Predicting El Niño with machine learning using past data

For centuries, the El Niño Southern Oscillation (ENSO) has played havoc with global weather. The climate event repeats every few years and can trigger droughts, floods, and hurricanes, but it is incredibly difficult to predict in advance.

Jinno et al. employed a reservoir computing-based model to predict the ENSO over a 24-month period, with a bandpass filter that will enable practical use.

Key to implementing a machine learning algorithm to analyze the ENSO is applying a frequency filter to isolate the relevant data. However, most existing filters cannot be applied to operational models.

“In past studies, prediction methods for ENSO using artificial neural networks typically rely on conventional low-pass filters, such as moving averages or Butterworth filters,” said author Takuya Jinno. “These smoothing techniques require data from both the past and the future to calculate the present value. Thus, strictly speaking, these prediction methods are not applicable to operational use, where only past time series data are available.”

As an alternative, the team developed a bandpass filter that performs a weighted moving average across past time steps, avoiding the use of any future information. Using this filter along with a reservoir computing model, they could predict the ENSO up to 24 months in advance.

The authors are excited about the possibility of applying this technique to other complex dynamical systems.

“We developed this method with the goal of predicting chaotic dynamics, including various atmospheric and oceanic phenomena,” said Jinno. “For example, the variability of tropical moist convection systems has multiple timescales in its intrinsic dynamics. Our approach may be beneficial for extracting specific temporal scales and investigating the prediction horizon and its characteristics as a low-order dynamical system.”

Source: “Long-term prediction of El Niño-Southern Oscillation using reservoir computing with data-driven realtime filter,” by Takuya Jinno, Takahito Mitsui, Kengo Nakai, Yoshitaka Saiki, and Tsuyoshi Yoneda, Chaos (2025). The article can be accessed at https://doi.org/10.1063/5.0261124 .

This paper is part of the Nonlinear Dynamics of Reservoir Computing: Theory, Realization and Application Collection, learn more here .