The Weather Company, North Sydney, NSW 2060, Australia

The Sydney Hailstorm of April 14, 1999: Synoptic descriptionand numerical simulation

Bruce W. Buckley, Lance M. Leslie, and Yuqing Wang

With 10 Figures

Received February 17, 2000Revised July 25, 2000

Summary

During the evening of April 14, 1999 an intense hailstormstruck the most densely populated region of Australia, theeastern suburbs of Sydney. This thunderstorm, whichtransformed into a high precipitation supercell when itmoved into a region of enhanced surface moisture con-vergence and increased helicity on the coast, maintained itsidentity for 5.5 hours. It produced the largest veri®ed hail inAustralia's history with the biggest stones being 11 cm indiameter. A microburst was recorded at Sydney Airport. Thedamage in¯icted by this hailstorm was immense with threedeaths, numerous injuries and insured losses exceeding $1.7billion Australian dollars. This storm is the most expensiveAustralian natural disaster since severe weather recordscommenced in 1975.

The thunderstorm initially formed from surface heatingof relatively dry air in a low shear environment but wasadvected by middle level west to south westerly winds into aregion where the surface to 500 hPa wind shear had increasedto 17 msÿ1 west south westerly. The storm relative helicity inthis region was ÿ180 m2sÿ2, in the layer between the surfaceand 700 hPa. Diagnostics from the 1500 Australian EasternStandard Time (AEST) radiosonde released from SydneyAirport, 150 km north of where the thunderstorm initiallyformed, are thought to be representative of the pre-stormenvironment. The Convective Available Potential Energy(CAPE) was moderately high at 1713 J/kg with a relativelylow Convective Inhibition (CIN) value of 50 J/kg. The TotalTotals Index (TT) was 55 and the Surface Lifted Index (SLI)was ÿ5.5, capable of supporting severe convection. Thefreezing level was at 2900 m, near average for the time of theyear. Convective cloud tops would be expected to reach thetropopause at 250 hPa. The coastal environment was assessedas being able to support a supercell thunderstorm.

A preliminary high resolution numerical simulation of thesevere thunderstorm has been conducted. The model wastriply nested, with its highest resolution grid spacing being1 km. It incorporates a multi (six) water ± ice phase micro-physics, enabling it to simulate hail growth associated withsupercell developmet. The initial generation and subsequentnorthward propagation of a hail-producing thunderstorm arecaptured in this simulation.

1. Introduction

Severe thunderstorms, most notably supercells,are a recurring cause of severe weather acrossNew South Wales (NSW), with signi®cantimpacts on Australia's largest city, Sydney (seeFig. 1 for location of places mentioned in thisstudy). Severe phenomena associated with thesestorms (tornadoes, hail 2 cm in diameter orgreater, wind gusts in excess of 48 knots(23 msÿ1), ¯ash ¯oods de®ned as a one in 10year one hour rainfall) permit only short lead timewarnings, issued by meteorologists who are severethunderstorm specialists.

April is not a month when large hail is likely inSydney. The hailstorm that is the subject of thisstudy occurred on April 14, 1999, outside theNSW severe thunderstorm season of October toMarch. The synoptic situation prevailing at thetime was signi®cantly different from other Sydneysupercell situations experienced over the past

Meteorol. Atmos. Phys. 76, 167±182 (2001)

decade. Figure 2 shows the synoptic situation for0900 Australian Eastern Standard Time (AEST)on the day of the very destructive January 21,1991 (Bureau of Meteorology 1997) supercell,more `̀ typical'' of a severe weather pattern overthe central NSW coast.

Late in the afternoon of April 14, 1999 athunderstorm was identi®ed on radar, south westof the town of Kiama, 120 km south of Sydney.Subsequent radar images (see Fig. 3a±c) showed

the storm moving towards the northeast, beingsteered by the west to south westerly middle levelenvironmental air¯ow that prevailed across thearea. The storm moved over the Tasman Seawhere its structure changed abruptly. The thunder-storm became a high precipitation supercell(Moller et al., 1990) and began to move atapproximately 30� to the left of the environmentalsteering ¯ow. The altered path of the supercellthunderstorm brought it over the populous easternsuburbs of Australia's largest city, Sydney. Thestorm also passed over Australia's principalairport, Sydney Airport, during the evening peakperiod (see Fig. 3b). It reached peak intensity soonafter (Fig. 4a), with the radar cross section(Fig. 4b) showing its structure at this time. In a®ve-and-a-half hour period giant hail of up to11 cm diameter from the storm created a path ofdestruction that led to the death of three peopleand produced an insured damage bill in excess of$1.7 billion Australian dollars, making thisthunderstorm the most expensive natural disaster,in dollar terms, in Australia's recorded history.

Although numerical model data has beenemployed in NSW to identify broad areas ofincreased potential for severe thunderstorm devel-opment (Mills and Colquhoun, 1999), to date nonumerical technique has been available thatdirectly simulates the growth of large hail. Therehave been numerous successful simulations ofsupercells in other countries such as the USA andEurope and these are discussed brie¯y below.However, the purpose of the present study is toconcentrate on the simulation of a supercell in theAustralian operational context. We thereforesimply point to a number of key references anddiscuss brie¯y how they differ from our aims.Lack of detailed surface and upper air observa-tions are a serious impediment to the successfulsimulation of locally severe convection in theAustralian environment. To achieve our goals,time is required to move this technique intooperations. The numerical modelling capacity isalmost available, and is expected to take a furthertwo to three years.

