Abstract

Several different cumulus parameterizations are compared in a 10-day regional model simulation over the tropical Americas in northern summer. A simple bulk diagnostic test is devised, comparing the model's preferred domain-mean wind divergence profile with “observed” drivergence. The latter is ob... Show moreSeveral different cumulus parameterizations are compared in a 10-day regional model simulation over the tropical Americas in northern summer. A simple bulk diagnostic test is devised, comparing the model's preferred domain-mean wind divergence profile with “observed” drivergence. The latter is obtained by a line integral of the normal wind component at the model's outer boundary, from the ECMWF reanalysis data used as lateral boundary conditions. The former is obtained from a line integral one grid point in from the boundary, a perimeter that encloses almost exactly the same region. Even though the model fields near the boundary are strongly nudged toward the ECMWF values, the difference is distinct, and indicative of systematic errors in the model's heating field throughout the interior of the domain. Heating reflects the effects of the convection scheme, both direct and indirect (e.g., through its impact on resolved condensation). A useful axis along which to characterize schemes appears to be overactive versus underactive. Underactive convective schemes tend to produce too little low-level convergence and upper-level divergence, while overactive schemes produce too much. This categorization is also reflected in rainfall fields, as overactive schemes produce widespread light convective rain while underactive schemes produce sparse occasional storms. For example, the Kain–Fritsch scheme is overactive with its default entraining-plume radius of 1500 m, a value optimized for midlatitudes over land. A value of 750 m makes the regional divergence magnitude about right, but makes the upper-tropospheric outflow altitude too low, illustrating a classic dilemma of entraining-plume models of convection. Schemes with other conceptual structures give widely varying divergence errors. The largest errors are found with the Anthes–Kuo scheme, while the smallest errors are found with the Betts–Miller–Janjic scheme, which has no consistent divergence bias over time. Diagnosis of other North American monsoon simulations supports the general underactive/ overactive characterization, but shows that the best scheme and parameters may depend on weather regime. Show less