Streamflow forecasting using artificial neural networks

A variety of ANN models are being tested for their performance in forecasting daily streamflow from rainfall measurements. In this study, a new ANN structure, called SOLO (Self Organizing feature map with Linear Output) is compared to the conventional time-delay neural network (TDNN) and recurrent neural network (RNN) structures. Results for the Leaf River watershed in Mississippi indicate that the SOLO structure provides equivalent or superior performance across the full range of flow levels (base flow recessions to peaks). Further, the SOLO structure is considerably less costly (in terms of effort and computational requirements) to identify and train.