Forecasting the Indian Ocean Dipole With Deep Learning Techniques

Jun Liu,Youmin Tang,Yanling Wu,Tang Li,Qiang Wang,Dake Chen

DOI: https://doi.org/10.1029/2021gl094407

IF: 5.2

2021-10-22

Geophysical Research Letters

Abstract:In the present research, Indian Ocean Dipole (IOD) prediction was explored using statistical methods based on deep learning techniques. First, convolutional neural network (CNN) models were trained using sea‐surface temperature anomaly (SSTA) maps of the Indian Ocean from 1854 to 1989, and the properly trained CNN models were then validated with the period from 1991 to 2019. The results indicate that the deep learning approach is capable of forecasting the IOD at lead times up to 7 months. The forecast skills of CNN are superior to those of the dynamic models in the North American Multi‐Model Ensemble (NMME). The CNN outperforms the NMME with lower sensitivity to predictability barriers and fewer systematic errors. Moreover, the gradient heat map analysis demonstrates that the triggering precursors selected by CNN models for IOD events are novel and physically sensible. These results suggest the CNN to be a new and effective tool for both IOD prediction and comprehension. The Indian Ocean Dipole (IOD) is a prominent climate variability observed in the tropical Indian Ocean, and affects the climate in many parts of the world via atmospheric teleconnection. Due to the substantial impacts of the IOD, extensive research has been conducted on its dynamic prediction using state‐of‐the‐art coupled climate models. However, the prediction of the IOD with dynamic models at lead times exceeding 3 months remains a challenge. Here, we propose a statistical method based on deep learning techniques to forecast the IOD, which is capable to provide skillful prediction at lead times up to 7 months. The convolutional neural network (CNN) models are less obstructed by the boreal winter and summer predictability barrier as well. The oceanic precursors for IOD development, being freely identified by CNN models, are physically reasonable. The results indicate these triggering precursors are mainly distributed in the negative sea‐surface temperature anomaly of the Southern Indian Ocean at long lead times. Statistical method based on deep learning techniques to forecast the Indian Ocean Dipole Convolutional neural network (CNN) can provide skillful prediction at lead times up to 7 months Triggering precursors selected by CNN are mainly distributed in the southern Indian Ocean Statistical method based on deep learning techniques to forecast the Indian Ocean Dipole Convolutional neural network (CNN) can provide skillful prediction at lead times up to 7 months Triggering precursors selected by CNN are mainly distributed in the southern Indian Ocean

geosciences, multidisciplinary