Sensitivity of Fractional Gaussian Noise Model to Hurst H Estimators in Drougth Analysis

The study of prolonged drought in the African subcontinent requires the formulation a FGN (Long Memory) model. Not much work seems to have been done in the parameter estimation of a long memory model. In this study, various estimators of Hurst h are evaluated to six annual flow series in the Congo basin. The sensitivity of a FGN model to Hurst h estimators is studied using simulation studies and statistical characteristics of droughts at variuos demand level are analysed. It is shown that an inappropriate choise of Hurst h estimator based on the principle of maximum likelihood should be used for formulating a long memory model.