A0782
Title: A global deep learning approach to forecasting monthly rainfall in Australia
Authors: Luyi Shen - The University of Melbourne (Australia) [presenting]
Guoqi Qian - The University of Melbourne (Australia)
Abstract: Understanding monthly rainfall patterns is part of an effort to enhance Australia's adaptability to climate change. Within this context, a spatiotemporal deep learning model is presented that integrates matrix factorization (MF) and temporal convolutional networks (TCN), incorporating year-month covariates and key climatic drivers to analyze and forecast monthly rainfall. TCN-MF model uses a fused dataset of ground-based rain gauge and satellite estimates, covering 1391 Australian grid locations from April 2000 to March 2021. To prioritize regions prone to flooding, empirical dynamic quantiles (EDQ) are applied to rank locations by rainfall dynamics. TCN-MF is applied to high-ranking locations, while forecasts at remaining sites are approximated via spatial interpolation. Using data from April 2000 to March 2021, rainfall is forecasted for April 2021 to March 2022, demonstrating strong performance in capturing significant rainfall events, especially in flood-prone areas. This model contributes to the field of continent-wide rainfall forecasting, providing valuable insights that would enhance the climate change adaptability of Australia.
