There is accumulating evidence that the question of tropical cyclogenesis is decided by convective features and mesoscale processes. The lack of observational data of these processes over the oceans and insufficient model parameterizations make skillful dynamical model forecasts of tropical cyclogenesis difficult. This paper explores the development of a statistical model that would provide near real-time forecasts of tropical cyclogenesis.

In order to develop such an operational forecast system, data must be regularly available over the entire region of interest. Furthermore, the data must be of high enough quality to find the important differences between developing and non-developing events. These requirements dictate that any operational tropical cyclogenesis model must evaluate the large-scale environment of the incipient cloud cluster, since high-quality mesoscale data is not regularly available. Most crucial to the system is an analysis algorithm that exploits those differences to yield a skillful forecast. It appears that a non-linear classification scheme, such as a neural network, will provide a viable method for data analysis.

All developing and non-developing cloud clusters that formed during the 1998-2000 Atlantic hurricane seasons are examined. Large-scale environmental predictors are derived from the NCEP-NCAR reanalysis as a training set. A simple linear discriminant analysis produces a 75% accurate classification if only three of the more important large-scale predictors (latitude, surface relative vorticity tendency, and the daily genesis parameter) are considered. If it can be shown that a more robust analysis algorithm can significantly improve that figure, than a skillful, near real-time model may be trained from large-scale operational data.