RAINFALL-RUNOFF MODELING OF A HILLY CATCHMENT

Abstract

Over the last few decades many approaches have been presented to made runoff simulation in river basins. They are deterministic or stochastic in nature and involve conceptual or statistical understanding. Of late, techniques based on modeling of the data, rather than those for the underlying physical process, seem to have become popular following the advent of computational methods. In situations where the information is needed only at specific sites in a river basin and where adequate meteorological or topographic information are not available, site specific and simple artificial neural networks (ANNs) seem attractive alternatives to apply. The Artificial Neural Network (ANN) is a method of computation inspired by studies of the brain and nervous systems in biological organisms. A neural network method is considered as a robust tools for modeling many of complex non-linear hydrologic processes. It is a flexible mathematical structure which is capable of modeling the rainfall-runoff relationship due to its ability to generalize patterns in imprecise or `noisy' and ambiguous input and output data sets. This study describes the application of three layer feed forward ANN predict 10-daily runoff as a function of rainfall, previous day runoff and maximum and minimum temperature for the Tehri Dam catchment area. Selection of input data for development of the model is done on the basis of autocorrelation and cross-correlation between observed values of rainfall, minimum temperature, maximum temperature and runoff data from period Jan.1996 to Dec.2005 at various lags and without lags. The data on minimum and maximum temperature is considered for analysis due to the fact that the Pace Abstract study catchment received major part of runoff from snowmelt. Inclusion of maximum and minimum temperature with rainfall as input variables have improved simulation results . significantly. In this particular case, where snowmelt significantly contributes to the runoff, introducing temperature as an input, attempts to capture the snowmelt processes to some extent as observed in various model performances. The number of hidden nodes considered in ANN structure varied from 4 to 12. Performance of different ANN models is evaluated by comparing RMSE, R, and DC. Based on comparative analysis both for learning/training and testing on independent data sets, the performance of the model — 10 having model architecture as 6-5-1 is adjudged best and recommended for use for study catchment. Page