Using species abundance models as indicators of habitat disturbance in tropical forests



1. In temperate communities, species abundance distributions have been used to detect ecosystem disturbance: in undisturbed habitats, distributions are claimed to generally fit log-normal models, whereas in disturbed habitats, distributions fit log-series models.

2. There is a growing literature on the effects of habitat disturbance in tropical ecosystems and several studies suggest that species abundance models may be useful in detecting disturbance, although data are lacking.

3.Nummelin (1998) claims that these models are not universal indicators of forest disturbance, but we highlight a number of problems with the data presented in Nummelin's study and conclude that it is too soon to dismiss these models. We discuss several important points arising from Nummelin's study which need to be considered if species abundance models are to be used appropriately.

Nummelin (1998) claims that species abundance models cannot be used to detect tropical forest disturbance. We doubt, however, whether this study can be used either to support or refute this claim, although it does raise important issues related to measuring disturbance in tropical forest ecosystems.

Current concern over the consequences of destruction and degradation of tropical forests has resulted in a growing literature on the effects of habitat disturbance on tropical forest communities. These studies have used a wide variety of indices to examine the effects of disturbance on species diversity, but species abundance models have received surprisingly little attention in this context (Tokeshi 1993). Species abundance distributions of communities in undisturbed habitats are claimed generally to fit log-normal models, whereas those in disturbed habitats generally fit log-series models (Preston 1948; May 1975; Stenseth 1979; Sugihara 1980; but see references in Magurran 1988). One consequence of this pattern is that species abundances have been used as indicators of ecosystem disturbance in temperate regions (see references in Nummelin 1998), although data from tropical ecosystems are lacking.

In two recent studies in Indonesia (Hill et al. 1995; Hamer et al. 1997), we observed that species abundance distributions of butterflies fitted log-normal distributions in undisturbed forest but fitted log-series models in disturbed forest. A similar pattern has also been shown for birds in logged and undisturbed Malaysian forests (Putman 1994). In view of the urgent need for a means to assess rapidly the impact of habitat disturbance on tropical forest communities, we suggested that deviation from a log-normal distribution might provide an ‘instantaneous’ measure of habitat disturbance, and we highlighted the need for further work focusing on species abundance distributions of tropical communities in disturbed and undisturbed habitats. To this end, studies such as that of Nummelin (1998), which test models on independent, real data sets are to be encouraged, although the following issues need to be addressed to avoid inappropriate use of species abundance models in this context.

1. Sample sizes and statistical methods

Log-series and log-normal models cannot be differentiated unless the community has been adequately sampled. If not, most of the abundance distribution will be concealed and data from the tail of the distribution may equally well be described by log-series or log-normal models (illustrated by Magurran 1988, p. 28). Not only should communities be sampled adequately (usually assessed with species accumulation curves) but Kilmogorov–Smirnov one-sample tests should be used to fit species abundance data to log-normal and log-series distributions (biomass rather than numbers of individuals may be a more appropriate measure of abundance in some cases; Tokeshi 1993). Chi-squared tests are routinely used, although they lack power and are inappropriate for analysing small communities (Tokeshi 1993). In cases where a substantial proportion of recorded species have not been sampled, goodness-of-fit to a truncated log-normal distribution will be more appropriate (Magurran 1988). The number of species in the community must also be large enough to reveal the shape of the distribution and in cases where communities are small (e.g. Acridoidea in Nummelin 1998), species abundance models will not be appropriate.

2. Taxonomic level of study

Definitions of communities or assemblages are subjective but must be made at a level appropriate to the study and the ecology of the group in question. Much discussion of species abundance models has arisen through different researchers’ ideas of what constitutes a community or assemblage, with decisions on limits to group size being influenced by which species abundance model is preferred (Hughes 1986). However, species abundance models will be easiest to interpret when applied to fairly limited, well defined taxonomic groups (e.g. birds, termites, butterflies) because, as a statistical consequence of large numbers, species abundance data will almost certainly fit log-normal distributions irrespective of the degree of habitat disturbance if samples or taxonomic groups are indiscriminantly combined (Magurran 1988). The choice of taxa is also important, with the taxonomy of some groups (e.g. birds and butterflies) being more straightforward and so avoiding potential taxonomic biases (e.g. pooling sibling species).

