Machine Learning Emulation of Gravity Wave Drag in Numerical Weather Forecasting.

Matthew Chantry,Tim Palmer,Sam Hatfield,Peter Dueben,Inna Polichtchouk

doi:10.1029/2021ms002477

Abstract

We assess the value of machine learning as an accelerator for the parameterization schemes of operational weather forecasting systems, specifically the parameterization of nonorographic gravity wave drag. Emulators of this scheme can be trained to produce stable and accurate results up to seasonal forecasting timescales. Generally, networks that are more complex produce emulators that are more accurate. By training on an increased complexity version of the existing parameterization scheme, we build emulators that produce more accurate forecasts. For medium range forecasting, we have found evidence that our emulators are more accurate than the version of the parametrization scheme that is used for operational predictions. Using the current operational CPU hardware, our emulators have a similar computational cost to the existing scheme, but are heavily limited by data movement. On GPU hardware, our emulators perform 10 times faster than the existing scheme on a CPU.

Highlights

Numerical weather prediction has a proud history of saving lives and protecting property in societies vulnerable to extremes of weather
We assess the value of machine learning as an accelerator for the parameterization schemes of operational weather forecasting systems, the parameterization of nonorographic gravity wave drag
Our networks have similar computational cost to the existing scheme when coupled to the Integrated Forecasting System (IFS) on a CPU-based architecture

Summary

Introduction

Numerical weather prediction has a proud history of saving lives and protecting property in societies vulnerable to extremes of weather. As these extremes become more extreme still under the influence of climate change, it is important that numerical weather prediction systems improve further. One way to enhance the skill of numerical weather prediction is to increase model resolution. Will higher resolution directly enhance the ability of models to simulate small-scale extreme events, and enables the information in observations to be better assimilated at initial time, and will reduce the dependence of models on inaccurate parameterized processes, and in this way will reduce systematic errors (Palmer, 2020). We must find ways of improving the computational efficiency of our models, so that valuable computing resources can be targeted where they will be most effective

Objectives

Results

Discussion

Conclusion