Seasonal prediction of Australian region tropical cyclones: the respective influences of the subtropical Indian Ocean and ENSO

The seasonal predictability of tropical cyclones (TCs) in the Australian region is analyzed for the 45-year period 1970 to 2014. Principle component and correlation analyses reveal six sea surface temperature anomaly (SSTA) regions to be potential predictors for forecasting seasonal TC counts (TCC). Three are well-known: Niño 3.4, Niño 4 and the Indian Ocean Dipole (IOD). The three other regions identified are the South Indian Ocean (STSIO), the subtropical South Pacific Ocean (STSPO), and a region immediately to the north of Australia (Local). The correlation pattern between TCC and SSTA in the Indian Ocean resembles the Subtropical Indian Ocean Dipole, which emerges as the second rotated principle component loading in that basin. It has a strong influence on TCs that form in the western part of the Australian region during the early part of the TC season. For the Pacific Ocean, the largest modulation of TCC occurs in conjunction with warming/cooling near the Dateline, rather than the canonical ENSO pattern. Three-month averaged SSTAs from January-February-March (JFM) to August-September-October (ASO) are used predictors for seasonal TCC. Both Poisson regression (PR) and support vector regression (SVR) methods are assessed. PR gives moderate to high correlations (>0.6), starting from JJA, thereby providing a skillful predictability lead-time of 2 months prior to TC season start (November 1). SVR and AR-SVR double the lead-time with correlations >0.6, beginning with AMJ. SVR methods are found to be superior to PR from JJA to SON. The results are interpreted physically by composting atmospheric parameters that are known to modulate TCC.