Long Range Prediction of Indian Summer Monsoon Rainfall Using Data Mining and Statistical Approaches

Abstract

The Indian subcontinent is mainly dependent on South-West monsoon for her fresh water needs. The variability of South-West monsoon decides the state of the economy of this region. This rainfall in the Indian context is also known as Indian Summer Monsoon. From the literature available, it may be noted that people of India have been aware of the reversal of winds over the Arabian sea from the turn of Christian era as the important information regarding the wind regime was understood by the traders due to the movement of ships for trade across the Indian Ocean. Nevertheless, since ancient times, there has been a great demand for accurate forecasting of Indian Summer Monsoon Rainfall (ISMR). Predictors are the parameters that have high influence on rainfall patterns. In the past, several predictors have been proposed by hard-core meteorologists for ISMR prediction. However, over the times, the role of the predictors for ISMR prediction is drastically changing due to climate shift and new issues like; pollution, global warming etc. In this regard, a holistic approach is essential to use popularly available computing techniques to study the correlation between various existing predictors and to define a new set for efficient prediction. It is also true that compared to linear prediction techniques the use of recent approaches such as probabilistic and ensemble methods are expected to give improved performance. New approaches use inherent nonlinear relations among the data for better prediction. Using the techniques like fuzzy logic, a new scenario may be defined in prediction, where prediction value can be more precise and quantitative than conventional range prediction. In the present work the ISMR data provided by reputed data acquisition agencies like; National Centers for Environmental Prediction - National Center for Atmospheric Research (NCEP-NCAR), India Meteorological Department (IMD) and Indian Institute of Tropical Meteorology (IITM); over a period of 37 years (1969-2005), have been used to predict the rainfall. Data mining approaches, like correlation analysis and association rule mining, are used to identify highly influential predictors. Sophisticated state-of-the-art classifiers such as Neural Networks, Neuro-Fuzzy systems are used to predict ISMR using highly influential predictors. An approach of fuzzy logic has been applied, to quantitatively predict rainfall in a small geographical area of South India. The proposed method is used to analyse the effect of South-Western and North-Eastern monsoons on the Indian peninsular region. The findings of the present research are observed to be highly efficient, compared to the existing traditional prediction approaches including the ones used by IMD; the official government organization responsible for ISMR prediction.