Articles | Volume 13, issue 5
https://doi.org/10.5194/gmd-13-2185-2020
https://doi.org/10.5194/gmd-13-2185-2020
Development and technical paper
 | 
08 May 2020
Development and technical paper |  | 08 May 2020

Coupled online learning as a way to tackle instabilities and biases in neural network parameterizations: general algorithms and Lorenz 96 case study (v1.0)

Stephan Rasp

Data sets

Lorenz-Online S. Rasp https://github.com/raspstephan/Lorenz-Online

Model code and software

Lorenz-Online S. Rasp https://github.com/raspstephan/Lorenz-Online

Download
Short summary
Subgrid parameterizations are largely responsible for uncertainties in climate models. Recently, several studies tried to improve the representation of subgrid processes by learning parameterization directly from high-resolution modeling data. In this paper, the current state of the art of this research direction is summarized, and an algorithm is proposed to combat major problems with existing approaches, namely instabilities and biases.