absence of R. makuta along the transect was not significantly different from a random distribution in 1997 (runs test, z = –0·38, P = 0·7), but occupied stations had a significantly clumped distribution in 1996 (z = –2·40, P = 0·017).
Table 4. Rainfall data from Danum Valley Field Centre during March–August in 1996 and 1997 (data kindly supplied by the Hydrology Project, University of Manchester, UK)
Monthly rainfall (mm)
Number of raindays per month SD
There was no difference between logged and unlogged habitats in the proportion of stations where R. makuta was present/absent both years and stations where it disappeared/appeared in 1997 (P > 0·3 in both cases). Stations where R. makuta had been present in both years (n = 10) were compared with stations where it had disappeared in 1997 (n = 16) using logistic regression to investigate the importance of four habitat variables [percentage cover of Selaginella, distance to nearest stream, forest density (PCA factor 1) and tree size (PCA factor 2)]. The regression model predicted 77% of presence/absence correctly (chi-square = 16·40, 1 d.f., P < 0·001) and R. makuta was more likely to have disappeared from a station in 1997 if the station was far from a stream. The probability of disappearance (p) is described by the following equation:
In contrast, stations where R. makuta was present in 1997 only (n = 5) had a higher cover of larval food plant compared with stations where butterflies were absent in both years (n = 34; chi-square = 4·36, 1 d.f., P = 0·037; model predicted 90% presence/absences correctly, although these results should be treated with caution given the small sample size in one category). The probability of appearance (p) is described by the following equation;
There were significant differences in vegetation structure between logged and unlogged areas showing that the effects of disturbance were still evident 8–9 years after selective logging. Unlogged forest had significantly larger trees (measured as girth at breast height) and greater canopy cover, and two PCA factors measuring forest density and tree size were significantly higher in unlogged forest, compared with selectively logged areas. There was also significant heterogeneity in vegetation structure between transects within logged forest. Logging records indicated that similar timber volumes were extracted from areas around transects 3 and 4 (75–95 m3 ha–1; Costa & Karolus 1992), but these data were averaged over areas of approximately 2000 ha, which obscures small-scale differences in logging intensity at a scale relevant to this study. Selective logging typically results in a mosaic of forest disturbance ranging from areas of severe disturbance along tractor tracks and around timber collection points, to areas of minimal disturbance and relict patches of primary forest (Whitmore 1991). Transect 3 was close to streams and passed through a river catchment area containing areas that were less severely logged (Steyshen Baru; Douglas et al. 1992). Differences among transects in logged forest in this study thus probably largely reflected this variation in disturbance, and the presence of smaller trees on transect 4 suggests that a larger volume of timber was extracted from around transect 4. However, transects in logged forest were also slightly further apart than in unlogged forest (Fig. 1), and although the distance between transects in logged forest was small (< 500 m at the nearest point), this may also have contributed to observed heterogeneity between transects within logged forest. Levels of disturbance within logged forests are also related to other topographical and landscape features, and high habitat heterogeneity is typical of selectively logged areas throughout the tropics (Whitmore 1991).
The importance of sampling widely to take account of habitat heterogeneity has been recognized (Sparrow et al. 1994). In this study, butterflies were recorded from an area of approximately 10 ha (combining areas of observation stations and walks between stations), and the use of transects (total length 8·5 km) meant that a wide variety of habitats and microclimates (riverine, canopy gaps, different aspects, etc.) was surveyed. Results from this study highlight the importance of sampling over a large area for a relatively sedentary subcanopy species such as R. makuta, and sampling widely may be even more important for more mobile species.
Transect data showed that R. makuta adults were always recorded in areas of closed-canopy forest, and within this habitat they preferred sites in damper areas (close to streams) with high cover of Selaginella and, to a lesser extent, sites with less dense forest (low PCA factor 1 score). There was no difference between logged and unlogged habitats in spatial distribution or abundance of butterflies, and also no difference in cover of Selaginella or proximity to streams, although logged forest had lower PCA factor 1 scores. However, abundance of R. makuta differed among transects and was highest on transect 3 in logged forest, where stations were closer to streams, but was lowest in more severely logged forest (transect 4), indicating that the effects of selective logging may have been apparent in this species in some areas 8–9 years after logging. In this study, measuring vegetation structure and availability of larval resources in different areas within each habitat was crucial to understanding the responses of R. makuta to habitat disturbance. These results stress the importance of quantifying vegetation structure in order to determine the severity of forest disturbance and to allow interpretation of the effects of disturbance on different species.
