Indian Ocean Dipole (IOD) forecasts based on convolutional neural network with sea level pressure precursor

Abstract

Forecasting the Indian Ocean Dipole (IOD) is crucial because of its significant impact on regional and global climates. While traditional dynamic and empirical models suffer from systematic errors due to nonlinear processes, convolutional neural networks (CNN) are nonlinear in nature and have demonstrated remarkable El Niño Southern Oscillation (ENSO) and IOD forecasting skills based on oceanic predictors, particularly sea surface temperature and heat content. However, it is difficult to measure heat content and easily introduces uncertainties, prompting the need to explore atmospheric predictors for IOD forecasts. Based on sensitivity prediction experiments, we identified the sea level pressure (SLP) signal as a crucial predictor, which forecasts IOD at a 7 month lead. In addition, the CNN model improves monthly forecasting accuracy while reducing errors by 13.43%. Utilizing the heatmap analysis, we elucidated that the multi-seasonal predictability of the IOD primarily originates from mid-latitude climate variability. Besides ENSO signals in the Pacific Ocean, our study highlights the significant impact of remote climate forcing in the South Indian Ocean, tropical North Indian Ocean, and Northwest Pacific Ocean on IOD forecasts. By introducing the SLP precursor and extratropical zones into IOD forecasts, our study offers fresh insights into the underlying dynamics of IOD evolution.