Date of Award

May 2019

Degree Type

Thesis

Degree Name

Master of Science

Department

Atmospheric Science

First Advisor

Paul Roebber

Second Advisor

Clark Evans

Committee Members

Jonathan Kahl

Keywords

Evolutionary Programming, Tropical Cyclone

Abstract

An innovative statistical-dynamical tropical cyclone (TC) intensity model is developed from a large ensemble of algorithms through evolutionary programming (EP). EP mimics the evolutionary principles of genetic information, reproduction, and mutation to develop through selective pressure a population of algorithms with skillful predictor combinations. From this process the 100 most skillful algorithms as determined by root-mean square error on cross-validation data is kept and bias corrected. Bayesian model combination is then used to assign individual weights to a subset of ten algorithms from the 100 best algorithms list, which are chosen to minimize mean-absolute error (MAE) and maximize mean-absolute difference across the selected algorithms. This results in combining both skillful and diverse algorithms, which together produce a forecast that is superior in skill to that from any individual algorithm. Using these methods and a perfect-prognostic approach, two similar but distinctly separate TC intensity models are developed to forecast for TC intensity every 12 h out to 120 h, with one forecasting TC intensity for the North Atlantic basin and the other for the east/central North Pacific basins. Results show improvements as defined by MAE over the “no skill” Decay Statistical Hurricane Intensity Forecast (OCD5) climatology/persistence model in the North Atlantic basin out to 96 h. In the east/central Pacific basins performance over the 12-24 h lead-time is similar to the OCD5, while at later lead times performance drops below that of OCD5. Specific case studies are analyzed to give more insight into the behavior and performance of the models.

Download

Share

COinS