In Sect. 2, the climatology of severe thunder-storm events in NSW, with particular attentiongiven to the Sydney area, is summarized in orderto better understand the rarity of the April 14hailstorm. The details of the hailstorm event andits impacts are outlined in Sect. 3, with the

Fig. 1. Location map of south eastern Australia

Fig. 2. Synoptic situation for the January 21, 1991 supercellthunderstorm

168 B. W. Buckley et al.

synoptic in¯uences that triggered the stormdiscussed in Sect. 4.

The hailstorm was poorly predicted in real-timeby the Bureau of Meteorology. In an attempt toincrease the lead-time and predictability of futurestorms of this type, a high resolution post-eventnumerical simulation was carried out for theeastern seaboard of central NSW, covering the fulllife cycle of the severe thunderstorm. Thisinvestigation constitutes the ®rst attempt to do so

for eastern Australian supercell thunderstorms.The numerical simulations are described in Sect.5. Conclusions and areas requiring furtherresearch are discussed in Sect. 6.

Elsewhere, notably in the USA, Europe andAsia, mesoscale numerical models models havebeen used for several decades to simulate severeconvective systems. For brevity we con®ne ourshort summary of these simulations to researchand operations in the USA. Amongst the earliest

Fig. 3. Radar plan position indicator images from theLetterbox radar showing the severe thunderstorm and otherconvection during April 14, 1998 at the following times: a1700 AEST; b 1900 AEST; and c 2100 AEST. The redoutlines are nowcast positions at 10 minute intervals

The Sydney Hailstorm of April 14, 1999: Synoptic description and numerical simulation 169

mesoscale simulations of severe convection, werethose of Klemp and Wilhelmson (1978), andZhang and Fritsch (1986). Since that time muchprogress has been made in the modelling of severestorms. In the context of this paper, modelling ofsevere convection is divided into two categories.The ®rst category is the very high resolutionmodelling (1 km down to 100s of metres) of thelife cycles of speci®c severe convective systemssuch as individual supercells, squall lines andlocal circulations that produce very heavy rainfall.Excellent results have been achieved and as thestudies are far too numerous to mention, we referonly to a very small selection of these. We refer to

only a small selection of these (e.g., Jewett andWilhelmson, 1996; Davis et al., 2000; and Guoet al., 2000). The second category, within whichthe model used by the present authors ®ts, is themesoscale modelling of severe convection onhorizontal scales of approximately 5 to 15 km.This category is probably best referred to asmesoscale NWP. The aim of this approach is theroutine prediction, in real-time, of severe convec-tion over relatively large areas. In many cases themodels are either located at major NWP centres orare community models such as MM5, ARPS,RAMS, COAMPS, Meso-ETA and MASS mod-els. These model acronyms are so well known thatthere is no need to further de®ne them or toprovide individual references for each of thesesystems. Mass and Kuo (1998), and Kuo (2000)provide an overview of these models.

Our study ®ts squarely into the second categoryof a mesoscale NWP model adapted for real-timeoperational usage. As such, as far as the authorsare aware, this simulation is the ®rst successfulattempt of this kind to be carried out using amodel developed in Australia. We report on it herefor that reason and also because of the singularimportance of the hailstorm and as one of theearly products of our program of research andoperations in the modelling of severe convectionin Australia.

2. Climatology of severe thunderstormsin New South Wales

Severe thunderstorm reports used to generateclimatologies, including hail size, are largelysupplied by a 1,300 strong volunteer storm spotternetwork. This network, sparse away from themain urban areas, is quite substantial across thecity of Sydney itself. Additional reports aresupplied by the Bureau's paid observers withwind and rain information provided by a networkof approximately 100 automatic weather stations.April is a month when severe thunderstorms areuncommon in New South Wales, as illustratedin the histogram in Fig. 5a, which shows theoccurrence of severe thunderstorms for NSW, ofall types, by month. This is particularly the casefor Sydney. The severe thunderstorm season forNew South Wales, including Sydney, is de®ned asthe start of October through until the end ofMarch. The marked seasonality of large hail for

Fig. 4. Radar plan position indicator and correspondingrange height indicator images from the Letterbox radarshowing a plan, and b elevation views of the severehailstorm at peak intensity (1920 AEST)

170 B. W. Buckley et al.

Sydney is clearly evident from the ®gure. Theclimatology of hail size, showing the relativenumber of events where large or giant hail hasoccurred in the Sydney metropolitan area, is givenin Fig. 5b. Large hail is de®ned as having adiameter of 2 cm or greater. Giant hail has adiameter of 5 cm or greater. Giant hail is reportedon approximately 25% of severe hailstorm eventsin Sydney. The peak month for large hail in NSWis December with the greatest number of reports ofgiant hail occurring one month earlier in Novem-ber. The tendency for large hail to be reported inthe ®rst half of the severe thunderstorm season isattributed to the fact that incursions of middlelevel cold pools of polar origin in a moderatelysheared environment are more common during thetransition period from winter through to summer.During this time NSW falls under the in¯uence ofthermal troughs associated with mid-latitude

frontal systems that progressively retreat south-wards as summer approaches. This also is the timeof the year when the highest values of CAPE(Moncrieff and Green, 1972) are likely. The latterhalf of the severe thunderstorm season is charac-terized by a low shear environment with few coldpools present. Prior to this event there had beenonly one report of large hail in Sydney during themonth of April, a report of 2.5 cm hail in 1994.There have only been three reports of giant hail inall of NSW during the month of April. However,severe hailstorms are a recurring feature of theweather in the Sydney basin, particularly duringthe warmer months. The hailstorm was theseventh to produce hail with a diameter of 6 cmor greater in Sydney this decade, with otherstorms occurring on March 6 and 18, 1990,January 21, 1991, February 12, 1992, October 28,1995 and November 12, 1997. Unlike the sixother severe thunderstorms that initially formedon the ranges inland from the eastern coast ofAustralia and then moved north east acrossSydney, this storm moved towards the north northeast, approaching Sydney from over the TasmanSea. There has been no other severe hailstorm inSydney's recorded history that has followed asimilar track.

