Evaluation of a Stochastic Mixing Scheme in Kilometer-scale Simulations of Tropical Oceanic Deep Convection

Abstract:

Simulations of deep moist convection (DMC) at kilometer-scale horizontal grid spacings (∆_h) are known to produce an updraft size spectrum that is too narrow with sizes that are too large compared to observations and large eddy simulations. Updrafts that are too wide experience weak horizontal gradients of scalar quantities and vertical velocity which produce underprediction of updraft mixing and dilution that have the potential to bias predicted DMC properties. A method to introduce variable mixing has been developed through a stochastic framework implemented within the Weather Research and Forecasting model. This approach allows the horizontal diffusion coefficient within the Smagorinsky-type mixing scheme to vary in time and space with a prescribed spatiotemporal autocorrelation scale of the stochastic perturbations.

The stochastic mixing scheme is evaluated in ∆_h = 3 km simulations of a real-case tropical oceanic deep convection event that transitioned from discrete cellular convection into an organized mesoscale convective system. We analyze the sensitivity of precipitation structural evolution to the spatiotemporal autocorrelation scale of the stochastic perturbations in which one ensemble employs a relatively “short” scale (10 min and 10 km) while another employs a relatively “long” scale (~3 hrs and 300 km). In the “short” autocorrelation scale ensemble, all simulations systematically reduce total rainfall, increase domain-mean convective rain rates, increase the size of convective reflectivity cores, and decrease total precipitating area. This domain-wide systematic behavior suggests convection properties are more sensitive to enhanced rather than reduced mixing. Conversely, no systematic domain-wide changes from the control are evident in the ensemble employing the “long” autocorrelation scale. However, statistical correlations in these “long” autocorrelation scale simulations indicate that regions of net enhanced mixing over a limited-area subdomain produce less rainfall while regions of net reduced mixing produce more rainfall. Interestingly, this response is inconsistent with simulations employing domain-wide constant modifications to mixing, suggesting possible scale interactions between the stochastic autocorrelation scale and evolution of deep convection.