Seasonal forecasting of tropical cyclone formation in the Australian region

The hazard of tropical cyclones (TCs) is a very relevant topic to the Australian economy and to the welfare of its northern population. Australia's climate and the interannual variability of Australian region TC formation (genesis; TCG) is strongly dominated by the ocean-atmosphere interannual climate pattern El Niño-Southern Oscillation (ENSO) and to a lesser extent by other climate modes of variability. This thesis investigates new ways of seasonal forecasting Australian region TCG counts and distribution by identifying potentially skilful climatological predictors and applying more advanced statistical modelling methods than previous models for the region. ENSO is known to be the most important predictor of seasonal variations in TCG for the Australian region. To investigate the ENSO-independent effects of the Indian Ocean Dipole (IOD) on Australian TCG, a simple, but effective method has been developed to separate the IOD from ENSO. Results demonstrate, that there is reasonable individual forecast skill afforded by the influence of the isolated IOD. In combination with common ENSO metrics, however, the IOD does not significantly improve seasonal forecasting of seasonal TCG counts in the Australian region or subregions. A Poisson regression model using Bayesian inference and the Markov chain Monte Carlo (MCMC) method was developed to forecast seasonal TCG counts in the Australian region. The final three-predictor model based on derived indices of subtropical Central Pacific June-July-August average convective available potential energy (CAPE), the tropical northeast Pacific May-June-July average meridional winds at 850 hPa (v850) and subtropical central South Pacific June-July-August geopotential height at 500 hPa performs best with the corresponding correlation coefficient between observed annual TCG totals and cross-validated model hindcasts of r = 0.73 over the 40-year record between 1968/89-2007/08. The model is adaptable for hindcasting seasonal TCG totals in Australia's Eastern (Coral Sea) TC subregion, while it lacks skill in the Western (eastern Indian Ocean) TC subregion (r = 0.79 and r = 0.38 respectively). To improve forecasts of annual TCG counts in the Western region (90⁰-135⁰E), a separate model was developed with correlations between cross-validated hindcasts and observed annual TCG count of r = 0.67 using the June-July-August tropical Central Pacific sea level pressure (SLP) and the above used index of v850 as predictors. A logistic regression approach applied in the Bayesian seasonal forecast model was found to be successful in forecasting spatial probabilities of Australian region TCG on a 2.5⁰ x 2.5⁰ grid for the upcoming season. The most skilful model is based on the SLP, NINO4 and v850 indices, combined with spatial information from CAPE and shows an average improvement over the climatological average of 25%. The average distribution of TCG probabilities over the study period, as well as the hindcasted strong variations of probabilities and distribution of TCG during ENSO events match remarkably well against observations over most of the study domain. Results demonstrate that the combination of dynamic with synoptic and/or thermodynamic features is most useful to identify climatic influences on the seasonal frequency and spatial distribution of TC development in the Australian region. Independent forecasts using the three introduced models and comparisons with current operational models demonstrate the relatively high skill of the models presented in this thesis.