[A-TT35] Machine Learning Techniques in Weather, Climate, Ocean, Hydrology and Disease Predictions

[ATT35-06] Operational forecasting of El Niño-Southern Oscillation (ENSO) Niño3.4 index using ensembles of Convolutional neural networks

*Kalpesh Ravindra Patil¹、Takeshi Doi¹、Jayanthi Venkata Ratnam¹、Swadhin Behera¹ (1.APL, VAiG, JAMSTEC)

キーワード：El Niño-Southern Oscillation (ENSO), Niño3.4 index, Operational forecasts, Convolutional neural network (CNN), CNN ensemble, Long lead-time forecasts

El Niño-Southern Oscillation (ENSO) forecasts are vital for various sectors such as agriculture, health, and more, to enhance preparedness and planning against extreme climate and weather anomalies. Leading global forecasting centers issue ENSO forecasts based on state-of-the-art dynamical and statistical models. Recent studies have shown that deep learning models based on the Convolutional Neural Network (CNN) outperform traditional forecasting models in predicting ENSO at long lead times. At JAMSTEC, we have recently started issuing ENSO forecasts (Niño3.4 index) for research purposes, utilizing CNN-based deep learning model ensembles (Patil et al., 2023). The CNN ensembles are trained on historical global oceanic surface and subsurface anomalies, employing an extreme value loss function to effectively capture the extremes of ENSO events. Operational forecasts are generated for lead times ranging from 2-months to 2-years, using a ten-member CNN ensemble. The forecasts are available at https://www.jamstec.go.jp/aplinfo/sintexf/e/seasonal/outlook.html.
It is planned to integrate these forecasts with those from the International Research Institute (IRI). We aim to present our experiences with operational ENSO forecasting using deep learning models, as well as our future plans to improve ENSO predictions through various deep learning techniques.
Reference
Patil KR, Doi T, Jayanthi VR and Behera S (2023) Deep learning for skillful long-lead ENSO forecasts. Front. Clim. 4:1058677. doi: 10.3389/fclim.2022.1058677