Reservoir Computing as a Tool for Climate Predictability Studies

Reduced-order dynamical models continue to play a central role in developing our understanding of predictability of climate. In this context, the Linear Inverse Modeling (LIM) approach (closely related to Dynamic Mode Decomposition DMD), by helping capture a few essential interactions between dynamical components of the full system, has proven valuable in being able to give insights into the dynamical behavior of the full system. While nonlinear extensions of the LIM approach have been attempted none have gained widespread acceptance.

We demonstrate that Reservoir Computing (RC), a form of machine learning suited for learning in the context of chaotic dynamics, provides an alternative nonlinear approach that comprehensively outperforms the LIM approach. We do this in the example setting of predicting sea surface temperature in the North Atlantic in the pre-industrial control simulation of a popular Inter-governmental Panel for Climate Change (IPCC) class earth system model (Community Earth System Model version 2) so that we can compare the performance of the new RC based approach with the traditional LIM approach both when learning data is plentiful and when such data is more limited. The improved perdictive skill of the RC approach over the LIM approach in both these settings reiterates the use of the new machine learning approach in future climate predictability studies. Additionally, the potential of the RC approach to capture the structure of the climatological attractor and to continue the evolution of the system on the attractor in a realistic fashion long after the ensemble average has stopped tracking the reference trajectory is highlighted by considering the RC approach in the context of the Lorenz '63 system.

Finally, comparisons to other feedforward and recurrent deep learning methods are made and a broader perspective on the use of machine learning in understanding climate predictability is offered.