The Application of Time Delay Neural Networks to River Level Forecasting

There are many different artificial neural networks (NN) (Shepherd, 1997) but as mentioned in Chapters 2 and 9, NN for forecasting river flow are almost always trained using backpropagation (BPNN). This may be due in part to the fact that BPNN were the first successful models to be implemented (Rumelhart et al., 1986), and because the algorithm is simple to program and apply. Within hydrology there are many examples of the successful application of these network types, e.g. as rainfall-runoff models (Abrahart & Kneale, 1997; Campolo et al., 2003; Minns & Hall, 1996; Salas et al., 2000; Shamseldin, 1997; Smith & Eli, 1995), for predicting water quality (Brion et al., 2001; Gumrah et al., 2000, Maier & Dandy, 1996) and in estimating rainfall (Dell’Acqua & Gamba, 2003; French et al., 1992; Hsu et al., 2000).