FireweatherclimatologydatasetforVictoriaSupplementalReportforDatasetVersion2

SupplementalReporttotheVictoriaDepartmentofEnvironment,LandandWaterPlanning

June2016(AppendixArevisionAugust2016)

Timothy Browna,b, Graham Millsb, Sarah Harrisb, Domagoj Podnara, Hauss Reinbolda andMattFearonaaDesertResearchInstitute,RenoNevadaUSA,tim.brown@dri.edubMonashUniversity,ClaytonVictoriaAustralia,sarah.harris@monash.edu

i

EXECUTIVESUMMARYSincethereleaseoftheoriginaldataset,extensiveworkandevaluationwasundertakentoimprove the distributions for surface temperature, relative humidity and wind speed.Emphasis was given to improving the bias correction of the distribution tails. Quantilemapping was still used for this version, except that empirical cumulative distributionfunctionswereemployedratherthantheoreticaldistributions.Thisallowedforbetterdirectquantile matching between observations and model. Stricter quality control procedureswere used on the observational data to remove additional erroneous values discoveredafterfurthertesting.Specific methodology modifications were undertaken that provided substantialimprovementstothedataset.First,acarefulQCoftheobservationdatasetwasrequiredtoremove unrealistic or unlikely values, since these would heavily influence the empiricaldistributiontails.Second,itwasnecessarytomakesurethatobservationlocationswerenotrelatedtomodelwatergridpointssincethesewouldbiasthecorrection.Third,becauseofdiurnal,seasonal,andlocalphysicalcharacteristicsoftheobservationstations(e.g.,terrain),it was important to develop mapping functions for each hour by month. Fourth, muchemphasis in assessing the correction was given the distribution tails, since this region isquiteimportantforbushfireanalyses.Fifth,thespatialinterpolationofcorrectionacrossthegrid required a balance of weighting the station grid point to account for localcharacteristicsanditsinfluenceonneighbouringstations.This supplemental report describes the methods used to generate Version 2 of the Fireweather climatology dataset for Victoria. Version 2 replaces Version 1, and users shouldonly work with this new version of the dataset given the substantial improvements thatwere made, particularly in the reduction of bias at the highest wind speeds and lowestrelative humidity extremes of their distributions, and consequently substantially higherextremevaluesoftheForestFireDangerIndex(FFDI).

TABLEOFCONTENTS

EXECUTIVESUMMARY...................................................................................................................i

1 BACKGROUND...........................................................................................................................12 METHODS...................................................................................................................................12.1 BiasCorrection...............................................................................................................................2

3 RESULTS...................................................................................................................................143.1 VERSION2OFTHEDATASET.................................................................................................143.2 CLIMATOLOGY..............................................................................................................................14

4 SUMMARYANDCONCLUSION...........................................................................................165 DATASTORAGEANDACCESS............................................................................................17

6 ACKNOWLEDGEMENTS.......................................................................................................18

7 REFERENCES...........................................................................................................................188 APPENDICES...........................................................................................................................19AppendixA:TableofAWSstationsusedinthebiascorrectionanalysis...........................19AppendixB.Datasetdescription:FireWeatherClimatologyforVictoria..........................21

1

1 BACKGROUNDClimatologydataoffireweatheracrossthe landscapecanprovidescience-basedevidencefor informingstrategicdecisionstoamelioratetheimpacts(attimesextreme)ofbushfiresoncommunitysocio-economicwellbeingandtosustainecosystemhealthandfunctions.Along-termclimatology requires spatialand temporaldata thatareconsistent to representthe landscape in sufficient detail to be useful for fire weather studies andmanagementpurposes.Toaddressthis inhomogeneityproblemforanalysesofavarietyoffireweatherinterestsand toprovideadataset formanagementdecision-support, ahomogeneous41-year (1972-2012), hourly interval, 4 km gridded climate dataset for Victoria has beengenerated using a combination of mesoscale modelling, global reanalysis data, surfaceobservations, and historic observed rainfall analyses. Hourly near-surface forecast fieldswere combined with Drought Factor (DF) fields calculated from the Australian WaterAvailabilityProject(AWAP)rainfallanalysestogeneratefieldsofhourlyfiredangerindicesfor each hour of the 41-year period. A quantilemapping (QM) bias correction techniqueutilizingavailableobservationsduring2003-2012wasusedtoameliorateanymodelbiasesin wind speed, temperature and relative humidity. Extensive evaluation was undertakenincluding both quantitative and case study qualitative assessments. The final datasetincludes 4-km surface hourly temperature, relative humidity,wind speed,wind direction,Forest FireDanger Index (FFDI), anddaily drought factor (DF) andKeetch-ByramDroughtIndex(KBDI),anda32-levelfullthree-dimensionalvolumeatmosphere.Brownetal. (2015)describes indetailVersion1of thedataset.Thisdocumentshouldbereferred to forotherdetailsabout theprojectandmethods,aswellas importantcaveatswhenusingthedataset.

