ENSO prediction based on Long Short-Term Memory (LSTM)

Abstract El Niño-Southern Oscillation (ENSO), as a global climate event with cyclical characteristics, often causes global climate anomalies and produces non-negligible economic and social impacts. Therefore, the prediction and research of ENSO events are important for understanding and solving global climate change issues. It has important scientific and practical significance. Previous research on ENSO events mainly used traditional statistical analysis and numerical simulation methods. This study explores the use of deep learning to improve the accuracy of El Niño-Southern Oscillation (ENSO) prediction. Based on long-term and short-term memory neural networks, the time series of meteorological and marine elements were analyzed. In the meantime, the sea surface temperatures (SST) and sea level pressure were predicted to calculate the Southern Oscillation Index (SOI) to reflect the ENSO phenomenon. Finally, this article takes the Niño3.4 regional data from the National Centers for Environmental Prediction (NCEP) dataset as an example, and uses the model proposed in this paper to compare with traditional statistical regression methods. The results show that the Long Short-Term Memory (LSTM) has a good effect in the prediction of ENSO events, and has certain scientific significance and practical value for the prediction of ENSO events.