Mining data and expert intuition to develop a nonlinear tropical cyclone intensification model

Linear models of tropical cyclone (TC) intensity response to their environments have proven useful for short-term forecasting, but fail to capture characteristic profiles of intensification and decay when used to simulate intensity time series along the full length of TC tracks. In the present work, we use a Bayesian approach to develop a nonlinear model for TC intensity evolution that is better suited for simulating complete maximum wind speed histories. Prior models for the nonlinear response of intensity to environmental parameters like vertical wind shear and maximum potential intensity are elicited from the intuition and understanding of experienced TC scientists. This information is refined via Bayesian updating using historical time series of intensity from historical TCs and reanalysis data on their local, contemporaneous environments. The derived posterior can be interpreted in terms of the environmental influences on TC intensification, the variability across storms and conditions in these influences, and the uncertainty in the environment/intensification relationships. The posterior predictive distribution also provides a simulation model, which is useful in catastrophe risk modeling contexts.