2 METHODSFigure 1 shows the process flowchart for creation of the Victoria gridded climatologydataset. TheWRFmodel is initializedwith global reanalyseswith the primary outputs ofsurface temperature, relative humidity and wind speed, along with numerous otherelements and the full 3-D atmospheric volume. From these data empirical cumulativedistribution functions (ECDF)weredetermined forgridpointscorresponding toAutomaticWeather Station (AWS) hourly observations. ECDFswere also computed for each station.Stricter quality control (QC)methodswere applied to the observed data than in the firstversion. Next, quantile mapping (QM) and bias correction was done for each stationlocation,andspatiallyinterpolatedacrossthegrid.Allofthesestepsaredescribedfurtherbelow.

2

Fig.1.ProcessflowchartforcreationoftheVictoriagriddedclimatologydataset.

2.1 BiasCorrectionAstatisticalbias(meanerrorofpredictedminusobserved)fromatmosphericmodelsisnotuncommonduetocombinationsofphysicsparameterizations,spatialresolutionandinputdata(initialboundaryconditionsincludingmodeltypeanddataassimilation).Thebiasmayrunconsistentlyaboveorbelowameanobservedvalue,andthuscanbeaccountedforinthefinalmodeloutputusingastatisticalcorrection.However,atmosphericmodelbiascanalsobeafunctionofseasonandhourofthedayduetobothmodelphysicsandthe localcharacteristics of the station siting (e.g., elevation, aspect), requiring a more nuancedstatistical correction approach. Figure 2 highlights the diurnal and seasonality oftemperaturebias(°C)fortwoAutomaticWeatherStations(AWS)inVictoria,Essendon(top)and Bairnsdale (bottom). For Essendon, the overall magnitude of the bias ranges fromapproximately -0.5 to +2.0°C, though nearly all months and hours have a positive bias.Bairnsdale shows amuch greater range of bias from approximately -2.75 to +2.1°C. Theseasonalityisespeciallyhighlightedwithacoolbiasduringthecoolseason(May-November)for the hours around 0800 through 2100 UTC, which are the local time night and earlymorning hours. But during thewarm season, the highestwarm bias is during 0000-0300hours local time. These examples clearly highlight the need for hourly and seasonal biascorrection.

WRF

Globalreanalysis Observa2ons QCoutliers/grosserrors

EmpiricalCDFdataatsta2onloca2ons(byhour,bymonth)

Quan2lemappingatsta2onloca2ons

Biascorrec2onatsta2onloca2ons

Interpolatesta2onnetworktoregularWRFgrid

CorrectedWRFgrids

AWAPrainfall

Droughtfactor,KBDI FFDI

T RH W

EmpiricalCDFdata(byhour,bymonth)

Victoriagriddeddatacrea,onprocess

SurfaceT,RH,W3Dvolume

3

Fig.2.Hourly (UTC)andmonthly temperaturebias (°C) forEssendon (top)andBairnsdale(bottom).Thecoloredsymbolsandlinescorrespondtoaparticularmonth.Quantile mapping (QM) for statistical adjustment of bias has been an acceptedmethodology for many years (e.g., Panofsky and Brier 1968), and has been used fornumerousglobalclimatemodelprojectionbiascorrections(e.g.,Maureretal.,2010,2014;Thrasher et al., 2012). QM adjusts a model value by mapping quantiles of the modeldistribution onto quantiles of the observation distribution. Figure 3 provides a visualschematic of the QM process (adapted from Pierce et al., 2015). The blue line shows ahypothetical normal distribution cumulative distribution function (CDF) of observedtemperaturesandtheircorrespondingquantiles.TheredlinerepresentsahypotheticalCDF

4

of model temperatures and their corresponding quantiles. The black horizontal linehighlights the 0.2 quantile value. The upward blue dashed line indicates that 10°C is themodel0.2quantile.The leftpointingshortarrowshowswhere the0.2quantile intersectsthe observed temperature value, and the downward pointing arrow shows the observedvaluethatshouldbeusedasthebiascorrectedvalue.Inthiscase,theWRF10°Cbecomes8.9°C.

