1

EASTERN REGION TECHNICAL ATTACHMENTNO. 98-8

SEPTEMBER, 1998

VIL Density as an Indicator of Hail across Eastern New York and Western New England

Jonathan L. BlaesCarl S. Cerniglia Jr.Michael A. Caropolo

NOAA / National Weather Service Forecast OfficeAlbany, NY

Editor’s Note: The lead author’s current affiliation is NWSFO Raleigh, NC.

1. INTRODUCTION

The deployment and continued utilization ofthe Weather Surveillance Radar - 1988Doppler (WSR-88D) network across theUnited States has resulted in an increase insevere local storm warning lead times andimproved verification (Polger et al. 1994). Asexperience grows with the radar, newtechniques are being developed to betterutilize radar products and understand systemlimitations. Determining the likelihood ofsevere hail, currently defined as � 0.75 inches(19 mm) in diameter, remains a significantforecast problem facing operationalmeteorologists monitoring the potential ofsevere local storms. Amburn and Wolf (1997)first compared maximum VerticallyIntegrated Liquid water content (VIL) valuesand associated Echo Tops (ET) to computeVIL densities (VIL / ET) which weresuccessfully used as a predictor of large hailacross the northeastern Oklahoma -northwestern Arkansas region.

A sample of hail events over 5 convectiveseasons from 1994 through early 1998 wasexamined at the NEXRAD Weather Service

Forecast Office (NWSFO) in Albany, NY(ALY) using KENX WSR-88D radar datafrom Berne, NY. The study sought todetermine if established VIL densitytechniques from other parts of the countrywere effective across eastern New York andwestern New England. The study alsoinvestigated whether new information andenhancements available in software Build 9 ofthe WSR-88D, such as Cell Based VIL(CBVIL) and Storm Top (ST), can improvethe effectiveness of VIL density. Finally,with the proposed change in the severe hailsize criteria (Purpura 1998) from � 0.75inches (19 mm) in diameter increasing to �

1.0 inch (25 mm) in diameter, we sought todetermine how this change would affect theusefulness of VIL density.

2. VERTICALLY INTEGRATED LIQUID

Greene and Clark (1972) proposed thatVertically Integrated Liquid water content(VIL) could be a useful tool for determiningthe potential for hail in the central Plains[hereafter, VIL and Grid Based VIL (GBVIL)refer to the same value]. They suggested that

2

hail may produce false values of liquid waterdue to enhanced radar return. Operationalforecasters have learned to use VIL in part todetermine the potential severity ofthunderstorms which are likely to producesevere hail.

There are several limitations in using VIL asan indicator of severe hail. Often, forecastershave no idea of what VIL value willcorrespond to severe size hail on a given day.Meteorologists often have to wait for groundreports of hail and then compare the locationand time of the report to VIL. Based upon thecorrelation between the hail size reported andthe VIL values displayed on the radar,forecasters can then estimate the occurrenceand size of additional hail events. Forecastersat NWSFO ALY have observed that a colder,drier atmosphere can support severe hail withrelatively low VIL values (20-30 kg m-2),while a warmer, more moist atmosphere oftenrequires larger VIL values (50-60 kg m-2) toproduce severe hail. This atmospheric andseasonal variation in VIL values associatedwith severe hail has led to various techniquesand procedures to determine a representativeVIL value that would produce severe hail fora particular event.

In an effort to anticipate thunderstorms whichmay produce severe hail, several studies haveattempted to predict what VIL value on aparticular day would result in severe hail.These values are commonly referred to as the“VIL of the Day.” There are limitationsinherent in using the VIL of the Day (VOD)approach. First, VOD values may beunrepresentative of the environment across alarge forecast area with varying atmosphericconditions. Since the VOD requires forecastdata, it is also subject to the biases andconstraints of model forecasts. Finally, theVOD approach relies upon predictive data,failing to utilize real time observational data.

