SARIMA approach for modelling and forecasting of monthly stream flow of Godavari river in India

Abstract

In the present study, time series modelling of monthly stream flow of Godavari river was carried out using SARIMA (Seasonal Auto Regressive Integrative Moving Average Models). Monthly stream flowdata collected at Polavaram site by the National Water Development Agency from the year 1966 (June) to 2004 (May) were used in this study. Monthly stream flow datafrom the year 1966 (June) to 2000 (May) were taken for model development and rest data were used for validation.The monthly runoff series was made stationaryby applying the Box-Cox transformation and differncing. The best fit SARIMA (10,1,2)x (2,1,2)12 model was selected based on the ACF(autocorrelation function) and PACF (partial autocorrelation function) and the minimum values of AICC (Akaike’s Information Criterion) and BIC (Bayesian Information Criterion). Performance and validation of the models were evaluated based on various statistical measures.The mean and standard deviation of forecasted series were found close to the observed values. Nash-Sutcliffe coefficient also indicated high degree of models fitness to the observed data. Using the developed SARIMA model,forecasting of monthly stream flow time series for 37 years (June, 2004 to May, 2040) was carried out.