Forecasting monthly rainfall of Sub-Himalayan region of India using parametric and non-parametric modelling approaches

Rainfall being a complex phenomenon governed by various meteorological parameters is difficult to model and forecast with high precision. For hilly regions such as state of Sikkim and adjoining areas of West Bengal, rainfall acts as lifeline. Several parametric models such as seasonal autoregressive integrated moving average (SARIMA) and exponential autoregressive (EXPAR) are very popular and extensively used to model and forecast rainfall. Owning to complex nature of rainfall series, non-parametric time delay neural network (TDNN) model has also gained substantial amount of attention by researchers. This study uses these two broad class of models and applies them to the monthly rainfall of Sub-Himalayan West Bengal and Sikkim. The models were compared based on their forecasting efficiencies and pattern prediction ability.