|
Abstract:
|
Cyclogenesis is the process of tropical cyclone formation. It has been recognized that certain environmental states of the atmosphere and ocean can create favorable/unfavorable environments for cyclogenesis. By using several key environmental variables as predictors, regression models have previously been developed for cyclogenesis prediction. Here, we adapt previous approaches to develop a model for simulation. This application requires finer space-time resolutions, which presents a challenge for inference because it renders historical cyclogenesis events – already rare in a short historical record - extremely sparse on the finer model grid. Our model is a semiparametric logistic regression model of cyclogenesis probabilities across both space and time. The nonparametric part of the model is empirically-determined and represents the expected cyclogenesis pattern across space and time. The parametric part of the model is a logistic regression using environmental variables as predictors, in which the Firth Method enables Maximum Likelihood Estimation of the regression coefficients on our sparse dataset. Environmentally-driven simulations reproduce the historical space-time patterns.
|
