J4.7 THE WDSS-II SUPERCELL IDENTIFICATION AND ASSESSMENT ALGORITHM Richard Jason Lynn *, 1, 2 1NOAA/NCEP/Storm Prediction Center, Norman, Oklahoma 2Cooperative Institute for Mesoscale Meteorological Studies, Norman, Oklahoma 1. INTRODUCTION an effort to prom ote this heightened level of situation awareness, SIAA is designed to m im ic the feature-based interrogation process Supercell thunderstorms account for a small that a human meteorologist would follow in determining storm percentage of convective activity, and yet they are responsible type, and assessing the priority level at which the storm should for a disproportionate am ount of severe weather dam age. receive increased attention. Long track, violent tornadoes, giant hail, and extrem e surface wind gusts are m ore likely to accompany supercell 2. ALGORITHM DESCRIPTION thunderstorms than m ost other types of organized convection (Browning 1977; Burgess and Lem on 1991; Moller, et al. Because of its intuitive use of membership, fuzzy set 1994). Because of this propensity to harm life and property, theory is used as a proxy for the human process of classification supercell thunderstorms draw increased attention from in determining the degree to which a thunderstorm is National W eather Service warning forecasters. Faced with a supercellular. There are several advantages gained by using severe weather situation, a warning forecaster must rely on fuzzy sets and fuzzy logic for this application. Many of the his/her data interpretation skills, situation awareness, and the following observations fit well with the fuzzy set description laid ability to integrate numerous sources of data, using human forth in (Zadeh 1965). First, within the field of meteorology, there expertise to m ake effective and timely warning decisions. is no univers ally accepted quantitative definition of a supercell However, interrogating a potentially severe thunderstorm thunderstorm. The boundaries between storm type classes are requires a large allocation of the forecaster’s temporal and not crisp. Also, uncertainty often exists about the presence, or cognitive resources. The presence of several potentially significance of a feature, which im plies a degree of membership dangerous storm s within the County W arning Area further of an object to som e class, or classes. Fuzzy set theory is well strains the forec as ter’s ability to investigate the severity of equipped to deal with a spectrum of m em bership of an object to each storm , decreasing his/her interrogation efficiency m ultiple classes. (Bunting 1998; Durso and Gronlund 1999). The addition of Second, fuzzy set theory facilitates a feature-bas ed autom ation through signature detection algorithms has shown approach towards the intuitive classification of an object, provided to im prove the interrogation efficiency a forecaster can exhibit that the class features can be well defined, as with supercells. in a severe weather situation (Lemon, et al. 1992), but such Fuzzy sets allow the m easurem ent of feature memberships to tools can promote algorithm dependency. idealized classes, and combination of these memberships to arrive P revious methods of automated thunderstorm at easily understood classification decisions. interrogation have focused on the independent detection of Finally, fuzzy logic supports the im plem entation of a rule single features (or single set of features) c ommonly base and the ability to describe the situation in linguistic term s. associated with severe thunderstorms, leaving the These rules m ay be constructed to m odel the relative im portance classification process as an exercise solely for the hum an of features in the classification decision, and can provide both user. Research has shown that supercell thunderstorms supporting and refuting evidence to place an object in a certain sim ultaneously possess m any of these features. Thus, a class. In addition, the use of fuzzy set operators yields fuzzy feature-based object recognition scheme can be used to rules and classifiers that are very flexible and easy to change, or identify such storm s. Since supercell thunderstorms tune (union, intersection, com plem ent, and aggregation; Yager frequently produce severe weather, the recognition that a 1988; Klir and Folger 1988; Zadeh 1965). storm is adopting supercell-like characteristics is beneficial to the warning forecaster. 