3. Which taxa to study?

Further studies investigating the use of species abundance models in detecting tropical forest disturbance must be carried out using sensible choices of taxonomic study groups. Certain groups, such as butterflies which are known to include many species dependent on forest (Collins & Morris 1985), are better choices than others (Kremen 1992). Grasshoppers, for example, which in temperate areas are associated with early successional grassland and occur at very low diversity in undisturbed tropical forests (as illustrated in Nummelin 1998), will be a poor choice in the context of these models, although these taxa may have a useful role in other types of study of forest disturbance.

4. Sampling methods

The problems of using attraction traps (e.g. moth light traps) to sample communities have been discussed (e.g. Southwood 1978) and the fact that abundance data from this type of sampling method are unlikely to represent relative abundances in the natural community has been recognized (Tokeshi 1993 and references therein). Any method involving attracting individuals to traps is unlikely to reflect underlying natural abundance because some species will be more likely to be attracted than others, and such methods should not be used to investigate community structure or effects of habitat disturbance on tropical forest systems. Sampling methods need to be chosen carefully to remove these potential biases. For example, lack of knowledge of species ecology may mean that ground-based surveys are not representative of the entire community (Stork 1991) and the efficiency of methods such as sweep netting (used by Nummelin 1998) may be related to habitat disturbance because forest floor vegetation is typically denser in logged forests (Hill et al. 1995). Other methods such as recording free-flying individuals or hand-sorting samples for soil-dwelling species, although labour-intensive, avoid many potential biases (e.g. Eggleton et al. 1995; Hamer et al. 1997). The choice of sampling method must be appropriate to the study in question.

5. Temporal changes

The practical problems of working in tropical areas have resulted in few long-term, replicated studies and most studies have compared disturbed forest with adjacent undisturbed forest, assuming no beta diversity between sites. The goodness-of-fit of species abundance distributions to various models may be related to time-since-disturbance. For example, Nummelin's study (1998) took place 20 years after logging and species abundance distributions from disturbed habitats fitted log-normal distributions, whereas in our study of forests logged 5 years previously (Hill et al. 1995), species abundances fitted only log-series distributions. Studies of temporal changes in species abundance distributions following forest disturbance and subsequent recovery of communities are lacking and this topic requires further study.

6. Are tropical forests at equilibrium?

Log-normal distributions are supposedly characteristic of equilibrium communities (Stenseth 1979; Ugland & Gray 1982). Undisturbed forest, however, may not necessarily be at equilibrium (Blau 1980), and so there may be situations where communities in undisturbed habitats do not fit log-normal models. Conversely, disturbed habitats may be at equilibrium if the perturbation is maintained. This needs further study.

Despite Nummelin's claims (1998), we believe that it is far too soon to dismiss the usefulness of species abundance models in the context of tropical forest disturbance. Alternative methods of assessing species abundance distributions (e.g. Cotgreave & Harvey 1994; Bulla 1994) may also prove useful. However, all models must be used appropriately. For example, they cannot help in understanding community processes in relation to disturbance and are not able to address important questions such as whether certain species (particularly those of high conservation value) are more vulnerable than others to habitat disturbance (e.g. Bowman et al. 1990; Spitzer et al. 1997; Hill et al. 1995; Hamer et al. 1997). Nonetheless, species abundance models are a powerful tool for describing community structure and we encourage the wider use of these models in tropical ecosystems. Insects, in particular butterflies, respond rapidly to forest disturbance (Kremen 1992; Kremen et al. 1993) and species abundance models of insects may be particularly powerful as indicators of tropical forest disturbance.

Received 30 May 1997