Seasonal prediction of North Atlantic sea surface temperature anomalies using the LSTM machine learning method

Abstract Sea surface temperature anomalies (SSTAs) over the North Atlantic (NA) have a significant impact on the weather and climate in both local and remote regions. This study first evaluated the seasonal prediction skill of NA SSTA using the North American multi-model ensemble and found that its performance is limited across various regions and seasons. Therefore, this study constructs models based on the long short-term memory (LSTM) network machine learning method to improve the seasonal prediction of NA SSTA. Results show that the seasonal prediction skill can be significantly improved by LSTM models since they show higher capability to capture nonlinear processes such as the impact of El Niño-Southern Oscillation on NA SSTA. This study shows the great potential of the LSTM model on the seasonal prediction of NA SSTA and provides new clues to improve the seasonal predictions of SSTA in other regions.