3. The hailstorm of April 14, 1999

April 14, 1999 began with calm winds andscattered middle and high-level cloud. Surfacedew points were moderate at around 18 �C at 0600AEST. There was no pre-existing convectionevident in Sydney or the surrounding regions onthe weather watch radars that provide weathersurveillance across this region.

Weather radar coverage of the greater Sydneyregion in provided by two radars. The ®rst is a1.8� beam-width S Band radar located at Letter-box Hill, some 60 km to the south of the city. Thisradar provides continuous weather watch cover-age for the greater Sydney area. This coverage iscomplemented by a 0.9� beam-width C BandDoppler radar located at Kurnell, 15 km south ofthe city. The latter radar is newly installedspeci®cally to provide enhanced monitoring ofevolving wind and precipitation ®elds nearSydney, with particular emphasis on monitoringthe pre-storm environment. They both operate in avolumetric scan mode, scanning at 15 different

Fig. 5a. Histogram illustrating the climatology of severethunderstorms in the state of New South Wales; b Histogramillustrating the climatology of hail size across Sydney

The Sydney Hailstorm of April 14, 1999: Synoptic description and numerical simulation 171

radar tilt angles, with a 10 minute repetitionperiod.

The earliest radar signature for this thunder-storm was at 1625 AEST on the afternoon of April14, 1999. The storm was located to the south westof the town of Kiama, with Fig. 6 illustrating theradar derived path of the storm cell as de®ned bythe surface projection of the volumetric 35 dBZre¯ectivity contour. This contour was selected soas to eliminate less signi®cant convective activityfrom the display, a technique routinely used insevere thunderstorm forecasting in NSW. Thecentral box provides a simple measure of theintensity of the thunderstorm for each volumescan. If the 49 dBz radar re¯ectivity contour, asmeasured by the S Band radar at Letterbox,reaches 8 km above sea level the box is shaded,otherwise it is left open. This criterion wasselected because locally it has been identi®ed thatthe threshold for large hail occurs when the49 dBZ contour attains a height of 8 km above sea

level, with the re¯ectivity contours being dis-played in 3 dBZ intervals on operational displays.It should be noted that there were two radar scanswhich depicted the initial stages of convection thatare not shown on this track due to the lack of a35 dBZ contour at this stage of the storm'sdevelopment. It was the ®rst thunderstorm todevelop that day, although there had been short-lived towering cumulus development over the landareas east of the ranges throughout the afternoon.The storm initially moved towards the north eastuntil it reached the warm waters of the TasmanSea. Sea surface temperature (SST) measurementsfrom an instrumented wave rider buoy moored3 km off the Sydney coast indicated the 1 metredepth water temperature was 23.6 �C. At this pointthe thunderstorm structure transitioned rapidlyinto that of a high precipitation supercell, asshown in the radar image from the Letterbox radarin Fig. 3a. Its movement altered towards the northnorth east. The thunderstorm, now severe, moved

Fig. 6. Path of the thunderstorm,derived from the surface projec-tion of the most extensive35 dBZ radar re¯ectivity contouras viewed by the S Band weatherwatch radar located at Letterbox

172 B. W. Buckley et al.

across the town of Bundeena then over Sydney'sInternational Airport (see Fig. 1). Hail up to 8 cmin diameter fell in this area (see Fig. 6), rendering23 commercial aircraft insuf®ciently airworthy forcommercial ¯ight operations. The storm reachedpeak intensity as it moved over the Sydney suburbof Kensington towards the city centre. TheLetterbox radar Plan Position Indicator (PPI) andRange Height Indicator (RHI) scans for this timeare shown in Fig. 4a,b. Peak re¯ectivity valuesin the vicinity of 65 dBZ were recorded at thistime and the 49 dBZ contour attained a height of10.5 km above ground level. The extensive sus-pended hail and overhang are quite clearly visiblein the RHI. It was in this area that the largestcon®rmed hail over to fall in Australia wasreported, with a maximum diameter of 11 cm.

A signi®cant regeneration occurred on the left¯ank of the thunderstorm as it moved to thenorthern side of Sydney Harbour. The very earlystages of this cell can be seen as the small celllocated approximately 10 km to the north west ofthe major cell in Fig. 4. No additional surface orupper air information is available to fully explainthe mechanisms behind the regeneration of thisnew but less severe supercell. However, carefulstudy of the sequence of radar images hasconcluded that this regeneration was triggeredby the enhanced convergence from the con¯uenceof the downdraft produced by the large hail-fallwith the warm, moist air associated with theleading edge of the low level southeasterlychange. This left ¯ank development would alsohave been favoured by additional lift provided by

the rising topography in this region. Although lessintense from this point onwards, the stormcontinued to produce large hail until the thunder-storm eventually moved offshore near BrokenBay, some ®ve and a half hours after it was ®rstdetected. Such longevity of an individual thunder-storm is unusual for New South Wales, even for asupercell.