2.1 Storm type classification W ithin the W arning Decision Support System - Integrated Inform ation (W DSS-II), the Supercell Identification The concept of a long-lived, quasi-steady, intense, and Assessment Algorithm (SIAA) has been developed to rotating thunderstorm has been a topic of vigorous investigation integrate the results of m ultiple signature detection algorithms, for several decades, even predating the term “supercell” creating a m ore holistic approach to storm type classification. (Browning 1962). The advent of more sophisticated data By sim ultaneously searching for several features known to be processing techniques and radar technology, especially Doppler related to severe potential in thunderstorms, and common to radar, has im proved the understanding of supercell structure and supercell thunderstorms, the am ount of cognitive processing evolution. Through the past fifty years, research has shown that required by the human is reduced, and the likelihood of the despite assuming a wide variety of configurations and user applying a relevant conceptual model is increased. In orientations, supercells can possess common identifiable features. Although there is no widely accepted quantitative definition of a supercell, several of these features are mentioned *Corresponding author address: R. Jason Lynn, Storm Prediction in most, if not all supercell definitions. Such features include: Center, 1313 Halley Circle, Norm an , OK 73069, Email: [email protected]. 1) deep, persistent shear (the mesocyclone; Davies-Jones 1984; Brooks, et al. 1993) with the following brief discussion about each term: 2) possession of a bounded weak echo region (Chisholm 1973; Lemon 1980), fz_meso - represents the mesocyclone strength fuzzy 3) deviant motion, with respect to the mean wind (Browning m em bership value; com puted linearly using the 3-D meso 1963; Lemon and Doswell 1979; Klemp and W eisman strength rank param eter for the current shear signature from M DA 1983; Moller, et al. 1994). (Stum pf, et al. 1998), with a membership range of 0 to 4; assigned the largest weight because the m esocyclone is the m ost SIAA functions as a m esocyclone-based algorithm, c o mmon feature mentioned across supercell processing inform ation only for those storms that are definitions/discussions (B rowning 1963; Lem on and Doswell associated with som e sort of four dim ensional azim uthal 1979; Klemp and W eisman 1983; Moller, et al. 1994) shear in the Doppler radar radial velocity data. SIAA depends on the NSSL M esocyclone Detection Algorithm (MDA; fz_meso_life - represents the m esocyclone longevity fuzzy Stum pf, et al. 1998) to identify these shear regions. m em bership value; com puted linearly using the num ber of Following a fuzzy logic scheme sim ilar to that used m inutes since the first detection of the current shear signature, in (Lakshmanan and W itt 1997), classification of a storm as with a m em bership range of 0 to 25 minutes; assigned the next a Supercell (fz_max), M arginal Supercell (fz_marg), or Non largest weight because the persistence of the 3-D shear signature Supercell (fz_min) is done using both actual and hedged is a distinctive attribute of supercells (Browning 1977; M oller, et values of two fuzzy m em bership values, fz_yes and fz_no, al. 1994) in the following fuzzy classifiers: fz_bwer - represents the significance of any bounded weak echo fz_max = (fz_yes AND “not” fz_no) region associated with the current shear signature; retrieved from OR the NSSL Bounded W eak Echo Region detec tion algorithm (“very much” fz_yes AND “very much” “not” fz_no) (Lakshmanan and W itt 1997); assigned a relatively small weight because of detection difficulty (Chisholm 1973; Lemon 1980) fz_marg = (fz_yes AND fz_no) OR fz_deviant_motion - represents the m esocyclone deviant (“very much” fz_yes AND “very much” fz_no) m otion fuzzy m em bership value; computed linearly using the angular difference between the current shear signature’s motion fz_min = (“not” fz_yes AND fz_no) vector and the 0-6 km shear vector, with a m em bership range of OR 0 to 25 degrees (Maddox 1976; W eisman and Klemp 1984); (“not” “very much” fz_yes AND “very much” fz_no) assigned the smallest weight because of MDA dependency where “not” represents the fuzzy com plem ent of the argument Note that the fz_no value in the fuzzy c lassifiers is (1 - fz_val), “very m uch” represents an overestim ating hedge com puted using the fuzzy com plem ent of fz_meso; this form ula of the argument (1 - e^ (-1.5*fz_val))/(1 - e^(-1.5)), and is used because a storm is not a supercell if it is not associated OR/AND represent the logical or/and operators, that return with a reasonably strong 3-D shear signature. the max/min value of the expression. The structure of these rules highlights the feature-based The base fuzzy m em bership values of fz_yes and approach of this recognition scheme. The presence of any one feature in tem poral or spatial isolation, no matter how strong, is fz_no are determined by the following calculations: not sufficient to classify the storm as a supercell. fz_yes = (fz_sp_previous AND fz_sp) OR 2.2 Attention worthiness classification