Other studies have used VIL values incombination with other variables, to predictthe occurrence of hail. Billet et al. (1997)developed a logistic regression model to act asa yes/no predictor of hail severity based onvalues of VIL, 850-hPa temperature, freezinglevel, and low-level storm inflow. Althoughthis approach includes observed VIL values,it still relies upon forecast data or twice dailysounding data which may be unrepresentativeof atmospheric conditions in the vicinity ofthunderstorms. Recent research has centeredon methods to eliminate the variabilityassociated with VIL by “normalizing” VILwith the ET; which would result in a valuethat should be airmass independent.

3. VIL DENSITY

Amburn and Wolf (1997) first demonstratedthe use of VIL density (VIL / ET) as apredictor of large hail across the northeasternOklahoma - northwestern Arkansas region.Other studies have investigated VIL densityacross varying geographic locations. Turnerand Gonsowski (1997) explored the utility ofVIL density across northwestern Kansas,Roeseler and Wood (1997) investigated if VILdensity was a useful predictor of severe hailacross the northwestern Gulf of Mexico, andTroutman and Rose (1997) investigated VILdensity across middle Tennessee.

Amburn and Wolf (1997) defined VIL densityas the quotient of VIL (kg m-2) divided by theET (m) and then multiplied by 1000 to yieldunits of g m-3.

VIL density (g m-3) = VIL (kg m-2) / ET (m)& 1000

Amburn and Wolf (1997) hypothesized theVIL density concept after noting that high-topped thunderstorms with high VIL values

3

do not always produce large hail, while low-topped thunderstorms with low VIL valuesoccasionally do produce large hail. Theysuggested that dividing the VIL by the ETwould “normalize” VIL with the ET. Thiswould result in a value that can be used toidentify thunderstorms which contain highreflectivities relative to their height andtherefore possess a likelihood to produce largehail. By normalizing VIL with the ET, theVIL density value should be airmassindependent and should be a useful predictorregardless of the thunderstorm’s actual VIL orheight and the local atmosphericcharacteristics.

4. METHODOLOGY AND DATA

This study examined 154 thunderstormswhich occurred from April 1994 through May1998 in the NWSFO ALY County WarningArea (CWA). Thunderstorms ranged from 10to 72 nm (19 to 133 km) from the radar; VILvalues ranged from 15 to 77 kg m-2, and ETvalues ranged from 17,000 to 52,000 feet (5.2to 15.8 km). Of these 154 events, 97produced severe hail [� 0.75 inches (19 mm)in diameter], with 57 events producing non-severe [< 0.75 inches (19 mm) in diameter] orno hail. Hail sizes ranged from zero or no hailto 3.0 inches (76 mm) in diameter. Therewere two types of events included in thisstudy, severe hail [� 0.75 inches (19 mm) indiameter] and non-severe hail [< 0.75 inches(19 mm) in diameter] events.

Severe hail events were identified by reportsof severe size hail. This information wasretrieved from local storm reports, localsevere weather logs, and Storm Data (U.S.Department of Commerce 1994-1997). Non-severe hail cases were identified from radardata and severe weather logs and thenchecked with Storm Data to ensure the events

were of non-severe size. They were includedin the study only if non-severe hail or no hailwas observed over a significantly populatedarea during the time of day in which severeweather reports would normally be expected(0800 through 2000 local time). These eventswere included in the study if archived radardata was available.

Radar data used in this study was principallyfrom archive IV data at the Berne, NY(KENX) WSR-88D. Echo Top data from theTaunton, MA (KBOX), Brookhaven, NY(KOKX) and Binghamton, NY (KBGM) radarsites were used for a few cases to fill inincomplete data sets. To be included in thestudy, the radar data had to be free ofanomalous propagation and beam blockage.To be consistent with earlier studies,thunderstorms with maximum VIL values lessthan 15 kg m-2 were not included. Because ofthese methodology constraints, the totalnumber of cases examined was significantlyreduced from the overall number of hail casesavailable.