The damage the hailstorm produced during itspassage of the eastern suburbs of Sydney washuge. Figures from the Insurance Council ofAustralia (Henri 1999, personal communication,July 1999) placed the insured damage bill inexcess of $1.7 billion Australian dollars. Threedeaths were attributed to this thunderstorm. Anestimated 170,000 people were adversely affectedby the hailstorm. The emergency clean-up opera-tion was one of the largest in Australia's history.Volunteers from the State Emergency Service unitin NSW were fully committed for several weeks,with the personnel from the NSW Rural FireService contributing 215,000 hours of effort to theclean-up. The response effort was estimated bythe emergency services to be eight times greaterthan the previous largest severe thunderstormresponse effort required in New South Wales.

4. Synoptic description

The broad scale synoptic situation during April14, 1999 was relatively benign, as is shown in the1600 AEST mean sea-level pressure analysis ofFig. 8a. The pressure gradient across the centralNSW coastline was weak. The tail end of a cold

Fig. 7. Photograph comparing hail, of diameterup to 8.5 cm with a tennis ball and damagedapple (photo courtesy of Mike De Salis)

The Sydney Hailstorm of April 14, 1999: Synoptic description and numerical simulation 173

front that was moving across the Tasman Sea wasbrushing the southern NSW coastline. A ridgefrom a 1030 hPa high pressure system locatedover the Southern Ocean to the south of theAustralian continent was developing along the

southern NSW coastline in the wake of the weakcold front. Ahead of the weak frontal system wasa deep layer of moderate strength west to southwesterly winds, extending from 1,000 m throughto the tropopause, located near 10,000 m. Figure

Fig. 8a. Australian region sea-level pressure (SLP) analysis for 1600 AEST April 14, 1999; b Time series of upper level windsfrom selected aircraft operating through Sydney Airport between 0900 AEST and 1900 AEST April 14, 1999. Aircraft reportsare routinely reported in feet above mean sea level; c Upper level temperature and humidity pro®le for Sydney Airport in thepre-storm environment at 1500 AEST April 14, 1999

174 B. W. Buckley et al.

8b shows a time series of upper level winds,routinely reported in feet above sea level, fromselected aircraft ¯ights at approximately hourlyintervals operating into and out of Sydney Airportduring the afternoon and evening of April 14,1999.

Temperatures ahead of the cold front were afew degrees warmer than normal for April.Maxima in the western suburbs of Sydneyclimbed to around 26 �C. However, as the 1500AEST upper air ¯ight from Sydey Airport shows(Fig. 8c), there was only a very shallow layer ofsurface moisture, with the middle and upper levelsremaining moderately dry, apart from a layer ofmoisture between 3,000 m and 4,000 m in the pre-storm environment. However, there was a moder-ately high CAPE (Moncrieff and Green 1972) atthis time with a computed value of 1713 J/kg andthe TT Index was a high 55. The SLI of ÿ5.5indicated the environment had the potential tosupport severe convection. The presence of aweak low-level isothermal layer resulted in theCIN value of a relatively low 50 J/kg. It was onlyin the region just inland of Kiama that the strongsurface heating was suf®cient to break throughthis stable layer to form a thunderstorm. Thecritical changes in low-level wind structure thatled to the transition of the ordinary thunderstorminto a well organized supercell are well illustratedin the wind pro®les of Fig. 8b. This was assistedby warmer than normal SSTs, with an instrumen-ted buoy moored 3 km east of Sydney's coastlinerecording a value of 23.6 �C at 1 m depth. Theseanomalously warm seas provided an additionalsource of moisture for the thunderstorm. Therewas an increase in helicity caused by a subtlestrengthening and deepening of the pre-frontaleasterly that preceded the arrival of the weak coldfront. The ®rst evidence of an easterly can beseem from the 1200 AEST aircraft wind pro®le,with the depth of the easterly deepening toapproximately 1,500 m (5,000 feet) by 1900AEST. At 1500 AEST the storm relative helicityat Sydney Airport for the layer from the surface to3000 m was ÿ180 m2sÿ2, a value that continuedto increase throughout the afternoon. Thisincreased helicity is an indication of the increasedability of the vertical wind structure to sustainorganized up-drafts and down-drafts in the storm.The boundary layer moisture convergence alsoincreased as this change approached, with the

moist easterly air¯ow reaching its maximumdepth of 1500 m immediately before the stormarrived. The moisture and wind shear available tothe thunderstorm was therefore at a maximumright on the change. A weak and shallow seabreeze formed shortly before midday on April 14,1999. Easterly winds were only about 6 knots(3 m/s) and con®ned to the lowest 300 m of theatmosphere. The sea breeze had the effect ofmasking the movement and intensity of thesynoptic scale low level south easterly changethat was concurrently moving up the NSWcoastline. By 1500 AEST the sea breeze hadincreased only slightly, averaging 10 knots (5 m/s)and extending 600 m into the atmosphere. Theenvironmental ¯ow immediately ahead of thethunderstorm is illustrated by the aircraft derivedwind pro®le at 1900 AEST (Fig. 8b). An increaseof the wind at 900 m to 16 knots (8 m/s) is evident,as is the sudden increase in the depth of theboundary layer easterly air¯ow to 1500 m. Postanalysis of the available data showed that ashallow mesoscale low had formed on a prefrontaltrough which preceded the weak frontal system.This development occurred on the edge of theescarpment immediately adjacent to the coastbetween Wollongong and Sydney, further enhan-cing the low-level convergence and assisting inthe transition of the thunderstorm into a supercell.

