Reservoir Computing as a Tool for Climate Predictability Studies

  Reduced-order dynamical models play a central role in developing our
  understanding of predictability of climate irrespective of whether
  we are dealing with the actual climate system or surrogate climate
  models. In this context, the Linear Inverse Modeling (LIM) approach,
  by helping capture a few essential interactions between dynamical
  components of the full system, has proven valuable in being able to
  provide insights into the dynamical behavior of the full system.

  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 improves on 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 earth system model, the Community
  Earth System Model version 2 (CESM2) 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 useful predictive skill of the RC approach over a
  wider range of conditions---larger number of retained EOF
  coefficients, extending well into the limited data regime,
  etc.---suggests that this machine learning approach may have a use
  in climate predictability studies. While the possibility of
  developing a climate emulator---the ability to continue the
  evolution of the system on the attractor long after failing to be
  able to track the reference trajectory---is demonstrated in context
  of the Lorenz-63 system, it is suggested that further development of
  the RC approach may permit such uses of the new approach in settings
  of relevance to realistic predictability studies.