The algorithmic detection of pulse thunderstorms within a large, mostly non‐severe sample

Abstract

The accurate differentiation of pulse thunderstorms from benign weakly forced thunderstorms (WFTs) is both a historical and contemporary forecasting challenge. Little research has been directed toward WFTs, and the few existing efforts are characterized by small sample sizes and inflated proportions of pulse thunderstorms. The purpose of this study was to determine whether pulse thunderstorms can be successfully differentiated from non‐severe WFTs within a large, mostly non‐severe sample, a more operationally realistic scenario. Random forests, a decision‐tree‐based machine learning technique, was applied to radar, total lightning and environmental parameters of >84,000 WFTs, of which <1% were pulse thunderstorms. In particular, differential reflectivity (ZDR) fields were mined for occurrences of ZDR troughs, suggested by recent work as a reliable indication of impending downbursts. The random forest identified pulse thunderstorms from the 84,000 storm sample with a critical success index (CSI) of 0.29. The CSI improved to 0.50 in a subset of the most active geographical regions and convective environments, outperforming the US National Weather Service (CSI = 0.35) for warnings issued on pulse thunderstorms. Unfortunately, the presence of ZDR troughs contributed little skill to the random forest. Although forming at higher rates in pulse thunderstorms (61%) than non‐severe WFTs (5.1%), trough‐possessing non‐severe WFTs were roughly 9× more common than pulse thunderstorms with the same feature. Overall, the random forest shows promise as a potential decision aid for identifying pulse thunderstorms. Performance may be further improved if collinearity amongst the radar parameters and probable under‐reporting of severe weather can be overcome.

Weakly Forced Thunderstorms in the Southeast US Are Stronger Near Urban Areas

“…Each WFT is then characterized by its flash total per volume scan per pixel (flash density) as well as its total lifetime flash count. The incorporation of these additional radar‐based and lightning metrics into the Miller and Mote (2017a) data set are described in Miller and Mote (2018a).…”

“…Five of these groups holistically matched the morphological expectations for WFTs (i.e., small, stationary, diurnally driven) with comparisons to morning‐of radiosonde launches confirming that these WFT‐like classes indeed formed on high‐instability, weakly sheared days (Miller & Mote, 2017a). Model proxy soundings from Miller and Mote (2018a) determined the average WFT environment for this data set was characterized by 1,658 J kg −1 of mean‐layer (ML) convective available potential energy (CAPE) and 7.2 m s −1 of 0–6‐km wind shear. In total, the Miller and Mote (2017a) data set includes 885,496 WFTs between May‐September during 2001–2015, with each WFT associated with an (a) initiation time, the time elapsed from 1200 UTC daily until the WFT was first detected; (b) duration, the time elapsed between the first and last detection in min; (c) max pixel size, the number of pixels comprising the WFT at its largest; (d) max reflectivity, the greatest composite reflectivity associated with the WFT; (e) max pixel ratio, the WFT's mean size in pixels per volume scan versus its maximum size in any single scan (uniformly circular storms have values near one while spatially variable storms have values near zero); and (f) total pixel count, the number of pixels associated with the WFT over its entire lifespan.…”