Return to search

Evaluating Operational and Newly Developed Mesocyclone and Tornado Detection Algorithms for Quasi-Linear Convective Systems

Tornadoes in the southeastern United States frequently occur with quasi-linear convective systems (QLCSs, also referred to as squall lines). Studies have shown that the non-descending mode of tornadogenesis (i.e., when strong rotation with a tornado builds upward through the storm) is especially common with the QLCS. Due to the frequent non-descending mode of tornadogenesis associated with QLCS tornadoes (hereafter referred to as tornadoes), it is difficult to issue timely tornado warnings for them. 86 tornadoes associated with QLCS systems that occurred during 22 separate severe weather episodes were studied using archived Warning Surveillance Radar – 1988 Doppler (WSR-88D) data to recreate mesocyclones and Tornado Vortex Signatures (TVSs) identified by the WSR-88D operational algorithms. Storm Data reports were used to create separate datasets of tornadic, pre-tornadic, and non-tornadic (null) mesocyclone and TVS detections. Selected diagnostics from the Mesocyclone and TVS Detection Algorithms (MDA and TDA, respectively) were compared between the tornadic and non-tornadic datasets. Low Level Delta Velocity (LLDV), Low-Level Rotational Velocity (LLROTV), and Strength Index (SI) showed some skill at discriminating between the tornadic and non-tornadic mesocyclones; however, the remainder of the selected mesocyclone parameters showed no significant skill, and none of the TVS parameters showed significant skill at discriminating between the tornadic, non-tornadic, or pre-tornadic cases. For mesocyclones, a correlation coefficient of 0.803 between LLDV and Maximum Delta Velocity for the pre-tornadic detections suggested that most of the tornadoes followed the non-descending tornadogenesis paradigm. Selected MDA and TDA diagnostics were combined using regression trees to determine whether these combinations would improve the skill at discriminating between tornadic and non-tornadic mesocyclones and TVSs. Ten-fold cross validation tests were performed on the selected diagnostics to ensure that the regression trees would generalize to independent data. Vertically Integrated Rotational Velocity and LLDV most often were selected for the regression trees. Some combinations of predictors produced detection success rates exceeding 30%, thereby exceeding the success rates when single MDA or TDA parameters were used. Various sized arrays of Spectrum Width (SW), Azimuthal Shear (AZ), and Reflectivity (RE) were combined with the MDA and TDA diagnostics to determine if these predictors contributed additional skill at discriminating tornadic mesocyclones and TVSs from the non-tornadic variety. These parameters had not been considered in previous studies. Qualitative analysis showed that the largest values of SW occurred with the non-tornadic dataset, but the differences between this dataset and the tornadic datasets were too small to be useful in an operational setting. Two sample Kolmogorov-Smirnov goodness-of-fit tests revealed that, in general, statistically significant differences did not exist among tornadic, pre-tornadic, and non-tornadic datasets for the different array sizes. Generally, qualitative differences were smallest with the largest arrays. SW arrays were included in the list of selected MDA and TDA diagnostics to create regression trees that would determine whether these combinations would yield predictive skill comparable to the best performing MDA and TDA diagnostics. For the mesocyclone detections, the same predictors (VIROTV and LLDV) dominated every SW array size. For the TVS comparisons, only the larger SW arrays appeared in some trees. This suggests that the larger SW arrays had predictive ability that was comparable to the four TDA predictors used in the regression trees Azimuthal Shear (AZ) frequently was found to exhibit negative values, even at the times of tornado occurrence. This result was unexpected since mesocyclones and TVSs are almost always associated with positive shear. Different array sizes of AZ showed strong correlations with each other, which was expected since the linear least squares derivative that is applied to the Velocity data has an a priori smoothing effect. Two-sample Kolmogorov-Smirnov goodness-of-fit tests showed that the most statistically significant differences in AZ occurred between the non-tornadic mesocyclones (NTM) and both the tornadic mesocyclones (M0) and the tornadic/pretornadic mesocyclones (MALL). Results indicated that AZ exhibited less ability than SW as a sole predictor at discriminating between tornadic and non-tornadic mesocyclones. However, when used in regression trees together with MDA and TDA parameters, AZ exhibited utility comparable to SW for TVS detections and was superior to SW for mesocyclone detections. When both SW and AZ arrays were used to create regression trees, SW always emerged as the dominant predictor. A surprising finding was the preponderance of negative AZ values in the tornadic TVS detections. Increasing array sizes of Reflectivity Variance (REV) appeared to have an upper limit of utility. Contrary to what was hypothesized, pre-tornadic mesocyclone and TVS detections exhibited larger variances than tornadic detections. Based on two-sample Kolmogorov-Smirnov goodness-of-fit tests, only the non-tornadic mesocyclone (NTM) versus pre-tornadic mesocyclone (MPRE) datasets exhibited statistically significant differences. None of the other comparisons was statistically significant. REV did not exhibit skill comparable to the best Severe Storms Analysis Package (SSAP) diagnostics for the mesocyclone detections in the regression trees that were constructed. However, for TVS detections, the larger REV arrays appeared in regression trees for the NTT versus TALL comparisons, and smaller arrays appeared in regression trees for the NTT versus TPRE comparisons. Regression trees combining REV arrays with the best-performing MDA and TDA predictors showed that REV exhibited better predictive ability for the TVS datasets compared to SW and AZ. These results suggest that SW, AZ, and REV all show utility as predictors when used in combination with some of the already existing MDA and TDA diagnostic parameters. Incorporating these additional data sources into the algorithms therefore could improve their performance. / A Thesis Submitted to the Department of Meteorology in Partial Fulfillment of the
Requirements for the Degree of Master of Science. / Summer Semester, 2007. / April 5, 2007. / Tornadoes, Quasi-Linear Convective Systems, Algorithms / Includes bibliographical references. / Henry E. Fuelberg, Professor Directing Thesis; Jon E. Ahlquist, Committee Member; Paul H. Ruscher, Committee Member; Andrew I. Watson, Committee Member.

Identiferoai:union.ndltd.org:fsu.edu/oai:fsu.digital.flvc.org:fsu_175987
ContributorsTurnage, Thomas James (authoraut), Fuelberg, Henry E. (professor directing thesis), Ahlquist, Jon E. (committee member), Ruscher, Paul H. (committee member), Watson, Andrew I. (committee member), Department of Earth, Ocean and Atmospheric Sciences (degree granting department), Florida State University (degree granting institution)
PublisherFlorida State University, Florida State University
Source SetsFlorida State University
LanguageEnglish, English
Detected LanguageEnglish
TypeText, text
Format1 online resource, computer, application/pdf
RightsThis Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s). The copyright in theses and dissertations completed at Florida State University is held by the students who author them.

Page generated in 0.0027 seconds