Habitat requirements and disturbance
Habitat requirements of R. makuta did not differ between logged and unlogged areas, as measured by the proportion of stations where the butterfly was both predicted to occur and was recorded present. This suggests that a lack of any observed differences in abundance and distribution between logged and unlogged areas was not due to butterflies changing their habitat requirements in response to disturbance. There was, however, a significant difference among transects, and transect 3 in logged forest had a larger number of stations where the prediction of presence/absence of butterflies did not agree with the recorded distribution. These cases included both stations where the butterfly was predicted present but was absent, and stations where the butterfly was present but predicted absent, making any effects of logging on habitat requirements difficult to interpret. Although sample sizes were small, there is some evidence that butterflies were occupying stations in more marginal habitats in logged forest, but that they were not moving into areas of very unsuitable habitat (predicted occurrence was significantly higher at stations where the butterfly was present but predicted absent compared with stations where the butterfly was both predicted and recorded absent). This indicates some degree of flexibility in habitat requirements in this species. Such flexibility could potentially reduce the impacts of selective logging, but any effect in this study was relatively small and broad habitat requirements did not differ between habitats.
Effects of drought
A marked decrease in rainfall in 1997 corresponded with significant declines in both abundance and local distribution of R. makuta. Compared with a year of more normal rainfall, populations contracted to damper areas close to streams and to areas with high cover of the larval host-plant, although there was no difference between logged and unlogged areas in the effects of the drought. Similar patterns of population contraction in response to a reduction in humidity and moisture availability have been observed in other satyrid species from temperate areas (Sutcliffe et al. 1997) and from tropical areas with marked wet/dry seasons (Braby 1995), but have not been previously recorded from less seasonal tropical regions. Compared with a year of more normal rainfall, butterfly distribution was not clumped in 1997, and stations with butterflies were further from other occupied stations, suggesting increased fragmentation and isolation of local populations. In temperate regions, small isolated populations are more prone to local extinction (Thomas, Thomas & Warren 1992) and, although the degree of isolation in this study was relatively small (the maximum distance between occupied stations was 1·1 km), the reported poor flight of R. makuta (Corbet & Pendlebury 1992) suggests that it may be vulnerable to population fragmentation at this scale. Few studies investigating the effects of disturbance have taken place over sufficient time to take account of widespread changes in environmental conditions, and whether populations in logged and unlogged areas recover at an equal rate from such events has not been investigated.
Butterfly conservation in malaysia
The consequences of degrading large forest areas for the survival of insect species are poorly understood (Sayer & Whitmore 1991), and the ability of different species to survive and reproduce in disturbed forest is therefore of great interest. In the near future, most remaining production forest in Malaysia will have been through one cycle of selective logging and will be in a period of regeneration before being selectively logged again (Lambert 1992). In Indonesia, butterfly diversity and biogeographic distinctiveness were significantly lower 5 years after selective logging compared with undisturbed forest (Hill et al. 1995), and results from this study show that 8–9 years after logging differences in vegetation structure were still evident and butterfly distribution and abundance were lower in the most severely disturbed areas. Further work is needed to see if populations and communities come to resemble those in unlogged forest after longer periods of recovery, or whether repeated selective logging will result in a reduction of diversity and loss of species with high biogeographic and/or taxic distinctiveness (Vane-Wright, Humphries & Williams 1991), a more important consideration than diversity per se in terms of conservation of global biodiversity.
This work was in collaboration with Dr Chey Vun Khen, Forestry Research Centre, Sepilok. I thank Keith Hamer, Tom Sherratt and Helen White, and also Andrew Davis and all at Danum Valley for making my stay so enjoyable. I also thank Yayasan Sabah (Forestry Upstream Division), the Danum Valley Management Committee, the State Secretary (Internal Affairs and Research Office), Sabah Chief Minister's Department, and the Economic Planning Unit of the Prime Minister's Department of Kuala Lumpur for permission to conduct research in the Danum Valley, Sabah. This study is project number DV157 of the Danum Valley Rainforest Research and Training programme and was supported by a British Ecological Society research travel grant. This is publication A/224 of the Royal Society South-East Asia Rainforest Research Programme.
Received 16 May 1998; revision received 5 May 1999