A number of studies have considered the occurrence of intense tropical vortices ("tropical storms") in low, moderate and high resolution GCMs over the past several decades. Recently, Vitart et al. (1997) analyzed the structure, behavior and interannual variability of such storms in a 10-year, 9 member ensemble simulation using an AGCM developed at GFDL, configured at T42L18 resolution. Though their results show some biases and other problems related to the model's ability to simulate real tropical storms, their findings clearly suggest that GCMs could be brought to bear on the important problem of seasonal prediction of tropical storm activity in some key regions - for example, the tropical north Atlantic. We are in the process of examining the behavior of a similarly configured GCM (the ECHAM3 model from the Max Planck Institute for Meteorology configured at T42L19 resolution) in a 10-member ensemble covering the period 1950-98.
Initial results for 5 members over 5 years (1982, 1983, 1988, 1990, and 1995) are quite promising. It is suggested that ensemble forecasts of the tropical storm season configured at much higher resolution (e.g., T106; see Bengtsson, 1995) would likely produce even more useful results. However, such utility can only be realized if the necessary extended retrospective ensemble simulations can be performed and rigorously analyzed for validation.