Skill of ACCESS-S2 in predicting rainfall bursts over Australia

Skill of ACCESS-S2 in predicting rainfall bursts over Australia

Precipitation falling in multi-day bursts is an important water source for the northern Australian tropics in summer and southern Australia in the cool autumn and winter months. Primary producers of crops and livestock need to plan for these events beyond the 7-day deterministic timeframe during the northern wet season (October – April) and southern wet season (April – November), to aid in their business operations. To coincide with the operational release of a multi-week burst potential forecast product in mid-2022, this report examines the skill of the Bureau's latest sub-seasonal-to-seasonal forecast system, the Australian Community Climate Earth-System Simulator – Seasonal version 2 (ACCESS-S2), released in October 2021, in predicting burst events 1 to 4 weeks in advance, predominantly focusing on northern Australia. We use the Symmetric Extremal Dependence Index (SEDI) as a metric to evaluate the skill of ACCESS-S2 in predicting rainfall bursts, given these are often rare events, particularly throughout the rangelands and sub-arid regions of northern Australia. As well as evaluating the hindcast skill of ACCESS-S2, we also apply the SEDI score to ACCESS-S1, to assess if there has been an appreciable improvement in burst prediction skill between forecast system versions.

As expected, the skill of ACCESS-S2 in predicting rainfall bursts across northern Australia (north of 30°S) is highest in the first week of the forecast and decreases with lead time. Skill is generally stronger and more significant over Queensland than the Northern Territory and northern Western Australia in the second week of the forecast between October to December. The peak skill months, in terms of magnitude and area of significance, occur from January to March, with the highest skill scores in March across all lead times (weeks 1 to 4). The ACCESS-S2 forecast system is also more skilful at predicting lower threshold bursts (e.g., 20 or 30 mm in 3-day events) than the more extreme, rarer burst events (e.g., 50 or 70 mm in 3-day events). This is likely as a result of sampling rather than real skill. Given that ACCESS-S2 shows good skill out to four weeks in February and March over the southern Top End, the Victoria River District, the Kimberley and Cape York, even for 50 mm in 3-day events, the burst potential forecast product has the potential to be useful to producers in these regions. This four-week prediction skill is in line with ACCESS-S2's prediction skill of the Madden Julian Oscillation (MJO). For southern Australia, model skill is only seen in lower thresholds, and there is good model skill out to two weeks in March to July over the Murray-Darling Basin. Likewise for southwest Western Australia, southern Victoria and Tasmania, there is prediction skill out to two weeks from June through to November. Beyond two weeks, there is very little accuracy. This two-week timescale is in line with current limits of skill for predicting atmospheric blocking and the Southern Annular Mode on the daily/sub-weekly timescale; they are two key drivers of rainfall variability over southern Australia.

Based on regional averages of the SEDI scores for '30 mm in 3-day' burst events over northwest and northeast Australia, ACCESS-S2 is generally more skilful than ACCESS-S1 across all lead weeks in the peak monsoon months of December to March. Armed with skill maps, examples of which are provided in the Appendix, and observed burst climatology maps, primary producers will be able to judge whether a burst event forecast is: (1) likely to be accurate; and (2) the outcome is more or less likely than they would expect for that time of year.

Files associated with this item cannot be displayed due to copyright restrictions.