These methodology constraints limited thedatabase significantly but they were necessaryto eliminate sources of error in the dataanalyses. For example, thunderstorms withhigh VIL values may not have acorresponding verifying report of severe hailif they occurred in an area with a lowpopulation density. Additionally, non-severehail events were included only if theyoccurred in a location and time of day inwhich severe weather reports would normallybe expected. The absence of a verifyingreport does not necessarily mean that severehail did not occur. As noted by Wyatt andWitt (1997), the largest source of potentialerror in a hail study arises from complicationsand errors inherent in spotter and verificationreports. An accurate study requires thatspotters accurately report the time, location,

4

and size of the hail. Most thunderstormswithin 15 nm (28 km) of the radar site and inareas subject to anomalous propagation aswell as beam blockage were excluded sinceVIL and ET values would be unrepresentativedue to “cone of silence” sampling limitationsand clutter contamination, respectively.However, a few low topped thunderstormsthat were within 15 nm (28 km) of the radarsite were included after noting that the “coneof silence” was not a factor in observing thesestorms.

For each thunderstorm to be investigated, thetime and location of each event, and the hailsize was determined from local severeweather logs and Storm Data. The maximumobserved VIL was recorded over the locationof each hail report. The highest VIL valueobserved during the volume scan nearest thetime of the report or during the previousvolume scan was used. For this part of theinvestigation, the VIL value was specificallythe WSR-88D GBVIL and was determinedusing mid point values of the data rangesunless the storm in question was producingthe maximum observed VIL for that volumescan. In the latter case (approximately one-third of the events in this study), the VILvalue that was explicitly displayed on theradar was used. The radar calculates GBVILby estimating the liquid water content of theatmosphere for a vertical, square column, 2.2nm (4.0 km) on a side. VIL values for eachelevation angle, over a particular square, aresummed to determine the VIL for a particulargrid. This GBVIL value is oftenunrepresentative for fast moving and tiltedstorms.

After the location, time, hail size, and value ofthe GBVIL was determined, the WSR-88DET was recorded for the same time at thesame pixel or the next pixel down stream ofthe GBVIL. The down stream pixel was used

to account for storm tilt and/or stormmovement. Like the GBVIL, the ET has a 2.2nm (4.0 km) resolution. The ET value is thehighest sample volume that contains aminimum reflectivity value of 18.5 dBZ orgreater. It is derived from each elevationangle, which makes it susceptible to severalsources of error. For example, the ET issubject to truncation errors, especially atlonger ranges from the radar and while theradar is using the VCP-21 scan strategy. Thisoccurs where the true ET value lies betweentwo radar scans (elevation slices).Interpolation is not used in the calculation ofthis product, and the ET is assigned to thelower radar beam height. The traditional VILdensity value as defined by Amburn and Wolf(1997) was then computed for each event bydividing the GBVIL by the ET (GBVIL / ET).

Made operationally available with the Build 9software installation in November 1996, theWSR-88D Storm Cell Identification andTracking (SCIT) Algorithm (Witt 1990, Wittand Johnson 1993) continuously scansreflectivity data to identify storm cells andcalculate various parameters of these cells.The algorithm calculates numerous values foreach cell it identifies including; storm motion,maximum reflectivity of the storm, stormvolume, CBVIL, and ST. Many of thesevalues are displayed on the Principle UserProcessor (PUP) attribute table includingCBVIL and ST. The PUP is a graphicsdisplay computer system that NationalWeather Service meteorologists use to viewWSR-88D products. The attribute tableprovides an easy to read table of pertinentradar based values that are used bymeteorologists at the radar to quicklydetermine the intensity of a storm. The VILdensity of a particular storm cell could beeasily calculated by dividing the CBVIL andST values which are readily displayed on thePUP attribute table.

5

The CBVIL is calculated for each storm cellidentified by the SCIT algorithm. Thealgorithm calculates CBVIL by verticallyintegrating maximum reflectivity values of anidentified cell’s components. Thesecomponents are vertically correlated bycomparing the centers of each component inadjacent elevation angles. If a component’scenter overlaps a component in the adjacentelevation angle, they are linked and a cell iscreated. The CBVIL is then calculated usingthis algorithm defined cell rather than justusing reflectivities stacked vertically above afixed point. This method of cell identificationand VIL calculation enables the CBVIL to beadjusted for storm tilt and storm motion,unlike the GBVIL.

