Large-eddy simulations with ClimateMachine v0.2.0: a new open-source code for atmospheric simulations on GPUs and CPUs

Sridhar, Akshay; Tissaoui, Yassine; Marras, Simone; Shen, Zhencai; Kawczynski, Charles; Byrne, Simon; Pamnany, Kiran; Waruszewski, Maciej; Gibson, Tom; Kozdon, Jeremy E.; Churavy, Valentin; Wilcox, Lucas C.; Giraldo, Francis X.; Schneider, Tapio

doi:10.5194/gmd-15-6259-2022

Cited by 15 publications

(16 citation statements)

References 90 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The a posteriori learning strategy clearly leads to much better scores. Topping, 2021; Ramadhan et al, 2020;Sridhar et al, 2021) naturally supports our contribution. Besides, deep differentiable emulators (Hatfield et al, 2021;Kasim et al, 2021;Nonnenmacher & Greenberg, 2021) that learn a differentiable approximation of a non-differentiable forward solver or of its adjoint may also open new avenues for the development of SGS parametrizations for state-of-the-art ESMs with a posteriori learning strategies.…”

Section: Discussionsupporting

confidence: 76%

A Posteriori Learning for Quasi‐Geostrophic Turbulence Parametrization

Frezat

Sommer

Fablet

et al. 2022

J Adv Model Earth Syst

View full text Add to dashboard Cite

The representation of unresolved processes is a key source of uncertainty in weather and climate models. Climate science and weather forecasting indeed heavily rely on numerical simulations of the Earth's atmosphere and oceans (Bauer et al., 2015;Neumann et al., 2019). But even the most advanced applications are currently far from resolving explicitly the wide variety of space-time scales and physical processes involved. This will likely remain the case for the foreseeable future because of the nonlinearity of fluid dynamics and thermodynamics, and because of the finite nature of computational resources (Fox-Kemper et al., 2014;. Weather and climate models will therefore keep relying on approximated representations of the effect of unresolved processes in the form of subgrid parametrization schemes (Fox-Kemper et al., 2019;Schneider, Lan, et al., 2017). Parametrization schemes accounting for the impact of turbulence in the atmosphere and oceans at various scales will in particular remain essential components of these models.

show abstract

Section: Discussionsupporting

confidence: 76%

A Posteriori Learning for Quasi‐Geostrophic Turbulence Parametrization

Frezat

Sommer

Fablet

et al. 2022

J Adv Model Earth Syst

View full text Add to dashboard Cite

show abstract

“…More specifically, St1 had an estimated top value close to 3 m 2 s −1 , whereas inner areas showed more moderate values (~1 m 2 s −1 ). Indeed, St1 is situated close to the lagoon mouth, at the ocean border, where the water flow is the most intense (the current intensity reaches peak values close to 1.5 m s −1 [27][28][29]). Applying the sensitivity analysis to the salinity showed that St1 presented the lowest sensitivity index relative to the baseline.…”

Section: Discussion and Concluding Remarksmentioning

confidence: 99%

“…In the end, to close the turbulence, the so-called Reynolds-stress tensor is defined [22][23][24]. The LES is a high-resolution turbulence model sensitive to the fine details of the equations and the meshes used to represent the flow [25][26][27][28][29][30]. The SAS introduces the von Karman length scale [31][32][33] to avoid dependency determining of the RANS/LES interface.…”

Section: Introductionmentioning

confidence: 99%

“…Figure3aand Figure3bpresent, respectively, the monthly time series for the salinity and temperature for June 2001.The salinity presents values close to the ocean ones, ~34 PSU, reflecting the low river flow regime and the tidal domination. The semi-diurnal oscillation is mostly evident for St2, which shows significant amplitude oscillation, ranging within(29)(30)(31)(32)(33)(34) PSU, between the peak and trough. The fortnightly oscillations are mostly evident for St1, St5, and ST6, with amplitude oscillation ranging within (32-34) PSU.…”

mentioning

confidence: 98%

See 1 more Smart Citation

Using Different Classic Turbulence Closure Models to Assess Salt and Temperature Modelling in a Lagunar System: A Sensitivity Study

Lopes

2022

JMSE

View full text Add to dashboard Cite

Turbulence modelling is an important issue when dealing with hydrodynamic and transport models for better simulation of the transport of dissolved or suspended substances in a body-water. It controls processes involving physical balances (salt and water temperature) and, therefore, the ecosystem equilibrium. The study arises from the need to model the turbulence more efficiently when dealing with extreme situations on the Ria de Aveiro (Portugal), a coastal lagoon shallow water system dominated by tidal transport. Because the turbulence model is coupled to the hydrodynamic and transport models, a correct estimation of the eddy viscosity is important in simulating the salt and the heat transports. The aim is to assess the performance of four turbulence schemes/models (k, k-ε, Smagorinsky’s, and k-ε/Smagorinsky’s (k-ε/Sma), where k is turbulent kinetic energy and ε the dissipation rate of the turbulent kinetic energy) associated to a coupled hydrodynamic and transport models to simulate the eddy viscosity, the salinity, and the temperature. Overall, the results point out that among the different models/schemes used, the is the one which provides a more realistic value of the eddy viscosity within the range (1–6) m2 s−1, but most probably (1–3) m2 s−1. The application of the sensitivity analysis to some non-universal k-ε/Sma parameters evidenced significant sensitivity for the eddy viscosity and the salinity and moderate sensitivity for the water temperature. A 100% adjustment of the parameter values relative to the reference, translated into variations within the range of (1, 4) m2 s−1, (0, 13) PSU, and (1, 2.20) °C, for the eddy viscosity, salinity, and water temperature, respectively.

show abstract

“…We test the new implementation for both continuous and discontinuous elements (e.g., see [29] for how this can be achieved in the same source code). Other models that use either continuous or discontinuous spectral elements for atmospheric flows are, e.g., CESM2 [15], E3SM [5], both via the CAM-SE dycore [16], and ClimateMachine [30].…”

Section: Introductionmentioning

confidence: 99%