Performance of Multi Model Canonical Correlation Analysis (MMCCA) for prediction of Indian summer monsoon rainfall using GCMs output

Abstract

The emerging advances in the field of dynamical prediction of monsoon using state-of-the-art General Circulation Models (GCMs) have led to the development of various multi model ensemble techniques (MMEs). In the present study, the concept of Canonical Correlation Analysis is used for making MME (referred as Multi Model Canonical Correlation Analysis or MMCCA) for the prediction of Indian summer monsoon rainfall (ISMR) during June-July-August-September (JJAS). This method has been employed on the rainfall outputs of six different GCMs for the period 1982 to 2008. The prediction skill of ISMR by MMCCA is compared with the simple composite method (SCM) (i.e. arithmetic mean of all GCMs), which is taken as a benchmark. After a rigorous analysis through different skill metrics such as correlation coefficient and index of agreement, the superiority of MMCCA over SCM is illustrated. Performance of both models is also evaluated during six typical monsoon years and the results indicate the potential of MMCCA over SCM in capturing the spatial pattern during extreme years.