3. RACORO-FASTER: Single-Column Model Simulations and Bias Attribution

Description

Climatically important low-level clouds can be both underestimated and misrepresented by climate models. The FAst-physics System TEstbed and Research (FASTER) project has constructed three cases over the Atmospheric Radiation Measurement (ARM) Climate Research Facility’s Southern Great Plain site during the RACORO Campaign [Routine ARM Aerial Facility (AAF) Clouds with Low Optical Water Depth (CLOWD) Optical Radiative Observations] to facilitate the research of model representation of boundary-layer clouds (see the Vogelmann et al. poster). Simulations using the single-column Community Atmosphere Model version 5 (SCAM5) of the multi-day continental shallow cumulus case exhibit the same traits of underestimation and misrepresentation; the causes are investigated in this work. To address the uncertainty associated with the forcing data, multiple observationally constrained large-scale forcings are used to drive the SCAM5 simulations. The model’s shallow cumulus convection scheme, which is intended for this very cloud process, tends to significantly under-produce clouds during the time when shallow cumulus activity prevails in the observations, while other physical schemes in the model have a strong tendency to misrepresent (over-trigger) the low-level clouds throughout the day. Large-eddy simulations that are driven by the same large-scale forcings (see the Endo et al. poster) can reasonably capture the shallow cumulus activity, and are used to investigate the links between model biases and the underlying assumptions of the shallow cumulus scheme used in the model. It is found that the weaker boundary layer turbulence simulated is directly responsible for the weaker cumulus activity. A shallower model boundary layer along with insufficient convective ventilation results in excessive stratiform cloud production, especially during nighttime. The results manifest the challenge of having a modularized treatment of intimately related physical processes for cloud representation, since biases in one module can propagate into and become amplified in other modules.