Accelerating Radiation Computations for Dynamical Models With Targeted Machine Learning and Code Optimization. (2nd December 2020)

• Record Type: Journal Article
• Authors: Ukkonen, Peter; Pincus, Robert; Hogan, Robin J.; Pagh Nielsen, Kristian; Kaas, Eigil
• Is Part Of: Journal of advances in modeling earth systems. Volume 12:Number 12(2020)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-12-02
• Subjects: machine learning -- radiation -- atmospheric model -- parameterization; Geological modeling -- Periodicals; Climatology -- Periodicals; Geochemical modeling -- Periodicals; 551.5011
• DOI: 10.1029/2020MS002226
• Languages: English
• ISSNs: 1942-2466

Abstract

Abstract: Atmospheric radiation is the main driver of weather and climate, yet due to a complicated absorption spectrum, the precise treatment of radiative transfer in numerical weather and climate models is computationally unfeasible. Radiation parameterizations need to maximize computational efficiency as well as accuracy, and for predicting the future climate many greenhouse gases need to be included. In this work, neural networks (NNs) were developed to replace the gas optics computations in a modern radiation scheme (RTE+RRTMGP) by using carefully constructed models and training data. The NNs, implemented in Fortran and utilizing BLAS for batched inference, are faster by a factor of 1–6, depending on the software and hardware platforms. We combined the accelerated gas optics with a refactored radiative transfer solver, resulting in clear‐sky longwave (shortwave) fluxes being 3.5 (1.8) faster to compute on an Intel platform. The accuracy, evaluated with benchmark line‐by‐line computations across a large range of atmospheric conditions, is very similar to the original scheme with errors in heating rates and top‐of‐atmosphere radiative forcings typically below 0.1 K day −1 and 0.5 W m −2, respectively. These results show that targeted machine learning, code restructuring techniques, and the use of numerical libraries can yield material gains in efficiency while retaining accuracy. Plain Language Summary: Solar and terrestrial radiation interact with Earth's atmosphere, surface, and clouds and provide the energy which drives climate and weather. Simulating these radiative flows in climate and weather models is crucial and can also be very time‐consuming. One possible way to model radiative effects more efficiently is to use neural networks or similar machine learning algorithms, but predictions are not guaranteed to be realistic because such models do not use physical equations. Here we investigate using neural networks to replace only one part of traditional radiation code, where the optical properties of the atmosphere are computed. We have found that this approach can be several times faster, while still being accurate in various situations, such as simulating future climate. Key Points: Neural networks (NNs) were trained to predict the optical properties of the gaseous atmosphere Training data were generated with a recently developed radiation scheme for dynamical models (RRTMGP) RRTMGP‐NN is roughly 3 times faster than the reference code and has a similar accuracy, also in future climate scenarios