Analysis of a bistable climate toy model with physics-based machine learning methods

TL;DR

This paper introduces a framework combining bifurcation analysis and neural ODEs to study and predict the behavior of a multistable climate toy model, enhancing understanding of complex climate dynamics.

Contribution

It presents a novel multistable climate toy model and demonstrates how neural ODEs can predict its future states within different attractors.

Findings

Successful identification of attractors using Monte Carlo Basin Bifurcation Analysis

Detection of the Melancholia state separating attractors

Effective prediction of system states with Neural ODEs

Abstract

We propose a comprehensive framework able to address both the predictability of the first and of the second kind for high-dimensional chaotic models. For this purpose, we analyse the properties of a newly introduced multistable climate toy model constructed by coupling the Lorenz '96 model with a zero-dimensional energy balance model. First, the attractors of the system are identified with Monte Carlo Basin Bifurcation Analysis. Additionally, we are able to detect the Melancholia state separating the two attractors. Then, Neural Ordinary Differential Equations are applied in order to predict the future state of the system in both of the identified attractors.