Data-driven characterization of spatiotemporal chaos using ensemble reservoir computing

TL;DR

This paper introduces an ensemble reservoir computing approach to characterize spatiotemporal chaos in high-dimensional systems, providing both robust predictions and dynamical insights from data.

Contribution

It develops an ensemble multiplex reservoir computing method that improves prediction robustness and extracts dynamical information from uncertainty measures.

Findings

Ensemble reservoir computing enhances prediction robustness.

Uncertainty correlates with dynamical features like frozen positions and diffusion coefficients.

The method aligns uncertainty measures with Lyapunov spectra and power spectra.

Abstract

Spatiotemporal chaotic systems are difficult to characterize in a model-free manner because of their high dimensionality, strong nonlinearity, and sensitivity to initial conditions. Coupled map lattices, as a representative class of extended nonlinear systems, exhibit diverse regimes such as frozen random pattern, defect chaotic diffusion, and fully developed turbulence. In this work, we propose an ensemble version of multiplexing local reservoir computing for the data-driven characterization of spatiotemporal chaos. By constructing multiple base learners with randomized hyperparameters and combining their outputs, the method improves prediction robustness and quantifies predictive uncertainty through ensemble spread. More importantly, we show that this uncertainty contains direct dynamical information. It identifies frozen positions in frozen random pattern, supports the estimation of defect diffusion coefficients in defect chaotic diffusion, and provides an effective indicator of chaotic intensity in fully developed turbulence. Analyses of the spatial power spectrum and Lyapunov exponent spectrum further support the consistency between the uncertainty field and the intrinsic dynamical properties of the system. These results show that ensemble reservoir computing can serve not only as a prediction tool but also as a data-driven framework for the dynamical characterization of high-dimensional nonlinear systems.