Once the thunderstorm air¯ows were organizedby the environmental ¯ow, it propagated to theleft of the environmental steering ¯ow, main-taining its supercell structure for over ®vehours. On two occasions during its life amiddle-level mesocyclone was detected by theKurnell Doppler weather radar. Other thunder-storms subsequently developed, with one thatclosely followed the track of the major storm. Itproduced a short lived burst of hail up to 3 cm indiameter. However, none of the subsequentthunderstorms displayed a severity close to thatof the initial storm.

5. Numerical simulation of the April 14,1999 hailstorm

a) Motivation for the numerical studyof the hailstorm

The mode of operation of the severe thunderstormwarning service in Australia has two phases. The

The Sydney Hailstorm of April 14, 1999: Synoptic description and numerical simulation 175

®rst phase is the identi®cation of general atmo-spheric conditions conductive to the formation ofconvective weather, with special attention given tosubareas where severe convection is likely. Thisevaluation normally takes place early in the dayand provides advance identi®cation of the regionsthat are most likely to require closest attention byskilled operational forecasters as the day pro-gresses. A decision tree approach developed byColquhoun (1998) is used to re®ne further theprocess, to identify sub-regions conductive to theoccurrence of supercells, tornadoes, microburstsand storms with a high ¯ash ¯ood potential. Thistechnique has been incorporated into the Austra-lian Bureau of Meteorology's Mesoscale LocalArea Prediction System, MesoLAPS (Puri et al.,1998), to provide geographical representations ofthe areas with severe thunderstorm potential(Mills and Colquhoun, 1999). On this occasionthe assessment, based upon available observationsand numerical guidance, was that the environmentwas too dry, the shear too low and the steeringwinds directed too strongly offshore for severethunderstorms to be of concern for the SydneyBasin. The second phase of the warning servicerelies heavily on the identi®cation of severethunderstorm signatures on weather watch radar,complemented by a range of other informationsources including storm spotter reports, satelliteimagery, surface weather observations fromtrained observers or from automatic weatherstations, upper level observations from aircraftand weather balloons, and information from alightning detection network that covers the easternpart of Australia. For greatest success the ®rst phaseof the warning service must be as accurate aspossible, as it ensures that severe weatherspecialists are available to monitor the availablerealtime information and to issue short lead time,location speci®c, severe thunderstorm warningsand advices.

In the case of the April 14, 1999 hailstorm, theoperationally available guidance did not indicatean environment capable of supporting severethunderstorms over the land. The second phaseof the warning service was therefore ineffective.The very ®ne scale of the meteorological factorsthat were critical in triggering the transition of theinitial non-severe thunderstorm into a supercellthunderstorm, in particular the strength, depth andtiming of the south easterly change and the

erosion of the low-level temperature inversion,dictated that higher resolution numerical simula-tions than those currently operationally availablewere required. Land ± sea and terrain interactionshave been found by operational forecasters to beof importance in triggering severe thunderstormevents along the NSW coast line. A very highresolution numerical weather prediction modelwhich included a recently developed multi (six)water±ice phase microphysics parameterisationscheme was therefore selected for this study.The complex cloud microphysics package wasobviously necessary for the prediction of largehail producing convection.

It was noted above in the Introduction that therehas been very limited success in modeling signi-®cant convective weather events in Australia todate unlike, for example, the USA and Europe. Inthese regions modellers have achieved success intheir simulations of various severe weatherphenomena. However, none have adopted theapproach taken for this study. The sparse surfaceand upper air observing networks over the largeocean areas surrounding Australia have been asigni®cant contributing factor working against thesuccessful operational simulation of ®ne-scalesevere weather events. One earlier study of a¯ash ¯ood produced by severe convection on theNSW east coast was conducted by Speer andLeslie (1998). It was captured well because themain ingredients were moisture availability andstrong topographic forcing. The severe thunder-storm studied here developed relatively late in theafternoon with the ®rst radar signature evident at1625 AEST. Currently, operational forecastersrely heavily on their previous experience of severeweather events to estimate their time of develop-ment. The ability of the numerical model toprovide guidance on the time of initiation of thethunderstorm is of great interest. These factorsprovided the motivation for the very high resolu-tion numerical simulation of this storm. Thenumerical model chosen for the present investiga-tion was the model used in the Speer and Leslie(1998) study, but enhanced by the explicit cloudphysics. It is referred to as the High Resolutionnumerical model, HIRES, developed at theSchool of Mathematics, The University of NewSouth Wales, Sydney (see, for example, Leslie andSpeer (1998a, 1998b), Leslie and LeMarshall(1997), Leslie and Purser (1997), Leslie and

176 B. W. Buckley et al.

Skinner (1994)). In a recent study, Buckley andLeslie (2000) have used HIRES at a resolution of2 km to simulate detailed temperature and moist-ure gradients across the Sydney basin, showing itsability to resolve features important for theprediction of convective scale features.

b) HIRES con®guration

The dif®culty in accurately predicting the genesisareas of severe convection in Australia is wellknown to all operational meteorologists. For thisreason the model initialisation time was selectedat 0000 UTC (1000 AEST) on April 14, 1999, thetime of maximum density of surface and upper airobservations in Australia. This would provide thenumerical model with the most comprehensivesurface and upper air analysis ®elds upon which tobase its predictions and an adequate lead time ofabout 6 hours.

