Using Bayesian Statistical Post-Processing Techniques to Improve Tropical Cyclone Track and Intensity Forecasts
Abstract
This thesis examines the use of statistical post-processing techniques involving Bayesian estimation and Markov Chain Monte Carlo methods to aid in the reduction or elimination of tropical cyclone track and intensity forecast errors. The results of this research showed an improvement in the forecasts for intensity and total track error over the European Centre for Medium-Range Weather Forecasts (ECMWF) ensemble mean for all forecast times. These findings indicate that applying Bayesian statistical post-processing to forecasts made by the ECMWF ensemble can reduce the overall track and intensity error and result in more accurate forecasts. The most significant forecast improvement resulted from larger sample sizes and creative grouping schemes. By increasing the number of storms used and altering the manner in which the data is grouped, a more accurate forecast can be obtained. Future research using a larger sample size that spans several decades is indicated, but any significant physics alterations to the models over time, as well as more specific ways of grouping the data, must be taken into consideration.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2018
- Accession Number
- AD1059816
Entities
People
- Sabrina L. Cummings
Organizations
- Naval Postgraduate School