Two-Dimensional Convective Boundary Layer: Numerical Analysis and Echo State Network model

Florian Heyder, Juan Pedro Mellado, Jörg Schumacher

arXiv preprint · 13 August 2021

In one sentence. Echo state network trained on direct numerical simulations reproduces convective boundary-layer buoyancy flux statistics and generalizes to unseen flux ratios without solving the equations of motion.

Abstract

The numerical study of global atmospheric circulation processes requires the parametrization of turbulent buoyancy fluxes in the lower convective boundary layer which typically cannot be resolved by the coarse-scale computational grids. In the present work, a two-dimensional model of a shallow convective boundary layer in the Boussinesq limit is investigated by direct numerical simulations. A series of simulation runs evaluates the turbulent transport properties as a function of the ratio between the prescribed buoyancy fluxes at the top and the bottom of the layer. Our model is able to reproduce essential properties of the lower convective boundary layer. The resulting data records are subsequently used to train and test a recurrent neural network which is realized by an echo state network with a high-dimensional reservoir. It is shown that the echo state network reproduces the turbulence dynamics and the statistical properties of the buoyancy flux across the layer very well and is thus able to model these transport processes without solving the underlying highly nonlinear equations of motion. Focus is given to the generalization properties of the echo state network, i.e., its ability to model unseen data records with a different flux ratio.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.