The practical value of modelling relative abundance of species for regional conservation planning: a case study

https://doi.org/10.1016/S0006-3207(00)00139-7Get rights and content

Abstract

Statistical modelling of species presence/absence data in relation to mapped environmental predictors has been widely used to predict distributions of species for use in regional conservation planning. This paper evaluates the extent to which predictive mapping of habitat suitability might be refined by modelling relative abundance or density of a species instead of presence/absence. We use data collected at field survey sites in north-east New South Wales to develop models predicting the abundance of vascular plant and vertebrate fauna species as a function of regional-scale environmental variables. The predictive accuracy of these models is then evaluated using survey data collected at independent evaluation sites. A number of ‘direct’ abundance modelling techniques were evaluated including generalised linear and generalised additive Poisson regression, and zero-inflated negative binomial regression. We also evaluated the performance of predicted probability of occurrence generated by logistic regression modelling as an ‘indirect’ index of abundance. Both the direct and indirect modelling techniques generally failed to provide consistently reliable predictions of abundance. Reasonably accurate models were produced for only 12 of the 44 species evaluated. A further key finding was that, for all 12 of these species, predictions from direct abundance models performed no better as a relative index of abundance than predicted probabilities of occurrence generated by logistic regression modelling. Implications of these results for the use of predictive modelling in regional conservation planning are discussed.

Introduction

Regional conservation planning requires basic information on the distribution of animal and plant species throughout the region of interest. In Australia, nationally agreed criteria for the establishment of forest conservation reserves include criteria relating to both forest ecosystems and individual species, stating specifically that reserves should be designed to “maximise the area of high-quality habitat for all known elements of biodiversity wherever practicable, but with particular reference to the special needs of rare, vulnerable or endangered species; groups of species with complex habitat requirements, or migratory or mobile species; areas of high species diversity, natural refugia for flora and fauna, and centres of endemism; and those species whose distributions and habitat requirements that are not well correlated with any particular forest ecosystem.” (Commonwealth of Australia, 1997).

Proper implementation of the above criteria requires mapping of ‘high-quality habitat’ for selected species across vast forested regions. To date, such mapping has generally been performed by modelling species presence/absence data, collected at field survey sites, in relation to mapped environmental predictors using logistic regression or related modelling techniques. These models are then applied to environmental layers held in a geographical information system (GIS) database to extrapolate predicted likelihood of occurrence across the entire region of interest. Pearce et al. (unpublished manuscript) have demonstrated that, in general, such models can discriminate reasonably well between occupied and unoccupied areas, when evaluated using independent survey data.

In recent years, there has been increasing interest in the potential for modelling of relative abundance or density data instead of presence/absence data as a means of improving predictive mapping of habitat quality and delineation of high quality habitat. This interest is based on the assumption that relative abundance is likely to be a good indicator of habitat quality, reflecting key factors such as reproductive success, longevity, carrying capacity and susceptibility of populations to extinction (Kellner et al., 1986, Hobbs and Hanley, 1990).

A positive correlation between habitat quality and species abundance has been assumed in several fauna and flora studies in Australia (e.g. Lindenmayer et al., 1991, Stockwell et al., 1990), and elsewhere (e.g. Feber et al., 1996, Lavers and Haines-Young, 1996, Walsh and Harris, 1996, Leathwick ,1998). However, a number of authors caution against this assumption (e.g. Van Horne, 1983, Kellner et al., 1986, Hobbs and Hanley, 1990). Van Horne (1983), in particular, identifies six environmental and species characteristics that may reduce the probability that species density will be positively correlated with habitat suitability. Interactions among these characteristics and variability in the effectiveness of survey techniques adds further uncertainty to the assumption of a positive correlation between habitat quality and species abundance (Hobbs and Hanley, 1990). Even given an effective survey technique, the detectability of a species will strongly influence recorded abundance or density. Detectability may vary with the age of an individual, the habitat type, prevailing environmental conditions affecting the level of faunal activity, season, time of day, and observer skill and bias (Southwood, 1987).

