**Appendix: Brief description of the developed statistical models**

#### **A. Linear discriminant analysis (LDA) model**

Linear discriminant analysis (LDA) model is currently used as an operational model for TC seasonal prediction by a number of organisations including the Australian Bureau of Meteorology which utilise NIÑO3 and SOI indices as LDA models' inputs [7]. Examining prospects for improving skill of operational TC seasonal forecasting, Kuleshov et al. [15] demonstrated that 5VAR index performs better than NIÑO3 and SOI. Consequently, the LDA model for annual total occurrences of TCs in the Australian region (*AR*) has been modified to use 5VAR and also incorporate time trend variable (*T*) as predictors in the region:

where 0, 2 . For a detailed mathematical description of the developed LDA model, see [15].

#### **B. Support sector regression (SVR) model**

Support vector regression (SVR) has been identified as a skilful machine learning algorithm for application to TC seasonal prediction [9]. Using non-parametric and non-linear regression approach, annual total number of TCs expected to be formed in the coming season (*Y*) has been generated using nine variables as the model's input. Selected input variables (*X*1–*X*9) were the following indices: *X*1, Dipole mode index; *X*2, NIÑO4; *X*3, NIÑO3.4; *X*4, NIÑO3; *X*5, NIÑO1.2; *X*6, El Niño Modoki index; *X*7, 5VAR index; *X*8, multivariate ENSO index; and *X*9, SOI. For a detailed mathematical description of the SVR model, see [17].
