Implementation of a machine-learned gas optics parameterization  in the ECMWF Integrated Forecasting System: RRTMGP-NN 2.0

Ukkonen, Peter; Hogan, Robin J.

doi:https://doi.org/10.5194/gmd-16-3241-2023

Articles | Volume 16, issue 11

https://doi.org/10.5194/gmd-16-3241-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/gmd-16-3241-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 16, issue 11

Development and technical paper

|

09 Jun 2023

Development and technical paper |

| 09 Jun 2023

Implementation of a machine-learned gas optics parameterization in the ECMWF Integrated Forecasting System: RRTMGP-NN 2.0

Peter Ukkonen and Robin J. Hogan

Model code and software

peterukk/rte-rrtmgp-nn: 2.0 P. Ukkonen https://doi.org/10.5281/zenodo.7413935

Code and extensive data for training neural networks for radiation, used in "Implementation of a machine-learned gas optics parameterization in the ECMWF Integrated Forecasting System: RRTMGP-NN 2.0" P. Ukkonen https://doi.org/10.5281/zenodo.6576680

Optimized version of the ecRad radiation scheme with new RRTMGP-NN gas optics P. Ukkonen https://doi.org/10.5281/zenodo.7148329

Short summary

Climate and weather models suffer from uncertainties resulting from approximated processes. Solar and thermal radiation is one example, as it is computationally too costly to simulate precisely. This has led to attempts to replace radiation codes based on physical equations with neural networks (NNs) that are faster but uncertain. In this paper we use global weather simulations to demonstrate that a middle-ground approach of using NNs only to predict optical properties is accurate and reliable.