Application of neural networks for the prediction of total phosphorus concentrations in surface waters

This paper describes the application of artificial neural networks (ANNs) for the time series modeling of total phosphorous concentrations in the Odra River. Data from the monitoring site Police in the lower part of the Odra were used for training, validating and testing the models. Two models are proposed to prove the satisfactory forecast of phosphorus concentrations: a simpler one with a single input variable and a more complex one with 14 input variables. Both ANN models show a high ability to predict from the new data set. On the basis of sensitivity analysis the relationships between phosphorus concentrations and other water quality variables were established.