Vincent E. Larson

The Coriolis effect can deflect turbulent momentum flux and anisotropy of turbulent kinetic energy. This Coriolis turbulent deflection is usually omitted in parameterizations of atmospheric turbulence and convection. However, previous large‐eddy simulations (LES) found that the nontraditional Coriolis terms (NCTs) can enhance vertical velocity variance with easterly flow and suppress it with westerly flow. To parameterize this effect in the atmosphere, this study introduces the Coriolis terms to the Cloud Layers Unified By Binormals (CLUBB) parameterization of clouds and turbulence. We design three single‐column benchmarks to test the implementation of the NCTs in CLUBB. First, assuming that all except the Coriolis terms vanish on the right‐hand side of the prognostic equations of the second‐order moments of velocity, we derive an analytical solution in which the anisotropy of the moments rotates back to the initial state with a certain period. Our CLUBB results match this analytic benchmark. Second, assuming dry and neutrally stratified atmosphere with constant geostrophic winds, our CLUBB and LES results suggest that the NCTs make the Ekman layer deeper with geostrophic easterlies but shallower with westerlies. Third, under realistic conditions for trade‐wind shallow convection, our CLUBB results show that the cloud layer becomes shallower if we omit the NCTs or flip the geostrophic winds from easterly to westerly. These benchmarks not only validate the Coriolis terms in CLUBB but also form a model hierarchy that aids our understanding of the effects of the NCTs on turbulence and convection.

The Coriolis effect can deflect turbulent momentum flux and anisotropy of turbulent kinetic energy. This Coriolis turbulent deflection is usually omitted in parameterizations of atmospheric turbulence and convection. However, previous large-eddy simulations (LES) found that the nontraditional Coriolis terms (NCTs) can enhance vertical velocity variance with easterly flow and suppress it with westerly flow. To parameterize this effect in the atmosphere, this study introduces the Coriolis terms to the Cloud Layers Unified By Binormals (CLUBB) parameterization of clouds and turbulence. We design three single-column benchmarks to test the implementation of the NCTs in CLUBB. First, assuming that all except the Coriolis terms vanish on the right-hand side of the prognostic equations of the second-order moments of velocity, we derive an analytical solution in which the anisotropy of the moments rotates back to the initial state with a certain period. Our CLUBB results match this analytic benchmark. Second, assuming dry and neutrally stratified atmosphere with constant geostrophic winds, our CLUBB and LES results suggest that the NCTs make the Ekman layer deeper with geostrophic easterlies but shallower with westerlies. Third, under realistic conditions for trade-wind shallow convection, our CLUBB results show that the cloud layer becomes shallower if we omit the NCTs or flip the geostrophic winds from easterly to westerly. These benchmarks not only validate the Coriolis terms in CLUBB but also form a model hierarchy that aids our understanding of the effects of the NCTs on turbulence and convection.

Parameterizing boundary layer turbulence in Earth System Models (ESMs) introduces a key source of uncertainty. Here, we explore the impact of implementing a prognostic formulation (PMF) for momentum flux, which permits countergradient (CG) fluxes. We implement PMF in the turbulence parameterization in the Community Atmosphere Model version 6 (CAM6). CG fluxes in PMF reduce globally‐integrated momentum flux in CAM6. A similar response in globally‐integrated momentum flux is observed when reducing a tunable diffusivity parameter in the existing diagnostic formulation for momentum flux (DMF). Both methods of reducing effective eddy diffusivity produce qualitatively similar responses in the modeled general circulation, namely an enhanced upward branch of the Hadley cell, increased equatorward low‐level flow, and strengthened meridional temperature and moisture gradients. However, the reduction in effective eddy diffusivity from CG fluxes in PMF produces a higher‐magnitude response in the general circulation compared to reducing the diffusivity parameter in DMF.

In this study, we extend the Cloud Layers Unified by Binormals (CLUBB) turbulence scheme within the GFDL atmospheric model (AM4) by implementing direct momentum‐flux prognosis and a multiscale turbulent lengthscale, to improve the simulation of nocturnal precipitation and associated Low‐Level Jets (LLJs) over the Great Plains (GP). Toward this aim, we set up four AM4‐CLUBB configurations: diagnosed momentum flux, prognosed momentum flux, diagnosed momentum flux with a multiscale turbulent lengthscale, and prognosed momentum flux with a multiscale turbulent lengthscale. Simulations are evaluated against the AM4 control, the Integrated Multi‐satellitE Retrievals for GPM (IMERG), and the Doppler wind radar profiles from the Atmospheric Radiation Measurement program. Results show that all AM4‐CLUBB configurations improve the precipitation timing from the unrealistic midday peak seen in the AM4 control simulation toward the satellite‐observed nocturnal maximum. The configuration that prognoses momentum flux and uses a multi‐scale turbulent lengthscale, best matches the timing and intensity of GP precipitation rate. This configuration is also that which more accurately simulates the ARM‐observed nocturnal LLJ wind profiles, while increasing the frequency of counter‐gradient momentum fluxes near the LLJ core compared to prognosing momentum fluxes with the original AM4‐CLUBB turbulent lengthscale. Momentum budget analysis attributes this increase to a nearly fivefold enhancement in the buoyancy production term when using the multiscale formulation, and leads to stronger nocturnal convective activity, as diagnosed from the greater vertical velocity skewness and plume asymmetry.

