<p>&#160; Reduced-order dynamical models play a central role in developing our<br>&#160; understanding of predictability of climate irrespective of whether<br>&#160; we are dealing with the actual climate system or surrogate climate<br>&#160; models. In this context, the Linear Inverse Modeling (LIM) approach,<br>&#160; by helping capture a few essential interactions between dynamical<br>&#160; components of the full system, has proven valuable in being able to<br>&#160; provide insights into the dynamical behavior of the full system.</p><p>&#160; We demonstrate that Reservoir Computing (RC), a form of machine<br>&#160; learning suited for learning in the context of chaotic dynamics,<br>&#160; provides an alternative nonlinear approach that improves on the LIM<br>&#160; approach. We do this in the example setting of predicting sea<br>&#160; surface temperature in the North Atlantic in the pre-industrial<br>&#160; control simulation of a popular earth system model, the Community<br>&#160; Earth System Model version 2 (CESM2) so that we can compare the<br>&#160; performance of the new RC based approach with the traditional LIM<br>&#160; approach both when learning data is plentiful and when such data is<br>&#160; more limited. The useful predictive skill of the RC approach over a<br>&#160; wider range of conditions---larger number of retained EOF<br>&#160; coefficients, extending well into the limited data regime,<br>&#160; etc.---suggests that this machine learning approach may have a use<br>&#160; in climate predictability studies. While the possibility of<br>&#160; developing a climate emulator---the ability to continue the<br>&#160; evolution of the system on the attractor long after failing to be<br>&#160; able to track the reference trajectory---is demonstrated in context<br>&#160; of the Lorenz-63 system, it is suggested that further development of<br>&#160; the RC approach may permit such uses of the new approach in settings<br>&#160; of relevance to realistic predictability studies.</p>