HIRES was ®rst run across the entire Australianregion domain at a horizontal resolution of 50 kmand sixteen levels in the vertical. It was then selfnested at 10 km then 1 km over successivelysmaller domains (see Fig. 9). The vertical levelswere most closely spaced in the boundary layerand the resolution of the orography was set at 2minutes. This approach was used to ensure therewould be no boundary problems in the area ofinterest caused by nesting the high-resolutionmodel directly in another prediction system. Theresolution of the topography around the eastern

sea-board was re-interpolated with the increase inmodel resolution from the original 20 topographi-cal data-set to ensure the key topographicinteractions were adequately simulated. The50 km resolution run of HIRES was nested inthe Australian Bureau of Meteorology LimitedArea Prediction System, LAPS, (Puri et al., 1998).This model domain covered the area from 10� S to45� S and 105� E to 170� E. The 10 km domainwas reduced to 29� S to 44� S and 137� E to155� E. This domain was selected as it is suitablefor future operational purposes, given a suf®-ciently fast computing platform, as it is broadenough to incorporate key mesoscale meteorolo-gical factors important to the development of thesouth easterly change that was critical in thisinstance. The dynamic effects of the cold frontmoving across the Tasman Sea and the ridge thatwas developing in its wake needed to beaccurately modelled. This domain also minimisesthe in¯uences of any unwanted boundary effectsover the key southern New South Wales coastalregion. The ®nal, 1 km resolution, simulation useda much smaller domain of 32.5� S to 35.5� S and149.5� E to 152.5� E.

The in¯uence of the warm sea surface tem-peratures in the East Australian Current that ¯owssouthwards on the western edge of the TasmanSea were well captured by the input of seven dayaverage sea surface temperatures with 2 minuteresolution into the model. Fundamental to theprediction of the severe convection was the

Fig. 9. Diagram showing the three domains for the triplenesting of HIRES as employed in this study

Table 1. Key model parameters used in the high resolutionnumerical prediction model, HIRES, for this simulation

HIRES model feature Details

Horizontal resolution 50 km, 10 km, 1 kmNumerical scheme Split explicitVertical levels 31, concentrated in boundary

layer and near tropopauseAssimilation 6 hour cyclingInitialisation Diabatic, dynamicOrography 2 minuteSurface layer scheme Mellor±Yamada 2.5Boundary-layer scheme Louis scheme (ECMWF)Radiation scheme Fels±SchwarzkopfPrecipitation scheme Explicit microphysicsSea surface temperatures 7 day average, 2 minute

resolutionBoundary conditions Bureau of Meteorology LAPS

The Sydney Hailstorm of April 14, 1999: Synoptic description and numerical simulation 177

inclusion of a complex convection scheme intoHIRES that contained the all important water andice phase processes into the numerics. Thisconvection scheme was used in the model tosimulate hail growth in any convective cellsdepicted by the simulation. HIRES employs ahigh order split ± explicit numerical scheme, a 6hour cycling assimilation scheme and a level 2.5Mellor±Yamada surface layer scheme. The Fels±Schwarzkopf scheme was used to simulateradiation effects. Table 1 provides a summary ofthe key model parameters used in this very high-resolution numerical simulation.

c) Parameterisation of microphysicalprecipitation processes

The key to the successful simulation of large hailin the April 14, 1999 event was the inclusion of acomprehensive suite of schemes representing thecloud droplet and precipitation growth processes.For complete details see Wang (1999).

d) Results of HIRES run initialized0000 UTC April 14, 1999

The HIRES simulation of this extreme hail eventwas initialised with the 0000 UTC April 14, 1999LAPS analysis. Although not an ideal con®gura-tion for future operational application, the largerdomain of the LAPS model ensures the broaderscale developments were accurately captured bythe higher resolution simulations. This was thelatest time a very high resolution model could berun given the Australian operational schedule,because there are only two major upper airobservation times each day, at 0000 UTC and1200 UTC. It should also be remembered thatthe major impact of the severe thunderstorm thatwas to develop occurred between six and tenhours after this initialisation time. For the modeloutput to be useful to assist in the delivery of atimely warning service, the very high resolutionmodel results must be available prior to the initialonset of convection, which was at 0625 UTC inthis case. Suf®cient time is required for severeweather forecasting staff to be brought in tooperate the radar based short lead time warningservice.

The results from the numerical simulation wereencouraging. A narrow band of liquid precipita-

tion of convective origin was generated with anorientation parallel to the coast, as is shown in Fig.10a. A precipitation rate of up to 25 mm/hour isdepicted, close to the precipitation rate recordedduring the initial stages of development of thestorm. It is evident immediately that the orienta-tion of this precipitation is different to the middlelevel steering winds, indicating that it was beingin¯uenced by low level land-sea and topographi-cal in¯uences. Another area of precipitation isshown over the west of the Sydney basin. Thisprecipitation is clearly less well organised. Inreality this corresponds to the area where othernon-severe thunderstorms, not described in thispaper, formed midway through the major event.The amount of precipitation actually recorded inthis area was less than that in the simulation.Closer scrutiny of the hourly model outputshowed that the organised coastal band of preci-pitation commenced as a localised region ofintense convection in the Kiama region atapproximately 0600 UTC (see Fig. 10b), veryclose to where the initial thunderstorm was ®rstidenti®ed on radar and at a similar time of the day.It should be recalled that 0625 UTC was theearliest radar detection of the convective cell thatwas to become the supercell. The region oforganised precipitation extended progressively tothe north north east, analogous in behavior to thesupercell that formed on the afternoon of April 14.This modelled precipitation had a high icecontent, with an area of giant hail predicted nearWollongong, the area where the thunderstorm ®rstturned into a supercell. Fig. 10c shows a smallarea where hail up to 8 cm at ground level wasmodelled, which would have alerted operationalmeteorologists of the potential for a severe hailproducing thunderstorm to form.

