This paper provides an assessment of the performance of the National Climate Centre (NCC) seasonal forecast model using cross-validated hindcast data for the period from 1950 to 1993. Hindcast probabilities of rainfall occurring in three categories defined by the rainfall tercile values were generated by an optimal linear combination model using climatology, persistence, and the phases of the Southern Oscillation Index (SOI) and Darwin pressure anomalies. The performance of the consensus hindcasts was assessed using the linear error in probability space (LEPS) scoring method. A further assessment was made of the combination model after the inclusion of the probability forecasts from a sea-surface temperature eigenvector forecast scheme. It was found that the addition of the SST scheme provided an increase in skill over the hindcast period across the ‘predictability barrier’ of the austral autumn and enhanced and enlarged the areas of hindcast skill over eastern and inland Australia during the winter and spring.
Trevor M. Casey (Mon,) studied this question.