Spontaneous aggregation of deep convection is a common feature of idealized numerical simulations of the tropical atmosphere in Radiative-Convective Equilibrium (RCE). However, at coarse grid resolution where deep convection is not fully resolved, the occurrence of this phenomenon is extremely sensitive to subgrid-scale processes. This study focuses on the role played by mixing and entrainment, either provided by the turbulence model or the implicit numerical dissipation. We have analyzed the results of two different models, WRF and SAM, and compared different configurations by varying the turbulence models, the initial conditions and the horizontal spatial resolution. At coarse grid resolution (3 km), the removal of turbulent mixing prevents the occurrence of Convective Self-Aggregation (CSA) in models with low numerical diffusivity, while it is preserved in models with high numerical diffusivity. When the horizontal grid resolution is refined to 1 km (thus reducing the implicit numerical dissipation), CSA is only achieved by increasing the explicit turbulent mixing. In this case, CSA was found to occur even with a small amount of shallow clouds. Therefore, this study suggests that the sensitivity of CSA to horizontal grid resolution is not primarily due to the corresponding decrease in shallow clouds. Instead, it is found that turbulent mixing and dissipation at small scales regulate the amplitude of initial humidity perturbations introduced by convection in the free troposphere: the greater the dissipation at small scales, the greater the size and the strength of humidity perturbations in the free troposphere that can destabilize the RCE state.

Spontaneous aggregation of deep convection is a common feature of idealized numerical simulations of the tropical atmosphere in Radiative-Convective Equilibrium (RCE). However, at coarse grid resolution where deep convection is not fully resolved, the occurrence of this phenomenon is extremely sensitive to subgrid-scale processes. This study focuses on the role played by mixing and entrainment, either provided by the turbulence model or the implicit numerical dissipation. We have analyzed the results of two different models, WRF and SAM, and we have compared different configurations by varying the turbulence models, the numerical schemes and the horizontal spatial resolution. At coarse grid resolution (3 km), removing turbulent mixing prevents the occurrence of Convective Self-Aggregation (CSA) in low numerical diffusion models, while delaying it in high numerical diffusion models. When the horizontal grid resolution is refined to 1 km (thus reducing the implicit numerical dissipation), CSA is achieved only by increasing the explicit turbulent mixing. In this case, CSA was found to occur even with a small amount of shallow clouds. Therefore, this study suggests that the sensitivity of CSA to horizontal grid resolution is not primarily due to the corresponding decrease in shallow clouds. Instead, it is found that turbulent mixing and dissipation at small scales regulate the amplitude of humidity perturbations introduced by convection in the free troposphere: the greater the dissipation at small scales, the greater the size and the strength of humidity perturbations in the free troposphere that can destabilize the RCE state.

The Radiative-Convective Equilibrium (RCE) of two models exhibiting convective aggregation has been compared. The goal of the work, following the suggestion from the Radiative-Convective Equilibrium Model Intercomparison Project (RCEMIP), is to identify key parameters controlling self-aggregation in RCE for both models and discuss the processes controlled by these parameters in order to find the simulations similarities and to test their differences. The two models studied, the SAM (System for Atmospheric Modeling) and the ARPS (Advanced Regional Prediction System), have different physical and numerical formulations. This allowed us to compare the sensitivity to processes related to self-aggregation. When self-aggregation occurs, the two models present similar statistics for what concerns precipitation, warming, and drying of the atmosphere and anvil cloud area reduction (leading to an “Iris effect’), within the spread of the RCEMIP values. On the other hand, they differ both in the degree of organization and the organization feedback: SAM is strongly organized (is on the highest quartile of the RCEMIP for the Iorg Index) and the convective organization is achieved by cloud-radiative feedback; ARPS is weakly organized (on the multi-model average of the RCEMIP for the Iorg Index) and the moisture-convection feedback is leading to the convective organization. The prevalence of one mechanism over the other has been found in the interaction between the microphysics and the sub-cloud layer properties. This comparison suggests that, in order to have a robust measure of climate sensitivity, climate models should include both types of convective organization mechanisms as shown by the two models.