Sub-seasonal to seasonal prediction of rainfall extremes in Australia

Scientific journal article


Seasonal forecasts normally provide probabilities of how a given month or season will compare to long-term average rainfall (the usual climatology). Since large scale climate drivers change states slowly and are known to impact rainfall extremes, it should be possible to predict extremes in rainfall in addition to deviations from averages. This research evaluated using an existing seasonal forecasting model, ACCESS-S1, to forecast metrics of rainfall extremes.

Scientists at the University of Melbourne and the Bureau of Meteorology have tested the ability to use ACCESS-S1 to predict daily-scale rainfall extremes. The number of wet days could be predicted as accurately as mean rainfall. Forecast performance was poor for measures of extreme events that are rare, such as rainfall days above the 90th percentile. The most accurate predictions were across Queensland and the Northern Territory. A case study of the 2010 La Nia demonstrates the model performed well during a strong phase of this climate driver. The metrics that were poorly predicted were spatially varied, such as maximum 1-day rainfall in a season and were driven by smaller scale processes that are more randomly driven. ACCESS-S1 showed decreasing forecast performance with increasing lead time, particularly across the warm season months. The model gave the most accurate forecasts when events were within a week to 10 days of initialisation.