ENSO Forecast using the Deep Learning

Figure 1. The Nino3.4 time-series from 2023 to 2026. The black line denotes the observed, and the red line denotes the CNN forecast result starting on September 1st 2024. The shading denotes the correlation skill during the validation period (i.e. 1984-2017)

Description of the statistical ENSO forecast model using Convolutional Neural Network (CNN)

The CNN-based statistical model for ENSO forecasts is formulated using the CMIP5 historical simulations and the SODA reanalysis from 1871 to 1973. The CNN has three convolutional layers and two max-pooling layers between them. A third convolutional layer is linked to the neurons in the fully connected layer, which is linked to the final output (Ham et al., 2019).

References

 - Ham, Y. G., Kim, J. H., & Luo, J. J. (2019). Deep learning for multi-year ENSO forecasts. Nature, 573(7775), 568-572.
 - Barnston, A. G., Tippett, M. K., Ranganathan, M., & L’Heureux, M. L. (2019). Deterministic skill of ENSO predictions from the North American Multimodel Ensemble. Climate Dynamics, 53, 7215-7234.