novel deep learning method for forecasting enso

Ability of predicting climate phenomena enables international organization and governments to manage natural disasters such as droughts. el niño sothern oscillation (enso) is one the most influential and crucial phenomenon follows with large scale climatic events and can be used for predicting droughts and floods all around the world. due to such a great importance, a new convolutional neural network method based on augmented data (acnn) for predicting enso on a relatively long period is developed in this research. the method is developed based on cnn to forecast enso six month earlier. sea surface temperature (sst) anomaly maps are given to the model as the predictors and niño 3.4 index is the predictand. the method applies convolutional tensors to extract features from the maps, and delivers them to a fully connected neural network to discover connections between niño index and the features. a tricky augmentation process is used to increase the number of input data to compensate lack of observations. the model’s skill correlation is over 0.83 for january-february-march season, while, the original cnn method’ correlation is 0.71. the model can be executed on gpus of a laptop without any need to super computers. the feature that makes it a great tool for predicting enso even for research institutions in low income countries.