The arrival of Build 9 also resulted in animprovement to the ST value which has beenavailable since the installation of the WSR-88D radar. The SCIT algorithm installed withBuild 9 more accurately identifies individualstorm cells which results in an improvementin all the derived cell attributes, including ST.This is especially true of lines or clusters ofstorms which the earlier software wouldfrequently identify as a single, large cellinstead of individual storms. The ST value isa quantity somewhat similar to the ET,however the intensity threshold for this valueis higher. Storm Top is defined as the heightof the beam center point at the center of thehighest component detected for eachidentified storm. The minimum reflectivityvalue used to create a component is 30 dBZ,therefore the ST value is the highest radaridentified echo that contains a reflectivityvalue of 30 dBZ or greater (U.S. Departmentof Commerce 1991). Storm top values willgenerally be lower than ET values exceptwhen truncation errors occur. In thissituation, which occurred in less than 8 % ofthe events examined, the two values will benearly the same. The ST values were

recorded for events investigated subsequent tothe release of Build 9.

A database containing hail size, GBVIL, ET,as well as CBVIL and ST (when available)was created for the 154 investigated events.A traditional VIL density value (GBVIL / ET)was computed for each of these events. VILdensity values were also calculated using cellspecific data (CBVIL and ST) when available.This resulted in three additional methods tocalculate VIL density using a combination ora portion of CBVIL and ST (GBVIL / ST,CBVIL / ET, and CBVIL / ST). Variousstatistical parameters were then calculated, forboth traditional and cell specific VIL densityvalues, in order to determine the thresholdsthat possess the greatest skill in forecastingsevere hail events.

5. LIMITATIONS

As noted earlier, the largest source ofpotential error in a hail study arises fromcomplications and errors inherent in spotterand verification reports (Wyatt and Witt1997). The collection of data can introduceerrors into the data set through various means.First, the exact time and location of the hailevent may be inaccurate, making it difficult tocorrelate the event with radar data. Second,hail sizes are nearly always estimated, ofteninaccurately. Even if the location, time andsize of the hail event is accurate, the reportmay not be representative of the actualseverity of the particular thunderstorm.Larger hail may have fallen elsewhere andgone unobserved or unreported. Similarly, athunderstorm without a report of hail orspecifically severe hail, does not necessarilymean that hail or severe hail did not occur.These potential sources of error can belimited, but not eliminated, by carefullycorrelating the hail reports with the radar data

6

and using storms that occur over a reasonablypopulated area at a time of day in whichreports normally would be expected.

Due to design limitations of the radar system,the actual radar data may contain errors. Theradar is designed to scan the atmosphere atdiscrete elevation angles allowing onlyportions of a thunderstorm to be sampleddirectly. Portions of a thunderstorm betweenthe elevation angles are often not sampled andresult in missing data. The missing dataaffects the calculation of volume productssuch as VIL and results in truncation errors inproducts such as ET and ST (U.S.Department of Commerce 1991).

At short distances from the radar, generallyless than 15 nm (28 km), thunderstorm topsmay exceed the highest elevation scan of theradar leading to unrepresentative calculationsof GBVIL and ET since the radar is unable tosample the entire storm. Since VIL densitynormalizes reflectivities (GBVIL) with theheight of the storm (ET), the truncation ofGBVIL and ET values should generally notpose a problem. However, if the reflectivitycore exceeds the highest elevation scan, theVIL density value would be unrepresentativeand result in a VIL density value that is toolow.

Grid Based VIL values are oftenunderestimated in thunderstorms that tiltexcessively or move rapidly. In thesesituations, portions of a thunderstorm may bepresent in more than one of the 2.2 nm (4.0km) grid boxes used in the calculation of theGBVIL product. This results in diminishedGBVIL values that are unrepresentative of thethunderstorms severity. This problem isespecially noticeable in cool season, fastmoving convective events which often occurin a strong synoptic scale wind regime.