The band of simulated liquid precipitation wasdisplaced to the west of where it actually occurredby approximately 15 kilometres. However, thehail swath was more accurately depicted. Themodelled hail swath in Fig. 10c and the radarderived path of the hail bearing supercell in Fig. 6are quite close. Moreover, the model was runwithout the bene®t of a data assimilation schemethat could ingest the detailed asynoptic data thatwas crucial in identifying the ®ne-scale structureof the event. Further investigation is planned,incorporating more comprehensive data analysistechniques and the use of synthetic data in an

178 B. W. Buckley et al.

attempt to determine the extent to which thesepredictions may be improved.

In Fig. 10d, a vertical time section is presentedof the high resolution modelled cloud ice content(displayed in units of 10ÿ3 g m3) for the grid pointcentred on Sydney Airport is presented. SydneyAirport was selected rather than an area furthersouth where the modelled storm intensity wasgreater, as the airport is the location where acontinuous record of surface observations aremade by quali®ed weather observers. The simula-

tion shows the main cell crossing the airport in thespace of 40 minutes, commencing at 1815 AEST.It is noted that evaporation is occurring beneaththe cells, with no surface impact expected fromthe ®rst of the three cells depicted. The verticalextent of the cells is under-predicted at thislocation, as the accumulated surface hail output inFig. 10c shows. Further modelling experimentsare planned to see if the intensity of the thunder-storm can be maintained for a longer period thanthe current model predictions. In reality, this

Fig. 10a. One kilometre horizontal resolution HIRES model output showing the accumulated liquid precipitation (mm) for the6 hour period from 1300 AEST to 1900 AEST April 14, 1998. This interval encompasses the period of the severe hailstorm; bOne kilometre horizontal resolution HIRES model output showing the liquid precipitation rate (mm/hour), for the hour ending1600 AEST (0600 UTC) April 14, 1999; c One kilometre horizontal resolution HIRES model output showing the accumulatedmaximum hail size, at the surface, for the 10 hour period ending 2200 AEST (1200 UTC) April 14, 1999; d One kilometrehorizontal resolution HIRES time series of cloud ice content �10ÿ3 g m3� for Sydney Airport spanning the time period of thesevere hailstorm event of April 14, 1999

The Sydney Hailstorm of April 14, 1999: Synoptic description and numerical simulation 179

modelled storm cell arrived approximately oneand one half hours earlier than the actual storm.The existence of a weak leading convective celland a trailing convective cell of intermediateintensity arriving one hour after the main stormcell was similar to the sequence observed at theairport.

6. Discussion and conclusions

This paper describes the key synoptic featuresassociated with the supercell thunderstorm thatformed on the afternoon of April 14, 1999 on theeast coast of Australia and moved across easternSydney. Hail of up to 11 cm in diameter wasproduced by this thunderstorm, being the largestcon®rmed hail every to be recorded in Australia.In dollar terms, it was the most expensive naturaldisaster in Australia's history with a damage billexceeding $1.7 billion. A climatology of severethunderstorms in NSW was prepared spanning the10 year period from January 1988 to mid 1998.The climatology shows severe thunderstorms to be

rare for April, although a recurring phenomena forthe Sydney region at other times of the year. Inparticular the April 1999 storm was unprece-dented in its severity for April over more than 200years since records began. The climatology alsocon®rmed the strong seasonality of severe thun-derstorm activity across the state of NSW. Mostsevere thunderstorms are shown to occur betweenthe months of November to February with theshoulder months of October and March beingmuch reduced but still signi®cant. With a life spanof ®ve and a half hours, this supercell was foundto be long lived in comparison to other supercellsthat have impacted upon the Sydney region duringthe past decade.

The synoptic analysis revealed the conditionsprevailing during the morning and early afternoonwould not support a supercell thunderstorm due toa lack of low level moisture and very limitedvertical wind shear. The diagnosis was con®rmedby the formation of a single thunderstorm that wasnon-severe in its early stages. This thunderstormeventually became the supercell. Increased ver-

Fig. 10 (continued)

180 B. W. Buckley et al.

tical shear and a good low level moisture supplyare essential for supercell formation. The analysisrevealed the development of a shallow southeasterly change that was not well forecast becauseit was heavily masked by the pre-existing, butweak, easterly sea breeze. This change produced asudden increase in vertical wind shear and,coupled with the anomalously warm SST overthe Tasman Sea, increased the low-level moistureavailability in the region near Wollongong. Thethunderstorm experienced a rapid transition into asupercell when it encountered the region of in-creased vertical wind shear and low-level moistureover the Tasman Sea.

