PubMed2001
Only recently, modelling has been accepted as an interesting and powerful tool to support river quality assessment and management. The 'River Invertebrate Prediction and Classification System' (RIVPACS), based on statistical modelling, was one of the first and best known systems in this context. RIVPACS was developed to classify macroinvertebrate community types and to predict the fauna expected to occur in different types of watercourses, based on a small number of environmental variables. The prediction is essentially a static 'target' of the fauna to be expected at a site with stated environmental features, in the absence of environmental stress. Therefore this system is rather limited to apply in river assessment and management. Models that offer a prediction of faunal responses to changes in environmental features (e.g. changes in discharge regime, dissolved oxygen level, ...) would be of considerable value for river management. In this context, models based on classification trees, artificial neural networks and fuzzy logic were developed and applied to predict macro-invertebrate communities in the Zwalm river basin located in Flanders, Belgium. Structural characteristics (meandering, substrate type, flow velocity, ...) and physical-chemical variables (dissolved oxygen, pH, ...) were used as inputs to predict the presence or absence of macroinvertebrate taxa in the headwaters and brooks of the Zwalm river basin. In total, data from 60 measurement sites were available. Reliability and particular strengths and weaknesses of these techniques were compared and evaluated. Classification trees performed in general well to predict the absence or presence of the different macroinvertebrate taxa and allowed also to deduct general relations from the dataset. Models based on artificial neural networks (ANNS) were also good in predicting the macroinvertebrate communities at the different sites. Sensitivity analyses related to ANNs allowed to study the impact of the input variables on the presence or absence of macroinvertebrate taxa and to determine the major variables that affect the ecosystem quality and should be taken under direct consideration in the management of river basins. Performance of the fuzzy logic models was significantly related to the methods that were used to set up the membership functions and the reliability of the information that was available. Fuzzy logic did not perform as well as the other two techniques with regard to short term predictions. Fuzzy logic appeared to be better and more robust for long term predictions, because of the easy and pragmatic integration of general expert knowledge and data derived rules in the transparent inference engine. The overall conclusion of our study is that all three techniques, classification trees, artificial neural networks and fuzzy logic appeared to be reliable to predict macroinvertebrate communities in polluted streams.