Quantifying 3D Gravity Wave Drag in a Library of Convection-Permitting Simulations: What Do We Need for Data-driven Parameterization of Convective Gravity Waves?

Atmospheric gravity waves (GWs) exhibit a wide range of length scales that cannot be directly resolved by general circulation models (GCMs). Yet, they play critical role in the Earth system. Consequently, parameterization schemes are employed to represent these under-resolved and unresolved GWs in GCMs. Recently, machine learning (ML) techniques have emerged as promising approaches for developing such parameterization schemes. In the widely used supervised learning approach, accurate extraction of the true representation of gravity wave forcing (GWF) terms from high-fidelity data, such as GW-resolving simulations, is crucial for achieving high-quality ML-based parameterizations. However, this task is challenging, and the effectiveness and accuracy of the estimated GWF directly influence the performance of the ML-based parameterizations, and hence, the GCMs.

In this study, we compare three methods for extracting three-dimensional GW fluxes and the resulting GWF from a library of convection-permitting simulations, based on Helmholtz decomposition and spatial filtering techniques, enabling the computation of the Reynolds stress and full sub-filter scale stress. Unlike previous studies that focused on vertical fluxes caused by GWs, we also evaluate the contribution of lateral momentum fluxes to the overall GWF. Our results reveal that horizontal momentum fluxes can have a significant impact on the total GWF, thereby influencing the large-scale circulation within GCMs. Furthermore, we show that it is essential to consider the interactions between missing GWs and the flow in GCMs when estimating GWF. To inform the development of data-driven parameterizations, we also explore the sensitivity of our findings to different filter types and length scales, which depend on the resolution of the GCM. These findings underscore the importance of a scale-aware perspective in the development of data-driven parameterizations, promoting a sub-filter scale mindset rather than a sub-grid scale mindset when addressing the missing driving forcing of GCMs.

Reference:

Sun, Y. Q., Hassanzadeh, P., Alexander, M. J., & Kruse, C. G. (2023). Quantifying 3D gravity wave drag in a library of tropical convection-permitting simulations for data-driven parameterizations. Journal of Advances in Modeling Earth Systems, 15, e2022MS003585. https://doi.org/10.1029/2022MS003585