AUTHOR=Song Dan , Su Xinqi , Li Wenhui , Sun Zhengya , Ren Tongwei , Liu Wen , Liu An-An 

TITLE=Spatial-temporal transformer network for multi-year ENSO prediction

JOURNAL=Frontiers in Marine Science

VOLUME=Volume 10 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2023.1143499

DOI=10.3389/fmars.2023.1143499

ISSN=2296-7745

ABSTRACT=The El Niño-Southern Oscillation (ENSO) is a quasi-periodic climate type that occurs near the equatorial Pacific Ocean. Extreme periods of this climate type can cause terrible weather and climate anomalies on a global scale. Therefore, it is critical to accurately, timely and effectively predict the occurrence of ENSO events. Most existing research methods relied on the powerful data fitting capability of deep learning, which does not fully consider the spatio-temporal evolution of ENSO and its quasi-periodic character, resulting the neural networks with complexity structure but poor prediction. To solve this problem, we propose a  spatio-temporal transformer network to model the inherent characteristics of sea surface temperature anomaly map and heat content anomaly map along with the changes of space and time by designing the effective attention mechanism.  In addition, to better conduct long-term forecasting, the previous prediction is treated as the prior knowledge to enhance the reliability of long-term prediction. Moreover, we also innovatively integrate the temporal index into feature learning procedure to model the influence of seasonal variation on forecasting EI Niño. Extensive experimental results show that our model can provide 18-month valid ENSO forecasts, which validates the effectiveness of our method.