A data-driven multi-cloud model for stochastic parametrization of deep convection

Area of Science:

• Atmospheric Science
• Climate Modeling
• Computational Fluid Dynamics

Background:

• General circulation models (GCMs) struggle with medium-term errors and lack of variability due to unresolved subgrid-scale processes.
• Deep convection representation is a persistent challenge, hindering accurate climate predictions.
• Stochastic parametrizations offer a promising approach to address these limitations.

Purpose of the Study:

• To construct a data-driven stochastic subgrid model for deep convection.
• To emulate computationally expensive deep convection-resolving large-eddy simulations (LESs).
• To improve the representation of cloud variability and spatial distribution in climate models.

Main Methods:

• Utilized a data-driven stochastic parametrization methodology.
• Developed a stochastic model based on a finite number of discrete cloud states.
• Employed Markov chains to model transitions between cloud states, conditioned on large-scale variables.
• Introduced local spatial coupling into Markov chains, creating stochastic cellular automata.

Main Results:

• The developed model emulates LES data in a computationally efficient manner.
• Conditional Markov chains accurately reproduce the time evolution of cloud fractions.
• Introducing local spatial coupling enhanced the fidelity of cloud type variability and spatial distribution compared to LES data.

Conclusions:

• The data-driven stochastic subgrid model effectively captures deep convection processes.
• Spatially coupled Markov chains (stochastic cellular automata) provide a more faithful representation of subgrid-scale cloud dynamics.
• This approach offers a computationally inexpensive yet accurate method for improving climate model performance.