Discipline: Mathematics and Statistics
Subcategory: Mathematics and Statistics
Ian Livengood - North Carolina A&T State University
Co-Author(s): Liping Liu, North Carolina A&T State University, Greensboro, NC; Justin Riley, North Carolina A&T State University, Greensboro, NC; Yuh-Lang Lin, North Carolina A&T State University, Greensboro, NC
Tropical cyclones (TC) are dangerous because they produce destructive winds, heavy rainfall with flooding, and damaging storm surges. It is valuable to understand tropical cyclogenesis: a meteorological word used to describe TC formation, and its strengthening due to the atmosphere. This study focuses on tropical cyclogenesis frequency and variation and the prediction of the number of tropical cyclones that form in the North Atlantic Basin. Previous studies from physics analysis identified four factors that affect the tropical cyclogenesis: potential intensity, vertical shear, relative humidity, and absolute vorticity. In this study, we include other factors in addition to the above four variables. We obtain the data for various variables from the European Centre for Medium-Range Weather Forecast (ECMWF) ERA-Interim reanalysis dataset. Some variables (e.g. potential intensity) are computed from the ECMWF data. From the National Hurricane Center’s track maps, we compile the data on the number of TCs that formed each month from 1979 to 2011 in the North Atlantic Basin. The first part of this study investigates a genesis potential index (GPI) as a function of the four factors. The data is organized in Excel, and the GPI is computed for each month. Numerous plots are generated to compare the GPI with the cyclogenesis frequency. The second part of this study employs machine learning methods on the variables that could be possibly linked with cyclogenesis. The methods considered include the support vector machine, random forest, naïve Bayes classifier, and nonlinear regression. It is discovered and verified that the GPI matches well with the cyclogenesis frequency for most months except in August and September when the GPI prediction is higher than the actual frequency. The machine learning methods reach a similar major conclusion: the most significant factors for tropical cyclogenesis in the North Atlantic Basin are vertical shear and absolute vorticity. More details on the significance of the other factors can be found in the full paper. An effective model for the prediction of the number of tropical cyclones is also provided. The effectivity and efficiency of the machine learning methods are provided in the study. This study provides effective methods to predict the number of TCs. The insights of this study help people with preparation for the events and thus reduce the damage. For future work, other factors can be included, and other machine learning methods can be explored for a better predictive model and a more thorough understanding of cyclogenesis. References: • Bruyère, C.L., Holland, G.J. & Towler, E., Investigating the Use of a Genesis Potential Index for Tropical Cyclones in the North Atlantic Basin. J. of Climate 25.24 (2012): 8611-626. • Camargo, S.J., Sobel, A.H., Barnston, A.G. & Emanuel, K.A., Tropical Cyclone Genesis Potential Index in Climate Models. Tellus A 59.4 (2007): 428-43.
Funder Acknowledgement(s): This research project is funded by the HBCU-UP ACE Data Science and Analytics Program – National Science Foundation HRD#1719498.
Faculty Advisor: Liping Liu, firstname.lastname@example.org
Role: I did all of this research with the following exceptions: 1. I did not obtain the data for various factors from the European Centre for Medium-Range Weather Forecast (ECMWF) ERA-Interim reanalysis dataset. 2. I didn't calculate factors like potential intensity from the ECMWF data.