Modelling ecological interrelations in running water ecosystems with artificial neural networks

The assessment of properties and processes in running water ecosystems is a major issue in basic and applied aquatic science and has consequences for environmental management. However, knowledge of the system functions, e.g. temporal and spatial dynamics of physical, chemical, hydro-morphological and biological processes, and species-habitat interrelations are still insufficient. An integrative and prognostic ecological assessment of running waters thus is presently not available (e.g. Bayerisches Landesamt für Wasserwirtschaft 1998; Resh et al. 1994; Statzner et al. 1994; Townsend and Hildrew 1994; Townsend 1989; Vannote et al. 1980). The analysis of running water ecosystems and prediction with deterministic and stochastic models are limited. However, studies on water quality assessment have improved the methodology, which can also be applied to basic science. The high complexity and the spatial and temporal system dynamics are examples of typical non-linear relationships of abiotic and biotic variables with often low amounts of non-normally distributed data. This limits the application of traditional statistics. Artificial neural networks (ANNs) provide an alternative tool to analyse and model ecological relationships. Their most important features are multi-dimensionality, non-linearity, the ability to learn from examples and to generalise. General aims of our modelling approach are: Application and development of ANNs to 1. visualise and test data reliability 2. model ecological relations 3. test the suitability of different ANN types and pre-processing methods 4. detect the most important input variables 5. visualise the network status and the activation of neurons 6. develop neural modelling techniques initiating further research on bioindication and ecological prediction In basic and applied running water ecology detection and description of unknown interrelations and generation of hypotheses identification of the most relevant variables for modelling, e.g. indicator species prediction of ecological properties of lotic ecosystems prediction of the assemblage of benthic communities in disturbed and undisturbed streams generalisation of interdependencies. In an interdisciplinary research project of mathematicians, computer scientists, ecologists, and engineers, the suitability of various types of ANNs was tested. They were used to model temporal dynamics of water quality based on weather, urban storm-water run-off and waste-water effluents, bioindication of lotic ecosystem properties using benthic macroinvertebrates, and long-term population dynamics of aquatic insects depending on environmental and ecological variables.