Climate Processes in CMIP5: Deep Convective Onset Statistics as Process-Oriented Diagnostics for Climate Models

Deep convection plays a crucial role in tropical climate. The physical processes that control the transition from shallow to deep convection are known in principle but constraining them with sufficient accuracy to improve the representation of these in climate models remains challenging. A set of convective onset statistics, including precipitation pickup, occurrence probability, and the onset boundary on an empirical thermodynamic surface discussed below, can be used as process-oriented diagnostics for climate models. Observed statistics of deep convective onset show a sharp pickup in precipitation beyond a threshold value of column water vapor, referred to as the critical value. The onset and the onset boundary on a temperature-water-vapor thermodynamic plane can be compared to similar statistics from climate model output (although this requires supplementary fields not included in the standard CMIP5 archive). Plume computations coupled to an updraft velocity equation help to show how the onset behavior (e.g., as inferred from isolines of entraining convective available potential energy on the aforementioned empirical thermodynamic plane), depends on the entrainment assumptions. Only the plumes with sufficient entrainment through the lower troposphere produce deep convection onset characteristics similar to observations.

Both high resolution (0.5°) uncoupled version of the Community Climate System Model (CCSM) and the standard 1° CMIP5 resolution of the coupled version show good agreement with the observed onset statistics, strong entrainment in the deep convection scheme being the key. Numerical experiments altering the entrainment in the convective instability computations of the convection scheme of the model do very poorly at capturing the deep convective onset boundary at low entrainment, confirming the role inferred from the diagnostics and plume model. This helps to illustrate the role of process models in informing process-oriented diagnostics, as well as the role of direct numerical experimentation in confirming process postulates.