A three-dimensional nonhydrostatic mesoscale model, the Penn State University (PSU) / National Center for Atmospheric Research (NCAR) MM5,is used to simulate the evolution of convective systems over the IFA (Intensive Flux Array) during TOGA COARE (Tropical Ocean Global Atmosphere Coupled Ocean-Atmosphere Response Experiment) 19-26 December, 1992. The model is driven by a time-varying `IFA mean forcing' based on the average advective tendencies of temperature and moisture over the IFA. The domain-averaged horizontal wind is kept close to the observed IFA mean using Newtonian relaxation. Periodic lateral boundary conditions are imposed. Simulations with three horizontal grid spacing, 2 km, 15 km and 60 km, are conducted. With 15 km and 60 km resolution, subgrid-scale cumulus convection is parameterized while mesoscale convective organization is explicitly resolved over a 600 km x 600 km domain. With 2 km resolution, convection is fully resolved over a 210 km x 210 km domain.
Despite their different horizontal resolution and different treatment of moist convection, the simulations all produce very similar temporal variability in domain-averaged temperature and relative humidity profiles. They also closely resemble each other in various statistical properties of convective systems. A comprehensive comparison of the 15 km and 2 km model results against observations is performed. The domain-averaged cloud amount and precipitation agree well with observations. Some shortcomings are noted. During suppressed convective periods, the model tends to have greater areal coverage of rainfall and more cirrus anvil clouds than observed. Over the eight-day period, both models produce mean temperature drifts about 2 K colder than observed. A histogram of modeled cloud top temperature captures the observed breaks between convective episodes but shows excessive and persistent cold cirrus clouds. A radar reflectivity histogram shows that the 15 km model slightly overpredicts radar reflectivity, and that the 2 km model has too high and temporally homogeneous reflectivities. The model-simulated cloud cluster size is somewhat smaller than the observed. Surface sensible and latent heat fluxes are overestimated by 50-100%, due both to shortcomings in the surface flux calculations in the model and model-produced mean temperature and humidity biases. Downwelling solar flux at surface is underestimated mainly because of the simple shortwave radiation scheme.
This study suggests that large-domain simulations using the MM5 with 15 km resolution can be a useful tool for further study of tropical convective organization and its interaction with large- scale circulation.