Fig.3.Exampleillustrativeschematicofthequantilemappingmethodology.Ourbiascorrectionprocesswasamultistepapproach(Fig.4).Thefirststepwastogatherhourly AWSobservations for stations in and borderingVictoria. Data quality control (QC)wasappliedbyfirstcheckingforobviousunrealisticvalues(i.e.,negativerelativehumidityandwindspeed; relativehumidity>100%).Next,anexploratorydataanalysis typecheckwasapplied to remove temperatureandwind speedoutliers thatwere likely inerror.Aninterquartilerangewascalculatedbasedonthe20thand80thpercentiles,andmultipliedby3;thisvaluewasthenaddedtothe80thpercentilenumberprovidingathresholdforwhichanyobservationexceedingthisvaluewasremoved.Thisremovedextremevaluesthatweredeemedanobservationerror. Checkingof results from thismethod revealed that indeedquestionable values were properly flagged and subsequently removed. As an additionalprecaution for a temperature check, we checked that no station exceeded the officialBureau of Meteorology maximum and minimum state records. This process yieldedmaximumandminimumvaluesallowablebyhourandmonthforeachstation.Onlystationsthathadhourlydata forat least10yearswereusedyielding75stationsavailable for theQM process (Fig. 5). Appendix A provides the list of stations corresponding to the mapstationindex.SteptwowassimplytomatchtheAWSlocationtothenearestWRFlandgridpoint.TestingshowedthatusingwatergridpointsintroducedanomalousvaluesintheQM;therefore,itisimportanttouseonlylandgridpointsforthecumulativedistributionfunctions.Step three was the computation of the climatological empirical cumulative distributionfunctions (ECDF) for each AWS and corresponding matched WRF grid point. Theoretical

5

distributions(i.e.,normal,beta,Weibellfortemperature,relativehumidityandwindspeed,respectively)wereoriginallytested,but itwasdeterminedthatthesedidnotsatisfactorilyprovide estimates in the distribution tails during the mapping process. ECDFs weredeterminedforeachhourforeachmonth(288functionseachforobservationandmodel).Forexample,ifastationhad16yearsofhourlydata,thenanECDFwascomputedforhour00overallofthe16yearsofJanuary.TheRsoftwarepackage(RCoreTeam,2013)wasusedtoempirically fit thedataanddetermine thequantile values. Figure6 showsanexampleECDF for January 0400 UTC for Melbourne airport. The red (observed) and blue (WRF)curvesarefortheperiod2003-2012.Thepurple(observed)andyellow(WRF)curvesarefortheperiod1972-2002.Thus,thesecurvesshowboththedifferencebetweentheobservedandWRF ECDFs and between themost recent decade versus the previous 31 years. Theobserved early period is cooler than the recent; this is partially reflected in the highertemperaturesforWRFstartingaround0.6quantile.

Fig.4.Flowchartforthebiascorrectionprocess.

QCobserva2onsandorganizedataRemoveunrealis2cvaluesEDAcheckforoutliersApplystatewiderecordtempboundaryDeterminedhourlybymonthmax/minvaluesforeachsta2on’selements

Victoriagriddeddatabiascorrec,onprocess

Mapsta2ondatatoclosestWRFlandgridpoint

ComputeempiricalCDFforeachsta2on(OBS)andassociatedWRFgridpoint

MatchCDFvaluesbetweenOBSandWRF(quan2lemapping)

ReplacemodelvaluewithCDFmatchedsta2onvalue

Createdifferencedatabetweenoldandnewmodelvaluesatallsta2onloca2ons

Spa2allyinterpolatedifferencestoen2reWRFgridusingIDW(1.5)

Subtractinterpolatedgridfromoriginalgridtocreatebiascorrectedgrid

QCthecorrectedgrid

BiascorrectedgridT,RH,W

Requirethatmax/minvaluescannotexceedthemax/minsta2onobservedforthatmonth/hour

AdjustCDFduevaryinglengthObs.records

6

Fig.5.AWSlocationsforhourlystationsusedinthedevelopmentofthequantilemappingforbiascorrection.Thenumbersindicateastationindex(seeBrownetal.,(2016)forstationinformation). Appendix A provides the list of stations corresponding to the map stationindex.

