Show simple item record

dc.contributor.authorMiller, Paul Wesley
dc.date.accessioned2018-06-07T04:30:12Z
dc.date.available2018-06-07T04:30:12Z
dc.date.issued2017-12
dc.identifier.othermiller_paul_w_201712_phd
dc.identifier.urihttp://purl.galileo.usg.edu/uga_etd/miller_paul_w_201712_phd
dc.identifier.urihttp://hdl.handle.net/10724/38124
dc.description.abstractWeakly forced thunderstorms (WFT), convection forming in stagnant summer air masses, are a historical forecasting challenge for operational meteorologists. Pulse thunderstorms, defined by this dissertation as WFTs that produce severe weather, closely resemble their nonsevere counterparts, thwarting forecaster efforts to issue accurate severe weather warnings. This dissertation seeks to overcome the apparent similarities by developing a large, custom dataset of WFTs and applying machine learning techniques to accurately distinguish nonsevere WFTs from pulse thunderstorms as well as the convective environments that enhance WFT intensity. The WFT dataset (885,496 storms) is extracted from 15 years of warm season (May-September) composite reflectivity radar imagery from 30 collection sites in the Southeast, U.S., an active WFT region. Further, output from a high-resolution weather model, the Rapid Refresh, is used to characterize the convective environment of all WFTs between 2012–2015 (228,363 storms), and thirteen additional radar-derived and lightning-related parameters are recorded for WFTs during June and July of this subset (84,664 storms). Pulse thunderstorms, WFTs associated with Storm Data severe weather reports, constitute 0.60%, 0.65%, and 0.97% of each subset, respectively. The results of this dissertation show that the spatial maximum in pulse thunderstorm activity, the Blue Ridge Mountains, is displaced from the overall WFT maximum in Florida and the Gulf Coast. Only two convective environmental parameters, vertical totals (VT) and total totals (TT), appreciably differentiate days with pulse thunderstorm activity from days with only nonsevere WFTs. When VTs (TTs) exceed 25.1°C (47.3°C), severe wind days are roughly 5x more likely. Meanwhile, severe hail days became roughly 10x more likely when VTs (TTs) exceed 26.0°C (49.2°C). A decision-tree-based machine learning algorithm, random forests, struggles to distinguish pulse thunderstorms from nonsevere WFTs in the broadest sample, but performs satisfactorily in a subset of the most active geographic regions and convective environments mentioned above. The critical success index (CSI) is 46.0%, which out-performs the U.S. National Weather Service CSI (34.8%) for severe thunderstorm warnings issued on pulse thunderstorms. Likely under-reporting of pulse thunderstorm-related severe weather is hypothesized to impede identification of clearer differences between pulse thunderstorm and nonsevere WFT environments and radar behavior.
dc.languageeng
dc.publisheruga
dc.rightsOn Campus Only Until 2019-12-01
dc.subjectSevere Weather, Weakly Forced Thunderstorms, Disorganized Convection, Machine Learning, Southeast U.S.
dc.titleDetecting hazardous weather potential in low signal-to-noise ratio settings:
dc.typeDissertation
dc.description.degreePhD
dc.description.departmentGeography
dc.description.majorGeography
dc.description.advisorThomas Mote
dc.description.committeeThomas Mote
dc.description.committeeJ. Marshall Shepherd
dc.description.committeeLynne Seymour
dc.description.committeeJohn Knox


Files in this item

FilesSizeFormatView

There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record