Review of "Reconstructing coupled time series in climate systems by machine learning"
2020
This manuscript investigates the potentialities of reconstructing time series using machine learning (ML) techniques.This approach is applied on a set of simple systems, and then applied to the interaction between the Tropical surface temperature and the Northern extra-tropical surface temperature.Different configurations of the machine learning approaches are explored, the reservoir computing, the long short-term memory, but also a simplified version of the latter and back-propagation.The authors use the correlation (for linear systems) and the convergent cross mapping (for nonlinear systems), CCM, as tools to evaluate the ability of the machine learning approaches to reproduce the original time series.Although I find the idea of putting in parallel the CCM with the ability of reconstructing time series based on ML very interesting, the description of the tools and the results is confusing, the presentation is quite poor and