7

Fig 6. Example empirical cumulative distribution function for Melbourne airport January0400 UTC. The red (observed) and blue (WRF) curves are for the period 2003-2012. Thepurple(observed)andyellow(WRF)curvesarefortheperiod1972-2002.Thenextstepsaretheactualbiascorrection.Eachofthesestepswasappliedtoallavailablehours of station data. In step 5, the model value was replaced with the ECDF matchedstationvaluepertheprocessdescribedforFigure4.Thisstepcreatednewmodelvaluesforeachstationlocation.Step6createdadifferencedatasetbetweentheoldandnewmodelvalues at all station locations. This allowed for spatially interpolating a difference fieldacrossthe4-kmgrid(step7).The“dsgrid2”inversedistanceweightingspatialinterpolationalgorithm with a power coefficient of 1.5 in the NCAR Command Language (NCL, 2012)softwarepackage Inversedistanceweighting (IDW)wasused for the spatial interpolationmethod.Thepowercoefficientof1.5wassubjectivelychosenbaseduponexaminingoutputmaps. It was felt that 1.5 allowed for reasonable weighting of nearby stations, but stillretainedsufficientlocalinformation.Higherpowercoefficientvalueslocalizetheweighting,while lower valuesdistribute station influencemorewidely. Step8 simply subtracted the

8

interpolatedgridfromtheoriginalgridtocreatethebiascorrectedgridforthathour.Figure7showsexamplemapsoftwodifferencegridsrepresentinghours0500and1700localtimefor 7 February 2009. Note in this example that the earlymorning hour of 0500 requiresgreatercorrectionsespeciallyinthecomplexterrainregionthanforthelateafternoonhourof 1700. This is primarily due to themodel’s nighttimephysics and the local topographicinfluencesofcomplexterrain.

Fig. 7. Spatial interpolation examples from the quantile mapping bias correction for 7February20090500localtime(top)and1700localtime(bottom).ShadedvaluesareWRFuncorrectedgridpointvaluesminusWRFbiascorrectedgridpointvaluesfortemperature(°C).

9

The final step in the process was a check of the final grid requiring that themaximum/minimumvaluesofthenewgriddidnotexceedthemaximum/minimumvaluesofstationobservedvalues.Thisstepkeptthebiascorrectedvaluesasestimatesacrossthegrid from exceeding the known observed values. It certainly is possible that locations inbetween stations could achieveobserved values exceedingobserved, but given there arenot observations everywhere, itwas felt that this conservative approach still allowed forrealisticclimatologicalrepresentationswhilekeepingthemodelvalueswithinrangesknownmeasurements.The results of thebias correction canbepartially assessedby examining example stationboxplots and scatterplots. Figure 8 (top) shows an example for the January 2003-2012hourly (UTC) distribution of WRF minus observed temperature (°C) for the Melbourneairportstationbeforebiascorrection.Overall, thebias isquite low(approximately±0.5°C.The outlying points are due to likely due to a timing issue of fronts and other localizedcirculation patterns or phenomena thatWRFmissed, and these cannot be corrected forstrictly from thebias correctionprocess. Thebottomplotof Fig.8 shows thedistributionafter the bias correction. Though not all medians fall on the zero line, the interquartiledistribution ismore centered around zero degrees. Themajority of the outliers areWRFover-predictingtemperature.

10

Fig. 8. Boxplots of January 2003-2012 hourly (UTC) distribution of WRF minus observedtemperature(°C)forMelbourneairportbeforebiascorrection(top)andWRFbiascorrectedminusobserved(bottom).Figure 9 (top) shows the original bias for Melbourne airport 2003-2012 January relativehumidity(%).Themedianbiasisgenerallysmall,withtheexceptionhours15-19UTC,whichare earlymorning hours in local time. The bias corrected values are also shown in Fig. 9(bottom), and show good centering of the interquartile range around zero with medianvaluesof 0-1%. Similar to temperature, thereare someoutliers from frontal timing, localprecipitationevents,orother factors forwhich thecorrectionmethodcannotadjust. ThemajorityoftheseoutliersareWRFunder-predictingrelativehumidity.

11

Fig. 9. Boxplots of January 2003-2012 hourly (UTC) distribution of WRF minus observedrelative humidity (%) for Melbourne airport before bias correction (top) and WRF biascorrectedminusobserved(bottom).Figure10(top)showstheoriginalbiasforMelbourneairport2003-2012Januarywindspeed(knots). Themedian original bias is in the 0-2 knot range, though overallWRF tended tounder-predictwindspeed.SimilartoFigs.8and9,thereareoutliersrelatedtoothermodelandobservation factors beyond just bias. Figure 10 (bottom) showsoverall improvementfromthebiascorrection.Thoughthere isstillaslightunder-prediction,themedianvaluesarenow0-1knots,andtheoveralldistributionismorecondensed.

