Soil Pore Water Pressure and River Suspended Sediment Modelling using Artificial Neural Networks

Prediction of water resources variables required for design and modelling purposes is
subjected to various degrees of uncertainties, ambiguities and challenges. This lack of
clarity is because of inappropriate or partial implementation of nonlinear
mathematical modelling techniques, lack of relevant data and the degree of
complexity involved in representing the physical process in a predictive modeL
Artificial Neural Network (ANN) is a powerful and robust computational technique
and has recently been widely and successfully used as a prediction tool in various
fields including water resources engineering. ANN models are mostly preferred as a
predictive tool because of their ability to map nonlinear patterns between the
dependent and independent variables without a mathematical description of the
physical process involved.
This study reports the attempts and associated findings in predicting some
hydrologic variables using the A1--.'N technique. Since the number of hydrologic
variables relevant to water resources engineering is very large, the number of
variables investigated in this study was restricted to two namely, the soil pore water
pressure and river suspended sediment concentration. These two variables were
investigated on the premise that (i) sufficient field data on the time series of these two
dependant variables and associated independent variables were obtainable; (ii) it was
concluded from literature review that no attempt has yet been made to predict soil
pore water pressure responses to rainfall using ANN and the issue has great relevance
to slope stability and hill-slope hydrological studies (iii) even though attempts for
suspended sediment prediction using ANN have been reported in literature, hardly
any attempt has been made to assess the suitability of different ANN training
algorithms in predicting river suspended sediment concentration.