Project without external funding

Modellierung von Wirkungszusammenhängen in Fliessgewässer-Ökosystemen mit Künstlichen Neuronalen Netzwerken

Project Details
Project duration: 06/199508/2000

Modelling ecological properties in lotic ecosystems using Artificial Neural Networks The assessment of states and processes of running waters is a major issue of aquatic environmental management. Because system analysis and prediction with deterministic and stochastic models is often limited by the complexity and dynamic nature of these ecosystems, supplementary or alternative methods have to be developed. We tested the suitability of various types of artificial neural networks for system analysis and impact assessment in different fields: a) temporal and spatial dynamics of water quality as influenced by the weather, urban storm-water run-off and waste-water effluents b) colonisation patterns of benthic macro-invertebrates in relation to water quality and habitat structure c) prediction of population dynamics of aquatic insects. Specific pre-processing methods and neural models were developed and allowed the assessment of relations among complex parameters (weather, discharge, water-quality, habitat-structure, benthic community structure) with high levels of significance. Special data processing employing a dynamic evaluation of data fed into the network improved the performance of the networks. Sensitivity studies with a special back-propagation-variant allowed a quantitative assessment of the input data which was not possible with standard neural modelling techniques. Time dependent neural networks, which allow for a consideration of past events, proved rewarding and compensated for investment into the analysis and prediction of time series. The results demonstrate that neural networks can conveniently be applied to running water ecology and sanitary engineering including the prediction of effects in the community. In particular neural networks can be used to reduce the complexity of data sets by identifying important (functional) inter-relationships and key variables. Thus, complex systems can be visualised in easily surveyed models with low measuring and computing effort. Examples for this are the diurnal variation of oxygen concentration (modelled from precipitation, water temperature and oxygen concentration of the preceding day; r^2 = 0.79), population dynamics of emerging aquatic insects (modelled from discharge, water -temperature and abundance of the parental generation; r^2 = 0.78), and water quality as indicated by a few "sensitive" benthic organisms (e.g. conductivity, using 5 out of 248 species; r^2 = 0.86). Estimation of the influence of complex environmental stress variables and prediction of the course of development of running waters from given conditions can be developed to greater detail than in conventional methods. Future studies on artificial neural networks and their applicability to bioindication and monitoring in running water systems are intended.

Principal Investigator

Last updated on 2017-11-07 at 13:36