Daily prediction of total coliform concentrations using artificial neural networks

Current water quality monitoring systems depend on the analysis of in situ grab samples. This is both cost- and labor-intensive, which makes it difficult to conduct daily monitoring tests. One possible way of overcoming this problem is to use a modeling approach. This paper describes the use of an artificial neural network, Self-organizing Linear Output (SOLO), as a modeling approach to predict total coliform concentrations from rainfall and streamflow data. Six different input scenarios are tested to check the efficiency of the SOLO approach, and the results show that the prediction of total coliform concentrations is possible if rainfall events occur. However, poor estimation results are obtained when there is no rain. The model performance improves slightly during periods of no rain if streamflow data are incorporated into the input. However, the model requires more input variables for no-rain periods, because the streamflow data do not enable observed variations to be fully predicted.