12

Fig. 10. Boxplots of January 2003-2012 hourly (UTC) distribution ofWRFminus observedwind speed (knots) for Melbourne airport before bias correction (top) and WRF biascorrectedminusobserved(bottom).Figure11showsexamplescatterplotsforWRFpredictedversusobservedtemperatureforJanuary0000UTC.ThebluecirclesshowbeforeandtheredcirclesafterQMbiascorrection,respectively.Figure11(top)isforFallsCreekwherebothscatterplotsshowawell-definedlinearrelationship,butthebeforecorrectionscatterclearlyhighlightsWRFover-predictingtemperature. The QM bias correction shows the values nicely aligned along the 45°diagonal.As shown in thepreviousboxplotexamples, there remain somepoints inwhichtheerrorscouldnotbesubstantiallyreduced.Figure11(middle)showsJanuary0000UTCscatterplots for Horsham Aerodome. In this case there is little difference between thepredicted versus observed points before and after correction, indicating that little QMcorrectionwasneeded tobeginwith.As a final example, Fig. 11 (bottom) shows January0000UTCscatterplotsforMountHotham,wheretherewasaclearneedforbiascorrection,andseveralpointsshowlargeerrorsbetweenpredictedandobserved.Manyoftheselargererrors were reduced after correction, and generally the points align better along thediagonal.

13

Fig.11. ScatterplotsofWRFpredictedversusobserved temperature (°C) for Januaryhour0000UTC for the years 2003-2012 for the stations Falls Creek (top), HorshamAerodome

14

(middle), and Mount Hotham (bottom). Blue and red circles are before and after,respectively,biascorrection.

3 RESULTS3.1 VERSION2OFTHEDATASETThisversionofthedatasetincludeshourlytemperature,relativehumidity,windspeed,winddirection,andFFDI,anddailyDFandKBDIata4-kmspatialresolutionfortheperiod1972-2012. The bias correction method described in Section 2.1 above was applied totemperature,relativehumidity,andwindspeed.BecauseofqualityissueswithprecipitationdiscussedBrownetal.(2015),AWAPdatawereusedinsteadtocalculatedailyDFandKBDIandthenthesewereused(incombinationwithbiascorrectedhourlytemperature,relativehumidity,andwindspeed)tocalculatehourlyFFDI.

3.2 CLIMATOLOGYNumerous climatology analyses can be undertakenwith this dataset, both for diagnosticandsummaryassessments.Hereweshowafewexamples.Thehomogeneous4kmgridinthedatasetallowsforavarietyofannual,seasonal,monthly,daily, and hourly spatial analyses. Figure 12 shows the average January temperature for0600UTC(upperleft)and1800UTC(lowerleft),andcorrespondingrelativehumidityintheupper right and lower right, respectively. These times represent approximately the dailymaximumandminimumvaluesofthediurnalcycleduringJanuary.Thereisacleardiurnalcycle, with the topography modifying the absolute values of temperature, and a strongmodification of the airmass south of the ranges by the effects of land-sea contrast. Therelativehumiditydistributionanddiurnalvariationapproximatelyfollowsthetemperaturedistributionduetothenegativerelationshipofrelativehumiditywithtemperature,buttheeffectsoftopographyaresomewhatdifferent,probablyduetothenormalhydrolapse.

15

Fig.12.AverageJanuarytemperatureandrelativehumidityforhour0600UTC(topleftandright) and hour 1800 UTC (bottom left and right). The geographic boundaries representBushfireRisklandscapeboundariesusedbyDELWP.Figure 13 shows December-February all-time maximum temperature (left map) andminimumrelativehumidity(rightmap)fortheyears1972-2012.Thehighesttemperaturesareseeninthenorthwestportionofthestate,andthecoolest inthehigherelevationsoftheGreatDividingRangeintheEast,andshowsthesharptemperaturecontrastalongthecoast. For most of Victoria, the all-time lowest humidity values are below 15%. It isinterestingthatthereislessvariationacrossthestateintheseextremevaluesthanthereisin the average 0600 UTC (January) fields shown in Fig. 12, indicating that even in therelativelybenignclimates,suchasthatofeastGippsland,veryhightemperaturesandverylowrelativehumiditiescanoccur.ComparingFig.13with theequivalentgeographicaldistributions forversion1of thedataset (Fig.33ofBrownet al 2015) shows relatively subtledifferences for the temperatures,butinVersion2,thelowestrelativehumidityisloweroveralmostthewholestatethanwasthat of Version 1. This indicates the reduction in moist bias seen at the lowest relativehumidityvaluesinVersion1.