The second component of the investigation wasto apply a high resolution numerical weatherprediction model, suitably con®gured withrecently developed multi-phase cloud physics, tosimulate the life cycle of the supercell. A modelresolution of 1 kilometre was selected so that thekey boundary layer and convective processescould be adequately resolved. A supercell thun-derstorm was produced by the simulation. Themodelled hail was not quite as large as thatrecorded but nevertheless the simulated thunder-storm had a high ice content, with hail up to 8 cmdiameter predicted at the surface close to thelocation where the storm turned into a supercell. Aband of liquid precipitation was displaced to thewest of where it actually occurred. Signi®cantly,the modelled surface hail swath was very close tothe radar identi®ed storm track. The timing ofonset of the initial convection and arrival time ofthe main convective cell at Sydney Airport waswithin a couple of hours of reality. Furtherinvestigations are planned, incorporating morecomprehensive data analysis techniques and theuse of synthetic data in an attempt to re®ne themodel simulations. The ultimate aim is to run theHIRES system operationally in real-time andcomputational demands suggest that this will bepossible in about two to three years.

Acknowledgements

The efforts of Lixin Qi and the Bureau of Meteorologydrafting section in assisting with the preparation of the®gures is sincerely appreciated. This research was partlyfunded by The Australian Research Council (ARC) LargeGrant Scheme. Two of the authors (LML and YW) aresupported by the US Of®ce of Naval Research Grant No.N00014-94-1-0556.

References

Buckley BW, Leslie LM (2000) Sudden temperaturechanges in the Sydney Basin: climatology and case stu-dies during the Olympic months of September and Octo-ber. Int J Climatol 20: 417±441

Bureau of Meteorology (1997) NSW Severe ThunderstormDirective. Bureau of Meteorology, Darlinghurst, Australia

Bureau of Meteorology (1995) The January-21-1991 SydneySevere Thunderstorm. Bureau of Meteorology, Darlin-ghurst, Australia

Davis C, Warner T, Astling E, Bowers J (2000) Developmentand application of an operational, relocatable, meso-gamma-scale weather analysis and forecasting system.Tellus 51A: 710±727

Guo Y-R, Kuo Y-H, Dudhia J, Pasons D, Rocken C (2000)Four-dimensional variational data assimilation of hetero-geneous mesoscale observations for a strong convectivecase. Mon Wea Rev 128: 619±643

Henri C (1999) Insurance Council of Australia (personalcommunication)

Jewett BF, Wilhelmson RB (1996) Numerical simulation ofsevere squall line thunderstorms. Preprints, 18th Confer-ence on Severe Local Storms. Am Meteor Soc, 268±272

Klemp JB, Wilhelmson RB (1978) The simulation of three-dimensional convective storm dynamics. J Atmos Sci 35:1070±1096

Kuo Y-H (2000) Mesoscale numerical weather prediction.Proc International Conference on Mesoscale ConvectiveSystems and Heavy Rain in East Asia, 1±7

Leslie LM, Purser (1997) A semi-Lagrangian NWP Modelfor real-time and research applications: Evaluation insimple- and multi-processor environments. Atmosphere± Ocean 35: 75±101

Leslie LM, Speer MS (1998a) Atmospheric particulatetransport modelling in a controlled burning event. MeteorAppl 5(1): 17±24

Leslie LM, Speer MS (1998b) Short-range ensemble fore-casting of explosive Australian east coast cyclogenesis.Wea Forecasting 13: 822±832

Leslie LM, LeMarshall JL (1997) The generation andassimilation of cloud drift winds in numerical weatherprediction. J Meteor Soc Japan 75: 1±11

Leslie LM, Skinner TCL (1994) Real-time forecasting of theWestern Australian summertime trough: Evaluation of anew regional model. Wea Forecasting 9: 371±383

Mass C, Kuo YH (1998) Regional real-time numericalweather prediction: Current status and future potential.Bull Amer Meteor Soc 79: 253±263

Mills GA, Colquhoun JR (1999) Objective Prediction ofSevere Thunderstorm Environments: Preliminary ResultsLinking a Decision Tree with an Operational RegionalNWP Model. Wea Forecasting 13(4): 1078±1092

Moller AR, Doswell III CA, Przybylinski R (1990) High-precipitation supercells: a conceptual model and documen-tation. Preprints, 16th Conference on Severe Local Storms,Kananaskis Park, Alberta. Am Meteor Soc, 52±57

Moncrieff MW, Green JSA (1972) The propagation ofsteady convective overturning in shear. Quart J RoyMeteor Soc 98: 336±352

The Sydney Hailstorm of April 14, 1999: Synoptic description and numerical simulation 181

Puri K, Dietachmeyer GD, Mills GA, Davidson NE, BowenRA, Logan LW (1998) The new BMRC Limited AreaPrediction System, LAPS. Aust Meteor Mag 47(3): 203±223

Speer MS, Leslie LM (1998) Quantitative precipitation fore-casting for an extreme mesoscale ¯ash ¯ood event. Proc 16thConf on Weather Analysis and Forecasting, AMS AnnualMeeting, 11±16 January 1998, Phoenix, Arizona, 144 pp

Wang Y (1999) Cloud parameterisations for high resolutionnumerical modelling. Bureau of Meteorology, PO Box1289K, Melbourne Vic 3001 Australia. BMRC ResearchReport No 74

Zhang D-L, Fritsch JM (1986) A case study of the sensitivityof numerical simulation of mesoscale convective systems.Mon Wea Rev 114: 2418±2431

Authors' addresses: Bruce W. Buckley, The WeatherCompany, L 2, 7 West St, North Sydney, NSW 2060Australia (E-mail: bbuckley@theweather.com.au); LanceM. Leslie, University of New South Wales Sydney, Austra-lia; Yuqing Wang, Department of Meteorology, Universityof Hawaii, Hawaii, USA

182 B. W. Buckley et al.: The Sydney Hailstorm of April 14, 1999: Synoptic description and numerical simulation