6. RESULTS

Traditional VIL density values (GBVIL / ET)as defined by Amburn and Wolf (1997) areshown in a scatter diagram (Fig. 1) of ET(thousands of feet) versus GBVIL (kg m-2)and was created using 97 severe hail cases (:)and 57 non-severe hail cases (q). A VILdensity threshold of 3.50 g m-3 correctlyidentified 82 % (80 of 97) of severe hail casesin this study and incorrectly identified 7 % ofnon-severe cases (4 of 57) as severe. In anoperational setting, forecasters would find theVIL density value of 3.80 g m-3 significantsince it serves as the upper limit of non-severecases; 99 % (69 of 70) of events with a VILdensity value greater than or equal to 3.80 gm-3 were of severe size. Conversely, the VILdensity value of 3.28 g m-3 serves as a lowerlimit for severe size hail; only 11 % (5 of 45)of events less than 3.28 g m-3 were severe.Fig. 2 illustrates the distribution of severe andnon-severe hail frequencies versus VILdensity.

These calculations have led to the subjectivedevelopment of a nomogram (Fig. 3) for useby meteorologists at NWSFO ALY asguidance during potential severe weatherevents. The nomogram utilizes a VIL densityvalue of 3.28 g m-3 as the upper limit of non-severe hail events while a VIL density valueof 3.50 g m-3 is used as the lower limit ofsevere hail events. The area on the nomogrambetween the 3.28 g m-3 and the 3.50 g m-3

values is a region in which the severe hailpotential is indeterminate. This nomogramshould be used as a part of the warningdecision making process and not usedexclusively to predict the occurrence of severehail.

Our results compare to the Tulsa study byAmburn and Wolf (1997) in which a VILdensity of 3.50 g m-3 correctly identified 90 %

7

of severe hail events and falsely identified 5.5% of non-severe hail cases as severe. TheAlbany study investigated 154 total casesduring 5 convective seasons. The Tulsa studyinvestigated 221 total cases (185 severe cases)in one convective season. The betterperformance of the 3.50 g m-3 VIL densitythreshold in the Tulsa study as compared toour investigation was likely enhanced by thegreater number of severe cases investigated byAmburn and Wolf. The Tulsa study wascomprised of many events with a generallylarger hail size than our study. Our studyinvestigated only 43 events with a hail size of1.0 to 1.75 inches (25 to 45 mm) compared to63 events of similar size in the Tulsa study.The Tulsa study also included 6 cases withhail � 2.5 inches (63 mm) compared to onlyone case of similar size in our investigation.The larger hail size included in the Tulsainvestigation compared to our study wouldallow an easier identification of severe hail,resulting in a better performance of VILdensity in the Tulsa study.

Our results are more similar to aninvestigation by Troutman and Rose (1997)across middle Tennessee in which a V ILdensity of 3.50 g m-3 correctly identified 81 %of severe hail cases. Our findings arepreferable to the Roeseler and Wood (1997)study across the northwest Gulf Coast inwhich a VIL density threshold of 3.50 g m-3

correctly identified 72 % of severe hail cases.At the same threshold our study correctlyidentified 82 % of severe hail cases. TheTurner and Gonsowski (1997) study acrossnorthwestern Kansas used a VIL densitythreshold of 3.25 g m-3 to correctly identify 91% of severe events. However, this thresholdfalsely identified 43 % of the non-severeevents as severe.

Our findings are consistent with studies byTroutman and Rose (1997), Roeseler and

Wood (1997), and Turner and Gonsowski(1997) which indicate that VIL density isoperationally useful across the climaticregimes of eastern New York and westernNew England, northwestern Kansas,northwestern Gulf of Mexico, and acrossmiddle Tennessee respectively. The VILdensity approach appears to be effective as anindicator of hail across various geographicallocations, across areas of varying climatology,and in both cool and warm seasons.

Previous VIL density studies showed that asthe VIL density increased, the maximumreported hail size also increased. Resultsfrom this investigation indicate that as theaverage VIL density (averaged for eachparticular hail size), increased from 3.95 g m-3

to 4.49 g m-3, the average hail size generallyincreased from 0.75 inches (19 mm) to 1.5inches (38 mm) in diameter (Fig. 4). AverageVIL density values fluctuated somewhat ashail size increased above 1.5 inches (38 mm)to 3.0 inches (76 mm); this is likely a result ofthe limited number of cases investigated witha hail size greater than 1.5 inches (38 mm).