Fig. 13. December-February all-time maximum temperature (left map) and minimumrelativehumidity(rightmap)fortheyears1972-2012.

16

Figure 14 shows the 1972-2012 December-February maximum FFDI at each gridpoint..There is some small-scale variability in the northwest, and understanding this requiresfurtherinvestigation,butitappearstobeassociatedwithveryisolatedhighwindspeedsintheWRFdataset,togetherwiththeextremesensitivityoftheFFDItovariationsinitsinputswhenthevaluesreachhighlevels.Onabroaderscale,thehighestFFDIvaluesareseeninthenorthwesternportionofVictoria,extendingeastwards tocentralVictorianorthof theranges, and southwards in the farwestof the state.TherearealsonotableareasofhighvaluessouthofthecentralhighlandsjustwestofMelbourneandextendingwestwardalongtheBarwonValley,andalsoareasofrelativemaximuminwestandcentralGippsland–boththese areas appear associatedwith lee effects of the ranges. The lowest values are seenacross the Great Dividing Range in the east where elevation effects keep temperatureslower,andinfar-eastGippsland.Comparing thesevalueswith those fromVersion1of thedataset (Fig.34ofBrownetal2015) shows similar patterns, but with generally higher values, in Version 2 (note thediffering data ranges in the two figures). This is the effect of the new QM betterrepresenting the high wind speed and low relative humidity tails of the respectivedistributions, with consequent effects on the resulting FFDI values. There are also somedifferencesbetween thepatterns along the coastline,wheregreater attention to station-gridpoint interpolation issues has led to changes that provide a more realisticrepresentation.

Fig.14.December-FebruaryhighestvalueofFFDIfortheyears1972-2012fromthegriddedclimatology.

4 SUMMARYANDCONCLUSIONVersion2ofthedatasetissubstantiallyimprovedoverVersion1,andthusreplacesVersion1.TheimprovementswereduetostricterQCandthresholdsintheobservationaldata,andapplyingempiricalcumulativedistributionfunctionsforthequantilemapping.Thisallowedfor improved bias correction in the tails of the distribution, and thus have better

17

representedextremesinboththeindividualparametersandalsointhederivedFFDIvalues.This is important for fire analyses since it is extreme fire events that are often of mostinterest.

5 DATASTORAGEANDACCESSAtthetimeofwritingthisreport,thecompletedatasetisstoredatMonashUniversitywitha backup and at DRI. Two file groups are available, the original uncorrected fields for alllevels,andbiascorrectedsurfacefieldsbasedonthequantilemappingmethoddescribedinthisreport.ThefileformatisnetCDF,whichcontainsrelevantmetadataaspartofthedatastructure.Derived fields include the Drought Factor (DF), Keetch-Byram Drought Index (KBDI), andForest FireDanger Index (FFDI). Daily precipitation from theAustralianWater AvailabilityProject(AWAP)datasetwasusedfortheDFandKBDIcalculations.FFDIwascalculatedusingthese outputs along with temperature, relative humidity and wind speed from theWRFdataset.ThesederivedfieldsarealsointhenetCDFformat.The full dataset (including theupper-air fields) is approximately 53TB in size. The surfacefieldscommonlyusedforbushfireanalyses (e.g., temperature,relativehumidityandwindspeed)aremuchsmaller, and formatted innetCDFallowing fordirect input intoPhoenix.Metadatadescriptionforthedatasetisavailableuponrequest.Questionsandcommentsregardingthedatasetmaybedirectedto:Dr.TimBrown;tim.brown@dri.edu,orDr.SarahHarris;sarah.harris@monash.edu

18

6 ACKNOWLEDGEMENTSThisprojectwouldnothavebeenpossiblewithout the support andencouragement fromLiamFogarty at theDepartment of Environment,Water, Land andPlanning (DEWLP) andAlenSlijepcevic (CountryFireAuthority; formerlyDEWLP).WeverymuchappreciateAndyAckland at DEWLP for testing output data in Phoenix. This work was funded under theVictoria Department of Sustainability and Environment Project 92c – Weather ResearchForecastingModel–VictoriaFireClimatologyGriddedDataset.

