Abstract:
The rainfall-runoff process is a highly non-linear complex process due the spatial and temporal variability of precipitation patterns and watershed characteristics. The understanding and modelling of rainfall-runoff transformation on watershed scale has attracted the hydrologists for water management, stream flow estimation, water supply, irrigation, drainage, fl ood control, water quality, power generation, recreation, and wild life protection and propagation. A number of modelling approaches have been developed in past to simulate the process accurately and efficiently.
In this study, a back propagation Artificial Neural Network (ANN) modelling approach has been formulated in FORTRAN language. Model uses the gradient descent optimization technique considering pattern learning process with different normalization techniques and applied for monthly rainfall-runoff modelling of the Tawi river catchments up to Tawi Bridge at Jammu. The model uses the monthly rainfall and runoff data from 1992 to 2002 which is pre-analyzed for general behavior of process on annual basis. Models were calibrated and validated considering whole data set in four different ways with different normalization techniques and by considering both three and four layers system with different numbers of nodes in hidden layers. The models were also evaluated considering four different statistical testing techniques. Three-layered feed forward network structure was better than a four layered structure. In all four combinations, adopted for the modeling, none was found effective uniformly in all calibration, cross-validation and verification periods and may be probably due to quality of the data.
