The potential of canonical correlation analysis in multivariable screening of climate model

Abstract

Abstract The statistical downscaling model (SDSM) been used to analyse the potential changes of local climate trend in the long term. The difficulty of the SDSM model in selecting the best predictors group which having good association to the local climate. Even the SDSM provides screening process to analyse the predictor-rainfall relationship, however it has limited ability in analysing multiple variables from 26 predictors with 10 rainfall stations around Kedah state, Malaysia. In this regard, the Canonical Correlation Analysis (CCA) been used to analyse the multi predictor-rainfall relationships. The concept of canonical coefficient is sufficient to show the capability and reliability of the predictors based on the percentages of variance that can explained in the dependent variable using the independent variable. There were 10 predictors’ group have been developed and one predictor’s group was built based on the CCA result. The performances of these predictors groups were tested using statistical analyses. Results revealed that the predictors group selected by the CCA method has produced smaller values of MAE and MSE for all stations except at station of Ladang Tanjung Pauh. The box plot’s results, which generated from one hundred simulated samples, indicated that the performance of CCA method was remarkable. The presence of discrepancies in the HadCM3-A2 and HadCM3-B2 scenario simulations were relatively small and considered acceptable.