This study also sought to determine whetherenhancements and new cell specific valuesand information (CBVIL, improved ST, andenhanced SCIT algorithm) providedsubsequent to the delivery of Build 9 couldimprove the effectiveness of VIL density.VIL density values were calculated using acombination or a portion of CBVIL and ST(GBVIL / ST, CBVIL / ET, and CBVIL / ST).It was hoped that VIL density calculationsusing cell specific data would result inimproved VIL density utility.

Analysis of VIL density values using GBVIL/ ST, CBVIL / ET, and CBVIL / ST indicatethat there is little if any improvement overtraditional VIL density calculations using cellspecific values. VIL density calculations

8

using GBVIL / ST showed the least skill of allof the VIL density calculations using cellspecific data. There appears to be no definitetrend in the distribution of non-severe hailcases [< 0.75 inches (19 mm)] in the data set.The majority of severe hail cases [� 0.75inches (19 mm)] occurred with a VIL densityvalue of 5.0 g m-3 or greater.

VIL density calculations using CBVIL / ET(Fig. 5) and CBVIL / ST (Fig. 6) performedbetter than the GBVIL / ST but still withoutthe skill present in the traditional VIL densitycalculations (GBVIL / ET). In both theCBVIL / ET and CBVIL / ST calculations, thenumber and occurrence of non-severe eventsdecreases as VIL density increases. For bothof these VIL density calculations, the numberof severe cases increases to a maximum as thenumber of non-severe cases decreases. Thisis a pattern similar to calculations using thetraditional VIL density methodology (Fig. 2).However, there is significant overlap ofsevere and non-severe cases for a given VILdensity, resulting in an unacceptable rate ofincorrect identifications of non-severe hailevents as severe. For this reason, VILdensities calculated using cell specific dataappear to be ineffective in identifying severehail events.

This study contained only 55 events (31severe and 24 non-severe) in which the cellspecific data needed to calculate VIL densitiesusing a combination or a portion of CBVILand ST was available. The relatively smallnumber events appears to be too limited foran effective analysis. We will continue toexpand the data set with additional post Build9 cases in order to effectively analyze thepotential utility of VIL density calculationsusing cell specific data.

The criteria for severe hail [current criteria �

0.75 inches (19 mm) in diameter] is expected

to increase to a new criteria by the spring of2000 [future criteria � 1.0 inch (25 mm) indiameter] (Purpura 1998). Traditional VILdensity values (GBVIL / ET) are shown in ascatter diagram (Fig. 7) of ET (thousands offeet) versus GBVIL (kg m-2) using the futuresevere hail criteria [� 1.0 inch (25 mm) indiameter]. The scatter plot includes 44 severehail cases (:) and 110 non-severe cases (q).Results from this study indicate that a VILdensity threshold of 3.70 g m-3 correctlyidentified 91 % (40 of 44) of the future severehail events. However, the 3.70 g m-3

threshold falsely identified 48 % (37 of 77)cases of non-severe hail as severe. In fact, thefalse alarm rate does not drop below 40 %until the VIL density value increases to 4.80g m-3. The 3.50 g m-3 VIL density thresholdassociated with the current severe hail criteriaperformed much better than the 3.70 g m-3

VIL density threshold associated with thefuture severe hail criteria. Fig. 8 illustratesthe decrease in the occurrence of non-severeevents as VIL density values increase. Fig. 8also shows that the number of severe eventsgradually increases as VIL density increasesup through a VIL density value of 4.50 g m-3.Beyond the 4.50 g m-3 VIL density value, thetrend is not clear; likely a result of a limiteddata set.

