Harun Rashid: Assessing Australian rainfall teleconnections using linear and machine learning models
Analysing climate and earth system model simulations and comparing them with observations is crucial for model development and scientific understanding of climate phenomena. Comprehensive climate models produce complex outputs, and simple to sophisticated analysis techniques are needed to understand the processes behind phenomena of interest. Classical univariate and multivariate (linear) statistical methods have been used to analyse climate observations and simulations for years, but they have limitations in explaining the nonlinear aspect of climate variability.
This study investigates teleconnections between Australian rainfall variability and key climate drivers, e.g., the Indian Ocean Dipole (IOD), El Niño–Southern Oscillation (ENSO), and the Southern Annular Mode (SAM). The performance of traditional multiple linear regression (MLR) is compared with machine learning approaches, specifically Random Forests and Artificial Neural Networks (ANNs), in representing these relationships. Machine learning methods are expected to better capture nonlinear interactions and improve predictive skill. Models are trained using historical simulations from the CMIP6 large ensemble and evaluated against observed Australian rainfall time series. Details of the methodology and results will be discussed in the presentation.