A fuzzy inference model for rainfall prediction

Rainfall prediction is one of the complex nonlinear dynamic phenomena. This is due to uncertainties associated with the climatic parameters used for rainfall prediction. Fuzzy system has the capability to deal with the uncertainties and is efficient when the conventional linear statistical models are not able to perform well due to the nonlinear nature of the climatic parameters. In the present study, a data driven Fuzzy Inference System for high-dimensional data is developed to predict rainfall of the Indian subcontinent. Indian monsoon is an important climatic phenomenon due to its direct impact on socio-economic growth. The parameters Sea Surface Temperature, Sea Level Pressure, El Nino-Southern Oscillation, Indian Ocean Dipole Mode and the Equatorial Indian Ocean Oscillation have been used for analyses and prediction. The variability of Indian rainfall is considered for the period of 25 years from 1990-2014 and the possibility of prediction is explored using Fuzzy Inference System. In fuzzy inference system the membership functions are the building blocks and computing its range is a crucial task. We have used triangular membership function and in order to define the range of membership function, this study proposes two methods, divisive method for input parameters and clustering based method for output parameter. The experimental results obtained using the proposed fuzzy inference system is compared with Multiple Linear Regression and Multiple Adaptive Regression Splines. The proposed Fuzzy based predictive model shows better results in terms of the accuracy with 84% and correlation 0.78 between actual and predicted rainfall.