# BPCPPDYSOE21 – wissenschaftliches Programm

## Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

# DY: Fachverband Dynamik und Statistische Physik

## DY 2: Fluid Physics 1 - organized by Stephan Weiss and Michael Wilczek (Göttingen)

### DY 2.2: Vortrag

### Montag, 22. März 2021, 09:20–09:40, DYa

**Reservoir Computing of Dry and Moist Turbulent Rayleigh- Bénard Convection** — •Florian Heyder, Sandeep Pandey, and Jörg Schumacher — TU Ilmenau, Ilmenau, Germany

Reservoir Computing (RC) is one efficient implementation of a recurrent neural network that can describe the evolution of a dynamical system by supervised machine learning without solving the underlying nonlinear partial differential equations. We apply such a neural network to approximate the large-scale evolution and the resulting low-order turbulence statistics of two-dimensional dry and moist Rayleigh-Bénard convection. We acquire training and test data by long-term direct numerical simulations (DNS). They are postprocessed by a Proper Orthogonal Decomposition (POD) with the snapshot method. The training data comprise time series of the first 150 POD modes, which are associated with the largest total energy amplitudes and thus the large-scale structure of the flows. Feeding the data to the Reservoir Computing model and optimizing the reservoir parameters results in predictions for the evolution of the dry and moist convection flows. The prediction capabilities of our model are comprehensively tested by a comparison with DNS and test data, the latter of which are reconstructed from the most energetic POD modes. Vertical profiles of mean thermodynamic fields and their mean vertical transport show good agreement. We find that RC is capable to model the large-scale structure and low-order statistics of dry and moist turbulent convection. This shows potential for subgrid-scale turbulence parameterization in large-scale atmospheric circulation models.