7 REFERENCESBrown, T.,G.Mills, S.Harris,D. Podnar,H. Reinbold, andM. Fearon, 2015: FireWeatherClimatolology Dataset for Victoria. Final Project Report to the Victoria Department ofEnvironment,LandandWaterPlanning,August2015,134pp.

19

8 APPENDICES

Appendix A: Table of AWS stations and years used in the bias correctionanalysis.

MapID Stationname

StationID

Startyear

Endyear

#ofyears

0 CANBERRAAIRPORT 70014 1996 2010 15.01 WAGGAWAGGAAMO 72150 1996 2013 17.32 MILDURAAIRPORT 76031 1996 2013 17.53 EASTSALEAIRPORT 85072 1996 2013 16.74 ESSENDONAIRPORT86038 2003 2013 10.55 MELBOURNEREGIONALOFFICE 86071 1997 2013 13.26 MOORABBINAIRPORT 86077 1996 2013 17.37 MELBOURNEAIRPORT 86282 1996 2013 17.68 LAVERTONRAAF 87031 1996 2013 16.99 MANGALOREAIRPORT 88109 1996 2013 17.010 BALLARATAERODROME 89002 2000 2013 12.911 CAPEOTWAYLIGHTHOUSE 90015 1996 2013 17.212 BRAIDWOODRACECOURSEAWS 69132 1996 2013 17.013 COOMAAIRPORTAWS 70217 1996 2013 17.214 BOMBALAAWS 70328 1996 2013 16.915 BENDIGOAIRPORT 81123 1996 2013 16.916 CERBERUS 86361 1996 2013 15.517 RHYLL 86373 1996 2013 17.218 GROVEDALE(GEELONGAIRPORT) 87163 1996 2011 15.619 SHEOAKS 87168 1996 2013 16.720 HAMILTONAIRPORT 90173 1996 2013 16.521 PORTFAIRYAWS 90175 1996 2013 17.322 MORTLAKERACECOURSE 90176 1996 2013 16.723 BEGAAWS 69139 1996 2013 16.724 GOULBURNAIRPORTAWS 70330 1996 2013 17.025 ALBURYAIRPORTAWS 72160 1996 2013 17.126 WANGARATTAAERO 82138 1996 2013 17.027 MALLACOOTA 84084 1996 2013 16.728 COMBIENBARAWS 84143 1996 2013 16.929 COLDSTREAM 86383 1996 2013 17.130 AVALONAIRPORT 87113 1996 2013 16.931 WALLAN(KILMOREGAP) 88162 1996 2013 15.432 PORTLAND(CASHMOREAIRPORT) 90171 1996 2013 16.933 CAPENELSONLIGHTHOUSE 90184 1997 2013 16.134 TUGGERANONG(ISABELLAPLAINS)AWS 70339 1996 2013 16.935 CABRAMURRASMHEAAWS 72161 1996 2013 16.0

20

36 KHANCOBANAWS 72162 1996 2013 15.937 SHEPPARTONAIRPORT 81125 1996 2013 16.838 HUNTERSHILL 82139 1996 2013 17.039 MOUNTBULLER 83024 1996 2013 14.040 FALLSCREEK 83084 1996 2013 16.641 MOUNTHOTHAM 83085 1996 2013 16.042 GELANTIPY 84142 1996 2013 17.143 MOUNTNOWANOWA 84144 1996 2013 17.344 MOUNTBAWBAW 85291 1996 2013 14.245 MOUNTMOORNAPA 85296 1996 2013 17.046 SCORESBYRESEARCHINSTITUTE 86104 1996 2013 16.647 FRANKSTONAWS 86371 1996 2013 17.348 FERNYCREEK(DUNNSHILL) 86372 1996 2011 15.549 EILDONFIRETOWER 88164 1996 2013 17.150 COONAWARRA 26091 1997 2013 14.951 HAYCSIROAWS 75175 1996 2007 10.152 WALPEUPRESEARCH 76064 1998 2013 14.553 SWANHILLAERODROME 77094 1997 2013 15.954 LONGERENONG79028 1997 2013 16.055 STAWELLAERODROME 79105 1996 2013 17.256 POINTWILSON 87166 1996 2008 11.657 REDESDALE 88051 1997 2013 14.458 LOOKOUTHILL 89105 1996 2007 10.959 AIREYSINLET 90180 1996 2013 17.360 MERIMBULAAIRPORTAWS69147 1998 2013 14.961 THREDBOAWS 71032 1998 2013 14.862 YANCOAGRICULTURALINSTITUTE 74037 1999 2013 13.463 YARRAWONGA 81124 2002 2013 11.064 RUTHERGLENRESEARCH 82039 1998 2013 15.165 DINNERPLAIN(MOUNTHOTHAMAIRPORT)83055 2000 2013 12.866 ORBOST 84145 2000 2013 12.667 BAIRNSDALEAIRPORT 85279 1997 2013 15.268 MORWELL(LATROBEVALLEYAIRPORT)85280 1997 2013 16.069 VIEWBANK(ARPANSA) 86068 2000 2013 13.070 DENILIQUINAIRPORTAWS 74258 1997 2013 16.071 NHILLAERODROME 78015 2003 2013 10.172 HORSHAMAERODROME 79100 1997 2013 15.373 COLAC(MOUNTGELLIBRAND)90035 2000 2013 12.874 WARRNAMBOOLAIRPORTNDB 90186 1998 2013 14.9

