Species spatial distribution and richness of stream insects in south-western France using artificial neural networks with potential use for biosurveillance

The major goal of the PAEQANN European project is to provide tools to aquatic ecosystem managers, by using aquatic communities as ecological indicators and ANNs as modelling techniques. In lotic ecosystems the species composition of benthic communities depends on the diversity and stability of the stream habitats (Cummins 1979, Ward and Stanford 1979) which provide the possibilities of development (Malmqvist and Otto 1987). Therefore, benthic invertebrates are widely used as indicators of short- and long-term environmental changes in running waters (Hellawell 1978, Lenat 1988, Smith et al. 1999, Hawkins et al. 2000). Because they are ubiquitous, basically sedentary, with a large number of species, and strongly influenced by many natural and/or anthropogenic disturbances, aquatic invertebrates are by far the most commonly used indicators for the assessment of freshwater ecosystem quality (Rosenberg and Resh 1993). However, the very high diversity of aquatic invertebrates - 70% of the overall animal species recorded in European continental waters (Illies 1978) - and the difficulty to obtain specific identifications make quantitative approaches using macroinvertebrates unsuitable for the assessment of long term or large-scale changes in water quality. In the Adour-Garonne drainage basin (SW France) these quantitative studies have often been restricted to a single valley or range of mountains (Décamps 1968, Vinçon and Thomas 1987, Vinçon and Clergue 1988, Giudicelli et al. 2000), and were usually based on a single taxonomic group (e.g., one insect order). An important development for water management is the generation of practical tools which provide accurate biological assessments of river conditions without requiring a high level of expertise, effort and time for their users. These "rapid assessment" aproaches are designed to fulfil two objectives (Resh and Jackson 1993). First, reducing the effort (and cost) in sampling, sorting, and identification procedures. This can be achieved for exemple by considering only a fraction of the macroinvertebrates collected. A second objective is to summarize the results of site surveys by using single-score measures that can be understood by non-specialists. Species richness is such a measure, and is commonly used as an integrative descriptor of the community (Lenat 1988). It is influenced by a large number of environmental factors which can determine gradients in stream species richness (Vannote et al. 1980, Minshall et al. 1985) and it is also strongly influenced by natural and/or anthropogenic disturbances (Rosenberg and Resh 1993), which may lead to spatial discontinuities of these predictable gradients (Ward and Stanford 1979, 1983) and losses of taxa (Brittain and Saltveit 1989). Resh and Jackson (1993) observed that species richness measures were sensitive to the impact of human activities on stream ecosystems, and this was particularly true of some aquatic insects, e.g., Ephemeroptera, Plecoptera or Trichoptera (EPT), which can be considered as good biological indicators of disturbance in streams. Thus, the species richness of a restricted number of selected taxonomic groups is a good descriptor of the influence of disturbance upon the biota (Lenat 1988). An a priori framework for developing biological indicators is a stream classification based on macroinvertebrates, to characterize how ecosystems differ in terms of species assemblage. An interest of such classifications is that the stability of species assemblages may be used to define representative and/or reference sites for biological surveillance (Hughes et al. 1986), as any structural change in population features can indicate environmental changes in streams from a given region or a longitudinal section. At a large geographic scale, such stream classifications detecting several sub-regions associated to their characteristic macroinvertebrate assemblages are basically necessary to calibrate biological indicator measures. Using macroinvertebrates, we deal with ecological data that are bulky, nonlinear and complex, showing noise, redundancy, internal relations and outliers (Gauch 1982, Jongman et al. 1995). Great changes can also appear in variables, and complex interactions can occur between explanatory and response variables (Jongman et al. 1995). Traditionally, conventional multivariate analyses have been applied to solve these problems (Bunn et al. 1986, Ludwig and Reynolds 1988, Legendre and Legendre 1998). With these nonlinear and complex ecological data, however, nonlinear analysing methods should be preferred (Blayo and Demartines 1991). One of these methods is artificial neural networks (ANNs), which are versatile tools to extract information out of complex data, and which could be effectively applicable to classification and association. This paper describes how ANN methods can be used: i) to contribute to the understanding of large-scale geographic patterns in aquatic macroinvertebrate assemblages; ii) to obtain taxa richness predictions, with simple environmental attributes as input variables; iii) to replace or complement existing tools for water quality biosurveillance and management (Fig. 4.8.1). The results of recent studies, which focused on macroinvertebrates from four orders of aquatic insects (EPTC) in the Adour-Garonne stream system (South-Western France) are used to highlight the concepts.