Graphics Processing Units (GPUs) are becoming increasingly central to high-performance computing (HPC), but fair comparison with central processing units (CPUs) remains challenging, particularly for applications that can be subdivided into smaller workloads. Traditional metrics such as speedup ratios can overstate GPU advantages and obscure the conditions under which CPUs are competitive, as they depend strongly on workload choice. We introduce two peak-based performance metrics, the Peak Ratio Crossover (PRC) and the Peak-to-Peak Ratio (PPR), which provide clearer comparisons by accounting for the best achievable performance of each device. Using the performance of the Cloud Layers Unified by Binormals (CLUBB) standalone model as a case study, we demonstrate these metrics in practice, show how they can guide execution strategy, and examine how they shift under factors that affect workload. We further analyze how implementation choices and code structure influence these metrics, showing how they enable performance comparisons to be expressed in a concise and actionable way, while also helping identify which optimization efforts should be prioritized to meet different performance goals.

We extend the Cloud Layers Unified by Binormals (CLUBB) turbulence scheme within the GFDL atmospheric model (AM4) by implementing direct momentum-flux prognosis and a multiscale turbulent length scale, to improve the simulation of nocturnal precipitation and associated Low-Level Jets (LLJs) over the Great Plains (GP). Towards this aim, we set up four AM4-CLUBB configurations: diagnosed momentum flux, prognosed momentum flux, diagnosed momentum flux with a multiscale turbulent lengtshcale, and prognosed momentum flux with a multiscale turbulent lengtshcale. Simulations are evaluated against the AM4 control, the Integrated Multi-satellitE Retrievals for GPM (IMERG), and the Doppler wind radar profiles from the Atmospheric Radiation Measurement (ARM) program. Results show that all AM4-CLUBB configurations improve the precipitation timing from the unrealistic midday peak seen in the AM4 control simulation toward the satellite-observed nocturnal maximum. The configuration that prognoses momentum flux and uses a multi-scale turbulent length scale, best matches the timing and intensity of GP precipitation rate. This configuration is also that which more accurately simulates the ARM-observed nocturnal LLJ wind profiles, while increasing the frequency of counter-gradient momentum fluxes near the LLJ core compared to prognosing momentum fluxes with the original AM4-CLUBB turbulent lengthscale. Momentum budget analysis attributes this increase to a nearly fivefold enhancement in the buoyancy production term when using the multiscale formulation, and leads to stronger nocturnal convective activity, as diagnosed from the greater vertical velocity skewness and plume asymmetry.

Accurately simulating tropical cyclones (TCs) in global climate models remains a key challenge due to not only multiscale interactions that govern storm genesis, intensity, and structure, but computational constraints that limit model grid spacing. Here we evaluate TC representation in a development version of the Community Atmosphere Model, version 7 (CAM7), which introduces several major updates, including higher vertical resolution, a revised Zhang-McFarlane deep convection scheme, and a new prognostic formulation for turbulent momentum fluxes in the boundary-layer scheme CLUBB. Using a suite of globally-uniform and variable-resolution simulations at 0.25deg grid spacing, we assess both large-scale statistics (global and basinwise TC frequency) and storm-scale characteristics (inflow angle, inflow depth, and wind structure).CAM7 with prognostic momentum fluxes produces improved spatial patterns of TC activity and more realistic intensity metrics compared to prior CAM generations. However, default configurations overproduce storms by roughly a factor of two. By increasing the parameterized CAPE consumption by deep convection during TC genesis, we achieve a ~40% reduction in global TC counts and improved agreement with observed basin distributions. At the storm scale, we reduce boundary-layer diffusivity, leading to stronger tangential winds, larger inflow angles, and shallower inflow layers, consistent with idealized f-plane sensitivity experiments and more in line with both observations and large-eddy simulations.These results demonstrate that targeted parameter tuning in deep convection and boundary-layer turbulence schemes can substantially improve both the frequency and structure of simulated TCs in CAM7, advancing its capability for high-resolution climate and weather prediction applications.

Graphics Processing Units (GPUs) are becoming increasingly central to high-performance computing (HPC), but fair comparison with central processing units (CPUs) remains challenging, particularly for applications that can be subdivided into smaller workloads. Traditional metrics such as speedup ratios can overstate GPU advantages and obscure the conditions under which CPUs are competitive, as they depend strongly on workload choice. We introduce two peak-based performance metrics, the Peak Ratio Crossover (PRC) and the Peak-to-Peak Ratio (PPR) which provide clearer comparisons by accounting for the best achievable performance of each device. Using a case study into the performance of the Cloud Layers Unified by Binormals (CLUBB) standalone model, we demonstrate these metrics in practice, show how they can guide execution strategy, and examine how they shift under factors that affect workload. We further analyze how implementation choices and code structure influence these metrics, showing how they enable performance comparisons to be expressed in a concise and actionable way, while also helping identify which optimization efforts should be prioritized to meet different performance goals.

Graphics Processing Units (GPUs) are becoming increasingly central to high-performance computing (HPC), but fair comparison with central processing units (CPUs) remains challenging, particularly for applications that can be subdivided into smaller workloads. Traditional metrics such as speedup ratios can overstate GPU advantages and obscure the conditions under which CPUs are competitive, as they depend strongly on workload choice. We introduce two peak-based performance metrics, the Peak Ratio Crossover (PRC) and the Peak-to-Peak Ratio (PPR) which provide clearer comparisons by accounting for the best achievable performance of each device. Using a case study into the performance of the Cloud Layers Unified by Binormals (CLUBB) standalone model, we demonstrate these metrics in practice, show how they can guide execution strategy, and examine how they shift under factors that affect workload. We further analyze how implementation choices and code structure influence these metrics, showing how they enable performance comparisons to be expressed in a concise and actionable way, while also helping identify which optimization efforts should be prioritized to meet different performance goals.