21

AppendixB.Datasetdescription:FireWeatherClimatologyforVictoriaReadmeversiondate:22June2016Datasetversion“June2016”.ThisdatasetreplacesVersion1.Substantialimprovementstothedatasetweremade,especiallyinthedistributiontailsdueto1)modificationsofthequantilemappingbiascorrectionmethod,and2)additionalqualitycontrolcheckingonthesurfacestationobservationsusedforthebiascorrection.ThisdatasetincludesthesurfaceelementsproducedfromtheFireWeatherClimatologyforVictoriaprojectdocumentedinthereportbyBrownetal(2015;2016).Pleaseseethatreportforthedetailsoftheconstructionandapplicabilityofthesedata.ThedatasetperiodisJanuary1972–December2012.Producedusingmodelversion“WRFv3.5.1”Thetemporalscaleishourly.Thehorizontalresolutionis4x4km.Twofilegroupsareavailable,originaluncorrectedfieldsandbiascorrectedfieldsbasedonthequantilemappingmethoddescribedinthereport.ThefileformatisnetCDF,whichcontainsrelevantmetadataaspartofthedatastructure.DerivedfieldsincludetheDroughtFactor(DF),Keetch-ByramDroughtIndex(KBDI),andForestFireDangerIndex(FFDI).DailyprecipitationfromtheAustralianWaterAvailabilityProject(AWAP)datasetwasusedfortheDFandKBDIcalculations.FFDIwascalculatedusingtheseoutputsalongwithtemperature,relativehumidityandwindspeedfromtheWRFdataset.ThesederivedfieldsarealsointhenetCDFformat.Thedatasetwasconstructedprimarilyforclimatologicalanalyses;hourlydatashouldbeexaminedcarefullyforspecificcasestudyanalyses.WRFsurfaceoutputinthisdataset:Variable=Temperature(T)FillValue=-32767.0projectionType="MERCATOR"level="SFC"units="C"gridType="SCALAR"lonCentre=145.4

22

Variable=RelativeHumidity(RH)FillValue=-32767.0projectionType="MERCATOR"level="SFC"units="%"gridType="SCALAR"lonCentre=145.4Variable=Windspeed(WSPD)FillValue=-127.0projectionType="MERCATOR"level="SFC"units="kts"gridType="VECTOR"lonCentre=145.4Variable=Winddirection(WDIR)FillValue=-127.0projectionType="MERCATOR"level="SFC"units="degrees"gridType="VECTOR"lonCentre=145.4Variable=Precipitation(PPT)FillValue=-32767.0projectionType="MERCATOR"level="SFC"units="mm"gridType="SCALAR"lonCentre=135.0Whenusingthesedata,pleasecitetheBrownetal(2015;2016):Brown,T.,GMills,S.Harris,D.Podnar,H.Reinbold,andM.Fearon,2015:FireweatherclimatologydataforVictoria.Finalprojectreport,July2015.Brown,T.,GMills,S.Harris,D.Podnar,H.Reinbold,andM.Fearon,2016:FireweatherclimatologydataforVictoria.SupplementalreportforDatasetVersion2,June2016.Forquestionsorcomments,pleasecontact:TimBrown,tim.brown@dri.eduOrSarahHarris,sarah.harris@monash.edu