Long-lead forecasts of seasonal precipitation in the tropical pacific islands using CCA

Abstract

A potentially operational system for 3-month total precipitation forecasts for island stations in the tropical Pacific has been developed at NOAA's Climate Prediction Center using the statistical method of canonical correlation analysis (CCA). Routine issuance of the forecasts could begin during 1996; presently they are issued experimentally. The levels and sources of predictive skills have been estimated at lead times of up to one year, using a cross-validation design. The predictor fields, in order of their predictive value, are quasi-global sea surface temperature, Northern Hemisphere 700-mb height, and prior values of the predictand precipitation itself. Four consecutive 3-month predictor periods are used to detect evolving as well as steady-state conditions. Modest forecast skills are realized for most seasons of the year; however, moderate skills (correlation > 0.5) are found for certain stations in the northern Tropics at lead times of 3 months or less in late northern winter, especially in the western Pacific. CCA generally outperforms persistence, even at short leads. The El Niño-Southern Oscillation (ENSO) phenomenon is found to play the dominant role in the precipitation variability at many tropical Pacific islands. During especially the late northern winter of mature warm (cold) episodes, precipitation is suppressed (enhanced) in a horseshoe-shaped region surrounding (to the north, west, south) the central and eastern equatorial zone, which is anomalously wet (dry). A secondary source of predictive skill, most important for northern summer, is a pattern with like-signed SST anomalies over the Tropics of all three ocean basins. While this pattern may encompass ENSO episodes, it varies at lower frequencies than the ENSO phenomenon on its own.