Authors: Rupsa Bhowmick*, PhD Candidate, Department of Geography and Anthropology, Louisiana State University, Jill C Trepanier, Associate Professor, Department of Geography and Anthropology, Louisiana State University, Alex M Haberile, Assistant Professor, Department of Geography and Anthropology
Topics: Climatology and Meteorology, Spatial Analysis & Modeling, Australia and New Zealand
Keywords: Tropical cyclone development, aerosol components, machine learning, spatial regression
Session Type: Paper
Start / End Time: 9:55 AM / 11:35 AM
Room: Marshall East, Marriott, Mezzanine Level
Presentation File: No File Uploaded
Tropical cyclones (TC) over the Southwest Pacific Ocean (SWPO) cause major disasters in Eastern Australia (EA). The ability to anticipate these TC development would be useful to many lives in EA. This paper uses machine learning and geophysical variables from 1980 to 2017 during the active TC season in the SWPO basin to determine if a data mining approach can accomplish this task. Several atmospheric variables (specific humidity, geopotential height, etc.) and aerosol components (extinction aerosol optical depth of dust, etc.) are considered based on previous research that has suggested their importance in TC development. TC initiation points (n=345) in the SPEArTC dataset are stratified into two subsets based on whether or not they eventually developed into tropical storms. The environmental data are extracted from the MERRA-2, NCEP/NCAR and ERA interim reanalysis dataset based on the time and location of the initiation points. A decision tree is trained with the labels of 0 (non-developing) and 1 (developing) using 80% of the dataset (n = 276). Assessing the performance of the trained decision tree on the testing dataset (n = 69) suggests an accuracy of 69.57%, although some spatial variability in accuracy was noted. The trained decision tree suggests that certain aerosol (dust, and organic carbon) and atmospheric variables (air temperature at 850 mb, and relative humidity at 300 mb pressure level) were important for classification decisions. Further, geographically weighted regression is used to quantify the spatial variation and the relative importance of the covariates on TC development over the basin.