Modeling Weekly Rainfall Using Long Memory Models

Modeling rainfall has become a demanding assignment since the complexity of rainfall pattern is changing day by day particularly in Asian countries. Accurate inferences on rainfall prediction facilitate filling the noticeable gap with respect to climate monitoring to reduce climate stress in the country. A lesser number of efforts have been taken to model the rainfall by accounting for long-range dependency features with conditional heteroskedasticity. This research attempts to model weekly rainfall using long memory models as it exhibits the blend features of long memory and time dependence variance. The long memory model parameters were estimated using maximum likelihood estimators. A comparison of forecasting power was assessed of four long memory models using an independent sample, and the best model with the highest forecasting strength was selected via certain selection criteria. The result derived from the best-fitted long memory model gives good conformity with the real weekly rainfall series, and the forecasting performance is not much diluted with the increase of the forecasting length. This indicates that the selected model performs better in forecasting weekly rainfall distribution.