Forecasting of rainfall using ocean-atmospheric indices with a fuzzy neural technique

Forecasting of rainfall is imperative for miffed agriculture of and and semi-arid regions of the world where agriculture consumes nearly 80% of the total water demand Fuzzy-Ranking Algorithm (FRA) is used to identify the significant input variables for rainfall forecast A case study is carried out to forecast monthly rainfall in India with several ocean-atmospheric predictor variables Three different scenarios of ocean-atmospheric predictor variables are used as a set of possible input variables for rainfall forecasting model (1) two climate indices i e Southern Oscillation Index (SOI) and Pacific Decadal Oscillation Index (PDOI) (2) Sea Surface Temperature anomalies (SSTa) in the 5 degrees x 5 degrees grid points in Indian Ocean and (3) both the climate indices and SSTa To generate a set of possible input variables for these scenarios we use climatic indices and the SSTa data with different lags between 1 and 12 months Nonlinear relationship between identified inputs and rainfall is captured with an Artificial Neural Network (ANN) technique A new approach based on fuzzy c-mean clustering is proposed for dividing data into representative subsets for training testing and validation The results show that this proposed approach overcomes the difficulty in determining optimal numbers of clusters associated with the data division technique of self-organized map The ANN model developed with both the climate indices and SSTa shows the best performance for the forecast of the monthly August rainfall in India Similar approach can be applied to forecast rainfall of any period at selected climatic regions of the world where significant relationship exists between the rainfall and climate indices (c) 2010 Elsevier B V All rights reserved