The overall poorer performance of the � 1.0inch (25 mm) in diameter hail threshold likelyresults from at least two factors. First, thenumber of cases included in this study withhail � 1.0 inch (25 mm) in diameter is limited.Second, spotters, law enforcement officials,and verification procedures are all biased tothe � 0.75 inches (19 mm) in diameterthreshold. For example, a report of hail 0.88inches (22 mm) in diameter would satisfycurrent verification requirements. Because ofworkload restrictions during severe weather,NWS personnel may elect not to seek outreports of larger hail, even though larger hail

9

may have occurred, since the verificationrequirement has already been met. Due to thelimitations previously mentioned, it isdifficult to develop reliable thresholds or evena nomogram to predict the occurrence ofsevere hail � 1.0 inch (25 mm) withoutadditional cases and further research.

7. SUMMARY

Results from this study indicate that there is astrong correlation between VIL density asdefined by Amburn and Wolf (1997) and theoccurrence of severe hail in eastern New Yorkand adjacent western New England. Theseresults indicate that meteorologists duringpotential severe weather events could use VILdensity as a significant input into the warningdecision making process. A VIL densitythreshold of 3.50 g m-3 correctly identified 82% (80 of 97) of severe hail cases in this studyand incorrectly identified 7 % of non-severecases (4 of 57) as severe. The VIL densityvalue of 3.80 g m-3 is significant since itserves as the upper limit of non-severe cases;99 % (69 of 70) of events with a VIL densityvalue greater than or equal to 3.80 g m-3 wereof severe size. Conversely, the VIL densityvalue of 3.28 g m-3 serves as a lower limit forsevere size hail; only 11 % (5 of 45) of eventsless than 3.28 g m-3 were severe. A resultingnomogram (Fig. 3) has been used successfullyby meteorologists at NWSFO ALY, duringboth severe and potential severe weatherevents.

VIL density also appears useful as a means toapproximate the size of the severe hail event.Although indisputable evidence is lacking,findings from this study and other studiesnoted earlier indicate that there is a correlationbetween increasing VIL density andincreasingly large hail. However, some

caution should be used when using VILdensity to predict the size of severe hail.

Further investigation is required to morecompletely determine the role, if any, that cellspecific values (CBVIL and ST) can play inimproving the VIL density concept. In ourstudy, VIL density calculations using CBVIL/ ET and CBVIL / ST preformed the bestamong all of the VIL density calculationsusing cell specific data. However, these cellspecific VIL density calculations were inferiorto the traditional VIL density calculationsusing GBVIL and ET.

The criteria for severe hail is expected toincrease to � 1.0 inch (25 mm) in diameter bythe spring of 2000. First, our study indicatesthat there are far fewer events of hail � 1.0inch (25 mm) in diameter than � 0.75 inches(19 mm) in diameter. Second, the VILdensity technique had much less skill with thelarger severe hail criteria. A VIL densityvalue of 3.70 g m-3 correctly identified 91 %(40 of 44) of the � 1.0 inch (25 mm) indiameter severe hail events but falselyidentified 48 % (37 of 77) cases of non-severehail as severe. The overall inferiorperformance of the � 1.0 inch (25 mm) indiameter hail threshold is likely a result of thelimited number of cases included in this studywith hail � 1.0 inch (25 mm) in diameter.Continued investigation is also necessary todetermine the viability of using VIL density inthe future after a larger severe hail criteria [�

1.0 inch (25 mm) in diameter] is enacted.

10

ACKNOWLEDGMENTS

We would like to thank George MaglarasNWSFO Albany for his helpful insight intostatistics and probability, and Laurie HermesERH SSD for her thorough and timely reviewof this paper. We would also like to thank therest of the staff at NWSFO Albany, NY fortheir helpful suggestions used in thecompletion of this study.

REFERENCES

Amburn, S., and P. Wolf, 1997: VIL densityas a hail indicator. Wea. Forecasting,12, 473-478.

Billet, J., M. DeLisi, and B.G. Smith, 1997:Use of regression techniques topredict hail size and the probability oflarge hail. Wea. Forecasting, 12,154-164.

U.S. Department of Commerce, 1991:Doppler Radar MeteorologicalObservations Part C. FederalMeteorological Handbook 11, 105 pp.[Available from the Office of theF e d e r a l C o o r d i n a t o r f o rMeteorological Services andSupporting Research, NOAA, USDC,Washington, D.C.]

