use of the canonical correlation analysis for the prediction of rainfall in central america

Abstract

A statistical model based on canonical correlation analysis was used to explore the predictability of early May, June, July (MJJ) and late August, September, October (ASO) rainfall seasons in Central America. Explanatory variables are seasonal Atlantic and Pacific Ocean surface sea temperature (SST) for the region inside 112.5 ºE-7.5 °W and 7.5 ºS-62.5 °N during 1958-1998. For the early rainfall season, MJJ, positive (negative) tropical Atlantic SST anomalies were associated with positive (negative) rainfall anomalies over a broad area located at the north of the studied region. For the late season, ASO, opposite sign anomalies in the Atlantic and Pacific oceans that surround Central America were associated with a strong rainfall signal, in which positive & negative (negative & positive) SST anomalies in the Pacific & Atlantic Tropical oceans, tend to be associated with lobes of negative (positive) rainfall anomalies mainly at Central American Pacific slope. The model results were cross validated, showing significant skill values over an important portion of the studied region.