A new algorithm has been developed for predicting seasonal rainfall at specific Pacific island locales. The algorithm uses a hybrid artificial intelligence technique which combines the genetic algorithm and genetic programming paradigms. The algorithm basically searches the input database, in this case sea surface temperature and the multi-variate ENSO index (MEI), and optimizes locations (i.e. variables) and the functional relationship among locations that best predict seasonal rainfall with a 2 month or greater lead time. The functions produced are specific to the predictand and can be ported to individual Pacific Meteorological Offices for their use in providing forecasting services.

Algorithm output identifies relevant SST locations and the function specific to each island's forecasted rainfall. The accuracy of the resulting forecast has been accessed using data from 1982-1991. The algorithm has been trained using data from 1991 - present.