Water is one of the most vital natural resources for sustaining life on earth. The rapid population growth, urbanization, and fast growing industrialization are under increasing threat of water demands. The demand for water is increasing every day, and the total availability of water remains practically constant rather decreasing with passage of time. It is therefore, vital to estimate, conserve, and scientifically manage water available in hydrological system for survival of mankind and necessary to meet the future needs of water. Traditional hydrologic models that are based on mathematical representation of watershed process have been applied to streamflow predictions. These models typically require a significant effort of data collection including rainfall, stream flow and watershed characteristics. Additional efforts are needed for assessing model parameters, and performing model calibration and verification. Hydrologic models devoted to stream flow predictions range from hourly to daily forecast of stream/flood flows (Bertoni et al. 1992; Shamseldin 1997; Rajurkar and Chaube 2002.)
A large number of hydrological analyses require mapping and modeling of non-linear systems data. Traditionally such mapping is performed with the help of conceptual models or statistical tools such as regression and curve fitting (Raju, 2008). However, when the underlying physical laws are unknown or not precisely known, it is rather difficult to model the phenomenon adequately. Attempts have made to develop a technique that does not require algorithm or rule development and thus reduces the complexity of the software. One such technique is known as neurocomputing, and the networks laid out with many parallel processing elements to do this neurocomputing are called artificial neural networks (ANNs).
