Implementation of a machine-learned gas optics parameterization in the
ECMWF Integrated Forecasting System
Abstract
Radiative transfer parameterizations are physically important but
computationally expensive components of weather and climate models. In
previous work, it was demonstrated that the gas optics module of a
radiation scheme, which traditionally rely on look-up-tables, can be
replaced with neural networks (NN) to improve speed while retaining a
high degree of accuracy. However, the evaluation of the NN version of
the RRTMGP gas optics scheme (RRTMGP-NN) was based solely on offline
radiation computations. In this paper, we describe the implementation
and prognostic evaluation of RRTMGP-NN in the Integrated Forecasting
System (IFS) of the European Centre for Medium-Range Weather Forecasts
(ECMWF). This was carried out by incorporating the gas optics scheme
into ecRad, the modular radiation scheme used in the IFS. New NN models
were trained on RRTMGP k-distributions with reduced spectral resolution.
A hybrid loss function helped reduce radiative forcing errors. Four
1-year coupled ocean-atmosphere simulations were performed for each
configuration. The results show that RRTMGP-NN and RRTMGP produce very
similar model climates, with the differences being smaller than those
between existing gas optics schemes, and statistically insignificant for
zonal means of single-level quantities such as surface temperature. The
use of RRTMGP-NN speeds up the radiation scheme by roughly a third
compared to RRTMGP, and is also faster than the older and less accurate
RRTMG which is used in the current operational cycle of the IFS.