Artificial Neural Network Modeling of Dissolved Oxygen in a Wetland Pond: The Case of Hovi, Finland

Abstract

Artificial neural networks (ANNs) are flexible tools from neuroinformatics that have performed well in a number of hydrologic applications so far. They tend to be particularly useful when applied to complex processes, the details of which are not well understood. The dissolved oxygen regime in constructed wetland ponds is, in turn, such a complex process, governed by a considerable number of hydrologic, hydrodynamic, and ecological controls which operate at a wide range of spatiotemporal scales. This paper reports on the results from a study conducted to test the adequacy of artificial neural networks in modeling near-bottom concentrations of dissolved oxygen in the Finnish free water surface wetland at Hovi. Various different networks of the multilayer perceptron (MLP) type of ANN were developed. The application proved successful, and in particular it was observed that MLPs were able to “learn” the mechanism of convective oxygen transport quite well. The ANN was also used to determine the relative influence of flow rate and wind shear on near bottom oxygen saturation.