Privacy Policy Disclaimer
  Advanced SearchBrowse




Journal Article

Long-term ENSO prediction with echo-state networks


Hassanibesheli,  Forough
Potsdam Institute for Climate Impact Research;


Kurths,  Jürgen
Potsdam Institute for Climate Impact Research;


Boers,  Niklas
Potsdam Institute for Climate Impact Research;

External Ressource
No external resources are shared
Fulltext (public)

(Publisher version), 5MB

Supplementary Material (public)
There is no public supplementary material available

Hassanibesheli, F., Kurths, J., Boers, N. (2022): Long-term ENSO prediction with echo-state networks. - Environmental Research: Climate, 1, 1, 011002.

Cite as: https://publications.pik-potsdam.de/pubman/item/item_27911
The El Niño-Southern Oscillation (ENSO) is a climate phenomenon that profoundly impacts weather patterns and extreme events worldwide. Here we develop a method based on a recurrent neural network, called echo state network (ESN), which can be trained efficiently to predict different ENSO indices despite their relatively high noise levels. To achieve this, we train the ESN model on the low-frequency variability of ENSO indices and estimate the potential future high-frequency variability from specific samples of its past history. Our method reveals the importance of cross-scale interactions in the mechanisms underlying ENSO and skilfully predicts its variability and especially El Niño events at lead times up to 21 months. This study considers forecasts skillful if the correlation coefficients are above 0.5. Our results show that the low-frequency component of ENSO carries substantial predictive power, which can be exploited by training our model on single scalar time series. The proposed machine learning method for data-driven modeling can be readily applied to other time series, e.g. finance and physiology. However, it should be noted that our approach cannot straightforwardly be turned into a real-time operational forecast because of the decomposition of the original time series into the slow and fast components using low-pass filter techniques.