Abstract

Abstract. Recent advances in statistical and machine learning have opened the possibility of forecasting the behaviour of chaotic systems using recurrent neural networks. In this article we investigate the applicability of such a framework to geophysical flows, known to involve multiple scales in length, time and energy and to feature intermittency. We show that both multiscale dynamics and intermittency introduce severe limitations to the applicability of recurrent neural networks, both for short-term forecasts as well as for the reconstruction of the underlying attractor. We suggest that possible strategies to overcome such limitations should be based on separating the smooth large-scale dynamics from the intermittent/small-scale features. We test these ideas on global sea-level pressure data for the past 40 years, a proxy of the atmospheric circulation dynamics. Better short- and long-term forecasts of sea-level pressure data can be obtained with an optimal choice of spatial coarse graining and time filtering.

Highlights

  • The advent of high-performance computing has paved the way for advanced analyses of high-dimensional datasets (Jordan and Mitchell, 2015; LeCun et al, 2015)

  • The motivation for this study came from the evidence that a straightforward application of ESNs to high-dimensional geophysical data does not yield the same result quality obtained by Pathak et al (2018) for the Lorenz 1963 and Kuramoto–Sivashinsky models

  • We stick to the Euler method for similarity to the climate data used in the last section of the paper

Read more

Summary

Introduction

The advent of high-performance computing has paved the way for advanced analyses of high-dimensional datasets (Jordan and Mitchell, 2015; LeCun et al, 2015) Those successes have naturally raised the question of whether it is possible to learn the behaviour of a dynamical system without resolving or even without knowing the underlying evolution equations. Evolution equations are difficult to solve for large systems such as geophysical flows, so that approximations and parameterizations are needed for meteorological and climatological applications (Buchanan, 2019). These difficulties are enhanced by those encountered in the modelling of phase transitions that lead to cloud formation and convection, which are major sources of uncertainty in climate modelling (Bony et al, 2015). Machine learning techniques capable of learning geophysical flow dynamics would help improve those approximations and avoid running costly simulations resolving explicitly all spatial/temporal scales

Objectives
Methods
Results
Conclusion
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call