These factors have led Van Horne (1983) and others to suggest that detailed demographic and resource utilisation information is vital to an understanding of habitat suitability and hence quality. Unfortunately, this information is prohibitively intensive and expensive to collect for conservation planning activities that require both immediate action and a consideration of the requirements of large numbers of species over large geographical areas. In such circumstances there may be no choice but to assume a correlation between abundance and habitat quality. Assuming this, habitat quality for a species can then be mapped by recording relative abundance at a sample of field survey sites within the region of interest, modelling the abundance data in relation to mapped environmental predictors and then using these models to extrapolate relative abundance across the entire region.

In this paper, we develop models of relative abundance using fauna and flora survey data from forested north-east New South Wales. The abundance data consist of counts of individuals for the fauna species and an index of relative cover abundance for the flora species. A number of ‘direct’ abundance modelling techniques are evaluated including generalised linear and additive Poisson regression and zero-inflated negative binomial regression. We also evaluate the performance of predicted probability of occurrence generated by existing logistic regression models as an ‘indirect’ index of abundance.

Throughout this evaluation, it is assumed that models of species abundance are required to provide maps identifying high quality habitat within north-east New South Wales. The performance of models will therefore be assessed in terms of their predictive accuracy when applied to new data within the region.

Section snippets

Survey data

Abundance data were collected for small reptiles, arboreal marsupials, vascular plants, and diurnal birds. The data used in this evaluation were collected during a number of biological surveys conducted in north-east NSW between 1991 and 1997: the North East Forests Biodiversity Study, NEFBS (NSW NPWSm, 1994a, NSW NPWS, 1994b); Natural Resources Audit Council studies, NRAC (NSW NPWS, 1995a, NSW NPWS, 1995b); the Joint Old Growth Forests Project, JOGFP (Clode and Burgman, 1997); and the

Performance of logistic regression as an indirect abundance modelling technique

The performance of logistic regression modelling as an indirect or surrogate abundance modelling technique was evaluated by testing the correlation between predicted probabilities of occurrence and observed abundance at independent evaluation sites in north-east New South Wales. The results of this evaluation are presented in Table 2.

A significant (P< 0.05) positive correlation between predicted probabilities of occurrence and observed abundance was obtained for 65 (89%) of the 73 species

Discussion

Although statistical models describing the abundance of species within north-east NSW were successfully developed for most species examined using GLM, GAM and ZIP techniques, these models had low predictive power for all but a handful of species when evaluated using independent survey data. The predictions from logistic regression models also performed poorly as a relative index of species abundance within north-east New South Wales for the majority of fauna and flora species examined. No clear

Acknowledgements

The work described in this paper was performed as part of two consultancies funded by the Australian Nature Conservation Agency and Environment Australia. We thank Andrew Taplin and Dave Barratt from these agencies for their support and encouragement. We also thank David Meagher for commenting on an early draft of the manuscript. The hard work and dedication of the survey teams that collected the data used in this study is greatly appreciated.

References (36)

  • W.J Conover

    Practical Nonparametric Statistics

    (1980)
  • S Derksen et al.

    Backward, forward and stepwise automated subset selection algorithmsfrequency of obtaining authentic and noise variables

    British Journal of Mathematical and Statistical Psychology

    (1992)
  • R.E Feber et al.

    The effects on butterfly abundance of the management of uncropped edges of arable fields

    Journal of Applied Ecology

    (1996)
  • K.J Gaston et al.

    Effects of scale and habitat on the relationship between regional distribution and local abundance

    Oikos

    (1990)
  • I Hanski et al.

    Three explanations of the positive relationship between distribution and abundance of species

  • Heilbron, D.C., 1994. Zero-altered and other regression models for count data with added zeros. Biometrical Journal 36,...
  • S.E Henke

    The effect of multiple search items and item abundance on the efficiency of human searchers

    Journal of Herpetology

    (1998)
  • N.T Hobbs et al.

    Habitat evaluationdo use/availability data reflect carrying capacity?

    Journal of Wildlife Management

    (1990)
  • Cited by (0)

    View full text