Greene, D. R., and R. A. Clark, 1972:Vertically integrated liquid water - Anew analysis tool. M o n . W e a .Rev., 100, 548-552.

U.S. Department of Commerce, 1994-1997:Storm Data, National Climatic DataCenter, Asheville, NC. [Availablefrom National Climatic Data Center,Asheville, NC 28801.]

Polger, P. D., B. S. Goldsmith, R.C.Przywarty, and J.R. Bocchieri, 1994:National Weather Service warningperformance based on the WSR-88D.Bull. of the Amer. Meteor. Soc., 75,203-214.

Purpura, James, 1998: ChangesR e c o m m e n d e d f o r S e v e r eThunderstorm Warning Criteria.Aware, National Weather Service,Office of Meteorology. [Availableo n - l i n e f r o m http://www.nws.noaa.gov/om/public.html.]

Roeseler, C. A., and L. Wood, 1997: VILdensity and associated hail size alongthe Northwest Gu l f C oas t .Preprints, 28th Conf. On RadarMeteorology, Austin, Amer. Meteor.Soc., 434-435.

Troutman, T. W. and M. Rose, 1997: Anexamination of VIL and echo topassociated with large hail in MiddleTennessee. NWS Southern RegionTechnical Attachment SR/SSD 97-15.Fort Worth, TX.

Turner, R. J., and D. M. Gonsowski, 1997: Areview of VIL density performance atNWSO Goodland, Kansas. Preprints,28th Conf. On Radar Meteorology,Austin, Amer. Meteor. Soc., 370-371.

Witt, A., 1990: A hail core aloft detectionalgorithm. Preprints, 16th Conf. onSevere Local Storms, KananaskisPark, Alberta, Amer. Meteor. Soc.,232-235.

11

_____, and J. T. Johnson, 1993: An enhancedstorm cell identification and trackingalgorithm. Preprints, 26th Intl.Conf.on Radar Meteorology, Norman,Amer. Meteor. Soc., 141-143.

Wyatt, A. L. and A. Witt, 1997: The effect ofpopulation density on ground-truthverification of reports used to score ahail detection algorithm. Preprints,28th Conf. on Radar Meteor., Austin,Amer. Meteor. Soc., 368-369.

12

Figure 1. Scatter diagram of Echo Top (thousands of feet) versus Grid Based VIL (kg m-2) for 97 severe hail (:) cases and 57 non-severe hail cases (q). A few data points are composed of multiple events. Severe hail criteria is � 0.75 inches (19 mm).

13

Figure 2. The number of severe and non-severe events versus VIL density (Grid Based VIL / Echo Top) indicating a decrease in the occurrence of non-severe events as VIL density increases. Severe hail criteria is � 0.75 inches (19 mm).

Figure 3. Nomogram for operational use depicting Echo Top (thousands of feet) and Grid Based VIL (kg m-2) with the derived warning thresholds. Area in white between severe and non-severe regions represents indeterminate events with limited confidence. This nomogram should be used as guidance only and not used exclusively in the warning decision making process. Severe hail criteria is � 0.75 inches (19 mm).

14

Figure 4. Average VIL density (Grid Based VIL / Echo Top) versus hail size in inches. VIL density generally increases as hail size increases up through 1.5 inches (38 mm).

Figure 5. The number of severe and non-severe events versus VIL density (Cell Based VIL / Echo Top) indicating a significant decrease in the occurrence of non-severe events as VIL density increases. Severe hail criteria is � 0.75 inches (19 mm).

15

Figure 6. The number of severe and non-severe events versus VIL density (Cell Based VIL / Storm Top) Severe hail criteria is � 0.75 inches (19 mm).

16

Figure 7. Scatter diagram of Echo Top (thousands of feet) versus Grid Based VIL (kg m-2) for 44 severe hail (:) cases and 110 non-severe hail cases (q). A few data points are composed of multiple events. Severe hail criteria is �1.0 inch (25 mm).

17

Figure 8. The number of severe and non-severe events versus VIL density indicating a decrease in the number of non-severe events as VIL density increases. Severe hail criteria is � 1.0 inch (25 mm).