3rd Conference on Artificial Intelligence Applications to the Environmental Science


Probabilistic tropical cyclogenesis prediction

Christopher C. Hennon, Ohio State University, Columbus, OH

On average, approximately 100 tropical waves and other organized convective events move across the Atlantic basin during the hurricane season of summer and autumn. Of these, only about 10 develop into named tropical storms. Though the physical process of tropical cyclone formation (or "tropical cyclogenesis") is not understood, there are several well known large-scale factors that must be in place for this rare event to occur.

A new database of tropical cloud clusters (precursors to cyclogenesis) that formed in the Atlantic during the 1998-2000 hurricane seasons is examined to ascertain if tropical cyclogenesis can be predicted with skill up to two days prior to the event. Cloud clusters were binned into "developing" and "non-developing" categories and eight large-scale predictors were derived from a global reanalysis dataset. This information was fed into a feed-forward backpropagation neural network configured to yield the probablility of tropical cyclogenesis within a certain time frame. Results will be compared to a linear discriminant analysis that was performed on the same dataset.

Session 3, Neural Networks
Tuesday, 11 February 2003, 8:30 AM-12:15 PM

Next paper

Browse or search entire meeting

AMS Home Page