Stream−flow forecasting for Jakham river of Rajasthan using stochastic models

Abstract

Stochastic modeling is a widely−adopted method of stream−flow forecasting that is preferred over process−based method due to easy implementation, less data requirement, rapid development time and better economics of the former. This study employed a standard and comprehensive methodology for developing stochastic models, i.e. auto−regressive moving average (ARMA) to forecast monthly stream− flow of Jakham river located in Rajasthan. Prior to stochastic modeling, normality, homogeneity, and stationarity are examined in 40−year (June, 1975 to May, 2015) stream−flow series, and deterministic components of periodicity and trend are identified and removed. Then, residual series is used for developing stochastic models and estimating model parameters. Performance of the developed models is evaluated using six goodness−of−fit criteria. The best−fit model is found to be ARMA (1,3) with the highest value of correlation coefficient (0.81), modified Nash−Sutcliffe efficiency (MNSE) (0.53), and modified index of agreement (MIA) (0.77), and the lowest values of root mean square error (RMSE) (20.52 × 106 m3), Akaike information criterion (AIC) (361.53), and Bayesian information criteria (BIC) (372.68).