Forecasting El Niño Six Months in Advance Utilizing Augmented Convolutional Neural Network

One of the most significant phenomena related to large-scale climatic changes is the El Niño Southern Oscillation (ENSO), which can be instrumental in predicting dissasters across various regions of the globe. This study introduces a novel method aimed at predicting ENSO over a longer duration. This approach utilizes Convolutional Neural Networks (CNN) to anticipate ENSO occurrences up to six months ahead. The model utilizes anomaly maps as input data. Augmented Heat Content (HC) in conjunction with SST which is the surface temperature of the oceans are predictors, while the Niño 3.4 Index serves as the target for predictions. By applying convolutional tensors, the method extracts features from anomaly maps and processes them through a fully connected neural network to uncover relationships between the Niño Index and these features. A strategic data augmentation technique is employed to boost the quantity of merely HC input data, addressing the issue of limited observations related to recharged thermocline. The model has shown to provide consistent predictions when validated against observations over an extended period, confirming its reliability. Furthermore, the method's relatively low computational demand makes it a valuable tool for predicting ENSO and its related impacts.