FORECASTING INDIAN MONSOON RAINFALL INCLUDING WITHIN YEAR SEASONAL VARIABILITY

Indian annual rainfall is divided into three seasons namely, pre-monsoon, southwest monsoon and northeast monsoon (or post- monsoon). The total seasonal precipitation in the month of June, July, August and September is generally known as southwest monsoon (SWM) rainfall. January to May is the pre-monsoon period and the northeast monsoon is from October to December. Maximum amount of rainfall occurs during SWM. During this season, the variation about the long term expected value is as high as 40-50% in some parts of the country. The distress caused by droughts and floods due to extreme variations of the monsoon can be mitigated to some extent if the rainfall time series can be modeled efficiently for simulation and forecasting of SWM data. Rainfall data is a strongly non-Gaussian time series exhibiting slowly varying oscillatory trends. Artificial neural network (ANN) models are known to be versatile in handling complex unstructured data. In this paper a new ANN model which includes within year (that is inter-seasonal) variation to model SWM data is developed. The model is found to be efficient in explaining nearly 94% of the data variance. One year ahead forecast on a set of observations, independent from the training period is shown to perform well, and hence can be taken as validation of the new model