An assessment of rainfall seasonal forecasting skill from the statistical model SCOPIC using four predictors

Abstract

The statistical model SCOPIC (Seasonal Climate Outlook for Pacific Island Countries) has been used to produce seasonal forecasts in ten Pacific Island nations since mid-2007 to improve their seasonal forecasting capacity and to provide timely warnings to changes in rainfall. However, to date there has been no detailed hindcast validation study to compare the forecast skill from the different predictors used to produce the seasonal forecasts at different stations from across the Pacific region. Here, we compare the rainfall forecasts created by the linear discriminant analysis model within SCOPIC using the four predictors: the Southern Oscillation Index (SOI); empirical orthogonal functions of sea surface temperature anomalies (SST1&9) and the NINO3.4 and the 5VAR index. This indicates that skill varies from season to season across the Pacific, with the highest skill in the austral summer and lowest skill in the austral winter. This study using tercile hit rates and LEPS percentage scores shows the 5VAR index has slightly superior skill compared to the NINO3.4, SOI and the SST1&9 indices, but results will vary depending on the station location, analysis period and the number of months used to calculate the predictor value.